WORLD BANK LATIN AMERICAN AND CARIBBEAN STUDIES 35348 Poverty Reduction Growth: and Virtuous and Vicious Circles Guillermo E. Perry · Omar S. Arias · J. Humberto López William F. Maloney · Luis Servén POVERTY REDUCTION AND GROWTH: VIRTUOUS ANDVICIOUS CIRCLES THE PRINCIPAL AUTHORS OF THIS BOOK ARE AS FOLLOWS: Chapter 1: Guillermo E. Perry, J. Humberto López, and William F. Maloney Chapter 2: William F. Maloney Chapter 3: J. Humberto López Chapter 4: J. Humberto López Chapter 5: J. Humberto López Chapter 6: J. Humberto López Chapter 7: William F. Maloney Chapter 8: Omar S. Arias Chapter 9: Omar S. Arias POVERTY REDUCTION AND GROWTH: VIRTUOUS ANDVICIOUS CIRCLES Guillermo E. Perry Omar S. Arias J. Humberto López William F. Maloney Luis Servén THE WORLD BANK Washington, D.C. © 2006 The International Bank for Reconstruction and Development / The World Bank 1818 H Street NW Washington DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org E-mail: feedback@worldbank.org All rights reserved 1 2 3 4 5 09 08 07 06 This volume is a product of the staff of the International Bank for Reconstruction and Development / The World Bank. The findings, interpretations, and conclusions expressed in this volume do not necessarily reflect the views of the Executive Directors of The World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgement on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this publication is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The International Bank for Reconstruction and Development / The World Bank encourages dissemination of its work and will normally grant permission to reproduce portions of the work promptly. For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance Center Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; telephone: 978-750-8400; fax: 978-750-4470; Internet: www.copyright.com. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, The World Bank, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2422; e-mail: pubrights@worldbank.org. ISBN-10: 0-8213-6511-8 ISBN-13: 978-0-8213-6511-3 eISBN-10: 0-8213-6512-6 eISBN-13: 0-8213-6512-0 DOI: 10.1596/978-0-8213-6511-3 Library of Congress Cataloging-in-Publication Data Poverty reduction and growth : virtuous and vicious circles / Guillermo E. Perry ... [et al.]. p. cm - (World Bank Latin American and Caribbean studies) Includes bibliographical references and index. ISBN-13: 978-0-8213-6511-3 ISBN-10: 0-8213-6511-8 1. Poverty-Latin America. 2. Latin America--Economic conditions--1945­ 3. Poverty--Government policy--Latin America. 4. Latin America--Economic policy. I. Perry, Guillermo. II. World Bank. III. Series. HC130.P6P72 2006 339.4'6098--dc22 2005057764 Cover art: Remedios Varo, "Spiral Transit" © 2005 Artists Rights Society (ARS), New York / VEGAP, Madrid. For more information on publications from the World Bank's Latin America and the Caribbean Region, please visit www.worldbank.org/lacpublications (o en Español: www.bancomundial.org/publicaciones). Contents Foreword xi Acknowledgments xiii Acronyms and Abbreviations xv Chapter 1: From Vicious to Virtuous Circles 1 Poverty as a multidimensional and dynamic concept 1 The twin disappointments: Destiny or choice? 2 The link from growth and development to income-poverty reduction 4 Closing the virtuous circle: The link from poverty to growth 5 Global convergence clubs 6 Does poverty matter for growth? 7 Regional convergence clubs 8 Household-level poverty traps 9 Implications of the report 11 Pro-growth poverty reduction 15 Notes 19 Chapter 2: Dimensions of Well-Being, Channels to Growth 21 Income poverty 21 Beyond income and consumption 27 Why not just ask them? 29 Snapshots vs. movies: Life-cycle welfare, mobility, and risk 31 Intergenerational mobility 37 Conclusion 40 Annex 2A: Estimating the monetary value of mortality changes 41 Annex 2B: A tractable welfare measure that captures income, mobility, and risk 42 Annex 2C: Intergenerational mobility in Latin America: Country comparison 42 Notes 42 Chapter 3: How Did We Get Here? 45 Per capita income in Latin America: A long-run comparative perspective 46 Long-run inequality 53 Notes 56 Chapter 4: The Relative Roles of Growth and Inequality for Poverty Reduction 57 The relative roles of growth and income distribution for poverty reduction 59 Growth and inequality: Bringing country specificity into the picture 63 Concluding remarks 70 Annex 4A: Testing for lognormality of income 71 Notes 72 v C O N T E N T S Chapter 5: Pro-Poor Growth in Latin America 75 Are all pro-growth policies equally pro-poor? 76 Does the composition of growth matter? 89 The role of taxes and transfers in reducing income inequality 92 Concluding remarks 100 Annex 5A: Simulating the impact of pro-growth policies on poverty 101 Notes 102 Chapter 6: Does Poverty Matter for Growth? 103 A poverty-traps view of the development process 104 Empirical evidence on poverty traps 108 What is the empirical evidence on poverty's impact on growth? 115 Transmissions channels from poverty to growth? 118 Concluding remarks 123 Annex 6A 124 Notes 126 Chapter 7: Subnational Dimensions of Growth and Poverty 129 What is spatial inequality, how is it measured, and what are the regional trends? 129 Identifying spatial concentration 130 Why do we observe regional convergence clubs? 135 Does migration work as an equilibrating mechanism? 138 The link back to growth and policy issues 139 Conclusions 143 Notes 143 Chapter 8: Microdeterminants of Incomes: Labor Markets, Poverty, and Traps? 145 The distribution of earnings: The role of worker endowments and labor markets 146 Microdrivers of changes in the income distribution 151 Determinants of income dynamics: Lessons from rural El Salvador 152 Implications for policies 159 Annex 8A: Data and methodological details 160 Notes 162 Chapter 9: Breaking the Cycle of Underinvestment in Human Capital in Latin America 165 The educational transition in the region: Slow and unbalanced progress 166 Poverty and human capital: A two-way relationship 167 Human capital formation: Sources of underinvestment traps 169 The educational ladder in Latin America: A persisting educational divide 171 Liquidity constraints, family factors, and educational investments: A sneak preview 178 The private value of schooling: How much does it pay? To whom? 181 Short-term or long-term poverty: Which is more pressing for schooling investments? 190 Implications for human capital formation policies 194 Investing now: The demographic window of opportunity 196 Annex 9A: Data and methodological details 197 Notes 199 Bibliography 203 Index 217 Boxes Chapter 2 2.1 Income poverty lines 23 2.2 National accounts and household surveys-based growth: How different are they? 25 2.3 Inflation inequality: What really happened to LAC poverty and inequality 26 2.4 Mobility and poverty traps 32 2.5 Is it inequality or risk? Maybe Latin America has less inequality than we thought 34 2.6 . . . Or maybe more: Inequality and demographics 35 vi C O N T E N T S Chapter 4 4.1 Decomposing poverty into growth and income distribution effects 60 4.2 The size distribution of income 64 4.3 Total growth elasticities of poverty and the efficiency of growth 65 Chapter 5 5.1 Trade policy and income risk 82 5.2 Taxes, transfers, and inequality 96 5.3 Conditional cash transfers in Colombia 98 Chapter 6 6.1 Education and technology 107 6.2 Is Latin America different? 117 Chapter 7 7.1 Tools to detect spatial association 131 7.2 Will trade liberalization increase regional disparities? NAFTA and Mexico 136 7.3 Trade-offs in regional policy: The Spanish experience 140 7.4 Rural roads and poverty reduction in El Salvador 142 Figures Chapter 1 1.1 Per capita income relative to the OECD, 1870­2000 2 1.2 Gini coefficient for Latin America, 1950­2000 2 1.3 Poverty rates in Latin America, 1950­2000 2 1.4 Low educational traps persist across generations among the poor and excluded 3 1.5 Although they stand to gain the most from education, poor people actually have low returns 3 1.6 Gini coefficients for market and disposable incomes 5 1.7 Indicators for poor and rich countries 7 1.8 Convergence clubs in life expectancy throughout the world 8 1.9 Poverty and investment throughout the world 8 1.10 Regional income dynamics in Brazil: The persistence of two convergence clubs 9 1.11 The sharp educational divide between the poor and the rich in Latin America 10 1.12 Total tax revenue versus per capita income, throughout the world 18 Chapter 2 2.1 Poverty in selected Latin American countries 22 2.2 The evolution of Latin American poverty during the 1990s 24 2.3 Gini coefficient for Latin America, 1950­2000 25 2.4 Income poverty profile for Bolivia: Self-rated by head of household versus data driven 30 2.5 Elasticity of son's income relative to father's income 38 2.6 Mobility indicators 39 Chapter 3 3.1 Per capita GDP for eight major Latin American countries, 1850­2000 48 3.2 Per capita growth and initial income levels in eight major Latin American countries 48 3.3 Cross-country dispersion of per capita GDP in Latin America, 1870­2000 49 3.4 Aggregate per capita income in Latin America, 1850­2000 49 3.5 Per capita income of five groups relative to the United States, 1850­2000 51 3.6 Incomes in Spain and peripheral Europe relative to OECD countries 51 3.7 GDP per capita in Latin America relative to several country groupings, 1850­2000 52 3.8 Latin American per capita GDP relative to Western Europe, 1500­2001 53 3.9 Income inequality in the United States and Spain, 1910­90 55 3.10 Income inequality in the United Kingdom and France, 1910­90 55 vii C O N T E N T S Chapter 4 4.1 Growth, inequality, and poverty reduction throughout the world 58 4.2 Decomposition of poverty into growth and distribution effect 60 4.3 Share of changes in poverty explained by growth and inequality 62 4.4 Share of changes in Latin American poverty explained by growth and inequality 63 4.5 Empirical and theoretical quintiles 67 4.6 Iso-poverty curves for headcount poverty 68 4.7 Mapping Latin American countries in the income inequality space 69 Chapter 5 5.1 Policies, growth, distributional change, and poverty reduction 76 5.2 Incidence of public spending in Latin America 84 5.3 Enrollment rates for secondary education relative to per capita GDP, for selected Latin American countries 88 5.4 Institutions and per capita income levels 88 5.5 Rural and urban headcount poverty rates 89 5.6 Potential spillovers between rural and nonrural GDP 90 5.7 Relative labor intensity per sector 91 5.8 Poverty changes and labor-intensive growth throughout the world 91 5.9 The impact of public transfers on income inequality 92 5.10 Gini coefficient in selected countries before and after taxes and transfers 94 5.11 Total tax revenue versus per capita income, throughout the world 95 5.12 Social protection spending mix in Latin America 99 5.13 Impact of social insurance and social assistance programs on inequality 99 5.14 Incremental tax rate needed to halve poverty in 10 years 100 Chapter 6 6.1 Traditional view of the growth-poverty relationship 105 6.2 Poverty-traps view of the growth-poverty relationship 105 6.3 Multiple equilibriums in the presence of increasing returns to scale 105 6.4 Interest rate spreads in Latin America, 2003 106 6.5 Growth in developed (OECD) and developing countries, 1963­2000 108 6.6 Income in Latin America relative to the OECD countries, 1960­2002 108 6.7 Histograms for per capita income, 1960s versus the 1990s 110 6.8 Histograms for per capita income in Latin America, 1960s versus the 1990s 112 6.9 Twin peaks 112 6.10 Equilibrium and distribution in 1999 113 6.11 Latin American states: One peak? 113 6.12 Convergence clubs in life expectancy 114 6.13 Income, poverty and investment 118 Chapter 7 7.1 Variation in regional poverty rates in Latin America 130 7.2 Income dynamics and space in Brazil 132 7.3 Income dynamics and space in Brazil at the municipal level 133 7.4 Income dynamics and space in Chile 134 7.5 Income dynamics and space in Mexico 134 7.6 The distribution of municipal incomes and life expectancy in Brazilian municipalities 135 7.7 Social indicators in Mexico, by period 135 7.8 Poverty rates versus poverty densities in Brazil 141 Chapter 8 8.1 Productivity and wages go hand in hand 147 8.2 Earnings gap between the formal and informal sectors in Bolivia, 2002 149 8.3 Transitions between the formal and informal sectors, and between salaried employment and self-employment in Mexico, 1987­2001 151 8.4 Complementarities in the income generation process in rural El Salvador 156 8.5 Sources of persistent poverty and low incomes in rural El Salvador 158 8A.1 Differences in returns to education 161 8A.2 Changes in returns over time 161 viii C O N T E N T S Chapter 9 9.1 Latin America is in a slow educational transition 166 9.2 Most Latin American countries show deficits in secondary and tertiary enrollments 167 9.3 Poverty is higher in families in which parents have little education 168 9.4 Children and youth in poor families have low educational attainment 168 9.5 Educational attainment of working age population, by country 172 9.6 Educational attainment for the poorest 30 percent and the richest 30 percent in Argentina, Mexico, Brazil, and El Salvador 174 9.7 Educational attainment for urban and rural areas in Nicaragua, El Salvador, Brazil, and Bolivia 175 9.8 Educational attainment for three age groups in Argentina, Colombia, Mexico, and El Salvador 176 9.9 Low educational attainment is reinforced in current cohorts 177 9.10 Poor children and youth stay out of school because of high costs and low benefits 179 9.11 Opportunity costs and schooling gaps get larger for secondary to post-secondary school-age children 180 9.12 Low education continues for generations, especially among the poor 181 9.13 Average rates of return for education increase at the tertiary level 182 9.14 The returns to education differ for urban and rural labor markets 184 9.15 Differences in returns to education in Brazil largely reflect unequal human capital and a secondary effect of skin color 185 9.16 Returns to each level of education for the three tiers of the earnings distribution 187 9.17 Correlation between returns to each level of education and poverty 188 9.18 Returns to education are generally lower for workers at the bottom of the earnings scale 189 9.19 Education quality differences lead to differential returns to education in Brazil 191 9.20 Factors that have an impact on moving up the educational ladder 192 9.21 The demographic transition and human capital accumulation--an opportunity that should not be missed 197 9A.1 Labor force by educational level 197 Tables Chapter 1 1.1 Growth rates needed to compensate for a 1-percentage-point increase in inequality 4 Chapter 2 2.1 Poverty in Latin America 22 2.2 Economic growth in Latin America 24 2.3 Welfare gains from increased longevity 29 2.4 Welfare comparisons: Argentina and Mexico 36 2.5 Intergenerational transition matrix for Colombia, 1997 38 Chapter 3 3.1 Economic growth in eight major Latin American countries 47 3.2 Aggregate per capita growth in Latin America 49 3.3 Economic growth in several reference groups 50 3.4 Inequality in Latin America 1950­2000, as measured by Gini coefficients 54 Chapter 4 4.1 Poverty, growth, and redistribution in Latin America 61 4.2 Growth and inequality elasticity of poverty 66 4.3 Impact on poverty of different growth scenarios 68 4.4 Growth rates needed to compensate for a 1 percent increase in inequality 69 Chapter 5 5.1 Economic policies and growth: Review of the evidence 77 5.2 Economic policies and income inequality: Review of the evidence 79 5.3 Growth and inequality regressions 85 5.4 Net growth elasticities of poverty to selected policies 86 5.5 Institutional quality in Latin America 89 5.6 Poverty reduction and sectoral growth 91 5.7 How much is Latin America undercollecting? 96 5.8 Results of simulations of income-neutral growth rate and incremental tax rate 100 ix C O N T E N T S Chapter 6 6.1 Median income in Latin America and the Caribbean relative to the industrial countries 109 6.2 Median income of convergence clubs 111 6.3 Does financial sector development play a role in the poverty-investment interaction? 119 6.4 Does poverty lead to lower secondary education? 120 6.5 The impact of risk on growth 123 Chapter 7 7.1 Typology of appropriate actions according to poverty rate and density 141 7.2 Public investment effects in Mexico, 1970­2000 142 Chapter 8 8.1 Decompositions of poverty and inequality changes in Argentina, 1992­2001 152 8.2 Decompositions of poverty and inequality changes in Peru, 1997­2002 152 8.3 Determinants of rural individual wages, El Salvador 153 8.4 Determinants of rural per capita family incomes, El Salvador 154 8.5 Permanent and transitory poverty in rural El Salvador, 1995­2001 157 Chapter 9 9.1 Average years of schooling in the "1­12" educational system and excess years spent in school, 6­18 age range, circa 2000 178 x Foreword L ATIN AMERICA'S DEVELOPMENT IN THE PAST incomes close to Latin America. Their achievement of more few decades has been characterized by two dis- egalitarian social outcomes is good news: Even without appointments: lagging growth and persistent fundamental shifts in economic structure, policies target- poverty and inequality. Set against the perfor- ing the poor can go a long way towards ameliorating social mance of other regions, notably China and injustice. India, and the East Asian miracles before them, Latin That such investments in the poor are good business for America's average annual growth of 4.2 percent in 2005 is society as whole is a central theme of the report. Poverty at best modest, and at worst, inadequate to tackle poverty itself hampers the achievement of high and sustained quickly. And the region's poverty remains acute, with one growth rates, completing a variety of vicious circles. For quarter of Latin Americans with incomes of under $2 a day, instance, poor students, faced with substandard schools and and the highest measures of inequality in the world. volatile returns to their human capital, underinvest in edu- Over the past decade, the World Bank, through the flag- cation. Poor entrepreneurs, excluded from capital markets, ship publications of the Latin American and Caribbean underinvest in good projects. Poor regions, lacking infra- Region, has sought to understand these issues individually. structure, fail to attract investment, and have fewer citizens In the area of growth, we have looked at the impact of able to adopt, manage, and generate new technologies. structural reforms, at the promise and constraints of natural Poor countries, unable to moderate income disparities, find resource abundance, and at the burden of educational ethnic or racial tensions exacerbated that, in turn, thwart and technological shortfalls. On the issue of poverty and the establishment of a healthy business climate. inequality, we have examined the root causes and impacts To move to a virtuous circle of growth and poverty of poverty and inequality, and the social implications of reduction will take action on many poverty fronts and an income insecurity. approach that not only considers how the poor can benefit This, our eighth flagship, takes a fresh look at how from growth, but also how they can contribute to it. Key growth and poverty are interlinked, and makes new recom- among these is investment in human capital. Here the mendations on how to boost growth and reduce poverty at report emphasizes that an integrated strategy, taking into the same time. The report revisits how growth can reduce account barriers to getting education and the entire life- poverty and how much emphasis should be placed on cycle of students, is essential. For example, educating rural growth relative to distribution, given a country's income children will pay greater dividends if improved infrastruc- and inequality levels. It also reopens the question of how ture attracts firms who can employ their enhanced skills. much policy can influence how "pro poor" the growth Social safety nets that mitigate labor market risk increase process is. Latin America's inequality is undeniably partly the perceived return to education. Improved access to due to the results of inherited economic structures and financing for college, where the returns to education are resource endowments, but it is also the case that the United highest, gives impetus to finishing secondary school. At Kingdom and Sweden have distributions of market the national, regional, and household levels, and on the xi F O R E W O R D health, trade, and financial sector fronts, policies that build Bank are committed to enriching, supporting, and learning on these interrelationships have been shown to be more from this debate, a debate that is critical to the design of effective in fighting poverty. These and many other find- policies conducive to enhancing welfare in all its dimen- ings and recommendations throughout the report are sions among the poor of Latin America and the Caribbean. grounded in detailed analysis and examples and should provide additional insights to policy makers and develop- ment practitioners in the different countries of the region. We believe this year's flagship, Poverty Reduction and Growth: Virtuous and Vicious Circles to be a valuable contri- Pamela Cox bution to the intense current regional debate on poverty Vice President for Latin America and the Caribbean and growth. As a development institution, we at the World The World Bank xii Acknowledgments P OVERTY REDUCTION AND GROWTH: VIRTUOUS AND VICIOUS CIRCLES IS THE PRODUCT OF A collaborative effort by a number of professionals from within and outside the Bank. The report was prepared under the guidance and direction of Guillermo Perry by a core team comprising Humberto López, William Maloney, Omar Arias, and Luis Servén. Other significant contribu- tors to the drafting of the report included Mariano Bosch (LSE), Cesar Calderón (World Bank), Anna Fruttero (World Bank), and Edwin Goñi (World Bank). Background papers were prepared by Patricio Aroca (Universidad Católica del Norte, Chile), Monserrat Bustelo (World Bank), Ana María Diaz Escobar (World Bank), Maurizio Bussolo (World Bank), Maria Victoria Fazio (World Bank), Leonardo Gaspariani (CEDLAS and Universidad Nacional de la Plata), Federico Gutierrez (CEDLAS and Universidad Nacional de la Plata), Tom Krebs (Syracuse University), Pravin Krishna ( Johns Hopkins University), Norman Loayza (World Bank), Alex Mariana Marchionni (Universidad de la Plata), Denis Medvedev (World Bank), Leandro Prados de la Escoura (Universidad Carlos III), Claudio Raddatz (World Bank), Lucas Siga (University of Cal- ifornia, San Diego), Walter Sosa (Universidad de San Andres), and Leonardo Tornarolli (CEDLAS and Universidad Nacional de la Plata). Emmanuel Skoufias (World Bank), Kathy Lindert (World Bank), and Joseph Shapiro (World Bank) also shared with us many of the results of the regional study Redistributing Income to the Poor; Public Transfers in Latin America and the Caribbean. Patricia Macchi (Boston University), and Guillermo Beylis (World Bank) provided excellent research assistance at different times during the project. The report has also benefited from comments by Nancy Birdsall (Institute for International Eco- nomics), Nora Lustig (Universidad de las Américas), Nohra Rey de Marulanda (Inter-American Develop- ment Bank), and John Williamson (Institute for International Economics), and by our two principal advisers: Francisco Ferreira and Roberto Zagha. Finally, Elena Serrano and Catherine Russell coordinated the report's publication and dissemination activities, working closely with Dana Vorisek and Susan Graham in the World Bank's Office of the Publisher. xiii Acronyms and Abbreviations CCT conditional cash transfers IV instrumental variable CPI consumer price index LAC Latin America and the Caribbean ECLAC Economic Commission for Latin LISA local indicators of spatial associations America and the Caribbean NAFTA North American Free Trade Agreement FUSADES Fundación Salvadoreña para el OECD Organisation for Economic Desarrollo Económico y Social Co-operation and Development GATT General Agreement on Tariffs and PPP purchasing power parity Trade PWT Penn World Tables GDP gross domestic product RER real exchange rate GIS Geographical Information Systems SA social assistance GMM Generalized Methods of Movement SEDLAC Socio Economic Database for Latin i.i.d. independent and identically distributed America and the Caribbean IPEA Instituto de Pesquisa Economica SI social insurance Aplicada (Brazil) Note: All dollar amounts are U.S. dollars unless otherwise indicated. xv CHAPTER 1 From Vicious to Virtuous Circles That raising income levels alleviates poverty, and that economic growth can be more or less effective in doing so, is well known and has received renewed attention in the search for pro-poor growth. Less well explored is the reverse channel: that poverty may, in fact, be part of the reason for a country's poor growth performance. This more elaborated view of the devel- opment process opens the door to the existence of vicious circles in which low growth results in high poverty and high poverty in turn results in low growth. This report is about the existence of those vicious circles in Latin America and about the ways and means to convert them into virtuous circles in which poverty reduction and high growth reinforce each other. L ATIN AMERICA'S TWIN DISAPPOINTMENTS OF inequality, it would have been more pro-poor. Second, even relatively weak economic growth and persis- when inequality remains unchanged, economic growth is tent poverty and inequality are longstanding less effective in reducing poverty in countries with less and intimately related. That raising income equal distributions of income: To attain the same reduction levels alleviates poverty, and that economic of poverty, unequal countries must grow more than more growth can be more or less effective in doing so, is well equal ones. Given the region's acute growth divergence known and has received significant attention in the search during the lost decade of the 1980s and the slowdown from for pro-poor growth. Less well explored is the reverse chan- 1998 to 2003, as well as lack of progress on the inequality nel--poverty may, in fact, be part of the reason for a region's front, it is not surprising that income poverty has been so poor growth performance, creating vicious circles where low persistent since 1980 (figure 1.3). Though the report dis- growth results in high poverty and high poverty in turn cusses important caveats in traditional comparisons across results in low growth. This report is about finding ways of countries and across time, it remains true that, with the converting this negative cycle into a virtuous circle of exception of Chile, there has been little poverty reduction poverty reduction, in which broad-based attacks on poverty beyond the gains of the 1950­80 period, and in many feed back into higher growth that in turn reduces poverty. countries growth has not been especially pro-poor. Latin America's economic performance in the last 50 years has been disappointing. Growth lagged behind core coun- Poverty as a multidimensional and tries of the OECD (Organisation for Economic Co- dynamic concept operation and Development), at a time when East Asia and These conclusions broadly hold when a broader view of Spain, the madre patria on the periphery of Europe, were poverty and welfare is taken (chapter 2). As the literature quickly catching up (figure 1.1). Income inequality has increasingly stresses, poverty is a concept that spans a range remained very high in Latin America over the past 50 years of dimensions, such as health, mortality, and security, that (figure 1.2), posing a double impediment to poverty reduc- may be uncorrelated with conventional measures of income tion. First, had growth been accompanied by reduced poverty. Further, a complete concept of well-being needs to 1 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 1.1 FIGURE 1.3 Per capita income relative to the OECD, 1870­2000 Poverty rates in Latin America, 1950­2000 Ratio Percent 0.8 Spain 70 0.6 60 LAC 0.4 50 0.2 40 East Asia 0 30 0 0 1870 1880 1890 1900 1913 1925 1929 1938 195 1960 1970 1975 1980 199 2000 20 Source: Authors' calculations based on Prados de la Escosura (2005) and Maddison (2005). 10 Note: LAC Argentina, Brazil, Chile, Mexico, República Bolivariana de Venezuela, and Uruguay. East Asia South Korea, 0 Taiwan (China), Hong Kong (China), and Singapore. 1950 1960 1970 1980 1990 2000 Source: Authors' calculations for 1950­1980; Gasparini, Guitierrez, and Tornarolli (2005) for 1990 and 2000. Note: We used a poverty line of US$2 a day; poverty rates for 1950­1980 are estimated using a lognormal approximation. FIGURE 1.2 Gini coefficient for Latin America, 1950­2000 0.60 growth. However, intergenerational mobility remains lower in Latin America and the Caribbean than in the worst of 0.55 the OECD countries. Recent evidence indicates that the 0.50 children of poor families and of parents with low education face a relatively high probability of achieving low educa- 0.45 tional levels, obtaining lower returns for their education, 0.40 and remaining poor (figures 1.4 and 1.5). The fact that 1950 1960 1970 1980 1990 2000 Chile is one of the most mobile societies in the region sug- Source: Authors' calculations based on Altimir (1987) and Londoño gests that the modernization of the country across the last and Szekely (1997). Note: Based on data for Brazil, Chile, Mexico, and República decades has offered more opportunities to the less well-off. Bolivariana de Venezuela. Finally, as documented in the World Bank's Latin Ameri- can region flagship Securing Our Future in a Global Economy (de Ferranti and others 2000), the high economic volatility in the region implies that the poor are subject to higher incorporate income movements across lifetimes or even risks than the poor in other regions. Although macroeco- generations, which means that issues of risk and mobility nomic volatility was reduced in the 1990s after peaking in through the income distribution must be examined. Ignor- the 1980s, it still remains exceptionally high, and labor ing these considerations leads to large distortions in the market volatility remains substantially higher than it is in concepts of poverty and inequality. the United States, for example. Although the limited existing data on these aspects of As later chapters show, all these dimensions not only poverty do not permit the kind of global comparisons that provide a more complete view of poverty, they also consti- measures of income inequality and headcount poverty tute channels back to growth. numbers do, the picture they sketch is only somewhat more optimistic. It is true that mortality rates have fallen far The twin disappointments: Destiny or choice? more than income levels would predict and account for Is there something intrinsic to the region that has left it large improvements in welfare in those countries with little with relatively low growth and high levels of inequality 2 F R O M V I C I O U S T O V I RT U O U S C I R C L E S FIGURE 1.4 Low educational traps persist across generations among the poor and excluded Colombia Brazil Children's years of education Average years of education of adult sons 20 20 40% poorest 20% richest Whites Pretos Pardos 16 16 12 12 8 8 4 4 0 0 0 to 6 6 to 11 more than 11 0 to 4 5 to 8 9 to 11 more than 11 Mother's years of education Father's years of education Source: Authors' estimates based on household survey data. Note: Average years of school for adults aged 24­65 is determined by their parents' years of school. FIGURE 1.5 Although they stand to gain the most from education, poor people actually have low returns Chile Nicaragua Wages relative to level of education Wages relative to level of education 1.7 1.7 1.5 1.5 1.3 1.3 20% richest 20% richest 1.1 1.1 0.9 0.9 0.7 0.7 20% poorest 20% poorest 0.5 0.5 0.3 0.3 0.1 0.1 Complete Some Complete Some Complete Complete Some Complete Some Complete primary secondary secondary university university primary secondary secondary university university Source: Authors' estimates based on household survey data. Note: Average schooling returns for workers from families in the bottom and top quintiles of the income distribution; from Mincer earnings regressions, controlling for work experience, gender, and urban residence. and poverty? The World Bank's Latin American region least until the late 1800s and thus had adverse conse- flagship Inequality in Latin America: Breaking with History? quences for growth and inequality for a long time. (de Ferranti and others 2004) argued that exclusionary In chapter 3, we show that indeed Latin America was institutions set up during the European conquest to exploit well behind the advanced economies in the mid-1800s, existing mineral wealth and indigenous populations, and when the region's per capita income levels represented the particular crops suited to the region's climate (such as about 60 percent of the U.S. levels and 55 percent of those sugar plantations based on a slave workforce), led to highly in the broader OECD group. More important, we also show unequal access to land, education, and political power at that a significant part of the current development gap in 3 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S the region dates from the middle of the 20th century, when TABLE 1.1 other regions took more advantage of the rapid pace of Growth rates needed to compensate for a 1-percentage-point global expansion. Latin America's relative retardation in increase in inequality this period was in all likelihood related to the extreme inward-looking policies instituted then and to the lack of Compensatory Compensatory Country growth rate Country growth rate macroeconomic prudence that led to the devastating debt crisis of the 1980s. Although policies are importantly con- Argentina 2.5 Peru 1.6 ditioned by historical context, more promising roads were Chile 2.4 St. Lucia 1.5 Brazil 2.3 Guatemala 1.5 not taken. Mexico 2.1 Paraguay 1.5 The same appears true in the realm of income distribu- Costa Rica 2.1 El Salvador 1.4 tion. The report shows that as the 20th century began, Colombia 2.1 Venezuela, 1.2 Trinidad and Tobago 2.0 R.B. de France, Spain, the United Kingdom, and the United States Dominican Republic 1.9 Ecuador 1.1 all had high levels of income inequality. Yet they managed Panama 1.9 Nicaragua 1.1 Belize 1.8 Guyana 1.1 to lower income inequality dramatically during the century Uruguay 1.8 Bolivia 1.0 and over relatively short periods of time (two to three Jamaica 1.7 Honduras 0.8 decades). Such achievements appear related to the universal provision of basic education and health services and the Source: Authors' calculations. establishment of highly redistributive welfare states. Note: The table reports the growth rates that would leave poverty unchanged when the Gini coefficient increases by 1 Both Latin America's loss in relative income position in percent. Higher values indicate that inequality plays a more the last 50 years and the OECD's ability to sharply reduce important role in poverty reduction. inequality are, perhaps counterintuitively, good news: our history is not our destiny--choices of policies and institu- industries show large differences in labor intensity (agricul- tions can lead to major improvements along both dimen- ture and construction are generally more labor intensive sions. Breaking with history is indeed difficult, but it is by than manufacturing and services, and the latter are more no means impossible. labor intensive than mining and utilities); and poverty reduction is stronger when growth has a labor-intensive The link from growth and development inclination. The chapter also finds that policies such as to income-poverty reduction increased access to education and infrastructure have had Chapter 4 of the report concentrates on the effect of growth direct positive impacts on growth, inequality, and poverty and changes in inequality on income-poverty reduction in reduction, while others, such as trade opening, have had countries with different characteristics. It shows that positive effects on growth but have tended to increase achieving the greatest reduction in poverty may imply inequality and even poverty in the short run. In the long placing differing relative emphasis on growth versus redis- run, however, all pro-growth policies tend to reduce income tribution depending on the individual country's initial poverty. conditions: poor countries (such as Bolivia, Haiti, and Chapter 5 also discusses the importance of transfers as a Honduras) and relatively equal countries that, bluntly put, means of sharing the fruits of growth by investing in the have little to distribute, need first and foremost high and poor. Bringing the historical discussion above into the pres- sustained growth, even at the expense of some increases in ent, the chapter shows that roughly half of the stark differ- inequality; this might be called the China model. In ence in income inequality between Latin America and contrast, relatively richer and more unequal countries-- contemporary OECD countries results from differences in most of Latin America, and especially Argentina, Brazil, returns to factors of production--the result of the unequal Colombia, and Mexico--need both higher growth and distribution of human and other capital in Latin America. significant redistribution if they want to make a fast and But the other half results from the generally unprogressive significant dent in poverty reduction (table 1.1). nature of Latin America's system of transfers. The core Chapter 5 examines how different policies and different OECD countries use transfers from the rich to the poor, and sectoral patterns of growth affect income-poverty reduc- extensive pension schemes that distribute income from the tion. It finds that sectoral composition matters: different those working today to those retired tomorrow, to lower 4 F R O M V I C I O U S T O V I RT U O U S C I R C L E S FIGURE 1.6 Gini coefficients for market and disposable incomes Gini market incomes Gini disposable incomes Latin America Ireland United Kingdom Canada Portugal Finland Denmark Italy Greece EU15 United States Spain Belgium Sweden Germany France Luxembourg Netherlands Austria 0.60 0.55 0.50 0.45 0.40 0.35 0.30 0.25 0.20 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 Source: Authors' calculations. the Gini (the standard measure of inequality) by about especially low. More important, although Latin American 15 percentage points (from, for instance, 0.53 in the public expenditures underwrite large, progressive items United Kingdom to 0.35).1 Transfers in a typical Latin (basic education and health), they also fund large regressive American country, in contrast, alter the Gini by 2 percent- items (subsidies to pensions, tertiary education, and age points or less, although there are a few exceptions such energy), which offset the progressive spending. An encour- as Chile, which managed to reduce the Gini by twice as aging recent development is the introduction of successful much (figure 1.6). policies such as Progresa/Oportunidades in Mexico, Familias Whether the pure transfers of the magnitudes discussed en Acción in Colombia, and Bolsa Escola in Brazil, that com- above for Europe have been optimal from a growth point of bine fiscal transfers to the poor with incentives for them to view is debatable, as is their wisdom or political feasibility build human capital through both health and education in Latin America. Arguably, for a variety of reasons, and in investments from early childhood. particular to be consistent with growth objectives, redis- tributive policy probably should focus on equalizing Closing the virtuous circle: The link from poverty opportunities through more equal access to assets, such as to growth human capital, rather than on equalizing outcomes mea- The more novel thesis of the report is that Latin America's sured as incomes per se. What is clear, however, is that persistent poverty may itself be impeding the achievement Latin America has not made the efforts to mobilize the of higher growth rates--that there are reinforcing vicious resources to attack poverty that it could. First, the region's circles that keep families, regions, and countries poor and tax collections are below those in similar countries (when unable to contribute to national growth. The now-expansive benchmarked by income per capita), with a few exceptions literature on poverty traps has elaborated a large number of such as Brazil and Nicaragua, and collections for progres- channels that may perpetuate poverty. The emphasis we sive taxes, such as personal income and property taxes, are place on the multidimensionality of poverty and on lifetime 5 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S and intergenerational considerations in welfare measure- lower aggregate growth. Such vicious circles can lead to ment further enriches the universe of channels through "convergence clubs"--richer and poorer countries, regions, which poverty impedes growth. To list just a few we discuss: or households tend to converge to different income or wel- fare levels even in the long run. Whether these are, in fact, · Poor people often have limited access to financial mar- poverty traps that cannot be escaped without intervention, kets or other necessary complements to private invest- or whether it simply takes much longer to transition to ment (such as property rights and infrastructure) higher-income states, is to us a distinction of secondary essential to the accumulation of physical and knowl- importance, particularly when political economy issues are edge capital and participation in the growth process. considered. What we do argue is that smart investments in · Poor people are often in poor health, which reduces the poor can lead to virtuous circles and that the issue of their productivity and impedes their ability to man- "pro-growth poverty reduction" should perhaps be as age and generate knowledge. important a policy concern as traditional concerns with · Poor people attend low-quality schools and the low "pro-poor growth." In other words, investing in the poor is and late returns to education and diminished good business for society as a whole, not just for the poor. prospects for mobility deter the accumulation of Tracing these reinforcing circles implies necessarily human capital essential for growth. Education enhances moving away from static concepts of poverty and studying earnings potential, expands labor mobility, promotes the dynamics of poverty at every level, and this report the health of parents and children, and reduces fertil- aspires to break new ground in this area. It provides evi- ity and child mortality. dence on the existence of convergence clubs at the house- · Poor people may face more labor market risk, or may hold, regional, and international level and in several cases be less able to hedge against it, and thus find returns shows that these appear to reveal the evidence of poverty- to investing in human capital adjusted for risk to be trap dynamics. less attractive. Further, the inability to diversify risk prevents specialization in agriculture or movements Global convergence clubs to off-farm activities, for example, that would lead to Do poorer countries grow less than richer countries? The greater productivity. Since the poor are typically evidence presented in chapter 6 suggests that, with a few more risk averse than the rich because losses hurt notable exceptions, they do. Panel a of figure 1.7 suggests them more severely, in the absence of well-functioning that, apart from two short periods (one in the second half of insurance and credit markets, the poor skip profitable the 1970s and another in the early 2000s), the typical investment opportunities that they deem too risky. developing country (and Latin America is not an exception Once again, societies with high poverty rates show a here) has always experienced lower growth rates than the tendency to underinvest. typical rich country. Over the 1963­2003 period, median · Poor regions and countries have fewer individuals per capita growth in industrial countries outpaced median capable of adopting, managing, and generating new growth in developing countries by an average of more than technologies that would contribute to productivity. 1 percent per year. · Poor regions may lack the infrastructure or human The difference in per capita growth rates between the capital that would make them attractive to extra- developed and developing countries has led to an expanding regional investment or the resources to develop them gap between rich and poor countries over time (figure 1.7, and that would facilitate sectoral and territorial labor panel b). In the early 1960s the median Latin American mobility in search of higher income opportunities. country had an income level that was slightly less than one- · Poor countries with poor regions may find ethnic or third the income of the median developed country; today racial tensions exacerbated by income disparities lead- that gap is less than 20 percent. Globally speaking, the typ- ing to interregional tensions that make both regions ical developing country had an income level about 12 per- and the country as a whole riskier to invest in. cent that of the richer countries in 1960; and today it is closer to 5 percent. There is little to support the conver- In each case, poverty in itself prevents taking actions gence hypothesis that poorer countries will tend to catch that would facilitate the exit from poverty and results in up with the richer ones. Rather, as panel c of the figure 6 F R O M V I C I O U S T O V I RT U O U S C I R C L E S FIGURE 1.7 Indicators for poor and rich countries a. Growth rates b. Relative incomes Percent Income relative to OECD 5 0.35 4 0.30 3 0.25 2 0.20 1 0.15 0 0.10 1 0.05 2 0 2 5 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 200 1960 1963 1966 1969 1972 197 1978 1981 1984 1987 1990 1993 1996 1999 2002 Developing Developed Latin America Latin America Developing c. World d. Latin America Number of countries Number of countries 25 10 20 8 15 6 10 4 5 2 0 0 0.4 0.7 1.1 1.8 3 5 8 13 22 35 60 0.4 0.7 1.1 1.8 3 5 8 13 22 Per capita income, US$ thousands Per capita income, US$ thousands Source: Authors' calculations. suggests, the poor stay poor, while the rich get richer. The whereas the mass of the high peak increases (worldwide life histogram for the world in 1999 suggests a trimodal distri- expectancy has increased and is slowly converging). bution, with a low peak at $1,100; a second at between $5,000 and $8,000, and a third peak around $35,000 form- Does poverty matter for growth? ing poor, middle-income, and rich convergence clubs. Are high poverty levels to blame for the disappointing (Chapter 7 shows that since 1960 there has been conver- growth performance of poorer countries? A bimodal distri- gence within these clubs but divergence among them.) bution in income or life expectancy levels does not, in Panel d shows that Latin America as a region is unimodal itself, prove that poverty is a brake on growth, and with its single peak at about $8,000 and belongs to the chapter 6 finds only mixed evidence for the extreme case of middle cluster that is slowly separating both from the very poverty traps. However, the chapter does identify several poor and, distressingly, from the very rich. self-reinforcing mechanisms that may retard growth and Convergence clubs at the cross-national level are also cause poverty to persist, and these may be more relevant evident, though much less so, when nonincome dimensions from a policy point of view. Looking across countries, of welfare are considered. For example, figure 1.8 presents poverty does appear to deter growth and investment (fig- the cross-national life expectancy histograms for 1960 and ure 1.9), especially when the degree of financial develop- 2002. These histograms indicate the presence of a two- ment is limited. More specifically, we estimate in chapter 6 peaked pattern in both periods, but it is also evident that that, for the average country, a 10-percentage-point the mass of the low peak declines between 1960 and 2002, increase in income poverty lowers the growth rate by about 7 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 1.8 Convergence clubs in life expectancy throughout the world 1960 2002 Number of countries Number of countries 35 60 30 50 25 40 20 30 15 20 10 5 10 0 0 35 40 45 50 55 60 65 70 75 40 45 50 55 60 65 70 75 80 85 Life expectancy at birth, years Life expectancy at birth, years Source: Authors' calculations. measured by research and development expenditures) and FIGURE 1.9 the accumulation of human capital (see below), both of Poverty and investment throughout the world which are additional channels through which poverty influ- % of people living in poverty Rate of investment ences aggregate growth. 70 24 Investment (% of GDP) 60 22 Regional convergence clubs 50 20 Chapter 7 finds an unusual combination of converging 40 18 income among subnational units, but increased spatial con- 30 centration within countries. Modern spatial econometric 16 20 tools show that within Brazil, Chile, and Mexico, there are Poverty (%) 10 14 clear convergence clubs of rich and poor regions, that appear 0 12 to be drifting increasingly apart (figure 1.10). This finding 1 2 3 4 5 6 7 8 9 10 (poorest) (richest) is consistent with the New Economic Geography literature World income ranking by decile that has focused on how larger, already established regions Source: Authors' calculations. enjoy scale economies while lagging regions are less produc- tive and hence less attractive to factors of production. These dynamics, and those discussed for national poverty 1 percent, holding other determinants of growth constant. traps in chapter 6, apply to national or subnational units Further, we estimate that a 10-percentage-point increase in equally. Two considerations are particular to the latter, income poverty reduces investment by 6­8 percentage however. The first is that within countries, labor can legally points of gross domestic product (GDP) in countries with move freely. In practice it does not, leaving large wage gaps underdeveloped financial systems. These results validate of often 50 percent among regions. Evidence from Chile the predictions of theory: that poverty may limit growth and Mexico suggests that this phenomenon is partly the when financial sectors are imperfect because the poor, who result of another poverty-trap dynamic--the poor cannot lack access to credit and insurance, will not undertake muster the savings or liquidity to migrate and hence can- many socially profitable investments, thus depressing the not leave. But other evidence suggests that this story may aggregate level of investment and growth. The report also be incomplete. Nonincome measures of poverty, such as finds evidence that poverty limits the level of innovation (as mortality, show convergence within countries, much the 8 F R O M V I C I O U S T O V I RT U O U S C I R C L E S low-productivity economic activities. The poverty-traps FIGURE 1.10 literature emphasizes insufficient asset holdings (including Regional income dynamics in Brazil: The persistence of two convergence clubs human capital), thresholds in the returns to those assets, fixed costs of productive transitions, and limited access to Relative income credit or insurance among the poor as main determinants of 2.5 their inability to take advantage of growth opportunities. 2.0 Of particular importance is the ability of the poor to use 1.5 their labor (their most abundant asset) in wage jobs, self- 1.0 employment, or their own microenterprises. Labor earnings 0.5 often account for more than two-thirds of total household 0 3.0 income of the Latin American poor. The pricing of labor 2.0 reflects productivity differentials across workers and jobs, 1.0 2.5 3.0 1.5 2.0 sector and regional supply-demand imbalances, and non- Country relative, 0 0.5 1.0 period t 0 market factors. Low-earnings traps can arise from deficien- Country relative, period t 10 cies in the endowments that enhance the productivity Source: Authors' calculations. (quality) of labor assets (such as human capital and infra- Note: Figure shows relative state income distributions at time t and ten years later for the period 1955­2000. It suggests little structure) and from earnings differentials unrelated to movement in states' relative positions and a persistent two humped distribution. skills (such as ethnic discrimination and location) that arise from barriers to mobility in the labor market. way they do internationally, suggesting that the welfare Chapter 8 examines some of the mechanisms that may gap broadly considered may be less dramatic. Further, sim- prevent the Latin American poor from participating in the ply asking people how poor they feel reveals some provoca- growth process and lead to persistent poverty. Unfortu- tive anomalies. The poorest group in the Bolivian altiplano nately, the limited long-span panel data prevent in-depth (largely indigenous) self-rates as the least poor in Bolivia, analyses of the duration of poverty and its main determi- while inhabitants of the rich province of Buenos Aires rate nants throughout Latin America. The chapter draws on the themselves as the poorest in Argentina. These findings sug- limited, though highly consistent, evidence available on gest that "congestion externalities"--the negative aspects these issues and reaches two main conclusions. First, low of living in concentrated urban areas--may be important, levels of productivity, rather than labor market segmenta- that relative income disparities may be more brutally tion, is the overwhelming driver of low earnings. Most apparent in urban contexts, or simply that researchers are poverty is thus not generated directly by labor market fail- missing key dimensions of well-being that are uncorrelated ures but by deficiencies in workers' productive endow- with income. ments, especially education, combined with the low levels Second, laggard regions in general have low levels of of overall productivity of their local economy. This effect is education and infrastructure that require special efforts to exacerbated by high volatility and the inability to insure bring them toward the country average. However, to the against shocks, much more so than in developed countries. degree that agglomeration externalities--the economies of Second, detailed analyses of rural El Salvador and consistent scale that may arise from concentrating economic activity-- evidence from other countries suggest that poverty traps dictate that poor regions have lower growth potential and surrounding the accumulation of these productive assets lower returns to investment, governments may be con- are a phenomenon of practical relevance in the region. fronted eventually with a trade-off between aggregate Chapter 9 then takes on one of the central channels that growth and geographical equity. can support a two-way causality between poverty and eco- nomic growth: the accumulation of human capital. Human Household-level poverty traps capital, proxied by education or health levels, is generally The fundamental building block underlying the interna- believed to be one of the key determinants of long-term tional and regional analyses discussed above is the household. growth, while cross-country empirical evidence suggests Addressing persistent poverty requires an understanding of that poverty may affect education levels (see chapter 8). the factors preventing poor families from moving out of Chapter 9 investigates the micromechanisms that could 9 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S support this double causality, so that specific actions to cycle in most countries suggests that lack of school facili- increase the educational attainment of the poor could ties is not the main driving factor, although in some coun- ignite a virtuous circle of faster growth and poverty reduc- tries physical access constraints remain a problem. tion in the region. Second, returns to schooling tend to increase with the The chapter begins with a well-known fact: families with level of education, a finding consistent with a skill bias in little education (specifically those with less than secondary labor demand caused by technological change in the schooling) tend to be poor, and in turn they tend not to invest region, as detailed in the World Bank's Latin American enough in their and their children's education to escape region flagship Closing the Gap in Education and Technology poverty. The chapter documents several pieces of evidence on (de Ferranti and others 2003). Schooling returns are flat self-reinforcing mechanisms driving this vicious circle. during the basic and secondary cycles and increase after First, despite the region's recent progress toward univer- completion of secondary education; in some cases, the full sal primary enrollment, there is a clear and persistent educa- return materializes only after completion of tertiary educa- tional divide in educational attainment. The population tion. That is, schooling returns become attractive just as sorts into two groups: individuals with low-education the opportunity cost, in terms of wages forgone by the stu- attainments (typically less than secondary education) and dent, becomes most acute for poor families. In addition, the individuals with secondary education and above (fig- chapter strikingly shows that in most countries poor fami- ure 1.11). Rural residents and the poorest families, includ- lies face below-average returns to tertiary (and sometimes ing disadvantaged ethnic groups, are predominantly secondary) education, plausibly due to low-quality schools trapped in the low educational group. This divide continues as well as disadvantages arising from family background and replicating itself among the current cohort of students in attitudes toward education (see figure 1.4). Poor families high rates of repetition and dropout of these same groups. have to juggle current subsistence needs against schooling The smooth decline in enrollments during the secondary investments with a remote, uncertain, and less-attractive FIGURE 1.11 The sharp educational divide between the poor and the rich in Latin America Argentina Brazil Percent Percent 45 45 40 40 35 35 30 30 25 25 20 20 15 15 10 10 5 5 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18+ 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18+ Years of education Years of education 30% poorest 30% richest Source: Authors' estimates based on household survey data. Note: Distribution of the working-age population across schooling levels from families in the bottom and top quintiles of the income per capita distribution. 10 F R O M V I C I O U S T O V I RT U O U S C I R C L E S payoff. The statistical evidence describing the low incen- account their direct and indirect effects on growth tives and barriers to accumulating human capital is corrob- and poverty reduction. This awareness introduces orated by the responses that poor children and youth give new but necessary levels of complexity in the eval- for dropping out of school: high opportunity costs at older uation of policy options on both agendas. As a ages, perceived low benefits in the 1­12 grade schooling simple but important example, conditional cash cycle, and physical access constraints. transfer programs have an impact on poverty that In sum, the completion of a secondary education neces- goes beyond the increased incomes for poor sary for poor families to move out of poverty remains out of households provided by straight transfer policies. reach and children's education remains strongly correlated Conditional transfer programs also relieve credit with that of their parents. The educational divide is self- constraints on and provide a further incentive to the reinforcing across generations and is a critical underlying accumulation of human capital that raises income driver of the vicious circles of poverty observed at the both at the household and, eventually, at the econo- household, regional, and national levels. mywide level. · Third, pro-growth policies that have short-run Implications of the report adverse impacts on distribution and poverty, as A number of implications emerge from the analyses appears to be the case with trade opening, may described above. We discuss them along two main dimen- actually create a drag on growth creation (see sions: strategic and policy levels. chapter 5). However, when combined with com- plementary policies such as improved access to Strategic implications education and infrastructure, the short-run adverse The report uncovers several lessons that have implications poverty effect can be mitigated, enhancing both for the way we view poverty reduction. the direct and indirect effects on growth. Further, compensatory actions to offset some of these effects 1. Pro-poor growth and pro-growth poverty reduction. The (for example, support to small farmers in noncom- existence of virtuous circles between growth and petitive sectors during trade opening) gain a new poverty reduction enriches the debate on optimal rationale in increasing the efficiency of reform poverty reduction strategies in several ways. policies in addition to those justifications related · First, the debate about whether strategies should to social protection. emphasize pro-growth or pro-poor policies now · Finally, transfer programs should always seek to appears somewhat less germane. Strategies that do directly stimulate the accumulation of assets that not focus on growth forswear perhaps the most will advance the growth process, as programs like potent weapon for improving human well-being at Oportunidades in Mexico, Bolsa Escola in Brazil, and our disposal, especially in light of the likely limits Familias en Acción in Colombia do. of explicitly pro-poor policies discussed above. Yet 2. Pro-poor growth vs. pro-poor government policy. The find- failing to take account of the constraints facing the ing that at most half of the difference in inequality poor in participating in and contributing to between Latin America and OECD countries arises growth undermines its generation. For example, from differences in the distribution of market liquidity constraints, risk, and indivisibilities or incomes implies two things. First, while efforts need lumpiness in human capital investments appear to to be made to improve both the endowments of the prevent the poor from acquiring the education that poor and the returns to them offered by the market, would move them out of poverty and fuel growth. there appear to be limits to what can be done. For Redressing these constraints gives rise to an under- example, Sweden, a country well known for its con- examined dimension of policy analysis that might cerns with equity and human capital formation, has a be called pro-growth poverty reduction. market distribution that is very similar to that of · Second, the bidirectional relationship between many countries in Latin America, suggesting that growth and poverty reduction suggests that ide- even states that put equity high on their policy ally consideration of policies should take into agenda may end up with high levels of inequality in 11 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S market incomes. Second, much of the heavy lifting of In short, policy makers need to consider more equalizing incomes in the OECD countries appears comprehensive measures of poverty and inequality to have been done by their expansive transfer systems not only to get a more accurate view of the evolution that dwarf anything found in the Latin American of societal well-being but to better understand and region to date, although the report suggests that, take advantage of the channels back to growth. here too, there are limits posed by political economy and efficiency. In short, policies designed to obtain 4. Nonlinear thinking: Humps and black holes, agglomera- equal opportunities for development of human capi- tion externalities, and complementarities. One critical tal, and hence more equal market incomes, need to be insight of the poverty-traps literature is that the complemented with redistribution through taxes and response to policy is nonlinear: it may vary depend- transfers. ing upon the magnitude and comprehensiveness of 3. Multiple dimensions of poverty, multiple channels to the effort. growth. The narrowness of the traditional focus on · There are thresholds (or humps) below which income poverty becomes increasingly unsatisfactory effort may have no impact; in such cases policy in the context of tracing feedbacks to growth. As makers are effectively throwing resources down a examples: black hole. For example, the fact that the returns · The strong gains in longevity in the region are to secondary education often materialize only only weakly correlated with income growth. In upon completion--or, worse, upon completion of some countries where incomes have remained stag- tertiary education--implies that it is not worth it nant, welfare has risen substantially because of for households to invest beyond primary school. improvements in health care and disease preven- Programs that seek to create incentives to invest tion. As noted above, health is linked to produc- in education may have a greater impact on poverty tivity growth, and policies dedicated to redressing if they are designed to get the student "over the this dimension of poverty are thus both pro-poor hump"--through the end of secondary school and and pro-growth. not just to the next grade level. · The prospect of moving out of poverty or upward · The literature suggests that the returns to assets, in the income distribution is a major motivation such as human capital, depend greatly on other for the accumulation of human capital. However, public assets that are complements, such as roads, the lower, late, and uncertain rates of return to communications systems, and credit markets. education of the poor, for the reasons discussed Major investment in education, for example, may above, foreclose such mobility and discourage have limited payoff if individuals cannot com- individuals and their children from accumulating mute to a job that uses the higher level of skills. In this capital. Clearly, one lesson is that redressing the same way, a pro-growth policy of building these disincentives both improves social indicators roads in a region may have a greater impact if the that more completely measure poverty and sti- population has the human capital to work in mulates growth. But a second lesson is that anti- emerging industries than if they are sick, illiter- poverty policy must take a life-cycle view, with ate, or constrained by language. policies that look at the barriers to mobility in a · Policies toward lagging regions may be complicated comprehensive way. by the fact that concentrations (agglomerations) of · The risk associated with unanticipated mobility-- economic activity are self reinforcing--that is, high volatility in wages, for example--is also a they are more economically dense. Richer areas disincentive to long-term investments in human may have intrinsic dynamism and yield higher capital. Clearly, reducing the high macroeconomic returns to capital and labor than poorer areas volatility of the region, as well as designing mech- where there is no natural equilibrating tendency anisms to mitigate the various types of risk-- toward geographical equality over the long run. health or income, for example--reduces poverty in There seems to be ample scope for policies that all its dimensions and has pro-growth impacts. would facilitate growth and labor mobility in 12 F R O M V I C I O U S T O V I RT U O U S C I R C L E S regions whose citizens have had particularly low America: Breaking with History? (de Ferranti and others levels of access to markets, education, and infra- 2004) showed, the poor were the primary beneficiaries of structure. Yet, as discussed in the World Bank's efforts within the region in the 1990s to provide universal Latin American region flagship Beyond the City: basic education and health services and to expand some The Rural Contribution to Development (de Ferranti public services, such as access to safe water and electricity and others 2005), investing excessive state re- (that were already provided to rich and middle-income sources in some of these areas could lower overall groups). Going forward, care must be taken to guarantee aggregate growth, and thus governments may that the poor continue to benefit from efforts to expand eventually face a growth-equity dilemma. Even in coverage of secondary and tertiary education (which up to such cases, however, a smart combination of con- now have benefited more middle- and high-income groups) ditional cash transfers for the poor and payments and to improve educational quality. In the same vein, for environmental services can enhance both poverty future investments in infrastructure must benefit laggard reduction and long-term growth. regions and increase the poor's access to those services where past expansions primarily benefited rich and middle- Policy implications income groups (telecommunications and access to the These considerations have important implications for spe- Internet, for example). cific policies. The report does not offer universal recipes to In addition, under a broad definition of poverty, two break the vicious circle between low growth and poverty. other areas have the complementary potential to reduce For one thing, different countries will likely have different poverty and promote growth. First, improvements in policy priorities; policy makers in poorer and more equal health have important impacts on welfare and demon- countries should focus mainly on growth, whereas those in strated positive effects on growth. Second, the report pro- richer and more unequal countries should try to balance vides conceptual grounds for treating the income, health, growth-enhancing objectives with policies to reduce and other risks that households face as a critical dimension inequality. Nonetheless, the following examples emerge of poverty. The macroeconomic instability arising from from the report as illustrative. unsound policy therefore has a direct impact on the well- being of the poor and a documented adverse impact on Making growth more pro-poor growth. There is no doubt that economic growth has to be at the There are, however, other pro-growth areas where Latin center of the development strategies, and numerous studies America needs to make progress but where there may be conducted by the Latin American Region of the World potential trade-offs with inequality and even with poverty Bank have explored constraints on growth that the region reduction goals in the short run, according to the results faces. For example, both the 2002 and 2003 World Bank's discussed in chapter 5. Indeed, several previous studies Latin American region flagships (de Ferranti and others, have found that trade openness (an area of particular rele- 2002, 2003) stressed the need to address the gaps in edu- vance given potential liberalization efforts) may lead to cation (particularly secondary schooling) and innovation to higher inequality through greater divergence of wage get the most out of its existing endowments and to develop incomes.2 This result appears to be related to the very dynamic new areas of comparative advantage. Similarly, the desirable adoption of technologies that tend to be skill World Bank's Latin American regional study The Limits of biased and thus enhance the returns and the demand for Stabilization: Infrastructure, Public Deficits, and Growth in education. This phenomenon, found globally, nonetheless Latin America (Easterly and Servén 2003) stressed how the leaves the poor, and often poor regions, behind in the short region's wide gaps in infrastructure implied significant lost run. Chapter 5 argues that governments may need to take opportunities in growth and welfare. complementary policies behind the border--facilitating This report offers suggestive evidence that investments access to education, expanding infrastructure to lagging in these areas have, in fact, been highly efficient in both areas with potential to tap into the benefits of liberaliza- promoting growth and allowing the poor to connect with tion, and providing conditional transfers for poor peasants that process over the last 40 years, providing a classical who may lose out in the transition. Such policies permit a "win-win" situation (see chapter 5). As Inequality in Latin country to take full advantage of the opportunities brought 13 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S about by trade opening, and thus significantly mitigate the provision of public goods (such as rural roads, health and inequality effects and considerably enhance the growth education, research and development, and extension ser- effects of trade liberalization. A parallel argument could be vices) and when policy biases against labor mobility (such made based on concerns that greater trade openness will as fiscal generosity for capital-intensive activities and stiff increase the risk that workers face. To date, little evidence labor markets) are removed. has emerged to suggest that this is true, but were it the Nor does this report delve into policies to stimulate case, income support programs could mitigate the impact more "labor intensity" within all sectors, apart from mak- on poverty and the disincentive effects on human capital ing sure that potential biases against labor use are removed. accumulation. However, the previous discussion suggests that one would Although chapter 5 suggests that financial deepening have to carefully weigh the potential adverse effects on effi- over the past 40 years appears to have had adverse impacts ciency and growth of more "active" policies in this regard on inequality and even on poverty in the short term, chap- against potential short-term gains in poverty reduction. ter 6 finds that it is precisely in countries with low access to Given the potential short-term adverse effects of trade financial services where poverty may become more of a drag opening on poverty and the negative effects of poverty on for investment and growth. Chapters 8 and 9 reinforce this growth, an area of future research regards the desirability of conclusion at the household level. Thus, even if past lim- attempting to keep undervalued exchange rates in the early ited advances in financial deepening in the region may have phases of trade opening, as long as inflationary pressures are left most of the poor behind, it is essential that future kept at bay, as Chile did after 1984 and China is currently efforts guarantee that the poor gain access to both credit practicing. and insurance markets. Now that Latin America has appar- ently succeeded in achieving more resilient financial sectors Pro-poor government policy to avoid the costly crises of the past, extending access to In the end, the relatively young literature on pro-poor credit and insurance markets appears as a key policy agenda growth has not given us a feel for how much it is possible to strengthen the virtuous circles between poverty reduc- to engineer growth in order to promote income distribu- tion and growth. tion. That the differences in the distributions of market Another strand of the literature has explored the impact incomes between Latin American and OECD countries on poverty of the structure of growth. In particular this lit- explain at most only 50 percent of differences in dispos- erature argues that the higher the representation of sectors able incomes suggests the important complementary role that use unskilled labor, the more the favorable effect on of taxes and public expenditures to ensure that the fruits poverty. Findings reported in chapter 5 give support to this of growth are broadly distributed. Chapter 5 argues that view. The potential conceptual conflict is that policies that Latin America has made relatively modest use of these induce a sectoral bias in growth may conflict in the long tools. Although recent trends toward universal basic edu- run with pursuit of a country's natural comparative advan- cation and health and the introduction of targeted condi- tage, leading to growth-impeding inefficiencies. While tional transfers (among others) are likely to have had a this report does not delve deeply into the complex (country- progressive impact on the distribution of income, many specific) issues surrounding the sources of growth and big-ticket items continue to be highly regressive: the high interlinkages across sectors or into the political economy of subsidies to pensions do not benefit the poor since they are government intervention, the evidence provided here and seldom covered; since the poor seldom finish secondary in de Ferranti and others (2005) suggests that interventions education, they do not benefit from subsidized universi- to induce strong sectoral biases are probably ill advised. A ties; gasoline, electricity, and other goods and services different matter is to ensure that policy biases and ineffi- subsidized by the state are mostly consumed by the well- ciencies against rural development, for example, are lifted to-do. and that growth opportunities are enhanced by the efficient Achieving a more redistributive and efficient pattern of provision of public goods and national and sectoral "inno- public expenditures similar to the OECD patterns would vation" policies. Incomes of the poor, including those from greatly reduce poverty and inequality. However, given the agriculture and off-farm activities, thrive with higher trade centrality of growth to the goal of poverty reduction, policy openness, when public rural expenditures focus on the makers may wish to ensure that state efforts of such 14 F R O M V I C I O U S T O V I RT U O U S C I R C L E S magnitude have favorable effects on growth. Vehicles that New Economic Geography, the case for major reorientation condition cash transfers on the acquisition of human capital of resources to disadvantaged zones becomes less clear, and could be substantially expanded. The forthcoming World the literature to date has been very circumspect on policy Bank's Latin American regional study The Redistributive prescriptions. Fundamentally, if the existing agglomera- Impact of Transfers in Latin America and the Caribbean finds tion externalities imply that those regions that are already that conditional cash transfers tend to be well targeted and most advanced are also those with the highest potential make a strong marginal contribution to social welfare, out- for growth, concentrating all types of costly infrastructure ranking not only social insurance schemes but also most of investments on poor regions may decrease national growth. the existing social assistance programs. However, the cen- Unfortunately, the literature offers little guidance on tral thesis of this report is that, in addition to conditional whether the externalities relative to agglomeration or those cash transfers, there are numerous other areas where inter- leading to dispersion of activity are more important, so we ventions to aid the poor would also be pro-growth. Some of cannot know whether existing agglomerations are too big these interventions are reviewed in the next sections. or too small. However, as indicated in Beyond the City: The First, we should emphasize once more that the relative Rural Contribution to Development (de Ferranti and others weight of different instruments depends on initial condi- 2005), some policies targeted to rural areas, such as tions in individual countries. As mentioned above, poor improved rural education and access to communications, (and more equal) countries should concentrate on achieving are clearly win-win solutions: they would increase produc- increased growth, even at the expense of some increases in tivity in agriculture and other rural activities and at the inequality, while middle-income countries with high same time increase labor mobility toward more productive inequality should aim for policies that achieve a better bal- activities and toward richer areas with higher growth ance of pro-growth and pro-poor effects (including redistri- potential. bution through conditional transfers). A more subtle use of geographic information can atten- uate the potential trade-offs to some extent. In many countries--the report looks specifically at Bolivia and Pro-growth poverty reduction Brazil--lagging regions frequently have the highest The report presents some of the first empirical evidence poverty rates, but larger urban areas actually contain the that poverty adversely affects growth at economywide most poor people. Therefore, the theoretical trade-offs, pro- levels. As noted above, a central channel appears to work viding existing agglomerations are not too large already, through underdeveloped financial sectors--more specifi- may be less important than initially thought: a large chunk cally, through the poor's lack of access to credit. This lack of the poor are, in fact, in areas with potentially higher may arise from institutional failures that make contract growth. In addition to those advanced regions with no enforcement difficult and do not address the problems of poverty, three different spatial categories emerge that information asymmetries and the poor's lack of collateraliz- imply distinct policies, some of which allow investment in able wealth. The search for efficient means and innovations potential high-growth areas with large numbers of poor to overcome information asymmetries (including credit people. bureaus) and enforcement constraints and to convert the scarce wealth of the poor into collateralizable assets are key · Areas with high poverty rates but low poverty density lack priorities for policy and further research. economies of scale arising from agglomeration exter- nalities and are unlikely to develop substantial eco- Addressing spatial concerns nomic dynamism. Policies thus need to focus more All the concerns that could potentially lead to lower eco- on direct poverty alleviation and on programs that nomic growth at the national level hold for low growth in will impart skills useful in other more dynamic subnational regions as well, and a case can be made for poli- regions. Conditional cash transfer programs or other cies analogous to those discussed above. Further, regional education and health initiatives, agricultural research inequalities correlated to ethnic, linguistic, or religious di- and development, and payments for environmental visions provide fertile ground for internal conflict that can services would be most appropriate in these circum- undermine economywide growth. Yet in the world of the stances (see de Ferranti and others 2005). 15 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S · In areas with low poverty rates but high poverty density, One of the findings of the report in this area is that public often urban or relatively dense rural areas where investments and policies in one area may have different agglomeration forces have already taken place, policies impacts depending on the existing level of assets and other aimed at fostering growth have a good chance of reach- initial conditions affecting the poor. Ensuring that poor ing the poor and translating into important poverty households have access to minimum bundles of assets (such reductions. The major problem is to ensure that as education, health, or access to infrastructure) is essential wealthy groups do not capture the flow of resources. for their capacity to exploit growth opportunities. For this reason, self-targeting mechanisms, such as On the human capital front, demographic forces offer those envisaged in the Argentine and Colombian many countries in the region a unique opportunity to workfare programs, are particularly appropriate. That translate the human capital accumulation of young cohorts said, conditional cash transfer schemes, such as those into a more productive labor force and faster reduction in in Colombia and Mexico where targeting is quite poverty. There is a need for integrated, long-term strate- good, perform well in this type of situation. gies for skills development that go beyond narrow educa- · Areas with high poverty rates and high poverty density tional policies and exploit the synergies in the life-cycle have the potential to take advantage of projects with human capital accumulation process in which both economies of scale with low levels of leakage of families and schools play a central role. This calls for resources to the nonpoor. Infrastructure investments actions to correct deficiencies in early-childhood develop- such as rural roads may be a good example of the type ment of poor children, strengthen degree completion and of projects for these kinds of areas. schooling transitions, upgrade education quality for the poor, and improve the fluidity of labor markets. The main From a practical point of view, the increasing use of specific implications for human capital formation poli- detailed poverty maps to identify poor groups and target cies are: poverty policies may yield high dividends. History suggests, however, that policy makers often · Leveling the initial playing field for children at risk. The either judge that current agglomerations are too big or unequalizing impact of deficiencies in early- allow other considerations to lead them to resist abandoning childhood development and deficient parenting on entire regions to low levels of economic activity and exten- poor children's educational attainment and returns to sive conditional cash transfer programs. In fact, as several education as adults needs to be addressed. Almost recent World Bank reports have noted, Latin America has half of the countries in the region are off track on substantial experience with ambitious regional develop- meeting the UN Millennium Development Goal of ment programs that have met with mixed success. The now halving malnutrition by 2015. Early-childhood vast OECD literature on the effects of public investment interventions and other policies that strengthen the policies generally finds a positive impact on growth and capacities of families to create early human capital sometimes inequality, although, as the Spanish case sug- should be given more attention. For example, condi- gests, they do not necessarily maximize national growth. tional cash transfer programs should systematically The evidence for Latin America is thinner but generally incorporate health and nutritional components for concurs. mothers and infants. The experience with the Head What should be emphasized, however, is that traditional Start program in the United States and similar inter- regional policy has not focused enough on the complemen- ventions elsewhere in the world may merit considera- tary roles of human capital, knowledge transmission, inno- tion for replication in the region. vation, and improved economic environments, all of which · Strengthening the full option value of education for the poor. consistently emerge as correlated with differences in Education policies should aim to strengthen transi- regional income. tions to secondary school and enable opportunities for tertiary education for the poor. While spending Addressing household concerns and reform priorities must be set according to bind- Coordinated policies are needed to reverse the vicious ing constraints, acting at all levels of the education cycles of poverty and low asset accumulation in the region. system, even on a small scale, is crucial to signal 16 F R O M V I C I O U S T O V I RT U O U S C I R C L E S low-income families that their educational invest- effects on the accumulation of human capital that, in turn, ments have better chances of maturing in higher slow down growth. Income security policies, such as unem- grades. Where returns are high and basic infrastruc- ployment insurance, workfare programs, or conditional ture is deficient, the construction and upgrading of cash transfers as used in Colombia, therefore become both schools and roads are of paramount importance. The pro-poor and pro-growth. Policies to improve access to jobs development of multigrade schools, learning from may be needed that include enacting and enforcing antidis- best practices such as the Colombian Escuela Nueva crimination laws and establishing labor market intermedi- and the Chilean MECE Rural, can address supply ation services that help well-educated ethnic and racial constraints cost-effectively; when appropriate, public- populations gain greater access to better-quality jobs. private partnerships and other modalities such as Some of the best policies from a social cost-benefit calcu- distance education should be considered. Schemes to lation, such as early-childhood interventions and overhauls use conditional cash transfers to the poor for encour- of the educational system, may be complex to implement aging completion of full courses of education (basic for reasons of political economy. However, considering the or lower secondary) may hold promise to reduce positive spillovers on technology adoption, productivity, dropouts especially of children from poor families and growth from a labor force with a minimum level of and parents with little education. Also needed are education, it is hard to overstate the critical importance of policies to promote the development of the tertiary overcoming political failures that prevent pushing "educa- education market, including student loan programs tion for all" (see de Ferranti and others 2003). This is criti- and well-designed (means-tested and merit-based) cal to the region's long-term human capital accumulation university scholarships. and prospects for sustained growth. In many countries, the · Making education count for the poor. Increasing or level- demographic window of opportunity is closing; the time to ing the returns to educational investments of the invest is now. poor is key to encourage them to move up the educa- Bridging the gaps in both the quantity and quality of tion ladder. Well-informed actions to improve the education and other productive characteristics of workers scholastic performance of poor children are needed. can go a long way toward reducing the wide earnings dis- These may include removing automatic promotion parities in the region, but it will not be enough to reduce policies in early grades, offering special programs to poverty significantly. In most countries, low levels of labor address learning deficiencies resulting from a poor productivity are a chief constraint to earnings potential. learning environment at home, and addressing fail- Policies that promote an economic and institutional envi- ures in the instruction process such as inadequate ronment conducive to productivity growth are thus impor- teaching and large class sizes. Effective interventions tant to reduce the incidence of low-paid jobs and in turn include decentralizing school management to get make investments in skills more attractive. parents more involved in their children's school For example, rural investments seem to correlate posi- progress, offering incentives to encourage qualified tively with rural household characteristics, indicating a need teachers to work in disadvantaged schools, adapting to increase access to markets through expansion of basic innovations to improve learning environments in infrastructure while simultaneously strengthening the disadvantaged schools and communities, upgrading capacity of households to ensure a minimum level of wealth textbooks and school aids, providing teacher train- and education skills. ing, expanding computer education in secondary Rural development could be made more inclusive with schools, and consistently using international stan- some minimum coordination of rural investments and dardized tests to assess performance progress. Some programs--such as education, the construction of roads to targeted and performance-based increases in public markets, the establishment of microcredit schemes, and the expenditures, particularly at the secondary level, provision of agricultural extension--to ensure that all might be needed in some countries. the potential returns to these investments are realized and the conditions of the rural poor improved. A minimum Finally, chapter 9 shows that the higher levels of labor coordination of public interventions in poor areas can help market risk found in the region have strong disincentive exploit synergies and overcome the associated potential 17 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S poverty traps that may affect households with a bundling of a trade-off between targeting and coverage: the greater the unfavorable characteristics. number of poor covered by a program, the more difficult it is to avoid leakages. A careful review of existing social pro- How are we going to pay for these interventions? grams, however, can result in significant savings that may This report offers a relatively large number of areas that be redirected to priority areas. Even more important, may require additional attention if the vicious circle although they would require politically difficult reforms, between growth and poverty is to be converted into a virtu- highly regressive subsidies--of pensions for the well-to-do, ous circle. For example, it urges that the levels of human of university students from wealthy families or who pay capital and public infrastructure in the region be expanded, back educational credits, and of the consumption of energy in particular by increasing the poor's access to quality edu- by the middle class and the rich--offer huge opportunities cation and infrastructure. Similarly, it argues that an to reallocate expenditures. expansion of conditional cash transfer programs (especially Once these potential gains have been tapped, and once in richer countries) would likely have a sustained impact on efforts to curtail tax evasion have been stepped up, policy poverty reduction and growth. But what are the real possi- makers can consider increasing tax rates. In this regard, bilities the region has for financing these interventions, chapter 5 argues that most countries in the region (with a which in some cases can be quite expensive? few exceptions such as Brazil and Nicaragua) have tax col- It is crucial that policy makers step up efforts toward lections that are below what would be expected from their improving the efficiency of the system and achieving better per capita income (figure 1.12). This, too, is a window of targeting before they increase public spending. For exam- opportunity because bringing Latin America in line with ple, as noted in chapter 5, a number of big-ticket items the international experience in tax collections would allow such as tertiary education are highly regressive. Moreover, some extra space to finance part of the expenditure priori- many public transfer programs such as pensions or unem- ties of the region. One related issue discussed in chapter 5 ployment insurance are typically poorly targeted and do is that countries aiming at increasing tax collections should not reach many of the poor. Policy makers are likely to face avoid, to the extent that it is possible, tax structures with FIGURE 1.12 Total tax revenue versus per capita income, throughout the world Total tax revenue (% GDP) 45 LAC Selected countries throughout the world 40 Italy France 35 30 Estonia Spain Uruguay 25 20 Brazil Chile United States Nicaragua Costa Rica Honduras Peru 15 Dominican Rep. Mexico Argentina Paraguay Bolivia Colombia 10 El Salvador Guatemala 5 0 4.5 5.5 6.5 7.5 8.5 9.5 10.5 11.5 Log, per capita GDP Source: Authors' calculations. 18 F R O M V I C I O U S T O V I RT U O U S C I R C L E S high efficiency costs. Latin American countries tend to have Converting the state into an agent that promotes equal- especially low levels of collections from personal income ity of opportunities and practices efficient redistribution is, and property taxes--the very taxes that may have some perhaps, the most critical challenge Latin America faces in redistributive effect without large costs to economic implementing better policies that simultaneously stimu- growth. Thus well-designed systems could increase tax col- late growth and reduce inequality and poverty. lections while keeping the impact on growth low. Also, the region's value added and income tax productivity is signifi- Notes cantly lower than it is in the OECD countries, and most 1. The Gini coefficient is a standard measure of inequality that ranges between 0 and 1. A value of 0 would indicate a perfectly equal Latin American countries maintain a large set of exemp- distribution. As inequality increases, the Gini coefficient also tends tions that significantly reduce the tax base. Thus the elimi- to increase. nation of exemptions combined with additional efforts to 2. See, for example, de Ferranti and others (2003); Lederman, enforce compliance would likely increase collections. Maloney, and Servén (2005); and World Bank (2005c). 19 CHAPTER 2 Dimensions of Well-Being, Channels to Growth This chapter reviews recent trends in poverty and inequality in Latin America and the Caribbean, along with the well- known concerns about the implications of static measures of poverty and inequality. The review shows that such concerns are not merely conceptual curiosities--incorporating them in the analysis can and does lead to very different conclusions about the evolution of welfare in the region and complicates inferences about the effect of growth on the welfare of the poor. As important, however, these more complete measures of welfare open several additional channels through which poverty or inequality can affect growth. T HE PERSISTENCE OF HIGH LEVELS OF the reverse causality may occur and thus prevents the poverty remains the central disappointment fullest understanding possible of the virtuous circles of the last 20 years in Latin America. This between poverty reduction and growth. As is generally the chapter begins by presenting the standard case with these reports, we aim not to provide the final indicators of income poverty and inequality word, but rather to contribute some new ideas or, in this for the region--the share of the population living below $2 case, some new evidence on old ideas, to the debate. a day and Gini coefficients--their recent evolution, and some caveats surrounding the conclusions we draw from Income poverty them. Table 2.1 suggests that the rate of income poverty in Latin However, it has long been acknowledged that such indi- America is 24.6 percent, based on a poverty line of $2 a day cators are very imperfect measures of well-being, both of in purchasing power parity (PPP) weighted by population the poor and of the society as a whole.1 Many of the points and using the latest available surveys.2 It is somewhat made in this chapter were foreshadowed in Kuznets's semi- higher in Central America and Mexico (30 percent) and the nal "Economic Growth and Income Inequality," published Andean Community (31 percent) and lower in the coun- in 1955; others were made by Sen (1985). Yet in the con- tries of the Southern Cone (around 19 percent), which text of understanding the reinforcing relationship between nonetheless have a larger number of the poor by virtue of growth and poverty reduction, these points gain renewed their larger populations. The sample does not have compa- importance. First, to understand how growth may affect rable measures for the Caribbean as a whole, but the two the poor, we need to understand the channels through most populous countries (excluding Cuba) have poverty which different characteristics of growth affect the quality rates of 16.4 percent (Dominican Republic) and 44.1 per- of life of individuals across dimensions of well-being, across cent (Jamaica). Very similar patterns emerge when working their lives, and across generations. with unweighted averages, which are more relevant when Second, excessive narrowness in understanding poverty the analysis requires taking the country as the unit of can lead to overlooking important channels through which analysis rather than the individual.3 21 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S TABLE 2.1 Poverty in Latin America (US$2 a day headcount poverty) FIGURE2.1 Poverty in selected Latin American countries Early Early Last 1990s 2000s survey Change Argentina Region (i) (ii) (iii) (iii) ­(i) Bolivia A. Southern Cone Brazil Poverty (weighted) (%) 23.6 19.0 18.8 -4.9 Poverty (unweighted) (%) 18.1 16.2 17.1 -1.1 Chile Population (million) 204.4 244.4 246.4 42.1 Number of poor (million) 48.3 46.5 46.2 -2.1 Colombia B. Andean community CostaRica Poverty (weighted) (%) 24.8 34.9 31.4 6.6 Poverty (unweighted) (%) 30.6 37.2 34.0 3.4 DominicanRep. Population (million) 94.4 118.3 118.0 23.6 Number of poor (million) 23.4 41.3 37.1 13.7 Ecuador C. Central America and Mexico ElSalvador Poverty (weighted) (%) 30.5 29.2 29.2 -1.3 Poverty (unweighted) (%) 36.5 30.0 30.1 -6.4 Honduras Population (million) 112.7 140.4 139.6 26.8 Number of poor (million) 34.4 41.0 40.8 6.4 Jamaica Latin America (A+B+C) Mexico Poverty (weighted) (%) 25.8 25.6 24.6 -1.2 Poverty (unweighted) (%) 29.3 28.1 27.4 -1.9 Nicaragua Population (million) 411.5 503.1 504.0 92.6 Panama Number of poor (million) 106.1 128.8 124.1 18.0 Paraguay Source: Gasparini, Gutierrez, and Tornarolli (2005). Peru Note: Weighted refers to population-weighted averages. Uruguay Figure 2.1 offers a closer examination of the great vari- R.B.deVenezuela ety of poverty levels across countries. Chile and Uruguay have the lowest poverty rates (about 5 percent) followed 0 10 20 30 40 50 60 70 80 very closely by Costa Rica (9 percent). At the other Percent extreme, despite the significant progress made over the past Livingon$2orlessperday few years, poverty in Nicaragua remains at levels of 50 per- Livingbelowthenationalpovertyline cent. Although comparable numbers for Haiti are not Source: Gasparini,Gutierrez,andTornarolli(2005). available, other sources show it to have the most extreme Note: Basedonthelatestavailablesurvey. poverty, at between 73 percent and 83 percent.4 These are followed by several countries with poverty levels around 40 percent (including Bolivia, Ecuador, El Salvador, 15 years. The weighted average poverty rate declined by Guatemala, Honduras, and Jamaica). Among the most only 1.2 percentage points between the early 1990s and the populated countries, poverty rates are slightly above last available survey, and of this decline a significant com- 30 percent in Mexico, Peru, and República Bolivariana de ponent was probably related to the recent recovery of the Venezuela; about 20 percent in Brazil and Colombia; and regional economy in 2003 and 2004.5 Again, there are sub- about 16 percent in Argentina. stantial regional differences. Poverty fell slightly in Central Nationally defined poverty tends to be higher than the America (from 30 to 29 percent), increased in the Andean measure of $2 a day in most of the countries, although the Community (from 25 to 31 percent, with a peak of 35 per- differences between these two measures are not uniform cent in the early 2000s), and declined in the Southern Cone across countries (box 2.1). area (from 24 to 19 percent).6 In the Caribbean, Jamaica Table 2.1 also suggests that the region has made rela- experienced a decline in poverty of 15 percentage points tively little progress in reducing poverty over the past between the early 1990s and early 2000s, while the 22 D I M E N S I O N S O F W E L L - B E I N G , C H A N N E L S T O G R O W T H BOX 2.1 Income poverty lines Income poverty is defined as the inability to achieve a the Caribbean), which in some cases helps governments to certain minimum income level, known as the poverty calculate the national poverty lines. Despite some similari- line. Even this limited definition can be contentious ties, methodologies for estimating national poverty levels because there are neither normative nor objectively clear differ substantially across nations so they are not compara- arguments for setting the line at a particular value below ble. Some countries, such as Mexico, use expenditures; which everybody is poor and above which everyone is others, such as Argentina, use incomes; and still others, nonpoor (Deaton 1997). Despite this central conceptual such as Bolivia, use a mix of income and expenditures. ambiguity, reducing poverty is still a deliberate policy Both international and national measures of poverty objective for governments around the world and has been are useful. Measurements that use national poverty lines embraced as a Millennium Development Goal by the take into consideration the different criteria societies use international community. to identify the poor, while international poverty lines Because of the fundamental arbitrariness in defining are indispensable instruments for comparing absolute poverty, different authors and agencies use different poverty levels and trends across countries and providing poverty lines. The international poverty line is set at $1 a regional and world poverty counts. day per person at purchasing power parity (PPP) prices. Nationally defined poverty tends to be higher than $2 That measure is meant to define an international norm to a day in most of the countries in Latin America, although gauge the inability to pay for food needs. The $1-a-day the differences are not uniform across countries. More- line was formally proposed in Ravallion, Datt, and van de over, in three countries--Jamaica, Ecuador, and Walle (1991) and is generally used in the World Bank's Nicaragua--the national poverty lines are lower than the 1990 World Development Report. It is a value measured in internationally defined poverty line. As a result, the 1985 international prices and adjusted to local currency poverty ranking in the LAC region changes significantly using purchasing power parities to take local prices into when one focuses on national poverty lines. Based on account. The $1 standard was chosen as being representa- national poverty lines, poverty is highest in Honduras tive of the national poverty lines found among low- (above 70 percent), Colombia and Peru (about 55 per- income countries. The line has been recalculated in 1993 cent), and Mexico (51 percent) and lowest in Chile, Costa PPP terms at $1.0763 a day (Chen and Ravallion 2001). Rica, and Jamaica (around 20 percent). This value is multiplied by 30.42 to get a monthly Comparison of the comparable international and poverty line. Although the $1-a-day line has been criti- national poverty figures indicates that in some countries cized, its simplicity and the lack of reasonable and easy- like Argentina, Colombia, Honduras, and Mexico, the to-implement alternatives has made it the standard for national definition of poverty is quite generous (people international poverty comparisons. It is, for example, the are being classified as poor in these countries who might basis of the United Nations' Millennium Development not be considered poor in other countries of the region). Goal 1, which calls for eradicating extreme poverty and In contrast, Chile, Costa Rica, El Salvador, and Paraguay hunger by halving between 1990 and 2015 the propor- appear to use poverty concepts that are very exclusive tion of people whose income is less than $1 a day. A $2-a- (people who are not considered poor in these countries day line is also extensively used in comparisons across might qualify as poor in others). It is worth noting that middle-income countries and is periodically presented in in some cases the deviations from the regression line are the World Bank's World Development Indicators. quite important. For example, in Honduras the national Most Latin American countries calculate two poverty poverty rate is 35 percentage points above the interna- lines: national extreme poverty, which is based primarily on tionally comparable poverty rate, whereas in Jamaica it is the cost of a basic food bundle, and moderate poverty, com- 21 percentage points below. puted from the extreme lines using the Engel/Orshansky ratio of food expenditures. This methodology is also used by ECLAC (Economic Commission for Latin America and Source: Gasparini, Gutierrez, and Tornarolli (2005). 23 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S at annual rates above 4 percent per capita over the FIGURE 2.2 1990­2003 period), growth in Latin America during the The evolution of Latin American poverty during the 1990s 1990s was low. Per capita growth for the region as a whole Percent averaged about 1 percent between 1990 and 2003 (see 30 box 2.2 for a discussion of differences in the measures of 29 growth). At this growth rate, per capita GDP doubles every 28 27 65 years. That implies that on a continuous trend, the 26 region would need about 150 years to reach the per capita 25 income level of the United States today. The median 24 growth rate for the region during the 1990­2003 period 23 Early 1990s Mid-1990s Early 2000s was also around 1 percent, indicating that the poor perfor- mance is not the result of a few of the most populated coun- Source: Authors' calculations. Note: The data refer to unweighted poverty rates. tries displaying low economic growth. In fact, only half of the countries in the region managed to grow at rates above 1 percent. Similarly, fewer than one in four countries aver- Dominican Republic sustained an 8-percentage-point aged per capita growth above 2 percent. increase over the same period. Inequality trends were dealt with in great detail in our Figure 2.2 suggests that the decadal averages, in fact, flagship report Inequality in Latin America and the Caribbean, obscure important dynamics.7 The regional poverty rate Breaking with History? (de Ferranti and others 2004); here may have fallen by almost 4 percentage points between the we offer only a historical view of the evolution of the early and mid-1990s, a period of expansion, and increased regionwide Gini coefficients since 1950 (figure 2.3). After by almost 3 percentage points between the mid-1990s and some progress in the 1960s and 1970s, inequality levels early 2000s following the financial crises of East Asia in rose during the lost decade of the 1980s; this increase was 1997 and Russia in 1998. not reversed during the 1990s and may, in fact, have con- The lack of progress on the poverty front since 1980 is tinued. As chapter 4 discusses in detail, the level of inequal- caused both by low average economic growth rates during ity is an important factor in how "pro-poor" growth is. the period (table 2.2) and by the high and generally stag- As box 2.3 suggests, however, this picture of inequality nant levels of income inequality in the region. Despite may be overly pessimistic. Poverty lines need to be some success stories such as Chile (which managed to grow adjusted for inflation across time, and Goñi, Lopez, and TABLE 2.2 Economic growth in Latin America Region 1990­93 1993­97 1997­2000 2000­03 1990­2003 A. Southern Cone Growth (weighted) 2.27 2.85 0.32 -0.52 1.35 Median 3.22 3.16 -0.55 -1.38 0.99 B. Andean community Growth (weighted) 0.95 1.84 -1.79 -0.40 0.27 Median 0.58 1.83 -0.55 0.87 0.52 C. Central America and Mexico Growth (weighted) 1.41 0.76 3.21 -0.95 1.07 Median 3.30 1.14 2.47 -0.37 1.38 Latin America Growth (weighted) 1.78 2.08 0.77 -0.61 1.08 Median 2.08 1.76 0.37 0.46 1.04 Source: Authors' calculations. 24 D I M E N S I O N S O F W E L L - B E I N G , C H A N N E L S T O G R O W T H Servén (2005) show that standard inflation numbers corre- FIGURE 2.3 spond to the consumption basket of the very well-off and Gini coefficient for Latin America, 1950­2000 greatly overstate the level of inflation relevant to the poor. 0.60 Hence, deflating poverty lines, or each income share com- prising the Gini, by the common consumer price index 0.55 (CPI) imparts a strongly antipoor bias to the summary 0.50 statistics during this period. The implications of these findings are far reaching. To 0.45 begin, Latin America is doing better than was initially 0.40 thought on the poverty and distribution fronts, and hence 1950 1960 1970 1980 1990 2000 concerns about the negative distributional impacts of Source: Authors' calculations based on Altimir (1987) and Londoño reforms have probably been overstated. Second, real figures and Szekely (2000). Note: Based on data for Brazil, Chile, Mexico, and República obtained using incorrect deflators may potentially confuse Bolivariana de Venezuela. the relationship between different types of growth strategies BOX 2.2 National accounts and household surveys­based growth: How different are they? In a joint analysis of poverty and growth, one issue that growth rates, with national accounts data usually produc- must be considered is the source of the data used to com- ing higher estimates than household surveys (see Deaton pute the growth rates. The Latin American growth 2005 for a discussion). trends reviewed here are based on the evolution of The figure plots the growth rates based on surveys national accounts (NA) data, whereas poverty rates are against those based on the national accounts. Two large computed on the basis of household surveys. If the outliers are apparent in this figure, one in the southwest implied growth rates of the NA and the surveys were the quadrant (PRY, or Paraguay) and the other in the south- same, then using survey-based poverty rates and national east quadrant (DOM, or Dominican Republic). The accounts growth rates to analyze the evolution of poverty regression line in this chart has an associated slope of and growth over time would not be misleading. In prac- 0.97 and an intercept of about -0.9. While the estimated tice, however, surveys and NA tend to generate different slope suggests an almost one-to-one relationship between the growth rates derived from the two sources, the nega- tive intercept indicates that national accounts growth Survey-based income growth versus national accounts­based income growth rates tend to be much higher (almost 1 percentage point) than survey-based estimates. Income growth according to household surveys, % What does this difference imply in practice? First, 6 since changes in poverty are related to changes in house- JAM CRI 4 NIC BOL CHL hold survey­based income growth, it could be perfectly ECU SLV 2 possible that an increase in poverty associated with a PER BRA HND 0 COL PAN national accounts­based growth episode would be 2 MEX URY observed (especially at low growth levels). Instead of 4 VEN ARG reflecting an antipoor growth episode, the increase in y 0.9766x 0.8646 6 poverty would just capture the existing statistical dis- 8 PRY DOM crepancy between two different data sources. Second, if 10 the difference between national accounts and household 12 survey­based data results from a bias in the survey data, 3 2 1 0 1 2 3 4 5 Income growth according to national accounts, % then the poverty statistics will be biased upward. 25 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S BOX 2.3 Inflation inequality: What really happened to LAC poverty and inequality Rich and poor families consume different baskets of Individual inflation by decile and average annual inflation by goods, and the inflation rates of these baskets can differ viniventiles greatly. Goñi, Lopez, and Servén (2005) show that using Peru, 2001­3 the aggregate CPI can greatly mislead policy. For one % inflation thing, tax brackets, pensions, social transfers, and mini- 2.0 mum wages are often indexed to the CPI, and using an 1.9 inappropriate aggregate index can lead to real transfers 1.8 among income classes that were not intended. In addi- 1.7 tion, the picture of the evolution of poverty and inequal- Pi 1.6 ity can be sharply distorted by assuming that deflators 1.5 are similar across income classes, either by working 1.4 Pih with undeflated nominal baskets of goods, or by using 1.3 aggregate deflators, and contaminating inference about 1.2 the relationship between these variables and growth or 1.1 policy. 1.0 In Latin America and the Caribbean, as in the OECD, 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95100 most officially reported inflation rates correspond to the % of population inflation rates of the very rich--defined as those with income between the 80th and 90th percentiles; for the centage points a year. These patterns persist even after very rich, inflation is relatively high, as the figure for adjusting for quality change bias and after recomputing Peru shows. In Brazil (1988­96) the inflation differential Paasche indexes to control for potential substitution between the highest and lowest viniventiles (5 percentile effects. intervals) is 7 percentage points a year and in Colombia Since most inequality indexes are calculated using (1997­2003), Mexico (1996­2002), and Peru (2001­3), nominal expenditures, such inflation differentials lead to the difference is a lower but still noticeable 0.5­0.7 per- apparent movements in nominal inequality without any Distribution effects of inflation Inequality t1 Inequality t2 Price Quantity Period (Gini) (Gini) Change (%) change change Brazil 1988­96 0.54 0.55 1.60 2.17 -0.58 Colombia 1997­2003 0.53 0.50 ­5.49 1.92 ­7.41 Mexico 1984­89 0.50 0.50 ­0.20 2.77 ­2.97 1989­94 0.50 0.49 ­1.85 1.38 ­3.23 1994­96 0.49 0.46 ­6.88 ­1.30 ­5.57 1996­2002 0.46 0.49 6.32 1.42 4.90 Peru 1995­99 0.46 0.50 9.91 1.28 8.63 1999­2001 0.50 0.49 ­2.72 1.05 ­3.78 2001­03 0.49 0.48 ­1.21 0.47 ­1.67 26 D I M E N S I O N S O F W E L L - B E I N G , C H A N N E L S T O G R O W T H real movement, much the way nominal growth rates may inequality measures overstated the changes in real rise even if there is no real growth. To measure the inequality and importantly so. In six of the eight cases magnitude of these distortions, we first recalculate the (Brazil 1988­96, Colombia 1997­2003, Mexico expenditure of each household in the first period with 1984­89 and 1989­94, and Peru 1999­2001 and prices of the second period to get the "real" changes in 2001­3), the change in prices offset the effect of changes inequality. Analogously, the difference in the inequality in quantities. In Brazil (1988­96) the real distribution of index caused by revaluing the first-period bundle using income improved despite an apparent increase in the second-period prices gives us "nominal" changes in Gini. Similarly, in Mexico (1984­89) the Gini showed a inequality. small improvement in inequality (-0.2), whereas the real The table shows the distribution effects of inflation decline was much larger (-2.97). Finally, there are two and suggests that these distortions are very important. cases (Mexico 1996­2002 and Peru 1995­99) where First, in only one of the nine time spans do prices exert a price and quantity effects reinforced each other to exag- negative contribution on nominal inequality (Mexico, gerate worsening inequality, with prices contributing 1994­96): during the tequila crisis, inflation was 23 percent and 12 percent, respectively, of the total vari- antipoor and led to a lower reduction in real inequality ation in nominal inequality. than suggested by the standard inequality figures. How- ever, in all the other cases, the changes in the standard Source: Based on Goni, Lopez, and Servén (2005). ~ and their impact on poverty. For instance, liberalizations expensive car, the value of their consumption will appear to and devaluations, by their design, have the goal of changing rise. Since the consumption share of the poor is falling and relative prices of goods within the economy. When assess- that of the rich is rising, the Gini will appear to worsen ing the impact of trade liberalization on the poor, for exam- even though, in real terms across the course of their lives, ple, one needs to ask not only what the impact is on the distribution has without question improved. The example production side--labor income--but also on the specific highlights both the desirability of working in real terms basket of goods consumed by the poor. Liberalization of and the need to introduce the intertemporal considerations trade in corn in Mexico under NAFTA (the North Ameri- discussed below. can Free Trade Agreement) could have led to lower prices that reduced the income of poor corn producers. But one Beyond income and consumption must also take into account the decline in the cost of maize, It has long been acknowledged that measures of income or a key element in the consumption basket of the poor. As a consumption poverty and distribution capture well-being result, the CPI of the poor falls relative to that of the well- only very imperfectly. Sen's celebrated "capacities" off, which is what the national CPI measures. The poor, approach to poverty analysis stresses the centrality of often both urban and rural, are in fact better off than the national overlooked dimensions of deprivation. In his book Develop- CPI would suggest. In a symmetrical way, an increase in ment as Freedom, for example, Sen (1999) argues that the price of cars caused by new export opportunities would Europe's favorable measures of income inequality relative affect the bundle of the rich far more than that of the poor to those in the United States are offset to an important who consume them less. degree by high unemployment rates in Europe that inhibit The striking fact is that, in both cases, if the price participation in the labor market and associated social changes do not lead to major substitutions away from these networks. In another example, he notes that despite their goods, the Ginis will move in unexpected directions even if relatively high money incomes, African American men calculated correctly. If the poor save the money gained from have lower average life spans than Chinese, Costa Ricans, or buying maize more cheaply, their nominal consumption Jamaicans. Deaton and Paxton (2001) and Becker, Philip- will appear to fall, and if the rich borrow to buy the more son, and Soares (2005) document this fact more rigorously: 27 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S clearly there is a component of the health dimension of into an average monetary gain of roughly $1,365 per capita, well-being that is uncorrelated with income and thus needs or roughly half the monetary gain (table 2.3). But as impor- to be somehow integrated separately into comparisons of tant, progress in income and longevity has not always been welfare. Since the Millennium Development Goals focus highly correlated, and in some countries--Bolivia, El attention on deprivation in multiple dimensions, this Salvador, Honduras, and Peru--the greater part of the wel- agenda is extraordinarily relevant. fare gains has been in longevity, with life expectancy However, it is far from trivial to operationalize.8 Mar- increasing 20 years while incomes remained relatively kets for some proposed attributes of poverty--longevity, stagnant. the provision of public goods, security, even freedom and Improvements in life expectancy during this period literacy--are imperfect or do not exist and thus provide took place across different age groups and causes of death, little guidance on their relative values to the poor.9 As but most were concentrated at early and old ages and were Atkinson and Bourguignon (1982) show, adding just one driven by reductions in mortality from infectious diseases, dimension (in their case, adding mortality to income) raises respiratory and digestive diseases, congenital anomalies the complexity of welfare comparisons significantly: the and perinatal period conditions, and heart and circulatory conclusions about how much and in which direction wel- diseases. These in turn appear to be driven by improve- fare changed for 61 countries between 1960 and 1970 ments in health infrastructure and large-scale immuniza- depend heavily on what particular form of the social wel- tions that increased substantially across the period. Soares fare function is used to combine the two dimensions. (2005) finds similar patterns looking across Brazilian The same indeterminacy emerged in rural Brazil when municipalities. Life expectancy gains were largely inde- Bourguignon and Chakravarty (2003) sought to combine pendent of income, but represented between 22 and income poverty and "educational poverty" measures, which 35 percent of welfare gains across municipalities. More moved in opposite directions.10 Recent ferment in this than half of these gains, 51 percent, can be explained by literature has generated numerous techniques for multidi- improved access to water and sanitation and greater mensional comparisons, and a careful discussion is beyond literacy. the scope of this report.11 What is clear, however, is a Soares (2004) also looks at how an environment of inse- consensus that researchers need to look beyond traditional curity and violence affects welfare. He calculates that, glob- income measures and that nonincome dimensions of ally, reducing violence rates to zero would add an average of poverty are of important magnitudes and can radically one-third of a year in life expectancy at birth that would change the view of the evolution of well-being. have a lifetime value of approximately 15 percent of GDP. One approach to quantifying these magnitudes is offered For Colombia, Soares calculates that violence reduces life by Becker, Philipson, and Soares (2005), who convert life expectancy by 2.2 years, representing a welfare loss on the span into monetary values to calculate a measure of total order of 100 percent of current GDP; for Brazil, the welfare welfare gain by calculating how much people would pay for loss is 38 percent. an additional year of life (annex 2A). Globally, convergence Although these calculations depend on assumptions in life expectancy has been impressive compared with con- that may be debated, at a minimum they suggest that these vergence of incomes, with the "longevity Gini" halving dimensions of well-being are not well captured by income from 0.13 to 0.07 even as the income per capita Gini and are of sufficient magnitudes that they cannot be omit- decreased only slightly. Looking at Latin America and the ted from the picture of the well-being of the poor. And Caribbean more specifically, Soares (2004) argues that both longevity and violence potentially have important longevity and hence welfare have increased substantially impacts on growth. The issues related to health are dis- despite continued political instability and almost perma- cussed in chapter 7. Those related to violence have been nent crisis over the last 25 years. Between 1960 and 2000, reviewed by Bourguignon (2001) and Londoño and Guer- average per capita income in the region doubled, from rero (2000) and will not be developed further here.12 In $3,419 to $6,865 (in 1996 international prices). At the sum, not only are direct impacts on welfare obtained from a same time, average life expectancy at birth increased by focus on a broader measure of poverty, but these then can 13 years, from 57 to 70 years, an increase that translates feed back into growth. 28 D I M E N S I O N S O F W E L L - B E I N G , C H A N N E L S T O G R O W T H TABLE 2.3 Welfare gains from increased longevity Value of Health share Life expectancy life expectancy of welfare Income per capita (US$) at birth (years) gains (US$) gain (%) Region/country 1960 2000 1960 2000 1960­2000 1950­2000 Europe and Central Asia 6,813 13,864 68 73 1,454 17 East Asia and Pacific 1,319 5,667 47 70 2,600 37 Middle East and North Africa 1,911 4,898 48 68 1,719 37 North America 12,378 31,761 70 77 2,804 13 South Asia 888 2,269 44 62 635 31 Sub-Saharan Africa 1,442 1,583 41 47 73 34 Latin America and the Caribbean 3,419 6,865 57 70 1,365 28 Argentina 7,386 11,201 65 74 1,071 22 Barbados 6,007 15,850 65 75 2,174 18 Bolivia 2,152 2,701 43 63 881 62 Brazil 2,514 6,989 55 68 1,380 24 Chile 3,919 9,591 58 76 2,383 30 Colombia 2,481 5,393 57 71 951 25 Costa Rica 3,514 5,597 62 78 850 29 Dominican Republic 1,698 4,967 53 67 1,157 26 Ecuador 2,100 3,413 54 70 668 34 El Salvador 3,411 4,339 52 70 1,130 55 Guatemala 2,613 4,005 46 65 1,288 48 Honduras 1,682 2,082 47 66 468 54 Jamaica 2,301 3,286 65 75 283 22 Mexico 3,976 8,391 58 73 1,941 31 Nicaragua 3,204 1,672 48 69 399 -35 Panama 2,453 6,134 61 75 926 20 Paraguay 2,053 4,545 64 70 277 10 Peru 3,179 4,479 49 69 1,482 53 Trinidad and Tobago 3,922 10,557 64 73 1,394 17 Uruguay 5,835 9,919 68 74 624 13 Venezuela, R.B. de 4,480 6,279 60 73 1,062 37 Source: Becker, Philipson, and Soares (2005) calculations. Why not just ask them? tive responses contain real content and that a wide variety Given the difficulties in combining nonmonetary mea- of factors go into the consideration of being poor, consis- sures, a reasonable question might be: "Why not just ask tent with a multidimensional poverty approach. Third, people whether they regard themselves as poor?" This has probit analyses by Arias and Sosa-Excudero for Bolivia sug- recently been done in Argentina (Lucchetti 2005), Bolivia gest that these characteristics appear to be highly similar in (Arias and Sosa-Escudero 2004), and the Dominican their influence on both subjective and objective measures Republic (World Bank 2005b), generating some striking (figure 2.4). conclusions. First, the subjective surveys and income mea- Finally, there are some notable exceptions to these gen- sures generate similar numbers of households in poverty, eralizations; we offer four examples: with roughly 65 percent of the households falling under First, in Argentina, being unemployed has an effect on the poverty line also reporting that they are poor. Second, self-rated poverty that is four times higher than would be in all cases, many and varied household characteristics carry predicted by the objective poverty line. This is consistent a very high statistical significance as determinants of with Sen's idea that being effectively excluded from the subjective poverty. This finding suggests both that subjec- workforce has impacts on well-being extending beyond 29 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 2.4 Income poverty profile for Bolivia: self-rated by head of household versus data driven Education Employment Years of school 17 Self-employed 16 48% self- 48% self- 15 Employee rated poor rated poor 14 Employer 13 55% income 55% income poor poor 12 Underemployed 11 10 Out of labor force 9 Unemployed 8 7 Employed 6 5 Private 4 Public 3 2 White collar job 1 Blue collar job 0 0 20 40 60 80 0 20 40 60 80 Percent Percent Demographics Living conditions Women Electricity Men Toilet 48% self- rated poor Computer Married 55% income poor 48% self- Radio Divorced rated poor Television Single 55% income poor Refrigerator Spanish Telephone Quechua Good Quality Roof Aymara Average Quality Roof Other Poor Quality Roof 0 20 40 60 80 0 20 40 60 80 Percent Percent Income poor Self-rated poor Source: Arias and Sosa Escudero (2004). Note: Income poverty measures are based on household income per capita for urban areas and rural per capita expenditures. The self-rating was done by the head of household, who was 18 years or older. 30 D I M E N S I O N S O F W E L L - B E I N G , C H A N N E L S T O G R O W T H immediate income. In Bolivia indigenous groups are twice fundamentally human behavior attests, individuals are con- as likely as the average Bolivian to rate themselves as poor cerned with their welfare across their entire life span, not if they are unemployed. just at any instant. Yet the scarcity of longitudinal (panel) Second, in Bolivia, informal, self-employed workers feel data sets in developing countries has made a life-cycle per- less poor than their incomes would predict, indicating, per- spective difficult to introduce into welfare measures. This haps, that there is a premium on flexibility or on being one's absence severely distorts our picture of poverty and own boss as some of the recent literature on informality sug- inequality. As an extreme example, imagine a country gests (Maloney 2003). In the Dominican Republic there is where every young person begins earning wages that place no difference between self-employed and other workers, them below the poverty line, but where the returns to each suggesting that the self-employed feel no special vulnerabil- additional year of experience (accumulated human capital) ity relative to salaried workers, while in Argentina, where are so large that everyone dies a millionaire. Despite the high rates of unemployment may have increased the share of fact that everyone has equal lifetime welfare, the staggered involuntarily self-employed, the reverse is the case--the distribution of ages in the population will reveal substan- self-employed do feel more vulnerable. tial poverty and inequality in a single cross-section.13 Third, some of the largest discrepancies are among Ignoring this mobility renders static measures of poverty regional and ethnic groups. Bolivian Quechuas tend to rate and distribution deeply suspect, as Kuznets (1955, 2) themselves as poorer than suggested by income poverty pro- bluntly argued: files, while the converse is true for Bolivian Aymaras. Even though Gran Buenos Aires is the second richest region in To say, for example, that the "lower" income classes Argentina, its inhabitants feel especially poor, perhaps gained or lost during the last twenty years in that reflecting larger observable income differentials among their share of total income increased or decreased has households, or congestion externalities in a larger city. meaning only if the units have been classified as As a final example, Velez and Nunez (2005) attempt to members of the "lower" classes throughout those explain the apparent increase in reported subjective well- 20 years--and for those who have moved into or out being in Colombia where the share of the poor ranking their of those classes recently, such a statement has no living conditions as "good," the top of the scale, rose by significance (italics added). 16 percent from 1997 to 2003. Given the deep recession across the period, income is not driving the ranking. Calcu- The appropriate focus on welfare across the life cycle intro- lations using eight different techniques to measure two- duces two new elements into discussions of distribution and dimensional poverty indicators capturing income plus poverty and their link to growth: mobility and risk. security and income plus home crowding still showed wors- ening poverty. Income plus educational gains did show Mobility declining poverty for many techniques, although the results The link between the snapshot Gini we see and true long- were again very ambiguous when these two factors were term income inequality is mobility through the income dis- combined with security in a three-dimensional poverty tribution. This need not be unidirectional, as in the example indicator. In the end, Velez and Nunez speculate that their above. Atkinson and Bourguignon (1982) and Shorrocks indicators may be missing expectations of a much improved (1993) stress that reversals of position--a poor person security situation in light of the dramatic changes in policy becoming a millionaire and vice versa--make lifetime since 2002 and perhaps redistributive programs that dou- incomes more equal and hence can be seen as improving bled as a percent of GDP across the 1990s. social welfare. But beyond this income equalization angle, mobility is seen as reflecting the equalization of opportunities, Snapshots vs. movies: life-cycle welfare, mobility, a conception that links to Sen's concern with capabilities for and risk individual progress and to Roemer's (1998) concern with As the literature has also frequently noted, together, per the leveling of "circumstances" lying beyond the control of capita income and measures of distribution or poverty in a the individual but critically affecting the outcome of his or single moment in time offer an incomplete vision of her efforts. Benabou and Ok (2001) argue that these greater well-being. As economic theory suggests, and more opportunities engender a greater tolerance for inequality, in 31 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S some sense formalizing Hirschman's (1981) famous tunnel vicious circles, where individuals, communities, or even allegory where stalled motorists sit patiently watching the nations are found to be unable to escape poverty or a low next lane of traffic advance, only because they see that as a level of development because they lack human, physical, or sign that sooner or later they too will move. Even earlier, social assets.14 This topic is taken up at length beginning Friedman (1962) argued that a lack of mobility in the in chapter 6 and is only sketched out here. United States was probably a greater cause for concern than A large literature (see Fields and Ok 1996 for a review of was adverse distribution. These considerations of equality of some) has studied indexes of mobility and, increasingly, opportunity underlie the 2006 World Development Report: general patterns of income dynamics including poverty Equity and Development. traps (box 2.4). The need to gather long-term panel data The possible structural absence of mobility also lies has meant that studying mobility is a reasonably new behind the now-established literature on poverty traps or endeavor for Latin America. As an example, Fields and BOX 2.4 Mobility and poverty traps Two possible dynamics can lead to poverty traps, as sug- Myriad varieties of poverty traps have been discussed gested by the figure, taken loosely from Lokshin and in the literature. The efficiency wage hypothesis of Ravallion (2004). In the left panel, there are increasing Mirrlees (1975) and Stiglitz (1976) stresses that below a returns to scale up to Yu and decreasing returns to scale certain level of consumption, individuals are too under- thereafter. Households below Yu earn less and less, pro- nourished to work and hence find themselves further pelled toward zero while households above Yu are pushed malnourished. Lokshin and Ravallion (2004) also postu- away from it toward Ys. Yu is therefore an unstable equi- late that a minimum level of expenditure may be needed librium, and households below it or falling below it are to participate in society, for instance, getting a job, hav- stuck in a poverty trap. Lumpy investment opportunities ing a fixed address, or having adequate clothing. They also pose a trap, as shown in the right panel. For a house- argue that consuming below this point creates "social hold earning Y1, any change that raises income will pro- exclusion." Mehlum, Moene, and Torvik (2005) posit the pel the household toward higher levels of income, and existence of a poverty trap based on violence. any negative shock could push the family below, into a poverty trap.a a. Paraphrased from Antman and McKenzie (2005), written for this report. Poverty traps a. Caused by increasing returns to scale b. Caused by lumpy investment requirement Yt 1 Yt 1 Yt Yt 1 f(Yt) 0 Yu Ys Yt 0 Y1 Y2 Yt 32 D I M E N S I O N S O F W E L L - B E I N G , C H A N N E L S T O G R O W T H others (2005), looking at panel data for Argentina, Mexico, unpredictable and risky, and with the exception of gam- and República Bolivariana de Venezuela, examine changes blers, people tend to dislike risk. Generally speaking, peo- in individual earnings during positive and negative growth ple would rather take a smaller income with certainty than periods. They find limited evidence in Mexico and none a larger average one where they might receive much more in the other countries for what they term "divergent or might earn less and fall into poverty. mobility"--that those starting in the best economic posi- Risk has moved center stage in discussions of welfare tion to begin with experience the largest earnings gains or and poverty. The importance of risk to welfare was a central smallest losses; this finding would suggest overall conver- argument of Rodrik's (1997) discussion about whether gence and perhaps little evidence of poverty traps. How- globalization had gone too far; and concerns about the high ever, a problem plaguing the use of these data is their economic volatility of Latin America and the Caribbean design as short-term labor market surveys spanning no and the means to reduce it and mitigate its effects were the more than two years (Argentina) rather than the longer subject of the 2000 World Bank Latin American regional term. This means that they disproportionately capture flagship Securing Our Future in a Global Economy (de Ferranti measurement error or short-term movements in incomes.15 and others 2000). The World Development Report: Attacking Lokshin and Ravallion (2004) examine income dynamics in Poverty (World Bank 2001b) specifically included "secu- Hungary and Russia using six-year and four-year panels rity," meaning low risk, as a central dimension of poverty. respectively and propose a simple way of identifying The expanding literature on "vulnerability" goes beyond poverty traps.16 They find no evidence of poverty traps the concern with a family's current position to the likeli- for these two countries, although Rodriguez-Mesa and hood (risk) that they may find themselves in a worse posi- Gonzalez-Vega (2004), using a similar methodology, find tion, perhaps falling into poverty.17 some evidence for poverty traps in El Salvador. Risk also can affect measures of inequality (box 2.5). Numerous authors have recently explored techniques for First, income distribution measures are contaminated by extracting longer-term movements from short series such risk: one cannot tell if the Gini is showing the distribution as the ones in Latin America (see Glewwe 2004; Luttmer of differing incomes that are constant across time, or, at the 2002; and Krebs, Krishna, and Maloney 2004). One other extreme, whether everyone, on average, earns the approach proposed by Antman and McKenzie (2005) for same income over time but with those incomes varying this report was to create pseudo panels that effectively aver- greatly around that average. Either way, a cross-section age out transitory shocks across an entire cohort. These shows that inequality and higher measured inequality cohorts are then tracked over repeated cross-sectional sur- could reflect either an increase in true inequality or veys where the average of the cohort approximates a type of increased volatility: for example, the increase in inequality individual moving across time (see Deaton 1985 for a com- in the United States over the last decades is evenly divided plete discussion). Comparing the raw transitions to the between real increased inequality and increased volatility. pseudo panels, they find that correcting for measurement Kuznets may have been the first to link measures of error significantly reduces measured mobility, but in nei- inequality with risk when he asked if the apparently ther case do they find substantial evidence of poverty traps. declining inequality in the advanced countries might not The issue of mobility and poverty traps recurs through- result in part from workers moving into jobs with fewer out the chapters of this report--first in the mobility of "transient disturbances."18 A related issue, as Deaton and countries in the international distribution (chapter 6), then Paxton (1994) note, is that the observed cross-sectional of regions within countries (chapter 7), and finally of fami- measures of inequality are in fact combinations of the dis- lies and individuals (chapters 8 and 9). tributions of successive age cohorts, which, given that ran- dom life events cause incomes to diverge, should show Risk increasing dispersion with age. That is the case in Costa Although on the surface, mobility would seem to be good, Rica, as box 2.6 shows. whether it is in fact good or not depends to an important degree on the predictability of the movements. If an Relating mobility and risk income reversal occurs randomly, it would still mitigate That mobility and risk are, to an important degree, two sides life-cycle inequality, but it also makes incomes more of the same coin was recognized by Hart (1981, 11), who 33 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S BOX 2.5 Is it inequality or risk? Maybe Latin America has less inequality than we thought . . . There is a long-established concern that inequality mea- smoothable, have the same impact on measured inequal- sures are not measuring true inequality in lifetime ity as those shocks arising from true ability or opportu- incomes or opportunities and that instead they may nities captured by the first term, which is, in fact, the largely be picking up short-term fluctuations in term we care most about. As Solon (2002) shows, how- income. For example, consider the following equation ever, if one were measuring the distribution of true dis- log yibt = ib + (t - b) + ibt, where y captures the real counted lifetime earnings, the transitory variations annual earning in year t of individual i born in year b; would nearly completely vanish. Hence, measured cross- captures more or less permanent characteristics of the sectional inequality of current incomes is distorted by individual such as intelligence, motivation, and inter- almost the entire value of the transitory component personal skills; is the growth rate of wages across the (Lillard 1977; Shorrocks 1981). As Krebs, Krishna, and life cycle after reflecting, for instance, the accumulation Maloney (2004) show, the transitory component of vari- of experience; and represents transitory deviations of ance across time using panels is roughly two-thirds of measured earnings from the life-cycle earnings trajec- the total variance, suggesting that these distortions can tory including both short-term fluctuations and measure- be large. ment error. If one assumes that the three components of These distortions can be important. The table shows income are independent and that transitory shocks are that measured inequality among the self-employed in uncorrelated across time, then the observed variance of various Latin American countries is roughly double that incomes in the sample can be expressed as Var(log yibt) = of salaried workers, much of it attributable to the intrin- 2 + b + 2.2 2 sically higher risk of the sector. Since the share of self- From this one sees that if earnings inequality is employment decreases with level of development, the measured for the entire labor force, part of that inequal- number of self-employed may be of some importance ity simply arises from the second terms and reflects the (Maloney 2000). Were Bolivia to have U.S. levels of self- intercohort variation in stage of the life cycle at any employment, that is, 10 percent instead of 56 percent, year t. As Paglin (1975), and implicitly Kuznets the level of inequality as measured by the Theil index for (1955), note, this variation need not imply inequality in all workers would fall almost 30 percent. any meaningful sense. Across the life cycle, all are equal. Second, transitory shocks, while important if not Source: This discussion draws heavily on Solon (2002). Earning inequality decomposition for salaried and self-employed workers Argentina Bolivia Chile Colombia Uruguay Venezuela, R.B. de Self-employed share (%) 26 56 29 33 26 37 Theil Index: All workers 0.362 0.642 0.735 0.667 0.398 0.34 Self-employed 0.484 0.819 0.867 0.972 0.499 0.47 Salaried 0.295 0.43 0.411 0.433 0.35 0.264 Within and between group inequality, with groups defined by type of employment Within group 0.355 0.642 0.639 0.653 0.395 0.34 Between group 0.007 0.001 0.096 0.013 0.004 0 Source: Maloney and Wodon (1999). Analysis for all workers with incomes above zero in 1995. 34 D I M E N S I O N S O F W E L L - B E I N G , C H A N N E L S T O G R O W T H BOX 2.6 . . . Or maybe more: Inequality and demographics How do demographics affect measures of inequality? As Again, Kuznets (1955) foreshadowed this finding in Deaton and Paxton (1994) note, the observed cross- arguing that inequality comparisons should take a cross- sectional measures of inequality are in fact averages of the section of units at the prime earning phase of the life distributions in successive age cohorts, which, if the per- cycle and avoid the phases of youth or retirement. manent income hypothesis is correct, should show very Preliminary regressions of Ginis on measures of the different distributions of income and consumption. The age of population suggest that these effects are not small reason is that the accumulation of positive and negative in the aggregate. The right panel of the figure graphs the shocks to income as individuals age leads the incomes of cross-national partial correlation between the share of age cohorts to diverge. Deaton and Paxton demonstrate people below age 14 and the Gini, and its negative and that in Taiwan (China), the United Kingdom, and the statistically significant trend line. Were Latin America to United States, any changes in aggregate inequality are have aging Europe's demographic structure as a bench- many times smaller than the changes in age-cohort mark, its Ginis might be 4 percentage points higher; inequality. This appears to be the case in Costa Rica as Ginis in comparatively youthful Bolivia, Guatemala, well (see left panel of the figure). Thus, it is possible for Honduras, and Nicaragua could be up to 7 percentage substantial changes in the distribution of aggregate points higher. income to be driven purely by demographic changes. Inequality and age of population Standard deviation of incomes by age cohort, 2004, Costa Rica Inequality measures versus age of population, world Adjusted Gini 1.1 20 15 y 0.2687x 0.0331 1.0 10 0.9 5 0 0.8 5 0.7 10 0.6 15 20s 30s 40s 50s 60 12 10 8 6 4 2 0 2 4 6 8 10 Age cohort Adjusted share of young argued that "a society with zero correlation [in income levels between mobility and risk has emerged only recently (see across time] and very high mobility would be too unstable Gottshalk and Spolaore 2002). The complications involved for most people so there is an optimal level of correlation can be suggested by asking what happens if the unexpected somewhere between zero and one." The link is also implicit shocks to income occur symmetrically: that is, what happens in recent discussions of the new opportunities and increased if, on average, an individual experiencing an unexpected insecurity arising in economies transitioning to a more income shock has as much chance of moving up as down. In market-based economic system (Birdsall and Graham 1998). this case, there can be lots of apparent mobility, but on aver- However, a more rigorous discussion of the relationship age, and on expectation across the life cycle, everybody stays 35 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S in the same place. There is no narrowing of expected lifetime · Unpredicatable mobility (risk). Like mobility, risk also income differentials but only more risk, and society is neces- has potentially strong feedbacks to growth. As an sarily worse off. In this view, only the predictable elements of example, using cross-country data, Flug, Spilimbergo, mobility can positively affect welfare. and Wachtenheim (1998) find that income volatility Krebs, Krishna, and Maloney (2005a) offer one possible adversely affects educational attainment. As later way of calculating welfare that captures the various ele- chapters tell in greater detail, simulations suggest ments that are discussed above and dealt with in subse- that were Mexicans to face the same level of income quent chapters. They argue that the welfare measure of the risk as workers in the United States, they would distribution of expected lifetime consumption adjusted for increase their investment in human capital (health, risk needs to incorporate measures of: education, on-the-job training) by roughly 2.5 per- cent of GDP. Further, the poor appear to face more · Initial income position of the individual or group. If this income volatility than the middle class (Krebs, initial income were considered the permanent and Krishna, and Maloney 2005b, 2005c). unvarying status of an individual or group, then it would be more or less captured by traditional mea- Annex 2B offers a tractable method for combining all sures of poverty and inequality. Welfare can clearly be these elements in one measure of welfare, and the results for altered by transfers among these individuals or groups, Argentina and Mexico are presented in table 2.4. Although and the feasibility of engineering significant changes income distribution statistics are generally calculated using through this mechanism is addressed in chapter 5. data divided into quintiles or deciles, the need to estimate · Predictable mobility. These measures encompass pre- a measure of the permanent component of risk (the part dictable movements of individuals or groups from that cannot be easily smoothed) limits us to three education their initial income position both absolutely and rel- categories, with "primary" proxying broadly for the poor. ative to others. Perhaps the most discussed driver of The first line of table 2.4 tabulates the share of the pop- such mobility is the accumulation of human capital, ulation found in each education category. The second, which in turn is central to growth. Chapters 8 and 9 third, and fourth rows in the table capture the components show that investment in education for the poor yields of expected lifetime utility for each. The fifth calculates relatively low rates of return in Latin America and this level of utility (increasing as it becomes less negative), hence the poor do not make the push to complete sec- and the sixth combines the three different levels of utility ondary schooling. Failure to complete secondary into one measure of social welfare. Unsurprisingly, in both school typically prevents the poor from escaping the countries the poor show lower levels of welfare, and cycle of poverty. Argentina, with both higher levels of initial income and TABLE 2.4 Welfare comparisons: Argentina and Mexico Argentina Mexico education categories education categories Primary Secondary Tertiary Primary Secondary Tertiary Share in population () 0.352 0.405 0.243 0.606 0.207 0.187 Predictable income growth (µ) 0.010 0.017 0.026 0.009 0.012 0.023 Initial income level [c(i, 0)] 428 595 904 279 348 546 Income risk (2) 0.056 0.045 0.052 0.064 0.046 0.075 Utility -2.780 -1.966 -1.525 -3.871 -2.734 -2.544 Welfare -2.059594892 -3.301187884 Difference 0.389245076 Source: Krebs, Krishna, and Maloney (2005a). Note: Difference is measured in the equivalent difference in first period consumption. 36 D I M E N S I O N S O F W E L L - B E I N G , C H A N N E L S T O G R O W T H lower levels of risk (although slightly lower levels of erations is a central issue, both for understanding welfare growth), shows a higher level of total social welfare. and for growth. Looking at predictable income growth of the primary- educated group relative to the other subgroups suggests Further, if one may add a final touch to what is that in neither country are the poor catching up; there is beginning to look like a statistical economist's pipe little predictable upward mobility of this class in the dis- dream, we should be able to trace secular income lev- tribution. It is straightforward to calculate (not shown) els not only through a single generation but at least that were the poor to share the same rate of growth as the through two--connecting the incomes of a given rich, perhaps from an increased investment in education or generation with those of its immediate descen- a higher return to schooling for the poor, the poor in both dants. . . . If living members of society--as produc- countries would gain 32 percent in utility measured in ers, consumers, savers, decision-makers on secular initial consumption, and society as a whole would gain problems--react to long-term changes in income 13 percent in Mexico and 21 percent in Argentina. To levels and shares, data on such an income structure determine relative mobility, one could ask what would are essential.19 happen if the growth rate of the poor were raised 1 percent at the expense of the growth rate of the two other groups so The last decade has generated substantial new research that overall growth were unchanged. Making growth more on measuring intergenerational mobility for Latin America pro-poor in this way would increase total welfare by 1.6 per- and the Caribbean and, to a lesser degree, identifying its cent in Argentina and 9 percent in Mexico. correlates and causes. Again, the question is whether peo- Changes in risk also yield large, although opposite, ple can move out of poverty or whether there may be inter- changes in overall welfare. Mexico appears to have a higher generational poverty traps where the poor, or some level of income risk for every income group than does particular groups of poor, simply replicate their parents' Argentina, and its aggregate risk measure is 0.073 com- status ad infinitum. pared with 0.048 for Argentina and 0.023 for the United The most common strategy for measuring the degree of States (see Krebs, Krishna, and Maloney 2005a for intergenerational mobility is similar to that for intragener- Argentina and Mexico; Meghir and Pistaferri 2004 for the ational mobility: studying the correlation of a generation's United States). Were Mexico to lower its aggregate risk to well-being with that of its progeny, generally measured as Argentine levels, it would improve its aggregate welfare in the elasticity of children's earnings or education level rela- an amount equal to an increase in the income growth rate tive to that of their fathers.20 This elasticity is expected to of roughly 0.6 percent or a 15 percent rise in average con- increase with the strength of intrinsic qualities such as sumption levels. In both countries, the poor are addition- genetics or social connectedness of families and decrease ally hit because they have higher risk than the middle class. with the progressivity of government investment in chil- If the poor had the same risk levels as the middle class, the dren's human capital that would allow children to over- utility gain for the poor in Mexico would be equivalent to come their families' position in the social structure.21 an increase of 0.7 percent in the income growth rate and 19 Numerous studies have postulated, for example, that the percent in consumption; for Argentina the figures are 1.3 lower elasticities in Canada and Sweden arise from their percent for income growth and close to 30 percent for greater efforts in public education.22 Conceptually, it is not consumption. While these calculations suggest that mea- hard to integrate credit constraints as barriers to accumu- sures of poverty and welfare would indeed change greatly lating the desired level of children's education and the by introducing a measure of risk, they are in the realm of expected volatility of the children's income as being impor- those calculated in the mainstream literature for the tant to these investment decisions. United States. Comparisons across countries are difficult because of dif- ferences in methodology, data sets, and units of compari- Intergenerational mobility son, but a fairly consistent picture is emerging. Grawe The welfare measure captures the distribution of individual (2002) attempts a very consistent classification of elasticities welfare across his or her life span. But again, the omniscient for a sample that includes two Latin American countries Kuznets (1955, 2) argued that, in fact, mobility across gen- (figure 2.5). The United Kingdom and the United States, 37 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S of sibling educational attainment: if parental characteris- FIGURE 2.5 tics have no impact, there should be no correlation, and if Elasticity of son's income relative to father's income determinant, then children should have identical attain- 1.2 ment. In some cases, the rankings do shift importantly. Mexico goes from high mobility to relatively poor mobil- 1.0 ity, El Salvador from mid-level to bottom; Argentina from 0.8 top to middle; Costa Rica from middle to top. Despite this shifting around, a general pattern emerges: Latin America 0.6 is consistently less mobile than the United States and, 0.4 therefore, most of the advanced countries. And within the region, Chile, Paraguay, and Uruguay show relatively high 0.2 mobility; Brazil, Guatemala, and Nicaragua are generally very low. 24 0 ia s As is always the case in measuring mobility, such simple Peru State stan any Nepal Ecuador Kingdom Malays Paki CanadaGerm indicators also hide important information, in particular United about differing patterns across units. For this reason, in United looking at mobility of countries, subnational units, and Source: Grawe (2002). individuals, it is common to report transition matrices showing transitions among a limited number of categories with values between 0.5 and 0.6, show little intergenera- or kernel density plots, using continuous variables as their tional mobility relative to Canada and Germany, but Peru analogue. The transition matrix for Colombia, given in at 0.67 is substantially worse and Ecuador at slightly above table 2.5, shows, for example, that the probability that a 1.0 winds up being the country with the least mobility. child of parents with primary education (generally the Although studies conflict, the literature seems to be con- poor) will obtain tertiary education is 10.5 percent; the verging on the United States as being among the least probability of that child even finishing secondary school is mobile advanced countries, and it is this reference point only 14 percent. Only 61 percent of those children whose that the available comprehensive studies of Latin American parents had some secondary education completed secondary and Caribbean countries benchmark against (see figure 2.6 school. These findings are suggestive of a low-education and annex 2C).23 poverty trap that perpetuates a family's poverty across time. In general, the focus of studies on specific Latin American Constructing earnings matrices for Brazil, Guimarães and countries has been on education because of both the greater Veloso (2003) find sharp differences by regions, races, and reliability of the measure and the apparent consensus, con- cohorts, and in all cases, mobility is lower for sons of low- sistent with the framework above, that educaton is the wage fathers than for sons of middle-wage fathers. critical driver of intergenerational mobility. Behrman, Birdsall, and Székely (1999) tabulate the correlation TABLE 2.5 between parents' and children's schooling and find that Intergenerational transition matrix for Colombia, 1997 Brazil, Colombia, Mexico, and Peru all do worse than the United States, with a coefficient above .4, as is common in Education of children the literature. The finding holds both in urban areas and Education of Primary Some overall, with correlation coefficients for Brazil and Colombia parents or less secondary Secondary Some higher above 0.6. Andersen (2001) calculates a social mobility index that uses a measure of the schooling gap--what is Primary or less 51.2 24.2 14.1 10.5 Some secondary 12.6 26.2 25.4 35.9 attained versus what is expected for an individual of a cer- Secondary 9.1 17.3 25.4 48.2 tain age--and finds a similar ranking, with the exception Higher education 2.2 6.5 14.2 77.1 of Peru, whose ranking improves somewhat. Behrman, Total 41.7 23.2 16.2 18.8 Gaviria, and Székely (2001) and Dahan and Gaviria (1999) use another measure of parental influence--the correlation Source: Behrman, Gaviria, and Székely (2001). 38 D I M E N S I O N S O F W E L L - B E I N G , C H A N N E L S T O G R O W T H FIGURE 2.6 Mobility Indicators a. Correlation of schooling between parents and children b. Social mobility index based on teenagers (13­19 years old) 1. Chile 2. Argentina* United States 3. Uruguay* 4. Peru 5. Mexico Mexico 6. Paraguay 7. Panama 8. Venezuela, R.B. de 9. Dominican Rep. Peru 10. El Salvador 11. Honduras 12. Colombia Colombia 13. Costa Rica 14. Nicaragua 15. Ecuador 16. Bolivia Brazil 17. Brazil 18. Guatemala 0 0.2 0.4 0.6 0.8 1 0.70 0.75 0.80 0.85 0.90 0.95 All Urban Social mobility index for teenagers (point estimate and 95% confidence interval) *Based on urban samples only. Source: Behrman, Gaviria, and Székely (2001). Source: Andersen (2001). c. Intergenerational school mobility in Latin America and in d. Social mobility in the Americas the United States United States 1998 United States Paraguay 1998 Costa Rica Panama 1999 Peru Uruguay 1998 Uruguay* Jamaica 1998 Chile 1998 Paraguay R.B. de Venezuela 1999 Chile Dominican Rep. 1998 Argentina* Peru 2000 R.B. de Venezuela Honduras 1999 Dominican Rep. Colombia 1999 Panama Costa Rica 1998 Bolivia 1999 Brazil Argentina 1998 Bolivia Mexico 1998 Nicaragua Ecuador 1998 Ecuador Brazil 1999 Colombia Guatemala 1998 Nicaragua 1998 Mexico El Salvador 1998 El Salvador 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 *Based on urban samples only. Source: Behrman, Gaviria, and Székely (2001). Source: Dahan and Gaviria (1999). 39 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S Theory predicts that borrowing constraints, discrimina- can and does lead to very different conclusions about the tion, spatial segregation, and marital sorting--all typically evolution of welfare in a region and about the relationship mechanisms of exclusion--are among the principal factors of poverty and inequality to growth. that inhibit mobility. Although the thin empirical literature So far, the data remain limited for generating compre- broadly supports this hypothesis, most studies also suggest hensive indicators of well-being that are comparable across that greater educational expenditures improve mobility. countries in Latin America. The good news is that progress Behrman, Birdsall, and Székely (1999) argue that for a typi- is being made in the region on these fronts. Looking even cal country doubling the share of public expenditures on at simple static measures, better techniques for deflating education as a share of GDP would increase mobility by poverty and distribution series are available, and the litera- 25 percent. They also find that higher spending per school- ture on multidimensional and subjective poverty measures age child on primary education and better quality primary is ballooning. Since Kuznets wrote in 1955, the macroeco- and secondary schooling are positively associated with inter- nomics literature has erected elegant architecture for ana- generational mobility, while relatively greater public spend- lyzing income dynamics and thinking through life-cycle ing on tertiary education may actually reinforce the impact welfare issues. The increased availability of panel data in of family background and reduce intergenerational mobility. recent years and the development of techniques for elimi- Consistent with these findings, Andrade and others (2003) nating measurement error and transitory income fluctua- find evidence that credit constraints increase the persistence tions have made feasible serious, if still limited, mappings of immobility found among poor groups. At the aggregate of mobility, testing for poverty traps, and calculations of level, results offer less clarity. Andersen (2001) finds a posi- the variance measures necessary for dynamic welfare mea- tive correlation between his measure of social mobility and sures. Numerous papers have sought to evaluate the magni- urbanization and level of development (GDP) and none with tudes and determinants of intergenerational mobility. From measured inequality. Behrman, Gaviria, and Székely (2001) these efforts, several findings appear. find no correlation with GDP or trade openness, leaving the question about whether mobility and economic growth are · Measurements that use the correct deflators show that related somewhat up in the air. Behrman, Birdsall, and for the majority of episodes studied, Latin America Székely (1999) find that macroeconomic conditions--in and the Caribbean have reduced poverty and inequal- particular those related to the extent of internal market ity more than conventional indicators suggest. development--significantly shape intergenerational mobil- · Health, longevity, and other indicators of welfare ity by loosening the strong link between parents' back- have improved much more than the incomes of the ground and children's education. poor would suggest. Some countries saw substantial As with the intragenerational mobility discussed in the improvements in welfare despite stagnation in previous section, the message is that measures to encourage incomes. human capital accumulation--certainly in education and · At the same time, mobility, measured as the ability in all likelihood across several dimensions--are critical to to move out of poverty across generations, seems redressing poverty and improving social welfare in a much lower and income risk much higher than they dynamic context, as are measures to reduce impediments to are in advanced countries, suggesting that in relative accumulation of human capital, such as risk and liquidity welfare terms, Latin America and the Caribbean are constraints. doing substantially worse than standard poverty indicators may suggest. Conclusion This chapter has elaborated on Kuznets's "economic statis- A stronger data effort across the region in all these dimen- ticians pipe dream," reaffirming his now 50-year-old doubts sions will further enrich our picture of poverty in the region. about how well the common measures of poverty and A broader conception of poverty also enriches the dis- inequality really capture welfare and extending the laundry cussion surrounding pro-poor growth and, in turn, what list of considerations that need to go into a comprehensive might be called pro-growth poverty reduction. At the most welfare measure. We have shown that these considerations elementary level, correctly deflating welfare statistics is, in are not merely conceptual curiosities--incorporating them principle, essential for understanding their links to growth 40 D I M E N S I O N S O F W E L L - B E I N G , C H A N N E L S T O G R O W T H and policy reforms that, by their design, alter relative where y(t) and c(t) are the income and consumption at t, r is prices. More profoundly, an expanded concept of poverty the interest rate, and is the subjective discount factor. also forces policy makers to take a broader look at the chan- Consider a given individual at two points in time ( denotes nels running in each direction. Progress in health, security, the second period). The inframarginal income W(T, T) education, and risk reduction is correlated with income that would give this person the same utility level observed growth, but not so tightly as to obviate the need for impor- in the second period but with the life expectancy observed tant antipoverty efforts independent of those promoting in the first is defined by V(Y + W(T, T), T) = V(Y, T). income growth per se. Consider a hypothetical life-cycle individual who receives These dimensions of poverty form the reverse channel of the municipality's income per capita in all years of life and a virtuous circle, as chapter 6 shows, and thus affect income lives to the age corresponding to the municipality's life growth. Education and, to a lesser degree, health make reg- expectancy at birth. Assume that = r, so that optimal ular appearances in the ubiquitous growth regressions, consumption is constant and equal to the constant income while labor market risk affects the accumulation of human flow [c(t) = c = y]. In this case, the indirect utility function capital and hence offers a separate channel to growth. Peo- can be expressed in terms of the yearly income y as in: ple's prospects for mobility and for the advancement of V(y, S) = u(y)A(T), where A(T) = (1 - e-rT)/r. Define their children also offer incentives to accumulate human w(T, T) as the yearly income. Therefore, w satisfies u[y + capital. From a growth point of view, poverty reduction in w(T, T)]A(T) = u(y)A(T). these dimensions is good business. The monetary value of the total gains in welfare To some degree, however, we can only sketch a longer- observed in the period, when measured by yearly income, term research agenda. In the short run, global databases of can be denoted as (y - y) + w. The lifetime value of these poverty and inequality statistics are not ideally deflated, changes is the present discounted value of this annual flow. multidimensional analysis is available for only a few coun- The contribution of health to the total gain in welfare is the tries, calculation of income risk is data-intensive, and panel fraction w/[(y - y) + w]. Inverting the instantaneous utility data coverage is similarly extremely limited. Yet subjective function u(.), w turns out to be poverty indicators suggest that income--even when the u(y)A(S) data are incomplete--is not a poor proxy for well-being, (A2.2) w = u-1 A(S) - y(*). meaning that many pending questions in pro-poor growth and antipoverty policy can be fruitfully approached with Two dimensions of u(.) affect the willingness to pay for the data on hand. The next three chapters do this, largely at extensions in life expectancy: the substitutability of con- the macroeconomic and regional levels. sumption in different periods of life (that is, the intertem- poral elasticity of substitution), and the value of being alive Annex 2A relative to being dead. To capture both, a particular defini- tion of u(c) is calibrated, u(c) = c1-1/ 1 - 1/ + , where deter- Estimating the monetary value of mortality mines the level of annual consumption at which the changes individual would be indifferent between being alive or dead, Becker, Philipson, and Soares (2005) convert life span into arising from the normalization of the utility of death to zero. monetary values to calculate a measure of total welfare gain If the intertemporal elasticity of substitution is larger than by calculating how much people would pay for an addi- 1, then is negative. With expression (*) and this functional tional year of life: form, a closed form solution for w is obtained.25 Assume the existence of a perfect capital market and consider the indirect utility function V(Y, T) of an individ- Annex 2B ual living in a municipality with life expectancy T and life- time income Y: A tractable welfare measure that captures income, mobility, and risk T (A2.1) V(Y, T) = max e-tu(c(t)) dt subject to Krebs, Krishna, and Maloney (2005a) assume that incomes {c(t)} 0 T T evolve over time according to log yit = t + txit + uit. Y = e-rt y(t) dt = e-rt c(t) dt, Income is driven by time-changing shifts in levels, , and 0 0 41 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S returns to human capital (x), . The parameter u captures mobility index, which is defined as SMI = 1 - factor individual income changes that are caused by changes in inequality weights of the family background variables of the observable worker characteristics. In turn, u is composed of following specification: Schooling Gap = + 1Max(Sf,, Sm) + a permanent shock, , that follows a random walk, and a 2Yh + iCONi + e, where the gap is the disparity between transitory component that captures both temporary actual years of education and the potential; Sf,, Sm repre- income shocks and measurement error. It is straightforward sents the schooling variable for father and mother, respec- to show that the greater the variance of the permanent tively; Yh measures the household income and CONi are shocks to income, 2, the lower the covariance of the control variables; Max(Sf,, Sm) and Yh are the family back- unpredictable component of incomes, that is, the greater ground variables, and the factor inequality weight is the the unpredictable component of mobility. This component product of the coefficient estimate for each variable, the of mobility is pure risk and hence negatively affects wel- standard deviation for the same variable, and the correla- fare. Krebs (2002) shows that given a one-period utility tion between the same variable and the dependent one. function given by u(c) = c1- 1 - , 1, the expected lifetime These factors are necessary inputs to perform the Fields utility of an individual or subgroup facing the above variance decomposition. income process is Panels C and D report results based on sibling correlations: c1i- F Bf (g­f - g­)2 (B2.1) Ui = , (1 - )(1 - (1 + µ)1 exp(.5( - 1)2)) - g = f =1 Bg­(1 - g­) , where g­f is the average value of gsf in family where ci is initial consumption levels; µ is the predictable f, Bf is the number of teenage siblings in family f, g­ s is the average value of g in the entire sample, B is the number of part of income growth, perhaps arising from accumulated human capital; is the coefficient of risk aversion; and is individuals, and F is the number of families. This index corresponds to the R2 obtained by regressing gsf (defined as the discount factor. A Generalized Methods of Moments a dummy variable capturing whether individual s of family (GMM) technique is used to separate permanent from tem- porary shocks.26 f has more years of schooling than the median individual of his or her cohort), on a set of dummy variables for all fami- To capture the fact that societies dislike inequality and lies in the sample. Since g could yield positive values even hence weight the utility of the poor more than those of the if family background is inconsequential, as is the case, for rich, the individual expected utilities are combined into an instance, when children are assigned to families randomly, a overall welfare function: 1 modified version of the previous index is used: a = 1 - (1 - W = U1j - 1- (B2.2) j , g)B B - 1 - F (the index a yields positive values only if the pre- j vious index g is greater than would be expected purely by where is the share of the subgroup in the total popula- chance). Differences among results on both panels (C and tion, and is the social aversion to inequality. For the dis- D) emerge more from the more recent data used by cussion in the text, = = 1.5 and = 0.95. Behrman, Gaviria, and Székely (2001) than from the mea- sures per se. Annex 2C Notes Intergenerational mobility in Latin America: 1. Poverty and inequality analysis has, for the most part, focused on capturing changes in income or consumption measured as a basket Country comparison of goods. The poverty line itself is generally defined in terms of a bas- Two sets of rankings comparing intergenerational mobility ket of goods satisfying minimum caloric intake requirements. This different from those proposed by Solon (2002) are reported definition, as Thorbecke (2005) highlights, is in itself not trivial, as in figure 2.6. Panel A shows the correlation of schooling it immediately raises the problem of what should be in that basket: between parents and children captured by in the follow- should that same common basket be used across all countries and ing first-order Markov model: Sit = + Sit + wt, where S subnational regions, as suggested by Ravallion and Bidani (1994) -1 and Ravallion (1998), or should the basket be tailored to each coun- is schooling, i indexes each family, t is the generation of the try's tastes, preferences, and relative prices. sons, t - 1 is the generation of the parents, and w is a sto- 2. Unless otherwise noted, the poverty figures refer to the head- chastic term. Panel B shows Andersen's (2001) social count index and a poverty line set at $2 per capita purchasing power 42 D I M E N S I O N S O F W E L L - B E I N G , C H A N N E L S T O G R O W T H parity. The poverty figures reviewed in this chapter come from a 12. The Economist estimates that the region pays a cost of 13­15 background paper for this report by Gasparini, Gutierrez, and percent in security; see "The Backlash in Latin America: Gestures Tornarolli (2005). The calculations are based on the results of pro- against Reform," Economist, Nov. 30, 1996, p. 19. cessing 57 household surveys for 18 Latin American countries (which 13. In fact, if capital markets were perfect, then individuals could represent around 92 percent of the region's population) covering the perfectly smooth consumption across their lives, and consumption 1990s and early 2000s. might be completely equalized across individuals at any period in time. 3. Population-weighted averages are more useful to assess 14. See, for example, Rosenstein-Rodin (1943), Nurkse (1953), poverty rates when the region is treated as a single entity and hence Nelson (1956), and Basu (1997). Our thanks to Gary Fields for when individuals in different countries are given the same relevance. pointing out these references. To a large extent, population-weighted average poverty rates are dri- 15. On the first point, Lokshin and Ravallion (2004) caution that ven by the poverty rates of the most populated countries. For exam- measurement error is likely to cause spurious negative correlation ple, Brazil's weight would be about 0.35 whereas Jamaica's would be between income changes and initial income levels. On the second only 0.005. Unweighted averages, in contrast, are more useful to point, short-term variation--for instance, the variation that arises assess poverty when interest centers on countries rather than individ- from universally volatile self-employment--is not very interesting uals (that is, when the country is the unit of analysis). Proportion- from a life-cycle point of view while it is hard to identify whether ately, poor individuals living in smaller countries are given more households really do bounce back from shocks given the likely longer relevance in this second measure. duration of the recovery process. See, for instance, Fajnzylber, 4. See Egset and Sletten (2004) for the former, and World Devel- Maloney, and Montes (2005) and Bosch and Maloney (2005) on opment Indicators (2005f ) for the latter. short-term variation. 5. In fact, between the early 1990s and the early 2000s, the 16. They estimate the degree to which the relationship between change was a mere 0.2 percentage point, as a consequence of the income today and yesterday involves a cubic function in income, the regional slowdown after the Russian crisis. The evolution of headcount empirical structure that would generate a pattern such as seen in the poverty based on a $1 a day poverty line would show an even lower figure in box 2.4. decline, from 11.2 percent in the early 1990s to 10.8 percent now. 17. See Ligon and Schecter (2002) and Gamanou and Morduch 6. These results are reversed for Central America and the South- (2002) for a review of the literature. For applications to specific coun- ern Cone area when looking at the unweighted means, which sug- tries, see Maloney, Cunningham, and Bosch (2004) for Mexico; gests that poverty would have dramatically declined in Central Glewwe and Hall (1998) for Peru; and Contreras, Cooper, and America (by 6 percentage points) and remained basically constant Heman (2004) for Chile. The disconnect between discussions of risk (-1 percentage point) in the Southern Cone area. To a large extent, and mobility is exemplified by the fact that the Maloney, Cunningham, this is just a reflection of Brazilian trends (the most populated coun- and Bosch paper uses the same Mexican panels for studying income try of the region), where poverty declined significantly, and Mexican shocks as Fields and others (2005) use to study mobility, yet neither trends (the largest country in the Central America region), where work mentions the other concept. poverty remained unchanged. 18. "Do the distributions by annual incomes properly reflect 7. Figure 2.2 presents estimates of the (unweighted average) trends in distribution by secular incomes? As technology and eco- regional poverty rate in the mid-1990s, together with those already nomic performance rise to higher levels, incomes are less subject to discussed above for the early 1990s and early 2000s. The period from transient disturbances. If in the earlier years the economic fortunes of the early 1990s to the mid-1990s corresponds to an economic expan- units were subject to greater vicissitudes--poor crops for some farm- sion, whereas the period from the mid-1990s to the early 2000s rep- ers, natural calamity losses for some nonfarm business units--if the resented a mix of expansion and recession. It must be noted that the overall proportion of individual entrepreneurs whose incomes were different country coverage of the samples raises some comparability subject to such calamities was larger in earlier decades, these earlier issues between the different periods. distributions of income would be more affected by transient distur- 8. Generally, as Sen (1972) shows, it is hard to squeeze many bances." Kuznets (1955, 6) dimensions of social well-being such as freedom or the ability to get 19. Kuznets continues: "An economic society can then be judged a job into conventional social welfare function analysis. by the secular level of the income share that it provides for a given 9. See Thorbecke (2005) for a discussion of these issues. generation and for its children. The important corollary is that the 10. These were, in particular, the relative weights on each measure study of long-term changes in the income distribution must distin- of poverty and the substitution assumed between them. guish between changes in the shares of resident groups--resident 11. Several excellent papers covering the topic were included in a within either one or two generation--and changes in the income recent conference sponsored by the U.K. Department for Interna- shares of groups that, judged by their secular level, migrate upward tional Development, Instituto de Pesquisa Economica Aplicada or downward. . . ." (IPEA) in Brazil, the International Poverty Center, and the United 20. This is generally taken as the coefficient in an OLS (ordinary Nations Development Programme. See Anderson, Crawford, and least square) regression of a log linear regression of a son's earning (or Liecester (2005); Deutsch and Silber (2005); Duclos, Sahn, and education) on a father's earning, with age controls for both genera- Younger (2005); and Thorbecke (2005). tions. Solon (2004), extending the canonical framework by Becker 43 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S and Tomes (1979), argues that such a specification can be theoreti- make careful transitive comparisons, concludes that the United States cally motivated in a framework that shares a close kinship with the and the United Kingdom are substantially less mobile than, say, standard permanent income hypothesis used for analyzing intragen- Canada, Finland, and Sweden. erational mobility. Parents are assumed to divide their income 24. Brazil's low mobility is confirmed by Dunn (2003), Ferreira between investing in their children and their own consumption, and Veloso (2003), and Bourguignon, Ferreira, and Menendez maximizing welfare across generations so that there are increases (2003). both in today's consumption and in children's income. Children 25. The set of parameters (, , r) needed to compute w can be cal- effectively receive endowments that are determined by genetics, the ibrated from other parameters more commonly estimated in the reputation and connectedness of their families, correlates of race, "value of life" and consumption literatures. More precisely: = 1 values placed on learning and the like, which are then augmented c1 -1/ 1 u(c)c - 1-1/ , where = is the elasticity of the instantaneous u(c) by educational expenditure. utility function. In particular, U.S. parameters are employed as the 21. Roemer (2005) argues that "equality of opportunity" in some ones for Brazil are not available. Murphy and Topel (2003, 23) esti- circumstances does not necessarily imply zero correlation across mate that = 0.346, and Browning, Hansen, and Heckman (1999, generations--innate abilities and inherited values imply correlated 614) suggest that is slightly above unity. Using = 1.250, = outcomes. 0.346, and c = $26,365, the value of is calculated to equal -16.2. 22. This approach also offers insights into cross-sectional inequal- (The value of consumption is the value of U.S. per capita income in ity. The variance of log earnings depends not only on the same fac- 1990 in the Penn World Tables 6.1 data set, matching the year in tors, with the same sign as mobility, but also on the variance of the which Murphy and Topel 2003 estimate .) innovations to the process of inheritability of endowments. Hence, 26. Numerous authors (Glewwe 2004; Luttmer 2002) have two countries with the same intergenerational elasticity might differ stressed the need to deal with the problem of separating income risk in inequality if they had differing degrees of heterogeneity of ability from measurement error, that is, the need to extract the correct com- or endowments. ponent of risk from the sample. Krebs, Krishna, and Maloney (2004) 23. Checchi and Dardanoni (2002), using a wide variety of have discussed the problems of extracting the correct measure of risk indexes on both job quality and education for many OECD countries from the noisy panel data that are available. We are less interested in and a few developing countries, consistently found the United States the transitory shocks, which even relatively poor households can and the United Kingdom to be the most mobile, and Brazil the least smooth over, than in permanent shocks, which the poor cannot mobile. However, Solon (2002), using other studies in an attempt to smooth out. 44 CHAPTER 3 How Did We Get Here? The existing differences in development between Latin America and the advanced economies of the world did not appear overnight. In fact, they are likely the result of historical processes that in some cases trace back to the colonial period. That opens the door to several interesting questions: How much has the region grown economically since its independence from colonial rule? How much did Latin America lag behind the more advanced economies in the 19th century? Has that gap widened steadily over time? How has inequality in Latin America evolved historically and how has it evolved elsewhere in the world? Is today's high inequality a permanent feature of modern Latin America? In short, how did we get here? M OST OF THE COUNTRIES IN THE Differences in income distribution are also dramatic. Lev- Latin American region are middle- els of inequality in the region are well above those of the income countries, and some of the developed countries. As noted in the World Bank's Latin richer ones have per capita income American Region 2004 flagship, Inequality in Latin America levels that are close to those of the and the Caribbean: Breaking with History? (de Ferranti and poorer industrial countries and were even higher in the others 2004), the Gini coefficient for the region is about past. For example, in 2003 Argentina's per capita GDP 0.55, compared with 0.37 for the developed countries, and is was about two-thirds of Portugal's, but in 1930 Argentina the highest in the world together with that for Sub-Saharan boasted the seventh largest economy in the world, with Africa.1 The negative impact that this higher inequality has per capita income higher than that in Canada or France, on the observed income poverty levels is significant: if Latin and nearly as high as that in the United States. Yet the America had the level of inequality of the developed world, region as a whole still has a long way to go before achiev- its income poverty levels would be closer to 5 percent than to ing the living standards of the advanced economies. Today the actual rate of 25 percent estimated in chapter 2.2 the per capita income of Latin America is about 30 per- Clearly, the existing differences in development between cent of the per capita income of the developed world, on the region and the developed world did not appear the basis of population-weighted averages, and about overnight. In fact, they are likely the result of historical 25 percent of U.S. levels. Even if Latin America manages processes that in some cases go back to the colonial period. to double the growth rates it experienced during the For example, de Ferranti and others (2004) argued that to 1990­2003 period, the region as a whole would still need understand the region's contemporary situation, one needs about 70 years to reach the current levels of development to recognize the role played by the colonial inheritance of its northern neighbor. (characterized by the extremely high inequality that This chapter relies heavily on a background paper for this report, "Growth and Poverty in Latin America: A Historical View," by Leandro Prados de la Escosura. 45 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S emerged soon after the Europeans began to colonize) and While those initial conditions help explain the magni- the institutional framework put in place at the time (which tude of the region's current development gap, authors such allowed a small group of elites to protect the large rents as Prados de la Escosura (2005) have also stressed the role they were enjoying and excluded most of the population played by developments during the second half of the 20th from access to land, education, and political power). That century, when Latin America seems to have lost significant report also noted that both the initial inequality and the ground relative to most of the reference groups that one institutions that appeared were shaped more by the factor might consider, including the United States, the developed endowments, found by the colonial powers, that favored nations in the OECD, East Asia, peripheral Europe the establishment of large plantations and extractive activ- (Greece, Ireland, Portugal, and Spain), and Spain itself. In ities relying on forced labor rather than by the nature of the fact, the Latin American development gap relative to the colonial powers themselves. developed countries may have opened by between 15 and This type of argument is put forward by, among others, 20 percentage points since 1950. Engerman and Sokoloff (2006), who argue that the impact In this chapter, we review how and when the Latin of the colonial inheritance can be observed not only in the American development gap appeared and pose some basic current high levels of income inequality but also in the per- questions. How much has Latin America grown economi- sistent poverty. This is so because institutional arrange- cally since its independence from colonial rule? How much ments that place the economic opportunities created in the did it lag behind advanced economies in the 19th century? development process beyond the reach of broad segments of Has that gap widened steadily over time? How has society are likely to result in reduced growth rates, as mod- inequality in Latin America evolved historically, and how ern economies require broad participation in entrepreneur- has it evolved elsewhere in the world? Is today's high ship and innovation.3 Thus Engerman and Sokoloff note inequality a permanent feature of modern Latin America? that the gap in per capita incomes between Latin America In short, how did we get here? and the richer countries began in the 18th and 19th cen- Clearly, accurate answers to these questions depend turies. largely on data; hence to set the debate, one needs to try to Haber (1997), for example, finds that from 1800 to the measure the evolution of living standards (per capita early 1900s, per capita GDP grew one and one-half times income or production and its distribution across the differ- in Mexico and not at all in Brazil. Over the same period, ent households or individuals). This chapter is foremost a per capita income in the United States grew sixfold. Put contribution to that effort in that it presents, discusses, and another way, whereas U.S. per capita income in 1800 was compares with other countries and regions the long-run not quite twice that in Mexico and roughly the same as trends (1850­2000) of Latin American per capita income in Brazil, in the early 1900s it was about four times that and inequality. of Mexico and seven times that of Brazil. Similarly, Coatsworth (1998) suggests that Latin America fell into Per capita income in Latin America: relative backwardness between roughly 1700 and 1900. At A long-run comparative perspective the beginning of that period, the Latin American There are two main steps in assessing the evolution of Latin economies (which still were Iberian colonies) were roughly America's income levels over time. The first is assembling as productive as those of British origin. For most of the historical time-series data on which to base the debate. The subsequent 200 years, however, the Latin American second is acknowledging that the exercise of assessing the economies stagnated whereas those of North America evolution of the region is comparative in nature and there- achieved sustained increases in income levels. fore that it requires deciding which country or region to According to the evidence presented in this chapter, in use as the benchmark. We address these two issues in turn. the early 1900s Latin America had per capita income levels that were about 35 percent of the U.S. level and between Historical per capita GDP estimates 40 and 50 percent of the level of a broader group of devel- for Latin America oped countries. Thus even a century ago, the gap between Research in the quantitative economic history of Latin Latin America and the rich countries was already quite America still has a long way to go, and we lack complete sets significant. of homogeneously constructed GDP estimates that would 46 H O W D I D W E G E T H E R E ? TABLE 3.1 Economic growth in eight major Latin American countries (percent on an annual basis) Time span Argentina Brazil Chile Colombia Mexico Peru Uruguay Venezuela, R.B. de 1850­70 -- 0.2 1.7 -- 0 -- -- -1.2 1870­90 3.3 0.2 2 -- 2 -- 0.4 2.6 1890­1900 -0.8 -0.9 1.2 -- 1.5 -- 0.8 -1.5 1900­13 2.5 2.2 2.3 1.8 1.9 1.4 3.1 2.6 1913­29 0.9 1.4 0.9 3.9 0.4 3.6 0.9 6.8 1929­38 -0.8 1 -0.8 1.4 0.4 0.1 0.1 0.5 1938­50 1.7 1.6 1.3 1.5 3.5 1.2 1.5 4.3 1950­60 1.1 3.7 1.5 1.6 2.3 2.9 0.6 3.4 1960­70 3.9 3.1 1.9 2.2 3.4 2.3 0.8 2.4 1970­80 2.1 5.8 0.9 2.9 2.5 1.7 2.1 0.1 1980­90 -2.4 -0.2 1.2 1.1 -0.1 -3.3 -0.2 -1.9 1990­97 5.0 1.5 6.1 1.3 1.0 3.0 3.2 1.1 1997­2000 -1.2 0.0 0.9 0.6 -0.5 0.8 -2.0 -3.2 1870­29 1.8 0.8 1.6 1.5 1.5 1.3 1.2 3.0 1938­80 2.9 4.5 1.8 2.7 3.9 2.6 1.7 3.5 1980­2000 0.4 0.4 2.9 1.1 0.2 -0.5 0.7 -1.0 1870­1980 1.7 1.8 1.3 2.0 1.9 1.8 1.1 2.7 1870­2000 1.5 1.6 1.6 1.9 1.7 1.4 1.1 2.1 Source: Prados de la Escosura (2005). allow international comparisons across time. Recent inde- eight countries, Uruguay had the lowest per capita growth pendent attempts to build GDP series for Argentina, Chile, rate (1.1 percent), followed by Peru (1.4 percent) and Colombia, and Uruguay ease the problem of assessing Latin Argentina (1.5 percent). Brazil and Chile were intermediate America's performance quantitatively over time.4 Yet for cases, both with an estimated per capita growth rate of 1.6 most Latin American countries, product or income data are percent per year. At this growth rate, per capita GDP dou- not available before 1900 and, to the best of our knowledge, bles roughly every 45 years, so today per capita GDP for no Latin American country has reliable comparable data these countries would be about eight times the observed before 1850 (that is, direct comparisons with the first half of level in the late 1800s. One interesting issue that emerges the 1800s are not possible).5 from the table regards the low variance of the average Considering these limitations, table 3.1 compares the growth rates over the 1870­2000 period. In fact, excluding per capita growth rates of eight major Latin American Uruguay and República Bolivariana de Venezuela, the countries with a combined population that represents growth rates of the remaining countries ranged within half almost 90 percent of the whole region's population in a percentage point, from 1.4 percent to 1.9 percent. 2003. These growth rates are presented at roughly decadal As for the evolution of per capita growth over time, benchmarks for the period 1850­2000 (although admit- table 3.1 suggests that for most of the countries, the tedly for four of the countries we do not have access to reli- 1938­80 period was the most productive. This was espe- able growth rates for the 1850­70 period). The estimates cially true for Brazil and Mexico, where per capita growth come from Prados de la Escosura (2005), who in a back- for the period is estimated at 4.5 and 3.9 percent, respec- ground paper for this report, constructs comparable histor- tively. The exception is Chile, where average per capita ical income and inequality series for a number of Latin growth over 1938­80 was 1.8 percent, compared with 2.9 American countries. percent over 1980­2000. Table 3.1 indicates that over the 1870­2000 period, Except for Chile, however, the last two decades of the República Bolivariana de Venezuela had the highest per 20th century were not very positive (Peru and República capita growth rate (2.1 percent a year), followed closely by Bolivariana de Venezuela actually experienced negative per Colombia (1.9 percent) and Mexico (1.7 percent). Of the capita growth rates) due to two negative episodes. The first 47 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S is the lost decade of 1980s following the Latin American FIGURE 3.2 debt crisis. The second is the period following the Asian Per capita growth and initial income levels in eight major Latin financial crisis of 1997 and the Russian financial crisis in American countries 1998. Had it not been for the positive performance of the Per capita growth, % region during 1990­97, when all eight countries under 2.2 R.B. de Venezuela consideration enjoyed substantial positive growth (and half 2.0 Colombia of them enjoyed per capita growth rates that more than 1.8 Mexico doubled their historical trends), the last part of the 20th 1.6 Brazil Chile Argentina century would have been much more dramatic than it actu- Peru 1.4 ally was.6 1.2 Uruguay The growth rates in table 3.1, when combined with 1.0 recent estimates of the level of per capita GDP, can be used 5.0 5.5 6.0 6.5 7.0 7.5 Initial income, log to assemble historical trends in per capita GDP. Estimates of the per capita incomes levels circa 1900 for the eight Source: Authors` calculations. countries covered in table 3.1 show that Uruguay was the richest with a per capita GDP of $1,645 (in 1980 Geari- convergence hypothesis of income levels between the Latin Khamis PPP $). It was followed by Argentina ($1,375), American countries. That is, over the past century or so, Chile ($1,209), Mexico, ($1,141), Peru ($491), Brazil have countries that were initially poorer managed to grow ($444), Colombia ($427), and República Bolivariana de faster than those that were initially richer? To explore the Venezuela ($407). empirical evidence on this issue, figure 3.2 compares the Figure 3.1 plots the per capita GDP trends for the eight average annual growth rates experienced by the different Latin American countries in question (in Geari-Khamis PPP countries between 1870 (or the earliest date available) and 1980 $).7 Although the figure shows some dispersion in 2000 with their corresponding (logged) initial per capita the GDP levels (especially toward the end of the sample), income level in 1870. The figure clearly shows a negative the parallelism in the evolution of the income levels of the correlation between these two variables. The estimated slope different countries is remarkable. of the regression line is -1.3, and it has an associated stan- dard error of 0.30. Although one has to be careful extrapo- Income convergence in Latin America lating results based on only eight countries, the evidence One interesting question regards whether the evidence that presented here would indicate that initially poorer countries emerges from the estimated long-run trends supports the in the late 1800s grew faster over the ensuing 130 years than the initially richer countries. This, in turn, would lend FIGURE 3.1 some support to the hypothesis of convergence of incomes Per capita GDP for eight major Latin American countries, across the Latin American countries during this period. 1850­2000 Figure 3.3 changes the focus of the analysis somewhat 1980 Geari-Khamis PPP $ and plots the cross-country standard deviation of logged per 7,000 capita income. This is a measure of income dispersion that 6,000 5,000 can be understood as an alternative way to explore the pos- 4,000 sibility of convergence.8 This figure suggests that disper- 3,000 sion of cross-country per capita income increased during the 2,000 first epoch of globalization (1870­1913) and then decreased 1,000 during the deglobalization of the interwar years, whereas 0 185018601870188018901900191319251929 19381950196019701975198019902000 between the late 1930s and 1970, the dispersion of cross- country per capita income increased once more before Argentina Brazil Chile Colombia falling in 1980 to its historical low. Overall, figure 3.3 Mexico Peru Uruguay R. B. de Venezuela suggests a convergence in per capita income levels over the 1870­2000 period, albeit with a number of ups and downs Source: Prados de la Escosura (2005). suggesting periodic increases in cross-country inequality. 48 H O W D I D W E G E T H E R E ? FIGURE 3.3 FIGURE 3.4 Cross-country dispersion of per capita GDP in Latin America, Aggregate per capita income in Latin America, 1850­2000 1870­2000 US$ PPP Log scale 5,000 0.7 4,000 0.6 3,000 0.5 2,000 0.4 0.3 1,000 0.2 0 0.1 185018601870188018901900191319251929 19381950196019701975198019902000 0 LA4 LA6 LA10 LA14 LA20 38 70 1870 1880 1890 1900 1913 1925 1929 19 1950 1960 19 1975 1980 1990 2000 Source: Authors' calculations. Source: Authors` calculations. Note: See table 3.2 for the list of countries in each group. Long-run per capita GDP trends in Latin America Having reviewed the evidence for several individual coun- tries, we now move to analyze the evolution of per capita weighted measures of regional real per capita GDP growth income at the regional level. The results are shown in (table 3.2) and regional real GDP income levels (figure 3.4) table 3.2 and in figure 3.4, which report population- over the past 150 years. In addition to the eight major countries discussed above, we now introduce several other TABLE 3.2 Latin American economies in the time periods for which Aggregate per capita growth in Latin America historical data are available. Clearly, the lengthier the (percent) coverage, the lower the number of countries covered. A number of features can be pointed out regarding the Time span LA20 LA15 LA10 LA6 LA4 aggregate performance of Latin America. First, the picture 1850­70 0.2 of Latin America's performance seems quite robust (this is 1870­90 1.7 1.4 in part a result of the low variance of growth rates across 1890­1900 0.4 0.5 countries). After a slow start in the mid-1800s when per 1900­13 2.3 2.2 1.8 1913­29 1.2 1.2 1.0 0.9 capita income growth was probably well below 1 percent, 1929­38 0.1 0.2 0.1 0.4 growth in Latin America appears to have risen significantly 1938­50 2.1 2.1 2.3 2.6 1950­60 2.3 2.3 2.3 2.4 3.0 during the 1870s and 1880s, slowed during the 1890s, and 1960­70 2.9 2.9 3.0 3.2 3.2 accelerated in the early 1900s. It then slowed again because 1970­80 3.3 3.3 3.3 3.4 3.7 of World War I and came to a halt during the Great 1980­90 -0.5 -0.5 -0.4 -0.5 -0.2 1990­2000 1.3 1.3 1.3 1.5 1.3 Depression. 1870­1929 1.4 1.2 From the late 1930s up to 1980, however, Latin America 1938­80 2.9 2.6 2.6 2.7 3 began displaying robust growth. Over this period, depend- 1980­2000 0.4 0.4 0.4 0.5 0.6 ing on the sample under consideration, growth appears to 1870­1980 1.8 1.9 1870­2000 1.6 1.7 have hovered around 2.5­3.0 percent (with this growth, per capita income doubles every 25 years or so). The 1980s, Source: Authors' calculations. however, saw a reversal of fortunes with per capita income Note: LA20 = population-weighted average of Latin American declining by 0.5 percent a year on average (a cumulative countries; LA15 = population-weighted average of LA10 + Costa Rica, El Salvador, Guatemala, Honduras, and Panama; decline of 5 percent in per capita income levels). Finally, LA10 = population-weighted average of LA6 + Colombia, Cuba, one can also clearly observe the recovery that took place Ecuador and Peru; LA6 = population-weighted average of LA4 + Argentina and Uruguay; LA4 = population-weighted average during the 1990s, which as mentioned previously extended of Brazil, Chile, Mexico, and Venezuela. to the end of the decade. 49 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S On the whole, Latin American per capita income levels TABLE 3.3 are now between eight and nine times the observed level in Economic growth in several reference groups 1850, about six times the level in 1900, and about two and a (percent) half times the level in 1950. With this information in hand, we are now in a position to compare the relative evolution of Time span United States Spain OECD PE EA GDP in the region against different reference groups. 1850­70 2.2 0.5 1.5 0.8 1870­90 1.6 1.5 1.4 1.2 0.8 1890­1900 1.8 0.9 1.5 0.9 0.7 Comparative perspective 1900­13 1.9 1.0 1.6 0.9 0.7 How does Latin America's per capita GDP perform in 1913­29 1.6 1.7 1.3 1.6 1.9 1929­38 -0.5 -4.8 0 -1.7 2.4 comparison with other countries and regions of the world? 1938­50 4.7 1.8 3.2 0.6 -4.4 Typically, historical comparisons of Latin America have 1950­60 1.7 3.6 2.7 3.3 3.6 1960­70 2.9 7.4 3.4 6.3 5.6 taken the United States as reference. Over the 19th century, 1970­80 2.1 3.7 2.4 3.9 7.0 however, even in western European economies, per capita 1980­90 2.1 2.9 2.1 2.4 5.7 1990­2000 1.9 2.4 1.9 2.7 4.8 GDP lagged behind the United States. That suggests com- 1850­1900 1.9 1.0 1.5 1.0 paring Latin America with only the United States may bias 1870­1929 1.7 1.3 1.4 1.2 1.0 the assessment in that the United States was the leading 1938­80 2.9 3.9 2.9 3.4 2.5 performer during this period and hence serves as a very nar- 1980­2000 2.0 2.6 2.0 2.5 5.2 1870­1980 2.0 1.8 1.9 1.8 1.7 row reference. To try to control for this possibility, we take a 1870­2000 2.0 1.9 1.9 1.9 2.3 broader view and consider the performance of several different groups. These include the group of developed Source: Authors' calculations based on Maddison (2005). countries that today are part of the OECD; Spain, a country Note: Peripheral Europe (PE) includes Greece, Ireland, Portugal, with which Latin America shares some institutional back- and Spain. East Asia (EA) consists of Hong Kong (China), Republic of Korea, Singapore, and Taiwan (China). ground; peripheral Europe, which includes countries known for quickly catching up with European Union levels; and East Asia (covering Hong Kong, China; the Republic of the same level as East Asia, both Spain and peripheral Korea; Singapore; and Taiwan, China) to take account of the Europe also outperformed the United States and the OECD "Asian miracle." Table 3.3 reports the growth rates these group. Even the OECD group seems to have performed rel- reference groups have experienced since 1850. atively better than the United States over the second half of This table indicates that during the second half of the the 20th century. Thus whether the Latin American experi- 19th century, the United States was the fastest-growing ence over this period is considered a success depends to a economy, with per capita GDP growth of almost 2 percent large extent on the countries and regions being considered on an average annual basis (reaching 2.2 percent over as a reference group. 1850­70). OECD's advanced economies grew at 1.5 per- Figure 3.5 graphically illustrates the evolution of cent, and Spain and the peripheral Europe group each grew income trends (relative to the United States) for a group of at about half the U.S. rate (1 percent in both cases).9 four Latin American countries (Brazil, Chile, Mexico, and Although we do not report data for the four East Asian República Bolivariana de Venezuela) and for the other four economies until 1870, the existing estimates suggest that groups under analysis. Several messages emerge from the this group also was growing at a much slower pace than the figure. First, in 1850 Latin America's per capita GDP was United States (the estimates for the Asian economies in already about 60 percent of the U.S. level, whereas Spain's table 3.3 over the 1870­1900 period would suggest an was about 80 percent, and peripheral Europe's was 75 per- average per capita growth rate of less than 1 percent a year). cent. The OECD group as a whole was richer than the Thus, as already noted, the United States performed signif- United States (107 percent). For East Asia the first avail- icantly better than all other regions under consideration able estimates correspond to 1870. Then it was the poorest during this period. among those considered here with per capita income levels Starting in the 1960s, however, East Asia became the representing only 25 percent of the U.S. levels. fastest-growing group, with per capita growth rates in the Interestingly, 110 years later, in 1980, the situation con- 6­7 percent range until the 1980s. Moreover, while not at tinued to be very similar, the result of all the groups under 50 H O W D I D W E G E T H E R E ? FIGURE 3.5 FIGURE 3.6 Per capita income of five groups relative to the United States, Incomes in Spain and Peripheral Europe relative to OECD countries 1850­2000 Ratio Ratio 0.8 Spain 1.2 0.6 1.0 0.8 0.4 Peripheral 0.6 Europe 0.2 0.4 0.2 0 0 9 0 1850 1860 1870 1880 1890 1900 1913 1925 1929 1938 1950 1960 1970 1975 1980 1990 2000 18501860 1870 1880 189019001913 1925192 1938 195 1960 1970 197519801990 2000 Source: Authors' calculations. Latin America United States Spain East Asia Peripheral Europe Spain and peripheral Europe were also moving up toward Source: Authors' calculations. U.S. levels, and more significantly toward OECD levels (figure 3.6). consideration sharing some trends relative to the United Admittedly, the trends observed in Spain, peripheral States. First, they all lost significant ground in the second Europe, and East Asia during the 1980s and 1990s were to half of the 19th century, Second, they all lost some ground in a large extent a continuation of those observed since 1950. the first half of the 20th century. And third, they all This is shown in figure 3.7, which presents the evolution of regained some of the lost ground in the 1950­80 period. In population-weighted average per capita income levels for fact, in 1980 the OECD group was still leading our four Latin America relative to the different reference groups. comparison groups, although its relative income levels had Looking first at panel a, which compares Latin America fallen to about 80 percent of those of the United States. Per with the OECD, the picture indicates that the region was capita income levels in Spain and peripheral Europe were losing ground during the last part of the 19th century. 50 percent of those in the United States, while in Latin However, panel a also indicates that Latin America America they were 30 percent, and in East Asia they were experienced a significant decline over the second half of the close to but still below 30 percent. Thus, over the 20th century. For example, Latin America's per capita 1850­1980 period, mobility was quite limited in our income levels fell from about 45­50 percent of OECD's country groupings. In relative terms, those groups that levels in 1950 to about 30 percent in 2000. Thus Latin started poor compared with the United States remained America may have experienced the paradox of fast growth poor and those that started rich (also compared with the (recall that 1950­80 was the fastest-growing experience of United States) remained rich. the region with per capita growth rates in the 3 percent Does this lack of mobility mean that countries cannot range) while losing ground relative to the advanced break with history and therefore that states of development economies. are given and immutable? Well, the answer is that coun- When the region is compared with Spain (panel b), the tries and regions can indeed break with history--as a series picture is somewhat different. Over the 1850­1930 period, of developments since 1980 confirm. East Asia more than Latin America's per capita income remained basically con- doubled its relative income during the last two decades of stant relative to Spain, and if anything it increased. The the 20th century, moving from 27 percent of U.S. levels in central years of the 20th century, resulting from Spain's 1980 to 55 percent in 2000 (see figure 3.5). Put another civil war and autarkic aftermath, witnessed a dramatic way, in just 20 years, the four East Asian economies moved recession in Spain (Latin American income levels were in from last in our relative classification to levels comparable this period about 20 percent higher than Spain's). How- with those observed for Spain and peripheral Europe. This ever, as Spain reengaged in the world economy in the achievement is even more remarkable when one considers 1950s, the country began regaining lost ground. Spain 51 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 3.7 GDP per capita in Latin America relative to several country groupings, 1850­2000 a. Latin America relative to OECD b. Latin America relative to Spain 0.60 1.30 0.55 1.20 1.10 0.50 1.00 0.45 0.90 0.40 0.80 0.35 0.70 0.30 0.60 0.25 0.50 0.20 0.40 25 1850 1860 1870 1880 1890 1900 1913 1925 1929 1938 1950 1960 1970 1975 1980 1990 2000 1850 1860 1870 1880 1890 1900 1913 19 1929 1938 1950 1960 1970 1975 1980 1990 2000 c. Latin America relative to East Asia d. Latin America relative to Peripheral Europe 3.5 1.4 3.0 1.2 2.5 1.0 2.0 0.8 1.5 0.6 1.0 0.4 0.5 0.2 0 0 38 25 1870 1880 1890 1900 1913 1925 1929 19 1950 1960 1970 1975 1980 1990 2000 1850 1860 1870 1880 1890 1900 1913 19 1929 1938 1950 1960 1970 1975 1980 1990 2000 LA4 LA6 LA10 LA14 LA19 Source: Authors' calculations. Note: See table 3.2 for the list of countries in each group. grew faster in the 1950s than Latin America did and expe- well between 1850 and 1950. From 1950 onward, how- rienced exceptional growth in the 1960s and early 1970s.10 ever, things changed, and Latin America's performance Moreover, despite near stagnation during the "transition to declined sharply over the next five decades relative to those democracy" (1975­85), Spain's growth was above the groups. OECD average during the last two decades of the 20th cen- The relevance of the second half of the 20th century for tury. By the 1980s incomes in Latin America were at about understanding the magnitude of Latin America's current the same levels relative to Spain as they had been 100 years development gap relative to several country groupings is earlier. also apparent from figure 3.8. This figure is based on the In a similar fashion, putting Latin America side by side regional estimates of per capita income levels in Maddison with peripheral Europe (panel c) and East Asia (panel d), (2005), which go back in some cases to 1500. According one would also conclude that Latin America performed to figure 3.8, between 1820 and 1870, Latin America 52 H O W D I D W E G E T H E R E ? Factor endowments, technology, and relative scarcity of FIGURE 3.8 resources have had important implications for the initial Latin American per capita GDP relative to Western Europe, 1500­2001 inequality levels. For example, in Latin America the char- acteristics of the colonies favored the establishment of large Relative GDP plantations (such as sugar) and mining activities that 0.7 employed forced labor. As a result, a social structure 0.6 emerged where a privileged few were in control of most of 0.5 the profitable activities and where, most importantly, most of the population was excluded from access to land, educa- 0.4 tion, and political power. In contrast, the colonial powers 0.3 in North America soon learned that there was no gold, few indigenous peoples to exploit, and soils and climates that 0.2 would not support the production of crops based on large 0.1 slave plantations. In fact, unlike in the South, in the North land was cheap and labor scarce. In addition, fewer health 0 problems affected European settlements in North America. 1500 1820 1870 1913 1950 1973 2001 Year Such circumstances led to open competition among the earlier colonies to attract migrants by providing favorable Source: Maddison (2005). working conditions, something that in turn fostered a remarkable degree of equality.11 The issue of what created an initial level of high lost significant ground relative to Western Europe. Latin inequality is clearly different from the issue of why America's situation then improved markedly vis-à-vis inequality persisted over time. Inequality in Latin America Western Europe in the first half of the 20th century. By and the Caribbean: Breaking with History? argues that the 1950 Latin America's position was similar to the one it persistence of inequality during the colonial and early inde- held in 1820. After 1950, however, the region experienced pendence period occurred because the initial nexus of insti- a dramatic decline, with relative income falling from about tutions survived, as did the rationale for these institutions. 55 percent of that in Western Europe to about 30 percent. Given the disparities in resources that resulted from the On the whole, Latin America thus appears to have lost colonial period, the Creole elite who had benefited from ground since the mid-1800s relative to several other coun- those disparities during colonial times were able to quickly try groupings, and the downward slide seems to have been gain effective control of the independent countries and particularly fast in the last half of the 1900s. Breaking with determine the general structure of the institutions in ways this historic pattern will not be easy, but as East Asia, that favored their interest. Spain, and peripheral Europe have demonstrated, it can be Explaining the persistence of inequality over the 20th done, and countries can put themselves on an upward path. century is more problematic because significant social, eco- nomic, and political changes occurred during the 1900s. Long-run inequality Moreover, the increase in urbanization rates should have Together with Sub-Saharan Africa, Latin America has long somewhat mitigated the relevance of the highly inegalitar- been known as the region with the highest inequality in ian pattern of land ownership and its impact on income the world, with a Gini coefficient above 0.50 since the inequality. In addition, modernization moved most of the 1960s. What explains this high level of inequality? Various Latin American countries in the direction of more open and alternative interpretations have been offered, but to a large democratic societies. Inequality in Latin America and the extent they all follow the colonial inheritance argument Caribbean: Breaking with History? offers a number of conjec- coupled with the persistence of the initial institutions. tures in this regard, including slow increases in coverage and Inequality in Latin America and the Caribbean: Breaking low quality of education, a development strategy based on with History? (de Ferranti and others 2004) stressed the import substitution and isolation from world markets, and joint role played by factor endowments and institutions. imperfect financial markets that may have prevented those 53 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S at the bottom of the income distribution from exploiting four decades later. Similarly, El Salvador may have experi- economic opportunities by restricting their access to credit. enced a significant worsening in inequality over the Unfortunately, no quantitative assessment of long-run 1950­90 period, while Honduras saw some improvements. inequality validating these arguments has been carried out For the pre-1950 period, data availability prevents direct for Latin America. A good example is provided by the Bour- inequality comparisons. However, one can still explore guignon and Morrisson (2002) investigation of the historical empirically the evolution of income inequality using indi- trends in world income inequality. Conventional wisdom rect indicators and a handful of country studies that follow and lack of empirical evidence led them to assume that no that approach. One such study is the path-breaking work changes in income distribution had taken place in Latin by Bértola (2005) for Uruguay, which provides crude esti- America from independence to the mid-20th century. mates of income distribution and Gini coefficients that go Can we quantify trends in income inequality in modern back to the late 1800s. Also notable is the work by Latin America? It is possible to infer the evolution of Williamson (1999), who explored the consequences for inequality since 1950 on the basis of direct income distrib- inequality of the early phase of globalization (1870­1914). ution observations. For example, in table 3.4 we report On the basis of the wage­land rental ratio, he showed an Gini coefficients for several Latin American countries as increase of within-country inequality for Argentina and well as a population weighted regional average. This table Uruguay over that period. Bértola and Williamson (2003) indicates that inequality remained basically constant from follow up on that line of research and argue that inequality 1950 to 2000 at between 0.51 and 0.55 on the Gini index. trends reversed in the interwar period, when the observed Admittedly there is significant country heterogeneity. steep decline in the wage-rental ratio stopped, and then For example, the Gini index markedly increased in increased somewhat after the 1930s. Calvo, Torre, and Argentina, from 0.40 to 0.48 between 1950 and 1990, but Szwarcberg (2002) suggest that the extent of inequality it may have declined in República Bolivariana de Venezuela changed little during the century in Argentina, whereas from a high of 0.61 in the mid-20th century to about 0.45 Londoño (1995) argues that the inequality levels observed in Colombia during the 1990s were probably similar to those observed in 1938. TABLE 3.4 In a background paper for this report, Prados de la Inequality in Latin America 1950--2000, as measured by Escosura (2005) builds on Williamson (2002) to explore Gini coefficients (percent) the historical evolution of the ratio of GDP per worker to the unskilled wage between 1850 and 1950 (or earliest 1950 1960 1970 1980 1990 possible date) for Argentina, Brazil, Chile, Mexico, and Argentina 39.6 41.4 41.2 47.2 47.7 Uruguay. The rationale for this choice is that such a ratio Bolivia 53 53.4 54.5 compares the returns to unskilled labor with the returns to Brazil 57 57.1 57.1 57.3 Chile 48.2 47.4 53.1 54.7 all production factors, that is, GDP. Since unskilled labor is Colombia 51 54 57.3 48.8 50.3 the more evenly distributed factor of production in devel- Costa Rica 50 44.5 48.5 46 oping countries, an increase in the ratio suggests that Dominican Republic 45.5 42.1 48.1 El Salvador 42.4 46.5 48.4 50.5 inequality is rising. On that basis Prados de la Escosura Honduras 61.8 54.9 57 (2005) concludes that in Argentina, Chile, and Uruguay Mexico 55 60.6 57.9 50.9 53.1 Panama 50 58.4 47.5 56.3 income inequality does not seem to have changed much Paraguay 45.1 57 over the period whereas Brazil and Mexico may have expe- Peru 61 48.5 43 46.4 rienced some deterioration in the distribution of income. Uruguay 37 42.8 43.6 40.6 Venezuela, R.B. de 61.3 46.2 48 44.7 45.9 On the whole, all the evidence that emerges from these LAC4 50.5 53.2 53.1 49.1 50.7 studies indicates that, on average, Latin America entered LAC6 54.8 54.8 53.2 54.2 the 20th century with a very high level of inequality, which LAC15 53.9 51.9 53.2 persisted for the rest of the century, despite significant vari- Spain 45.7 36.3 34.7 ations by country in different periods. How do these trends compare to those observed in the Source: Altimir (1987); Londoño and Székely (2000). Note: See table 3.2 for LAC4, LAC6, LAC15 group definitions. advanced economies? Spain experienced a significant decline 54 H O W D I D W E G E T H E R E ? FIGURE 3.9 FIGURE 3.10 Income inequality in the United States and Spain, 1910­90 Income inequality in the United Kingdom and France, 1910­90 Gini index Income, top 1% 0.7 0.20 United States United Kingdom 0.6 Spain 0.15 France 0.5 0.10 0.4 0.05 0.3 0 1910 1920 1930 1940 1950 1960 1970 1980 1990 0.2 Source: Atkinson (2003). 0.1 0 to a large extent most of the decline took place between 1910 1920 1930 1940 1950 1960 1970 1980 1990 1940 and the late 1970s. Atkinson (2003) relies on income Source: Plotnick and others (1996) for the United States; Prados de tax statistics to construct estimates of the income shares of la Escosura (2005) for Spain. the wealthiest percentile in the United Kingdom. The esti- mates show that in the early 1900s the richest 1 percent in in income inequality between the 1970s and the 1990s, the United Kingdom shared almost 20 percent of total per- when the Gini coefficient fell by more than 10 percentage sonal income; in 1940 they had 17 percent; and in the late points (see the bottom row of table 3.4). Unfortunately, 1970s, when the declining trend in inequality was there are no direct estimates of the Gini coefficient for reversed, they held a mere 6 percent (figure 3.10). Spain before 1970. However, existing indirect indicators The results in Atkinson (2003) also indicate that income (Prados de la Escosura 2005) suggest that income inequal- inequality in France evolved in about the same way as it did ity has been declining in Spain since the 1950s, when Spain in the United Kingdom (at least until the 1980s). In the may have had inequality levels comparable to (if not higher early 1900s, the share of income of the richest percentile in than) those observed in Latin America. For 1950 Prados France was also about 20 percent, whereas in the 1980s it de la Escosura (2005) estimates a Gini coefficient for Spain was roughly 7 percent. The main difference between these above 0.50. Thus Spain appears to have lowered the Gini two countries is that most of the decline in French income coefficient by almost 15 percentage points between 1950 inequality took place between the 1920s and 1950. It is and 1980 and by around 20 percentage points between notable that Atkinson's estimates of the top percentile's 1950 and 1990 (figure 3.9). income share for both France and the United Kingdom are The estimates of the Gini index for the United States (see consistent with very high inequality levels at the begin- figure 3.9) indicate that from the turn of the century until ning of the century. In fact, if one were to assume that about 1930, inequality remained constant with a high Gini income approximately follows a lognormal distribution of 0.60 (Plotnick and others 1996). This relative stability (see chapter 4), income inequality in the two countries in was interrupted by World War I, which seems to have had a 1900 might have been around 0.60. brief equalizing effect, but starting about 1920 inequality Thus the empirical evidence reported in this section began to rise once more, reaching its pre-World War II high confirms to a large extent the finding of Inequality in Latin in 1929. From 1929 to 1951, income inequality fell dra- America and the Caribbean: Breaking with History? that matically from the prevailing Gini of 0.60 to about 0.40. inequality in Latin America has been persistent and stable The United Kingdom experienced a similar pattern. over the last century. It also confirms that inequality in Acemoglu and Robinson (2002) present evidence indicat- Europe and the United States seems to have declined sig- ing that the Gini coefficient for the United Kingdom could nificantly over the 20th century. In addition, the discussion have been around 0.55 in the 1890s. Then, for most of the notes that the levels of inequality in Latin America in the 20th century, inequality seems to have declined, although early to mid-1900s may not have been so much different 55 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S from those observed in France, Spain, the United Kingdom, 6. See Loayza, Fajnzylber, and Calderon (2005) for an analysis of and the United States, but while these countries signifi- the recent Latin American growth experience and the positive impact of the liberalization process of the 1990s on the growth performance cantly reduced their inequality at different moments in of the different countries. time, Latin America has yet to do so. The question remains: 7. Note that the data in figure 3.1 are in constant 1980 PPP If other countries have managed to break with their histo- dollars, so the per capita GDP ranking of the countries does not nec- ries on both the growth and income per capita fronts, then essarily coincide with rankings given in other parts of this report that why cannot Latin America also break with its history? use constant 1996 PPP dollars (when the source of data is the Penn World Tables (PWT6.1)) or constant 2000 PPP dollars (when the source of data is the World Development Indicators). Notes 8. Although now it would be -convergence rather than 1. See table A19 of the report. The figures refer to the mid- -convergence. See Barro and Sala-i-Martin (1995) for a discussion of 1990s, so the current levels may be different. the different concepts of convergence. 2. The inference is based on the results that emerge from using a 9. The OECD group used here consists of Australia, Austria, lognormal approximation for the distribution of income. See chap- Belgium, Canada, Denmark, Finland, France, Germany, ter 4 of this report for a discussion of that particular assumption. Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, the 3. See also chapter 6 of this report for a discussion of how social United Kingdom, and the United States. exclusion from the development process can result in lower GDP 10. In Spain, the year 1938 represents a trough in economic growth rates. performance. 4. See Cortés Conde (1994, 1997) and Della Paolera, Taylor, and 11. What mattered for the initial inequality level was not the Bózolli (2003) for Argentina; GRECO (2002) for Colombia; Díaz, identity of the colonizing power but rather the characteristics of the Lüders, and Wagner (1998) for Chile; and Bértola (1998) for colonies. The British colonies of British Honduras, Guyana, and Uruguay. Jamaica resulted in levels of inequality similar to most of those in 5. This is not to say, however, that there is no estimated data for Latin America. In contrast, in Argentina, Costa Rica, and Uruguay, the pre-1850 period. In fact, Maddison (2005) presents data going where there were few Native Americans, the social structure was not back to 1500. so unequal. 56 CHAPTER 4 The Relative Roles of Growth and Inequality for Poverty Reduction Growth is good for the poor, and growth that is accompanied by progressive distributional change is even better. But are the same type of policies appropriate for all countries that want to reduce poverty quickly? For example, should Chile and Nicaragua--two countries with similar levels of inequality but dramatically different income levels--try to strike a sim- ilar balance between growth-promoting and inequality-reducing policies? Similarly, should Uruguay and Brazil--which have similar levels of per capita income but are the least and most unequal countries in the region, respectively--follow sim- ilar policies in their attempts to reduce poverty? T HE LAST DECADE HAS WITNESSED A the distribution unchanged. There are two main reasons for booming literature on the links among this. One is that, in general, for a fixed level of income, pro- growth, inequality, and poverty reduction. gressive distributional change will shift resources from the As a result of this debate, a more or less broad richer to the poorer and thus lead to poverty reduction.1 consensus has emerged on a few findings. The other reason is that poverty is more responsive to First, nobody seems to doubt the importance of growth growth the more equal the income distribution. This point for poverty reduction. Countries that have historically is illustrated in panel c of figure 4.1, which plots the total experienced the greatest reduction in poverty are those that elasticity of poverty against the logged Gini index for a have experienced prolonged periods of sustained economic selected number of countries. The upward slope of the growth (panel a of figure 4.1). For example, over the regression line in this picture indicates that as inequality 1981­2000 period, China's poverty rate fell from more increases (that is, as one moves to the right of the horizon- than 50 percent to about 8 percent, thanks to an impressive tal axis), the growth elasticity of poverty becomes less neg- per capita growth rate of almost 8.5 percent a year. Simi- ative. Thus progressive distributional change will have, in larly, between 1993 and 2002 Vietnam cut its poverty rate addition to the one-shot instant impact on poverty derived in half, from about 58 percent to about 29 percent, by from the pure redistribution effect, a long-run effect growing at almost 6 percent a year. derived from an increase in the sensitivity of poverty to Second, progressive distributional changes are good for growth. poverty reduction (see figure 4.1, panel b). While it is diffi- The third finding is that there is no strong empirical cult to argue that poverty reduction can be achieved evidence suggesting a general tendency for growth as such through redistributive policies in the absence of economic to make income distribution more or less equal (figure 1, growth, growth associated with progressive distributional panel d). For example, Dollar and Kraay (2002) find that, changes will reduce poverty more than growth that leaves on average, the income of the poorest fifth of society rises This chapter is based on the background paper for this report "A Normal Relationship? Poverty, Growth and Inequality" by H. Lopez and L. Servén (2005a). 57 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 4.1 Growth, inequality, and poverty reduction throughout the world a. Poverty and growth b. Poverty and inequality Change in headcount poverty Change in headcount poverty 0.3 0.3 0.2 0.2 0.1 0.1 0 0 0.1 0.1 0.2 0.2 0.3 0.3 0.20 0.15 0.10 0.05 0 0.05 0.10 0.15 0.20 0.20 0.15 0.10 0.05 0 0.05 0.10 0.15 0.20 Per capita growth Change in inequality c. Growth elasticity and inequality d. Growth and inequality Efficiency of growth Change in inequality 6 0.2 4 0.1 2 0 0 2 0.1 4 6 0.2 3.2 3.4 3.6 3.8 4.0 4.2 0.20 0.15 0.10 0.05 0 0.05 0.10 0.15 0.20 Inequality (logged Gini index) Per capita growth Source: Computed on the basis of POVMONITOR data. proportionately with average incomes. Other studies con- advice is not very useful for policy purposes. For one thing, cluding that changes in income and changes in inequality the achievements of both growth and a more equal income are unrelated include Deininger and Squire (1996), Chen distribution are policy outcomes that are a challenge in and Ravallion (1997), and Easterly (1999). themselves. But beyond that, the discussion leaves unan- The Latin American countries analyzed in chapter 2 also swered a number of questions of extreme relevance for pol- fit this pattern: the linear correlations between changes icy making. For example, how much emphasis should in a given inequality index and income growth rates are policy makers place on achieving a high growth rate and always insignificant regardless of the inequality index and how much on achieving a balanced pattern of growth? the income variable (either survey-based or national What is more advisable from a poverty perspective: a high accounts-based). For example, the correlation between the growth rate that has an associated increase in inequality, or changes in the Gini for the distribution of household a lower growth rate that maintains inequality at a constant income and growth rates in that variable is just -0.02. level? Are there any conditions under which policy makers Growth would thus be good for the poor, or at least as good can accept a trade-off between growth and a deterioration as for everybody else in society.2 in the distribution of income? On the whole, the previous discussion suggests that a The answers to those questions are critical to strike the sensible development strategy should focus both on the right balance between growth-enhancing and inequality- quantity of growth (that is, on the achievement of a high reducing policies in a particular country. For example, if growth rate) and on the quality of growth (that is, on who growth is the main force behind poverty reduction in all benefits from that growth). Unfortunately, this general circumstances, then poverty reduction strategies should 58 T H E R E L AT I V E R O L E S O F G R O W T H A N D I N E Q U A L I T Y F O R P O V E RT Y R E D U C T I O N focus on growth, and policy makers should think twice line. This issue is important because a country can set its before implementing policies that, in the name of a better poverty line very high, so that large numbers of individuals income distribution, lead to a deceleration in growth. If, qualify as poor, or very low, so that the focus is on the poor- however, trends in relative incomes are found to account for est of the poor. Where a poverty line is set could thus deter- the lion's share of poverty changes, then development mine whether policy makers should focus on growth or strategies should also emphasize the pattern of growth, and poverty reduction when targeting different segments of the policy makers might be willing to accept a trade-off population. between fast growth and rapid poverty reduction.3 Clearly, The second way in which this chapter addresses the issue between these two extreme cases, one can expect to find a of the relative importance of growth and redistribution is continuum of possibilities where both growth and changes through the use of a particular functional approximation in inequality will be important, to varying degrees, for for the empirical income distribution. More specifically, we poverty reduction and where specific knowledge about the rely on a lognormal function to simulate how growth and relative importance of each component can prove useful for changes in inequality affect changes in poverty under dif- policy purposes. ferent scenarios and, more specifically, under different This chapter explores the types of questions posed above initial levels of inequality and development. One of the in two complementary ways. First, it applies standard virtues of this type of analysis is that the lognormal func- poverty decomposition techniques to identify the growth tion can easily be calibrated with observed values from and distribution components corresponding to the observed actual countries so that the discussion can move from some poverty changes for 18 Latin American countries. That is, basic generalizations to a country-specific assessment. for each particular country episode, the change in poverty The report makes two contributions on this front. First, that can be attributed to growth is separated from the even though parametric techniques have become very popu- change in poverty that can be attributed to changes in lar in poverty analysis (see, among others, Bourguignon income distribution. Then these variance decompositions 2004, and Kakwani and Son 2003), little effort has been are used to summarize the relative importance of the differ- spent to verify how well the approximations being used fit ent sources of poverty changes. the actual data. In this regard, we present new (and encourag- This type of exercise has been performed in a recent ing) results regarding the goodness of a fit of the lognormal paper by Kraay (2005), who finds that in a global sample of specification. The second contribution is a typology of Latin developing countries, growth in average incomes matters a American countries--grounded on the theoretical analysis-- great deal for poverty reduction. More specifically, Kraay that can be used as a guide to discriminate somewhat between estimates that over the short run, growth accounts for growth and inequality priorities at the country level. about 70 percent of the variation in poverty (as measured by a $1-a-day poverty line). As the time horizon lengthens, The relative roles of growth and income that proportion increases to above 95 percent. In other distribution for poverty reduction words, changes in poverty reduction are almost uniquely Changes in poverty can be related to two main sources: driven by growth in mean income. This finding would changes in mean income, and changes in relative incomes. probably justify development strategies that rely almost Following Bourguignon (2004), figure 4.2 graphically exclusively on growth as a tool for poverty reduction. illustrates this point for a particular measure of poverty, the The analysis in this report adds to this debate in two headcount index (see box 4.1 for a more formal discussion). main dimensions. First, it allows for a comparison between In the figure, poverty is simply the area under the density the Latin American countries and the global context. This function to the left of the poverty line, which in this case is comparison is interesting because, given the high levels of fixed at $1 a day. inequality in the region, one might expect that Latin When mean income or relative incomes, or both, change American development strategies would have to incorpo- from an "initial distribution" to a "new distribution," fig- rate both growth and inequality concerns. In addition, the ure 4.2 shows how the change in poverty can be decom- chapter also explores (within the Latin American context) posed using an intermediate step. First, one can simulate whether the results are sensitive to the choice of the poverty the impact of moving from the initial distribution to a 59 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S virtual distribution given by the horizontal translation of FIGURE 4.2 the original density. The movement to this intermediate Decomposition of poverty into growth and distribution effect density involves no change in relative incomes and hence Density (share of population) can be used to assess the impact of growth on poverty 0.6 Poverty New reduction (light gray in the figure). Notice that this is 0.5 line distribution (I) equivalent to asking about the change in poverty that 0.4 would have taken place if growth had been as observed but 0.3 the distribution of income remained constant. The second movement simulates the impact of moving from the virtual 0.2 Initial (I) density to the actual new distribution. It does not involve a distribution 0.1 (I) change in mean income and hence it captures only the 0 impact of changes in relative incomes on poverty (dark gray 0.1 1 10 100 in the figure). This is now equivalent to asking about the Income, $ a day, logarithmic scale impact of redistribution had per capita income levels Growth effect Distribution effect remained fixed. This simple decomposition provides a basic Growth effect Distribution effect on poverty on poverty statistical framework that can be used to analyze empiri- cally the relative contribution of growth and changes in Source: Bourguignon (2004). income distribution for poverty reduction on the basis of two household surveys. BOX 4.1 Decomposing poverty into growth and income distribution effects There is an identity linking poverty to mean income and poverty resulting from changes in mean income (the the distribution of that income across the different indi- growth component). The second term--P[y1,L1(p)] - viduals or households. It is possible to formally write P = P[y1,L0(p)]--captures the changes in poverty attributable P[y,L(p)], where P is a poverty measure (which for sim- to changes in the Lorenz curve when income levels plicity can be assumed to belong to the Foster-Greer- remain unchanged (distribution component). Thorbecke (FGT) 1984 class, such as headcount poverty, Note that this decomposition is not unique (although the poverty gap, or the squared poverty gap), y is per in principle the empirical differences between alterna- capita income, and L(p) is the Lorenz curve measuring the tives are not likely to be large). The changes of poverty relative income distribution. L(p) is the percentage of can be rewritten using as reference the poverty rate that income enjoyed by the bottom 100 × p percent of the would have occurred had income remained constant at y0, population. Changes in poverty between period 0 and 1 but the Lorenz had shifted to L1(p): can then be expressed as P0,1 = P[y1,L1(p)] - P[y0,L0(p)]. (4.2) P0,1 = P[y1,L1(p)] - P[y0,L0(p)] Adding and subtracting to the right-hand side of the = P[y1,L1(p)] - P[y0,L1(p)] previous expression the poverty rate that would have + P[y0,L1(p)] - P[y0,L0(p)]. resulted had income increased to the final level y1, but the Lorenz curve had remained constant at L0(p)--that In this alternative decomposition, the growth component P[y1,L0(p)]--it is possible to write: is captured by P[y1,L0(p)] - P[y0,L0(p)], and the distribu- tion component by P[y1,L1(p)] - P[y1,L0(p)]; in principle, (4.1) P0,1 = P[y1,L1(p)] - P[y0,L0(p)] = P[y1,L0(p)] - P[y0,L0(p)] these two components do not necessarily have to coincide + P[y1,L1(p)] - P[y1,L0(p)]. with P[y1,L0(p)] - P[y0,L0(p)] and P[y1,L1(p)] - P[y1,L0(p)]. The first term of the right-hand side of equation 4.1-- [P(y1,L0(p)] - P[y0,L0(p)]--measures the changes in 60 T H E R E L AT I V E R O L E S O F G R O W T H A N D I N E Q U A L I T Y F O R P O V E RT Y R E D U C T I O N TABLE 4.1 Poverty, growth, and redistribution in Latin America US$1-a-day poverty line US$2-a-day poverty line Total Growth Redistribution Total Growth Redistribution Country Time span (ii) (iii) (ii) (iii) Argentina 1992­98 1.8 0.0 1.8 4.1 0.1 4.2 1998­2002 6.4 3.2 3.3 15.3 10.9 4.4 2002­4 3.8 2.7 1.0 8.6 5.0 3.5 1992­2004 4.7 1.0 3.7 11.9 4.3 7.6 Bolivia (urban) 1993­97 6.2 5.1 1.1 13.4 12.6 0.7 1997­2002 2.8 1.0 1.8 4.4 1.8 2.6 1993­2002 3.4 4.4 1.1 9.0 10.7 1.7 Bolivia (national) 1997­2002 5.5 3.3 2.2 6.9 5.4 1.5 Brazil 1990­95 3.9 1.9 1.9 8.5 3.7 4.8 1995­2003 0.2 0.4 0.2 0.1 0.9 1.0 1990­2003 3.6 1.3 2.3 8.6 2.6 6.0 Chile 1990­96 1.8 1.3 0.5 7.6 7.3 0.3 1996­2003 0.1 0.2 0.1 1.6 1.4 0.3 1990­2003 1.9 1.6 0.4 9.3 8.4 0.8 Colombia (urban) 1992­2000 5.2 0.1 5.3 7.6 0.9 8.5 Colombia (urban) 2000­4 1.9 3.1 1.1 4.2 11.2 7.0 Costa Rica 1992­97 2.0 0.8 1.2 4.3 3.1 1.2 1997­2003 0.6 0.6 1.2 0.2 1.8 2.0 1992­2003 1.4 1.6 0.2 4.1 5.3 1.2 Dominican Republic 2000­4 1.4 3.6 2.1 7.6 8.5 0.8 Ecuador 1994­98 2.7 1.4 4.2 3.0 3.3 6.3 El Salvador 1991­2003 5.9 5.0 0.9 10.6 8.6 2.0 Honduras 1997­2003 2.3 1.1 1.2 3.6 1.6 2.0 Jamaica 1990­99 21.1 9.2 11.9 25.8 17.5 8.3 1990­2002 7.9 8.0 0.1 14.8 15.3 0.5 Mexico 1992­96 5.0 4.0 0.9 10.5 9.7 0.8 1996­2002 2.6 3.1 0.5 9.3 7.3 2.0 1992­2002 2.4 0.9 1.4 1.1 1.9 0.7 Nicaragua 1993­98 11.6 5.9 5.7 9.4 6.6 2.8 1998­2001 4.6 2.1 2.5 3.9 3.3 0.6 1993­2001 16.1 7.9 8.2 13.3 10.0 3.3 Panama 1995­2002 6.0 0.2 6.2 2.9 0.6 3.4 Paraguay 1997­2002 4.4 6.2 1.8 9.9 10.8 0.9 Peru 1997­2002 1.0 0.0 1.0 0.1 0.0 0.1 Uruguay 1989­98 0.5 0.1 0.7 0.2 1.3 1.5 1998­2003 0.2 0.7 0.9 1.6 3.8 -2.2 1989­2003 0.3 0.3 0.0 1.8 1.8 0.0 Venezuela, R.B. de 1989­95 3.7 1.0 2.7 11.4 3.1 8.3 1995­2000 0.8 3.9 3.1 0.9 7.5 6.6 2000­3 4.5 3.1 1.4 12.3 9.6 2.6 1989­2003 13.2 7.5 5.7 26.0 20.2 5.8 Source: Gasparini, Gutierrez, and Tornarolli (2005). Table 4.1 reports the results of decomposing headcount that if the distribution of relative incomes had remained con- poverty changes for two poverty lines ($1 a day and $2 a day) stant, then the poverty headcount ratio would have increased in 18 Latin American countries. For example, poverty (as by only 4.3 points. The remaining (7.6 points) was driven measured by the $2-a-day poverty line) increased 11.9 points by changes in the shape of the income distribution, which in in Argentina between 1992 and 2004. We estimate, however, the Argentine case, were unequalizing over the 1992­2004 61 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S period. Admittedly, distributional shifts affected poverty in a As an alternative, one can try to summarize the cross- different way before and after 2002. In fact, the income distri- country information using variance decomposition tech- bution deteriorated during the 1992­98 and 1998­2002 niques as in Kraay (2005). If the changes in poverty (P) are periods (and contributed to an increase in poverty), but it expressed as a growth component (Y) and a distributional improved over the 2002­4 period. component (D), then P = Y + D. Then the expression There are other countries where the distribution of income for the variance of the changes in poverty can be written: has also worked against the poor over the long run (taking Variance (P) = Variance(Y) + Variance(D) + 2 × Covari- the long run as the period between the first and last survey ance(Y, D). This expression can now be used to define regardless of the number of years spanned by the spell). One the proportion of poverty changes explained by growth as is República Bolivariana de Venezuela (1989­2003), where Variance(Y) + Covariance(Y, D)/Variance (P). about 6 percentage points of the 26 percent increase in What then are the relative roles played by growth and poverty was attributable to a deterioration of income inequal- changes in relative incomes in the Latin American region? ity. Urban Bolivia also experienced a deterioration in income Well, the results of this exercise suggest that the distribu- inequality over the 1993­2002 period, although it was tional component is likely to be a much more important fac- accompanied by a dramatic decline in poverty (9 percent) as a tor than the global data would suggest. In fact, the share of result of a significant growth component (-11 percent). Sim- variance of changes in poverty (now based on a $1 a day ilarly, poverty declined in Costa Rica (1992­2003) and in poverty line to ensure comparability with Kraay 2005) attrib- Jamaica (1990­2002), but it could have fallen even more if utable to growth would be about 50 percent in both the short income distribution had not changed for the worse. In con- and the long run (figure 4.3).4 Thus these results, if taken at trast, in Honduras (1997­2003) and Ecuador (1994­1998) the deterioration in income distribution was accompanied by FIGURE 4.3 increased poverty. The case of Ecuador is noteworthy because Share of changes in poverty explained by growth and inequality the contribution of the distributional component (6.3 per- Changes in poverty over the short run cent) was enough to tilt the balance from a decline in poverty of 3.3 percent to an increase of 3.0 percent. World Latin America In other countries the distributional component helped to accelerate poverty reduction. For example, had income distribution income remained constant in Brazil over the 1990­2003 period, poverty would have fallen by only 2.6 percent rather than the observed 8.6 points. Other countries where income distribution tended to favor the poor over the long run are Chile, the Dominican Republic, El Salvador, Mexico, Nicaragua, Panama, Paraguay, and Peru. Among this group, the only country where distribu- Changes in poverty over the long run tional changes were relatively important is Panama, which World Latin America experienced a 6 percent decline in poverty, as measured by US$1 a day. Had the distribution of income remained con- stant, poverty would have increased slightly (0.2 percent). These results indicate significant country heterogeneity in the Latin American sample. In some countries, such as Argentina, Ecuador, and Panama, the distributional compo- nent has been very important. In others, such as Bolivia, El Salvador, and Jamaica, the growth component has clearly predominated. In between are cases such as Brazil and Growth component Inequality component Nicaragua, where both components had similar effects. Source: Kraay (2005) and authors' calculations. Given the results of just this single exercise, reaching general Note: Poverty is defined here as living on $1 per day or less. conclusions that apply to most countries seems quite daring. 62 T H E R E L AT I V E R O L E S O F G R O W T H A N D I N E Q U A L I T Y F O R P O V E RT Y R E D U C T I O N face value, would suggest the need to focus on both growth- poverty. Regardless of the poverty line used, the distribu- enhancing and inequality-reducing policies simultaneously. tional component tends to account for a minimum of 25 per- Given the prevailing high inequality levels of the Latin cent of the variation of poverty changes and for as much as American region, our finding may not be surprising.5 50 percent. This is significantly higher than what is found Before jumping to the conclusion that growth and income in the sample of developing countries analyzed in Kraay distribution are equally important in the region, however, (2005) and is probably related to the high inequality levels notice that these results are extremely sensitive to the that prevail in the region. It must be noted, however, that choice of the poverty line used to compute the poverty fig- the choice of poverty lines is important. Typically, in coun- ures. In fact, the relevance of growth for poverty reduction tries with more inclusive poverty lines ($2-a-day or a dramatically increases as one moves from a $1-a-day to a national moderate line), growth appears to weigh more $2-a-day poverty line (that is, as the poverty concept than changes in income distribution; in those countries becomes more inclusive). The relevance of growth also with more selective poverty lines ($1-a-day or a national increases when one shifts from using international poverty extreme line), redistribution appears to play a bigger role lines to using national poverty lines, most likely because in reducing poverty. Reaching different segments of the countries tend to use more generous poverty lines (see fig- population will thus require different policies. ure 4.4, which focuses only on short-run changes). On the whole, the results reported here would under- Growth and inequality: Bringing country score the importance of both growth and changes in the specificity into the picture distribution of income for the evolution of Latin American The variance decomposition approach reviewed in the pre- vious section has highlighted some important elements regarding the relative roles played by growth and the distri- FIGURE 4.4 bution of income for poverty reduction. However, those Share of changes in Latin American poverty explained by growth and inequality results are probably less useful when interest centers on the relative importance of each component at the individual International poverty line country level and on the characteristics that determine that Living on less than US$1 Living on less than US$2 importance. For example, should Chile and Nicaragua-- two countries with similar levels of inequality but dramati- cally different income levels--try to strike a similar balance between growth-promoting and inequality-reducing poli- cies? Similarly, should Uruguay and Brazil--which have similar levels of per capita income but are the least and most unequal countries in the region, respectively--follow simi- lar policies in their attempts to reduce poverty? Or for any particular country, should policy makers implement the National poverty line same type of policies when they focus on the whole universe Extreme poverty Moderate poverty of poor than when they focus on a particular group, say, the poorest among the poor? Is the same strategy likely to have the same effect on everybody under the poverty line? To answer these questions, we have to rely on tools that go beyond statistical decomposition techniques and try to relate observed outcomes to some country characteristics that can be useful in discerning which type of policies might be appropriate in each country. One possible tool is a parametric analysis that approximates the actual distribution Growth component Inequality component of income with a more or less tractable functional form (that is, a mathematical model that can be related to some eco- Source: Authors' calculations. nomic variables to approximate the empirical distribution of 63 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S income). This functional form is then used to assess the role concepts), then this approach loses part of its appeal. The of country-specific conditions for the poverty-reducing second element is the degree to which the chosen parame- effects of growth and distributional change (that is, to see terization fits the data. Even if the selected functional form how changes in country conditions affect the impact on is tractable and provides an excellent theoretical framework poverty of growth and changes in relative incomes pre- to deal with the problem at hand, it could provide a very dicted by the model). To a large extent this is a theoretical poor approximation to the actual data and hence be empir- exercise that can be fully controlled and with which one can ically irrelevant. experiment. For our purposes, there is a functional form that appears Clearly, the usefulness of this approach depends on two to be a natural choice to approximate the size distribution critical elements. The first is the tractability of the used of income: the lognormal distribution. This is probably the approximation. If the selected functional form cannot be most standard approximation of empirical income distribu- related to country characteristics that are easily observable tions in the applied literature and seems to fulfill the two and can be used to discriminate among countries (or poverty criteria required for this approach to be useful (see box 4.2 BOX 4.2 The size distribution of income An abundant literature spanning more than a century-- Gibrat's work was followed by a large literature from Pareto (1897) to Gibrat (1931), Kalecki (1945), extending his basic framework and offering additional Rutherford (1955), Metcalf (1969), Singh and Maddala empirical evidence. Kalecki (1945) extended Gibrat's (1976), and more recently to Bourguignon (2003) and original setup by making negative income changes less Kakwani and Son (2003)--has attempted to approximate likely at low-income levels than at high ones and in that the distribution of income. They have used a variety of way accounted for the fact that the variance of log income functional forms: Beta, Gamma, Pareto, Champernowne, remained relatively constant over time. Rutherford Dagum, Singh-Maddala, displaced lognormal, and lognor- (1955) expanded Gibrat's model to introduce birth and mal. Among these, however, the most commonly used in death considerations. He also presented empirical experi- applied research is the lognormal function. Its use in the ments based on the comparison of theoretical and context of income was pioneered by Gibrat (1931), who observed quantiles of the distribution of income, search- noted that it offered a good empirical fit to the observed ing for a functional form that would improve upon the data and also provided a first theoretical justification based lognormal. The figure below illustrates how a lognormal on a model in which individuals' incomes are subject to distribution might look for different Gini coefficients. random proportionate changes. In his original explanation of why the logarithm of income could behave approxi- The look of the lognormal distribution for different Gini coefficients mately as a lognormal distribution, Gibrat (1931) described three conditions that must be present if the observed dis- 0.5 tribution is to approximate the lognormal form. First, the 0.4 distribution of income at any give time must be derived 0.3 from that of the previous period by assuming that the vari- 0.2 able corresponding to each member of the distribution is 0.1 affected by a small proportionate change. Second, the pro- 0 portions must differ for different members of the distribu- 6 tion. And third, these differences must be determined in a 1.01 1.1 1.33 1.53 1.75 2.01 2.30 2.64 3.03 3.48 3.99 4.58 5.26 6.03 6.92 7.94 9.12 10.5 12.0 random manner from a given frequency distribution. Gini 0.3 Gini 0.4 Moreover, Gibrat observed that whatever the distribution Gini 0.5 Gini 0.6 of income at the initial period, income would approach normality more and more as time passed. Source: López and Servén (2005a). 64 T H E R E L AT I V E R O L E S O F G R O W T H A N D I N E Q U A L I T Y F O R P O V E RT Y R E D U C T I O N for some historical perspective and for some brief back- 1 percent increase in income levels, holding inequality con- ground that can theoretically justify its use in practice). stant) and the partial inequality elasticity of poverty (that Regarding tractability, one of the appeals of the lognor- is, the impact on poverty of a 1 percent deterioration in mal distribution is its simplicity, since it can be written as income inequality, holding income levels constant). a function of mean income and the Gini coefficient. Given Thus, for given values of and G, one can map the per capita GDP and the Gini coefficient of an economy, one impact of growth and changes in inequality into poverty. can picture the probability of an individual having a partic- Moreover, under log normality the partial elasticities ular level of income. This in turn is all that is needed not and G can be shown to depend on just three familiar ele- only for a static assessment of the poverty situation for dif- ments: the level of per capita income, the poverty line, and ferent poverty lines but also for the analysis of how poverty the Gini coefficient (Lopez and Servén 2005a). Table 4.2 evolves when the parameters describing the distribution reports the growth and inequality elasticities of headcount change: poverty that result for various combinations of the Gini Change in Poverty (%) = × Income Growth (%) coefficient and the ratio of per capita income to the + G × Change in Gini (%), poverty line z. Inspection of this table confirms the well-known result where and G are, respectively, the partial growth (see, for example, Ravallion 1997, 2004; Bourguignon elasticity of poverty (that is, the impact on poverty of a 2003) that the growth elasticity is smaller (in absolute BOX 4.3 Total growth elasticities of poverty and the efficiency of growth The total growth elasticity of poverty is commonly Consider, for example, the case of two economies (coun- reported in the development literature as a measure of the tries, states, or regions) that are identical (that is, the poverty efficiency of growth. This is defined as the per- countries have similar values of and G so that differ- centage change in poverty for a given growth rate. For- ences in will result from differences in G and g. mally, denoting this elasticity by , growth by g, and the Assume also that over a given period of time, inequality log of poverty by P, can be expressed as = P/g. Thus a changes in the same fashion in both places but that the higher would indicate more effective poverty-reducing two economies have different growth rates (g1 > g2 > 0). growth. Intuitively poverty reduction performance could It is clear that if G > 0, the total growth elasticity be improved through two routes: by achieving high of the economy with the highest growth rate will be growth rates for a given elasticity; or by achieving a smaller (higher in absolute value). Thus one would be higher value (in absolute value) of for a given growth tempted to interpret this as one state being more pro- rate. growth and more pro-poor, when the only thing that is However, one has to be careful interpreting these fig- different in these economies is the growth rate. Similarly, ures. If one assumes that income follows a lognormal dis- if G < 0 in both economies (that is, inequality is tribution, we can express: falling), the total growth elasticity will be higher in absolute value in the economy with lower growth. Again, (1) P = g + GG. < 0, G > 0 one could be tempted to interpret this as a difference Thus poverty changes will be determined by the growth between the pro-poorness of the growth strategies: one component g and by the distribution component GG. economy experiences faster growth but at the apparent It then follows immediately that the gross growth cost of a lower growth elasticity of poverty whereas the elasticity of poverty can be rewritten as a function of other economy experiences lower growth, but with a the partial growth and inequality elasticities of poverty faster growth elasticity. and of the observed growth and observed changes in These somewhat extreme examples should highlight inequality: = P/g = + G G/g. This expression can the dangers of reading too much into a simple elasticity. now be used to analyze how changes with G and g. 65 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S TABLE 4.2 Growth and inequality elasticity of poverty (headcount index) Growth elasticity Inequality elasticity (Gini coefficient) (Gini coefficient) /z 0.30 0.40 0.50 0.60 /z 0.30 0.40 0.50 0.60 6 -6.05 -3.25 -1.95 -1.22 6 12.34 7.38 5.10 3.89 3 -3.94 -2.18 -1.33 -0.86 3 5.17 3.28 2.42 1.97 2 -2.80 -1.60 -1.01 -0.66 2 2.48 1.70 1.35 1.18 1.5 -2.06 -1.23 -0.80 -0.54 1.5 1.20 0.92 0.81 0.77 1 -1.16 -0.78 -0.55 -0.39 1 0.18 0.24 0.29 0.35 Source: López and Servén (2005a). value) the higher the level of inequality. For example, Similar results are obtained when one examines the way consider the case of a country whose per capita income lev- that income and inequality levels affect the inequality elas- els are about three times the poverty line (the row in table 4.2 ticity of poverty. Under most scenarios, higher inequality corresponding to /z = 3). In this country, if inequality (lower income) also lessens the impact of progressive distri- levels are low (say, a Gini of 0.3), a 1 percent growth rate butional change itself on poverty. As illustrated in would lead to almost a 4 percent decline in poverty. In con- table 4.2, the inequality elasticity falls as inequality rises trast, if inequality is high (say a Gini of 0.6), the same (income declines) for a given value of average income rela- growth rate would lead to a more modest decline in poverty tive to the poverty line (for a given Gini index). Note, how- (about 0.9 percent). Thus, inequality hampers the poverty- ever, that this relationship is highly nonlinear, and its sign reducing effect of growth, as stressed in the literature, and, is reversed at very low levels of development (captured in in highly unequal countries, justifies making a more bal- the table by values of /z close to 1), so that a higher Gini anced income distribution an important policy priority. coefficient is associated with a higher inequality elasticity Clearly, an improvement in the distribution of income has a (see the last line of table 4.2). double poverty-reducing effect. On the one hand, it has Clearly, before proceeding with this type of analysis, we a pure positive redistribution effect. On the other, it have to acknowledge that skeptical readers may question increases (in absolute value) the growth elasticity of whether the selected functional form provides a reasonable poverty and hence makes future growth more effective in approximation to the real world, particularly because the reducing poverty. existing empirical evidence in this regard is quite limited Table 4.2, however, also indicates that poverty itself (as and usually based on individual country studies. measured by low per capita income) is a barrier to poverty To narrow the existing gap between the empirical popu- reduction: for a given Gini coefficient, the growth elasticity larity of the lognormal distribution and the empirical of poverty declines rapidly (in absolute value) as average support for that distribution, Lopez and Servén (2005a) income declines in relation to the poverty line. For exam- compare the empirical distribution quintiles for almost ple, when the Gini is 0.4, for a country with per capita 800 country-year observations with those obtained theoret- income equal to six times the poverty line, the growth elas- ically using the lognormal approximation. They reason that ticity of poverty is about 3.25 percent, whereas for a coun- if the lognormal distribution provides a reasonable approx- try with per capita income equal to the poverty line, it imation, then any differences between the empirical and would be about 0.8 percent. This suggests that economic the theoretical distributions should not be dramatic. In growth also has a double poverty-reducing effect: first, the contrast, if the lognormal distribution provides a poor direct effect of income growth on the average level of approximation, then one would expect to find large differ- income; and second, the indirect effect that arises from the ences between theoretical and empirical distributions. higher average income via the correspondingly higher Figure 4.5 presents the scatter plots of the empirical growth elasticity of poverty. (vertical axis) and theoretical quintiles (horizontal axis) for 66 T H E R E L AT I V E R O L E S O F G R O W T H A N D I N E Q U A L I T Y F O R P O V E RT Y R E D U C T I O N FIGURE 4.5 Empirical and theoretical quintiles a. Full sample b. Income Empirical quintiles Empirical quintiles 0.40 0.40 0.30 0.30 0.20 0.20 0.10 0.10 0 0 0 0.05 0.10 0.15 0.20 0.25 0 0.05 0.10 0.15 0.20 0.25 Theoretical quintiles Theoretical quintiles c. Expenditure d. Gross income Empirical quintiles Empirical quintiles 0.40 0.40 0.30 0.30 0.20 0.20 0.10 0.10 0 0 0 0.05 0.10 0.15 0.20 0.25 0 0.05 0.10 0.15 0.20 0.25 Theoretical quintiles Theoretical quintiles e. Net f. Net income Empirical quintiles Empirical quintiles 0.40 0.40 0.30 0.30 0.20 0.20 0.10 0.10 0 0 0 0.05 0.10 0.15 0.20 0.25 0 0.05 0.10 0.15 0.20 0.25 Theoretical quintiles Theoretical quintiles Source: López and Servén (2005a). a number of samples depending on whether the original hypothesis of lognormality when the test is implemented data are income (net/gross), or consumption. The different on the distribution of per capita income, regardless of panels also present the 45-degree line (where all the obser- whether income is measured in gross terms (before taxes vations should be placed under the null). The figure sug- and transfers) or net terms (after taxes and transfers). gests that the lognormal distribution generally provides a Admittedly, even though the lognormal also seems to reasonable approximation to the actual data. More formally, approximate the consumption data quite well, the same Lopez and Servén (2005a) perform several statistical tests null hypothesis is unambiguously rejected when applied to on the data and find that the data cannot reject the null per capita consumption data (see annex 4A for details). On 67 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S headcount. Curves to the northeast of the graph correspond FIGURE 4.6 to higher levels of the poverty rate. The slope of these Iso-poverty curves for headcount poverty curves depicts the changing trade-off between growth and Gini coefficient redistribution. The steeper the slope, the bigger the decline 0.70 in the Gini coefficient required to keep poverty constant in P0 0.7 0.65 the face of a given decline in the ratio of mean income to P0 0.6 0.60 the poverty line. The curves become increasingly steep, and P0 0.5 0.55 closer to each other, as one moves downward along them. In P0 0.4 other words, the more equal and the poorer the economy (as 0.50 P0 0.3 reflected, respectively, by a lower Gini coefficient and a 0.45 P0 0.2 lower mean income/poverty line ratio), the bigger the 0.40 P0 0.1 change in the Gini coefficient required to offset a given 0.35 change in mean income relative to the poverty line--that 0.30 is, the more effective growth will be relative to redistribu- 0.25 tion in attacking poverty. As the economy becomes richer and more unequal (the northwest segment of the figure), 0.20 6.0 5.5 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 the curves become less steep, and therefore a smaller change Mean income poverty line in the Gini coefficient is now needed to offset a given Source: López and Servén (2005a). change in mean income relative to the poverty line. In other words, distributional change now plays a relatively larger role in poverty changes. the whole, the authors conclude that their results are An alternative analysis would exploit table 4.2 to encouraging for the use of parametric analysis based on the directly simulate the impact of alternative growth scenar- lognormal distribution for the analysis of poverty. ios. These results are reported in table 4.3. The left panel of On the basis of the previous discussion, we now perform the table reports the poverty impact of 1 percent growth two different exercises to illustrate how the parametric with no associated changes in inequality, whereas the right approach can be used to help gauge the relative priority of panel simulates the impact of 2 percent growth with an pro-growth and pro-redistribution policies when their associated increase in inequality of 1 percent. common objective is poverty reduction. First, consider fig- The shaded (no-shaded) cells in the right panel indicate ure 4.6, which plots a set of isometric poverty curves drawn that the poverty outcome of that panel is superior (inferior) under the hypothesis of lognormality for different values of to the poverty outcome in the left panel. The simulations the poverty headcount P0. Each of these curves depicts presented here clearly indicate that different countries may combinations of Gini coefficients and mean per capita require different types of policies. The scenario with higher income/poverty line ratios that yield a constant poverty growth and an associated increase in inequality tends to be TABLE 4.3 Impact on poverty of different growth scenarios Panel A. Neutral growth Panel B. Growth with inequality (Gini coefficient) (Gini coefficient) /z 0.30 0.40 0.50 0.60 /z 0.30 0.40 0.50 0.60 6 -6.05 -3.25 -1.95 -1.22 6 0.24 0.88 1.20 1.45 3 -3.94 -2.18 -1.33 -0.86 3 -2.71 -1.08 -0.24 0.25 2 -2.80 -1.60 -1.01 -0.66 2 -3.12 -1.50 -0.67 -0.14 1.5 -2.06 -1.23 -0.80 -0.54 1.5 -2.92 -1.54 -0.79 -0.31 1 -1.16 -0.78 -0.55 -0.39 1 -2.14 -1.32 -0.81 -0.43 Source: Authors' calculations. 68 T H E R E L AT I V E R O L E S O F G R O W T H A N D I N E Q U A L I T Y F O R P O V E RT Y R E D U C T I O N which report Gini indexes close to but still above the inter- FIGURE 4.7 national norm. Mapping Latin American countries in the income inequality space To what extent is it possible to create a typology of coun- Per capita income/poverty line tries for the Latin American region, based on their growth 6 and inequality-reducing priorities for reducing poverty? ARG Given the difficulties of clustering countries in a two- 5 TTO CHL dimensional space, we first reduce figure 4.7 to a single 4 URY CRI dimension by computing the growth rate that each of these MEX BRA COL countries would need to achieve to compensate for a 1 per- 3 BLZ DOM PAN VEN PER cent increase in the Gini coefficient and leave poverty PRY 2 LCA JAM SLV NIC unchanged (this statistic could be considered the marginal GUY ECU GTM rate of substitution between growth and changes in inequal- 1 HND BOL ity). A higher estimate for this compensatory growth rate 0 would indicate that inequality changes are very relevant for 0.35 0.40 0.45 0.50 0.55 0.60 poverty reduction in the country in question (given an Gini index increase in inequality, poverty will decline only when Source: Authors' calculations. growth is very high). In contrast, a low value for this com- pensatory growth rate would indicate the relevance of superior in poorer and more equal countries. In contrast, in growth (growth even if accompanied by a deterioration of richer and more unequal countries, policies that stimulate income distribution may lead to lower poverty). Note that lower growth with no associated deterioration in income the inverse of this statistic can also be interpreted as the would be a superior alternative. Moreover, as the unshaded maximum deterioration in the income distribution that portion of the right panel shows, the increase in inequality could occur for poverty to decline when growth is 1 percent. under this alternative scenario tends to dominate the Table 4.4 reports these statistics. The table indicates growth effect, and in several rich or highly unequal coun- that in a country such as Argentina, a 1 percent deteriora- tries, the final impact suggests an increase in poverty. tion in the Gini coefficient would require a compensatory Hence richer and very unequal countries will have to pay growth rate of 2.5 percent to maintain poverty at a con- significant attention to distributional concerns. stant level. Similarly, in Brazil, Chile, Colombia, Costa Figure 4.7 illustrates how the previous discussion can be Rica, and Mexico, growth would have to be above 2 percent used to highlight country policy priorities (whether these are growth-enhancing or inequality-reducing policies) on TABLE 4.4 the basis of different initial conditions. In this regard, it is Growth rates needed to compensate for a 1 percent increase useful to start mapping the Latin American countries into in inequality (percent) an income-inequality space comparable to the one used in tables 4.2 and 4.3.6 Given that this is a static exercise, we Compensatory Compensatory Country growth rate Country growth rate expand the sample of 18 countries in table 4.1 to add 5 additional countries (Belize, Guatemala, Guyana, St. Lucia, Argentina 2.5 Peru 1.6 and Trinidad and Tobago) for which we have at least one Chile 2.4 St. Lucia 1.5 measure of income distribution.7 Brazil 2.3 Guatemala 1.5 Mexico 2.1 Paraguay 1.5 As expected, this mapping shows a clustering of coun- Costa Rica 2.1 El Salvador 1.4 tries toward the high-inequality side of the figure (Gini Colombia 2.1 Venezuela, 1.2 Trinidad and Tobago 2.0 R.B. de larger than 0.5). This clustering is even more marked for Dominican Republic 1.9 Ecuador 1.1 the lower-income countries.8 The only countries that Panama 1.9 Nicaragua 1.1 Belize 1.8 Guyana 1.1 appear to depart from this norm of high-inequality levels Uruguay 1.8 Bolivia 1.0 are Uruguay and República Bolivariana de Venezuela and Jamaica 1.7 Honduras 0.8 three of the newly added countries (all three in the Caribbean: Guyana, St. Lucia, and Trinidad and Tobago), Source: Authors' calculations. 69 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S to compensate for a hypothetical deterioration in the with a given level of average per capita income. For exam- income distribution. Note that these countries are all ple, as noted in chapter 2, it is standard for countries to rely located in the northeast portion of figure 4.7 (that is, they on poverty figures computed according to at least two are all relatively rich and unequal). Note also that although poverty lines: a higher poverty line that measures moderate Brazil is more unequal than either Argentina or Chile, it poverty, and a lower poverty line that measures extreme would need a lower growth rate to compensate for a 1 per- poverty (the international counterparts of these concepts cent increase in the Gini index. In all these countries, could be the $2-a-day and $1-a-day purchasing power par- growth strategies that are accompanied by increases in ity poverty lines). inequality would probably lead to disappointing results on Our analysis can be twisted to explore how the appropri- the poverty front unless the deterioration in inequality is ate focus of the development strategy of any given country extremely modest or the growth rate very high. varies with the concept of poverty used. Given per capita At the other extreme of the table are Honduras and income levels, low poverty lines will result in a high mean Bolivia, where growth of 0.8 percent and 1 percent, respec- income/poverty line ratio (that is, low poverty lines will tively, would be enough to compensate for a 1 percent move a country toward the top of tables 4.2 and 4.3 and fig- deterioration in income inequality. Ecuador, Guyana, ure 4.7). Thus the analysis above of the relevance of growth and Nicaragua are close behind, each needing an esti- and distribution in relatively richer countries would apply mated compensatory growth rate of 1.1 percent. These low here. In contrast, a high poverty line will result in a low growth rates should highlight the importance of growth mean income/poverty line ratio (that is, a high poverty line for poverty reduction in these countries, where (political will push a country toward the bottom of tables 4.2 and 4.3 economy issues apart) poverty reduction seems to be and figure 4.7). Hence as the poverty line increases, the rel- mainly driven by growth, and where growth even if accom- ative importance of growth for reducing poverty goes up as panied by moderate increases in inequality will succeed in well, and other things equal, offers a rationale for shifting reducing poverty. poverty reduction priorities toward growth-oriented poli- Between the two extremes is a continuum of values cies and against redistributive policies. without apparent jumps, something that would indicate In essence, two main messages emerge from this analy- that there may not be well-defined clusters of countries sis. First, in any given country, the elements that underlie a with between-group differences and within-group similari- poverty reduction strategy should be highly dependent on ties. In any case, Belize, the Dominican Republic, Panama, the definition of poverty used. Given that national poverty Trinidad and Tobago, and Uruguay seem to be closer to the definitions deviate notably from the international norm group led by Argentina where reducing inequality is quite across countries, this analysis means that two countries that important for poverty reduction, whereas El Salvador, rely on different poverty lines but that are otherwise identi- Guatemala, Paraguay, Peru, St. Lucia, and República Boli- cal are justified in implementing different poverty reduc- variano de Venezuela seem closer to the group of countries tion strategies. Second, and probably more relevant for where growth appears as the main priority for poverty policy purposes, reaching different groups of poor people reduction. requires different sets of interventions that recognize their One final issue we address in this section regards the idiosyncrasies. In particular, this analysis indicates that the interpretation given to the ratio of mean income to the extreme poor (those below a relatively low poverty line) poverty line. So far we have implicitly viewed alternative probably require targeted interventions, whereas the mod- values of the mean income/poverty line ratio as reflecting erate poor (those below a relatively higher poverty line) different levels of average per capita income with a given require broader interventions that aim at raising incomes poverty line. This is probably the natural interpretation for all individuals in society. when comparing the impact of growth and income distrib- ution on poverty reduction across the different Latin Amer- Concluding remarks ican countries. This chapter started by posing several questions related to However, this ratio could also be interpreted the other the elements that should be at the center of any sensible way around, namely, as reflecting alternative poverty lines poverty-reducing strategy. Should such a strategy have a 70 T H E R E L AT I V E R O L E S O F G R O W T H A N D I N E Q U A L I T Y F O R P O V E RT Y R E D U C T I O N growth bias or instead concentrate mainly on reducing Aitchison and Brown (1966, ch. 11) show that lognormal- income inequality? Does a country's level of development ity implies matter for the chosen poverty reduction strategy? Which -1 1 + G strategy is better for poverty reduction: a high growth rate (4A.1) = 2 , 2 that has an associated increase in inequality, or a lower and growth rate that maintains inequality constant? Are there -1 any conditions under which policy makers can accept a (4A.2) L(p) = ( (p) - ), trade-off between growth and a deterioration in the distri- where (.) denotes the cumulative normal distribution. bution of income? Hence a change in the Gini coefficient, and thus in , must We find the answers to these questions depend on the be reflected in a matching change in the Lorenz curve. initial conditions in the individual country and on its con- On a cross-country basis, what is usually available to the cept of poverty. In countries with low per capita income researcher is some summary information on the shape of levels and relatively equal distribution, growth in mean the Lorenz curve. One such summary is provided by the income will be relatively more effective in reducing poverty income shares of the different quintiles of the population: than changes in the income distribution. In contrast, richer and more unequal countries will have to carefully balance (4A.3) Q20j L(0.2j) - L(0.2( j - 1)) for j 1,2,3,4. the growth and income distribution objectives, because in those cases even small increases in inequality may have a Given the one-to-one mapping between the Gini coeffi- dramatic negative impact on poverty. cient and the Lorenz curve that follows from equations As for the relevance of the concept of poverty that each 4A.1 and 4A.2, under lognormality there must also be a country uses, the chapter has argued that different poverty one-to-one mapping between the Gini coefficient and the concepts may require different strategies. In any given quintile shares (equation 4A.3). Thus, a test of the null country, if poverty is defined in a very inclusive way (that hypothesis of lognormality can be based on the comparison is, if a country relies on a very high poverty line where most of the empirical quintiles, say E20j, with their Gini-based of the population qualifies as poor), then strategies that rely theoretical counterparts Q20j. Following this approach, a on growth will be more appropriate for poverty reduction formal lognormality test can be performed on the basis of than strategies that stress redistribution. As the concept of the regression model: poverty becomes more restrictive (that is, as the poverty (4A.4) E20j = it + Q20j it j,it line declines and fewer people qualify as poor), the rele- vance of redistribution as a tool for poverty reduction rises where j = 1,2,3,4 denotes the income quintile; i = 1,2, . . ., N and the relevance of growth declines. is a country index, and t = 1,2, . . . Tidenotes the date of each On the whole, the main message that emerges from our income (or expenditure) survey available for country i. In analysis is that given the high income inequality levels pre- general Ti will differ across countries, resulting in an unbal- vailing in Latin America, it would seem appropriate to anced sample. In equation 4A.4, the theoretical quintiles focus on both growth and income distribution, although Qit20j are constructed on the basis of the observed Gini coeffi- the ideal balance between the two will differ from country cients Git, as implied by equations 4A.1­4A.3: to country. (4A.5) Q2 = it 1 1 + Git (0.2j) 2 -1 Annex 4A 0j 2 1 1 1 + Git Testing for lognormality of income (0.2(j 1)) 2 2 . To test the lognormality hypothesis of income, Lopez and Servén (2005a) exploit the one-to-one mapping that arises Testing for lognormality in model 4A.4 is equivalent to under lognormality between the Gini coefficient and the testing the joint null hypothesis: = 0; = 1. Lorenz curve L(p). Letting G and respectively denote the What are the results of formally testing that hypothesis? Gini coefficient and the standard deviation of log income, The table below presents the results of the estimation of 71 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S Annex table: Nested error component model-based lognormality tests Observed quintile All Income Expenditure Gross Net Net income 0.980 1.007 0.894* 1.009 0.960* 1.009 (0.015) (0.016) (0.012) (0.023) (0.016) (0.017) 0.002 0.001 0.013** 0.001 0.005** 0.001 (0.002) (0.002) (0.002) (0.003) (0.002) (0.002) Number of observations 3176 2420 756 1472 1484 892 Number of countries 130 98 65 75 97 55 0.0100 0.0124 0.0073 0.0141 0.0259 0.0086 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0027 0.0034 0.0021 0.0052 0.0019 0.0019 Hoa = 0; = 1 0.410 0.903 0.000 0.920 0.048 0.800 = 0 0.041 0.498 0.496 0.080 0.031 0.074 = 0 0.000 0.035 0.077 0.078 0.000 0.010 Source: Authors' calculations. Note: Robust standard errors are reported in parentheses. a. p-values are reported. *Ho: = 1 rejected at the 5 percent level. **Ho: = 1 rejected at the 5 percent level. model 4A.4 with the following nested structured for the degrees of freedom. As would be expected in light of the error term: j = it t it point estimates, the null can be rejected at the 5 percent i + i + j . The first thing that stands out in this table is that the level in the two samples in which expenditure-based obser- regression slopes and intercepts are very close to their vations represent a sizable share of the total number of data expected values under the null of 1 and 0, respectively. points. In contrast, the samples containing only income- Note that in the samples including expenditure observa- based observations show little evidence against the null-- tions (the first, third, and fifth columns), the estimated the p-values range from 0.41 to 0.92. In the full sample, slopes are slightly below 1, while they are slightly above 1 in which expenditure-based observations represent about in the regressions including only income-based observa- 20 percent of the total, we also fail to reject the null, with a tions. From a statistical perspective, we can formally reject p-value of 0.41. the null of unit slope in the expenditure and net subsam- Notes ples (third and fifth columns). In turn, the estimated 1. The exception is when per capita income levels are below the intercepts are positive in the samples including expenditure- poverty line, in which case progressive distributional change leads to based observations and negative in those including only increasing poverty. income-based observations. As with the slopes, in the 2. Admittedly, World Bank (2005e) presents evidence for expenditure and net subsamples we can reject the null of 14 countries suggesting a strong positive correlation between growth and changes in inequality during the 1990s. In particular, a 1 percent zero intercept. The bottom panel of the table reports Wald growth rate is associated with a 0.5 percent increase in the Gini coef- tests of the null hypothesis of lognormality. Under the null, ficient. The fact that growth and changes in inequality do not appear the test statistic follows a chi-square distribution with two to be correlated does not mean that inequality will not increase in a 72 T H E R E L AT I V E R O L E S O F G R O W T H A N D I N E Q U A L I T Y F O R P O V E RT Y R E D U C T I O N particular growth episode. It just means that having information on a 5. According to de Ferranti & others (2004), the only other country's growth rate does not add much to infer the possible change region that has inequality levels comparable to those observed in in inequality. Latin America is Sub-Saharan Africa. 3. This, of course, need not always be the case, since many poli- 6. The mean income/poverty line figures have been computed cies are likely to be both growth promoting and equality enhancing. using GDP per capita valued in 2000 constant US dollars PPP. The But some empirical evidence suggests that not all policies have this ratios roughly correspond to a poverty line of $2 a day in 2000 US feature (Barro 2000; Lundberg and Squire 2003; Lopez 2004), and dollars. some may force policy makers to face a trade-off between faster 7. Admittedly, the Gini coefficients for Belize, Guyana, St. Lucia, growth and increasing income inequality. and Trinidad and Tobago are more than 10 years old. 4. The short-run results are based on all possible episodes in a 8. Interestingly, there seems to be a negative correlation between country; the long-run results consider only the first and last surveys levels of income and levels of inequality. The correlation between per for each country. In countries with only two surveys, the short- and capita income/poverty line and the Gini coefficient for the 23 coun- long-run coincide. tries in the sample is -0.36 and significantly different from 0. 73 CHAPTER 5 Pro-Poor Growth in Latin America There is no doubt that growth must be at the center of any successful poverty reduction strategy. However, are all pro- growth policies equally pro-poor? Is it possible that some policies lead to higher growth but leave poverty unchanged or, even worse, lead to higher poverty? Similarly, does the composition of growth matter, or can all sectors be considered equally pro- poor? Finally, what is the role of taxes and transfers in this context? Should policy makers focus only on improving the dis- tribution of market incomes along with the growth process, or do they have to complement these actions with tax and transfer interventions that directly target disposable income? C HAPTER 4 ARGUED THAT FAST POVERTY policies are associated with higher income inequality. This reduction in the region would require the potential trade-off may in turn result in development strate- implementation of development strategies gies that may lead, on the one hand, to faster growth but, on that aim at simultaneously achieving fast the other hand, to no change in poverty or perhaps to even sustained growth rates and more equal soci- higher levels of poverty. Thus, if the objective is to reduce eties. This general advice, however, leaves unanswered sev- poverty, policies will have to be considered according to eral questions of critical interest for policy makers: are all their potential impact on both growth and inequality. pro-growth policies equally pro-poor? Is it possible that The chapter then adopts a sectoral perspective and some policies lead to higher growth but leave poverty focuses on whether growth in different sectors of economic unchanged or, even worse, lead to higher poverty? Will activity influences poverty in different ways. As discussed policy makers face a trade-off between faster growth and in Beyond the City: The Rural Contribution to Development higher inequality? Similarly, does the composition of (de Ferranti and others 2005), differences in labor intensities growth matter, or can all sectors be considered equally pro- in the location of economic activities or in sector-related poor? If the composition of growth does matter, should spillovers can result in growth in different sectors having policy makers aim at biasing growth toward some particu- different effects on poverty. To anticipate some of the empir- lar sectors? Finally, what role do taxes and transfers play in ical findings of this chapter, we find that, indeed, the sectoral this context? composition of growth matters for poverty reduction. This chapter explores these issues in three complemen- Finally, we also review the extent to which policies aimed tary ways. It addresses them first from a policy perspective at improving the distribution of market incomes (defined as and reviews what is known about the effect on inequality of the distribution of income among households determined a number of growth-enhancing policies. In many circum- by market rewards to the private assets and efforts of indi- stances the positive impact that a policy has on growth will viduals before government intervention) need to be comple- be reinforced by its positive impact on the distribution of mented with tax and transfer interventions that directly income. But it is also plausible that some pro-growth target disposable incomes (defined as the distribution of 75 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S income after taxes have been levied and transfers have been growth policy that has an associated increase in inequality paid). Disposable incomes, after all is said and done, are the will affect poverty. For example, if a policy or policy package relevant distribution to consider in poverty reduction strate- leads to a significant acceleration of growth of, say, 2 per- gies. The need to resort to taxes and transfers as a poverty cent, and simultaneously to a very slight deterioration in the reduction tool will depend largely on whether the distribu- distribution of income, one could possibly expect poverty to tion of disposable income is mainly driven by changes in the decline and hence consider the policy package an acceptable distribution of market incomes or alternatively by govern- alternative even if it leads to a higher dispersion of incomes. ment interventions using the tax-and-transfer instrument. In contrast, if a policy package leads only to modest growth but increases inequality substantially, then one would have Are all pro-growth policies equally pro-poor? to be wary of a potential increase in poverty associated with If policies could be easily categorized as growth enhancers that package. Moreover, since growth and changes in or inequality reducers, then policy makers could target a inequality affect poverty in different ways from country to growth-inequality objective by selecting a set of policies country, depending on initial incomes and inequality levels, expected to promote high growth and a second set aimed at then similar pro-growth policies can be expected to have reducing inequality. In practice, however, things are likely different poverty effects in different countries. to be more complex not only because of the inherent diffi- These problems are further complicated by the dynam- culties of selecting appropriate policies tailored to an indi- ics and time lags involved in the adjustment processes of vidual country's specific situation but also because in most income levels and income inequality following the imple- cases policies are likely to affect growth and inequality mentation of a particular policy. Those lags may generate simultaneously and in some circumstances even produce intertemporal poverty dynamics. Consider a pro-growth conflicting outcomes. Figure 5.1 illustrates this point with policy package that has a negative impact on inequality. If a simple representation of the links between policies and the growth and inequality effects become apparent at sub- poverty reduction. It shows that a policy's effect on poverty stantially different times, then the policy intervention may reduction depends not only on its effect on income growth increase poverty in the short run and decrease it in the long and the way that growth translates into poverty reduction, run. This would be the case if the inequality effect of the but also on the policy's simultaneous effect on income policy is felt immediately but the growth effect is not felt inequality and the way inequality changes are translated for some time. This section explores these issues. into poverty reduction. From the discussion in chapter 4, it should be clear that The simultaneous impact of policies policies that contribute to faster growth and lower inequal- on growth and inequality ity will reduce poverty. However, it is far less clear how a The past few years have witnessed an explosion of works ana- lyzing the way different policies affect growth. According to Durlauf and Quah (1999), the number of determinants of FIGURE 5.1 growth considered in the literature is greater than the num- Policies, growth, distributional change, and poverty reduction ber of countries in the standard growth data set, and a review of all these determinants is outside the scope of this report. Income growth Instead, table 5.1 presents a partial survey of policy areas where progress is typically considered as pro-growth, the indicators typically used to assess progress, and some of the Policy Change in empirical works that have analyzed its relevance. For exam- reform poverty ple, the existing literature largely supports the idea that countries tend to grow faster when they have a higher capital stock, a more-developed financial sector, better institutions, Changes in income distribution more trade openness, smaller governments, better public infrastructure, and good macroeconomic management. Two disclaimers need to be made here. The first regards Source: Authors. the unanimity of these results: in almost all of the areas 76 P R O - P O O R G R O W T H I N L AT I N A M E R I C A TABLE 5.1 Economic policies and growth: Review of the evidence Policy area Indicator category Econometric results I. Structural policies and institutions Education Enrollment rates, years of education [+]: Barro (1991, 2001); Mankiw, Romer, and Weil (1992); Loayza, Fajnzylber, and Calderón (2005) Quality of education [+]: Barro and Lee (2001) Allocation of talents [+]: Murphy, Shleifer, and Vishny (1991) R&D investment [+]: Coe and Helpman (1995) Financial development Private domestic credit (% GDP) [+]: Levine, Loayza, and Beck (2000); Loayza, Fajnzylber, and Calderón (2005) Liquid liabilities (% GDP) [+] via total factor productivity growth: Beck, Levine, and Loayza (2000) [+] only for countries with well-developed financial systems: Rioja and Valev (2004). Government burden Distortionary taxation [-]: Kneller, Bleaney, and Gemmell (1999) for OECD, Gupta and others (2005) for developed countries Corporate taxes [-]: Lee and Gordon (2005) Labor income tax, marginal tax rates [0]: Lee and Gordon (2005) Government consumption [-]: Loayza, Fajnzylber, and Calderón (2005) Infrastructure Infrastructure stocks [+]: Sanchez-Robles (1998); Bougheas. Demetriades, and Mamuneas (2000); Easterly (2001); Esfahani and Ramírez (2003); Calderón and Servén (2004) Infrastructure quality [+]: Calderón and Servén (2004) Governance Institutional quality (Business [+]: Knack and Keefer (1995) Environment Risk Intelligence; International Country Risk Guide) Absence of corruption [+]: Mauro (1995) Kauffman et al. indicators [+]: Dollar and Kraay (2003); Acemoglu, Johnson, and Robinson (2001, 2002); Hall and Jones (1999) Trade openness Exports and imports (% GDP) [+]: Ben-David (1993); Edwards (1998); Dollar and Kraay (2003) Index of outward orientation / [+]: Dollar (1992); Sachs and Warner (1995); Wacziarg and openness Welch (2003) Openness adjusted by geography [+]: Frankel and Romer (1999); Loayza, Fajnzylber, and Calderón (2005) II. Stabilization policies Macroeconomic CPI inflation rate [-]: Fischer (1993); Loayza, Fajnzylber, and Calderón (2005) stabilization [-] for high-inflation periods: Bruno and Easterly (1998); Fischer, Sahay, and Végh (2002) External imbalances Real exchange rate overvaluation [-]: Dollar (1992); Easterly (2001); Loayza, Fajnzylber, and Calderón (2005) [-] and larger impact the higher the overvaluation: Collins and Razin (1999); Aguirre and Calderón (2005) Financial turmoil Systemic Banking Crises [-]: Kaminsky and Reinhart (1999); Dell'Arriccia, Detragiache, and Rajan (2005); Loayza, Fajnzylber, and Calderón (2005) Source: Authors. Note: [+] implies a positive and significant relationship between growth and the corresponding economic policy. [-] reflects a nega- tive and significant relationship, and [0] denotes no statistical relationship between the variables. 77 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S included in table 5.1, at least one work raises serious perpetuate wealth inequalities in the presence of indivisible doubts about the robustness of the results. A classic exam- investments. Poor entrepreneurs--having no collateral, ple usually cited is the work by Levine and Renelt (1992), credit history, or connections--are especially affected by which examines whether the conclusions from existing asymmetries of information, transaction costs, and contract growth studies are robust to small changes in the condi- enforcements costs, as well as other imperfections in the tioning information set. They conclude that almost all capital markets. These capital market imperfections may results are indeed quite fragile (the exceptions are the hinder the allocation of capital to poor entrepreneurs with investment rate, the ratio of international trade to GDP, high-return projects (they may, for example, postpone and the initial level of income of the country in question). investment in human capital) and further increase inequal- The second disclaimer is that table 5.1 should not be ities (Banerjee and Newman 1994; Galor and Zeira 1993). construed as implying that countries trying to achieve fast, In this case, financial development would reduce poverty sustained growth should aim at making progress in each not only through higher growth--by improving the alloca- and all of these areas simultaneously. In fact, in World tion of capital--but also through a more egalitarian distri- Bank (2005c), it is argued that while sustained growth bution of income--by relaxing market imperfections and depends on key elements that need to be fulfilled over granting the poor access to credit markets. These effects time--such as the accumulation of human and physical appear to play a critical role in explaining the results in capital, the efficient allocation of resources in the economy, chapter 6 regarding the negative impact of poverty on the adoption of technology, and the sharing of the benefits growth. of growth--the importance of each of these elements However, it is also possible to argue that financial devel- depends on the particular country and particular period. opment may worsen income inequality (at least in the ini- That is, countries should probably aim at making progress tial stages of economic development). The development of in the areas that are more relevant to their specific context domestic financial intermediaries may benefit primarily the and initial conditions. Progress in areas that do not have rich since poorer sectors of the economy rely mostly on much relevance for the particular country and period may informal banking and family connections to finance their lead to disappointing results. projects. For example, Greenwood and Jovanovic (1990) The literature is far less unanimous on how progress in have argued that the relationship between financial devel- the pro-growth areas listed in table 5.1 is expected to affect opment and income inequality varies according to the stage income inequality. As table 5.2 suggests, there is some con- of economic development. At earlier stages of develop- sensus in some areas. For example, progress on the educa- ment, financial development may increase inequality since tion, governance, infrastructure, and macroeconomic only rich people have access to the financial sector. Such stability fronts is typically associated with declines in access requires an initial set-up cost that poor households income inequality (see also de Ferranti and others 2004). In cannot afford. As financial intermediaries develop, growth other words, policies supporting progress in those areas and savings increase, and the inequalities rise. At later could be considered win-win policies where the inequality stages, the proportion of people that have access and can impact reinforces the growth impact of the policies. profit from financial development increases. The distribu- However, in at least three other areas the findings are tion of income across agents stabilizes, and growth con- more mixed and subject to some controversy. These regard verges to a higher level than the initial one. the roles played by the financial sector, international trade, What does the empirical evidence suggest on this front? and the size of the government in determining income Unfortunately a quick review of table 5.2 indicates that the inequality. We now pause to review in more detail what is empirical evidence is also mixed. On the one hand, Beck, known about the way progress in these three areas affects Demirguc-Kunt, and Levine (2004) evaluate the relation- income distribution. ship between financial development, inequality, and poverty using a cross-section of countries and find that Financial development financial development raises the growth rate of income of Theoretically, the effect of financial development on the poor more than proportionately, thus exerting an inequality and poverty remains ambiguous. Theoretical impact beyond the effect of financial development on aver- models consider that financial market imperfections can age income growth--that is, approximately half of the 78 P R O - P O O R G R O W T H I N L AT I N A M E R I C A TABLE 5.2 Economic policies and income inequality: Review of the evidence Policy area Indicator category Evidence I. Structural policies and institutions Education Education levels [-] for schooling levels and [+] for schooling inequality: Educational inequality Adelman and Morris (1973); Ahluwalia (1976); De Gregorio and Lee (2002) Financial development Private domestic credit (% GDP) [-]: Beck, Demirguc-Kunt, and Levine (2004); Li, Squire, and Zou (1998) [-] by reducing child labor: Dehejia and Gatti (2005) [+] Bonfiglioli (2004) [+]: Bourguignon (2001) Stock market liberalization [0/+] in countries with larger nonagricultural sectors: Clarke, Xu, and Zou (2003) [+]: Das and Mohapatra (2003) Government burden Public employment [-]: Milanovic (2000) Transfers (% GDP) [-]: Milanovic (2000) Targeted spending [-]: Kakwani and Pernia (2000); Iradian (2005) Progressive tax sytems [-]: Iradian (2005) Government consumption [-]: Li and Zou (2002) [+]: Dollar and Kraay (2002) [0]: Kraay (2005) Infrastructure Infrastructure stocks [-]: Estache and Fay (1995); Gannon and Liu (1997); Smith and others (2001); Leipziger and others (2003); Galiani, Gertler, and Schargrodsky (2005) Infrastructure quality [-]: Calderón and Servén (2004) Governance Institutional quality (Business [+] at earlier stages and [-] at later stages of development: Enviromental Risk Intelligence; Chong and Calderón (2000); Li, Xu, and Zou (2000) International Country Risk Guide) Trade openness Exports and imports (% GDP) [+]: Barro (2000), Lundberg and Squire (2003) [+] in countries with abundant skilled labor: Spilimbergo, Londoño, and Székely (1999) [0]: Dollar and Kraay (2002, 2004) Tariffs [0]: Edwards (1997); Milanovic and Squire (2005) [+]: Milanovic (2005) Trade liberalization [+]: Morley (2000); [+] on wage differentials: Behrman, Birdsall, and Székely (2003) II. Stabilization policies Macroeconomic CPI inflation rate [+] and more detrimental for countries with high or stabilization hyperinflation: Easterly and Fischer (2001), Bulir (2001), Li and Zou (2002) Financial turmoil Systemic banking crises [+]: Baldacci. De Mello, and Inchauste Comboni (2002); Honohan (2004) Source: Authors. Note: [+] implies a positive and significant relationship between inequality and the corresponding economic policy, [-] reflects a negative and significant relationship, and [0] denotes no statistical relationship between the variables. overall impact of financial development on the growth rate This positive influence of financial development on of income of the poor is not explained by the impact of inequality and poverty at the aggregate level is consistent financial development on average growth. Not only are with country-case studies that show persistent poverty lev- their estimates significant but they also suggest a large eco- els among households that lack access to credit markets. nomic impact. Jacoby (1994) and Jacoby and Skoufias (1997) find that in 79 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S the presence of adverse shocks, households in Peru and sector. The accompanying Stolper-Samuelson theorem India tend to reduce human capital investments in their predicts that this change in product prices will be trans- children. Similarly, Dehejia and Gatti (2005) indicate that lated into an increase in the wages of workers in the skill- child labor rates are higher in countries with underdevel- intensive sector of the economy. Liberalization should then oped financial systems.1 Specifically, they find that child reduce wage differentials if product market changes shift labor is inversely related to financial development and is production toward a country's comparative advantage, particularly sizable among low-income countries. which within the assumptions of the classical framework A second strand of the empirical literature argues that would seem to benefit less-schooled workers relative to since the less-favored sectors of the population hold only a more-schooled workers in most developing countries. small fraction of the country's assets, financial development On the other hand, a number of possible countereffects may not affect income inequality and poverty. In general, a could result in higher wage dispersion. For example, the disproportionate concentration of financial institutions and preliberalization framework might have protected unskilled services in the main metropolitan areas of a country, more workers who find themselves unemployed following the specifically in its capital, is observed in many Latin American implementation of the liberalization agenda. Capital goods countries. This fact may lead some to think that the link may become cheaper, allowing entrepreneurs to substitute between poverty and access to credit at the regional level capital for labor. Moreover, since workers with more school- may be different from the evidence obtained from cross- ing tend to complement physical capital, the demand for country studies. Even from aggregate results, there is evi- skills could increase and eventually lead to skill-biased tech- dence that the impact of financial development on poverty nological change. For example, de Ferranti and others may be different across activities or regional groups. An (2003) argue that the observed increases in the wage of interesting aggregate result from Clarke, Xu, and Zou skilled workers in Latin America were probably transmitted (2003) claims that financial development may reduce income through trade, foreign direct investment, and licensing inequality, with the impact being larger (in absolute value) if from the United States and other OECD countries. financial development guarantees access to people working Thus it is possible to find sensible theoretical arguments in agriculture. They argue that giving access to credit to the suggesting that inequality can move in one or the other poorest of the poor--typically poor people in rural areas-- direction with trade opening. So, what does the empirical will improve the distribution of income and reduce poverty. evidence say in this regard? Once again, the empirical evi- Bonfiglioli (2004), argues that financial development dence is quite segmented. In one of the first studies at the may affect inequality in different ways. First, it improves aggregate level for developing economies, Edwards (1997) risk sharing, thereby reducing income volatility for a given evaluates whether income inequalities are higher in open size of the risky sector. Second, it raises the share of popula- economies and whether trade liberalization leads to a less tion that is exposed to earnings risk. The first effect tends egalitarian distribution of income. Using data on tariffs to reduce inequality, while the second boosts it. When and nontariff barriers, he finds that inequality is higher in Bonfiglioli empirically validates the model, she finds a result countries with more distortions in their external sector and in line with the Greenwood and Jovanovic (1990) predic- that trade reforms do not appear to have a significant tions. Inequality rises with the level of financial develop- impact on the distribution of income. Similarly, Dollar and ment until it reaches a certain level and then it declines. Kraay (2003) find no evidence that trade affects inequality. A different picture emerges from Milanovic and Squire Openness to international trade (2005), who provide a critical review on the issues of Trade liberalization and openness to trade are usually whether trade liberalization increases wage inequality and viewed as key elements of successful growth strategies. from Lundberg and Squire (2003) and Barro (2000) who However, trade policy may induce countervailing forces on estimate the impact of trade on the Gini coefficient. Most income distribution and poverty alleviation. On one hand, of the studies in this strand of the literature find that trade in a two-sector economy with different skill intensities, reforms have a negative, although modest, effect on the dis- the Heckscher-Ohlin model of international trade predicts tribution of income. Milanovic and Squire also examine the that trade reform in a skill-abundant country will increase effects of tariff reductions on inequality among occupations the relative price of goods produced in the skill-intensive and find that a 1 point decrease in the average tariff rate is 80 P R O - P O O R G R O W T H I N L AT I N A M E R I C A associated with an annual increase of 5.7 percent in interoc- plays is negligible. Second, health technologies, to some cupational inequality (thus implying an annual increase of extent, have features corresponding to public goods. Ineffi- 1.2 points in the Gini coefficient for a country with an aver- cient private provision may lead to the implementation of age interoccupational Gini of approximately 24). several public health programs. This implies that diffusion Similarly, Milanovic (2005) evaluates the impact of trade of health technologies goes beyond the embodiment of new liberalization on the distribution of income and finds that technologies. increased trade openness reduces the income share of the bot- Second, a recent strand of the literature evaluates the tom eight deciles and raises the income share of the top two impact of international trade openness on poverty through deciles (in other words, poor and middle-income groups its impact on income risk. Trade reforms may affect individ- seem to be hit harder the more their country's economy is ual risk by reallocating capital and labor across firms and integrated into world goods markets). Only when the level of sectors, thus raising short-run individual labor risk, and by income reaches a certain threshold (which Milanovic esti- increasing the elasticity of goods and the derived labor mates at about $8,000 in purchasing power parity) does demand functions. If shocks create larger fluctuations in openness appear to benefit the poor and the middle class. wages and employment because of higher demand elastic- Milanovic illustrates the economic significance of his results ity, tariff reductions may lead to increased individual by considering the impact on income distribution of a 0.2 income risk. Conversely, greater openness may reduce increase in the trade-to-GDP ratio, from 0.7 to 0.9, which income risk by reducing the volatility of goods prices that was the world average increase between 1985 and 2000. In a an autarkic economy may face relative to an economy inte- country with a mean income of $2,000 where the second grated into the world economy. In sum, economic theory decile's mean income is $800, higher trade openness would does not provide a clear indication of the nature of the rela- reduce the income share of that decile of the population by tionship between openness and income risk, and the empir- 3.8 percent, to a mean income of $760 (Milanovic, 2005, 33). ical work is ambiguous. On the one hand, Fajnzylber and Beyond income poverty, trade openness may have addi- Maloney (2005d) find no evidence that increased openness tional impacts on poverty, broadly construed through chan- increases labor demand elasticities in Colombia and Chile nels touched upon in chapter 2. First, international trade and weak evidence for Mexico. On the other hand, Krebs, may affect poverty through its influence on the rate of Krishna, and Maloney (2005) find that trade policy affects mortality. Improved health programs in developing coun- permanent income risk and argue that the welfare magni- tries may be explained by the transmission of health tech- tudes are significant (see box 5.1). nologies from industrial economies. The idea behind this argument is that the health sector in the developing coun- Size of the government tries becomes more productive by implementing new tech- A third area of possible conflict between the growth and nologies embodied in their imports of capital goods. For inequality objectives derives from the way the government instance, Papageorgiou, Savvides, and Zachariadis (2005) uses fiscal policy in the fight against poverty; a more spe- find that higher imports from countries responsible for cific issue is the relationship between inequality and the medical research and development in the world are related size of the government. Despite the significant role that to lower mortality rates. governments can play in the provision of public goods and Soares (2005) argues that although the diffusion of services, governments may also be a drain on private activ- productive technologies may partly explain the process of ity. This is likely to be the case if governments impose high the diffusion of health technologies, there are some crucial taxes, assume roles more appropriate for the private sector, aspects that are specific to the sector. First, some aspects of and maintain ineffective public programs and a bloated health (such as personal hygiene, food preparation and han- bureaucracy. Thus in principle, larger governments are dling, and water treatment, among others) are outcomes of likely to harm growth prospects. On this aspect, it can be the household production process. Absorption of health said that the empirical growth literature shows a certain technologies, in this case, may depend on the accumulation degree of consensus. of knowledge of households. In addition, to the extent that The effect of the size of the government on inequality is health improvements do not depend on specific medical less clear, however. One factor influencing that effect is the interventions, the role that embodied technological change structure of spending. For example, whether the bulk of 81 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S BOX 5.1 Trade policy and income risk Although a large body of literature deals with the impact model. The direct impact of tariff reduction is an increase of trade liberalization on levels of wages or income, Krebs, in individual income risk of 0.005 (from a mean level of Krishna, and Maloney offer the first attempt to estimate 0.008 to 0.013), and the corresponding welfare cost is empirically the effects of trade policy on individual income 0.98 percent of permanent consumption if the coefficient risk, as well as the welfare consequences of changes in of relative risk aversion (CRAA) is equal to 1 (under log- income risk induced by trade policy changes. Using house- arithmic preferences). For higher levels of risk aversion hold surveys and manufacturing data for Mexico during (a coefficient equal to 2), the welfare cost of higher income 1987­98, the authors find that tariff levels do not affect risk would increase to 1.96 percent of lifetime consump- income risk but that tariff changes do. Individual income tion. A 10 percent real appreciation would raise the risk may increase more than 30 percent in the event of a income risk from 0.008 to 0.011 with a 10 percent tariff, 5 percent reduction in tariffs. In addition, the authors find and the welfare costs are 0.59 and 1.18 percent of life- that the impact of other macroeconomic shocks on income time consumption if the coefficient of risk aversion is risk is affected by trade policy. For instance, a 10 percent equal to 1 and 2, respectively. For lower tariffs (5 per- appreciation of the real exchange rate (RER) would raise cent), individual income risk increases to 0.014, and the income risk by 35 percent if tariffs were 10 percent, and by corresponding welfare costs are 1.18 and 2.36 percent for 60 percent if tariffs were 5 percent. In contrast, a decline the different levels of risk aversion. A drop in output of of 5 percent in GDP growth would raise income risk by 5 percent would lead to higher income risk (from 0.008 25 percent if the tariff is 10 percent, and by 60 percent if to 0.01) with welfare costs of 0.39 percent of lifetime the tariff is 5 percent. In sum, trade reforms increase the consumption if the coefficient of relative risk aversion is sensitivity of income risk to macroeconomic shocks. This equal to 1, and 0.78 percent if it is equal to 2. If tariffs result is consistent with the prediction of Newberry and were lowered to 5 percent, income risk rises to 0.013, Stiglitz (1984) that negative productivity shocks would and the welfare costs are higher--0.98 and 1.96 percent have smaller equilibrium effects on output and employ- of lifetime consumption. In sum, the impact on individ- ment in a closed economy than in an open economy. ual income risk of trade reforms through their direct and Krebs, Krishna, and Maloney then calculate the wel- indirect effects in amplifying the impact of macroeco- fare effects using a simple dynamic general equilibrium nomic shocks are economically significant. Welfare effects of trade reform Changes in Welfare change Welfare change Simulation individual income risk CRRA = 1 CRRA = 2 Trade reform Tariff reduction of 5 percent 0.005 0.98 1.96 (0.002) (0.39) (0.79) Macroeconomic factors Tariff level of 10 percent GDP growth lower by 5 percent 0.002 0.39 0.78 (0.001) (0.20) (0.40) RER appreciation of 10 percent 0.003 0.59 1.18 (0.001) (0.20) (0.39) Tariff level of 10 percent GDP growth lower by 5 percent 0.005 0.98 1.95 (0.001) (0.29) (0.59) RER appreciation of 10 percent 0.006 1.18 2.36 (0.002) (0.40) (0.80) Source: Krebs, Krishna, and Maloney (2005d). Note: Numbers in parentheses are standard errors. 82 P R O - P O O R G R O W T H I N L AT I N A M E R I C A public spending is devoted to the social sectors and other by a higher level of transfers as a ratio to GDP--tend to be programs, such as infrastructure, from which the poor are associated with lower inequality (Milanovic 2000). Simi- likely to benefit has an impact on the evolution of inequal- larly, Li and Zou (2002) also find that higher government ity. Moreover, the structure of spending within social sec- spending is usually associated with lower inequality. But tors also matters. For example, figure 5.2 shows absolute Dollar and Kraay (2002) find that the incomes of the poor incidence curves of several public spending programs in the decline with greater government spending even after con- Latin American region. Each curve has been computed as trolling for average income levels (that is, the size of the the average of country-specific incidence curves; upward- government is associated with increases in income inequal- sloping lines indicate that richer quintiles benefit more ity). Kraay (2005) finds that government spending does not than poorer quintiles. Downward-sloping curves indicate have a significant effect on the Gini coefficient. progressive spending. This figure indicates that while public spending on Pro-growth, pro-poor: Is there a trade-off? health, primary education, and cash transfer programs ben- On the whole, the previous discussion indicates that in a efits people in the lower part of the distribution more than number of policy areas, progress is likely to be a win-win people in the higher part, other types of social spending, situation in that it will lead to faster growth and lower such as on tertiary education, pensions, unemployment inequality (and hence lower poverty). Yet there are some insurance, and electricity subsidies, are highly regressive. areas where a potential conflict can appear. The three areas In particular, the first quintile of the population does not reviewed above that potentially lead to growth-inequality seem to benefit at all from public spending on tertiary edu- trade-offs are especially important for Latin America. Fur- cation and pensions, whereas more than half of all spending ther financial deepening appears as a critical ingredient of in these two areas benefits the top quintile. Clearly, similar sustained development in Latin America. Trade issues have levels of aggregate social spending may have dramatically received significant attention given ongoing liberalization different impacts on income inequality depending on the efforts in the region. Similarly, as argued below, the size of social programs being implemented; substantial gains in Latin American governments is smaller than one would reducing inequality could be achieved by simply reallocat- expect, even controlling for level of development. ing resources within a given budget envelope. Unfortunately, just knowing that progress in a particu- At the same time, it is also possible to argue that if pub- lar policy area may create some growth-inequality trade- lic spending is a burden for the economy and growth, then offs is of limited use in inferring the impact on poverty. the government is likely to be more predatory than benev- Moreover, studies that estimate the simultaneous impact olent. And a predatory government may be motivated by a of policies on growth and inequality, so that one can com- desire to direct rents to specific groups, which typically are pare outcomes associated with the same inputs more or not the poor. Even where governments are benevolent in less accurately, are very rare (Li and Zou 2002; Lundberg character, a retrenchment of the public sector can lead to and Squire 2003), and none of them consider the joint cuts in programs that benefit the poor. And if public impact on poverty reduction. To begin to address these employment plays a safety-net role (by overstaffing public shortcomings, we now build on a recent study of the units, perhaps to gain the support of particular groups), World Bank's Latin American region by Norman Loayza, then retrenchment may lead to increasing inequalities. Fur- Pablo Fajnzylber, and Cesar Calderón, Economic Growth in thermore, there is some evidence indicating that in general Latin America and the Caribbean: Stylized Facts, Explanations, governments tend to pay premium salaries (above market and Forecasts (2005). rates) to unskilled workers at the expense of higher-grade Before proceeding, however, we would like to make a employers' salaries. Clearly, this policy is not likely to lead clarification. Dealing with these issues is extremely com- to efficiency gains by any standard, but it admittedly has an plex. Indeed, as some development practitioners argue, if income distribution component. the economics and the development professions more gen- On the empirical front, the literature is again quite erally still do not have a completely clear picture of what divided, with results for all possible tastes. Some empirical works and what does not work for economic growth, it evidence suggests that larger governments--measured might seem pretentious to address not only how a policy either by a higher share of workers in the public sector or affects the growth rate but also how that policy affects the 83 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 5.2 Incidence of public spending in Latin America Health Primary education Percent Percent 30 40 30 20 20 10 10 0 0 1 2 3 4 5 1 2 3 4 5 Quintile Quintile Secondary education Tertiary education Percent Percent 30 60 20 40 10 20 0 0 1 2 3 4 5 1 2 3 4 5 Quintile Quintile Electricity subsidies Unemployment insurance Percent Percent 30 40 30 20 20 10 10 0 0 1 2 3 4 5 1 2 3 4 5 Quintile Quintile Pensions Cash transfer programs Percent Percent 80 50 40 60 30 40 20 20 10 0 0 1 2 3 4 5 1 2 3 4 5 Quintile Quintile Source: Author calculations using data provided by Lindert, Skoufias, and Shapiro (2005). Note: Each graph reports the incidence of a public spending item. In each case, based on data availability, the curve is computed as the average of the country-specific incidence curves of Argentina, Brazil, Chile, Colombia, Dominican Republic, Guatemala, Mexico, Peru, and Uruguay. 84 P R O - P O O R G R O W T H I N L AT I N A M E R I C A TABLE 5.3 by the Gini index) to the same set of policy determinants, Growth and inequality regressions excluding those aimed at capturing income convergence and cyclical reversion and including lagged inequality to Variable Growth Change in logged Gini capture the possibilities of inequality convergence and a dynamic adjustment. The second column of table 5.3 Lagged inequality -0.242 (13.32) reports the results of estimating this second model. This Initial GDP per capita -0.018 combined exercise now allows us to explore the simultane- (3.80) ous impact on growth and inequality of progress on the dif- Initial output gap -0.237 (8.52) ferent policies. Education 0.017 -0.022 The estimates in table 5.3 indicate that consistent with (6.7) (2.77) the earlier discussion, several policy areas may present Financial depth 0.006 0.014 (4.28) (2.83) growth-inequality trade-offs. More specifically, while a Trade openness 0.01 0.024 more developed financial sector, an economy more open to (3.14) (3.04) international trade, and a smaller government may all be Government burden -0.015 -0.018 (3.18) (2.71) associated with faster growth, they also seem to be associ- Public infrastructure 0.007 -0.016 ated with higher levels of income inequality. (2.71) (3.32) How do these results feed into poverty changes? To Governance -0.001 0.005 (0.68) (1.74) explore whether there is a growth-poverty trade-off associ- Price stability -0.005 0.008 ated with the potential growth-inequality trade-off of these (1.89) (2.16) policies, we use the results of table 5.3 with growth and Cyclical volatility -0.277 0.112 (3.76) (1.41) inequality elasticities estimated under the assumption of External imbalances -0.006 -0.002 lognormality for income levels (see chapter 4). Recall that (3.90) (0.32) under lognormality, the impact on poverty of changes in Banking crisis -0.029 -0.021 (7.42) (4.02) growth and inequality depends on the country's initial per External conditions 0.072 0.051 capita income and inequality levels. Thus, table 5.4 pre- (4.98) (1.87) sents the result of the simulation for different values of the Gini index and different levels of per capita income relative Source: Loayza, Fajnzylber, and Calderón (2005); Lopez (2004). to the poverty line. This table also differentiates between Note: Numbers in parentheses are t-statistics. the short-run and the long-run impact of the policies on patterns of growth. We stress that we are not aiming to set poverty, something that may generate poverty dynamics any particular debate on how specific policies may affect when the speeds of adjustment of per capita income levels poverty. Our purpose here is simply to explore the practical and inequality are different. relevance of potential trade-offs between economic growth Several messages emerge from this exercise. First, the and inequality when poverty reduction is the overarching policies have a distinctly different impact on poverty over policy objective. the long run than they do in the short run. Over the long To be more specific on the way these simulations have run, progress in the three policy areas is estimated to con- been performed, we build on Loayza, Fajnzylber, and tribute to poverty reduction, but in the short run there Calderón, who relate cross-national growth rates to the pol- is the possibility of a growth-poverty trade-off (that is, icy areas in tables 5.1 and 5.2, plus other controls such as growth accompanied by higher poverty caused by the par- transitional convergence, cyclical reversion, and external allel deterioration of income distribution). The table also conditions (see also annex 5A). The first column of table 5.3 shows that the estimated orders of magnitude of the reports the results that are obtained from their empirical short-run impacts are much smaller than the orders of regression model. It suggests that countries that have magnitude of the long-run impacts, something that shown progress on the variables described above as growth should give perspective to the short-run costs and long- determinants have tended to grow more. run benefits of the different policies. That said, however, The second step in this exercise is reestimating a similar we do not want to minimize the potential negative model that now relates changes in inequality (as measured impact, even if it is only temporary, that some policies can 85 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S TABLE 5.4 Net growth elasticities of poverty to selected policies Short-run impacts Long-run impacts Gini coefficient Gini coefficient PL/pc income 0.3 0.4 0.5 0.6 0.3 0.4 0.5 0.6 Financial sector development 0.16 0.14 0.09 0.06 0.05 -1.32 -0.65 -0.36 -0.17 0.33 0.05 0.03 0.03 0.02 -1.03 -0.54 -0.29 -0.18 0.5 0.02 0.01 0.01 0.01 -0.79 -0.43 -0.25 -0.16 0.66 0.00 0.01 0.01 0.01 -0.63 -0.35 -0.22 -0.12 0.9 0.00 0.00 0.00 0.00 -0.44 -0.28 -0.18 -0.11 1.1 0.00 0.00 0.00 0.00 -0.33 -0.23 -0.16 -0.12 Trade liberalization 0.16 0.25 0.15 0.11 0.08 -2.17 -1.07 -0.59 -0.27 0.33 0.08 0.06 0.04 0.04 -1.71 -0.89 -0.48 -0.30 0.5 0.03 0.02 0.02 0.02 -1.31 -0.72 -0.42 -0.27 0.66 0.01 0.01 0.01 0.01 -1.05 -0.58 -0.37 -0.20 0.9 0.00 0.00 0.00 0.01 -0.74 -0.46 -0.29 -0.18 1.1 0.00 0.00 0.00 0.00 -0.55 -0.38 -0.26 -0.19 Government burden 0.16 -0.14 -0.09 -0.07 -0.05 4.21 2.18 1.27 0.70 0.33 -0.03 -0.03 -0.02 -0.02 2.95 1.59 0.90 0.60 0.5 0.00 -0.01 -0.01 -0.01 2.15 1.21 0.73 0.49 0.66 0.01 0.00 0.00 -0.01 1.66 0.93 0.61 0.36 0.9 0.01 0.01 0.00 0.00 1.14 0.72 0.47 0.30 1.1 0.01 0.01 0.00 0.00 0.83 0.58 0.40 0.31 Source: Lopez (2004). Note: PL/pc income is the ratio of the poverty line to per capita GDP. The tables is computed under the assumption that income follows a lognormal distribution. have on poverty, especially when temporary may mean sev- compensatory mechanisms along with policies that have a eral years. growth-inequality trade-off effect. Second, different countries may react to the same policy in different ways. Table 5.4 indicates that even if the same Complementarities and nonlinearities policy had the same effect on growth and inequality, its in the development process impact on poverty reduction would be different depending Do these findings imply that poverty reduction strategies on the country. As discussed in chapter 4, poverty in richer should tend to avoid policies that involve potential and more unequal countries is relatively more reactive to growth-inequality trade-offs? The answer to this question changes in inequality than to changes in mean income. At is unequivocally no. There is now some evidence (Gallego the same time, poverty in poorer and more equal countries and Loayza 2002; Calderón and Fuentes 2005; Loayza, is relatively more reactive to growth than to changes in Oviedo, and Servén 2005) that from an economic develop- income inequality. This finding implies that in the absence ment point of view not only does the "quantity" of an of compensatory mechanisms or complementary policies, implemented policy matter but so does the overall policy policy makers may be better placed to implement policies mix, something that the models used in the simulation involving growth-inequality trade-offs in poorer and more exercise cannot capture. In fact, one important limitation equal countries. In richer and more unequal countries, pol- of our simulations is that they are based on simple linear icy makers may need to consider implementing adequate relationships that implicitly assume that policy makers can 86 P R O - P O O R G R O W T H I N L AT I N A M E R I C A obtain a desired outcome on the growth or inequality fronts between just two policies or growth determinants. Among by making progress in a single policy area without address- those that have received significant attention are education ing other potential constraints on the economy. and institutions. In practice, however, it seems foolhardy to assume that a poverty reduction strategy can be uniquely based on win- Policy complementarities and education win types of policies without addressing bottlenecks in The role of education as an important policy complement other areas such as the financial sector or external trade dis- in the growth process is clear: education is not only an tortions, especially if progress in those areas can potentially input in the production process, it can also determine the lead to higher income inequality. Consider, for example, a rate of technological innovation and facilitate the absorp- country that liberalizes capital flows but does not show any tion of technologies. For example, de Ferranti and others respect for property rights. It would be surprising if that (2003) argue that the interaction between technology and country managed to realize the benefits of potential foreign skill is critical in determining growth, productivity, and direct investment, and it is perhaps more likely that the distribution of earnings across individuals. That report domestic capital would flee the country. also points to evidence suggesting that low levels of skill Simple linear models cannot account for complementar- can constrain the acquisition of technology through trade ities, understood as the interactions that take place among and foreign direct investment. and between policies and existing conditions of the country, The academic literature has also devoted significant region, or individual, but they can nonetheless be attention to the topic. For example, Levin and Raut (1997) extremely important. For example, Gallego and Loayza show the high degree of complementarity that exists (2002) estimate the "extra bonus" enjoyed by good between human capital and growth in the export sector for performers that jointly implement a series of growth- a sample of semi-industrial countries. They note that the promoting measures and eliminate bottlenecks in different export sector is likely to be able to use human capital more areas at more than 1 percentage point of their growth rate. efficiently than can the rest of the economy. This would At a more practical level, Lederman, Maloney, and be the case, for example, where educated workers are able to Servén (2005) argue that the effects of NAFTA varied adapt more quickly to the sophisticated technology and widely among different types of workers, firms, and rapid production changes required for competitiveness in regions in Mexico. Workers with higher skills and educa- world markets. Similarly, Borensztein, De Gregorio, and tion seem to have benefited more than workers with lower Lee (1998) present evidence of complementarity between skills. Large firms also seem to have benefited more than foreign direct investment and human capital. They argue small and medium-size ones, probably because of the that foreign direct investment contributes to higher pro- greater availability of credit to larger firms after the finan- ductivity and higher economic growth only when the host cial crisis of 1994. Similarly, commercial agricultural country has a sufficient capability to absorb the advanced producers with access to irrigated land seem to have expe- technologies. rienced significant productivity gains, whereas smaller This education complementarity to growth is important producers experienced no effect. Finally, states with higher for Latin America. For although the region's record on net initial levels of education, better infrastructure, and better primary enrollment rates is quite encouraging, most Latin local institutions accelerated their income convergence American countries have massive deficits in net enroll- toward the United States, but there was little or no move- ments in secondary education (figure 5.3). These educa- ment toward convergence among Mexico's poorer southern tional deficits are apparent even after controlling for states. income levels. Controlling for per capita income levels, the Are some policy complementarities more critical to suc- secondary enrollment deficit for the region is estimated at cessful poverty reduction than others? Several attempts about 19 percent. For tertiary education, the estimated have been made in the literature to assess the relevance of deficit is lower but still above 10 percent. policy complementarity for growth, although most of these Thus not only is the low stock of skilled human capital studies have focused on the possible complementarity in Latin America limiting the possibility of technology 87 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S result, financial resources may end up allocated to activities FIGURE 5.3 that are not the most productive. In those cases, it should Enrollment rates for secondary education relative to per capita GDP, for selected Latin American countries be no surprise if financial sector liberalization fails to meet expectations (and even results in a crisis). Argentina At the academic level, Calderón and Fuentes (2005) Brazil Chile have explored whether the empirical evidence supports this Colombia view and conclude that institutional quality seems to play a Costa Rica significant role in understanding the impact on growth of Dominican Rep. both financial sector liberalization and openness to trade. Ecuador Moreover, not only do these policies have a greater impact Mexico Nicaragua on growth impact when institutions are good, but in coun- Paraguay tries with low institutional quality, the impact on growth Peru may actually be negative. One example is a financial sector Uruguay liberalization that ends in crisis through lack of oversight. R.B. de Venezuela Similarly, Loayza, Oviedo, and Servén (2005) estimate that 50 40 30 20 10 0 10 high levels of regulation are associated with higher macro- Percent economic volatility, lower growth, and more informality in Source: de Ferranti et al. (2003). labor markets. However, this effect is observed mainly in countries with low institutional quality. As the quality of institutions improves, the negative impact of regulation on adoption, but it may also be affecting the way other poli- macroeconomic performance and growth disappears. cies such as trade or capital account liberalization influence Is this type of policy complementarity relevant in the the growth process. Latin American context? Figure 5.4 plots the average for the six indexes contained in the Kaufman, Kraay, and Mastruzzi Policy complementarities and institutions (2004) database of institutional quality measured against A second area that has received significant attention as a the log per capita income level of each country. The figure potential policy complement is institutional quality. Insti- indicates that a very close association between per capita tutions, understood as the rules and norms constraining human behavior (North 1990), basically establish the rules FIGURE 5.4 of the game for a society. The importance of institutions in Institutions and per capita income levels the process of development has long been understood-- going back at least to the writings of Adam Smith. More Institutional quality, index 2.5 recently, it has been argued that growth-enhancing poli- 2.0 cies, including in the areas of human capital accumulation 1.5 and trade openness, are less likely to be effective where 1.0 political and other institutions are weak. As a result, these 0.5 arguments continue, the adverse effects of weak institu- 0 tions on economic performance are reinforced by their 0.5 interaction with other policies. 1.0 For example, World Bank (2005c) notes that the effec- 1.5 tiveness of financial liberalization on growth depends to a 2.0 large extent on the underlying institutions: intermediaries; 2.5 3.0 3.5 4.0 4.5 5.0 markets; and the informational, regulatory, legal, and judi- Per capita GDP, log cial framework. When supervision and financial regulation World Latin America Linear (World) are weak, liberalization may encourage domestic financial institutions to build up excessive risk by borrowing exces- Source: Authors' calculations based on Kaufmann, Kraay, and Mastruzzi (2004) data. sively and expanding lending to overly risky activities. As a 88 P R O - P O O R G R O W T H I N L AT I N A M E R I C A TABLE 5.5 Institutional quality in Latin America Country Institutional quality Country Institutional quality Argentina -0.34 Honduras -0.51 Bolivia -0.43 Jamaica -0.05 Brazil 0.01 Mexico 0.04 Chile 1.25 Nicaragua -0.32 Colombia -0.55 Panama 0.16 Costa Rica 0.77 Paraguay -0.78 Dominican Republic -0.25 Peru -0.35 Ecuador -0.66 Trinidad and Tobago 0.3 El Salvador -0.06 Uruguay 0.54 Guatemala -0.65 Venezuela, R. B. de -0.97 Haiti -1.59 Median -0.32 Source: Authors' calculations using data from Kaufman, Kraay, and Mastruzzi (2004). income levels and institutional quality, something that in FIGURE 5.5 turn suggests that a comparison of institutional quality Rural and urban headcount poverty rates based on absolute indicators may be misleading. To address this issue in part, table 5.5 tabulates the relative performance Bolivia of countries in the region controlling for income levels. More Brazil specifically, the table reports the difference between the Chile observed institutional index and its expected value. Costa Rica Table 5.5 indicates that two-thirds of the countries in Dominican Rep. the sample have a negative sign, indicating institutional Ecuador underperformance. The countries with a clear positive sign El Salvador are Chile, Costa Rica, Panama, Trinidad and Tobago, and Honduras Uruguay. Brazil, El Salvador, and Mexico are clustered Jamaica around the regression line, and the rest are well below it. Mexico Haiti, Paraguay, and República Bolivariana de Venezuela Nicaragua have the strongest negative signs. Clearly, as in education, Panama many Latin American countries may be limiting the effec- Paraguay tiveness of some of their poverty reduction policies by not improving the effectiveness of their institutions. Peru 0 10 20 30 40 50 60 70 80 Does the composition of growth matter? Percent In the previous section we addressed several policy issues Urban Rural related to pro-poor growth. Determining the effects of Source: Gasparini, Guitierrez, and Tornarolli (2005). growth on poverty can also be addressed from a sectoral Note: Poverty is defined here as living on $2 or less per day. point of view. Beyond accounting issues related to the rela- tive size of the sector in question, there are a number of rea- sons why growth in some sectors may alleviate poverty in sectors located where the poor live would likely have a more than growth in other sectors.2 One reason is the rela- large impact on poverty alleviation (see chapter 7 for a dis- tionship between the geographic location of a sector's pro- cussion of spatial mobility, poverty, and growth). duction and the incidence of poverty in the area. According The existing empirical evidence seems to give some to this argument, in the absence of spatial mobility, growth support to this view. Figure 5.5 illustrates the different 89 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S poverty rates of urban and rural areas in a selected number Latin America, and high-income developed countries. This of Latin American countries. The figure reveals that rural panel shows that developing countries, including Latin poverty rates tend to be much higher than urban poverty America, experienced positive effects emanating from rates. In Peru, for example, about two-thirds of the rural growth in the rural sector. On average, a 1 percent increase population is poor, compared with 4 percent in urban areas. in rural activities would translate into a 0.12 percent The median rural poverty rate for the 14 countries in the increase in nonrural activities. figure is 52 percent; the median urban poverty rate is 19 per- Conversely, panel B of this figure shows that growth in cent. Thus is growth in the agricultural sector more pro- the nonrural sector would have a very modest (and statisti- poor than growth in the nonagricultural sector? In Beyond cally insignificant) impact on Latin American rural growth. the City: The Rural Contribution to Development, de Ferranti In other developing countries and in high-income devel- and others (2005) found that, on average, the expansion of oped countries, growth in the nonrural sector is associated agricultural activities in Latin America would contribute with a shrinking of rural output, something known as the less to overall poverty reduction than the expansion of the "pull effect." Generally speaking, one effect of this asym- nonagricultural sector. To a large extent, however, this metry is that different sectors lead to different rates of result was a consequence of the agricultural sector's smaller poverty reduction even if they have similar shares of GDP size. In fact, relative to its size, agricultural growth in Latin and a similar impact on poverty, controlling for growth. America tends to be more pro-poor than overall growth in The labor intensity of growth may also explain why nonagricultural sectors. growth in different sectors seems to have different effects A second explanation for why some sectors have a larger on poverty. Loayza and Raddatz (2005) stress that differ- impact on poverty reduction than others is related to the ences in the relative labor intensities of various sectors help potential spillovers between sectors. If one sector acts as a explain why their effects on poverty alleviation are not the locomotive for other sectors, then growth in the locomotive same. How different, then, is relative labor intensity across sector would be expected to have a larger impact on sectors and across countries? Is the pattern of sectoral poverty. Figure 5.6 illustrates this issue for a two-sector growth elasticities of poverty consistent with relative labor economy (rural and nonrural) using results from Bravo- intensities? Ortega and Lederman (2005). More specifically, panel A of Figure 5.7 presents box-plots for the cross-country dis- figure 5.6 shows the estimated percent increase in the non- tribution of relative labor intensities corresponding to six rural sector associated with a 1 percent increase in rural GDP economic sectors. Agriculture is clearly the most labor- for Latin America, other developing countries excluding intensive sector: the ratio of median labor intensity to FIGURE 5.6 Potential spillovers between rural and nonrural GDP a. Rural b. Nonrural 0.20 0.02 0 0.15 0.02 0.10 0.04 0.05 0.06 0.08 0 0.10 0.05 0.12 0.14 0.10 0.16 0.15 0.18 Developing Latin High-income Developing Latin High-income countries America countries countries America countries Source: Bravo-Ortega and Lederman (2005). 90 P R O - P O O R G R O W T H I N L AT I N A M E R I C A TABLE 5.6 FIGURE 5.7 Poverty reduction and sectoral growth Relative labor intensity per sector 1. Partially Partially constrained, 2. Sector growth Unconstrained constrained robust 3. Agriculture growth -15.228 -15.952 -13.08 (-1.80) (-2.37) (-2.03) 4. Mining growth 4.575 4.521 4.256 (1.17) (1.39) (1.40) 5. Manufacturing -2.051 -1.235 -1.241 growth (-1.42) (-1.64) (-1.68) 6. Utilities growth 5.463 4.521 4.256 (0.86) (1.39) (1.40) Construction growth -1.477 -1.235 -1.241 0 0.5 1.0 1.5 2.0 (-0.33) (-1.64) (-1.68) Ratio Services growth -0.480 -1.235 -1.241 (-0.19) (-1.64) (-1.68) 1. Mining 2. Utilities 3. Services 4. Manufacturing Source: Loayza and Raddatz (2005). 5. Construction 6. Agriculture Note: The dependent variable is the change in headcount poverty. t-statistics are in parentheses. Growth rates are share Source: Loayza and Raddatz (2005). weighted. Note: Excludes outside values. FIGURE 5.8 sector size is nearly 1.4 and most corresponding country Poverty changes and labor-intensive growth throughout the world values are larger than 1. Construction, manufacturing, and services can be grouped in another category of labor inten- Growth of poverty headcount index 0.2 ETH sity, with median ratios surrounding 1. The construction sector is notable in that its cross-country distribution of YEM relative labor intensities is quite dispersed around the 0.1 COL CHN IND mean. Mining and utilities are the least labor-intensive sec- PRY LSO GHA LKA MYS SLV NGA tors, with median ratios around 0.5 and moderately con- 0 PER EGY VEN PAN ZMB MAR UGA THA TUN centrated distributions. ECU HND IDN MEX KEN BRA The notion that relative labor intensity determines a PHL 0.1 PAK sector's influence on poverty alleviation is consistent with VNM the pattern of coefficients on sectoral growth in table 5.6. 0.2 This table presents the results of regressing changes in 0.005 0 0.005 headcount poverty on sectoral growth interacted with the Labor intensity­weighted sectoral growth coef 11.440578, (robust) se 5.2829909, t 2.17 share of the sector in total value added. Given the some- Source: Loayza and Raddatz (2005). what small sample size available and relatively large disper- sion across countries, three different specifications are used. The first is a fully unrestricted specification. The second utilities does not seem to help reduce poverty, once growth pulls together sectors that appear to have similar effects on in other sectors is controlled for. Thus agriculture, the most poverty. The third also controls for the impact of extreme labor intensive-sector, presents the largest growth elasticity observations or outliers. of poverty, while mining and utilities carry the lowest elas- The table indicates that growth in agriculture appears to ticities for poverty reduction. Manufacturing, services, and have a clear, significant poverty-reducing effect. Growth in construction can be found in the middle of both labor inten- manufacturing, construction, and services also appears to sity and poverty reduction effects. have a poverty-reducing effect, which is statistically signif- The relevance of sectoral labor intensity is also apparent icant at marginal levels. In contrast, growth in mining and from figure 5.8, which shows a partial-regression plot linking 91 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S the change in poverty to sectoral growth weighted by labor FIGURE 5.9 intensity. This figure confirms a negative pattern that is The impact of public transfers on income inequality well established by most observations in the sample. Thus, it appears that in addition to the size of growth, the degree Argentina of labor intensity in that growth is statistically and eco- Brazil nomically relevant for explaining poverty reduction. Chile We emphasize, however, that these results should not be used as a rationale for adopting industrial policies that bias Colombia the sectoral composition of growth toward some sectors in Dominican Rep. the name of a higher growth elasticity of poverty. Such Guatemala policies may result in the country moving away from its Mexico comparative advantages because policy makers may face a trade-off between a higher growth elasticity of poverty and Peru a lower growth rate for the economy as a whole. 2 1 0 1 2 3 4 Removing bias, especially against the agricultural Change in Gini, % sector, and overcoming underinvestment in public goods Source: Lindert, Skoufias, and Shapiro (2005). (such as education and infrastructure) in rural areas are Note: A positive entry indicates that inequality declines when transfers are taken into account. completely different issues, however. According to de Ferranti and others (2005), overall public expenditures in Latin America are allocated with an apparent pro-urban bias. Similarly, the results discussed in this section also support the removal of biases against labor, whether policy induced Similarly, according to World Bank (2005d), income or not, so that effective opportunities can be created for the inequality is unaltered in El Salvador regardless of whether poor in growing economic activities. it is estimated before or after government transfers. Finally, analysis undertaken for this report indicates that public The role of taxes and transfers in reducing transfers would contribute to declines in the Gini coeffi- income inequality cient of about 4 points in Bolivia; 2­3 points in Costa Rica; So far we have focused on the impact that different policies 1­2 points in Ecuador, Nicaragua, Uruguay, Paraguay, and and sectors have on poverty reduction through their effect República Bolivariana de Venezuela; and 1 point or less in on market incomes. In practice, however, the relevant dis- Honduras. tribution for poverty purposes is that of disposable income, As for the impact of taxation on the distribution of which takes into account the redistributive role of the gov- income, Engle, Galetoviv, and Raddatz (1998) estimate ernment through taxes and transfers. Thus, what is the role that in 1996 the after-tax Gini coefficient for Chile was of the government budget and, more specifically, of taxes 0.496, compared with the before-tax Gini of 0.488--this and transfers in explaining the distribution of disposable despite the fact that Chile's tax system is the most effective income in Latin America? And what are the possibilities of in Latin America, collects the most from personal income making progress on this front? taxes, and has the highest marginal rates. Moreover, these In a recent paper Lindert, Skoufias, and Shapiro (2005) researchers estimate that even if tax allowances were elimi- present estimates of the Gini coefficient of eight Latin nated from the personal income tax and underreported American countries before and after transfers. Their findings income was taxed, the improvement in the Gini index indicate that in seven of the countries, public transfers would be only marginal, and they argue that the more (defined as social assistance plus social insurance) help mod- unequal the distribution of market incomes, the less estly to lower levels of income inequality. In Peru transfers the redistributive effect of progressive taxation. Although have the opposite effect and contribute to higher inequality the evidence that emerges from these studies is clearly very (see figure 5.9) The average change in the Gini coefficient of limited, it indicates that in most Latin American countries, household income as a result of public transfers for the eight market income inequality does not likely differ much from countries in figure 5.9 is around 1 percentage point. disposable income inequality. 92 P R O - P O O R G R O W T H I N L AT I N A M E R I C A In contrast, the role played by the tax and transfer and the United States have basically the same levels of instrument in developed countries is apparently much inequality both before and after taxes and transfers. more significant. For example, according to Atkinson A natural question that emerges from this discussion is (2003), the Gini coefficient of market income in the United of great relevance for fiscal policy, namely, from a redistrib- Kingdom is around 0.53 whereas the Gini coefficient of dis- utive point of view, is the role played by taxes more impor- posable income is much lower: around 0.35. That is, taxes tant or less important than the role played by transfers in and transfers reduce income inequality in the United King- the EU countries? To address this issue, panel B reports dom by 18 percentage points as measured by the Gini coef- the Gini coefficient of market income before and after taxes ficient. Atkinson makes similar estimates for Canada, and social security contributions. This panel suggests that Finland, Germany, and Sweden. He does not provide the the coefficient does not change much for the EU15 overall, elements to compare the role of taxes and transfer in the falling just 2 points after taxes; in some countries-- United States, but according to the U.S. Census Bureau, Denmark, Finland, and Sweden, it even increases. The rea- the Gini coefficient of income before taxes and transfers is son for this is apparent from panel C, which indicates that 0.47, whereas the OECD estimates a Gini of 0.34 for the Gini coefficient of taxes is very similar to the Gini coef- disposable income in the United States. ficient of market incomes across the different European A similar picture emerges from data provided by countries. If taxes are a constant proportion of income at all EUROMOD, a source of harmonized microdata on the dif- points in the distribution (that is, if it is a flat tax), the Gini ferent income components before and after redistribution coefficient will not change at all after taxes. through the tax-benefit system for 15 members of the However, the story from panel D, which compares the European Union (EU).3 As can be observed in panel A of Gini coefficients of transfers and market incomes, is radi- figure 5.10, the EUROMOD data provide estimates of the cally different. For the EU15 overall, the Gini of transfers is Gini coefficient that are virtually identical to those pro- a low 0.04, indicating an almost perfectly equal allocation vided by Atkinson (2003) for the countries where there of transfers along the income distribution. Thus, to a large is overlap; the exception is the market income Gini for extent most of the redistribution observed in the EU coun- Sweden, where the estimate is now 0.45. tries comes from the transfer component rather than from Panel A also indicates that the Gini coefficient of market the tax component. This is not to say that taxes are not incomes for the United Kingdom and Ireland are similar: important. In fact, since they finance the transfers, they a high 0.53 and 0.52, respectively. Surprisingly, even the are critical. However, the relevance of taxes for reducing Nordic countries of the EU15, which are traditionally income inequality would appear to be more related to the praised for their levels of equality, also show very high tax level than to the structure (in fact, the correlation coef- inequality in market incomes. The Gini indexes for ficient between redistribution and tax level as a percentage Denmark, Finland, and Sweden are 0.49, 0.49, and 0.45, of GDP is 0.41). respectively. The most equal countries in terms of market What can we learn from this? First, the evidence pre- incomes are Austria and Netherlands with Gini coefficients sented above indicates that redistribution takes place of 0.38 and 0.39, respectively. According to the EURO- largely through transfers rather than through taxes. Taking MOD data, the population-weighted average Gini of the into account the potential negative impact of taxes on eco- EU15 countries before taxes and transfers is 0.47. nomic efficiency, this finding suggests that policy makers After taxes and transfers, however, the Gini coefficient is interested in the use of the tax-benefit instrument to substantially lower in all the countries.4 For the EU15 as a address income inequality and poverty concerns should first wholeitis0.33.Thatis,intheEU15taxesandtransferslower address the composition and structure of existing transfer theGinicoefficientby14points.Thisdeclineisevenlargerin programs, and when in need of additional resources use Denmark and Ireland where taxes and transfers lower the taxes to increase collections while minimizing economic Gini by 20 and 19 points, respectively. Even the countries distortions. that distribute the least through the tax-benefit system Second, the data also suggest that this is an area where (Greece, Italy, and Portugal) still manage to lower their Gini Latin America can make progress. Even if one assumes that index by more than 10 points. One final point: even though the Latin American market income Gini coefficient is 4 per- there may be some comparability issues, the EU15 as a whole centage points above the disposable income Gini (which on 93 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 5.10 Gini coefficient in selected countries before and after taxes and transfers Panel a Panel b Austria Austria Belgium Belgium Denmark Denmark Finland Finland France France Germany Germany Greece Greece Ireland Ireland Italy Italy Luxembourg Luxembourg Netherlands Netherlands Portugal Portugal Spain Spain Sweden Sweden United Kingdom United Kingdom EU15 EU15 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Gini index Gini index Market income Disposable income Market income Market income after taxes and social security contributions Panel c Panel d Austria Austria Belgium Belgium Denmark Denmark Finland Finland France France Germany Germany Greece Greece Ireland Ireland Italy Italy Luxembourg Luxembourg Netherlands Netherlands Portugal Portugal Spain Spain Sweden Sweden United Kingdom United Kingdom EU15 EU15 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Gini index Gini index Market income Taxes and social security contributions Market income Transfers (including pensions) Source: Authors' calculations using EUROMOD (2004) data. 94 P R O - P O O R G R O W T H I N L AT I N A M E R I C A the basis of the evidence presented above would seem a total tax revenues as a percentage of GDP are generally low. high estimate), fully half of the differences in disposable This is so whether one measures revenue in absolute terms income inequality between Latin America and Europe (or (in 2000, Latin American countries were collecting, on the United States) are attributable to the different effec- average, half as much as industrial countries were) or as the tiveness of tax and transfer systems.5 level of per capita GDP of the individual countries. However, several caveats are in order when moving from Figure 5.11 indicates that only three Latin American coun- this stylized fact to the design of policy. First, these calcula- tries have tax revenues above the regression line (Honduras, tions do not include in-kind transfers such as those pertain- Nicaragua, and Uruguay), while only one (Brazil) has rev- ing to public health, education, or housing, the bias of enue on the regression line. The rest of the region is col- which we have not examined in this report. Second, the lecting less than would be expected given their level of level of taxes and transfers may itself affect the observed development--dramatically less in some cases--notably, level of market income inequality, an effect that is difficult Argentina, at 12 percent of GDP, and Colombia, El to follow without careful modeling. Finally, at this point Salvador, and Paraguay, at 8 percent of GDP. we cannot separate transfers from the well-off to the poor This regional underperformance is particularly relevant from pensions, which are intertemporal transfers from the because even though the structure of the taxes may not be well-off now to themselves (or others like them) during the most relevant factor from a redistributive point of view, retirement when incomes are low. Thus, it is possible that the quantity of taxes does matter both as a factor that miti- our analysis overestimates the magnitude of the redistribu- gates fluctuation in market incomes (see box 5.2) and as a tion effect of transfers. determinant of the overall budget envelope available for use on the spending side. Why do Latin America's taxes and transfers have A natural question is whether the poor performance of the such a low redistributive impact? region on the revenue front is generated by the poor perfor- Several reasons may explain why taxes and transfers have mance of one particular tax category or whether it is caused such a low redistributive impact in Latin America. First, by problems common to the overall taxation framework. FIGURE 5.11 Total tax revenue versus per capita income, throughout the world Total tax revenue % of GDP 45 LAC Selected countries throughout the world 40 Italy France 35 30 Estonia Spain Uruguay 25 20 Brazil Chile United States Nicaragua Costa Rica Honduras Peru 15 Dominican Rep. Mexico Argentina Paraguay Bolivia Colombia 10 El Salvador Guatemala 5 0 4.5 5.5 6.5 7.5 8.5 9.5 10.5 11.5 Per capita GDP, log Source: Authors' calculations. 95 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S BOX 5.2 Taxes, transfers, and inequality The overall impact of the government budget depends on Suppose now that inequality increases in the market the combined effect of taxes and expenditure. A progres- incomes of the nonpoor population, leaving the mean sive transfer system financed by a proportional tax is unaffected, so that the same tax t finances the same trans- progressive overall. Moreover, personal taxation may fer. A given increase in the Gini coefficient for market dampen disequalizing changes in the market distribu- income translates into an increase in the inequality of dis- tion, even where the tax system is purely proportional. A posable income of 1 t as much. With a tax rate of simple example may help to illustrate this point. Suppose 50 percent, an increase of market inequality of 5 percent- there is a group, referred to for convenience as the poor, age points corresponds to an increase of 2.5 points in that makes up a proportion p of the population and has disposable income inequality. Thus countries with low zero market income. The poor receive a state transfer, b, taxation levels will find a close mapping from changes in financed by a proportional tax at rate t on the income of market income inequality and disposable income the rest, 1 p, of the population. The transfer is revenue inequality, whereas this association will be much lower in neutral in that the sum of market incomes is equal to the countries with higher tax levels. sum of net incomes after taxes and transfers. Source: Based on Atkinson (2004). TABLE 5.7 How much is Latin America undercollecting? (percent of GDP) Country Total Corporate Personal Goods and services International trade Property Argentina -12.3 -1.2 -4.4 -3.4 -1.1 -0.3 Bolivia -3.6 -1.5 -1.5 1.5 -2.7 1.1 Brazil -0.7 -1.3 -3.7 -0.8 -1.9 -0.5 Chile -3.6 -2.4 -4.0 2.9 -0.4 -0.5 Colombia -8.6 1.6 -2.7 -1.7 -1.7 -0.3 Costa Rica -3.3 -1.0 -3.0 0.0 -0.1 -0.4 Dominican Republic -4.0 -1.2 -1.1 -1.6 2.9 -0.3 El Salvador -7.7 -1.1 -1.5 -0.5 -1.6 -0.3 Guatemala -9.4 -1.0 -2.0 -1.5 -1.8 -0.3 Honduras 1.4 -- -- -- -- -- Mexico -5.2 -2.4 -3.6 1.0 -1.9 -0.5 Nicaragua 3.2 -2.3 -0.7 4.5 -2.4 -0.2 Panama -3.5 -1.0 -3.4 -3.2 0.4 -0.1 Paraguay -8.0 -0.6 -2.6 -1.5 -1.0 -0.1 Peru -4.6 -0.7 -2.0 1.4 -1.4 -0.2 Uruguay 1.8 -0.7 -3.4 1.9 -0.8 0.8 Venezuela, R.B. de -6.4 6.0 -3.5 -3.4 -0.9 -0.1 Median -4.0 -1.0 -2.9 -0.6 -1.2 -0.3 Source: Authors' calculations. Note: -- = not available. A negative entry indicates that the country is collecting less than it should, taking into account its per capita income level. To address this issue, table 5.7 reports how much each of - into account). The table clearly shows that the median coun- several Latin American countries is undercollecting, control- try in the region is collecting 4 percentage points of GDP ling for per capita income levels (defined as the difference less than one would expect, with Argentina, Colombia, El between the actual tax revenue collection in each country and Salvador, Guatemala, and Paraguay showing collection lev- its predicted value once differences in income levels are taken els that are 7.5 percentage points below the predicted value. 96 P R O - P O O R G R O W T H I N L AT I N A M E R I C A The table shows that the region is undercollecting no In some cases, social insurance programs, such as pen- matter whether the tax is on personal income, property, sions and unemployment insurance, have much larger unit corporate income, goods and services, or trade. It is note- values, but these programs tend to be regressive, mainly worthy that in the case of the personal income tax, not a because they are accessible only through employment in single country is collecting above or in line with expecta- the formal labor market.7 Since poor households tend to tions. In effect, the only tax that Latin America seems to be work in the informal labor market, they do not have access collecting more or less in accordance with the international to these benefits. Nonetheless, these programs constitute experience is the goods and services tax.6 a significant portion of total public spending, much of Moving to the spending side, the first aspect to mention which is financed by general taxation (due to deficits in is that not all transfers are the same. In fact, given the dif- contributions). In most cases, even the net subsidies to ferences in incidence and unit values, we have divided social insurance (those financed by general taxation, net of "social protection transfers" into two broad categories: contributions) are still several times higher than spending on targeted social assistance programs (figure 5.13). · Social insurance (SI): transfers for which beneficiaries Thus, there is scope for both fiscal savings and improve- make contributions that involve some degree of ments in equity by reducing pensions deficits and im- "risk pooling," but the benefit they receive is not proving accessibility by poor and informal workers. necessarily directly proportional to what they con- The redistributive and poverty impacts of well-targeted tribute; and programs, such as conditional cash transfers, could be · Social assistance (SA): transfers for which beneficiaries enhanced through broader coverage and higher unit trans- do not make a direct "risk-pooling" contribution. fers, provided that these reallocations are accompanied by design incentives to promote work efforts and link benefi- Within this second group particularly attractive vehicles ciaries to complementary services to help them get beyond are the conditional cash transfer (CCT) programs such as cash assistance. Bolsa Escola in Brazil, the Subsidio Unico Familiar (SUF) and Solidiario programs in Chile, Familias en Acción in Colombia Simulating redistributive packages (see box 5.3), Programa de Asignación Familiar in Honduras, In the previous sections we discussed the structure of taxa- and Oportunidades, previously known as Progresa, in Mexico. tion in Latin America and reviewed the situation regarding Under these programs, the receipt of the transfers is condi- public transfers, but so far we have not addressed the tioned on the household investing in the education and required fiscal effort the region should make to reduce health status of their members. This type of program has poverty through the tax and transfer instrument. The the benefit of contributing to an immediate reduction in answer to this question is critical for assessing both the inequality and poverty through the cash transfer compo- practical possibilities of achieving fast poverty reduction nent and to a sustained decrease in poverty over the through redistribution over the short run and the potential medium-to-long run through the associated accumulation for improvement on this front over the long run. of human capital by the beneficiaries. In that sense these This section takes a first pass at this issue with the pur- transfers are not a trade-off between growth and redistri- pose of illustrating the order of magnitude of the required bution, as could be argued with more traditional pure cash efforts. It presents the results of simulating the incremental transfers. tax rates associated with reducing poverty by 25, 50, and The low impact of transfers on income inequality occurs 75 percent over a 10-year horizon under a simple tax and even though Latin American social assistance programs, transfer scenario. The redistributive policy we consider and in particular conditional cash transfer programs, tend would tax all income at the same rate and allocate the rev- to be well targeted. The problem is that their unit values enues in equal amounts per capita. Here, the resulting are small (figure 5.12), which considerably limits their decline in the Gini coefficient is similar to the tax rate. ability to redistribute income. In Peru, for example, the This simple redistributive policy, although not targeted to unit value of social insurance transfers (pensions) is about the poor, is not far from the actual fiscal system of several 10 times higher than the value of food-based social assis- countries (including those of the EU15 reviewed above), tance programs. where taxes are approximately proportional and per capita 97 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S BOX 5.3 Conditional cash transfers in Colombia The Familias en Acción program is a conditional cash trans- for each child who met the primary school attendance fer program that has successfully increased human capital requirements and US$10 monthly for each child who accumulation in low-density, high-poverty regions of met the secondary enrollment requirements. Colombia. The program was initiated in 2001 amid high An impact evaluation using a randomized sample unemployment, slow economic growth, increasing armed design showed that after two years, the Familias en Acción conflict, and increased poverty rates. Although impact program had significant impacts on health: evaluations from Mexico's Oportunidades suggested that · Food consumption, especially of proteins and dairy, this program design could be effective, the Colombian increased; doubters argued that Familias en Acción would create a cul- · Vaccinations increased by 7­12 percentage points; ture of dependency and crowd out adult labor, that the cash · Children's height increased by 0.62­0.75 centime- would be diverted to adult consumption, that fertility ters, and their weight increased by 0.32­0.48 kilo- rates would increase, and that the human capital impacts grams; observed in Mexico were an anomaly that could not be · Illness dropped by 11 percentage points; replicated in Colombia. A well-designed and imple- and on education: mented program, coupled with carefully designed impact · Secondary school attendance increased by 4.6­10.1 evaluations, showed not only that the critics were wrong percentage points; and but also that such a program had potential in poor, rural · Primary school attendance increased by 3 percentage zones. The objectives of the Familias en Acción program points. were to complement the income of extremely poor families with children under age 18 by The program did not generate the adverse incentive effects that were feared. The evaluations showed: · Reducing the nonattendance and desertion rates of students · Child labor declined by an average of 80 hours a · Improving health outcomes of children under age 7 month; · Improving health care practices for children, · Adult labor increased by 3.6­6.5 percentage points; including improving nutrition and early stimula- · Participants were 2.5 percentage points less likely tion and curbing family violence. to migrate; Familias en Acción was implemented in 631 municipal- · Birth rates declined by 9­13 percentage points; and ities, covering 58 percent of all low-density areas, and · Alcohol, tobacco, and other adult consumption did benefited nearly 1 million children in 340,000 families. not increase. Before the program began, the target population had Given these positive results, the future of the program monthly household expenditures below US$30 per capita, looks bright. The government has implemented the pro- 10 percent of the children were severely malnourished, gram in pilot urban areas to determine its effectiveness in nearly 50 percent of the children under age 6 were ill, and high-density, high-poverty zones and, depending on the 9 percent of primary school children and 37 percent of sec- results from future impact evaluations, plans to expand ondary school children were not attending school. coverage to the entire poor population by the year 2019. Eligible families were those who were indigent poor On a larger scale, the Familias en Acción program shows and living in the target municipality. Families with chil- that successful conditional cash transfer programs, such dren younger than age 7 were eligible for a bimonthly as Mexico's Oportunidades and Brazil's Bolsa Escola pro- transfer equivalent to US$17 if they complied with the gram, can be replicated elsewhere. Careful evaluation has growth and development control appointments for their provided a new data point that supports the human capi- children over the two-month period. Mothers of school- tal accumulation power of conditional cash transfers, age children received the equivalent of US$5.50 monthly with few of the efficiency losses predicted. 98 P R O - P O O R G R O W T H I N L AT I N A M E R I C A public expenditures do not vary substantially with income. FIGURE 5.12 Our simulations assume that there are no efficiency costs Social protection spending mix in Latin America (that is, the increase in taxes and transfers does not affect Ecuador 2004 Social insurance growth), something that admittedly may be unrealistic Nicaragua 1999 Social assistance given the typical inefficiency costs associated with taxation. Dominican Rep. 2002* For comparison purposes, we also estimate the growth Honduras 1998 Guatemala 2000* rates that would be required to achieve the same poverty Colombia 2002* reduction objectives when growth is not accompanied by El Salvador 2003 any distributional change, as well as the required tax Mexico 2002* R.B. de Venezuela 2000 increases needed when growth averages 3 percent a year Paraguay 2000 over the 10-year horizon. Table 5.8 reports the results of Peru 2003* Costa Rica 1999 the first two simulations, and figure 5.14 shows the incre- Chile 2000* mental tax rate associated with the third simulation. Argentina 2002* For example, according to table 5.8, Costa Rica has to Brazil 2004 cons.* Uruguay 1998 grow at an annual rate of 2.6 percent for the next decade to reduce poverty by 25 percent, assuming no changes in Latin America inequality. The corresponding growth rates for the targets OECD 1995 United States 1995 of reducing poverty by 50 and 75 percent are 6.1 percent Continental Europe 1995 and 14.2 percent, respectively. Notice that even though 0 5 10 15 20 25 fast poverty reduction in the region requires a significant % of GDP acceleration in observed growth rates, the estimates in Source: Lindert, Skoufias, and Shapiro 2005. table 5.8 are not completely unrealistic. The median per Note: *Data are from the most recent year available. capita growth rate of the estimates associated with reduc- ing poverty by 25 percent is 2.4 percent, whereas that of the second target is 5.5 percent. The third target--reduc- ing poverty by 75 percent--would require a less realistic growth rate of about 10 percent. Looking now at the incremental tax rates required to FIGURE 5.13 reduce poverty through redistribution alone, the estimates Impact of social insurance and social assistance programs on inequality in table 5.8 indicate that if the objective is cutting poverty in half over a 10-year period, the tax rates of the region Argentina should increase by between 5 percent (Chile) and 33 percent Brazil (Nicaragua). The median values associated with the three Chile poverty reduction targets in our simulations are 11, 20, and 29 percent. Over a 10-year period, these incremental tax Colombia rates would produce the same poverty reduction as would Dominican Rep. the neutral growth rates we estimate in the first simulation. Guatemala Needless to say, such high tax increases seem unrealistic from a practical point of view. Moreover, with these incre- Mexico mental tax rates, it would be very difficult to maintain our Peru assumption of no efficiency costs associated with the tax 2 1 0 1 2 3 4 and transfer policy--in practice if one allowed for some Change in Gini, % negative impact on income growth, one would expect the Social insurance Social assistance necessity for an even higher incremental tax rate. Obviously, the two simulations shown in table 5.8 are Source: Lindert, Skoufias, and Shapiro (2005). extreme cases. Figure 5.14 attempts to illustrate the benefits Note: Positive values indicate a reduction in inequality. 99 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S TABLE 5.8 Results of simulations of income-neutral growth rate and incremental tax rate Neutral growth Redistribution Income growth rate Incremental tax rate Country 25% 50% 75% 25% 50% 75% Argentina (2004) 2.2 5.0 10.2 5.6 10.5 15.8 Bolivia (2002) 3.5 8.6 20.8 19.0 31.4 40.9 Brazil (2003) 2.4 5.4 12.8 5.3 9.9 15.9 Chile (2003) 1.4 3.4 8.1 2.3 4.8 8.7 Colombia (2004) 3.1 9.9 17.3 8.2 17.0 25.1 Costa Rica (2003) 2.6 6.1 14.2 4.5 8.5 13.3 Dominican Republic (2004) 1.5 3.4 6.3 4.4 8.7 13.2 Ecuador (2003) 2.3 5.4 10.5 14.5 25.2 34.3 El Salvador (2003) 2.7 6.4 13.9 19.0 31.8 42.2 Honduras (2003) 2.3 5.6 9.8 12.8 22.9 30.1 Mexico (2002) 2.6 7.2 25.9 10.9 21.4 32.8 Nicaragua (2001) 2.5 5.5 9.8 20.9 33.2 42.0 Panama (2002) 2.2 5.3 10.1 4.4 8.8 12.9 Paraguay (2002) 3.7 9.5 27.2 16.8 28.0 37.3 Peru (2002) 2.4 5.5 10.2 11.1 19.6 26.9 Uruguay (2003) 1.2 2.6 5.4 2.7 5.5 9.5 Venezuela R. B. de (2000) 2.1 4.8 9.2 14.4 25.3 34.8 Source: Gasparini, Gutierrez, and Tornarolli (2005). poverty in half over a 10-year period with a per capita FIGURE 5.14 growth rate of 3 percent a year. Even if the tax increases are Incremental tax rate needed to halve poverty in 10 years much lower than those reported in table 5.8, they are still Bolivia quite significant and in most cases above the tax level for Paraguay each country given its per capita income level. For exam- El Salvador ple, Bolivia, El Salvador, Nicaragua, and Paraguay would Nicaragua Ecuador need tax increases in excess of 12 percent. Mexico On the whole, one message that emerges from this Colombia analysis is that even though taxes and transfers can comple- Honduras R.B. de Venezuela ment growth in Latin American development strategies, Peru assuming that this instrument can substitute for growth to Argentina reduce poverty in the medium run seems unrealistic. Thus Brazil Costa Rica policies that address the evolution of market income in Panama terms of growth and its distribution will have to be central Dominican Rep. to the development strategies of the region. Chile Uruguay Concluding remarks 0 2 4 6 8 10 12 14 16 18 20 Percent In this chapter we have explored a number of issues of par- ticular interest for policy makers preparing poverty reduc- Source: Gasparini, Gutierrez, and Tornarolli (2005). Note: This projection assumes a 3 percent growth rate. tion strategies. First, we have argued that there are several pro-growth areas where Latin America needs to make that would appear from strategies based both on growth progress and where there may be potential trade-offs with and on improvements in the distribution of income inequality and even with poverty reduction goals in the through taxes and transfers. The figure reports the esti- short run. For example, several studies have found that mated incremental tax rates that would be needed to cut trade openness (an area of particular relevance given ongoing 100 P R O - P O O R G R O W T H I N L AT I N A M E R I C A liberalization efforts in the region) may lead to higher capital-intensive activities, stiff labor markets) are inequality through greater divergence of wage incomes. To removed. a large extent this result may be related to the very desir- Finally, the chapter explored the extent to which poli- able adoption of technologies that tend to be skill biased, cies aimed at reducing poverty through market incomes thus enhancing the returns to and the demand for educa- must be complemented with taxes and transfers. It con- tion, a phenomenon found globally. Nonetheless, the poor cluded that achieving a more redistributive and efficient and poor regions might be left behind in the short run. In pattern of public expenditures along OECD patterns would the long run, however, the evidence presented in this chap- significantly reduce poverty and inequality. Given the cen- ter suggests that all pro-growth policies will lead to lower trality of growth to the goal of poverty reduction, however, poverty regardless of their impact on inequality. policy makers may wish to ensure that efforts on that front We also argued that these results indicate that govern- have impacts favorable to growth. That would imply deal- ments may need to adopt complementary policies behind ing first with some of the shortcomings in public spending, the border--facilitating access to education, expanding including the regressive nature of some big-ticket items infrastructure to lagging areas with the potential to tap such as tertiary education, subsidies to electricity, and pen- into the benefits of liberalization, and offering conditional sions. It is worth stressing once more that the highest level transfers for poor peasants who may lose out in the transi- of targeting toward the poor comes from social assistance tion. These complementary policies can significantly miti- programs, especially conditional cash transfer programs, gate the inequality effects while considerably enhancing which in addition to ranking among the most progressive the growth effects, permitting the country to take full in Latin America, combine a transfer with the condition of advantage of the opportunities brought about by trade engaging in the accrual of human capital. Finally, on the opening. A parallel argument could be made based on con- tax front, first items in the agenda would be strengthening cerns that greater trade openness will increase the risk that anti-tax evasion programs and addressing the existing high workers face. Although little evidence has emerged to sug- level of exemptions. gest that this is true, were it the case, income support pro- grams could mitigate the impact on poverty and the Annex 5A disincentive effects on human capital accumulation. Another question explored in the chapter is whether dif- ferences in sectoral growth affect the impact that growth Simulating the impact of pro-growth policies on has on poverty reduction. We concluded that the composi- poverty tion of growth does matter for poverty reduction, and we The empirical models used to asses the impact of pro- stressed that policies that induce a sectoral bias in growth growth policies on poverty take the following form: yit - may conflict in the long run with pursuit of a country's nat- yit-1= yit -1 + xit + i + t + it, and git - git -1 = git -1 + ural comparative advantage, leading to growth-impeding xit + µi + t + it, where y is the log of per capita income, inefficiencies. That is, policy makers aiming at biasing g is the log of the Gini coefficient, x represents the set of growth toward sectors with a high growth elasticity of explanatory variables other than the lagged measure of poverty may have to face a trade-off between a high growth income or inequality, and µ are unobserved country- elasticity of poverty and higher growth. specific effects, and are time-specific effects, and and Another matter is to make sure that policy biases and are the error terms. The subscripts i and t represent coun- inefficiencies against, for example, rural development are try and time period. lifted and that growth opportunities are enhanced by the Beyond expressing the impact that the coefficients of the efficient provision of public goods and national and sectoral different policies may have on growth and inequality, these "innovation" policies. Incomes of the poor, including those models can be employed to obtain estimates of how poverty from agriculture and off-farm activities, thrive with higher changes would be associated with a change of x in policy j. trade openness, when rural expenditures focus on the provi- The presence of dynamics allows us to differentiate sion of public goods (rural roads, health and education, between the immediate impact that a change in a given research and development, extension services), and when policy has on both income and inequality and the long-run policy biases against labor mobility (fiscal generosity for impact that results from the dynamic feedback. For example, 101 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S changes to policy j will lead in the short run to: tor and that the sector thus contributes only slightly to poverty reduction. dp (5A.1) 3. http://www.econ.cam.ac.uk/dae/mu/emodstats/index.htm dxj = j × + j × , 4. Excluding the social security contributions does not change where p is the log of the poverty measure. 8 In the long run much the results. these changes will lead to:9 5. As noted in the text, the Gini coefficient of disposable income for the EU15 is 0.33 (about 20 percentage points lower than that of dpL R j j (5A.2) × . Latin America). In contrast, the Gini coefficient of market incomes is dxj = - × - 0.47 (about 10 percentage points lower than that in Latin America when we assume that the Gini of market income inequality is cut by Clearly, if the dynamics in the original models are similar 4 percentage points through taxes and transfers). Thus overall, of the (that is, if is similar to ), then equation 5A.1 reduces to 20-percentage-point difference in the Ginis of disposable income, 5A.2 scaled up to = . But if one of the variables adjusts 10 percentage points are attributable to higher market income much faster than the other, one should also expect to find inequality and the rest to the role of the government interventions. dynamics in poverty. In 5A.1 and 5A.2, and are the Clearly, estimates of market income Gini coefficients that are less growth and inequality elasticities of poverty that can be than 4 percentage points above disposable income Ginis for Latin America would imply an even higher relevance of taxes and transfers. obtained from assuming that income follows a lognormal For example, if, for the region as a whole, taxes and transfer lowered distribution. the Gini only 1.3 percentage points (the average for the countries in figure 9), then taxes and transfers would account for about two-thirds Notes of the differences in disposable income inequality levels between 1. There are several mechanisms through which the development Europe and Latin America. of credit markets might affect child labor. At the household level, 6. Although the deviation from the predicted value is smaller in credit constraints can prevent households from optimally trading off the case of the property tax, the volume of the property tax tends to a child's contribution to current household income against future be much smaller than the volume of the goods and services tax. returns from her schooling. In particular, households may resort to 7. The evidence in Lindert, Skoufias, and Shapiro (2005) indi- child labor to smooth transitory income shocks. Credit markets also cates that the richest quintiles of the population tend to receive a potentially affect the demand for child labor through their impact on higher share of total social insurance spending, whereas in general the firms' development. poorest quintiles receive a higher share of social assistance. 2. From an accounting point of view, it is likely that growth in 8. Strictly speaking, one should also consider an error term bigger sectors of economic activity has a larger impact on poverty emerging from using a discrete approximation to an infinitesimal reduction than growth in smaller sectors. Intuitively, if a sector interval. accounts for a small share of economic activity, then it is likely that 9. This assumes that 0 and 0. If the parameter controlling few people (both poor and nonpoor) benefit from growth in that sec- the dynamics is 0, all the adjustment would take place immediately. 102 CHAPTER 6 Does Poverty Matter for Growth? There is ample evidence that growth reduces poverty. This justifies having a pro-growth package at the heart of any poverty reduction strategy. However, is it also the case that poverty reduction is good for growth? Is there a possibility of entering a virtuous circle by which growth lowers poverty and in turn lower poverty results in faster growth? T HE PREVIOUS CHAPTERS HAVE EXPLORED dent in poverty. This in fact may be the root problem the link between growth and poverty by because as some development practitioners argue, existing focusing on the poverty-reducing effect of global poverty levels are probably more related to the insuf- growth and the factors that shape it. It was ficient growth experienced by developing countries over argued that in poorer and more equal coun- the past decades than to particularly anomalous patterns of tries, development strategies aimed at poverty reduction growth. Today the annual median per capita income in should emphasize growth. As countries become richer or developing countries is $3,000, a figure that indicates only more unequal, however, policy makers should try to bal- modest progress since 1975, when the median income level ance growth and distribution concerns because in those cir- was about $2,500. Over this same time period, median per cumstances poverty may be much more sensitive to capita income in developed countries increased from about changes in relative incomes than to changes in mean $15,000 to more than $25,000. income.1 We also addressed whether the pattern of growth Against this background and given that the achieve- associated with specific policies and sectors is more pro- ment of growth--any type of growth--is a big challenge in poor in some circumstances than in others. We concluded itself, should a discussion on growth and poverty reduction, that even though over long-run horizons most pro-growth or pro-poor growth, focus first on how to achieve growth policies will also be pro-poor (in the sense that the poor and only then consider how to ensure that its pattern is receive some benefit from the particular policy), in princi- pro-poor? This chapter argues that, on the contrary, the dis- ple one can expect that growth will have differing effects on appointing growth performance of developing countries poverty in the short run depending on the policies with makes the growth-poverty link even more critical. Not which it is associated. only does low growth mean that even small deteriorations A debate on the pro-poorness of a particular pattern of in income inequality may lead to higher poverty (see Cord, growth can be very appealing from an intellectual viewpoint Lopez, and Page 2005 for a discussion). It also means that but of little practical relevance if there is no growth--of poverty per se may be a barrier to growth, as suggested by any type--to start with or if growth is too low to make a several theoretical models developed in the economics This chapter relies heavily on the background paper "Too Poor to Grow," prepared for this report by H. Lopez and L. Servén (2005b). 103 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S literature. In other words, countries do not grow fast the evidence presented here suggests a bimodal income dis- because they are too poor to grow. This direction of causal- tribution, with countries showing a tendency to cluster ity from poverty to growth in turn opens the door to the around either a high-level efficient equilibrium or a low- existence of poverty traps, in which poverty and growth level inefficient equilibrium. This clustering is consistent interact in a vicious circle where high poverty leads to low with one of the predictions of poverty-traps models. growth and low growth in turns leads to high poverty. Of particular interest here are the findings for the cross- The theoretical appeal of poverty-traps models is clear: country distribution of incomes in the Latin American these models explain a number of stylized facts on the region. In contrast to the global data, this distribution growth-poverty link (such as the disappointing growth appears to be roughly unimodal, implying that most Latin performance of developing countries relative to the devel- American countries belong to the same convergence club oped world or the existence of convergence clubs2) for and thus share the same dynamics of the development which the traditional neoclassical growth model is inappro- process in the region. When we also ask to which country priate. Beyond the theoretical appeal, however, several cluster the region belongs--the rich or the poor--the aspects related to the poverty-traps view of the develop- results are mixed. On the one hand, it is difficult to argue ment process are likely to have important policy implica- that the region is stuck in the low, inefficient equilibrium tions. First, at a strategic level, the existence of poverty (although admittedly some weak evidence suggests that a traps should mitigate the debate on whether development few countries in the region--namely, Bolivia, Honduras, strategies should rely more on pro-growth or pro-poor and Nicaragua--could be trapped in the poor-countries policies, because strategies that do not take into account club).3 On the other hand, the region does not seem to the bidirectional relation between poverty and growth will belong to the rich-countries club either. On the whole, the likely lead to disappointing results: poverty will not region would be better described as in an intermediate decline without growth, but growth will be difficult unless state somewhere between the very poor and the very rich. the constraints affecting the poor are also addressed. Sec- Finally, the chapter presents new empirical evidence sug- ond, if a country is trapped in a bad equilibrium, then mar- gesting that poverty deters investment and growth, espe- ket policies may not be enough to break the vicious circle cially where the degree of financial development is limited. between poverty and growth, and policies that change the This result appears consistent with stylized theoretical state of development may be needed. In this regard, country- models in which financial market imperfections prevent the specific analytical work that blends growth and poverty poor from taking advantage of their investment opportuni- analyses into a single entity and tries to uncover the poten- ties, and it suggests a particular mechanism through which tial complex set of interactions operating in a given country poverty affects growth. Admittedly, this mechanism is not would be a first step toward determining exactly which necessarily exclusive; moreover, there are other channels, policies are needed to break the poverty trap. Third, at a such as education, health, and innovation, through which more operational level, one implication of the potential high poverty can potentially feed back into lower growth existence of poverty traps is that the biggest payoff to rates. In any case, we emphasize here that this chapter, and growth (and hence to poverty reduction) would likely more generally this report, does not aim at setting the result from policies that not only promote growth but also debate on the existence of poverty traps (defined as the exis- exert an independent, direct impact on poverty--thereby tence of multiple steady states); admittedly the empirical reducing the drag of poverty on growth. evidence on this question is mixed at best. Instead, its main This chapter elaborates on these issues. It motivates the concern is whether the empirical evidence supports a weaker discussion by briefly reviewing arguments put forward in version of the predictions derived from poverty-traps mod- the literature suggesting how poverty can become self- els, namely, that poverty tends to hold back growth. reinforcing and potentially lead to multiple equilibriums. The chapter then examines the empirical evidence on the A poverty-traps view of the development process dynamics of per capita income. First, it reviews the recent The past few years have witnessed the emergence of a growth experience in the developed and developing worlds, booming theoretical literature aimed at explaining why concluding that the developing world has underperformed poverty may be self-reinforcing and therefore why coun- systematically relative to the developed countries. In fact, tries that start out being poor continue to be persistently 104 D O E S P O V E RT Y M AT T E R F O R G R O W T H ? FIGURE 6.1 FIGURE6.3 Traditional view of the growth-poverty relationship Multiple equilibriums in the presence of increasing returns to scale a. Growth model with decreasing returns to scale Country characteristics, Pattern Poverty 0.8 institutions, and of growth policies 0.7 d k 0.6 0.5 s y 0.4 FIGURE 6.2 Poverty-traps view of the growth-poverty relationship 0.3 0.2 0.1 k Country 0 characteristics, Poverty 0 2 4 6 8 institutions, and policies Percapitacapitalstock b. Growth model with increasing returns to scale 0.8 0.7 Pattern 0.6 of growth 0.5 0.4 s y 0.3 poor over the long run (see Azariadis and Stachurski 2005 0.2 d k for a survey). In the traditional view of development 0.1 (presented schematically in figure 6.1), country constraints kL k kH 0 (institutions, policies, internal and external shocks, and the 0 2 4 6 8 like) are considered to be largely exogenous (that is, they Percapitacapitalstock are not determined within the system). In contrast, the Source: Authors. poverty-traps literature stresses the possibility that poverty has feedback effects on growth, a dynamic that has the potential to create poverty traps and that results in a very different picture of the development process (figure 6.2). stock (d × k). If, however, the production function experi- One critical difference between the two development ences a technological jump (discussed in more detail later), views is that in the poverty-traps view, different equilibri- there would be two steady states, and countries would tend ums may exist that are stable and self-reinforcing so that the toward one or the other equilibrium depending on their initially poor may stay poor and the initially rich stay rich. initial position. The lower equilibrium could be thought of Figure 6.3 illustrates this point, comparing the results of the as a poverty trap. Countries with capital below kL would standard neoclassical growth model with decreasing returns initially grow and converge toward the steady-state kL. to scale (panel A) with a model that exhibits increasing Countries between kH and k would converge toward kH. returns to scale (panel B). In the case of the standard neoclas- Thus initially poor countries would converge toward the sical growth model, the equilibrium is uniquely determined low, inefficient equilibrium whereas initially rich countries by the intersection of per capita savings and investment would tend toward the high, efficient equilibrium, produc- (s × y) with the rate of depreciation of the per capita capital ing a bimodal cross-country distribution of income. 105 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S In these circumstances policies aimed at eliminating investment and then repay the loan out of the returns of the market distortions that prevent the economy from moving investment. toward its equilibrium may be highly effective at achieving However, in real life--and especially in developing their objective. The problem is that the economy may be countries--capital and financial markets are plagued with headed toward an inefficient equilibrium. Thus poverty- imperfections. In many economies large segments of the traps models have the ability to explain both why poor population may not have access to credit at all. In some economies may have a tendency to underperform richer cases, access to credit is denied because the poor do not economies and why the benefits of good policies fail to have the necessary collateral. In other cases, financial opera- materialize. What are the mechanisms that lead to this tors may find it difficult to enforce contracts, and an indi- type of feedback from poverty to growth? Several channels, vidual's access to credit will likely be constrained by his or typically in the form of departures from the basic neoclassi- her initial wealth; those with low or no initial wealth may cal model, have been explored in the literature.4 We briefly be excluded from capital markets. Moreover, even those discuss three of those channels here. with access to credit may encounter significant constraints. Since deposit rates tend to be much lower than borrowing Increasing returns to scale and poverty traps rates (figure 6.4), the opportunity cost of capital is lower for As suggested earlier, one mechanism that may potentially those who need to borrow less. For example, the average lead to poverty traps is the existence of increasing returns interest rate spread (lending minus deposit) for 2003 in the to scale (this is the issue illustrated in panel B of fig- ure 6.3). Increasing returns may appear when the adoption of newer and more efficient technologies has an associated fixed cost. For example, Murphy, Shleifer, and Vishny FIGURE 6.4 (1989) argue that even if modern technologies are freely Interest rate spreads in Latin America, 2003 available to poor countries, when the size of the domestic Argentina market is small relative to the fixed costs required to adopt Bolivia the new, more efficient technology, firms may not have the Brazil right incentive to do so. As a result, initially richer Chile economies may enter a virtuous circle, whereas initially poorer economies may end up stuck with less-efficient Colombia technologies and lower income levels. Increasing returns Costa Rica may also appear in the presence of complementary produc- Dominican Rep. tion processes that act as an incentive for firms to match Ecuador workers of similar skills, in which case the incentive to Guatemala educate increases as the initial pool of skilled workers Haiti increases (Kremer 1993). Honduras Jamaica Market failures and poverty traps Mexico A second mechanism that may generate poverty traps is Nicaragua related to the existence of potential market imperfections Panama in credit and insurance markets. With perfect capital mar- Paraguay kets, investment decisions in physical or human capital Peru depend on the expected returns (probably adjusted by R.B. de Venezuela risk) of the investment and on the associated cost. When 0 5 10 15 20 25 30 35 40 45 50 the returns are higher than the cost of capital, an individ- Percentage points ual would have the same incentive to invest regardless of Source: WDI database. his or her initial income level: theoretically, poor people Note: The figure reports lending minus deposit rates. could always borrow the capital they need to make the 106 D O E S P O V E RT Y M AT T E R F O R G R O W T H ? sample of Latin American countries included in figure 6.4 For example, Engerman and Sokoloff (forthcoming) argue is about 10 percentage points, but in specific countries that institutions that place economic opportunities beyond (Brazil and Paraguay), it is more than 30 points. Thus, if the reach of broad segments of society are likely to result in both a rich and a poor person face a similar rate of return on reduced growth rates because modern economies require a project, it is likely that the rich person will invest much broad participation in entrepreneurship and innovation. In more than the poor person. In other words, the opportuni- addition, a natural tendency for those who hold power to ties and costs of borrowing can be very different for rich try to perpetuate that power results in path dependence and and poor people and play against the latter group. persistence for the institutional framework. These two ele- Imperfect capital markets coupled with fixed costs ments together help explain the tendency for poverty and imply that important segments of the population are bad institutional arrangements to coexist and persist over excluded from investment opportunities. For example, time. Banerjee and Newman (1994) stress the effect that an indi- Similarly, Mauro (2002) considers low economic growth vidual's initial wealth has on the level of physical invest- in countries with persistent corruption and notes that ment when there are credit constraints. Thus high poverty some countries appear to be stuck in a bad equilibrium rates might result in low investment rates and hence in characterized by pervasive corruption with no sign of lower growth. improvement. He argues that one reason why rooting out Galor and Zeira (1993) make a similar argument. They widespread corruption is so difficult may be that it just note that people at the bottom of the income distribution does not make sense for individuals to attempt to fight it, may not be able to cover the expense of education or access even if everybody would be better off if corruption were to the financial sector to borrow for that purpose. In this case be eliminated. For example, if corruption is widespread in high poverty rates result in low educational outcomes an administration, civil servants might find it difficult to because poor individuals likely opt out of the education sec- decline bribes in exchange for favors because their superiors tor and work at unskilled, low-return labor. Note that this may expect a portion of the bribe for themselves. In con- effect goes beyond the lower supply of education possibili- trast, in bureaucracies that are generally honest, a real ties in poorer countries and focuses on the demand side. As threat of punishment deters individual civil servants from argued in de Ferranti and others (2003), education levels are behaving dishonestly. This is an example of a strategic a vital complement for technological advance and are thus a complementarity, whereby if one agent does something it critical element in understanding growth rates (box 6.1). becomes more profitable for another agent to do the same thing. The tendency of corruption to persist, together with Institutional mechanisms and poverty traps the negative impact of corruption on growth (Mauro The theoretical literature also stresses the role played by 1995), would then explain why some countries may be the institutional framework in generating poverty traps. caught in inefficient equilibriums. BOX 6.1 Education and technology Productivity differences between countries and between of the successful natural resource­based economies. firms within countries are profoundly affected by differ- Within Latin America, the best-performing country, ences in skills and technology. It is therefore no surprise Chile, concurrently had positive increases in productiv- that the East Asian tigers--Hong Kong (China), Repub- ity, substantial skill upgrading, and increases in all lic of Korea, Singapore, and Taiwan (China)--which indicators associated with technology transfer and inno- exhibit well-above-average rates of total factor produc- vation. tivity growth, also outperform Latin America on mea- sures of technology and skills. The same is true for some Source: de Ferranti and others 2003. 107 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S In summary, a variety of mechanisms that typically do Over the 1963­2003 period, median per capita growth in not fit the assumptions underlying the neoclassical model industrial countries has outpaced median growth in devel- may both cause poverty and perpetuate it over the long oping countries by an average of more than 1 percent a run. Moreover, many of these mechanisms may well inter- year.6 Moreover, there are two extended periods of time-- act with and reinforce each other. For example, corruption the 1960s and early 1970s, and the mid- to late 1980s-- may exacerbate credit access problems if the public sector where the differences are consistently in the range of subsidizes or guarantees credit to some privileged groups 2 percent a year. in society at the expense of poorer segments of the Latin America does not seem to be an exception among population. Similarly, institutional frameworks with weak developing countries; the growth performance of the region enforcement of the rule of law may discourage investment over the 40-year period was fairly consistent with the in sectors where intellectual property rights have a high performance observed in other developing countries. The value for the firm. That in turn can lower the demand for differences between Latin America and all developing coun- skilled workers and hence the incentives for individual tries were notable for three periods: the early 1980s, when workers to invest in skill acquisition. The next section Latin America was badly hit by the debt crisis and recorded reviews some existing empirical evidence on the practical median growth rates below -1 percent; the early 1990s, relevance of these models. when the region did much better than the rest of the devel- oping countries; and the late 1990s, when once again Latin Empirical evidence on poverty traps America experienced a significant deceleration following the Over the last decades, the world has become increasingly financial crises in East Asia in 1997 and in Russia in 1998. divided into two clubs--one of rich countries, the other of The underperformance of the developing world relative poor countries. Figure 6.5 plots median per capita growth to the developed world appears even more dramatic when rates for industrial and developing countries between 1963 one looks at the evolution of median per capita income lev- and 2003.5 It also plots median per capita growth rates for els over time (figure 6.6). Because developing countries Latin America. The figure indicates that, apart from one have been experiencing lower growth rates for prolonged short period in the second half of the 1970s and another in periods of time, the gap between the per capita income the early 2000s, the typical developing country has experi- levels of rich and poor countries has been steadily increas- enced lower growth rates than the typical rich country. ing. In the early 1960s, the income level of the median FIGURE 6.5 FIGURE 6.6 Growth in developed (OECD) and developing countries, 1963­2000 Income in Latin America relative to the OECD countries, 1960­2002 Percent Income relative to OECD 5 0.35 4 0.30 3 0.25 2 0.20 1 0.15 0 0.10 1 2 0.05 2 5 0 84 0 1963 1966 1969 197 197 1978 1981 19 1987 199 1993 1996 1999 2002 1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 Developing countries Latin America Developed countries Latin America Developing countries Source: WDI database. Source: Authors' calculations. Note: The chart reports the 3-year moving average of the median per capita growth for each group of countries. 108 D O E S P O V E RT Y M AT T E R F O R G R O W T H ? developed country was six times greater than the income very significant. Take the case of Argentina, the richest level of the median developing country; today income in country of the region in 1960 with a per capita income the median developed country is close to nine times greater level that was close to the level of industrial countries (representing a 50 percent increase in the gap). More dra- (85 percent). Forty years later Argentina's relative income matically, in 1960 the income of the richest country at the has declined to 43 percent of the industrial countries' level. time, Switzerland, was about 50 times the income of the Similarly, the relative per capita income of Nicaragua has poorest country, Malawi. Today, the richest country is declined from 49 percent in 1960 to about 12 percent in Luxembourg, which has a per capita income level that in 2000. Today three countries in Latin America (Bolivia, purchasing power parity is almost 120 times that of Sierra Haiti, and Honduras) have PPP-adjusted per capita GDP Leone, now the poorest country. levels that are less than 10 percent of the income of the The use of the median as a summary statistic is some- developed countries. In 1960 no country in the region had what limited because it does not show the significant het- a relative income level below 20 percent. erogeneity that exists at the country level. Yet, even if we On the whole, this evidence is at odds with the conver- focus on the evolution of income on a country-by-country gence predictions of the simple neoclassical model and basis (table 6.1), the majority of the Latin American coun- instead is more consistent with what World Bank tries (the exception is the Dominican Republic) have economist Lant Pritchett (1997) refers to as "divergence income levels today that are lower than they were in 1960 big time": "Whichever way the debate about whether relative to the income of OECD countries. Not only have there has been some `conditional' convergence in the recent the majority of Latin American countries lost ground over period is settled, the fact remains that one overwhelming the past 25 years but in some cases the decline has been feature of the period of modem economic growth is massive TABLE 6.1 Median income in Latin America and the Caribbean relative to the industrial countries Country 1960 1970 1980 1990 1998 2003 Argentina 0.85 0.72 0.64 0.40 0.52 0.43 Bolivia 0.22 0.15 0.14 0.10 0.10 0.09 Brazil 0.30 0.28 0.38 0.29 0.29 0.27 Chile 0.37 0.30 0.26 0.26 0.36 0.36 Colombia 0.32 0.27 0.27 0.27 0.25 0.24 Costa Rica 0.46 0.37 0.38 0.29 0.32 0.34 Dominican Republic 0.21 0.18 0.22 0.19 0.22 0.24 Ecuador 0.22 0.16 0.19 0.16 0.14 0.13 El Salvador 0.38 0.32 0.24 0.17 0.18 0.17 Guatemala 0.26 0.22 0.23 0.16 0.15 0.15 Guyana 0.30 0.23 0.20 0.16 0.17 0.15 Haiti 0.31 0.18 0.18 0.11 0.07 0.06 Honduras 0.20 0.15 0.15 0.11 0.10 0.09 Jamaica 0.29 0.28 0.19 0.18 0.15 0.14 Mexico 0.42 0.39 0.44 0.34 0.33 0.32 Nicaragua 0.49 0.46 0.26 0.13 0.12 0.12 Panama 0.26 0.28 0.29 0.21 0.24 0.24 Paraguay 0.25 0.21 0.27 0.21 0.19 0.17 Peru 0.41 0.35 0.30 0.18 0.19 0.19 Trinidad and Tobago 0.49 0.46 0.53 0.32 0.32 0.38 Uruguay 0.62 0.43 0.43 0.33 0.37 0.29 Venezuela, R.B. de 0.69 0.54 0.38 0.26 0.25 0.17 Latin America 0.31 0.28 0.26 0.19 0.21 0.19 Source: Authors' calculations using GDP per capita ($2,000 PPP) from the World Development Indicators for various years. Data before 1975 has been computed using available per capita growth rates for the period 1960­75 and the per capita GDP level of 1975. 109 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S divergence of absolute and relative incomes across be viewed as a single entity? We now address these ques- countries, a fact which must be grappled with in a fully tions in turn. satisfactory model of economic growth and development." Convergence clubs in absolute income levels Convergence clubs The first question concerns the dynamics of cross-national What explains this apparent divergence between developed per capita income levels and the existence of convergence and developing countries? Could it be attributable to the clubs. Panel a of figure 6.7 presents the histograms of per existence of multiple states of development toward which capita income for 1960 and 1999 computed for 102 coun- different countries converge, creating convergence clubs? If tries using data from the Penn World Table (PWT6.1). so, where is the Latin American region in this picture? Are The histograms suggest that whereas in the early 1960s there also regional convergence clubs that will result in the distribution of income appeared to be unimodal in the regional clusters of development or, instead, can the region early 1960s, by the late 1990s it had become trimodal, FIGURE 6.7 Histograms for per capita income, 1960s versus the 1990s Panel a 1960 1999 Number of observations Number of observations 25 25 20 20 15 15 10 10 5 5 0 0 0.4 0.7 1.1 1.8 3 5 8 13 22 0.4 0.7 1.1 1.8 3 5 8 13 22 35 60 Per capita income, US$ PPP Per capita income, US$ PPP Panel b 1960 1999 Number of observations Number of observations 18 18 15 15 12 12 9 9 6 6 3 3 0 0 0.4 0.7 1.1 1.8 3 5 8 13 22 0.4 1.1 3 8 22 60 Per capita income, US$ PPP Per capita income, US$ PPP Low-low Low-high High-high Source: Penn World Tables (PWT) 6.1. Note: The top panel reports the histograms of the cross country per capita income distribution (102 countries) in 1960 and 1999. The bottom panel presents the transitions of three groups of countries: low-low shows countries that in both 1960 and 1999 had per capita income levels below $3,400; high-high shows countries that in both 1960 and in 1999 had per capita income levels above $3,400; low-high shows countries that in 1960 were below $3,400 and in 1999 were above $3,400. 110 D O E S P O V E RT Y M AT T E R F O R G R O W T H ? with a low peak at $1,100; a second peak between $5,000 especially if the groups are in a transition toward a steady and $8,000, and a third peak around $35,000.7 state. Where, then, is each of these groups heading? The In panel b we attempt to discriminate between conver- annex to this chapter discusses a simple procedure that can gence clubs and present the histograms for three groups of be used to estimate the steady state for each group. Imple- countries. Here we follow an approach similar to the one mentation of this procedure suggests convergence but to used by Mayer-Foulkes (2003) in his study of convergence three dramatically different steady states. For the low-low clubs in life expectancy and divide the sample into four group, the estimated equilibrium for per capita income is groups. The first group includes those countries whose per around $1,700. For the low-high group, the equilibrium is capita income levels were below $3,400 in both 1960 and around $11,000, and for the high-high group, the point 1999. This is the per capita income level of the poorest estimates suggest an equilibrium well above current levels. industrial country in 1960 (Portugal) and is very close to How does the Latin American region fare in this con- the observed peak in 1960. We refer to that group as low- text? Is the apparent bi- or trimodality of the world distri- low. The second group includes countries with per capita bution also observed in the region, or do all the countries in income levels above $3,400 in both 1960 and 1999. This the region belong to a single cluster? To answer these ques- is the high-high group. The third group (low-high) com- tions, figure 6.7 plots a histogram similar to the one in prises countries with per capita income levels below $3,400 panel A of figure 6.6 but restricts the sample to Latin in 1960 and above $3,400 in 1999. No country falls in the American countries. In contrast to the full sample, the esti- fourth group, which notionally corresponds to a high-low mated cross-country distributions of per capita income for group, and the numbers of countries in each of the other Latin America appear to be unimodal for both the early three groups are quite balanced. 1960s and the late 1990s. The peak in 1960 is around Panel b shows three markedly different behaviors. The $3,000, which is fully consistent with the global data. The initially rich countries present the highest per capita peak in 1999 is around $8,000, which implies average growth rates. The median income of the high-high club annual growth in the 2.5 percent range, approximately increased from about $7,500 in 1960 to about $22,000 in halfway between the growth rates for the global high-high 1999 (table 6.2). The transition countries (the low-high and low-high groups. group) also show considerable growth (from a median How do we interpret these results? Well, it depends on income of about $2,400 in 1960 to about $5,400 in 1999), whether we see the glass as half full or half empty. As a but the average annual growth rate is lower than in the half-full glass, it seems difficult to argue that the region is high-high group by almost 0.7 percentage point. Finally, stuck in the low, inefficient equilibrium (the one corre- the low-low group shows very low growth. The median sponding to the equilibrium around $1,700). More likely, income for the 37 countries in this group increased from taking into account the initial starting point and the evolu- about $1,050 in 1960 to just $1,300 in 1999, which tion of income levels over the 1960­99 period, the region implies an average annual increase of about half a percent. is better characterized as belonging to the low-high transi- Clearly, the peaks in the histogram for 1999 may not tion group (for which the estimated equilibrium for correspond to the equilibriums for the different groups, income per capita is in the $11,000 range). As a half-empty glass, the region does not seem to belong to the high-high equilibrium either. On the whole, the region would be TABLE 6.2 better described by an intermediate state somewhere in Median income of convergence clubs between the very poor and the very rich. One issue needs to be highlighted before we continue, Median income however. Careful observation of figure 6.8 indicates that Annual Club Countries 1960 1999 increase (%) the dispersion of regional income in 1999 is significantly higher than it was in 1960. This results from the relatively Low-low 37 1,046 1,277 0.51 good performance of some of the economies that were Low-high 33 2,395 5,442 2.13 High-high 32 7,417 21,632 2.78 richer to begin with (Chile, Mexico, and Uruguay) and the modest performance of some of the poorer economies Source: Authors' calculations. (Bolivia, Honduras, and Nicaragua), which initially 111 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 6.8 Histograms for per capita income in Latin America, 1960s versus the 1990s 1960 1999 Number of observations Number of observations 10 10 8 8 6 6 4 4 2 2 0 0 0.4 0.7 1.1 1.8 3 5 8 13 22 0.4 0.7 1.1 1.8 3 5 8 13 22 Per capita income, US$ PPP Per capita income, US$ PPP Source: Penn World Tables (PWT) 6.1. Note: The figure reports the histograms of the cross-country per capita income distribution (18 countries) for the Latin American region in 1960 and 1999. experienced average annual growth rates below 0.5 percent FIGURE 6.9 (Nicaragua's average annual growth rate was in negative Twin peaks territory). At least three countries in the region appear to have a performance that is more consistent with that Frequency 0.5 observed for the low-low group in figure 6.7, and these Expanded data countries could potentially be trapped in the low equilib- 0.4 Quah (1993) rium. In other words, behind figure 6.8 there could be a 0.3 bimodal distribution, with a second steady state toward the 0.2 lower end of the distribution that is not apparent because 0.1 the associated probability mass is very low (that is, because 0 only a few countries belong to that group). 1 2 3 4 5 State Convergence clubs in relative incomes Source: Quah (1993) and authors' calculations. An alternative way to look at the cross-national distribu- tion of income is based on an analysis of relative income levels and on the probability that a country moves between between the world average and twice the average, and those states of development. In the technical annex to this chap- with incomes above twice the average, respectively. ter, we review some methodological details and present Figure 6.9 plots the equilibrium as computed by Quah some empirical results that can be used to estimate equilib- (1993) on the basis of data spanning 1962­84, and it also rium values for the distribution of income. Figure 6.9 plots the equilibrium that results when the analysis is reports results for five states of relative development. In based on an expanded sample covering 1960­99. A num- state 1 are the poorest countries of the world: those with ber of interesting points are revealed in this figure. First, per capita income levels below 25 percent of average world both samples suggest the presence of convergence clubs at per capita income. In state 2 is a group of richer but still either end of the income distribution: there is a cluster of relatively poor countries: those with per capita income lev- poor countries around a low per capita income equilibrium els between 25 and 50 percent of average world per capita and a second cluster around the high per capita income income. State 3 includes economies that have income levels equilibrium (that is, the poor tend to stay poor and the rich between 50 percent and the world average. States 4 and 5 tend to stay rich). However, while the 1962­84 sample cover the richest countries: those with per capita incomes results in a picture of the world that is divided almost 112 D O E S P O V E RT Y M AT T E R F O R G R O W T H ? FIGURE 6.10 FIGURE 6.11 Equilibrium and distribution in 1999 Latin American states: One peak? Frequency Frequency 0.5 0.5 Distribution in 1999 0.4 Equilibrium 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0 0 1 2 3 4 5 1 2 3 4 5 State State Source: Authors' calculations. Source: Authors' calculations. symmetrically, the 1960­99 sample produces a distribu- estimated long equilibrium (figure 6.11). As in figure 6.8, tion that is clearly skewed toward the lower equilibrium the obtained results for the region do not show evidence of (that is, the cluster of poor countries has more members). bimodality. Instead, there seems to be a long-run equilib- In other words, while evidence of some type of bimodal- rium around state 3. The cross-country distribution of ity still exists, the expected long-run frequency of countries income, however, is not symmetrical, and long-run equilib- in the first state increases by almost 20 percentage points rium computed on the basis of the estimated transition (from 0.24 to 0.43) by expanding the sample. This finding matrix places 80 percent of Latin American countries in implies that the updated estimates predict more countries states 2 and 3; these are countries whose relative income falling behind (at least relative to the world average). This ranges from 25 percent of the world average to the world is further explored in figure 6.10, which compares the dis- average. tribution in 1999 to the estimated equilibrium. The figure These results are largely consistent with those of the suggests that unless there are changes in the transitional previous analysis and show the region on an equilibrium dynamics of the growth process, the number of countries in that is well below the world average. The estimates also the first state can be expected to increase. show a disturbing tendency for Latin American countries Unlike our previous analysis where the empirical evi- to cluster around the lower tail of the equilibrium. Here dence pointed toward a three-club characterization, this the only thing we can do is to speculate that a relatively body of evidence is more consistent with the existence of small group of countries in the region do not belong to the two convergence clubs. One is composed of very poor coun- state 3 equilibrium and instead converge around state 2. tries, apparently with loose rules of admission; on the basis of the data to 1999, more than 40 percent of the countries Convergence clubs in other dimensions of poverty belong to this club. The second club--the rich-countries So far we have focused on the cross-national distribution of club--is much more exclusive, and our estimates suggest per capita income. However, there is no reason to constrain that only about 20 percent of the countries belong to it. the analysis to the income dimension of welfare. Conver- (The remaining 40 percent of the countries lie somewhere gence clubs may also involve specific health phenomena. in between these two convergence clubs.) For example, the theory of efficiency wages in Dasgupta and The difference between having two or three clubs is key Ray (1986) implies the possibility of a low-productivity, for Latin America, given our earlier conclusion that the low-nutrition trap. Mayer-Foulkes (2003) argues that the region fell somewhere between the low and the high equi- existence of convergence clubs is also apparent in life- librium. To explore this issue, we replicate the previous expectancy dynamics. Figure 6.12 presents cross-national exercise but use data only for Latin America. The results life-expectancy histograms for 1960 and 2002. These his- suggest that there are important differences in the tograms indicate the presence of a two-peaked pattern in 113 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 6.12 Convergence clubs in life expectancy 1960 2002 Number of countries Number of countries 35 60 30 50 25 40 20 30 15 20 10 5 10 0 0 35 40 45 50 55 60 65 70 75 40 45 50 55 60 65 70 75 80 85 Life expectancy at birth, years Life expectancy at birth, years Source: Authors' calculations. both periods. It is also evident that the mass of the low With these ideas in mind, Bloom, Canning, and peak declines between 1960 and 2002, whereas the mass of Sevilla (2003) move beyond the pure description of the the high peak increases. These figures are basically a replica cross-national income distribution and find that the exis- of those in Mayer-Foulkes (2003) and indicate that the tence of twin peaks in the data is more likely attributable cross-country data on life expectancy are consistent with the to multiple equilibriums than to fundamental forces. This, presence of three convergence clubs (with a different num- in turn, supports the hypothesis that poverty traps with ber of members): one for the low equilibrium, one for the low and high equilibriums underlie the dynamics of per high equilibrium, and one for a third transitional group. capita income. An alternative way to determine the existence of poverty Formal tests of the poverty-traps hypothesis traps is to investigate specific sources of multiple equilibri- The empirical evidence discussed here is supportive of a ums. One such approach is the calibration of models consis- multimodal distribution in cross-national per capita tent with the poverty-trap hypothesis. Once a model has income levels, which is consistent with the predictions been calibrated, its empirical relevance can be assessed. For of poverty-traps models. However, as Azariadis and example, Graham and Temple (2004) calibrate a two-sector Stachurski (2006) argue, one has to be extremely careful to general equilibrium model and then explore the extent avoid taking these empirical findings as evidence of to which this model is able to explain the real data. The poverty-traps phenomena. In fact, in a recent study, Bloom, model considers a traditional agricultural sector with Canning, and Sevilla (2003) stress that a multimodal dis- diminishing returns and a nonagricultural sector with tribution in cross-country income levels is also consistent increasing returns (in the vein of our earlier discussion with the existence of fundamental differences between about poverty traps in the presence of increasing returns to countries that result in different but unique equilibriums scale). As it turns out, the degree of increasing returns is for each country. Thus, in principle one has to be able to one of the key parameters underlying the simulations, and determine whether bimodality results from two "similar" depending on its assumed value, the model can explain countries having completely "different" states of develop- between 15 and 60 percent of the variance of incomes. ment (that is, poverty traps) or from fundamental differ- The Graham and Temple analysis has the same limitations ences between the two countries. in the Latin American context, however. In particular, as 114 D O E S P O V E RT Y M AT T E R F O R G R O W T H ? the authors recognize, the model appears to explain the data for Russia and Hungary. They find no evidence to existing income differences between the low- and middle- support the poverty-traps hypothesis (although they do income countries better than it explains the differences find that the adjustment of income to shocks is nonlinear). between middle-income and developed countries. Thus Their results indicate that households tend to bounce back while the results they obtain offer some ideas of why from transient shocks, although the adjustment process is African countries are so poor, they have much less to say slower for poorer individuals. Jalan and Ravallion (2002) about the current positions of Latin America relative to the use household panel data from China, however, and find industrial countries. that aggregate physical and human capital endowments Kraay and Raddatz (2005) also calibrate simple aggre- play a significant role in household consumption growth, a gate models capable of generating poverty traps through finding that they argue is consistent with the existence of low savings or low technology at low levels of develop- regional poverty traps. ment.8 The basic idea behind these models is that if either On the whole, it must be acknowledged that the empir- the saving rate or productivity increases above a certain ical evidence on the existence of poverty traps is, at best, threshold of development, it would then be possible to find mixed. How then do we explain the existence of conver- poverty-trap-like features in the data. To assess the empiri- gence clubs alongside the relative lack of evidence on the cal relevance of these models, Kraay and Raddatz explore existence of poverty traps? One possibility is that poverty whether saving rates exhibit the sort of nonlinear relation- traps do exist and that the econometric models used to test ship implied by the model for the existence of poverty such hypotheses are unable to capture the dynamics behind traps, and whether scale effects on productivity are of a the data. An alternative possibility is that poverty traps do magnitude consistent with the theoretical model. Unlike not exist in the strict theoretical sense (multiple equilibri- Graham and Temple's findings, their results do not lend ums created, for example, by increasing returns to scale or much support to the existence of poverty traps based on any other mechanism), but that poverty is still a barrier to these mechanisms. In particular, their technology-based growth by which poorer countries find it more difficult to model suggests that for a poverty trap to exist, the esti- grow than richer countries. In this regard, Azariadis and mated returns to scale would have to be in the 1.4 to 2.5 Stachurski (2006) use a much more general definition and range. This interval is much higher than is typically found classify any self-reinforcing mechanism that causes poverty in the literature, where most studies report constant to to persist as a poverty trap. Note that with this alternative moderate increasing returns. definition in mind, the important question is not whether Another strand of the empirical poverty-traps literature the development process is characterized by the existence of has explored the existence of nonconvexities by exploit- multiple equilibriums but rather how persistent and self- ing existing microeconometric evidence. For example, reinforcing the mechanisms are that lock in poverty over McKenzie and Woodruff (2004) examine the empirical rel- time frames that matter from a policy perspective. But is evance of the assumptions that minimum start-up costs are there any empirical evidence suggesting that poverty may high relative to wealth and that returns to capital are low represent a barrier to growth? The next sections explore at low investment levels (see Banerjee and Newman 1993). this issue. Using microenterprise data for Mexico, McKenzie and Woodruff show that the median investment levels of new What is the empirical evidence on poverty's firms in some sectors are very low (about US$100, or less impact on growth? than half of the monthly earnings of even a low-wage The past few years have witnessed a renewed interest in worker). They also show that the marginal return to capital both the theoretical and the empirical relationship between is quite high even for low levels of invested capital (in the inequality and growth. At the theoretical level, two main $200 range), concluding that the Mexican evidence does types of arguments have been put forward: sociopolitical not support this particular mechanism as a candidate to economy arguments and credit constraint­factor accumula- justify the existence of poverty traps. tion arguments. Similarly, Lokshin and Ravallion (2004) test for the The sociopolitical economy arguments stress the role existence of a threshold effect in household incomes using that high inequality may play in the decisions of various 115 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S agents and how these decisions may influence growth. For The main results of that work are the following: example, Alesina and Rodrik (1994) suggest that high inequality may lead to lower growth if the level of taxation · Poverty has a consistently negative impact on growth has a negative impact on capital accumulation, if taxes are that is significant both statistically and economically. proportional to income but the benefits of public expendi- · This negative growth effect seems to work through ture accrue equally to all individuals (implying that an investment in the sense that high poverty deters individual's preferred levels of taxation and expenditure are investment, which in turn lowers growth. inversely related to her income), and if the tax rate selected · The data suggest that this mechanism operates only by the government is the one preferred by the median at low levels of financial development, consistent voter. Similarly, Alesina and Perotti (1996) argue that with the predictions of theoretical models that highly unequal societies create incentives for individuals to underscore financial market imperfections as a key engage in activities, such as crime, that are outside normal mechanism of poverty traps. markets and that sociopolitical instability discourages accumulation because of current disruptions and future We now review each of these findings in some detail. uncertainty. In both cases, high levels of inequality may lead to lower future growth. Poverty is bad for growth The credit constraint­factor accumulation argument Lopez and Servén (2005b) begin with the observation that emphasizes the possibility that some individuals will be if poverty hampers growth, then countries with higher excluded from the economic process because they have initial poverty should grow less rapidly than comparable neither the resources nor the means to borrow them to countries with lower initial poverty, all else being equal. engage in potentially profitable economic activities. For This hypothesis is a weaker version of the predictions example, as discussed earlier, Galor and Zeira (1993) argue derived from the analytical models on poverty traps, in that that the process of development is characterized by comple- to support it one does not need to find evidence of multiple mentarity between physical and human capital so that equilibriums but simply empirical proof that poverty tends growth increases as investment in human capital increases. to hold back growth. Using a standard growth model aug- However, credit constraints may prevent poorer individuals mented to include a suitable poverty measure among the from investing in education and thus affect growth explanatory variables, the authors find that after control- prospects by reducing the number of individuals who are ling for other factors, poverty has a negative and strongly able to invest in human capital. Similarly Aghion, Caroli, significant impact on growth, which is also economically and García-Peñalosa (1999) show that if there are decreasing significant. On average, a 10 percent increase in poverty returns to individual capital investments and if credit reduces annual growth by 1 percentage point. This finding imperfections mean that individual investments are an is robust to a number of basic departures from the basic increased function of initial endowments, then the concen- specification in Lopez and Servén (2005b),9 including: tration of investment in fewer richer people will negatively affect growth. · The use of alternative poverty lines. The estimated Admittedly for a given level of income, higher inequal- impact on growth of a change in headcount poverty ity will lead to higher poverty. But note that the credit con- is very similar regardless of the poverty line ($2, $3, straint­factor accumulation argument is more a poverty or $4 a day) used in the computation of the poverty argument than an inequality argument. Yet, to the best of index. Changes to the poverty line have an impact on our knowledge, the hypothesis that countries suffering the estimated coefficient of poverty of around 0.01. from higher levels of poverty grow less rapidly than those · The use of different sets of control variables. Changing countries with less poverty has remained untested. To fill controls seems to have only a moderate effect on the that gap, in a background paper for this report, Lopez and estimated negative impact of poverty on growth. Servén (2005b) make a first attempt to provide a direct Depending on the control set used, a 10 percent empirical assessment of the impact of poverty on growth increase in headcount poverty reduces growth (see the technical appendix). prospects by between 0.7 and 1.3 percent. 116 D O E S P O V E RT Y M AT T E R F O R G R O W T H ? · The use of different poverty measures (headcount, poverty · Adding inequality to the regression models. When gap, squared poverty gap). Changing the definition of inequality is added to the empirical models, the sign, poverty does affect the estimated coefficients of significance, and magnitude of the poverty effect poverty, which are not comparable across definitions. decline somewhat in absolute value, and the estimate However, the coefficients continue to be negative and is less accurate. It remains highly significant, how- statistically significant; in absolute value, the coeffi- ever, suggesting that the poverty variable does cap- cients of the poverty gap and square poverty gap tend ture a true poverty effect rather than an inequality to be larger than the coefficient corresponding to the effect. This result is also robust to adding inequality headcount definition. and squared inequality to control for the likely non- · The use of alternative estimation methods. One of the linear relation between poverty and inequality. problems dealing with highly persistent endogenous data is that the standard GMM estimation method In principle, the finding that poverty lowers growth does based on internal instruments may not fully elimi- not necessarily rule out the convergence of cross-national nate the potential reverse causality bias. To control incomes (conditional convergence in this case) predicted by for this problem, Lopez and Servén (2005b) also pre- the neoclassical model, but the empirical estimates in Lopez sent results based on cross-sections that should not and Servén (2005b) do imply the existence of a threshold suffer from reverse causality (although admittedly poverty level beyond which divergence would occur. For they may suffer from fixed-effects bias). The results example, with the baseline estimates in Lopez and Servén, also confirm the negative impact of poverty on there would be divergence for levels of the poverty head- growth. count (with a $2-per-day poverty line) above 10 percent. BOX 6.2 Is Latin America different? Although the Lopez and Servén (2005b) results do not for Latin America are always negative and significant explicitly consider whether the impact of poverty on (in other words, poverty would reduce growth more in growth varies by geographic region, extending the model Latin America than in the typical country of the world). to test this possibility is relatively simple. In fact, we The magnitude of the Latin American dummy declines have reestimated their basic models to allow Latin significantly in absolute value as the poverty line used in American poverty levels to have an impact on growth the computation of headcount poverty increases, from that is different from the average for the group (that is, 0.23 under a $2-a-day poverty line to about 0.10 we are allowing the Latin American region to be "differ- under a $4-a-day poverty line (although admittedly the ent"). The table below reports the results of this exercise. standard error in the former case is also much larger than This table suggests that Latin America may indeed be in the latter). different. In particular, the estimates of the coefficients Poverty and growth: Is Latin America different? Poverty line $2 a day $3 a day $4 a day All LAC dummy All LAC dummy All LAC dummy Parameter -0.114 -0.237 -0.128 -0.165 -0.140 -0.098 (0.02) (0.08) (0.02) (0.05) (0.02) (0.03) Source: Authors' calculations. 117 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S Transmissions channels from poverty to growth FIGURE6.13 What are the channels through which poverty might influ- Income, poverty, and investment ence growth? A quick review of the theoretical literature Panela suggests a number of potential channels including invest- ment, human capital (both education and health), innova- Income(logged) 10.5 tion and mobility, and risk. 10.0 9.5 Poverty and investment 9.0 8.5 Several theoretical models on poverty traps exploit the 8.0 result that high poverty levels (typically coupled with 7.5 7.0 credit constraints) are likely to affect the investment rate 6.5 negatively. But what do we actually know about the rela- 6.0 1 2 3 4 5 6 7 8 9 10 tionship between poverty and investment? Although the Incomeranking literature has paid significant attention to the impact of income levels on the investment rate (see, for example, Panelb Ben-David, 1995), little is known about the impact of Poverty,% poverty on investment. As a first pass at the issue, we 70 60 ranked 99 countries for which we have income, poverty, 50 and investment data according to their per capita income 40 in the mid-1990s.10 Then we partitioned these countries 30 into 10 groups of 10 countries each (the last group has only 20 9 countries). The poorest countries in the sample are in the 10 first group, the next poorest 10 countries are in the second 0 group, and so on; thus the 9 richest countries form the 1 2 3 4 5 6 7 8 9 10 Incomeranking tenth group. For each group, figure 6.13 plots median (log) income in Panelc panel A, poverty ($2 poverty line) in panel B, and gross Investment,%ofGDP fixed capital formation relative to GDP in panel C.11 25 Inspection of this figure reveals a clear nonlinear pattern in 23 the relationship between income, poverty, and investment. 21 For example, headcount poverty falls dramatically between 19 the first and fourth groups--from about 66 percent to less 17 than 8 percent, but after that it declines much more mod- 15 estly as one moves up the income-group classification. Sim- 13 ilarly, investment increases from 14 to about 22 percent of 1 2 3 4 5 6 7 8 9 10 GDP between the first and fourth groups, and then remains Incomeranking virtually constant between the fourth and tenth groups. Source: LópezandServén(2005b). Note that these nonlinearities are not driven by the under- Note: Thepictureplotsmedianincome,headcountpoverty ($2povertyline),andinvestment(grossfixedcapitalformation lying income data (panel A), whose association with invest- asapercentageofGDP)bygroupofcountries.Countrieshave beenrankedbytheirincomeinthe1990sandthengroupedin ment seems to be well described by a linear pattern. 10groupsof10countrieseach(exceptforthelastgroup,which has9countries)foratotalof99countries.Thepoorestcountries The figure suggests a closer association between poverty wouldbeingroup1andtherichestingroup10. and investment than between income levels and invest- ment. In fact, the correlation coefficient between the between the investment series and the poverty series shown income series in panel A and the investment series in panel in panel B is -0.77. C is about 0.55 (that is, investment tends to be higher in Does this apparent close association between poverty and richer countries), whereas the correlation coefficient investment withstand econometric scrutiny? Apparently 118 D O E S P O V E RT Y M AT T E R F O R G R O W T H ? yes. Lopez and Servén (2005b) estimate the impact of The role of the financial sector poverty on investment using a simple accelerator model and As noted above, theoretical models on poverty traps tend to find that, all else being equal, a 10 percent increase in exploit the joint impact of high poverty and credit con- poverty is likely to be associated with a decline in invest- straints on growth. The basic idea is that poverty is likely ment of about 6­8 percentage points. This result is robust to have a greater effect on investment when financial sector to the use of different poverty lines and alternative measures development is limited. Thus, one would expect to find of investment. that the impact of poverty on investment is affected by the This finding suggests a potential explanation for poverty's degree of financial sector development. negative effect on growth: a higher poverty rate leads to a Table 6.3 reports the results of estimating an empirical lower investment rate, which leads to lower growth. In fact, investment equation (based on the simple accelerator when one econometrically explores the impact of poverty on model) augmented with two variables aimed at capturing growth controlling for investment, the investment rate turns any potential difference in the effect of poverty on invest- out to belong to the growth equation, but poverty does not ment in countries with a highly developed financial sector enter significantly in the various specifications (that is, the (PovertyHFD) and in those with a less developed financial sec- impact of poverty on growth is captured by the investment tor (PovertyLFD).13 The results of this exercise indicate that, variable).12 as expected, investment rates tend to be highly persistent, to TABLE 6.3 Does financial sector development play a role in the poverty-investment interaction? GFCF GCF Variable (1) (2) (3) (4) (5) (6) Investment (t - 1) 0.721 0.716 0.735 0.653 0.656 0.674 (0.04) (0.04) (0.05) (0.03) (0.03) (0.03) Income (in logs) (t - 1) -0.005 -0.011 -0.010 -0.005 -0.006 -0.002 (0.00) (0.00) (0.01) (0.00) (0.00) (0.01) Growth (t) 0.524 0.507 0.498 0.620 0.616 0.612 (0.04) (0.04) (0.04) (0.04) (0.05) (0.05) PPP (t - 1) -0.004 0.001 -0.001 0.000 0.000 -0.001 (0.00) (0.00) (0.00) (0.01) (0.01) (0.01) Terms of Trade (t) 0.079 0.089 0.100 0.071 0.078 0.079 (0.02) (0.02) (0.02) (0.02) (0.02) (0.03) PovertyHFD ($2) (t - 1) 0.031 0.016 (0.03) (0.03) PovertyLFD ($2) (t - 1) -0.055 -0.057 (0.03) (0.02) PovertyHFD ($3) (t - 1) -0.002 0.011 (0.03) (0.03) PovertyLFD ($3) (t - 1) -0.059 -0.038 (0.02) (0.02) PovertyHFD ($4) (t - 1) 0.003 0.025 (0.03) (0.03) PovertyLFD ($4) (t - 1) -0.039 -0.010 (0.03) (0.03) Source: Lopez and Servén (2005b), table 9. Note: Numbers in parentheses are standard errors. The table reports the results of regressing investment on the variables in the first column. In columns 1, 2, and 3, we use the ratio of gross fixed capital formation (GFCF) to GDP as the measure of investment. In columns 4, 5, and 6, we use the ratio of gross capital formation (GCF) to GDP. PPP is a measure of the price of capital goods, and PovertyLFD and PovertyHFD are the poverty headcounts of countries with low and high financial sector development, respectively. The poverty line used for each variable is given in US$. 119 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S be procyclical, and to negatively depend on the cost of cap- others 2003). For the region the deficit is estimated at ital goods. Moreover, the impact of poverty on investment about 19 percent, but in some countries it is much higher. In is more adverse in countries with less developed financial Brazil, for example, the secondary school enrollment deficit sectors. In fact, poverty does not seem to affect investment is estimated at 36 percent and in República Bolivariana de at high levels of financial sector development when credit Venezuela at 42 percent. constraints for the poor may not be so relevant. However, as discussed in detail in chapter 9, poverty These findings are consistent with those in Giuliano and may also affect education levels so that the relationship Ruiz-Arranz (2005) who analyze the impact on investment between poverty reduction and education is one of double and growth of foreign workers' remittances. Giuliano and causality. In table 6.4 we present the results of estimating Ruiz-Arranz find that remittances typically have a positive a simple econometric model for the years of secondary impact on investment but that this impact declines with schooling using cross-country data.14 In addition to the the level of financial sector development. In other words, lagged dependent variable, it includes among the explana- remittances seem to alleviate the credit constraints on the tory variables the following indicators: per capita income to poor and through that channel contribute to capital accu- control for the country's level of development, the pupil-to- mulation and growth. teacher ratio to capture quantity and quality efforts at the country level, and poverty (as measured by the headcount Poverty and education index using the $2-, $3-, and $4-a day poverty lines). There is a clear relationship between education and poverty Table 6.4 shows that, as expected, secondary education reduction. Education has a very strong impact on earning is highly persistent. It also indicates that richer countries potential, expands labor mobility, promotes the health of (as measured by per capita income levels) have more-edu- parents and children, and reduces fertility and child mor- cated populations, and that a lower quality of education (as tality. For example, the World Bank's 2005 poverty assess- measured by a higher pupil-to-teacher ratio) is associated ment for El Salvador (World Bank 2005) estimated that the with less-educated populations. Finally, higher poverty per capita income of a household whose head had a primary levels typically result in lower average years of secondary education was 13 percent higher, on average, than that of a education. household with an uneducated head. The gain from a On the whole, this discussion highlights the possibility household head with a secondary school education was that poverty and growth interact through the education about 26 percent relative to a head with a primary school education, whereas the average gain from a household head TABLE 6.4 having a university education was about 38 percent. Does poverty lead to lower secondary education? Similarly, the Bank's poverty assessment for Honduras in 2001 (World Bank 2001a) reported that in urban areas Dependent variable is average years of secondary education during the 1990s, workers with 7 years of school Secondary education (t - 1) 0.95 0.94 0.94 increased their labor income by 9 percent over workers (0.00) (0.00) (0.00) with 6 years of school, whereas an increase from 15 to 16 Income 0.12 0.11 0.08 years resulted in additional income of 14 percent. The (0.01) (0.01) (0.01) Pupil/teacher ratio -0.01 -0.01 -0.01 income gains in rural areas from comparable improve- (0.00) (0.00) (0.01) ments in schooling were estimated at 11 and 18 percent, Poverty ($2 a day) -0.08 (0.04) respectively. Poverty ($3 a day) -0.09 Education is also crucial to achieve sustained economic (0.03) growth and hence sustained poverty reduction. As noted in Poverty ($4 a day) -0.16 (0.03) chapter 5, human capital plays a central role in long-run growth. Education directly contributes to worker produc- Source: Authors' calculations. tivity and to more rapid technological adaptation and inno- Note: Numbers in parentheses are standard errors. The table vation. This point is particularly relevant for growth in reports the results of regressing the average years of secondary education on the variables. Although not reported here, the Latin America because most Latin American countries have standard specification tests do not indicate any particular prob- massive deficits in secondary enrollment (de Ferranti and lem with the estimated model or the instruments used. 120 D O E S P O V E RT Y M AT T E R F O R G R O W T H ? channel. As the literatures on both growth and micro- For example, Fogel (1994) argues that nutrition and economic determinants of poverty stress, higher education health have a significant influence on labor productivity levels result in higher growth and higher household and estimates that when labor is adjusted for intensity income levels and therefore in lower poverty. At the same (measured by calories), improved gross nutrition explains time, lower poverty levels feed back into the system and about one-third of economic growth in the United King- result in higher education, creating the potential for a vir- dom since 1800. Similarly, Boucekkine, de la Croix, and tuous circle between growth and poverty. Licandro (2003) estimate that the observed improvements in adult mortality since the 18th century account for 70 Poverty and health percent of the growth acceleration that occurred before the Poorer countries have much worse health indicators than industrial age. They argue that exogenous improvements in richer countries, most likely because of the bidirectional adult mortality between 1600 and 1800 increased individ- causality between income and health. On the one hand, ual incentives to build human capital and, as a consequence, empirical evidence indicates that higher income levels lead investment in education rose, which in turn exerted a posi- to better health indicators For example, Pritchett and tive effect on economic growth. Summers (1996) estimate that the long-run income elastic- Mayer-Foulkes (2001) has studied the long-term impact ity of infant and child mortality in developing countries of health on economic growth in Latin America. Although lies between 0.2 and 0.4. On the basis of those estimates, he is unable to disentangle the relative contribution of such they calculate that more than 500,000 child deaths in the factors as nutrition and adult mortality, his results indicate developing world in 1990 alone could be attributed to the that typical health improvements for adults may be associ- poor economic performance in the 1980s. ated with a permanent incremental increase in annual On the other hand, there are a number of channels growth of between 0.8 and 1.5 percent. Thus poverty can through which health can affect growth and income levels. also affect growth through the health channel. High poverty may result in worse health, which feeds back into · Productive efficiency. Healthier workers are more pro- lower growth, creating the possibility of a vicious circle. ductive. When health improves, more output can be produced with any given combination of skills, phys- Poverty and innovation ical capital, and technological knowledge. One way The discussion so far has suggested that poverty can hamper to think about this effect is to take health as another economic growth by choking an economy's ability to accu- component of human capital, analogous to the skill mulate various forms of productive capital. Another poten- component. tial link between poverty and growth exists, however, one · Learning capacity. Health plays an important role in that concerns an economy's ability to innovate and thus determining the rate of return to education. Children improve the productivity or efficiency of capital, labor, and who are well nourished and alert gain more from a other factors of production. Moreover, poverty's negative given amount of education. effect on capital accumulation can itself hamper innovation · Creativity. Just as a healthier person is more efficient when capital investments are required to cover the costs of in producing goods and services, so is the person innovation. For instance, introducing new export products likely to be more efficient in producing new ideas and can require investments to understand market regulations hence in his or her ability to innovate (see also below). and product standards, or simply to experiment with various · Life expectancy. Increases in life expectancy have a business plans to achieve an efficient production process. direct effect on the average skill level of the popula- Similarly, more sophisticated innovations with commercial tion. This is a consequence of two forces. When the value can be achieved only through investments in research probability of dying young is high, the discount rate and development. And both types of innovations can require is also high, making it optimal for people to start at least a minimum amount of education. Consequently working early in their life and not to stay at school poverty, which is associated with low levels of human and too long. Similarly, when life expectancy is short, the physical capital, can be associated with lower levels of inno- depreciation rate of human capital is high, making vation at the national level (for a given level of national its accumulation less profitable. income per capita). In other words, poverty can effectively 121 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S limit the number of potential innovators, not because com- At the same time, mobility through the income distrib- munity members are not talented, but because poverty pre- ution may have impacts that promote growth. Hart (1981, vents them from undertaking the necessary investments to 9), for example, argues that "it is mobility which provides bring about economically meaningful innovations. the sticks for those who do not wish to move down the dis- While the links between poverty and innovation remain tribution and the carrots for those who wish to move up." understudied, and our understanding of the drivers of inno- More generally, the accumulation of human capital that is vation and technical progress in general is quite modest, so critical to intergenerational mobility has effects on recent research by Klinger and Lederman (2005) sheds growth; a greater possibility for moving up the income lad- some light on this important issue. These authors studied der stimulates greater investment, which in turns leads to the determinants of two types of innovations, namely, the higher growth. introduction of new export products and patenting activity Mobility is also seen as an indicator of efficiency: high across countries and over time. This study reports the so- levels of income fluctuations may be seen as evidence that called marginal effects of population and poverty, and their individuals are moving fluidly from one position to interaction on the number of new products exported by a another, responding to changes in supply and demand for sample of 70 countries during 1994­2003. It also presents labor. Labor legislation that leads to segmented labor mar- the same marginal effects, but for patenting activity during kets where certain classes of workers are therefore rationed the 1980s and the 1990s. It is worth highlighting that out of good jobs, liquidity constraints that prevent individ- these analyses controlled for numerous other variables that uals from migrating to more prosperous regions, or defi- might also affect innovative activity.15 cient financial markets that deny good entrepreneurs the In any case, Klinger and Lederman find that the median resources they need to grow both restrict mobility and lead (or typical) effect of poverty on export "discoveries" is about to poor allocation of resources. They can also be elements of 0.02; for patenting activity, it is about 0.06. In other poverty traps, which are explicitly about the inability of words, for each 1 percent increase in a country's poverty low-income groups to move up in the distribution. rate, the number of export innovations falls by 0.02 percent However, chapter 2 argued that the unpredictable ele- and the number of patents falls by 0.06 percent. Since the ment of mobility constitutes risk that adversely affects wel- monetary value of exports and patents can be quite high, fare. For this reason, advanced societies have developed the economic consequences of poverty through these inno- insurance and other mechanisms to reduce the risk that vation channels should be worrisome. Perhaps more inter- individuals and families face. Simulations that measure esting, the empirical evidence also suggests that poverty how risk-averse people are suggest that these welfare effects affects innovation by affecting the number of potential are large. In addition, a recent strand of the literature innovators within a country. For both export discoveries (Krebs 2003) argues that risk also has negative impacts on and patenting, the effect of population size on innovation growth. As chapter 9 discusses, individuals' decisions to activity declines with poverty. A plausible explanation for invest in education are strongly dependent on the perceived this result is that poverty reduces the number of people long-run gains in income. But like any other investment, with sufficient human and physical capital needed to pro- the riskier the expected return, the less attractive it duce innovation. becomes. Cunha, Heckman, and Navarro (2005) argue that college attendance is lower than expected given the rela- Poverty, mobility, and risk tively high average return to education because roughly According to de Ferranti and others (2000), volatility is 40 percent of the observed variability in postcollege considerably higher in all developing regions than in incomes is unpredictable: if individuals could make their industrial economies. The less-diversified economies in decisions based on their actual incomes, 25 percent of high lower-income countries, as well as limited access to external school graduates would rather be college graduates and financing, expose these countries to higher risk and thus 31 percent of college graduates would have stopped at high greater volatility. This then translates into higher volatility school. Hence, "uncertainty about future outcomes greatly in aggregate wage measures and unemployment rates. Thus affects schooling choices, and there is plenty of scope for poverty seems to lead to higher risk. ex-post regret," the three write (54). In countries where 122 D O E S P O V E RT Y M AT T E R F O R G R O W T H ? workers face large shocks to their labor incomes, because economy, and, more fundamentally, simple algebraic mod- of either frequent bouts of unemployment or high earnings els cannot capture all the very subtle effects. Nonetheless, volatility caused, perhaps, by inflation, or where frequent the exercise suggests that the magnitudes of effects arising illness prevents working, the incentive to invest in educa- from the presence of high risk in Latin America are large tion may fall even more. The resulting lower levels of edu- and that risk thus needs to be treated as an important cation in turn dampen growth. dimension of an effective poverty reduction and growth Here, then, is another example where two dimensions of strategy. Not only are policies to ameliorate risk beneficial poverty--health and risk--undercut growth, and the mag- from a pure vulnerability point of view, they may also be nitudes appear large. Krebs, Krishna, and Maloney (2005) central to growth. make an attempt to assess empirically the effect on human capital accumulation and growth of declines in the level of Concluding remarks income risk of Argentina and Mexico to the U.S. levels. This chapter explored the possible existence of links Their findings indicate that if Mexico could lower its between growth and poverty reduction by which growth labor market risk to Argentine levels, it could potentially lowers poverty and lower poverty in turn contributes to increase its growth rate permanently by almost half a per- faster growth. We reviewed several possible theoretical centage point (table 6.5). The amount that growth would arguments that support the existence of such links. Among have to increase to increase the total welfare measure by an the most prominent are those arguments in the poverty- equivalent amount has two components. The first is the traps literature that suggest that the countries of the world direct loss that is attributable to workers' and families' are increasingly divided into two convergence clubs--the dislike of risk; this effect is worth the equivalent of a rich and the poor. Membership in the poor club is consid- 0.59 percent permanent loss in yearly growth. The second ered a huge handicap for growth and hence for poverty component is the additional effect that arises because risk reduction. also makes workers and their families invest less in human The chapter then assessed the empirical evidence on this capital; this has a direct impact on welfare of 0.48 percent. front and found mixed results. On the one hand, we pre- On the whole, the effect of lowering Mexico's risk to sented evidence of convergence clubs in both absolute and Argentine levels is equivalent to increasing growth by relative income levels: richer countries converging toward slightly more than 1 percent, a huge amount in a country the rich-club equilibrium, and poorer countries toward the where growth rates hover around 2 percent. If Argentina poor-club equilibrium. By these measures, Latin America could reduce its risk to U.S. levels the effect would be less seems to be a homogeneous entity that is converging toward dramatic--growth would increase only about 0.2 percent-- an equilibrium somewhere between the rich and the poor but still important over the long run. clubs. On the other hand, we also reviewed several empirical These are only ballpark estimates. Clearly, the Mexican works that have formally tested whether the bimodality in and Argentine economies are not identical to the U.S the cross-national distribution of income is driven by poverty traps. In this regard, most, although not all, of the studies tend to reject the poverty-traps hypothesis. TABLE 6.5 The impact of risk on growth Finally, we posed one simple question. Even if there is no evidence of poverty traps in the strict sense, is it still possi- Factor United States Argentina Mexico ble that poverty is a barrier to growth? We addressed this question from two different directions. First, we reviewed Income risk 0.15 0.18 0.21 the empirical evidence contained in a background paper for Growth rate (%) 2.00 1.81 1.33 In education (%) 28.12 25.8 21.8 this report, which found that countries with higher poverty Direct loss due to risk (%) 0.59 levels tend to grow less than countries with lower poverty Loss due to lower growth (%) 0.48 levels. The estimates presented in this chapter suggest that Total welfare loss (%) 1.07 an additional 10 percentage points in the headcount poverty index cut growth prospects by about 1 percentage point. Source: Krebs, Krishna, and Maloney (2005b) for Argentina and Mexico; Meghir and Pistaferri (2004) for United States. Second, we explored a number of potential channels through 123 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S which poverty might lead to lower growth. This evidence could expect the countries in the group to cluster around indicated that in countries with higher poverty rates, accu- the equilibrium values over the long run. In contrast, val- mulation of both physical and human capital (education and ues of > 0 would indicate divergence, and one would health) is lower. Evidence also suggests that countries with expect to observe that the dispersion in the cross-country higher poverty levels have lower rates of innovation (a criti- distribution of per capita income increases as time goes by. cal contributor to growth) and higher risk. Finally, for = 0 there is neither convergence nor diver- It must be noted that in many of these channels the gence. This simple model can be used to estimate the financial sector may play a very significant role, either by expected value of income over time when < 0, which is imposing a binding financial constraint on the poor that given by -µ/. may prevent them from undertaking investments in The table below reports the results of estimating the human and physical capital or by preventing them from previous model for the full sample of countries and for the hedging against risk. Thus, the development and operation three clubs discussed in the text (low-low, low-high, and of the financial sector also appear to matter for the potential high-high). The first noteworthy point is that, not surpris- feedback effect from poverty to growth. ingly, in view of figure 6.7, the full sample presents diver- Overall, the results of this chapter suggest two main gence ( > 0). However, when we reestimate the model for messages. First, the focus of the growth-poverty discussion each of the three clubs we obtain convergence, the point needs to be shifted from the possible effects of growth on estimates of are always negative (although admittedly for the poor (on which ample evidence has already been col- the high-high group, the estimate is not significant, which lected) to the relationships between growth and poverty. in turn may suggest that although there is no divergence, That shift in focus should mitigate the debate on whether there may not be convergence either). development strategies should rely more on pro-growth or pro-poor policies, because strategies that do not take into Convergence clubs account the bidirectional relation between poverty and growth will likely lead to disappointing results: poverty Parameter Equilibrium will not decline without growth, but growth will be diffi- Club µ US$ cult unless the constraints affecting the poor are also addressed. Second, at a more operational level, considering All 0.0033* -0.007 Divergence (0.0017) (0.014) poverty and growth as part of the same problem suggests Low-low -0.0117* 0.087* 1,717 that the biggest payoff to growth (and hence to poverty (0.003) (0.024) reduction) is likely to result from policies that not only Low-high -0.0178* 0.165* 10,600 (0.0069) (0.053) promote growth, but also exert an independent, direct High-high -0.006 0.07* 120,000 impact on poverty--hence reducing the drag of poverty on (0.004) (0.036) growth. Source: Authors' calculations. Annex 6A *Significant at the 5 percent level. Convergence clubs and long-run equilibriums Convergence clubs and country transitions One way to estimate the long-run per capita income equi- To explore the distribution of income levels across countries, librium for each convergence club is based on the concept Quah (1993) takes each country's income level relative to of -convergence (see Barro and Sala-i-Martin 1995). This the world average; allocates each observation to one of five concept relies on the estimation of the following simple states: 0­0.25, 0.25­0.5, 0.5­1, 1­2, and 2 and above (that model: is, the first state includes the poorest countries and the fifth state the richest); computes a transition matrix measuring (6A.1) [ln(Y1999) - ln(Y1960)]/39 = µ + ln(Y1960), the probability that a country in one state changes state by where Y denotes per capita income and the subscript refers averaging the observed one-year transitions over every year to the year in question. Values of < 0 would indicate con- from 1962 to 1984; and evaluates the long equilibrium vergence (-convergence, to be more precise), and one consistent with the stationary distribution. 124 D O E S P O V E RT Y M AT T E R F O R G R O W T H ? When we replicate all these calculations but use data for about 99 percent, the same probability for Latin America is 1960­99, we obtain the following transition matrix: estimated at 87 percent and 85 percent, respectively. Thus the Latin American region seems to display more mobility 0.987 0.013 0 0 0 at the extremes of the distribution than does the global dis- 0.038 0.935 0.026 0 0 tribution: both getting out of extreme poverty and getting M = 0 0.033 0.936 0.031 0 , out of extreme richness seems easier in Latin America than 0 0 0.032 0.954 0.014 in the rest of the world. 0 0 0 0.009 0.991 The second difference regards the probability of moving ahead for a Latin American country in state 3 or 4; that where a typical element mij measures the probability that a probability appears to be lower than it is in the rest of the country in state i shifts to state j. So, for example, the prob- world. In particular, a Latin American country in state 3 ability that a country in the first state remains in its state is has about half the probability of moving to state 4 as do almost 99 percent, whereas the probability that it moves to state 3 countries in the global sample (1.6 percent and the second state is about 1 percent. Similarly, the probabil- 3.1 percent, respectively). More dramatically, the estimated ities that a country in the second state remains in the same probability of moving from state 4 to state 5 is nil in Latin state, progresses to the third, and returns to the first state America. These differences would result in a regional equi- are 93 percent, 2.6 percent, and 3.8 percent, respectively; librium given by 0.052, 0.33, 0.47, 0.14, and 0. thus suggesting that the probability that an economy in state 2 falls behind is slightly larger than the probability of the same economy going ahead. This type of asymmetric Estimating the impact of poverty on growth behavior also applies to countries in state 3 and more The empirical strategy that Lopez and Servén (2005b) use markedly to those in state 4 where the probability of falling to explore the links between poverty and growth in the behind is more than double the probability of advancing. data is based on the addition of a suitable measure of Using the transition matrix M, it is now possible to poverty to an otherwise standard empirical cross-nation compute the associated long-run equilibrium for the distri- growth regression: bution of income levels by allowing the time horizon of the (6A.2) iterations to expand. This exercise results in the following (yit - yit ) = yit -1 -1 + xit + pit + i + it, -1 equilibrium values for each of the five states under consider- where y is the log of per capita income, p is a measure of ation: 0.43, 0.15, 0.12, 0.12, and 0.18. poverty, x represents a set of control variables other than lagged income (discussed shortly), i is a country-specific Convergence clubs and country transitions effect, and it is an i.i.d. (independent and identically dis- in Latin America tributed) error term. However, several aspects of this The previous exercise can be replicated using data only for empirical strategy require attention. Latin America. The resulting transition matrix in this case is as follows: Estimation issues 0.875 0.125 0 0 0 Estimation of the previous equation poses two main chal- 0.02 0.928 0.052 0 0 lenges, namely, the presence of country-specific effects and MLAC = 0 0.036 0.948 0.016 0 . the possible simultaneity of some of the explanatory vari- 0 0 0.055 0.945 0 ables with growth. These problems are addressed by using 0 0 0 0.154 0.846 a GMM estimator (Arellano and Bover 1995 system esti- mator) that relies on internal instruments. Admittedly, There are at least two important differences between with highly persistent instruments, that estimation MLAC and M. First, M displays more persistency in the first method may not fully eliminate the potential bias related and fifth states than MLAC does (the estimated persistency to reverse causality. To control for this problem, Lopez and of states 2, 3, and 4 is very similar in both cases). Whereas Servén (2005b) also present results based on cross-sections, the estimated probability that an economy in either state 1 which should not suffer from reverse causality. In this or state 5 of the global sample continues in the same state is 125 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S regard, changing the estimation method does not dramati- 2. By convergence club, we refer to a tendency of countries to cally affect the results. converge to different equilibriums for per capita income levels. For example, Quah (1993), among others, finds evidence suggesting that the cross-country distribution of income may be well characterized Control variables by a twin-peak structure with poor and rich countries clustering The empirical growth literature has experimented with so around two different equilibriums. many alternative sets of explanatory variables that accord- 3. Although not included in the sample, it is likely that Haiti ing to Durlauf and Quah (1999), by 1998 the number of also belongs to this group. individual regressors that had been considered as potential 4. See Azariadis and Stachurski (2005) for a complete survey, and Lustig, Arias, and Rigolini (2003) for a nontechnical review. explanatory variables in growth regressions exceeded the 5. For the purposes of this report, the industrial, or developed, number of countries in the standard growth data set. countries group covers the OECD economies that are not eligible for Rather than adding to the already huge variety of growth lending from the International Bank for Reconstruction and Devel- models, Lopez and Servén (2005b) use a baseline specifica- opment. Figure 6.5 was constructed as follows. First, for each year we tion that relies on the controls used by Perotti (1996), compute the median growth rate for all the countries in the relevant Forbes (2000), Banerjee and Duflo (2003), and Knowles group for which the annual World Development Indicators report data. Then we apply a three-year, backward-moving average filter to (2005). However, Lopez and Servén also experiment with smooth the series. two alternative sets to check whether the results are sensi- 6. Admittedly, if the analysis were to take into account popula- tive to changes in the controls. The basic finding is that tion weights, the story for the 1990s would be different: per capita changing controls does not significantly affect the esti- growth would be approximately the same in both the developing and mated impact of poverty on growth. developed worlds. China and India account for much of this evening out, not only because they had almost 40 percent of the world's pop- Missing variables ulation during the 1990s, but also because India and especially China had excellent growth records. These differences are a reflection of the The problem of missing variables is quite standard in this different ways in which economic performance can be measured. If type of analysis. However, one variable in this context-- individuals are the preferred unit of analysis, then weighted averages inequality--needs particular attention. A relatively exten- are probably more useful. If, instead, the unit of analysis is the coun- sive literature already relates inequality and growth. For try (as is the case when one focuses on country policies and country example, Alesina and Rodrik (1994) and Perotti (1996) performance), then medians seem more appropriate. find a negative relationship between inequality and growth 7. Admittedly, it would be possible to argue that the 1960s dis- tribution has two peaks: one around $3,000 and the other around on the basis of cross-section data, but Li and Zou (1998) $13,000. and Forbes (2000) obtain the opposite result using aggre- 8. For savings, Kraay and Radatz (2005) use a representative gate panel data. Barro (2000) finds that inequality may agent framework, something that rules out the possibility of credit affect growth in different directions depending on the market failure. In the Solow framework they use, the roles of jumps country's level of income, while Banerjee and Duflo (2003) in saving and jumps in technology are more or less interchangeable. conclude that the response of growth to inequality changes 9. Overall the results are backed by almost 90 robustness checks. 10. This approach is similar to that of Ben-David (1995) who has an inverted U-shape. Given the relation between focuses on the impact of income levels on investment. We pick the inequality and poverty, excluding inequality from the 1990s because it is the period over which more poverty observations equation could lead to the poverty variable capturing a are available. pure inequality effect rather than a poverty effect. The 11. The results remain virtually unchanged if one uses gross capi- empirical findings in this regard confirm that the estimated tal formation (GFC) as the investment measure. impact of poverty on growth does not result from poverty 12. This result is robust to the use of different measures of the investment rate. acting as a proxy for inequality either in a linear or in a 13. PovertyHFD is equal to the poverty headcount when the stock nonlinear fashion. of credit to the private sector in the country/year in question is larger Notes than the sample median and zero otherwise. PovertyLFD equals the poverty headcount when the stock of credit to the private sector in 1. Clearly, given the aversion of societies to high income inequal- the country/year in question is smaller than the sample median and 0 ity levels (see de Ferranti and others 2004), one could also justify the otherwise. Clearly, PovertyHFD + PovertyLFD = Poverty. need to pay attention to distributional issues on the basis of political 14. Estimation is performed using the GMM system estimator economy arguments. with internal instruments. This estimator therefore controls for 126 D O E S P O V E RT Y M AT T E R F O R G R O W T H ? unobserved fixed effects and potential endogeneity of the explanatory tion, past innovation activity, expenditures in research and develop- variables. The data are the same as in Lopez and Servén (2005b), ment (in the case of patents granted by the U.S. Patent and Trade- except for the pupil-to-teacher ratio and expenditure in education, mark Office), and exports to the United States (in the case of patents which come from the World Development Indicators. granted by the U.S. Patent and Trademark Office). These authors 15. Klinger and Lederman (2005) control for GDP per capita, obtained similar results when using the share of the population export growth, population size, the sectoral concentration of innova- with less than a high school education, but they were unable to dif- ferentiate between the effects of poverty on both human capital and physical capital reducing the effective share of the population capa- ble of undertaking productive innovations. 127 CHAPTER 7 Subnational Dimensions of Growth and Poverty Poverty rates within Latin American countries differ as much as those across countries. Moreover, some groups of subna- tional units seem to behave as convergence clubs, suggesting the existence of regional poverty traps. The presence of agglom- eration externalities and relatively weak equilibrating mechanisms, especially through migration, creates important trade-offs in policies toward lagging regions. C HAPTER 6 EXPLORED HOW THE REGION mobility in Latin America. We focus primarily on Brazil, fares in the overall distribution of world Chile, and Mexico, which have generated the most careful income and concluded that, with some data and analytical work to date. For Brazil and Mexico, we important exceptions, the region is situated also consider regional convergence of nonincome measures in an intermediate position between the of well-being. We then turn to some possible explanations high-income countries and the really poor. However, com- for the existence of regional convergence clubs, the failure paring regions within countries reveals differences in pros- of intranational income-equilibrating mechanisms, and perity that are staggering and of the magnitudes seen finally to selected policy issues. internationally. For example, in 2000, income per capita in the poorest municipality in Brazil was barely 10 percent of What is spatial inequality, how is it measured, that in the richest; in Mexico, per capita income in Chiapas and what are the regional trends? was only 18 percent of that in the capital. The mobility of To capture the relevance of geography, traditional indexes subnational units across the income distribution has been of income inequality can be decomposed along the spatial studied as much as the movement of countries and individ- dimension and poverty rates calculated for each of the spatial uals across the global income distribution. There is also a units.1 Compared with a time series in which the ordering similar concern with the existence of poverty traps, of data points is given naturally, the definition of the rele- although with some policy twists particular to the geo- vant spatial unit--the state, department, province, munici- graphical level of analysis. pality, or perhaps even finer disaggregations--is more The 2005 World Bank regional flagship report for Latin arbitrary. As Shorrocks and Wan (2005) show, looking America, Beyond the City: The Rural Contribution to Develop- across several countries, the component of inequality due ment (de Ferranti and others 2005) provided compelling to differences between geographical regions averages around evidence that the quantity and quality of jobs are highly 12 percent of overall inequality, with a maximum of 51 per- influenced by regional characteristics and argued that there cent depending on the subdivisions of the data used. This is was scope for a territorially targeted development policy. broadly consistent with Kanbur and Venables' (2005) con- Building on that work, we first focus on the evidence for clusion that the available empirical evidence suggests that geographic inequality, spatial concentration, and regional spatial inequality may account, at most, for one-third of 129 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S units are contiguous, perhaps forming regional clusters, or FIGURE 7.1 whether poor municipalities or states are randomly distrib- Variation in regional poverty rates in Latin America uted among rich ones. We also would like to know whether Uruguay such spatial patterns are persistent--are we dealing with Chile regions "spatially" trapped in a vicious circle of low growth­ Costa Rica low investment­low growth, as explained in chapter 6? An Argentina Dominican Rep. emerging spatial econometrics literature provides the tools Panama and indicators to begin to analyze these questions (box 7.1), Brazil and recent studies have measured the spatial distribution of Colombia Mexico incomes and how it has evolved over time in Brazil, Chile, R.B. de Venezuela and Mexico. Ideally, we would examine average household Peru incomes or poverty rates rather than per capita state Honduras Ecuador incomes, but the long spans of data required are not avail- EI Salvador able for these variables, so we work primarily with state- Paraguay level GDP per capita. Bolivia Jamaica For each of the three countries, we present a set of com- Nicaragua parable figures and statistics (see box 7.1) to assess the 0 10 20 30 40 50 60 70 80 degree of spatial clustering, as well as the mobility patterns Headcount ratio of states within the national income distribution. The upper Source: Authors` calculations. panel in each of the figures 7.2, 7.4, and 7.5 presents the standard deviation that is used in the literature to capture "sigma" convergence among log incomes per capita of the total interpersonal inequality (that is, inequality between subnational units together with Moran's I, which captures individuals); in other words, the majority of inequality the spatial concentration (clustering) of that income. The occurs within spatial units. middle panel shows the Moran scatter plots that offer a A similar pattern is sound in Gasparini, Gutierrez, and visual presentation of whether states are clustered in "neigh- Tornarolli (2005) for Latin America and the Caribbean. borhoods" with similar levels of income--high- or low-level Regional differences account for more than 20 percent of convergence clubs--or whether they are more or less ran- inequality in Paraguay and Peru and for more than 10 per- domly distributed for the beginning and end of the sample cent in the Dominican Republic and República Bolivariana period. Finally, the bottom panel presents the "stochastic de Venezuela. For most of Latin America, the regional dif- kernels," or three-dimensional mobility plots, introduced ferences appear to contribute substantially less. However, by Quah (1997) to study income dynamics.2 The advantage this finding seems to say much more about how very large of these kernels over simple plots of income distribution is the idiosyncratic differences are between people than about precisely that one can see changes of position that might be how small differences in well-being are across spatial units. hidden by identical "snapshot" distributions. Each kernel Figure 7.1 shows that variation of the poverty rate across presents state income relative to the country ("country- regions is very large for many Latin American countries. In relative") in time t on the Y axis and in time t + 5 or t + 10 Bolivia, Honduras, Mexico, Paraguay, and Peru, the differ- on the X axis. Information on each state's position within ence in poverty counts among regions is more than 40 per- the country's income distribution across many different centage points. The fact that some regions of Peru have multi-year periods is integrated to form each kernel. If there counts of under 10 percent while others hover above 70 per- is no movement at all among states, the kernels would con- cent speaks for itself about the importance of integrating sist of a single vertical plane along the 45-degree line spatial considerations into poverty analysis. shown. The fact that there is some mobility--states do change relative position--gives the kernel its volume. Were Identifying spatial concentration there are a lot of mobility but no convergence (in other Beyond knowing that poverty is concentrated in particular words, if states were just switching places), one would see an geographic units, we would also like to know if these inverted bowl or half sphere. Slicing the volume parallel to the 130 S U B N AT I O N A L D I M E N S I O N S O F G R O W T H A N D P O V E RT Y BOX 7.1 Tools to detect spatial association In the spatial statistics literature, a number of methods for all observations is proportional to the global indicator and indicators have been proposed to capture the interre- of spatial association (Anselin 1995) latedness of geographical areas (Anselin 1988, 1995; N Griffith 1996). The extent of spatial dependence of a Ii = Nzi j wijzj . given variable among a set of spatially distributed units, N z2i such as regional per capita income for the Brazilian states, i=1 can be assessed by computing a global spatial dependence Extra help with the interpretation of the local statis- statistic such as Moran's I, which reads as follows: tics is provided by the Moran scatter plot, which is a graphical complement to LISA that can be used to visual- N ize local (in)stability. The Moran scatter plot shows the I = N wijzizj , ij S values of Wzi versus zi, where W is the row-standardized N z2i i=1 (that is, rows sum to 1), first-order contiguity matrix, where N is the number of regions, wij are the elements of and zi are the standardized values of per capita income. In a (N × N) binary contiguity matrix W (taking the value 1 the current context, we plot the standardized log of per if regions i and j share a common border and 0 if they do capita income of a state against its spatial lag (standard- not), S is the sum of the elements of W, and zi and zj are ized as well), which corresponds to the weighted average normalized vectors of the log of per capita income of each income (per capita and logarithmic) of a state's neigh- state. Positive values of Moran's I indicate positive spatial bors. The Moran scatter plot divides the x-y space into dependence, which indicates a clustering of similar four distinct areas, corresponding to four types of possi- attribute values, whereas negative values are associated ble local spatial associations between a state and its with clustering of dissimilar values. To further explore neighbors. In quadrant I rich states coincide with rich the spatial pattern of the data, it is important to investi- neighbors; in quadrant II poor states have rich neighbors; gate not only whether the overall regional income distri- in quadrant III poor states are surrounded by poor neigh- bution of a country is spatially concentrated but also in bors; and in quadrant IV rich states have poor neighbors. which specific states this concentration occurs and States located in quadrants I and III represent the associ- whether high- or low-income values are clustered. We ation of similar values (positive spatial correlation), focus our analysis on local indicators of spatial association whereas states located in quadrants II and IV show the (LISA), as developed by Anselin (1995), and on the inter- association of opposite values (negative spatial correla- pretation of the Moran scatter plot (Anselin 1993). tion). The concentration of states in quadrants I and III is Two properties of LISA are important to note. First, to be expected in a scenario in which rich and poor states the value of a local statistic for each observation indicates cluster separately, generating differentiated areas of high the extent of (significant) spatial clustering of similar val- and low income. If states were located randomly, occupy- ues around that observation. This means that the local ing the four quadrants without a discernible pattern, spa- indicator Li enables us to infer the statistical significance tial dependence would be nonexistent. Notwithstanding of the pattern of spatial association at that location. Sec- an identifiable clustering, local instabilities may still be ond, the sum of the local indicators of spatial association found for individual observations. X axis reveals the distribution of states at each initial income Brazil: Slow overall convergence and clear signs ten (or in the case of Mexico, five) years later. Significant of spatial polarization income convergence would result in a rotation of the kernel Brazil presents a case where there has been an overall toward the Y axis: states with lower incomes in t would have decrease of the standard deviation of state per capita higher relative incomes in t + 5, and vice versa. Divergence incomes, implying a process of convergence (see figure 7.2). would lead to the reverse. At the same time, the evidence (Moran's I) strongly rejects 131 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S the idea that incomes are randomly distributed across states. FIGURE7.2 The scatter plots confirm this by showing that most states IncomedynamicsandspaceinBrazil are found in quadrants I and III: rich states are found in rich 0.8 neighborhoods (their spatial lag), and poor among poor. The local Moran statistics that offer a parametric measure of the 0.7 spatial relationship of a state to its immediate neighborhood 0.6 show that income is concentrated in two well-defined spatial 0.5 clusters: the low-income northeast region--Piauí (PI), Ceará 0.4 (CE), Rio Grande do Norte (RN), Paraíba (PB), Pernam- buco (PE), and Bahia (BA)--and the more prosperous 0.3 Moran'sI Standarddeviation southeast region comprised of Rio de Janeiro (RJ), São 0.2 Paulo (SP), Paraná (PR), and Minas Gerais (MG). 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 Looking across time reveals two important findings. Spatiallag,1970 First, a comparison of the 1970 and 2000 scatter plots 3 shows a clear, substantial persistence in the relative posi- 2 SC tions of states; these patterns are found, in slightly weaker MG 1 PR SP form, as far back as the data allow us to look--1939. ES RJ AM MI RS Second, at the same time that state incomes appear to be 0 AL SE PB RN converging in Brazil, the data suggest, somewhat counter- PI MA BA PA 1 CE GO PE intuitively, that spatial clustering has increased across the 2 same period. The kernel further clarifies what is occurring. The relatively modest convergence in incomes does not 3 3 2 1 0 1 2 3 impart any noticeable rotation off the diagonal of the clus- Incomepercapita,1970 ter, and the overall narrowness of the kernel suggests rela- Spatiallag,1998 tively little mobility among states. Further, there are 3 two-well defined humps, suggesting convergence clubs 2 similar to the "twin peaks" pattern detected by Quah MG SC (1997) for the world distribution of incomes (along with a 1 MI PR SP RJ ES very rich outlying minipeak around 2.5 times average RS 0 MA ALSEPA AM national incomes) that Moran's I suggests is growing more PB RN BA GO 1 PI defined with time.3 CE PE More disaggregated data at the municipality level allow 2 an even clearer definition of this pattern. The left panel of 3 figure 7.6 shows that the bell-shaped 1970 income distrib- 3 2 1 0 1 2 3 Incomepercapita,1998 ution has given way to a bimodal, or "two-humped," distri- bution in 2000. The scatter plots of the municipal data (figure 7.3) suggest that there were fewer outliers in 2000 2.5 than in 1970, and hence a lower overall dispersion. But the 2.0 diagonal concentration has split into two distinct groups, 1.5 with the richer municipalities and neighborhoods pulling 1.0 away from the poorer municipalities in poor neighbor- 0.5 0 hoods. This is less clearly seen in the state-level scatter 3.0 plots: São Paulo and Rio are less extreme than before as 2.0 other states have caught up, but the cluster of moderate- 1.0 2.5 3.0 1.5 2.0 income states in the middle is missing. That the action is at Countryrelative, 0 0 0.5 1.0 periodt Countryrelative,periodt 10 the state level is confirmed by other evidence, however: In 1970, 60 percent of the inequality among municipalities Source: Authors' calculations. was attributable to differences among states that they are 132 S U B N AT I O N A L D I M E N S I O N S O F G R O W T H A N D P O V E RT Y FIGURE 7.3 Income dynamics and space in Brazil at the municipal level Spatial lag, 1970 Spatial lag, 2000 4 4 3 3 2 2 1 1 0 0 1 1 2 2 3 3 4 4 4 3 2 1 0 1 2 3 4 4 3 2 1 0 1 2 3 4 Income per capita, 1970 Income per capita, 2000 Source: Authors' calculations. part of; in 2000 that figure had risen to 72 percent. The Mexico: Openness, divergence, and spatial dramatic decrease in inequality between 1970 and 2000 concentration has been almost entirely (98 percent) due to decreases in Mexico shows a case of increasing income disparities across within-states inequality. states combined with increased spatial clustering--the reversal of a process of convergence and declustering that Chile: Divergence and spatial concentration began around the period of unilateral liberalization (1987) Aroca and Bosch (2000) find similar strong evidence of spa- and continued through the signing of the NAFTA treaty tial clustering in Chile (figure 7.4). In particular they find a (1995). As in Brazil and Chile, there is clear evidence in the low-income cluster comprising the southern regions VIII, various Moran statistics of convergence clubs and polariza- IX, and X that was also evident in the 1960s. Again, there tion in Mexico; again, the kernel suggests little mobility is overall convergence in regional incomes at the same among states and the emergence of another case of twin time that one sees evidence of more spatial concentration in peaks (figure 7.5). Aroca, Bosch, and Maloney (2005) show the 1990s, a period of rapid overall growth of the Chilean that much of the increase in both dispersion and spatial economy. The impressive increase in the overall indicator of concentration is explained by the adjoining states of spatial dependence was caused by the emergence of a cluster Oaxaca, Guerrero, and Chiapas, which have fallen behind of high income per capita in the north of the country, espe- and been unable to take advantage of new economic oppor- cially around regions I, II, and III, although the economic tunities, thus consolidating a longstanding low-income forces driving each state do not seem closely related. How- cluster in the far south. ever, this time the kernel does not show such a clear conver- The increased dispersion in per capita incomes does not gence-club story, partly because the relatively few seem to be driven by the emergence of a strong northern observations do not permit clear definition of the kernel. region in Mexico: the frontier states have benefited from But overall, there appears to be a one-hump (unimodal) dis- their proximity to the United States, but beyond these tribution with some outliers. Again, the lining up of the frontier states, there appears to be little evidence of a steep- kernel along the 45-degree axis and its overall narrowness ening gradient in state incomes, and there is almost a ran- suggests relatively little movement among states. In sum, dom distribution of incomes and growth rates in the Chile until 1995 was another case of income convergence middle of the country. To the degree that there is an emerg- with increased spatial concentration. Recently, however, ing cluster, it appears to be forming among a group of both forces are moving in the same direction--toward states closer to Mexico City. Nor is it obvious that distance divergence. from the United States should condemn the southern states 133 FIGURE7.4 FIGURE7.5 IncomedynamicsandspaceinChile IncomedynamicsandspaceinMexico 0.8 1.0 0.45 0.28 0.9 0.26 0.7 0.8 0.40 0.24 0.6 0.7 0.22 0.35 0.6 0.20 0.5 0.5 0.18 0.30 0.4 0.4 0.16 0.3 0.25 0.14 0.3 Standarddeviation Moran'sI 0.2 Standarddeviation Moran'sI 0.12 0.2 0.1 0.20 0.10 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 1965 1970 1975 1980 1985 1990 1995 2000 2005 Spatiallag,1970 Spatiallag,1970 3 2.5 2.0 BCs 2 1.5 BC I 1.0 1 XI CU SO II 0.5 YU MI SL TA IV DU SI III RM XII ZC TL MO 0 CO V 0 AG NL VII GE HI NA CL QI VI 0.5 OA QU JA CH PU GU MX 1 VC 1.0 IX X 1.5 2 VIII 2.0 3 2.5 3 2 1 0 1 2 3 2.5 2.0 1.5 1.0 0.5 0 0.5 1.0 1.5 2.0 2.5 Incomepercapita,1970 Relativegrossdomesticproductpercapita,1970 Spatiallag,1998 Spatiallag,2002 3 3 2 I 2 III II YU 1 1 XI BC IV ZC SL BCs CO NL MI DU TA SO 0 0 CU V XII TL HI GE NA MO JA AG QI VII RM OA GU QU PU CL 1 VI 1 VC MX IX CH 2 X 2 VIII 3 3 3 2 1 0 1 2 3 3 2 1 0 1 2 3 Incomepercapita,1998 Domesticproductpercapita,2002 2.5 2.5 2.0 2.0 1.5 1.5 1.0 1.0 0.5 0.5 0 5 0 4 3.0 3 2 2.0 Country 2.5 1 4 5 2.0 1.0 relative, 0 1.5 2 3 1.0 period t 1 Countryrelative, 0.5 1 0 1 0 periodt 0 Country relative, period t 10 Countryrelative,periodt 5 Source: Authors' calculations. Source: Authors' calculations. 134 S U B N AT I O N A L D I M E N S I O N S O F G R O W T H A N D P O V E RT Y FIGURE 7.6 The distribution of municipal incomes and life expectancy in Brazilian municipalities 0.45 0.45 Distribution 1970 0.40 0.40 0.35 0.35 Distribution 2000 Distribution 2000 0.30 0.30 Distribution 1970 0.25 0.25 0.20 0.20 0.15 0.15 0.10 0.10 0.05 0.05 0 0 4 3 2 1 0 1 2 3 4 4 3 2 1 0 1 2 3 4 Country relative income per capita Country relative life expectancy Source: Authors' calculations. to their traditional position at the bottom of the distribu- literature focuses on the interplay of agglomeration exter- tion, given the proximity of the southeast coast to the port nalities resulting from the availability of specialized labor of Miami and the substantial rail links throughout south- or intermediate inputs and technology spillovers, on the ern Mexico to the port of Veracruz; all other things equal, one hand, and transportation costs on the other (see Krug- the southern states should have been well positioned to man 1991, 1993a; and Fujita, Krugman, and Venables enjoy a boom from trade liberalization (box 7.2). 1999). Once agglomeration has started in a particular place, for whatever reason, even a historical accident as Nonincome welfare measures Krugman (1993a) points out, reinforcing forces are at play That said, as chapter 2 suggested, income is only one that perpetuate the situation. Lack of agglomeration effects dimension of welfare, and focusing on it excessively may obscure the evolution of welfare more fully considered. Figure 7.6 shows that the distribution of life expectancy FIGURE 7.7 Social indicators in Mexico, by period in Brazilian municipalities does not follow the same pat- tern of increasing bimodality that is found in incomes. A % of population similar finding emerges for Mexico, as shown in figure 7.7. 30 1970 1980 1990 The dispersion of rates of infant mortality, mortality, liter- 1995 2000 25 acy, and school attendance shows a steady decreasing trend across the last 30 years, despite the convergence and then 20 divergence of incomes. Both cases suggest, first, that distri- bution trends in regional welfare may be improving. Sec- 15 ond, they suggest an important role for policies that fight 10 poverty independent of those dedicated to growth per se. 5 Why do we observe regional convergence clubs? Chapter 6 reviewed the literature on why convergence 0 clubs emerge among countries, and much of the same logic Infant Mortality Literacy School Years of mortality rates attendance schooling applies to regions as well. Two views receive particular atten- Source: Authors' calculations. tion in the literature. First, the New Economic Geography 135 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S BOX 7.2 Will trade liberalization increase regional disparities? NAFTA and Mexico Much of the work that analyzed the impacts of trade market may be further energized by the increased access reform and that predicted the impact of NAFTA on to cheaper and higher-quality inputs from abroad and the Mexico examined the potential response of specific indus- lowered risk implied by, especially, the NAFTA agree- tries, but was silent on how their location might be ment. Further, the location of some potential growth affected. In one possible scenario, Hanson (1997) sug- industries is clearly driven by immobile endowments not gested, along the lines of the new economic geography, necessarily concentrated on the border. NAFTA poten- that firms might choose to locate nearer the U.S. market, tially has a stimulative impact on nonborder areas with on the border, and shift away from the traditional Mexico natural endowments with its elimination of import City agglomeration centrally positioned to serve the restrictions to the United States on mangos (produced in domestic market. The benefits of proximity to the border Guerrero and Michoacan), pineapples (Veracruz, Oaxaca, would likely dissipate with distance and, as some have Tabasco), and grapes in 1994 and as it phases out restric- argued, lead to increased dispersion of welfare between tions on tomatoes ( Jalisco) and avocados (Michoacan) by north and south. 2008. Both agricultural production and exports made But there are other elements to consider as well. To large gains in the post-NAFTA period. begin, the new economic geography is not without theo- Further, other forms of nonroad transport may offer retical ambiguity: Behrens and Gaigne (2003), for exam- low-cost transport to the U.S. market for nonborder ple, suggest that the finding that trade liberalization regions. The two largest airports after Mexico City are increases geographic polarization depends critically on found in Jalisco (center-south) and Yucatan (south). Air- the specific modeling of internal transport costs. Second, lift capacity, along with its high level of human capital Krugman and others (see Head and Mayer, forthcoming, and good governance, was critical to Intel's plant location for a review) have noted the remarkable persistence of in Costa Rica, south of Mexico. Yucatan also benefits patterns of industry distribution over very long periods of from the shallow water port of Progreso that offers easy time and large changes in economic environment. This access to U.S. ports in the Gulf of Mexico as well as those persistence may arise from the power of accumulated in Central and South America and the Caribbean. It is agglomeration externalities sparked initially by often perhaps not surprising that in 2003, Yucatan had the sec- trivial historical accident, in Krugman's view, or perhaps, ond-highest concentration of maquila employment of a the importance after all of natural advantages that anchor nonborder state, exceeded only by Jalisco. The port of industries to their existing locales. In both the new eco- Veracruz, the entry point for Mexico's first globalizing nomic geography and Heckscher-Ohlin-Vanek (HOV)- influence in the 16th century, remains the country's most based views, it is not clear whether the sudden increase in important, with extensive road and rail networks that demand from abroad, and an increase in supply of connect the central and southern states, again to the Gulf cheaper and better quality inputs, may lead to the dis- of Mexico ports. Given this ready water access, all other placement of existing nonborder growth poles, or to their endowments equal, it seems as plausible to find a southern reenergizing. pole or a southeastern corridor enjoying the same benefits In Mexico, these types of considerations suggest that of proximity as it would to see the region being left the postintegration geographical patterns of economic behind. In fact, to date, there is very little evidence that performance may be more subtle and hard to predict. The either the 1985 unilateral trade liberalization or NAFTA higher costs of exporting from established central indus- has led to a correlation of growth with distance from the trial locales, such as Queretaro, Aguascalientes, or border. Guadalajara, might be offset by their well-trained work- forces and lower levels of congestion. Domestic and potential foreign firms in these areas serving the Mexican Source: Aroca, Bosch, and Maloney (2005). 136 S U B N AT I O N A L D I M E N S I O N S O F G R O W T H A N D P O V E RT Y also drives the reverse pattern: remote indigenous commu- effects of the private assets. Whether one believes that nities may have few workers to attract industry, a small local being asset-poor in this fashion constitutes a poverty trap market to produce for, and hence few economies of scale. In strictly defined, the logic of Lopez and Servén (2005b), between can be found the smaller islands of the Caribbean described in chapter 6, that poverty in these dimensions where there are few economies of scale in infrastructure, and others hinders growth resonates here as well. governance, or even diversification against adverse shocks Although evidence to date is limited, these asset deficits and where the small pool of qualified labor can make these may also dampen the transmission of growth impulses countries less attractive to foreign investors.4 from dynamic areas to poorer ones. The dampening effect Second, natural advantages anchor industries to their can work through numerous channels (De Vreyer and existing locales. Davis and others (1997) argue that Spielvogel 2005): producers establish supply links with traditional endowment-based trade theories such as the firms in other regions; growing markets in the dynamic HOV framework, perform so well as a theory of the loca- hub create new market opportunities for firms in neighbor- tion of production in Japanese regions that the New Eco- ing localities; new technologies or ideas are copied or other- nomic Geography literature actually adds little to our wise disseminated.6 These spillovers are the subject of an understanding. Ellison and Glaeser (1999) find that only emerging literature on "spatial externalities," which are 21 percent of U.S. industries exhibit levels of geographical captured by a measure of the degree of spillover, called the concentration significantly higher than those predicted by "spatial multiplier" (Anselin 2003). Current estimates of natural advantages such as weather or natural resources. average multipliers are fairly small. For Mexico, Bosch Redding and Vera-Martin (2004) show that both theoreti- (2003) finds that a 10 percent increase in growth in one cally and in 45 regions of Europe, factor endowments are state leads to a 1.5­6.5 percent increase in growth in the important in determining the location of production.5 neighboring states.7 For Brazil, De Vreyer and Spielvogel Both views can contribute to explaining the very high (2005) find that a 10 percent increase in the average income persistence of patterns of concentration of economic activity per capita of a Brazilian municipality raises the growth rate documented above and in prominent cities of the region. of the neighboring municipalities by 2.6 percent; a finding Medellin, Colombia, São Paulo, Brazil, and Monterrey, consistent with Bosch, Aroca, Fernandez, and Azzoni Mexico, all grew around a natural resource industry, usually (2003).8 These are average measures that may overstate mining, but the cities later diversified, often to very differ- spillovers to poorer regions; moreover, they suggest that ent industries. Both views also may help explain differences growth impulses from Mexico City or São Paulo are in what Jalan and Ravallion (2002) term "geographic capi- unlikely to have much stimulative effect on the peripheral tal," which may determine whether households enjoy a ris- regions of their countries. ing or stagnating standard of living. The elements of this That a positive growth shock to one state rapidly dissi- capital include roads, technological spillovers from pates is consistent with the observation of areas of high and advanced producers to those less so, and health care. The low economic activity in the same country. What is less evidence on the importance of these factors is mixed for clear is why earnings and hence levels of poverty differ Latin America and the Caribbean. Duarte, Ferreira, and across regions where movement of capital, labor, and Salvato (2003) argue that income differences among technology should, in theory, equalize earnings and hence regions in Brazil largely reflect different levels of human poverty rates. Lucas (1990) offered an explanation for why capital, more than differing returns that might arise from capital does not flow to poor countries based on differences complementarities with other regional endowments such as in levels of human capital, and a similar logic holds within roads; they estimate that if the northeast states had the countries. For example, evidence from rudimentary data for same educational endowment as those in the southeast, the Mexico suggests that foreign direct investment tends to average income gap would almost completely close. On the pass over areas with low levels of literacy such as Mexico's other hand, in Peru, Escobal and Torero (2005) find strong southern states (Aroca and Maloney 2002). A World Bank complementarities between private assets (human capital) report on Mexico's southern states (World Bank 2003) and public assets (transport, telephones, sewerage): the points to additional missing complements to foreign increase in expenditures by families in response to a cluster investment, including a lack of proper infrastructure, weak of interventions to build public assets often multiplied the financialsystems,unclearpropertyrights,andanatmosphere 137 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S of conflict. Knowledge flows, stressed as critical to growth market variables such as wages, unemployment levels, and and prosperity in Closing the Gap in Education and Technology transport costs affect migration in predictable ways. (de Ferranti and others 2003), are strongly related to However, the responsiveness to wage differentials is not capital and educational accumulation, and require an even large enough to equalize differentials.11 In seeking to more sensitive set of conditions to foster (Maloney and explain low elasticities and low mobility generally, a long Rodriguez-Clare 2005). Given that the capital cities of literature identifies liquidity constraints--the inability to the region lag in the effectiveness of their national innova- borrow against the gains that would occur if a family tion systems, even less can be expected from the lagging migrated--and there is clear evidence of this effect in regions. Mexico.12 Both liquidity constraints and the risks associ- ated with moving can be mitigated to some degree by the Does migration work as an equilibrating existence of networks of established migrants in the desti- mechanism? nation; a now-expansive literature documents that Perhaps more surprising is that migration from region to migrants to the United States tend to come from areas that region appears relatively limited as an equilibrating mech- have long been sources of migration. There can also be anism. Generally, migration is thought to be induced crowding of urban labor markets and an expansion of through the labor market, which makes state-level wages, poverty pockets in near-urban areas (Lucas 1988). Further rather than GDP per capita differences, the more relevant impediments may include poorly defined property rights in measure. Although wages show somewhat less variance the sending region and language or cultural barriers. than GDP per capita in Mexico, persistent gaps exist, and Another provocative explanation is put forward by the southern states remain at the bottom of the distribu- Aroca (2005b), who notes that in Chile from 1993 to 2003, tion, with their wages only 50 percent of those of the states there was essentially no correlation between unemploy- with the highest average wages. Overall in Latin America, ment and growth at the subnational level, while there was these wage gaps often range between 15 and 40 percent a clear and significant negative relationship at the national after controlling for worker characteristics, but they can be level. This could partly be explained by the fact that the even higher in countries with sharp geographical differ- percentage of individuals who live in one region but work ences.9 Much of the migration is, in fact, rural to urban, in another i.e., commuters is roughly double the percent- and data for Bolivia, Brazil, Colombia, and Peru reveal that age of migrants on an annual basis. Further, commuting to the urban wages are often two to three times higher than a destination seems closely related to inflows of foreign rural wages in these countries. direct investment to the destination region and negatively Migration flows have been less than what might be related to housing costs in that area. Thus, it may be that in expected given these wage differentials.10 In Mexico net terms of real income net of local costs, commuting is actu- migration from the impoverished Chiapas, Guerrero, and ally preferred to migration and constitutes a significant but Oaxaca states amounts to 2­2.5 percent of the population heretofore understudied equalization mechanism. over a period of five years; similar rates are found in the lag- Finally, consistent with our argument in favor of multi- ging regions VIII, IX, and X in Chile. A quick comparison dimensional approaches to welfare and the discussion of indicates that this dearth of migration may have an impact converging social indicators above, it may be that money on wage gaps: In the Dominican Republic, where the earn- isn't everything after all. Arias and Sosa-Escudero (2004) ings gap between some rural and urban areas is less than 10 find that, after controlling for socioeconomic characteristics percent, migrants make up 44 percent of the urban labor and access to basic services, rural residents in Bolivia no force; in Bolivia, where the regional earnings gap is 50 per- longer considered themselves poorer than the urban popu- cent, migrants make up less than 10 percent of urban lation despite remaining more likely to be income-poor. workers. Although Chuquisaca, a region with a very high fraction of Trying to understand the determinants of these flows, indigenous population, is the second poorest region as mea- Aroca and Hewings (2002) and Aroca and Maloney (forth- sured by income, its residents rated themselves the least poor coming) find that the determinants of interregional migra- in the country. Thus, geographical and cultural attractions tion flows for Chile and Mexico, respectively, are broadly in may offset income poverty and prevent further arbitraging line with the mainstream literature on migration. Labor of spatial earnings differentials.13 Further, life at the 138 S U B N AT I O N A L D I M E N S I O N S O F G R O W T H A N D P O V E RT Y "destination" may be less attractive than incomes suggest. to some measure. Figure 7.8 displays two maps of Brazil, As mentioned in chapter 2, residents of the province of one showing poverty rates and the other showing poverty Buenos Aires, the second richest province in Argentina, densities, or the number of poor people. The maps clearly rated themselves as poorer than virtually every other region show that the more rural northern states have the highest of the country. This self-rating may reflect negative poverty rates, while the big cities, both north and south, agglomeration (congestion) effects of living in big urban show the highest concentrations of poor people. The same areas, or a greater awareness of relative poverty in the pres- is true of Bolivia, where the border regions with Argentina ence of stark income differentials. and Chile have the highest proportions of poor people but not very many of them, while the developed regions with The link back to growth and policy issues high growth potential--La Paz, Cochabamba, and Santa What do these persistent inequalities in spatial income Cruz--have the highest numbers of poor people. Therefore, (if less obviously welfare) and the lack of labor mobility provided that existing agglomerations are not already too imply for growth and policy? The growth issue is, in fact, large, the theoretical trade-offs may be less important than less straightforward than it appears at first sight, and that, we initially thought--in other words, a large chunk of the in turn, complicates the policy debate. At the level of the poor are, in fact, in areas with potentially higher growth. subnational unit, all the arguments outlined in chapter 6 Chomitz's observation allows us to define four different showing that poverty-related factors may slow growth spatial categories (table 7.1) that imply distinct policies, hold, and a case can be made for policies to ameliorate some of which allow investment in potential high-growth them. In addition, Kanbur and Venables (2005), among areas with large numbers of poor people. others, have stressed that regional inequalities correlated to Areas with high poverty rates and low poverty density ethnic, linguistic, or religious divisions provide fertile capture the essence of Chomitz's trade-off. In areas of low ground for internal conflict that can undermine economy- population density, the cost of infrastructure per person is wide growth.14 higher, or, alternatively, the returns to investment are low Yet in the world of the new economic geography, the relative to areas of greater density, which can reap case for reorienting resources to disadvantaged zones economies of scale. The high-poverty-rate, low-poverty- becomes less clear, and the literature to date has been very density area is unlikely to develop substantial economic circumspect on policy prescriptions. Fundamentally, this dynamism, and policies thus need to focus more on direct literature argues that if existing externalities mean that poverty alleviation and on programs that will impart skills the current agglomerations actually show the highest useful in other, more dynamic regions. Conditional cash potential for growth, then focusing on poor regions will transfer programs or other education and health initiatives actually decrease national growth. The goal must be to or, perhaps, agricultural research and development would find a way to move people and resources to the existing be most appropriate in these circumstances. rich centers. Box 7.3 suggests that such a trade-off In areas with low poverty rates and high poverty density, between equity and growth appears to have been impor- often urban or relatively dense rural areas where agglomer- tant in Spain. Unfortunately, more generally the literature ation forces have already taken place, policies aimed at fos- offers little guidance on whether it is the externalities rel- tering growth have good chances of reaching the poor and ative to agglomeration or those leading to dispersion of translating into important poverty reductions. The major activity that are more important, so we do not know problem is to ensure that wealthy groups do not capture whether existing agglomerations are too big or too small. the flow of resources. For this reason, self-targeting mecha- As an example of the reigning agnosticism, Krugman (1999, nisms, such as those envisaged in the Argentine and 160) remarks: "One may have opinions--I am quite sure in Colombian workfare programs, are particularly appropri- my gut, and even more so in my lungs, that Mexico City is ate. That said, conditional cash transfer schemes, such as too big--but gut feelings are not a sound basis for policy." Familias en Acción in Colombia or Oportunidades in Mexico, where targeting is quite good, have been used in this type Poverty rates vs. poverty density of situation.15 Chomitz (2005), however, argues that a more subtle use of Areas with high poverty rates and high poverty density spatial information can attenuate these potential trade-offs have the potential to take advantage of projects with 139 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S BOX 7.3 Trade-offs in regional policy: The Spanish experience De la Fuente (2002) estimates that the European cohesion return for the country as a whole would have occurred had funds meant to remedy regional inequalities within the the funds gone toward the most developed regions. EU contributed significantly to the growth of poorer De la Fuente (2003) further simulates the conver- regions of Spain and to the reduction of regional dispari- gence of Spain toward the European mean of incomes ties. However, he also points out that there has been an and the convergence of the Objective 1 regions toward opportunity cost in terms of overall efficiency for the coun- the Spanish mean income under three possible scenarios: try. This is suggested in the figure below, which presents the actual relative incomes (BASE); the resulting relative the return to (marginal product of) infrastructure in the incomes in the absence of cohesion funds (SIN); and the Spanish regions in 1995. Objective 1 regions are those result of distributing the funds efficiently among all poorer regions that were targeted by the cohesion funds, the Spanish regions according to the marginal returns to virtually all of which show below average returns. It is infrastructures. The results again suggest that cohesion clear that the highest returns are found in Madrid, Catalo- funds helped the targeted regions converge toward the nia, and Balearic Islands that were not objective 1 regions national mean, as well as Spain's convergence toward and, in fact, are the richest. In other words, a much higher the European income level. In reality, the income gap between Spain and the EU15 closed by 2.9 points between 1993 and 2000 and the gap in relative incomes How the European cohesion funds benefited the different Spanish regions, 1995 between Objective 1 regions and the rest of Spain decreased 2.2 points. In the second scenario, the conver- Percent 100 gence toward the European mean was only 1 point and 75 the gap between Objective 1 regions and the others rose 50 5.6 points. Finally, had the cohesion funds been distrib- Objective 1 regions 25 uted efficiently among all the regions, the overall growth 0 of the Spanish economy would have caught up quicker 25 with the other members of the European Union (closing Nonobjective 1 regions 50 the gap by 3.9 points). However, the gap between the Va Ga Objective I regions and the rest of Spain would have Cana Mu Cant Cyl An As Ex Ba Pv Na Ar Ri C-M Ma Cat increased by 7.4 points, even more than the gap would Source: De la Fuente (2002). Note: Percentage deviations from the national average. have been in the absence of the European funds. TABLE 7.1 Typology of appropriate actions according to poverty rate and density Type of area Type of project Low poverty density High poverty density Low poverty rate No special programs needed Investments that boost labor demand Self-targeting antipoverty projects High poverty rate Investments with no scale economies Rural roads, other infrastructure Agricultural research and development Education Cash transfers Source: Chomitz (2005). 140 S U B N AT I O N A L D I M E N S I O N S O F G R O W T H A N D P O V E RT Y FIGURE 7.8 Poverty rates versus poverty densities in Brazil BOX 7.4 Rural roads and poverty reduction in El Salvador El Salvador has a high-density population in rural areas distance to the market place in rural areas. Both indicators that corresponds with the high-poverty-rate, high-poverty- are closely linked to the poverty level. The poorest house- density category in table 7.1. The country increased its holds live almost double the distance to a paved road, and rate of investment from 1 percent of GDP in 1998­99 to have 25 percent longer travel time to market, as do non- 1.9 percent of GDP in 2002­3. The increase was mostly poor households. Over the 1999­2001 period, significant concentrated in the rehabilitation of the primary road improvements in both indicators were reported for network after the 2001 earthquake, paving of main sand extremely poor households: travel time was reduced from roads, and maintenance. Roughly 26 percent of the 2,200 53 to 46 minutes, roughly the level of moderately poor cantons around the country directly benefited from the households. A systematic study of the impact on poverty improvements. of these improvements suggests that extreme poverty fell Rural roads are thought to contribute to poverty 8.8 percent in the control group, while in the cantons reduction through access to education and health, and where roads improved, poverty fell 13.9 percent. The net expansion of markets for agricultural products. To mea- contribution of better rural roads to extreme poverty of sure the improvement in access, Yepes (2004) estimated 5 percent seems remarkable for such a short period of time. two indicators using a rural panel of households: the aver- age distance from households to paved roads, and the Source: Yepes (2004). economies of scale and be subject to low levels of leakage of targeting poverty policies yields dividends. Elbers and resources to the nonpoor. Infrastructure investments such others (2004) showed that in Cambodia, Ecuador, and Mada- as rural roads may be a good example of successful projects gascar, allocating funds to geographically defined subgroups for these areas (box 7.4). of the population according to their relative poverty status From a practical point of view, the increasing use of could achieve the same degree of poverty reduction with detailed poverty maps to identify poor groups and then 40 percent fewer resources than traditional methods require. 141 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S TABLE 7.2 Public investment effects in Mexico, 1970­2000 Public investment Growth GDP per capita Infant mortality Years of schooling Productive Activities Industry 0.0068 Agriculture 0.0302 Infrastructure and Communications 0.0394 0.0224 Social Investment Education 0.0043 0.0870 0.0052 Health 0.0018 0.0211 0.0022 Source: Bosch and Cobacho (2005). Note: All coefficients are significant at the 5 percent level. The GDP coefficient includes both direct effects and indirect effects through the other two variables. And if Mexico City is, in fact, too big? a simultaneous equations framework allows them to model History suggests, however, that policy makers often judge the cross-effects of the different types of investment.19 that present agglomerations are too big, or that other con- Table 7.2 shows that investment in productive activities siderations lead them to resist abandoning entire regions to (industry and agriculture) positively affects growth. Con- low levels of economic activity and extensive conditional sistent with Calderón and Servén (2004), public spending cash transfer programs. In fact, as Beyond the City and other in infrastructure and communications do so as well, but recent World Bank reports have noted, Latin America has part of the effect comes through a channel of reducing substantial experience with ambitious regional develop- infant mortality by improving access to the water supply. ment programs that have met with mixed success, and this Social investment in education and health increases the report will not attempt a comprehensive survey of the liter- years of schooling and lowers infant mortality, and these ature.16 The now-vast OECD literature on the effects of effects also feed back through the overall increase in public investment policies generally finds a positive impact growth. The estimates also suggest that these policies have on growth and sometimes inequality although, again, as been responsible for the observed convergence in nonin- the Spanish case suggests, these policies do not necessarily come measures of poverty at a time when per capita state maximize national growth.17 The evidence for Latin Amer- incomes were diverging.20 ica and the Caribbean is thinner but generally concurs.18 What does merit emphasis, however, is that traditional Conclusions regional policy has, to some degree, neglected discussion To sum up, regional disparities in poverty and income are about the role of human capital, knowledge transmission, large and persistent. In two of the three countries studied, innovation, and improving economic environments--the overall dispersion in per capita state incomes is falling, very factors that emerge consistently as correlated with while in all three cases, the spatial distribution moves the regional income differences (see chapter 6). other way toward becoming more concentrated. Generally, In an attempt to capture the development impact of a the natural equilibrating flows of factors, especially migra- broader set of interventions, Bosch and Cobacho (2005) tion, do not operate with enough vigor to equalize incomes, model the direct and indirect effects of five types of so policy makers need to articulate region-based policies. Mexican regional federal investment (industry, agriculture, The trade-off posed by the new economic geography infrastructure and communication, education, and health) between investing in those agglomerations with high rates not only on GDP growth, but also on broader measures of of return versus those poorer areas that would yield less welfare such as infant mortality and education; working in aggregate growth needs to be kept in mind as a particular 142 S U B N AT I O N A L D I M E N S I O N S O F G R O W T H A N D P O V E RT Y policy wrinkle specific to the regional level of analysis. But directly and possibly indirectly through remittances. See Tannuri- whether policy chooses to focus on already advanced areas Pianto, Pianto, and Arias (2004). For Mexico, see Taylor (2001) and Taylor, Yúnez-Naude, and Cerón (2004). with well-designed antipoverty programs for areas of low- 11. Following the technique developed by Gabriel, Shack- density poverty, or to attempt a comprehensive strategy for Marquez, and Wascher (1993) for examining the same question in developing such low-density areas, the lessons from chap- the United States, Aroca and Hewings (2002) and Aroca (2005a) ter 6 pertain: a comprehensive approach that keeps in mind conclude that, for plausible values of the local labor demand and sup- the feedbacks directly back to growth that accrue from ply elasticities, only a proportion of the shock in wages is arbitraged attacking poverty across a broad front is likely to have more by migration. 12. See Aroca and Maloney (forthcoming). Traditional specifica- success than more traditional approaches focusing on tions have entered the wage of both the destination and origin wages narrow incentives to production. with the latter generally entering insignificantly. However, if wages are entered as both a relative wage, wj/wi, and a free-standing initial Notes wage term capturing liquidity constraints, both variables enter very 1. See Shorrocks and Wan (2005). To measure the contribution to significantly and are of expected sign. inequality, we simply partition the sample into a set of geographical 13. Urban migrants often initially settle in ethnically similar regions and then calculate the two components of aggregate inequal- neighborhoods, which suggests that networks lower the effective cost ity; a weighted average of regional inequality (within-group com- of moving and that a minimum agglomeration may be needed to ponent) and the between-group component term, which captures the elicit larger-scale migration. inequality attributable to variations in average incomes across regions. 14. An emerging empirical growth literature has documented the 2. For a detailed description on how to compute and interpret impact of fragmentation indexes and polarization on growth. See the kernels, see Quah (1997). Easterly and Levine (1997), Rodrik (1999) and Brock and Durlauf 3. Laurini, Andrade, and Valls Periera (2004) confirm these twin (2001), Alesina and others (2003). humps at the municipal level. 15. See Gertler, Martinez and Rubio (2005). They show that in 4. For a thorough treatment of the challenges facing the Mexico, CCTs led to long-term rises in living standards that per- Caribbean, see World Bank (2005f). sisted after the termination of the program and that the return on 5. Theoretically they show this should be the case regardless of investment was quite high and that households are both liquidity the degree of factor mobility. Working in a similar tradition, and credit constrained. Bernstein and Weinstein (2002) reintroduce the importance of trans- 16. As an example, Brazil's high-profile programs of fiscal incen- port costs as a means of anchoring the indeterminacy intrinsic to tives for regional development have generally been thought disap- HOV when the number of goods exceeds the number of factors. pointing for a variety of reasons, including inefficiencies and poor 6. See Bottazi and Peri (2003) for a study of regional spillovers in management. These efforts also have been dwarfed by lending, for Italy. instance, by the Brazilian Development Bank (BNDES), based on 7. But after allowing for growth effects of neighboring states to nonregional criteria such as export promotion. Recent studies sug- work through these variables, particularly literacy, the spillover gest that regional subsidies to the north and northeast represent only impact is reduced to only 0.6 percent. 12 percent of total subsidies for export promotion and industrializa- 8. The spatial effect of explanatory variables is consistent with tion, which tend to favor the industrialized regions of the south. See Chomitz (2005), who shows what appear to be positive spillover Calmon (2003) and World Bank (2005a). effects on wages and employment from income growth in nearby 17. Easterly and Rebelo (1993) find a positive relationship regions. His estimates for nonmetropolitan areas show that a 10 per- between public investment in transportation and communications cent income increase in close neighborhood regions is associated with and overall growth using a sample of 100 countries. Knight, Loayza, a 7 percent increase in a region's wages and a 2 percent increase in and Villanueva (1993) also find positive effects on investment on employment. growth for OECD countries. As noted above, De la Fuente (2002) 9. See background studies summarized in the next section and shows that in Europe the structural and cohesion funds have played World Bank poverty assessments for other countries. an important role in reducing or at least maintaining disparities 10. In fact, countries differ in ways that we poorly understand. In within countries but also warns of the possible dangers of ineffi- Bolivia and the Dominican Republic, for example, interurban migra- ciently allocating scarce resources. More recently, Calderon and tion dominates (especially to larger cities), although seasonal and Servén (2004) show how public infrastructure has been a determinant temporary migration to the rural sector in Bolivia is on the order of factor in promoting growth and reducing inequality. Foster and migration to the city in the first place. The idea that migration is a Araujo (2001) find positive effects of improvements in basic services one-way flow thus seems seriously incomplete. In both countries, infrastructure (electricity, water supply, telecommunications) for earnings were improved by migration. That is, despite a potential poverty reduction in Guatemala. lack of contacts and urban know-how, migrants got competitive 18. Ramirez and Nazmi (2003), using a cross-section of Latin urban jobs for their skills. Thus, migration likely reduces poverty American countries, find positive effects of public investment on 143 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S growth. Rodriguez-Oreggia and Rodriguez-Pose (2004), using a by years of education and investments in health and infrastructure. cross-section of Mexican regions, find a significant effect across Therefore, public expenditure in education has a direct effect on edu- 1970­85 that disappears in the period 1985­2000. cation and an indirect effect on growth and infant mortality. Infra- 19. There are three main equations in the model. Growth in GDP structure may affect growth directly and through its effects on the per capita is determined by education, infant mortality, the different social variables. kinds of investments, and a number of control variables. Similarly, 20. Further, as suggested by World Bank (2003), a multipronged years of education depend directly on investments in education, approach that attacked health and education directly probably would infrastructure, and other controls. Finally, infant mortality is affected also have growth dividends. 144 CHAPTER 8 Microdeterminants of Incomes: Labor Markets, Poverty, and Traps? The preceding chapters focused on the cross-national and spatial aspects of the coexistence of high and persistent poverty and low rates of economic growth in Latin America. The next two chapters amplify that analysis through the lens of households and individuals. This chapter examines the role that labor and other assets and their market returns play in generating persistent low earnings and inequality in the region. It concludes that public investments and policies to foster the poor's accumulation of assets (including equitable returns to their investments) would facilitate their mobility and would exploit complementarities in the generation of income that are essential for ensuring that the poor benefit from and participate in the growth process. T HE PERSISTENCE OF POVERTY ARISES FROM earnings traps can result from deficiencies in the endow- the inability of certain population groups to ments that enhance the productivity (quality) of labor increase their long-term income generation assets (such as human capital or infrastructure) as well as potential. Addressing this situation requires from earnings differentials that arise from barriers to an understanding of the factors that prevent mobility in the labor market (such as discrimination or poor families from moving out of low-productivity eco- impediments to migration) and that are unrelated to nomic activities. The poverty-traps literature emphasizes skills. that the main determinants of the poor's inability to take This chapter examines some of the mechanisms that advantage of growth opportunities are insufficient asset may prevent the Latin American poor from participating holdings, thresholds in the returns to those assets, fixed or in the growth process, thus keeping them in persistent switching costs of productive transitions, and limited poverty. Unfortunately, little long-span panel data has been access to credit or insurance.1 Of particular importance is collected for the region, which prevents in-depth analyses the ability of the poor to use their labor (their most abun- of the duration of poverty and its main determinants dant asset) in wage jobs, self-employment, or their own throughout the region.3 The chapter instead relies on the microenterprises. Labor earnings often account for more limited, though highly consistent, evidence that is avail- than two-thirds of total household income of the Latin able on these issues. Drawing from cross-section survey American poor.2 The pricing of labor reflects productivity data, the chapter discusses the variation in the level and differentials across workers and jobs, sector and regional growth path of labor earnings across individuals of dif- supply-demand imbalances, and nonmarket factors. Low- ferent skills, demographics, and job characteristics, with This chapter draws from the studies by Arias and Diaz (2004), Gasparini, Gutierrez, and Tornarolli (2005), Sosa-Escudero and Lucchetti (2004), and Sosa-Escudero and Cicowiez (2005), and from background analyses for this report by Bustelo (2005), Tannuri-Pianto, Pianto, and Arias (2005) and Sosa-Escudero, Marchionni, and Arias (2005). 145 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S attention to the quantitative importance of potential barri- Policies to improve the functioning of labor markets, ers to mobility (segmentation across sectors, occupations, including sound regulations and institutions, should facili- or locations) as a source of low earnings and poverty traps. tate productivity growth while guarding equity in the The chapter then analyzes the main determinants of labor market. The poor are generally disadvantaged in sev- income growth and poverty persistence, drawing primarily eral dimensions. Public investments and policies in one on analytical work from a unique panel household survey area (such as credit or roads construction) may have hetero- in rural El Salvador and evidence from other countries. geneous impacts depending on the level of assets and other The chapter pays special attention to complementarities initial conditions affecting the poor. A minimum coordina- (threshold or "bundling" effects) between publicly pro- tion of public interventions in poor areas can help exploit vided assets and household characteristics (observed and synergies and overcome the associated potential poverty unobserved) as drivers of family income growth. traps that may affect households with a bundling of unfa- The chapter reaches two main conclusions. First, labor vorable characteristics. market segmentation is a second-order source of low earn- ings in the region relative to low levels of productivity. The distribution of earnings: The role of worker Most low earnings and thus poverty are not generated endowments and labor markets directly by the labor market, but largely reflect differences There are two distinct perspectives on how labor markets in workers' productive endowments (chiefly education) affect poverty and inequality (Fields 2004). In one view, and overall productivity levels in the countries of the earnings are mainly determined by the interplay of the sup- region. The reduction of earnings disparities specifically ply and demand of labor in competitive, frictionless labor associated with gender, ethnicity and race, the informal markets. Differences in wages arise from differences in mar- economy, occupation, sector of employment, or geographic ginal labor productivity and workers' preferences, which in location would have a larger immediate impact on turn depend on individual characteristics either observed inequality than on poverty, particularly in the poorest (such as education and work experience) or unobserved countries in the region. The feedback effects of inequality (such as unmeasured skills or industriousness) and the in the pricing of labor on human capital accumulation quality of the economic and institutional environment (discussed in chapter 9) and the unequalizing role of that determines overall productivity levels. In this view, unmeasured worker characteristics (such as education low labor productivity--resulting, for example, from low quality, labor market ability, and family connections) human capital or technological innovations--is the main deserve greater attention as potential sources of poverty reason for persistent low earnings. A number of researchers traps. adhere to an alternative view of labor pricing in developing Second, a detailed analysis of rural El Salvador and countries that is best characterized by segmented, dualistic consistent evidence from other countries suggest that markets where earnings differences between workers of household-level poverty traps are a phenomenon of practi- similar skills result from discrimination (ethnicity or gen- cal relevance in Latin America and the Caribbean. Not der) or barriers to mobility across occupations (such as everyone benefits equally from growth: often individuals informal/formal jobs), sectors (subsistence agriculture/off- and families with bundles of favorable characteristics farm jobs), and locations (rural/urban areas). These barriers (observed and unobserved) reap faster-than-average income can be related to labor market institutions such as union- growth--this is especially true of the more mobile. Impor- ization, minimum wages, and other labor regulations, and tant complementarities between public investments and to labor market connections and geographic mobility costs. household characteristics mean that poor families often In this second view, labor markets per se generate unequal lack the minimum level of private and public assets advantage and low-earnings traps. required to exploit growth opportunities fully. While lack While analytically useful, this distinction is artificial. of family endowments is the main driver behind persistent Inequality in the pricing of skills has feedback effects to the low incomes and poverty, high volatility and the inability incentives to invest in skills and innovation. As discussed in to ensure against shocks are also important sources of chapter 9, lower returns to schooling associated with exclu- variation in incomes, much more so than in developed sion can help sustain low-education poverty traps. Recent countries. studies find that the process of job reallocation contributes 146 M I C R O D E T E R M I N A N T S O F I N C O M E S : L A B O R M A R K E T S , P O V E RT Y, A N D T R A P S ? 15 to 50 percent of productivity growth in an economy 2002). High levels of education are needed to escape from (IDB 2004). For instance, informality can trap significant poverty in most countries in Latin America. As discussed in resources in low-productivity activities. Lacking access to detail in chapter 9, on average, Latin American workers capital, many micro- and small enterprises cannot capitalize with a university diploma earn one and a half to three times productivity gains through scale economies and innovation as much as uneducated workers, while those with a sec- and may be trapped in a bad equilibrium: because of low pro- ondary degree earn up to one and a half times as much. ductivity, they cannot afford the costs of participating in for- Moreover, returns to schooling tend to be higher (often by mal institutions, but informality in turn limits the potential 2 to 4 percentage points) for workers located higher up in for productivity growth. Hence, A fluid labor market is the earnings distribution given observed characteristics, so important for sustainable increases in productivity in the the payoff to education may depend on a worker's endow- region. ment of unobserved characteristics. Considerable evidence indicates that unobserved hetero- Earnings also depend on demand factors and, more gen- geneity among individuals with the same human capital, erally, a country's economic and institutional environment. sector of work, and demographic characteristics is very Labor productivity trends mimic the region's lukewarm important in explaining earnings levels and earnings differ- overall productivity growth, measured by total factor pro- entials in Latin America and the Caribbean. A large portion ductivity, which was negative in the 1980s and meager in (around 40­60 percent) of earnings inequality in the region the 1990s. In contrast, East Asia experienced a sustained remains "unexplained" by measured worker characteristics.4 increase in productivity and labor earnings during this Factors unobserved by the analyst such as the quality of edu- period. Achieving significant poverty reduction is harder in cation, family background, labor market connections, and countries with a low earnings base (where unskilled workers individual industriousness are distributed unevenly across earn very little), a point illustrated in figure 8.1. The figure workers. These characteristics may grant an advantage in access to high-paying jobs, affecting the returns to skills and the price of labor in the labor market. Workers from poor FIGURE 8.1 families may be disproportionately disadvantaged in these Productivity and wages go hand in hand unobserved earnings determinants. With these issues in Low-wage jobs and productivity mind, this chapter review what is known about the main Nicaragua sources of the level and differences in earnings in the region, and the links to poverty and overall income inequality.5 EI Salvador Peru Earnings and productivity: Education and the Bolivia quality of the economic environment Guatemala A key factor behind the persistent low levels of earnings in Brazil the region is low and stagnant productivity. Real wages Uruguay moved one-for-one with labor productivity between the mid- Chile 1980s and early 2000s (IDB 2004), but labor productivity Costa Rica stagnated during this period, with half of the countries Panama exhibiting a decline. Thus, the scope for sustained earnings gains has been limited, a reflection in part of the region's Argentina sluggish skills accumulation and overall productivity trends. Mexico Education is the single most important individual 0 10 20 30 40 50 60 determinant of earnings, accounting for about one-third of % of workers earning less than $1 PPP an hour overall earnings inequality in the region. One study found Improving economic environment that disparities in educational endowments and in returns Universalizing secondary education to education as one of the main factors driving differences Actual in poverty and income inequality between Brazil, Mexico, Source: Drawn from IDB (2004). and the United States (Bourguignon, Ferreira, and Leite 147 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S reports hypothetical simulations for a sample of 12 coun- adjusted for workers' schooling, parental education, and tries where earnings of unskilled workers are made to match school quality, a typical nonwhite worker with a secondary those of analogue Mexican workers--the country with the education faces a 16 percent lifetime average-earnings dis- largest unskilled hourly wages (as measured in purchasing advantage; while significant, this is far short of the 50 per- power parity dollars) in this particular sample of countries. cent unadjusted earnings gap. Contrary to findings for In the poorest countries, the fraction of low-wage jobs gender, differences in returns to schooling across ethnic and would fall by more or at least as much in this scenario as in racial groups are significant (often 1 to 3 points). Whether a scenario where the labor force had universal secondary they reflect gaps in school quality or labor market discrimi- education at prevailing earnings levels. While highly nation, these unequal returns may discourage skills accumu- artificial, these results highlight that addressing low overall lation by the nonwhite population (see chapter 9). productivity through improvements in the economic and Evidence indicates there may be greater pay discrimina- institutional environment (for example, with policies to tion at higher-salary jobs for any given skill level.10 For foster private investment and technological change) can go a instance, the earnings of the best-paid pardos in Brazil are long way in lowering poverty rates in the region.6 similar to those of the best-paid white workers, but when comparing workers at the bottom of the salary scale pardos Earnings disparities unrelated to skills and pretos face the same earnings disadvantage relative to Differentials in earnings adjusted for human capital are whites. Thus the gradient of skin color affects mobility quantitatively important in the region. Earnings disparities opportunities, so that the saying in Brazil "money whitens" associated with gender and ethnic or racial background are applies only to pardos. In Chile, the gender wage gap often attributed to labor market discrimination. Sectoral, increases from 10 percent to about 40 percent as women occupational, and location earnings inequality may reflect move up the earnings distribution. The returns to experi- segmentation that impedes labor mobility to higher- ence are similar for women and men in the lower part of the paying jobs or earnings differentials related to fringe or earnings distribution, but are significantly lower in the top nonmonetary characteristics of jobs. of the distribution. Thus, labor market discrimination While women likely experience some degree of discrim- seems more likely to occur when workers cannot be denied ination in the labor market, it does not seem to be of first the higher-paying jobs within occupations on the basis of order. The gender gap in average earnings (adjusting for their observed productive attributes (Darity and Mason education and potential experience) ranges from 12 percent 1998). in Mexico to 47 percent in Brazil, and improved during the The poor are often employed in agriculture, construc- 1990s to almost match the gender gap in the United States, tion, retail-trade sectors, and informal occupations, and they which nevertheless is still wider than the gender gap in tend to live in laggard areas, all of which cause their wages to most other OECD countries. The gender gap in Latin be lower regardless of skills.11 As noted in chapter 7, America also reflects the effect of women's role in the regional earnings gaps within Latin America are also quanti- household on their labor force participation and occupa- tatively important given that poorer regions lack natural tional choice.7 Moreover, women do not generally face a resources as well as agglomeration externalities in skills, disadvantage in the returns to investments in schooling. infrastructure, and other factors of production. Raceandethnicityareamoresignificantsourceofearnings Of particular interest are earnings gaps between formal disadvantage.8 The indigenous population in the region on and informal jobs. Salaried workers in the informal econ- average earns 46 to 60 percent of the earnings of the non- omy and the self-employed account for 25 to 70 percent of indigenous population, while pardos (mixed race) and pretos employment across countries in the region. The average (blacks) in Brazil earn just over half of average earnings for earnings gap between workers in small firms (a proxy for whites. Poverty rates are also higher for indigenous popula- informal wage employment) and those in large enterprises is tions in Bolivia, Guatemala, and Peru and among African about 30 percent (similar to the gap in the United States) descendants in Brazil. The limited evidence suggests that and ranges from 17 to 51 percent across countries (IDB these higher poverty rates arise largely from the disadvantage 2004). Average earnings for the self-employed (most of nonwhites face in human capital (quantity and quality) whom are also informal) are typically far less than those of and its returns.9 In Brazil, after racial earnings gaps are formal salaried workers. The informal-formal earnings gaps 148 M I C R O D E T E R M I N A N T S O F I N C O M E S : L A B O R M A R K E T S , P O V E RT Y, A N D T R A P S ? primarily stem from low skill endowments despite unequal Overall, the evidence summarized above suggests that rewards to skills. Around two-thirds of the informal-formal earnings differentials unrelated to skills are a second-order average earnings gap is explained by differences in worker source of low earnings relative to differences in workers' skill endowments, and the rest by a lower remuneration to productive endowments. While debate continues about the these endowments in the informal sector.12 policy significance of these earnings differentials, it is clear Moreover, the pattern of informal-formal remuneration that facilitating labor mobility is key if the poor are to gaps along the earnings scale is consistent with a two-tier escape their condition. This issue is discussed next. informal sector. This is illustrated in figure 8.2 for Bolivia. It decomposes the informal-formal earnings gap into a por- tion attributable to differences in measured characteristics Market segmentation and mobility across workers in each sector and a component attributable The applied literature on what makes growth more pro-poor to differences in how each sector rewards such characteris- has focused on how the pattern of growth affects poverty. As tics for workers in the 10th, median, and 90th earnings noted in chapter 5, studies have shown that growth brings percentiles in each sector. The latter component is often about more poverty reduction when it extends to the geo- taken, although not without question, as a measure of seg- graphical areas or sectors where the poor are concentrated mentation. The results suggest that segmentation might so as to make more intensive use of unskilled labor. This exist for informal salaried workers in low- to average- report does not deal with the complex issues--such as the paying jobs and for the self-employed at low-paying jobs sources of growth or the political economy of government for their skills set. At the best-paid jobs for any skill level, intervention--surrounding "industrial" (or selective) poli- the returns to skills are similar between sectors so that cies to induce a sectoral bias in growth. In any event, the evi- these workers can move between sectors with little wage dence provided here and in the 2005 regional flagship report penalties. Similar patterns are found in Argentina, Brazil, Beyond the City: The Rural Contribution to Development (de and the Dominican Republic. Ferranti and others 2005) points in another direction. The FIGURE 8.2 Earnings gap between the formal and the informal sectors in Bolivia, 2002 Workers in the informal sector paid in the Self-employed workers paid in the formal sector (using formal sector returns) formal sector (using formal sector returns) Log earnings gap Log earnings gap 1.00 1.20 1.00 0.80 0.80 0.60 0.31 0.25 0.00 0.60 0.76 0.40 0.40 0.51 0.20 0.39 0.43 0.20 0.37 0.32 0.22 0 0 10th quantile Median 90th quantile 10th quantile Median 90th quantile (low-pay jobs) (average-pay jobs) (top-pay jobs) (low-pay jobs) (average-pay jobs) (top-pay jobs) Due to difference in worker characteristics Due to difference in sector prices Source: Based on Tannuri-Pianto, Pianto, and Arias (2004). Note: Earnings regressions controlled for education, work experience, economic activity, gender, ethnicity, demographic and regional effects, and corrected for differences in the probabilities of self-selection into each sector. 149 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S incomes of the poor thrive when the poor are able to diversify who are unemployed and those out of the labor force) in to more viable economic activities. Argentina, Brazil, and Mexico, Bosch and Maloney (2005) Since development involves a shrinking agricultural find significant evidence supporting the latter view. Fig- sector and increasing urbanization, longer-term poverty ure 8.3 illustrates this for Mexico. Patterns of movements reduction depends crucially on the ability of the poor to across sectors are consistent with the sectors showing a fair engage in dynamic (competitive) economic activities. In degree of integration and transitions not solely driven by some cases market segmentation may prevent mobility earnings differentials, although informal jobs take on more because workers in low-earnings sectors, occupations, and slack during downturns. regions face high costs or barriers to mobility. In others, However, as noted earlier, a nonnegligible fraction of differences in nonmonetary benefits of jobs mean that informal workers face earnings penalties that are too large observed mobility may be lower than one would expect and that are not offset by nonmonetary benefits; these earn- given observed earnings differentials. ings penalties may be related to low-productivity traps One important issue concerns movements out of subsis- resulting from lack of skills or credit constraints. Moreover, tence agriculture to higher-yield crops or to nonfarm rural since access to social protection (such as health care or activities. As stressed in the 2005 flagship report, evidence pensions) in most of the region remains tied to a formal from country studies underscores the critical importance employment contract and since informal workers face for poor households of a minimum bundle of asset holdings higher unemployment risk, they may be disinclined to (chiefly, human capital and rural roads) and risk protection upgrade their skills and diversify to more promising occu- (such as remittances and safety nets) so that they can under- pations (both formal and informal). take productive diversification strategies. For instance, Recognizing the considerable heterogeneity in the using panel data for El Salvador, Tannuri-Pianto, Pianto, informal sector, researchers are beginning to agree that the and Arias (2005) find that more-educated households and informal sector has two distinct components: workers who those with other asset holdings such as stable access to elec- choose this sector voluntarily and conform more closely to tricity and proximity to a paved road are more likely to rely entrepreneurship motives, and those who use this sector as heavily on off-farm activities for their income generation. employment of last resort. The relative size of each tier Moreover, these effects are multiplicative. Closer proximity depends on country-specific contexts, particularly on the to rural roads increases the chances that individuals with level of productivity in the formal sector, the demographic more initial asset holdings will shift from agriculture to and skills composition of the labor force, and the incentives nonfarm employment compared with individuals with resulting from tax and labor regulations. fewer assets. Remittances reinforce the impact of education Finally, as discussed in chapter 7, the spatial pattern of on the probability of leaving agriculture. This means that economic growth can influence the effect that poverty families lacking a minimum bundle of assets and risk reduction has on a given growth rate, especially if trans- mitigation capacity are less likely to benefit directly from portation and market connectivity are low and migration off-farm employment opportunities induced by rural costs are high. That chapter highlighted some of the issues investments. related to geography and cultural factors that may con- In urban areas, a key question is the extent to which tribute to persistent spatial earnings differentials and thus informal and formal sector participation reflects segmenta- be a source of poverty traps. Country case studies of house- tion or voluntary choice. The conventional view of the infe- hold determinants of migration indicate that the young, riority of informal jobs has been questioned (Maloney moderately educated (secondary or primary), women, and 2004). An alternative view points out that many informal smaller families are more likely to migrate to urban locali- salaried and self-employed workers (especially youth, mar- ties, but that individuals from the poorest locations and the ried women, and the unskilled) may voluntarily choose this indigenous are more prone to rural-to-rural migration sector as an entry point to the labor force and to enjoy non- (Tannuri-Pianto, Pianto, and Arias 2004; de Ferranti and monetary benefits such as greater flexibility, the ability to others 2005; see also Taylor, Yúnez-Naude, and Cerón exploit entrepreneurial skills to improve mobility, and 2004, and Taylor 2001 for Mexico). The persistence of avoidance of burdensome regulations. In studying patterns regional earnings gaps and small migration flows should of transitions across employment states (including those receive more attention in the region's policy agenda. 150 M I C R O D E T E R M I N A N T S O F I N C O M E S : L A B O R M A R K E T S , P O V E RT Y, A N D T R A P S ? FIGURE 8.3 Transitions between the formal and informal salaried sectors, and between salaried employment and self-employment in Mexico, 1987­2001 Probability of self-employment Probability of salaried employment Probability of informal sector Probability of formal sector 0.12 0.40 0.08 0.07 0.10 0.35 0.07 0.06 0.30 0.08 0.06 0.05 0.25 0.06 0.05 0.04 0.20 0.04 0.04 9Q1 1987Q1 1989Q1 1991Q1 1993Q1 1995Q1 1997Q1 199 2001Q1 1987Q1 1989Q1 1991Q1 1993Q1 1995Q1 1997Q1 1999Q1 2001Q1 Probability of Probability of Probability of Probability of self-employment salaried employment informal sector formal sector Source: Bosch and Maloney (2005). Microdrivers of changes in the The selected countries reflect a variety of trends in income distribution poverty and inequality in the region. Argentina suffered a In this section, we ascertain the quantitative impor- dramatic increase in income poverty and inequality during tance of the numerous earnings determinants in driving the 1990s. Peru also saw a significant increase in both vari- the growth path of earnings for individuals with different ables between 1997 and 2002. Bolivia experienced a mod- characteristics. We do this by isolating the quantitative est reduction in urban poverty during the first half of the contribution of the different factors to past changes in the 1990s, followed by an increase during 1997­2002. The income distribution. This exercise also helps illustrate the Dominican Republic saw little change in poverty during profile of workers who have been benefiting from growth this period, and a large increase in the inequality of labor as well as the profile of those who have been left behind. incomes. Tables 8.1 and 8.2 illustrate the main results for We look particularly at changes in poverty and inequality Argentina and Peru. in a few selected countries. Recent studies for Argentina Overall, changes in poverty and income inequality in (1992­2001), Bolivia (1993­2002), the Dominican the region during recent episodes of economic growth and Republic (1997­2002), and Peru (1997­2002) used fairly downturn reflect several microforces, some reinforcing, comparable microsimulations of counterfactual income others counteracting each other. Forces that lead to distributions that allow unobserved worker skills to affect unequalizing income growth have dominated and explain the returns to the worker's characteristics.13 The analysis the disconnect between the performance of the overall here extends the simple growth-redistribution decomposi- economy and incomes of families at the lower end of the tions in chapter 4. The main goal is to find answers to the distribution in several countries. question: what would the level of poverty (inequality) have Particular note should be taken of two common forces. been in the country if factor X (such as education or its First, the unequalizing effect of a moderate upgrading of returns) had not changed? The question is answered by the educational level of the workforce is fairly visible and simulating the distribution of income that results from accentuated by the rise in the returns to higher education. changes in the relevant factor while all others are kept Unskilled earnings, primarily in agriculture, tend to lag unchanged, that is, by estimating a counterfactual distrib- behind and prevent many rural families from benefiting ution (see annex 8A). from growth and escaping poverty. Second, researchers are 151 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S TABLE 8.1 education returns is magnified by the uneven change in the Decompositions of poverty and inequality changes in Argentina, returns across workers at different points of the adjusted 1992­2001 earnings distribution (except in Argentina). This finding may suggest that among well-educated workers, those from Inequality Poverty better-quality schools or with better connections have been Effect Ginia FGT(0) FGT(2) able to cling to the better-paid jobs. Furthermore, the generally small contribution of indi- Observed change in hours of work 8.0 17.3 17.3 vidual factors to changes in poverty and inequality points to 1. All the coefficients 2.9 7.5 2.1 the inadequacy of single explanations for the sources of dis- Returns to education 0.8 -1.2 -0.3 Gender gap 0.1 -0.8 -0.2 tributional change. Individuals with some bundling of Returns to regions -0.1 -0.4 -0.1 favorable characteristics are more likely to take advantage of Number of children -0.1 1.0 0.4 2. Structure education -0.3 -0.4 -0.2 better employment opportunities throughout the growth 3. Structure children -0.1 -0.1 -0.1 process. Evidence on this is presented in the next section. Observed change in earnings 7.5 7.6 7.6 Determinants of income dynamics: 1. All the coefficients 1.6 1.1 0.7 Returns to education -0.4 -1.1 -0.4 Lessons from rural El Salvador Gender gap 0.7 0.4 0.1 Our previous discussion of the main sources of labor earn- Returns to regions -0.1 0.0 0.0 Returns to sectors 0.7 0.4 0.1 ings differences in the region and their evolution over time 2. Structure education 0.8 -0.9 -0.2 relied on cross-section data; this approach presumes that the growth path of earnings (and its determinants) for any Source: Based on Bustelo (2005). given individual and his relative position in the earnings a. Based on equivalent household income. F.G.T. = Foster, Greer, Thorbecke indicator distribution is well represented by the growth path of average earnings and the rank of a typical individual with similar characteristics. For example, the change in average earnings of a typical college-educated worker is taken as a proxy for the increase in earnings experienced by all TABLE 8.2 workers with a college education. Decompositions of poverty and inequality changes in Peru, As discussed in chapter 2, this approach may not pro- 1997­2002 vide adequate answers to questions such as whether poverty is transitory or permanent. Nor does it reveal the factors Inequality Poverty that make poverty transitory for some individuals and per- Effects Ginia FGT(0) FGT(2) manent for others. Answering these questions requires Observed 1997­2002 3.5 6.3 2.7 longitudinal data sets that are rarely available in Latin Returns to education 1.0 0.3 0.2 America. In the following discussion, we examine in some Gender wage gap -1.3 -1.1 -0.9 detail the empirical relevance of some of the mechanisms Returns to experience 6.5 -8.3 -4.1 Education 1.0 -0.2 -0.1 that may lead to poverty traps by using a unique panel data Regions 1.5 -0.1 0.3 set of close to 500 rural households in El Salvador Sectors -0.7 -0.9 -0.2 (FUSADES­Ohio State University, hereafter dubbed BASIS) continuously followed during a six-year period Source: Based on Sosa-Escudero and Lucchetti (2004). a. Based on equivalized household income. (1995­2001; see the annex 8A). Although six years is not a great time span, it is a major improvement over the one-to two-year panels that have been used to study mobility in increasingly recognizing the importance of unmeasured Latin America. This data set also allows more careful analy- worker skills for labor market performance; these skills sis of the confluence of unfavorable characteristics that may include school quality, labor market connections, and conspire to generate persistent poverty and inequality. We unmeasured individual ability (such as spunk or industri- rely on existing studies using these data and new analysis of ousness). The effect on income inequality of changes in the main microdeterminants of growth in incomes, 152 M I C R O D E T E R M I N A N T S O F I N C O M E S : L A B O R M A R K E T S , P O V E RT Y, A N D T R A P S ? accounting for the role of unobserved heterogeneity of complementarities between income determinants (observed households and individuals in rural El Salvador.14 and unobserved). In addition to the availability of better-quality data, El The BASIS data confirm that determinants of income Salvador offers a promising context in which to study these growth are fairly similar to those entering cross-sectional issues. The country achieved considerable improvements in earnings functions. Numerous analyses with this data set poverty and other indicators of living conditions during indicate that assets endowments (land, education), access the 1990s. Rural poverty fell by 20 points according to the to markets and infrastructure (road, credit), household national household survey and by 28 points using the risk-coping strategies (productive diversification, microen- BASIS data, which provide information on rural incomes in terprise development, remittances), and household demo- greater detail and probably greater precision (see World graphics (size, composition, gender) all affect family Bank 2005e). Much of the progress in rural areas is related income growth. Tables 8.3 and 8.4 present the results of to a significant economic diversification away from tradi- random effects (RE) and fixed effects (FE) regressions of tional agriculture such as basic grains, coffee, and sugar to individual wages and per capita household incomes on off-farm productive activities; important investments in relevant socioeconomic characteristics (Tannuri-Pianto, rural infrastructure that improved access to markets; and an Pianto, and Arias 2005).15 The FE results are presented for important inflow of international remittances. Yet half of three quantiles of the earnings-income distribution to Salvadorans in rural areas remain poor, and a quarter live in investigate whether the returns to observed characteristics mere subsistence. While the findings of one country study depend on unobserved (unmeasured) income determinants. clearly cannot be directly extrapolated to the entire region, More detailed results are discussed in the next section. they do offer important insights into the mechanics of The main overall findings are: income and poverty dynamics in a context of significant poverty reduction driven by private strategies and · Nonfarm jobs carry a large wage premium, which public investments. We first discuss the findings on the varies with gender and a worker's initial education determinants of income growth and the importance of level.16 Switching to a nonfarm activity increases TABLE 8.3 Determinants of rural individual wages, El Salvador Individual earnings equations Mean regressions Quantile fixed effects regressions Fixed effects Random effects 25th 50th 75th Effect on log hourly wages, in 2001 colones Coefficients Coefficients Education 0.011 0.021*** 0.008 0.013 0.014* Experience 0.028** 0.014*** 0.018 0.018 0.022 Experience^2 -0.033 -0.021** -0.002 -0.005 -0.013 Head household 0.114** Female -0.018 Nonfarm main sector 0.135* 0.205*** 0.156* 0.252* 0.316* Distance from bus stop (km) -0.047*** -0.045*** -0.031* -0.039* -0.038* Distance from bus stop^2 0.0017* 0.0023*** 0.0003 0.0012 0.0017 Distance from bus stop * education 0.002 -0.001 0.003* 0.001 0.000 Nonfarm * Female -0.236* -0.153** -0.118 -0.157 -0.228* Nonfarm * Education 0.023* 0.043*** 0.013 0.008 0.009 Constant 1.203*** 1.144*** 1.150* 1.230* 1.220* Regional and year dummies Yes Yes Yes Yes Yes Source: Based on Tannuri-Pianto and others (2005). *Significant at 10 percent. **Significant at 5 percent. ***Significant at 1 percent. 153 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S TABLE 8.4 Determinants of rural per capita family incomes, El Salvador Household income equations Mean regressions Quantile fixed effects regressions Fixed effects Random effects 25th 50th 75th Effect on log yearly per capita income, in 2001 colones Coefficients Coefficients Average education workers 0.011* 0.026*** 0.004 0.007 0.012* Log number of workers in household 0.118*** 0.131*** 0.075* 0.086* 0.108* Log number of children and elderly -0.052** -0.073*** -0.036* -0.042* -0.084* Distance to paved road -0.005 -0.006* -0.001 -0.002 -0.006* Distance to paved road^2 0.0002 0.0002** 0.0000 0.0000 0.0002* Electricity -0.058 0.023 -0.029* -0.020 -0.021 Formal credit 0.153*** 0.110** -0.008 0.008 0.040 Other credit -0.005 -0.007 -0.030 -0.022 -0.023 Remittances (*10,000) -0.0122 -0.0115 -19.3000 -9.2500 9.1500 Subsidies (*1,000) 0.0012 -0.0010 0.8320 2.6000 3.9300 Activity diversification index 0.001 0.000 0.001 0.001* 0.000 Number of Microenterprises 0.076*** 0.084*** 0.077* 0.062* 0.089* Non traditional farm sector 0.013 0.046 -0.006 0.019 0.034 Non farm sector 0.163*** 0.185*** 0.145* 0.147* 0.156* Constant 9.367*** 9.301*** 9.210* 9.300* 9.450* Regional and year dummies Yes Yes Yes Yes Yes Source: Based on Tannuri-Pianto and others (2005). *Significant at 10 percent. **Significant at 5 percent. ***Significant at 1 percent. average wages for males by 14 percent. Meanwhile, tional levels over the panel. The income gains from only well-educated females benefit from joining the education for a household that remains predomi- nonagriculture sector, those women with below- nantly on the farm are lower (1.1 percent) than if average education (three and five years in the tradi- they had switched to nonfarm activities, although tional agriculture and nonfarm sectors, respectively) again this finding may be downward biased (the can even experience wage losses. effect is twice as large in the random effects regres- · Households that engage more intensively in non- sion). Workers that are closer to markets earn higher agriculture activities accrue a 17 percent income wages, perhaps because they incur lower transaction gain. Surprisingly, there is no significant income costs (in time and money) associated with engaging difference between traditional and nontraditional in the market economy. Earnings decrease with dis- agricultural households, suggesting that partial tance from the market, declining by about 4 percent diversification to nontraditional crops fails to boost a kilometer and reaching a maximum penalty of agricultural incomes once one controls for household 27 percent for workers at about 10 kilometers from characteristics and idiosyncratic effects that affect a bus stop (more than 80 percent of workers are at activity choice. least that far away). Similarly, households that get · Returns to education are seemingly low in the agri- closer to a paved road also derive higher per capita cultural sector and at least twice as large in nonfarm incomes, the effects being very similar to those on employment, as identified through workers who labor earnings. switch sectors. Changes in education do not corre- · Having or gaining access to formal credit positively late significantly with mean earnings, likely as a affects incomes by 15 percent, while informal credit consequence of very little real variation in educa- has no discernible effect on average family incomes. 154 M I C R O D E T E R M I N A N T S O F I N C O M E S : L A B O R M A R K E T S , P O V E RT Y, A N D T R A P S ? · Families' capacity to diversify risks has a mixed level may be needed for potentially profitable investments impact on family incomes. Income diversification and income diversification to materialize. (measured by the Simpson Diversification Index), These issues can be examined in two ways: first, by remittances, and subsidies do not affect average including nonlinear terms and interactions between rele- family per capita income.17 However, opening a vant observed characteristics in the income (labor earnings) microenterprise increases income by about 8 percent. regressions shown in tables 8.3 and 8.4; and, second, by allowing the returns to observed characteristics to depend on the conditional income or earnings quantile of the These results offer some comfort that the conclusions household or worker, that is, on its rank in the income derived from cross-section income differentials are a good (earnings) distributions that would obtain if all workers approximation of the drivers of income growth. Next we had the same measured characteristics (for methodological focus on the role of complementarities between public details, see the annex 8A).18 The conditional quantile of a investments and household characteristics (observed and household or worker depends on unobserved characteristics unobserved) that lead to lower income growth for many such as school quality, or work ethic, or differences in poor families. These effects can rarely be isolated with household productivity, such as differences in cropping cross-section data given the high colinearity between methods or soil yield. Coefficients that increase (decline) socioeconomic characteristics (such as the high confluence significantly over the quantiles indicate that unobserved of unfavorable characteristics among the poor), a problem income determinants operate as complements to (substi- that is overcome by the time variation in a panel context. tutes for) the relevant measured characteristic. For exam- ple, households with idiosyncratically low productivity When it rains it pours: complementarities may benefit less from having access to credit or being closer and initial conditions matter to markets, in which case the returns to credit and rural One of the main mechanisms behind poverty traps is the roads will be lower at the bottom quantiles of the condi- existence of minimum thresholds and strategic comple- tional income distribution. mentarities caused by externalities or coordination failures The results indicate that complementerities play an in production or income generation. These can arise under important role in determining which rural families share limited capacity to face catastrophic shocks, credit market fully in income growth opportunities (Tannuri-Pianto, restrictions (resulting from imperfect credit information Pianto, and Arias 2005). Individuals and households with and low collateral), and fixed costs of carrying an invest- bundles of favorable characteristics observed or unobserved ment that households cannot amortize in the short term. reap faster income growth, especially those moving out Households may be unable to borrow or save the minimum of agriculture. Some of these findings are illustrated in amount necessary to go beyond the fixed cost or the outlay figure 8.4. The main conclusions are summarized here: required for an investment to be profitable, be it the adop- tion of a more modern cropping technique or investments · Often a minimum level of education (an average of in higher education. In other words, convex or lower initial six years among family members) is needed for returns to investments may prevent making investments households to fully exploit the income gains from that become profitable only beyond a given investment improvements in access to roads and credit and to threshold. Strategic complementarities occur when indi- leverage remittances. vidual decisions or private rates of return to investments · The impact of road proximity and human capital on depend on a family's initial assets and the broad capital income growth depends on unobserved income deter- stock. For example, whether a household benefits from the minants. Closer road proximity does not affect paving of rural roads may depend on its level of assets and incomes of households at the bottom 25 percent of human capital and on its access to credit. This interdepen- the income distribution given their observed charac- dency can give rise to coordination failures that prevent teristics, while those in the top 25 percent reap the entire regions or population groups from diversifying to highest income gains. economic activities with higher returns. Minimum coordi- · Higher remittances correlate with increases in labor nation of investments at the national, regional, or group income only among households with more education 155 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 8.4 Complementarities in the income generation process in rural El Salvador Impact of nonformal credit on annual growth of average Association between remittances and annual growth in average per capita income by household education level per capita nonremittance income by household education level % change in annual per capita income % change in annual per capita income 9 0.5 6 0.3 0.1 3 0 0 0.1 3 0.3 6 0.5 9 0.7 0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10 Years of education Years of education Impact of formal credit on annual growth of average Earnings gains of moving from agriculture to a nonfarm per capita income by distance to paved roads activity by quantile of the earnings distribution % change in annual per capita income Change in earnings, % 32 40 28 35 24 30 20 16 25 12 20 8 15 4 0 10 0 1 2 3 4 5 6 7 8 9 10 Lowest quantile Middle quantile Highest quantile Distance to paved roads Source: Based on Tannuri-Pianto and others (2005). Note: The impact of remittances is illustrated for marginal changes of 1,000 colones. and higher idiosyncratic productivity. For the less unfavorable characteristics are more likely to sink in low- educated, the regression correlation is negative, wage farm and off-farm jobs. Informal credit, remittances, which may suggest that remittances serve as a safety and unobserved income determinants all complement a net (they smooth negative income shocks) or that household's human capital in generating income. In many they may induce negative labor supply effects (by cases, a minimum of primary education appears to be neces- increasing the reservation wage at which individuals sary for households to fully exploit the benefits from credit accept work). and remittances. Moreover, road access partially substitutes for lack of education (and vice versa) so that rural transporta- These results uncover evidence of threshold and interac- tion investments have a greater benefit for more-isolated and tion (bundling) effects that prevent the poor from benefiting less-educated households, which are more likely to be poor. fully from rural investments and their own diversification Similar results were found in studies for Peru (Saavedra and strategies and may also discourage them from undertaking Torero 2004) and for other countries in Central America potentially profitable investments. Individuals with largely (World Bank 2004c). A minimum coordination of public 156 M I C R O D E T E R M I N A N T S O F I N C O M E S : L A B O R M A R K E T S , P O V E RT Y, A N D T R A P S ? interventions in rural areas is needed to exploit these syner- and continuing more recently with work by Geweke and gies and overcome the associated threshold effects that con- Keane (2000), this literature has focused on developed strain the incomes of households with a bundling of countries where relatively longer panel data allow examina- unfavorable characteristics. tion of long-term income persistence. The second strand is Now that we have established the main microdetermi- more common in developing countries and regions like nants of income growth and found that they usually inter- Latin America where short panels or cross-section data have act in reinforcing or offsetting ways, it is natural to ask been used to examine the link between poverty and the whether the dynamics of the income generation process are inability to insure risks (see, for example, Chaudhuri, such that low-income status and thus poverty tend to per- Jalan, and Suryahadi (2002); Chaudhuri (2000); Pritchett, sist over time. That is, what are the chances that low- Suryahadi, and Sumarto (2000); Jalan and Ravallion (1999); income families in El Salvador in 1995 will still be and Ravallion and Chaudhuri (1997). Vulnerability arising low-income families in 2001? How much of this persistent from high volatility requires interventions to reduce and poverty hinges on idiosyncratic and transitory characteris- insure risks, while vulnerability arising from low endow- tics of families (measured and unmeasured), and how much ments calls for policies to support the accumulation of on external shocks or fortune? We turn to these questions endowments and long-term income potential. next. Table 8.5 illustrates the transitory (vulnerability) and permanent (persistence) aspects of rural poverty in El Income and poverty persistence: Shocks, Salvador. The BASIS data reveal the considerable income observed and unobserved endowments volatility faced by rural Salvadorans.19 In any given year, Income dynamics are best understood under the "permanent- the results show, the poverty rate hides continuous transitory income hypothesis" of Friedman and Kuznets movements in and out of poverty of different individuals. (1954), which assumes two components in the determination Around 6 out of 10 rural households fell into poverty tem- of incomes over time. One is a permanent component that porarily during 1995­2001, although more than half of reflects an individual or family long-term income potential these had an income stream above the poverty line for related to productive characteristics such as human capital, most of the period. In addition to the inherent risk other assets, and unmeasured skills. The second is a transitory attached to rural incomes, this volatility reflects a series of component that captures external factors, such as economic aggregate shocks including two earthquakes and the swings, individual-specific shocks, or plain measurement impact of declining world coffee prices on coffee produc- error, that cause incomes to depart from their permanent ers. At the same time, almost 4 out of 10 households never level. In subsequent empirical work, the issue of income and TABLE 8.5 Permanent and transitory poverty in rural El Salvador, 1995­2001 poverty persistence has been studied from the perspective of intergenerational income mobility and more recently of poverty vulnerability. In essence both views ask how likely Percent with Percent with average per average per it is that a household of given characteristics will find itself capita incomes capita incomes over the period over the period in poverty at a given future time. The answer ultimately Percent of below the above the depends on the household's long-term consumption States households poverty line poverty line prospects and the consumption volatility it faces. In theory Permanent poor 25.1 25.1 n.a. a household can be continuously poor because its endow- (all 4 periods) ments yield only low-income potential or because it is sys- Transient poor 61.9 27.9 33.9 3 of 4 periods 24.8 21.1 3.8 tematically affected by income shocks that it is unable to 2 of 4 periods 19.7 6.0 13.8 smooth. Each of these factors depends on the state and evo- 1 of 4 periods 17.2 0.9 16.4 lution of household characteristics (observed and unob- Nonpoor 13.1 n.a. 13.1 Percent of 100 53.0 47.0 served) and on the aggregate environment. The literature households on intergenerational income mobility has emphasized the first aspect. Starting with the classic work in the United Source: Based on Beneke de Sanfelíu and Shi (2004). States by Lillard and Willis (1978) and MaCurdy (1982) Note: n.a. = not applicable. 157 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S crossed the poverty mark: one-quarter of all households FIGURE8.5 remained poor the entire period, while 13 percent always SourcesofpersistentpovertyandlowincomesinruralElSalvador stayed above the poverty threshold. This finding points to the significance of the structural determinants of poverty Impactofselectedcharacteristicsonpovertyvulnerability (changeinriskrelativetobeingchronicallypoor) in rural El Salvador. Relativeriskratio Which factors make poverty transitory for some individ- 1.6 uals and permanent for others? What is the role of "uncon- 1.4 trollable" factors such as economic shocks or unexploited 1.2 externalities such as a lack of public goods? Recent studies 1.0 0.8 with the El Salvador data point to some valuable answers, 0.6 illustrated in figure 8.5. With respect to poverty vulnera- 0.4 bility, human capital of the family, its proximity to mar- 0.2 kets, and its reliance on subsistence agriculture (proxy of 0 Averageeducation Distancetopaved Basicgrainscrop risk aversion or the inability to self-insure from risk) all ofworkers roads(km) (dummy) increase the probability that a rural Salvadoran household Vulnerablepoor Vulnerablenonpoor Nonpoor remained permanently poor during 1995­2001. The level of human capital was a particularly strong factor in deter- SourcesofvariabilityinincomesinruralElSalvador mining whether families were likely to sink into poverty or Percent become highly vulnerable to falling into poverty. 100 In a study using the El Salvador data, Rodriguez-Meza 80 and Gonzalez-Vega (2004) found evidence that the risks faced by households to materialize its future consumption 60 prospects given its current characteristics (observed and 40 unobserved) are a possible cause of poverty traps. Their 20 study showed that recovery from an income shock is quick 0 for the relatively rich in rural areas but much lengthier for Model1 Model4 the poor. This result, however, might be somewhat sensi- tive to estimation methods since they use a short time span Timevariance Individualspecificvariance to identify highly nonlinear income dynamics. In a background study for this report, Sosa-Escudero, SourcesofpersistencyofincomesinruralElSalvador Marchionni, and Arias (2005) used a different approach Percent 60 that focuses on the sources of income persistency. Their evidence shows that transitory income shocks are the 50 major source of variation in incomes across rural families 40 in El Salvador, much more so than in developed coun- 30 tries.20 However, the correlation of bad shocks is relatively 20 low (0.24) in these data. Over a lifetime, good shocks and 10 bad shocks cancel each other out so that transitory shocks 0 are not as important in determining whether an individ- Model1 Model4 ual's or a household's income stays the same as are endow- Timepersistency Individualspecificpersistency ments, including unobserved income determinants. Totalpersistency Indeed, about two-thirds of the persistency in low- and high-income states is attributable to idiosyncratic differ- Source:BasedonSosa-Escudero,Marchionni,andArias(2005);and BenekedeSanfelíuandShi(2004). ences between families, including unobserved heterogene- Note:Model1 timedummies,andmodel4addscontrolsfor ity. Observed income determinants, chiefly education, familycharacteristicsincludinginteractionseffects.Measuresof timeandindividualpersistencydonotadduptototalpersistency account for about half of this income persistence. Conse- andshouldnotbecompared.Therelevantcomparisonisthe changeineachcomponentmovingfrommodel1tomodel4. quently, low income potential is a strong predictors of low 158 M I C R O D E T E R M I N A N T S O F I N C O M E S : L A B O R M A R K E T S , P O V E RT Y, A N D T R A P S ? incomes later in life. In other words, while a large propor- labor productivity. Although of second-order impor- tion of total cross-section inequality (as measured by the tance, reducing the portion of these earnings gaps variance of logarithmic incomes) is explained by income associated with discrimination and labor market fric- instability, life-cycle inequality results largely from the tions can boost the incentives for disadvantaged permanent income component, particularly from the rela- groups to invest in skills acquisition and facilitate tively time-invariant productive characteristics of families the mobility of workers. and their members. · Bridging the gaps in education (both quantity and Being persistently poor in rural El Salvador thus seems quality) and other productive characteristics of work- more likely to result from the lack of endowments needed ers can go a long way toward reducing the wide earn- to escape a lifetime of low income than from the inability ings disparities in the region. But it will not be to ensure against income shocks. Many of these endow- enough to reduce poverty significantly. In most coun- ments can be influenced by policy interventions, although tries, low levels of labor productivity are a chief con- not always in the short term; in particular, it takes one straint to earnings potential. Thus policies that to two decades for a family to accumulate levels of human promote an economic and institutional environment capital sufficient to escape poverty. conducive to productivity growth are important for reducing the incidence of low-paid jobs and making investments in skills more attractive. Implications for policies · Labor market interventions, including changes in The findings reviewed in this chapter suggest several policy labor legislation and its application, should focus on approaches that could improve the prospects for more equi- achieving a better balance between protecting work- table growth and poverty reduction: ers and unleashing the potential for productivity growth in the region. This calls for actions aimed at · Most of the earnings differentials, and thus poverty reducing discriminatory practices or location-specific and income inequality in Latin America and the biases and facilitating the mobility of workers such as Caribbean, are not generated by earnings differentials more effective enforcement of equal pay and merit in the labor market; instead, these differentials reveal promotion regulations, labor market intermediation what firms and workers bring to the market. Many of services, more flexible work schedules, and establish- the poverty and earnings disparities in the region ment of child care centers. reflect the level of productivity of firms and differences · The evidence from rural El Salvador indicates that in workers' productive endowments; distortions in the despite considerable persistence in individuals' and allocation of workers and jobs are of second-order households' sectors of specialization, there is room for importance. What is important are the feedback public policies to encourage mobility. Education and effects to human capital accumulation the labor mar- access to services (such as electricity and water) and ket creates through the pricing of labor (earnings markets (roads) affect the probabilities of transition- returns). There is a need to reverse the unequalizing ing from the farm to the nonfarm sector and vice role of unmeasured worker characteristics (such as versa. deficiencies in early-childhood development, educa- · The poor are generally disadvantaged in several tion quality, and labor market connections) in com- dimensions. We find significant evidence of impor- manding higher wages. This is discussed in more tant complementarities between rural investments detail in chapter 9. and rural household characteristics (observed and · Labor markets do not seem to operate with pervasive unobserved) in determining the probability of sector segmentation. The reduction of residual earnings dis- participation and the returns to their income-deriving parities associated with gender, ethnicity and race, endowments. Public investments and policies in one informality, occupation, sector of employment, and area (such as credit access or road construction) may geographic location would have a larger impact on have heterogeneous impacts depending on the initial reducing overall inequality than on reducing poverty conditions affecting the poor, particularly their levels, a finding that is symptomatic of overall low observed and unobserved productive endowments. 159 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S · Access to markets can be increased through invest- Estimation methods ments in basic infrastructure, which contribute to a A primer on quantile regressions household's ability to attain the minimum level of The technique of quantile regression (Koenker and Bassett wealth, educational skills, or credentials needed to 1978) is used extensively in the background studies for this move to modern occupations. Rural development chapter and chapter 9 because it provides a rich characteriza- could be made more effective with some minimum tion of the effect of the explanatory variables on the condi- coordination of rural investments and programs-- tional distribution of the dependent variable (such as the such as education, road construction, and the estab- distribution of earnings). When there is sizable unobserved lishment of microcredit schemes--so that they heterogeneity in the data, mean linear regression models pro- benefit the more-isolated and poorest families. vide only a limited characterization of this distribution and of · Policy interventions that generate synergies and the role of explanatory factors. Quantile incomes regression break the mutually reinforcing mechanisms that lead analysis is useful given the income inequality in Latin Amer- to poverty traps could ignite a virtuous cycle between ica and the Caribbean, as well as the limitations of existing growth and broad poverty reduction. National devel- surveys in collecting all relevant earnings determinants. opment policies need to maintain a long-term per- For example, we can estimate regression lines for various spective to give the investments needed to break low percentiles of the adjusted (conditional) wage distribution, incomes and poverty persistence (for example, in that is, the distribution of earnings that results if all workers human capital formation) time to mature and trans- have the same observable characteristics. For instance, late into significant improvements in family incomes. median regression (the 50th quantile) splits the sample in half (half of the residuals above and half below the regression Annex 8A line) and gives the same results as Ordinary Least Squares (OLS, mean regression) when the wage distribution is sym- Data and methodological details metric. This allows unobserved wage determinants to inter- act with measures of observed skills. This interaction is Data captured by regression coefficients that vary across per- Most of the new analysis for this chapter relies on the rural centiles of the adjusted wage distribution. This way we can panel survey conducted by the Fundación Salvadoreña recover different impacts of the explanatory variables para el Desarrollo Económico y Social (FUSADES) in El throughout the entire distribution without imposing any Salvador and the Rural Finance Program at Ohio State prior assumptions such as normality or constant variance of University, in Columbus Ohio. The survey investigates regression errors. Results are also robust to outliers in wage demographic, occupational, and physical assets (such as data. infrastructure, land, and housing) among other character- Suppose that X is a dummy variable for gender istics that affect the income dynamics of rural households (women = 1). The quantile regression coefficient measures and their strategies for coping with risk. The panel data set the gender wage gap between a woman and a man with is composed of four biennial observations for the years similar education and experience at the same conditional 1995, 1997, 1999, and 2001. The main sample used in our quantile of the wage distribution. For example, the coeffi- analysis is 449 households that were observed in all four cient in the 90th percentile yields the wage disadvantage years. The attrition rate (individuals dropping from the faced by women in the top 10 percent of best-paid jobs for panel) is about 30 percent and largely occurred from the any given level of observed skills while the 10th per- first to the second wave when it was decided the survey centile coefficient yields the gap for women in the bottom would be continued as a panel. The evidence from previ- 10 percent of jobs on the earnings scale. Now suppose ous studies indicates that attrition does not appear to that X consists of years of formal education. OLS provides have a significant effect on either the sample composition a single estimate of the returns to education, the average or the validity of statistical inference from this sample for the whole population. Individual returns to education, (see Rodriguez-Mesa and Gonzalez-Vega 2004 for more however, may depend on some unobservable factors, like details). quality of education, unmeasured skills, or labor market 160 M I C R O D E T E R M I N A N T S O F I N C O M E S : L A B O R M A R K E T S , P O V E RT Y, A N D T R A P S ? Explaining changes in income distribution FIGURE 8A.1 Microeconometric simulations of counterfactual distribu- Differences in returns to education tions are helpful to characterize past distributional changes w and to simulate the distributional impact of changes in QRu economic factors and public policies. The idea is to simu- late the distribution of labor income at time t as a function x x of individual observable characteristics affecting wages and x x OLS x employment, the parameters that determine the effects of x x x these characteristics on market hourly wages and employ- x x x x x QRb ment outcomes (participation and hours of work), and x x x x x x x x x unobservable characteristics. A counterfactual distribution x x in time t1 is generated by taking some of its determinants x x x (parameters or distribution of characteristics) as if they x were those of time t2 and then comparing this counterfac- E1 E2 E tual distribution to the actual distribution observed in t1. Source: Authors` calculations. The difference between the two distributions can be attrib- Note: OLS ordinary least squares; and QR quantile regression. uted to the change in the selected determinants between t1 and t2. This method isolates the contribution of changes in observed household characteristics (endowments), the FIGURE 8A.2 returns to those characteristics, and unobserved hetero- Changes in returns over time geneity in the returns. w Four studies--Gasparini and others (2004) for Bolivia; Sosa-Escudero and Lucchetti (2004) for Peru; Sosa- QRut2 Escudero and Cicowiez (2005) for the Dominican Repub- x QRut1 lic; and Bustelo (2005) for Argentina--use these methods x x x x OLSt1 OLSst2 to estimate regressions for a reduced form of a labor supply x x x model with two equations, one for the number of hours of x x x x work and one for wages. The explanatory variables include x QRbt1 x x the typical measures of workers' human capital (education x x x x and experience, proxied by age and its square), demo- x x x x graphic characteristics such as gender and ethnicity, job x x QRbt12 x x x x characteristics (sector of activity and labor-informality indicators), and geographical location. The earnings equa- E tions are estimated separately for household heads and non- Source: Authors` calculations. heads, both in rural (except in Argentina) and urban areas. Note: OLS ordinary least squares; and QR quantile regression. The decompositions are carried out for one or two periods in the 1990s and early 2000s using national household sur- vey data. connections, and hence may differ across workers (fig- The decomposition analysis is enriched with estimates of ure A8.1). In fact, recent studies for several countries quantile earnings equations that are used to generate coun- suggest that returns are higher for workers at the top of terfactual distributions when the whole family of returns to the distribution. Moreover, it is possible for the returns to education (varying across quantiles) changes or for changes education to increase for workers in the upper quantiles of in each of the return quantile coefficients. This procedure, the wage distribution and decline for those in the used throughout the report, may provide a richer character- bottom quantiles, leaving the average return unchanged ization of past and predicted changes in the income distrib- (figure 8A.2). Quantile regressions allow an assessment of ution generated by economic and social changes or policy these important potential differences. interventions. Particularly, when investigating changes in 161 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S educational structure, we can simulate the new individual "state dependence" of the shocks. Consistent estimation of wage from upgrading education according to the wage-edu- all the parameters is done relying on the method of cation profile of the particular percentile to which the indi- moments as in Baltagi (2001, 82­83). vidual belongs. See each study for details. The empirical strategy consists of the following: Quantile regression for panel data · Implement the Bera, Sosa-Escudero, and Yoon (2001) Tannuri-Pianto, Pianto, and Arias (2005) estimate recent robust test for the presence of unobserved hetero- extensions of quantile regressions to longitudinal data geneity, state dependence, or both, based on a "null" allowing individual specific effects. The analogue in least model of no persistency (plain pooled OLS). squares regression is a fixed effects model estimated for a · Estimate the dynamic model using instrumental vari- balanced panel of households. Koenker (2004) considers ables to obtain some relevant parameters and corrobo- the following model for the conditional quantile functions rate the validity of the Lillard-Willis specification. of the response of the jth observation on the ith individ- · Implement the Lillard-Willis approach: estimate a ual yij base model to measure overall persistency (only yi , t-1 (8.A1) Qyij ( | xij) = ij + xij () j = 1, . . . , mi, as regressor); control for xi,t and xi,t ; and control for -1 i = 1, . . . , n. ui and then for the presence of xi,t and xi,t -1 under serially correlated errors. Four model specifications In this formulation the s have a pure location shift effect are considered: model 1 is only time dummies; model on the conditional quantiles of the response. The effects of 2 adds basic educational and demographic character- the covariates, xij are permitted to depend upon the quan- istics and geographic controls; model 3 adds credit, tile, , of interest, but the s do not. With least squares market access, and other economic characteristics; methods, one can transform y and X to deviations from and model 4 adds interactions between the latter individual means, and then compute ^ from the trans- characteristics. See Sosa-Escudero, Marchionni, and formed data. This decomposition of projections is not avail- Arias (2005) for more details. able for quantile regression, and we are required to deal directly with the full problem and the computational com- Notes plexities associated with it. For this we use the algorithm 1. For studies based on an asset-based approach to poverty persis- proposed by Koenker (2004) and rely on the bootstrap (300 tence, see Carter and Barrett (2005), and Attanasio and Székely replications) to obtain standard errors for the regression (2002) for Latin America. quantile coefficient estimates. 2. See De Ferranti and others (2004) and World Bank country poverty assessments available at www.worldbank.org\lac\poverty. 3. For recent studies for Africa, see Barrett Carter, and Little Analysis of income persistency (forthcoming). In their study of income persistency in rural El Salvador, 4. The R2 of earnings regressions controlling for all of these char- Sosa-Escudero, Marchionni, and Arias (2005) applied the acteristics are typically 0.4 to 0.6 (in Brazil). linear panel model with first-order serial correlation of the 5. For far more comprehensive surveys of earnings studies in the classic work of Lillard and Willis (1978). This is a linear region, see de Ferranti and others (2003, 2004) and IDB (2004). dynamic model for household income with first-order auto- 6. See IDB (2004), de Ferrranti and others (2003), and recent World Bank poverty assessments for Bolivia, Brazil, Dominican Repub- correlation: lic, Ecuador, and Peru, for example, for country-specific studies of the (8A.2) yit = xit + µi + it. importance of productivity for escaping poverty and low earnings. (8A.3) it = i,t 7. Females tend to have more intermittent labor force participa- -1 + it, || < 1, tion (rates in the region average 48 percent compared with 52 per- where µi ~ iid (0, 2µ), it ~ iid (0, 2), independent of each cent in East Asia and 70 percent in the United States). Women's other and of xit. In this specification the potential sources of actual labor market experience is lower than men's for a number of persistence are xit, µi and the presence of serial correlation reasons, particularly child bearing. Married women often participate in informal sector jobs that grant more time flexibility, so their lower in the observation-specific error process. µi represents pay may partly reflect a flexibility premium. See Kim and Polachek individual-specific "unobserved heterogeneity," and the (1994), Cox Edwards, Duryea, and Ureta (2001), and Cunningham serially correlated structure in the error term represents (2001). 162 M I C R O D E T E R M I N A N T S O F I N C O M E S : L A B O R M A R K E T S , P O V E RT Y, A N D T R A P S ? 8. See, for example, the studies in Hall and Patrinos (2005) and and Gonzalez-Vega (2004); Lanjouw (2001); and other references Arias, Yamada, and Tejerina (2004). therein. 9. Differences in schooling and other characteristics account for 15. The latter are robust to omitted variable biases since the effects over 70 percent of ethnic earnings gaps in Bolivia; Guatemala, and are identified from the within-period covariation between socioeco- Ecuador and about 50 percent in Peru. nomic variables (such as workers who switch sectors or changes in dis- 10. For ethnicity and race, see Arias, Yamada, and Tejerina (2004) tance to roads) and incomes or wages. However, the FE results for for Brazil; Gasparini and others (2004) for Bolivia; and Sosa-Escudero variables with little time variability such as education (a small frac- and Lucchetti (2004) for Peru. For gender in Chile, see Montenegro tion of workers remain in school) may be biased downward (because of (2001). higher signal-to-noise ratios). In this case RE are preferred, since they 11. Sector earnings differentials average 10 to 15 percent in the reflect both cross-section and within-period variation. region (after falling with economic restructuring), not unlike those in 16. The sectoral classification of individuals and households-- the United States; some differentials reach more than 40 percent in traditional and nontraditional agriculture and nonfarm--is based on some sectors and countries, however (IDB 2004). primary occupation and the number of hours spent in each sector. See 12. See Tannuri-Pianto, Pianto, and Arias (2004a) for Bolivia; Tannuri-Pianto, Pianto, and Arias (2005) for details. Carneiro and Henley (2002) for Brazil; World Bank (2005b) for the 17. The diversification index is created by counting each different Dominican Republic; and Bustelo (2005) for Argentina (although source of income weighted by its contribution to total household Bustelo does not correct for self-selection into the informal and for- income; it captures the ability of households to diversify the eco- mal sectors). nomic activities (such as crops cultivated, variety of microenterprises) 13. The studies are Gasparini and others (2004) for Bolivia, Sosa- in which their members engage. Escudero and Lucchetti (2004) for Peru, Sosa-Escudero and Cicowiez 18. This approach relies on recent developments in quantile (2005) for the Dominican Republic, and extensions of the analyses by regression for longitudinal data (Koenker 2004). See the annex. Bustelo (2005) for Argentina. See Bourguignon, Ferreira, and Lustig 19. Beneke de Sanfeliu and Shi (2004) report that about 80­85 per- (2005) for similar microsimulation studies. cent of households moved at least one decile upward or downward 14. We rely on Tannuri-Pianto and others (2005); the back- and 30­45 percent moved two deciles or more from period to period. ground paper for this report by Sosa-Escudero, Marchionni, and 20. Using a similar methodology, Freije and Souza (2002) report Arias (2005); Beneke de Sanfeliu and Shi (2004); Rodriguez-Mesa similar results for Venezuela. 163 CHAPTER 9 Breaking the Cycle of Underinvestment in Human Capital in Latin America Human capital is essential for enhancing the productivity of the Latin American poor and accelerating growth and poverty reduction. Why are the Latin American poor not accumulating enough human capital? What main policies can ensure they get the minimum level of skills required to break the cycle of poverty and low human capital? This chapter finds that an educational divide keeps the poorly educated in persistent poverty. That divide is caused by a combination of liquidity con- straints and lumpy and uneven returns to schooling. H UMAN CAPITAL, IN ITS BROADEST of the region. In particular, it aims to improve the under- sense, encompasses the levels of educa- standing of the main barriers to and opportunities for tion, health, and nutrition of the popu- significantly boosting the pace of educational progress and lation. Despite some uncertainty sur- poverty reduction in Latin America and the Caribbean. rounding the results from cross-country The chapter begins with a well-known fact: families empirical studies, human capital (proxied by education or with less than secondary schooling tend to be poor, and health levels) is generally considered one of the key they tend not to invest enough in education for their chil- determinants of growth. In a previous report in this series, dren to escape poverty. Several questions then become for example, de Ferranti and others (2003) described how central: Is this situation perpetuating across generations? educational investments are crucial for increased productiv- Can market forces be expected to break down this ity, rapid technological adaptation, and innovation, all poverty­low-education cycle, say, with sustained economic essential for sustained growth. Chapter 8 illustrated how growth? Or are there self-reinforcing mechanisms that tend sufficient levels of education are critical if poor Latin to reproduce the cycle? If so, what are they, and what sorts American families are to benefit fully from growth oppor- of public policy interventions are needed to address them? tunities and to reduce earnings inequality in the longer term. Chapter 7 pointed to cross-country empirical evi- · The chapter shows that Latin America is divided dence showing that poverty may affect education levels, between individuals who are highly educated and thus opening the possibility of a two-way causality in this those who have little education, and this divide is relationship. simultaneously a source and a result of subsistence This chapter investigates the mechanisms that could incomes across generations. Since parental education support this double causality and their bearing on the dis- and income are strongly correlated with children's appointing level of skills upgrading and persistent poverty educational attainment, the educational divide is also This chapter is based on background analyses for this report by O. Arias, A. M. Diaz, and V. Fazio. 165 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S self-reinforcing across generations. The dominant educational policies to encompass integrated strategies for mechanism in most countries is a function of a vic developing long-term skills that correct deficiencies in ious investment dynamic: returns to schooling are early-childhood development of poor children, strengthen low when it is cheaper to invest and become attrac- grade transitions and degree completion, upgrade educa- tive when the costs of schooling are hard to afford. tion quality for the poor, and improve the operation of labor markets. We corroborate these findings in ten countries, showing that: The educational transition in the region: Slow and unbalanced progress · Returns to schooling are essentially flat when stu- As a starting point, we illustrate two relevant findings of dents are in primary and secondary school and the 2003 flagship report on education and technology (de increase only with and after completion of secondary Ferranti and others 2003). First, skills upgrading through education. This pattern is consistent with a skill bias formal education, the so-called educational transition, has in labor demand from technological change in the been much slower in Latin America and the Caribbean than region (de Ferranti and others 2003). in East Asia, although both regions started with similar · Opportunity costs (forgone family income from chil- educational attainment in 1960 (figure 9.1). Second, the dren's potential earnings) and direct costs are larger transition in most Latin American countries has followed a for poor families with children in their final high school years and at the tertiary level, thus making liquidity constraints more binding. FIGURE 9.1 · In some cases the full return to educational invest- Latin America is in a slow educational transition ments materializes only around completion of Transitions in education attainment, 1960­2000 secondary or tertiary education. · In most countries, poor families face below-average Argentina returns to tertiary (and sometimes secondary) educa- Chile tion, perhaps because of disadvantages in family fac- tors needed for skills development at home (such as Peru family background or attitudes toward schooling) Colombia and lack of access to quality schools or high-pay jobs. Mexico These findings suggest that the value options of a sec- Brazil ondary or university diploma alone cannot be expected to Dominican Rep. break Latin America's educational divide. Poor families Bolivia have to juggle current subsistence needs against invest- El Salvador ments in schooling that carry a remote and uncertain pay- off. The end result: they invest in climbing the educational Nicaragua ladder while it is cheap, but stop when it becomes more Korea costly and when the full return to the investment cannot be Taiwan (China) realized because of the children's poor academic perfor- mance or the inability to buy higher-quality education. Of Singapore course, families are guided by other strong nonmonetary Hong Kong (China) considerations when investing in their children's educa- 20 10 0 10 20 30 40 tion. But the harsh economic reality of poverty too often Change in % of adult population becomes preponderant. Comprehensive policies are needed to break the vicious Secondary education Tertiary education cycle of poverty and low educational attainment in the Source: Based on de Ferranti et al. (2003). region. These policies must move beyond typical narrow 166 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A FIGURE 9.2 Most Latin American countries show deficits in secondary and tertiary enrollments Average educational attainment and educational deficits in LAC Deficit, % Years of education 30 12 20 10 10 8 0 10 6 20 4 30 2 40 50 0 re a) Rep. Haiti Rica ragua Brazil Peru Chile tigers Korea Salvador Colombiacan Jamaica Bolivia EcuadorMexico UruguayTobago Panama Malaysia (Chin countries Guatemala Honduras Nica Paraguay El Venezuela Costa & SingapoArgentina Asian de Kong Domini rces B. East R. Trinidad Hong resou Natural Secondary net enrollment deficit Tertiary gross enrollment deficit Years of schooling Source: Based on de Ferranti et al. (2003). Note: Deficit is defined by the gap between a country's actual educational attainment and what is expected from its per capita income. Data are circa 2000. pyramid distribution, with smaller numbers of people with Poverty and human capital: A two-way secondary education than with primary education. In con- relationship trast, East Asia moved to a distribution with higher num- Poverty can be related to the accumulation of human capi- bers of secondary-educated workers than of those with tal as both cause and effect. That higher educational attain- primary or tertiary education. Some Latin American coun- ment during youth leads to higher incomes later in life is tries, such as the Dominican Republic and El Salvador, probably the most documented finding in empirical micro- even funded tertiary schools at the expense of secondary economics.1 At the same time, poverty leads to lower schools and so developed an even larger "missing middle" human capital formation through various mechanisms dis- of secondary education. As a result, most of the region has cussed below. Figures 9.3 and 9.4 illustrate the two-way significant deficits in secondary and tertiary schooling relationship between poverty and schooling for our sample (figure 9.2) and a lower accumulation of average years of of Latin American and Caribbean countries, ranked by their education, a first-pass measure of skills. overall educational development (see annex 9A). The 2003 flagship report and the recent regional Figure 9.3 shows that in all countries the fraction of companion to the World Development Report: Making Services poor individuals falls systematically as the education level Work for the Poor (World Bank 2004d) analyzed institu- of the head of family rises.2 In fact, a typical family head tional factors affecting educational markets and the provi- requires at least a high school diploma to make a significant sion of education in the region. In this chapter we focus on dent in poverty. Poverty rates are 25 to 40 percentage the specific links between education and poverty and its points lower among families headed by high school gradu- intergenerational transmission. ates compared with those whose head has not completed 167 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 9.3 Poverty is higher in families in which the parents have little education Poverty headcount ratio by level of education of household head % of poor individuals 80 70 60 50 40 30 20 10 0 Argentina Chile Peru Colombia Mexico Brazil Dominican Bolivia El Salvador Nicaragua Rep. Less than primary Primary graduates Some secondary High school graduates Some tertiary Tertiary graduates Source: Based on data from SEDLAC. primary education. Only a college education secures an FIGURE 9.4 income level that makes ends meet: in almost all countries Children and youth in poor families have low educational attainment less than 10 percent of individuals are in poverty when the family is headed by a college graduate. Income poverty Gap in educational level between the nonpoor and poor, age 6­25 regressions in numerous World Bank country poverty assessments corroborate that households with main earners Argentina (heads and spouses) who have secondary education and Chile above are typically two to three times less likely to be poor.3 Peru Figure 9.4 illustrates the reversed stream of the cycle: Colombia poor families invest much less in human capital of their offspring. A much lower proportion of Latin American Mexico children and youth from poor families reach secondary Brazil and tertiary education than do children of richer fami- Dominican Rep. lies. The fraction with only primary education is 20 to 30 points higher among the poor; the college education Bolivia gap reaches 20 percentage points or more among coun- El Salvador tries like Argentina, Bolivia, Chile, Colombia, and Peru. The achievement gap between the poor and nonpoor is Nicaragua much smaller at the secondary level, although it still 40 30 20 10 0 10 20 30 40 ranges from 15 to 20 percentage points in Brazil, Difference in percentage of individuals Mexico, and Nicaragua. The relatively more egalitarian Primary Secondary Tertiary distribution of high school students reflects the already noted failure to expand secondary education massively in Source: Based on data from SEDLAC. the region. 168 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A Thus, the acceleration of educational development in education markets. Among chief supply factors, low acces- the region requires filling in the missing middle of the sibility of schools offering required grades and deficiencies educational pyramid through a more egalitarian skills in the educational system can limit the school progression upgrading. History indicates that under current national of children and youth. On the demand side, family income progression rates, it may take two to four decades to erase or wealth, parental education, the number of offspring, and the schooling gaps between the poor and nonpoor in these unequal access to higher-paying jobs can affect access to countries.4 Several self-reinforcing mechanisms could higher-quality schools, attitudes and family time devoted to prevent this catch-up from happening and lead to persis- schooling, and ultimately child scholastic performance and tent underinvestment or a slowing down in human capi- the returns to schooling. The poverty-traps literature points tal formation and to poverty traps. These are discussed to several self-reinforcing mechanisms that can lead to slug- below. gish school transitions coupled with persistent poverty in entire economies or certain population groups (Azariadis Human capital formation: Sources and Stachurski 2005; Bowles, Durlauf, and Hoff 2004; and of underinvestment traps Mayer-Foulkes 2004). These mechanisms and their empiri- Human capital formation is a synergistic process that starts cal implications are described below. very early in life. A large body of literature documents the importance of adequate health and nutrition for developing cognitive capacity, readiness to learn at school, and greater Credit constraints and increasing, lumpy returns: productivity in adult life.5 With the acquisition of formal Too poor to afford schooling schooling and training from childhood to adulthood, The inability to afford education is the most recognized these early investments crystallize in the development of inhibitor of human capital formation. Credit restrictions marketable skills (Heckman 1997, 2000). The number and indivisibilities in human capital investments can lead of years of education are therefore only a first-pass measure to self-sustaining underinvestment and poverty traps even of the skills embodied in individuals. The productivity if the returns to education are high (Galor and Zeira 1993; content of an individual's educational level depends on the Ljungqvist 1993). This can happen especially when fami- quality of family and school formation during infancy, lies must invest in their children's schooling for a span of childhood, and adolescent years. many years before education becomes a profitable endeavor. The determinants of human capital investments are cap- Educational investments are the prime example where tured in the well-known Becker (1967, 1975) model of adverse selection, moral hazard, and the lack of acceptable human capital and household behavior. Parents make collateral can lead to suboptimal investment by the poor. schooling decisions for their children to maximize the Several studies show that the main cost factor making school- welfare of all household members by allocating family ing investments unattractive to very poor families is the resources (including time in the home) among consump- opportunity cost of the children and young people who can tion, work, schooling, and leisure. Education is an invest- work at home or receive pay in the labor market (Basu 1999; ment with associated costs made in exchange for future Strauss and Thomas 1995). This situation is aggravated in benefits, that is, on the basis of net expected returns. The families with many small children (Behrman, Pollak, and costs include direct outlays such as school fees and other Taubman 1989; Haveman and Wolfe 1995) and in rural or related expenditures and the indirect opportunity cost of periurban areas with remote public schools and a deficient time (including forgone earnings from work), as well as any basic infrastructure. Direct costs, such as school fees, become nonmonetary costs related to aptitude and readiness to relatively more binding on poor families at the postsecondary learn. Private benefits from higher levels of education are level. Liquidity constraints and the inability to borrow generally future higher earnings in the labor market but against future higher earnings lead to underinvestment. also include increased capabilities to function in a modern Moreover, many poor families may underinvest in society. schooling because the full benefits of the investment are too The costs and benefits of schooling are influenced by remote. The probability of getting to the tertiary level is supply and demand factors related to household characteris- lower for children of poor families, so they may face both a tics, public investments, and the functioning of labor and lower expected return and more uncertainty in realizing 169 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S income gains from schooling. This can happen when the its quality content), while labor market abilities refer to returns to education increase markedly with the level of the skills needed to learn and adapt to different tasks and education, as has been widely documented in Latin America problem-solving environments. The lay terms for these (de Ferranti and others 2003; IDB 2004; Bourguignon, abilities are "book smarts" and "street smarts." In the labor Ferreira, and Lustig 2005). There also may be a diploma, or market these abilities result in higher returns to whatever "sheepskin" effects, whereby much of the schooling earn- level of education an individual acquires.7 ings premium accrues to those who have completed a high Children born into disadvantaged families are at higher school or hold a university degree.6 In this case the option risk of experiencing malnutrition, illnesses, and home value of completing secondary school and going to the uni- environments less conducive to learning, and they tend to versity is the main incentive to attend school in the first receive a lower quality of schooling. They therefore tend to place. For a poor family the rate of return to education may develop less motivation and readiness to learn, as well as compensate for the cost of delaying present consumption to have lower levels of the noncognitive skills complemen- (their discount rate) only when children can complete a tary to education. It is difficult to remedy fully the impact minimum level of education (such as primary or secondary that these deficiencies in a child's early years can have school). Hence, poor children are more likely to drop out of on the development of skills during youth and adulthood school once or before they reach education levels where liq- through formal schooling or training.8 Poor children there- uidity constraints become more binding, as is the case in fore can face important long-term learning constraints even the transition from secondary to university education. We in the absence of short-term liquidity constraints to attend- next discuss some mechanisms that may lower the returns ing school. These deficiencies can lead to more grade repe- to schooling for the poor. tition, delayed progression, lower expected returns to schooling as adults, and ultimately little transition to higher education grades. Intergenerational and agglomeration effects: Social exclusion caused by overt discrimination or biases Too poor to benefit from more schooling in public investment allocations can prevent poor families Multiple failures in the skills development process can from taking advantage of human capital production exter- inhibit the development of the scholastic and labor market nalities (such as spatial or labor market spillovers). Resi- abilities of poor children and youth and thus lower both dential segregation can lead to dismal funding for schools their educational attainment and returns to schooling. in poor communities and to negative sociological factors Human capital formation is a long-term process subject to such as the absence of role models and externalities for important intergenerational and agglomeration externali- learning ("peer group" effects), trapping children of poor ties. Families and community environments have a key role families in low levels of education.9 Lack of labor market to play in the early development of cognitive and noncog- connections or discrimination may hinder their access to nitive skills critical to the schooling process. Failures in the higher-paying jobs available for their level of schooling. developing these skills either at home or in the first grades Although discriminatory practices can hurt the efficiency of school accumulate and hinder a child's readiness to learn. of profit-maximizing firms, there is evidence that the The quality of schools is, of course, central to developing effects of exclusion on human capital formation and socioe- basic cognitive and problem-solving skills that complement conomic status can persist for generations, impervious to education and readily translate into higher productivity in competitive market pressures (Borjas 1992; Heckman the labor market. These multiple skills crystallize in 1997). an individual's "scholastic ability" (readiness to learn at There are also externalities in human capital formation school) and "labor market ability" (capacity for on-the-job related to interdependencies between private investments acquisition of skills). in skill and broader capital formation, particularly skills While scholastic and labor market abilities are corre- agglomeration and technological innovation. Countries or lated, they can lead to different schooling and labor market regions lacking a minimum skill level (typically workers outcomes. Scholastic abilities are reflected in academic with some secondary schooling) are less likely to attract scores and lead to higher educational attainment (including more technology and domestic or foreign investments in 170 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A technology and areas that require research and development levels. The evidence also shows persistent delayed (R&D) skills.10 Lack of technology investments holds back transitions to higher grades, closely related to family the growth in the demand for skills and thus the ability to income and exclusion. maintain attractive private returns to higher levels of edu- · Increasing and heterogeneous returns to education. cation under a massive educational expansion. The ensuing Particularly notable are returns that become signifi- slowing down in the transitions to higher educational cantly more attractive at higher levels of education; grades in turn continues to hinder technology upgrading show significant spikes for graduation grades (sheep- and reinforces the low-skill, low-innovation cycle. skin effects); and are lower for individuals from The upshot of all the mechanisms described here is to poor, lower-ability, and disadvantaged families and alter the poor's expectations of the likely returns to long- regions. term schooling investments. Even if average returns to · Strong intergenerational effects in human capital education are high, at any education level, there may be formation, chiefly, strong effects of liquidity con- considerable variation in returns to schooling for new straints (such as low family income and high family entrants to the labor market. While the evidence points to size) and long-term family-limiting factors (such as a pro-cyclical relationship between macroeconomic crises low parental education and family effects on educa- and educational enrollment in the region (since the lower- tion returns) on the educational progression of ing of opportunity costs dominates liquidity constraints), children and youth. less is known about the impact of the region's ever-present volatility on long-term investments in secondary and In examining these hypotheses, we rely on recent living college attendance.11 This and other sources of uncertainty conditions household surveys to estimate for each country a in returns can trap the poor in suboptimal education full set of Mincerian returns to education. These measure levels despite decisive public efforts to expand their variation across education levels and workers' observed access to schooling by removing infrastructure and credit and unobserved characteristics (see annex 9A). They also constraints. track microdeterminants of grade progression for individu- als in the 6­25 age range, with a focus on the effect of fam- Identifying human capital underinvestment traps: ily factors on grade-to-grade transition probabilities while In search of the smoking guns accounting for the sequential nature of schooling invest- How can we examine the empirical relevance of these mech- ment decisions (see annex 9A). anisms for explaining the slow educational transitions of Evidence supporting a combination of these elements many Latin American countries? The data requirements for would make a stronger case for the existence of human cap- conducting proper empirical tests of the relevant hypotheses ital underinvestment traps. For example, underinvestment are prohibitive--namely, a long panel data set covering a traps are more likely at play when educational attainment representative sample of families, including clean indicators is low despite high returns to schooling (at all levels of edu- of nutrition, health, and cognitive and noncognitive abili- cation and for all workers) and when liquidity constraints ties of children and adults, along with standard socioeco- affect progression to higher education grades. Poverty traps nomic characteristics. In a recent detailed study for Mexico, may also arise when the low- and high-education divide Mayer-Foulkes (2004) relied on evidence from a specialized occurs at a level of education insufficient to make ends health household survey and income and expenditure cross- meet. For each country we take a hard look at the evidence section surveys to examine mechanisms generating human to draw conclusions about the quantitative importance of development traps. Building on his analysis, we uncover the the underlying mechanisms. supporting evidence for the following empirical regularities in the ten countries we are focusing upon: The educational ladder in Latin America: A persisting educational divide · A multipeaked education distribution (grade cluster- Educational transitions can be thought of as climbing a ing) that shows a persistent divide between those ladder, where at each step, or grade, individuals and their with low levels of education and those with high families decide whether to move up to the next step. If 171 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S educational attainment were determined solely by an indi- variety of observed educational progressions. These depict vidual's liking for schooling, the percentage of people at the percentage of individuals at each step of the educational each step would not vary significantly by income or other ladder.12 demographics. Figures 9.5, 9.6, and 9.7 show the distribu- Figure 9.5 starts with the national distributions for tions of the educational attainment of the working-age Bolivia, Chile, Mexico, and Nicaragua. These help visualize population (ages 15­65) across income groups, location, the overall clustering of individuals around specific and cohorts for four countries chosen to represent the grades (taller bars) and also offer grand summaries of the FIGURE 9.5 Educational attainment of working age population, by country Chile Mexico Percent Percent 30 30 25 25 20 20 15 15 10 10 5 5 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Years of education Years of education Bolivia Nicaragua Percent Percent 30 30 25 25 20 20 15 15 10 10 5 5 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Years of education Years of education Source: Authors' estimates based on household survey data. 172 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A skills-matching possibilities faced by firms. Two different roughly 20 to 30 percent of workers have no schooling. grade clusters stand out in Chile (those completing basic Barely 15 percent reach lower high school in Mexico, and education and high school graduates) and Mexico (those only about 10 percent finish a full course of secondary with complete primary and those reaching up to lower sec- school in Chile, Colombia, and Peru.13 Hence, poor rural ondary). A third, much smaller cluster in these countries is families unable to migrate to urban centers can hardly apparent for those completing tertiary schooling and count on education as a means of mobility to better jobs. beyond. The distribution of skills is more diluted in Except for the more educationally developed countries, Bolivia and Nicaragua (with peaks centered at very low the educational divide of the population seems to be sus- grades), a sign of failures in completing diploma-granting tained over time, with prolonged and unequal educational grades and of delayed grade transitions (overage students). transitions still the norm among younger individuals. The grade clustering in Argentina, Brazil, and Colombia Figure 9.8 illustrates the typology of education transitions mimics that in Mexico, with one of the peaks at secondary for three birth cohorts (ages 15­25, 26­40, and 41­65) that completion and with higher dispersion in Argentina and attended school during the last 60 years (each spanning Brazil. Peru closely resembles the Chilean grade distribu- roughly two decades) in Argentina, Colombia, El Salvador, tion, but with a higher density of university graduates; and Mexico.14 Despite steady progress in educational while the Dominican Republic and El Salvador mimic the attainment, clustering at grades below secondary comple- grade distribution in Bolivia and Nicaragua. It would be tion is still prominent in many countries. harder for firms in the latter four countries to match work- In the less educationally developed countries, progress ers to more technology-intensive investments. in educational attainment is not yet strongly visible in the The clustering of educational achievement crystallizes younger labor force, and attainment of higher grades in an educational divide of the population strongly related remains sparse. For example, 20 percent of the young to income class and area of residence. Figure 9.6 presents Salvadoran workforce still has no schooling whatsoever, the educational distributions for the poorest 30 percent and only slightly less than older cohorts there. Colombia has a the richest 30 percent in the representative cases of balanced transition with a single peak at secondary comple- Argentina, Brazil, El Salvador, and Mexico. The two educa- tion, while postsecondary education is still rare for the two tional grade groupings noted for Chile and the modest younger cohorts. That is, they show signs of moving clustering in tertiary education are strongly reinforced toward a diamond-shaped educational distribution. Chile across income classes in Argentina as well as in Mexico, and Peru show a similar pattern. The schooling ladder in except that completion of lower secondary education is not Mexico remains largely twin-peaked for the youngest an income-schooling divide for Mexicans. High school cohort, with clustering at lower secondary completion completion is the sharp dividing line between the poorest becoming more pronounced (30 percent of the youth). and richest in Brazil, while few of the very poor working- Argentina is the only case where the youth appear to be in age Salvadorans have finished primary education. The a balanced educational transition that breaks the postsec- income­school grade groupings in Chile, Colombia, and ondary education barrier and points to an inverted-pyramid- Peru are similar to Brazil's, although with varying degrees shaped education distribution. However, about 20 percent and more visible college graduate clusters. Nicaragua and, of prime-age Argentines and 30 percent of the older cohort to a lesser degree, Bolivia and the Dominican Republic hold only a basic education degree. mimic El Salvador's groupings. The richest Latin Ameri- The data for children and youth currently in school indi- cans do not stand out as university-goers. The best per- cate that these patterns of educational transitions are being formers are in Argentina, Colombia, and Mexico, where reinforced. Figure 9.9 presents net enrollment rates of indi- around one-third of individuals from the richest families viduals in the 6­18 age range for most countries in the obtain a university degree, compared with more than half region. The demand for schooling, signaled by almost uni- of all adults in the United States and Canada. versal enrollment rates, is strong up to age 13, which corre- The slicing of educational groupings for urban and rural sponds to the completion of primary education in most workforces is even more startling (figure 9.7). In Brazil, countries. Net enrollment rates begin falling fast beyond Bolivia, El Salvador, and Nicaragua, the bulk of the rural this age, with the exception of Argentina, Chile, and workforce has not gone beyond primary education, and Jamaica, where dropout rates accelerate only after the first 173 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 9.6 Educational attainment for the poorest 30 percent and the richest 30 percent in Argentina, Mexico, Brazil, and El Salvador Argentina Mexico Percent of population Percent of population 45 45 40 40 35 35 30 30 25 25 20 20 15 15 10 10 5 5 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Years of education Years of education Brazil El Salvador Percent of population Percent of population 45 45 40 40 35 35 30 30 25 25 20 20 15 15 10 10 5 5 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Years of education Years of education 30% poorest 30% richest Source: Authors' estimates based on household survey data. years of high school (15­18 age range). Further analysis of One common reason for the sharp decline in enrollment these data in numerous country studies shows that the drop is that Latin American children experience delayed transi- in enrollment rates is generally more marked among chil- tions mainly due to grade repetition. Figure 9.9 also shows dren and youth from poor families.15 The smooth decline in the dismal performance of the region in ensuring high rates enrollments during the secondary cycle in most countries of on-time progression to the next grade. This low on- suggests that lack of secondary school facilities is not the time progression to the next grade, combined with high main driving factor. enrollment, results in substantial numbers of children who 174 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A FIGURE 9.7 Educational attainment for urban and rural areas in Nicaragua, El Salvador, Brazil, and Bolivia Nicaragua El Salvador Percent of population Percent of population 35 35 30 30 25 25 20 20 15 15 10 10 5 5 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Years of education Years of education Brazil Bolivia Percent of population Percent of population 35 35 30 30 25 25 20 20 15 15 10 10 5 5 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Years of education Years of education Urban Rural Source: Authors' estimates based on household survey data. are overage for the grade they are in. For example, in many each country it compares a measure of average years spent Central American countries, 40 to 50 percent of children in school (the "1­12" educational system, 6­18 age range, are two or more years overage when they reach secondary proposed by Urquiola and Calderón 2004) with the actual education (World Bank 2005b). number of grades that children have completed, on aver- Table 9.1 illustrates the poor record of most countries age. The first column captures the expected number of in the region in turning children's and youth's contact years that a child will spend in school given the country's with the educational system into years of schooling. For current enrollment patterns. It provides a convenient 175 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 9.8 Educational attainment for three age groups in Argentina, Colombia, Mexico, and El Salvador Argentina Colombia Percent of population Percent of population 40 40 35 35 30 30 25 25 20 20 15 15 10 10 5 5 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Years of education Years of education Mexico El Salvador Percent of population Percent of population 40 40 35 35 30 30 25 25 20 20 15 15 10 10 5 5 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Years of education Years of education 15­25 26­40 41­65 Source: Authors' estimates based on household survey data. summary of the resources (in a time scale) spent by coun- Latin American children stay, on average, two to four tries to keep children in school.16 The gap with respect to extra years in school than needed to complete a full course of the actual grades completed (third column) indicates how secondary education. The countries with lower educational effectively educational systems turn average years in school attainment--Belize, Brazil, and Nicaragua, for example-- into average number of grades completed. tend to be among the worst performers on this indicator. 176 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A FIGURE 9.9 Low educational attainment is reinforced in current cohorts Age-specific net enrollment rates, circa 2000 Percent 100 90 80 70 60 50 40 30 20 Argentina Brazil Colombia Dominican Rep. El Salvador Honduras Nicaragua Paraguay R.B. de Venezuela Bolivia Chile Costa Rica 10 Ecuador Guatemala Mexico Panama Peru 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Grade in school Source: Based on Urquiola and Calderón (2004). Note: Data for Argentina include only urban areas. Students progressing through education system on time Percent 100 90 80 70 60 50 40 30 20 Argentina Chile Ecuador Mexico Peru Uruguay 10 Brazil Colombia Guatemala Panama Dominican Rep. 0 1 2 3 4 5 6 7 8 9 Grade in school Source: Based on Cabrol (2002). Note: Data for Argentina include only urban areas. However, countries like the Dominican Republic, Jamaica, delaying full entry into the labor market), and likely and Uruguay, which stand out in keeping children in increases the risk of eventually dropping out. school, are fairly inefficient in the production of years of To summarize, Latin America's success in improving schooling. This low on-time progression slows down the average educational levels, with close to universal primary accumulation of skills, lowers the returns to education (by enrollment, has not been sufficient to reverse the persisting 177 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S TABLE 9.1 Why aren't poor families leading their offspring to a level Average years of schooling in the "1­12" educational system and of education sufficient to better their chances of escaping excess years spent in school, 6­18 age range, circa 2000 this potential intergenerational poverty cycle? As noted before, liquidity constraints, deficient infrastructure, and Average number Average excess low returns to education may be to blame. These are in turn Average years of grades years spent Country spent in school completed in school linked to both short-term (income, for example) and long- term family factors. Chile 12.1 10.4 1.7 Argentinaa 12.1 9.8 2.3 Panama 11.5 9.5 2.0 Peru 11.1 9.0 2.1 Liquidity constraints, family factors, and Bolivia 11.2 8.9 2.3 educational investments: A sneak preview Jamaica 11.7 8.8 2.9 The reasons Latin American children and youth reveal for Ecuador 10.4 8.7 1.7 Mexico 10.6 8.7 1.9 being out of school consistently point to a combination of Uruguaya 11.4 8.7 2.7 high opportunity costs, perceived low benefits, and access R. B. de Venezuela 11.0 8.6 2.4 Colombia 10.5 8.4 2.1 constraints.17 Figure 9.10 illustrates how the relative Paraguay 10.7 8.4 2.3 emphases on each factor vary by age, gender, and poverty lev- Dominican 11.8 8.3 3.5 els in four selected countries. The following patterns emerge: Republic El Salvador 10.0 8.0 2.0 Costa Rica 10.5 7.8 2.7 · Work-related reasons (opportunity and direct costs) Brazil 11.4 7.3 4.1 Belize 10.6 6.6 4.0 tend to be the most pressing in all countries, espe- Honduras 8.6 6.2 2.4 cially among boys, youth of postsecondary school Haitia 8.8 5.9 2.9 age, and the poor. Nicaragua 9.7 5.9 3.8 Guatemala 8.2 5.5 2.7 · Low benefits are more important among the poor, boys, and children of primary and secondary school Source: Based on Urquiola and Calderón (2004). age, particularly in Bolivia and El Salvador. a. Data for urban areas only. · Other reasons, including pregnancy, family prob- lems, or other idiosyncrasies, are more prevalent among girls, at younger ages, and among the rich, educational divide in the population except in the more particularly in Chile and Colombia. educationally advanced countries. Only Chile, Colombia, · Limited physical access appears to be a less-pressing and Peru show signs of moving fast toward a diamond- factor overall, but is evident mostly among primary- shaped educational distribution. Argentina appears to be school-age children, particularly in Chile, Colombia, moving toward this pattern as well, although on a some- the Dominican Republic, and Nicaragua. what longer horizon, while Brazil shows delayed but steady progress. The population in the other countries sorts into Figure 9.11 shows that the relationship between educa- two groups, one of individuals with low schooling (typically tional investments and proxies of some of the above factors less than secondary education) and the other with more- are largely consistent with self-assessments. The cost of educated individuals (secondary and above). These patterns schooling appears to be pressing largely for youth of post- of educational attainment emerge strongly across income secondary school age. The top panel in the figure shows that and regional lines, with rural residents and the poorest fam- the opportunity cost, proxied by the contribution of youth's ilies predominantly trapped in the low-education group. earnings to total family incomes, of sending young children Patterns of school progression of current student cohorts to school is negligible in most countries. While the forgone indicate that this educational divide repeats itself as a result income increases for adolescents of secondary education age, of high repetition and dropout rates. Since completion of at it still represents less than 10 percent of family incomes. least a secondary education is needed for typical poor fami- The greater concern of poor families for present rather lies to have a real chance of escaping subsistence levels, this than future consumption (that is, a higher discount rate) is educational divide might be self-reinforcing and induce very likely to make liquidity constraints binding in the persistent poverty across family generations. transition from secondary to tertiary school. The income 178 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A FIGURE 9.10 Poor children and youth stay out of school because of high costs and low benefits Chile Peru Percent Percent 100 100 90 90 80 80 70 70 60 60 50 50 40 40 30 30 20 20 10 10 0 0 6­14 ­18 ­24 15 19 6­14 ­18 ­24 15 19 6­14 ­18 ­24 15 19 6­14 ­18 ­24 15 19 6­14 ­18 ­24 15 19 6­14 ­18 ­24 15 19 6­14 ­18 ­24 15 19 6­14 ­18 ­24 15 19 Women Men 40% poorest 20% richest Women Men 40% poorest 20% richest Colombia El Salvador Percent Percent 100 100 90 90 80 80 70 70 60 60 50 50 40 40 30 30 20 20 10 10 0 0 6­14 ­18 ­24 15 19 6­14 ­18 ­24 15 19 6­14 ­18 ­24 15 19 6­14 ­18 ­24 15 19 6­14 ­18 ­24 15 19 6­14 ­18 ­24 15 19 6­14 ­18 ­24 15 19 6­14 ­18 ­24 15 19 Women Men 40% poorest 20% richest Women Men 40% poorest 20% richest Limited access Low benefits Work related Insecurity Other Source: Authors' estimates based on household survey data. loss for poor families that invest in postsecondary schooling costs of higher-quality private secondary schools and univer- is more significant than for the relatively rich, ranging from sities and should be weighted against the promise of high 10 to 17 percent for very poor youth and from 14 to 22 per- returns to postsecondary schooling.18 Even for poor youth cent for the moderately poor (except in the Dominican with access to free public schools, the high dependence of Republic). The income loss is in addition to the high tuition their families on their earnings to make ends meet almost 179 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 9.11 Opportunity costs and schooling gaps get larger for secondary to post-secondary school-age children Opportunity cost of schooling in LAC by age groups Earnings of youth as % total family incomes 25 Lowest 20% Middle 20% Highest 20% 20 15 10 5 0 il il Chile Rep. Rep. Rep. Brazil Chile Braz Chile Braz Colombia Salvador Colombia Salvador Colombia Salvador El Nicaragua El Nicaragua El Nicaragua Dominican Dominican Dominican Age 12­14 Age 15­18 Age 19­25 Gap in average years of schooling between nonpoor and poor by age group Difference in years of education 4.5 Age group 4.0 6­12 13­18 3.5 19­25 3.0 2.5 2.0 1.5 1.0 0.5 0 ico ivia Chile Peru Rep. ombia Mex Brazil Bol Argentina Col Salvador El Nicaragua Dominican Source: Authors' estimates based on household survey data. certainly deters transitions to higher education grades. No for Brazil, Colombia, and the Dominican Republic, illus- wonder poor Latin American children start to fall signifi- trating the strong correlation between parental education cantly behind the nonpoor in average years of schooling in and educational attainment and how this is mediated by their teenage and young adult years (bottom panel). income levels, school access (proxied by area), and race. Finally, as noted in chapter 2, well-educated parents tend Parental education compensates for low incomes and lack of to have better-educated children. Figure 9.12 portrays this access in Colombia and the Dominican Republic. In Brazil, 180 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A FIGURE 9.12 Low education continues for generations, especially among the poor Colombia Colombia Children's average years of education Children's average years of education 20 20 Rural Urban 40% poorest 20% richest 16 16 12 12 8 8 4 4 0 0 0 to 6 6 to 11 more than 11 0 to 6 6 to 11 more than 11 Mother's years of education Mother's years of education Dominican Republic Brazil Children's average years of education Average years of education of adult sons 20 20 40% poorest 20% richest Whites Pretos Pardos 16 16 12 12 8 8 4 4 0 0 0 1 to 8 9 to 12 more than 12 0 to 4 5 to 8 9 to 11 more than 11 Mother's years of education Father's years of education Source: Authors' estimates based on household survey data. The graph for Brazil is taken from Arias, Yamada, and Tejerina (2004). pretos (blacks) are caught in an intergenerational low- and primary education remained sluggish or declined education trap. Differences in returns to schooling may be over the 1990s in most countries (de Ferranti and oth- behind this unequal educational mobility. We turn to these ers 2003, 2004; IDB 2004; Bourguignon, Ferreira, differences next. and Lustig 2005). · The trends in returns to schooling are largely attrib- The private value of schooling: How much uted, although not indisputably, to relative demand does it pay? To whom? shifts--caused by trade liberalization and parallel The numerous studies estimating returns to education in technical change--that favor more skilled workers Latin America and the Caribbean point to several stylized (Bourguignon, Ferreira, and Lustig 2005; de Ferranti facts: and others 2003, 2004; IDB 2004). · Overall, average returns are relatively high compared If the returns to tertiary education are high and increas- with other regions of the world, but there is signifi- ing, why do we not see many more Latin American chil- cant variation in returns across countries in the dren (including more of the rich) moving up to the top of region (Psacharopoulos and Patrinos 2004). the education ladder? A detailed analysis of returns to · Education contributes significantly to rising earnings schooling in our sample of ten countries suggests that the inequality: the average return to tertiary education pattern of returns may be an important part of the story rose, while returns to those completing secondary behind the persisting educational divide in several of these 181 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S countries. There are two main findings. First, returns to Figure 9.13 presents a snapshot of various measures of education are lumpy, and diplomas often matter a great the average returns to education in the ten countries. deal--in many cases education seems attractive only when These indicators answer distinct questions about the edu- the long-term investments needed to complete at least a cation investment process. The top left panel shows the full course of secondary and some tertiary education can be evolution of the average earnings premium for schooling realized. Second, in most countries the high average returns as individuals move up each step of the education ladder to tertiary education are not available to everyone alike; in from no schooling to university completion, while the top particular, poor families tend to accrue returns to their right panel simply presents the per year returns that result investments in higher levels of education that are signifi- from dividing this by the number of grades completed. cantly below the average market return. The two panels are informative of the cumulative increase FIGURE 9.13 Average rates of return for education increase at the tertiary level Earnings premium to each education grade, Rate of return to education per year compared to uneducated worker Log wage difference relative to no education Percent 3.0 26 Argentina Chile Argentina Chile 24 Peru Colombia Peru Colombia 2.5 Mexico Brazil 22 Mexico Brazil Dominican Rep. Bolivia Dominican Rep. Bolivia 20 El Salvador Nicaragua El Salvador Nicaragua 18 2.0 16 14 1.5 12 10 1.0 8 6 0.5 4 2 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Years of education Years of education Cumulative marginal rates of return within level Earnings premium to each education level Percent Log wage difference relative to no education 25 2.5 Argentina Chile Primary incomplete Peru Colombia Secondary incomplete Mexico Brazil Secondary complete 20 2.0 Dominican Rep. Bolivia Tertiary incomplete El Salvador Nicaragua Tertiary complete 15 1.5 10 1.0 5 0.5 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 ico il Chile Peru mbia Rep. Mex Braz can Bolivia Argentina Colo El SalvadorNicaragua Source: Authors' estimates based on household survey data. Domini 182 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A in average earnings of a successful school progression. Brazil, Mexico, and Peru, the annual average returns to In other words, for the family of a child just starting investment in the basic education cycle are below 10 per- school, it answers the question, "On average, how much cent. Argentina, Chile, the Dominican Republic, and El will she make if she reaches grade A (total and per year Salvador have notably low average returns to basic educa- completed)?" tion, ranging from 2 to 4 percent a year. Marginal returns The bottom left panel shows the cumulative change in are generally higher for those obtaining some or complet- the marginal annual returns within each education level, ing tertiary education (bottom panels), while the average computed as a moving average of the grade-to-grade differ- returns to completing a full course of secondary schooling ence in the return coefficients shown in the top right panel. are more meaningful in Brazil, Colombia, and Mexico These reveal the additional average earnings gains from and negligible in Bolivia, the Dominican Republic, and completing each subsequent grade of primary, secondary, El Salvador. In Argentina, Brazil, El Salvador, and and tertiary education and may be the relevant indicators Nicaragua, the full value of a college education accrues for the family in deciding whether or not their child should only after getting a diploma or completing a full four- to continue in school for an additional year given that she has five-year course at a university. Those planning to work reached grade A. Finally, the bottom right panel depicts and study to finance college have a harder time doing so the average earnings premium for each level of education in these countries. defined according to the educational system of each coun- A key conclusion is that, barring liquidity and access try. The vertical distance between the points gives the constraints, the value option of getting a secondary or uni- marginal mean returns to each education level. These versity diploma may be the strongest incentive for poor returns reflect the actual average value ascribed by local Latin American youth to break the educational divide. The labor markets to a degree and thus capture any labor low and flat returns to basic education in all countries and market signaling effect of degree completion.19 For the to high school education in the less-advanced countries child just starting school, the underlying marginal returns suggest that workers who do not finish these cycles, say, answer the question: "On average, how much more will she workers with four to eight or nine to twelve years of school- make if she reaches/completes education level X?" Cross- ing, are highly substitutable in the labor market. It is the country comparisons of the data in this panel should be completion of successive higher grades that makes earlier treated with caution due to variations in the structures of school investments more rewarding. Unless the prospects educational systems. of reaching higher education grades are good, poor youth As one moves up the education ladder, average returns have few incentives to continue beyond basic education. to schooling increase fairly similarly across countries, Yet do these average returns to education give a fair although the differences widen considerably at higher indication of the incentives to invest in education for every- grades. The average of annual returns in the ten countries one? There are two reasons why the answer might be no. studied is about 6 percent for completion of eight years of First, returns to education can vary across workers accord- basic education, 7.5 percent for secondary school graduates, ing to gender, race and ethnicity, residential location, and and 11 percent for university graduates.20 The lowest and other unobserved (unmeasured) characteristics such as highest returns in the sample are consistently observed in quality of education, family background factors, and indi- Brazil and El Salvador, ranging from 2.3 percent a year in vidual spunk.22 Second, to the extent that individuals and El Salvador to 9.8 percent in Brazil for an eight-year course families act on the expected returns to education in making of basic education, from 3.4 to 11.8 percent for a secondary their schooling decisions, estimates of average returns to degree, and from 8 to 16.9 percent for a five-year course of education may not accurately represent the actual return tertiary education. to those not currently in school. For example, the returns to Several telling patterns are noticeable. The marginal tertiary education could reflect the average quality (ability) returns to each subsequent grade stay constant or decline of those who already have a college education. We now for the first eight years of basic education, increase in the explore the empirical relevance of these issues (except for first years of secondary education, and soar with and gender, which is less correlated with poverty and access beyond completion of secondary education.21 Except for constraints to schooling). 183 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S Variations in returns to schooling: Rural and racial areas, but in Bolivia, Mexico, and Peru, they are much dimensions higher for rural workers over the whole range of levels of Earning incentives for rural workers are similar to--in education. Other countries, including Colombia, show no some countries even higher than--those for urban workers. gaps between urban and rural workers. These results reflect Figure 9.14 illustrates that there are few differences in the the growing importance of nonfarm occupations in rural returns to education in urban and rural labor markets, and economies. The majority of uneducated rural workers when the differences are more visible, they favor rural throughout Latin America are employed in agriculture, workers. where education is less productive, while the more skilled In Brazil, Chile, and Nicaragua, education returns, par- hold nonfarm jobs. Since incomes in rural areas start from a ticularly to secondary education, are mildly larger in urban lower base, workers in nonfarm jobs get a larger earnings FIGURE 9.14 The returns to education differ for urban and rural labor markets Chile, 2001 Bolivia, 2002 Returns, % Returns, % 3.0 3.5 2.5 3.0 2.5 2.0 2.0 1.5 1.5 1.0 1.0 0.5 0.5 0 0 0.5 0.5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Years of education Years of education Colombia, 2003 Mexico, 2003 Returns, % Returns, % 3.5 3.5 3.0 3.0 2.5 2.5 2.0 2.0 1.5 1.5 1.0 1.0 0.5 0 0.5 0.5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Years of education Years of education National Urban Rural Source: Authors' estimates based on household survey data. 184 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A kick from education than do comparable workers in urban FIGURE 9.15 areas. Hence, a lack of earnings dividends from schooling Differences in returns to education in Brazil largely reflect unequal should not be a first-order deterrent for rural families to human capital and a secondary effect of skin color invest in education except for those unable to engage in Advantage of white men in the returns to education; growing rural economic activities. adjusted by parental education, Brazil The influence of racial inequality on returns to educa- Percentage points per year tion is stronger, although labor market discrimination may 3.0 not be the main culprit. Several studies find that indige- 2.5 nous and Afro-descendant populations are restricted in access to the better-paying jobs. In Bolivia, Brazil, 2.0 Guatemala, and Peru, these populations have average 1.5 returns to schooling that are 1 to 3 percentage points lower than whites.23 That compounds a disadvantage in educa- 1.0 tional attainment that ranges from an average of two to 0.5 three full years of schooling. There is evidence that differences in other components 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 of human capital have a significant bearing on these results. Low-paying jobs Average-paying jobs Best-paid jobs Studies for Bolivia and Brazil (Mercado, Andersen, and Quantiles of the wage distribution Muriel 2003; Arias, Yamada, and Tejerina 2004) show that the lower education quality and parental education of non- Pretos (unadjusted) Pretos (adjusted) Pardos (unadjusted) Pardos (adjusted) whites can explain more of the gap in returns than labor market discrimination. Differences in the formal education Source: Based on Arias, Yamada, and Tejerina (2004). of parents in Brazil account for 1 percentage point of the Note: Data for urban areas only. edge in average returns of white men relative to pretos and 0.5 percentage point of the gap relative to pardos (mixed- race Brazilians) (figure 9.15). The fact that whites attend the sources of ethnic and racial earnings inequality in the school in states with relatively better-quality education region, these populations do face lower incentives to invest further accounts for half of their remaining lead in the in schooling that should be addressed by human capital returns to education. Overall, after factoring in racial dif- and labor market policy interventions. ferences in the quantity and quality of individual education and family background, the average earnings gap between Unobserved abilities and the returns white and nonwhite Brazilian workers falls from 46 percent to the marginal labor market entrant to a 16 percent earnings disadvantage unrelated to workers' A flurry of studies shows that returns to education can vary productive potential. among individuals with the same race, gender, labor mar- However, labor market inequality related to skin color ket experience, or sector of employment because of the imposes larger earnings penalties on blacks in the higher- complementarity between education and unobserved earn- paying jobs of any given skills. As shown in figure 9.15, ings determinants.24 The latter are related to the multiple while pretos and pardos located at the bottom of the salary skills that constitute an individual's scholastic and labor scale enjoy a similar payoff to education, the best-paid market abilities that may create more channels for acquir- quintile of pardos have a schooling return advantage of ing higher levels of education as well as the higher-paying about 1 percentage point over the best-paid quintile of jobs for any given level of education. Data on the quality of pretos with similar observed skills. This finding is consistent schools, family background, labor market connections, and with studies showing that labor market discrimination is characteristics of communities in early childhood can serve more likely when nonwhite workers cannot be denied as proxies for these abilities but are often absent in house- access to the higher-paying jobs within occupations on the hold surveys. Nonetheless, it is important to factor in these basis of their observed productive attributes (Darity and and other sources of variation in the costs and benefits of Mason 1998). While further research is needed to ascertain schooling across families and individuals. 185 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S Average returns to education can misrepresent the become more schooled. However, returns would be higher actual incentives faced by less-schooled individuals to move for more affluent families to the extent they have an edge in up the education ladder. Two distinct possibilities are rele- producing higher scholastic aptitude and labor market vant (Carneiro and Heckman 2003; and Card 2001). First, skills. While there are bright and industrious individuals there may be "cream skimming," in which the best-quality in both poor and rich families, the factors affecting a child's students (those with higher abilities or higher returns) are readiness to learn, quality of schooling, and labor market more likely to get a university (or secondary) education, connections tend to lower the returns to educational invest- while the less talented (low returns) are more prone to join ments of poor families. These disadvantaging effects should the pool of the less educated. In this case the returns for be compounded and thus be more visible at higher educa- individuals with low propensity to attend university (the tion levels. less talented) will be lower than the average return for To examine the importance of these issues, we estimate a the already college educated. Second, faced with binding series of returns to education and assess whether returns are liquidity constraints, many talented high school dropouts lower for poor families. We fitted earnings functions may be unable to attend university despite high expected through 10 different percentiles of the conditional wage dis- earnings gains. That is particularly true for those in the tribution in each country, that is, for workers located at the best-paid unskilled jobs who face higher forgone earnings if bottom to the top of the salary scale adjusted by their demo- they opt to continue their schooling. Thus, many marginal graphics and skill levels. Figure 9.16 illustrates the returns entrants to college may actually have returns above the to each level of education for workers in the 20th, 50th and average return to current college graduates. Which effects 80th wage percentiles in selected countries, which represent predominate depends on the strength of the correlation the schooling returns to the low- average- and best-paid between schooling costs and benefits along income lines workers at jobs of any skill level.26 Taking the position of and education levels. In either case the average returns to workers in the adjusted salary scale as a proxy of their unob- secondary or university education are insufficient to assess served ability, differences in returns along the salary scale the schooling investment incentives for youth randomly reflect variations in their unmeasured skills. selected from the population or for those from disadvan- In most countries returns to schooling, particularly at the taged families. tertiary level, are higher for workers who have the best-paid How does this issue bear on the question of underinvest- jobs for their skills. The differences are quite large in Chile, ment in human capital and poverty? The second case above El Salvador, and Nicaragua, where the top-rank (high abil- is a clear-cut example of schooling underinvestment caused ity) college workers enjoy returns to tertiary education that by credit constraints that may be addressed through condi- are 30 to 40 percent larger than the returns for the college- tional cash transfers or student loan programs. In the first educated in jobs with lower pay. Returns for basic and sec- case more evidence is needed on the role of long-term fam- ondary education are similar to the average return except in ily factors or other externalities in generating low returns Brazil and Chile, where returns to completion of secondary to education to assess the case for underinvestment. For education are 30 to 40 percent larger for the best-paid example, if returns are low in general because of a deficient workers. Only in the Dominican Republic and Peru are the school system or unsound economic policies, then from a returns roughly similar throughout the earnings scale. private perspective, families' schooling investments may be There is further evidence that the poor tend to benefit "just right" for existing returns.25 The most appropriate less from higher education. Figure 9.17 illustrates the policies to promote more education need not bear a direct results of following a procedure that maps the schooling link with poverty. returns of workers (implicitly reflecting rankings of unmea- Differences in empirical measures of schooling returns sured human capital) to the rankings of per capita incomes across income groups can be informative about whether of their families (see annex 9A). Returns to a university short-term liquidity constraints or long-term family effects education (complete or incomplete) tend to be higher for are more significant. Since very poor families should face the richest families in all countries where we observed more binding liquidity constraints, their measured returns significant differences. The gaps in returns between the rich to education could be higher because at the margin only and the poor are somewhat muted (20 percent in Chile and the more talented (with very high expected returns) 40 percent in Nicaragua, for example), reflecting the fact 186 FIGURE 9.16 Returns to each level of education for the three tiers of the earnings distribution Chile, 2002 Argentina, 2003 Returns, % Returns, % 2.0 2.0 1.8 1.8 1.6 1.6 1.4 1.4 1.2 1.2 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 Primary Secondary Secondary Tertiary Tertiary Primary Secondary Secondary Tertiary Tertiary complete incomplete complete incomplete complete complete incomplete complete incomplete complete Years of education Years of education Brazil, 2002 El Salvador, 2002 Returns, % Returns, % 2.0 2.0 1.8 1.8 1.6 1.6 1.4 1.4 1.2 1.2 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 Primary Secondary Secondary Tertiary Tertiary Primary Secondary Secondary Tertiary Tertiary complete incomplete complete incomplete complete complete incomplete complete incomplete complete Years of education Years of education Nicaragua, 2002 Colombia, 2003 Returns, % Returns, % 2.0 2.0 1.8 1.8 1.6 1.6 1.4 1.4 1.2 1.2 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 Primary Secondary Secondary Tertiary Tertiary Primary Secondary Secondary Tertiary Tertiary complete incomplete complete incomplete complete complete incomplete complete incomplete complete Years of education Years of education 20th earnings percentile 50th earnings percentile 80th earnings percentile Source: Authors' estimates based on household survey data. Note: The three tiers (20th earnings percentile, 50th earnings percentile, and 80th earnings percentile) are adjusted by experience, gender, and area of residence. 187 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 9.17 Correlation between returns to each level of education and poverty Chile Chile Primary complete Tertiary complete Centile average of return Centile average of return 0.22 1.35 1.30 0.20 1.25 0.18 1.20 0.16 1.15 5 15 25 35 45 55 65 75 85 95 5 15 25 35 45 55 65 75 85 95 100 quantiles of_n 100 quantiles of_n Unconditional per capita household income centiles Unconditional per capita household income centiles Bolivia Argentina Tertiary incomplete Secondary complete Centile average of return Centile average of return Centile average/prediction Centile average/prediction 1.04 0.62 1.02 0.61 1.00 0.60 0.98 0.59 0.96 0.94 0.58 5 15 25 35 45 55 65 75 85 95 5 15 25 35 45 55 65 75 85 95 100 quantiles of_n 100 quantiles of_n Unconditional per capita household income centiles Unconditional per capita household income centiles Brazil Nicaragua Tertiary incomplete Tertiary complete Centile average of return Centile average/prediction Centile average of return 0.80 2.0 1.8 0.75 1.6 1.4 0.70 1.2 0.65 1.0 5 15 25 35 45 55 65 75 85 95 5 15 25 35 45 55 65 75 85 95 100 quantiles of_n 100 quantiles of_n Unconditional per capita household income centiles Unconditional per capita household income centiles Centile average Prediction Source: Authors' estimates based on household survey data. 188 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A that some workers from poor families also benefit from Bolivia is one of the few countries in the sample where unobserved labor market abilities that are complementary marginal schooling returns are higher for workers at the to schooling. However, the poor face lower returns to ter- bottom of the job ladder; this happens in the transition tiary education returns, since they tend to have a dispropor- from primary to secondary school and for completion of ter- tionate disadvantage in the production of skills at home and tiary education. Recall that marginal returns to having school. Note that returns to basic and secondary education some tertiary education are lower for the low-ranking are fairly constant along family income lines and, in some Bolivian workers. That is, the few low-ranking Bolivian countries like Argentina, Colombia, and El Salvador, may workers who reach tertiary education enjoy a relatively slightly favor the poor. The low and flat returns to lower larger boost in earnings along the way but end up with levels of education offer similar investment disincentives to similar returns to the investment once they get a university the poor and the rich. diploma. The latter is highly suggestive that liquidity con- Figure 9.18 shows that differences in schooling returns straints hinder transitions to higher grades in Bolivia. blur the incentives to make additional investments in sec- What lies behind these differences in the returns to ondary and higher education for workers that rank low in schooling? As noted earlier, education and incomes may be the adjusted salary scale (those with lower unmeasured highly correlated across generations. The poor are also con- skills). For example, in Bolivia and Chile the marginal strained by longer-term family factors that affect both educa- earnings gains from having some tertiary education are tional achievement and adult earnings, such as home close to 80 percent for the best-paid workers, but only schooling, family wealth (which buys quality schooling), and 30 to 40 percent for those who end up in the less-well-paid family connections. Family background and school quality-- jobs. The differential returns are less staggering in other information rarely collected in survey data--remain countries like El Salvador and Nicaragua but are still unaccounted for in the analysis, which may cause us to mis- significant and add up to overall low marginal returns to represent the returns to education, as well as the impact of tertiary education. short-term liquidity constraints in educational attainment. FIGURE 9.18 Returns to education are generally lower for workers at the bottom of the earnings scale Marginal returns to schooling along the earnings scale Marginal returns to schooling along the earnings scale Log wage gains from reaching each level of education Log wage gains from reaching each level of education 0.80 0.80 0.70 0.70 0.60 0.60 0.50 0.50 0.40 0.40 0.30 0.30 0.20 0.20 0.10 0.10 0 0 Low-paid jobs Average pay Best paid Low-paid jobs Average pay Best paid Brazil (secondary complete) Peru (tertiary incomplete) Bolivia (tertiary complete) Argentina (primary complete) Bolivia (tertiary incomplete) Argentina (secondary incomplete) Mexico (primary complete) Argentina (primary incomplete) Chile (tertiary incomplete) Chile (secondary incomplete) Bolivia (secondary incomplete) El Salvador (primary complete) El Salvador (tertiary incomplete) Bolivia (secondary complete) Brazil (tertiary complete) El Salvador (secondary incomplete) Nicaragua (tertiary incomplete) Source: Authors' estimates based on household survey data. 189 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S An examination of the data in Colombia and the The spare evidence on the impact of school quality in Dominican Republic, the only two countries with reliable Latin America suggests that it is a significant source of parental education data in recent household surveys, indi- variation in the returns to education. The Arias, Yamada, cates that the offspring of more-educated parents do enjoy and Tejerina (2004) study for Brazil measured the impact higher earnings, but returns to education are not vastly of education quality on schooling returns from cross-state overstated as a result. Failing to purge parental effects on and intercohort variations in pupil-teacher ratios--proxies the earnings of their offspring overstates education returns for education quality. Figure 9.19 illustrates its main find- by only 3­7 percent, except for the returns to primary edu- ing: workers educated in states with a lower pupil-teacher cation for low-paid workers and to tertiary education for ratio (say, by 10 students) have higher average returns to the best-paid workers, which are overstated by 40 percent education (by 0.9 percentage point for each year of school- and 11 percent, respectively. Similarly, as shown in figure ing). Large class sizes are not uncommon for Latin Ameri- 9.18, Arias, Yamada, and Tejerina (2004) found that the can poor children, especially those in marginal urban returns to education in Brazil are about 10 percent over- schools. The pupil-teacher ratio is also correlated with stated due to the joint impact of family background on the other key inputs of the educational process, such as instruc- education of children and youth and their earnings as tional time, educational materials, and teachers' education adults. While still sparse, this evidence is remarkably con- and experience. In another study for Brazil, Albernaz, sistent with the consensus of the literature that estimates Ferreira, and Franco (2002) found that other indicators of returns to education in the United States to be slightly school quality, such as teachers' educational level and overstated by about 10 percent. This suggests that the esti- school infrastructure, have significant effects on children's mated returns to education shown here are not severely educational performance. Mizala and Romaguera (2002) misrepresenting the earnings-schooling relationship.27 summarize the evidence for other countries in the region. In addition to the well-known positive effect on chil- Therefore, differences in education quality could plausibly dren's educational attainment, the education of parents account for an important portion of the gaps in returns to boosts the earnings of sons and daughters. In Colombia and education between the poor and nonpoor in the region. the Dominican Republic, children's earnings are increased This highlights the critical importance of enhancing the by 20­35 percent (7­15 percent) for each parent with a col- quality of the educational supply for the poor. lege (high school) education compared with a parent with To summarize, the high value ascribed to a university primary education. This could reflect an impact on returns education in Latin America is not available to everyone. to education that is difficult to isolate with cross-section College-educated workers with lower unmeasured human data. Using longitudinal data, Altonji and Dunn (1996) capital, particularly the poor, do not receive the same returns found that returns to schooling are higher for children of to their education as do other workers with college educa- more-educated parents. In Brazil, Arias, Yamada, and tion. Long-term family factors, particularly education Tejerina (2004) found substantial earnings payoffs to quality and parental education, appear to be important higher levels of parental education that vary across race determinants of the productivity of schooling investments groups. Father's education generates more significant earn- and earnings as adults. While the total returns to tertiary ings gains for whites, while mother's schooling was more education for the poor are still significant, even mild liquid- important to boost the earnings of nonwhites. The authors ity constraints could quickly take children and youth from interpret these as suggestive that father's education plausi- disadvantaged families off the path to reaching higher educa- bly proxies wealth and thus school quality and family tion grades. In the next section we weigh the evidence on the connections in the labor market. Meanwhile mother's relative contribution of short-term and long-term poverty schooling more closely captures differences in the home factors to Latin America's persistent educational divide. production of skills in light of the low female labor force participation at the time workers were schooled. This Short-term or long-term poverty: Which is more means that effects of parental education need to be pressing for schooling investments? accounted for before interpreting a correlation between low We discern the relative importance of liquidity constraints family incomes and low educational attainment as evidence and long-term family factors in preventing Latin Ameri- of short-term liquidity constraints. can children from getting sufficient schooling to escape 190 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A FIGURE 9.19 Education quality differences lead to differential returns to education in Brazil Pupil-teacher ratios in Northeast and South Brazil, 1940­90 A decline of the pupil-teacher ratio by 10 increases the average return to education by almost 1 percentage point per year of schooling % nonwhite workers among total educated Returns (adjusted) 50 10 45 40 5 35 30 0 25 20 5 15 1938 1942 1946 1950 1954 1958 1962 1966 1970 1974 1978 1982 1986 10.36 7.88 Pupil/teacher ratio (adjusted) Northeast Brazil South of Brazil White Nonwhite Source: Arias, Yamada, and Tejerina (2004). Note: The left figure is based on administrative school data from the Anuario Estatistico do Brasil, various years. The right figure shows the fitted regression of estimates of the average education returns by state, cohort, and race and the associated pupil-teacher ratios. Both variables are depicted as deviations from their means within cohort. Returns were estimated for male household heads with Mincer earnings regressions controlling for parental education. poverty by means of survival (hazard) regressions (see along the region's gradient of educational development. annex 9A). This analysis is common in clinical studies of Figure 9.20 illustrates the main results. The findings are the effect of a new drug treatment on patients' chances of summarized below. "survival" from a disease after a certain time has elapsed. · Family effects do matter a great deal. Compared with hav- We examine how child and family characteristics affect the ing a college-educated mother, having a mother with only risk that children and adolescents (6­25 age range) fail to primary education increases the risk of school dropout by as enroll in school (primary, secondary, or tertiary) at a given much as 160 percent in Chile and 60 percent in El grade (a proxy of school dropout) given the number of Salvador. A father with low education additionally increases grades already completed, thus capturing the sequence of the risks of school failure by up to 140 percent in Chile and the entire schooling investment process.28 Incomes, prox- 40 percent in the Dominican Republic. These are substan- ies of physical access, returns to education, and family tial impacts given the high degree of assortative mating in demographics could be considered "treatments" to the the region. These risks are cut by one-half or two-thirds when extent they can be manipulated by specific policy inter- the parents have a secondary education; children of Central ventions. School variables that affect the learning and American fathers with high school education have the same schooling process are not explicitly part of the analysis chance as children of college-educated fathers to move up due to lack of data, so their effect is captured by family the educational ladder. In Colombia, having grandparents socioeconomic characteristics that influence the capacity to with little education increases the risk of school failure of access better-quality schools. The analysis is conducted children and youth even when parental education, incomes, for Brazil, Chile, Colombia, the Dominican Republic, and other family characteristics are accounted for. That is, El Salvador, and Nicaragua to illustrate the effects low educational attainment in Colombia tends to persist 191 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S FIGURE 9.20 Factors that have an impact on moving up the educational ladder Risk of school dropout and mother's education Risk of school dropout and father's education % change in risk compared to college-educated fathers % change in risk compared to college-educated mothers 160 140 140 120 120 100 100 80 80 60 60 40 40 20 20 0 0 il il Chile Braz Chile Rep. Braz Rep. Salvador Nicaragua Salvador Dominican Colombia El Nicaragua Colombia El Dominican Mother with primary education Father with primary education Mother with secondary education Father with secondary education Risk of school dropout and poverty Risk of school dropout, ethnicity and infrastructure % change in risk compared to middle-income families % change in risk compared to whites and paved road access 60 40 70 20 50 0 30 20 10 40 10 60 30 Ethnicity Unpaved roads dor Brazil Chile Rep. aragua Salva Dominican Nic Colombia El Pardo (Br) Indigenous (Br) Preto (Br) Amarelo (Br) Nonwhite (Col) Poorest 20% Wealthiest 20% Dominican Rep. Nicaragua Risk of school dropout and gender Risk of school dropout and school access % change in risk compared to female children % change in risk compared to children in rural areas 60 0 50 10 40 20 30 30 20 40 10 0 50 il il Chile Chile Rep. aragua Braz aragua Braz Rep. Salvador Colombia Colombia Salvador El Dominican Nic Nic El Dominican Source: Own estimates based on household survey data. Note: These are the risk ratios in percentage terms from a hazard regression of school attainment on a set of family characteristics. In Colombia and the Dominican Republic, regressions include grandparents' education. 192 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A strongly across three family generations. The effects of the below a subsistence threshold interferes with school education of the mother and grandmother are remarkably progression. For example, in Nicaragua only children from similar but are lower for grandfather's than for father's edu- the poorest 40 percent of families face a higher risk of cation (see figure 9.19). It seems plausible that simultane- school failure than the richest families. In El Salvador and ous conditioning on current income and parental and Nicaragua, international remittances--which in this con- grandparents' education yields cleaner measurements of text are a relatively more exogenous income source--lower liquidity constraints and the quality of skills development the risk of school failure, although modestly. Boys, irre- at home and schools. Thus the results affirm that both spective of whether they are rich or poor, face a much short-term and long-term family factors (family back- higher risk of dropout than do girls in Brazil, the Domini- ground, liquidity constraints) and the quality of schools can Republic, and Nicaragua (40­60 percent), and a mod- are central to accelerate human capital formation in the estly higher risk in Chile and Colombia (13­17 percent). region. Moreover, each additional young sibling (age 6­12) There is further evidence that higher expected returns to increases the risk of school failure for any one of the siblings education (at higher grades) and a better-quality home envi- (by 4 to 22 percent across countries), while more children ronment correlate with more stable school progressions. of secondary school age actually lower the risk in Brazil, the Children from nonwhite families face a higher risk of leav- Dominican Republic, and El Salvador (by 5 to 22 percent). ing school early: 20 percent in Colombia and Nicaragua, All of these effects were obtained controlling for parental and 17 to 52 percent for pardos, pretos, and indigenous peo- education and proxies of family returns to education and ple in Brazil. Moreover, children and youth in female- are thus highly suggestive that liquidity constraints are headed households face a three to four times bigger risk of binding, to different degree, in all of the countries. dropping out of school in Brazil, Chile, and the Dominican · Physical access constraints remain operative, binding most Republic, although school dropout is not affected by when returns are higher. The risk of school failure is 40 per- whether the mother is a salaried employee or self-employed. cent higher in the rural areas of Brazil, Colombia, and Since we are purging income and family background fac- Nicaragua (all countries with higher returns to education) tors, the latter effects plausibly reflect the lower expected and 20­30 percent higher in the rural areas of the Domini- returns to education for nonwhites, as well as the constraints can Republic and El Salvador (with the lowest returns). on single parents in providing quality school supervision of Deficient infrastructure (proxied by unpaved roads) children (such as doing homework) and role models. increases the risk of school dropout by 80 percent in Although the effects are small, some evidence shows that Nicaragua and by 30 percent in the Dominican Republic. the expectation of higher returns to education at higher The poorest regions in Brazil, Chile, and Colombia, where grades also encourages more-even school progressions. basic infrastructure is generally more deficient, show Using a proxy, albeit imperfect, of the differential schooling higher risks of school failure, but these become weaker or return that children might face due to abilities inherent to even reverse signs after adjusting for family socioeconomic their families, we find that those with higher family returns characteristics. Migrants are at higher risk for dropping out for secondary completion (Brazil and Nicaragua) and college of school in Colombia (15 percent, in part perhaps captur- education (Chile and Nicaragua) are less likely to drop out ing violence-related displacement), Nicaragua (45 percent), of school.29 Altogether, these findings further reaffirm the and the Dominican Republic (70 percent); only in Brazil role of long-term family factors in enhancing the productiv- do they face lower risk (5 percent). However, school supply ity and incentives for schooling investments. does not seem the most prevalent consideration for migra- tion. For instance, only 14 percent of Dominicans age 3­22 · Liquidity constraints play a relatively smaller but signifi- who migrated in the past five years stated school-related cant role. Children and youth from the poorest 20 percent of reasons; a similar fraction sought income opportunities. families face a higher risk of school failure compared with What conclusions can we draw from these results and those from middle-class families: the difference ranges from the preceding analyses? The main lesson is that long-term 55 percent in Brazil to 20 percent in Chile. This risk is half family factors, liquidity, and school access constraints con- as large for families in the second quintile and then tapers spire, in different degrees, to generate human capital off the richer a family becomes, suggesting that being underinvestment traps that hinder sustained and balanced 193 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S educational progression in the region. The two main ele- · Brazil's low schooling attainment and high educa- ments interacting in the resulting vicious cycle are a pat- tional inequality likely arise from the interplay of tern of schooling returns that makes it unattractive for multiple sources: high but very unequal returns many poor families to invest in education, namely, returns (which are lower for the poor) to secondary and ter- that are low and flat in the eight-year basic education cycle, tiary education, persistent intergenerational family rise significantly at the tertiary level but are lower for poor effects, pressing liquidity constraints, and localized families and only occasionally mature fully when a degree supply bottlenecks. Mexico's sharp educational is completed; and liquidity constraints stemming from divide reflects a similar though less marked situation, subsistence incomes and borrowing constraints. as Mayer-Foulkes (2004) has more fully documented. The extent of underinvestment traps and the relative · Finally, the acceleration of educational transitions in weight of the intervening factors varies across countries. A the Dominican Republic and El Salvador is con- few patterns can be identified that are likely responsible for strained mainly by exceedingly low returns to educa- reinforcing educational divides within countries: tion on top of already low overall earnings, largely related to poor readiness to learn (a result itself of low · Chile, Colombia, and Peru are the countries rela- parental schooling) and particularly deficient educa- tively better positioned in our sample to experience a tion quality.31 Thus, liquidity constraints in these faster transition toward a diamond-shape (broad sec- countries do not appear to be as important as increas- ondary base) educational distribution; these three ing the incentives of families to make sustained countries are favored by relatively high and smoother investments in education. returns to schooling and a relatively lower fraction of the prime-age population with very low education These are not intended as exhaustive explanations of the (less so in Colombia). Potential limiting factors are low educational attainment in these countries, but as unequal schooling returns (especially in Chile) and important links to poverty and its intergenerational trans- liquidity and learning constraints related to family mission. Similar patterns may be operative in other Latin educational and wealth endowments and ethnicity, American countries where poor children and youth do not which result in home and school quality gaps for the succeed in completing higher grades. Each merits detailed poor's offspring. examination in specific country studies incorporating insti- · Bolivia's unequal educational transition and tutional analyses of the educational systems. Nicaragua's very low educational attainment result from a similar set of limiting factors, with a strong Implications for human capital role played by liquidity constraints exacerbated by formation policies relatively high returns that materialize fully only near This chapter examined how Latin America's educational or upon degree completion and by larger gaps in sec- divide between two groups of low and highly educated ondary school infrastructure. The low levels of skills individuals is simultaneously a source and a result of in these two countries pose a high risk that they will subsistence incomes across generations. As for any invest- fall into a self-reinforcing cycle of low technology, low ment, the confluence of opportunity (attractive returns) and demand for skills, and low innovation and skills possibility (liquidity, quality schools, and home environ- investments. ments) is essential to human capital accumulation. Poor · In Argentina, a high fraction of poor families with Latin American families lack elements from both in differ- low parental education, low returns to the primary ent degrees. The main overall implication of the results dis- and secondary education cycle, uncertain tertiary cussed here is the need for integrated, long-term strategies returns (maturing with degree completion), and high for skills development that exploit the synergies in the life- discounting of the future may be preventing poor cycle human capital accumulation process in which both children from sharing in the fast transition of recent families and schools play a central role. Specific implica- age cohorts to largely free secondary and tertiary pub- tions for human capital formation policies (nutrition and lic education.30 health, education, and training) are: 194 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A · Leveling the initial playing field for children at risk. It is other modalities such as distance education can be consid- imperative to address the unequalizing impact of deficien- ered when the preconditions for their success exist. cies in early-childhood development and deficient parent- Liquidity constraints have been the main motivation for ing on the educational attainment of poor children and cash transfers to the poor tied to school attendance, as in their capacity to command higher returns to education as the Oportunidades program in Mexico, Bolsa Escola in Brazil, adults. Although nutritional failures are very hard to rem- and similar programs in Central America and the Andean edy after the child's first two years, almost half of Latin region. The opportunity cost of children's school atten- American and Caribbean countries are not on track to meet dance does not seem very binding until the child completes the UN Millennium Development Goal of halving malnu- primary school or reaches the lower secondary grades. trition by 2015. Schemes that encourage investments throughout full Well-targeted interventions to strengthen the capacities courses of basic education or lower secondary education (for of families to create early human capital should be priori- example, a lump-sum grant for those graduating from high tized. For example, conditional cash transfer programs can school) may hold substantial promise for reducing dropouts be used to induce parents to devote more attention to chil- and inducing poor parents to invest more time helping dren's health and nutrition by conditioning transfers on their children succeed in school. maternal and infant health care. The experience with the Well-designed (means-tested and merit-based) univer- Head Start program in the United States and similar inter- sity student loan programs and scholarships also have a ventions elsewhere in the world can serve as a guide for role in facilitating access for low-income and high- more systematically targeting infants at long-term risk. performing students. These should build in features to Although costly, these interventions are very likely to pass ensure their sustainability, such as delegation of loan pro- rigorous cost-benefit assessments because of their demon- cessing and recovery to private banks with partial govern- strated long-term impacts on children's readiness to learn ment guarantees on the repayment. These loan programs and socioeconomic success as adults. may be more feasible with the gradual development of · Strengthening the full option value of education for the poor. individual credit registries that increase the long-term Since families factor in the promise of the payoff to higher costs of a default. Moreover, a strategic partnership with education in their investment decisions, educational poli- the private sector (including private universities) and cies should adopt a systemic view. Fragmentary educational civil society is needed to fund and operate these programs policies, focused solely on ensuring narrow objectives such through competitive biddings. Needed also are policies to as primary completion or coverage goals, are no longer as promote the development of the tertiary education effective in the global economy where a minimum of market, such as those discussed in de Ferranti and others secondary education is needed to compete for above- (2003). subsistence wages. While scarce resources and political · Making education count for the poor. The take-up rate on capital require setting spending and reform priorities, student loans--or for that matter enrollment in free public removing binding supply and demand constraints at all universities--may be low because eligible persons perceive levels of the education system, even on a small scale, is cru- that their expected returns to tertiary education do not com- cial to signal low-income families that their educational pensate for the forgone earnings. Gaps in enrollment in sec- investments have better chances of maturing with ondary schools and above persist in Argentina, Brazil, and improved access to higher grades. Mexico, where public university is largely free. Thus, poli- Where education returns are high and basic infrastruc- cies are needed to increase the returns to education for the ture is deficient, public investments in the construction poor to encourage them to move up the education ladder. and upgrading of schools and roads are essential. The devel- The main challenge is to gain a better understanding of opment of multigrade schools, learning from best practices how to reduce grade repetition among the poor. The role of such as the Colombian Escuela Nueva and the Chilean automatic promotion policies in the early grades, learning MECE Rural, can address supply constraints cost effec- deficiencies due to poor learning environments at home, tively. Public-private partnerships to exploit good-quality and failures in the instruction process, including inade- private urban secondary schools with excess capacity and quate teaching and large class sizes, should be analyzed 195 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S with data on schools, children, and family characteristics Investing now: The demographic window through vigorous impact analyses. of opportunity Possible policies include decentralizing school manage- Demographic forces offer many countries in the region a ment to get parents more involved and committed to their unique opportunity to translate the human capital accumu- children's school progress, offering incentives to encourage lation of young cohorts into a more productive labor force qualified teachers and principals to work in disadvantaged and a faster reduction in poverty. Most countries are in the schools, adapting innovations to improve learning environ- midst of a demographic transition where the "dependency ments in disadvantaged schools and communities, upgrad- ratio" (the fraction of the population that is too young or ing textbooks and school aids, providing teacher training, too old to work) is declining. This is illustrated in fig- and expanding computer education in secondary schools. ure 9.21 for Bolivia; Bolivia and Haiti are the only Latin The consistent application of international standardized American countries just beginning the first stage of demo- tests to assess performance progress should become com- graphic transition. As countries go through this transition, mon practice. Unfortunately, there is not a well-tested labor force participation is expected to rise. Because the recipe to follow, but rather a host of international experi- share of younger cohorts in the working-age population ences, both failures and successes to learn from. will rise faster, older and poorly educated workers can be Some targeted and performance-based increases in pub- replaced with younger workers at a fast pace. Most Central lic expenditures, particularly at the secondary level, might American countries just recently started this process and be needed in some countries. While overall education can still reap most of these benefits, while the rest of the expenditures in most countries in the region are not low region is much more advanced but still has a decade or so to and increases in spending do not always translate into take advantage of the transition. better outcomes, there might be limits to what can be As the bottom panel of figure 9.21 shows, changes in achieved with pure efficiency gains unless expenditures in fertility in most of the region are favorable to human capi- education are increased. Countries such as the Dominican tal accumulation. In almost every Latin American country Republic and others in Central America have clear expen- today, fertility rates are falling, families are having fewer ditures deficits and are already relatively output efficient, so children, and women are increasingly joining the labor a sustainable increase in education expenditures is needed. market. This means more resources to invest in quality Other policies to improve access to jobs may include education for children as well as lower costs of making the enacting and enforcing antidiscrimination laws and estab- investments. But patience is required. This is a gradual lishing intermediation services that help well-educated transition, and it will take more than a decade for skill ethnic and racial populations obtain greater access to investments to translate into a more productive labor force better-quality jobs. Where returns to education are too and improvements in national and family incomes. low, the best medium-term policies lie in promoting Human capital formation, including schooling, is an technology-intensive investments that demand skills. This extremely time-dependent process. For families unable to is actually a precondition to ensuring a country's ability to do it at the right time, the opportunity is gone. In maintain attractive private returns to higher levels of Argentina, 30 percent of workers ages 41­65 and 20 per- education under a massive educational expansion. cent of prime-age workers are stuck with a basic education · Interventions to fill minimum instructional gaps of the adult that puts those heading families at high risk of poverty. population. Given the strong family effects we have shown These families have to wait a decade or more before any here, especially of parental education, there is a role for pro- schooling bequests to their young children can lift family grams targeted at improving the educational level and incomes significantly. Further taking into account the pos- skills of the adult population. Recent experiences in Chile itive spillovers of a labor force with rising minimum levels and Mexico in support of lifelong learning hold some of education on technology adoption, productivity, and promise. For instance, the national Chile Califica program is growth, it is hard to overstate the critical importance of designed primarily to strengthen the link between what is pushing the "education for all" agenda. In many countries, taught in the latter years of secondary schools and what the the demographic window of opportunity is closing; the time labor market demands. to invest is now. 196 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A Annex 9A FIGURE9.21 The demographic transition and human capital accumulation--an opportunity that should not be missed Data and methodological details Data Window of opportunity (dependency ratio) We employ household living conditions and labor force a. Bolivia and Latin America surveys for 10 countries chosen to represent the different Dependencyratio levels of educational development in the region. Below are 1.5 the countries, the national household survey data sets used 1.4 in the report, and their educational ranking: 1.3 1.2 Andean Countries: Bolivia, ECH-MECOVI 2002; 1.1 Colombia, ECV 2003; Peru, ENAHO 2002. Bolivia 1.0 Central America and the Caribbean: El Salvador, EHPM 0.9 2002; Mexico, ENIGH 2000; Nicaragua, EMNV 2001; LatinAmerica Dominican Republic, ENCOVI 2004. 0.8 0.7 South America: Argentina, EPH 2003; Brazil, PNAD 2002; Chile, CASEN 2001. 1950 1960 1970 1980 1990 2000 2010 2020 2030 2040 2050 b. Number of children Estimation of returns to education Changeinfertility We rely on Mincer earnings functions: Ln Wij = aj + bj educij 1 + qj Xij + eij, with j = quantile of the earnings distribution. 0 We are primarily interested in the bj (returns to education), controlling for some demographic characteristics 1 2 FIGURE 9A.1 3 Labor force by educational level 4 Nicaragua 5 El Salvador 0 2 4 6 8 10 Bolivia Fertilityrateininitialperiod Dominican Rep. Restoftheworld LACcountries Brazil Source: IDB(2004),andauthors' estimatesbasedoncross-country data. Mexico Note: Dependencyratio (populationage65andolderor15and Colombia under)/populationage15to64. Peru Chile The best policies, in terms of a social cost-benefit calcu- Argentina lation, may not be the most palatable for short political 0 10 20 30 40 50 60 70 80 90 100 horizons or for political economy reasons. Such is the case Percent with early-childhood interventions and major reforms of No education Primary incomplete the educational system. Overcoming political failures that Primary complete Secondary incomplete prevent consensus around the need to address the large Secondary complete Tertiary incomplete achievement gaps between poor and nonpoor children is Tertiary complete critical to the region's long-term human capital accumula- Source: Authors' estimates based on household surveys. tion and prospects for sustained growth. 197 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S (Xij = gender, urban/rural). Education is specified as a set of second-stage regressions are weighted to account for the year dummies for the last grade completed (a total of 18) standard errors of the estimates of quantile return coeffi- and as 6 dummies for the maximum level of education cients from the first stage Mincerian equations. (incomplete and complete). Marginal returns to education are derived from the difference in log wages between two Estimation of hazard "schooling-progression" consecutive grades or education levels. We estimate these functions equations for 20 percentiles of the earnings distribution To ascertain the role of factors affecting the cost and using quantile regression to assess the consistency of the expected benefits from continuing schooling, we employ behavior of returns along the distribution and report Cox hazard regression methods (see Cox Edwards and Ureta results for selected quantiles. 2003 for a first application to school attainment). Hazard (risk) analysis is synonymous with time-to- Linking returns to education to poverty event analysis, which studies a variable that measures the To map the quantile returns to education to the per capita duration between a particular starting time (entrance to family income distribution, we employ the following school) and a particular end time of interest (school methodology, used by (Arias (2004) and Tannuri-Pianto, dropout), and a set of independent variables thought to be Pianto, and Arias (2005). related to the end-time variable (school dropout). In gen- Quantile regression allows us to measure heterogeneity eral, censored observations arise whenever the dependent in the returns to education that is not related to measured variable represents a time to event, and the duration of the worker characteristics. The ranking of workers in the con- study is limited in time. In this case, the time to event is ditional earnings distribution can be taken as a proxy of the time between completion of a one-year study period their level of "ability," or unmeasured earnings determi- and the time the child drops out of school. Subjects that nants. We would like to link these conditional returns to are not enrolled at the time of the survey and did not com- workers' positions in the (unconditional) per capita house- plete a full education course (primary to college) are hold income distribution. assumed to have dropped out. The individuals who are To do this, we first identify the conditional quantile of enrolled represent censored observations, since they have each worker in the wage distribution. We perform quantile not yet completed their entire education spells. The regressions for 20 quantiles and then identify the quantile method of analysis takes the censoring into account and to which each worker belongs as the quantile for which the correctly uses the censored observations as well as the worker is predicted to have the smallest wage residual (in uncensored observations. absolute value). That is, the conditional quantile of worker Assuming away reentry after a temporary absence from i given by i is determined as i = arg min(s ), where = i school, schooling attainment is the last grade completed 0.05, 0.2, . . . , 0.95. We assign to each worker the esti- before the failure to enroll, that is, the years of completed mated education coefficient for his or her level of education schooling. The event that schooling attainment G takes the and given quantile i. value g is equivalent to the event that the child drops out of Next we compute household-specific returns for each school after achieving g grades. Thus, the probability of level of education by averaging the return coefficients failing to enroll in g + 1 matches the probability of attain- across workers who belong to the same household. This ing g years of schooling, conditional on past enrollment implicitly averages the level of "ability" of working mem- decisions. From this we can derive the risk, or hazard rate, bers to obtain the family return to education. We then of dropping out of school after completion of grade g and regress the household-specific returns for each education before the completion of grade g + 1, given that the child level on the (unconditional) household's percentile, Ci, in has continuously been in school up to the g + 1 enrollment the per capita household income distribution. The samples time. A "failure" event here is to drop out after grade g, in these second-stage regressions are composed of house- which exactly corresponds to the failure of enrolling in holds with positive returns. We consider two specifications, g + 1 at the beginning of the school year. the first including only dummies for the five quintiles and The hazard rate in this case is the probability that an the second, the unconditional percentile and its square. To individual will drop out of school at a certain point in time properly gauge the statistical significance of the results, the (at risk of dropping out), that is, the rate at which dropouts 198 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A occur. The aim of the analysis is to determine how the inde- the unobserved component of the family return to educa- pendent variables (covariates) described below affect the tion. For the first we use two specifications, one controlling school dropout rate. For example, if a child has a hazard for the log of income per capita, and another including a set rate of 1.2 at six years of education and a second child has a of dummies for the family's income quintile. The latter is hazard rate of 2.4 at the same time, then the second child's useful since liquidity constraints are consistent with non- risk of dropping out would be two times greater at six years linear income effects; in other words, the poorest house- of education. holds (first income quintile) should be relatively more We use the Cox's Proportional Hazard model, which constrained than, say, the not-so-poor or the middle-class assumes that independent variables exert a proportional households, especially for sending their children to private effect on the baseline hazard rate of school failure. Cox's schools. For the second issue, we use the average return to regression model is a semi-parametric technique that education of each family computed from all working models: household members (ages 26­65) and their rank in the conditional wage distribution (that is, the return at the (9A.1) h (t), (z1, z2, . . . , zn) = h0(t)e(b1z1 + + bmzm) , percentile at which they fall in the distribution), as where h[(t), (z1, z2, . . . , zn)] denotes the hazard ratio, given explained above. The latter is a proxy, albeit imperfect, of the values of the covariates. The term h0(t) is known as the the expected differential return that a child or youth might baseline hazard, that is, the hazard for the respective indi- face from each level of education due to the abilities inher- vidual when all independent variable values are equal to ited from his or her family. These returns for each level of zero. This can be estimated through a linear model of the education (primary to college completion), adjusted by form: their estimated standard errors, are included in some of the hazard regression specifications. School variables are missing from the schooling regres- (9A.2) log h (t), (z1, z2, . . . , zn) = b1z1 + + bmzm sion analysis. This means that family background variables, that is, parental education, also capture family wealth The estimated coefficients can be interpreted as relative effects that allow access to better-quality schools. Also risk ratios. The baseline survival curve is shifted up or missing are variables capturing the scholastic ability of down by each of the covariates. The proportional hazard children. In an uneasy truce with available data, we hope technique estimates a coefficient for each independent vari- any biases are ameliorated by the controls for family back- able that indicates the direction and degree of flexing that ground variables (especially in Colombia and the Dominican the predictor has on the survival curve. A coefficient equal Republic, where grandparents' education is included) and to 0 (relative risk ratio of 1) means that a variable has no by imputed measures of family earnings abilities. effect on the baseline hazard; a positive coefficient (risk ratio greater than 1) implies that larger values of the vari- able are associated with a greater risk of school dropout; Notes and a negative coefficient (risk ratio less than 1) means a 1. Card (1999), Lemieux (2004), and Heckman and Todd (2004) lower risk. offer a comprehensive review of the literature. Psacharopoulos and Patrinos (2004) provide a large set of cross-country empirical results. The hazard regressions include a full set of family char- 2. The poverty rates are based on national poverty lines and acteristics: gender of the child, area of residence, family per therefore should not be used to make comparisons or rankings across capita income, international remittances when available, countries. the number of children ages 6­12 and the number ages 3. The most recent reports can be found in www.worldbank.org\ 13­17, education of the father and mother (for the sample lac\poverty. that still live with their parents), whether the household is 4. As reported in de Ferranti and others (2003), high-performing countries in East Asia increased their average schooling by just under headed by a female, whether the mother and father work as five years between 1960 and 2000, while most countries in Latin salaried workers or are self-employed, some interactions of America and the Caribbean increased theirs by two to three years. these variables, and regional control dummies. during this period. Two important features of the analysis are examinations 5. See Mayer-Foulkes (2004) for a review of numerous studies, of the effect on enrollment of liquidity constraints and of and also the 2005 World Development Report. 199 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S 6. This happens when employers regard these workers as more Peru, and 20­30 percent in Mexico. Annual tuition costs are almost talented (more productive) so that a diploma acts as a signal of their equivalent to per capita income in Brazil and Colombia and 30 to productivity, as illustrated in the job market signaling model of 50 percent of per capita income in Argentina and Chile (de Ferranti Nobel Prize winner Mike Spence. and others 2003). 7. This important distinction is present in the various studies of 19. The surveys in Chile, Colombia, the Dominican Republic, Heckman and coauthors on human capital accumulation and hetero- Mexico, and Nicaragua contain information to distinguish individu- geneous returns to schooling. See, for example, his Nobel Prize lec- als with a tertiary (university) diploma. In the other countries, ter- ture (2000). tiary completion was assigned to those with five years of tertiary 8. See Mayer-Foulkes (2004) and Heckman (2000) for empirical education or more. evidence from numerous studies. 20. The estimates of returns are comparable to those reported in 9. When parents cannot set a higher educational bar by example, de Ferranti and others (2003), being within 2 percentage points dif- children and youth could turn to relatives, peers, or mentors. Durlauf ference in some countries, but differ from those reported in IDB (1996), Bénabou (1994), Manski (2000), Akerlof and Kranton (2004), which are generally much larger. The difference stems from (2002), and sources in Bowles, Durlauf, and Hoff (2004) show how surveys, samples (IDB 2004 is restricted to prime-age men), and these mechanisms can generate low human capital formation and measurement methodology (treatment of incomplete and complete poverty traps. Lalive and Cattaneo (2004) present evidence of the degrees). We impose fewer restrictions on the sample and estimating impact of social interactions on schooling decisions. equations. 10. See, for example, Lucas (1988), Azariadis and Drazen (1990), 21. The high school graduation effects are weaker in the Domini- Kremer (1993), and Acemoglu (1997) for growth and poverty-traps can Republic, Mexico, and Peru. In Brazil and Colombia, returns models of skill agglomerations, and De Ferranti and others (2003) for jump in the 11th grade, the last year of secondary school in these empirical evidence on the correlation between technological and countries. skills investments in Latin America and the Caribbean. 22. Education alone accounts for up to one-third of overall earn- 11. Behrman, Duryea, and Székely (1999) conclude that 80 per- ings differentials in Latin America and the Caribbean. The fraction of cent of the slowdown in educational progress in the region in the the variance in earnings explained by education, gender, and region 1980s and 1990s was associated with macroeconomic volatility. of residence is as high as 0.48 in Brazil and Colombia and as low as Carneiro, Hansen, and Heckman (2003) find supporting evidence of 0.05­0.10 in rural areas of El Salvador and Nicaragua (given that a negative effect of variation (uncertainty) in the returns on college there is little variance in earnings differentials in rural areas). Other attendance in the United States. factors, including differences in education returns, contribute to 12. The patterns tend to persist between families given the high earnings inequality in the region. degree of assortative mating on the basis of education (de Ferranti 23. See Hall and Patrinos (2004) and Jiménez and Landa (2004) and others 2004). Distributions by gender reveal that girls and boys for Bolivia; Trivelli (2004) for Peru; Larrea and Montenegro (2004) have about the same level of school attainment in most countries. for Ecuador; Arias, Yamada, and Tejerina (2004) for Brazil. 13. Only the Dominican Republic shows a relatively equal, flat 24. This strand of studies is growing exponentially. See, for exam- distribution of schooling for both the rural and urban labor force. ple, Carneiro, Heckman and Vytlacil (2001); Carneiro (2003); Argentina's household survey does not collect data for rural areas. Carneiro, Hansen, and Heckman (2001, 2003); Carneiro and 14. The cohorts cover those individuals born in 1980­90, Heckman (2003); and Arias, Sosa-Escudero, and Hallock (2001) for 1965­79, and 1940­64. the United States. See Blundell, Dearden, and Sianesi (2005) for 15. See, for example, World Bank (2004a) for Central America as European countries. For numerous Latin America countries, see well as recent country poverty assessments. In a few countries, such as World Bank (2004); Arias, Yamada, and Tejerina (2004); Arab- the Dominican Republic, the income-enrollment gaps are modest sheibani, Carneiro, and Henley (2002); Lopez-Acevedo (2001); Mon- (World Bank 2005b). tenegro (2001); and Saavedra and Maruyama (1999). 16. It is obtained by cumulatively adding age-specific net enroll- 25. From a social standpoint, there could still be a case for public ment rates. For example, if the net enrollment rate in a given country intervention to address underinvestment given the positive external- is 86 percent at age 6 and 93 percent at age 7, the average 7-year-old ities of education in the form of lower fertility, crime, and the like. in the country has spent 1.79 years in school. See Urquiola and 26. The grade-specific return profiles are similar to those in the Calderón (2004) for more details. top left panel of figure 9.12, that is, returns are relatively constant in 17. Some examples of answers in each category are: (1) Work the transitions between education levels. related: need to work, economic difficulties, and help at home; (2) 27. Card (1999). Other, somewhat dated, studies for Brazil (Lam low benefits: not interested, low grades, and too old; (3) limited and Schoeni 1993), Panama (Heckman and Hotz 1986), and Peru access: remote school, difficult to get to, and lack of slots; (4) other: (Behrman and Wolfe 1984) report higher upward biases in education sickness, pregnancy/maternity, military service, and miscellaneous. returns after purging the effects of parental education and other fam- In Colombia, insecurity includes those reporting they stay home ily variables on earnings and educational attainment. Their findings because of insecure streets and being displaced. might suggest this effect may depend on the stage of educational 18. The private sector accounts for more than half of the univer- development of the country. Another issue is that controls for sity market in Brazil and Colombia, close to 40 percent in Chile and variables highly correlated with own schooling such as parental 200 B R E A K I N G T H E C Y C L E O F U N D E R I N V E S T M E N T I N H U M A N C A P I TA L I N L AT I N A M E R I C A education may exacerbate a downward bias in the estimated returns where all working members (ages 26 to 65) fall in the conditional to education when people misreport their education. individual earnings distribution (that is, their ranking in unobserved 28. Cox Edwards and Ureta (2003) first applied these methods to earnings determinants). study school transitions in El Salvador; Raymond and Sadoulet 30. Herrán and Van Uythem (2001) show that students who drop (2003) recently used it to study impacts of the Mexican Oportunidades out often belong to families where the parents have no more than a program. primary education, while parents of those staying at school have 29. The effects are small given the little range of variation in completed more than nine years of education. imputed returns. An average return to each education level is 31. World Bank (2005b). imputed to each family using the education returns at the percentile 201 Bibliography Acemoglu, D. 1997. "Training and Innovation in an Imperfect Labor Alesina, A., and D. Rodrick. 1994. "Distributive Politics and Eco- Market." Review of Economic Studies 64 (3): 445­64. nomic Growth." Quarterly Journal of Economics 109: 465­90. Acemoglu, D., S. Johnson, and J. Robinson. 2001. "The Colonial Altimir, O. 1987. "Income Distribution Statistics in Latin America Origins of Comparative Development: An Empirical Investiga- and their Reliability." Review of Income and Wealth 33 (2): 111­55. tion." American Economic Review 91: 1369­1401. Altonji, J., and T. Dunn. 1996. "The Effects of Family Characteris- ------. 2002. "Reversal of Fortune: Geography and Institutions tics on the Return to Education." Review of Economics and Statistics in the Making of the Modern World Income Distribution." The 78 (40): 692­704. Quarterly Journal of Economics 117 (4): 1231­94. Andersen, L. 2001. "Social Mobility in Latin America: Links with Acemoglu, D., and J. Robinson. 2002. "The Political Economy of Adolescent Schooling." Washington, DC: Inter-American Devel- Kuznets Curve." Review of Development Economics 6 (2): 183­203. opment Bank. Adelman, I., and C. Morris. 1973. "Economic Growth and Social Equity Anderson, G., I. Crawford, and A. Liecester. 2005. "Statistical Tests in Developing Countries." Stanford, CA: Stanford University Press. for Multidimensional Poverty Analysis." Paper presented at the Agence Française de Développement, Bundesministerium für Wirt- International Conference on the Many Dimensions of Poverty, schaftliche Zusammenarbeit und Entwicklung, U.K. Department Brasilia, August 29­31. for International Development, and World Bank. 2005. Pro-Poor Andrade, E., F. Veloso, R. Madalozzo, and S. Ferreira. 2003. "Do Growth in the 1990s: Lessons and Insights from 14 Countries. Opera- Borrowing Constraints Decrease Intergenerational Mobility in tionalizing Pro-Poor Growth Research Program. Washington, Brazil? A Test Using Quantile Regression." IBMEC Working DC: World Bank. Paper 2003-3, Brazil. Aghion, P., E. Caroli, and C. García-Peñalosa. 1999. "Inequality and Anselin, L. 1988. Spatial Econometrics: Methods and Models. Dordrecht, Economic Growth: The Perspective of the New Growth Theo- Netherlands: Kluwer Academic Publishers. ries." Journal of Economic Literature 37: 1615­60. ------. 1993. "The Moran Scatterplot as an ESDA Tool to Assess Aguirre, A., and C. Calderón. 2005. "Real Exchange Rate Misalign- Local Instability in Spatial Association." Paper presented at the ment and Economic Performance." Working Paper 315, Central GISDATA Specialist Meeting on GIS and Spatial Analysis, Bank of Chile, Santiago, April. Amsterdam, December. Ahluwalia, M. 1976. "Income Distribution and Development: Some ------. 1995. "Local Indicators of Spatial Association: LISA," Stylized Facts." American Economic Review 66: 128­35. Geographical Analysis 27: 93­115. Aitchison, J., and J. Brown. 1966. "The Lognormal Distribution." ------. 2003. "Spatial Externalities, Spatial Multipliers and Spatial Cambridge: Cambridge University Press. Econometrics." International Regional Science Review 26: 153­66. Akerlof, G., and R. Kranton. 2002. "Identity and Schooling: Some Antman, F., and D. McKenzie. 2005. "Poverty Traps and Nonlinear Lessons for the Economics of Education." Journal of Economic Income Dynamics with Measurement Error and Individual Literature 40 (4): 1167­1201. Heterogeneity." Unpublished paper, Stanford University and Albernaz, A., F. Ferreira, and C. Franco. 2002. "Qualidade e equidade World Bank. no ensino fundamental brasileiro." Pesquisa e Planeamiento Economico Arabsheibani, G., F. Carneiro, and A. Henley. 2002. "Human Capital 32 (3): 453­76. and Earning Inequality in Brazil 1988­1998: Quantile Regres- Alesina, A., A. Devleeschauwer, W. Easterly, and S. Kurlat. 2003. sion Evidence." Policy Research Working Paper 3147, World "Fractionalization." Journal of Economic Growth 8 (2): 155­94. Bank, Washington, DC. Alesina, A., and D. Perotti. 1996. "Income Distribution, Political Arellano, M., and O. Bover. 1995. "Another Look at the Instrumen- Instability, and Investment." European Economic Review 40 (6): tal Variable Estimation of Error-Components Models." Journal of 1203­28. Econometrics 68: 29­52. 203 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S Arias, O. 2004. "Poverty in El Salvador during the 1990s: Evolution Baldacci, E., L. De Mello Jr., and M. Inchauste Comboni. 2002. and Characteristics." Background chapter for the 2004 World "Financial Crises, Poverty, and Income Distribution." IMF Work- Bank Poverty Assessment of El Salvador, Washington, DC. ing Paper 02/4, International Monetary Fund, Washington, DC. Arias, O., and J. Diaz. 2004. "Earnings Differentials in Latin Baltagi, B. 2001. A Companion to Theoretical Econometrics. Oxford, America and the Caribbean: Determinants, Trends and Links to U.K.: Blackwell. Poverty and Inequality." Background paper for the World Bank Banerjee, A., and E. Duflo. 2003. "Inequality and Growth: What Labor Stocktaking in the Latin American and Caribbean Region, Can the Data Say?" Journal of Economic Growth 8: 267­99. Washington, DC. Banerjee, A., and A. Newman. 1993. "Occupational Choice and the Arias, O., and W. Sosa-Escudero. 2004. "Subjective and Objective Process of Development." Journal of Political Economy 101: 274­98. Poverty in Bolivia." Background paper for the 2005 World Bank ------. 1994. "Poverty, Incentives, and Development." American Bolivia Poverty Assessment, Washington, DC. Economic Review 84 (2): 211­15. Arias, O., W. Sosa-Escudero, and K. Hallock. 2001. "Individual Het- Barrett, C., and B. Swallow. 2005. "Fractal Poverty Traps." SAGA erogeneity in the Returns to Schooling: Instrumental Variables Working Paper, Cornell University, Ithaca, NY. Quantile Regression Using Twins Data." Empirical Economics Barrett, C., M. Carter, and P. Little. Forthcoming. "Understanding 26 (1): 7­40. and Reducing Persistent Poverty in Africa." Journal of Development Arias, O., G. Yamada, and L. Tejerina. 2004. "Education, Family Studies. Background, and Racial Earnings Inequality in Brazil." Interna- Barro, R. 1991. "Economic Growth in a Cross-Section of Countries." tional Journal of Manpower 25 (3­4): 355­74. Quarterly Journal of Economics 106: 407­43. Aroca, P. 2005a. "The Effect of Interstate Migration on Wage ------. 2000. "Inequality and Growth in a Panel of Countries." Inequality among Mexican States." Unpublished paper, Universidad Journal of Economic Growth 5: 5­32. Catolica del Norte, Chile. ------. 2001. "Human Capital and Growth." American Economic ------. 2005b. "Regional Labor Markets, Migration and Commut- Review 91: 12­17. ing." Unpublished paper, Universidad Catolica del Norte, Chile. Barro, R., and J. Lee. 2001. "Schooling Quality in a Cross-Section of Aroca, P., and M. Bosch. 2000. "Crecimiento, Convergencia y Countries." Economica 68 (272): 465­88. Espacio en las Regiones Chilenas: 1960­1998." Estudios de Barro, R., and X. Sala-i-Martin. 1995. Economic Growth. New York: Economía 27 (2): 199­224. McGraw Hill. Aroca, P., M. Bosch, and W. Maloney. 2005. "Spatial Dimensions of Basu, K. 1997. Analytical Development Economics. Cambridge, MA: Trade Liberalization and Economic Divergence: Mexico 1985­ MIT Press. 2002." World Bank, Washington, DC. Basu, S. 1999. "Procyclical Productivity: Increasing Returns or Aroca, P., and G. Hewings. 2002. "Labor Market Adjustment and Cyclical Utilization?" Working Paper 5336, National Bureau of Regional Growth in Chile, 1982­1992." Annals of Regional Science Economic Research, Cambridge, MA. 36: 197­218. Beck, M. 2003. "Spatial Growth Multiplier in Mexico." Unpub- Aroca, P., and W. Maloney. Forthcoming. "Migration, Trade, and For- lished paper, World Bank, Washington, DC. eign Direct Investment in Mexico." World Bank Economic Review. Beck, T., A. Demirguc-Kunt, and R. Levine. 2004. "Finance, Atkinson, A. 2003. "Top Incomes in the United Kingdom over the Inequality, and Poverty: Cross-Country Evidence." NBER Working Twentieth Century." Nuffield College, University of Oxford, U.K. Paper 10979, National Bureau of Economic Research, ------. 2004. "Increased Income Inequality in OECD Countries Cambridge, MA. and the Redistributive Impact of the Government Budget." In Beck, T., R. Levine, and N. Loayza. 2000. "Finance and Sources of Inequality, Growth and Poverty in an Era of Liberalization and Growth." Journal of Financial Economics, 261­300. Globalization, ed. G. Cornia. Oxford: Oxford University Press. Becker, G. 1967. "Human Capital and the Personal Distribution of Atkinson, A., and F. Bourguignon. 1982. "The Comparison of Income: An Analytical Approach." Woytinsky Lecture 1. Ann Multi-Dimensional Distributions of Economic Status." In Social Arbor: University of Michigan, Institute of Public Administration. Justice and Public Policy. London: Harvester Wheatsheaf. ------. 1975. "Human Capital: A Theoretical and Empirical Atkinson, A., F. Bourguignon, and C. Morrison. 1992. Empirical Analysis." New York: Columbia University Press for the National Studies of Earnings Mobility. Chur, Switzerland: Harwood Academic Bureau of Economic Research. Publishers. Becker, G., T. Philipson, and R. Soares. 2005. "The Quantity and Attanasio, O., and M. Székely. 2002. "Wage Shocks and Consump- Quality of Life and the Evolution of World Inequality." American tion Variability in Mexico during the 1990s." RES Working Economic Review 95 (1): 277­91. Paper Series 451, Research Department, Inter-American Devel- Becker, G., and N. Tomes. 1979. "An Equilibrium Theory of the opment Bank, Washington, DC. Distribution of Income and Intergenerational Mobility." Journal Azariadis, C., and A. Drazen. 1990. "Threshold Externalities in Eco- of Political Economy 6: 1153­89. nomic Development." Quarterly Journal of Economics 105 (2): 501­26. Behrens, K., and C. Gaigne. 2003. "Interregional and International Azariadis, C., and J. Stachurski. 2005. "Poverty Traps." In Handbook Trade: Seventy Years after Ohlin." CEPR Discussion Paper 4065, of Economic Growth, ed. P. Aghion and S. Durlauf. North Holland. Center for Economic Policy Research, London. 204 B I B L I O G R A P H Y Behrman, J., N. Birdsall, and M. Székely. 1999. "Intergenerational Birdsall, N., and C. Graham. 1998. "New Ways of Looking at Old Mobility in Latin America: Deeper Markets and Better Schools Inequities: Market Reforms, Social Mobility, and Sustainable Make a Difference." In New Markets, New Opportunities? ed. Growth (Latin America in Comparative Context)." Working N. Birdsall and C. Graham. Washington, DC: Brookings Institu- paper, Brookings Institution and the Inter-American Develop- tion Press. ment Bank, Washington, DC. ------. 2003. "Economic Policy and Wage Differentials in Latin Bloom, D. E., D. Canning, and J. Sevilla. 2003. "Geography and America." Working Paper 29, Center for Global Development, Poverty Traps." Journal of Economic Growth 8: 355­78. Washington, DC. Blundell, R., L. Dearden, and B. Sianesi. 2005. "Evaluating the Effect Behrman, J., S. Duryea, and M. Székely. 1999. "Schooling Invest- of Education on Earnings: Models, Methods and Results from the ments and Macroeconomic Conditions: A Micro-Macro Investiga- National Child Development Survey." Journal of the Royal Statisti- tion for Latin America and the Caribbean." Working Paper 407, cal Society. Series A (Statistics in Society) 168 (3): 473­512. Inter-American Development Bank, Washington, DC. Bonfiglioli, A. 2004. "Equities and Inequality." Unpublished paper, Behrman, J., A. Gaviria, and M. Székely. 2001. "Intergenerational HES, Stockholm University, Stockholm. Mobility in Latin America." Inter-American Development Bank, Borensztein, E., J. De Gregorio, and J.-W. Lee. 1998. "How Does Washington, DC. Foreign Direct Investment Affect Economic Growth?" Journal of Behrman, J., R. Pollak. and P. Taubman. 1989. "Family Resources, International Economics 45 (1): 115­35. Family Size, and Access to Financing for College Education." Borjas, G. 1992. "The Intergenerational Mobility of Immigrants." Journal of Political Economy 97 (2): 398­419. NBER Working Paper 3972, National Bureau of Economic Behrman, J., and B. Wolfe. 1984. "The Socioeconomic Impact Research, Cambridge, MA. of Schooling in a Developing Country." Review of Economics and Bosch, M., P. Aroca, I. Fernandez, and C. Azzoni. 2003. "Growth Statistics 66 (2): 296­303. Dynamics and Space in Brazil." International Regional Science Bénabou, R. 1994. "Human Capital, Inequality, and Growth: A Review 26 (3): 393­418. Local Perspective." European Economic Review 38: 817­26. Bosch, M., and B. Cobacho. 2005. "Effects of Federal Public Invest- Benabou, R., and E. Ok. 2001. "Mobility as Progressivity: Ranking ment in Mexico, 1970­2000." Unpublished paper, World Bank, Income Processes According to Equality and Opportunity." Washington, DC, and Universidad Politecnica de Cartagena, NBER Working Paper 8431, National Bureau of Economic Mexico. Research, Cambridge, MA. Bosch, M., and W. Maloney. 2005. "Labor Market Dynamics in Ben-David, D. 1993. "Equalizing Exchange: Trade Liberalization Developing Countries: Comparative Analysis Using Continuous and Income Convergence." Quarterly Journal of Economics 108: Time Markov Processes." World Bank, Washington, DC. 653­79. Bottazi, L., and G. Peri. 2003. "Innovation and Spillovers in Regions: ------. 1995. "Convergence Clubs and Diverging Economies." Evidence from European Patent Data." European Economic Review Working Paper 40/95, Foerder Institute for Economic Research, 47: 687­710. Tel Aviv University, Tel Aviv. Boucekkine, R., D. de la Croix, and O. Licandro. 2003. "Early Beneke de Sanfeliu, M., and M. Shi. 2004. "Dinámica del Ingreso en Mortality Declines at the Dawn of Modern Growth." Scandinavian El Salvador." Serie de Investigación 2. Antiguo Cuscatlán: Depar- Journal of Economics 105: 401­18. tamento de Estudios Económicos y Sociales (DEES)/FUSADES. Bougheas, S., P. Demetriades, and T. Mamuneas. 2000. "Infrastruc- Bera, A., W. Sosa-Escudero, and M. Yoon. 2001. "Tests for the Error ture, Specialization, and Economic Growth." Canadian Journal of Component Model in the Presence of Local Misspecification." Economics 33: 506­22. Journal of Econometrics 101 (1): 1­23. Bourguignon, F. 2001. "Crime As a Social Cost of Poverty and Bernstein, J., and D. Weinstein. 2002. "Do Endowments Predict the Inequality: A Review Focusing on Developing Countries." In Location of Production? Evidence from National and Interna- Facets of Globalization International and Local Dimensions of Develop- tional Data." Journal of International Economics 56: 55­76. ment, ed. S. Yusuf, S. Evenett, and W. Wui, 171­92. Discussion Bértola, L. 1998. "El PBI de Uruguay 1870­1936 y otras estima- paper series WDP 415, World Bank, Washington, DC. ciones." Documento de Trabajo 43, Unidad Multidisciplinaria, ------. 2003. "The Growth Elasticity of Poverty Reduction; Facultad de Ciencias Sociales, Universidad de la República, Explaining Heterogeneity across Countries and Time Periods." In Montevideo. Inequality and Growth: Theory and Policy Implications, ed. T. Eicher ------. 2005. "A 50 años de la curva de Kuznets, una reivindicación and S. Turnovsky. Cambridge, MA: MIT Press. sustantiva: distribución del ingreso y crecimiento económico en ------. 2004. "The Poverty-Growth-Inequality Triangle." Unpub- Uruguay y otros países de nuevo asentamiento desde 1870." Insti- lished paper, World Bank, Washington, DC. tuto Laureano Figuerola de Historia Económica, Working Paper Bourguignon, F., and S. Chakravarty. 2003. "The Measurement of Series No. 05-04, Universidad Carlos III de Madrid. Multi-Dimensional Inequality." Journal of Economic Inequality 1 (1): Bértola, L., and J. G. Williamson. 2003. "Globalization in Latin 25­49. America before 1940." NBER Working Paper 9687, National Bourguignon, F., F. Ferreira, and P. Leite. 2002. "Beyond Oaxaca- Bureau of Economic Research, Cambridge, MA. Blinder: Accounting for Differences in Household Income 205 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S Distributions across Countries." Policy Research Working Paper ------. 2001. "Estimating the Return to Schooling: Progress 2828, World Bank, Washington, DC. on Some Persistent Econometric Problems." Econometrica 69 (5): Bourguignon, F., F. Ferreira, and N. Lustig. 2005. The Microeconomics 1127­60. of Income Distribution Dynamics in East Asia and Latin America. Carneiro, F., and A. Henley. 2002. "Long Run Determinants and Washington, DC: World Bank. Short Term Dynamics of Aggregate Wages in Brazil." Discussion Bourguignon, F., F. Ferreira, and M. Menendez. 2003. "Inequality of Paper 9610, Department of Economics, University of Wales, Outcomes and Inequality of Opportunities in Brazil." Texto para Aberystwyth. discussao 48, Departmento de Economia, Pontifícia Universidade Carneiro, P. 2003. "Evaluating Education Policies When the Rate of Católica do Rio de Janeiro (PUC-Rio). Return Varies Across Individuals." Cuadernos de Economia 40 (121): Bourguignon, F., and C. Morrisson. 2002. "Inequality among World 516­29. Citizens." American Economic Review 92 (4): 727­44. Carneiro, P., K. Hansen, and J. Heckman. 2001. "Educational Attain- Bowles, S., S. Durlauf, and K. Hoff. 2004. Poverty Traps. Princeton, ment and Labor Market Outcomes: Estimating Distributions of NJ: Princeton University Press. the Returns to Educational Interventions." Paper presented as Bravo-Ortega, C., and D. Lederman. 2005. "Agricultural and Klein Lecture, University of Pennsylvania, Philadelphia. National Welfare around the World: Causality and International ------. 2003. "Estimating Distributions of Treatment Effects with Heterogeneity since 1960." Policy Research Working Paper an Application to the Returns to Schooling and Measurement of 3499, World Bank, Washington, DC. the Effects of Uncertainty on College." Working Paper 2003: 9, Brock, W., and S. Durlauf. 2001. "Discrete Choice with Social Inter- Institute for Labour Market Policy Evaluation (IFAU). Uppsala, actions." Review of Economic Studies 68 (2): 235­60. Sweden. Browning, M., L. Hansen, and J. Heckman. 1999. "Micro Data and Carneiro, P., and J. Heckman. 2002. "The Evidence on Credit General Equilibrium Models." In Handbook of Macroeconomics, Constraints in Post-Secondary Schooling." Economic Journal 112: ed. J. Taylor and M. Woodford. Vol. 1, 543­636. Amsterdam, 989­1018. New York, and Oxford: Elsevier Science. ------. 2003. "Human Capital Policy." IZA Discussion Paper 821, Bruno, M., and W. Easterly. 1998. "Inflation Crises and Long-Run Institute for the Study of Labor (IZA), Bonn, Germany. Growth." Journal of Monetary Economics 41: 3­26. Carneiro, P., J. Heckman, and S. Lee. 2004. "Comparative Advantage Buchinsky, M. 1994. "Changes in the U.S. Wage Structure and Schooling." Working paper, University College, London. 1963­1987: An Application of Quantile Regression." Econometrica Carneiro, P., J. Heckman, and E. Vytlacil. 2001. "Estimating the Rate 62 (2): 405­58. of Return to Education When It Varies among Individuals." Paper Bulir, A. 2001. "Income Inequality: Does Inflation Matter?" IMF presented as the Economic Journal Lecture at the Royal Economic Staff Paper 48, 139­59, International Monetary Fund, Washing- Society meeting, Durham, England, and as the Review of Econom- ton, DC. ics and Statistics Lecture, Harvard University, Cambridge, MA. Bustelo, M. 2005. "Tracing Out the Changes in Inequality and Carter, M. R., and C. B. Barrett. 2005. "The Economics of Poverty Poverty in Argentina Using Micro Econometric Decompositions Traps and Persistent Poverty: An Asset-Based Approach." Work- Techniques (1992­2001)." CEDLAS Working Paper 13, available ing paper, University of Wisconsin, Madison. at www.depeco.econo.unlp.edu.ar/cedlas/. Chaudhuri, S. 2000. "Empirical Methods for Assessing Household Cabrol, M. 2002. "Los Desafíos de la Educación Secundaria. ¿Qué Vulnerability to Poverty." School of International and Public nos Dice el Análisis de Flujos?" Economic and Sectorial Affairs, Columbia University, New York. Studies Series RE-02-004, Inter-American Development Bank, Chaudhuri, S., J. Jalan, and A. Suryahadi. 2002. "Assessing House- Washington, DC. hold Vulnerability to Poverty: A Methodology and Estimates for Calderón, C., and R. Fuentes. 2005. "Cuanto Explican las Reformas y Indonesia." Department of Economics Discussion Paper 0102-52, la Calidad de las Instituciones el Crecimiento Chileno? Una Columbia University, New York. Comparación Internacional." Working Paper 314, Central Bank Checchi, D., and V. Dardanoni. 2002. "Mobility Comparisons: Does of Chile, Santiago. Using Different Measures Matter?" Working Paper 15.2002. Calderón, C., and L. Servén. 2004. "The Effects of Infrastructure Dipartimento di Economia Politica e Aziendale, Università degli Development on Growth and Income Distribution." Policy Studi di Milano. Research Working Paper 3400, World Bank, Washington, DC. Chen, S., and M. Ravallion. 1997. "What Can New Survey Data Tell Calmon, P. 2003. "Evaluation of Subsidies in Brazil, and Overview." Us about Recent Changes in Distribution and Poverty?" World Unpublished paper, Universidad de Brasilia, Barasdilia. Bank Economic Review 11 (2): 357­82. Calvo, E., J. Torre, and M. Szwarcberg. 2002. The New Welfare ------. 2001. "How Did the World's Poorest Fare in the 1990s?" Alliance. Buenos Aires: Universidad di Tella, Department of World Bank, Washington, DC. Political Science. Chiquiar, C. Forthcoming. "Why Mexico's Regional Income Conver- Card, D. 1999. "The Casual Effect of Education on Earnings." Hand- gence Broke Down." Journal of Development Economics. book of Labor Economics. Vol. 3, 1801­63. Amsterdam: Elsevier Chomitz, K. 2005. "Development Policy for a Heterogeneous Science, North-Holland. Space." In Estratégias para o Nordeste e a Amazônia, coord. João 206 B I B L I O G R A P H Y Paulo dos Reis Velloso (Cadernos Fórum Nacional 2). Instituto Dasgupta, R., and D. Ray. 1986. "Inequality as a Determinant of Nacional de Altos Estudos, Rio de Janeiro. Malnutrition and Unemployment: Theory." Economic Journal 96 Chong, A., and C. Calderón. 2000. "Institutional Quality and (384): 1011­34. Income Distribution." Economic Development and Cultural Change Davis, D., D. Weinstein, S. Bradford, and K. Shimpo. 1997. "Using 48: 761­86. International and Japanese Regional Data to Determine When Clarke, G., L. Xu, and H. Zou. 2003. "Finance and Income Inequality: the Factor Abundance Theory of Trade Works." American Economic Test of Alternative Theories." Policy Research Working Paper Review 87 (3): 421­46. 2984, World Bank, Washington, DC. Deaton, A. 1985. "Panel Data from Time Series of Cross-Sections." Coatsworth, J. 1998. "Economic and Institutional Trajectories in Journal of Econometrics 30:109­26. Nineteenth-Century Latin America." In Latin America and the ------. 1997. "The Analysis of Households Surveys: A Micro- World Economy Since 1800, ed. J. H. Coatsworth and A. M. Taylor, economic Approach to Development Policy." Washington, DC, 23­54. Cambridge, MA: Harvard University Press. World Bank. Coe, D., and E. Helpman. 1995. "International R&D Spillovers." ------. 2005. "Measuring Poverty in a Growing World (or Measur- European Economic Review 39: 859­87. ing Growth in a Poor World)." Review of Economics and Statistics Collins, S., and O. Razin. 1999. "Real-Exchange-Rate Misalignments 87 (1, February): 1­25. and Growth." In The Economics of Globalization: Policy Perspectives Deaton, A., and C. Paxton. 1994. "Intertemporal Choice and from Public Economics, 59­81. Cambridge: Cambridge University Inequality." Journal of Political Economy 102 (3): 437­67. Press. ------. 2001. "Mortality, Income and Income Inequality over Time Contreras, D., R. Cooper, and J. Heman. 2004. "Dinámica de la in Britain and the United States." NBER Working Paper 8534, Pobreza y Movilidad Social: Chile 1996­2001." University of National Bureau of Economic Research, Cambridge, MA. Chile. de Ferranti, D., G. Perry, F. Ferreira, and M. Walton. 2004. Inequality Cord, L., H. Lopez, and J. Page. 2005. "`When I Use a Word' Pro- in Latin America and the Caribbean: Breaking with History? Poor Growth and Poverty Reduction." In Attacking Poverty: What Washington, DC: World Bank. Makes Growth Pro-Poor? ed. M. Krakowski. Hamburg: Nomos de Ferranti, D., G. Perry, I. Gill, J. Guasch, W. Maloney, C. Sánchez- Verlagsgesellschaft. Páramo, and N. Schady. 2003. Closing the Gap in Education and Cortés Conde, R. 1997. La economía argentina en el largo plazo (Siglos Technology. Washington, DC: World Bank. XIX y XX). Buenos Aires: Editorial Sudamericana-Universidad de Ferranti, D., G. Perry, I. Gill, and L. Servén, with F. Ferreira, de San Andrés. N. Ilahi, W. Maloney, and M. Rama. 2000. Securing Our Future in Cortés Conde, R., with M. Harriague. 1994. "El PBI argentino, a Global Economy. Washington, DC: World Bank. 1875­1935." Unpublished paper. Universidad de San Andrés, de Ferranti, D., G. Perry, D. Lederman, and W. Maloney. 2002. From Buenos Aires. Natural Resources to the Knowledge Economy. Washington, DC: Cox Edwards, A., S. Duryea. and M. Ureta. 2001. "Women in the World Bank. LAC Labor Market: The Remarkable 1990's." Working Paper de Ferranti, D., G. Perry, D. Lederman, A. Valdes, and W. Foster. Series 500, William Davidson Institute, University of Michigan, 2005. Beyond the City: The Rural Contribution to Development. Ann Arbor, Michigan. Washington, DC: World Bank. Cox Edwards, A., and M. Ureta. 2003. "International Migration, De Gregorio, J., and J. Lee. 2002. "Education and Income Inequality: Remittances, and Schooling: Evidence from El Salvador." Journal New Evidence from Cross-Country Data." Review of Income and of Development Economics 72 (2): 429­61. Wealth 48: 395­416. Cunha, F., J. Heckman, and S. Navarro. 2005. "Separating Uncertainty Dehejia, R., and R. Gatti. 2005. "Child Labor: The Role of Financial from Heterogeneity in Life Cycle Earnings." Oxford Economic Papers Development and Income Variability across Countries." Economic 57: 191­261. Development and Cultural Change 53: 913­32. Cunningham, W. 2001. "Breadwinner Versus Caregiver: Labor Force Deininger, K., and L. Squire. 1996. "A New Data Set Measuring Participation and Sectoral Choice Over the Mexican Business Income Inequality." World Bank Economic Review 10: 565­91. Cycle." In The Economics of Gender in Mexico, ed. E. Katz and De la Fuente, A. 2002. "The Effect of Structural Fund Spending on M. Correia. Washington, DC: World Bank. the Spanish Regions: An Assessment of the 1994­99 Objective Dahan, M., and A. Gaviria. 1999. "Sibling Correlations and Social 1 CSF." UFAE and IAE Working Papers 538.02, Unitat de Mobility in Latin America." Inter-American Development Bank, Fonaments de l'Anàlisi Econòmica (UAB) and Institut d'Anàlisi Washington, DC. Econòmica (CSIC). Darity, W., and P. Mason. 1998. "Evidence on Discrimination in ------. 2003. "El impacto de los Fondos Estructurales: Convergencia Employment: Codes of Color, Codes of Gender." Journal of real y cohesión interna." UFAE and IAE Working Papers 572.03, Economic Perspectives 12: 63­90. Unitat de Fonaments de l'Anàlisi Econòmica (UAB) and Institut Das, M., and S. Mohapatra. 2003. "Income Inequality: The Aftermath d'Anàlisi Econòmica (CSIC), Barcelona, Spain. of Stock Market Liberalization in Emerging Markets." Journal of Della Paolera, G., A. M. Taylor, and G. Bózolli. 2003. "Historical Empirical Finance 10: 217­48. Statistics." In A New Economic History of Argentina, ed. G. Della 207 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S Paolera and A. M. Taylor, 376­85, plus a CD-ROM. New York: Easterly, W., and L. Servén. 2003. The Limits of Stabilization: Infra- Cambridge University Press. structure, Public Deficits, and Growth in Latin America. Washington, Dell'Ariccia, G., E. Detragiache, and R. Rajan. 2005. "The Real DC: World Bank. Effect of Banking Crises." IMF Working Paper WP/05/63, Inter- Edwards, S. 1997. "Trade Policy, Growth, and Income Distribution." national Monetary Fund, Washington, DC. American Economic Review 87: 205­10. Deutsch, J., and J. Silber. 2005. "The Order of Acquisition of Durable ------. 1998. "Openness, Productivity and Growth: What Do We Goods and the Multidimensional Measurement of Poverty." Paper Really Know?" Economic Journal 108: 383­98. presented at the International Conference on "The Many Dimen- Egset, W., and P. Sletten. 2004. "La pauvreté en Haïti. Un profil de la sions of Poverty," Brasilia, August 29­31. pauvreté en Haïti à partir des données de l'enquête ECVH." Fafo De Vreyer, P., and G. Spielvogel. 2005. "Spatial Externalities between Institute for Applied International Studies, Oslo. Brazilian Municipios and their Neighbors." Working paper Elbers, C., T. Fujii, P. Lanjouw, B. Ozler, and W. Yin. 2004. "How 2005/11, Développement Institutions et Analyses de Long Terme Much Does Disaggregation Help?" World Bank, Washington, DC. (DIAL), University of Lille, France. Ellison, G., and E. Glaeser. 1999. "The Determinants of Geographic Díaz, J., R. Lüders, and G. Wagner. 1998. "Economía chilena 1810­ Concentration." American Economic Review 89 (2): 311­16. 1995: evolución cuantitativa del producto total y sectorial." Engerman, S., and K. Sokoloff. 2004. "Factor Endowments, Institu- Documento de Trabajo 186, Pontificia Universidad Católica de tions, and Differential Paths of Growth among New World Chile, Santiago. Economies: A View from Economic Historians of the United Dollar, D. 1992. "Outward-Oriented Developing Economies Really States." NBER Working Paper H0066, National Bureau of Do Grow More Rapidly: Evidence from 95 LDCs, 1976­85." Economic Research, Cambridge, MA. Economic Development and Cultural Change 40: 523­44. ------. Forthcoming. "The Persistence of Poverty in the Americas: Dollar, D., and A. Kraay. 2002. "Growth Is Good for the Poor." The Role of Institutions." In Poverty Traps, ed. S. Bowless, Journal of Economic Growth 7: 195­225. S. Durlauf, and K. Hoff. ------. 2003. "Institutions, Trade, and Growth." Journal of Monetary Engle, E., A. Galetoviv, and C. Raddatz. 1998. "Taxation and Economics 50: 133­62. Income Distribution in Chile: Some Unpleasant Redistributive ------. 2004. "Trade, Growth, and Poverty." Economic Journal 114 Arithmetic." NBER Working Paper 6828. National Bureau of (493): 22­49. Economic Research, Cambridge, MA. Duarte, A., P. Ferreira, and M. Salvato. 2003. "Regional or Educa- Escobal, J., and M. Torero. 2005. "Adverse Geography and Differ- tional Disparities? A Counterfactural Exercise." Ensayos Econom- ences in Welfare in Peru." In Spatial Inequality and Development, icos 532, Fundacao Getulio Vargas, Rio de Janeiro. ed. R. Kanbur and A. Venables. Oxford: Oxford University Press. Duclos, J., D. Sahn, and S. Younger. 2005. "Robust Multidimensional Esfahani, H., and M. Ramírez. 2003. "Institutions, Infrastructure and Poverty Comparisons." Paper presented at the International Economic Growth." Journal of Development Economics 70: 443­77. Conference on "The Many Dimensions of Poverty," Brasilia, Esquivel, G. 2000. "Geografia y Desarrollo Eeconomico en Mexico." August 29­31. Latin American Research Network Working Paper 389, Dunn, C. 2003. "Assortative Matching and Intergenerational Research Department, Inter-American Development Bank, Mobility in Family Earnings: Evidence from Brazil." Unpub- Washington, DC. lished paper, University of Michigan, Ann Arbor, Michigan. Estache, A., and M. Fay. 1995. "Regional Growth in Argentina and Durlauf, S. N. 1996. "A Theory of Persistent Income Inequality." Brazil: Determinants and Policy Options." Unpublished paper, Journal of Economic Growth 1: 75­93. World Bank, Washington, DC. Durlauf, S., and D. Quah. 1999. "The New Empirics of Economic Euromod. 2004. http://www.econ.cam.ac.uk/dae/mu/emod.htm. Growth." In Handbook of Macroeconomics, ed. J. Taylor and Fajnzylber, P., and W. Maloney. 2005. "Labor Demand and Trade M. Woodford. Amsterdam: North-Holland. Reform in Latin America." Journal of International Economics 66: Easterly, W. 1999. "Life during Growth." Journal of Economic Growth 423­46. 4: 239­76. Fajnzylber, P., W. Maloney, and G. Montes. 2005. "LDC Micro-Firm ------. 2001. "The Lost Decades: Developing Countries' Stagna- Dynamics: How Similar Are They to Those in the Industrialized tion in Spite of Policy Reform 1980­1998." Journal of Economic World? Evidence from Mexico." Unpublished paper, World Growth 6 (2): 135­57. Bank, Washington, DC. Easterly, W., and S. Fischer. 2001. "Inflation and the Poor." Journal of Ferreira, S., and F. Veloso. 2003. "Intergenerational Mobility of Money, Credit and Banking 33:160­78. Earnings in Brazil." Brazilian Development Bank (BNDES) and Easterly, W., and R. Levine. 1997. "Africa's Growth Tragedy: Policies Universidade Candido Mendes (UCAM), Brazil. and Ethnic Divisions." Quarterly Journal of Economics 111 (4): Fields, G. 2004. "A Guide to Multisector Labor Markets." Paper pre- 1203­50. pared for the World Bank Labor Market Conference, November Easterly, W., and S. Rebelo. 1993. "Fiscal Policy and Economic 18­19, Washington, DC. Growth: An Empirical Investigation." Journal of Monetary Economics Fields, G., R. Duval, S. Freije, and M. Puerta. 2005. "Earnings Mobil- 32: 417­58. ity in Argentina, Mexico, and Venezuela: Testing the Divergence 208 B I B L I O G R A P H Y of Earnings and the Symmetry of Mobility Hypotheses." Unpub- econometric Approach." Background paper for the 2005 World lished paper, Cornell University, Ithaca, NY. Bank Bolivia Poverty Assessment, Washington, DC. Fields, G., and E. Ok. 1996. "The Meaning and Measurement of Gasparini, L., F. Gutierrez, and L. Tornarolli. 2005. "Growth and Income Mobility." Journal of Economic Theory. Income Poverty in Latin America and the Caribbean: Evidence Fischer, S. 1993. "The Role of Macroeconomic Factors in Growth." from Household Surveys." World Bank, Washington, DC. Journal of Monetary Economics 32: 485­511. Gertler, P., S. Martinez, and M. Rubio. 2005. "Investing Cash Trans- Fischer, S., R. Sahay, and C. Végh. 2002. "Modern Hyper- and High fers to Raise Long Term Living Standards." Unpublished paper, Inflations." Journal of Economic Literature 60: 837­80. University of California at Berkeley. Flug, K., A. Spilimbergo, and E. Wachtenheim. 1998. "Investment Getis, A., and K. Ord. 1992. "The Analysis of Spatial Association by in Education: Do Economic Volatility and Credit Constraints Use of Distance Statistics." Geographical Analysis 24: 189­206. Matter?" Journal of Development Economics 55 (2): 465­81. Geweke, J., and M. Keane. 2000. "An Empirical Analysis of Income Fogel, R. 1994. "Economic Growth, Population Theory, and Phy- Dynamics among Men in the PSID: 1968­1989." Journal of siology: The Bearing of Long-Term Processes on the Making of Econometrics 96: 293­356. Economic Policy." American Economic Review 84 (3): 369­95. Gibrat, R. 1931. Les inégalités économiques. Paris: Sirey. Forbes, K. 2000. "A Reassessment of the Relationship between Giuliano P., and M. Ruiz-Arranz. 2005. "Remittances, Financial Inequality and Growth." American Economic Review 90 (4): 869­87. Development and Growth." Working Paper, International Mone- Foster, J., J. Greer, and E. Thorbecke. 1984. "A Class of Decompos- tary Fund, Washington, DC. able Poverty Measures." Econometrica 52: 761­66. Glewwe, P. 2004. "How Much of Observed Economic Mobility Is Foster, V., and C. Araujo. 2001. "Does Infrastructure Reform Work Measurement Error? A Method to Remove Measurement Error, for the Poor? A Case Study from Guatemala." World Bank, Latin with an Application to Vietnam." Unpublished paper, World America and the Caribbean Region, Finance, Private Sector, and Bank, Washington, DC. Infrastructure Department, Washington, DC. Glewwe, P., and G. Hall. 1998. "Are Some Groups More Vulnerable Frankel, J., and D. Romer. 1999. "Does Trade Cause Growth?" to Macroeconomic Shocks than Others? Hypothesis Tests Based American Economic Review 89: 379­99. on Panel Data from Peru." Journal of Development Economics 56: Freije, S., and A. Souza. 2002. "Earnings Dynamics and Inequality 181­206. in Venezuela: 1995­1997." Working Paper 0211, Vanderbilt Goni, E., H. Lopez, and L. Servén. 2005. "Getting Real about University, Department of Economics, Nashville, TN. Inequality." Unpublished paper, World Bank, Washington, DC. Friedman, M. 1962. "Capitalism and Freedom." Princeton, NJ: Gottschalk, P., and E. Spolaore. 2002. "On the Evaluation of Eco- Princeton University. nomic Mobility." Review of Economic Studies 69 (1): 191­208. Friedman, M., and S. Kuznets. 1954. "Income from Independent Graham, B., and J. Temple. 2004. "Rich Nations, Poor Nations: How Professional Practice." National Bureau of Economic Research, Much Can Multiple Equilibria Explain?" IIIS Discussion Paper 17, New York. Institute for International Integration Studies, The University of Fujita, M., P. Krugman, and A. Venables. 1999. The Spatial Economy: Dublin, Ireland. Cities, Regions and International Trade. Cambridge, MA: MIT Press. Grawe, N. 2002. "Quantile Measures of Mobility in the US and Gabriel, S., J. Shack-Marquez, and W. Wascher. 1993. "Does Migra- Abroad." In Generational Income Mobility in North America and tion Arbitrage Regional Labor Market Differentials." Regional Europe, ed. Miles Corak. Cambridge: Cambridge University Press. Science and Urban Economics 23: 211­33. GRECO (Grupo de Estudios de Crecimiento Económico). 2002. El Galiani, S., P. Gertler, and E. Schargrodsky. 2005. "Water for Life: crecimiento económico colombiano en el siglo XX. Bogotá: Banco de la The Impact of the Privatization of Water Services on Child República-Fondo de Cultura Económica. Mortality." Journal of Political Economy 113 (1): 83­120. Greenwood, J., and B. Jovanovic. 1990. "Financial Development, Gallego, F., and N. Loayza. 2002. "The Golden Period for Growth in Growth and the Distribution of Income." Journal of Political Chile: Explanations and Forecasts." Series on Central Banking, Economy 98: 1076­1107. Analysis, and Economic Policies, 6, Central Bank of Chile, Santiago. Griffith, D. A. 1996. Practical Handbook of Spatial Statistics." Boca Gallup, J., A. Gaviria, and E. Lora. 2003. "Is Geography Destiny?: Raton, FL: CRC Press. Lessons from Latin America." Stanford University Press. Guimarães, S., and F. Veloso. 2003. "Intergenerational Mobility of Galor, O., and J. Zeira. 1993. "Income Distribution and Macroeco- Earnings in Brazil." Brazilian Development Bank (BNDES) and nomics." Review of Economic Studies 60 (1): 35­52. Instituto Brasileiro de Mercado de Capitais (IBMEC), Brazil. Gamanou, G., and J. Morduch. 2002. "Measuring Vulnerability to Gupta, S., B. Clements, E. Baldacci, and C. Mulas-Granados. 2005. Poverty." Discussion Paper 2002/58, United Nations University "Fiscal Policy, Expenditure Composition, and Growth in Low- (WIDER), Helsinki. Income Countries." Journal of International Money and Finance 24: Gannon, C., and Z. Liu. 1997. "Poverty and Transport." Unpub- 441­63. lished paper, World Bank, Washington, DC. Haber, S. H. 1997. "Introduction." In How Latin America Fell Behind. Gasparini, L., M. Cicowiez, F. Gutiérrez, and M. Marchionni. 2004. Essays on the Economic Histories of Brazil and Mexico, 1800­1914, "Simulating Income Distribution Changes in Bolivia: A Micro- ed. S. Haber. Stanford, CA: Stanford University Press. 209 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S Hall, G., and H. A. Patrinos, eds. 2004. Indigenous People, Poverty, and ------. 2002. "Geographic Poverty Traps? A Micro Model of Con- Human Development in Latin America: 1994­2004. New York: Pal- sumption Growth in Rural China." Journal of Applied Econometrics grave Macmillan. 17: 329­46. Hall, R., and C. Jones. 1999. "Why Do Some Countries Produce So Jaramillo, C., and D. Lederman, with A. Mason, D. Gould, M. Bussolo, Much More Output per Worker than Others?" Quarterly Journal of R. Tejada, and N. Fiess. 2005. DR-CAFTA: Challenges and Economics 114: 83­116. Opportunities for Central America. Washington, DC: World Hanson, G. 1997. "Increasing Returns, Trade, and the Regional Bank. Structure of Wages." Economic Journal 107: 113­33. Jiménez, W., and F. Landa. 2004. "Poverty and Indigenous Peoples in ------. 2001. "Scale Economies and the Geographic Concentration Bolivia, 1989­2002." Background paper for the report Indigenous of Industry." Journal of Economic Geography 1: 255­76. People, Poverty and Human Development 1994­2004. World Bank, Hart, P. 1981. "The Statics and Dynamics of Income Distributions: A Washington, DC. Survey." In The Statics and Dynamics of Income, ed. N. A. Klevmarken Kakwani, N., and E. Pernia. 2000. "What Is Pro-Poor Growth?" and J. A. Lybeck. Clevedon: Tieto. Asian Development Review 18: 1­16. Haveman, R., and B. Wolfe. 1995. "The Determinants of Children's Kakwani, N., and H. Son. 2003. "Pro-Poor Growth: Concept, Attainments: A Review of Methods and Findings." Journal of Measurement, and Application." Unpublished paper, University Economic Literature 33 (4): 1829­78. of New South Wales, Australia. Head, K., and T. Mayer. Forthcoming. "The Empirics of Agglomera- Kalecki, M. 1945. "On the Gibrat Distribution." Econometrica 13: tion and Trade." In Handbook of Regional and Urban Economics, vol 4. 161­70. Amsterdam: Elsevier Science, North Holand. Kaminsky, G., and C. Reinhart. 1999. "The Twin Crises: The Causes Heckman, J. 1997. "The Value of Quantitative Evidence on the of Banking and Balance of Payments Problems." American Economic Effect of the Past on the Present." American Economic Review 87 (2): Review 89: 473­500. 404­8. Kanbur, R., and A. Venables. 2005. "Spatial Inequality and Develop- ------. 2000. "Policies to Foster Human Capital." Research in ment: Overview of UNU-WIDER Project." Cornell University, Economics 54 (1): 3­56. Ithaca, NY, and University of London. Heckman, J., and J. Hotz. 1986. "An Investigation of the Labor Kaufmann, D., A. Kraay, and M. Mastruzzi. 2004. "Governance Market Earnings of Panamanian Males: Evaluating the Sources of Matters III: Governance Indicators for 1996, 1998, 2000, and Inequality." Journal of Human Resources 21(4): 507­42. 2002." World Bank Economic Review 18 (2): 253­87. Herrán, C., and B. Van Uythem. 2001. "Why Do Youngsters Drop Kim, M., and S. Polachek. 1994. "Panel Estimates of Male-Female Out of School in Argentina and What Can Be Done about It?" Earnings Functions." Journal of Human Resources 29: 406­28. Inter-American Development Bank, Washington, DC. Klinger, B., and D. Lederman. 2005. "Sectoral Concentration in the Hirschman, A. 1981. "The Changing Tolerance for Income Inequality in Search for Innovations and Export Booms." Unpublished paper, the Course of Economic Development." In Essays in Trespassing, Eco- World Bank, Washington, DC. nomics to Politics and Beyond. New York: Cambridge University Press. Knack, S., and P. Keefer. 1995. "Institutions and Economic Perfor- Hobjin, B., and D. Lagakos. 2003. "Inflation Inequality in the mance: Cross-Country Tests Using Alternative Institutional United States." Staff Report 173, Federal Reserve Bank of Measures." Economics and Politics 7: 207­27. New York, New York. Kneller, R., M. Bleaney, and N. Gemmell. 1999. "Fiscal Policy and Honohan, P. 2004. "Financial Development, Growth and Poverty: Growth: Evidence from OECD Countries." Journal of Public How Close Are the Links?" Policy Research Working Paper Economics 74 (2): 171­90. 3203, World Bank, Washington, DC. Knight, M., N. Loayza, and D. Villanueva. 1993. "Testing the IDB (Inter-American Development Bank). 2004. Good Jobs Wanted: Neoclassical Theory of Economic Growth: A Panel Data Labor Markets in Latin America. Economic and Social Progress Approach." IMF Staff Paper 40 (3), International Monetary Fund, Report. Washington, DC. Washington, DC. Iradian, G. 2005. "Inequality, Poverty, and Growth." IMF Working Knowles, S. 2005. "Inequality and Economic Growth: The Empirical Paper WP/05/28, International Monetary Fund, Washington, Relationship Reconsidered in the Light of Comparable Data." DC, February. Journal of Development Studies 41 (1): 135­59. Jacoby, H. 1994. "Borrowing Constraints and Progress through Koenker, R. 2004. "Quantile Regression for Longitudinal Data." School: Evidence from Peru." Review of Economics and Statistics Journal of Multivariate Analysis 91: 74­89. 76 (1): 151­60. Koenker, R., and G. Bassett. 1978. "Regression Quantiles." Economet- Jacoby, H., and E. Skoufias. 1997. "Risk, Financial Markets, and rica 46 (1): 33­50. Human Capital in a Developing Country." Review of Economic Kraay, A. 2004. "When Is Growth Pro-Poor? Cross-Country Evi- Studies 64 (3): 311­35. dence." IMF Working Paper 4-47, International Monetary Fund, Jalan, J., and M. Ravallion. 1999. "Income Gains to the Poor from Washington, DC. Workfare: Estimates for Argentina's Trabajar Program." Policy ------. 2005. "When Is Growth Pro-Poor? Evidence from a Panel Research Working Paper 2149, World Bank, Washington, DC. of Countries." Journal of Development Economics. 210 B I B L I O G R A P H Y Kraay, A., and C. Raddatz. 2005. "Poverty Traps, Aid, and Leipziger, D., M. Fay, Q. Wodon, and T. Yepes. 2003. "Achieving the Growth." Policy Research Working Paper 3631, World Bank, Millennium Development Goals: The Role of Infrastructure." Washington, DC. Policy Research Working Paper 3163, World Bank, Washington, Krebs, T. 2002. "Growth and Welfare Effects of Business Cycles in DC, November. Economies with Idiosyncratic Human Capital Risk." Working Lemieux, T., and K. Milligan. 2004. "Incentive Effects of Social Paper 2002-31, Brown University, Department of Economics, Assistance: A Regression Discontinuity Approach." NBER Providence, RI. Working Paper 10541, National Bureau of Economic Research, ------. 2003. "Human Capital Risk and Economic Growth" Cambridge, MA. Quarterly Journal of Economics 118 (2): 709­43. Levin, A., and L. Raut. 1997. "Complementarities between Exports Krebs, T., P. Krishna, and W. Maloney. 2004. "Trade Policy, Income and Human Capital in Economic Growth: Evidence from the Risk, and Welfare." NBER Working Paper 11255, National Semi-industrialized Countries." Economic Development and Cultural Bureau of Economic Research, Cambridge, MA. Change 46 (1): 155­74. ------. 2005a. "Income Dynamics and Welfare." Unpublished Levine, R., N. Loayza, and T. Beck. 2000. "Financial Intermediation paper, World Bank, Washington, DC. and Growth: Causality and Causes." Journal of Monetary Economics ------. 2005b. "Income Risk and Human Capital in LDCs." 46 (1): 31­77. Unpublished paper, World Bank, Washington, DC. Levine, R., and D. Renelt. 1992. "A Sensitivity Analysis of Cross-Coun- ------. 2005c. "Labor Market Risk and Human Capital Investment try Growth Regressions." American Economic Review 82 (4): 942­63. in Developing Countries." Unpublished paper, World Bank, Levy, S., and S. Van Wijnbergen. 1995. "Transition Problems in Washington, DC. Economic Reform: Agriculture in the North American Free Trade ------. 2005d. "Trade Policy, Income Risk, and Welfare." Policy Agreement." American Economic Review 85 (4): 738­54. Research Working Paper 3622, World Bank, Washington, DC. Li, H., L. Squire, and H. Zou. 1998. "Explaining International and Kremer, M. 1993. "The O-Ring Theory of Economic Development." Intertemporal Variations in Income Inequality." Economic Journal Quarterly Journal of Economics 108 (3): 551­75. 108 (446): 26­43. Krugman, P. 1991. "Increasing Returns and Economic Geography." Li, H., L. Xu, and H. Zou. 2000. "Corruption, Income Distribution, Journal of Political Economy 99 (3): 483­99. and Growth." Economics and Politics 12 (2): 155­82. Krugman, P. R. 1993a. "First Nature, Second Nature and Metropol- Li, H., and H. Zou. 1998. "Income Inequality Is Not Harmful for itan Location." Journal of Regional Science 33: 129­44. Growth: Theory and Evidence." Review of Development Economics ------. 1993b. "On the Number and Location of Cities." European 2 (3): 318­34. Economic Review 37 (1): 293­98. ------. 2002. "Inflation, Growth, and Income Distribution: A ------. 1993c. "On the Relationship between Trade Theory and Cross-Country Study." Annals of Economics and Finance 3: 85­101. Location Theory." Review of International Economics 1:102­22. Ligon, E., and L. Schechter. 2002. "Measuring Vulnerability." Dis- ------. 1999. "The Role of Geography in Development." Interna- cussion Paper 2002/86, United Nations University (WIDER), tional Regional Science Review 22 (2): 142­61. Helsinki. Kuznets, S. 1955. "Economic Growth and Income Inequality." Lillard, L. 1977. "Inequality: Earnings vs. Human Wealth." American American Economic Review 45 (1): 1­28. Economic Review 67 (2): 42­53. Lalive, R., and A. Cattaneo. 2004. "Social Interactions and Schooling Lilliard, L., and R. Willis. 1978. "Dynamic Aspects of Earnings Decisions." University of Zurich, Switzerland. Mobility." Econometrica 46 (5): 985­1012. Lam, D., and R. Schoeni. 1993. "Effects of Family Background on Lindert, K., E. Skoufias, and J. Shapiro. 2005. "How Effectively Do Earnings and Returns to Schooling: Evidence from Brazil." Journal Public Transfers in Latin America Redistribute Income?" LACEA of Political Economy 101 (4): 710­40. Working paper, World Bank, Washington, DC. Lanjouw, P. 2001. "Nonfarm Employment and Poverty in Rural Ljungqvist, L. 1993. "Economic Underdevelopment: The Case of El Salvador." World Development 29 (3): 529­47. a Missing Market for Human Capital." Journal of Development Larrea, C., and F. Montenegro. 2004. "Indigenous People and Poverty in Economics 40 (2): 219­39. Ecuador." Background paper for the report Indigenous People, Poverty Loayza, N., P. Fajnzylber, and C. Calderon. 2002. "Economic Growth and Human Development 1994­2004, World Bank, Washington, DC. in Latin America and the Caribbean: Stylized Facts, Explanations Laurini, M., E. Andrade, and P. Valls Pereira. 2004. "Income Con- and Forecasts." Unpublished paper, World Bank, Washington, DC. vergence Clubs for Brazilian Municipalities: A Non-Parametric ------. 2005. Economic Growth in Latin America and the Caribbean. Analysis (English version of WPE-6/2003)." Working Paper Washington, DC: World Bank. WPE 41, Business School, São Paulo. Loayza, N., A. Oviedo, and L. Servén. 2005. "Regulation and Macro- Lederman, D., W. Maloney, and L. Servén. 2005. Lessons from NAFTA economic Performance." Policy Research Working Paper 3469, for Latin America and the Caribbean. Stanford, CA: Stanford World Bank, Washington, DC. University Press. Loayza, N., and C. Raddatz. 2005. "The Composition of Growth Lee, Y., and R. Gordon. 2005. "Tax Structure and Economic Matters for Poverty Alleviation." Unpublished paper, World Growth." Journal of Public Economics 89 (5­6): 1027­43. Bank, Washington, DC. 211 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S Lokshin, M., and M. Ravallion. 2004. "Household Income Dynamics Maloney, W., W. Cunningham, and M. Bosch. 2004. "Who Suffers in Two Transition Economics." Studies in Nonlinear Dynamics and Income Shocks during Crises? An Application of Quantile Analy- Econometrics 8 (3). sis to Mexico: 1992­95." World Bank Economic Review 18 (2): Londoño, J. L. 1995. "Distribución del Ingreso y Desarrollo 155­74. Económico: Colombia en el Siglo XX." Bogotá: TM Editores. Maloney, W., and A. Rodriguez-Clare. 2005. "Innovation Shortfalls." Londoño, J. L., and R. Guerrero. 2000. "Violencia en America Latina: Unpublished paper, World Bank and Inter-American Develop- epidemologia y costos." In Asalto al desarrollo: Violencia en America ment Bank, Washington, DC. Latina, ed. R. Guerrero, A. Gaviria, and J. L. Londoño. Washing- Maloney, W., and Q. Wodon. 1999. "Self-Employment as an Expla- ton, DC: Inter-American Development Bank. nation for Latin Income Inequality." Unpublished paper, World Londoño, J. L., and M. Székely. 2000. "Persistent Poverty and Excess Bank, Washington, DC. Inequality: Latin America, 1970­1995." Journal of Applied Mankiw, N., D. Romer, and D. Weil. 1992. "A Contribution to Economics 3 (1): 93­134. the Empirics of Economic Growth." Quarterly Journal of Economics Lopez, H. 2004. "Pro-Poor-Pro-Growth: Is There a Trade Off?" 107 (2): 407­37. Policy Research Working Paper 3378, World Bank, Washington, Manski, C. 2000. "Economic Analysis of Social Interactions." NBER DC. Working Paper 7580, National Bureau of Economic Research, Lopez, H., and L. Servén. 2005a. "A Normal Relationship? Poverty, Cambridge, MA. Growth and Inequality." World Bank, Washington, DC. Martins, P. S., and P. T. Pereira. 2004. "Does Education Reduce ------. 2005b. "Too Poor to Grow." World Bank, Washington, DC. Wage Inequality? Quantile Regressions Evidence from Fifteen Lopez-Acevedo, G. 2001. "Evolution of Earnings and Rates of Returns Countries." Labour Economics 11 (3): 355­71. to Education in Mexico." Policy Research Working Paper 2691, Mauro, P. 1995. "Corruption and Growth." Quarterly Journal of World Bank, Washington, DC. Economics 110 (3): 681­712. Lucas, R. E. 1988. "On the Mechanics of Economic Development." ------. 2002. "The Persistence of Corruption and Slow Economic Journal of Monetary Economics 22 (1): 3­42. Growth." IMF Working Paper 02/213, International Monetary ------. 1990. "Why Doesn't Capital Flow from Rich to Poor Coun- Fund, Washington, DC. tries?" American Economic Review 80 (2): 92­6. Mayer-Foulkes, D. 2001. "The Long-Term Impact of Health on Lucchetti, L. 2005. "Pobreza Subjetiva en la Argentina." Unpub- Economic Growth in Latin America." World Development 29 (6): lished paper, World Bank, Washington, DC. 1025­33. Lundberg M., and L. Squire. 2003. "The Simultaneous Evolution of ------. 2003. "Convergence Clubs in Cross-Country Life Growth and Inequality." Economic Journal 113 (487): 326­44. Expectancy Dynamics." In Perspectives on Growth and Poverty, ed. Lustig, N., O. Arias, and J. Rigolini. 2003. "Poverty Reduction and R. van der Hoeven and A. Shorrocks, 144­71. Tokyo: United Economic Growth: Two-Way Causality." Inter-American Devel- Nations University Press. opment Bank, Washington, DC. ------. 2004. "The Human Development Trap in Mexico." Centro Lustig, N., B. Bosworth, and R. Lawrence, eds. 1992. Assessing the de Investigacion y Docencia Economicas (CIDE), Economic Impact of North American Free Trade. Washington, DC: Brookings Division, Mexico. Institution. McKenzie, D., and C. Woodruff. 2004. "Do Entry Costs Provide an Luttmer, E. 2002. "Measuring Economic Mobility and Inequality: Empirical Basis for Poverty Traps? Evidence from Mexican Disentangling Real Events from Noisy Data." Unpublished Microenterprises." Unpublished paper, Stanford University, paper, University of Chicago, Chicago. Stanford, CA. MaCurdy, T. 1982. "The Use of Time Series Processes to Model the Meghir, C., and L. Pistaferri. 2004. "Income Variance Dynamics and Error Structure of Earnings in a Longitudinal Data Analysis." Heterogeneity." Econometrica 72 (1): 1­32. Journal of Econometrics 18 (1): 83­114. Mehlum, H., K. Moene, and R. Torvik. 2005. "Crime-Induced Maddison, A. 2005. "Measuring and Interpreting World Economic Poverty Traps," Journal of Development Economics 77 (2): 325­40. Performance 1500­2001." Review of Income and Wealth 51 (1): Mercader-Prats, M., and H. Levy. 2004. "The Role of Tax and Trans- 1­35. fers in Reducing Personal Income Inequality in Europe's Regions: Maloney, W. 2000. "Self-Employment and Labor Turnover in Develop- Evidence from EUROMOD." EUROMOD Working Paper ing Countries: Cross Country Evidence." In World Bank Economists' EM9/04, EUROMOD, University of Essex, Colchester, Essex, Forum, ed. S. Devarajan, F. Rogers, and L. Squire. Washington, DC: U.K. World Bank. Mercado, A., L. Andersen, and B. Muriel. 2005. "Discriminación ------. 2003. "Informality Revisited." World Bank, Washington, étnica en el sistema educativo y el mercado de trabajo de Bolivia." DC. Latin American Journal of Economic Development. ------. 2004. "Informality Revisited." World Development 32 (7): Metcalf, C. 1969. "The Size Distribution of Personal Income during 1159­78. the Business Cycle." American Economic Review 59 (4): 657­68. Maloney, W., and R. Azevedo. 1995. "Trade Reform, Uncertainty Milanovic, B. 2000. "Determinants of Cross-Country Income and Export Promotion." Journal of Development Economics 48 (1): Inequality: An `Augmented' Kuznets Hypothesis." In Equality, 67­89. 212 B I B L I O G R A P H Y Participation, Transition, ed. V. Franicevic and M. Uvalic. New York: United States." In The Cambridge Economic History of the United St. Martin's Press. States, Vol. II, ed. S. Engerman and R. Gallman. Cambridge, ------. 2005. Worlds Apart: Measuring International and Global U.K.: Cambridge University Press. Inequality. Princeton, NJ: Princeton University Press. Prados de la Escosura, L. 2005. "Growth, Inequality, and Poverty in Milanovic, B., and L. Squire. 2005. "Does Tariff Liberalization Latin America: Historical Evidence, Controlled Conjectures." Increase Wage Inequality? Some Empirical Evidence." NBER Economics History and Institutions Working Papers wh 054104, Working Paper 11046, National Bureau of Economic Research, Universidad Carlos III, Departamento de Historia Económica e Cambridge, MA. Instituciones, Madrid. Mirrlees, J. 1975. "A Pure Theory of Underdeveloped Economies." Pritchett, L. 1997. "Divergence, Big Time." Journal of Economic In Agriculture in Development Theory, ed. L. Reynolds. New Haven, Perspectives 11 (3): 3­17. CT: Yale University Press. Pritchett, L., and L. Summers. 1996. "Wealthier Is Healthier." Mizala, A., and P. Romaguera. 2002. "Equity and Educational Per- Journal of Human Resources 31 (4): 841­68. formance." Journal of the Latin American and Caribbean Economic Pritchett, L., A. Suryahadi, and S. Sumarto. 2000. "Quantifying Association 2 (2): 219­62. Vulnerability to Poverty. A Proposed Measure Applied to Montenegro, C. 2001. "Wage Distribution in Chile: Does Gender Indonesia." Policy Research Working Paper 2437, World Bank, Matter? A Quantile Regression Approach." Policy Research Washington, DC. Report on Gender and Development, Working Paper 20, World Psacharopoulos, G., and H. Patrinos. 2004. "Returns to Investment in Bank, Washington, DC. Education: A Further Update." Education Economics 12 (2): 111­34. Morley, S. 2000. "The Effects of Growth and Economic Reform on Quah, D. 1993. "Empirical Cross-Section Dynamics in Economic Income Distribution in Latin America." CEPAL Review 71: 23­40. Growth." European Economic Review 37 (2­3): 426­34. Murphy, K., A. Shleifer, and R. Vishny. 1989. "Industrialisation and ------. 1997. "Empirics for Growth and Distribution: Stratifica- the Big Push." Journal of Political Economy 97 (5): 1003­26. tion, Polarization and Convergence Clubs." Journal of Economic ------. 1991. "The Allocation of Talent: Implications for Growth." Growth 2 (1): 27­59. Quarterly Journal of Economics 106 (2): 503­30. Ramirez, M., and N. Nazmi. 2003. "Public Investment and Eco- Murphy, K., and R. Topel. 2003. "The Economic Value of Medical nomic Growth in Latin America: An Empirical Test." Review of Research." In Measuring the Gains from Medical Research: An Development Economics 7 (1): 115­26. Economic Approach, ed. K. Murphy and R. Topel. Chicago: Univer- Ravallion, M. 1997. "Can High Inequality Development Countries sity of Chicago Press. Escape Absolute Poverty?" Economics Letters 56 (1): 51­7. Nelson, R. 1956. "A Theory of the Low-Level Equilibrium Trap in ------. 1998. "Poverty Lines in Theory and Practice." LSMS Work- Underdeveloped Economies." American Economic Review 46 (5): ing Paper 133, World Bank, Washington, DC. 894­908. ------. 2004. "Pro-Poor Growth: A Primer." Policy Research Newberry, D., and J. Stiglitz. 1984. "Pareto Inferior Trade." Review of Working Paper 3242, World Bank, Washington, DC. Economic Studies 51 (1): 1­12. Ravallion, M., and B. Bidani. 1994. "How Robust Is a Poverty North, D. 1990. Institutions, Institutional Change, and Economic Perfor- Profile?" World Bank Economic Review 8 (1): 75­102. mance. New York: Cambridge University Press. Ravallion, M., and S. Chaudhuri. 1997. "Risk and Insurance in Nurkse, R. 1953. Problems of Capital Formation in Developing Countries. Village India: A Comment." Econometrica 65 (1): 171­84. New York: Oxford University Press. Ravallion, M., G. Datt, and D. van de Walle. 1991. "Quantifying Openshaw, S., C. Brundson, and M. Charlton. 1991. "A Spatial Absolute Poverty in the Developing World." Review of Income and Analysis Toolkit for GIS." Proceedings of the Second European Con- Wealth 37 (4): 345­61. ference on Geographical Information Systems (EGIS), 788­96. Raymond, M., and E. Sadoulet. 2003. "Educational Grants Closing Utrecht: EGIS Foundation. the Gap in Schooling Attainment between Poor and Non-Poor." Openshaw, S., A. Cross, and M. Charlton. 1990. "Building a Proto- Economic Studies and Policy Analysis Division of the Department type Geographical Correlates Exploration Machine." International of Finance, Ottawa, Canada; and Department of Agricultural and Journal of Geographical Information Systems 4: 297­311. Resource Economics, University of California at Berkeley. Paglin, M. 1975. "The Measurement and Trend of Inequality: A Redding, S., and M. Vera-Martin. 2004. "Factor Endowments and Basic Revision." American Economic Review 65 (4): 598­609. Production in European Regions." CEPR Discussion Paper 3755. Papageorgiou, C., A. Savvides, and M. Zachariadis. 2005. "Interna- Centre for Economic Policy Research, London. tional Medical Technology Diffusion." Department of Economics, Rioja, F., and N. Valev. 2004. "Does One Size Fit All? A Reexamina- Louisiana State University, Baton Rouge; Oklahoma State Uni- tion of the Finance and Growth Relationship." Journal of Develop- versity, and University of Cyprus, Nicosia. ment Economics 74 (2): 429­47. Pareto, W. 1897. Cours d'Economie Politique 2. Lausanne: F. Rouge. Rodriguez-Clare, A. 2001. "Costa Rica's Development Strategy Perotti, R. 1996. "Growth, Income Distribution and Democracy." Based on Human Capital and Technology: How It Got There, Journal of Economic Growth 1 (2): 149­87. the Impact of INTEL, and Lessons for Other Countries." Paper Plotnick, B., E. Smolensky, E. Evenhouse, and S. Reilly. 1996. "The written for the Human Development Report 2001, San Jose, Twentieth-Century Record of Inequality and Poverty in the Costa Rica. 213 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S Rodriguez-Mesa, J., and C. Gonzalez-Vega. 2004. "Household Shorrocks, A. 1981. "Income Stability in the United States." In The Income Dynamics and Poverty Traps in El Salvador." Paper writ- Statics and Dynamics of Income, ed. N. A. Klevmarken and J. A. ten for AAEA (American Agriculture Economists Association) Lybeck, 175­94. Clevedon: Tieto. annual meeting, Ohio State University, Columbus. ------. 1993. "On the Hart Measure of Income Mobility." In Indus- Rodriguez-Oreggia, E., and A. Rodriguez-Pose. 2004. "The trial Concentration and Economic Inequality, ed. M. Casson and Regional Returns of Public Investment Policies in Mexico." J. Creedy. Cheltenham, UK: Edward Elgar. World Development 32 (9): 1545­62. Shorrocks, A., and G. Wan. 2005. "Spatial Decomposition of Rodriguez-Pose, A., and J. Sanchez-Reaza. 2004. "Economic Polar- Inequality." Journal of Economic Geography 5 (1): 59­81. ization through Trade: Trade Liberalization and Regional Growth Singh, S., and G. Maddala. 1976. "A Function for Size Distribution in Mexico." In Spatial Inequality and Development, ed. R. Kanbur of Incomes." Econometrica 44 (5): 963­70. and A. J. Venables. Oxford: Oxford University Press. Smith, D., A. Gordon, K. Meadows, and K. Zwick. 2001. "Liveli- Rodrik, D. 1997. "Globalization, Social Conflict. and Economic hood Diversification in Uganda: Patterns and Determinants of Growth," The 1997 Raul Prebisch lecture delivered at the UN Change across Two Rural Districts." Food Policy 26: 421­35. Conference on Trade and Development (published in World Soares, R. 2004. "The Welfare Costs of Violence." Unpublished Economy, March 1998). paper, University of Maryland, College Park. ------. 1999. "Where Did All the Growth Go? External Shocks, ------. 2005. "Health and Inequality in Latin America and the Social Conflict, and Growth Collapses." Journal of Economic Growth Caribbean." Unpublished paper, University of Maryland, College 4 (4): 385­412. Park. Roemer, J. E. 1998. Equality of Opportunity. Cambridge, MA: Harvard Solon, G. 2002. "Cross-Country Differences in Intergenerational University Press. Earnings Mobility." Journal of Economic Perspectives 16 (3): 59­67. ------. 2005. "Intergenerational Justice and Sustainability under ------. 2004. "A Model of Intergenerational Mobility Variation the Leximin Ethic." Cowles Foundation Discussion Papers 1512, over Time and Place." In Generational Income Mobility in North Cowles Foundation, Yale University, New Haven, CT. America and Europe, ed. M. Corak, 38­47. Cambridge: Cambridge Rosenstein-Rodin, P. 1943. "Problems of Industrialization in Eastern University Press. and South Eastern Europe." Economic Journal 53: 202­11. Sosa-Escudero, W., and M. Cicowiez. 2005. "Exploring the Determi- Ruiz-Castillo J., E. Ley, and M. Izquierdo. 2002. "Distributional nants of Poverty and Income Distribution in the Dominican Aspects of the Quality Change Bias in the CPI: Evidence from Republic." Unpublished paper, Centro de Estudios Distributivos, Spain." Economic Letters 76 (1): 137­44. Laborales y Sociales (CEDLAS), Universidad Nacional de La Rutherford, R. S. G. 1955. "Income Distributions: A New Model." Plata, La Plata, Argentina. Econometrica 23 (3): 277­94. Sosa-Escudero, W., and L. Lucchetti. 2004. "Exploring the Determi- Saavedra, J., and J. Diaz. 1999. "Distribución del Ingreso y del Gasto nants of Poverty and Income Distribution in Peru: A Micro- Antes y Después de las Reformas Estructurales (Perú)." ECLAC econometric Approach." Unpublished paper, World Bank, Economic Reform Series 34. European Commission for Latin Washington, DC. America and the Caribbean. Sosa-Escudero, W., M. Marchionni, and O. Arias. 2005. "Exploring Saavedra, J., and E. Maruyama. 1999. "Los Retornos a la Educación Sources of Income Persistency in Rural El Salvador." Unpublished y a la Experiencia en el Perú, 1985­1997." In Pobreza y Economía paper, World Bank, Washington, DC. Social: Análisis de una Encuesta ENNIV-1997, ed. Richard Webb Spilimbergo, A., J. Londoño, and M. Székely. 1999. "Income Distri- and Moisés Ventocilla. Lima, Peru: Instituto Cuánto. bution, Factor Endowments, and Trade Openness." Journal of Saavedra, J., and M. Torero. 2004. "Labor Market Reforms and Their Development Economics 59 (1): 77­101. Impact over Formal Labor Demand and Job Market Turnover: Stiglitz, J. 1976. "The Efficiency Wage Hypothesis, Surplus Labor The Case of Peru." In Law and Employment: Lessons from Latin and the Distribution of Income in LDCs." Oxford Economic Papers America and the Caribbean, NBER Conference Report series. 28 (2): 185­207. Chicago: University of Chicago Press. Strauss, J., and D. Thomas. 1995. "Empirical Modeling of Household Sachs, J., and A. Warner. 1995. "Economic Reform and the Process of and Family Decisions." Paper 95-12, Reprint Series, RAND, Global Integration." Brookings Papers on Economic Activity 1995 Santa Monica, California. (1): 1­118. Tannuri-Pianto, M., D. Pianto, and O. Arias. 2004a. "Informal Sanchez-Robles, B. 1998. "Infrastructure Investment and Growth: Employment in Bolivia. A Lost Proposition?" Paper 149, Econo- Some Empirical Evidence." Contemporary Economic Policy 16 (1): metric Society, New York. 98­108. ------. 2004b. "Rural-Urban Migration in Bolivia: An Escape Sen, A. 1972. "Control Areas and Accounting Prices: An Approach Boat?" Background paper for the 2005 World Bank Bolivia to Economic Evaluation." Economic Journal 82 (325): 486­501. Poverty Assessment, Washington, DC. ------. 1985. Commodities and Capabilities. Amsterdam: North Tannuri-Pianto, M., D. Pianto, O. Arias, and M. Beneke de Sanfeliu. Holland. 2005. "Determinants and Returns to Productive Diversification ------. 1999. Development as Freedom. Oxford: Oxford University Press. in Rural El Salvador." Background paper for the Beyond the City: 214 B I B L I O G R A P H Y The Rural Contribution to Development, World Bank, Washington, World Bank. 1990. World Development Report 1990: Poverty. DC. Washington, DC: World Bank. Taylor, J. E. 2001. "Microeconomics of Globalization: Evidence from ------. 2001a. "Honduras Poverty Assessment." Report 20531- Mexico, China, El Salvador, and the Galapagos Islands." Report HD. World Bank, Washington, DC. to the Latin America and Caribbean Regional Office, World ------. 2001b. World Development Report: Attacking Poverty. Bank, Washington, DC. Washington, DC: World Bank. Taylor, J. E., A. Yuñez-Naude, and H. Cerón. 2004. "The Effects ------. 2003. "Mexico Southern States Development Strategy." of Rural Mexico Household Assets on Rural Activity Washington, DC: World Bank. Selection, Incomes, and Technology Adoption." World Bank, ------. 2004a. Central America: Education Strategy Paper. Vol. 1. Washington, DC. Washington, DC: World Bank. Taylor, J. E., A. Yunez-Naude, and C. Hazael. 2004. "The Effects of ------. 2004b. "Drivers of Sustainable Rural Growth and Poverty Mexico Household Assets on Rural-Non-Rural Activity Selec- Reduction in Central America." Report 31191, Washington, DC: tion, on Incomes and Technology Adoption." World Bank, World Bank. Washington, DC, and PRECESAM, El Colegio de México. ------. 2004c. Education Strategy Paper for Central America. Green Thisse, J-F., and M. Fujita. 2000. "Economics of Agglomeration." Cover Report. Washington, DC: World Bank. In The Economics of Cities, ed. J.-M. Huriot and J.-F. Thisse. ------. 2004d. World Development Report: Making Services Work for the Cambridge, UK: Cambridge University Press. Poor. Washington, DC: World Bank. Thorbecke, E. 2005. "Multi-Dimensional Poverty: Conceptual and ------. 2005a. "Brazil Regional Economic Development--(Some) Measurement Issues." Paper presented at the international confer- Lessons from Experience." Finance, Private Sector and Infrastruc- ence on "The Many Dimensions of Poverty," Brasilia, August ture Management Unit, Latin American and Caribbean Region, 29­31. World Bank, Washington, DC. Trivelli, C. 2004. "Indigenous Poverty in Peru: An Empirical Analy- ------. 2005b. Dominican Republic Poverty Assessment: Achieving More sis." Unpublished paper, Instituto de Estudios Peruanos (IEP). Pro-Poor Growth. Report 32422-DO. Washington, DC: World Lima, Peru. Bank. Urquiola, M., and V. Calderón. 2004. "Apples and Oranges: Educa- ------. 2005c. DR-CAFTA: Challenges and Opportunities for Cen- tional Enrollment and Attainment across Countries in Latin tral America, Vol. 1. Working paper 32953, World Bank, America and the Caribbean." Unpublished paper, Columbia Washington, DC. University, New York. ------. 2005d. Economic Growth in the 1990s: Learning from a Decade Velez, C., and J. Nunez. 2005. "Multidimensional Poverty in of Reforms. Washington, DC: World Bank. Colombia: Explaining Improved Well-Being Despite Economic -------. 2005e. "El Salvador Poverty Assessment: Strengthening Recession and Violence." Unpublished paper. Inter-American Social Policy." Report 29594-SV. World Bank, Washington, DC. Development Bank, Washington, DC, and Universidad de Los ------. 2005f. A Time to Choose: Caribbean Development in the 21st Andes, Bogotá, Colombia. Century. Washington, DC: World Bank. Wacziarg, R., and K. Welch. 2003. "Trade Liberalization and ------. 2005g. World Development Indicators. Washington, DC: Growth: New Evidence." NBER Working Paper 10152, World Bank. National Bureau of Economic Research, Cambridge, MA, ------. 2005h. Pro-Poor Growth in the 1990s: Lessons and Insights from December. 14 Countries. Washington, DC: World Bank. Williamson, J. G. 1999. "Real Wage Inequality and Globalization ------. 2006. World Development Report: Equity and Development. in Latin America before 1940." Revista de Historia Económica 17 Washington, DC: World Bank. (special issue): 101­42. ------. Forthcoming. The Redistributive Impact of Transfers in Latin ------. 2002. "Land, Labor, and Globalization in the Third World, America and the Caribbean. Washington, DC: World Bank. 1870­1940." Journal of Economic History 62 (1): 55­85. Yepes, T. 2004. "Infrastructure and Poverty Reduction in Rural Wodon, Q., R. Ayres, M. Barenstein, N. Hicks, K. Lee, W. Maloney, El Salvador." Unpublished paper, Poverty Reduction and Economic P. Peeters, C. Siaens, and S. Yitzhaki. 2000. "Poverty and Policy Management and Human Development Sector Management in Latin America and the Caribbean." Technical Paper 467, Units, Latin America and the Caribbean Region, World Bank, World Bank, Washington, DC. Washington, DC. 215 Index Argentina subnational dimensions of growth and poverty, poverty-traps view of development process development gap, 45, 47, 48, 54 9t, 129, 130, 131­133, 132f, 133f, 135f, and, 110­114, 110f, 111t, 112­114f, education and human capital in, 168, 173, 137, 138, 139, 141f 124­125 174f, 176f, 183, 187f, 188f, 189, transfer programs in, 5 regional/subnational convergence clubs. See 194, 195 transfers and income inequality, 97 subnational dimensions of growth and effects of poverty on growth, 109 poverty indicators of poverty in, 22, 23b, 29, 31, 33, Chile relative income levels, 112­113f 36­37, 36t, 38 development gap, 47, 48, 50, 54 transitions of countries between convergence labor and earnings differentials, 149, 150, education and human capital in, 168, 172f, clubs, 124­125 151, 152t 173, 178, 179f, 183, 186, 187f, 188f, Costa Rica poverty reduction, economic growth, and pro- 189, 191, 193, 194, 195, 196 indicators of poverty in, 22, 23b, 33, 35b, 35t, 38 gressive distribution, 70 effects of poverty on growth, 107b, 111 poverty reduction, economic growth, and pro- subnational dimensions of growth and indicators of poverty in, 22, 23b, 24, 38 gressive distribution, 62 poverty, 139 labor and earnings differentials, 148 transfers and income inequality, 99 transfers and income inequality, 95, 96 poverty reduction, economic growth, and pro- Cox hazard regression methods, 198­199 gressive distribution, 62, 63, 70 credit constraint--factor accumulation argument Becker model of human capital and household subnational dimensions of growth and poverty, for effects of poverty on growth, 116 behavior, 169 129, 130, 133, 134f, 138, 139 Cuba, 21 Belize, 69, 70, 176 transfers and income inequality, 92, 97, 99 Bolivia Colombia de Ferranti, David, editor, World Bank Latin country income--level of, 104 development gap, 47, 48 American region flagship publications. education and human capital in, 168, 172f, education and human capital in, 168, 173, See under World Bank 173, 175f, 183, 188f, 189, 194, 196 176f, 178, 180, 181f, 183, 189, 190, decomposition. See variance decomposition effects of poverty on growth, 109, 111 191, 193, 194, 195 approach indicators of poverty in, 22, 23b, 28, 29­31, indicators of poverty in, 22, 23b, 26b, 27b, decomposition of effects of economic growth, and 30f, 35b 28, 38t progressive distribution on poverty labor and earnings differentials, 148, 149, 151 subnational dimensions of growth and poverty, reduction, 60b, 60f, 61t poverty reduction, economic growth, and pro- 137, 138, 139 demographics gressive distribution, 62, 70 transfer programs in, 5 education and human capital, 196­197, 197f subnational dimensions of growth and poverty, transfers and income inequality, 95, 96, 97, 98b income inequalities and, 33, 35b, 35t 130, 138, 139 consumption levels as indicator of poverty, development gap in Latin America/Caribbean, Brazil 20­27 45­56 development gap, 46, 47, 48, 50, 54 convergence hypothesis colonial period, historical origins in, 45­46 education and human capital in, 168, 173, absolute income levels, 110­112, 110f, 112f global comparisons 174f, 175f, 176, 180, 181f, 183, 186, development gap in Latin America/Caribbean income inequalities, 55f 187f, 190, 191, 193, 194, 195 and, 48f, 49f per capita income, 50­53, 50t, 51­53f effects of poverty on growth, 107 global convergence clubs, evidence contradict- income inequalities, 53­56, 54t, 55f indicators of poverty in, 22, 26b, 27b, 28, 38 ing, 6­7, 7f per capita income, 46­53 labor and earnings differentials, 148, long-run equilibriums for convergence convergence hypothesis, 48f, 49f 149, 150 clubs, 124 global comparisons, 50­53, 50t, 51­53f poverty reduction, economic growth, and pro- nonincome welfare measures, applicability to, historical estimates, 46­48, 47t gressive distribution, 62, 63, 69, 70 113­114, 114f long-run trends, 49­50, 49f, 49t 217 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S development process. See economic growth poverty reduction, economic growth, and pro- education, 120­121, 120t discrimination gressive distribution, 62, 70 empirical evidence of, 115­116 in education, 170, 184­185, 185f subnational dimensions of growth and poverty, estimating, 125­126 labor and earnings differentials due to, 148 141 financial sector development, 119­120, distribution of income. See income inequalities; education, 165­201 119t, 124 transfers data and methodologies, 197­199, 197f formal tests of, 116­117 divergent mobility, poverty traps as, 32­33, demographic factors, 196­197, 197f health, 121 32b, 32t discrimination in, 170, 184­185, 185f innovation, 121­122 Dominican Republic effects of poverty on growth, 120­121, 120t investment rates, 118­119, 118f education and human capital in, 167, 173, gender affecting, 193 mobility and risk/security factors, 122­123, 177, 179, 180, 181f, 183, 186, 190, human capital, failure of poor to accumulate, 123t 191, 193, 194 165­166 poverty-traps view of development and, effects of poverty on growth, 109 human capital underinvestment traps and, 104­115 indicators of poverty in, 21, 24, 25b, 29, 31 169­171 convergence hypothesis, 110­114, 110f, labor and earnings differentials, 149, 151 inability to afford, 159­170 111t, 112­114f, 124­125 poverty reduction, economic growth, and pro- labor and earnings differentials empirical evidence of, 108­110, 108f, 109t gressive distribution, 62, 70 effects of level of education on, 147­148, formal testing for, 114­115 subnational dimensions of growth and 151, 154, 155, 159 increasing returns to scale, 105f, 106 poverty, 130 unmeasured skills, effects of, 185­190, institutional mechanisms, 107­108 187­189f investment rates, 118 earnings. See labor and earnings differentials liquidity constraints affecting, 178­181, market factors, 106­107, 106f, 107b East Asia, Latin America/Caribbean relative to 179­180f, 193 traditional view of development vs., 105 development gap, 50­53 parents' level of education affecting children's, special regional effects of, 117b, 117t education and human capital, 166­167 180, 181f, 190, 191­193, 192f transmission channels for, 118­123 effects of poverty on growth, 107b persistent patterns and clusters in educational El Salvador indicators of poverty, 29t transitions, 171­178, 172f, 174­177f, development gap, 54 labor and earnings differentials, 147 178t education and human capital in, 167, 173, poverty reduction, economic growth, and pro- physical access constraints, 193 174f, 175f, 176f, 183, 186, 187f, 189, gressive distribution, 57 policy implications, 194­196 191, 193 vicious circle of high poverty/low economic pro-poor economic growth policies, 87­88, 88f indicators of poverty in, 22, 23b, 28, 38 growth, 1, 2f, 14 productivity differentials and, 107b labor and earnings differentials, 146, East Asian financial crisis of 1997, 24, 48, 108 quality of education, effects of differences in, 152­159, 153­154t, 156f, 157t, 158f Economic Commission for Latin America and the 190, 191f poverty reduction, economic growth, and pro- Caribbean (ECLAC), poverty indicators reasons for low attainment levels, 178­181, gressive distribution, 62, 70 used by, 23b 179­181f subnational dimensions of growth and poverty, economic crises relationship of poverty and low levels of, 141b East Asia, 1997, 24, 48, 108 167­169, 168f transfers and income inequality, 92, 95, 96 Mexico, 1994, 87 relative importance of different constraints on elasticities of poverty Russia, 1998, 24, 48, 108 achievement of, 190­194, 192f poverty reduction, economic growth, and pro- economic growth returns on, 181­190, 182f gressive distribution, 65­66, 65b, 66t effects of poverty on. See effects of poverty on data and methodologies, 198 pro-poor economic growth policies, 86t growth quality of education, effects of differences Engel/Orshansky ratio of food expenditures, 23b income inequality and, 4t, 57­58, 58f, 69t in, 190, 191f ethnicity and race low growth's relationship to poverty. See racial and rural variations, 184­185f education, variation in returns on, 184­185, vicious circle of high poverty/low unmeasured labor market skills, effects of, 185f economic growth 185­190, 187­189f labor and earnings differentials, 148 nexus between transfers, growth, and poverty skills crucial to success in schooling, problems EUROMOD, 93 reduction. See poverty reduction, in development of, 170­171 Europe. See OECD countries, Latin America/ economic growth, and progressive slow pace of educational transition in region, Caribbean relative to; Spain and periph- distribution 166­167, 166f, 167f eral Europe, Latin America/Caribbean poverty-traps view of development and. See unmeasured skills, effects of, 185­190, relative to under effects of poverty of growth 187­189f pro-poor forms of. See pro-poor economic vicious circle of high poverty/low economic financial crises growth policies growth and, 3f, 10­11, 10f, 16­17 East Asia, 1997, 24, 48, 108 rates in Latin America/Caribbean, 24t, 25b effects of poverty on growth, 103­127 Mexico, 1994, 87 subnational dimensions. See subnational convergence hypothesis, 110­114, 110f, 111t, Russia, 1998, 24, 48, 108 dimensions of growth and poverty 112­114f, 124­125 financial sector development Ecuador credit constraint--factor accumulation argu- effects of poverty on growth, 119­120, indicators of poverty in, 22, 23b, 38 ment for, 116 119t, 124 218 I N D E X poverty and income inequalities affected by, lognormal function used to measure and policy implications, 159­160 78­80, 79t express, 64­65, 64b, 64f, 71­72, 72t productivity and earnings, relationship formal vs. informal labor markets, earnings dif- mapping Latin American/Caribbean countries between, 147f ferentials in, 148­149, 149f in income inequality space, 69t race and ethnicity, 148 funding as means of reversing vicious circle of pro-growth policies and. See pro-poor regional variations, 148, 150 high poverty/low economic growth, economic growth policies segmentation theory, 146, 159 18­19, 18f reasons for persistence of, 4 unmeasured worker skills, effects of, 152, spatial/geographic, 129­130, 130f 185­190, 187­189f gender transfers to address, 4­5, 5f vulnerability to and persistence of poverty over education affected by, 193 income risk. See risk/security factors time, 157­159, 157t, 158f labor and earnings differentials affected by, 148 income transfers. See transfers liberalization of trade. See trade liberalization Guatemala indicators of poverty, 20­44 life expectancy indicators of poverty in, 22, 35b, 38 income/consumption levels, 20­27 convergence hypothesis applied to, 113­114, labor and earnings differentials, 148 inflation inequality, 26­27b, 26t 114f poverty reduction, economic growth, and pro- life expectancy, 28, 29t, 41 effects of poverty on growth, 121 gressive distribution, 69, 70 mobility, 31­33 as indicator of poverty, 28, 29t, 41 transfers and income inequality, 96 NA (national accounts) and household subnational dimensions of growth and poverty, Guyana, 69, 70 surveys-based data, 25b, 25t 132, 135f nonincome measures, 27­28 local indicators of spatial association (LISA), Haiti, 22, 109, 196 risk/security factors, 33­37, 34b, 41­42 131b health and effects of poverty on growth, 121. self-assessments of, 9, 29­31, 30f lognormal function used to measure and express See also life expectancy single-moment vs. long-term measures, 31 income inequalities, 64­65, 64b, 64f, Honduras vicious circle of high poverty/low economic 71­72, 72t country income--level of, 104 growth formed by, 41 longevity. See life expectancy development gap, 54 inflation inequality, 26­27b, 26t effects of poverty on growth, 109, 111 informal vs. formal labor markets, earnings dif- market differentials. See labor and earnings indicators of poverty in, 22, 23b, 28, 35b ferentials in, 148­149, 149f differentials poverty reduction, economic growth, and pro- innovation, poverty as limiting, 121­122 market factors and poverty traps, 106­107, 106f, gressive distribution, 62, 70 institutional quality 107b subnational dimensions of growth and as poverty trap, 107­108 measures of poverty. See indicators of poverty poverty, 130 as pro-poor economic growth policy, 88­89, Mexico transfers and income inequality, 97 88f, 89t development gap, 46, 47, 48, 50, 54 household surveys data as indicator of intergenerational mobility, 37­40, 38f, 38t, 39f, education and human capital in, 168, 171, poverty/economic growth, 25b, 25t 42, 157­159 172f, 173, 174f, 176f, 183, 194, human capital isometric poverty curves, 68f 195, 196 educational divide perpetuating poor's failure effects of poverty on growth, 111 to accumulate, 165­166. See also Jamaica, 21, 22, 23b, 62, 177 financial crisis of 1994, 87 education indicators of poverty in, 21, 22, 23b, 26b, 27, policy implications, 194­196 Kuznets, Simon, 21, 31, 33, 34b, 35b, 37, 40, 27b, 33, 36­37, 36t, 38 relationship of poverty and accumulation of, 157 labor and earnings differentials, 148, 150, 167­169, 168f 151f underinvestment traps, 169­171 labor and earnings differentials, 145­163 NAFTA, effects of, 27, 87, 136 vicious circle of high poverty/low economic complementarities and initial conditions, rele- poverty reduction, economic growth, and pro- growth and, 10­11, 16 vance of, 155­157, 156f gressive distribution, 62 Hungary, Latin America/Caribbean compared to, data and methodologies, 152, 160­162 subnational dimensions of growth and poverty, 33, 115 differentials in, different perspectives on 129, 130, 133­135, 134f, 135f, 136, mechanisms behind, 146­147 137, 138, 139, 142t income as indicator of poverty, 20­27 driving factors in, 151­152, 152t transfers in, 5, 97 income inequalities education microdeterminants convergence hypothesis, evidence contradict- effects of level of, 147­148, 151, 154, education. See education ing, 6­7, 7f 155, 159 human capital. See human capital demographics and, 33, 35b, 35t unmeasured skills, effects of, 185­190, labor and earnings. See labor and earnings dif- development gap and, 53­56, 54t, 55f 187­189f ferentials economic growth and, 4t, 57­58, 58f, 69t El Salvador case study, 152­159, 153­154t, vicious circle of high poverty/low economic heterogeneity between Latin American/ 156f, 157t, 158f growth at household level Caribbean countries regarding, 54 formal vs. informal labor markets, 148­149, poverty traps, 3f, 9­11, 10f indicators of, 24­25, 25f 149f reversal strategies, 16­18 inflation inequality, 26­27b, 26t gender gap, 148 migration as equilibrating mechanism on sub- in Latin America/Caribbean, 1950­2000, 1, 2f mobility, 149­150, 151f national inequalities, 138­139, 150 219 P O V E RT Y R E D U C T I O N A N D G R O W T H : V I RT U O U S A N D V I C I O U S C I R C L E S Millennium Development Goals, 16, 23b, 28, 195 parametric analysis of poverty reduction, isometric poverty curves, 68f Mirrlees, efficiency wage hypothesis of, 32b economic growth, and progressive distri- lognormal function used to measure and mobility bution, 59, 63­64, 68 express income inequalities, 64­65, 64b, indicators of poverty, 31­33 permanent-transitory income hypothesis, 157 64f, 71­72, 72t intergenerational, 37­40, 38f, 38t, 39f, 42, Peru mapping Latin American/Caribbean countries 157­159 development gap, 47, 48 in income inequality space, 69t labor and earnings differentials, 149­150, 151f education and human capital in, 168, 173, parametric analysis, use of, 59, 63­64, 68 poverty traps and, 32­33, 32b, 32t 178, 179f, 183, 186, 194 relative importance of growth vs. redistribu- risk/security factors, 33­37, 41­42, 122­123, indicators of poverty in, 22, 23b, 26b, 27b, tion, 58­63, 62f, 63f 123t 28, 38 variance decomposition approach, 60b, 60f, 61t mortality. See life expectancy labor and earnings differentials, 148, 151, 152t poverty traps poverty reduction, economic growth, and pro- convergence clubs and, 110­114, 110f, 111t, NAFTA (North American Free Trade gressive distribution, 62, 70 112­114f Agreement), 27, 87, 136 pro-poor economic growth policies, 80 development process from point of view of, national accounts (NA) data as indicator of subnational dimensions of growth and poverty, 104­106, 105f poverty/economic growth, 25b 130, 137, 138 as divergent mobility, 32­33, 32b, 32t New Economic Geography literature, 8, 15, 135, transfers and income inequality, 97 dynamics leading to, 32b, 32t 136b, 137, 142 policy implications empirical evidence of, 108­110, 108f, 109t Nicaragua education and human capital, 194­196 formal testing for, 114­115 country income--level of, 104 growth policies aimed at poverty reduction. at household level, 3f, 9­11, 10f education and human capital in, 168, 172f, See pro-poor economic growth policies human capital underinvestment, 169­171 173, 175f, 176, 183, 186, 189, 191, 193 labor and earnings differentials, 159­160 increasing returns to scale, 105f, 106 effects of poverty on growth, 109, 111 public investment, 142 institutional mechanisms and, 107­108 indicators of poverty in, 22, 23b, 35b, 38 subnational dimensions of growth and poverty, investment rates and, 118 poverty reduction, economic growth, and pro- 139­142 market factors, 106­107, 106f, 107b gressive distribution, 62, 63, 70 trade liberalization. See trade liberalization PPP (Purchasing Power Parity), 21, 23b transfers and income inequality, 99 vicious circle of high poverty/low economic pro-poor economic growth policies, 75­102 nonincome welfare measures growth, 13­15 complementarities and nonlinearities between convergence hypothesis applicable to, poverty growth and poverty, 86­89 113­114, 114f defining, 23b conflicts and trade-offs between growth and as indicators of poverty, 27­28 density vs. rates, 139­141, 140t, 141f poverty, 83­86, 85t, 86t, 139 of subnational dimensions of growth and education/human capital accumulation and, education, 87­88, 88f poverty, 132, 135f 167­169, 168f financial development, effects of, 78­80, 79t North American Free Trade Agreement effects on growth. See effects of poverty on institutional quality, 88­89, 88f, 89t (NAFTA), 27, 87, 136 growth pro-growth policies that fail to reduce, or elasticities of increase, poverty, 75­76, 76f OECD countries, Latin America/Caribbean poverty reduction, economic growth, and sectoral distinctions, 89­92, 89f, 90f, 91f, 91t relative to progressive distribution, 65­66, 65b, 66t simulating impact of, 100­102 development gap, 45­46 pro-poor economic growth policies, 86t simultaneous impact of policies on growth and effects of poverty on growth, 108f, 109 indicators of. See indicators of poverty income inequalities, 76­78, 77t labor and earnings differentials, 148 low economic growth and. See vicious circle of size of government and public spending, poverty reduction, economic growth, and pro- high poverty/low economic growth 81­83, 84f gressive distribution, 50­53, 55f, 56 multidimensional aspects, 1­2 trade liberalization, 80­81 pro-poor economic growth policies, 93­95, rates in Latin America/Caribbean, 1950­2000, transfers, role of, 92­100 94t, 95f 1, 2f, 20­24, 21f, 21t, 24f low impact of current Latin American/ public investment policies and subnational subnational dimensions. See subnational Caribbean regimes, 95­97, 95f, 96b, dimensions of growth and poverty, 142 dimensions of growth and poverty 96t, 99f vicious circle of high poverty/low economic poverty line, 23b, 63 simulating effective redistributive packages, growth, 1, 2f, 4, 5f, 18f poverty reduction, economic growth, and 97­102, 100f, 100t progressive distribution, 57­73 progressive distributional changes. See transfers Panama, 62, 70 alternative growth scenarios, 68­69, 68t public spending and size of government, effects Paraguay choice of poverty line, significance of, 63 of, 81­83, 84f effects of poverty on growth, 107 country-specific considerations, 63­70 Purchasing Power Parity (PPP), 21, 23b indicators of poverty in, 23b, 25b, 38 elasticities of poverty, 65­66, 65b, 66t poverty reduction, economic growth, and pro- empirical and theoretical quintiles, 66­68, 67f race and ethnicity gressive distribution, 62, 70 importance of economic growth, 57, 58f education, variation in returns on, 184­185, subnational dimensions of growth and importance of redistribution measures, 57, 58f 185f poverty, 130 income inequality reduction not strongly labor and earnings differentials, 148 transfers and income inequality, 95, 96 linked to economic growth, 57­58, 58f redistribution of income. See transfers 220 I N D E X regional dimensions of growth and poverty. See labor and earnings differentials, 148, 150 variance decomposition approach subnational dimensions of growth and Mexico case study, 133­135, 134f, 135f indicators of poverty, 34f, 42 poverty migration as equilibrating mechanism, labor and earnings differentials, 149, 151, República Bolivariana de Venezuela. See 138­139 152t, 161, 162 Venezuela, República Bolivariana de nonincome welfare measures, 132, 135f poverty reduction, economic growth, and risk/security factors policy implications of, 139­142 progressive distribution, 59­63, 60b, effects of poverty on growth, 122­123, 123t poverty rates vs. poverty density, 139­141, 60f, 61t mobility and, 33­37, 41­42, 122­123, 123t 140t, 141f spatial inequality and subnational dimensions poverty measures and, 33­37, 34b, 41­42 public investment policies, 142t of growth and poverty, 129 trade policy and, 82b, 82t reasons for, 135­138 Venezuela, República Bolivariana de rural areas special/geographic inequalities, 129­130, 130f development gap, 47, 48, 50, 54 education, variation in returns on, 184­185, 184f trade liberalization, effects of, 136 indicators of poverty in, 22, 33 mobility, labor, and earnings differentials, 150 poverty reduction, economic growth, and pro- pro-poor economic growth policies, sectoral taxes, redistributive. See transfers gressive distribution, 62, 69, 70 factors in, 89­92, 89f, 90f Theil index, 34b subnational dimensions of growth and subnational dimensions of growth and poverty, trade liberalization poverty, 130 141b effect on poor of, 27 vicious circle of high poverty/low economic Russia, Latin America/Caribbean compared to, as pro-poor economic growth policy, 80­81 growth, 1­19 24, 33, 48, 108.115 risk factors, 82b, 82t convergence Russian financial crisis of 1998, 24, 48, 108 subnational inequalities and, 136 global convergence clubs, evidence contra- transfers dicting, 6­7, 7f sectoral distinctions in pro-poor economic growth income inequalities addressed via, 4­5, 5f regional convergence clubs, evidence of, policies, 89­92, 89f, 90f, 91f, 91t low impact of Latin American/Caribbean 8­9, 9f security. See risk/security factors regimes, 95­97, 95f, 96b, 96t, 99f education and, 3f, 10­11, 10f, 16­17 segmentation of labor market, 146, 159. See also as poverty reduction strategy. See poverty effects of growth on poverty levels, 4­5 labor and earnings differentials reduction, economic growth, and effects of poverty levels on growth, 5­6 self-assessments of poverty, 9, 29­31, 30f progressive distribution evidence of, 7­8, 8f Sen, Amartya, 21, 27, 29, 31 pro-poor economic growth policies and, funding reversal mechanisms, 18­19, 18f size of government and public spending, effects 92­100 at household level of, 81­83, 84f role in reducing income inequality, 92­95, 92f poverty traps, 3f, 9­11, 10f social indicators. See nonincome welfare measures simulating effective redistributive packages, reversal strategies, 16­18 social security systems. See transfers 97­102, 100f, 100t human capital and, 10­11, 16 Spain and peripheral Europe, Latin subnational areas with low poverty rates and indicators of poverty forming, 41 America/Caribbean relative to high poverty densities, 139 multidimensional aspects of poverty, 1­2 development gap, 46 Trinidad and Tobago, 69, 70 policy implications of, 13­15 poverty reduction, economic growth, and pro- pro-growth poverty reduction to transform, gressive distribution, 50­53, 55f United Nations Millennium Development Goals, 15­19 pro-poor economic growth policies, 93, 94t 16, 23b, 28, 195 reasons for persistence of, 2­4 subnational dimensions of growth and poverty, United States. See OECD countries, Latin spatial concerns and reversal strategies, 15­16 139, 140b America/Caribbean relative to strategic implications of, 11­13 vicious circle of high poverty/low economic urban areas transfers addressing, 4­5, 5f growth, 1, 2f, 4, 5f education, variation in returns on, 184­185, St. Lucia, 69, 70 184f welfare payments. See transfers Stiglitz, efficiency wage hypothesis of, 32b mobility, labor, and earnings differentials, 150 World Bank subnational dimensions of growth and poverty, pro-poor economic growth policies, sectoral Latin American region flagship publications 129­144 factors in, 89­92, 89f, 90f (de Ferranti and others) Brazil case study, 131­133, 132f, 133f, 135f Uruguay 2000, 2, 33, 122 Chile case study, 133, 134f development gap, 47, 48, 54 2002, 13 conflicts and trade-offs between growth and education and human capital in, 177 2003, 10, 17, 80, 87, 88, 107, 120, 138, poverty, 139 effects of poverty on growth, 111 165, 166, 167, 170, 181, 195 evidence for, 8­9, 9f indicators of poverty in, 22, 38 2004, 3, 13, 24, 45, 53, 78, 126 identifying spatial concentrations, 130­131, poverty reduction, economic growth, and pro- 2005, 13, 14, 15, 75, 90, 92, 129, 149, 150 131b gressive distribution, 63, 69, 70 poverty line used by, 23b 221 hat raising income levels alleviates poverty, and that economic T growth can be more or less effective in doing so, is well known and has received renewed attention in the search for pro-poor growth. Less well explored is the reverse channel: that poverty may, in fact, be part of the reason for a country's poor growth performance. This more elaborat- ed view of the development process opens the door to the existence of vicious circles in which low growth results in high poverty and high poverty in turn results in low growth. Poverty Reduction and Growth: Virtuous and Vicious Circles is about the existence of those vicious circles in Latin America and the Caribbean and about the ways and means to convert them into virtuous circles in which poverty reduction and high growth reinforce each other. Through its analy- sis of fresh data and the attention it pays to issues such as the persistent inequality in the region, the role played by various microdeterminants of income, and the potential existence of human capital underinvestment traps, Poverty Reduction and Growth: Virtuous and Vicious Circles should be a valuable contribution to the current regional debate on poverty and growth, a debate that is critical to the design of policies conducive to enhancing welfare in all its dimensions among the poor of Latin America and the Caribbean. ISBN 0-8213-6511-8