WPS8135 Policy Research Working Paper 8135 Global Inequality in a More Educated World Syud Amer Ahmed Maurizio Bussolo Marcio Cruz Delfin S. Go Israel Osorio-Rodarte Development Economics Development Prospects Group June 2017 Policy Research Working Paper 8135 Abstract In developing countries, younger and better-educated representing almost 90 percent of the world population. cohorts are entering the workforce. This developing world- The findings under alternative assumptions suggest that led education wave is altering the skill composition of the global income inequality will likely decrease by 2030. This global labor supply, and impacting income distribution, at increasing educated labor force will contribute to the clos- the national and global levels. This paper analyzes how this ing of the gap in average incomes between developing and education wave reshapes global inequality over the long high income countries. The forthcoming education wave run using a general-equilibrium macro-micro simulation would also minimize, mainly for developing countries, framework that covers harmonized household surveys potential further increases of within-country inequality. This paper is a product of the Development Prospects Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at sahmed20@worldbank.org, mbussolo@worldbank.org; marciocruz@worldbank.org; dgo@worldbank.org; and iosoriorodarte@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Global Inequality in a More Educated World1 Syud Amer Ahmed, Maurizio Bussolo, Marcio Cruz, Delfin S. Go, and Israel Osorio-Rodarte2 JEL Classification: D31, J11, J31, E24 KEYWORDS: Global Inequality, Education, Demographic trends, Structural change 1 We thank Christina Calvo, Francisco Ferreira, Ayhan Kose, Maryla Maliszewska, Hans Timmer, Philip Schellekens, Jos Verbeek, and participants of seminars organized by the World Bank’s Equity and Public Policy Practice Group, Development Prospects Group, Graduate Program in Economics at the Federal University of Parana, Brazil, and the 17th Annual Conference on Global Economic Analysis in Dakar, Senegal, for their useful comments. Kyun Chang, Nathaniel Russel, and Ivan Torre provided valuable research assistance. We also thank the support from the Knowledge for Change Partnership multi-donor trust fund. The views expressed in this paper are the authors’ only. 2 Authors’ contacts: sahmed20@worldbank.org, mbussolo@worldbank.org; marciocruz@worldbank.org; dgo@worldbank.org; and iosoriorodarte@worldbank.org. 1. Introduction The global labor market is undergoing a fundamental transformation with important consequences on global inequality. In the decades ahead, younger and better-educated cohorts will enter the global workforce while older, less educated ones leave. With better education, workers have better skills (hereafter, skilled workers in this paper refer to those having nine or more years of education, and the two terms, skilled workers and better-educated workers, are used synonymously.) The new entrants of better-educated workers will come mainly from developing countries (figure 1). This paper analyzes the effects of this forthcoming demographic and education transition on global inequality using a general-equilibrium macro-micro simulation framework that covers harmonized household surveys for almost 120 developing countries. Our findings suggest that a more educated labor force in developing countries will contribute to a reduction in global income inequality by 2030. Figure 1 Future sources of educated and skilled working-age population, new entrants w.r.t 2015 600 500 High income countries Eastern Europe & Central Asia 400 Middle East & North Africa millions 300 East Asia & Pacific 200 Latin America & Caribbean Sub-Saharan Africa 100 South Asia 0 2020 2025 2030 2035 2040 2045 2050 -100 Source: Authors’ projections. Notes: Skilled is defined as workers with more than nine years of education. Projections are based on UN (2015) and education information from harmonized household and labor surveys from 117 countries. The impending change in the skill composition of labor is the latest wave of demographic and education trends that have been shaping the global labor market. In the first recent wave, a truly global labor market took shape almost all at once in the 1990s, when China, India, and the former Soviet bloc joined the global economy, increasing the size of the labor pool from 1.46 billion workers to 2.93 billion workers. This “great doubling” of global labor, as Freeman (2008 and 2007) called it, came from new entrants who were low skilled and low wage. The impact of this great doubling on global inequality is not immediately clear, as it depends on changes in the dispersion of incomes within as well as between countries. Freeman argued that China, India and the former Soviet bloc would not contribute much to the global capital stock, and therefore the 2 capital-labor ratio at the global level would be greatly reduced by about 60% of what it would have been without the great doubling. This factor ratio would cause an increase in the returns to capital vis-à-vis to wages and, since physical and financial assets are more concentrated than human capital, it would tend to push inequality up.3 Yet, other mechanisms also influence the results, so that global inequality fell over the first wave despite the increase in inequality within many countries (World Bank 2006 and 2007, Lakner and Milanovic 2016). The key facts about the second wave of upcoming changes in the global labor market are also striking: this time, a wave of skilled workers from developing countries will take center stage. On current trends, based on UN population projections (UN 2015) and present rates of educational enrollment (conservatively kept constant into the future), the world will see the number of skilled workers rising from 1.66 billion in 2011 to 2.22 billion by 2050, an increase of about 560 million or 33 percent. As in the case of the great doubling, the role of developing countries is crucial. Due to their investments in education and their growing populations, developing countries will contribute all of the additional workers to the world pool of educated workers (see figure 1). The number of skilled workers in high-income countries is projected to decline, from 603 million in 2011 to 601 million in 2030 and 594 million in 2050. Not exactly another great doubling, but still a dramatic change. While in 2011, each skilled worker in high-income countries was sharing the global market with two skilled workers in developing countries; by 2030 this ratio will be one to three. The increase in the supply of skilled workers will likely drive down the education premia these workers enjoy (other things being equal), and unlike the first wave, it may affect inequality within countries in a beneficial way. This kind of result has, for example, already been observed in developing countries in Latin America and the Caribbean (Cruz and Milet, 2017 and Lopez-Calva and Lustig, 2010). Note that, because of trade links, wages of skilled workers in high-income countries will also come under pressure even if their domestic supply will not be increasing. Signs of this pressure have also been already identified and attributed to trade by recent research (Autor et al. 2016). An important factor may, however, have a counteracting role. As in the past, technological progress may be skill-biased and thus offset the supply (education) effect. Research has shown that in high-income countries like the United States, the premia for skilled workers remained firmly high even during expansions of the skilled labor supply (Acemoglu 1998 and 2009). Conditional on what will happen on the demand for labor (i.e. on what happens to technological change), the possible decrease in skilled and sectoral earnings premia may have profound consequences for: i) international trade patterns; ii) global and local economic growth; iii) inequality across countries; and, as mentioned iv) inequality within countries (see Edwards and Hertel-Fernandez 2010; OECD 1999; Shimer 2001; Freeman 2007). 3 Indeed, several authors (e.g. Atkinson 2015, Bourguignon 2015, Galbraith 2012, Piketty 2014, and Stiglitz 2012) have called attention to the recent issue of rising capital share in total value added and to the concentration of income and wealth at the top of the distribution in many countries. 3 Our analysis focuses on forces of the educational wave that shape future supply and demand in the labor markets and their ensuing effects on global income distribution. On the supply side, it considers demographic shifts, improvement in education achievements, and policies that increase access to education and enable inter-sectoral mobility. On the demand side, it accounts for technological change, sectoral patterns of growth, and trade. It then draws the effects of these forces on global inequality by 2030, which is the target year of the Sustainable Development Goals (SDGs) as well as the World Bank’s goals of ending extreme poverty and boosting shared prosperity.4 Because it usually takes longer than 15 years for the stock of skilled workers to show significant improvement from the new inflow of younger and more educated workers reaching the labor market, the time horizon to 2030 will present only a partial effect of the education wave, hence a very conservative scenario. This paper uses two well-tested global economic models, the LINKAGE global general equilibrium model and the Global Income Distribution Dynamics (GIDD) microsimulation framework. As such, the approach allows for a systematic quantitative analysis of the global effects of educational attainment and demographic trends, which hitherto have been relatively neglected. For example, Ahmed et al. (2016) looks only at the demographic dividend of the Africa region. In looking at the effect of changes in skill premia and sector mobility of labor on income inequality, the approach relaxes the assumption of distribution-neutral growth found in other studies (e.g. Ravallion 2013) to analyze global poverty. Like the first wave, our results suggest that the next wave of education and demographic trends will also reduce global inequality between countries. Moreover, the second wave may ameliorate the overall increase in inequality within countries. Indeed, the baseline scenario suggests a decrease the Gini index for most countries relative to the case of no education wave.5 The remainder of this paper is organized as follows: Section 2 presents observed patterns of global inequality in the last decades and explains how the education wave may impact inequality through the labor markets. Section 3 presents the methodology, particularly the GIDD modeling framework and descriptive statistics of the underlying microdata. Section 4 shows simulation results along with robustness checks. The last section concludes with policy implications. 2. Recent changes in global inequality and labor force composition 2.1 Evolution of global inequality To analize the evolution of inequality across the world, Milanovic (2013) suggests three concepts of global inequality: 1) inequality across countries based on averages for GDP per capita or 4 The topic of this paper is directly related SDG 10 (Reduce inequality within and among countries). 5 These results, particularly within country, face several uncertainties regarding the possibility of disruptive technological changes related to automation and jobs’ polarization (Autor, 2015). 4 household income; 2) inequality across countries based on population-weighted GDP per capita or household income; and 3) global inequality using household-level data from surveys. The first two concepts refer to inequality between countries. Relative to concept 1, inequality as measured by concept 2 reflects the weight of relative population size in addition to discrepancies in average incomes. The within-country component of inequality is nonetheless critical. It is possible, for instance, that a low-income country could catch up on average per capita income at the expense of rising inequality among its residents. This effect will neither be captured by concept 1 or concept 2. Among the three concepts of global inequality, the individual-based concept 3 is the only one that takes into account inequality between as well as within countries. It requires, however, a large amount of harmonized micro data to perform any attempt for international comparison. In this paper, we analyze global inequality using the individual-based (concept 3) approach. From 1950 to 2000s, inequality between countries using concept 1 increased, while inequality using concept 2 decreased. However, during the last decade, both concepts show a dramatic reduction in inequality (Milanovic 2013). The individual-based inequality (concept 3) also showed a decreasing trend between 2000 and 2010 (see Milanovic 2013; Lakner and Milanovic 2016), but at a slightly smaller reduction when compared to the pattern of concepts 1 and 2. Since concepts 1 and 2 are about between-country inequality, the individual-based result somewhat suggests that within-country component of inequality may have increased in this period. Lakner and Milanovic (2016) show that (individual-based) global inequality, as measured by the Gini index, dropped from 72.2 in 1988 to 70.5 in 2008. This recent decline in global inequality can be largely explained by the economic progress during the last decades in low- and middle-income countries, particularly by the sustained growth of populous countries like China and India.6 Increased trade benefited low-skilled workers in China, India and partly from the Soviet bloc. Demand for the goods they produced went up, and so did their wages. The opposite happened for low-skilled workers in high-income countries. Also, increased global economic integration accelerated diffusion and adoption of technology, which in turn supported the economic boom in developing countries before the recent Great Recession.7 The resulting economic prosperity reduced the gaps in income per capita between high-income and developing countries, which was the main driver of the reduction in global inequality. Notwithstanding this recent reduction, the world as a whole remains very unequal. In fact, if the world were a country it would have a highest Gini in 2008. Also, global convergence has been accompanied with increasing national divergences. Lakner and Milanovic (2016) estimates that the Gini indices have increased for many countries and regions from about 1988 to 2008: 38.2 to 41.9% for mature economies; 32.0 to 42.7% for China; 31.1 to 33.1% for India; and 53.5 to 58.3 6 World Bank (2011) and Bussolo et al (2012). 7 See, for examples, World Bank (2011) for the multipolarity of global growth and Arbache et al. (2010) for the case of Africa. 5 for Sub-Saharan Africa. Another study, Alvaredo et al. (2013), finds that the share of income held by the top 1% in the USA rose dramatically. 2.2 The engine of the education wave: Education and demographic transitions by region Looking forward, what does an increase of skilled labor supply driven by developing countries mean for global inequality? The first step in the analysis of the impact of the forthcoming education wave is to assess the geographically diverse increases in the supply of skilled workers. These increases depend on two key factors: the intergenerational education gap – i.e. the difference in education attainments of the old cohorts that are exiting the labor market vis-à-vis the young entering cohorts – and the intergenerational demographic gap – i.e. the difference in population size between the young and the old cohorts. Large intergenerational education and demographic gaps, and thus a large increase in the supply of skilled workers, are observed for countries that have boosted their investment in education and that have still high population growth, i.e. a large share of young people. Conversely, aging countries where most of their population has already entered advanced secondary and tertiary education will not experience increases in the supply of skilled workers. Figure 2: Education transition by region a) ECA, LAC, SSA, and High-income b) EAP, MENA, and South Asia Education gap Education gap between age groups 20−24 and 60−64 between age groups 20−24 and 60−64 6 6 5 difference in years difference in years 4 3 4 2 Europe & Central Asia 2 Latin America & Caribbean East Asia & Pacific Sub−Saharan Africa Middle East & North Africa High Income South Asia 0 1 1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 Source: Barro Lee Source: Barro Lee Source: Authors elaboration based on Barro and Lee (2013) Note: Old cohort refers to the population between 60 and 64 years old. Young cohort refers to the population between 20 and 24 years old. 6 Figure 2 summarizes the educational transition by region from 1960 to 2010. It shows the remarkable and diverse developments in education attainment across these regions. The average years of schooling for the young cohorts (20-24 years old) has improved when compared to the old cohorts (60-64 years old) in all developing regions except Eastern Europe and Central Asia. In the 1960s, young workers in Latin America and the Caribbean, Middle East and North Africa, South Asia, and Sub-Saharan Africa had about one additional year of education than their contemporary older workers. By 2010, this difference had increased to about four or more years of education, or the equivalent of almost a full cycle of secondary education. This trend is marginally different for countries in East Asia and the Pacific which started with a slightly higher intergenerational gap and progressed a bit more slowly. The lines in Figure 2 are regional averages, and there is some variation across countries within each region. In Latin American and the Caribbean (LAC), for example, Bolivia, started with an intergenerational education gap of 3.5 years and, by 2005, the gap rose to over five years. This trend is followed by Brazil, Mexico, and Chile. These increases are evident both in average years of schooling as well as in tertiary schooling. In Bolivia, the difference in the average number of years of tertiary education between young and old was initially close to zero, but the difference rose to about a quarter of a year by 2005. For Benin, in Sub-Saharan Africa (SSA), the difference in the average years of schooling for 25-29-year-olds as compared to 60-64-year-olds was almost non- existent (0.51 years vs. 0.48 years) in 1980. By 2010, this difference rose to 1.5 years of schooling.8 Indeed, this trend is followed by other countries such as Rwanda, Gambia, and Kenya. Most of the gains in rising educational attainment in Africa, however, are in primary and secondary education. There has been little progress in increasing the rates of tertiary education and beyond (except in a few countries, like Gabon). Because China and India have been major drivers of recent declining global inequality as well as contributors to the global labor force, a pressing question is whether future generations will be able to supply enough skilled workers. In China, the overall rate of educational attainment and the supply of young workers have stabilized. These numbers are reflected in Figure 2 in the performance of East Asia and the Pacific, where the average years of schooling for the young cohorts had improved significantly up to the 1980s, with a leveling off of the intergenerational gap after that. India’s educational attainment levels have been rising, although starting from a lower level than China’s. The difference in years of schooling between young and older cohorts in India was 1.7 in 1980 and 3.0 in 2005 –with only a small increase in tertiary education, which looks weak in comparison to other middle-income countries. This observation is consistent with the large gap in average years of schooling observed in South Asia (SAR). Two regions stand out because of their very different historical performance: the high- income countries group and Europe and Central Asia. Both regions start off in the 1960s with a higher (vis-à-vis developing regions) gap between young and old schooling achievements, due to 8 The direction and magnitude of these differences are robust to wider cohort definitions. 7 their earlier investment in education. This gap increases in the first two or three decades, but then decreases and, towards the end of the period, the gap has almost disappeared. This trend should not be considered as lack of progress. On the contrary, once all young people achieve a desired (high) level of schooling, and this is maintained for all new cohorts, the difference in education between old and the young would disappear. However, an intergenerational education gap close to zero means that the region-wide average education will not increase. Some countries, notably the Russian Federation, are even experiencing a negative intergenerational gap in education: older cohorts tend to have more years of education than the young. Once the old workers retire, average education may go down. Assuming that for each old worker exiting the labor market a new, young replaces him or her, the average education level of the labor force would increase in line with the trends shown in figure 2. In contrast, some high-income OECD and East European countries are either in a stable or even a downward trend in terms of educational attainment. Therefore, this education wave will not be uniform across regions, and developing countries will play the leading role in this process, especially those with demographic bonuses. For this reason, it is critical to add education by age cohorts, as well as important characteristics such as gender, as demographic dimensions in long- term prospective economic analysis (see Lutz, Goujon, and Doblhammer-Reiter 1998; Lutz and Goujon 2003). Remarkably, the largest gap on years of education between generations is currently in regions that will contribute most to the growth of global working age population, namely LAC, Middle East and North Africa (MENA), SAR, and SSA. Moreover, the average years of schooling of young cohorts in these regions, still lag behind high-income countries (HIC), particularly SAR and SSA—even without getting into important issues related to quality. Results from the PISA test (OECD, 2015) suggests that the gap regarding education in developing countries is not only related to school attainment but more importantly to the quality of education when comparing the performance of students using standard tests. While 83% of students from households in the bottom quintile of income distribution in Singapore demonstrate the basic competencies in math, the corresponding numbers are below 20% for students from households in the bottom quintile of income distribution in Latin American countries (e.g. Colombia, Brazil, and Peru). In addition to the younger generation becoming more educated than the older one, the magnitude of its effect on the supply of skilled workers is enhanced by a growing cohort’s size (of that working-age group). In a similar way to the previous figure, Figure 2 shows the intergenerational demographic gap. This is expressed as the ratio of the population size of two age cohorts: the young, aged 20 to 24 and the old aged 60 to 64. Note that these ratios vary both in levels and trends. For advanced economies, and ECA countries the ratio at around 2 is low and slightly declining in the period 1960-2010. This means that new younger cohorts are still larger 8 than older ones, but since the education level of the young is not much higher than that of the old, average education is not expected to increase.9 Figure 3: Demographic transition by region a) ECA, LAC, SSA, and High-income b) EAP, MENA, and South Asia Population share ratio Population share ratio between age groups 20−24 and 60−64 between age groups 20−24 and 60−64 6 5.5 Europe & Central Asia East Asia & Pacific Latin America & Caribbean Middle East & North Africa Sub−Saharan Africa South Asia 5 High Income 5 Ratio Ratio 4 4.5 3 4 2 3.5 1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 Year Year Source: Authors elaboration based on Barro and Lee (2013) An interesting region is LAC, where countries are beginning their aging process as the intergenerational demographic ratio shows a kind of inverted U in Figure 2, but this is compensated by the rapidly increasing educational gap. A similar pattern is observed for EAP, with a decisively stronger aging trend. Figure 2 shows that the other regions – MENA, Sub-Saharan Africa, South Asia – are in earlier phases of the demographic transition. Their increasing demographic ratios will boost the impact of the intergenerational education gap and push up average education level strongly. population will be shaped by a younger and more educated cohort coming from developing countries, especially regions that are lagging behind on economic growth and per capita income level. Figures 2 and 3 can be extended to the future using population projections and some simple assumptions on education of future young cohort. Indeed, going forward, the proportion of skilled workers will increase about 43.7% in low- and middle-income (LMI) countries between 2015 and 2050; it will decline approximately 3.3% in high-income countries during the same period. Moreover, the share of the population from LMI countries was approximately 82.7% in 2015. By 2050 they will represent about 85.7% of total population. 10 Figure 1 in the introduction summarizes how the increasing in the number of workers with more than nine years of schooling will be predominantly driven by countries in LMI regions. The forthcoming wave of changes in the global labor market will, therefore, come from the combination and growing intensity of the demographic and education transitions just described. 9 Ahmed et. al. (2016) provides further description of different trends of demographic change across the word. 10 United Nations (2015), medium variant scenario. 9 As the global population ages in the future, younger and better-educated cohorts will enter the workforce while older, less educated, ones leave. The average schooling of the working age population will tend to increase, even without any additional efforts to improve educational attainment rates – a pipeline effect or natural progression as students move up from one educational grade to the next over time. The larger intergenerational education gap in developing countries, combined with the large size and growing pool of younger cohorts relative to the older ones in those countries, forms the key and significant dynamics of the second wave that is analyzed in this paper. In the next section, we describe the methodology to quantify the effects of this and other related factors on global inequality. 3. Methodology 3.1 The global economic models The global distributional impacts of the education wave depend on changes in per capita incomes between countries and changes in inequality within countries. To capture the full – between and within countries – distributional change, one needs a framework that captures both convergences at the macro level (country averages) and the evolution of factor markets at the micro level (dispersion). This paper adopts a dynamic global microsimulation approach which allows taking into account both these average and dispersion effects. The origin of dynamic microsimulation can be traced back to the 1950s seminal work of Orcutt (1957) whose contributions aimed at overcoming the limitations of models available at that time. These models could be used to predict the aggregate impact, but could not describe the distributional impact of policy reforms or the effects on inequality of long-term trends, such as demographic change. Data availability and modeling have significantly advanced since then, but dynamic microsimulations remain the main tool to study distributional change, and still, provide the unique perspective of projecting samples of population forward in time (simulating different scenarios). This paper uses the World Bank’s global microsimulation framework labeled Global Income Distribution Dynamics (GIDD) model in combination with the global computable general equilibrium (CGE) model called LINKAGE. Both tools have been described in detail in other papers (e.g. Bourguignon et. al. 2008 and Bussolo et. al. 2010; the full technical documentation of the LINKAGE is provided by van der Mensbrugghe, 2011), but it is still useful to briefly describe their structure in this section. The ultimate focus of analysis is household welfare, using household per capita income as its indicator. The distribution D of household per capita income y in year t can be expressed as the product of the joint distribution of all relevant household or individual characteristics X and the distribution of income conditional on these characteristics: 10 ( )= = … ( ) ( | ) ( ) (1) where ( ) is the density function of the distribution of income and the summation is over the domain C(X) on which X is defined. Household per capita income (y) can be modeled as a function of: (i) household members’ characteristics or endowments (X), (ii) the market reward for those characteristics ( ), and (iii) the intensity in how those endowments are used as captured by a set of parameters defining labor force participation and occupation status ( | ); and, finally, (iv) unobservable components ( ): , = , , , , , , (2) The income distribution D for a population of N individuals or households in year t can be represented by a vector , … , … , , where each Yit can be defined as in (2) in terms of endowments, prices, labor status and unobservables: = , … , = , , , , , , … , , , , , , (3) How does this distribution change dynamically, for example from year t to year t+k? This framework allows distinguishing two sources that affect the dynamic change of distribution D, both of which are relevant for the assessment of the distributive impact of the education wave. The first source consists of the changes in either the parameters or , namely the market rewards to the characteristics X and parameters affecting occupational decisions. This means, for example, that inequality for distribution D can go down if the skill premia / is reduced; or if a change in labor demand in sectors with higher wages (a change in ) affects the decision to move to these sectors for some individuals working in sector with lower wages. The second source of dynamic shift is represented by changes in the distribution of individual and household characteristics (X). Alterations of the structure of the population in terms of age and education, and changes in the size and composition of households, will all affect the distribution of income of that population.11 Both sources of distributional change matter to the impact of the education wave. In fact, the GIDD allows generating a scenario that includes the education wave, which is then compared to a counterfactual where education achievements remain stable at the levels observed in t. _ _ Comparing the distributions and derived from the two scenarios in effect isolates the distributional impact of the education wave. Defining the contrasting values of endowments, prices, and labor status to build the two s can be quite challenging, especially when done globally for many countries. To do so, the functional form of equation (2) is defined in a simple fashion using only independent variables available for all countries in the sample. In this 11 These two sources of dynamic change are not independent one from the other and, in the real world, they are simultaneously determined. The problem of estimating and running a fully simultaneous microsimulation framework is overcome by making some simplifying assumptions. 11 paper, the right-hand side of equation (2) includes age, education endowments and sector of employment workers (country subscripts excluded for simplicity), so equation (2) can be re- written as: , = + , ( , , )+ , ( , , )+ , ( , , )+∑ , , + , (4) where and are dummy variables identifying whether workers are skilled or unskilled, respectively; and are dummy variables taking the value of 1 if the worker is employed in the agricultural sector or in the non-agricultural sector, respectively; captures the proportion of household members in each of the k age cohorts. The s are rewards (prices) to education endowments conditional on the sector of employment,12 and the s are prices associated with household composition. Finally, includes all other income determinants not included in equation (4). The counterfactual expression to (4) for year t+k is: , = + , , , , + , , , , + , , , , +∑ , , + , (5) where the demographic characteristics, endowments, and returns to these endowments have been modified in accordance with the counterfactual scenarios of the education wave and no wave. Using (5) – and the relevant simulated S, E, F, s, and s – the two distributions _ _ and can be generated. To close the model, one needs to figure out how the characteristics, represented by the various dummies, evolve through time as well as how the rewards, represented by the s, change in the scenarios. Characteristics and rewards should be jointly determined; however, this is not the case for the GIDD, nor for the majority of microsimulation models.13 In practice, the GIDD implements the sequential approach described in Figure A1 in the appendix. In the first step, changes in the number of skilled and unskilled workers are estimated on the basis of demographic change and the pipeline effect described before.14 In a second step, these aggregates are used as: (i) constraints for the reweighting procedure that is applied to change the distribution of characteristics at the household level; and (ii) as inputs for the LINKAGE general equilibrium model. The equation here below summarizes the reweighting procedure: 12 Note that unskilled workers employed in agriculture, i.e. when these dummies , , are equal to one, are the reference category, so they are excluded from the equation. 13 See Bourguignon and Bussolo (2011) for a discussion. 14 These changes are assumed to be exogenous and not affected by the rewards. Bourguignon and Bussolo (2011) describes more complex structures, where changes in characteristics affect the rewards, and then these affect the supply. [redraft] 12 , = a , , , = , ( ) , and , , the supply of skilled or unskilled workers in year t+k, are the constraints of the reweighting procedure which consists of finding multipliers ai,s,t+k that change the original (i.e. for year t) household weights wi,s. Once these multipliers are determined, the dummies , can be recomputed and used to calculate a ‘partial’ change in the distribution. Formally, equation (5) can be re-written as: ′, = + , , ′, , + , , ′, , + , , ′, , +∑ , , + , (5’) This new equation (5’) represents a change in the incomes, and thus in the distribution, that is due to a pure demographic (or quantity) education effect. Other characteristics and rewards have not been changed yet.15 The change in rewards, in this specific case, the change of the skill premia, is determined in the LINKAGE general equilibrium model. This model calculates wages of skilled and unskilled workers by combining firms’ demand for labor with the aggregate change in supplies of these two types of workers. In addition to the changes in rewards, the computable general equilibrium model also calculates overall economic growth, sectoral reallocation of labor, and a new vector of consumer prices. Changes of these variables – the Linkage Aggregate Variables (LAVs) – are used as inputs in the final step of the microsimulation. This step consists of changing the s in equation (5’).16 3.2 The micro data underlying the models and the analysis The methodology combines a large data set from three different sources, which we describe further to highlight the innovation. First, we use population projections from UN (2015). Second, we use input-output matrices from the Global Trade Analysis Project (GTAP, 2011) for the CGE scenarios. Third, we use the GIDD database, which is composed by harmonized household surveys 15 Note that the constraints of the reweighting procedure are in terms of (age cohorts and) education levels of the population in year t+k, and not in terms of sectoral employment. However, because skilled workers may initially be more concentrated in the non-agricultural sectors, increasing the weight of skilled workers in the total population generates also an increase in the share of non-agricultural employment. This is why, in equation (5’), sectoral dummies are labelled as ′ , ′ , with a prime (’) sign. These do not necessarily represent the final sectoral employment which is determined by the CGE model. 16 In this last step, workers move across sectors to achieve the proportions of employment in agriculture and non- agriculture calculated by the CGE model. Note that these inter-sectoral movements are net of the sectoral shifts already generated by the reweighting procedure as mention in footnote 15. The microsimulation procedure to select which specific worker moves is based on a probabilistic model described in detail in Bussolo et al (2010). 13 for a large among of countries. The GIDD dataset is derived from household surveys harmonized by the World Bank.17 It provides a cross section of surveys using 2011 as a base.18 The sample covers 10.45 million individuals in 127 countries, constituting 83.4% of the global GDP and 86.3% of the global population19 (see Table 1). The coverage of the data in GIDD is also large for most specific regions, both in terms of population and GDP. Table 1: GIDD database coverage GDP PPP$, B Population, M Microdata Region Total % Total % obs. Low and middle-income countries 13,781.5 90.8 5,788.3 87.8 5,320,925 East Asia & Pacific 5,698.5 96.5 1,991.8 94.6 918,891 Europe & Central Asia 1,287.5 80.9 269.8 67.0 347,714 Latin America & Caribbean 3,379.2 98.2 585.6 98.0 1,472,436 Middle East & North Africa 843.6 19.5 344.1 19.4 70,402 South Asia 1,703.2 99.3 1,675.0 98.2 782,726 Sub-Saharan Africa 869.6 92.3 922.1 79.4 1,728,756 High-income countries 40,979.1 81.0 1,277.2 79.6 5,134,448 World 54,760.6 83.4 7,065.5 86.3 10,455,373 Source: Authors’ elaboration using available data from GIDD-dataset. From the household surveys, we extract information on individuals’ age, employment status, sector of activity, years of schooling and wage as the starting points to project the population growth by school attainment and the sectoral wage bill database (Cruz, Go and Osorio- Rodarte, 2017). These pieces of information feed into the LINKAGE model to establish consistency between the skill definition and volume of workers between the CGE model and the microsimulations.20 Table 2 shows the share of workers by broad sectors (agriculture and non- agriculture) and skill level for 2012. The tabulation indicates that the share of skilled workers is much larger in high-income countries and the region of Europe & Central Asia (ECA). The three regions of East Asia & Pacific (EAP), South Asia (SAS), and Sub-Saharan Africa (SSA) together had approximately 96% of the people living below poverty line in 2015. In all of these regions, 35% or more of the paid workers were unskilled working in agriculture. An important channel through which changes in the composition of labor supply may affect income inequality and income distribution, in the long run, is the movement of workers across different categories delineated by the cells - agriculture, non-agriculture, skilled workers, and unskilled workers. 17 The data include household surveys harmonized by the Poverty Group Global Practice of the World Bank and the International Income Distribution Database (I2D2). The I2D2 is a World Bank project to collate, harmonize and make accessible comparable information from household surveys held by the World Bank, for poverty, inequality, education, demographics, and labor market analysis. For further details on I2D2, see Montenegro and Maximilian (2008). 18 For each country, the closest survey to 2011 has been chosen to be consistent with the database used in the LINKAGE model (Global Trade Analysis database v.8). From this dataset we derived the GIDD Sectoral Wage Bill database (Cruz, Go, and Osório-Rodarte, 2017). 19 If Japan and Switzerland are included with aggregated information generated from household surveys, using the Luxemburg Income Study data set, the GDP coverage increases to approximately 93%. 20 One issue is the comparability of school attainment to define skill-level because of differences in quality of education across countries. However, the differences of wages by skill, sector, and country indirectly account for the differences in respective productivity. 14 Table 2: Share of workers by region, sector of activity and skill level (percent) Agricultural (%) Non-agricultural (%) Workers, millions Region Skilled Unskilled Skilled Unskilled Skilled Unskilled Total Low & middle-income countries 3.5 28.0 28.6 39.9 637 1,351 1,989 East Asia & Pacific 2.7 29.1 23.4 44.8 259.8 737.1 997.0 Europe & Central Asia 4.9 6.9 66.4 21.8 48.9 19.7 68.5 Latin America & 1.7 10.2 51.5 36.5 134.5 118.0 252.5 Caribbean Middle East & North 0.2 19.0 19.4 61.5 2.6 10.6 13.2 Africa South Asia 6.2 36.1 24.4 33.3 145.2 329.9 475.1 Sub-Saharan Africa 3.0 33.9 22.3 40.8 46.1 136.2 182.3 High-income countries 2.0 0.8 87.4 9.9 416.0 49.4 465.3 World 3.2 22.8 39.7 34.2 1,053 1,401 2,454 Source: Authors’ elaboration using available data from GIDD-dataset. Table 3: Average wage and skill premia by region (2011 US$ PPP) Wages, monthly US$(PPP) Labor, millions Skill Wage Premia Region Agri Non-Agri Agri Non-Agri Non- Unskilled Skilled Unskilled Skilled Unskilled Skilled Unskilled Skilled Agri Agri Total East Asia & Pacific 152 244 281 409 2.2 0.4 26.1 9.7 1.6 1.5 1.5 Europe & Central Asia 281 282 380 523 0.2 0.2 0.6 1.8 1.0 1.4 1.4 Latin America & Carribean 223 492 359 786 1.1 0.2 3.3 4.9 2.2 2.2 2.4 Middle East & North Africa 205 296 257 293 0.2 0.0 0.4 0.1 1.4 1.1 1.2 South Asia 93 124 108 230 7.8 1.5 7.5 4.9 1.3 2.1 2.1 Sub-Saharan Africa 72 155 138 327 4.2 0.5 3.2 2.1 2.2 2.4 2.9 High-Income Countries 685 1,129 1,037 1,849 0.1 0.3 1.9 14.6 1.6 1.8 1.8 Low-Income Countries 70 113 107 195 3.2 0.3 3.3 1.3 1.6 1.8 2.0 Lower Middle- Income Countries 98 130 158 268 10.5 1.9 11.7 8.2 1.3 1.7 1.9 Upper Middle- Income Countries 237 420 303 581 1.8 0.5 26.1 14.1 1.8 1.9 1.9 High-Income Non- OECD 361 595 678 943 0.0 0.1 0.5 2.6 1.6 1.4 1.4 High-Income OECD 782 1,311 1,153 2,045 0.1 0.3 1.5 12.0 1.7 1.8 1.8 World 114 289 281 987 15.8 3.1 43.0 38.1 2.5 3.5 4.0 Source: Authors’ elaboration based on GIDD-dataset Note: Wage skill premia are calculated by dividing average wages of skilled to unskilled. This table aggregates workers across countries and regions. 15 Table 3 shows that there are significant differences in the average wages of workers (conditional on their level of skill) between sectors and regions.21 An average unskilled worker in non-agriculture in high-income countries has a wage that is significantly higher than an average skilled worker in a non-high-income economy. Also, skill premia in the wages by sector are sharply characterized. 22 3.3 The scenarios Using the methodology just described, we first define the “education wave” scenario as the baseline or reference case. On current global trends, the education wave foretells substantial changes in the global labor market. Around 2012, one skilled worker from a high-income (OECD) country was competing, in the global labor market, with two skilled workers from developing countries. In less than a generation, by 2030, that ratio will be 1 to 3, confirming that the global pool of skilled workers will originate mostly in developing countries as a direct consequence of the education and demographic transitions (as discussed in section 2). This trend is, in essence, the global shock of the education wave that will likely transform once again, as the ‘great doubling’ did in the first wave, the world’s labor market. Digging beyond this global picture, the education wave will also affect distribution within countries, a local shock that will vary from country to country. Figure 4 illustrates the geographic dispersion of the education wave with the growth rates of skilled labor (y-axis) against those for unskilled labor (x-axis). Most of the dots, which represent countries or regions, are above the 45- degree line, meaning that skilled workers will grow faster than that of unskilled workers. Yet, there is heterogeneity across countries. First, the growth rate of unskilled workers is often negative in ‘older’ countries and region. Compare, for example, the reduction rates of 30% or 15% for unskilled populations in Europe or China, versus the expansion rates of 53% for Nigeria, or 65% for the Sub-Saharan region. Second, very few countries or regions experience growth rates that are similar for the two groups of workers (that is, no education wave): the only dots in the figure that are (reasonably) close to the 45-degree line are those for the US, Russian Federation, and Sub-Saharan Africa. For most of the other countries, however, the differences can be quite large. In Turkey, the gap between the rates of expansion of skilled and unskilled populations is of almost 50 percentage points. For Brazil and Nigeria, the gap is close to 40 percentage points, for Indonesia, India, China is between 25 and 33 percentage points. The Russian Federation is the only country found below the 45-degree line, albeit close to it. In this country, older generations tend to have more education than the younger ones. This fact, combined with 21 Heterogeneity is large within regions and within countries, which are accounted for in the components of global income inequality. 22 The wages for skilled workers in agriculture are higher than the wages for unskilled workers in the same sector. The same happens in non-agriculture activities. 16 an aging population, generates a rate of reduction of skilled workers slightly higher than that of unskilled workers, resulting in a (negative) gap of 3 percentage points. Figure 4: Growth rates of labor by skill, sector and country/region 90.0 NGA SSA 70.0 MNA 50.0 TUR IDN IND SAS LAC MEX Skilled BRA ZAF 30.0 LKA EAP CHN 10.0 ECA USA YHI -10.0 EUR RUS -30.0 -30.0 -10.0 10.0 30.0 50.0 70.0 90.0 Unskilled Source: Authors’ calculations based on GIDD projections. Note: the growth rates are expressed as the cumulative growth for the period 2030-2012. The red dots represent regions; the blue dots are individual countries. These changes in the labor supply of different education levels constitute the exogenous shock defined in the baseline scenario. While labor supply is likely to respond to wage changes, the strong assumption that labor supply is exogenous and that it follows demographic developments is adopted here. In fact, most economic models likewise treat population growth as exogenous and, in this exercise, labor supply by education level is indirectly derived from the aging of the population and by assuming that the education attainment of the young cohorts in the future will adhere to the same attainment rate observed in the current data. This assumption is conservative because some countries will likely do better. That is, their education policy may bring about better educational attainment among the young generations in the future when compared to those in the present period. The education wave, if such improvement is accounted for, would be even larger than what is currently defined in the baseline. Table 4 summarizes the scenarios used for this analysis. Against the baseline (education wave), a “no education wave” is set up as the main counterfactual simulation, which is a what-if scenario in which the education trend of the baseline is removed. Two additional scenarios aim to address potential concerns related to technological changes that may be biased towards skilled 17 workers. Since the baseline and what-if counterfactual simulations contain similar underlying economic projections and parameters, their differences minimize the effects of common assumptions and isolate the effects of the critical factor of each counterfactual simulation. The simulation period is for a 20-year span, from 2011 to 2030, which makes it more comparable to the first wave. This time horizon also makes the analysis more conservative because it will not cover the full impact of the education wave beyond 2030. Table 4: CGE baseline and counterfactual scenarios Scenario Key features Purpose Baseline – education wave Population projections from UN Establish a business-as-usual or (2015) medium fertility variant reference case for comparison scenario; economic growth with counterfactual scenarios. projections from World Bank (2015); the share of skilled workers grows assuming constant education attainment rates. No education wave Same as baseline except for the Provide a counterfactual scenario fact there are no changes in the without the “education wave.” share of skilled workers. Higher elasticity of The baseline plus a higher This scenario tests the sensitivity substitution between elasticity of substitution of 3 of the results for changes in the skilled labor and capital (previously 2) between skilled substitutability between skilled labor and capital. labor and capital to mimic the consequence of a biased technological change Alternative nesting The baseline with an alternative This scenario tests the sensitivity structure of labor and nesting structure. Skilled and of the results for technological capital unskilled labor are substitutable changes that may affect the with each other, and together also substitutability between skilled substitutable with capital. We use and unskilled workers. elasticities of substitution of 2 for the two relevant nests. Source: Authors’ elaboration. 4. Results 4.1 Results at macro level: Growth differentials and wage premia In the simulations, the productivity factor for each country is set exogenously to replicate the evolution of economic growth to 2030 based on the World Bank projections reported in the Global Monitoring Report 2015/16 (World Bank, 2015). The catching up of productivity of developing countries to the levels of high-income countries generates faster growth rates for the developing countries. The growth differential vis-à-vis high-income countries is quite pronounced for the East Asia and the Pacific region as well as for South Asia region. These differentials bring 18 about convergence in incomes per capita and are a key driver for the reduction of the between- country component of global inequality. The increasing of the working age population shares in many developing countries, vis-à-vis its reduction in high income countries also contributes towards this convergence (Cruz and Ahmed, 2016). Figure 5: Projected economic growth by geographical region Annual average growth rates of income per capita (2012-2030) 9.0 8.7 8.0 7.0 6.0 Percent 5.0 4.4 4.0 3.3 3.1 3.0 2.6 2.7 2.5 2.0 1.0 0.0 East Asia & Europe & Latin America Middle East & South Asia Sub-Saharan High Income Pacific Central Asia & Caribbean North Africa Africa Source: CGE and GIDD simulations results. Is the education wave contributing to the reduction of global inequality from the perspective of this between-country convergence? The answer is yes, but it does not seem much in the aggregate as shown in Figure 6. The contribution of the education wave – vis-à-vis a scenario where skilled and unskilled workers grow at the same rate – comes exclusively from the higher productivity of skilled workers who are becoming more abundant in the wave scenario. However, other drivers of growth and convergence, namely the catching up of productivity and supplies of other factors (land and capital), do not change across simulations. For these reasons, the education wave boosts convergence by, at best, about one percentage point in East Asia and the Pacific and Latin America and the Caribbean regions, but its impact is more limited in other regions. The convergence differentials should be considered a lower bound, given that more abundant skilled workers are likely to generate externalities and an additional productivity uptick that is not simulated here. In addition, we have very conservative education projections and we a looking at a relatively short period of time to fully capture the effects of the education wave. 19 Figure 6: Incomes convergence - education versus no education wave 50.0 48.1 49.3 Incomes per capita as % share of high 40.0 36.2 income countries (2030) 35.1 33.0 33.7 30.0 20.0 13.6 13.9 11.5 11.7 10.0 8.1 8.3 0.0 East Asia & Europe & Central Latin America & Middle East & South Asia Sub-Saharan Pacific Asia Caribbean North Africa Africa no-wave wave Source: CGE and GIDD simulations results. One final point on the regional aggregate growth rates. The education wave, as discussed above, is a scenario where the growth of the global pool of skilled workers is almost entirely due to the expansion of education in developing countries. Indeed, by 2030 for all developing countries incomes per capita are higher in the wave versus than in the no-wave scenario, while for the rich countries incomes are unchanged. Further to the rising inequality in high-income countries observed by several authors (see footnote 4), the micro results below suggest that increasing pressure on the middle-income classes of high-income countries will continue. Changes in the skill and sector premium of wages are the other crucial macroeconomic results of the simulations that affect global inequality through its effects on incomes dispersion within each country. Wages will follow the interaction of the demand and supply of labor by skill and sector as well as the economy-wide repercussions of other factors in general equilibrium. For each country, a higher supply of skilled workers in 2030 relative to 2012 will lead to a larger reduction in the skill premium (Figure 7). The regression that relates the relative size of the skill premium against the relative supply of skilled labor has a well-defined negative slope. The causality in this framework runs from the changes in the quantities of workers to the changes in the wages because all labor supplies are exogenous. Turkey, Brazil, and South Africa are amongst the countries with the largest drops in the skill premium. In Turkey, the skill premium by 2030 is reduced by 20 percent of what it was in 2012. At the opposite end, the premium increases by about 5 percent in the Russian Federation. In general, the skill premium will tend to decrease across countries. 20 Figure 7: The skill premium of wages versus supply of skilled labor 2030 relative to 2012, in percent change, all sectors Skill premia increase Skill premia decrease Source: GIDD simulations. Note: Relative skill premium changes are calculated as: [(Wage_skill_2030/Wage_unsk_2030) / (W_skill_2012/W_unsk_2012) – 1] x 100, and the same formula is used for relative labor supplies. Individual results are presented for selected countries: Brazil (BRA), China (CHN), India (IND), Indonesia (IDN), Japan (JPN), Mexico (MEX), Nigeria (NGA), Russia (RUS), South Africa (ZAF), Sri Lanka (LKA), and the United States (USA). Other countries are aggregated by regions, such as Sub-Saharan Africa (SSA) and European Union (EUR). For other regions, we follow the World Bank classification, including East Asia and Pacific (EAP), East Europe and Central Asia (ECA), Latin America and the Caribbean (LAC), Middle East and North of Africa (MENA), and South Asia (SAR), combined to the demographic typology described in Ahmed (2016), which divide countries according to their potential for demographic dividend by 2030, such as as: early-dividend (ed), late divided (lt), and post-divided (pd). This disaggregation is based on the CGE macro- simulation. Changes in the premium of wages by sector also affect distributional shift. By sector, we refer to urban (non-agriculture) and rural (agriculture) areas. In high-income countries, both skilled and unskilled workers are perfectly mobile across sectors so that a single, skill-specific, economy-wide wage clears each labor market. In developing countries, on the other hand, skilled workers are perfectly mobile while unskilled workers are segmented by the two broad sectors. A single wage clears for skilled workers clears across sectors but not the case for unskilled workers. In the latter case, a migration function that allows unskilled workers to move from the rural to the urban sector defines the supplies of unskilled workers across sectors, in the tradition of Lewis (1954) and Harris and Todaro (1970). Interactions with the sectoral demands for unskilled workers then determine the wages, with a premium prevailing for urban (non-agriculture) unskilled workers relative to their counterparts in the rural (agriculture) sector. Unlike the skill composition of labor, the sector distribution of unskilled workers is endogenous. 21 As each country urbanizes, production and income growth, as well as the allocation of unskilled workers, will all shift towards non-agricultural sectors.23 Even so, the urban premium for unskilled workers will still tend to go down due to the relative intensities of factor use in the production function across sectors. That is, agriculture, which is more intensive in the unskilled factor, will release many more unskilled workers than those needs by the expanding non- agriculture sectors, resulting in a relative oversupply of unskilled workers, driving down the premium for unskilled workers in the urban sector. Figure 8 depicts the reduction of the urban premium of wages in 2030 relative to 2012 against the relative sector supply of unskilled workers. Figure 8: The urban premium of wages and the sector supply of labor for unskilled workers, 2030 relative to 2012, in percent change Source: GIDD simulations. Note: Relative urban premium changes are calculated, just for the unskilled, as: [(Wage_non-agri_2030/Wage_agri_2030) / (Wage_no-agri_2012 / Wage_agri_2012) – 1] x 100, and the same formula is used for relative employment. Individual results are presented for selected countries: Brazil (BRA), China (CHN), India (IND), Indonesia (IDN), Japan (JPN), Mexico (MEX), Nigeria (NGA), Russia (RUS), South Africa (ZAF), Sri Lanka (LKA), and the United States (USA). Other countries are aggregated by regions, such as Sub-Saharan Africa (SSA) and European Union (EUR). For other regions we follow the World Bank classification, including East Asia and Pacific (EAP), East Europe and Central Asia (ECA), Latin America and the Caribbean (LAC), Middle East and North of Africa (MENA), and South Asia (SAR), combined to the demographic typology described in Ahmed (2016), which divide countries according to their potential for demographic dividend by 2030, such as as: early-dividend (ed), late divided (lt), and post-divided (pd). This disaggregation is based on the CGE macro-simulation. 23 Demand for workers by the different sectors is in line with firms’ production plans which in turn depend on consumers, export and other final demands. Due to income growth – and an elasticity with respect to income above one of consumption of non-agricultural goods – demand for manufacturing goods and services (i.e. demand for non- agricultural goods) increases more rapidly than demand for agricultural commodities. This shift of demand is transmitted to production, and consequently demand for labor in non-agricultural sectors increases faster than that in agriculture. 22 4.2 Results at the micro level Between- and within- country evolution of inequality Using the macro results about average income convergence and changes in wage premia, the microsimulation part of the economic framework estimates the impact of the education wave at the household level.24 The microsimulation results confirm that the world will become more equal by 2030 as it becomes more educated (Table 5). The (individual-based) Gini index falls from 65.8 in 2012 to 62.6 in 2030, while the Theil-L index is reduced from 90.7 to 76.6. Compared to recent patterns, these results suggest a continuation of the reduction in global inequality. During the great doubling of the global labor force, global inequality decreased by 2.3 percentage points in a 20-year interval from 1988 to 2008 (Lakner and Milanovic 2016). Our education wave scenario shows a comparable reduction of 3.2 percentage point, with the global Gini index falling during the period 2012 to 2030. The Theil-L index also suggests a similar reduction of close to 14 percentage points by 2030. Compared to Lakner and Milanovic (2016), our initial measure of Gini and Theil-L indices are smaller, which may be explained by differences in benchmark year, data sources, country coverage, and the slight improvement in equality when the education wave is measured by 2012.25 Table5: Global inequality will go down in a more educated world 2030 - Education Wave 2030 - No Inequality measures 2012 Full Wave Demographic simulation Gini index 65.8 65.5 62.6 63.2 Theil-L 90.7 91.0 76.6 78.6 Theil Decompositions: Between regions (%) 51.7 48.0 41.4 41.0 Within regions (%) 48.3 52.0 58.6 59.0 Between countries (%) 57.2 53.6 49.1 48.6 Within countries (%) 42.8 46.4 50.9 51.4 Percentile 75 / Percentile 25 5.5 5.4 6.7 6.6 Mean, $(PPP) 416.9 430.3 835.2 827.4 Coeff. of variation 3.1 3.3 2.4 2.5 Source: Authors’ calculations based on GIDD projections. What are the factors explaining the projected reduction in inequality? Using the Theil index, it is possible to decompose inequality between and within groups. The central panel 24 The microsimulation exercise is done at the country level using household surveys, but results in this section are aggregated to analyze inequality at the global level. The CGE macro results provide yearly simulations from 2011 to 2030. The microsimulation uses 2012 as a reference year for the harmonized household surveys. Thus, this section uses projections from 2012 to 2030 for the microsimulation exercises. 25 The latest year available in Lakner and Milanovic (2016) is 2008 for which the Gini is 70.5. Our initial year is 2011, for which our Gini index is 65.8. Thus, we focus the comparison on relative changes. Ferreira et. al. (2015) describe the differences across several cross-national inequality databases. 23 of Table 5 shows the results of this decomposition when the groups are represented by either regions or countries. 26 A clear answer emerges: global inequality decreases mainly because, on average, poorer countries are catching up. At the beginning of the period, the contribution of the ‘between-groups’ component to total inequality is more than 50 percent (52 percent when using regions, 57 percent when using countries as groups). However, by the end of the period, the between-country component drops to the less than 50 percent (due to income convergence) while the within-countries component shows a larger contribution to global inequality. This result does not imply that inequality within countries is increasing (more on this below), but that, at least, inequality between countries is decreasing at a faster pace. Poor and large countries, such as India and China, are growing fast to reshaping the global income distribution and contributing to the reduction of global inequality. The importance of the education wave in the dynamics of inequality within countries can be seen by comparing the results of the education wave with those of the no-wave scenario in the last two columns of Table 5. The larger decreases of the skill premium in the education wave scenario pushes down inequality within countries, while this is not the case in the no-wave scenario. As a result, the within-group component in the no-wave scenario, as well as total inequality, are higher than those in the education wave. Comparing the global growth incidence curves (GICs) for the education wave and the no- wave scenarios is another way of illustrating the change in the global distribution (Figure 10). The GICs differ from the elephant curve of Lakner and Milanovic (2016) in that they do not have the trunk part. The reasons are: the microsimulations are based on labor incomes,27 and there are no corrections for top incomes (which are not recorded in the underlying household surveys). Moreover, our 2 sectors by 2 levels of skilled workers do not provide enough heterogeneity to capture effects on the top income. 26 The regions are: Sub-Saharan Africa, Middle-East and North Africa, South Asia, East Asia and the Pacific, Europe and Central Asia, Latin America and the Caribbean, High Income countries. These regions correspond to the World Bank definitions. 27 Note that household surveys often, if not always, record non-labor incomes. The microsimulations apply the country-specific average growth rate to change these non-labor incomes. The change in the returns of capital from the global CGE could be also used, but this was not done in the current exercise. A main reason is that most of the non-labor incomes are (public and private) transfers and thus do not represents incomes from capital. 24 Figure 9: Global Growth incidence curves The education wave versus the no-wave scenario 5 annual percentage change, 2012−2030 Per capita income/consumption 3 4 2 0 20 40 60 80 100 Education Wave No Wave Source: Authors’ calculations based on GIDD projections. These GICs highlight several interesting points. First, the education wave provides its highest benefits for the population with incomes between the bottom 20 and top 20; growth rates for the groups at the two extremes of the distribution are 1 to 2 percentage points lower than for the group in the middle. Second, the no- education wave rates of income expansion are below those of the education wave scenario for everyone with incomes up to about the 90th percentile. This is expected as the education wave is mainly a wave in the developing world. Third, the distance between the two lines appears small but, for the middle of the distribution and the bottom 5 percent, the difference should not be underestimated. In fact, half a percentage point gap in growth rates accumulates to 10 percent larger incomes after 20 years, a non-trivial difference. We also analyze the changes in the country composition of the bottom 20- and the top-20 percent of the global income distribution by 2030 (table A1 of the appendix). In 2012, eight countries made up three-quarters of the population in the bottom 20 of the global distribution. And India and China alone accounted for almost 50 percent of it. Most of the other countries in the bottom were from Sub-Saharan Africa. For some of these countries almost their entire national populations (or large shares of them) belonged to the global bottom 20. While only 15 percent of Chinese nationals occupied the global bottom 20, 97 percent of the South African citizens resided in the bottom 20, 82 per cent of those from Congo, and 60 percent of Nigerians. The composition changes significantly by 2030. Notably, China disappears while India still contributes one-quarter to the total world population in the bottom 20. India’s number corresponds to 23 percent of the total population of India, not 30 percent as in 2012. A similar evolution can be described for the top 20. With the arrival of China in 2030, a large percentage of citizens from the US and from Western European countries will lose their top positions. 25 Select countries The education wave also has strong influences on within countries distributions. For example, a steep downward sloped incidence curve for the education wave is observed in Turkey (Figure 10) while an upward sloping one is registered for other countries, such as Russia. However, the growth incidence curve of Turkey for the full shock (the solid line) seems rather flat. The main reason is that the demographic shock in Turkey is regressive.28 This demographic effect generates the dotted regressive incidence curve shown in Figure 10.29 The fact that the demographic impact is regressive should not be too surprising. Deaton and Paxson (1998) show that there is a quite strong age effect on inequality, meaning that as cohorts age their inequality increases. Thus, older societies (which is the case for Turkey in 2030 vis-à-vis 2012, and in fact for all countries) are expected to be more unequal. The effect of the wage premia – that is driven by the education wave – shows, as expected, a progressive GIC. This effect can be inferred by estimating the difference between the demographic and the full effect GICs. Yet, the ‘net’ GIC for the no-wave scenario is also flat (Figure A2 in the appendix). Figure 10: Growth incidence curves for Turkey 28 We reweight the households so that the simulated population matches, in terms of its shares of age and education groups, the projected 2030 demographic structure is not distributional neutral. In fact, households with older and more educated members get reweighted more heavily than other households. 29 Note that the dotted GIC also includes the effect of the inter-sectoral reallocations of unskilled workers and a distributionally neutral residual growth effect (so that the average growth for the population as a whole is the actual growth of the period). It is possible to decompose this GIC further and isolate the pure demographic effect, but the observation in the main text will not change since the sectoral reallocation effects are of negligible importance. 26 Figure 11: Growth incidence curve for China In the case of China, the growth incidence curve is shaped as an inverted U due to the demographic effect, and the change in skill and urban premium are jointly and slightly regressive (Figure 11). The expansion of the skilled workers in China will occur in the middle of the distribution (see Figure A3 in the appendix). For each percentile of the population, we compare the share of workers that are skilled and unskilled both at the beginning of the period and at the end of the period during the education wave (Figure A3a). This comparison tells us that most of the skilled are towards the top and most of the unskilled towards the bottom, but also that the shift caused by the education wave is concentrated in the middle of the income distribution. We also find that skilled workers expand in the middle and substitute for the contraction of the unskilled, also in the middle of the income distribution (Figure A3b). In the case of China, it is mainly a story of quantity, more than of prices. A final graph for the US (Figure 12) illustrates what may happen to high income countries. The graph shows the density distribution for the US population by plotting the percentages of the US population that are found at different levels of income. These levels of income are shown in the graph not as actual USD amounts, but as the corresponding percentiles of the global income distribution. So clearly, most of the US population is found at the top of the world distribution. However, it is interesting to note that the US will lose its dominance at the top and, by 2030, a larger share of the US citizens will be in lower positions. This is a result of both income convergence across countries (citizens from other countries will start occupying the higher positions in the world distribution of income), but also of the pressure on the upper middle-income class in the US. 27 Figure 12: Distribution of the US population across the world income distribution, in 2012 and 2030 .1 .08.06 Density .04 .02 0 10 20 30 40 50 60 70 80 90 100 Percentiles of Global Income Distribution 2012 2030 Source: Authors’ calculations. Notwithstanding the potential pressures on high income countries, overall the education wave will be pushing down inequality within countries. The Gini index is lower for almost all countries in the education wave when compared to the case of the no education wave. Figure 13 plots the difference in the Gini index by 2030 between the education wave and no education- wave scenarios for each country. A negative number denotes less inequality (an improvement) brought about by the education wave. For most of the 117 countries covered, changes in the within-country inequality are clearly more favorable with the education wave. Although the improvement did not completely reverse the increase in the overall within- country inequality by 2030 (as shown in Table 6), it clearly ameliorates the deterioration.30 30 Even if the Gini indices in both scenarios would have deteriorated in 2030 from 2011, as long the index for the education wave is less than the other case, the result would come out as favorable towards reducing inequality. 28 Figure 13: Difference in the Gini Index between Education and No Wave, by Country, 2030 Gini, change w.r.t no education wave −4 −2 0 2 Portugal China Italy United Kingdom Iraq France Netherlands Sweden Maldives Honduras Paraguay Iceland Canada Thailand Uruguay Costa Rica Spain Guatemala Germany Turkey Vietnam Sri Lanka Morocco Panama Denmark Belize South Africa Peru Brazil Ecuador Greece Bosnia and Herzegovina Colombia Ireland Argentina Mexico Estonia Nicaragua El Salvador Venezuela, RB Bolivia Ukraine Pakistan Finland Austria Senegal Albania Kiribati Bhutan Haiti Czech Republic Poland Indonesia Mauritius Slovenia Swaziland Ghana Dominican Republic Madagascar Vanuatu Norway India Mongolia Gabon Moldova Belgium Jamaica Botswana Philippines Nigeria Seychelles Angola Slovak Republic Liberia Djibouti Luxembourg Guinea−Bissau Tanzania Burkina Faso Mauritania Congo, Rep. Chile Cameroon São Tomé and Principe Papua New Guinea United States Gambia, The Kenya Ethiopia Russian Federation Guyana Nepal Zambia Rwanda Fiji Chad Bulgaria Georgia Armenia Hungary Mozambique Malawi Montenegro Niger Lithuania Cyprus Tonga Timor−Leste Latvia Kazakhstan Guinea Sudan Romania Congo, Dem. Rep. Australia Côte d’Ivoire Solomon Islands High income countries East Asia & Pacific Eastern Europe & Central Asia Latin America & Caribbean Middle East & North Africa South Asia Sub−Saharan Africa Source: Authors’ calculations. Note: Results based on microsimulation for individual countries. The re-weight procedure is based on projected population in 2030, taking into account changes in the size and number of skilled workers by age cohort, for each country between 2011 and 2030. 4.3 Sensitivity tests The sensitivity of the results to important factors, such as alternative elasticity of substitution between labor types and different nesting structure between labor and capital was tested. If skilled and unskilled workers are imperfect substitutes, an increase in the relative supply of skilled workers, without offsetting skill-biased changes in demand or other factors in supply, will unavoidably reduce the wage premium of skill. In the growth literature, one way to account for the lack of effects of capital accumulation on the rental-wage ratios in advanced economies is to introduce labor-augmenting productivity (due to having more capital goods available per worker.) Another way is to raise the elasticity of substitution between the factors of production as suggested by de la Grandville and Solow (2009). Others, like Houthakker (1955) and Jones (2005), distinguished between short-term and long-term production possibilities. In modern economic growth theory, Acemoglu (2009) emphasized the importance of directed or biased technological change and the endogenous nature of technology to explain that, in over 60 years since 1939, the U.S relative supply of skills has increased rapidly, but there, college premium increased over the period.31 In the sensitivity tests (table 6), the elasticity, along with a few other factors, was changed. When the elasticity of substitution between skilled labor and capital is raised (relative to the baseline), the mean income increases. Raising the elasticity just between skilled labor and capital 31 See in particular Chapter 15 and figure 15.1 in Acemoglu (2009). 29 means that capital accumulation would tend to raise demand for skilled labor in the same nest. The effect would tend to maintain the wage premium of skill (similar to the effects regarding K and L in the growth literature). Inequality in all dimensions (global, regional, and country) is relatively similar to the education wave scenario. The findings somewhat mimic the effects of a biased technological change. The worst-case scenario for inequality under education wave is the alternative nesting structure of inputs (ST-3). Putting skilled and unskilled labor in the same nest has the same effect as raising the elasticity of substitution between the labor types against capital. The effect would more directly dampen any fall in premia between skilled and unskilled workers. The result is to raise global inequality between and within country in all the measures against the baseline, but it is still slightly better than the no-education wave scenario. Table 6 - Effect of education wave on global inequality for different scenarios Education No education ST-1 -Higher ST-2-Alt. wave wave elasticity nesting Gini index 62.6 63.2 62.6 63.0 GE(1) (Theil) 76.6 78.6 76.5 77.8 Between - GE(1) (Theil) by region 31.7 32.3 31.7 32.6 Within - GE(1) (Theil) by region 44.9 46.4 44.8 45.1 Between - GE(1) (Theil) by country 37.6 38.2 37.6 38.6 Within - GE(1) (Theil) by country 39.0 40.4 38.9 39.2 Percentile ratio 75 / 25 6.7 6.6 6.8 6.7 Mean, $(PPP) 835.2 827.4 841.7 801.1 Coeff. of variation 2.4 2.5 2.4 2.3 Source: Authors’ calculations. Note: ST1: Higher elasticity. The baseline with a higher elasticity of substitution between skilled labor and capital of 3, instead of 2. Skilled labor and capital are bundled together in the same nest, and together substitutable to unskilled labor; ST2: Alternative nesting structure. The baseline with an alternative nesting structure of production inputs. Skilled and unskilled are directly substitutable with one another in the same nest, and together substitutable with capital. 5. Conclusion The sudden integration of China, India, and the former Soviet bloc of countries in the global economy has brought about the ‘great doubling’ (Freeman 2007) of the global labor markets and triggered significant changes in the global income distribution. This paper analyzes, in an ex-ante fashion, what is the impact of another ‘great one-and-a-half times’ shock to the global economy. Because their younger (and larger) populations and the still large positive gap in education of their young cohorts vis-à-vis their old ones, developing countries will soon be the sole contributors to the expansion of the global pool of skilled workers. Even without any improvement in education effort by 2030, the ratio of skilled workers from developing countries to skilled workers from high income (OECD) countries will be 3 to 1, up from 2 to 1 in 2012; a ‘one-and-a-half times’ increase that this paper labels the education wave. The main result from the analysis of the impact of this education wave is that global inequality will 30 likely decrease. This is driven by a reduction of inequality between and within countries. Convergence of incomes per capita between countries, mainly depending on closing up of productivity gaps between high income and developing countreis, is further boosted by the education wave. In addition, despite its increasingly larger weight in determing global inequality, within-country inequality would descrease in most countries. Education, especially for the developing world, can still play the role of the great equalizer. The results also show that the pressure on high income countries may still be on in the future. References Acemoglu, D. (2009). Introduction to Modern Economic Growth. Princeton University Press. Acemoglu, D. (1998). Why do New Technologies Complement Skills? Directed Technical Change and Wage Inequality. The Quarterly Journal of Economics 113(4), 1055–1089. Acemoglu, D. (2002b). Technical Change, Inequality, and the Labor Market. Journal of Economic Literature 40(1), 7–72. Ahmed, S.A., M. Cruz, D.S. Go, M. Maliszewska, and I. Osorio-Rodarte (2016). “How significant is Sub-Saharan Africa’s demographic dividend for its future growth and poverty reduction?” Review of Development Economics doi:10.1111/rode.12227. Ahmed, S. Amer, Cruz, Marcio, Quillin, Bryce, and Schellekens, Philip (2016). "Demographic change and development: a global typology," Policy Research Working Paper Series 7893, The World Bank. Alvaredo, F., A. B. Atkinson, T. Piketty, and E. Saez (2013). The top 1 percent in international and historical perspective. Technical report, National Bureau of Economic Research. Arbache, J., D.S. Go, and J. Page (2010). “Is Africa’s economy at a turning point?” in D.S. Go and J. Page (eds.), Is Africa’s Economy at a Turning Point? Growth, Aid, and External Shocks. Washington DC: World Bank, pp. 13-85. Atkinson, A. (2015). Inequality – What can be done? Cambrige: Harvard University Press. Autor, David H. 2015. "Why Are There Still So Many Jobs? The History and Future of Workplace Automation." Journal of Economic Perspectives, 29(3): 3-30. Autor, D., Dorn, D., Hanson, G, and Majlesi, K. (2016). "Importing political polarization? the electoral consequences of rising trade exposure." NBER Working Paper 22637. Barro, R. J. and Lee, J. W. A new data set of educational attainment in the world, 1950—2010. Journal of development economics, n. 104, 2013. Blanchflower, D. and Freeman, R. (1999) Preparing Youth for the 21st Century: The Transition from Education to the Labour Market. Proceedings of the Washington D.C. Bourguignon, F. (2015). The Globalization of Inequality. Princeton: Princeton University Press. Bourguignon, F., Bussolo, M. and Da Silva, L.P. (2008) The impact of macroeconomic policies on poverty and income distribution: macro-micro evaluation techniques and tools. World 31 Bank. Bussolo, M., De Hoyos R. and Medvedev D. Economic growth and income distribution: linking macroeconomic models with household survey data at the global level. International Journal of Microsimulation, pages 92-103, 2010. Bussolo, M. Rafael E. De Hoyos, Medvedev, D. and Mensbrugghe D. Global Growth and Distribution: Are China and India Reshaping the World? The World Bank Policy Research Working Paper, n. 4392, November 2007. Bussolo M., de Hoyos, R. E. & Medvedev, D., and Van der Mensbrugghe D., 2012. "Global Growth and Distribution: China, India, and the Emergence of a Global Middle Class," Journal of Globalization and Development, De Gruyter, vol. 2(2), pages 1-29, January. Card, D. and T. Lemieux (2001). Can Falling Supply Explain the Rising Return to College for Younger Men? A Cohort-Based Analysis. The Quarterly Journal of Economics 116(2), 705–746. Cruz, M. and Milet, E. (2017) Skill premia in Developing Countries. World Bank, Manuscript. Cruz, M., Go, and Osório-Rodarte, I. (2017) Modeling the poverty and distribution effects of human capital formation across countries: Using a new database on labor skills and wages. World Bank, Manuscript. Cruz, M. and Ahmed, S. A. 2016. "On the impact of demographic change on growth, savings, and poverty," Policy Research Working Paper Series 7805, The World Bank. Deaton, A. and C. Paxson (1998). “Economies of Scale, Household Size, and the Demand for Food.” The Journal of Political Economy, 106 (5): 897-930. De la Granville, O. and R.M. Solow (2009). “Capital-Labour Substitution and Economic Growth.” In O. de la Granville, Economic Growth – A Unified Approach. Cambridge: Cambridge University Press. Ferreira, F.H., Lustig, N. & Teles, D. Journal of Economic Inequalality (2015) 13: 497. Freeman, R. B. (2008). "The New Global Labor Market." Focus 26(1) Summer-Fall 2008, pp. 1-6. Freeman, R. B. (2007). "Labor Market Institutions around the World," NBER Working Papers 13242, National Bureau of Economic Research, Inc. Freeman, R. B. (2007). "The Great Doubling: The Challenge of the New Global Labor Market." In J. Edwards, M. Crain, and A.L. Lalleberg, eds., Ending Poverty in America: How to Restore the American Dream. New York: The New Press. Galbraigth, J.K. (2012). Inequality and Instability. Oxford: Oxford University Press. Harris, J.R. and M. Todaro (1970). “Migration, Unemployment and Development: A Two-Sector Analysis.” American Economic Review 60: 126-42. Houthakker, H.S. (1955). “The Pareto Distribtuion and the Cobb-Douglas Production Function in Activity Analysis.” Review of Economic Studies 23: 27-31. Jones, C.I. (2009). “The Shape of Production Functions and the Direction of Technical Change.” Quarterly Journal of Economics 2:517-549. 32 Katz, L. F. and K. M. Murphy (1992). Changes in Relative Wages, 1963-1987: Supply and Demand Factors. The Quarterly Journal of Economics 107(1), 35–78. Lakner, C. and B. Milanovic (2016). Global income distribution: from the fall of the berlin wall to the great recession. World Bank Economic Review, vol. 30 (2), pp. 203-232. Lewis, A. (1954). “Economic Development with Unlimited Supplies of Labor.” The Manchester School of Economics and Social 28: 139-91. López-Calva, Luis Felipe, and Nora Claudia Lustig, eds. Declining inequality in Latin America: A decade of progress? Brookings Institution Press, 2010. Maloney, William F.; Molina, Carlos A. (2016). Are automation and trade polarizing developing country labor markets, too? Policy Research Working Paper; no. WPS 7922. Washington, DC, World Bank Group. Milanovic, B. (2013). Global income inequality in numbers: in history and now. Global Policy 4 (2), 198-208. Orcutt, G.H. (1957). “A New Type of Socio-Economic System.” Review of Economics and Statistics. 39(2):116-23. Piketty, T. (2014). Capital in the Twenty-First Century. Cambridge: The Belknap Press of Harvard University Press. Ravallion, M. (2013). "How Long Will It Take to Lift One Billion People Out of Poverty?" World Bank Research Observer, World Bank Group, vol. 28(2), pages 139-158, August Stiglitz, J. (2012). The Price of Inequality –How Today’s Divided Society Endangers Our Future. New York: W.W. Norton & Company. Van der Mensbrugghe, D.,2013. Modeling the Global Economy - Forward Looking Scenarios for Agriculture. In P.B. Dixon & D.W. Jorgenson (Eds.), Handbook of Computable General Equilibrium Modeling. North Holland Elsevier B.V., pp. 933-994. Van Der Mensbrugghe, D. (2011). Linkage Technical Reference Document, version 7.1. Development Prospects Group, The World Bank, Washington DC. World Bank (2015). Global Monitoring Report 2015/16: Development Goals in an Era of Demographic Change. Washington DC: World Bank. World Bank (2011). Global Development Horizons 2011 – Multipolarity: The New Global Economy. The World Bank, Washington DC. World Bank (2007). Global Economic Prospects 2007 – Managing the Next Wave of Globalization. The World Bank, Washington DC. World Bank (2006). World Development Report 2006 – Equity and Development. The World Bank, Washington DC. Young, A. Inequality, the urban-rural gap, and migration. The Quarterly Journal of Economics, 128(4):1727{1785, 2013. 33 Appendix Table A1 - Change in the composition by country of the global bottom 20 and top 20 groups Bottom 20 Top 20 2012 2030 2012 2030 % of % of % of % of % of % of % of % of total country total country total country total country India 32 30 India 25 23 United States 22 84 China 40 40 China 17 15 Nigeria 11 58 China 13 11 United States 18 71 Nigeria 8 57 Congo, Dem. Rep. 7 87 Russian Federation 7 54 Russian Federation 5 52 Congo, Dem. Rep. 5 82 Pakistan 7 40 Germany 6 96 Germany 4 76 South Africa 4 97 Ethiopia 6 60 France 5 96 France 4 75 Indonesia 4 20 South Africa 4 98 Brazil 5 29 United Kingdom 3 63 Ethiopia 3 45 Tanzania 4 66 United Kingdom 5 89 Brazil 3 19 Pakistan 2 16 Philippines 3 30 Italy 4 87 Iraq 2 62 Spain 3 84 Uganda 2 53 Turkey 2 32 Mozambique 2 78 Canada 2 64 Madagascar 2 89 Mexico 2 17 Tot 76 Tot 75 Tot 76 Tot 76 Source: Authors’ calculations based on GIDD projections. Figure A1 - GIDD framework Source: Authors’ elaboration based on Bussolo, De Hoyos, and Medvedev (2010). 34 Figure A2 - Net wage skill premia effect in the growth incidence curve for Turkey Figure A3 - Education wave quantity effects in China a) Skill level by income decile Skill Level by Income Decile, CHN .6 % share of total population .2 0 .4 0 20 40 60 80 100 Percentiles of Per Capita Income Unskilled in Non Agri 2012 Skilled in Non Agri 2012 Unskilled in Non Agri 2030 Skilled in Non Agri 2030 b) Difference in Skill level by income decile Difference in Skill Level by Income Decile, CHN .05 0 % share of total population −.1 −.05 −.15 0 20 40 60 80 100 Percentiles of Per Capita Income Difference in Unskilled Non Agri Difference in Skilled Non Agri Source: Authors’ calculations. 35