e - -. ZIa - . - i aluOW-'3'oipne'j'.11!g-.s J ! ' WS,q 'oaiv uasul Wov suossal pue sanbiua -iJ Aq pap AAJO°d--oqel '6ul-e3-- -- -r i 0'91-9 -0.- '- Crafting Labor Policy Techniques and Lessonsfrom Latin America Cover design by Patricia Hord Graphik Design. Photos provided by (front cover, left to right) Curt Carnemark/World Bank Photo Library, Photodisc, and Truman Packard; (back cover, left to right) Thomas Sennett/World Bank Photo Library, Corbis, and Curt Carnemark/World Bank Photo Library. Crafting Labor Policy Techniques and Lessons from Latin America Edited by Indermit S. Gill Claudio E. Montenegro and Dorte Domeland A copublication of the World Bank and Oxford University Press C 2002 The International Bank for Reconstruction and Development / The World Bank 1818 H Street, NW Washington, DC 20433 All rights reserved. 1 2 3 4 04 03 02 A copublication of the World Bank and Oxford University Press. Oxford University Press 198 Madison Avenue New York, NY 10016 The findings, interpretations, and conclusions expressed here are those of the author(s) and do not necessarily reflect the views of the Board of Executive Directors of the World Bank or the governments they represent. The World Bank cannot 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 on the part of the World Bank any judgment of the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this work is copyrighted. No part of this work may be reproduced or trans- mitted in any form or by any means, electronic or mechanical, including photocopying, recording, or indusion in any information storage and retrieval system, without the prior written permission of the World Bank. The World Bank encourages dissemination of its work and will normally grant permission promptly. For permission to photocopy or reprint, please send a request with complete informa- tion to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA, telephone 978-750-8400, fax 978-750-4470, www.copyright.com. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, World Bank, 1818 H Street NW, Washington, DC 20433, fax 202-522-2422, e-mail pubrights@worldbank.org. ISBN 0-8213-5112-5 hardcover ISBN 0-8213-5111-7 paperback Library of Congress Cataloging-in-Publication Data Crafting labor policy: techniques and lessons from Latin America / edited by Indermit S. Gill, Claudio E. Montenegro, Dorte Domeland. p. cm. Includes bibliographical references and index. ISBN 0-8213-5112-5 - ISBN 0-8213-5111-7 (pbk.) 1. Labor policy-Latin America. 1. Gill, Indermit Singh, 1961- II. Montenegro, Claudio. III. Domeland, Dorte, 1971- HD8110.5 .C73 2002 331.12'042'098-dc2l 2002019991 Contents Tables and Figures vii Foreword xi Acknowledgments xv Acronyms and Abbreviations xvii 1 Labor Reform in Latin America during the 1990s Diirte Domeland and Indermit S. Gill 1 PART I: PUBLIC AND PRIVATE SECTOR LABOR MARKETS IN BRAZIL 33 2 Improving Public Sector Efficiency in Brazil Ricardo Paes de Barros, Miguel N. Foguel, Indermit S. Gill, and Rosane Silva Pinto de Mendon,ca 37 3 Assessing the Impact of Regulations on Informal Workers in Brazil Edwardj Amadeo, Indermit S. Gill, and Marcelo C. Neri 67 PART II: LABOR SUPPLY AND LABOR DEMAND IN ARGENTINA 97 4 Understanding Labor Supply Dynamics in Argentina Carola Pessino, Indermit S. Gill, andJose Luis Guasch 101 v Vi CONTENTS 5 Increasing Labor Demand in Argentina Carola Pessino, Indermit S. Gill, and Jose Luis Guasch 129 PART III: EARNINGS AND EMPLOYMENT OUTCOMES IN CHILE 155 6 Responding to Earnings Differentials in Chile Indermit S. Gill and Claudio E. Montenegro 159 7 Dealing with Employment Instability in Chile Indermit S. Gill, Erik Haindl, Claudio E. Montenegro,. and Claudio N. Sapelli 191 8 Some Guidelines for the New Decade: Summary and Conclusions Indermit S. Gill and Claudio E Montenegro 215 Technical Annexes A. Technical Annex to Chapter 2: The Public-Private Wage Gap 237 B. Technical Annex to Chapter 5: Wage and Output Elasticities of Labor Demand 241 C. Technical Annex to Chapter 6: The Quantile Regression 247 D. Technical Annex to Chapter 7: Haindl's Labor Flows Model 253 Contributors 259 Index 263 Tables and Figures Tables Chapter 1 Labor Reform in Latin America during the 1990s 1-1 Changes in Temporary Contract Legislation 7 1-2 Changes in How Severance Pay Increases with Tenure 14 1-3 Changes in Severance Payment Laws and Cause of Dismissal 17 1-4 Changes in Severance Compensation Fund Legislation 20 Chapter 2 Improving Public Sector Efficiency in Brazil 2-1 Sectoral Shares in Employment, Selected Years, 1981-95 41 2-2 Share of Workers, by Class of Worker Distinction, 1995 42 2-3 Average Monthly Earnings, Selected Years, 1981-95 43 2-4 Public Sector Wage Premium 48 2-5 Public Sector Wage Premium 49 2-6 Estimates of Log Wage Gap Due to Differences in Composition and Controlled Wage Gap for Metropolitan Areas 53 2-7 Measures of the Public Sector Earnings Premium by Sector and Class, September 1995 56 2-8 Estimated Monthly Pensions 59 2-9 Effects of Reducing Public Employment on Average Earnings, Transfer of Obligations, and DPV of Savings 61 vii Viii TABLES AND FIGURES 2-10 Measures to Reduce Public Payroll Expenses While Maintaining or Improving Labor Market Efficiency 64 Chapter 3 Assessing the Impact of Regulations on Informal Workers in Brazil 3-1 Labor Turnover Rates, Brazilian Formal Labor.Market, 1985-93 72 3-2 Total Cost of Firing a Worker 76 3-3 Stylized Facts of Wage Indexation Regimes, 1980-2000 78 3-4 Wage and Nonwage Labor Costs 79 3-5 Wage Earners Paid above the Minimum Wage 85 3-6 Wages Exactly Equal to One Minimum Salary 86 3-7 Wages Exactly Equal to One Minimum Salary 87 3-8 Proportion of Wage Adjustments Exactly Equal to Minimum Wage Adjustments at Dates When the Minimum Salaries Were Changed 90 3-9 Proportion of Workers with Hours Worked Below or Equal to Maximum Hours Limit 91 3-10 Effective Workweek before and after New Constitution 92 3-11 Brazil: Some Measures of Conformity with Labor and Social Security Laws 93 Chapter 4 Understanding Labor Supply Dynamics in Argentina 4-1 Argentina: Labor Force, Employment, and Unemployment, 1980-95 103 4-2 Labor Force Participation, by Gender 105 4-3 Unemployment Rates, Inflows, and Duration, October 1987 and May 1995 108 4-4 Distribution and Duration of Unemployment, by Age, Gender, and Previous Labor Market Status, May 1995 111 4-5 Labor Force Participation: Cyclical and Trend Factors Specification (1), Females and Males 113 4-6 Labor Force Participation: Cyclical and Trend Factors Specification (2), Females 115 TABLES AND FIGURES iX 4-7 Labor Force Participation: Cyclical and Trend Factors Specification (3), Females 116 4-8 The Predictive Power of Alternative Models of Labor Force Participation 120 4-9 Female Labor Force Status, by Husband's Employment Status 122 Chapter 5 Increasing Labor Demand in Argentina 5-1 Cobb-Douglas Technology 139 5-2 CES Technology: Cost Function Approach 140 5-3 CES Technology: Production Function Approach 142 5-4 Single Factor Demand Equations 143 5-5 Generalized Leontief Technology 145 5-6 Transcendental Logarithmic Technology 146 5-7 Translog Technology, Nonneutral Technical Change 148 5-8 Estimated Elasticities: Own-Price, Output, and Substitution 149 5-9 Years Needed to Reduce the Unemployment Rate to 8 Percent 151 Chapter 6 Responding to Earnings DiJfierentials in Chile 6-1 Average Labor Earnings and Changes, in Percent Workers Aged 14-65 165 6-2 Measures of Wage Inequality 168 6-3 Mean Years of Completed Schooling, Population Aged 14 to 65 171 6-4 Returns to Education by Schooling Level at Different Points of the Conditional Wage Distribution, Full-Time Male Earners 172 6-5 Education Level of Parents by Type of School Attended by the Child 177 6-6 Decomposing the Poor and Nonpoor Earnings Gap and Simulating Payoffs to Alternative Policies 179 6A-1 Trends in Hourly Wage Inequality 186 6A-2 Returns to Education 187 X TABLES AND FIGURES Chapter 7 Dealing with Employment Instability in Chile 7-1 Estimates of Labor Force Entrants and Employment Flows, 1962-95 197 7-2 Labor Force Participation Rates, Females Aged 14 to 65 198 7-3 Estimates of Unemployment Rates and Duration, 1962-95 199 7-4 Average Unemployment Rates, Men Aged 14to65 202 7-5 Average Unemployment Rates, Women Aged 14 to 65 203 7-6 Income Support Programs for the Unemployed: Findings and Policy Implications 207 Figures 2-1 Share of Employment and Wage Bill of Different Public Employment Sectors 45 2-2 Annual Growth from 1992 to 1995 in Employment, Average Earnings, and Wage Bill 46 3-1 Wage Distribution of Formal and Informal Male Workers, Brazil 1998 88 3-2 Proportion of Wage Adjustments Exactly Equal to Minimum Wage Adjustments, Legal and Illegal Employees 89 4-1 Employment and Labor Force 124 6-1 Trends in Wage Inequality: Whole Sample 169 6-2 Trends in Wage Inequality: White-Collar Workers 170 6-3 Trends in Wage Inequality: Blue-Collar Workers 170 6-4 Return to Education: Mean and Median Estimates, Whole Sample 173 6-5 Return to Education: Quantile Estimates, Whole Sample 174 6-6 Return to Education: Quantile Estimates, White-Collar Workers 174 6-7 Return to Education: Quantile Estimates, Blue-Collar Workers 175 7-1 Labor Force Flows 194 7-2 Expected Job Tenure 196 7-3 Unemployment Rate and Duration, Greater Santiago 200 7-4 Probability of Finding a Job, Greater Santiago 200 Foreword THIS VOLUME PRESENTS a set of papers on labor market issues in Argentina, Brazil, and Chile, based on work sponsored by the World Bank. Some of the leading labor economists in the three countries covered here have coauthored these papers with the editors. An introduc- tory chapter provides the context, summarizing labor market reforms in Latin America since the late 1980s. A concluding chapter summarizes the main results and policy implications for these and other countries. The papers attempt to answer important labor policy questions in a quantitative manner. They have two features in common: each uses an estimation technique-sometimes one that is fairly complex-to do this, and the papers all have a strong policy focus. Thus, * For Argentina, we use a labor force participation model to see if the structure of the labor market is such that the labor force grows dur- ing periods of economic slowdown and leads to exaggerated unem- ployment fluctuations, and wage and output elasticities of demand for labor to determine whether unemployment can be reduced sub- stantially without labor market reforms to lower the price of labor relative to capital. * For Brazil, we calculate public-private compensation differentials to determine the public employment reforms that would lower the fis- cal deficit in Brazil in a sustainable manner, and identify specific labor laws that influence earnings and employment conditions by using statistical techniques to identify the pressure points in labor legislation. xi Xii FOREWORD For Chile, we use a labor flow model to determine if employment has become more precarious even as unemployment has fallen, and quantile regression techniques to examine if the labor market rewards to skills and effort differ for workers from different socioeconomic groups. We study labor issues in these countries during 1995-98. In macro- economic terms, of course, they were at very different stages of structural adjustment: * Argentina's experience was analyzed during a stage of moderate growth and high unemployment, having undergone stabilization in the early 1990s and when it was midway through a fiscal adjustment. * Brazil was examined during a stage of sluggish growth and low but rising unemployment, while still struggling to affect a fiscal adjust- ment and sustain the stabilization begun in 1994. * Chile was studied during a stage of sustained growth and low unem- ployment, having completed its stabilization and fiscal adjustment in the mid-1980s. The book looks at labor policy from three different angles: * In the case of Argentina, the focus is structural: that is, labor demand and labor supply. * In the case of Brazil, the approach is sectoral: that is, public and pri- vate employment and earnings. * In the case of Chile, we focus on the main labor market outcomes: earnings and employment. Finally, the study contains material at three levels of technical sophistication: * Readable, nontechnical, and policymaker-friendly introductory and concluding chapters, written with the broadest audience in mind: policymakers, researchers beginning work on Latin America, and those with a temporary or casual interest in labor market issues in de- veloping countries. * Readable but somewhat more technical policy chapters on each country, written with a narrower target audience: policymakers who have a deeper interest in labor market issues and labor market specialists. * Technical annexes that summarize the main techniques and models used in the empirical analysis in some of the policy chapters which FOREWORD Xiii serve as a brief introduction to these techniques, aimed at researchers interested in empirical labor market issues and students of economic policy in developing countries. We hope this book advances labor market reform-which has correctly been referred to as the "unfinished" economic reform in Latin Arnerica- in this and other parts of the world through quantification of some im- portant issues, so that the debate is increasingly based on objective argu- ments instead of rhetoric. We will consider this and similar ventures worth the effort if and when labor reforms in emerging markets are upgraded in status from mosdy unfinished to merely tardy. Acknowledgments WE WOULD LIKE TO THANK Homi Kharas, Danny Leipziger, and John Underwood who, as lead economists during the time that this work was initiated, provided valuabl guidance in identifying labor market issues for closer scrutiny and determining policy-focused treatment of these subjects. We are also grateful to Jose Luis Guasch, Al- berto Vald&s, and Gautam Datta for their help in carrying out these ideas; as task managers of World Bank reports for which the studies collected here were commissioned, they encouraged analytical rigor in our inquiry even when time was short and budgets tight. Of course, all errors remain ours. We would like to express our gratitude to Gobind Nankani and Suman Bery for encouraging us to write this book even when other tasks could have had priority. We would also like to thank Ana-Maria Arriagada, Amit Dar, Robert Holzmann, and Zafiris Tzannatos for financial support for the participation of D6rte Domeland and Claudio E. Montenegro. The comments of Norman Hicks and three anonymous referees have helped us improve the contents of this book, and Diane Stamm has improved its form. Finally, we would like to thank Thaisa Tiglao of the World Bank's Office of the Publisher for her balanced advice and encouragement, and Cindy Fisher for shepherding the book through the final stages. xv Acronyms and Abbreviations AR i First-order autoregressive AR2 Second-order autoregressive BCRA Banco Central de la Republica Argentina (Central Bank of the Republic of Argentina), Argentina CAGED Cadastro Geral de Empregados e Desempregados (Official Register of Employment and Unemployment), Brazil CASEN Encuesta de Caracterizaci6n Socioeconomica Nacional (National Socioeconomic Population Survey), Chile CES Constant elasticity of substitution CIEPLAN Corporaci6n de Investigaciones Econ6micas para America Latina (Latin American Economic Research Corporation), Chile CPI Consumer price index CLT Consolidac,o das Leis do Trabalho (Consolidated Labor Code), Brazil CRS Constant returns to scale EC European Community ECLAC Economic Commission for Latin America and the Caribbean EFTA European Free Trade Association xvii XViii ACRONYMS AND ABBREVIATIONS EPH Encuesta Permanente de Hogares (Continuing Household Survey) FAT Fundo de Aparelho de Trabalhadores (Fund for Work Assistance Programs and Unemployment Insurance), Brazil FGTS Fundo de Garantia do Tempo de Servico (Severance Fund ) FIEL Fundaci6n de Investigaciones Econ6micas Latinoamericanas (Latin American Economic Research Foundation), Argentina GBA Greater Buenos Aires GDP Gross domestic product GNP Gross national product IADB Inter-American Development Bank IDB International Development Bank ILO International Labour Organization INCRA Instituto Nacional de Colonizac,o e Reforma Agraria (Federal Land Reform Agency), Brazil INDEC Instituto Nacional de Estadistica y Census (National Statistics and Census Institute), Argentina EPH Encuesta Permanente de Hogares (Permanent Household Survey), Argentina INPC Indice Nacional de Precos ao Consumidor (National Index of Consumer Prices), Brazil INSS Instituto Nacional de Seguridade Social (National Institute of Social Security), Brazil IPEA Instituto de Pesquisa Econ6mica Aplicada (Institute for Applied Economic Research), Brazil IPI Industrial production index IV Instrumental variable JSA Job search assistance KLEM Capital, labor, energy, materials (short for an estimation model that includes these four variables) LF Labor force ACRONYMS AND ABBREVIATIONS xix LFP Labor force participation MARE Ministerio da Administra,ao Federal e Reforma do Estado (Ministry of Federal Administration and State Reform), Brazil OECD Organisation for Economic Co-operation and Development OLS Ordinary least squares PASEP Programa de Forma,co de Patrim6nio do Servidor Pdblico (Program of Assistance to Civil Servants), Brazil PDV Programa de demissao voluntaria (Voluntary Dismissal Program), Brazil PIS Programa de Integra,co Social (Social Integration Program), Brazil PK Price of capital PL Price of labor PME Pesquisa Mensual de Emprego (Monthly Employment Survey), Brazil PNAD Pesquisa Nacional por Amostra de Domicilios (National Household Survey), Brazil PUC Pontificada Universidade Cat6lica (Catholic Pontifical University), Brazil QLM Quantile regression method QRM Quintile regression method R$ Reals-Brazilian national currency RAIS Relacao Annual de Informa,6es Sociais (Annual Report of Social Indicators), Brazil RGPS Regime Geral da Previdencia Social (General Regime for Social Security), Brazil RJU Regime Juridico Unico (Pension Regime of Government Workers), Brazil SEBRAE Serviqo de Apoioas Pequenas Empresas (Assistance Program for Small Enterprises), Brazil SENAC Servi,o Nacional de Aprendizagem Comercial (National Apprenticeship Program in Commerce), Brazil xx ACRONYMS AND ABBREVIATIONS SENAI Servico Nacional de Aprendizagem Industrial (National Apprenticeship Program in Industry), Brazil SENCE Servico Nacional de Capacitaci6n y Empleo (National Formation and Employment Program), Chile SESC Servico Social do Com6rcio (Social Service Program for Industrial Workers in Commerce), Brazil SESI Servico Social da Industria (Social Service Program for Workers in Industry), Brazil SIMCE Sistema de Medici6n de la Calidad de la Educaci6n (Educational Quality Measurement System), Chile SINE Sistema Nacional de Emprego (National Employment System), Brazil SURE Seemingly unrelated regression estimator TFP Total factor productivity UI Unemployment insurance V-T Technical and vocational WPI Wholesale price index WTO World Trade Organization Labor Reform in Latin America during the 1990s DORTE DOMELAND AND INDERMIT S. GILL AFTER THE 1980s, a period characterized by high eco- nomic volatility and stagnation combined with worsening income in- equality, the 1990s promised better days for Latin American workers. In- deed, macroeconomic indicators did improve during the decade, despite the Tequila Crisis during 1994-95 and the financial turmoil following the East Asia and Russia crisis during 1997-98. Gross domestic product (GDP) growth increased from less than 3 percent in the second half of the 1980s to more than 4 percent during 1991-97 before falling again. Infla- tion rates fell to single digits in many countries. Fiscal deficits fell. Even the rising trend in aggregate volatility in Latin America as measured by volatil- ity in GDP growth, wages, and consumption was reversed during the 1990s. It was expected that economic growth induced by macroeconomic stabilization and structural reforms would generate new employment. But this did not happen. Wages grew, but real wage levels in some coun- tries still do not exceed the levels workers had attained in the early 1980s. In some countries, the increase in the supply of skilled labor did not keep up with the demand for skilled labor, leading to increases in the real wages of highly skilled, wealthier workers relative to those of the less skilled. Again, while the gains from the macroeconomic reforms did translate into produc- tivity increases in most countries, they did not appear to generate the sort of jobs people had in mind. Argentina, Brazil, Mexico, and Uruguay experi- enced periods of growth, but little job crearion in the formal, or regulated, sector. Compared with the second half of the 1980s, the rate of job creation in these four countries, particularly in the formal sector, actually fell. 1 2 CRAFTING LABOR POLICY The supply side of the labor market featured high labor force growth, largely due to increasing labor force participation of women. This, com- bined with the low job creation rate in the formal sector, implied both an increase in the share of the informal employment sector and in unem- ployment rates. The regional unemployment rate rose to about 10 percent by the end of the decade. In more than half the countries studied by de Ferranti and others (2000), unemployment rates remained above the levels of the 1970s. In Argentina, Paraguay, Peru, and the Republica Boli- variana de Venezuela unemployment rates increased to levels that are high even in absolute terms. Colombia and Uruguay faced decreases in unemployment in the early 1990s, which were sharply reversed by the end of the decade. These high recorded rates of unemployment are worrisome. Given the large informal sector and the absence of widespread formal unemployment insurance, even low unemployment rates do not necessarily imply the absence of em- ployment problems in most Latin American countries. Informal sector employment in Latin America and the Caribbearn region increased from 52 percent in 1990 to more than 57 percent in 1996, and is likely to be close to 60 percent today. So understanding Latin American earnings and employment clearly requires a careful look at the informal sector, even more so now than a decade ago. During the first half of the 1990s, rates of informality increased in every Latin American country (Lora and Marquez 1998), and according to the International Labour Organization and the Inter-American Development Bank, 80 out of 100 jobs in the 1990s were created in the informal sector.' However, informal firms show high rates of mortality and income vari- ance (Levenson and Maloney 1998). Hence, it is not surprising that the jobs they offer are usually of "low quality" in the sense that they imply low salaries, long working hours, low job stability, and no pension benefits.2 Given these problems, it is not surprising that many Latin Americans harshly judge the developments during the last decade, and view the fu- ture somewhat pessimistically. Using data from the Latinobarometro sur- vey, Lora and Marquez (1998) find that most Latin Americans considered unemployment to be the number one problem. At the time of the survey, 7 out of 10 Latin Americans were concerned about becoming unemployed within the next year. Even in countries with relatively low unemployment rates, such as Chile and Uruguay, this number was 5 out of 10. This con- cern may have increased since the Asian and Russian crises. Workers con- sider education and low wages the next two most important issues.3 On the other hand, employers worry more about the deficiencies in labor sup- LABOR REFORM IN LATIN AMERICA DURING THE 1990s 3 ply and inadequate worker training, fearing that the educational system does not meet the needs of a more competitive economy. A strikingly large percentage of Latin Americans now call for more government action: 88 percent of respondents in the Latinobarometro survey believe it is govern- ment's responsibility to give work to everyone, and 85 percent believe gov- ernments have to provide a decent standard of living for the unemployed. The design of labor laws in many Latin American countries shares some of the responsibility for these poor outcomes. Labor legislation in Latin America was originally enacted and designed to "protect" male household heads who worked for the government and other large- and medium-size employers. Over time, with the increases in female labor force participa- tion and the size of the informal sector, the proportion of workers not cov- ered by the legislation has increased, thus gradually increasing the tension between legislation and labor market realities. With stabilization and trade liberalization during the late 1980s and the 1990s, labor laws may have come under additional strain as the costs of "protection" of some workers increased. These pressures may have made Latin American labor markets ripe for reform in that the dissatisfaction may be peaking. But the desired direction of change is far from obvious. The aim of this book is to help inform policymakers and others about the most promising directions of change in labor market regulation. Some Latin American governments have already reacted to these con- cerns by introducing more or less profound changes in their labor regu- lations. In most Latin American countries, constitutions or labor codes, which are based on the civil law system, regulate the type and duration of labor contracts and the conditions for their enforcement.4 For the pur- poses of this book, we define labor market reforms as changes in labor laws and regulations. In this chapter, we describe the basic labor market re- forms that took place during the 1990s, and which are related to the top- ics that seem to most concern workers in the region-unemployment, job stability, and informality. The labor laws we focus on in this chapter are those that affect job security while workers are employed, and that provide support-such as severance payments, individual severance funds, and unemployment in- surance-during unemployment. We discuss the pros and cons of the measures taken and summarize existing evidence on the effect of these re- forms on the welfare of Latin American workers. We survey the reforms that are likely to affect labor turnover in the formal sector, such as tem- porary contracts, job security legislation, and income support for the un- employed.5 Public sector reforms may have underlined the pessimism in 4 CRAFTING LABOR POLICY Latin America, as some workers who held secure jobs suddenly faced the prospect of joblessness. For this reason we also provide a short summary of the most important public sector reforms. Given the large share of in- formal workers in the Latin American labor market, we explicitly deal with the issue of coverage and noncompliance. The final section concludes with an introduction to the six country chapters in this book-two each for Argentina, Brazil, and Chile-and a brief discussion of how this book addresses some of the main concerns of Latin American workers. Reforms in Job Security Legislation Job security legislation in Latin America varies considerably across coun- tries.6 Among the few common elements are limitations on temporary hir- ing and tenure-related severance payments that raise the cost of hiring "permanent" workers. These regulations came under attack during the 1990s, when it became clear that economic growth by itself would not be enough to increase job creation and job quality.7 The trade, financial, and fiscal reforms that were introduced during the 1 990s reduced inflation and increased competition. Inflation had been an important instrument for adjusting wages during the 1980s. Proponents of labor market reform claim that the decrease in inflation exacerbated the rigidity in real wages, and forced an adjustment through formal employ- ment instead. Further, it is argued that increased competition aggravated the negative effects of this rigid labor legislation, as firms required in- creasingly more flexibility to adjust their workforce when economic con- ditions changed in a competitive environment. Changes in relative prices and more intensive use of machinery in the tradable sectors in the wake of trade reforms and stabilization led to an increased share of employment in the nontradable goods sectors. This reallocation of labor among the sec- tors reflected changes in the size and composition of the workforce at the plant level. The restrictions on the firing of workers increased the costs of these adjustments, and may have prolonged the transitional phase and hindered an efficient allocation of workers. They may also have reduced job creation, because high dismissal costs affect the hiring decisions of firms, especially in an environment with intrinsically greater economic uncertainty. Finally, proponents of reduced job security pointed out that rigid labor legislation was likely to encourage noncompliance, and hence increase the size of the informal sector. For some firms this implies pressures to pro- duce at a smaller (possibly less efficient) scale in order to avoid detection LABOR REFORM IN LATIN AMERICA DURING THE 19905 5 by tax and labor authorities. According to this argument, reducing job se- curity might increase the number of firms that offer formal contracts, and hence the proportion of workers covered by labor legislation. Further, it might increase productivity by not distorting the decision of the optimal size for the firm. To summarize, deregulating the labor market would lower labor costs, which would result in more labor-intensive production and higher job creation in the formal sector and enhanced compliance, leading to a reduction of the informal sector and, hence, an improvement in the quality of jobs. But this line of argument must answer certain questions. First, if job se- curity legislation is so bad, why was it implemented in the first place? There is some evidence that job security decreases the volatility of em- ployment and improves the welfare of certain groups who are likely to be better represented in the political process. For them, many of the regula- tions that are considered distortionary actually increase the quality of jobs in terms of wage, work amount, and job stability.8 Those who suffer from the regulations and who could benefit from deregulation are often politi- cally less well represented. Second, while it is claimed that job security reduces the level of em- ployment, where is the evidence? Actually, the theoretical and empirical ev- idence on the net effects of firing costs on employment and unemployment is ambiguous. Hence, benefits of reforms are often not well understood, which may introduce some reluctance on the part of policymakers to undertake reform. However, high labor market flexibility increases the volatility of employment, leading to higher firing relative to hiring during recessions, and increased hiring relative to firing during booms. This may contribute to the fear of workers of introducing labor market flexibility in an environment that is already characterized by great economic volatility. Saavedra (1999) relates a telling anecdote about the popularity of job se- curity in Peru. When the Peruvian public was informed in 1996 that a draft law reducing mandatory severance payments was being considered, it created such an uproar that the Peruvian government quickly recanted, claiming that the purported reduction in benefits was in fact a typograph- ical error. It ended up increasing mandated severance payments instead! The fact that deregulation of labor market reforms is difficult to imple- ment might be reflected in the low number of countries in Latin America that actually undertook serious reforms. While 23 out of 26 countries in the region carried out profound trade reform and 24 notably liberalized their financial sectors, only 6 have introduced substantial changes to their labor legislation since the late 1980s: Argentina (1991, 1995, 1999); Brazil 6 CRAFTING LABOR POLICY (1988); Colombia (1990); Guatemala (1990); Panama (1995); and Peru (1991). And while it is generally believed that macroeconomic, financial sector, and trade reforms have been steps in the right direction, it is far from obvious whether changes in labor laws qualify as improvements. Temporary Contracts To promote stable employment, 14 out of 26 countries in Latin America restrict the use of temporary contracts or fixed-term contracts. Until the 1990s, legislation either explicitly prohibited the use of temporary con- tracts for permanent activities, or restricted the duration and number of times these contracts could be renewed. Temporary contracts were usually allowed only for work temporary in nature, such as harvesting or con- struction. During the 1990s, Argentina, Brazil, Colombia, Ecuador, and Peru introduced new rules for contracts, especially temporary contracts. Essentially, the duration of periods during which the work relationship was legally classified as "temporary" was extended. During this time these contracts have lower dismissal costs. In addition, the employers are some- times required to pay lower payroll taxes for such workers, making them "special" as well as temporary. Advocates of the more extensive use of temporary contracts in Latin America point out that temporary contracts reduce the cost of employ- ment adjustment, especially in an environment characterized by high dis- missal costs. Temporary contracts allow firms to hire workers during peri- ods of high labor demand, such as seasonal work or during booms. Using temporary contracts for this purpose was not generally forbidden by leg- islation in most Latin American countries, but the administrative costs of using temporary contract provision (for example, requiring labor union consent in each case) may have been high. The second argument often put forward is that firms might use temporary contracts as a screening device. This improves the allocation of workers, leading to efficiency gains. A final argument is that temporary contracts might open up new possibili- ties for workers who traditionally have low employment possibilities, such as women and youths. Women and young workers have the highest un- employment rates in Latin America. Permanent contracts may induce dis- crimination against women if there are higher implicit costs of hiring women due to maternity leave and allowance. Temporary contracts might help young workers gain experience, and thus facilitate their incorpora- tion into the workforce. Opponents of temporary contracts claim that they reduce job stability. Although temporary contracts facilitate adjustment of a firm's workforce, LABOR REFORM IN LATIN AMERICA DURING THE 1990S 7 these contracts might imply that the workers subject to them bear the en- tire brunt of adjustment, further increasing the rents of workers with per- manent contracts. Screening good workers from bad is unlikely to be the primary objective of temporary contracts in Latin America, since most Latin American countries have probationary periods, ranging from 2 months in Colombia to 12 months in Brazil and Chile, even for regular contracts. Whether young workers really benefit from temporary contracts is also not clea-, because the short time. temporary workers are in the firm reduces incentives for the firm to invest in their human capital. A last crit- ical argument regarding the use of temporary' contracts in Latin America concerns the unstable political environment. If employers fear that the existing labor regime might change, they might use temporary contracts extensively, because they would not incur new, unpredictable obligations if the law were changed. Argentina, Brazil, Colombia, Ecuador, and Peru introduced new for- miulas for temporary contracts. The most important changes were proba- bly introduced in Argentina, Colombia, and Peru (see Table 1-1). In Argentina, temporary contracts were illegal in all cases, except where ex- plicitly allowed before the 1991 reform.9 Temporary contracts could not TABLE 1-1 Changes in Temporary Contract Legislation Country Year Chahge Argentina 1991 Introduction of fixed-term and training contracts to promote employment and training with lower mandatory severance payment and payroll taxes. 1995 Introduction of trial period provision. Special employment promotion contracts, special conditions for small firms streamlining use of temporary contracts. 1998 Abolition of most of the exemptions from severance payments and payroll taxes and reduction of probation period to one year. Colombia 1990 Reduction of minimum duration of temporary (fixed-term) contracts. Peru 1991 Introduction of different forms for temporary contracts. Extension of eligibility criteria. Reduction of administrative costs of implementing temporary contracts. 8 CRAFTING LABOR POLICY exceed five years in duration and required a severance payment. The 1991 reform introduced new types of contracts, called modalidadespromovidas, with the explicit objective of promoting reemployment of former public sector workers who had lost their jobs during public sector reform in Ar- gentina. To facilitate the incorporation of young workers in the labor mar- ket, contracts were created that aimed at providing training and work ex- perience for those with previous training. Each contract was applicable only to specific workers (unemployed, laid off as a consequence of public sector cutback, or less than 24 years of age) or firms (new firms, new lines of production, or services in existing establishments). Duration of these new types of contracts ranged from six months to a maximum of two years. The contracts, however, applied only if trade unions approved them, and were required to be registered at the employment office. This may have severely reduced their expansion. Hopenhayn (2001) finds that less than 5,000 new employment contracts were registered every month for the entire country. Guasch (2000) points out that only the largest compa- nies with the greatest turnover rates actually took advantage of the law. In 1995, a special employment promotion contract for workers over 40 years of age was introduced that did not require payment of severance at termination of contract, and lowered the labor taxes that employers had to pay. Labor taxes and severance payments of the training promotion con- tract increased. A new provision introduced a three-month trial period for all new contracts, during which time no severance payments for termina- tion had to be paid. This provision applied to all new contracts. Small firms were allowed to use the previous employment promotion contracts, without the requirements of the approval of trade unions, the need to reg- ister the contract in the government employment agency, and to pay sev- erance. This new reform contributed to a more extensive use of temporary contracts. While at the beginning of 1996 the contracts did not surpass 5 percent of total employment, toward the beginning of 1998, 12 percent of formal employment were covered by them. Forty percent of the total net formal jobs that were generated between January 1996 and January 1998 were subject to these new contracts (Guasch 2000). Using the labor-cost elasticities presented in Chapter 5, "Increasing Labor Demand in Argentina," Guasch (2000) calculates that 13 percent fewer jobs would have been created in the absence of the new contractual forms. Hopenhayn (2001) shows that the probability of exiting employ- ment (that is, becoming unemnployed or leaving the labor force) increased during the trial period by about 40 percent. Hopenhayn does not find size- able effects for small firms and young workers. This may be explained by LABOR REFORM IN LATnN AMERICA DURING THE 1990S 9 the fact that temporary contracts implied lower taxes, which induced an in- crease in hiring, but at the same time replaced longer-term employment with short-term employment. Despite the economic benefits of such con- tracts, this reform was not politically popular. In May 1998, the Argentine Congress passed a law abolishing most of the exemption from severance payments and payroll taxes, and reduced the probation period to one year. In Peru temporary contracts were introduced in the 1970s. Many pro- vided social benefits but did not have an employment protection clause. Consequently, even before the reform in 1991, about 20 percent of work- ers in formal firms were subject to a temporary contract. However, the ad- ministrative costs of implementing temporary contracts were high, since they required the prior authorization of the Ministry of Labor. The 1991 reform reduced these costs and extended the reasons for hiring workers under temporary contracts. This resulted in a substantial increase in the use of temporary contracts from 1991 to 1996, cutting formal permanent wage employment from 80 percent in 1991 to 39 percent in 1997 (Pag6s 1999; Saavedra and Torero 2000). This example highlights something that employers and policymakers know well but economists need to be re- minded of: when assessing changes in job security legislation, the implied changes in administrative costs should not be ignored. In Colombia, temporary contracts-which did not require severance payment if they ended on the agreed date-were limited to a year's dura- tion. The reform of 1990 relaxed this minimum duration. At the same time, permanent contracts also became more flexible, as we will discuss below. For Colombia, Kugler (2000) finds that the labor legislation reform substantially increased the exit rates out of employment. After the reform the probability of exiting employment was 7 percent higher for temporary formal sector workers than for temporary informal sector workers, but only about 6 percent higher for permanent formal workers than for per- manent informal workers. This suggests that temporary contracts partially increased turnover of formal workers, but their effect on the increase in exit rates out of unemployment was relatively small. Given that permanent contracts have high dismissal costs, temporary contracts have helped in circumventing rigid labor laws. In the absence of such avenues, firms may evade compliance with labor legislation entirely, for example, by remaining small. Larger firms resort to legal methods of avoiding these restrictions through, for example, subcontracting smaller firms to deliver the services that workers might have provided if these laws were less onerous. Indeed, during the 1990s, subcontracting increased in most countries. The International Labour Organization (1999) argues 10 CRAFTING LABOR POLICY that the economic reforms increased the pressure on big firms to improve their productivity, leading to an increased tendency to subcontract for cer- tain activities. But this might also be due to new (unrelated) policies of re- structured firms to subcontract to, for example, firms that might them- selves subcontract cleaning and security services to smaller firms that hire employees informally (de Ferranti and others 2000). Temporary contracts have been an attractive instrument for introduc- ing labor market flexibility at the margin, and thus avoiding more pro- nounced labor reforms. While these contracts are beneficial to employers because they reduce adjustment costs, they do not necessarily increase worker welfare. Temporary contracts in Latin America might introduce an intermediate layer of job security regulation that lies between the formal and informal sectors. Their design will determine whether they place potentially formal workers into a category that lies between the formal and informal in terms of employment security and nonwage benefits, or whether they draw informal employees into the realm of regulated work- ers. This points to the fact that understanding the impact of labor legisla- tion on the informal sector is an important task in assessing the effect of labor market reforms on worker welfare. Chapter 3, "Assessing the Impact of Regulations on Informal Workers in Brazil," tries to do this. Severance Payments The level of mandated severance payment when a "permanent" contract is ended varies widely among Latin American countries, always increases with tenure, and is related to cause. The basic formula for calculating a severance payment is generally a proportion of monthly wages times years of service. Often a ceiling is imposed, and workers with high tenure usu- ally have additional rights, as will be discussed later. During the 1 990s, de- creases in the level of severance payment were introduced in Colombia, Guyana, Guatemala, Nicaragua, Peru, and the Reptiblica Bolivariana de Venezuela, while Brazil (1988), Chile (1991), and the Dominican Repub- lic (1992) increased the amount a firm had to pay upon dismissal of a worker. In Brazil the increase was substantial. Originally, severance payments were designed to protect workers against "unjust dismissal," not to provide dismissed workers with income support. This was achieved by imposing additional costs on a firm for fir- ing a worker. There is clear evidence that severance payments decrease the probability of a worker losing his or her job. For Brazil, Paes de Barros, Corseuil, and Bahia (1999) find that an increase in severance payments LABOR REFORM IN LATIN AMERICA DURING THE 19905 II increases the duration of employment spells. The Colombian reform in 1990 reduced severance payments, enlarged the definition of just dis- missal, extended the use of temporary contracts, and reduced the advance notice for mass dismissals. For Colombia, Kugler (2000) finds that the average tenure of formal workers (who are covered by the reform), relative to informal workers, decreased. This was largely due to a decrease in the average tenure of formal workers. Employment protection may have further positive impact on labor pro- ductivity. Firms will invest more in training if they expect employees to stay with them longer. Productivity improvements also crucially depend on a cooperative working environment, which is usually not achieved in a hostile atmosphere. Given the high frequency of litigation with respect to severance payments in Latin America, the question arises whether the de- sign of severance payments-which are mandatory, not voluntary-actu- ally increases the potential for conflict rather than contributes to a coop- erative environment. Opponents of high levels of job security point out that mandated severance payments might deter job creation because they not only reduce firing, but also hiring, by firms. Low job creation was one of the major employment problems during the 1990s. However, from an empirical viewpoint this claim is not unambiguously supported. Some empirical evi- dence for Latin America and the Caribbean indeed suggests a significant adverse effect of job security on unemployment rates (Heckman and Pag& 2000).'° But using a rank indicator for job security, Marquez (1998a) finds that job security is not significantly associated with lower employ- ment once other factors are controlled for. Nor does empirical analysis for specific countries that changed job security legislation reveal a clear picture (notably Kugler 2000 for Colombia; Maclsaac and Rama 2001 for Peru); To compound the problem, in Latin America high dismissal costs are combined with weak enforcement institutions. Weak enforcement insti- tutions may imply that (even formal) firms offer job security only to those workers who they expect to stay with the firm in the near future. In that case, having job security is more an indicator of the quality of a worker than a means of protecting workers against unjust dismissal. MacIsaac and Rama (2001) find that in Peru private sector workers who are legally' en- titled to severance payments earn lower wage rates than workers who are not. But they work longer hours, so their monthly earnings are higher than those of workers who do not receive severance payments. This im- plies that the objective of protecting workers against unjust dismissal is not served well in the first place."I I2 CRAFTING LABOR POLICY What is perhaps even more important is the effect of severance pay- ments on informality. As pointed out, most informal firms are actually small."2 High levels of severance payments are especially difficult to han- dle for small firms, because they are more likely to be credit constrained and face more uncertain production and sales conditions (Levenson and Maloney 1998). High dismissal costs therefore increase the premium for small firms to stay informal. But most of the jobs during the 1990s were created in these small firms, implying an increase in the number of work- ers not officially protected by labor legislation. To reduce the informality premium and thus increase the coverage of job security, one has to either lower the dismissal costs or increase the probabil- ity of being detected and fined in case of noncompliance.'3 While some countries opted for a decrease in severance payments, Argentina, Brazil, and Chile tried to strengthen their enforcement institutions during the 1990s. In Argentina a law was introduced in 1991 that established fines for the em- ployer in case a worker is not registered or is registered incorrectly. In 1991 the Chilean government increased the resources devoted to the enforcement of the labor legislation and carried out mass-media campaigns. Regulations that provide more effective access to the legal system for lower-income workers and legislation that allows more appropriate control of conmpliance were streamlined (Cortazar 1997). Since 1998 Brazil has stepped up efforts to tighten the enforcement of social security legislation and has increased penalties for noncompliance. While the effect of severance payments on aggregate employment is not unambiguous, design of severance pay legislation leads to a protection of some workers at the expense of others. i4 This is especially true if severance pay is tenure related. MacIsaac and Rama (2001) find, for example, that in Peru only one in five private sector workers and one in three private sec- tor wage-earners is actually entided to receive severance payments, and these covered workers tend to be wealthier.15 Entitlement to severance pay is nearly twice as high for the richest quintile group as for the poorest. This gap increases when coverage indicators are used, such as the presence of a written contract, coverage by social security, and employment in union- ized or large enterprises. On the other hand, it may segregate the labor market. But this is not the whole story. If high dismissal costs are com- bined with weak enforcement institutions, this may weaken the effect of high severance payments on aggregate formal employment, but is likely to increase the size of the informal sector. Severance pay, however, may affect not only the composition of em- ployment with respect to the informal sector, but may increase the formal LABOR REFORM IN LATIN AMERICA DURING THE I990S 13 employment rates of those protected by the legislation relative to those who are excluded. Apart from informal sector workers, these are lower- tenured workers, who are usually younger, less experienced, unskilled, and female. For Colombia, for example, Kugler (1999) finds that average un- employment spells of women, young and middle-aged with incomplete secondary and incomplete university education, decreased. These effects become stronger when severa-nce pay is related to tenure. Severance pay laws often intend to reward workers for long-term com- mitment to a firm. Accordingly, severance payments usually increase with tehure. Additionally, workers with certain years of tenure are given special rights. In Brazil, until the institution of individualized severance funds, workers with more than 10 years of tenure could not be fired except on rare disciplinary grounds. Before the 1990 reform in Colombia, workers with more than 10 years of tenure could sue the firm in case of dismissal. If the firm could not prove just cause, it had to rehire the worker. More- over, it had to pay the total amount of wages that the worker had foregone during the proceedings, and the firm had to take over the worker's pen- sion at the time of his or her retirement (Lora and Henao 1997). Similar rights of reinstatement also existed in Panama and Peru. As expected, tenure-related severance payments lead to lower employ- ment rates for young relative to older workers in Latin America, because they reduce hiring and increase firing of less-tenured workers relative to more-tenured workers. Using an index for job security, Heckman and Pages (2000) find that an increase in expected dismissal costs especially af- fected the employment rates of young workers. For Chile, Pages and Mon- tenegro (1999) provide evidence that tenure-based job security biases em- ployment and wages in favor of middle-aged and older workers and reduces long-term aggregate employment. They do not find a visible im- pact on unemployment, because the reduction in youth unemployment coincides with a reduction in labor force participation among the youth. Hence the authors recommend a flat severance payment, because this would have little effect on youth employment or on aggregate employ- ment and unemployment rates. This requires either reducing the maxi- mum amount a worker can receive as severance pay, or reducing the rate at which the severance pay increases with tenure. Tenured workers are made better off relative to untenured workers or those with lower tenure if the maximum ceiling increases (as in Chile) or if slope of the severance pay-tenure profile increases (as in Colombia, the Dominican Republic, and Peru). Only Argentina seems to have reduced severance costs for higher-tenure workers during the 1990s (see Table 1-2). 14 CRAFTING LABOR POLICY TABLE 1-2 Changes in How Severance Pay Increases with Tenure Country Year Change Argentina 1991 Maximum ceiling imposed.a Advance notice of one month independent of tenure. No severance payment for termination within trial period of three months. Brazil 1988 EmpLoyer must contribate every month the equivalent of 8 percent of each employee's current monthly wage, which is accumulated in a government-administered individual account. The penalty for dismissal without just cause was raised from 10 to 40 percent of this accumulated balance. Chile 1991 Maximum ceiling was raised from 5 to 11 months' wages. Workers with more than six years of tenure could opt out of job'security provision and participate in "unemployment fund." Colombia 1990 Workers with more than 10 years of tenure lost the right to sue for back pay and reinstatement, but cost of dismissing workers with more than 10 years of tenure increased. Dominican 1992 Severance payment increased for all new employees, Republic but to a larger extent for workers with more than five years of tenure. Nicaragua 1996 Severance payment for workers with more than three years of tenure increased. Maximum ceiling of five monthly wages was imposed. Panama 1995 Severance payment converted to 3.4 weeks per year of service if less than 10 years of service. Two additional weeks otherwise. Workers with more then 10 years of tenure receive 1.5 months less than before if fired because of economic cause. Peru 1991 Job security law eliminated for new contracts. No severance payment for workers with less than one year of service. LABOR REFORM IN LATIN AMERICA DURING THE 1990S I5 Country Year Change 1995 Job security eliminated for all contracts. Severance payment for workers with less than three years decrease compared with 1991. 1996 Severance payment increased for all workers. Maximum ceiling became available for workers with 8 years of tenure (in 1991, 12 years). Venezuela, 1997 Severance payment increased from one monthly R. B. de wage per year of service to two monthly wages per year of service. a. Such that maximum yearly payment cannot exceed the equivalent of the average of three months' salary determined in the collective bargaining agreement that applies to the worker. Note that the maximum ceiling existed until September 1989, when the Economic Emergency Law eliminated it. The changes may indicate that decreasing the rights to tenure is difficult because this is likely to reduce employment rates for older workers. Older workers exert pressure to maintain their level of job security provisions, and these workers are usually better represented in the political process. Policymakers can either try to pursue broad coalitions of the groups which will be favored by the reform, or apply changes only to new contracts. On the other hand, consideration should be given to the fact that older dis- placed workers might be especially hurt by layoffs due to their inability to update their skills, or by the unwillingness of firms to employ them for other reasons. In the 1990s, older workers in Peru displayed a higher prob- ability of becoming informal, and in Argentina and Uruguay unemploy- ment spells of older workers became longer (de Ferranti and others 2000). If the basic idea of introducing severance payments is to protect work- ers against unjust dismissal-for engaging in union activity, for exam- ple-it is understandable that labor legislation in most Latin American countries mandates that workers usually receive severance payments only when dismissal is not justified. Justified dismissal refers to serious mis- conduct as defined in the labor legislation. Further, consistent with the idea of protecting workers against unjust dismissal is the fact that eco- nomic problems of the firm are not considered a reason for just dismissal. However, it is often claimed that the objective of severance payments is stability. The exclusion of dismissals due to economic reasons from the list of justified reasons is consistent with the idea of worker protection. Where this is not true, as in Chile, it may provide an important loophole for i6 CRAFTING LABOR POLICY evading job security legislation. Cortazar (1997) points out that in Chile before 1990 about half of the termination of labor contracts was due to economic cause.16 Further, the fact that access to severance payments de- pends on cause implies that employees attempt to turn voluntary quits into dismissals in order to receive severance payments, which has led to high litigation costs. For example, Rama and Maloney (2000) point out that most of the complaints registered in Latin American labor courts are related to severance payments. In most Latin American countries no unemployment insurance system exists. Precautionary savings are low and workers are largely credit con- strained. Because severance payments provide funds at the moment of firing, they also help protect workers from income loss in the period im- mediately following unemployment. Consequently, severance payments play the role of unemployment insurance. But as a means of providing income support for the unemployed, their relation to cause-and the re- sulting uncertainty about the amount the worker will receive-contri- butes to their inefficiency. Employment distortions introduced by sever- ance payments could be evaded if workers would forego part of their earnings when employed, which they could use in case of unemployment independent of the cause of job separation. This is the idea underlying compensation funds. Considering severance payments as a type of unemployment insurance implies then that the uncertainty about obtaining payments should be re- duced. Table 1-3 indicates again that no clear trend emerged among Latin American countries. However, as we will see, during the 1990s Chile, Colombia, Panama, Peru, and the Republica Bolivariana de Venezuela in- troduced compensation funds, and these were the countries which actu- ally also seemed to increase access to severance payments. Whether mandated severance payments are likely to be useful tools for unemployment insurance is doubtful. First, coverage may be low and richer workers are more likely to be covered by severance pay legislation, largely because of the exclusion of the informal sector. MacIsaac and Rama (2001) find that in Peru only one in five private sector workers and one in three private sector wage earners is actually entitled to receive severance payments. 17 Wealthier workers are covered by the legislation: eligibility for severance pay is twice as high for the richest quantile compared to the poorest. This gap increases when the authors use coverage indicators. Second, in Latin America unemployment rates are especially high for female, young, and low-skilled workers (Marquez 1998b), and these are the groups whose employment rates are likely to be depressed by the sev- LABOR REFORM IN LATIN AMERICA DURING THE I990S I7 TABLE 1-3 Changes in Severance Payment Laws and Cause of Dismissal Counny Year Change Argentina 1991 If dismissal is due to just cause, severance payment is half of the payment for unjust cause. Economic reasons now considered just cause. Bolivia 1990 Access to same severance payment for quits and layoffs after five years on the job. Brazil 1988 In case of dismissal without just cause, employer has to pay 40 percent of severance fund balance; this was 10 percent until 1988. Chile 1991 Clause stipulating that economic reasons are a just cause was reinstated. Failure to prove economic cause raises legal severance by 20 percent. Failure to prove just cause raises legal severance by 50 percent. Workers participating in unemployment fund have access to fund independent of cause. Colombia 1990 Just cause extended to include any dismissal for failure to comply with firm regulation and instructions from one's supervisor. Nicaragua 1996 Workers who quit receive a severance payment. Severance payment paid in case of economic cause increases. In both cases this accounts to monthly wage times years of service up to the third year, and increases thereafter. Panama 1995 One-fourth of monthly wage per year of service available for new contracts if worker quits (onty after 10 or more years of service). Compensation for dismissal in case of economic causes decreases for all new contracts. Peru 1991 The definition of "unjustified dismissal" made more restrictive. Access to compensation fund independent of cause. Venezuela, 1997 Access to compensation fund independent of cause. R. B. de I8 CRAFTING LABOR POLICY erance legislation. Further, given the relation of severance to tenure and wages, they also receive relatively low payments. Third, in general sever- ance payments do not link the amount received and unemployment du- ration, so severance pay may largely serve to bridge short-term unemploy- ment. Again, female, young, and low-skilled workers have the highest unemployment duration and least benefit from severance pay. Fourth, the relation of severance payment to cause limits its ability to ensure against idiosyncratic risk. The fact that the firm does pay them and that often eco- nomic cause is excluded does not allow insurance against aggregate risk. De Ferranti and others (2000) point out that before trade reform these risks were effectively pooled over a larger population because higher prices mean that consumers effectively subsidized these firms. Given that more firms sell closer to world prices as trade regimes are liberalized, this is no longer possible. Finally, as pointed out, distortionary effects of severance payments are likely to be large. One way to decrease these effects is to re- duce uncertainty and direcdy link contributions and benefits. This is the basic idea of severance compensation funds. Compensation Funds Compensation funds are individualized accounts to which workers regu- larly transfer a part of their salaries. Workers can draw on their accounts when they are separated from their jobs. Sometimes partial withdrawals from the fund are possible for housing, education, or health expenses. These funds provide a form of forced self-insurance. In contrast to sev- erance payments and unemployment insurance, they provide no net transfers to workers who lose their jobs. Mandatory self-insurance may be welfare-improving if the government cannot credibly state that it will not intervene if individuals become unemployed. It evades the employment distortions introduced by the severance pay regulation because it in- troduces a direct link between payments and contributions. Compensa- tion funds are not designed to increase job stability; they are created to protect workers against the income loss they face at the moment they be- come unemployed. Compensation funds have been introduced in Chile, Panama, Peru, and the Republica Bolivariana de Venezuela. Colombia reformed its exist- ing compensation fund. This may reflect a higher demand for unemploy- ment insurance and the fact that it is easier to introduce compensation funds than to change severance payment legislation. This is reflected in the fact that in Colombia, Ecuador, Peru, and the Repiublica Bolivariana LABOR REFORM IN LATIN AMERICA DURING THE I990S Ig de Venezuela, in the event of dismissal, workers receive both compensa- tion from the fund and severance payments. The relatively high coverage of the compensation funds is consistent with the evidence that workers do not opt out of them when given the choice. In Colombia 47 percent of urban workers are covered by the com- pensation fund. Kugler (2000) finds that in 1995 only 1.5 percent of workers in manufacturing and 0.6 percent of workers in commerce pre- ferred a higher salary in exchange for not being covered by the Colombian compensation fund, Fondo de Cesantia. The Fundo de Garantia do Tempo de Servifo (FGTS) in Brazil, which was created in 1966 and provides an individualized fund for each worker, is atypical in the sense that it combines the fund with an extra fine which can be assessed only in case of dismissal. Every month the employer con- tributes 8 percent of his or her employee's current wage to the fund. This implies that the worker's FGTS increases with tenure and that the amount accumulated in the Fund per year corresponds approximately to one monthly wage. In general, the FGTS can be accessed by the worker only upon dismissal (not if he or she quits) or upon retirement."8 When a worker is dismissed without just cause (semjusta causa) the employer must also pay a penalty that is proportional to the FGTS balance of the worker. Thus, when dismissed, the worker gains access to his or her accumulated balance in the FGTS, and additional compensation of 10 percent of the FGTS balance at the time of dismissal if it was judged as being without just cause. In 1988 this fine for unjust dismissal was increased to 40 per- cent of the FGTS (see Table 1-4 and Chapter 3). Unemployment Compensation The purpose of unemployment insurance is to provide short-term finan- cial support to workers if they are laid off from their jobs, and not to change the firing decisions of firms. Unemployment insurance systems may vary in their eligibility criteria, the level and length of benefits, and the way they are administered, depending on country-specific labor mar- ket characteristics and national consensus on secondary objectives. The programs may prevent unemployed workers from sliding into poverty, which means that unemployment insurance could provide a safety net. On the other hand, unemployment insurance can provide a worker with the opportunity to search for a job that better corresponds to his or her skills and interests, and may also induce workers to hold out for higher- wage jobs. An unemployment insurance system may also encourage work- 20 CRAFTING LABOR POLICY TABLE 1-4 Changes in Severance Compensation Fund Legislation Country Year Change Brazil 1988 Fine for unjustified dismissal increased from 10 percent to 40 percent of worker's FGTS, which is equivalent to the worker's current monthly wage. Access to fund independent of cause. Chile 1990 All workers with tenure above six years can opt out of severance payment and participate in fund. Employer deposits 4.11 percent of gross wages in a bank account bearing worker's name. Colombia 1990 Employer contributes one month of yearly salary to a capitalized fund bearing the worker's name. Access independent of cause. Panama 1995 Firms are obliged to create a collective fund which is managed by a private entity. Employers contribute to this fund. Peru 1995 Employer deposits half a month's salary twice a year in a bank account bearing the worker's name. Access independent of cause. Venezuela, 1997 Employer deposits fund in a collective account. R. B. de Consequently, the severance payment does not equal the accumulated capital, but is last year's salary times years worked times the interest on the accumulated funds. If the firm cannot prove that the cause is justified, it must pay a supplementary severance. ers to apply for high-wage jobs with high unemployment risk. In response, firms may offer jobs with higher unemployment risk, higher capital-labor ratios, and higher wages, implying that moderate unemployment insur- ance may not only improve risk-sharing, but also increase output.19 Fi- nally, an unemployment insurance system may help as a macroeconomic stabilization tool. Countercyclical spending due to unemployment insur- ance may improve consumption levels and welfare during recessions and thus insure against both idiosyncratic and aggregate risk. To satisfy this LABOR REFORM IN LATIN AMERICA DURING THE I9905 2I objective, the unemployment insurance system requires covering a large part of the population and having high enough fiscal resources that allow paying benefits even during recessions. Opponents of unemployment insurance systems for Latin America point out its demanding administrative requirements costs and the com- mon moral hazard issue. Building up an unemployment insurance system is admittedly not an easy task. For example, in the early stages of the un- employment insurance system in Brazil, the delivery mechanism did not work very well, so some workers had to wait up to a year to receive their checks (Mazza 1999). But administrative capacity can be built and ad- ministrative effort reduced by, for example, relying on already existing in- frastructure and keeping the benefit structure simple. Moral hazard-the fact that the unemployed reduce their job search effort in the presence of unemployment insurance-can lead to an increase in the duration of un- employment. This issue can be tackled by keeping benefit levels low and the duration of benefits short, and by imposing certain work search re- quirements. This issue is discussed in more detail in Chapter 7, in the con- text of concerns in Chile that employment has become more precarious over the last decade. The existence of a large informal sector in many Latin American coun- tries introduces further complications concerning implementation of an unemployment insurance system. The most obvious problem is that a large informal sector makes it difficult to determine whether persons are actually unemployed or, rather, working in the informal sector. Argentina, for example, has tried to deal with this problem by increasing the cost of collecting unemployment benefits in terms of time (for example, having to collect payments in person, during peak business hours, and waiting in long queues). The existence of a large informal sector also makes it doubt- ful that the objectives an unemployment insurance system may fulfill are in fact achieved. First, the high proportion of informal workers who are not covered by unemployment insurance limits the usefulness of unem- ployment insurance systems as a safety net. Second, if workers do work in the informal sector while receiving unemployment benefits, this may un- dermine the function of unemployment to improve worker-firm matches. Unemployment insurance then may even contribute to an increase in the informal sector by creating disincentives to finding jobs in the formal sec- tor as long as unemployment benefits are paid. Finally, the existence of a large informal sector reduces the strength of unemployment insurance as a macroeconomic stabilization tool. A problem unrelated to the design of these schemes is that unemploy- ment insurance systems in Latin America were added on to existing man- 22 CRAFTING LABOR POLICY dated severance laws. With social security also being mostly funded by high taxes on labor income, the tax burden is already high. In Argentina, Brazil, Colombia, and Uruguay the costs of contribution to Social Secu- rity exceed 30 percent of wage costs. In Organisation for Economic Co- operation and Development (OECD) countries the costs are between 15 and 30 percent.20 What is more important is the problem of combining unemployment insurance with severance pay and individualized compen- sation funds. This lack of coordination and overlapping of different pro- grams may introduce substantial inefficiencies. In some Latin American countries severance payments exceed the amount a worker receives from unemployment insurance. This may introduce the perverse effect that un- employment insurance is converted into a type of unemployment subsidy, rather than a means of cushioning the drop in income. Thus, implemen- tation of an unemployment insurance system must consider the existing mix of termination and severance benefits. In Latin America, Argentina, Barbados, Brazil, Chile, Ecuador, Mexico, Uruguay, and the Republica Bolivariana de Venezuela have some kind of special unemployment assistance programs, nearly all of which were implemented during the 1980s and 1990s. Eligibility criteria are rather strict; In Argentina, Barbados, Uruguay, and the Repuiblica Bolivariana de Venezuela access to unemployment subsidies requires a certain number of months of contributions to social security or the unemployment fund. The worker must be immediately available to work in Argentina, Chile, and the Repuiblica Bolivariana de Venezuela, and the worker has to be laid off for just or economic cause to have access in Argentina, Brazil, Chile, and Uruguay. Coverage of these programs is rather low. This may be because the benefits are small, as in Chile and Ecuador, or the worker has to prove economic necessity, as in Brazil. Reforms took place from the end ofthe 1980s in Argentina (1991 and 1995), Brazil (1990), and the Reptiblica Bolivariana de Venezuela (1989 and 1998). In Chile, an unemployment insurance bill was introduced by the gov- ernment in 1993, but not implemented. Mandated individual accounts with some pooling of risks were introduced in Chile in 2001. In the Republica Bolivariana de Venezuela, the unemployment insurance system was enacted in 1989, reformed in 1999, but not yet implemented. In Ar- gentina, the Employment Law implemented an unemployment insurance system in March 1992.21 Coverage of the Argentine unemployment in- surance system is low. Nationwide, the system covers only 10 percent of the unemployed. The average beneficiary of the Argentina unemployment LABOR REFORM IN LATIN AMERICA DURING THE 19905 23 insurance program is a prime-aged male without family responsibility (Mazza 1999). Fifty-one percent of the applicants earned 300 to 600 pesos per month and 17 percent earned more than 1,000 pesos per month, which was three times more than the maximum unemployment insurance benefit level in the late 1980s. Only 6 percent earn less the 300 pesos per month, which indicates that the unemployment insurance systems does not meet the role of a safety net.22 The scheme is not integrated with other programs. A person who is laid off can get severance pay and unemploy- ment benefit emergency funds at the same time. Whether the unemployment insurance system in Argentina meets the role of macroeconomic stabilization is also questionable, given that the government decreased unemployment benefits in times of recession. In the wake of the Tequila Crisis in 1995, the government officially lowered the maximum amount of benefits. In Chapter 4, "Understanding Labor Supply Dynamics in Argentina," female labor force participation in Ar- gentina is shown to exhibit strong procyclical fluctuations. This behavior is explained by the added worker effect, which implies that women enter the labor market when the earnings of their husbands decrease due to a job loss or reduction in work hours. In Argentina this aggravates unem- ployment rates during recession. Cutting unemployment benefits during this period may even increase the cyclical response of female labor force participation. In Brazil the current unemployment system was created in 1986. It un- derwent major changes in 1990 when a larger fund, called the Fundo do Amparo ao Trabalhador (FAT), was created. It is the largest in Latin Amer- ica, with an average of up to 4 million beneficiaries per year. Unemploy- ment insurance is completely financed by the employer, who pays a tax of about 0.65 percent of total business sales to unemployment insurance. Length of benefit duration usually ranges between three and five months. Benefit levels increase only slightly with the duration of work experience. Benefit levels range from the minimum wage to twice the minimum wage. Workers who are laid off by their employer and have a formal contract are eligible for unemployment benefits. The worker must have worked six months during the last three years to receive the minimum benefit. Once a worker has obtained unemployment insurance, he or she is not eligible for additional insurance for another 16 months. The fact that a very short employment period is required may mean that the system is heavily used by young workers, a finding confirmed by Mazza (1999). About 50 per- cent of the workers are 30 years of age or younger. The average beneficiary 24 CRAFTING LABOR POLICY receives about 1.5 times the minimum wage, is a male 35 years of age or younger, is in the service or industrial sector, and has an eighth-grade ed- ucation (Mazza 1999).23 Evidence on unemployment insurance in Latin America is still weak. In Argentina and Brazil unemployment insurance systems seem to support middle-income groups and not the poorest workers, due to the fact that poor workers are usually in the informal sector and have low employment spells, which disqualifies them from having access to the system. Severance payments do not seem to be an adequate way of providing unemployment insurance due the distortions they introduce, the uncertainty about the amount transferred, and the high legal costs they imply. Whether compen- sation funds or unemployment insurance should be used as a means of in- come support best suited for a country will depend on the frequency and duration of unemployment, and the extent of the informal sector and its administrative capacity. Furthermore, it should be kept in mind that exist- ing instruments do not seem to be adequate measures for insuring low- wage earners and informal workers against long-term unemployment. Public Sector Employment and Earnings Proponents of labor market reforms claim that the only reforms that may have a negative effect on aggregate employment in the short term are those related to privatization and public sector downsizing. The 1990s was a decade of privatization for Latin America. Large privatization programs decreased the share of government employment from 15.3 percent in 1990 to 13.2 percent in 1995. Argentina and Mexico privatized about 2 percent of the value of their GDP from 1990 to 1994, and Brazil had equally ambitious privatizations between 1996 and 1999. In terms of value, 43 percent of privatizations in the region occurred in the area of utilities, which were traditionally not open for private participation. The sale of banks and similar institutions contributed a further 23 percent. Argentina reduced the number of federal administration and railroad workers during 1990-94, reducing public sector employment from 10 percent of the total in 1990 to approximately 5 percent by 1998. Re- structuring the workforce was considered an important component of public sector reform, because wages accounted for about 70 percent of federal expenditures, not including interest payments and transfers. Under the federal administration program 400,000 workers lost their jobs. Re- hiring was forbidden by law. About 73,000 railroad workers lost their jobs (Haltiwanger and Singh 1999). LABOR REFORM IN LATIN AMERICA DURING THE I990S 25 Peru performed a large public sector downsizing (of the civil service) that involved more than 225,000 job separations from 1991 to 1993. Both lump-sum severance payments and pension enhancements were used to induce departure. About 112,000 workers received severance packages averaging about $ 1,000. One of the major issues was that the program was targeted badly. This further aggravated pre-existing shortages in human capital, because many of the most qualified civil servants left the public sector. Furthermore, rehiring of separated workers was significant, imply- ing that reductions of employment in the federal government were offset by increases at the regional level. When the dust settled, 163,000 of the originally dismissed workers had been rehired (Haltiwanger and Singh 1999). Reduction in public sector employment may also have contributed to an increase in the informal sector. Pages (1999) points out that the pub- lic sector reform and the privatization process reduced employment in these sectors by about 12 percent. Workers who lost their jobs due to this process were on average about 50 years of age and had not completed sec- ondary education. Due to difficulties in finding jobs in the formal sector, most of them opted for self-employment, contributing to the growth of informality in Peru. To summarize, privatization included a reduction in the workforce of the privatized enterprises, but Lora and Pages (1996) conclude that, with the exception of Argentina, there is little evidence that this had an effect on aggregate employment. A Road Map for the Book For Latin America, the decade of the 1990s was largely characterized by economic growth but low formal job creation. Informality and unem- ployment levels increased. These employment problems triggered discus- sions on the distortionary effects introduced by existing labor laws that mostly had remained untouched since the 1950s, and moved labor reform into the spotlight. This chapter discussed the principal labor market re- forms related to job security and the provision of financial support to the unemployed. The picture that emerges is one in which labor market reg- ulations were generally not characterized by profound changes, but were at best improvements at the margin. In some cases, even this would be a generous characterization. But the way forward is much less clear. Labor legislation in Latin Amer- ica is distinguished more by its heterogeneity than by its homogeneity. And even if a law is identical in two countries, its effect on wages and em- 26 CRAFTING LABOR POLICY ployment may be different due to differences in employment structure (such as changes in the composition of the labor force or the size of the informal sector), political history, administrative capacity and incentives for compliance, and other aspects of each country's labor legislation. This implies that labor markets have to be studied country by country in order to identify the reforms that would most benefit workers, firms, and gov- ernments, and that is what this book does for three countries in Latin America. Chapters 2 and 3 are dedicated to Brazil, which underwent changes in labor legislation in 1988. With the excepiion of social security reforms, few reforms of the public sector were successful. This is striking since reducing the fiscal burden of public employment was viewed as central to sustain- ing the stabilization begun in 1994. Public sector employment in Brazil is approximately 10 percent of the total, similar to the percentage Argentina faced before its public sector reform. Large-scale public sector reforms may reduce employment levels in the short term, and are difficult to imple- ment. Hence, identifying the most-needed public sector reforms is impor- tant. Difficulties in implementing public sector reforms and the possible negative effect on employment require a thorough analysis of where inef- ficiencies exist in the public sector, in order to target the reforms. The ap- proach in Chapter 2, "Improving Public Sector Efficiency in Brazil," allows us to identify those sectors within the public sector where the problem was one of excess employment, and where the problem was one of higher-than- market compensation. The chapter finds that while not all government employees earn more than similar private sector employees, public-private pension differentials are large. During 1999-2000, Brazil reformed the social security system for private sector workers, which has increased the existing public-private pension gap even further. Labor legislation in Brazil requires formal sector employers to sign a standard contract ratified by the Ministry of Labor, which is called the carteira de trabalho, or work card. The household dataset we use in our analysis provides information on whether or not a worker has this signed work card. This provides a unique opportunity to identify informal wage earners, who are usually identified from the characteristics of the firm they work for, and allows us to study the effects of labor legislation on informal sector workers. It is usually assumed that labor laws that increase labor costs increase the size of the informal sector, but the direct effect of labor legislation on workers is generally not considered. Chapter 3, "Assessing the Impact of Regulations on Informal Workers in Brazil," examines whether labor market regulations, such as minimum wages and maximum LABOR REFORM IN LATIN AMERICA DURING THE 19905 27 hours worked, distort the wage and hour distribution of informal sector workers as well. The chapter illustrates the pitfalls of making generaliza- tions about informal sector work being precarious or unprotected. The average unemployment rate in Argentina over the last decade was among the highest in Latin America (aside from Panama and Nicaragua), peaking in 1995 at about 18 percent. At the same time, Argentina faced large increases in the female participation rate. Chapter 4, "Understand- ing Labor Supply Dynamics in Argentina," examines whether female labor force participation in Argentina exhibits noticeable cyclical patterns, and whether this aggravates or ameliorates unemployment rates during reces- sion. We can then speculate about whether cutting unemployment bene- fits during recessions, as the Argentine government did during the Tequila Crisis during 1994-95, affects the cyclical response of female labor force participation and hence unemployment rates. Benefits of reforms are often not well understood, which naturally in- troduces a reluctance to reform. Hence, understanding the impact of reforms on employment is important. Chapter 5, "Increasing Labor De- mand in Argentina," provides numerical estimates of employment growth that would occur if labor costs were reduced due to labor market reform, using the wage elasticity of employment. Argentina introduced substantial changes to labor legislation during the 1990s (relative to most other Latin American countries), but reversed some of them by the end of the decade. It relaxed labor legislation on temporary contracts, introducing labor mar- ket flexibility at the margin. It increased the monthly basis for calculating severance payments, but widened the definition of just cause and imposed a maximum ceiling for higher-tenure workers. At the end of the decade, these reforms did not appear to have led to a significant reduction in labor costs. In much of Latin America, the increase in the supply of skilled workers did not keep up with the demand for skilled labor, leading to increases in real wages in favor of especially highly skilled workers, and widening the differentials among skilled and unskilled workers. Chapter 6, "Responding to Earning Differentials in Chile," provides evidence on returns to educa- tion across different quantiles, analyzes whether rewards to education dif- fer by socioeconomic status, and quantifies the effect of policies that im- prove access and quality of education for poorer sections of Chilean society. Besides the concern of rising.wage inequality, a perceived increase in instability of employment worried observers in Chile in the mid-1990s. Chile had been exceptional in the region because in 1981 it had already introduced significant labor market reforms, which involved fewer job se- 28 CRAFTING LABOR POLICY curity restrictions. The government not only reformed labor legislation re- lated to individual contracts, but also introduced changes in legislation concerning unions and collective bargaining. In contrast to the other countries in the region, basic labor market indicators since the 1990s were promising. Chile's per capita output grew by about 5 percent, and unem- ployment rates fell from more than 20 percent in the early 1980s to less than 6 percent. Between 1984 and 1994, average real wages grew by 28 percent. By 1995, poverty fell to about half of what it was in the early 1980s. Chapter 7, "Dealing with Employment Instability in Chile," ex- amines whether, in spite of these favorable developments, the Chilean labor market has become a less reliable source of earnings, because of in- creased precariousness of employment. In this book, therefore, we analyze this wide range of different labor market topics for Argentina, Brazil, and Chile. The policy guidance pro- vided takes country-specific conditions into account, but we use standard statistical and econometric tools that can be easily applied to similar prob- lems in other countries. While focused on labor market analysis, the book considers macroeconomic conditions and labor market institutions where relevant. Implementation of labor reforms becomes more likely if-due to changes in macroeconomic and labor market fundamentals-societies be- come less willing to keep labor market restrictions in place. In Latin Amer- ica, changes in labor legislation were not profound during the 1990s, but it is encouraging that after several decades of inaction, the first steps have been taken. Perhaps the next decade will see more progress in this area of regulation, and Latin American workers will take home more of the ben- efits of reform in other areas of economic policy. Notes 1. Small firms are more likely to be informal both because ir is easier to hide from legislation and because they generally do not benefit from formal contract- ing, risk-pooling, and other programs that may be advantageous for larger firms. According to the definition of the World Trade Organization, firms with less than five employees are considered informal. In this book we follow country-specific de- finitions of informality. For example, in Brazil informality is defined as wage-earn- ers and self-employed workers who do not have a signed work card, implying that social security and other contributions are not made by them or their employers. 2. Of course, not all small enterprises are dedicated to simple survival. There are also informal firms with relatively high levels of technology and human capi- tal. There is evidence that the proportion of qualified personnel in microenter- prises increased during the 1990s, but remains below the levels in larger firms. LABOR REFORM IN LATIN AMERICA DURING THE 19905 29 "Low-quality" jobs may arise because of a low level of technology, low social pro- tection, or noncompliance with existing norms. 3. Poverty and social diseases such as crime, corruption, and political violence appear to be relatively minor concerns. 4. This does not apply for Caribbean economies, which we will not consider in the analysis that follows. 5. While we recognize that in Chile in the 1990s there were substantial re- forms in the areas of union and collective bargaining, we do not address those reforms here. 6. By job security we refer to all those provisions that increase the costs of worker dismissal. 7. A similar debate emerged in Europe during the 1980s after the economic downturn in the wake of oil shocks. 8. For example, full-time, indefinite contracts; mandatory severance payments; notice periods; limits on working hours; surcharges for overtime, night and holi- day work; social security contributions; and minimum wages. 9. Temporary contracts were allowed only in the case of force majeure, lack or reduction of work due to seasonal peak in work, organization of congresses and reunions, requirement of urgent security measures, and extraordinary and transi- tory necessities of firms (Bour 1994). 10. Heckman and Pages (2000) and Pages and Montenegro (1999) use a job security index that also includes advance notification and foregone wages during any trial in which a worker contests dismissal. In this sense it is not a pure mea- sure of the effect of severance payments. 11. Furthermore, job turnover may be partially reduced because job search is reduced. 12. See footnote 1. 13. Note that increasing coverage of job security legislation may also imply a reduction in labor demand. A rare study on the extension of job security provi- sion in developing countries was done by Fallon and Lucas (1991), who find a drop in labor demand in Zimbabwe. 14. Groups, for example, that are not covered, are the informal sector, family workers, and workers with low tenure. Domestic workers often receive lower sev- erance pay. 15. Note that in Peru less than two-thirds of private sector workers are wage earners. 16. Severance payments do increase job stability, but they may also reduce the intensity of job searches by employed workers. Furthermore, it creates a potential for conflict, which may induce large liability costs. Dismissals in Peru before 1991 had to be approved by the government, and if considered "unjustified" the worker could opt for being reinstated instead of receiving a severance payment. The need for government approval was abolished by the 1991 reforms, and the right of reinstatement was abolished in the 1995 reform. In 1995 Panama reduced the 30 CRAFTING LABOR POLICY wages firms had to pay from the time of firing until the court decision was handed down. 17. Note that in Peru, less than two-thirds of private sector workers are wage earners. Severance payments are rather high. MacIsaac and Rama (2000) find that unemployed workers who receive severance pay have consumption levels about 20 to 30 percent higher than workers who continue to work! 18. Withdrawals are possible for large health expenses or the acquisition of a home. 19. For a theoretical development of this idea see Acemoglu and Shimer (1998). 20. The actual tax burden in a country is difficult to quantify because of issues concerning tax compliance and the role of nontaxable wages. 21. The benefits paid out by the unemployment insurance system in Argentina are financed by a 1.5 percent payroll tax on employers and 1 percent on employ- ees. Except for workers in the construction, agriculture, and domestic service sec- tors, all workers are eligible for the new unemployment insurance system if they worked a minimum of one year on the last job and were laid off without cause. Payment duration ranges from 4 to 12 months, depending on the length of em- ployment in the previous job. Benefits range from 150 to 300 pesos per month and have a median level of 284 pesos per month. This compares to an average wage rate of 840 pesos in October 1993 for individuals in Greater Buenos Aires working more than 30 hours per week, but at the same time, 30 percent earn less than 400 pesos .(Pessino 1997). Compensation decreases over time, a common means of addressing the moral hazard issue. 22. At the same time the informal sector in Argentina increased from approx- imately 25 in 1988 to 40 percent in 1998 (Mazza 1999). 23. Ramos (1 999) finds that households with income between 20 and 30 min- imum wages receive 43 percent of unemployment benefit payments, while only 5.5 percent accrue to households with income between 3 and 5 minimum wages. In the case of FGTS accounts, households with income higher than 30 minimum wages make 47 percent of the withdrawals, while only 2.3 percent are made by households with incomes between 2 and 5 minimum wages. References Acemoglu, Daron,. and Robert Shimer. 1998. "Efficient Unemployment Insur- ance." Working Paper No. 6686. National Bureau of Economic Research, Cambridge, Mass. Bour, Juan Luis. 1994. "Regulaciones laborales y funcionamiento del mercado de trabajo en america latina: el caso argentino." In Gustavo Marquez, ed., Regu- lacion del mercado de trabajo en America Latzina. Coalici6n Costarricense de Iniciativas de Desarrollo (CINDE). LABOR REFORM IN LATIN AMERICA DURING THE 1990S 31 Cortazar, Rene. 1997. "Chile: The Evolution and Reform of the Labor Market." In Sebastian Edwards and Nora Lustig, (eds.), Labor Markets in Latin Amer- ica. Combining Social Protection with Market Flexibility. Washington, D.C.: The Brookings Institution. de Ferranti, David, Guillermo Perry, Indermit Gill, and Luis Serven. 2000. "Se- curing our Future in a Global Economy." Latin American and Caribbean Studies. World Bank, Washington, D.C. Fallon, Peter R., and Robert Lucas. 1991. "The Impact of Changes in Job Secu- rity Regulations in India and Zimbabwe." World Bank Economic Review 5 (3): 395-413. Guasch, Luis. 2000. Argentina: Labor Market in the New Millennium. Washing- ton, D.C: World Bank. Haltiwanger, John, and Manisha Singh. 1999. "Cross-Country Evidence on Pub- lic Sector Downsizing." World Bank Economic Review 31 (1): 23-66. Heckman, James, and Carmen Pages. 2000. "The Cost of Job Security Regula- tion: Evidence from Latin American Labor Markets." Working Paper No. 7773. National Bureau of Economic Research, Cambridge, Mass. Hopenhayn, Hugo. 2001. "Labor Market Policies and Employment Duration: The Effects of Labor Market Reform in Argentina." Working Paper No. 407. Inter-American Development Bank Research Department, Washington, D.C. International Labour Organization (ILO). 1999. "Decent Work and Protection for All: Priority of the Americas." Reporr of the Director General. Geneva. Kugler, Adriana. 1999. "The Impact of Firing Costs on Turnover and Unem- ployment: Evidence from the Colombian Labour Market Reform." Interna- tional Tax and Public Finance 6 (3): 389-410. . 2000. "The Incidence of Job Security Regulations on Labor Market Flexibility and Compliance in Colombia: Evidence from the 1990 Reform." Research Network Working Paper No. R-393. Inter-American Development Bank Research Department, Washington, D.C. Levenson, Alec R., and William Maloney. 1998. "The Informal Sector, Firm Dy- namics, and Institutional Participation." Policy Research Working Paper No. 1988. World Bank, Washington, D.C. Lora, Eduardo, and Carmen Pages. 1996. "La legislaci6n laboral en el proceso de reformas estructurales de Arnerica Latina y el Caribe." Office of the Chief Economist. Inter-American Development Bank, Washington, D.C. Lora, Eduardo, and Gustavo Marquez. 1998. "The Employment Problem in Latin America: Perceptions and Stylized Facts." Working Paper No. 371. Inter-American Development Bank Research Department, Washington, D.C. Lora, Eduardo, and Marta Luz Henao. 1997. "Colombia: The Evolution and Re- form of the Labor Market, Labor Markets in Latin America." In Sebastian Ed- wards and Nora Lustig, (eds.), Combining Social Protection with Market Flexi- bility. Washington, D.C.: The Brookings Institution. 32 CRAFTING LABOR POLICY MacIsaac, Donna, and Martin Rama. 2000. "Mandatory Severance Pay in Peru: An Assessment of its Coverage Using Panel Data." Paper prepared for the Re- gional Study on Economic Insecurity. Office of the Chief Economist. World Bank, Washington, D.C. . 2001. "Mandatory Severance Payment: Its Coverage and Effects in Peru." Development Research Group. World Bank, Washington D.C. Marquez, Gustavo. 1998a. "Protecci6n al empleo y funcionamiento del mercado de trabajo: una aproximaci6n comparativa." Inter-American Development Bank, Washington, D.C. Processed. _____. 1998b. "El Desempleo en America Latina y el Caribe a Mediados de los Afios 90." Office of the Chief Economist Working Paper No. 377. Inter- American Development Bank, Washington, D.C. Mazza, Jacqueline. 1999. "Unemployment Insurance: Case Studies and Lessons for Latin America and the Caribbean." RE2/SO2, Technical Study. Inter- American Development Bank, Washington, D.C. Paes de Barros, Ricardo, Carlos Corseuil, and Monica Bahia. 1999. "Labor Mar- ket Regulations and the Duration of Employment in Brazil." Texto para Dis- cussao No. 676. Institute of Applied Economic Research, Rio de Janeiro. Pages, Carmen. 1999. "Apertura, Reforma y Mercado de Trabajo: La experiencia de un decada de cambios estructurales en el Peril." Office of the Chief Econ- omist, Working Paper No. 397. Inter-American Development Bank, Wash- ington, D.C. Pages, Carmen, and Claudio Montenegro. 1999. "Job Security and the Age- Composition of Employment. Evidence from Chile." Office of the Chief Economist, Working Paper No. 398. Inter-American Development Bank, Washington, D.C. Pessino, Carola. 1997. "Argentina: The Labor Market During the Economic Transition." In Sebastian Edwards and Nora Lustig, eds., Labor Markets in Latin America. Combining Social Protection with Market Flexibility. Washing- ton, D.C.: The Brookings Institution. Rama, Martin, and William Maloney. 2000. "Income Support Programs for the Unemployed in Latin America." World Bank, Washington, D.C. Draft. Ramos, Lauro A. 1999. "Impacto Distributivo dos Gastos Sociais no Mercado de Trabalho." Institute of Applied Economic Research, Ministry of Planning and Budget. Brazil. Saavedra, Jaime. 1999. "Reformas Laborales en un Contexto de Apertura Eco- n6mica." Unpublished manuscript. Grupo de Analisis para el Desarrollo, Lima. Saavedra, Jaime, and Maximo Torero. 2000. "Labor Market Reforms and Their Impact on Formal Labor Demand and Job Market Turnover: The Case of Peru." Research Network Working Paper No. R-394, Inter-American Devel- opment Bank Research Department, Washington, D.C. PART Public and Private J Sector Labor Markets in Brazil CHAPTERS 2 AND 3 diagnose the main imbalances in the fumnctioning of public and private sector labor markets in Brazil, and propose policy remedies. Fiscal adjustment is required to complete the stabilization begun by Brazil with the Real Plan launched in 1994. Importantly, this involves public employment reform to reduce payroll-related expenses. For public employment reform to be sustainable, and to minimize the probability of reversals, it must be based on the principle of labor market efficiency. The most important rule in this regard is that compensation for government workers should be in line with pay and employment conditions in the pri- vate sector. Using a nationally representative household survey, Chapter 2 examines public-private differentials in salaries and pensions. Compared with equally qualified workers in the private sector, salaries of public em- ployees are 30 to 50 percent higher for federal administration, judicial, and legislative workers; 20 to 35 percent higher for employees in federal and state enterprises, and roughly the same for state civil servants; and 5 to 15 percent lower for municipal administrations and education and health sector workers. Compared with their private sector counterparts, pension levels of civil servants are 25 to 50 percent higher, depending on their salary level, gender, and occupation. These findings imply that efforts to reduce government payroll ex- penses should focus on lowering salary levels at the federal level, and a mix of salary restraint and reduced employment at the state level. At the mu- nicipal level, however, reductions of employment should be combined 33 34 CRAFTING LABOR POLICY with selected pay increases. Reductions in public-private pension differ- ences would significantly reduce the lifetime earnings gap between the two systems. Moreover, they would eliminate the significant hidden transfers of pension liability from states and municipalities to the federal social se- curity system that occur when subnational governments reduce employ- ment through voluntary severance programs. Chapter 3 takes up the issue of labor reform in some detail. Brazil's ex- perience shows that the economic and political history of a country is a critical determinant of which labor laws influence wages and employment, and which are not binding. Long periods of high inflation, workforce il- literacy, and biases in the design and enforcement of labor legislation bred by the country's socioeconomic history are all important in determining the reach of labor laws. Defying conventional wisdom, these factors are shown to affect labor market outcomes even in the sector of employment regarded as unregulated. Following accepted practice in Brazil, the authors distinguish regulated from unregulated workers by determining whether or not their contracts have been ratified by the Ministry of Labor. Chap- ter 3 then examines the degree of adherence to labor laws in the formal and informal sectors, and finds evidence of effects of the legislation re- garding minimum wage and work-hour restrictions in both the formal and informal sectors of the Brazilian labor market. The incentives to stay informal are naturally higher for workers who are assured of protection under labor legislation regardless of the nature of their contracts, which alters their financial relationship only with the gov- ernment. The findings of Chapter 3 imply that informality in Brazil is mainly a fiscal and not a "legal" phenomenon. But the manner in which these laws have been enforced is also a critical determinant of informality in Brazil: poor record keeping has strengthened the incentives to stay in- formal that are already built into the design of the main social security programs, and ambiguities in the design of labor legislation combined with slanted enforcement by labor courts have led to workers effectively being accorded the same labor rights whether or not they have ratified contracts. Chapter 3 concludes that informality in Brazil will remain high as long as labor laws remain ambiguous and are enforced with a clear pro- labor bias, and social security programs lack tight benefit-contribution linkages and strong enforcement mechanisms. In 1999 and 2000, Brazil made some headway in reforming its social security system, principally by tightening enforcement of pension rules for government workers and strengthening the linkage between social security contributions and benefits for private sector workers, and considerably re- PUBLIC AND PRIVATE SECTOR LABOR MARKETS IN BRAZIL 35 ducing the generosity of these benefits. These changes may help stem the increase in informality, but at the same time they have increased existing public-private pension gaps. No significant reform of labor legislation has been undertaken during the last,decade: payroll taxes and mandatory non- wage benefits remain high, rules for dismissals remain contentious, and enforcement of labor laws has changed little, if at all. The most recent change is in fact an increase in employer contribution rates to the indi- vidual severance (Fundo de Garantia do Tempo de Servico-FGTS) ac- counts and the penalties levied on employers for "unjust dismissal," forced by a judicial decision. Brazil enters the new millennium with a daunting agenda for labor reform. 2 Improving Public Sector Efficiency in Brazil RIcARDo PAES DE BARROS, MIGUEL N. FOGUEL, INDERMIT S. GILL, AND ROSANE SILVA PINTO DE MENDON9A BRAZIL IS IN THE PROCESS OF consolidating the macro- economic stabilization begun when the Real Plan was launched in July 1994. Tight money and a loose fiscal stance have helped improve poverty rates in the short term, but involve a longer-term compromise in the form of reduced growth. The public sector deficit almost doubled to about 8 per- cent of GDP between August 1996 and 1998. The financing of the public sector deficit through inflows of volatile foreign capital has made Brazil vul- nerable to exogenous shocks, as witnessed by the financial crises of Octo- ber 1997 and August-September 1998. Since deeper tax reforms involve addressing sensitive fiscal federalism issues, and Brazil's tax collections are already high at about 30 percent of GDP, much of the adjustment must take place on the expenditures side. Given the already squeezed capital budgets at all levels of government, reducing the fiscal burden of public employment is viewed as central to sustaining the successful stabilization. Alleviating the fiscal burden may be achieved by means of reductions in either public employment and public wages. In general terms, employ- ment in consolidated government (excluding public enterprises) in Brazil is neither extraordinarily high nor low by international standards. The most up-to-date study of public sector employment in 90 countries (Rama 1997) finds that the predicted level of government employment for Brazil is about 9.6 percent for the 1990s. This is close to the actual figures from 37 38 CRAFTING LABOR POLICY the nationwide annual household surveys (Pesquisa NacionalporAmostra de Domici'lios-PNAD), which ranged from 9 percent to 9.4 percent during 1990-95. There is no clear trend in this ratio, indicating that government employment has not risen faster or slower than private employment. In- cluding employment in Brazil's public enterprises, the share of government goes up to about 12 percent, which is also not unusually high by interna- tional standards. At the most aggregate level, therefore, the problem of high payroll expenses in Brazil seems due not to excessive government employ- ment, but to excessively high pay, pensions, and perks. However, this does not rule out the possibility of public sector redundancies in particular re- gions (for example, in some northeastern states), at particular levels of gov- ernment (for example, at the municipal level) or for particular classes of workers (for example, for judicial and legislative workers). This chapter identifies sectors where public employment is relatively high. Alternatively, the Brazilian government may reduce public sector wages. To understand whether this policy measure would improve the ef- ficiency of the public sector, we analyze public-private sector differentials in compensation.' To detect inefficiency in the public sector labor market, a benchmark has to be specified against which the outcome in the public sector can be compared. The ideal benchmark would be employment and earnings outcomes, which would exist if the decisionmakers in the public sector considered efficiency to be their only objective. In practice, the pri- vate sector serves as this benchmark. Furthermore, small differentials be- tween public and private sector compensation are attractive from an eq- uity point of view, which suggests that a worker should not become better or worse off simply by changing sector. Salaries and pensions are the main components of compensation: both labor market efficiency and equity concerns imply that their public-private differences should be small. Reducing the wage bill for civil servants in Brazil is a difficult task. The 1988 Constitution disallows layoffs of public servants without "just cause" and considerably limits the government's ability to control personnel ex- penditures. Reductions in employment are constrained by the presence of tenure or near-tenure rights of most public sector workers that, until re- cently, were guaranteed under the Constitution. The payment of generous pensions paid to civil servants results in their becoming a financial burden for life (or even longer because of generous survivor benefit schemes), but reduction of pensions paid or increases in time of contribution are diffi- cult because these rights are unassailable under the Constitution. Further- more, the Brazilian government's efforts to reduce the wage bill for civil servants by reducing their real salary levels have become increasingly dif- IMPROVING PUBLIC SECTOR EFFICIENCY IN BRAZIL 39 ficult since 1996 with the decline in inflation to single-digit levels. Nom- inal reductions are still forbidden by law. In order to deal with the high burden of public employment, the strat- egy of the government in the fiscal readjustment process in Brazil has been two pronged: The first is constitutional reform to reduce pension expen- ditures and allow mass dismissals of tenured civil servants in case of fiscal necessity, through the Social Security reform bill, partially approved by Congress in November 1998, and the Administrative Reform Bill. The Administrative Reform Bill, aimed at revoking job tenure rights for pub- lic servants was passed in early 1998, but awaits supporting legislation by states and municipalities. The second is based on inducing tenured public employees to voluntarily leave government employment. As a con- sequence, a number of redundancy programs, or programas de demissdo voluntaria (PDVs), were carried out in both federal and state administra- tions and enterprises especially between 1995 and 1997 (see Carneiro and Gill forthcoming). The size of severance packages required to induce em- ployees to leave depends upon public-private differences in earnings, pen- sions, and other benefits, and not on the level of public earnings. Much of the discussion surrounding fiscal adjustment in Brazil, however, has fo- cused on the latter. In this chapter, we present an overview of this issue and focus explicitly on public-private differentials in earnings and pensions: * Based on nationwide household surveys, we compute public-private differences in monthly earnings, adjusting for worker characteristics such as education, age, tenure, sex, and race. * Using experience, earnings profiles, and rules for determining pen- sions, we compute public-private differentials in pensions. Furthermore, we estimate the severance packages that are sufficient to compensate public servants for possible losses in earnings and pensions and identify their fiscal effects. We find that the overall public-private wage differential adjusted for hours with respect to the private sector wage is large, and that there is sig- nificant heterogeneity among different segments of the public sector. How- ever, differences in the composition of the labor force, with respect to edu- cation, race, and gender, explain most of the overall differentials. Once we control for individual worker characteristics, the large positive public- private differentials become smaller, and the absolute value of negative wage differentials for certain types of public sector workers increases. This indicates that not all segments of the public sector have significant wage 40 CRAFTING LABOR POLICY advantages when compared to similar private sector employees. However, differentials in pension levels are large. Compared to their private sector counterparts, pension levels of civil servants are 25 to 50 percent higher, depending on salary level, gender, and occupation. Combined with the greater job security (another benefit) that public sector workers enjoy, it may not be unreasonable to argue that, in the aggregate, government work- ers are overcompensated relative to similarly qualified workers in the pri- vate sector. This is consistent with the finding discussed earlier that while the share of government employment is not out of line with international benchmarks, the expenditures on government salaries and pensions are clearly higher in Brazil than in countries at similar stages of development. Employment and Earnings in the Public and Private Sectors Employment, 1981-95 PNAD surveys indicate that total employment in Brazil was 62.5 million in September 1995. Public employment-other than in state-owned firms-was approximately 5.6 million (2.2 million in public administra- tion, more than half at the municipal level; 0.6 million in the judiciary, legislature, and military; and 2.9 million in the education and health sec- tors). The private employment sector, including employment in public enterprises, had 56.9 million workers (16.1 million in agriculture, 8.4 mil- lion in industry, 28.4 million in services, and 4 million in the construc- tion and other private sectors). In Table 2-1 public employment is defined as the sum of civil servants and the military, and is disaggregated into employment at the three levels of direct administration (federal, state, and municipal), education and health, judiciary and legislative, and the military sector. We will call this the sectoral share decomposition. Table 2-1 shows that the public em- ployment share rose from 7.9 percent to 9.4 percent out of total employ- ment from 1981 to 1990. This change is largely due to an increase in mu- nicipal employment and in the education and health sector. From 1990 on, the public employment share has stayed at about 9 percent, and the shares of all subsectors remained constant. However, employment in- creased in all sectors during 1992-95. For example, judicial and legislative staff increased from 0.24 million to 0.31 million, and employment in ed- ucation and health increased from 2.65 million to 2.88 million. The desegregation of employment into public employment and private employment is not exhaustive, since it does not distinguish employment IMPROVING PUBLIC SECTOR EFFICIENCY IN BRAZIL 41 TABLE 2-1 Sectoral Shares in Employment, Selected Years, 1981-95 (percent of total) Employment 1981 1985 1990 1992 1993 1995 Public employment 7.9 8.4 9.4 9.0 9.3 9.0 Direct administration 2.9 3.3 3.8 3.6 3.6 3.5 Federal 0.5 0.4 0.3 0.3 0.3 0.3 State 1.4 1.5 1.5 1.4 1.4 1.3 Municipal 1.0 1.4 2.0 1.9 1.9 1.9 Education and health 4.0 4.2 4.6 4.5 4.7 4.6 Judiciary and legislative 0.4 0.4 0.4 0.4 0.4 0.5 Military 0.6 0.5 0.6 0.5 0.4 0.4 Private employmenta 92.1 91.6 90.6 91.0 90.7 91.0 Total 100.0 100.0 100.0 100.0 100.0 100.0 Total (millions) n.a.b n.a.b n.a.b 58.8 59.9 62.5 Note: For all of Brazil. a. Private employment includes employment in public enterprises. b. Total employment figures before 1991 are not accurately estimated. Source: PNAD (various years). in public enterprises from private employment. To overcome this issue we present results below, which use the class of worker distinction. This as- signs workers in public enterprises and autonomous agencies (fundacoes) to the public sector. Consequently, the share of the public sector increases to 11.5 percent during 1995. The class of worker distinction decomposes the employment at each level of government into public servants (estatu- tarios) and nonpublic servants, which subsumes employees in public en- terprises and autonomous agencies. Estatutarios are civil servants who are hired officially through formal selection exams and have tenure under cur- rent law. Nonpublic servants are divided into employees with a signed work card (com carteira assinada o CLTistas) and workers without a signed work card (sem carteira assinada). Workers without and with a signed work card conform in Brazil approximately to the informal and formal private sector, respectively.2 As can be seen in Table 2-2, almost all government employees are either public servants or have a signed work card. Only 1.4 percent of total employment consists of public sector workers without a work card. To analyze employment patterns in the Federal District and the six most populous states (Bahia, Minas Gerais, Pernambuco, Rio de Janeiro, 42 CRAFTING LABOR POLICY TABLE 2-2 Share of Workers, by Class of Worker Distinction, 1995 (percent) Public Nonpublic servants Total Workers servants With card Without card share Public sector workers - - - 11.5 Federal 0.9 0.7 0.1 1.7 Stare 3.5 1.1 0.4 5.0 Municipal 1.9 1.6 0.9 4.4 Military 0.4 n.a. n.a. 0.4 Private sector workers - - - 88.5 Salaried n.a. 28.9 21.1 50.0 Self-employed n.a. 0.0 23.6 23.6 Employers n.a. 0.0 4.1 4.1 Unpaid n.a. 0.0 10.8 10.8 Total 6.7 32.3 61.0 100 - Not available. n.a. Not applicable. Note: For all of Brazil. Source: PNAD (1995). Rio Grande do Sul, and Sao Paulo), we calculate the relevant shares with respect to total employment in each state. We find that public employ- ment is overrepresented in the Federal District and in Rio de Janeiro, which are among the states with the highest average wage. The share of public employment amounts to 23 percent in the Federal District and to 11 percent in Rio de Janeiro, while it is approximately 8 percent in Bahia, Pernambuco, Minas Gerais, Rio Grande do Sul, and Sao Paulo. Addition- ally, we calculate the share of public employment in each state with respect to the entire Brazilian public sector. We find that employment in the pub- lic sector is only slightly overrepresented in the poorest areas: 49 percent of public employment and 46 percent of private employment are in states with the average wage below the overall average, which does not support the argument that public employment serves as a safety net for the poorer northeastern states. Using the class-of-workers distinction, we find that the share of govern- ment employees rises to 29 percent in the Federal District, to 14 percent in Rio de Janeiro, and to approximately 10 percent in the other states. In all six states and the Federal District the share of public servants (estatuarios) IMPROVING PUBLIC SECTOR EFFICIENCY IN BRAZIL 43 in total employment is considerably higher than the respective share of workers in public enterprises and autonomous agencies. The share of work- ers in public enterprises and autonomous agencies ranges from 3 percent in Pernanbuco to more than 6 percent in Bahia and the Federal District. Government employees with a signed work card constitute about 1 percent of the total population, except in Bahia and the Federal District, where this share is twice as high. Much of this "informality" of government employ- ment is at the municipal level. This is especially true for the rwo north- eastern states (Bahia and Pernambuco). Concerning the degree of formal- ity of private employment, we find considerable variation across states. The share of workers with a signed card in private employment is 35 to 40 per- cent for southern states (Rio Grande do Sul, Rio de Janeiro, and Sao Paulo) and the Federal District, but only 15 to 20 percent for the northeastern states. Minas Gerais, with a ratio of about 30 percent, falls in the middle. Earnings, 1981-95 Using PNAD surveys and the commonly used fndice Nacional de Prefos ao Consumidor (INPC) deflator, we calculate average monthly earnings of public employees between 1981 and 1995. As can be seen in Table 2-3, in 1995 average earnings were highest for judicial and legislative workers, who earned almost R$1,500 per month. Earnings were 25 percent lower for federal workers, 50 percent lower for state and military personnel, 63 percent lower for education and health workers, and 75 percent lower for TABLE 2-3 Average Monthly Earnings, Selected Years, 198c1-95 (constant September 1995 realsa) Public employment 1981 1985 1990 1992 1993 1995 Direct administration Federal 876 1,155 1,059 1,056 1,150 1,126 State 675 780 783 555 621 728 Municipal 412 386 398 330 316 361 Education and health 546 606 621 476 512 551 Judiciary and legislative 1,043 1,247 1,362 1,058 1,235 1,473 Military 669 844 687 637 622 718 Note: For all of Brazil. a. The deflator used is INPC (Brazil). Source: PNAD (various years). 44 CRAFTING LABOR POLICY municipal workers. In the private sector, earnings were highest in produc- tive services, amounting to approximately R$950 per month. Compared to this, earnings in distributive services and manufacturing were about 45 percent lower, and in personal services and construction about 60 percent lower. Agricultural workers earned about 75 percent less. Changes in average real monthly earnings by sector from 1981 to 1995 show that real monthly earnings increased about 29 percent for federal employees and 8 percent for state employees, and fell by 12 percent for municipal employees. Earnings for judicial and legislative employees rose by more than 40 percent during these years, by about 7 percent for the military, and stayed roughly constant for education and health workers in the public sector. This indicates that judicial and legislative employees en- joyed the largest increase, while municipal employees had to cope with the largest decline in real earnings. In the private sector, earnings in agricul- ture fell by about 10 percent and in industry by 6 percent, but rose in dis- tributive and productive services by 6 percent, and in personal services and construction by about 16 percent. Considerable differences in average earnings also existed across the dif- ferent states. In 1995, average earnings in almost all sectors were highest in the Federal District and Sao Paulo, and lowest in Pernambuco and Bahia. Earnings in Minas Gerais were higher than in the two poorer northeastern states, but lower than in Rio Grande do Sul and Rio de Janeiro. Relative to earnings in the private manufacturing sector, state em- ployees received roughly 67 to 80 percent more in Bahia and Pernambuco, and 40 to 45 percent more in the other states. In contrast, relative earn- ings of municipal workers were highest in Rio Grande do Sul and Rio de Janeiro, somewhat lower in Minas Gerais and Bahia, and lowest in Per- nambuco and Sao Paulo. In absolute terms, municipal workers earned twice as much in Sao Paulo as in Pernambuco. In the private sector, earn- ings by sector relative to private manufacturing are strikingly uniform across regions. Only in Rio Grande do Sul are the earnings of agricultural workers comparatively high. The Wage Bill, 1992-95 Using employment and average earnings for each sector, we compute the sectoral wage bill in 1995. Figure 2-1 graphs the relative importance of each of the public employment sectors. Education and health workers, who constitute about 50 percent of all government workers, absorb about 45 percent of the wage bill. Municipal employees constitute about 20 per- IMPROVING PUBLIC SECTOR EFFICIENCY IN BRAZIL 45 FIGURE 2-1 Share of Employment and Wage Bill of Different Public Employment Sectors Percent 60 50 - Share of employment 4Ib Share of wage bill 40 30 20 10 0 Education State Judicial Municipal Military Federal and and health legislative Public employmenr sector Source: PNAD (1995). cent of total public employment, but absorb only 12.5 percent of the total wage bill. State employees, who constitute about 15 percent of employ- ment, absorb about 17 percent of the total wage bill. Federal and military workers are relatively small fractions of both total employment and the wage bill. The most dramatic difference between these two shares is for ju- dicial and legislative workers, who constitute about 5 percent of total em- ployment, but absorb almost 14 percent of the total wage bill. Thus, while employment of judicial and legislative workers may be relatively small, they are potentially critical for fiscal reasons. To examine the trends in wage-related expenditures in the public sec- tor, we computed the growth of the wage bill in the pubLic and private sec- tors. Because of inaccurate weighting before 1991, wage bill numbers are reliable only for the three PNAD surveys for 1992, 1993, and 1995. Fig- ure 2-2 graphs the annual growth rate for employment, average earnings, and the wage bill during 1992-95. In the public sector, the 20 percent an- nual growth of the wage bill for judicial and legislative workers dwarfs that 46 CRAFTING LABOR POLICY FIGURE 2-2 Annual Growth from 1992 to 1995 in Employment, Average Earnings, and Wage Bill Percent 25 - * Employment 20- j Earnings .~ I 1,,,,,221as Source: PNAD surveys. of other workers. This growth in personnel expenditures was due to both an increase in their number (about 7.5 percent annually) and increased av- erage earnings (about 12 percent annually). The growth of the state wage bill exceeded 9 percent, despite a small decrease in state employment. The wage bill for education and health workers increased by 7.5 percent an- niually, due to increases in both average real earnings and employment. The federal wage bill also increased by 4.5 percent because of both new hires and higher wages. Military employment stayed roughly constant, but earnings increased moderately. In the private sector average earnings grew, especially in personal and distributive services and construction. In contrast, industry and agricul- ture registered modest earnings growth, and employment actually fell in agriculture. In general, only employment growth in services exceeded the national average of about 2 percent during this period. In the public sec- tor, on the other hand, employment of federal, municipal, judicial and IMPROVING PUBUC SECTOR EFFICIENCY IN BRAZIL 47 legislative, and education and health workers rose faster than this average rate. Only state and military employment grew more slowly than the na- tional average. The Public-Private Wage Gap The Uncontrolled Wage Gap To provide evidence on the uncontrolled public-private wage gap, we start by estimating the overall wage gap between the private and public sector for the entire Brazilian labor market. We then calculate the wage gap be- tween different segments of the public sector and the overall private sec- tor to take into account the heterogeneity within the public sector. Finally, we analyze the impact of differences in spatial composition of the public and private labor forces on the wage gap. DATA. The empirical analysis conducted here is based on the PNAD survey for September 1995. The PNAD is the national household survey that is undertaken annually by the Brazilian National Statistical Institute. Each dataset has more than 360,000 observations. We use two measures of labor income. The first measure is the total monthly income normally received by a worker in his or her current main job. The second measure seeks to adjust for the number of hours usually worked. It is defined as the total monthly income normally received in the main job, divided by the number of hours usually worked per week in that job and multiplied by a standard workweek of 40 hours. Since the average number of hours worked by public employees tends to be significantly smaller than the cor- responding average for workers in the private sector, it does make a dif- ference for level of the wage gap between these sectors whether or not a standardization for hours of work is conducted. To ensure that the stan- dardized wage is a good approximation of what a worker would receive if he or she works 40 hours a week, workers working less than 20 hours per week were excluded from the analysis. We also exclude workers with zero labor income. THE OVERALL WAGE GAP. Table 2-4 presents estimates of the overall wage gap between public and private sectors in Brazil.3 The large positive estimates reveal that average earnings tend to be much higher in the pub- lic sector than in the private sector. The relative wage difference between the public and the private sectors amounts to 49 percent of the average 48 CRAFTING LABOR POLICY TABLE 2-4 Public Sector Wage Premium (percent) Public- Total private Public-Fornalprivate Measure Adjusteda Not adjusted Adjusted Not adjusted Baseline: Private sector wage 89 49 67 42 Average log-wage 70 45 38 18 Note: Unadjusted for sector, region, and worker attributes. a. Adjusted refers to earnings adjusted for differences in hours worked. Source: PNAD (1995). wage in the private sector. It increases to 89 percent when we adjust for hours, which indicates that workers in the public sector work fewer hours per week than private sector workers. Furthermore, Table 2-4 reports estimates of the gap between the wage of employees in the public sector and employees in the private formal sec- tor, which may be the part of the private sector where those leaving gov- ernment employment are more likely to seek employment.4 Table 2-4 re- veals that this wage gap is slightly smaller than the overall gap between the public and private sectors. More specifically, the wage gap decreases by 7 percent when we do not adjust for hours. The adjusted wage gap remains large, at 67 percent. Estimates of the public sector wage premium measured by the average of the log-wage are also provided in Table 2-4. Log-wage differences have the advantage that they are independent of the baseline wage, and hence eliminate the need to keep track of it. Moreover, they provide estimates for the wage gap that are between the two natural (using either private sector wage or public sector wage as baseline) measures of relative differences in wages. However, a drawback is that they are not as straightforward to in- terpret. We include the information on average log-wages in Table 2-4 (and Table 2-5) for reasons of comparison; we provide evidence on the public-private wage differential based on regression analysis, which is stated in terms of average log-wages. At this stage of the analysis it is difficult to meaningfully interpret these wage gaps. Three factors complicate such a comparison. First, as we have seen above, there are large differences in average earnings within the public (for example, between municipal and legislative workers) and private sec- tors (for example, between agricultural and service sector workers); second, IMPROVING PUBLIC SECTOR EFFICIENCY IN BRAZIL 49 TABLE 2-5 Public Sector Wage Premium (percent) Measure Baseline: Private wage By sector of activity Federal administration 162.5 State administrations 69.9 Municipal administrations -15.8 Judicial and legislative 231.0 Military 69.9 Education and health 18.0 By c/ass of worker Federal public servants 207.0 Federal other, with signed card 166.4 Federal other, no signed card 68.2 State public servants 60.5 State other, with signed card 95.3 State other, no signed card -1.1 Municipal public servants -2.8 Municipal other, with signed card -32.8 Municipal other, no signed card -41.7 Military 71.7 Note: Unadjusted for region and worker attributes. Source: PNAD (1995). there are regional differences in average earnings (for example, between state employees in Pernambuco and Sao Paulo); and third, there are large differences across workers by individual attributes (for example, workers in the public sector tend to be better educated and older). These three factors will be addressed in the following sections. Henceforth, only wage gaps ad- justed for hours worked will be discussed and reported. Interested readers can find more detailed results in Paes de Barros and others (1997). HETEROGENEITY IN THE PUBLIC SECTOR. To analyze the degree of het- erogeneity within the public sector, we use the same two decompositions for the public sector that we already proposed in the analysis of the em- ployment shares: the sectoral share decomposition, which divides public employment into the three levels of public administration (federal, state, and municipal), public health and education, legislative and judiciary ac- tivities, and the military; and the class of worker distinction, which pro- 50 CRAFTING LABOR POLICY vides information on both the type of labor contract and the level of gov- ernment. Table 2-5 presents estimates of respective wage gaps, which are calculated between the respective segment of the public sector and the overall private sector. The results indicate considerable heterogeneity within the public sector. Average wages are particularly high in federal ad- ministration, and in the legislative and judiciary sectors. The relative wage differences with respect to the overall average private wage amount to 162.5 percent and 231 percent, respectively. At the other extreme, work- ers in health and education and employees at the municipal level have earnings that are close to the average for the private sector, or even below it. The average wage in municipal administration is nearly 16 percent lower than the average in the private sector. In between are state admin- istration and military personnel; wages for these two groups are close to 70 percent above the average in the private sector. We obtain similar results when using the class of worker distinction of the public sector. Wages are particularly high among federal public ser- vants, and federal public employees with signed work cards, which largely refers to employees in public enterprises. In both cases the wage gap is greater than 150 percent of the average wage in the private sector. For mil- itary personnel, state public servants, and employees of state enterprises, the wage gap is about 75 to 100 percent. The average wage of municipal public servants and employees of municipal enterprises is below the aver- age for the private sector, with the wages of employees in municipal en- terprises being particularly low. Overall, average wages in all segments of the public sector tend to be above those paid in the private sector. The only exceptions are the wages paid by municipal administrations and en- terprises. Wages among different segments of the public sector reveal a large degree of heterogeneity in wages within the public sector, with the average wage in federal administration being more than 200 percent higher than the corresponding average paid in municipal administrations. REGIONAL DIFFERENCES IN THE PUBLIC-PRIVATE WAGE GAP. After having analyzed public-private differentials for different segments of the public sector, we now consider the role of spatial differences in the distri- bution of public and private employment. The fact that public employ- ment is overrepresented in the Federal District and in Rio de Janeiro, which are among the states with the highest average wage, implies that the wage gap would increase if public employment had the same spatial dis- tribution as private employment. In terms of the gap in average log-wage, this means an increase from 44 percent to 46 percent; that is, public em- IMPROVING PUBLIC SECTOR EFFICIENCY IN BRAZIL 51 ployment has a "locational disadvantage" of only 2 percent. Moreover, we find that the public-private wage gap tends to be considerably higher for metropolitan areas than for the corresponding states. This is especially true in the northeastern states of Pernambuco and Bahia, and in the State of Minas Gerais. Sao Paulo is the only state where the wage gap for the met- ropolitan area is smaller than for the corresponding gap for the entire state. The average gap between the public and private sectors is much higher in Brasilia than in any state. The average wage gap tends to be high in the northeast and lower in the more developed states in the South and Southeast of Brazil, particularly in Sao Paulo, Parana, and Santa Catarina. But it is also low in some poor states like Ceara and Alagoas (see Paes de Barros and others 1997). The Controlled Wage Gap The estimates we discuss above represent an important starting point for the wage analysis. However, they have serious limitations as a measure of the actual public-private wage differential, since they do not necessarily represent the actual difference in payment between equally productive workers in the public and private sectors. In fact, estimates of wage gaps, which do not control for worker characteristics, capture both differences in payment between equally productive workers in the two sectors and dif- ferences in the qualifications and characteristics of the labor force em- ployed in the two sectors. Wages may be higher in the public sector'sim- ply because the labor force in the sector is older and better educated. Or they may be lower due to the possible concentration of public employ- ment in the poorer Northeast and because women tend to be overrepre- sented in the public sector. Consequently, we are interested in estimating the wage gap between workers with identical observed characteristics such as gender, race, schooling, and age. Below we provide a thorough expla- nation of the methodology used to compare the wages of observably iden- tical workers in the public and private sectors. METHODOLOGY. The methodology used to compare the wages of ob- servably identical workers in the public and private sectors consists of four steps. In the first, we regress the log-wages of workers on their characteris- tics and on an indicator of whether they are in the public or private sector. This means that we assume that the impact of all personal characteristics on the level of wages is the same in the public and private sectors-that is, that the difference in the wage level between the public and private sectors 52 CRAFTING LABOR POLICY is the same for all types of worker. We refer to this specification as the basic model. Of course, we may think that, for example, the level of education affects wages to a different extent in the public sector than in the private sector. To address this issue, we also run regressions of log-wages on per- sonal characteristics for the public and private sector separately and, there- fore, allow personal characteristics to affect the wage level differently in the two sectors. We call this specification the general model. - In the second step, based on the estimates of the regression, we compute what would be the average wage of public employees if they were in the pri- vate sector; that is, we calculate the average log-wage that would prevail in the private sector if the labor force had the same average characteristics as the public sector workers. We refer to this type of wage gap as the con- trolled wage gap. Under the assumption of the basic model, which is that the wage gap between the public and private sector is the same for all types of workers, the log-wage gap between workers with identical observed char- acteristics is completely captured by the effect of the public-private sector indicator on wages. This is not true for the general model because this takes into account not only the effect of the public-private sector indicator on wages, but also captures the differences in the effect of individual charac- teristics on wages. Our estimation results reveal that no major difference exists between the two models. Consequently, we present only the results which are derived from the basic specification. Third, we decompose the overall wage gap into the wage gap, which is due to differences in the composition of the labor force, and the con- trolled wage gap. To analyze differences in the composition of the labor force, we use a basic set of observed characteristics, which includes gender, race, schooling, and age. We also work with a version of this set that in- cludes tenure at the current jdb. On average, workers in the public sector tend to have been in their current job for a longer period of time.5 Since to a considerable extent this is not a result of any differential merit be- tween employees in the public and private sectors, but one of the major advantages of jobs in the public sector, it is unclear whether we should control for this characteristic. Fourth, we estimate the wage gap between each segment of the public sector and the overall private sector controlling for worker-specific attri- butes, since the wage gap between the public and private sectors is likely to differ with sector of activity and class of worker. We perform the regres- sion analysis for the six major Brazilian metropolitan areas and Brasilia. The restriction of the analysis to these seven well-defined local labor mar- kets is useful to isolate the wage gap between the public and private sec- IMPROVING PUBLIC SECTOR EFFICIENCY IN BRAZIL 53 tors from possible spatial differences among local markets. However, it also has the disadvantage that the nature of public employment in large metropolitan areas is likely to be different from public employment in smaller urban areas. The contribution of federal and state employment is likely to be greater. Moreover, municipal jobs in metropolitan areas are more likely to be better paid. As a consequence, the wage gap estimated for these areas is likely to overestimate the gap for the entire urban labor market in the respective state. Role of Differences in Worker Characteristics The fundamental question we address here is: What fraction of the wage gap between the two sectors is due simply to differences in the character- istics of the labor force employed in the two sectors? Table 2-6 reveals that differences in the composition of the labor force are a major explanatory factor of the log-wage gap between the public and TABLE 2-6 Estimates of Log Wage Gap Due to Differences in Composition and Controlled Wage Gap for Metropolitan Areas Wage gap due to differences in composition Controlled wage gap Without With Without With tenure tenure tenure tenure Federal District 0.69 0.76 0.49 0.42 Recife (Pernambuco) 0.59 0.65 0.27 0.20 Salvador (Bahia) 0.60 0.69 0.19 0.10 Belo Horizonte (Minas Gerais) 0.63 0.71 0.14 0.07 Rio de Janeiro (Rio deJaneiro) 0.52 0.61 0.11 0.02 Sao Paulo (Sao Paulo) 0.47 0.53 -0.14 -0.20 Porto Alegre (Rio Grande do Sul) 0.56 0.65 0.13 0.04 Note: Results presented for the basic model. Wages are standardized by differences in work hours. Source: PNAD (1995). 54 CRAFTING LABOR POLICY private sector. As can be seen in Table 2-6, the specification without tenure reveals that the gap in average log-wages due to differences in the compo- sition of the labor force ranges from 0.47 to 0.69. Since public sector workers have longer tenure than private sector workers, and because wages increase with tenure, the public sector premium increases when we include tenure in our regression. Once we control for tenure the differences in the composition of the labor force are responsible for the average log-wage in the public sector being from 0.53 to 0.76 higher than in the private sec- tor. Comparison of the wage gap due to composition of the labor force and the controlled wage gap reveals that, apart from the Federal District, most of the overall wage gap between the public and private sectors reflects dif- ferences in the composition of their labor forces. Therefore, we conclude that measures for the wage gap that do not control for differences in the characteristics of the labor force are almost certainly very misleading indi- cators of the actual wage advantage of workers in the public sector. The wage gap, which is due to differences in the composition of the labor force, can be further decomposed in order to analyze rhe contribu- tion of each observed characteristic. In the specification without tenure we find that about 80 percent of the difference in log wages, which is due to differences in the composition, is explained by the higher educational level of the labor force in the public sector. Once we control for tenure this ef- fect decreases to about 70 percent. The remaining 30 percent is explained by the fact that public employees tend to be older and have longer tenure, with each of these two factors being responsible for 10 to 20 percent of the differential. The impact of gender and race is relatively small. The con- tribution of differences in race composition between the private and pub- lic sector ranges from 1 to 5 percent. The impact of gender differences ranges from 0 to -10 percent. In sum, differences in the composition of the labor force are a major explanation for the wage gap between the public and private sector. As a consequence the controlled wage gap is much smaller than the overall gap. Still, it remnains considerable for most states, even when we control for tenure: * In the Federal District the controlled wage gap amounts to a still very large public-private wage gap of 42 percent among workers with identical observed characteristics. * In Recife and Salvador workers in the public sector receive salaries that are 20 percent and 10 percent higher, respectively. * The public-private wage gap is 7 percent in Belo Horizonte, 4 per- cent in Porto Alegre, and 2 percent in Rio de Janeiro. IMPROVING PUBLIC SECTOR EFFICIENCY IN BRAZIL 55 Only in Sao Paulo is the pattern reversed: among equally qualified workers, workers in the public sector receive salaries that are 20 per- cent lower than private sector salaries. Note, however, that the nature of public employment in large metro- politan areas is likely to be different from public employment in smaller urban areas, and that as a consequence, the wage gap estimated for these areas is likely to overestimate the gap for the entire urban labor market in the respective state. This, however, is not true for Sao Paulo, which is the only region where the overall public-private sector wage gap is lower in the metropolitan area compared to the state. Hence, we are likely to underes- timate the wage gap for the entire urban labor market of Sao Paulo State. Controlled Wage Gaps by Sector and Class of Worker In this section we estimate the wage gap between each segment of the pub- lic sector and overall private sector controlling for differences in worker- specific attributes. This is perhaps the best estimate of the "pure" premium enjoyed by public sector employees over what they would have earned had they held private sector jobs. Table 2-7 presents estimates for the controlled wage gap between each segment of the public sector and the private sector for the country as a whole. It reveals large differences among segments of the public sector. The wage advantage of government workers, measured by the controlled log-wage gap, is larger at the federal level for both public admin- istration and enterprises. For these groups the average salaries are 23 to 46 percent higher than for comparable workers in the private sector. At the state level, the wage advantage is close to zero for public servants; for employees in public enterprises the wage advantage is between 10 and 18 percent. At the municipal level, public servants and employees in public enterprises get lower wages than workers with similar observed characteristics. Overall, the evidence corroborates the existence of a significant wage advantage in many segments of the public sector relative to similar private sector employees. Assignment to the public sector is not random. Individuals are assumed to make rational choices about working in the private or public sector. This choice is usually based on productivity-related characteristics that also affect earnings, such as motivation, taste for public service, or prefer- ence for nonrisky employment. Furthermore, access to the public sector in Brazil is determined either by personal connections or formal selection exams, which we also cannot observe in the data. Where sorting appears on the basis of these unobserved factors, estimates are biased, which means that they give a false picture of the relative earnings position of public sec- 56 CRAFTING LABOR POLICY TABLE 2-7 Measures of the Public Sector Earnings Premium by Sector and Class, September 1995 (percent) Measure Monthlya Hourrlya By sector of activity Federal administration 23.1 28.9 State administrations -7.8 -3.8 Municipal administrations -31.9 -22.4 Judicial and legislative 43.9 55.9 Military 2.3 5.7 Education and health -31.1 -15.6 By class of worker Federal public servants 40.8 46.3 Federal other, with signed card 27.0 36.3 State public servants -15.4 -5.5 State other, with signed card 10.3 18.0 Municipal public servants -30.5 -17.9 Municipal other, with signed card -31.5 -20.9 All workers without signed card -42.6 -19.7 Military 3.6 7.2 Note: Adjusted for worker characteristics: age, sex, race, tenure, and education. a. Monthly earnings premia are not adjusted for differences in hours worked. Hourly earnings comparisons are adjusted for differences in hours worked. Source: PNAD (1995). tor workers. The self-selection bias may be addressed by adding a third equation to the general model that explains the individual's decision to work either in the public or private sector. Unfortunately, the information available in our data does not allow us to arrive at a well-specified selec- tion equation. However, most of the studies on public-private wage dif- ferentials find that correcting for self-selection lowers the estimated size of the wage premium received by public sector workers. Public-Private Differences in Pensions Comparisons of monthly earnings of public and private workers reveal only part of the differences in the compensation between the two sectors. "Nonwage" compensation is often higher in the public sector due to more IMPROVING PUBLIC SECTOR EFFICIENCY IN BRAZIL 57 generous pension plans and other employee benefits. Most notably, the Brazilian pension system for public employees magnifies the differences between private and public sector lifetime income. As we will show, aver- age pension relative to average wage is much higher for civil servants com- pared to similarly paid workers in the private sector, although contribu- tions of civil servants are significandy lower. Private sector workers and workers in state-owned enterprises contribute to the National Security System (Regime Geral da Previdencia Social- RGPS), which is administered by the federal government's Instituto Na- cional de Seguridade Social (INSS). It is financed by payroll taxes on work- ers and employers.6 Employer's payroll taxes for social security amount to 22 percent of the wage bill. The self-employed contribute roughly 20 per- cent of earnings. Employee contributions are based on a progressive scale and, depending on the level of wage, they range from 8 percent to 11 per- cent. The INSS administers a number of benefit programs for participants including the Old Age and the Length of Service Programs. Under the Old Age Program the age of retirement is 65 for men and 60 for women who work in the urban sector. Those who retired before 1991 could apply for the Old Age Program after five years of contribution to the INSS. Recent legislation has increased this minimum period of contribution such that it increases by six months per year, reaching 15 years in 2011. In 1998 the minimum period of contribution to be eligible was 8.5 years. Under the Length of Service Program pensioners receive benefits irrespective of age after meeting a certain number of years of service. (This was changed to years of contributions under the reforms in 1998-99.) The minimum pe- riod of service for having access to the benefits is 30 years for men and 25 years for women. For both programs the minimum benefit in general has to be greater than one minimum wage. The maximum benefit cannot ex- ceed an amount that is usually close to 10 times the minimum wage (for a more detailed discussion see World Bank 2001). Workers in Brazil's public sector participate in mandatory pension plans, which are established under the so-called Regime Juridico Unico (RJU). Different institutions exist to run the pension plans for federal, state, and municipal public workers. In the mid-1990s pension contribu- tions were imposed for federal civil servants.7 The RJU is financed from payroll taxes on workers, and to a large part from transfers from govern- ment treasuries. Until the recent introduction of employee contributions, eligibility was based on years of employment, where employment also in- cluded time of service in the private sector and leaves of absence for study and other sabbaticals. To receive the unreduced pension levels, men and 58 CRAFTING LABOR POLICY women must be employed 35 and 30 years, respectively. While the RJUs have only 15 percent of Brazil's 20 million social security beneficiaries, they receive nearly 50 percent of all pension benefits paid in the country, and account for 75 percent of the consolidated pension deficits. In con- trast to the private sector, public employees face no upper limit on pen- sion amounts. Pensions of public employees are still based only on the salary of the last month, while in the INSS they were calculated on the basis of the monthly salary levels in the last three years of service, a period which is gradually being extended to cover almost all of the working life. This leads to higher pensions for public sector workers, because earnings usually increase with time of service, especially for relatively high-wage (skilled) workers. We take into account these basic features of the Brazilian pension sys- tems that existed until 1999 when estimating differences in pensions for public servants and those receiving social security benefits from the two main federal social security programs-the Length of Service and Old Age Programs of the INSS-and use the following techniques: * First, using PNAD survey data from September 1995 for the seven metropolitan areas (Belo Horizonte, Federal District, Porto Alegre, Recife, Rio de Janeiro, Salvador, and Sao Paulo), we estimate the pro- file of monthly earnings for males and females, which are divided further into four worker categories (public servants, private with signed work card, private without signed work card, and private self- employed), and three education groups (0 to 8 years, 9 to 11 years, and 12 or more years of schooling). As stated above, employees in Brazil without and with a signed work card correspond, approxi- mately, to workers in the informal and formal sectors. • Second, using these estimated earnings profiles, and assuming that all males retire after 35 years of service and all women after 30 years of service, we estimate the average pension level for each group. For public servants we use the 36th year's salary as the estimated monthly pension. Consistent with the fact that pension levels in the INSS are calculated on the basis of the monthly salary levels in the last three years of service, we use the 34th year's salary as the estimated pension for salaried workers with a signed work card and the self-employed. For workers without a signed work card, we assume that they receive minimum pensions which equal the minimum wage. Hence, their pension levels are assumed to be equal to one minimum wage in 1995, or R$100. IMPROVING PUBLIC SECTOR EFFICIENCY IN BRAZIL 59 * Third, we account for the fact that INSS pensions are capped at 10 minimum wages. This affects men with a signed work card who have more than 12 years of education; consequently, we restrict the pen- sions to be equal to R$1,000. * Fourth, using weights obtained from nationwide PNAD data for 1995 on the shares of employees with and without a signed work card, and self-employed workers, we compute the average expected pensions for private sector men and women. We do not take into ac- count pensions from the supplementary social security system (that is, closed or company funds).8 * Finally, these numbers are compared with expected monthly pen- sions for public servants with the same education level. Table 2-8 reports the results of these estimations. As can be seen, public- private pension differences are greater for men than for women, and great- est for the most educated male workers. This is largely due to the pension ceiling of 10 minimum salaries. We find that the average male private sec- tor worker with 0 to 8, 9 to 1 1, and 12 or more years of schooling receives, TABLE 2-8 Estimated Monthly Pensions (September 1995 Reals) Private With Without Self- Gender/education Public card card employed Total a Males 0-8 years schooling 447 430 100 407 335 9-l1 years schooling 772 758 100 686 558 12+ years schooling 1,747 1,793b n.a. n.a. 1,000 Females 0-8 years schooling 264 226 100 213 190 9-11 years schooling 448 447 100 362 326 12+ years schooling 927 995 100 832 655 n.a. signifies that the sample was too small to permit reliable estimation of experience- earnings profiles. Note: Men are assumed to retire after 35 years of service, and women after 30 years. a. Using weights derived from nationwide PNAD employment distribution data. b. Capped at 10 times the minimum monthly salary level, assumed to be R$100 in September 1995. Source: PNAD (various years) and IPEA-World Bank calculations. 6o CRAFTING LABOR POLICY respectively, 75, 73, and 57 percent of the monthly pension level of his estatutario counterpart. The corresponding numbers for women are 72, 73, and 71 percent. These large differentials in pension levels combined with the fact that pension contribution rates are 0 to 12 percent for public servants, but range between 8 and 11 percent for CLTistas, have important implications for the structure of optimal severance packages. As mentioned in the intro- duction, Brazil's 1988 Constitution disallows layoffs of public servants without "just cause," and considerably limits the government's ability to reduce employment in the public sector. To address this issue, between 1995 and 1997 the Brazilian government undertook a package of Consti- tutional reforms and introduced a number of voluntary severance pro- grams (PDVs). Given that the federal and state governments are cash strapped and can pay only relatively low benefits to PDV takers, these large differentials in pension benefits decrease the likelihood of success of current voluntary severance programs. Actuarial calculations reveal that if raising employee contributions were the only way to bring the state sys- tem into balance, even contribution rates as high as 75 percent of salaries would be insufficient to cover current benefits in some states. This indi- cates that in order to solve the financial problem, the government will have to increase pension contribution rates and simultaneously reduce the pension levels for civil servants. Voluntary Severance Programs The PDVs were implemented both at the federal and local government levels, in public administration as well as state-owned enterprises, to en- courage voluntary job separations of tenured estatutarios. PDVs imply savings in terms of a reduction in not only salary bills, but also reduce pen- sion bills. As shown above, pensions paid to public servants are substan- tially higher than those paid by the INSS. Consequently, government sav- ings in terms of the pension bill are higher when an estatutario takes a state-sponsored voluntary severance package than when a CLTista leaves public employment. We use a simple methodology that involves the comparison of experi- ence-earnings profiles of public and private employees to estimate the sav- ing in salaries and pensions for groups of workers according to their sex, ed- ucation, and tenure. Table 2-9 reports the results of this exercise for men and women with, respectively, 0 to 8, 9 to 11, and 12 or more years of educa- tion. In the case of men, it is assumed that the PDV taker has 17.5 years of IMPROVING PUBLIC SECTOR EFFICIENCY IN BRAZIL 6i TABLE 2-9 Effects of Reducing Public Employment on Average Earnings, Transfer of Obligations, and DPVa of Savings (thousand reals) Consolidated Transfer of government Education State government saving obligations saving level Salaries Pensions Total to INSS b Pensions Total Males, with 17.5years ofservice' 0-8 years 51.6 23.4 75.0 17.4 6.0 57.6 9-11 years 89.7 37.9 127.5 29.1 8.8 98.4 12+ years 215.5 91.6 307.0 41.9 49.7 265.1 Females, with 15 years ofservice' 0-8 years 27.1 21.2 48.3 14.5 6.7 33.8 9-11 years 51.4 36.0 87.4 25.1 10.9 62.3 12+ years 97.5 74.1 171.5 54.0 20.1 117.5 Note: These estimates are based on estimated earnings profiles of private and public sec- tor workers in a nationwide sample drawn from the September 1995 PNAD survey. a. A 6 percent discount rate was used to calculate present values. b. These transfers do not take into account the expected contributions of these workers (and their employers in the case of salaried employees) to the INSS system. c. Men are assumed to retire from public employment after 35 years of service and receive pensions for 20 years; women are assumed to retire after 30 years of service and receive pensions for 25 years. Source: PNAD (1995). service, and in the case of women, 15 years. These numbers are similar to actual tenure levels of participants in these programs (Carneiro and Gill forthcoming). Table 2-9 provides evidence on the savings to the state government in salaries and pensions due to a voluntary quit of an estatutario. Savings in salaries are lower for men compared to women and increase with educa- tion. The same is true for savings in pensions. For estatutarios who leave under the voluntary severance program but do not qualify for early retire- ment, there is an uncompensated transfer of pension obligations from the state administration to the national INSS scheme. This is because these employees carry over their years of service into the private system and, after a relatively short period of contributions, become eligible for private pen- sions. Thus, a male PDV participant with 17.5 years of service in the pub- 62 CRAFTING LABOR POLICY lic sector would be eligible for reduced INSS pensions after only 12.5 more years of work in the private sector, and for frill pensions after only 17.5 years. According to our estimates, the present value of these obligations for men (and women) leaving public employment mid-career are R$17,400 (R$14,500) for low-wage workers, R$29,100 (R$25,100). for medium- wage workers, and R$41,900 (R$54,000) for high-wage workers. Thus, for example, a state.PDV for 10,000 estatutarios drawn from all education- salary categories could imply a transfer of obligations of more than R$300 million in current terms from the state to the federal pension system. Considering the transfer of obligations to the INSS, savings in pensions decrease to 6,000, 8,800, and 49,700 reals for men; and 6,700, 10,900, and 20,100 reals for women with 0 to 8, 9 to 11 and 12 or more years of education. This implies that ignoring savings due to a reduction in the pensions bill leads to an underestimation of consolidated government savings of about 12 percent, 10 percent, and 23 percent, respectively, for men with 9 to 11 and 12 or more years of education. The corresponding numbers for women are 25 percent, 21 percent, and 21 percent, respec- tively. Note that pension-related savings are larger for highest-wage males because private sector pensions are capped at 10 times the minimum monthly salary, and the cap is binding only for this group. Other than this group, savings are greater for women because the expected length of re- tirement is greater for them. Savings due to pensions would be even higher if we account for the dif- ference between the public pension system and the INSS. While private sector workers contribute up to 20 percent of their salaries to pensions, civil- servants did not contribute anything until the 1980s (but now con- tribute up to 12 percent of their earnings). Consequently, the transfer of pension obligations to the INSS would to some extent be offset by payroll contributions. Under the current rules, however, the INSS system never recovers the "lost contribution" of these estatutarios-that is, what they and their employers would have contributed had they belonged to the INSS system during their years of service with the government. If one as- sumes that the INSS system is actuarially fair for workers who make full contributions for 30 to 35 years, this means that the INSS is "owed" em- ployee and employer contributions (which add up to about 30 percent of earnings) for their years of tenure. The estimated present values of "lost" contributions are R$16,000, R$42,000, and R$78,000 for men with 0 to 8, 9 to 1 1, and 12 or more years of schooling who take PDV packages after 17.5 years of service; and R$13,000, R$25,000, and R$54,000 for the corresponding groups of women with 15 years of service. Naturally, IMPROVING PUBLIC SECTOR EFFICIENCY IN BRAZIL 63 these numbers would be smaller for PDV takers with lower tenure levels, and greater for those more senior civil servants. Considering the generosity of public pensions, it is striking but not sur- prising to find that the potential benefits of PDV programs-in terms of reduced salary and pension bills-dwarfed the costs of existing programs. While our estimates in Table 2-4. indicate that the potential saving for these groups ranges between R$34,000 (for least-educated women) and R$265,000 (for educated men), the average benefits that were paid ranged from R$5,000 to R$25,000. One of the main reasons is that state gov- ernments are cash strapped and could not afford to pay more. As a conse- quence, though, take-up rates were low, and a promising source for re- ducing the long-term fiscal burden on state and federal governments remained largely untapped. Condusion Our analyses in this chapter corroborate the view that there are significant earnings gaps in many segments of the Brazilian public sector relative to similar private sector employees, but it is far from clear that public sector workers are systematically over- or undercompensated. In sharp contrast, differentials in pension levels were large and systematically higher for gov- ernment employees compared with similar private sector retirees. The es- timates in the mid-1990s indicated that compared to their private sector counterparts, pension levels of civil servants are 25 to 50 percent higher, depending on their salary level, gender, and occupation. Recent pension reforms for private sector workers may have widened these differentials. Based on this analysis, we recommend steps for reducing public sector payroll expenditures that differ across the various branches and levels of government. Table 2-10 presents the set of measures that we consider ap- propriate. Several points should be kept in mind in using this table. First, for efficiency and equity reasons the analysis is based on the principle that compensation should be the same for equally qualified workers in the public and private sectors. Second, the analysis presents strategies for re- ducing payroll expenses at all levels of government. Clearly, federal, state, and municipal priorities will dictate whether in fact downsizing efforts should be broad, or instead targeted to some workers (for example, lower- level administrative staff) while sparing others (for example, education and health workers). Finally, this analysis should be viewed as general guidance. Specific steps should be determined after more detailed analysis of public-private earnings differences at each level of government. 64 CRAFTING LABOR POLICY TABLE 2-10 Measures to Reduce Public Payroll Expenses While Maintaining or Improving Labor Market Efficiency Reduce Reduce salary/benefits Reduce pensions employment Judicial and legislative l/ / Federal administration / I State administration I / Municipal administration V Education and health / / Military Federal enterprises / / ? State enterprises / ? Municipal enterprises ? / signifies that this is a priority countrywide measure, though the measure may not be appropriate for some states, municipalities, or occupations. I signifies that this is a recommended countrywide measure. ? indicates that our analysis does not allow generalizatons. The measures suggested by the analysis are: * Judicial and legislative payroll expenses should be controlled by low- ering salary and pensions and by reducing employment levels. * Reducing personnel expenses in federal administration and federal and state enterprises mainly requires lowering salary and pension levels, though employment reduction in selected occupations and en- terprises may also be necessary. * Reducing personnel expenses in state administrations is best done by employment reduction preceded by tenure revocation and reforms to make state pension programs less generous. * Reduction of personnel expenses for education and health workers and municipal administrations and enterprises should be done mainly by reducing employment, preferably after revoking tenure for civil servants. Additionally, administrative reforms for public employment may in- clude reduction in tenure rights, enforcement of working hours, intro- duction of stricter performance monitoring, and the requirement for civil servants to make contributions for pensions at the same rate as for INSS benefits. This would reduce public-private differentials in earnings and IMPROVING PUBLIC SECTOR EFFICIENCY IN BRAZIL 65 pensions, and thus improve labor market efficiency. Moreover, reduction in public employment would become easier and cheaper to achieve. Concerning the pension levels, social security reform has become a pri- ority concern in Brazil's effort to carry out a lasting fiscal adjustment. In 1998, the overall pension deficit was 4.8 percent of GDP In the same year the Brazilian government responded to this challenge by introducing a set of reforms that put the social security system (RGPS) on a better actuar- ial base. Concerning access to pensions for civil servants, the government has changed the requirements for eligibility. It furthermore attempted to increase pension contributions by civil servants. This may be a first step in reducing the differences in compensation between public and private sec- tor workers and in building a sustainable and just social security system that better fits the country's budgetary constraints. Notes 1. Efficiency in this context means that the public employer minimizes the cost for a given output, and hence does not pay higher wages than necessary to attract sufficiently capable workers. 2. CLTistas are governed by the labor code for the private sector. This expres- sion derives from the Consolidafdo das Leis do Trabalho (CLT) of 1943 that forms the main body of Brazilian labor legislation. Among other things it determines that all workers must have a booklet where all individual labor contracts and changes to them over time are registered by the employers. By definition, a for- mal worker has a booklet signed by his employer. Formore information on Brazil- ian labor legislation, see Chapter 3. 3. For more information on the estimation of the overall wage gap, see Tech- nical Annex A. 4. Employees in the private formal sector are all employees in the private sec- tor who have a formaL labor contract. 5. For Brazil, using the PNAD September 1995, we find that tenure at the cur- rent job, even when adjusting for characteristics such as age, education, and gen- der, is longer for public sector workers in all states except Sao Paulo. 6. The INSS system comprises four major subsystems: time of service, disabil- ity, old age, and special programs. 7. Rio Grande do Sul introduced a 2 percent contribution rate in 1996. Fed- eral employees have had to contribute to their pensions since 1992, when contri- bution rates were set as a progressive scale. In 1997 this scale was replaced with a flat contribution rate of 11 percent. 8. The supplementary social security fund is privately managed, intended to supplement the RGPS benefits, and dominated mostly by pension funds of pub- 66 CRAFTING LABOR POLICY lic enterprises. The number of contributors is about 3 million (compared to 60 million in the RGPS). References Carneiro, Francisco G., and Indermit S. Gill. Forthcoming. "Effectiveness and Fi- nancial Costs of Voluntary Separation Programs in Brazil: 1995-1997." Re- vista Perspectiva Econ6mica. Pacs de Barros, Ricardo, Indermit S. Gill, Miguel Foguel, and Rosane Silva Pinto de Mendonca. 1997. "Labor Market Prospects of Public Employees in Brazil: An Empirical Evaluation." Economic Note No. 24. Country Department I, Latin America and the Caribbean Region, World Bank, Washington, D.C. PNAD (Pesquisa Nacional por Amostra de Domicilios). Various years. Brazil. Rama, Martin. 1997. "Downsizing and Severance Pay in Developing Countries." World Bank, Washington, D.C. World Bank. 2001. Brazil. Critical Issues in Social Security. A World Bank Coun- try Study. Washington D.C. 3 Assessing the Impact of Regulations on Informal Workers in Brazil EDWARD J. AMADEO, INDERMIT S. GILL, AND MARCELO C. NERI ASONE OF THE ALLEGED contributors to custo Brasil- the abnormally high costs of doing business in Brazil-labor legislation that both raises labor costs and makes them more uncertain has been a topic of some discussion for the last decade. The introduction in 1994 of the real, the new anchored currency, and the resultant decline in inflation, removed an instrument for keeping real wages flexible. Increased openness resulted in steadily rising unemployment rates (especially for more experi- enced industrial sector workers in areas such as Sao Paulo). These factors resulted in a steady buildup of pressure for labor market reforms. A sharp rise in open unemployment rates in 1998 precipitated by the Asian crisis of October 1997 brought labor market reforms to the forefront. As Brazil prepares to take on the difficult task of amending labor- related clauses in the Constitution and the labor code, a debate rages over whether inappropriate or outdated labor legislation is at all binding in de- veloping countries. The argument is that if, for example, minimum wages are set at above-market-clearing levels, these laws will simply be ignored rather than restricting employment. Similarly, if the workweek is specified at unrealistically low levels, employers and workers will agree on higher weekly hours, in effect annulling the mandate. If payroll tax rates are set at high levels and if the programs they fund are poorly designed, workers 67 68 CRAFTING LABOR POLICY and employers will conspire to avoid paying them altogether. One of the results of unsuitable mandates and inadequate enforcement capacity would be the emergence of an illegal, unregulated, or "informal" labor market, so that even if labor legislation changes were made, the effects on aggregate labor market outcomes would be insignificant. On the other hand, some proponents of labor reform argue that the excessively high taxes and labor market restrictions contribute to the increase in the infor- mal sector, because they provide incentives for evading formality. Using empirical evidence, they hint at a positive correlation between the level of payroll taxes or high firing restrictions and the level of informality, and hence recommend a reduction in taxes and labor market restrictions. This chapter shows that it is difficult to make sweeping generalizations on the question of whether labor laws matter and how they affect the size of and conditions in the informal sector. The more interesting and tractable question for policy purposes is, which laws matter, where, and why? This chapter suggests that the economic and political history of a country may be a critical determinant of the relative importance of legis- lation on wages and employment. A few examples of how the economic and political history of Brazil (and the manner in which labor laws are en- forced) has affected outcomes in both regulated and unregulated labor markets are: * Newcomers to Brazil are initially perplexed by the use of multiples of the legal minimum salary in individual and collective contract nego- tiations, in determining floor and ceiling pension benefits, and in of- ficial statistics on employment and earnings. The roots lie in a period of high inflation and widespread illiteracy of the workforce. Changes in the legally specified minimum salary-based on changes in the cost of living due to. inflation-seem to have provided a widely accepted and frequent signal to workers and employers to adjust nominal wages accordingly. We provide evidence that a significant number of work- ers are paid exactly the legal minimum wage, and adjustments in this wage are matched by salary adjustments even in Brazil's unregulated sector and, perhaps because of habit persistence, even in today's low- inflation environment. * Brazil's labor laws specify that workers must be paid within the first week of the month. We discuss evidence that this law-which was critical in periods of high inflation-appears to be obeyed by em- ployers in both the formal and informal sectors, even in today's low- inflation environment. ASSESSING THE IMPACr OF REGULATIONS ON INFORMAL WORKERS 69 Persistent high income inequality and a prolonged period of socialist policies has led to labor laws with a prolabor bias, and to labor courts acquiring a similar bias in their verdicts on disputes. Under Brazilian law, labor courts have policy-setting powers, in the sense that labor courts in judging a particular case are entided to formulate policies in areas where the labor code and the Constitution are ambiguous in the opinion of the court. Increased ambiguity of labor laws combined with a prolabor bias of dispute resolution has resulted in workers who are dismissed from either formal or informal employment being able to extract generous severance benefits from their former employers. These examples indicate that the subset of binding laws does not depend only on the design of labor legislation, but is also country specific.' Labor markets must be studied and understood country by country, not by com- piling and analyzing cross-country data designed to help arrive at a sum- mary verdict. In contrast to the general notion, we will show that labor legislation in Brazil not only affects the formal, but also the informal, sec- tor. Changes in labor legislation should take this into consideration.2 In this chapter, we first present some stylized facts about the Brazilian labor market and examine the evolution of labor market indicators, such as employment structure, job creation and unemployment, and labor in- come and earnings ratios in recent years. We briefly describe the evolution of labor legislation in Brazil from 1943 until 1998, explaining in detail the main labor laws, and discuss the main issues and objectives of the labor market reforms implemented and proposed in 1997-98. We examine the adherence to labor laws in the formal and informal sectors by looking for "pressure points" or evidence of clustering of labor market outcomes, such as wages and hours worked, exactly at the limits set by the law; and by dis- cussing whether changes in the labor legislation led to changes in these labor market outcomes. Relevant Labor Market Indicators The opening of the economy since the late 1980s and the sharp reduction of inflation since 1994 have changed the macroeconomic environment in Brazil.3 The analysis of relevant labor market indicators concentrates on the 1990s, with special emphasis given to the last four years after the launching of the Real Plan. In the case of variables that have more "struc- tural" determinants, such as unemployment and the level of informality of the labor market, the analysis goes back to the 1980s. 70 CRAFTING,LABOR POLICY Job Creation and Unemployment Despite the low rate of growth in the 1980s, neither job creation nor un- employment was a major problem during this decade. The level of em- ployment grew continually as more than 15 million new jobs were created. The labor force participation rate increased almost continuously from 1979 to 1987, then fell slighdy toward the early 1990s. The rate of open unemployment fluctuated during the 1980s, increasing during the reces- sion years of 1981-83, but never, surpassed 5 percent. Despite the adjust- ments associated with the opening of the economy, the early 1990s were not very different from the 1980s. In the recession of 1990-92, employ- ment decreased but then recovered, growing almost 10 percent between 1991 and 1996. Compared to the 1980s there has been a slight increase in the rate of unemployment. However, it has remained low and stable, os- cillating between 4 and 6 percent. Only in 1997 and 1998 did unem- ployment rise to above 6 percent, prompting concerns about joblessness. The Structure of Employment According to data of the Pesquisa Nacional por Amostra de Domicilios (PNAD-the annual nationwide household survey), the structure of em- ployment by type of working relation remained relatively stable during the 1980s. The share of formal workers remained around 32 percent, the share of informal workers around 22 percent, and the share of self-employed around 26 percent. Public servants accounted for approximately 12 per- cent, and nonpaid workers for around 7 percent. Data of the-Pesquisa Mensual do Emprego (PME-monthly household surveys in the six main metropolitan regions)4 show a persistent increase of the share of self-employed since 1986. Between 1986 and 1996 the share of self-employed went from 16.5 percent of the metropolitan labor force to 23 percent. The share of informal wage earners fell between early 1984 and 1989 and then started increasing, to reach almost 25 percent of the labor force in 1996. As a consequence, the share of the remaining wage earners (formal, public, and nonpaid workers) fell from an average of 58 percent during 1985-90 to less than 47 percent in 1996. Hence, around 10 percent of the labor force moved from legal wage earner status to either informal or self-employed status since the late 1980s. As for the sectoral structure of employment nationwide, PNAD data for the 1980s show a stable share of industry and manufacturing employ- ment, and a reduction in agricultural employment compensated by an in- ASSESSING THE IMPACT OF REGULAnIONS ON INFORMAL WORKERS 71 crease in service. Both PNAD and PME data for the 1 990s show a marked decrease in manufacturing employment and an increase in service sector employment. There is a clear movement in favor of job creation in the nontradable sector and against the tradable sector during the 1980s and the 1990s. Labor Turnover andJob Quality Looking at a selected set of countries in the early 1990s, the share of work- ers in the manufacturing sector with less than one year in service is 10 per- cent in Japan, 12 percent in Germany, 15 percent in France, 28 percent in the United States, and 33 percent in Brazil. For manufacturing workers with less than five years of service, the share in the labor force is 37 per- cent in Japan, 41 percent in Germany, 42 percent in France, 62 percent in the United States, and 71 percent in Brazil. Hence job tenure is greater in Japan, Germany, and France than in the United States and Brazil. Even when considering all sectors, workers change jobs frequently in Brazil and the United States and very infrequently in Germany and other European countries. In contrast, the duration of unemployment, while increasing, remains short in Brazil and the United States, and considerably higher in Germany and European countries. In Japan the frequency and duration of unemployment spells are both relatively low. Table 3-1 presents available data on labor turnover in the Brazilian for- mal sector. The data-should be read as follows. In 1985, on average, 2.8 percent of workers in all Brazilian legally registered firms with more than five employees changed their job every month. In 1989, 40 percent of the workers changed their jobs over the year. Thus, during 1989-93, 28 per- cent or more of workers in legally registered firms changed their jobs in the period of one year. Turnover is lower in the manufacturing and service sectors than in the trade sector where, in turn, turnover is lower than in the construction sector.5 The labor turnover statistics presented in Table 3-1 are based on the Relado Annual de Informapies Sociais (RAIS) and the Cadastro Geral de Empregados e Desempregados (CAGED) data, which are based on legally required self-reporting by firms. This sample comprises legally registered firms with more than five employees. Consequently, they do not cover the informal segment of labor markets. It is possible to build turnover mea- sures based on the longitudinal aspect of PME data for the six main Brazil- ian metropolitan regions. The turnover rate in the informal sector is three to four times greater than the already high turnover rate found for the for- 72 CRAFTING LABOR POLICY TABLE 3-1 Labor Turnover Rates, Brazilian Formal Labor Market, 1985-93 Year Monthly average Annual 1985 2.80 n.a. 1986 3.67 n.a. 1987 3.72 n.a. 1988 3.80 n.a. 1989 3.49 39.66 1990 3.26 38.20 1991 2.69 35.75 1992 2.26 28.05 1993a 2.73 32.81 n.a. Not applicable. a. January-October 1993. RAIS and CAGED data. Source: Ministry of Labor, Law 4923. Calculations, Amadeo and oth- ers (1994). mal segment of the labor market. The high level of labor turnover is some- times viewed as evidence of the flexibility of the labor market, especially if one takes into account illegal employees. However, high levels of turnover may also be associated with low levels of investment in firm-specific human capital (see Amadeo and others 1993). Furthermore, as we will dis- cuss below, the design of the Brazilian severance fund Fundo de Garantia por Tempo de Servico (FGTS) may artificially raise the level of turnover in the formal sector by providing strong incentives for workers to induce their dismissal in order to access their FGTS balances. Labor Income From 1980 to 1990, gross domestic product (GDP) increased by 17 per- cent while the labor force increased 40 percent, implying a decrease of 17 percent in average productivity. In a "competitive" labor market, wages would have fallen to accommodate the reduction in productivity. In fact, the average labor income per worker did fall 22 percent between 1980 and 1990.6 The path of the average labor income followed the oscillations of output, falling during the 1981-83 recession, increasing in the 1986 boom (the Cruzado Plan), then declining at the end of the decade. In the 1990s, real labor income also oscillated according to the output cycle in ASSESSING THE IMPACT OF REGULATIONS ON INFORMAL WORKERS 73 1991-92, falling by 17 percent between 1991 and 1992, then recovering by 35 percent between 1992 and 1996. Sectoral Earnings Ratios In the 1980s relative earnings moved strongly in favor of manufacturing wages and against nontradable sectors. This movement resulted from the increase in labor absorption in the service sector without a corresponding increase in output, on one hand, and the protectionist tariff and exchange rate policies which favored the manufacturing sector, on the other hand. The 1990s were marked by a symmetric movement with the opening of the economy and the pronounced appreciation of the exchange rate re- ducing the competitiveness of the tradable sectors, with negative effects on employment and relative earnings in manufacturing. Relative earnings between manufacturing and service sector workers fell from 1.6 during 1992-93 to 1.3 in 1996. Between manufacturing and trade workers it fell from 1.5 in 1992 to 1.27 in 1996. The Brazilian Labor Code We now turn to the main characteristics and elements of the Brazilian labor code and the main changes to it since it was created in the 1940s.7 The Consolidated Labor Code The main body of Brazilian labor legislation was introduced in the 1940s, and was consolidated into the Consolidac,o das Leis do Trabalho (CLT) in 1943. The CLT is a large, often overlapping set of rules that determines individual and collective rights and duties of workers, unions, and firms. The law mandates that all workers have a booklet in which all individual labor contracts and changes to them are registered by the employer. Con- sequently, a worker with a booklet signed by his employer (com carteira assinada) is considered a formal worker. Signing the booklet implies an obligation on the part of employer and employee to contribute to social security.8 Besides the obligation to sign the booklet, the law stipulates a set of min- imum conditions an employment relationship must meet. The most im- portant rules address maximum hours of work per week, maximum over- time working hours, minimum payment for overtime work, minimum wage, prepaid annual vacations. The rules include special protection clauses 74 CRAFTING LABOR POLICY for women and children that forbid the dismissal of pregnant women and provide for paid vacation before and after childbirth for the mother, special work conditions for night shifts, one-month prenotification of firing, and protection against unjustified dismissals. Between 1962 and 1988 the main changes were: • In 1962, introduction of a mandatory end-of-year bonus of one monthly wage (called the "thirteenth salary"). * In 1963, introduction of a mandatory family allowance. * In 1965, introduction of a wage-adjustment law which determined the minimum rate of wage adjustments of all workers in the economy. * In 1966, creation of an individualized severance fund (FGTS) in place of a clause forbidding dismissal of workers with more than 10 years of tenure. * In 1986, creation of an unemployment insurance program which today covers about 25 percent of the country's labor force. Reform of Brazilian labor regulations requires either changes in labor legislation or constitutional reform. In addition, reforms can be imple- mented immediately by the executive branch through executive decrees. Amendments to the Constitution require a two-thirds majority vote in both the upper and lower houses of Congress, twice. Changes in clauses of the CLT require a single vote in the upper and lower house and can be approved by simple majority. Examples of reforms that do not require votes in the two houses are the elimination of incompatibilities between the labor code and the introduction of occupation-specific contracts. They further may refer to minor changes, such as allowing shorter hours per week, or improvement of the inspection of occupational safety regulation. Extending the duration of unemployment benefits or allowing temporary layoffs without any severance payment can be approved by simple major- ity. Changing rules of union finance and membership, minimum wage- setting procedures, or reforms of the unemployment insurance system re- quire a Constitutional change. The Reforms of 1988 In 1988 a new federal Constitution was approved and new labor clauses introduced. The main changes in labor legislation concerning individual rights introduced in the Constitution of 1988 were: * Reduction of the maximum number of hours a worker worked per week before receiving overtime wages from 48 hours to 44 hours, ASSESSING THE IMPACT OF REGULATIONS ON INFORMAL WORKERS 75 and an increase in the minimum payment for overtime hours from 120 percent to 150 percent of the worker's regular wages. * Reduction of the maximum daily working time for continuous work shifts from eight hours to six hours. * Creation of a mandatory vacation bonus of one-third of the worker's wages. * Increase in the paid maternity leave for mothers to 120 days, and in- troduction of five days' paid childbirth leave for the father. * Increase in the firing costs for unjustified dismissals (which includes reductions in workforce due to economic reasons) from 10 percent of the FGTS balance to 40 percent. This is the list of the minimum individual rights for private sector and state enterprise workers. In general, therefore, the new Constitution clearly mandated higher nonwage benefits and made dismissals costlier for employers. Working conditions can be improved further through negoti- ations between the individual worker and the firm, or through collective bargaining. In 2001, the government introduced legislation that would allow collectively bargained contracts that provide nonwage benefits less than those stipulated in the CLT. Severance Rules and Unemployment Compensation Until 1965, to fire a worker without proper justification the employer had to pay one month's salary for each year of work in the firm. The com- pensation was calculated on the basis of the highest wage received during the work contract. It was a duty of the employer to prove the dismissal was justified, and the conditions for justified dismissals were clearly de- fined in the law. In 1966, this system changed and a severance fund was created, called the FGTS. The FGTS provides an individualized fund for each worker. Every month the employer contributes the equivalent of 8 percent of the employee's current wage to the fund. This implies that the worker's FGTS increases with tenure and that the amount accumulated in the Fund per year corresponds approximately to one monthly wage. In general, the FGTS can be accessed only by the worker upon dismissal without just cause or upon retirement.9 When workers are dismissed without just cause (sem justa causa) they also have the additional right to receive a penalty in proportion to their FGTS. Thus, until 1988, the dis- missed worker received access to his or her accumulated balance in the FGTS, and additional compensation of 10 percent of the FGTS balance at the time of dismissal if this dismissal was considered to be without just 76 CRAFTING LABOR POLICY cause. In 1988 this fine for unjust dismissal was increased to 40 percent of the FGTS. Besides this fine, the employer must notify the worker one month be- fore he or she is fired. This is the avisoprdvio law, or a mandate for the em- ployer to provide "advance notification" of firing. During the month the worker receives the advance notification of firing, he or she is allowed, ac- cording to the law, to take two hours a day to look for a new job. This im- plies a minimum cost of 25 percent of the worker's monthly wage. In fact, the cost is usually higher since firms end up paying the notification fee to workers and dismissing them immediately. The total cost of dismissal is thus 25 to 100 percent of the monthly wage plus 40 percent of the FGTS. The FGTS depends on the number of months the worker has worked for the firm. Table 3-2 shows the costs for the firm as a multiple of monthly wages, according to the number of years of the worker's contract, under the assumption that the firm dismisses the worker upon notification. For example, if the worker stayed with the firm one year, the cost of dismissal is, at most, 1.41 monthly wages. The cost to dismiss a worker who has been with the firm for five years is, at most, 3.19 monthly wages, and so on. According to PMEs in the six largest metropolitan regions during 1982-98, 72 percent of workers who reported separations from their jobs were in fact fired, 85 percent of whom received the FGTS. There are two points to note here. First, the share of dismissals among those exiting em- ployment rose from 67 percent to 76 percent after the Constitution raised firing fines from 10 percent to 40 percent. This result indicates that higher firing fines do not reduce dismissals but, on the contrary, may even in- crease them. Second, about 65 percent of the workers that reported vol- untary separation during 1982-98 also reported that they had access to the FGTS. This hints at illicit agreements between firms and workers re- garding access to FGTS funds. The fact that being fired is the main mech- anism which allows access to the FGTS may be a significant incentive to TABLE 3-2 Total Cost of Firing a Worker (multiple of monthly wages) Tenure I yr. 2yrs. 3yrs. 4yrs. 5yrs. lOyrs. 15yrs. 2Oyrs. FGTS fine 0.41 0.84 1.27 1.72 2.19 4.72 7.66 11.07 Aviso previo 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 Total 1.41 1.84 2.27 2.72 3.19 .5.72 8.66 12.07 ASSESSING THE IMPACT OF REGULATIONS ON INFORMAL WORKERS 77 induce dismissal. This may be exacerbated by credit constraints and by the fact that the fund is administrated by the government, which, because of poor management, typically generates real returns below market returns. Since 1986, besides advance notice, access to the FGTS, and the 40 percent fine for unfair dismissal, the worker also has the right to unem- ployment compensation benefits under the Seguro di Desemprego pro- gram. The unemployment compensation program offers partial coverage for up to four months of unemployment (extended to five months after 1996). To become eligible for the benefit, the worker must (a) have been dismissed without just cause, (b) have had a formal labor contract during the last six months or have been legally self-employed for at least 15 months, (c) be unemployed for at least seven days, (d) not receive any other pension, and (e) not have any other type of income sufficient to guarantee his own subsistence and that of his family. The value of the ben- efit cannot be lower than the value of the minimum wage, is adjusted monthly for inflation, and is related to the average wage received by the worker in the last three months in his or her previous job. Wage Laws An important change in the CLT was the introduction of the Wage Ad- justment Law in 1965. Before this date, wage adjustments were fixed through collective bargaining between workers and employer unions, and through individual negotiations between the individual worker and his or her employer. Only the minimum wage was determined directly by the president of the republic, although most of the time it automatically incorporated the prescriptions given by indexation clauses imbedded in the law. The Wage Adjustment Law gave the government the right to determine the minimum rate of adjustment of all wages in the formal sector of the economy. The first wage law stipulated that nominal wages should be ad- justed once a year, at a date which was specified for each occupation, fol- lowing a formula which took the past and expected future rate of inflation and the growth rate in GDP per capita as the base for the adjustments. The specific formula and the adjustment period changed many times over the years as the rate of inflation increased. In mid-1995, one year after the introduction of the Real Plan, the Wage Adjustment Law was abolished. Today, upward adjustment of wages is negotiated between employers and employees, but for all practical pur- poses downward adjustment of wages is prohibited by the Constitution. Attempts to do so open employers to lawsuits, which are generally re- TABLE 3-3 Stylized Facts of Wage Indexation Regimes, 1980-2000 Dec. March June Jan. May March Sept. Dec. July Starting Date 1979 1986 1987 1989 1989 1990 1991 1992 1995 Duration (months) 63 15 20 4 10 18 15 9 84 Stabilization plan - Cruzado Bresser Summer - Collor - - Real Transition phasea Instant Instant Gradual Instant Instant Instant Gradual Gradual - Rule typeb Time State Time Time Time - Time, R Time, R - Triggerpoint 6 m 20% 1 m - I m - 4.2 m 4.2 m - Average lagc 8 m _d 4 m - m - 4 m 4m - a. This attribute indicates whether the transition ro the new wage indexation regime was done in an instantaneous or a gradual manner. b. For time-dependent rules the trigger point is specified in terms of months between adjustments. For the state-dependent rule the trigger point was specified in terms of accumulated price index variation between adjustments. R means regressive adjustments, that is, lower wages get higher adjustments. c. Refers to the average lag between price rises and their incorporation into wages. d. In the state-dependent case the lag is endogenous. ASSESSING THE IMPACT OF REGULATIONS ON INFORMAL WORKERS 79 solved in favor of the worker. This was irrelevant during a time of high in- flation, but now quite possibly adds to the rigidity of the labor market. Payroll Taxes and Mandatory Contributions The CLT and the 1988 Constitution stipulate a comprehensive set of minimum standards any individual contract must follow. The rules do not provide much space for negotiation between employers and workers. The result is a rigid set of minimum rules, which reduces the flexibility of the labor contract in the face of changes in the economic environment. In ad- dition to the costs imposed by this inflexibility, there are more direct and obvious nonwage costs due to payroll taxes and mandatory benefits re- quired by the law. Table 3-4 shows the composition of the labor cost in Brazil. The cost of labor can be decomposed into four parts: * The basic contractual wage (60 percent of total cost = 100/165.4). * Mandatory benefits, which include the annual one-month bonus (dcimo terceiro saldrio), the contribution to the severance find FGTS, vacations, and other benefits (23 percent of total cost = 37.6/165.4). TABLE 3-4 Wage and Nonwage Labor Costs (monthly) Component Percent Total Basic Wage 100.0 Annual bonus 8.3 108.3 Vacations 11.3 119.6 Severance fund contribution (FGTS) 8.0 127.6 Other mandatory benefitsa 10.0 137.6 Total pay (basic wage + mandatory benefits) 137.6 SESI, SENAI, SEBRAE (employer associations) 3.1 140.7 INSSb + accident insurance + education + INCRA 24.7 165.4 Note: Normal number of hours = 44 weeks. a. These are benefits that cannot be calculated for all workers since they depend on gen- der, kind of work done, economic sector, and so forth. These indude family allowances, pregnancy leaves, and transport subsidies. b. Workers contribute 8, 9, or 10 percent of their wages to social security, depending on the wage. 8o CRAFTING LABOR POLICY *Contributions to the official training system (Servifo Nacional de Aprendizagem Industrial-SENAI, and Servifo Nacional de Apren- dizagem Comercial-SENAC), to finance an institution which assists small enterprises (Servifo de Apoioas Pequenas Empresas-SEBRAE), and a contribution paid by firms to finance a workers' assistance ser- vice (Servifo Social de Industria-SESI, or Servifo Social do Comer- cio-SESC) (2 percent of total cost = 3.1/165.4). * Contribution to the federal social security system (INSS) and to fund educational services (sa/drio educafao) and an on-the-job accident in- surance fee mandatory for all firms and proportional to the payroll (14.8 percent of total cost = 24.5/165.4). In addition to these contributions based on payroll costs, employers are also charged levies on revenues to pay for additional INSS-related obliga- tions (cofins), which in 1999 were raised from 1 to 2 percent. Additionally, employers have to pay contributions to the Fundo de Aparelho de Traba- Ihador (FAT), which funds unemployment compensation, job search as- sistance, and active labor programs such as training and microenterprise support schemes (PIS/PASEP). These labor-related levies can add up to be- tween 2 and 3 percent of employer revenues.10 Reforms Proposed and Implemented in the 1990s Brazil faces formidable challenges in the area of labor legislation. As pointed out, the set of laws that constitute the labor code have their basis in rules formulated in the 1 940s, with additional-sometimes overlapping or inconsistent-legislation added over the years in response to both gen- uine labor market concerns and for shortsighted political reasons. Today, the regulation of the labor market is a daunting task for the Ministry of Labor for the following reasons: * The plethora of laws has led to uncertainty about which regulations apply and under what circumstances, which results in frequent dis- putes between employers and employees." * These disputes are resolved by labor courts, which have earned the reputation of having a strong prolabor bias. Under Brazilian law, labor courts have policy-setting powers, in that labor courts-in judging a particular case-are entitled to formulate policies in areas where the law is ambiguous in the opinion of the court. 12 * No employment contract is strictly legal unless approved by the Min- istry of Labor, which leads to the Ministry having to devise and val- idate special contracts for specific working conditions, without which employers are left vulnerable to expensive lawsuits. ASSESSING THE IMPACT OF REGULATIONS ON INFORMAL WORKERS 8I * Collective bargaining between workers and empLoyers can be an in- strument for formulating more definite contracts, but collective bar- gaining rules in Brazil and the practices they have engendered are often insensitive to work-specific conditions. * The high rates of payroll contributions and the low benefits of pro- grams they fund encourage evasion and informality. Consequently the five fundamental aims of the government's reform agenda are to: * Reduce the uncertainty of labor costs for employers. * Create the conditions for more durable employee-employer relation- ships, so that both employers and employees voluntarily choose to stay together because the contract can be changed without friction in response to changing work and market conditions. * Create the environment for more representative collective bargaining. * Reform implementing institutions to ensure better enforcement of contracts. * Reduce the incentives to become informal. Addressing these issues, the government implemented some labor re- forms in 1997. They allowed the introduction of temporary contracts of employment with lower payroll taxes and considerably lower dismissal costs. In general, reforms that have been debated over the past few years include the following: * Eliminating contradictions between labor legislation and the constitu- tion. The Ministry has submitted a bill for consideration by Congress that seeks to eliminate contradictions between labor legislation and worker rights guaranteed under the dismissals to reduce the perverse incentives for workers to induce dismissal. * Severance laws. While reform of the FGTS has not been formally at- tempted, there have been proposals suggesting delinking the access to the fund from dismissals in order to reduce the perverse incentives for workers to induce dismissal. A 1998 Ministry of Labor proposal to re- duce the rate of employer contributions to fund the FGTS from 8 per- cent to 2 percent of payroll met with opposition within and outside the government, and was dropped. If seriously considered, this could be ac- companied by reform of the unemployment compensation system to function more like an unemployment insurance plan (that is, actuari- ally based). In 1998, the government proposed the allowance of tem- porary layoffs, which are funded by credits of the FAT. The employer has to repay these funds if he does not rehire the worker at the end of the layoff. 82 CRAFTING LABOR POLICY * Maximum hours worked. The government is considering a reduction in weekly work hours (jornada de trabalho) from 44 to 40.13 One of the proposals of the current administration is to reduce the hours of work that qualify a person for a full-time contract (which entitles the person to greater benefits than part-time work). The objective of this measure is to make the workweek more flexible by permitting daily work to range from 5 to 8 hours, and weekly hours from 26 to 40 hours. * Temporary contracts. In 1997, the government introduced legislation that would allow employers to hire new workers on temporary con- tracts, during which non-INSS-related payroll taxes would be waived and dismissals would be less costly. These measures were expected to lead to increased employment because the nonwage costs of labor would be reduced. * Union representation. For a single sector or occupation, Brazilian labor laws do not allow for more than one union per municipality- the unicidade sindical provision. For all practical purposes, this leg- islation outlaws plant-level collective bargaining. The Ministry of Labor has proposed that this law be altered to facilitate collective bar- gaining to reflect firm-level conditions. The Ministry has also pro- posed changes in the mechanisms by which unions are financed, making union fees voluntary rather than mandatory. In 2001, the government initiated efforts to allow union-negotiated contracts to set conditions "below" the floor set by the CLT. * Policymaking powers of labor courts. Brazilian labor laws give labor courts policy-setting powers (poder normativo) in that these courts can establish policy on issues which are left unclear by the CLT and the Constitution. As a result, labor court rulings have influence far beyond the case being arbitrated, in effect serving a policymaking role that should be the responsibility of the Ministry of Labor. Re- forms being contemplated to curb this policymaking role of labor courts while also reducing ambiguities in the labor law are thus likely to reduce the uncertainty regarding the full cost of labor, and hence result, in increased labor demand. * Minimum wage. Brazil has a nationwide minimum wage that is, at the same time, the minimum legal wage in the private sector and the minimum payment for pensions of the social security system. Ap- propriate reforms would attempt to (a) regionalize the minimum wage, (b) delink social security pensions from the minimum wage, and (c) delink public employee salaries from the minimum wage. ASSESSING THE IMPACT OF REGULATIONS ON INFORMAL WORKERS 83 The Effects of Labor Legislation on Informal and Formal Workers The informal sector is defined as the set of economic units that do not comply with taxes and regulations imposed by the government (Hernando de Soto 1989). Following accepted practice, throughout this chapter we distinguish formal from informal employment by determining whether or not the contract has been ratified by the Ministry of Labor; that is, whether the worker has a signed work card (com carteira assinada) or not (sem carteira assinada).14 For reasons of comparison between the formal and informal sector, our analysis concentrates on wage earners. We thus ignore the (informal) self-employed. In this section we analyze and con- trast the effect of various labor regulation schemes on the formal and in- formal segments of the labor market. We focus on labor regulations con- cerning minimum wages and maximum working hours. We provide evidence that informal jobs in Brazil are not necessarily un- regulated. In other words, we argue that institutions and the legal appara- tus affect both formal and informal workers. The distinction between legal and illegal employment appears rather to be associated with the incentives and costs for both the employer and the employee of maintaining a legal contract and, to a much lower degree, with the quality of jobs in terms of working conditions and wages. In this respect, the high proportion of in- formality of labor contracts on the one hand may be seen as emanating from poor design of the programs-the social security system, the sever- ance fund, the fund for providing unemployment compensation and more active labor programs, and various other schemes that are funded by high levels of mandatory contributions. On the other hand, it supports the view that, as Maloney (1998) puts it, the informal sector may be seen as a way to avoid inefficiency arising from labor market regulation, rather than the regulations themselves. Given that signing the work card requires pay- ing rather high contributions to the social security system, and that links between INSS benefits and contributions are low, some workers may have an incentive to evade labor legislation and seek informal employment. Brazil offers unique conditions for testing the effects of regulation on formal and informal labor market outcomes for the following reasons: * A substantial share of employees work in the illegal sector. * Labor market surveys in Brazil have traditionally explicitly asked whether or not employees possess working permits (carteira de tra- balho), allowing us to distinguish formal from informal employees. 84 CRAFTING LABOR POLICY * Large household surveys allow us to follow the same individuals through short periods of time. This longitudinal aspect allows us to analyze changes in several labor market outcomes at an individual level. * Perhaps most important, a regulated labor market exists and these regulations change from time to time, offering "natural experiments" to study the effects of regulation. The final point merits elaboration. Institutional effervescence in Brazil is due in part to adoption of the new Brazilian Constitution in 1988, and in part by the transition from hyperinflation before 1994 to almost zero inflation in 1998. The former offers the possibility of estimating the im- pact of various labor code provisions (for example, changes in maximum hours allowed, payroll taxes, and firing fines) on informal and formal wage employees by exploring the variation in labor market outcomes before and after the new Constitution. To assess to what extent labor legislation affects the formal and infor- mal sector, we start by looking at the percentage of workers in the formal and informal sector that comply with the limits set by the legislation. We then ask whether minimum wage and maximum hour legislation actually distort the wage and hour distribution in the informal and formal sector workers and answer this question by. first assessing the size of the cluster- ing exactly at the limits set by the law. * Wages. In the case of wages we assess how many individuals earn exactly one minimum wage. The idea here is that in the absence of regulation the wage distribution would be continuous, which means that each point of the distribution would have a zero mass. If the minimum wage law forces a change in the distribution, then we will observe a mass point at the minimum wage level. * Hours worked. Similarly, in the case of hours worked we assess the proportion of individuals who work the maximum number of hours allowed without requiring the payment of overtime. Second, we study the effects of changes in labor legislation on the distri- bution of wages and hours worked. This dynamic approach provides a check against habit formation in labor markets. As pointed out, history does matter, and the wage and hour level that we observe may be remi- niscences from the past. For example, people in the informal sector may frequently work 44 or 48 hours a week because it is a long-established tra- dition that also happens to be present in the labor code. To tackle this ASSESSING THE IMPACT OF REGUIATIONS ON INFORMAL WORKERS 85 issue, we test whether the relevant labor regulations are binding before and after the change is introduced. * Wages. We choose dates of minimum wage changes and assess the share of employees who adjust their earnings at the official minimum wage adjustment rates. We have already mentioned that changes in legally specified minimum salary had an important signaling effect during periods of high inflation. Now we show that wage adjust- ments are matched by salary adjustments in Brazil's unregulated sec- tor, even in today's low-inflation environment. * Hours worked. Similarly, we assess the proportion of individuals who work the maximum number of hours that are allowed for formal workers without requiring the payment of overtime. Minimum Wage Regulation Analysis of Distribution of Wage Levels Our empirical strategy to gauge the effects of wage regulation is the direct analysis of earnings distribution. The idea is to assess the relative concen- tration of the mass of the distribution of labor market variables at the min- imum wage level observed. If there is a point with positive mass at the minimum wage, the size of the spike of the frequency distribution at the minimum wage level constitutes a key statistic to quantify the effective- ness of the wage floor imposed. We start this analysis by providing evidence on the percentage of wage earners paid above the minimum wage (see Table 3-5). As can be TABLE 3-5 Wage Earners Paid above the Minimum Wage (percent) All Legal Illegal Recife 90 100 83 Salvador 91 100 83 Belo Horizonte 96 100 93 Rio de Janeiro 98 100 96 Sao Paulo 98 100 97 Porto Alegre 97 100 94 Note: By metropolitan region. Source: PME, September 1998. 86 CRAFTING LABOR POLICY seen, minimum wages are binding in the formal sector, but in the infor- mal sector the percentage of wage earners paid below the minimum wage level is strikingly low, ranging from 17 percent in Recife to 3 percent in Sao Paulo. The fact that we do not observe formal wage earners below the mini- mum wage does not necessarily mean that the minimum wage law affects formal sector wages. It may be just too low. If this is the case, then the dis- tribution of formal sector wages should in principle be continuous at each point in the distribution. If a minimum wage regulation is introduced and is binding, then all the workers that initially earn below or equal to the minimum wage should earn exactly the minimum wage after the intro- duction. Given that the minimum wage law is binding; the frequency dis- tribution at the minimum wage level indicates the share of people who would have earned the minimum wage or less without the legislative con- straint. This is not the correct interpretation for informal sector workers, as we still observe some workers earning less than the legal minimum. However, if we observe a point with positive mass at the minimum wage in this sector, then this indicates that some employers in the informal sec- tor respect minimum wage laws, and that consequendy minimum wage laws affect the earnings distribution of informal sector workers. Table 3-6 presents the relative concentration of wage levels for Febru- ary 1998 at exactly one minimum wage in the six main metropolitan re- gions according to PME. Ten percent of all employees earned exacdy one minimum wage. The point to be noted here is that the share of illegal em- ployees (14 percent) earning one minimum wage is double the share ob- served for legal employees (7 percent). As can be seen in Table 3-6, this TABLE 3-6 Wages Exacdy Equal to One Minimum Salary (percent) All Legal Illegal Recife 18 15 23 Salvador 24 20 29 Belo Horizonte 10 8 14 Rio de Janeiro 8 4 12 Sao Paulo 2 1 5 Porto Alegre 6 4 7 Total 10 7 14 Note: By metropolitan. region. Source: PME, September 1998. ASSESSING THE IMPACT OF REGULATIONS ON INFORMAL WORKERS 87 TABLE 3-7 Wages Exactly Equal to One Minimum Salary (percent) Legal Illegal Public North 9 21 11 Northeast 14 20 14 Southeast 6 15 6 South 6 15 4 All Brazil 8 18 9 Note: For all of Brazil. Source: Pesquisa NacionalporAmostra de Domic{lios (1996). greater share of illegal employees earning exactly one minimum wage is observed in all six metropolitan regions. Table 3-7 presents the share of total legal and illegal employees earning exactly one minimum wage in September 1996 for the whole country, ac- cording to the nationwide PNAD. Again, the share of minimum wage earners is highest in the illegal sector, compared to the formal and the pub- lic sector. In Tables 3-6 and 3-7 the share of employees earning exactly one minimum wage is larger in the illegal sector. This reflects the fact that the wage distribution of informal employees lies to the left of the distribution of the formal sector, which to some extent may be due to differences in worker characteristics such as age and education. Results of Maloney and Nufiez (2001) strongly support our findings. Using a sample from the PNAD,'5 the authors show, as can be seen in Fig- ure 3-1, that the minimum wage law strongly alters the wage curve in the informal sector. Analysis of Distribution of Wage Adjustments We have shown that minimum wage laws do affect the wage distribution of formal and informal sector workers. We now address the issue of whether this is a vestige of the past, or whether wages really move with changes in labor legislation. The indirect effect through which the mini- mum wage policy affects illegal employee earnings has been called the "lighthouse effect" (efeito-farol) of regulations. It is claimed that official wage rules would act as a "voluntary" reference price to illegal employees. However, as Maloney and Nufiez (2001) point out, the fact that inflation was moderate in 1998 indicates that the minimum wage is not only a sig- nal in a high-inflation country, but appears to be an important benchmark 88 CRAFTING LABOR POLICY FIGURE 3-1 Wage Distribution of Formal and Informal Male Workers, Brazil 1998 0.8 - Informal sector 0.6 - ., Formal sector 02 0.0 - 2 4 6 8 10 Log wage Source: PNAD (1998). for "fair" remuneration. A complementary, country-specific channel in Brazil is that illegal employees may get the benefits of labor market regu- lation by suing their employers for evading the minimum wage law. If a court decision is favorable to employees, firms have to pay all benefits de- nied to illegal employees. To shed light on the connection between illegal employee wage and wage law description, we take advantage of the longitudinal aspect of earn- ings information present in PME data. Our approach consists of deter- mining the share of employees with earnings adjustment rates exactly equal to the official minimum wage adjustment rates at the respective ad- justment dates. This allows us to describe how earnings adjustment rates evolve through time and identify their focal points. The idea is that em- ployers in the informal sector may frequendy pay minimum wages to some informal sector workers because it serves as a kind of reference point. But if we observe that changes in the legally specified minimum wages re- sult in changes in actual wages paid, then we can conclude that the mini- mum wage law is binding. Figure 3-2 presents the shares of legal and illegal employees that re- ceived earnings adjustments rates exacdy equal to the ones of the mini- ASSESSING THE IMPACT OF REGULATIONS ON INFORMAL WORKERS 89 FIGURE 3-2 Proportion of Wage Adjustments Exactly Equal to Minimum Wage Adjustments, Legal and Illegal Employees 25 - -- Legal 20 - ile-gal 15 10 Source: PME. mum wage at minimum wage adjustment dates that occurred during 1982-95. Table 3-8 summarizes the information presented in Figure 3-2 as aver- ages of selected periods. The data show that (a) the two adjustment dates after the launching of the Real Plan (September 1994 and May 1995) pre- sented a sharp rise in the influence of the minimum wage on salary levels in all three universes analyzed; (b) if we abstract from the post-Real Plan period there is not a clear trend in the degree of minimum wage influence in the legal employee universe analyzed; and (c) there is a sharp rise in the influence of minimum wages on the informal employee segment. All things remaining the same, in the first half of the 1980s less than 3 percent of illegal employees adjusted at the minimum wage adjustment rate, while this ratio rose to more than 10 percent in the first half of the 1990s. To compare the short-term movements of earnings at a monthly level between the two sectors, we proceed as follows. First, we calculate the ratio of a month's nominal earnings with respect to the earnings of the previous 90 CRAFTING LABOR POLICY TABLE 3-8 Proportion of Wage Adjustments Exactly Equal to Minimum Wage Adjustments at Dates When the Minimum Salaries Were Changed (percent) Period All Legal Illegal May 1980 to February 1986 6.00 6.38 2.75 March 1986 to February 1990 5.27 5.47 4.82 March 1990 to January 1994 7.04 6.86 10.29 September 1994 to May 1995 14.00 12.00 21.50 Average 6.25 6.21 7.03 Note: Average across selected periods. Source: PME. month (that is, one plus the monthly nominal earnings rate of change). For each month we plot the distribution of this ratio. We find that the main focal point corresponds to a nominal earnings ratio of one, which means that nominal incomes had been rigid between two consecutive months. It is interesting to note that in all graphs the illegal segment pre- sents a greater share of people with constant earnings. At the same time, illegal employees present, in general, a greater am- plitude of the rates of change of nominal earnings measured by the spread between the 3rd and the 97th percentile of the distribution. The main characteristics of the distribution of illegal employees' monthly adjust- ments distributions is the existence of plateaus or focal points equal to 1.5 and 2. Legal employees present these focal points during Novem- ber and December, which are probably connected to the payment of the mandatory annual bonus (d6cimo terceiro salario) in one or two receipts. Concerning illegal employees' distributions, however, these focal points are relatively more important and more spread out through the year. Regulations on Hours Worked Working houfs are an important determinant of job quality. Consistent with our analysis of minimum wages, we start by providing evidence on the percentage of workers who work hours below the maximum hours limit. Before the Constitution of 1988, the maximum number hours worked without the obligation of the payment of overtime was 48. In 1998 this limit changed to 44. As can be seen in Table 3-9, between 60 and 80 percent of formal sector workers work 44 hours or less. Compared to minimum wages, a rather large fraction works below the maximum ASSESSING THE IMPACT OF REGULATIONS ON INFORMAL WORKERS 9I TABLE 3-9 Proportion of Workers with Hours Worked Below or Equal to Maximum Hours Limit (percent) Metropolitan region All Legal Illegal Recife 60 60 59 Salvador 61 66 56 Belo Horizonte 63 65 61 Rio de Janeiro 66 67 66 Sao Paulo 67 71 64 Porto Alegre 73 80 66 Note: The legal limit is 44 hours. Source: PME (September 1998). legal limit. This may be due to the effect of overtime hours, which are not forbidden, but have to be remunerated differently. Since 1988, overtime has been paid at 1.5 times the worker's wage rate. Unfortunately, in the PME no information is available on whether or not overtime is paid. However, Table 3-9 shows that differences in the percentage of workers who work above the legal limit are strikingly small for Recife and Rio de Janeiro, but large for Porto Alegre and Sao Paulo. Using distributions of hours worked of legal and illegal employees on a yearly basis from 1982 to 1996, we find that the distributions of illegal employees are in most cases above those of legal employees. Except for the years 1985, 1989-91, and 1993, in terms of first-order stochastic domi- nance, illegal employees work more than legal employees. That is, the number of hours in any given percentile of the distribution of illegal employees is greater than the corresponding percentile in legal employ- ees hours distribution. The distribution of illegal employees hours also second-order stochastically dominates the corresponding distribution of legal employees in all years except 1982, 1988, and 1994. To analyze whether the maximum hours legislation distorts the hours distribution in the formal and informal sector, we evaluate the relative concentration of the mass of the distribution of hours effectively worked at the level of the parameters set by the law. Table 3-10 reveals that the main focal points found for legal employees hours distribution are also found for illegal employees distribution. There are clear plateaus at 40, 44, and 48 hours. The 40 hours focal point is not related to a legal limit, but rather corresponds to the weekly working week (jornada) of 40 hours. In 92 CRAFTING LABOR POLICY TABLE 3-10 Effective Workweek before and after New Constitution Legal employees Illegal employees Hours per week 1987 1990 1987 1990 0-30 9 11 9 9 30 5 5 4 6 30-40 5 7 4 5 40 30 31 21 22 40-44 1 1 1 1 44. 3 20 3 8 44-48 6 4 4 4 48 32 15 25 19 48-60 6 4 14 15 60 and more 3 2 15 11 All 100 100 100 100 Note: ANl Brazil. The legal maximum was lowered from 48 to 44 hours in 1988. Figures in bold are of particular interest. Source: PME (1987, 1990). 1987, the percentage of employees who worked 48 hours amounted to 32 percent in the formal sector, and only 7 percent less in the informal sec- tor. Consequently, the largest mass point is found at the legal maximum hours limit in both sectors, which hints at the fact that the hours legisla- tion distorts the hours distribution in the legal and the illegal sector. To understand whether changes in labor legislation affect the distribu- tion of hours worked in either sector, we exploit the change in the maxi- mum hours limit introduced by the Constitution in 1988. The reduction of the jornada de trabalho from 48 hours to 44 hours in the Constitution of 1988 is reflected in a decrease in the share of workers working 48 hours by 17 percent in the formal sector and 6 percent in the informal sector. Similarly, the share of workers working 44 hours increased from 1987 and 1990. This clearly reveals that changes in the legislation concerning the weekly working week did affect illegal employees, but to a lower extent than formal sector workers. Condusion In this chapter, we studied the effect of minimum wage and maximum working hours labor legislation on the wage and hours distribution of for- mal and informal wage earners. We found that labor legislation concern- ASSESSING THE IMPACT OF REGULATIONS ON INFORMAL WORKERS 93 ing minimum wages and maximum hours distorts the wage and hours dis- tribution not only in the formal labor market, but also in the informal labor market. This provides evidence that evading the costs for both the employer and employee of maintaining a legal contract, such as payroll taxes, social security contributions, and firing fines, does not necessarily imply that other labor legislation does not affect illegal employment. Table 3-11 summarizes our main results. That labor legislation affects the illegal labor market may be explained by the following reasons: First, concerning minimum wage, the argument of the "lighthouse effect" (efeito-farol) of regulations has been put for- ward, which claims that the official wage rules act as a "voluntary" refer- ence price to illegal employees. Second, the fact that inflation was moder- ate in 1998 indicates that the minimum wage is not only a signal in a high-inflation country, but appears to be an important benchmark for "fair" remuneration. A similar argument may apply to the number of work hours. Moreover, the high costs of social security for employees; the low TABLE 3-11 Brazil: Some Measures of Conformity with Labor and Social Security Laws Sector of employment Formal: Informal: with without Indicator signed card signed card Payroll taxes (% of workers whose firms ...) Paid INSS contributions 94.5 4.5 Paid FGTS contributions 95.0 5.0 Wage regulations (% of workers with ...) Paid exactly one minimum wage 7.0 14.0 Wage change = minimum wage increase March 1990 to January 1994 6.9 10.3 September 1994 to May 1995 12.0 21.5 Hours restrictions (% of workers) Workweek equal tojornada 1987 (before Constitution) 32.0 25.0 1990 (after Constitution) 20.0 8.0 Note: Following convention, formal employment implies having a carteira de trabalho assinada (working card or booklet signed by the employer), which signifies ratification of the contract by the Ministry of Labor. 94 CRAFTING LABOR POLICY link between contributions and benefits; and the low quality of many so- cial security benefits combined with the ambiguity in the design of labor legislation and the slanted nature of its enforcement by labor courts may increase the incentive to evade payment of social security contributions. Finally, Brazilian employees can take their employers to court-which has sweeping powers under current Brazilian law-to force them to pay for their legal working rights, whether or not their contract had been rat- ified by the Ministry of Labor. Given the high probability of the cases being resolved in favor of workers, employers accord these workers all the rights under the labor law even when they do not have legal contracts. The nature of enforcement of labor laws therefore endows informal sector workers "ex post legality," even though they are "ex ante illegal." Thus, all things being equal, the incentives to stay informal are higher for workers who are assured of protection under labor legislation regardless of the na- ture of their contract; labor legislation only alters their financial relation- ship with the government. Notes 1. A corollary of this, which is far more obvious, is that the current political economy of a country affects the possibility of successful labor reforms. 2. Recent research in Latin America (see, for example, Maloney 1998) has challenged the notion of the traditional dualistic view of the labor market, which considers the informal sector as a disadvantaged part of the labor market, which is segmented due to high labor costs in the formal sector. Some authors empha- size that given the high labor costs in the formal sector, some workers actually may choose to become informal. 3. This section draws on Amadeo and others (1993). 4. Note that the PNAD covers total labor population, while the PME is re- stricted to the six main metropolitan areas. 5. Turnover is relatively low in the formal service sector mainly because this sector includes workers employed in large public utilities and other state enter- prises. 6. Real labor income measures the income of all employed workers, including illegal (informal) wage earners and the self-employed. 7. This section draws upon Amadeo and others (1993) and Ministerio do Trabalho (1998). 8. Employee contributions for social security are based on a progressive scale and vary between 8 and 11 percent. 9. Withdrawals are possible for large health expenses or acquisition of a home. 10. PIS/PASEP rates are 0.65 percent of personnel costs of private sector firms and 1 percent of the wage bill of nonprofit establishments, but are 1 percent of the revenues of state enterprises. ASSESSING THE IMPACT OF REGULATIONS ON INFORMAL WORKERS 95 11. For example, under one part of the law a worker who worked for less than the full hours per week for a year is entitled to a proportional amount of paid va- cation and the mandatory Christmas bonus of a worker who worked full time all year, but is entided to fAll benefits under another part of the law. 12. Using administrative data from the CAGED, one researcher reports that in the early 1990s more than 1.5 million labor cases were disputed every year. Based on a sample for Minas Gerais, 96 percent of the disputes were for workers with one to five minimum monthly wages. Seventy percent of those disputes had less than two minimum wages. 13. That is, the maximum number of hours of work per week without the pay- ment of overtime wages. 14. A similar system exists in Peru. Note that this definition of informal sec- tor worker is more precise than the conventional definition of informality ini- tiated by the WTO, which classifies informal workers as nonprofessional self- employed workers in firms with less than five employees or domestic workers. 15. The sample they use consists of workers between 16 to 65 years of age who work between 30 and 50 hours a week. Informal salaried workers as defined as those working for firms with five employees or less, and formal salaried workers as those working for firms with more than five employees. References Amadeo, Edward, Ricardo Paes de Barros, Jose Marcio Camargo, and Rosane Silva Pinto de Mendon,a. 1993. "Institutions, the Labor Market and the Informal Sector in Brazil." Inter-American Development Bank, Washington, D.C. . 1994. "Institutions, the Labor Market and the Informal Sector in Brazil." Institute of Applied Economic Research and Department of Econom- ics, Catholic Pontifical University-Rio. Processed. Amadeo, Edward, Ricardo Paes de Barros, Jose Marcio Camargo, Rosane Silva Pinto de Mendon,a, Valeria Pero, and Andre Urani. 1993. "Human Resources in the Adjustment Process." Discussion Paper, No. 317. Rio de Janeiro: Insti- tute of Applied Economic Research. De Soto, Hernando. 1989. The Other Path. New York: Harper and Row. Maloney, William F. 1998. "Does Informality Imply Segmentation in Urban Labor Markets? Evidence from Sectoral Transitions in Mexico." World Bank Economic Review 13: 275-302. Maloney, William F., and Jairo Nufiez. 2001. "Measuring the Impact of Mini- mum Wages: Evidence from Latin America." Policy Research Working Paper No. 2597. World Bank, Washington D.C. Ministerio do Trabalho. 1998. Relatorio daforca-tarefa sobrepoliticas de emprego: diagnostico e recomendacoes. Governo Federal, Brasilia, August. PME (Pesquisa Mensual do Emprego). 1998. September. Brazil. PNAD (Pesquisa Nacional por Amostra de Domicilios). 1996. Brazil. PART Labor Supply I I and Labor Demand in Argentina CHAPTERS 4 AND 5 ANALYZE the determinants of labor demand and supply in Argentina using a quantitative approach. Chapter 4 takes up the question of variations in labor supply over the economic cycle. The approach employed in the chapter can be summa- rized as follows. First, the authors examine the basic patterns-trend (long-term) and cyclical (short-term)-in labor force participation of men and women. Then they determine the underlying cause of year-to-year fluctuations in labor force participation, distinguishing between the added worker and the discouraged worker effect. The added worker effect refers to an increase in labor force participation, when a demand-induced fall in employment arises, while the discouraged worker effect has the opposite implication. The likely explanations for these patterns are provided, con- trasting Argentina's experience with other countries. Next, the trends and structure of unemployment are analyzed to determine the extent to which increased labor force participation can account for the increased unem- ployment in recent years, and the extent to which reduced employment growth is responsible. Finally, the chapter discusses the policy implications of these findings, both in terms of what Argentina can expect over the longer term as benefits to labor reform, and the likely consequences of current labor supply patterns-for the sustainability of labor policies-if the proposed reforms are not implemented. Based on this analysis, Chapter 4 concludes that reforms to encourage efficient employment and wage-setting practices-while directly promot- ing sustainable growth of labor demand-are also likely to favorably in- fluence short-term labor supply patterns. In Argentina, high real wages 97 98 CRAFTING LABOR POLICY tend to be maintained during recessions, even though unemployment rates increase. An increase in unemployment rates leads to a drop in in- come for many households, which induces secondary workers, such as women, to enter the labor force. This results in. a strong added worker ef- fect, which may put even more pressure on unemployment. Hence, labor force participation of secondary workers in Argentina tends to be anti- cyclical, leading to exaggerated fluctuations in measured unemployment rates. Labor reforms that encourage wage flexibility will result in reduced cyclical volatility of unemployment rates, not just because employment fluctuations will be moderated by wage adjustments, but also because labor supply fluctuations will be less exaggerated. The discouraged worker effect will be strengthened, countering the significant added worker effect that leads to increased labor supply when the economy is weak. This would in turn allow the government to concentrate on the task of insti- tuting policies for raising long-term employment growth, instead of being distracted by temporary but sharp increases in unemployment, which are politically difficult to ignore. Chapter 5 provides quantitative estimates of the likely payoff-in terms of greater employment-to initiatives that improve labor market func- tioning. The analysis allows for a comparison between strategies that rely only on economic growth to reduce unemployment and those that com- bine these strategies with labor policy reform. Numerical estimates of em- ployment growth that would occur due solely to economic expansion, that is, without labor reforms which raise or lower labor costs, are approxi- mated by the output elasticity of aggregate employment, holding wages or unit labor costs constant. Similarly, the wage elasticity of employment can be used to quantify the effects of labor policy reforms on employment growth. Since the estimations require assumptions on issues such as the degree of substitutability between labor and capital in the production process, the chapter reports these results under a variety of assumptions. The estimated wage elasticity can reasonably be regarded as about -0.5, implying that a 10 percent fall in labor costs-other things remaining un- changed-will raise employment by 5 percent. Estimated output elastici- ties range between 0.1 and 0.4. The average estimate is about 0.25, im- plying that a 10 percent increase in output will result in a 2.5 percent increase in employment. With the output elasticity estimate being roughly half the absolute magnitude of the wage elasticity, the main implication is that Argentina cannot rely solely on economic growth to lower the high unemployment rates it has experienced since 1995, but must undertake labor reforms. LABOR SUPPLY AND LABOR DEMAND IN ARGENTINA 99 Argentina's monetary and exchange rate policies have resulted in low in- flation rates, but rhey also limit the policy instruments the government has at its disposal. Fiscal discipline and labor market deregulation are perhaps the two most important policy priorities. While comprehensive labor re- form is politically difficult to implement since it involves confronting pow- erful labor unions, Argentina did lower labor taxes in the mid-1990s and introduced legislation allowing for special labor contracts (modalidades promovidas). These contracts allowed the hiring of certain categories of workers without paying labor taxes or severance payments, and under longer probationary periods. When these laws were simplified, employers were quick to take advantage of the new law, and by 1998 more than 12 percent of formal employees had been hired under these more flexible terms. Employment growth accelerated to more than 3.5 percent annually between 1995 and 1998, and unemployment fell from 20 percent to about 12.5 percent. But in mid-1998, under political pressure during the run-up to elections in 1999, the government essentially revoked the special con- tracts law, reduced probationary periods, and made the centralized collec- tive bargaining laws even more centralized. By 1999, unemployment rates had begun to climb again, and were around 15 percent in 2000. Under pressure to tighten fiscal discipline, the Argentine government increased some labor taxes in 2001. After essentially reversing by the end of the decade the progress made in the mid-1990s, Argentina made a discourag- ing start with respect to labor policy reform in the new millennium. Three years of economic stagnation have resulted in unemployment rates rising to about 20 percent again. In late 2001, Argentina defaulted on its public debt, triggering a political crisis. 4 Understanding Labor Supply Dynamics in Argentina CAROLA PESSINO, INDERMIT S. GILL, AND JosE Luis GUASCH THE EVOLUTION OF A LABOR FORCE is influenced either by changes in the population size or changes in labor force participation (LFP). Changes in population size are determined by the rates of natural increase and net migration. These patterns-especially the rate of natural increase-are relatively slow to change. LFP may exhibit either long-term of short-term changes. Long-term tendencies in the change of LFP, such as increased participation of women, are due to sociological and other fac- tors, which are also slow to change. Speaking of "trend" labor force growth based on growth of the population base and these long-term changes in LFP is not unreasonable. But the size of the labor force also responds to shorter-term (or cyclical) changes, which operate largely through rapid- though often predictable-changes in the LFP rate. These changes have received considerable attention in recent labor policy debates in Argen- tina, and are the main concern of this chapter. In Argentina, unemployment rates have risen sharply since 1990, reaching about 18 percent in 1995. Since then they have remained at about 15 percent. Recent labor policy discussions about the origins of the high unemployment in Argentina have involved decomposition of changes in unemployment into a component related to labor demand (or an un- expected slowdown of employment growth) and another related to labor supply (or unexpected increases in LFP, especially of women). In joining this debate, we contend that the definition of "unexpected" fluctuations in IOI I02 CRAFTING LABOR POLICY labor supply employed so far ignores a fundamental attribute of the Ar- gentine labor market, that is, that LFP is positively correlated with the un- employment rate over the business cycle. The main contribution of this paper is to explain shorter-term LFP patterns in Argentina using estab- lished theory and econometric techniques, to re-examine the decomposi- tion of unemployment using these results, and to draw policy implications from these exercises. In Argentina, the change in female LFP explains most of the change in total LFP from 1974 to 1995. Female participation rates rose by 45 per- cent from 1974 to 1995, while male participation rates were the same in 1995 as in 1974. In this chapter, we examine variations in female LFP using data up to October 1995. We distinguish trend from cyclical be- havior and illustrate, through econometric techniques, the correlation be- tween LFP and unemployment rates. Based on this analysis we determine the extent to which higher LFP-or increased labor supply-can legiti- mately be blamed for the rapid increase in unemployment since 1990. The approach employed in this chapter can be summarized as follows: First, we examine the basic patterns-trend (long term) and cyclical (short term)-in LFP of men and women. Second, we determine the underlying cause of year-to-year fluctuations in LFP, decomposing the effects of demand-induced, reduction of employment on labor supply in "added worker" and "discouraged worker" effects: The added worker effect im- plies that LFP of secondary workers increases as unemployment rises, while the discouraged worker effect leads to the opposite conclusion. We provide likely explanations for these patterns, contrasting Argentina's ex- perience with that of other countries. Third, we examine the trends and structure of unemployment. This enables us to determine the extent to which increased LFP can account for the increased unemployment in recent years, and the extent to which reduced employment growth is re- sponsible. Finally, we discuss the policy implications of these findings. Trends in Participation and Unemployment Population, Labor Force, and Employment Before turning to the discussion about changes in LFP, we provide some evidence on the evolution of total population and employment. As can be seen in Table 4-1, total population in all 25 urban conglomerates in Ar- gentina increased from 23 million in 1980 to 30.7 million in 1995. Fol- lowing the increase in total population, the economically active popula- UNDERSTANDING LABOR SUPPLY DYNAMICS IN ARGENTINA 103 TABLE 4-1 Argentina: Labor Force, Employment, and Unemployment, 1980-95 Total Economically Unemployment Year population active pop. Employed Unemployed rate (%) 1980 23.00 8.73 8.48 0.25 2.81 1981 23.50 8.89 8.45 0.44 4.95 1982 24.00 9.10 8.59 0.51 5.60 1983 24.45 9.12 8.64 0.47 5.21 1984 24.90 9.32 8.84 0.47 5.10 1985 25.45 9.57 8.95 0.63 6.53 1986 26.95 9.89 9.29 0.60 6.12 1987 26.40 10.19 9.56 0.63 6.19 1988 26.90 10.41 9.73 0.68 6.54 1989 27.50 10.68 9.83 0.85 7.96 1990 28.00 10.72 9.92 0.79 7.42 1991 28.60 11.10 10.35 0.75 6.76 1992 29.10 11.44 10.63 0.82 7.12 1993 29.75 11.88 10.77 1.11 9.35 1994 30.35 12.08 10.73 1.34 11.14 1995 30.70 12.64 10.43 2.21 17.48 Note: All urban areas. Numbers are averages of April/May and October; estimates for 1995 are for May only. Source: "Secretaria de Programacion Economica," printed in Informe Econdmico, First Quarter, 1995. tion increased from 8.73 million in 1980 to 12.64 million in 1995. Em- ployment increased steadily from 1980 until 1993, ranging from 8.48 million to 10.77 million, and declined to 10.43 million in 1995. How- ever, employment rates decreased from 97.14 percent in 1980 to 82.52 percent in 1995. Unemployment increased steadily from 250,000 in 1980 to 850,000 in 1989, followed by a decrease until 1991. In 1995, the re- cession induced by the so-called Tequila Crisis reached its peak, which is reflected in a drastic increase in the number of unemployed. From 1993 to 1995, it nearly doubled, climbing to 2.21 million in 1995, which cor- responds to an unemployment rate of 17.48 percent. Between 1980 and 1991 Argentina's population grew at an annual rate of 1.4 percent. This growth in population-which is high compared to Europe (0.3 percent) and North America (1 percent)-is mostly due to the rate of natural increase. While the stock of migrants is important, 104 CRAFTING LABOR POLICY changes in the number of migrants are not large enough to significantly affect overall labor supply as well as employment, unemployment, or wages. Moreover, the number of legal immigrants from neighboring coun- tries has increased in absolute numbers, but not as a proportion of the population. ' Employment growth in Argentina has fluctuated considerably since 1974. According to Ministry of Economy household data for Greater Buenos Aires, employment grew at an annual rate of 1.1 percent between May 1974 and October 1995. Compared to other countries, GBA fared moderately well in terms of employment growth. Employment growth has been strongest in North America (1.7 percent annually since 1974) and weakest in the European Community (EC) (0.2 percent annually) and the European Free Trade Association (EFTA) (0.2 percent), with Argentina (1.1 percent), Oceania (1.2 percent), and Japan (1.1 percent) falling in the middle. (For a more detailed discussion see Pessino 1997.) However, when population growth is taken into account, Argentina fares the worst, with an annualized rate of growth in population of 1.3 percent (in GBA). Em- ployment growth minus population growth averages a 0.16 percent de- crease per year. In this respect, Argentina is much closer to the EC and EFTA. Using this index, North America and Japan fared the best, gener- ating the highest employment growth relative to population growth. Fur- thermore, in Argentina there was an increase in the overall LFP rate, im- plying that LFP growth was higher than population growth. The labor force grew at a 1.95 percent annualized rate during the period, which compares to an annual population growth of 1.3 percent. We will show that this growth was largely due to the increase in female LFP. Labor Force Participation LFP rates-the number of people .working or seeking work as a share of population aged 15 to 64-in Greater Buenos Aires rose by more than 10 percentage points (from 58.1 to 68.5 percent) between 1974 and 1995, as can be seen in Table .4-2.2 This increase was accounted for entirely by in- creased female LFP rates: female participation rates rose by 45 percent over 1974 levels (more than 16 percentage points), while male participation rates were about the same in 1995 as they were in 1974. Furthermore, while male LFP rates fluctuated only slightly around the 84 percent level during the two decades preceding 1995, changes in female LFP rates were much more pronounced. Female participation rates increased very rapidly from October 1991, peaked in May 1993 at 51 percent, then decreased by UNDERSTANDING LABOR SUPPLY DYNAMICS IN ARGENTINA 105 TABLE 4-2 Labor Force Participation, by Gender (percent) Year All Female Male 1974 58.1 36.6 85.3 1975 58.4 36.3 85.9 1976 58.3 35.7 86.2 1977 57.9 35.8 85.1 1978 .57.8 37.8 83.8 1979 58.1 37.3 84.3 1980 60.6 38.1 84.8 1981 60.1 38.5 83.6 1982 60.0 38.5 84.5 1983 58.6 36.6 83.0 1984 59.8 38.3 83.5 1985 60.6 39.6 83.6 1986 62.0 42.8 84.4 1987 62.6 43.4 84.3 1988 62.8 43.8 83.8 1989 64.0 45.5 84.3 1990 63.8 45.0 84.1 1991 63.2 44.4 83.2 1992 64.4 45.9 84.5 1993 66.5 49.7 84.8 1994 65.9 48.8 84.2 1995 68.5 53.0 85.1 Note-. For Greater Buenos Aires, people 15 to 64 years of age. two percentage points, peaked again in May 1995 at 53 percent, and de- clined again in October 1995. Participation rose relatively steadily between 1983 and 1989 (by between 1 and 2 percentage points per year), but has exhibited sharper fluctuations since 1990. From 1992 to 1993, and again from 1994 to 1995, female LFP rates rose by about 4 percentage points within a single year. As a consequence, understanding the trend and changes in LFP rates in Argentina requires understanding the trend and changes in female LFP. Not only aggregate and female LFP rates exhibited an upward trend from 1974 to 1995. Pessino (1997) shows that participation rates of teen- agers and older people increased from 1987 to 1995, as well. Using mi- croeconomic household data on male wage earners and the self-employed io6 CRAFTING LABOR POLICY aged 25 to 54 for Greater Buenos Aires, Pessino furthermore finds that ag- gregate real monthly. earnings dropped drastically from 1986 to 1990. Real wages behaved similarly. They increased substantially between 1991 and 1993, before starting to decrease again. Hence, from 1991 to 1993 aggregate participation rates and participation rates for women, teenagers, and older people increased. The change in labor supply due to change in the price of labor can be decomposed into two effects: the substitution effect and the income effect. Classical theory predicts that the substitution effect is positive, which means that the family member that receives higher wages will work more hours. The sign of the cross-substitution effect (for example, the effect of wages of husbands on the labor supply of their wives) is not determined by theory. The same holds for the income effect. If wages of a member of the household decrease, family income decreases. It is often assumed that a fall in family income induces members to work more. Given that the in- come effect is positive and dominates the cross-substitution effect,3 a de- crease in a husband's wages will increase the labor supply of his wife. With the exception of 1990 and 1991, female LFP increased during a time of rising unemployment rates. According to Lundberg (1985), the re- sponse of LFP rates to a demand-induced fall in employment consists of two components-the added worker effect and the discouraged worker ef- fect. The added worker effect usually refers to a temporary increase in the labor supply of married women whose husbands have become unem- ployed or face restriction in their work hours. The discouraged worker ef- fect refers to workers who drop out of the labor force or refrain from en- tering it, because they anticipate that wages will be lower or job search costs higher. The finding that female LFP increases with unemployment rates suggests that in Argentina the added worker effect dominates the dis- couraged worker effect. Compared to OECD countries, female LFP rates are quite low in Ar- gentina. In OECD countries, female LFP rates averaged 48 percent in 1973, and 60 percent in 1990.4 Hence, female LFP in Argentina during this period was smaller than in OECD countries.5 The increase in female LFP in Argentina is very similar to that of North America and Oceania, despite much higher population growth in Argentina. Other countries have an even higher growth trend in female LFP. In some sense, therefore, Argentina is lucky to have slower growth in LFP, since it would have gen- erated a much higher rate of unemployment, if one assumes that LFP is inelastic with respect to wages and income and that relative wages do not adjust to the increased labor supply of women. The difference between UNDERSTANDING LABOR SUPPLY DYNAMICS IN ARGENTINA 107 employment and labor force growth is negative in all OECD countries, leading to growing unemployment rates. The countries that fare best (in terms of low unemployment growth) are the United States and Japan. Ar- gentina faced a 0.8 percent shortfall per year in employment relative to labor force growth. High population growth combined with relatively low LFP growth compared to other countries hints at the fact that there exists a considerable reserve of potential workers in Argentina. This may be quite responsive to economic stimuli (for example, increasing wages and declining household wealth that accompany business cycles). The consid- erable year-to-year fluctuations exhibited by female LFP rates since 1990 are consistent with this observation. Summarizing, there are three noteworthy observations regarding female LFP rates. First, they remained largely unchanged from 1974 to 1983. The 45 percent rise in LFP rates between 1974 and 1995 is thus a phenome- non of the subsequent decade. Second, female LFP rates exhibit consider- able fluctuations. Third, compared with OECD countries, Argentina faces low female LFP rates and low LFP growth relative to population growth. Unemployment In Greater Buenos Aires (for which the most reliable and detailed data are available) the unemployment rate quadrupled from 5.1 percent in October 1987 to 20.2 percent in May 1995, as can be seen in Table 4-3. The rapid increase in the unemployment rate from October 1987 until May 1995 was similar for men and women. Table 4-3 illustrates that male and female unemployment rates increased by approximately 290 percent. Looking at worker groups by age, it can be seen that workers who were aged 35 to 49 faced a somewhat greater increase in unemployment rates than younger workers. This is in line with the fact that growth in the unemployment rate was lower for single workers compared to married workers. White-collar workers faced a higher growth in unemployment rates compared to blue-collar worker. Note, however, that all the groups that faced low unemployment growth had a relatively low level of unemploy- ment rate in October 1987. Differences among sectors and education in unemployment growth during this period are very small. Comparing increases in unemployment rates for two subperiods-October 1987 to October 1994, and October 1994 to May 1995-and for the period as a whole-October 1987 to May 1995-it can be seen that the rise in unem- ployment rates was even more uniform than during 1987-95 as a whole. Thus, while there were differences in the rate of increase of unemployment io8 CRAFTING LABOR POLICY TABLE 4-3 Unemployment Rates, Inflows, and Duration, October 1987 and May 1995 Ratea Inflowsb Duration' Oct. May Oct. May Oct. May 1987 1995 1987 1995 1987 1995 All 5.1 20.2 1.5 4.2 5.6 8.6 By Gender: Male 4.4 17.1 1.3 4.2 5.2 7.2 Female 6.3 24.8 1.8 4.1 5.8 10.0 By Age: Male 20-34 years 4.2 15.3 1.1 3.9 5.0 7.2 Male 35-49 years 2.9 11.9 1.0 3.4 4.8 6.2 Female 20-34 years 7.3 25.2 1.8 4.1 6.2 10.0 Female 35-49 years 4.2 19.0 1.5 2.9 4.4 11.0 By Sector: Manufacturing 5.5 19.0 2.1 3.5 4.0 9.6 Services 4.5 16.7 1.3 3.4 6.0 8.4 Construction 11.0 36.2 4.0 12.7 4.4 4.4 By Occupation: White collar 3.8 16.8 1.3 4.0 5.2 7.8 Blue collar 7.1 24.3 2.2 5.1 5.2 8.6 By Education: Primary incomplete 7.6 23.4 2.3 6.4 5.6 6.8 Tertiary complete 2.2 6.7 0;6 1.2 6.2 11.6 By Marital Status: Male single 8.6 28.6 2.2 5.9 5.4 8.2 Male married 2.8 11.6 0.9 3.3 5.2 6.4 Female single 8.4 30.0 2.0 5.0 6.2 10.0 Female married 5.3 22.6 1.8 3.8 5.4 10.2 Note: For Greater Buenos Aires. The groups in each category are not always exhaustive, and are chosen for illustrative purposes. a. Unemployment rates are percentage of labor force. b. Inflows are percent per month. c. Unemployment duration is in months, computed by multiplying the duration of in- complete spells by two. UNDERSTANDING LABOR SUPPLY DYNAMICS IN ARGENTINA IO9 across age, sector, occupation, education, and marital status, these numbers are more striking for their similarity than their differences. Thus, while these groups have quite different unemployment rates-for example, fe- male, younger, less-educated, blue-collar workers have higher unemploy- ment rates-this structure remained relatively unchanged before and after the reforms. Unemployment rates are the product of inflows into unemployment (that is, the fraction of workers that become unemployed in any month) and the duration of unemployment spells. Thus, for example, in 1995 construction workers had high inflow rates (about 12.7 percent compared with 3.5 for services and manufacturing), but relatively short duration of unemployment (about 4.5 months, and about 9 months for manufactur- ing and services). These numbers translate into an unemployment rate of 36 percent for construction, roughly twice the rate for manufacturing and services. The average inflow into unemployment in Greater Buenos Aires in- creased from 1.5 percent per month to 2.8 percent between 1987 and Oc- tober 1994, and rose to 4.2 percent in May 1995. This implies an overall increase of about 180 percent. There was more variation in inflow rates than in unemployment rates across categories of workers. For example, the percent change in inflows from October 1987 to May 1995 with respect to 1987 was 1.8 times higher for men than for women. Similarly, the per- cent change in the inflow rates of construction workers was more than 3 times higher compared to manufacturing, while the percent change of white-collar workers was 1.5 higher compared to blue collar workers. The average duration of unemployment spells increased less dramati- cally-from 5.6 months in October 1987 to 8.6 months in October 1994-which corresponds to an increase of 54 percent. It then remained unchanged. With the exception of older women and manufacturing work- ers, the increase in unemployment rates in 1995 was driven largely by in- creased inflows into unemployment, and not by longer unemployment spells. Since relative unemployment rates (for different worker groups) re- mained quite stable between 1987 and 1995, but relative inflows changed considerably, it stands to reason that unemployment duration of different workers would also have changed significantly as well. In fact, unemploy- ment rates of workers with different attributes tend to move together over the reference period. Another question that concerns policymakers is whether the unem- ployment increase that took place between October 1994 and May 1995 was an aberration, perhaps due to an unfortunate combination of an ex- 110 CRAFTING LABOR POLICY ternal shock (the Tequila Effect) and increased LFP, which was encouraged by the economy's strong performance between 1992 and 1994. Table 4-4 provides evidence of the composition of the unemployed according to their labor force status, age, and gender in May 1995. Of the 1 million unemployed in May 1995, 21 percent were reentrants, 14 percent were new entrants, and 65 percent were previously employed individuals. Among the latter group, 61 percent were men, while women dominated the groups of unemployed who entered or reentered the labor market (70 percent). Furthermore, unemployed new entrants, as expected, were mostly young. In contrast, males between 20 and 49 years of age formed the bulk of the unemployed with previous employment. While these find- ings are not conclusive evidence that increased LFP of women is not re- sponsible for increased unemployment, they do provide reason to be skep- tical of such claims. Using standard econometric techniques (probit analysis) to determine the characteristics that are associated with unemployment rates, we find mainly that being male, having more education, not being a new entrant into the labor force, working in larger firms, and working in nontradeable (service) sectors other than construction significandy lowers the probabil- ity of being unemployed. But being educated is no guarantee against un- employment: older, educated workers are more likely to be unemployed than older, less-educated workers. There also exists some evidence that having other sources of income allows workers to stay unemployed longer. Comparing probit analysis results for May of each year between 1992 and 1995, we find considerable stability of these relationships.6 Determinants of Labor Force Participation Methodology Unemployment can increase either because fewer jobs are available or be- cause more workers decide to look for jobs. Whether the rise in unem- ployment is largely demand induced or is the result of increased LFP has very different welfare implications for workers. This implies that under- standing the relation between the dramatic changes in female LFP in Argentina and the high level of unemployment has important policy im- plications. As explained, the response of LFP rates to a demand-induced fall in employment can be decomposed into an added worker effect and a discouraged worker effect. Added workers are, for example, women or school-aged children who increase their labor supply as the family head be- comes unemployed or faces restriction on his or her work hours. Discour- UNDERSTANDING LABOR SUPPLY DYNAMICS IN ARGENTINA III TABLE 4-4 Distribution and Duration of Unemployment, by Age, Gender, and Previous Labor Market Status, May 1995 Reentrants New Previously to laborforce entrants employed All Distribution of the unemployed (percent) Males 15-19 years 17.0 77.1 19.4 - 20-34 years 33.8 22.9 33.9 - 35-49 years 24.7 0.0 26.3 - 50-64 years 24.5 0.0 20.4 - Females 15-19 years 9.9 54.2 15.5 - 20-34 years 43.9 33.8 41.8 35-49 years 30.4 9.0 30.3 50-64 years 15.8 3.0 12.4 All 15-19 years 12.0 61.4 17.9 20-34 years 40.9 30.4 37.0 35-49 years 28.6 6.2 27.8 50-64 years 18.5 2.1 17.3 Males 30.1 31.1 61.0 Females 69.9 68.9 39.0 Duration of unemployment (months) Males 15-19 years 5.4 7.3 4.2 5.2 20-34 years 5.0 10.9 3.9 4.5 35-49 years 5.0 - 4.0 4.1 50-64 years 3.2 - 8.8 7.9 All 4.6 8.1 5.0 5.2 Females 15-19 years 3.6 6.5 4.7 5.4 20-34 years 5.1 9.8 6.1 6.4 35-49 years 8.3 8.2 7.7 7.9 50-64 years 8.9 12.4 7.3 8.2 All 6.5 7.9 6.5 6.8 All 6.0 8.0 5.6 6.0 - Not available. Source: INDEC-EPH. II2 CRAFTING LABOR POLICY aged workers may be women who refrain from entering the labor force be- cause their cost of looking for acceptable employment rises to such a de- gree, during a recession, that it no longer pays for them to make the effort. To address the question of how LFP rates respond to changes in un- employment rates, we assume that participation rates are functions of the cycle and a time trend. The time trend is included to reflect the impact of slowly changing social factors and other gradually moving variables omit- ted from the equation. The cycle may affect each demographic group in a different way. As a cycle measure we use the unemployment rate of prime- aged male workers, which is expected to influence the level of participa- tion since the costs of search are affected by the availability of jobs.7 Using male unemployment rates also allows us to test whether in Argentina the added worker effect or the discouraged worker effect applied during the recent years. We also analyze the determinants of female LFP decisions. Following the methodology used by Pencavel (1986), Killingsworth and Heckman (1986), Clark and Summers (1981), and Pessino and Gi- acchino (1994), we estimate variants of the following regression of LFP on trend (t) and cycle (UP) for Greater Buenos Aires during 1974-95 for eight gender-age groups. Li= a +Pup+yt+ -it (1) L; is the LFP of group i in year t, expressed as a percentage of total pop- ulation in group i, Up, is the unemployment rate of males aged 35 to 49 in year t. The superscript "P" on U designates this as the prime-aged ref- erence group. The responsiveness of the participation rate to the business cycle is measured by I, while y is the coefficient for the linear time trend. Note that P may vary across groups. The equation error is represented by et and the index i runs over eight gender-age groups. We assume that the errors follow an (AR1) process and present the results of estimating those equations by maximum likelihood techniques in Table 45 .8 Of course, the equations that we are estimating do not provide a de- tailed explanation of LFP of each of the groups. Rather, our purpose is to estimate a model that captures the relation between LFP and cyclical fluc- tuations in the unemployment rate. A positive coefficient on UP tells us that LFP of the respective group moves in the same direction as UP, and hence is evidence of the predominance of the added worker effect. If we find that a higher rate of unemployment of male workers is associated with lower LFP of this group, that is, that the coefficient on UP is nega- tive, we have evidence of a discouraged worker effect. UNDERSTANDING LABOR SUPPLY DYNAMICS IN ARGENTINA II3 TABLE 4-5 Labor Force Participation: Cydical and Trend Factors Specification (1), Females and Males a , 3 8 Constant Cycle Trend Autocorrelation Females 15-19 years 34.445a 1.257a -0.407a 0.372a (0.995) (0.272) (0.061) (0.145) 20-34 years 46.947a 1.067a 0.117a 0.365a (0.950) (0.262) (0.058) (0.145) 35-49 years 33.548a 0.745a 0.441a 0.517a (1.146) (0.251) (0.064) (0.134) 50-64 years 18.050a 0 544b 0.341a 0.430a (1.165) (0.294) (0.069) (0.141) All females 33.892a 0.821a 0.226a 0.615a (0.943) (0.170) (0.049) (0.123) Males 15-19 years 49 343a 0.453 -0.220a 0.508a (1.452) (0.324) (0.081) (0.135) 20-34 years 95.580a 0.310a -0.064a 0.044 (0.289) (0.109) (0.021) (0.156) 35-49 years 97.097a 0.154a -0.028 0.460a (0.314) (0.076) (0.018) (0.139) 50-64 years 75.8572 0.407 0.039 0.4412 (1.176) (0.292) (0.069) (0.140) All males 85.001a 0.274a -0.068a 0.290b (0.365) (0.110) (0.023) (0.149) Note: Biannual data from 1974/1-95/1. April 1986 is missing. Data are usually mea- sured in April/May and September/October. The number of observations is 42. The esti- mation method is the maximum likelihood method correcting for autocorrelation. Stan- dard errors are in parentheses. a. Significant at 5 percent level. b. Significant at 10 percent level. Source: INDEC-EPH. 114 CRAFTING LABOR POLICY To check the robustness of our results, we run three further regressions, which are variants of the standard specification described in equation (1). The other three regressions use the following specifications: * Trend, UP and industrial production index (IPI). Lit = °i + i Ut + Yt + 6log(IPI), + Fi, (2) * Trend, UP, period dummy for 1991-95 and period dummy inter- acted with cycle measure. Lit = °a; + i UP + yt + D*Dummy9l+O UP *Dummy9l + F-t (3) * Trend only. Lit = a + yt t + i(4) IPI refers to the index of industrial production and is introduced as a direct measure of the cycle. 5 is the coefficient on IPI. Dummy9l is a dummy which assumes value I if the period is 1991 or after, and equals 0 otherwise. D and f are the coefficients on Dummy91 and the Up inter- acted with Dummy9l, respectively. We will refer to the period following 1991 as the postreform period. Specification (3) allows us to test whether the added worker effect is, rather, a phenomenon of the period following 1991. The estimation results of specifications (1), (2), and (3) are pre- sented in Tables 4-5, 4-6, and 4-7. Specification (4) is used in the analy- sis of the predictive power of the models, which we will present below. Trend and Cyclical Effects As can be seen in Table 4-5, the coefficient on trend is negative for all males, which indicates that over the past two decades the participation rate of males decreased. This decline occurred mainly because of the de- crease of the LFP rate of teenage males of about 0.44 percentage points per year, and a decline in the participation rate for males aged 20 to 34 by 0.13 percentage points. For males, the estimates of the cycle coeffi- cient 3 show positive values for all age groups; however, only the positive values for the groups aged 20 to 34 and 35 to 49 are statistically signifi- cant at the 5 percent level. The value of , equal to 0.31 for men aged 20 to 34 indicates that a 1 percentage point decrease in the unemployment rate of prime-aged men increases the participation rate of men aged 20 to 34 by 0.31 percentage point. The response of males aged 35 to 49 is lower, but positive. Hence, we can conclude that the higher the unem- UNDERSTANDING LABOR SUPPLY DYNAMICS IN ARGENTINA 115 TABLE 4-6 Labor Force Participation: Cyclical and Trend Factors Specification (2), Females 1 8 p aX Prime-aged y Industrial Autoregressive Age Constant unemployment Trend production correction 15-19 88.328a -0.117 - 12.180a 0.701a (16.750) (0.085) (3.687) (0.111) 45.702a 1.185a -0.385a -2.510 0.429a (17.100) (0.299) (0.072) (3.802) (0.141) 20-34 61.562a 0.313a -3.345 0.5 loa (17.730) (0.058) (3.924) (0.134) 23.903 1.171a 0.085 5.108 0.250b (15.820) (0.267) (0.059) (3.517) (0.151) 35-49 28.417b 0.569a 1.040 0.486a (15.710) (0.049) (3.479) (0.136) 7.235 0.932a 0.389a 5.859b 0.482a (15.130) (0.268) (0.067) (3.366) (0.137) 50-64 -2.705 0.420a 4.543 0.478a (17.080) (0.052) (3.781) (0-137) -23.425 0.841a 0.258a 9.241a 0.348a (17.060) (0.293) (0.068) (3.794) (0.146) All Females 42.896a 0.389a -2,112 0.705a (11.330) (0.058) (2.494) (0.111) 14.443 0.986a 0.183a 4.328b 0.55la (10.220) (0.184) (0.049) (2.272) (0.130) Note: Biannual data from 1974/1-95/1. April 1986 is missing. Data are usually rnea- sured in April/May and September/October. The number of observations is 42. The esti- mation method is maximum likelihood method correcting for autocorrelarion. Standard errors are in parentheses. a. Significant at 5 percent level. b. Significant at 10 percent level. Source: INDEC-EPH. ployment rate for prime-aged males, the higher the participation rate for these groups. As we have seen, changes in female LFP dominate changes in aggregate LFP. Consequently, we gain more insight into the participation behavior by looking at female participation rates. Contrary to men, the trend in fe- ii6 CRAFTING LABOR POLICY TABLE 4-7 Labor Force Participation: Cyclical and Trend Factors Specification (3), Females Age a yOa Db p 5-19 35.331c 0.724c -0.405c 0.534c 0.158 (0.821) (0.329) (0.534) (0.220) (0.154) 35.312c 0.611 -0.388c 0.701 -1.043 0.154 (0.828) (0.445) (0.069) (0.497) (2.812) (0.154) 20-34 47.572c 0.761c 0. lod 0.357 0.388c (1.065) (0.342) (0.059) (0.258) (0.144) 47.442c 0.370 0.175c 0.965d -4.122 0.382c (1.045) (0.437) (0.074) (0.502) (2.922) (0.144) 35 -49 33.861c 0583d 0.436c 0.197 0.491c (1.194) (0.333) (0.063) (0.268) (0.136) 33.862c 0.586 0.436c 0.192 0.0350 0.491c (1.212) (0.428) (0.078) (0.501) (2.964) (0.136) 50-64 18.929c 0:125 0.326c 0.516d 0.395c (1.177) (0.375) (0.065) (0.284) (0.143) 18.747c -0.456 0.424c 1.390c - 6043d 0.467c (1.273) (0.467) (0.084) (0.544) (3.212) (0.138) All Females 34.357c 0.610c 0.215c 0.283 0.579c (0.924) (0.221) (0.046) (0.189) (0.127) 34.295c 0.305 0.270c 0.800c -4.010c 0.663c (1.059) (0.258) (0.058) (0.314) (1.906) (0.117) Note: Biannual data from 1974-95. April 1986 is missing. Data are usually measured in April/May and September/October. The number of observations is 42. The estimation method is maximum likelihood method correcting for autocorrelation. Standard errors are in parentheses. a. 0 is the interaction of Dummy91 and prime-aged unemployment. b. D is the coefficient of Dummy91. Dummy9l equals 1 if the period is 1991 or after, and equals zero between 1974 and 1990. c. Significant at 5 percent level. d. Significant at 10 percent level. Source: INDEC-EPH. male LFP was positive from 1974 to 1995. Except for teenagers, all age categories show a strong positive trend during this period. Prime-aged women (aged 35 to 49) increase their LFP by 0.9 percent a year, followed by older women (0.7 percent a year), and those aged 20 to 34 (0.2 percent a year). UNDERSTANDING LABOR SUPPLY DYNAMICS IN ARGENTINA 117 Female teenage LFP decreases by 0.8 percent per year. Table 4-5 further suggests that the aggregate response of female LFP to changes in UP amounts to 0.82, and that female participation rates are strongly counter- cyclical, at least if one measures the cycle by the unemployment rate of prime-aged males. Notice that the younger the group, the higher the re- sponse to the cycle. For female teenagers an increase by 1 percentage point in UP raises their participation by about 1.26 percentage points. These re- sults show that young individuals and females exhibit substantial cycle sensitivity. According to these estimates, over the past two decades, there has been a declining trend in the participation rate for all males, and an increasing trend in the participation rate for all females. With respect to males, the trend decline occurred mainly because of the decrease in male teenage par- ticipation of about 0.4 percentage points per year, and a decline in the par- ticipation rate for males aged 20 to 34 by 0.15 percentage points. The es- timates of P (the cycle coefficient) for males show positive values for all age groups; however, only the positive values for the groups aged 20 to 34 and 35 to 49 are statistically significant at the 5 percent level. This means that the behavior of these groups is countercyclical: the higher the unem- ployment rate for prime-aged males, the higher the participation rate for these groups.9 The estimation results of specification (2), which introduces the IPI as a direct measure of the business cycle, are presented in Table 4-6. When UP is excluded from the regression, the coefficient on log(IPI) becomes negative when it is significant. That is, using another measure of the cycle does not change our conclusion about the countercyclical behavior. Evi- dence becomes less clear when we include both UP and log(IPI). Teenagers respond countercyclically to both. Prime-aged, older females, and the en- tire sample of females continue to respond countercyclically to UP, but procyclically to log(IPI). The years of 1991-95 is the period of structural reform. It was charac- terized by a boom in terms of GDP and industrial production growth, while at the same time unemployment increased. Consequently, we expect differences in the structural parameters of the regressions. Regressing fe- male LFP on unemployment of prime-aged male workers (UP), a time trend, Dummy9l and the interaction of Dummy9l with UP, we find that, except for women aged 35 to 49, there is a significant effect of this inter- action. As can be seen in Table 4-7 the added worker effect grew stronger since 1991 for the less stable part of the female labor force. Table 4-5 shows that for teenagers, for example, a 1 percent increase in UP increases their LFP rate by 1.2 percentage points. Now we can show that until II8 CRAFTING LABOR POLICY 1990, a 1 percent increase in UP augments their participation by only 0.7 percentage points. The remaining 0.5 percentage point was added after 1991.10 Contrasting these results with similar estimations for the United States (see, for example, Pencavel 1986; and Killingsworth and Heckman 1986), we firid that our results differ mainly in that, while we find countercyclical LFP for both men and women, U.S. data indicate no cyclical effects for men, but procyclical LFP for some of the female cohorts. That is, it appears that in the United Kingdom and the United States in the analyzed period there is a prevalence of discouraged female workers during a recession, while in Argentina the phenomenon of added female workers prevails. We should not be surprised by the difference in cyclical patterns in LFP (especially of females) between Argentina and more developed countries, such'as the United Kingdom and the United States. First, females, on av- erage, have lower LFP rates in Argentina than in those countries, so there is a larger pool that can enter the labor market when facing higher unem- ployment rates for males. Second, Argentina does not yet have a widely available capital market to which families can turn to maintain consump- tion during a recession. Hence, it is not surprising that they turn to sec- ondary earners in the household-women and non-prime-aged men-to support the family if the primary earner becomes unemployed." Since the economic reforms of the Convertibility Plan in 1991,12 smooth wage ad- justments have been hampered by individual and collective bargaining laws and practices that evolved during a period of high inflation. Strong unions and centralized collective bargaining agreements hindered wage re- ductions. Attempts to reduce dismissal costs for all existing jobs have faced strong opposition. It is estimated that around 95 percent of the labor force are covered by these centralized collective bargaining agreements (Pessino 2001). Any reduction in wages is considered to be a layoff of the worker, and hence makes him or her eligible for severance payments. New collec- tive bargaining agreements cannot reduce benefits granted to workers in previous agreements, except if a firm declares a crisis, and if the new type of agreement is approved by the Ministry of Labor. Consequently, employment bears the brunt of the adjustment during a downturn. Furthermore, in Argentina the fixed costs of labor contracts are substantial due to high mandatory severance payment and high payroll taxes. Thus, wages-for those who still have jobs-remain high, prompt- ing a queuing for jobs, which is reflected in both higher LFP and unem- ployment. This last factor may be quite important in explaining the in- crease in observed LFP between October 1994 and May 1995 and, if confirmed, has important policy implications. UNDERSTANDING LABOR SUPPLY DYNAMICS IN ARGENTINA II9 Predictive Power of the Model Concerning the fit of our models, we find that specification (1) predicts LFP rates in May 1995 reasonably well, but does not capture well the peaks of 1989 and 1993.13 The predictions improve considerably when we add the industrial production index, log(IPI). For example, the below- trend LFP during 1980-86, which corresponds to years of stagnation in GDP and IPI, is better captured because of the procyclical nature of LFP with respect to the IPI measure of the cycle. Including the male unem- ployment rate interacted with the dummy for the reform-convertibility period increases the predictive power of the model even further. Estima- tions by age group may be more meaningful since different age groups be- have quite differently to changes in the economy. A handy test of the predictive power of the regressions was available, since at the time of estimation we had the results of the October 1995 wave of the INDEC-EPH. Using the four alternative models, we pre- dicted LFP for October 1995. The results are presented in Table 4-8. For females, specifications (1) and (2) are best, while for males, models (1), (2), and (4) do equally well. The average predictive power of the models appears to be good, indicating that the theoretical foundations and econo- metric formulation are consistent with the data. Using these estimates, we can examine the extent to which "abnor- mally" large increases in LFP are responsible for the observed increase in unemployment levels in 1995. Interpreting departures from a linear trend as "abnormal," actual participation levels for both females and males in May 1995 appear to be about 5 and 2 percentage points above their nor- mal trend levels. For females, this means an actual participation rate of 55 percent compared with the trend LFP of about 50 percent. Similarly for males, an actual LFP rate of 86 percent compares to a trend rate of 84 per- cent. But well-established theory suggests, and our empirical analysis for Argentina shows, that cyclical factors are important, as well. When cycli- cal factors are incorporated, the "surprise" component of the increase in labor supply in May 1995 diminishes significantly. May 1995 participa- tion rates are still abnormally high, but by much smaller margins: less than 1 percentage point for both males and females. Determinants of Female Labor Force Participation The facts about female LFP presented so far do not necessarily say much about the causal relation between male unemployment rates and female LFP To address this issue we provide further evidence using cross-section 120 CRAFTING LABOR POLICY TABLE 4-8 The Predictive Power of Alternative Models of Labor Force Participation Actual LFP Predicted LFP of model: Group (percent) 1 2 3 4 Average Females 15-19 years 27.1 29.8 29.4 30.8 25.6 28.9 20-34 years 63.1 62.9 63.5 64.2 59.3 62.5 35-49 years 58.7 60.1 61.0 60.6 57.7 59.9 50-64 years 39.7 38.2 39.7 39.9 36.6 38.6 All 51.7 52.0 52.7 52.7 49.7 51.8 Males 15-19 years 44.3 44.5 46.0 46.7 43.2 45.1 20-34 years 92.8 94.0 94.2 93.8 92.9 93.7 35-49 years 97.3 97.5 97.6 97.5 97.0 97.4 50-64 years 82.1 81.7 83.4 82.9 80.4 82.1 All 84.1 84.9 85.4 85.3 84.0 84.9 Note: The four models are: (1) trend and cycle represented by unemployment of prime- aged males, UP; (2) trend, UF, and IPI; (3) trend, u' and period dummy for 1991-95 in- teracted with cycle measure; and (4) trend only. Actual LFP calculated using October 1995 INDEC-EPH. data. Our approach is two-fold. We first estimate a linear probability model, which describes the determinants of the female LFP decision. Sec- ond, we compare the probability of a wife being employed conditional on having an unemployed husband-Prob(wife is employed|husband is unem- ployed)-with the probability of a wife being employed conditional on her husband being employed-Prob(wife is employedlhusband is employed). The linear probability model confirms that women do respond to the wages and employment prospects of their husbands by changing their labor supply. On the basis of 1,697 observations of married women aged 15 and 64, preliminary results for May 1995 show that: LFP = -0.500 + 0.024 Education (-3.3) (7.9) + 0.049 AGE - 0.00065 AGE2- 0.04 KIDS (6.5) (-7.4) (-3.7) - 0.00006 WAGE_H (-4.7) UNDERSTANDING LABOR SUPPLY DYNAMICS IN ARGENTINA 121 where KIDS is the number of children under 14 years of age, and WAGE_H is the log wage of the husband. The t-statistics are reported in parentheses. This indicates that LFP of working-age women increases when the wage of the husband decreases. With respect to husbands' em- ployment status and education level we find that women whose husbands are employed are less likely to participate in the labor force. LFP = -0.40 + 0.028 Education (-2.6) (7.3) + 0.048 AGE - 0.00065 AGE2 - 0.04 KIDS (6.5) (-7.3) (-4.2) - 0.061184 EMPL_H - 0.0109 EDUC_H (-1.9) (-2.9) EMPL_H equals one if husband is employed, and zero otherwise, and EDUC_H is education of husband. In a next step, we compare the probability of a wife being employed conditional on having an unemployed husband with the probability of a wife being employed conditional on her husband being employed. If the added worker effect dominates, we expect that, on average, wives of un- employed husbands are more likely to be employed than wives of em- ployed husbands. Hence, the difference between the two probabilities should be positive. As can be seen in Table 4-9, a wife whose husband is unemployed has a higher probability of participating in the labor force (both employment and unemployment) than a wife whose husband is em- ployed. We see that in most years the difference between columns (3) and (6) is positive, with the exception of 1986 and 1987, which were years of relatively low male unemployment. Differences in LFP increase between 1988 and 1991, drop in 92, and then increase from 1993 to 1994. Except in 1991, when we expected to have small differences, these data show signs of a strong added worker effect for married women in the peak un- employment years.'4 As a consequence, if there is an exogenous increase in overall unemployment (caused, for example, by sectoral variability in the demand for labor), the effect will be magnified by the accompanying rising unemployment of females that "added" to the labor force to help overcome this fluctuation in family income. During the period studied, Argentina had cycles of high inflation that tended to be followed by stabilization plans that opened the economy to both international capital movements and trade in goods. This resulted not only in lower inflation rates, but also in lower real exchange rates. The I22 CRAFTING LABOR POLICY TABLE 4-9 Female Labor Force Status, by Husband's Employment Status Husband Unemployed Employed Employed Unemployed In LF Employed Unemployed In LF Wife (1) (2) (3) (4) (5) (6) Oct. 1985 37.6 6.4 44.0 32.6 0.9 42.5 Oct. 1986 34.2 1.2 35.4 34.3 1.3 35.6 Oct. 1987 31.7 2.3 34.0 33.2 1.8 35.0 Oct. 1988 38.1 6.0 44.1 36.2 2.3 38.5 Oct. 1989 40.7 5.2 45.9 36.2 1.7 37.9 Oct. 1990 51.1 1.5 52.6 37.8 1.1 38.9 Oct. 1991 54.1 0.0 54.1 36.8 1.3 38.1 Oct. 1992 41.8 3.8 45.6 38.0 1.9 39.9 Oct. 1993 45.5 5.4 50.9 39.1 4.1 43.2 Oct. 1994 42.4 9.8 52.2 37.6 5.5 43.1 May 1995 37.7 15.6 53.3 37.4 10.2 47.6 change in real exchange rates introduced sectoral reallocations of resources and people. In particular, low real exchange rates correspond to a relative rise in wages in the service sector, which is the sector in which women are relatively highly represented. Furthermore, a change in the real exchange may increase unemployment.15 Consequently, periods of stabilization that are combined with an opening of the economy or, put differently, a com- petitive shock through lower real exchange rates, may lead to an increase in male unemployment. Given that the added worker effect dominates, the increase in male unemployment will be accompanied by an increase in female LFP. If simultaneously the change in the real exchange rate implies an increase of the reLative wage in the sector in which women are more likely to work, female LFP will be intensified, exerting additional pressure on unemployment rates. Summary and Policy Implications Main Findings First, LFP in Greater Buenos Aires rose by 18 percent from 1974 to 1985, fuelled entirely by increases in female participation rates, which increased substantially since 1990. Female participation rates rose by 45 percent UNDERSTANDING LABOR SUPPLY DYNAMICS IN ARGENTINA I23 from 1974 on, while male participation rates remained unchanged. Par- ticipation rates of teenagers and older people also increased (or did not de- crease) from 1987 to 1995, and increased sharply during 1994-95. Second, the increase in unemployment from 6 percent in 1987 to 20 percent in May 1995 cannot be explained by changes in female LFP. De- spite seemingly rapid growth of the female labor supply compared to 1974 levels, increased female LFP can at best account for less than 6 percentage points of the change in the unemployment rate. This is largely due to the low initial rates of female participation (relative to countries with similar income and education levels), so that absolute increases in female workers are not sufficiently large to confirm conjectures that increased supply, and not slowing demand, is responsible for the rapid rise in unemployment. Third, the aggregate response of female LFP to changes in prime-aged male unemployment indicates that female participation rates are strongly countercyclical. The response to the cycle is negatively related to age. Fur- thermore, a wife whose husband is unemployed has a higher probability of being in the labor force than a wife whose husband is employed. The difference between these two probabilities is higher in periods when un- employment peaked. These findings indicate that in Argentina the phe- nomenon of added female workers prevails. Fourth, additional evidence indicates that reentrants (including retir- ees) and new entrants to the labor force account for only about one-third of the total unemployed in GBA in October 1995: almost two-thirds of the unemployed were previously employed, and while about two-thirds of reentrants and new entrants were female, almost two-thirds of the unem- ployed who were previously employed were male. While these findings do not prove that unexpected increases in LFP are not mainly responsible for the high levels of unemployment since 1987, they do provide another rea- son to be cautious in making such claims. Finally, we find that analyses of current labor market trends that at- tribute recent unemployment increases largely to unexpected or abnor- mally large labor supply increases inadvertently understate the importance of policy reform in stabilizing and increasing labor demand.'6 Interpret- ing departures from a linear trend as abnormal, actual participation levels for both females and males in May appear to be about 5 and 2 percentage points above their normal trend levels. But established theory suggests, and empirical analysis shows, that using only trend values as expected LFP ignores an important feature of the Argentine labor market. In Ar- gentina LFP is systematically influenced by cyclical factors. Once we ac- count for cyclical factors, the surprise component of the recent increase in I24 CRAFTING LABOR POLICY labor supply diminishes significantly. May 1995 participation rates are still abnormally high, but by a margin of less than 1 percentage point for both males and females. Policy Implications The pattern of unemployment indicates that it is a widespread phenome- non and is not restricted to a few clearly identified worker groups. The rel- atively uniform increase in unemployment rates across worker categories points to falling labor demand as the main cause of growing unemploy- ment. In this chapter we show that LFP rates are positively correlated with unemployment rates in Argentina. This should be taken into account when drawing inference from changes in LFP on unemployment, and hints at the fact that the government should aim at increasing employ- ment growth. Figure 4-1 summarizes a simple exercise conducted to illus- trate the importance of measures to raise the trend rate of employment growth. Data on employment and labor force growth clearly show that employment is growing more slowly than the labor force by a widening FIGURE 4-1 Employment and Labor Force Millions 14 - 13 - 12 - 11 - 10 - -- 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 |-0- Employment -+-- Labor Force - - Linear Trend Labor Force ---- Linear Trend Employment Source: Ministry of Labor and Social Security, based on data from INDEC-EPH and Secretaria de Programacidn Econdmica (Department of Economic Policy). UNDERSTANDING LABOR SUPPLY DYNAMICS IN ARGENTINA I15 margin; the result, naturally, is higher unemployment. In the preceding chapter we showed that relying on output growth is not enough to in- crease employment. Labor market reforms are necessary. Hence, the gov- ernment would be advised to concentrate on policies that increase labor demand and employment, such as making wage- and employment-setting practices more flexible, reducing payroll taxes, and minimizing interven- tions that encourage workers to hold out for higher wages. Reforms that encourage efficient employment and wage-setting prac- tices-while promoting sustainable growth of labor demand-are likely to favorably influence short-term labor supply patterns as well. As we have ex- plained, in Argentina wage adjustments are hindered by individual and collective bargaining laws and practices, which evolved during a period of high inflation. Consequendy, employment bears the brunt of the adjust- ment during a downturn. Empirical studies indicate that in countries with relatively flexible labor markets, such as the United States, LFP is pro- cyclical. In Argentina, higher unemployment rates, which result in lower expected household incbme, induce additional workers to join the labor force. Consequently, the LFP of secondary workers tends to be anticyclical because of the added worker effect, leading to exaggerated fluctuations in measured unemployment rates. Labor reforms that encourage wage flexi- bility will result in reduced cyclical volatility of unemployment rates, not just because employment fluctuations will be moderated by wage adjust- ments, but also because labor supply fluctuations will be less exaggerated. This in turn will allow the government to concentrate on the task of instituting policies for raising long-term employment growth, instead of being distracted by temporary but sharp increases in unemployment, which are politically difficult to ignore. Notes I. It is believed that there are between 800,000 to I million foreigners work- ing illegally in Argentina, mostly as maids and construction and agricultural workers (Migrant News 1996). Between 1980 and 1991, the population born in foreign countries decreased by 14 percent for the country as a whole, and by 16 percent in Greater Buenos Aires. 2. Since data are not readily available for the rest of the country, we use data for Greater Buenos Aires for all estimations, unless stated otherwise. Greater Buenos Aires accounts for more than 50 percent of the GDP and nearly 40 per- cent of the population. 3. Of course, the income effect must not be larger than the cross-substitution effect, if the latter is positive. IZ6 CRAFTING LABOR POLICY 4. In North America female LFP rates averaged 69 percent in the 1990s. 5. In fact, one of the major consequences of development during this century has been the incorporation of women in high numbers into the labor force. Since Argentina has lagged in development with respect to OECD countries, it is not surprising to find that female LFP rates are lower in Argentina than in OECD countries, and have been growing at a slower pace. 6. Estimates of probit results for October 1991 and October 1994 can be found in Pessino (1997, pp. 196-7). 7. Alternatively, we may use a demographically adjusted unemployment rate as a measure for the cyclical conditions; however, because the quality of available jobs may vary over the cycle, we refrain from using this variable so as not to in- troduce simultaneity bias. 8. We control for autocorrelation since the Durbin-Watson statistic reveals se- rially correlated errors. Higher-order serial correlation was rejected by the data. Since the data are usually measured in April/May and September/October, we also introduced a seasonality dummy, since May and October are presumably dif- ferent in terms of LFP Only the participation rate for teenage females showed lower entrance into the labor force during October, without significantly altering the other coefficients. 9. Note that for men aged 35 to 49 we can have spurious correlation between LFP and UP and, given the correlation between UP and unemployment for males aged 20 to 34, results should be interpreted carefully. 10. Excluding the observations from 1993 onward, the estimates of a lose sig- nificance but do not switch signs. This evidently shows that the 1993 period is influencing the results, and that it was a period of strong countercyclicity in fe- male LFP. 11. In fact, using employment transition probabilities from household data, Lundberg (1985) found that there is a small but significant added worker effect for the United States, and attributes this finding mainly to the existence of credit constraints for some households. 12. The Convertibility Plan was an exchange rate-based stabilization program introduced in October 1991 after the end of the hyperinflation in 1990. The plan eliminated tariffs on imported capital goods, which substantially reduced the rel- ative price of capital. Further economic reforms included trade liberalization and extensive privatization of state-owned companies. 13. In October 1983, the sample size of the INDEC-EPH became much smaller, and the actual values of LFP are less precise. Note the large increase in LFP that is not captured by the estimates. 14. We have a large drop in the absolute number of husbands unemployed- from 68,800 to 42,700-and the relatively lower number of unemployed may have strong reasons to be unemployed, such as general disadvantages or sickness that make the wife the sole provider. 15. A lower real exchange rate means that imported goods are cheaper. Among other things, Argentina imports capital goods. A lower price of capital goods with UNDERSTANDING LABOR SUPPLY DYNAMICS IN ARGENTINA 127 respect to the higher price of labor implies a strong substitution effect toward cap- ital and against labor if production processes allow such adjustments. See Chap- ter 5 for moderate elasticities of substitution between labor and capital. 16. In Chapter 5, Pessino, Gill, and Guasch address labor demand issues, mea- suring the output and wage elasticities of labor demand during the same period as this paper, and imputing the benefits to labor reform in contrast to a policy of simply relying on economic growth to increase employment and reduce unem- ployment rates. References Clark, K. B., and L. H. Summers. 1981. "Demographic Differences in Cyclical Employment Variation." Journal ofHuman Resources 16: 61-79. Killingsworth, Mark R., and James J. Heckman. 1986. "Female Labor Supply: a Survey." In 0. Ashenfelter and R. Layard, (eds.), Handbook of Labor Econom- ics, Volume 1. Elsevier Science Publishers, pp. 103-204. Lundberg, Shelly. 1985. "The Added Worker Effect." Journal of Labor Economics 3 (1): 11-37. Migrant News. March 1996. Pencavel, John. 1986. "Labor Supply of Men: A Survey." In 0. Ashenfelter and R. Layard, (eds.), Handbook of Labor Economics, Volume I. Elsevier Science Publishers, pp. 103-204. Pessino, Carola. 1997. "Argentina: The Labor Market during the Economic Transition." Chapter 6 in Sebastian Edwards and Nora Lustig, (eds.), Labor Markets in Latin America: Combining Social Protection with Market Flexibility. Washington, D.C.: The Brookings Institution. _____. 2001. "Convertibility and the Labor Market." Paper prepared for the Conference The Currency Board's First Decade: Argentina and the Convertibil- ity Law. BCRA, April 5 and 6. Pessino, Carola, and Leonardo Giacchino. 1994. "Rising Unemployment in Argentina: 1974-1993." Paper prepared for the 13th Latin American Meeting of the Econometric Society, Caracas, Reptiblica Bolivariana de Venezuela. 5 Increasing Labor Demand in Argentina CAROLA PESSINO, INDERMIT S. GILL, AND JosE Luis GUASCH IN THE YEARS FOLLOWING the initiation of economic re- forms in the late 1980s, Argentina's employment growth has increasingly lagged behind output growth. Between 1990 and 1993, employment grew by about 2 percent annually, compared to a much higher economic growth rate of approximately 8 percent. The result was an increase in av- erage labor productivity by approximately 5.5 percent per year between 1990 and 1993 (Pessino 1997). In 1991, the Convertibility Plan-an exchange-rate-based stabilization program-was introduced, which also eliminated tariffs on imported capital goods. As a consequence, the rela- tive price of capital declined. Evidence on the combined effects of the Convertibility Plan, introduction of subsidies for domestic capital, and falling interest rates indicate that the price of capital relative to labor fell between 26 to 40 percent between 1990 and 1994 (Bour 1995). Thus, the "output effect" on employment triggered by the reforms was counteracted by a "substitution effect," leading to an increase in the price of labor rela- tive to capital. The effect of trade liberalization, combined with the ap- preciation of the peso, particularly hurt employment of unskilled labor, but led to an increase in dollar-dominated wages. This implied that aver- age real wages began to rise until 1994. From 1974 to 1995, total employment grew at an annual rate of 1 per- cent, and the population grew at approximately 3 percent. This implied that net employment growth-that is, employment growth minus popu- lation growth-decreased by 0.29 percent per year. Growth of labor force participation exceeded population growth by 0.5 percent, resulting in a I29 I30 CRAFTING LABOR POLICY difference between employment growth and labor force growth of 0.79 percent. The government introduced programs to encourage private em- ployment. Public programs to help displaced workers and new entrants to a labor market in transition were also initiated. Furthermore, the govern- ment responded with measures to lower labor costs and smooth adjust- ments by reforming wage-setting and employment regulations. But em- ployment growth remained sluggish. In 1994, in spite of GDP growth of about 6 percent, employment stagnated or even declined. In 1995, during the recession induced by the so-called Tequila Crisis, GDP growth and employment fell, and unemployment soared to about 20 percent. A likely explanation for the relatively slow employment growth in the early years despite healthy GDP growth is that the costs of labor remained high. Although productivity increased, it was not enough to compensate for the increase in average real wages. At the same time, Argentine labor legislation mandated that employer contributions to payroll taxes amounted to approximately 33 percent between 1990 and 1995. There is also the cost of a mandated 13th monthly salary, mandatory compensa- tion for work-related injuries, and severance payments. In the case of lay- offs "without just cause," severance payments are mandatory and advance notice of one or two months must be given to the worker. The Employ- ment Law passed in 1991 determined that severance compensation be equal to one month's salary per year of service, with a minimum payment of two months' wages and a maximum yearly payment of three times the average wage of the collective bargaining agreements that apply to the worker. To provide alternatives to the predominant form of contracting in Ar- gentina, which is by law of indefinite duration, the government introduced new forms of group-specific, fixed-term contracts. These contracts fea- tured lower payroll taxes and severance payments.' However, the law re- quired union approval of these contracts. But unions largely resisted them. Hence, it is not surprising that, as Hopenhayn (2001) shows, this reform seems not to have had a great impact. In 1995 a further reform introduced a trial period for all contracts and special employment promotion con- tracts. These contracts did not require the previous approval of trade unions. They induced increased hiring, but also reduced longer-term em- ployment. In 1999 the reform was reversed. GDP, productivity, and unemployment grew steadily between 1990 (the end of hyperinflation) and 1994. This contradicted the widespread belief that economic growth induced by macroeconomic stabilization would be a major contribution to job creation. The unemployment rate rose steadily from about 5 percent in October 1990 to about 10 percent INCREASING LABOR DEMAND IN ARGENTINA I31 in October 1993, and to 11 percent in May 1994. With the falloff in GDP growth in 1995, and a continued rise in the relative price of labor, employment growth fell. The unemployment rate rose sharply to almost 20 percent in May 1995, triggering a heightened interest in labor market issues. The fact that the relative price of labor did not adjust downward after the drop in GDP growth may be blamed on the slow progress of the labor reforms. The effects of the labor market rigidities had become worse in the more competitive, low inflation environment of the postreform pe- riod. While substantial reforms had taken place in monetary affairs and goods and services markets, no far-reaching labor market reform took place in the 1990s. Some claim that this asymmetry may have contributed to the high unemployment rates (Mondino and Montoya 2000). Changes in unemployment and wages depend on changes in both labor demand and supply. Issues related to labor supply and unemployment were discussed in the previous chapter. Here we examine the determinants of labor demand from a policy perspective. Classical theory predicts that-given the long-run neutrality of money-the main determinants of labor demand can be identified as output (GDP), since the demand of labor is derived from the demand of commodities and services, and the relative price of labor with respect to other input factors such as capital. The relative price of labor determines the costs of using labor in produc- tion. We use different assumptions for the technology of production to compute the own-wage and output elasticities of labor demand. Simply put, the own-wage elasticity of labor demand provides a numerical guide to the "payoff" of labor market reforms that lower the cost of labor in terms of an increase in employment. Estimates of the output elasticity of labor demand, holding wages and other labor costs constant, provide quantitative estimates of the effect of renewed output growth on increases in employment, given that labor market functioning is left unaltered. Our results can then be combined with estimates of the influence of different policy measures on labor costs, which are provided in World Bank (1996), to impute the employment effects of these policy reforms.2 We hope that quantitative estimates of the responsiveness of employment to, alterna- tively, higher growth and labor policy reform, will enrich the policy debate in Argentina and in other Latin American countries. Methodology, Technology, Data, and Issues In Argentina, beset by high inflation and concerns of macroeconomic sta- bility until the late 1980s, labor market issues were a secondary concern. Since the economic reforms in the early 1990s, however, employment, un- 132 CRAFTING LABOR POLICY employment, and wage policies have assumed importance. The paucity of labor-related data will increasingly hamper informed decisionmaking on critical issues. The construction of appropriate labor series is difficult, and application of econometric techniques to available labor-related data in Argentina is troublesome. Still, this chapter illustrates both the benefits of such exercises and the value of reliable data in facilitating informed deci- sionmaking. In what follows we briefly describe the methodology and data used. We then discuss the relevant issues. Methodology Assuming two factors of production labor (L) and capital (K), a constant returns to scale (CRS) production function and profit maximization, the own-wage elasticity of labor demand, holding output, and price of capital r constant, is given by £LL = -[1-s]7O, Cij>O, i, j = w, r (2) Using Shephard's lemma (see Varian 1992), we then write conditional labor demand as a function of the prices of labor and capital, output, and technology: dC/dw = L* = Ld (w, r, Y) (3) where dC/dw is the marginal effect of wages on total costs and L* is the profit-maximizing demand for labor. Capital demand is derived similarly. The own-wage elasticity of labor demand ELL equals equation (1). The only difference is that now s = wL/C, which is the share of labor in total costs. INCREASING LABOR DEMAND IN ARGENTINA 133 Representations of Technology The form of technology in an economy is generally not known. For this reason, we provide estimates of the elasticities of interest based on different production functions that represent a broad range of technologies. These technologies include Cobb-Douglas, constant elasticity of substitutions (CES), generalized Leontief, and transcendental logarithmic technology. (a) Cobb-Douglas Technology C = C(w, r, Y) = Zw'-P.rP.Y (4) where , is the share of capital and Z is a constant. Using Shephard's lemma, we get Ld = A.w(-P).rP.Y (5) InL = InA + (-,B)lnw + Ilnr + InY (6) A is a constant. The output elasticity and a, the elasticity of substitu- tion between L and K, are restricted to equal 1. The own-wage elas- ticity of labor demand SLL is given by (-,B). (b) Constant Elasticity of Substitution Technology C = Y[Qa'.w10 + (1a)UrbKl1/(l) (7) Using Shephard's lemma, the labor demand equation is: Ld= - ,w-y InL = alna-clnw + ELY InY (8) a is labor's share of revenue if the factors are paid their marginal prod- uct. a is the elasticity of substitution and is a constant, but does not necessarily equal 1. The own-wage elasticity of labor demand ELL now is given (-a). The output elasticity ELy is restricted to equal 1. (c) Generalized Leontief Technology C = Y[ajjw + 2a12J 5r0 5+ a22r] (9) Using Shephard's lemma, the labor demand equation becomes: Ld = a11 Y + a,2Y(w/r)0 5 (10) 134 CRAFTING LABOR POLICY Neither elasticity of substitution, output elasticity, nor own-wage elas- ticity of labor demand is restricted to be a constant; they vary with the ratio of input prices. (d) Transcendental Logarithmic Technology (Translog) In C = Ina, + a1lnw + 0.5a2[1n wl' + a)nw.lnr + 0.5b,.[Inr]3 + b)nr + b3}nY +b41nYlnr + a4lnY.lnw (11) Using Shephard's lemma, we arrive at the following labor demand function: wL/C = a, + a)lnw + a3lnr + a4jnY (12) Again the elasticities are not constants, but depend on the parameters and both factor prices. These technologies vary not only in their assumption with respect to the different elasticities, but also in the data required to estimate the re- spective parameters of interest. Given the data problems we face, using different functional forms provides an important device for checking the robustness of our results. Dat Sources and Concepts To estimate labor demand elasticities, we use all the data available for 1974-95. The main sources of information are publications of the Minis- terio de Economia (Ministry of Economy), the INDEC-EPH household surveys, and national accounts published first by the Central Bank, and then by the Ministry of Economy. The lack of reliable industry- and enterprise-level surveys is a serious shortcoming for our purposes. Worse, in Argentina no complete national accounts on the quantity of factors and their remuneration have been published since 1975. Consequently, homo- geneous and consistent aggregate data are not available, and the data we use in this study come from different private and public sources. Most of it had to be reconstructed, combining data from different sources, in order to cover the whole period analyzed. LABOR DEMAND. Employment data differ in coverage and sources. Concerning coverage, we have series for Greater Buenos Aires, for manu- facturing workers, and for wage workers, or combinations thereof. Gener- ally speaking, when coverage is taken into account, all series show very INCREASING LABOR DEMAND IN ARGENTINA 135 similar patterns. The two main sources are the INDEC-EPH surveys and the Informe Economico del Ministerio de Economia (Economic Report of the Ministry of Informe 1995). The different series for Greater Buenos Aires show very similar patterns. The main difference between Ministry of Economy and INDEC numbers arises in 1994 and 1995, when INDEC- EPH numbers show a greater decrease in employment than do Ministry of Economy numbers. We therefore use both series in our estimations to make sure that our results are not sensitive to the choice of series. Most of the results we present refer to the manufacturing sector. While total em- ployment reveals an upward trend, this is not necessarily for the manufac- turing sector, especially in the period following the Convertibility Plan. This may be due to the fact that in Greater Buenos Aires, the import- competing sectors in manufacturing, such as textiles and clothing, and metallic products, faced a concentration in the employment decrease, while the share of the service sector in total employment increased since 1991 at the expense of that of manufacturing.4 OUTPUT. GDP series are from the Ministry of Economy. Indexes of industrial production are from the Central Bank (1974-80) and the Fun- dacion de Investigaciones Econdmicas Latinoamericanas-FIEL (Latin Amer- ican Economic Research Foundation) (1980-95). Based on these data, we construct three series: real per capita GDP, real GDP, and industrial pro- duction. They display very similar patterns over the sample period. PRICE OF LABOR. Nominal wage data for 1970-79 are from FIEL, and data for 1980-95 are from the Ministry of Economy, FIEL, or INDEC- EPH. FIEL data are restricted to the manufacturing sector, while from 1980 on all sectors are included. Using these different sources makes the nominal wage series less than satisfactory, which implies that the real wage series (that is, nominal wages deflated by the consumer price index) are rather unreliable. Until the mid-1980s the two series are similar, but in 1987 they began to diverge. From 1990 on even the slopes differed. We define real labor costs as wages adjusted for payroll taxes and deflated by the wholesale price index. The different series for real labor costs track one another closely. PRICE OF CAPITAL. Constructing a reliable series on the price of capi- tal is a different task. Given the data constraints we had, we opted for the following solutions. We use the implicit price of gross domestic invest- ment from the national accounts, and deflate it either by the wholesale 136 CRAFTING LABOR POLICY price index (WPI) or implicit prices in GDP. Alternatively, we use the price of domestic and imported machinery deflated by the WPI. RELATIVE PRICE OF LABOR. More reassuringly, there is considerable consensus in the series for the price of labor relative to capital, which is most relevant for our estimation of labor demand. The exception is for 1995: three series show a decline, while two show a moderate increase. The divergence is again caused by the difference between labor costs that include an adjustment for compliance rate changes for payroll taxes and those that do not, and different deflators in the wage and capital series. Issues The objective of this chapter is to estimate the output elasticity of labor demand, holding wages and other labor costs constant, and the own-wage elasticity of labor demand. As shown in the methodological section, these are derived from the production or cost functions of the firms, which sug- gests the use of firm-level data. These data were not available for the esti- mation of long-term elasticities and, consequently, we have to rely on ag- gregated data. As noted in the literature (for example, Hamermesh 1993), there are several methodological concerns that arise when using aggregate data instead of establishment-level data. These imply that parameter esti- mates do not usually correspond to elasticities at the firm level. This has to be kept in mind when interpreting our results. For example, one series of our employment data refers to manufactur- ing employment in Greater Buenos Aires. Hence, because we do not esti- mate elasticities at the firm level, we cannot answer the question of how a firm would change its employment in response to a shock to its own wages. What we do estimate is a change in employment net of all shifts of employment that occur among firms in the manufacturing sector in Greater Buenos Aires, including those that are induced by changes in rel- ative costs across firms with different factor intensity. Ignoring biases that may result from other issues, such as measurement error and endogeneity, Hamermesh (1993) points out that estimates derived from studies using highly aggregated data will understate the true firm-level value. Further- more, using aggregate employment5 data introduces the standard problem of linear aggregation of nonlinear relationships. And it means that we simply add up the number of workers. This is a common procedure, but embodies the rather strong assumptions that workers are equally produc- tive and separable from other inputs.6 Using employment as a proxy for INCREASING LABOR DEMAND IN ARGENTINA I37 the quantity of labor may also lead to biases if worker hours are correlated with factor prices and outputs. Because we use time-series data and are in- terested in estimating long-term elasticities, we may ignore this problem due the fact that in the long term, variations in employment and worker- hours are generally highly correlated. A further data issue arises from the fact that the frequency of the data is annual, covering 1974-95. This gives us only 22 data points. There are some biannual data available for the same period, but only for some vari- ables. Since not all the variables could be included in the estimations, and there were other measurement and time aggregation problems, estimators tended to be less efficient using the biannual data, even after correcting for first-order autoregressive (AR1) and second-order autoregressive (AR2) specifications, including seasonal dummies and deseasonalizing the time series. However, the variable of the annual data set shows a high variance, which improves the efficiency of the estimation of the parameters of in- terest. Moreover, in our data, aggregate employment and output series dis- play strong serial correlation. The solutions adopted in this chapter are to choose output and employment series that are less aggregative-for exam- ple, industrial output, manufacturing sector employment, and wage em- ployment-and to apply econometric procedures that adjust for aurocor- relation in the data. Another important concern is whether to take quantities or prices as exogenous for the econometric implementation. Ideally, both the price and quantity of labor should be treated as endogenous. Therefore, one must be able to argue that supply of labor is either completely inelastic or completely elastic in response to exogenous changes in demand. So, the choice usually boils down to whether price or quantity is viewed as ex- ogenous. Input prices are likely to be endogenous variables because a change in wages, for example, affects demand of labor, but also its supply. For each point in time for which we have an observation, we observe a cer- tain combination of employment and wages. Comparing these outcomes, we can see changes in employment and changes in wages, but ex ante we do not know whether these changes are due to a change in demand or a change in supply of workers. One way to solve this issue is to use so-called instrumental variables, which in our case are variables that are correlated with input prices, but not with labor demand. Or to be more precise, we seek variables affecting the supply of labor, but not its demand. Conse- quently, we present results using instruments. The instruments we use are the logs of the real exchange rate (TRC2), real GDP per capita (Y1), real GDP index (Y2), urban population in Greater Buenos Aires (PBLURB), 138 CRAFTING LABOR POLICY and annual percent change in CPI (INFL). Unfortunately, in this context, good instruments are hard to find. With aggregate data, if the supply of labor is relatively inelastic in the long term, demand parameters are best estimated using specifications that treat the quantity of labor as exoge- nous. This recommends the use of a production function, such as the generalized Leontief and the translog production functions, which include factor quantities as regressors. Econometric Specification and Results As shown in the Methodology section, the own-wage and output elastic- ity of employment can be estimated either by relying on the production function directly, or by using the cost function approach and estimating factor demand functions. In what follows, we provide evidence of the estimates of these elasticticities for the technologies presented above. Fur- thermore, we will discuss the single factor estimation and its results. We use different time series for employment and output to check the robust- ness of our results and to control for autocorrelation. We address the issue of endogeneity of input prices by using instruments. Cobb-Douglas Technology We start with the analysis by estimating the own-wage elasticity of labor demand and the output elasticity of labor demand based on a Cobb- Douglas technology. For this technology the share of labor in total revenue of the firm (s = wL/pY) corresponds to the share of labor, which equals 1-,B. As (, the elasticity of substitution between labor and capital equals 1. It is easy to see from equation (1), that the own-wage elasticity of labor demand £LL equals minus the share of capital (-j3). Table 5-1 reports esti- mates of ELL, which are obtained from estimating equation (6), using dif- ferent output (Y) and employment (L) series. Diagnostic tests reveal the presence of autocorrelation. Taking into ac- count that there exists first order autocorrelation, these estimates imply that for employment in Greater Buenos Aires (EMl and EM2), the own- wage elasticity of labor demand ELL is about -0.65. This means that hold- ing output constant, a 1 percent increase in wages reduces employment net of all shifts of employment that occur among firms in Greater Buenos Aires by 0.65 percent. The elasticities for manufacturing employment (EM4) and wage employment in the manufacturing sector (EM6) are -0.74 and -0.71, respectively. Table 5-1 reveals ELL ranges from -0.50 to INCREASING LABOR DEMAND IN ARGENTINA 139 TABLE 5-1 Cobb-Douglas Technology (Estimates of ELL) L Used Y Used Specification e T-statistic EMI Y2 ARI -0.635 -3.19 EM2 Y2 ARI -0.674 -2.84 EM4 Y2 OLS -0.628 -8.21 EM4 Y3 OLS -0.472 -6.04 EM4 Y2 ARI -0:742 -6.95 EM4 Y3 ARI -0.493 -4.52 EM5 Y2 ARI -0.497 -2.31 EM6 Y2 OLS -0.640 -8.83 EM6 Y2 ARI -0.713 -7.25 EMI Employment in GBA (15 to 64 years), from INDEC-EPH. EM2 Employment in GBA (all ages), from Ministfio de Economia InfOrme Econdmico. EM4 Manufacturing employment, from INDEC-EPH and Ministdrio de Economia. EMS Total wage employment, from INDEC-EPH and Minist6rio de Economia. EM6 Wage employment in manufacturing, from INDEC-EPH and Ministdrio de Economia. Y2 Real GDP Index, from Ministrio de Economia (1980-95) and the Central Bank (1974-80). Y3 Industrial Production Index, from FIEL and Central Bank. See technical annex B for details. -0.75, which is consistent with estimates of the National Accounts in the 1970s (not updated since), and with a simple estimation of the share of labor. The share of labor can be estimated multiplying average wages (900 pesos) by aggregate employment (12 million employees) and dividing this product by GDP (300 billion pesos). This yields an estimate of the share of labor of 0.35, which corresponds to a share of capital j of 0.65. Our finding of the share of labor 1-1 is consistent with the results of other au- thors. Delfino (1984) finds a share of labor in manufacturing of approxi- mately 0.40 in 1973. Using data from the National Accounts of the Cen- tral Bank of Argentina, Elfas (1988) reports a series on the share of capital income in GDP, whose average during 1950-80 is close to 0.6. Constant Elasticity of Substitution Technology The CES technology continues to assume that the elasticity of substi- tution a is a constant, but loosens the restriction imposed in the Cobb- I40 CRAFTING LABOR POLICY TABLE 5-2 CES Technology: Cost Function Approach (Estimates of a, ELL, and sey) L Used Y Used Specification a0 £LY EM4 Y3 OLS W3 0.16 -0.10 1.08 (2.19) (-2.20) (0-36) EM4 Y3 IV* AR1 W3 0.35 -0.21 0.34 (3.28) (-3.24) (1.49) EM4 Y2 IV* ARI W3 0.51 -0.31 0.52 (1.77) (-1.76) (0.86) EM6 Y3 OLS W3 0.19 -0.11 0.13 (2.38) (-2.32) (0.52) EM6 Y3 IV* AR1 W3 0.41 -0.25 0.45 (3.70) (-3.65) (1.91) EM6 Y2 IV* ARI W3 0.62 -0.37 0.72 (2.03) (-2.00) (1.11) EM4 Manufacturing employment, from INDEC-EPH and Ministerio de Economla. EM6 Wage employment in manufacturing, from INDEC-EPH and Ministerio de Economfa. Y2 Real GDP Index, from Ministerio de Economfa (1980-95) and the Central Bank (1974-80). Y3 Industrial Production Index, from FIEL and the Central Bank. Note: To calculate SLL it was assumed that sL = 0.6. Instruments used were LTRC2, LY2, or LY3, LPBLURB, and INFL. Douglas technology that ca equals 1. Table 5-2 shows estimates of ay, the own-wage elasticity of labor demand 8LL and the output elasticity of labor demand ELY. The employment data used is manufacturing employment (EM4) and wage employment in the manufacturing sector (EM6). Speci- fications using the more aggregate categories of labor for all sectors in Greater Buenos Aires (EM 1, EM2, and EM3) did not produce tight esti- mates of the parameters of interest. The reason may be that the trend cap- tures most of the movements in these series and that using data for man- ufacturing and services sectors masks the true elasticities in each sector of the economy. To finesse this problem, most subsequent estimates rely on employment in the manufacturing sector. The effect of wages on manufacturing employment, as measured by the wage elasticity of labor demand, is stronger when using wage employment in manufacturing, which excludes self-employed and owners, compared to employment in manufacturing. As Table 5-2 shows, the elasticities are INCREASING LABOR DEMAND IN ARGENTINA 141 -0.37 and -0.31, respectively. Because input prices are likely to be endoge- nous, we choose instrumental variable estimations. In this context, we need instruments that are correlated with input prices and henceforth labor supply, but not with labor demand. The instruments we use here are the logs of the real exchange rate (TRC2), real GDP per capita (Y1), real GDP index (Y2), urban population in Greater Buenos Aires (PBLURB), and annual percent change in CPI (INFL). Using instrumental variables substantially increases the estimated elasticity of substitution, which im- plies that the absolute value of the own-wage elasticity of labor demand ELL increases. Instrumental variable estimates of the output elasticity, ELY lie between 0.34 and 0.72. The estimates are sensitive to the use of GDP or industrial production. They tend to be higher but of lower statistical significance for the real GDP index (Y2), relative to estimates based on the industrial production index (Y3). Note that CES technology restricts ELY to be equal to 1. Our estimates indicate, however, that the output elastic- ity is statistically different from 1. This does not necessarily tell us that firm technology is not CES, but may be due rather to the level of aggre- gation in our data. Under the assumption that firms make zero profits and operate in com- petitive markets, an alternative way of estimating the elasticity of substi- tution 6 based on CES technology can be derived, which allows us to check the robustness of our results. We proceed as follows. From the pro- duction function we take the first derivatives with respect to labor and capital and obtain: DY/aL = a (Y/L)'-P (13) DY/dK = [1-a](Y/K)'-P (14) Note that p =1-1/a. When equating (13) to w/p and (14) to r/p and adding an error term, we obtain the estimable equations: In(Y/L) = (xi + a1 In (w/p) + ul (15) and ln(Y/IK) = a2 + a2In (r/p) + u2 (16) where as1 and 62 are alternative estimates of 6 (see Berndt [1991] for a dis- cussion on these alternative estimates of a). Under the assumption of cost minimization the ratio of equations (15) and (16) is equal to the factor price. Taking logs and differentiating with respect to In(w/r) yields: -dln(K/L)/dIn(w/r) = a (17) 142 CRAFTING LABOR POLICY Hence an estimable form of this equation is: ln(K,L) = 03 + q3 ln(w/r) + U3 (18) As can be seen, estimation of equations (16) and (18) require a measure of capital input, and the price of its service r. Both are likely to be mea- sured with error, and hence estimations based on this approach should be less reliable than the results obtained from (15). Table 5-3 shows that 61, which is derived from (15), ranges from 0.5 to 0.7. Under the assumption that sL = 0.6, the own-wage labor demand elasticity of ELL lies between -0.3 and -0.5, which correspond to the above findings. Again, elasticities of substitutions are higher when using Y2 instead of Y3. And given the same output measure, oa is higher when wage employment in manufac- turing, instead of manufacturing employment, is the employment mea- sure. Difficulties in inferring _L when including an imperfectly measured price of capital can be overcome in a labor-demand equation by specify- ing w and r separately. We now turn to estimates based on the so-called factor-demand equations. TABLE 5-3 CES Technology: Production Function Approach (Different estimates of c) Variables Usedfor Output and Employment 01 o2 03 Real GDP index (Y2) 0.67 0.32 1.15 Manufacturing employment (EM4) (5.73) (2.08) (7.55) Industrial production (Y3) 0.48 0.28 Manufacturing employment (EM4) (3.90) (0.93) - Real GDP index (Y2) 0.70 - 1.21 Manufacturing wage employment (EM6) (5.95) - (7.75) Industrial production (Y3) 0.51 Manufacturing wage employment (EM6) (4.29) Note: Instruments used were log of the real exchange rate (TRC2), log of real GDP or industrial production (Y2 or Y3), urban population in GBA (PBLURB), December-to- December percent change in the CPI (INFL). See technical annex B for details. All esti- mates are ARI corrected instrumental variable estimates, with t-statistics in parentheses. INCREASING LABOR DEMAND IN ARGENTINA 143 Estimation of Single Factor Demand Equations To estimate Shephard's condition, equation (3) can be written in loga- rithmiic form as a log-linear equation. InL = a' + ELL lnw + LK lnr + ELY Iny + u (19) where a is a parameter and u represents an error term. ELK is the cross- elasticity of demand for labor in response to a change in. the price of cap- ital services. This specification allows us to obtain direct estimates of con- stant-output elasticity of demand for labor 5LL' ELK and the employment- output elasticity ELY7 Table 5-4 presents the results of this estimation. Again we find that eLL ranges from -0.4 to -0.5, which is consistent with the estimates of 6 and the labor share estimated above. However, the TABLE 5-4 Single Factor Deniand Equations (Estimates of £LL, ELK' and ELY) Variables usedfor emplo'yment, wages, and output ELL ELK SLY Manufacturing employment (EM4) -0.396 0.580 0.080 Nominal wages (W3), industrial production (Y3) (-4.03) (2.42) (0.35) Unrestricted Manufacturing wage employment (EM6) -0.472 0.718 0.127 Nominal wages (W3), industrial production (Y3) (-4.88) (3.05) (0.57) Unrestricted Manufacturing employment (EM4) -0.412 0.412 0.189 Nominal wages (W3), industrial production (Y3) (-4.30) (4.30) (1.05) Restricted Manufacturing wage employment (EM6) -0.493 0.493 0.272 Nominal wages (W3), industrial production (Y3) (-5.19) (5.19) (1.57) Restricted Note: Instruments used were log of the real exchange rate (TRC2), log of real GDP or industrial production (Y3), urban population in GBA (PBLURB), December-to- December percent change in the CPI (INFL). See technical annex B for details. All esti- mates are ARI corrected instrumental variable estimates. t-statistics in parentheses. I44 CRAFTING LABOR POLICY estimates of the cross-elasticity of demand for labor eLK and the output elasticity ELy appear to be inconsistent with the previous estimates. Given a constant returns to scale production function, profit maximizing behav- ior and competitive market, the cross-elasticity of demand for labor in re- sponse to a change in the price of capital equals the negative of the own- wage elasticity, which means that ELL + ELK = 0. An F-test of the estimates cannot reject the null hypothesis that ELK equals - ELL Consequently, we impose this restriction in the following estimations and find that the esti- mate of ELK decreases, while that of ELY increases in size and significance. Generalized Leontief Technology To estimate the generalized Leontief technology, we impose constant re- turns to scale and estimate equations (20) and (21) jointly, using a seem- ingly unrelated regression estimator (SURE) with the imposed cross-equa- tion restriction. We apply three-stage least squares estimation to correct for the simultaneous equation bias. This procedure provides the following results for the labor demand equation (10) and the capital demand equa- tion, using manufacturing employment (EM4) and real GDP index (Y2): L/Y = -0.0029 + 0.0098 (w/r)-05 (20) (-1.26) (4.12) K/Y = 0.0063 + 0.0098 (w/r)-05 (21) (2.47) (4.12) Using these estimates, the elasticity of substitution a, the own-wage elasticity of labor demand ELL, and the constant-output capital demand elasticity, EKK can be calculated (see Berndt 1991). These elasticities are no longer constant, but vary over the sample interval. As can be seen in Table 5-5, elasticities are much higher than the above-presented estimates. How- ever, the absolute value of ELL does not exceed 0.8, thus remaining below 1. The estimates appear to be sensitive to the way total costs are computed and the measure of the actual capital stock used. This is a drawback, be- cause measurement errors in capital stock are usually high. The implied labor share in these estimates is around 0.30, which is consistent with our previous results. Transcendental Logarithmic Technology In the two-factor case, the transcendental logarithmic (translog) technology has the advantage over the CES technology that the elasticities are not re- INCREASING LABOR DEMAND IN ARGENTINA 145 TABLE 5-5 Generalized Leontief Technology (Estimates of a, SLL, and EKK) Year E SLL ESx 1974 1.01 -0.71 -0.30 1975 0.99 -0.69 -0.29 1976 0.92 -0.65 -0.27 1977 0.93 -0.66 -0.28 1978 0.96 -0.67 -0.29 1979 0.99 -0.69 -0.30 1980 1.08 -0.76 -0.32 1981 1.08 -0.76 -0.32 1982 1.02 -0.72 -0.31 1983 1.07 -0.75 -0.32 1984 1.13 -0.80 -0.33 1985 1.04 -0.73 -0.31 1986 1.06 -0.74 -0.32 1987 1.04 -0.73 -0.31 1988 1.02 -0.72 -0.31 1989 1.04 -0.73 -0.31 1990 1.03 -0.72 -0.31 1991 1.05 -0.73 -0.31 1992 1.10 -0.77 -0.32 1993 1.11 -0.78 -0.33 1994 1.10 -0.78 -0.33 1995 1.08 -0.76 -0.32 Mean 1.04 -0.73 -0.31 stricted to be constant, but are allowed to vary with the values of the factor inputs or prices. Elasticities based on translog technology are usually esti- mated in a system of factor demands, known as KLEM (capital, labor, en- ergy, and materials). Again, we are constrained by the data at hand because information on energy and other materials is not available. This may intro- duce biases in the estimates we present in Table 5-6. We estimate the cost (11) and share (12) equations joindy using iterated three-stage least squares.8 Table 5-6 presents the estimates of the translog technology imposing different restrictions on the production function. The number of restric- tions increases as we move from left to right. The first column corresponds to the nonhomothetic case, the second to the homothetic, the third to the homogeneous, the fourth to the linear homogeneous, and the fifth to the 146 CRAFTING LABOR POLICY TABLE 5-6 Transcendental Logarithmic Technology Linear Cobb- Nonhomothetic Homothetic Homogeneous homog Douglas In ao -55.824 -72.021 -1.376 -4.462 -4.462 (2.01) (2.49) (2.30) (164.77) (271.54) a, 1.324 0.300 0.297 0.304 0.302 (2.05) (27.53) (27.40) (27.77) (33.54) a2 0.133 0.051 0.099 -0.029 (2.27) (1.02) (2.34) (0.56) b3 23.346 30.421 0.330 (1.96) (2.46) (2.55) a4 -0.223 (1.59) b5 -4.855 -6.404 (1.91) (2.43) Restrictions 0 1 2 3 4 Mean of SLL -0.256 -0.528 -0.368 -0.793 -0.698 Mean of EKK -0.114 -0.232 -0.163 -0.343 -0.302 Mean of a 0.369 0.760 0.531 1.136 1 Mean of ee 0.131 0.172 0.670 0 0 Note: Three-stage least squares estimates. Absolute value of ratio of estimated parameter to standard error in parentheses. Cobb-Douglas restriction. Using the likelihood ratio test, the hypothesis of homotheticity cannot be rejected at the 5 percent level.9 We reject the hypothesis of homogeneity, linear homogeneity, and Cobb-Douglas tech- nology. Under the preferred-homothetic-specification, the estimated average elasticity of labor demand with respect to wages is -0.528, and the elasticity of substitution a is 0.760. The elasticity of total costs with re- spect to output £CY is 0.828. 1- can be used as a measure of returns to scale ee, which takes a positive value if returns to scale exist, and a neg- ative value otherwise.'0 Under the assumption of homotheticity, ECY = ELY, which as a consequence assumes the value 0.828. This is a considerably higher value than our previous estimates. A summary of 42 estimates of INCREASING IABOR DEMAND IN ARGENTINA I47 this elasticity in the United States yields an average of 0.828, exactly the same as our estimate (Hamermesh 1993). For other countries, Hamer- mesh reports a mean of 0.767. Note that these estimates have high standard errors, and hence esti- mates are less precise than in previous cases. Again, this arises from the fact that data on actual costs are not available for Argentina. As can be seen in equation (12), the estimation of elasticities based on translog technology requires information on total cost, and consequently on the stock of cap- ital. As in the case of the generalized Leontief technology, the resulting estimates are sensitive to the specific indexes of capital and labor used and total costs of production. Without national accounts on payments to fac- tors, the best we could do was to use the "estimated" shares of labor, and reindex the data on capital and labor to obtain those shares for 1974. An- other reason for exercising caution in interpreting these numbers is that there is substantial variation in these elasticities over time. -CY fluctuates between -0.626 and 1.408. During 1974-90 it has an average value of 1.1. During 1991-95 it has an average value of -0.469. A negative value for this elasticity is worrisome, since it would imply that total costs fell with the level of output, and that returns to scale are low for low output levels, and very high for higher levels of output. Translog Estimations Incorporating Nonneutral Technical Change This finding points to the possibility of technical change, which we have ignored so far. Technical change implies that the cost function may have shifted down over time. Indeed, Hamermesh (1993) notes that the pres- ence of increasing returns (or economies of scale) may be due to the fact that econometricians cannot distinguish between the effects of labor- saving technical change and increasing returns. To address this issue, we estimated a translog system adjusting for the effects of biased technical change and total factor productivity growth (TFP). Since we have already rejected the nonhomotheticity restriction and do not have enough degrees of freedom to estimate the full model with technical progress, we estimate only two models: the first with homotheticity imposed, and the second which is further restricted to be linearly homogeneous. The results are presented in Table 5-7. We cannot reject the hypothesis that technical change is relatively labor saving and, by homogeneity, cap- ital using. Berndt and Khaled (1979) obtain the same result for the United States, but of a smaller magnitude. This points to the fact that labor- saving technical change is strong in Argentina, which in turn is likely to 148 CRAFTING LABOR POLICY TABLE 5-7 Translog Technology, Nonneutral Technical Change Homothetic Linear Homogenous In ao 11.548 -5.3204 (1.66) (10.46) a, 0.374 0.351 (13.76) (13.57) a2 0.153 0.099 (2.41) (1.70) b3 -5.25 (1-75) b5 1.170 (1.82) b6 0.042 0.131 (10.38) (1.93) b 7 -0.006 -0.004 (2.86) (1 .99) b8 -0.001 -0.004 (7.76) (1.98) Restrictions 1 3 Mean of ELL -0.183 -0.368 Mean of £KK -0.008 -0.159 Mean of SLK 0.264 Mean of ee 0.839 0 Mean of Ecy 0.009 0.001 Note: Three-stage least squares estimates, 1974-95. Absolute value of t-ratios in parentheses. be the result of an increasing price of labor relative to capital. Concerning estimates of elasticities, we find that e LL' KK, and eLK decrease in value when technical change is allowed for. Still, they remain greater than zero in absolute value. Finally, we again compute our measure of returns to scale ee and find that the degree of increasing returns to scale is large. With an estimate of ECY of 0.161, the measure of returns to scale ee equals 0.839. Note that this estimate of the output elasticity of demand ECY is not far from previous estimates with less-structured models. However, for the final part of the sample period (1990-95), the estimated ePC, amounts to 0.29.11 Delfino (1984) estimated a translog KLEM model for the manufactur- ing sector in Argentina for 1950-73, for which National Accounts data were INCREASING LABOR DEMAND IN ARGENTINA 149 available, estimating the shares of labor, capital, and materials. He finds a labor demand elasticity of -0.40 and a capital elasticity of -0.45 resulted. These elasticities are basically in accord with our previous results. However, Delfino (1984) obtains a relatively low estimate of eKL of about 0.24. Policy Implications In this chapter we provide evidence on the own-wage elasticity and the output elasticity of aggregate employment in Argentina. The own-wage elasticity measures the effect of a change in wages on employment, hold- ing output constant. At the risk of oversimplification, its estimate can be used to quantify the effects of labor policy reforms on employment growth, and hence on unemployment. The output-elasticity provides evi- dence on the effect of a change in output on employment, holding the wage or unit labor costs constant. It allows us to obtain numerical esti- mates of employment growth that would occur due solely to GDP growth, that is, without any labor policy reform, which raises or lowers labor costs. The main findings using the different series, technological specifications, and econometric techniques are summarized below. Estimates of Wage Elasticity of Labor Demand The results show that the constant-output wage elasticity of labor demand is significantly different from zero. In the preferred models estimated, it ranges from -0.3 to -0.8, giving a "best-guess" estimate of about -0.5. This would mean that if labor costs are reduced by 10 percent, all else re- maining the same, employment would increase by 5 percent. With an em- ployment and about 4 million people in Greater Buenos Aires, this would amount to an increase in employment of 200,000. As can be seen in TABLE 5-8 Estimated Elasticities: Own-Price, Output, and Substitution Wage Output Elasticity of Technology assumption elasticity elasticity substitution Cobb-Douglas -0.60 1.00 (Restricted) 1.00 (Restricted) Single factor demand -0.45 0 to 0.10 - CES -0.35 0.40 0.50 Generalized Leontief -0.75 1.00 (Restricted) 1.05 Translog -0.30 0.30 0.50 150 'CRAFTING LABOR POLICY Chapter 4, this is greater than the "normal" or trend increase in labor sup- ply, and is likely to lead to a significant and sustainable decrease in the un- employment rate in a relatively short period. If we assume at the outset a Cobb-Douglas specification, we obtain an elasticity of -0.6, given that the share of labor is approximately 0.4 as indicated by available national accounts (for the 1 970s). This lies not far from the results of our estimations. Compared to other findings, our esti- mates of the own-price elasticity are somewhat higher than those reported on average for the United States, where Hamermesh (1993) places his "best guess" based on aggregate data at -0.3 for the constant-output de- mand elasticity of labor. Estimates of Output Elasticity of Labor Demand The output elasticity is restricted to equal 1 under the Cobb-Douglas, CES, and generalized Leontief functions. Using the CES formulation and relaxing the restriction that the elasticity equals 1, the elasticity of output is estimated to be between 0 and 0.5. The estimate depends on the em- ployment and output series used, whether autocorrelation is corrected for, and whether an instrumental variables approach is applied. Using the sin- gle factor demand formulation, the output elasticity is estimated to be 0.2, but is not significant. The best guess, based on these results, would be an output elasticity of about 0.25. This estimate of the output elasticity is somewhat lower than estimates in other countries. Hamertmesh (1993) finds an average elasticity of 0.83 for the United States, while for other countries it is about 0.75. Revenga and Bentolila (1994) obtain lower long-term output elasticities of the em- ployment rate for 11 OECD countries, ranging from 0.21 for Japan to 0.49 for Germany. The estimate for Argentina implies that a 10 percent growth rate in GDP results in a 2.5 percent increase in employment. Table 4.9 sheds light on the years needed to reduce the unemployment rate from 16 percent to 8 percent for different levels of labor force and employment growth. Given an unemployment rate of about 18 percent of a total labor force of about 500,000 in Greater Buenos Aires, and assuming no net growth in the labor supply and an unrealistically high GDP growth rate of 10 percent, it would take about six years to bring the unemployment rate to 8 percent, if the government relies exclusively on output growth. As will be seen in the following chapter, the trend rate of labor force growth actually amounts to 1.5 percent. This implies that pushing the un- employment rate down to 8 percent will take more than 14 years. INCREASING LABOR DEMAND IN ARGENTINA 15I TABLE 5-9 Years Needed to Reduce the Unemployment Rate to 8 Percent (simulations using alternative rates of labor force and employment growth) Labor Force Growth (percent) 0.0 0.5 1.0 1.5 2.0 2.5 0.5 28.0 Employment 1.0 14.0 28.2 Growth 1.5 9.4 14.1 28.3 (percent) 2.0 7.1 9.4 14.2 28.4 2.5 5.7 7.1 9.5 14.3 28.6 3.0 4.7 5.7 7.1 9.5 14.3 28.7 Estimates of the Elasticity of Substitution In the case of Cobb-Douglas technology, the elasticity of substitution be- tween capital and labor is restricted to equal 1. In the CES case, it is re- stricted to a constant, which is estimated to equal 0.5. In the generalized Leontief case, where it is allowed to vary over time, the average elasticity is calculated to be more than 1. In the translog case, the estimated elastic- ity of substitution is about 0.5. In any case, it is clear that the elasticity of substitution between labor and capital in Argentina is significantly greater than zero, but less than 1. At the margin, changes in the price of labor rel- ative to capital will be met by proportional changes in the employment of these two factors. Summary of Findings Our analyses in this chapter produce the following findings. First, own- wage elasticities range from -0.3 to -0.8 for the preferred specifications. Based on these results, the own-wage elasticity of employment-as a sim- ple mean at the aggregate level-can reasonably be regarded as about -0.5, implying that a 10 percent decline in labor costs will, all else remaining the same, raise employment by 5 percent. Second, output elasticities range from 0.1 to 0.4 for the preferred specifications. Based on these results and those from an enterprise survey by FIEL (see World Bank 1996), the out- put elasticity of employment can reasonably be regarded as about 0.25. This implies that a 10 percent growth rate of GDP will result in a 2.5 per- cent increase in employment. Third, better data are required to narrow these ranges, but these estimates are in the general neighborhood of elas- I52 CRAFTING LABOR POLICY ticity estimates for the United States. Our estimates of wage elasticity provide strong evidence that policy reform that lowers the cost of labor relative to capital is urgently needed. To increase labor demand and em- ployment, it is not enough in Argentina to rely exclusively on faster out- put growth. To reduce unemployment, Argentina has little choice but to accelerate the process of labor market reform. Notes 1. For example, the "employment encouragement" contract applied to regis- tered unemployed persons. It was limited to a period of 6 to 18 months and al- lowed an exemption from 50 percent of employers' contributions and a 50 per- cent reduction in severance payments. Full severance payments had to be paid if the contract was terminated early. 2. The policy reforms discussed in World Bank (1996) are reduction of the payroll tax-related distortions, reform of collective bargaining laws, and a re- alignment of severance legislation. 3. Elasticity of substitution measures the ease at which a firm can relax the rel- ative use of the two input factors in response to a change in relative prices, with- out changing the level of output. 4. Self-employment increased by about 19 percent from October 1985 to Oc- tober 1991, which compares to an increase of 11 percent in total employment. From 1989 to 1991 the number of self-employed decreased by about 4 percent. This trend was reversed after the introduction of the Convertibility Plan, imply- ing an increase in self-employment of 1.5 percent between October 1991 and May 1994. During the same period total employment increased by about 2.3 per- cent (see Pessino 1997). 5. Alternatively, total employment hours worked may be used to capture labor demand. Using employment instead of work-hours may raise problems when using cross-section data if there is substantial heterogeneity among hours in dif- ferent firms or industries. Because we use time series data and are interested in es- timating long-term elasticities, this issue does not arise due to the fact that worker-hours are highly correlated with variations in employment over the long term. 6. Our analysis does not differentiate between human capital and labor. Pessino (1997) shows that from 1990 onward supply and real wages of individu- als with higher education increased, leading to increased differences in returns to education. Demand for less-skilled workers decreased during this period. This is likely to distort our results. A lower real exchange rate also means that imported goods are cheaper. Among other things, Argentina imports capital goods. A lower price of capital goods with respect to the higher price of labor implies a strong substitution effect toward capital and against labor if production processes allow INCREASING LABOR DEMAND IN ARGENTINA 153 such adjustments. But here we find evidence of moderate elasticities of substitu- tion between labor and capital. 7. Note that in the above approach, we calculated SLL using the estimate of C and the assumption that sL = 0.6. 8. This is identical to the SURE procedure, but with the fitted values of the endogenous variables regressed on all the instruments and exogenous variables. The disturbance covariance matrix is estimated using the original variables, not the fitted ones. A detailed description of the estimation procedure and the as- sumptions imposed can be found in technical annex B at the end of the book. 9. Notice, however, that the parameter a4 is only marginally significant. How- ever, we cannot reject the null hypothesis of a4 = 0, when we compare its asymp- totic standard error with the critical value of an x2 distribution with I degree of freedom, which assumes 6.63 at the I percent level. 10. Returns to scale p are computed as the inverse of the elasticity of cost with respect to output (1/ eC) and equal 1.208. 11. TFP results are interpreted as follows. In the context of a production func- tion, it has been traditional to measure TFP by the residual method, that is, the growth of output minus the growth in inputs. Berndr and Jorgenson (1975) re- lated such a measure to biased technical change and to total factor cost diminu- tion, the dual of TFP. They showed that under the assumption of constant returns to scale, the rate of TFP growth equals -(alnC)Iit, which represents the one- period reduction in costs due to disembodied technical progress given that prices and output are fixed. In the more general case of nonconstant returns to scale and nonhomotheticity, Berndt and Khaled (1979) show that ey, = qSC,/FCY. The value of 6ct turns out to be positive, meaning that, on average, TFP decreased. This is not surprising if the decrease is concentrated in the 1970s and 1980s, when Ar- gentina underwent a decrease in GDP and investment and collapsed into hyper- inflation. Indeed, the data show that the negative values for TFP are concentrated in the first 10 years until 1984. It then becomes less than 4 percent until the late 1980s, increasing monotonically until reaching almost 9 percent in 1995, which is surprisingly high. With the homothetic model, the behavior of TFP is more in line with our priors: negative values until 1988, zero in 1989, and monotonically increasing after that, reaching 3 percent in 1995. References Berndt, Ernst, 1991. The Practice of Econometrics: Cl.sic and Contemporary Reading, Mass. Addison-Wesley Publishing Co. Berndt, Ernst R., and D. W Jorgenson. 1975. "Energy, Intermediate Goods and Production in an Inter-Industry Econometric Model of the U.S., 1947-7 1." Paper presented at the World Congress of the Econometric Society, Toronto, August. 154 CRAFTING LABOR POLICY Berndt, Ernst R., and Mohamed S. Khaled. 1979. "Parametric Productivity Mea- surement and Choice Among Flexible Functional Forms." Journal of Political Economy 87 (6): 1220-45. Bour, Juan L. 1995. "Mercado de Trabajo y Productividad en la Argentina." Final version. FIEL, Buenos Aires. Delfino, Jose A. 1984. "Analisis economico de la tecnologia del sector manufac- turero Argentino." ("Economic Analysis of Argentina's Manufacturing Tech- nology," with English summary). Economica 30 (2-3): 149-202. Elias, Victor. 1988. "Productividad en el sector industrial argentino: 1935- 1985." ("Productivity in the Industrial Sector of Argentina, 1935-1985," with English summary). Economica 34 (2): 185-202. Hamermesh, Daniel. 1993. Labor Demand. Princeton, N.J.: Princeton University Press. Hopenhayn, Hugo. 2001. "Labor Market Policies and Employment Duration: The Effects of Labor Market Reform in Argentina." Research Network WP No. R_407. Inter-American Development Bank, Washington, D.C. Ministry of Economy. 1995. Informe Economico. Various issues. Argentina. Mondino, Guillermo, and Silvia Montoya. 2000. "The Effects of Labor Market Regulations on Employment Decisions by Firms: Empirical Evidence for Ar- gentina." Research Network WP No. R.391. Inter-American Development Bank, Washington, D.C. Pessino, Carola. 1997. "Argentina: The Labor Market during the Economic Tran- sition." Chapter VI in Sebastian Edwards and Nora Lustig, eds., Labor Mar- kets in Latin America: Combining Social Protection with Market Flexibility. Washington, D.C.: The Brookings Institution. Revenga, Ana, and Samuel Bentolila. 1994. "What Affects the Employment Rate Intenrsity of Growth?" Draft. Varian, Hal. 1992. Microeconomic Analysis. Third Edition. New York and Lon- don: WW. Norton and Co. World Bank. 1996. "Argentina Employment Study." Report No. 15643-AR. Washington, D.C. PART I VTT Earnings and 11 Employment Outcomes in Chile CHAPTERS 6 AND 7 ANALYZE trends in earnings distribu- tion and employment insecurity in Chile between 1970 and 1996. While the chapters do not study the effects of the economic slowdown during 1998-99 and the subsequent recovery in 2000, the analysis remains use- ful for providing a long-term perspective on the effects of labor market lib- eralization and growth on labor-related outcomes. Chapter 6 uses quantitative techniques to analyze whether the rewards to human capital differ by quantile and socioeconomic status, and to quantify the effects of policies to improve access to and quality of educa- tion for poorer sections of society. During 1987-92, average labor earn- ings increased by about 30 percent for the top and bottom quintile groups in Chile, and by about 20 percent for the middle classes. During 1992-94, average earnings fell for the top and bottom quintile groups, but rose modestly for the middle groups. Wage inequality, as measured by the Gini coefficient and the spread between the 90th and 10th percentile groups, rose between 1960 and 1987-88, but declined significantly until the mid-1990s. The authors argue that the change in direction has been missed or underemphasized by earlier studies. Rates of return to education behave in a similar way, rising from 1960 until 1987-88, and then de- clining until 1996. These findings suggest that the Chilean government's emphasis on ed- ucation as an instrument for combating inequality in recent years is ap- propriate, though this is tempered by the finding that rates of return to education are lower for poorer groups. Simple simulations suggest that '55 156 CRAFTING LABOR POLICY further improvements in access to schooling by the poor will have rela- tively modest effects on earnings inequality, and this only at higher levels of education (grade 9 and above). Improvements in the quality of educa- tion appear to be twice as effective as improvements in access to school- ing, and the returns to improvements in education quality are obtained largely at lower levels of schooling (0 to 8 years). Chapter 7 takes up the question of whether Chile's economic reforms (principally a reduced role of government as an employer combined with trade liberalization) have resulted in greater "precariousness" of employ- ment or job instability. This chapter uses a labor flows model to convert reported data into magnitudes that are better suited for addressing this question of precariousness, or volatility, of employment and earnings. The results appear to undermine the claims that employment and labor earn- ings have become more precarious in Chile. During 1990-97, the Chilean labor market became an increasingly reliable source of income: expected job tenure increased, unemployment rates and duration fell significantly, and long-term unemployment became practically nonexistent. The ap- proach followed in the 1980s and early 1990s paid dividends: poverty and wage inequality fell impressively since 1987, and direct interventions in the labor market (such as restrictions on firing or stringent prohibition of temporary contracts) to reduce poverty or inequality did not appear to be necessary. These findings have several important implications for labor market policies. Given the average tenure and unemployment duration prevailing during 1990-96, the mandated severance payment system seemed ade- quate to provide support for most formal sector workers if they become unemployed, though perhaps not suitable for a modern economy that is globally well integrated. Changes in labor legislation that make dismissals harder are likely to prove counterproductive. Additionally, using interna- tional experience as a guide-since Chile's programs have not been rigor- ously evaluated-the authors reason that public training programs are not likely to have a significant effect on earnings inequality of employed work- ers but, if these programs are well targeted, training programs and job search assistance might help some unemployed workers find jobs. The declining rates of return to education computed in Chapter 6 were for the first half of the 1990s. Since 1995-96, these rates appear to have risen somewhat. The economic slowdown in Chile at the end of the decade provided a reminder that mechanisms to provide income support during unemployment spells were still needed, since unemployment rates doubled from 5 to 6 percent in 1997 to about 11 to 12 percent by 2000. EARNINGS AND EMPLOYMENT OUTCOMES IN CHILE I57 The Chilean government has since decided that the appropriate income support program for the formal sector unemployed is a system of individ- ualized severance accounts, along the lines of the much-debated system of mandatory pension funds that Chile pioneered in the 1980s. Proposals to make dismissals more difficult and costly were opposed by private em- ployers, but public sector labor unions appeared to emerge from the decade stronger than before. Chile entered the new decade dealing with proposals to strengthen labor regulations that were initiated during a pe- riod of low unemployment and falling earnings inequality, many of which, however, must be resolved in a time of moderate to high unemployment rates and with signs of rising earnings inequality. 6 Responding to Earnings Differentials in Chile INDERMIT S. GILL AND CLAUDIO E. MONTENEGRO SINCE 1990, CHILE'S GDP PER CAPITA has grown by about 5 percent annually, and reached almost $4,200 in 1996. Open un- employment rates fell from more than 20 percent in the early 1980s to less than 6 percent in 2001. Between 1984 and 1994, average real wages grew by 28 percent. By 1995, aided by a system of targeted transfers, headcount poverty had fallen to about half of what it was in the early 1980s (Haindl 1996). Chile's performance in improving the most basic labor market in- dicators and in reducing poverty is stellar and-outside of East Asia- unfortunately rare. Chile's reliance on market-friendly, outward-oriented economic poli- cies has been credited with these successes. But these successes have also invited closer scrutiny of other outcomes of these policies. In particular, observers have questioned whether economic success as measured by aver- age magnitudes such as per capita income, average labor earnings, and job creation mask important socioeconomic failures, as indicated by growing inequality in the distribution of income and earnings and in the quality of social services such as education and health. In fact, some observers "see a real and frightening possibility of [Chile] in 20 years with a booming economy atop a sea of unskilled, underpaid citizens" (Atwood 1996). Ferreira and Litchfield (1998) provide systematic evidence that the dis- tribution of income did not worsen between 1987 and 1994, so these con- 159 i6o CRAFTING LABOR POLICY cerns appear to be either misplaced or to have arisen as a result of devel- opments in the late 1990s. Their findings indicate, however, that Chile re- mains a highly unequal economy, despite government efforts to provide targeted assistance to the poorest households. In this chapter, we investi- gate the extent to which the functioning of Chile's labor market is re- sponsible for the persistently high inequality in household income. It is believed that in Chile, as in other parts of Latin America, greater openness to trade has led to wider wage differentials between skilled and un- skilled workers. Wood (1997), for example, argues that this has happened in Latin America since the mid-1980s, and attributes it to the entry of China into the world market and to the adoption of new technology that is biased against unskilled workers. These trends seem to contradict both the- ory and the stylized facts from East Asia's experience with openness. The combination of an open economy and a flexible labor market is be- lieved to be the cause of many growing socioeconomic ills, including in- come inequality. It is argued that international trade has led,to widening differentials in earnings between Chile's poor and rich. Leiva (1996) writes that "savage deregulation" of labor markets under Pinochet was followed by a period of "legalized flexibility" since 1987 during which "the new jobs created tended to be low-paying, low-quality, short-term jobs concen- trated in agriculture, trade, services and construction, precisely those sec- tors characterized by greater instability in employment." Leiva suggests that "Chile's flexible-labor market not only deepens income inequality, but also acts as a mechanism for reproducing poverty as employment expan- sion takes place in low-paid, low-quality jobs." Increasing or high inequality of earnings can slow growth and might in- duce reversal of earlier reforms, especially the reform of labor legislation. To assist policymakers in identifying when corrective policy actions might be needed, and what the likely payoff to some policies being contemplated might be, we address three sets of issues: * First, we try to disentangle the links between labor markets, poverty, and inequality-that is, to understand to what extent the labor mar- ket is responsible for progress, or the lack of it, in fighting poverty and inequality. * Second, we analyze the labor market structure as we try to answer whether the labor market changed so that those with human capital (for example, high levels of education) are being rewarded propor- tionately more, and whether the structure of Chile's labor market im- plies that returns to human capital are lower for poorer workers. RESPONDING TO EARNINGS DIFFERENTIALS IN CHILE i6I Third, we analyze labor-related policy issues. Here we try to answer questions such as: Do recent developments warrant significant policy changes? Does the structure of the labor market in Chile warrant di- rect interventions? What are the likely consequences of the principal policies being considered to reduce inequality (for example, intro- duction of unemployment insurance, increases in mandated mini- mum wages, and initiatives to improve the access to and quality of education and training)? We use the latest available data and reliable econometric techniques to deal with these issues. University of Chile Household Surveys from June 1960 to June 1996 in Greater Santiago are used to provide a long-term per- spective on labor market trends, and Encuesta de Caracterizaci6n Socio- economica Nacional (CASEN) household surveys for 1987, 1990, 1992, and 1994 are used to confirm these findings for the country as a whole, and to further investigate these developments. The principal econometric techniques used are quantile regressions, which allow us to analyze returns to education at different points of the conditional wage distribution, and Oaxaca decompositions, used to quantify the effects of policies to improve access to and quality of education for poorer sections of society. Labor Market Issues, Methodology, and Data To keep our task manageable, we analyze only the labor market issues that are critical for explaining earnings inequality, and for evaluating the effects of policies being contemplated to address the concerns regarding income inequality. Main Labor Market Issues In principle, these topics relate directly to actual or perceived increases in inequality of access to employment and to earnings conditional upon being employed. While Chapter 7 addresses the question of whether em- ployment has become more precarious, this chapter deals with the follow- ing topics: * Earnings Inequality Trends. Have earnings become more unequal over time, especially since 1987? * Returns to Education. To what extent have changes in the distribution of earnings been reflected in changes in the returns to education? I62 CRAFTING LABOR POLICY * Levels of Education. Do the poor receive less and lower-quality education? * Policy Implications. Given the structure of the Chilean labor market, what is the likely result of the government's decision to rely on im- proving access to human capital? What would be the result of increas- ing the quality of education as a means of reducing income inequality? The Model and the Methodology The usual empirical human capital earnings function has the following empirical form: In y; = Po + ,13Sj + PA2E + E2+ u (1) where In y; is the log of earnings or wages for individual i; E. is the level of experience, E; is the square of the level of experience (included to ac- count for the commonly observed effect of a declining age earning profile for a given level of experience), ,1 is the rate of return of one additional year of schooling and u. is a random disturbance term that reflects unob- served characteristics.' In this specification, we expect f3] and f2 to be positive since the level of schooling and level of experience are positively correlated to wages for an individual. As educational attainment increases, wages should increase. As level of experience increases, wages should increase as well. We expect ,3 to be negative since the return to experience after a certain level is neg- atively correlated to wages for an individual. Equation (1) is based on some restrictive assumptions. It assumes individuals have equal abilities and equal opportunities,2 overlooks direct costs of schooling, neglects earnings while attending school, and presumes a constant return per year of schooling. We usually refer the estimate of PI as the return to school- ing. However, looking closely at equation (1) we find that the parameter P, is an estimate of the impact of schooling on wages rather than an in- ternal rate of return on investment. If it were an internal rate of return it would be a private one, since this specification disregards school subsidies and omits any positive or negative, externalities to schooling. An area of concern with equation (1) is the omission of the relevant variable, ability. Since ability is likely to be positively correlated with schooling, omitting this variable from the regression equation will bias the estimated returns to schooling upward. Regardless of this problem we choose to ignore measuring ability in our estimations because of the diffi- culty of conceptualizing and measuring it. There is also no consensus as to whether it is significant enough to differentiate earnings. RESPONDING TO EARNINGS DIFFERENTIALS IN CHILE I63 Traditionally, mean regression has been used to estimate equation (1). However, this method gives disproportional weights to large deviations and thus biases estimations. Hence, the method is not accurate when distur- bances are abnormal and when there is a relatively large proportion of out- liers. Furthermore, this method provides only a limited aspect when ana- lyzing changes in returns to education with heterogeneous populations because these changes may involve the entire conditional distribution. As a result of these reasons, study of the so-called 1-estimators has become wide- spread. The traditional mean regression method minimizes- the squared sum of errors and therefore allows one to identify the value of the parame- ters. The minimization implies that the fitted curve is the prediction of the mean of Y (the dependent variable) given the values of a certain vector of independent variables X. However, instead of squaring the sum of errors, it is possible to take their absolute value and to minimtze their sum. In this case the minimization implies that the fitted curve is the prediction of the median of Y given the values of a certain vector of independent variables X. Similarly, this can be regarded as fitting a curve with the implication that half of the errors will be positive and half negative. Consequently, this concept can be extended to the fitting of a curve that implies that 0 per- cent of the errors will be negative and (100-0) percent will be positive, and thus giving rise the quantile regression estimation method. In this context, when we refer to returns to schooling we mean the effect of schooling on the respective quantile of the conditional earnings distribution. The Data Two sources of data were used. Data for the Santiago metropolitan area were obtained from the University of Chile Household Surveys. These are com- parable and representative annual surveys for Santiago. Each survey includes between 10,000 and 16,000 people, and between 3,700 and 5,400 active labor force participants. During this period, Santiago represented about one-third of Chile's total population,3 and a higher proportion of GDP. Data at the national level was obtained from the CASEN for the years 1987, 1990, 1992, and 1994. The CASEN is a national and regional representa- tive survey carried out by the Ministry of Planning, with the objective of generating a reliable portrait of social and economic conditions across the country. Each survey includes between 94,000 and 170,000 people and be- tween 32,000 and 61,000 active labor force participants.4 Computing the same estimations for both data sets has two important payoffs: the Univer- sity of Chile surveys allow us to distinguish short- from longer-term trends, and the CASEN allows us to extend this analysis to the entire country, I64 CRAFTING LABOR POLICY For both surveys the samples include only male workers who worked at least 35 hours per week, which were either empleados (white-collar work- ers) or obreros (blue-collar workers). This implies that self-employed work- ers, domestic servants, and military personnel were eliminated from the sample. Self-employed workers were eliminated because the database did not allow the separation of income into returns to labor and returns to capital; domestic servants were eliminated because they are likely to mis- report labor income due to the fact that live-in domestic servants do not properly report the value of room and board, which is part of their labor income; and military personnel were eliminated because their salaries do not correspond to market productivity criteria. We do not include women in our sample because this would require accounting for their labor par- ticipation decision, which requires a different econometric model. For similar reasons, the unemployed and people who work in voluntary ser- vices are also excluded. The same definitions of variables and samples were used throughout the entire period. This gives us comparable year-to-year (and sample-to-sample) results. The dependent variable is defined as the log of the hourly wage. Are Labor Earnings Becoming More Unequal? To address the question of whether labor earnings are becoming more un- equal, we first reveal trends and changes of average labor earnings from 1987 to 1994 for different decile groups. In a second step, we use two dif- ferent measures-one unconditioned and the other conditioned on edu- cation and other factors-of earnings distribution to determine whether the distribution of earnings has become more unequal. Finally, we provide evidence on levels of and returns to education at different points of the wage distributions and how they evolved over time. Patterns and Trends in Labor Earnings Using decile groups by household per capita income, we first examine the trends in average labor earnings per employed worker. These are reported in Table 6-1. The trends in average earnings conform closely to those found for household income per capita and income per equivalent adult by Ferreira and Litchfield (1998). Between 1987 and 1990, real earnings rose by about 25 percent for the entire sample. Labor earnings for each decile group increased: by approximately 30 percent for the poorest 10 percent of households, by about 20 percent for the median class, and by TABLE 6-1 Average Labor Earnings and Changes, in Percent Workers Aged 14-65 (1994 prices) Decile Average Labor Earnings (thousands ofpesos) Change in Labor Earnings (percent) Group 1987 1990 1992 1994 1987-90 1990-92 1992-94 1987-94 1 22.57 29.09 37.12 35.57 28.87 27.60 -4.17 64.43 2 33.80 44.49 52.11 51.27 31.64 17.13 -1.63 54.19 3 38.21 52.68 59.45 58.85 37.87 12.84 -1.00 55.58 4 48.55 59.12 64.93 66.32 21.77 9.83 2.14 33.75 5 52.21 64.51 72.08 72.22 23.57 11.73 0.19 38.07 6 61.64 70.78 79.73 82.42 14.84 12.63 3.38 29.35 7 68.89 80.80 93.35 94.63 17.29 15.54 1.36 35.51 8 87.17 101.78 117.84 119.20 16.76 15.78 1.15 35.18 9 123.02 141.51 170.32 173.32 15.03 20.36 1.76 38.45 10 322.25 422.84 516.21 494.26* 31.22 22.08 -4.25 60.19 Average 99.73 124.65 144.89 142.32* 24.99 16.24 -1.77 45.28 Ratio of 10:1 14.3 14.5 13.9 13.9 * Indicates that the three households with the highest incomes are excluded. Note: Decile groups are calculated using household per capita income. Source: Authors' calculations using CASEN surveys. i66 CRAFTING LABOR POLICY about 30 percent for the richest 10 percent. The ratio of earnings of the richest to the poorest groups rose marginally to 14.5 in 1990. During 1990-92, real earnings rose at about the same annual rate; that is, by about 16 percent over the two years. Real earnings rose by 28 percent for the poorest group, by about 12 percent for the median group, and by 22 percent for the richest 10 percent. The ratio of earnings of the richest to the poorest groups fell to below 14 in 1992. Average labor earnings for all workers fell by 1.8 percent between 1992 and 1994. Earnings for the lowest three decile groups fell between 2 and 4 percent. The subsequent six decile groups enjoyed modest gains. Fol- lowing the suggestion of Ferreira and Litchfield (1998), we exclude the three households with the highest income.5 Our findings suggest that av- erage earnings of the richest 10 percent of households fell by about 5 per- cent and the ratio of earnings of the richest remained about 14 times higher than the earnings of the poorest group. This is the main difference between the patterns for labor earnings and the trends found for house- hold income per capita by Ferreira and Litchfield (1998), who find that average household incomes fell only for the poorest decile group. In summary, these findings indicate that inequality in earnings re- mained roughly constant, with some redistribution toward the richest 10 percent and poorest 30 percent of households between 1987 and 1992, and toward the middle and upper-middle classes between 1992 and 1994.6 These changes are reflected in a decrease of the earnings ratio of the richest relative to the poorest groups from 1987 to 1994. Measures of Inequality of Labor Earnings The impression obtained from Table 6-1, that inequality did not increase from the mid-1980s to mid-1990s, is also confirmed when we look at other measures of inequality. In what follows we provide evidence on changes in inequality by means of the Gini coefficient and another mea- sure, which Montenegro (1998) called the spread. The Gini coefficient measures inequality as the ratio to the mean of half the average over all pairs of the absolute deviations between people; there are N(N-1)/2 distinct pairs in all, so that the Gini is: PG= N(N - 1) - (2) Note that when everyone has the same, ,p, the Gini coefficient is zero, while if one person has Np, and everyone else zero, there are N-1 distinct RESPONDING TO EARNINGS DIFFERENTLALS IN CHILE 167 nonzero absolute values differences, each of which is N,u, so that the Gini is 1. A natural measure of dispersion is the difference between two quantiles of the wage distribution, as for example the ratio 10:1 presented in Table 6-1. Followinrg this idea and using the results from the quantile regression, we construct our second measure, the spread. The spread is defined as the difference between the 90th and the 10th quantile for given values of the covariates. Hence, in contrast to the 10:1 ratio presented in Table 6-1 and the Gini coefficient, we control for the observed characteristics of the in- dividuals. As mentioned in the section on methodology, the covariates in our specification refer to schooling, experience, and experience squared. We estimate the coefficients on these variable for the 10th and 90th quan- tiles and calculate the respective predicted wages,. evaluated at the mean value of the covariates. The precise definition of the inequality index used here is S = anti log(z2'90 w I) - 1 (3) This measure tells us that the "representative" agent will earn (SP+1)* wage if he or she were in the upper tail of the distribution instead of being in the lower tail (that is, Sp=2 means that the representative agent would earn three times as much in the upper tail as he or she would earn in the lower tail). Table 6-2 lists the two measures of wage inequality that are calculated using the data of the University of Chile. It shows that the Gini coefficient increases from 43 in 1960 to 58 in 1987-88, and decreases to 45 until 1996. Similarly, the 0.90-0.10 spread rises from 2.1 in 1960 to a high of 4.6 in 1987-88, and then declines to 3.3. Both measures, but especially the Gini, exhibit shorter-term sensitivity; for example, the Gini 'drops sharply during "bad times" (for example, in 1974-75 and 1982). The Gini coefficient and spread clearly show that 1987 marks a break from the past in both direction and association with the economy's health: wage in- equality begins to fall during a period of healthy economic growth. By 1996, the Gini coefficient for wage inequality was the same as during the years of the Allende administration, while the 0.90-0.10 spread was only marginally higher. Hence, we can conclude that wage inequality among male workers in Santiago has been marked by a significant decline since 1987.7 Figure 6-1 illustrates the results of Table 6-2.8 Montenegro (1998) also finds the same trends for both more- and less-experienced workers, for workers in both the private and public sectors, and for white-collar work- 168 CRAFTING LABOR POLICY TABLE 6-2 Measures of Wage Inequality Year Gini coefficient Spread 1960 42.5 2.1 1962 45.5 2.5 1969 48.9 2.9 1975 41.1 3.1 1980 49.1 3.3 1982 51.2 3.8 1985 51.5 3.3 1987 57.6 4.6 1990 53.9 4.2 1991 52.4 3.7 1992 47.4 3.8 1993 45.4 3.4 1994 45.9 3.3 1995 46.3 3.2 1996 45.4 3.3 Note: For males working 35 hours or more. Spread is the relative differ- ence between the predicted values of the 0.90 and 0. 10 percentile wage, eval- uated at the mean of schooling, experience, and experience squared. Source: Montenegro (1998) using University of Chile Household Surveys. ers. As can be seen in Figures 6-2 and 6-3, the difference in the level of the two measures relative to the normalization for the entire sample is low for white-collar workers, but high for blue-collar workers. However, when we compare the evolution of the two inequality measures, it is easy to see that the trends in the Gini coefficient and spread are very similar for blue- collar worker. This is true to a much lesser extent for white-collar workers. To check whether our results are robust, we calculate CASEN estimates of wage inequality for 1992 and 1994 and find that they are very similar to University of Chile estimates for the same years.9 Gini coefficients com- puted from the two sources for 1990 differ somewhat: the CASEN esti- mate is about 0.46, while the University of Chile surveys yield a coeffi- cient of 0.53. Although different, this result is not surprising given the fact that the University of Chile survey is basically an urban survey and thus does not include rural areas. This implies that some economic activities, including agriculture, mining, fishing, and forestry, are excluded from the sample. RESPONDING TO EARNINGS DIFFERENTLALS IN CHILE 169 FIGURE 6-1 Trends in Wage Inequality: Whole Sample Spread Gini coefficient 5.5 60 4.5 3.5 45 2.5 -40 35 1.5 30 0.5 I I I I I i I - 25 1960 1965 1970 1975 1980 1985 1990 1995 Source: Montenegro (1998); University of Chile Household Surveys for Greater Santiago. Levels of and Returns to Education Returns to education have been a topic of considerable interest in Chile. It is believed that the rates of return to education have increased because of structural changes in the economy (see for exarmple, Riveros 1990) and because international trade has rewarded educated workers relatively more. We do not attempt to summarize the literature here, but concen- trate instead on the issue of whether returns to education differ across in- come groups, and whether these differences have increased or decreased over time. Here, we pay special attention to the returns to education be- cause it is an important part of the government's strategy to address con- cerns regarding income inequality We start by providing some descriptive statistics on the level of education, and then present our estimates on re- turns to education. Table 6-3 reports the mean years of schooling by decile group (by house- hold per capita income). Not surprisingly, workers from poorer households also have lower levels of education. On average, poorer labor force partici- pants have no more than secondary schooling (about 6 to 7 years), which is a litde more than half the level for the richest group (11 to 12 years). Roughly speaking, this implies that while the average labor force partici- 170 CRAFTING LABOR POLICY FIGURE 6-2 Trends in Wage Inequality: White-Collar Workers 5 5 pread Gini coefficient 60 - . 1 Spread Gini |- 55 4.5 50 3.5 45 255 a 7 '40 35 1.5 30 0.5 l I l I l I l I l I l I l I - 25 1960 1965 1970 1975 1980 1985 1990 1995 Source: Montenegro (1998); University of Chile Household Surveys for Greater Santiago. FIGURE 6-3 Trends in Wage Inequality: Blue-Collar Workers 5 5 pread Gini coefficient 60 . m-Spread - Gini 5 55 4.5 50 3.5 4 ., ~~~~~~~~~~~~~~~45 2.5 40 35 1.5 0.5 25 1960 1965 1970 1975 1980 1985 1990 1995 Source: Montenegro (1998); University of Chile Household Surveys for Greater Santiago. RESPONDING TO EARNINGS DIFFERENTLUS IN CHILE 17I TABLE 6-3 Mean Years of Completed Schooling, Population Aged 14 to 65 Decile Group 1987 1990 1992 1994 1 6.23 6.65 7.10 7.08 2 6.51 7.01 7.36 7.38 3 6.75 7.17 7.53 7.85 4 6.88 7.51 7.75 8.11 5 7.09 7.55 8.06 8.30 6 7.51 7.95 8.46 8.71 7 7.82 8.51 8.95 9.30 8 8.44 9.28 9.56 10.22 9 9.76 10.36 10.57 11.20 10 11.19 11.75 12.10 12.46 Average 7.93 8.57 8.87 9.19 Difference 10:1 4.96 5.10 5.00 5.38 Note: Decile groups are calculated using household per capita income. Source: Authors' calculations using CASEN surveys. pant from the bottom third of the household income distribution has at most some secondary education, the average labor force participant from the top third has completed at least a few years of postsecondary education. Concerning the evolution of years of education from 1987 to 1994, Table 6-4 shows that schooling levels have increased across the board. Average schooling levels increased by a remarkable 1.25 years in the 8 years between 1987 and 1994, and the education levels increased for all groups in every survey year, except for the poorest 10 percent of house- holds between 1992 and 1994. Absolute differences in schooling across groups did not narrow between 1987 and 1994. For example, the differ- ence between the 10th and the 1st quantile increased from 5.0 in 1987 to 5.4 in 1994. However, when we take into consideration that the average increased at the same time, we find that the 10th and the 1st quantiles, as a percentage of the average, have decreased slightly from 0.63 to 0.59. Figure 6-4 shows the results of mean and median regression for Greater Santiago (reported in Montenegro [1998] and reproduced here).'0 It can be seen that changes in the rate of returns to education for full-time work- ing males dosely correspond to changes in wage inequality measures dis- cussed above. That is, the long-term trend is that rates of return rise from 1960 to 1986-87 (from about 11 percent to about 17 percent). They then 172 CRAFTING LABOR POLICY TABLE 6-4 Returns to Education by Schooling Level at Different Points of the Conditional Wage Distribution, Full-Time Male Earners (percent) 1987 1990 Level 0.1 decile 0.5decile 0.9decile 0.1 decile 0.5decile 0.9decile All Spline: 9.8 13.0 15.4 7.3 9.5 14.0 0-4 yrs. 2.8 2.7 0.0 3.0 3.4 6.6 5-8 yrs. 4.0 7.3 10.5 4.7 3.7 5.6 9-12 yrs. 9.3 12.7 15.6 6.6 8.8 15.9 13+ yrs. 17.9 21.6 24.5 15.9 23.5 25.2 1992 1994 Level 0.1 decile 0.5 decile 0.9 decile 0.1 decile 0.5 decile 0.9 deciLe All Spline: 7.2 10.2 14.7 8.5 10.7 14.8 0-4 yrs. 3.6 3.0 4.9 4.6 4.0 7.0 5-8 yrs. 4.3 5.1 6.3 5.4 5.4 9.0 9-12 yrs. 7.3 9.8 14.5 7.9 10.2 13.6 13+ yrs. 15.3 24.4 28.7 16.6 23.4 26.3 Note: Results for 1987 are calculated using monthly earnings; regressions for all other years use hourly earnings. Source: Authors' calculations using CASEN surveys. decline to about 13 percent in 1996. Cyclical patterns are also similar for wage inequality measures: rates of return fall during bad times (for exam- ple, during 1975-76 and 1981-82) and rise during periods of economic prosperity. Once again, 1987 marks a break from this pattern. This find- ing, combined with the close conformity of trends in the distribution of household income and labor earnings discussed in the above section, ap- pears to validate the government's strategy of concentrating on education as the instrument for reducing income inequality. To shed some light on changes affecting the entire distribution, we compute rates of return to education at different points of the conditional wage distribution. The main results of quantile regressions from both Uni- versity of Chile and CASEN data can be summarized as follows: * The rates of return to education are systematically higher for higher quantiles: in Greater Santiago, the rates of returns for males are about 10 percent for the 0.10 quantile, about 12 percent for the 0.25 quan- tile, about 15 percent for the 0.75 quantile, and about 18 percent for the 0.90 quantile regressions (see Figure 6-5). As can be seen in Table RESPONDING TO EARNINGS DIFFERENTIALS IN CHILE 173 FIGURE 6-4 Return to Education: Mean and Median Estimates, Whole Sample Percent 25- 25 - -a- Mean --Mda 20 1 5 10 O 5 1960 1965 1970 1975 1980 1985 1990 1995 Source: Montenegro (1998); University of Chile Household Surveys for Greater Santiago. 6-4, these patterns are confirmed using CASEN data, though rates of return are somewhat lower when the sample is extended to the entire country. In another study, Montenegro (2001) also found very similar patterns in the Chilean case, not only for men, but also for women, with one very important difference: the returns to education at the lower quantiles of the conditional wage distribution are significantly higher for women than for men, but relatively similar in the higher quantiles. * For all quantiles, rates of return are higher for higher education lev- els: CASEN data indicate that rates of returns for 0 to 4 years of ed- ucation are about 3 to 5 percent, rising to 5 to 8 percent, 8 to 15 per- cent, and 15 to 25 percent for groups with 5 to 8, 9 to 12, and 13 or more years of schooling, respectively. * As can be seen in Figures 6-6 and 6-7, the level of returns to educa- tion for blue-collar workers are lower than those of white-collar workers at all quantiles. Furthermore, differences among quantiles are smaller for blue-collar workers. This hints at the fact that white- collar workers constitute a more heterogeneous group than blue- collar workers. Furthermore, the rates of return to education for blue-collar workers reveal a downward trend, while the opposite is true for white-collar workers at all quantiles. Under the assumption that usually people with low levels of education choose blue-collar 174 CRAFTING LABOR POLICY FIGURE 6-5 Return to Education: Quantile Estimates, Whole Sample Percent 25 _0.10 -0.25 v0.75 0.90 20 1 5 10 5 1960 1965 1970 1975 1980 1985 1990 1995 Source: Montenegro (1998); University of Chile Household Surveys for Greater Santiago. FIGURE 6-6 Return to Education: Quantile Estimates, White-Collar Workers Percent 25 20 0.10 0.25 0.75 . 15 10 5 0 1960 1965 1970 1975 1980 1985 1990 1995 Source: Montenegro (1998); University of Chile Household Surveys for Greater Santiago. RESPONDING TO EARNINGS DIFFERENTIALS IN CHILE I75 FIGURE 6-7 Return to Education: Quantile Estimates, Blue-Collar Workers Percent 25 .in-0.10 -_- 0.25 -- 0.75 :-=- 0.90 20 15 10 5 0 1960 1965 1970 1975 1980 1985 1990 1995 Source: Montenegro (1998); University of Chile Household Surveys for Greater Santiago. jobs, this indicates that returns to low levels of education decreased from 1960 to 1995. Finally, as can be seen in Table 6-4, rates of return to education rise with the level of schooling at the first, fifth, and ninth percentiles. While the returns to up to four years of education ranged from 2.8 to zero percent in 1987, those of more than 13 years of schooling ranged from 18 to 25 percent. Similarly, in 1994, returns to educa- tion for the highest education groups were between 3.6 and 5.9 times higher than those of the lowest education group. Differences among returns to education for the highest and the lowest level of education did not increase between 1987 and 1994. Do Children from Poor Families Obtain Less and Lower-Quality Education? In the previous section, we found that poorer worker have lower levels of education and that returns to education are lower in the lower part of the conditional wage distribution. This raises the question of whether workers are poor because they have low levels of schooling, or whether they have low levels of schooling because they come from poor families. I76 CRAFTING LABOR POLICY This section discusses whether children from poor families obtain less and lower-quality education. Concerning the quality of primary education, evidence exists that for any level of schooling up to grade 8, schooling outcomes are lower for children from poorer households. While the poor attend municipal or government-subsidized schools, richer children are more likely to attend private schools. Private school students have significantly higher scores in nationai Sistema de Medicidn de la Calidad de la Educacidn (Educational Quality Measurement System-SIMCE) tests administered in grades 4 and 8, indicating either that they attend better primary schools or that their socioeconomic status makes them more likely to do better. Lower achievement scores reduce the probability of going on to higher levels of schooling. This means that children from municipal or government- subsidized schools are less likely to continue their education. Even if they go on to secondary school and higher levels of education, students from poorer households are more likely to enroll in vocational- technical education at the secondary level (see Cox Edwards and Dar 1995). However, returns to vocational-technical education-when es- timated rigorously-are lower over the working life than returns to humanistic-scientific education (see, for example, Butelmann and Ro- maguera 1993). Again, this hints at the fact that workers from poorer households receive education with lower returns to education. Table 6-5 shows that the parents of students in private humanistic- scientific secondary schools are about twice as schooled as those of students in vocational-technical schools. Parent education level is highest for stu- dents in unsubsidized private schools and lowest in municipal schools. Among the students who attend vocational-technical schools, those who go to corporation schools have relatively better-educated parents. At the same time, evidence on the labor market performance of students from cor- poration schools suggests the quality of education in corporation schools is better. The fact that parent education level is highly correlated with family income provides evidence that children from poorer families attend lower- quality schools. As for access to and type of tertiary education, children from poorer families are less likely to continue with postsecondary educa- tion, or at least to opt for cheaper programs, due to budget constraints. Policy Implications: Reducing Earnings Inequality Through Labor Market and Education Policies Given the importance of labor earnings in household income, and the im- portance of education in determining labor earnings, the link between ed- RESPONDING TO EARNINGS DIFFERENTIALS IN CHILE 177 TABLE 6-5 Education Level of Parents by Type of School Attended by the Child (years of education) Vocational-technical Type General Commerce Industry Agriculture Males Private Unsubsidized 14.2 Subsidized 11.3 7.9 7.6 7.2 Corporation 9.3 7.9 7.3 Municipal 9.9 8.4 6.6 4.5 Females Private Unsubsidized 14.2 Subsidized 11.3 7.3 7.2 4.2 Corporation 8.2 8.3 Municipal 9.2 7.8 7.2 Source: Cox Edwards and Dar (1995). ucational attainment and income inequality is obvious. This link has been explicitly tested by researchers in Chile for an explanation of differences in regional poverty and inequality trends using rate of returns to education. Furthermore, it is also the fundamental reason for the government's belief that education and training are likely to be the best instruments for re- ducing poverty and inequality. In this section we quantify the effectiveness of investments in education in addressing the problem of inequality. Given that even effective education policies will take time to yield results, we discuss which steps could be taken that yield more immediate results. Finally, we address medium- and long-term solutions. Quantifying the Likely Impact of Policy Measures For policy purposes, it is important to disentangle the effects on labor earnings of education-related factors such as education level and education quality. Here, we attempt to do this using a standard technique employed in labor economics that decomposes the earnings differential between rich and poor workers into two parts: (a) the component accounted for by ed- ucation levels, and (b) the component due to differences in rates of return to richer and poorer workers. We provide quantitative estimates of the I78 CRAFTING LABOR POLICY likely payoff-in terms of lower earnings inequality-of the government's initiatives to improve access to secondary education and training for poorer households. Moreover, we shed light on mreasures that increase re- turns to education for poorer workers, such as improvement in the qual- ity of education. To obtain these estimates, we divide a sample drawn from the 1994 CASEN survey into the poor and nonpoor. The sample consists of about 43,000 male workers who reported labor earnings and worked at least 35 hours per week. We use a standard Mincerian human capital earnings func- tion to characterize labor earnings; that is, one that uses completed school- ing-split into splines of 0 to 4, 5 to 8, 9 to 12, and 13 or more years- and potential experience (age minus schooling minus 6 years) as the human capital variables. In the next step we decompose the mean earnings differences between the nonpoor and the poor into the component due to different "endowments" of human capital, and the component due to dif- ferent rewards in the labor market given the same level of human capital. Table 6-6 lists the estimated coefficients and variable means. The main findings are the following. First, the difference in log labor earnings be- tween the poor and nonpoor is about 0.86 pesos, which corresponds to a difference in mean wages between nonpoor and poor (relative to the poor) of 1.38. This difference is reflected in lower levels of education and lower returns to education for any schooling level. Second, providing the poor with the same education levels as the nonpoor, while leaving unchanged the structure of rewards to education and experience faced by the poor, in- creases logwages of the poor by 0.30. Hence, under the assumption that the constant remains unchanged for the poor, their earnings would increase to 11.06. Consequently, increasing the education level of the poor would re- duce the wage gap between the poor and nonpoor by about one-third. Third, providing the poor with the same labor market rewards to education and experience as the nonpoor, while leaving unchanged the education lev- els and years of potential experience of the poor, increases logwages of the poor by 0.41, and implies a reduction in the wage gap of about 50 percent. These results imply that policies that increase education levels of the poor, and policies that increase returns to education, will help reduce the earnings gap between the poor and the nonpoor to a large extent. However, ensuring that the rewards to a year of schooling are the same for the poor and the nonpoor will have nearly twice the effect. Furthermore, much of the gains from improving access to schooling accrue at relatively high lev- els (9 or more years). In contrast, most of the gains from quality- and labor- related policy accrue at lower levels of schooling (0 to 8 years). Based on in- RESPONDING TO EARNINGS DIFFERENTIALS IN CHILE 179 TABLE 6-6 Decomposing the Poor and Nonpoor Earnings Gap and Simulating Payoffs to Alternative Policies Returns Means Policy simulations Education Quality & Variable Poor Nonpoor Poor Nonpoor accessa laborb Schooling Splines: 0-4 years 0.021 0.047 3.60 3.79 0.004 0.094 5-8 years 0.049 0.068 2.58 3.16 0.028 0.049 9-12 years 0.070 0.096 1.11 2.05 0.066 0.029 13+ years 0.025 0.187 0.07 0.59 0.013 0.011 Experience 0.042 0.056 20.93 20.64 -0.012 0.293 Experience Sq.* 100 -0.068 -0.078 665.0 678.4 -0.001 -0.067 Constant 10.052 10.279 Log (earnings) 10.800 11.668 11.06c 11.17c Earnings (pesos) 49,050 116,750 63,720 70,950 Note: All coefficients are significant at the 99 percent confidence level. The independent variable is the log of labor earnings. The sample consists of male workers aged 14 to 65. a. Education access policy payoffs refer to the gains to the poor if the poor are given the same years of schooling as the nonpoor, but the labor market reward structure is left un- changed. Technically, the equivalent of the payoff is the product of poor coefficients and the difference in mean endowments between the nonpoor and the poor. b. Quality and labor policy payoffs refer to the gains to the poor if the labor market re- ward structure faced by the nonpoor is now available to the poor, but their education en- dowments are left unchanged. The technical equivalent is the product.of the endowments of the poor and the difference in coefficients between the nonpoor and the poor. c. For education access, numbers refer to the constant of the poor plus the sum over the products of the poor coefficients and the respective mean endowment of the nonpoor. For quality and labor, the numbers refer to the constant of the poor plus the sum over the products of the nonpoor coefficients and the respective mean endowment of the poor. Source. Authors' calculations using CASEN 1994 data. ternational experience, improving the quality of primary and secondary schools is generally regarded as the best way to increase access to higher sec- ondary and university education (see Ziderman and Albrecht 1995). Con- sequently, our results lead to the straightforward conclusion that education- i8o CRAFTING LABOR POLICY related measures to reduce inequality are best focused on improving the quality of schooling at the primary and (junior) secondary levels. The findings of Riveros (1995) also confirm our results. Using data from CASEN 1992, Riveros imputes that for the poor, the predicted income gain by giving the poor the same quality of secondary education as the nonpoor results in a 50 percent increase in their earnings. For the rich, the increase in earnings was less than 25 percent, which clearly hints at the fact that poorer students benefit more from improvements in school quality. Differences in the quality of education obtained by the poor and the nonpoor may be only one reason why returns to education are low. An- other reason may be labor market discrimination against jobseekers from poorer households. While discrimination against less-privileged jobseekers occurs in every country, its extent is difficult to quantify. It would be best (but not precisely) measured by the residual differences in rewards to human capital after differences in schooling levels and quality between the poor and nonpoor have been accounted for. In countries such as Brazil, India, South Africa, and the United States, observable characteristics such as race, caste, or ethnic origin are normally associated with discrimination. Chile is a relatively homogeneous country, so the extent of discrimination may be statistically difficult to determine. However, anecdotal evidence exists that appears to indicate that discrimination generally takes the form of restricted access-conditional on skill levels-to high-wage jobs for workers from poor households. This differential in access to better jobs would, for any given education level, increase earnings differences. Given that discrimination exists, poor-nonpoor differentials in rewards to edu- cation can be lowered by reducing labor market discrimination against poorer workers. The first best policy to achieve this goal is to outlaw dis- crimination through equal employment opportunity legislation. But these laws are difficult to enforce. Second-best policies, such as affirmative hir- ing practices and minimum wage legislation, may also be considered. Immediate Solutions: Interventions in the Labor Market Measures to improve the quality and accessibility of education, even if im- plemented immediately, will help only those in the school cycle. For most workers, help must come in the workplace. Potential measures to change the distribution of earnings in the immediate to short term include mini- mum wage increases, public training programs, and introduction of an unemployment compensation system. We examine the likely effects of the first two here. Chapter 7 addresses issues relating to severance benefits and the unemployment compensation system. RESPONDING TO EARNINGS DIFFERENTIALS IN CHILE i8i Lopez (1998) estimates that using reasonable values for the relevant parameters such as wage elasticity of labor demand, the impact of the 44 percent rise of the real minimum wage between 1989 and 1994 was a reduction in poverty of about 4.4 percent, or about one-eighth of total poverty reduction."1 However, he cautions that this estimate is for a period when the real minimum wage was relatively low; that is, less than 30 per- cent of per capita GDP The effectiveness of the minimum wage as an instrument to reduce poverty declines as its level increases, because it may lead to an increase in unemployment, and hence may even become counterproductive-that is, it may increase poverty. This concern is strengthened by the strong corre- lation between unemployment rates of workers and the probability of their being in the lowest two decile groups. Lopez (1998) also finds, how- ever, that during 1987-94, increases in minimum wages did not have any discernible effect on the distribution of per capita income, as measured by the Theil Index. These results cast doubts on the usefulness-given cur- rent levels-of increases in minimum wages as a device to reduce income inequality. Combined with the finding that wage inequality has been de- creasing since the late 1980s, and that this can largely be attributed to eco- nomic growth combined with investments in human capital, these results form a strong case for not using minimum wage legislation as an inequal- ity-reducing device. Furthermore, using the University of Chile's surveys, Montenegro and Pages (2001) show that "unskilled workers in general- and the young unskilled in particular-see their probabilities of employ- ment reduced in relation to prime-male skilled workers" when the mini- mum wage increases. Given that the unskilled workers are usually the workers in poverty, this result casts additional doubts on the usefulness of the minimum wage as a device for reducing income inequality. Medium- and Long- Term Solutions: Postprimary Education Access and Quality The results of our policy simulation suggest that improvements in educa- tion access and quality are likely to result in improved income distribu- tion. Consequently, this paper supports the government's strategy of rely- ing on education as a main device for reducing income inequality. Still, there are reasons to be cautious. On the one hand, returns from invest- ment in education access and quality will take some time to accrue. Given that earnings inequality has been falling since 1990, reducing inequality is not as urgent as portrayed in the press and by some researchers. On the other hand, our results suggest that the payoff of improving education I82 CRAFTING LABOR POLICY quality is nearly twice as high as the payoff of increasing access to educa- tion. Improvement in education quality, however, is more difficult to achieve than improvement in access to education. At the secondary school level, perhaps the most direct impact of im- provements in education on poorer segments of the population will come through improved quality of vocational-technical education. Poorer groups constitute a large share of this stream of education, and this is unlikely to change in the near future. As can be seen in Table 6-5, all vocational- technical schools are subsidized, (while some general schools do not re- ceive subsidies), which is encouraging from an equity point of view. Among subsidized schools, vocational-technical schools receive larger per student subsidies, a pattern that is equity enhancing. Improvements in vo- cational education are therefore likely to benefit poorer households more than general secondary education. In this regard, it is important to note that the Estatuto Docente12 of 1991 is perceived as having slowed education reforms (see, for example, Cox Edwards 1997). For example, the statute assures tenure to teachers as long as their school is managed by the municipality. Consequently, teach- ers try to prevent municipal schools from being taken over by corpora- tions, and since 1991 fewer corporation schools have been created. In the meantime, it is believed that the quality of education in municipal schools continues to decline, and the gap in quality between private, unsubsidized and municipal schools remains high. In late 1996 the government con- ceded sizable wage increases to teachers. The effectiveness of this measure in improving the quality of primary and secondary schools-without ac- companying reforms to improve school management-is doubtful. There- fore, while all the evidence points to improvements in education quality at the primary and secondary level as necessary for sustainable reductions in wage inequality, the government faces serious institutional obstacles in making this happen. Given the government's obvious commitment to giving a higher priority to education and training, the remaining barriers are institutional rather than fiscal. And the Estatuto Docente is likely to be the main institutional barrier to improving educational outcomes and, through it, the distribution of earnings. Conclusion Between 1987 and 1992, average labor earnings increased by about 30 percent for the top and bottom quintile groups, and by about 20 percent for the middle classes. Between 1992 and 1994, average earnings fell for the top and bottom quintile groups, but rose modestly for the middle RESPONDING TO EARNINGS DIFFERENTIALS IN CHILE 183 groups. Wage inequality, as measured by the Gini coefficient and the spread between the 90th and 10th percentile groups, rose between 1960 and 1987-88, but has declined significantly since then. This change of di- rection-first pointed out by Montenegro (1998) and confirmed by Ruiz- Tagle (1999)-has been missed or underemphasized by earlier studies. These findings suggest that the government's emphasis on education as an instrument for combating inequality is appropriate; though this is tem- pered by the finding-based on quantile regression techniques-that rates of return to education have decreased slightly since 1987. Simple simula- tions using Oaxaca-type decomposition techniques suggest that further improvements in access to schooling by the poor will have relatively mod- est effects on earnings inequality, and this only at higher levels of educa- tion (grade 9 and above). Improvements in the quality of education ap- pear to be twice as effective as improvements in access to schooling, and these returns to improvements in education quality are obtained largely at lower levels of schooling (0 to 8 years). Using international experience as a guide, public training programs are not likely to have a significant effect on earnings inequality of employed workers. However, if these programs are well targeted, training programs and job search assistance may help some unemployed workers find jobs. This subject is taken up again in Chapter 7. Notes 1. This correspond to the standard Mincer (1974) wage equation. Refinements and revisions of the assumptions can be found in Willis (1986) and Polachek and Siebert (1993). 2. Equation (2) assumes perfect capital and labor markets, which. allows us to take earnings as a proxy for marginal productivity. 3. According to the censuses of 1960, 1970, 1982, and 1992, Santiago ac- counts, respectively, for 32, 35, 38, and 39 percent of the total population in each year. 4. A detailed description of the CASEN can be found in Ferreira and Litch- field (1998). 5. Ferreira and Litchfield (1998) claim that the incomes of the three richest households in the 1994 sample are true outliers. 6. It also appears that autonomous transfers prevented household per capita income from falling for decile groups 2 and 3, despite falls in real earnings per worker. 7. Using the same survey but different measures of income and several other measures of income inequality, Ruiz-Tagle (1999) arrives at the same conclu- sions-that is, that inequality declined significantly since 1987. 184 CRAFTING LABOR POLICY 8. Annex Table 6.A. 1 at the end of this chapter presents the same results for the whole available period and its breakdown by white- and blue-collar workers. 9. Hourly earnings are not available in the 1987 CASEN survey. 10. Annex Table 6.A.2 presents the returns to education by year, subsample, OLS, and QRM estimates. 11. During this period, average salaries rose by 22 percent, or half the rate of increase of minimum wages (Bravo and Vial 1997). 12. Estatuto Docente refers to the Law No. 19.070 (1991), which constitutes the labor regulation for the teaching staff and is a supplement to the Labor Code. References Atwood, Roger. 1996. "Clouds Mar Chile's Sunny Economic Horizon." Business World Update October 19-25: 12. Bravo, D., and J. Vial. 1997. "La Fijaci6n del Salario Minimo en Chile: Elemen- tospara una Discusi6n." Coleccidn de Estudios. CIEPLAN, 45: 117-51. Butelmann, Andrea, and Pilar Romaguera. 1993. "Educaci6n media general vs. tecnica: retorno econ6mico y deserci6n." Coleccidn Estudios Cieplan (38): 5-26. Cox Edwards, Alejandra. 1997. In Indermit Gill, Fred Fluitman, and Amit Dar, eds., Vocational Education and Training Reform: Matching Skills to Markets and Budgets, Chapter 10. Washington, D.C.: Oxford University Press for the World Bank. Cox Edwards, Alejandra, and Amit Dat. 1995. "Technical-Vocational Education in Competition: Evidence from the Chilean Reform of the Early 1980s." Poverty and Social Policy Department. World Bank, Washington, D.C. Draft. Ferreira, Francisco, and Julie Litchfield. 1998. "Calm After the Storms: Income Distribution in Chile, 1987-1994." Draft. Haindl, Erik. 1996. "Poverty and Income Distribution in Chile: A Survey." World Bank, Washington, D.C. Draft. Leiva, Fernando. 1996. "Flexible Labor Markets, Poverty and Social Disintegra- tion in Chile, 1990-1994: The Limitations of World Bank Policies." Paper presented at the World Bank, OXFAM draft. Lopez, Ramon. 1998. "Poverty in Chile: Analyzing the Role of Minimum Wages and Other Factors Using Regional Evidence." Economic Journal. Mincer, Jacob. 1974. "Schooling, Experience and Earnings." National Bureau of Economic Research, New York. Montenegro, Claudio E. 1998. "The Structure of Wages in Chile 1960-1996: An Application of Quantile Regression." Estudios de Economia 25 (1): 71-98. . 2001. "Wage Distribution in Chile: Does Gender Matter? A Quantile Regression Approach." Policy Research Report on Gender and Development No. 20, World Bank Development Research Group. World Bank, Washing- ton, D.C. Processed. Available at www.worldbank.org/gender/prr/wp.htm. RESPONDING TO EARNINGS DIFFERENTIALS IN CHILE I85 Montenegro, Claudio E., and Carmen Pages. 2001. "Who Benefits from Labor Market Regulations and Institutions? Chile 1960-1998." World Bank and the Inter-American Development Bank. Washington, D.C. Processed. Polachek, S., and W. S. Siebert. 1993. The Economics of Earnings. Cambridge: Cambridge University Press. Riveros, Luis. 1990. "The Economic Return to Schooling in Chile. An Analysis of its Long-Term Fluctuations." Economics ofEducation Review 9 (2): 111-121. . 1995. "The Economic Impact of Improving the Quality of Secondary Education in Chile." Report prepared for the World Bank. University of Chile, Santiago. Ruiz-Tagle, Jaime A. 1999. "Chile: 40 Afios de Desigualdad de Ingresos." De- partamento de Economfa, Documento de Trabajo No. 165. Universidad de Chile, Santiago. Willis, R. 1986. "Wage Determinants, A Survey and Reinterpretation of Human Capital Earnings Functions." In Orley Ashenfelter and Richard Layard, (eds.), The Handbook of Labor Economics 1: 525-602. Amsterdam: North Holland- El- sevier Science Publishers. Wood, Adrian, 1997. "Openness and Wage Inequality in Developing Countries: The Latin American Challenge to East Asian Conventional Wisdom." World Bank Economic Review 11 (1): 33-57. Ziderman, Adrian, and Douglas Albrecht. 1995. Financing Universities in Devel- oping Countries. Washington, D.C: The Falmer Press. 186 CRAFTING LABOR POLICY ANNEX TABLE 6A-1 Trends in Hourly Wage Inequality Spread Gini coefficient Total White Blue Total White Blue Year sample collar collar sample collar collar 1960 2.066 2.495 1.734 42.517 39.875 25.776 1961 2.492 3.205 1.959 45.158 40.612 26.619 1962 2.511 3.025 1.940 45.539 40.847 27.413 1966 2.200 3.251 1.822 45.188 43.392 25.738 1967 2.381 2.910 2.080 45.775 43.613 27.526 1968 2.323 3.702 1.938 48.145 45.695 28.541 1969 2.894 3.297 2.499 47.980 45.268 30.369 1970 3.174 3.786 2.119 47.542 44.069 31.436 1971 3.041 3.578 1.809 47.741 42.979 27.968 1972 3.048 3.443 2.262 43.088 39.059 28.640 1973 3.463 3.878 2.989 44.122 41.425 31.200 1974 2.595 3.341 2.044 40.743 40.428 27.354 1975 3.083 3.109 2.566 41.099 39.111 27.726 1976 3.299 3.783 2.278 47.162 44.100 29.495 1977 3.328 3.521 2.378 48.372 44.663 27.956 1978 3.084 3.616 2.186 49.789 46.233 27.885 1979 3.434 3.628 2.436 49.408 45.999 30.543 1980 3.341 3.954 2.293 49.133 45.358 27.948 1981 3.588 3.917 2.435 47.285 46.102 29.057 1982 3.821 4.496 2.207 51.163 48.243 29.582 1983 3.899 4.317 2.446 52.744 48.053 30.555 1984 4.329 4.689 2.367 54.159 48.456 30.686 1985 3.347 3.635 2.066 51.516 46.323 28.994 1986 3.732 3.384 2.039 48.695 44.001 26.855 1987 4.584 4.087 2.125 57.568 53.880 27.630 1988 4.852 4.911 2.453 53.660 51.180 28.499 1989 3.711 3.493 2.243 50.752 48.164 27.240 1990 4.199 5.358 2.014 53.916 51.145 26.294 1991 3.667 4.747 1.975 52.394 49.913 26.260 1992 3.843 3.970 1.804 47.369 46.620 25.820 1993 3.379 4.137 1.981 45.437 42.871 26.027 1994 3.346 3.436 2.043 45.886 44.472 27.078 1995 3.222 3.364 1.890 46.268 43.743 25.235 1996 3.256 3.716 2.152 45.362 42.136 27.305 Source: Unpublished data, Montenegro (1998). RESPONDING TO EARNINGS DIFFERENTIALS IN CHILE 187 ANNEX TABLE 6A-2 Returns to Education Total sample Year OLS Q=0.10 Q=0.25 Q=0.50 Q=0.75 Q=0.90 1960 0.114 0.091 0.101 0.109 0.124 0.138 1961 0.127 0.107 0.119 0.127 0.139 0.145 1962 0.127 0.096 0.111 0.129 0.142 0.151 1966 0.118 0.092 0.098 0.109 0.132 0.151 1967 0.118 0.093 0.102 0.113 0.129 0.146 1968 0.124 0.083 0.101 0.126 0.142 0.160 1969 0.124 0.098 0.106 0.118 0.144 0.149 1970 0.124 0.093 0.110 0.126 0.136 0.152 1971 0.134 0.101 0.121 0.137 0.150 0.159 1972 0.119 0.092 0.117 0.125 0.132 0.131 1973 0.108 0.082 0.095 0.116 0.125 0.120 1974 0.097 0.069 0.081 0.095 0.107 0.114 1975 0.101 0.085 0.089 0.097 0.103 0.120 1976 0.127 0.098 0.107 0.124 0.139 0.145 1977 0.135 0.111 0.119 0.132 0.142 0.152 1978 0.138 0.117 0.116 0.125 0.149 0.160 1979 0.142 0.120 0.122 0.141 0.152 0.163 1980 0.144 0.116 0.126 0.139 0.157 0.154 1981 0.128 0.109 0.116 0.127 0.135 0.144 1982 0.154 0.119 0.130 0.148 0.162 0.191 1983 0.154 0.117 0.132 0.153 0.167 0.176 1984 0.164 0.110 0.147 0.169 0.181 0.197 1985 0.153 0.109 0.131 0.155 0.172 0.170 1986 0.151 0.094 0.126 0.147 0.169 0.187 1987 0.173 0.121 0.128 0.167 0.191 0.200 1988 0.161 0.099 0.130 0.161 0.178 0.193 1989 0.150 0.090 0.115 0.143 0.169 0.179 1990 0.152 0.094 0.104 0.138 0.182 0.195 1991 0.149 0.088 0.110 0.144 0.166 0.184 1992 0.136 0.079 0.095 0.122 0.158 0.173 1993 0.131 0.083 0.102 0.126 0.147 0.165 1994 0.135 0.078 0.107 0.132 0.163 0.161 1995 0.135 0.080 0.101 0.139 0.159 0.162 1996 0.137 0.090 0.111 0.136 0.152 0.159 (Table continues on the following page.) I88 CRAFTING LABOR POLICY ANNEX TABLE 6A-2 (continued) White-collar workers Year OLS Q=0.10 Q=0.25 Q=0.50 Q=0.75 Q=0.90 1960 0.097 0.076 0.083 0.093 0.112 0.119 1961 0.111 0.087 0.099 0.120 0.124 0.136 1962 0.110 0.076 0.090 0.115 0.136 0.137 1966 0.117 0.079 0.091 0.119 0.132 0.156 1967 0.118 0.095 0.095 0.117 0.134 0.142 1968 0.127 0.087 0.114 0.130 0.151 0.164 1969 0.120 0.099 0.103 0.116 0.136 0.146 1970 0.117 0.093 0.103 0.122 0.124 0.140 1971 0.126 0.105 0.123 0.125 0.139 0.147 1972 0.113 0.094 0.116 0.123 0.126 0.114 1973 0.106 0.078 0.102 0.116 0.120 0.107 1974 0.103 0.081 0.099 0.103 0.108 0.112 1975 0.096 0.075 0.090 0.090 0.102 0.118 1976 0.127 0.087 0.118 0.128 0.146 0.143 1977 0.128 0.097 0.119 0.122 0.138 0.156 1978 0.137 0.103 0.113 0.135 0.157 0.156 1979 0.153 0.125 0.139 0.152 0.160 0.164 1980 0.140 0.105 0.116 0.149 0.156 0.149 1981 0.145 0.134 0.143 0.148 0.138 0.169 1982 0.168 0.146 0.151 0.165 0.191 0.206 1983 0.162 0.121 0.146 0.171 0.172 0.178 1984 0.174 0.130 0.178 0.170 0.180 0.188 1985 0.158 0.145 0.164 0.172 0.170 0.155 1986 0.143 0.110 0.127 0.156 0.164 0.177 1987 0.199 0.153 0.170 0.211 0.221 0.222 1988 0.187 0.170 0.181 0:186 0.200 0.220 1989 0.170 0.129 0.149 0.176 0.194 0.202 1990 0.187 0.132 0.149 0.201 0.224 0.212 1991 0.173 0.102 0.153 0.187 0.205 0.201 1992 0.160 0.114 0.134 0.171 0.185 0.185 1993 0.153 0.097 0.131 0.163 0.175 0.178 1994 0.165 0.132 0.156 0.176 0.186 0.180 1995 0.162 0.099 0.135 0.174 0.180 0.182 1996 0.160 0.145 0.154 0.174 0.178 0.161 RESPONDING TO EARNINGS DIFFERENTIAL IN CHILE 189 ANNEX TABLE 6A-2 Blue-collar workers Year OLS Q=0.1O Q=0.25 Q=0.50 Q=0.75 Q=0.90 1960 0.057 0.058 0.063 0.054 0.052 0.058 1961 0.058 0.074 0.065 0.053 0.057 0.051 1962 0.056 0.045 0.045 0.062 0.063 0.069 1966 0.049 0.052 0.044 0.050 0.046 0.054 1967 0.046 0.036 0.035 0.051 0.057 0.055 1968 0.050 0.045 0.047 0.048 0.056 0.065 1969 0.054 0.058 0.052 0.053 0.062 0.070 1970 0.054 0.048 0.055 0.063 0.057 0.052 1971 0.049 0.028 0.043 0.051 0.059 0.059 1972 0.058 0.061 0.062 0.065 0.050 0.054 1973 0.037 0.013 0.035 0.047 0.043 0.049 1974 0.038 0.043 0.040 0.034 0.042 0.031 1975 0.043 0.058 0.045 0.038 0.040 0.033 1976 0.045 0.038 0.037 0.045 0.042 0.051 1977 0.053. 0.060 0.054 0.052 0.047 0.053 1978 0.057 0.060 0.056 0.060 0.053 0.063 1979 0.057 0.057 0.053 0.049 0.065 0.076 1980 0.057 0.081 0.056 0.037 0.047 0.034 1981 0.039 0.043 0.033 0.048 0.044 0.033 1982 0.057 0.062 0.065 0.056 0.061 0.072 1983 0.059 0.054 0.052 0.061 0.051 0.051 1984 0.047 0.031 0.038 0.066 0.068 0.067 1985 0.051 0.027 0.031 0.062 0.069 0.072 1986 0.056 0.044 0.054 0.061 0.064 0.063 1987 0.041 0.019 0.032 0.041 0.048 0.064 1988 0.054 0.042 0.042 0.053 0.069 0.064 1989 0.051 0.046 0.038 0.043 0.055 0.071 1990 0.033 0.018 0.021 0.039 0.046 0.049 1991 0.049 0.037 0.042 0.042 0.060 0.058 1992 0.052 0.037 0.039 0.051 0.059 0.078 1993 0.043 0.031 0.035 0.037 0.062 0.058 1994 0.048 0.033 0.036 0.057 0.057 0.052 1995 0.042 0.034 0.040 0.038 0.044 0.047 1996 0.054 0.029 0.046 0.051 0.066 0.068 Source: Unpublished data, Montenegro (1998). 7 Dealing with Employment Instability in Chile INDERMIT S. GILL, ERIK HAINDL, CLAUDIo E. MONTENEGRO, AND CLAUDIO N. SAPELLI THIS CHAPTER ADDRESSES the question of whether the combination of an open economy and a flexible labor market increased uncertainty in employment and earnings in Chile. Some authors, such as Leiva (1996), argue that since 1987 "the new jobs created tended to be low-paying, low-quality, short-term jobs concentrated in agriculture, trade, services and construction, precisely those sectors characterized by greater instability in employment." Concerns about increased precarious- ness of employment were commonplace even before the Asian and Rus- sian financial crises, and are likely to gain even more attention today. Although earnings inequality in Chile did not become worse in the 1990s, and indeed improved over the last decade (see Chapter 6), em- ployment may nevertheless have become more precarious. That is, while workers enjoy higher earnings across-the-board, these earnings are be- lieved to be more unreliable. Indeed, it is common to hear that Chilean workers are more likely to lose their jobs-and related benefits-now than at any point in Chile's history. Moreover, if employment in low-paying jobs became more precarious, this change might largely affect poor peo- ple. These concerns have fueled the debate over both the need for greater mandated job security and an unemployment insurance system (see, for example, Sapelli 1996; and Coloma 1996), and have even been used as an argument for erecting trade barriers. I9I 192 CRAFTING LABOR POLICY In this chapter we provide evidence on the evolution of expected tenure and unemployment duration from 1962 to 1995. We use data from the University of Chile Households Survey and a model developed by Haindl (1985), which converts reported labor market stock data into flows. We will present basic issues and describe the methodology and the data used, discuss changes in expected tenure and unemployment duration and the relation evolution of unemployment rates across different income quan- tiles, and conclude with a discussion of policy implications. Issues, Methodology, and Data To assess whether employment has become more precarious in recent years, we use data from the University of Chile Household Surveys from 1962 to 1995. These are comparable annual surveys for Santiago. Each survey covers between 7,000 and 16,000 people, and between 3,700 and 5,400 active labor force participants. The sample is periodically updated to reflect the growth of the Greater Santiago Area, and the questionnaire has been modified in only minor aspects, making it possible to rely on fairly comparable information. The survey is conducted in June each year, and contains information on labor force participation (LFP) and employ- ment during the week preceding the survey. It does not provide informa- tion on worker tenure. Consequently, the data set reveals direct evidence on stocks, such as the number of people employed or unemployed, but not on flows into and out of unemployment.' Employment becomes more precarious if the probability of an em- ployed worker losing his or her job increases, or if the probability of an unemployed worker finding a job decreases. The first translates into a re- duction of the expected tenure on the current job (that is, the stability of employment). Accordingly, a decrease in the probability of finding a job corresponds to an increase in unemploymnent duration. Thus, the question of precariousness of employment refers to a change in employment status and, consequently, the relevant labor indicators measuring precariousness are flows, not stocks. Because our data set does not contain information on the relevant flows, we use a labor flows model, which converts data on current period stocks (for example, employment and unemployment) into current period flows. Furthermore, it allows us to calculate the probability of los- ing one's job when employed, and finding a job when unemployed, and the relevant prospective indicators, such as expected tenure and unem- ployment duration. The model was developed by and explained in Haindl (1985). See Technical Annex D for a summary. DEAUNG WITH EMPLOYMENT INSTABILITY IN CHILE 193 To understand the functioning of a labor market, looking at flows is much more instructive than looking at stocks. For example, a 5 percent unemployment level is consistent with 200,000 people (out of a popula- tion of 4 million) unemployed during one year; but it is also consistent with 3.2 million people working during the whole year and 800,000 peo- ple that are unemployed for an average of three months; and is also con- sistent with the 4 million workers unemployed for an average of 18 days a year. Obviously, the implications for the unemployed people are differ- ent in each case. Hence, to see how difficult it is to find a job in a given period, we need to complement the information given by the unemploy- ment rate with the expected time that people will spend unemployed, or with the probability of finding a job in a given period. The labor market is usually characterized by large flows of workers into or out of employment. Those unemployed consist of workers who either quit their job or were fired. When the flow of people entering employment exceeds the flow of people leaving employment, the employment level in- creases. In addition, the people who are unemployed are not the same throughout the year. On the contrary, people are continuously entering and leaving the pool of unemployed. Hence, unemployment increases if the flow of the number of people into unemployment exceeds the flow of people out of unemployment. Furthermore, there is also a continuous flow of people who are entering or leaving the labor force. The interaction of these flows drives the patterns of the unemployment and employment lev- els, and hence the unemployment rate. In mathematical terms we can de- scribe the evolution pattern in terms of a difference equation that relates the change in the stock variables to the relevant flows. Solving the system of equations in finite differences, we obtain flows from the stock variables. The interrelations among the flow variables and the stock variables in the labor market can be presented in a schematic diagram. Figure 7-1 shows that the employment level at the end of a period is equal to the stock of the people employed at the beginning of the period, plus the number of people hired during the period, minus the number of people leaving employment. In the same way, the unemployment level at the end of a period is equal to the number of unemployed people at the beginning of the period, plus the number of people that entered the unemployment pool minus the number of people that left unemployment. People who enter the unemployment pool are either people who quit their jobs or were fired, or people who entered the labor force but have not yet found employment. People who leave unemployment may either have found a job or have left the labor force. 194 CRAFTING LABOR POLICY FIGURE 7-1 Labor Force Flows OUT OF LABOR FORCE Flow of people who Flow of people into the labor force leave the labor force P ~~~UNEMPLOYMIENT LEVELi Flow of people hired Flow of people who were fired or quit their job EMPLOYMENT LEVEL Flow of people Labor Supply who change jobs Labor Demand ~~~~~~~~~~~~~~~~~~~~~~~~~~~ PRODUCTION OF GOODS AND SERVICES Goods and services dema-nd DEALING WITH EMPLOYMENT INSTABILITY IN CHILE I95 Has Chile's Employment Grown More Precarious? Precariousness of employment may be estimated by expected tenure on current job and unemployment duration. Expected tenure captures the stability of employment, which critically depends on the incidence of un- employment, that is, the probability of losing one's job. Unemployment duration, that is the time a worker is likely to be without work, depends on the probability of a worker exiting unemployment. This is estimated as the probability of finding a job in a fixed period of time, using the sam- ple of unemployed jobseekers. Is Expected Tenure Decreasing? To give a first insight about the evolution of tenure from 1962 to 1995, Figure 7-2 plots the expected tenure (in months).2 A first glance reveals that expected tenure increased until 1969, followed by a large downward trend, which was reversed at the end of the 1980s. Expected job tenure de- clined from more than seven years in 1969 to less than three years in the early 1980s. It increased to more than six years in 1992, and remained over four years in the following years. On average, expected tenure seems not to have decreased during the last years. Often concerns about increased precariousness of employment are based on statistics that show falling average tenure (on current job), which is not confirmed in our data. Moreover, it is true that changes in average tenure may be driven by structural and legislative changes in the economy, such as less restrictive firing laws or an increased creation of low-quality, short-term jobs. But this is only one part of the story. Business cycle fluc- tuations and changes in the composition of the labor force by age and gen- der may matter as well. Additional information on employment flows and labor force entrants is displayed in Table 7-1. Table 7-2 sheds light on the changes in the LFP of women. Recessions usually lead to a decrease in employment. This arises from the fact that although during recessions the number of quits decreases, lay- offs usually increase by a much higher amount. Furthermore, flow into employment decreases as fewer new jobs are created. Blanchard and Dia- mond (1990) show that for the United States, recessions are associated with large increases in job destruction, but small decreases in job creation. This suggests that we expect tenure to decrease during recessions. Chile faced two major recessions in 1975 and 1982-83, which reduced GDP growth significantly below its long-term trend. As can be seen in Table 7- 1, I96 CRAFTING LABOR POLICY FIGURE 7-2 Expected Job Tenure Months 100 90 80 70 60 50 40 30 20 10 0 1960 1965 1970 1975 1980 1985 1990 1995 2000 Source: Authors' calculations using University of Chile Household Surveys for Greater Santiago. employment decreased during the two recession years by more then 7 per- cent. These changes were largely driven by an increase in the flows out of employment. Tenure fell from 7.5 years in 1969 to 3.8 years in 1975, and reached its lowest point during the 1982 recessions, at 2.8 years. Changes in labor force composition may also affect expected tenure. For example, it is reasonable to expect that mean tenure of women is lower than that of men. Consequently, an increase in the participation of women in work is likely to decrease average tenure. As can be seen in Table 7-2, women's LFP rate increased from 26.4 in 1987 to 33.7 in 1994. The trend decline in expected tenure until 1985 may in part be ex- plained by a steady increase in first-time job seekers, which rose from about 25,000 to 100,000. Furthermore, the age composition of the labor force mattered. Pages and Montenegro (1999) show that in 1987 about 70 percent of male workers aged 15 to 25 had been on the job less than two years. The same was true for only 35 percent among male workers aged 26 to 50, and to 20 percent among male workers aged 51 to 65. This indi- cates that for men there exists a strong link between tenure and age. In 1996 the percentage of low-tenure workers was slightly higher among all age groups. From 1987 to 1990 mean tenure decreased, especially for older workers. This is likely to have negatively influenced average tenure. DEALING WITH EMPLOYMENT INSTABILITY IN CHILE 197 TABLE 7-1 Estimates of Labor Force Entrants and Employment Flows, 1962-95 Labor force entrants Employment flows First Total (% Net (% Expected time Reentrant of LFa) Outflow Inflow of totala) Tenure Year ('000) ('000) (%.)a ('000) (C00) (%1o)b (months) 1962 26 5 4.0 128 159 4.2 69.3 1965 30 20 6.2 104 152 6.2 89.3 1969 45 0 4.6 123 166 4.6 89.7 1975 73 -39 3.0 273 199 -7.2 45.1 1980 91 11 6.0 267 365 8.4 52.3 1982 70 -46 1.6 452 356 -7.5 34.0 1985 99 -10 2.0 324 433 8.2 49.0 1987 76 -22 3.2 389 438 3.2 46.6 1988 88 -39 2.8 451 525 4.8 41.5 1989 78 21 5.5 404 490 5.4 48.7 1990 74 -15 3.1 540 590 2.9 38.2 1991 62 -17 2.3 395 497 5.8 53.8 1992 94 44 6.9 293 432 7.4 76.6 1993 79 -51 1.3 491 512 1.0 49.1 1994 105 -19 4.0 496 556 2.9 49.1 1995 85 5 4.0 - - - 54.8 - Not available. Note: Calculated using Haindl (1985) labor flow model. a. As percentage of last year's labor force. b. As percentage of last year's employment. Source: Authors' calculations using University of Chile Household Surveys for Greater Santiago. However, from 1985 onward, employment rates for all age groups in- creased due to a period of sustained growth, but employment rates for older and middle-age groups increased at a faster rate. This should have had a positive affect on expected tenure. In another paper, Montenegro and Pages (2001) show that the gender composition of the working labor force also matters: women tend to have, on average, less tenure than men. This fact, plus the fact that female participation in the working labor force has been steadily increasing over time (Montenegro 1998), should have had a negative effect on expected tenure. To summarize, we find that expected tenure does not appear to be falling anymore. While expected tenure did fall until 1982, it has exhib- I98 CRAFTING LABOR POLICY TABLE 7-2 Labor Force Participation Rates, Females Aged 14 to 65 Decile Group 1987 1990 1992 1994 1 12.3 14.7 12.8 15.3 2 13.1 16.3 15.7 18.7 3 14.4 16.2 18.8 21.9 4 17.1 20.7 24.8 25.1 5 20.1 24.5 27.9 29.4 6 24.1 30.0 33.3 33.8 7 28.9 33.1 37.2 39.6 8 31.9 35.1 40.0 41.8 9 37.2 41.5 44.2 46.6 10 54.0 53.1 57.9 57.0 Average 26.4 29.3 32.2 33.7 Ratio 10:1 4.4 3.6 4.5 3.7 Note.: Decile groups are calculated using household per capita income. Source: CASEN surveys. ited an upward trend since then. This is true even though female LFP in- creased during this period and the percentage of low-tenure workers was slightly higher among young, prime-aged, and older workers, which is likely to have affected expected tenure negatively. While it is legitimate for workers, policymakers, and nongovernmental organizations to be con- cerned about job permanency, these findings provide a compelling argu- ment that changes in labor legislation to reduce involuntary turnover may not be necessary. Proponents of reform of labor laws may thus be trying to. fix a problem through government mandates that the market appears to have already fixed in the decade of economic prosperity since 1987. Is Unemployment Duration Increasing? Employment does not only become more precarious if expected tenure on a current job decreases, but also if unemployment duration, that is the time a worker is likely to be without work, increases. Frequent job changes do not necessarily imply unstable labor earnings, given that workers who change jobs may not enter the unemployment pool. Instability in earnings hence is determined by the probability of becoming unemployed, and the probability of exiting unemployment. To examine whether Chilean work- ers have experienced increases in instability of earnings, we analyze the DEALING WITH EMPLOYMENT INSTABIUTY IN CHILE I99 changes in average (expected) unemployment duration and the probabil- ity of exiting unemployment. The results are reported in Table 7-3 and plotted in Figures 7-3 and 7-4. The main findings are: Unemployment rates-plotted in Figure 7-3-rose steadily from 5 percent in 1962 to about 16 percent during 1974-75, fell until 1980, and peaked again at about 22 percent during 1982-83. Since then, unemployment rates have fallen sharply, and appear to have stabilized at about 7 percent. If unemployment rates were a guide to the precar- iousness of labor earnings, the last 15 years have seen a sizable decline in precariousness. CASEN surveys since 1987 show that while unem- ployment rates are higher for workers from poorer households (see Table 7-4), trends in unemployment appear to be the same for all workers. TABLE 7-3 Estimates of Unemployment Rates and Duration, 1962-95 Probability of Finding Job in: Expected Rate I mo. 3 mos. 6 mos. 1 yr. Length Year (%0) (%) (%) (%) (%) (months) 1962 5.1 26.7 60.6 84.4 97.6 2.7 1969 6.2 20.1 48.9 73.9 93.2 4.0 1975 16.1 9.6 26.2 45.5 70.3 9.4 1980 11.8 16.7 42.2 66.6 88.9 5.0 1982 22.1 8.9 24.5 43.0 67.5 10.2 1985 16.4 12.7 33.6 55.9 80.5 6.8 1987 12.2 15.6 39.8 63.8 86.9 5.4 1988 10.9 19.8 48.4 73.4 92.9 4.0 1989 9.1 21.3 51.3 76.2 94.4 3.7 1990 9.6 23.1 54.6 79.4 95.7 3.3 1991 7.4 23.4 55.0 79.8 95.9 3.3 1992 6.0 25.0 57.9 82.3 96.9 3.0 1993 6.3 26.8 60.9 84.7 97.7 2.7 1994 6.8 26.2 59.8 83.8 97.4 2.8 1995 6.8 26.1 59.7 83.8 97.4 2.8 Note: Calculated using Haindl's labor flow model. Unemployment rate is averaged for March, June, September, and December. Source: Authors' calculations using University of Chile surveys for Greater Santiago. 200 CRAFTING LABOR POLICY FIGURE 7-3 Unemployment Rate and Duration, Greater Santiago Rate/Months 22 - | Rate 12 | Length in months 17 - 7 2 1960 1965 1970 1975 1980 1985 1990 1995 2000 Source: Authors' calculations using University of Chile Household Surveys for Greater Santiago. FIGURE 7-4 Probability of Finding a Job, Greater Santiago -|- Month 1 - Month 3 E-a- Month 6 - Month 12 Probability 0.97 - -----_ 0.95- 0.93- 0.91 0 .8 9 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 1962 1969 1975 1982 1987 1990 1995 Source: Authors' calculations using Universiry of Chile Household Surveys for Greater Santiago. DEALING WITH EMPLOYMENT INSTABILITY IN CHILE 201 * Since 1962, unemployment duration has risen from less than three months, and it reached its highest level of 10.2 months during the recession of 1982-83. Since then it has declined steadily, reaching 2.8 months in 1995. As can be seen in Figure 7-3, unemployment duration closely tracks the pattern of unemployment rates. This hints at the fact that changes in unemployment rates are due rather to fluc- tuations in the time needed to find employment (that is, unemploy- ment duration), and to a lesser extent to changes in the likelihood of losing one's job in the first place (that is, unemployment incidence).3 * Table 7-3 and Figure 7-4 also provide evidence on the probability of finding employment within a certain time period. A first glance at Figure 7-4 reveals that the probability of finding a job dropped largely during the recession years. From 1582 the probability of find- ing a job increased continuously and came to a halt in 1993. For ex- ample, the probability of exiting unemployment in three months or less (that is, an average duration of about 1.5 months) was about 6 percent in 1962, falling to about 30 percent during 1982-83 and ris- ing to about 60 percent again. * The probability of a worker joining the ranks of the long-term un- employed (defined as being unemployed for more than one year) was negligible in 1962, and after rising significantly in the 1970s and early 1980s, declined to close to zero in 1995. These findings provide strong evidence that the labor market has become an increasingly re- liable source of earnings for labor force participants. Precariousness and the Poor An important question for policy reasons is whether unemployment is a phenomenon largely experienced by the relatively rich. But unemployment of workers-especially primary earners-may be an important determi- nant of a household's relative position in the per capita income distribu- tion. Marcel and Solimano (1994) suggest that changes in unemployment rates are important in explaining short-term fluctuations in the income distribution through losses in the share of the poorest households. Cowan and de Gregorio (1996) point out that, almost tautologically, unemploy- ment is highest among those with lowest incomes. Even for the employed, a rise in informal work during bad times leads to a deterioration of work conditions. Therefore, before discussing the policy implications of any findings on the problem of precarious employment, we briefly examine whether un- 202 CRAFTING LABOR POLICY employment rates differ for rich and poor households, and whether fluc- tuations in unemployment rates differ significantly for wealthy and poorer households. A person is considered unemployed if he or she did not work last week and looked for a job. Hence, we are not able to distinguish be- tween short-term and long-term unemployed. Unemployment rates are calculated using the microdata sets of the Caracterizacidn Socioecondmica Nacional (CASEN) for the years 1987, 1990, 1992, and 1994.4 Tables 7-4 and 7-5 report unemployment rates for men and women. During 1987-92, average male unemployment rates fell steadily from al- most 9 percent to 4.7 percent, and then rose to about 6 percent in 1994. Female unemployment rates are higher, but exhibit similar trends. They amounted to 10.5 in 1987 and decreased to 7.1 in 1992. Furthermore, Ta- bles 7-4 and 7-5 shed light on unemployment rates for different decile groups of household income. As expected, unemployment is significantly correlated with being in a poorer household for both men and women: workers in the poorest 10 percent of households are almost four times more likely to be unemployed than the average worker. This-ratio does not TABLE 7-4 Average Unemployment Rates, Men Aged 14 to 65 (percent) Decile Group 1987 1990 1992 1994 Changea 1 33.0 26.3 16.9 21.1 -36.1 2 13.9 11.3 7.7 8.1 -41.7 3 10.8 9.7 6.8 7.4 -31.5 4 9.6 8.6 4.2 6.7 -30.2 5 7.4 6.3 3.8 6.0 -18.9 6 7.5 6.2 4.3 4.5 -40.0 7 6.4 3.5 2.9 3.7 -42.2 8 5.3 3.8 2.8 3.3 -37.7 9 3.5 2.8 2.1 1.9 -45.7 10 1.4 1.9 0.7 1.9 35.7 Average 8.9 7.1 4.7 5.9 -33.7 Ratio 1: Average 3.7 3.7 3.6 3.6 Note: Unemployment is defined as, "Did nor work last week and looked for a job." Decile groups are calculated using household per capita income. a. Change from 1987 to 1994 is 80 percent with respect to 1990, but 500 percent with respect to 1987. Suurce: CASEN surveys. DEALING WITH EMPLOYMENT INSTABILIrY IN CHILE 203 TABLE 7-5 Average Unemployment Rates, Women Aged 14 to 65 (percent) Decile Group 1987 1990 1992 1994 Changea 1 38.7 35.2 28.1 36.8 -4.9 2 27.3 25.7 19.5 26.0 -4.8 3 22.0 16.9 16.6 20.3 -7.7 4 16.0 12.5 13.6 13.9 -13.1 5 14.5 10.8 7.3 10.3 -29.0 6 10.0 12.5 7.6 9.4 -6.0 7 6.2 7.2 6.1 7.4 19.4 8 7.9 5.2 3.9 7.0 -11.4 9 6.1 4.4 3.7 4.1 -32.8 10 3.8 1.2 0.1 1.8 -52.6 Average 10.5 9.2 7.1 9.7 -7.6 Ratio 1: Average 3.7 3.8 4.0 3.8 Note: Unemployment is defined as, "Did not work last week and looked for a job." Decile groups are calculated using household per capita income. a. Change refers to percent change from 1987 to 1994 with respect to 1987. Source: CASEN surveys. change much over the years, which may indicate that unemployment is al- ways critical in determining household income-the variable used to clas- sify households into decile groups. From 1987 to 1994 the average unemployment rate for men decreased by 33.7 percent, and for women by 7.6 percent. Looking at the different income deciles it can be seen that the change in unemployment ranged from 30 to 46 percent for most decile groups. For the highest decile group the unemployment rate increased, mainly driven by the change from 1992 to 1994. The median decile group realized a relatively low decrease in un- employment rate of only 18.9 percent. For women, the upper three and the fourth and fifth decile groups faced the highest decrease in unem- ployment rates. The lowest three decile groups also faced a reduction in unemployment rates, but to a lower extent: among these groups percent- age changes in unemployment rates range from 4.8 to 7.7 percent. Short-term fluctuations in unemployment rates by decile group defy simple characterizations, however. For men, unemployment rates decreased from 1987 to 1992 among all decile groups. Changes in unemployment rates range from 37 percent for the third decile group to 56 percent for the 204 CRAFTING LABOR POLICY fourth decile group. Women faced a negative change in unemployment rates throughout this period. However, the range of changes among decile groups is much larger among women. It extends from a 1.6 percent de- crease for the sixth decile group to a 97 percent decline for the highest decile group. The bottom three decile groups faced a decline in their un- employment rates of approximately 26 percent. The overall decrease in unemployment rates was reversed from 1992 to 1994. Unemployment rates increased for all decile groups, with the only exception being the eighth decile group of men. For men, the. increase in unemployment rates was lowest for the sixth decile group (4.6 percent) and highest for the ninth decile group (171.1 percent), while women faced in- creases ranging from 2.2 percent (fourth decile group) to 795 percent (sixth decile group).5 During the entire time period, no clear relation is visible between decile group and change in unemployment rate. However, the overall increase in unemployment rates from 1992 to 1994 may well have given rise to con- cerns. The question that needs to be answered is whether this requires changes in labor legislation related either to dismissals or unemployment compensation. Policy Implications: Reducing Precariousness Through Labor Market and Education Policies Since the early 1980s, expected tenure has increased, while unemploy- ment rates have declined steadily, and duration of unemployment has fallen significantly. Long-term unemployment-defined as being unem- ployed for more than one year-has declined to negligible levels. Future labor market reforms may require introducing unemployment insurance schemes and improving the education system for those who have difficul- ties finding jobs. Is an Unemployment Insurance System Needed? Concerns regarding the precariousness of employment have fueled a de- bate over whether an unemployment insurance system is needed in Chile. From the evidence on labor flows presented above, it appears that these concerns are largely unfounded. The probability of a worker being with- out earnings is low: the unemployment rate was less than 6 percent in 1997. For the average worker, unemployment duration was less than three months in 1995, and has been falling over the last five years. Long-term DEALING WITH EMPLOYMENT INSTABILITY IN CHILE 205 unemployment was not pervasive: more than 60 percent of job seekers find employment in less than three months, and almost all find work within a year. These results are consistent with the findings of Edwards and Cox Edwards (2000), who show that Chile experienced both a re- duction in persistence and in the natural rate of unemployment. In cases of unjustified dismissal, the labor code ensures that workers get a sever- ance payment of one monthly wage for each year of tenure, with a maxi- mum of 11 months. With expected tenure at 4.5 years and the expected duration of unemployment at three months, at least on average, therefore, this severance payment seems to be adequate in helping job seekers main- tain their standard of living while looking for work. But there is a demand for income support during unemployment even in labor markets where unemployment rates are low, and Chile's unem- ployment rates climbed into the double digits in 1999. As pointed out in Chapter 1, severance payments may be a rather inefficient and inequity- enhancing means of providing financial support to the unemployed. Transforming severance payments into a deferred compensation scheme would increase efficiency of the labor market. In 1990, the Aylwin gov- ernment introduced the option of an unemployment fund for workers with tenure above six years, as an alternative to the job security protection. If the workers chose the unemployment fund the employer made a monthly contribution of 4.11 percent of taxable monthly wages to a sav- ings account bearing the worker's name. While employees have no access to a severance package if dismissed with just cause, they can always with- draw the money accumulated in the unemployment fund independent of the reason for dismissal. In 2001, Chile replaced the traditional severance pay arrangements with individual severance accounts, essentially mandat- ing self-insuranice (with little pooling of risk), thus eschewing altern atives that involve substantial pooling of risk, such as conventional unemploy- ment insurance programs like those in all OECD countries. But for countries like Chile that have relatively balanced labor laws, the strong prejudice against insurance schemes that involve pooling of unem- ployment risk among formal sector workers should be reconsidered. The economic rationale for this is that pooling makes sense when the unem- ployment risk faced by the median worker is low and noncovariate. This appears to have been the case in Chile until the late 1990s (unemploy- ment rates rose from 6.1 percent in 1998 to 9.7 percent in 1999, before falling to about 9.0 in 2001). In our view, there are few strong theoretical reasons to think that Chile would be wrong to consider some pooling of unemployment risks, given its strong insurance fundamentals, that is, zo6 CRAFTING LABOR POLICY comprehensive macroeconomic reforms, generally balanced labor legisla- tion, and strong administrative capacity. (See de Ferranti and others 2000 for a discussion of these concepts.) If a private market for insuring against unemployment were to arise under the circumstances generally prevailing in Chile (low long-term unemployment, low to moderate incidence of un- employment), it would probably offer an instrument of the pooled type. So there is an argument that the government should correct this market failure with a similar instrument (the alternative being mandatory self- insurance such as individual accounts). See Table 7-6 for a summary. We could still justify arguments against such schemes for practical rea- sons, one of them being administrative capacity concerns. But adminis- trative capacity can always be built, and it is generally built while-not before-implementing a public intervention. Besides, the Chilean gov- ernment has shown that it can competently set up and manage social assistance schemes that, by most accounts, are more administratively de- manding than social insurance programs. Another argument against conventional unemployment insurance pro- grams is moral hazard-that is, the possibility that an employee might en- gineer a job separation to gain access to unemployment benefits. Two points are relevant here. First, any insurance scheme runs the risk of moral hazard that reduces labor market efficiency, but policymakers deal with this problem all the time because their public policy objective is welfare, not efficiency, maximization. Second, moral hazard can be minimized by keeping benefit levels low or by making insurance premia reflect risk lev- els by, for example, experience-rating.6 The presence of a large number of informal workers creates a problem for any formal unemployment insurance (UI) scheme. The most obvious one is that it becomes difficult for the UI administrator to determine whether a claimant is in fact unemployed, or simply working in an un- registered job. This is not strictly a moral hazard problem because it is not legally abusing the system. But it is a legitimate concern. Countries such as Argentina have tried to deal with this problem by making collecting these benefits costly in terms of time (for example, claimants have to collect payments in person, during peak business hours, and face long queues). The other problem may be one of adverse selection, where the low-unemployment-risk workers may choose to be informal, while high- risk workers choose to be formal. What is likely to be a more serious issue is that a formal UI system could discourage informal UI mechanisms. Guasch (1999) recommends combining individual savings accounts and conventional unemployment insurance with a part of worker contri- TABLE 7-6 Income Support Programns for the Unemployed: Findings and Policy Implications Nature of Advantages Measure instrument and disadvantages Policy implication Individual Self-insurance-no Low labor market efficiency costs, but Should be considered by countries that saving accounts pooling of risk welfare reduction especially for poorer have high unemployment, especially workers. where labor reforms are only a distant possibility. Severance pay Pooling over Almost no advantage. Little Possibly the worst form of unemployment small group- pooling of risk, entails labor market support in a globalized economy. globalization inefficiency, makes labor relations makes group contentious, and is administratively even smaller challenging. Public works Market-type Can reach informal sector workers and Should be considered for a part of and training .insurance the poor, but can entail high leakages workforce, but not a universal scheme. programs elements-implicit in the form of nonlabor costs when Permanent schemes allow for better pooling of risk investment element is made a priority. balance between consumption Training programs show less leakage smoothing and investment over the but also lower coverage potential. economic cycle. Unemployment Market-type Most pooling of risk can be used to Should be considered by governments insurance insurance-explicit address both idiosyncratic and aggregate that have carried out comprehensive pooling of risks risk, and hence serve as automatic fiscal economic reforms; labor market disin- stabilizer, generally politically popular. centive effects can be reduced by keep- May be administratively demanding. ing benefits frugal and mimicking the market as much as possible in design. Source; de Ferranti and others (2000). zo8 CRAFTING LABOR POLICY butions going toward a general fund to compleinent individual accounts of workers who have not reached levels of saving that would allow mini- mum unemployment benefits when dismissed from work. It appears that Chile has adopted precisely such a system. Since it is unlikely that even the strongest administrative efforts will considerably reduce informality in Chile, and so a good fraction of the workers will not be covered by the contributory insurance system, the gov- ernment should also think of other insurance mechanisms for such work- ers. Public works programs are the main candidates, but there may be oth- ers (such as transfers conditioned on participating in job search assistance programs or training programs). Obviously, these schemes will be non- contributory and hence financed through general taxation, but perhaps the government should be encouraged to think of such programs as social in- surance and not assistance (the main difference being that one tends to think of the latter as temporary-for example, existing in bad times-and the former as more permanent-that is, existing in both good and bad times). There is, of course, a significant antipoverty element in these pro- grams: de Ferranti and others (2000) report that Argentina's work-condi- tioned transfers and Mexico's training-conditioned transfers appear to help mainly labor force participants from the poorest 40 percent of households. There is also the option of providing cash or in-kind transfers to peo- ple below the poverty line: since unemployment appears to be particularly pernicious for the lowest two decile groups (see Tables 7-4 and 7-5), this would help the poorest among the unemployed. Another approach used in the past that proved to be quite successful was to provide low-wage work for the poor through, for example, the Programa de Empleo Minimo (Minimum Employment Program) and the Programa Ocupacional para Jefe de Hogar (Occupational Program for Households Heads), rather than to subsidize unemployment. Finally, the most disadvantaged job seekers, especially those who have been unemployed for long periods, may lack marketable skills or may require assistance with a job search. This provides an argument for helping job seekers acquire relevant skills through public retraining programs, and assisting them with job placement. Are Government-Sponsored Training Programs Helping the Poor? The Ministry of Labor's Serv'icio Nacional de Capacitacidn y Empleo (SENCE) is cited worldwide as an example of good practice in encourag- ing in-service training and in helping unemployed workers find jobs (see Gill, Fluitman, and Dar 1999). SENCE regulates the tax credit for train- DEALING WITH EMPLOYMENT INSTABILITY IN CHILE 209 ing, and also plays an active role in funding training courses for target groups (Cox Edwards and Dar 1995). In 1995, 430,000 people were trained under the tax credit scheme at a cost of $35 million ($24 million public, and $11 million private). The redistributive aspects of SENCE's activities are: * Tax Credit System. The highcr the salary of the trainee, the lower the tax credit to firms for training. For salaries below 10 minimum wages, the tax credit is for 100 percent of the training cost. For salaries above this, the tax credit is only 50 percent of the cost, and this ratio falls further as the salary increases. * Scholarships System. SENCE provides training scholarships for spe- cific groups, such as youth aged 15 to 24, ex-convicts, and rural women. These programs appear to help the relatively disadvantaged segments, but it is not clear that the participants are the worst off. In 1996, for examiple, about 55 percent of those participating in the training program for youth, Chile Joven, had completed at least 12 years of schooling, and another 27 percent had completed 9 to i 1 years of schooling. * Apprenticeship Schemes. Since 1994, SENCE has subsidized about 10,000 apprenticeships for disadvantaged job seekers in firms. Ini- tially, the government provided 100 percent of the wages during the apprenticeship period, which was about 8 to 10 months. Since 1995, SENCE has provided 40 percent of wages for the apprentices, with the firms paying the remainder. In general, SENCE's programs appear to be well designed and effi- ciently administered. However, none of them has been rigorously evalu- ated. The tax credit system is likely to be equality enhancing, but we are not aware of any systematic evaluation of the incidence of these expendi- tures. Similarly, it is difficult to determine whether the scholarship system is effective. Based on a survey of Chile Joven, Paredes (1996) recommends that the program be evaluated to resolve the tension between targeting the most vulnerable and ensuring high effectiveness. The same is true for the apprenticeship system. The program is quite new, and as a consequence no systematic evaluation exists yet. Rigorous evaluation of training programs and other similar labor mar- ket interventions is seldom found-not only in Chile, but across the world. Even in OECD countries, almost half of all evaluations are non- scientific and few contain a rigorous analysis of costs (Dar and Gill 1998). This is surprising because public retraining schemes are usually an impor- 2IO CRAFTING LABOR POLICY tant component of the total package of assistance for those laid off due to plant closure or restructuring. Furthermore, they include programs to as- sist other unemployed people. The main benefit of rigorous evaluations is that they are critical in resolving the contradiction between equity and ef- ficiency objectives. (See Gill and Dar [1995] for an illustration using eval- uations of public training programs in Hungary.) Scientific evaluations, when they are conducted, show poor results for retraining programs. Some programs resulted in modest gains in reem- ployment probabilities, but wage changes are generally insignificant. Eval- uations indicate that retraining programs are generally no more effective than jogi search assistance (JSA) in increasing either reemployment prob- abilities or earnings. Finally, retraining programs appear to be two to four times more expensive than JSA. Since JSA yields outcomes in terms of earnings and reemployment probabilities that are not worse than those of retraining programs and is also cheaper, it is more cost-effective than re- training in assisting displaced workers to get jobs. To illustrate the importance of doing rigorous analysis, we use Hun- gary's experience with retraining programs. A nonrigorous evaluation of Hungarian public training programs would have led to increased support for programs that recruited young and educated job seekers from better- off regions that have high placement rates and earnings. Rigorous evalua- tion yielded a different result. In Hungary, public training programs ap- pear to be substitutes for attributes that lead to higher reemployment probabilities in the absence of any intervention-for example, being younger, more educated, and from more dynamic regions. That is, the program's value added (in terms of improving labor market outcomes) is greater for relatively disadvantaged job seekers, such as older, less-educated workers from worst-off regions. Consequently, targeting job seekers who are relatively disadvantaged in terms of age and education, or who come from relatively backward regions, would appear to better serve both equity and efficiency objectives than simply ensuring support for programs whose retrainees have high reemployment probabilities and earnings gains.7 Concerning the effect of public training programs on earnings, ag- gregate results of the evaluation of Hungarian public training programs show that when differences in characteristics of program participants and control groups (such as age, sex, education, and skills) are controlled for, retraining is at best marginally successful. This leads to the conclusion that public training programs do not increase earnings. However, because re- training programs might increase reemployment probabilities, they might marginally reduce volatility- of earnings. DEALING WITH EMPLOYMENT INSTABILITY IN CHILE 21I In summary, experience in both OECD and middle-income countries does not appear to justify large-scale training programs for the unem- ployed. Rather, it indicates that retraining programs are more beneficial for some groups of job seekers than others, and hence should be carefully targeted to specific types of job seekers. However, it is difficult to predict who will benefit most from retraining. Since public training programs are expensive, modest pilot programs might be implemented first, and then rigorously evaluated. Retraining can then be tightly targeted to those for whom it is found most cost-effective. For both equity and efficiency rea- sons, Chile would do well to institute rigorous evaluation mechanisms for SENCE's training programs. Summary and Conclusions Concerning the evolution of expected tenure the main results are, first, ex- pected job tenure declined from about seven years in the 1960s to three years in the mid-1980s; since 1987, however, expected tenure has risen to 4.5 years and appears to be increasing. Thus while concerns of declining expected job tenure (increased precariousness) appeared to be well founded between 1960 and 1982-83, they are unwarranted since 1982-83. Sec- ond, in part, the trend decline in expected tenure until 1985 can be ex- plained by a steady increase in first-time job seekers, the number of which rose from about 25,000 to 100,000. First time job seekers can reasonably be expected to be more likely than reentrants to change jobs. With a de- cline in the absolute number of labor force entrants, it is likely that ex- pected tenure will continue to rise. Third, cyclical fluctuations in labor de- mand may result in shorter-term fluctuations in expected tenure. Thus, for example, expected tenure fell during the recession years of 1975-76 and 1982-83. Years of economic prosperity are associated with increases in ex- pected tenure, and this basic relationship has not changed over the years. Analyzing the changes in expected unemployment duration and the probability of exiting unemployment, we find that: a. Unemployment rates rose from 5 percent in 1962 to about 16 per- cent during 1974-75, and peaked at about 22 percent during 1982-83; since then, unemployment rates have fallen sharply to less than 7 percent by 1997, before rising to about 10 percent in 2000. b. Unemployment duration rose from less than 3 months to about 10 months in the two decades since 1962, and has since fallen to below 3 months. The duration of unemployment tracks unemployment 212 CRAFTING LABOR POLICY rates closely, implying that there have not been significant long-term changes in the incidence of unemployment. c. The probability of exiting unemployment in 3 months or less (that is, an average duration of about 1.5 months) was about 60 percent in 1962, falling to about 30 percent during the 1982-83 recession, and rising to about 60 percent again. d. The probability of a worker joining the ranks of the long-term un- employed (defined as being unemployed for more than a year) was negligible in 1962, rose significantly in the 1970s and early 1980s, but declined to almost zero by 1995. These findings indicate that the Chilean labor market has become an increasingly reliable source of earnings for labor force participants, and while expected tenure fell until 1982-83, it has exhibited an upward trend since then. The duration of unemployment rose until the late 1980s, but has fallen sizably since then. Unemployment rates decreased from 1982-83 on, increased from 1992 to 1994, fell again between 1995 and 1998, and spiked upward due to the global economic crisis in late 1998. These findings provide a compelling argument that government initiatives to address the problem of earnings instability for individual workers may be needed, but these should be instead of, and not in addition to, the mandated severance payments system that existed until 2000. A welfare- enhancing action by the government might be to provide facilities to in- sure against unemployment. In 2001, the Chilean government introduced an unemployment insurance program, which largely relies on mandated self-insurance. Notes 1. For more information on the data set see Chapter 6. 2. Even though it would be very interesting to see the tenure calculations con- ditional on labor force characteristics such as gender or age, this was impossible to achieve given the small size of the sample. Using a different Chilean survey, Pages and Montenegro (1999) and Montenegro and Pages (2001) look at this topic in more detail. 3. The expected unemployment duration is different from the one reported in Edwards and Cox Edwards (2000) due to the different methodologies used (Ed- wards and Cox Edwards used a duration model instead of a flow model). 4. The first CASEN survey was conducted in 1985, but it is widely reported to be less reliable; that is, it is of inferior quality than and less comparable with subsequent surveys. DEALING WITH EMPLOYMENT INSTABILITY IN CHILE 213 5. The increase in the 10th percent decile group of women exceeds 795 per- cent; however, it is not mentioned here because it is very susceptible to outliers. 6. Experience-rating means that unemployment tax rates vary according to the frequency with which an employer's workers have filed for unemployment bene- fits. This transfers the -problem of determining unemployment risk onto the em- ployer, but makes employees who have frequently filed unemployment insurance benefits less attractive to future employers. 7. Generally, training programs should not be targeted to a specific group. Which group or groups should benefit from training programs might depend on such factors as the country and its educational system. For example, in Mexico training programs appeared to help educated workers more than other groups. References Blanchard, Oliver Jean, and Peter Diamond. 1990. "The Cyclical Behavior of the Gross Flows of U.S. Workers." Brookings Papers on Economic Activity 2:85-142. Coloma, Fernando. 1996. "Seguro de desempleo: teoria, evidencia y una pro- puesta." Cuadernos de Economia 33 (99): 295-320. Cowan, Kevin, and Jose de Gregorio. 1996. "Distribution and Poverty in Chile Today: Have We Gained or Lost Ground?" Ministry of Finance, Chile. Draft. Cox Edwards, Alejandra, and Amit Dar. 1995. "Technical-Vocational Education in Competition: Evidence From the Chilean Reform of the Early 1980s." Poverty and Social Policy Department, World Bank, Washington, D.C. Draft. Dar, Amit, and Indermit Gill. 1998. "Evaluating Retraining Programs in OECD Countries: Lessons Learned." The World Bank Research Observer 13 (1): 79-101. de Ferranti, David, Guillermo Perry, Indermit Gill, and Luis Serven. 2000. Se- curing our Future in a Global Economy. Washington, D.C.: World Bank. Edwards, Sebastian, and Alejandra Cox Edwards. 2000. "Economic Reforms and Labour Markets: Policy Issues and Lessons from Chile." Economic Policy: A European Forum 0 (30):181-217. Gill, Indermit, and Amit Dar. 1995. "Costs and Effectiveness of Retraining in Hungary." Report No. IDP-155, Internal Discussion Paper, Europe and Cen- tral Asia Region. World Bank, Washington, D.C. Gill, Indermit, Fred Fluitman, and Amit Dar, eds. 1999. Skills and Change: Con- straints and Innovation in Reform of Vocational Education and Training. Wash- ington, D.C.: World Bank. Guasch, Jose Luis, 1999. "An Alternative to Traditional Unemployment Insur- ance Programs: A Liquidity-Based Approach Against the Risk of Earnings Losses." Unpublished note. World Bank, Washington D.C. Haindi, Erik. 1985. "Un modelo para la determinaci6n de flujos y parimetros dinamicos en el mercado del trabajo." Estudios de Economia 12, Marzo. Uni- versidad de Chile. 2I4 CRAFTING LABOR POLICY Leiva, Fernando. 1996. "Flexible Labor Markets, Poverty and Social Disintegra- tion in Chile, 1990-1994: The Limitations of World Bank Policies." Paper presented at the World Bank, OXFAM draft. Marcel, Mario, and Andres Solimano. 1994. "The Distribution of Income and Economic Adjustment." In Barry P Bosworth, Rudiger Dornbusch, and Raul Laban, eds. The Chilean Economy: Policy Lessons and Challenges. "Washington, D.C.: The Brookings Institution. Montenegro, Claudio E. 1998. "The Structure of Wages in Chile 1960-1996: An Application of Quantile Regression." Estudios de Economia 25 (1): 71-98. Montenegro, Claudio E., and Carmen Pages. 2001. "Who Benefits from Labor Market Regulations and Institutions? Chile 1960-1998." World Bank and the Inter-American Development Bank, Washington, D.C. Processed. Pages, Carmen, and Claudio E. Montenegro. 1999. "Job Security and the Age- Composition of Employment: Evidence from Chile." Office of the Chief Economist, Working Paper No. 398. Inter-American Development Bank, Wash- ington, D.C. Paredes, Ricardo. 1996. "Evaluaci6n Programa Chile Joven." Universidad de Chile, Santiago. Sapelli, Claudio. 1996. "Modelos para pensar el mercado de trabajo: Una revisi6n de la literatura chilena." Cuadernos de Economia 33 (99): 251-276. 8 Some Guidelines for the New Decade: Summary and Conclusions INDERMIT S. GILL AND CLAUDIO E. MONTENEGRO IN THE LAST DECADE, many Latin American countries have undertaken far-reaching economic reforms. In many countries the direct role of government in production has been scaled back through pri- vatization of state enterprises and functions. The region has led the world in reforming social security. Almost all countries have eased restrictions on capital flows within and across countries. Even land reform, while not widespread, has been carried out in some countries. Largely absent from this litany of achievements is far-reaching labor reform to match the changes in how product and factor markets now operate. The lack of far-reaching or comprehensive labor reform should not be viewed as arising from a judgment by Latin Americans that labor reform is relatively unimportant for economic outcomes, though that claim, too, is sometimes made. There are few things that affect the welfare of indi- viduals and households as directly and deeply as efforts to obey and avoid labor market rules and regulations. The apparent lack of success at the policy level in vigorously addressing labor issues is accompanied by widely expressed concerns about labor market outcomes: opinion polls in the region reveal, for example, that 3 out of every 10 Latin Americans re- gard poor employment and wage prospects as the most important prob- lem facing them. Many important labor market indicators, such as un- employment rate, informal sector size, and wage differentials, worsened 215 2I6 CRAFTING LABOR POLICY during the 1990s. The coexistence of widely expressed concern with pol- icy inaction appears to stem from disagreement on what changes would help, and the particularly difficult political economy of reform in this area. To narrow the divergence of views somewhat and move the debate from fiery rhetoric to cooler calculations of the payoffs and tradeoffs involved in labor reform, this book advocates a quantitative approach to sorne of the main labor market concerns facing countries in the region. Two points deserve mention here. First, while neither quantitative labor economics nor its application to analysis of developing countries is new, this book breaks some new ground in applying such analysis to a wide range of labor policy issues in developing countries. Second, while this book examines labor policies in only three countries in Latin America (Argentina, Brazil, and Chile), and hence provides policy recommendations for just the prob- lems faced by these countries, the policies illustrate techniques and ap- proaches that might be applicable in other countries in the region, for other times and, indeed, even in other parts of the developing world. Argentina, Brazil, and Chile in the Regional Context Following are some of the principal labor policy issues faced by Argentina, Brazil, and Chile in the 1990s, and their prospects as they entered the new decade. * Argentina faced a high level of unemployment-about 15 percent- for which both labor demand and supply factors could be responsi- ble. Under the macroeconomic constraints facing the country, labor legislation is identified as a major policy area for reform. * Brazil confronted high costs of public employment that have led to large government deficits and public debt. Onerous levels of taxa- tion, combined with problems of labor legislation, are central to the phenomenon of widespread and growing informality, which-by one measure-accounted for more than 50 percent of employment. * At the time of the study, Chile had policies that led to a decade of wage and employment growth, but which a large number of Chileans believe have led to growing income inequality and uncer- tainty of employment. For those not content with high average wages and low unemployment, labor reform held high promise as an effec- tive remedy. At the end of 1999 high unemployment rates of about 10 percent had compounded these concerns. SOME GUIDELINES FOR THE NEW DECADE: SUMMARY AND CONCLUSIONS 2I7 Readers familiar with other countries in the region and around the world will recognize some or all of these phenomena-high unemploy- ment, public-private labor market segmentation, informality of employ- ment, growing earnings inequality accompanying economic growth, and increased uncertainty of employment even when accompanied by low lev- els of unemployment.' Data compiled by the Economic Commission for Latin America and the Caribbean (ECLAC), ILO, IDB, and the World Bank suggest that these three countries are not unusual within the region in terms of broad labor market indicators in the 1990s: * Real wage growth during 1990-97 in the sample countries spans the full range in the region from the lowest (Brazil with almost -6 per- cent) to the highest (Chile with 32 percent). * In terms of employment growth, the sample countries have both rel- atively low employment growth (Argentina with 0.4 percent) and rel- atively high employment growth (Chile at 3 percent). * In terms of labor force growth, the sample countries are in the low range in the region: Brazil with 1.7 percent and Argentina and Chile with 2.1 percent. * In terms of unemployment and female participation rates, the sam- ple countries are again not different from the rest of Latin America. The particular labor market issues facing the three countries studied here during the 1990s are closely related to their general economic condi- tions. First, all three have reduced annual inflation rates from above 1,000 percent to less than 10 percent, shifting the focus from just macro imbal- ances to a mix of micro and macro concerns. This shift results in height- ened concerns about factor markets, including labor. Second, as countries move beyond stabilization and through fiscal adjustment, the nature of the concerns becomes predominantly microeconomic. Third, as fiscal ad- justment is completed and the economies hit a period of sustained growth, the concerns are almost exclusively related to micro imbalances. This shift in focus from macro to micro is illustrated in this book by examining three countries at different macroeconomic stages. The three countries studied faced different macroeconomic challenges in the 1990s. Brazil was studied during a stage of stabilization initiated in 1994, while it was still struggling to begin a serious fiscal adjustment. Ar- gentina was midway through fiscal adjustment, having undergone stabi- lization in the early 1990s. Chile was studied during a stage of sustained growth, having completed its stabilization and fiscal adjustment in the 218 CRAFTING LABOR POLICY 1980s. These countries therefore span a broad spectrum in terms of their macroeconomic performance: • In the early and mid-1990s, Chile had experienced sustained and high gross domestic product (GDP) growth rates of more than 8 per- cent per year, Argentina had more variable but still impressive growth rates of about 6 percent, but Brazil had experienced wildly oscillat- ing growth rates averaging less than 3 percent. * Chile and Brazil faced income inequality levels that were among the highest in the world, while Argentina had inequality levels that are average by Latin American standards but are high relative to OECD countries. * Brazil had averaged a high fiscal deficit of about 5 percent of GDP, Argentina's government accounts were relatively balanced, and Chile's completion of fiscal adjustment by the late 1980s shows up in con- sistent surpluses in the 1990s, rare among Latin American countries. * Chile enjoyed relatively low rates of inflation, having consolidated its stabilization through fiscal adjustment in the 1980s; Argentina had reduced its rate of inflation from triple to single digits during the early 1990s through some fiscal restraint, but mainly through mon- etary discipline imposed by adopting currency convertibility; and Brazil had just begun the process of consolidating the stabilization achieved by the adoption of a new currency and a tight monetary stance, by beginning a fiscal adjustment that threatened to be diffi- cult and protracted. * In terms of external sector performance, Argentina, Brazil, and Chile represented the middle ground in Latin America, with current ac- count deficits averaging 1.4, 0.4, and 2.8 percent of GDP respec- tively during 1990-97. It should be reemphasized that evidence that the countries and prob- lems studied here are representative of other countries in the region should not be interpreted as a claim that the policy solutions proposed here can be applied without additional analysis in, say, Mexico or Colombia, sim- ply because they face general economic conditions that seem similar to those in Argentina, Brazil, or Chile. In fact, we have made precisely the opposite point in this book. On the question of whether labor laws affect employment and earnings outcomes, for example, the general message of the book is that a country's economic and political history are critical de- terminants of the manner in which labor legislation affects wages and em- SOME GUIDELINES FOR THE NEW DECADE: SUMMARY AND CONCLUSIONS ZI9 ployment, so that labor markets must be studied and understood country by country to arrive at policy recommendations that have any chance of being implemented. Stabilization, Public Employment, and Informality in Brazil The two chapters on Brazil examine the public and private sector labor market during a period of stabilization and attempt to initiate fiscal ad- justment. The Real Plan, adopted in July 1994, brought down annual in- flation rates from about 3,000 percent in 1993 to less than 5 percent by 1998. But the currency-based stabilization depended on the government's ability to rein in public sector expenditures, a large part of which is per- sonnel related. Since personnel expenditures have been difficult to cut- due to a combination of constitutional guarantees of generous pensions and tenure for government workers and a lack of political consensus- attempts to control budget deficits have been centered on raising general tax revenues and earmarked contributions. At about 30 percent, the tax- revenue-to-GDP-ratio in Brazil is one of the highest in the developing world. A plethora of earmarked taxes are believed to raise the cost of labor in the formal, regulated sector by two-thirds of wage levels or more. The share of informality-measured as the fraction of workers who paid social security and severance fund contributions-rose from 40 percent in 1980 to almost 60 percent by the end of the 1990s. Chapter 2 tackles the issue of public employment. Based on the general finding that the ratio of public employment to the total (less than 9 per- cent in the mid-1990s) is not abnormally high by international standards, but the financial burden of public employment is quite high (more than 12 percent of GDP), the approach adopted is that the problem is not best solved by only cutting employment. A common problem with public em- ployment reform has been the high incidence of reversal of actions a few years after the reforms were undertaken. For public employment reform to be sustainable and to reduce the probability of later reversals, it should be based on the principle of labor market efficiency. The most important rule in this regard is that compensation for government workers should not be vastly out of line with pay and employment conditions in the pri- vate sector. In 1995, the conventional wisdom in Brazil was that government work- ers received considerably lower compensation than similar workers in the private sector-their pensions were more generous, but this was simply delayed compensation for a lifetime of relatively low pay. Chapter 2 sum- 220 CRAFTING LABOR POLICY marizes the first attempt in Brazil to systematically quantify public-private differences in compensation (pay, pensions, and stability of tenure). Quantification of salary differences was the most important, both because the magnitudes were hotly disputed, and because pension benefits are viewed almost as a continuation of salary payments upon completion of active service under the Brazilian pension regime for government work- ers.2 Using nationwide household survey data, Chapter 2 provides es- timates of human capital earnings functions designed to determine the premia attached to public employment. That is, the empirical analysis ex- amines whether, on average, people with similar observable skills and other characteristics such as education, tenure, age, gender, race, and lo- cation obtain significandy different compensation in government and pri- vate employment. These earnings functions were estimated for workers at all levels (federal, state, and municipal) and in all branches (public administration, education and health, military, judiciary, legislature, and public enterprises) of government. Results at this level of aggregation obviously miss many differences that might exist between public and private workers, but the level of detail at which empirical analysis was conducted was purposefully matched to ad- dress the questions being raised in the national debate on this subject, and was also constrained by tractability concerns. Pension and job stability dif- ferences between government and private sector workers were also esti- mated at the same level of detail. In qualitative terms the findings sur- prised some people, but the magnitudes surprised many more. Chapter 2 reports that in the mid-1990s, when Brazil initiated its fiscal adjustment efforts, public-private differentials in salaries and pensions were large: * Compared with equally qualified workers in the private sector, salaries of public employees were 30 to 50 percent higher for federal administration and judicial and legislative workers, 20 to 35 percent higher for employees in federal and state enterprises, roughly the same for state civil servants, and 5 to 15 percent lower for municipal administration and education and health sector workers. * Compared with their private sector counterparts, pension levels of civil servants were 25 to 50 percent higher, depending on their salary level, gender, and occupation. Despite the aggregate nature of the analysis, such findings can help to derive policy actions that explicitly consider labor market efficiency. The results suggested that efforts to reduce government payroll expenses in Brazil should focus on lowering salary and pension levels at the federal SOME GUIDELINES FOR THE NEW DECADE: SUMMARY AND CONCLUSIONS 221 level, a mix of salary restraint, reduced pensions, and reduced employment at the state level, and reductions of employment combined with pension restraint and selected pay increases at the municipal level. The focus on relative rather than absolute benefits also implied that, while reductions in public-private salary and pension differences and revocation of tenure rights of civil servants would both reduce the fiscal burden and increase labor market efficiency, so would the reform of selected private sector labor laws that introduce distortions, decrease labor demand, and raise in- formality in Brazil, which may in turn lower expected job stability, salary, and pension benefits in private employment. Chapter 3 tries to identify the aspects of labor legislation that most urgently require attention. As one of the contributors to Custo Brasil-the abnormally high costs of doing business -in Brazil-labor legislation that allegedly both raises labor costs and makes them more uncertain had been a topic of some dis- cussion for the last decade. The introduction of the real and the resultant fall in inflation removed an instrument for keeping real wages flexible, in- creased openness resulted in unemployment rates rising steadily (especially for more experienced industrial sector workers in areas such as Sao Paulo), and the nature of the stabilization plan left little recourse to devaluation as an alternative to wage flexibility. These factors resulted in a steady buildup of pressure for labor market reforms. A sharp rise in open unemployment rates in early 1998 precipitated by the Asian crisis of October 1997 brought the debate on labor market reforms to the forefront in mid-1998.3 Brazil's experience suggests that the economic and political history of a country is a critical determinant of which labor laws influence wages and employment, and which are not binding. Long periods of high inflation, illiteracy of the workforce, and biases in the design and enforcement of labor legislation bred by the country's socioeconomic history are all im- portant in determining the reach of labor laws.4 Contrary to conventional wisdom, these factors are shown to affect labor market outcomes even in the sector regarded as unregulated or informal. Following accepted practice in Brazil, Chapter 3 distinguishes regulated from unregulated employment by determining whether or not the con- tract has been ratified by the Ministry of Labor-that is, whether or not workers have a signed work booklet (carteira de trabalho). It then exam- ines the degree of adherence to labor laws in the formal and informal sectors, and finds "pressure points"-that is, evidence of the law on min- imum wage, work-hours, and payment timing being binding on out- comes-in both the formal and informal sectors of the Brazilian labor market. Some of the findings are: 222 CRAFTING LABOR POLICY * A significant number of workers are paid exactly the legal minimum wage, and adjustments in this wage are matched by salary adjust- ments in Brazil's unregulated sector, even in today's low-inflation environment. * Brazil's labor laws specify that workers are paid within the first week of the month. We provide evidence that this law-which was critical in periods of high inflation-appears to be obeyed by employers in both formal and informal sectors even today. • High income inequality and a prolonged period of socialist policies have led to labor laws with a prolabor bias, and to labor courts ac- quiring a similar bias in their verdicts on disputes. Combined with increased ambiguity of labor laws, this prolabor bias of dispute reso- lution has resulted in the ability of workers who have been dismissed from either formal or informal employment to extract generous sev- erance benefits from their former employers. Large backlogs in labor courts, ambiguities in labor laws, and the manner in which labor leg- islation is enforced have led to higher and more uncertain labor costs in both regulated and unregulated employment. The findings of Chapter 3 suggest that in terms of the design of legis- lation, informality in Brazil is mainly a fiscal, and not strictly a legal, phe- nomenon, a conclusion also reached byAmadeo and Camargo (1997). But the manner in which these laws have been enforced is also a critical deter- minant of informality in Brazil: poor record keeping by the government has strengthened the incentives for workers to stay informal that are al- ready built into the design of the main social security programs, and slanted enforcement by labor courts have led to workers effectively being accorded the same labor rights whether or not they have ratified contracts. The incentives to stay informal are naturally higher for workers who are as- sured of protection under labor legislation regardless of the nature of their contracts, which alters only their financial relationship with the govern- ment. The rise of informal self-employment should be of litde surprise: such conditions discourage people from employing anyone other than themselves. The chapter concludes that informality in Brazil will remain high as long as labor laws remain ambiguous and enforced with a clear pro- labor bias, and social security programs lack tighter benefit-contribution linkages and strong enforcement mechanisms. But the laws often viewed as culprits-minimum wage, severance legislation, collective bargaining mandates, and the level of payroll taxes-are, somewhat surprisingly, not the binding constraints to obtaining better labor market outcomes.5 SOME GUIDELINES FOR THE NEW DECADE: SUMMARY AND CONCLUSIONS 223 Some recent developments in Brazil deserve mention. In 1998, the com- bination of slow progress in social security and public employment reform, high levels of public debt, and currency crises in East Asia and Russia tested investor confidence in Brazil. The Brazilian Central Bank raised interest rates to unprecedented levels to defend the real against speculators, slowing economic growth. The slowdown continued after the real was allowed to float in January 1999, underwent a maxidevaluation in January 1999, and setded at levels 40 percent below predevaluation levels by the end of the year. GDP growth slowed from about 3 percent in 1998 to close to zero in 1999, showing some recovery toward the end of 1999. Metropolitan un- employment rates, while low compared with some of Brazil's neighbors such as Argentina, rose to historically high levels of more than 8 percent in mid- 1998, but have come down marginally since then despite continued slug- gishness in economic growth. In a simple model of the Brazilian economy, high interest rates would lead to a fall in the relative price of labor relative to capital (and to a substitution of labor for capital in response to this change in relative prices), but would also be associated with lower output growth (and a fall in the derived demand for labor). These offsetting substi- tution and output effects on labor demand could explain both the pattern of high but falling unemployment rates and sluggish wage growth in Brazil in 1999. Chapter 5 develops this approach in more detail for Argentina. Adjustment, Labor Demand, and Unemployment in Argentina In the years following the initiation of economic reforms in the late 1980s, Argentina's employment growth has increasingly lagged behind output growth. Despite healthy GDP growth of about 5 percent annually during the 1990s, average yearly employment growth has been less than 2 per- cent. With labor force participation (LFP) growth of 3 percent per year, the result was high unemployment, which reached almost 20 percent in 1995, and hovers around 15 percent as the country enters the new decade. In the latter half of the decade, there was improvement in measured unemployment, but this appears to have been driven in part by growing underemployment (World Bank 2000). In any case, concerns about high unemployment have dominated labor policy discussions in Argentina for the last five years. Patterns of unemployment growth do not show much evidence that this is due to a lack of relevant skills; a more likely explanation for the rise in unemployment in the early 1990s is that the costs of labor were high, prompting a shakeout of excess labor and the adoption of capital-intensive 224 CRAFTING LABOR POLICY methods of production by firms under increased pressure to improve competitiveness. Evidence on the combined effects of the Convertibility Plan, introduction of subsidies for domestic capital, and falling interest rates indicate that the price of capital relative to labor fell between 26 to 40 percent during 1990-94 (Bour 1995). Thus, countering the "output effect" on employment since the Convertibility Plan was initiated was a "substitution effect" of increases in the price of labor relative to capital. Chapter 5 formalizes this approach and quantifies the benefits to labor re- forms that lower the price of labor, in contrast to simply relying on eco- nomic growth to pull down unemployment rates. Of course, unemployment rates depend on both labor supply and labor demand, so understanding short-term variations in the supply of labor are crucial to any satisfactory treatment of the issue. With a relatively stable population, this is largely about variations in LFP rates. But more striking are the patterns in LFP, which is countercyclical in Argentina, in contrast to the United Kingdom and the United States, for example, where it is procyclical. This phenomenon has important implications for policy, and Chapter 4 examines the issue of short-term and long-term changes in labor supply in more detail. Based on an analytical framework that considers output, the relative price of labor, and technology as the main determinants of labor demand, Chapter 5 provides quantitative estimates of the likely payoff (in terms of greater employment) to Argentina's initiatives to improve labor market functioning. The analysis allows for comparisons between strategies that rely only on economic growth to reduce unemployment, and those that combine these strategies with labor policy reforms. The output elasticity of aggregate employment-holding wages or unit labor costs constant, and with an initial condition of high unemployment-helps in obtaining numerical estimates of employment growth that would occur solely due to economic growth; that is, without any labor policy reforms that raise or lower labor costs. The constant-output own-wage elasticity can be used to quantify the effects of labor policy reforms on employment growth. The authors estimate labor demand functions using a range of as- sumptions regarding technology, from relatively restrictive to the more flexible characterizations. Estimated wage elasticities range between -0.3 and -0.8 for the preferred specifications. Based on these results, the own- wage elasticity of employment can reasonably be regarded as about -0.5, implying that a 10 percent fall in labor costs will, all things being equal, raise employment by 5 percent. Estimated output elasticities range from 0.1 to 0.4. Based on these results, the output elasticity of employment can be regarded as about 0.25, implying that a 10 percent growth rate of GDP SOME GUIDELINES FOR THE NEW DECADE: SUMMARY AND CONCLUSIONS 225 will result in a 2.5 percent increase in employment. These results showed that far-reaching policy reform to lower the cost of labor relative to capi- tal was urgently required in 1996, and Argentina could not rely solely on faster output growth to noticeably increase labor demand and employ- ment. A World Bank study found that the situation had not changed sig- nificantly four years later (World Bank 2000). Women are often blamed when men face new and more serious prob- lems, and labor policy analysts are not immune to launching similar hunts for scapegoats. With Argentine unemployment rates above 15 percent in the mid-1990s, speculation that this rise was due to an abnormally large increase in labor supply of women became common. The quantitative re- sponse to this involved decomposing the changes in unemployment into a component related to labor demand or an unexpected slowdown of em- ployment growth, and another related to labor supply, or unexpected in- creases in LFP especially of women. While such exercises are largely useful only for accounting purposes-since any sensible policy measure for re- ducing unemployment must rely on stimulating labor demand and not discouraging labor supply-they can be important in determining whether labor reforms are urgent or unimportant. In joining this debate, Chapter 4 contends that common definitions of "unexpected" or "abnormal" fluc- tuations in labor supply employed ignored a fundamental attribute of the Argentine labor market; that is, that LFP is positively correlated with the unemployment rate over the business cycle. The main contribution of this chapter is to use a model of LFP to explain shorter-term LFP patterns in Argentina, reexamine changes in unemployment using these results, and draw policy implications from these exercises. Using this approach, Chapter 4 concludes: * First, during 1974-95, aggregate LFP rose by almost 18 percentage points, fuelled entirely by increases in female participation rates, which registered a sharp rise during 1994-95. Higher income usually brings about a shorter lifetime career through more schooling-de- laying the entry into the labor force-and earlier exit (retirement) from the labor force. * Second, in the first half of the 1990s, an increase in real wages ac- companied an increase in the unemployment rate in Argentina. This is the opposite of the experience of countries such as the United States, where increases in unemployment rates are accompanied by falling real wages, so there is both lower expected household income and a lower real wage for women during economic slowdowns, and the change in participation rates is ambiguous since the income effect 22b CRAFTING LABOR POLICY (the added worker effect) will imply higher participation, but the sub- stitution or wage effect (the discouraged worker effect) will imply lower participation. In the United Kingdom and the United States, the empirical finding is that the discouraged worker effect dominates, resulting in procyclical behavior of labor supply, so that fluctuations in unemployment rates are dampened. In Argentina, on the other hand, both the wealth effect and substitution effect increase partici- pation, and cyclical changes exacerbate the unemployment problem. * Third, increased female LFP could at best account for less than 6 per- centage points of the 14 percentage point increase in unemployment between 1987 and 1995 (from 6 percent to 20 percent). This is largely because of the low initial rates of female participation, so that absolute increases in female workers are not sufficiently large to con- firm conjecture that increased supply, not slowing demand, is re- sponsible for the rapid rise in unemployment. * Finally, analyses that attribute the observed rise in unemployment to abnormally large labor supply increases understate the importance of policy reform to stabilize and increase labor demand. Interpreting departures from a linear trend as abnortnal, participation levels for both women and men appear to be about 5 and 2 percentage points, respectively, above their normal levels. But both theory and empiri- cal analysis suggest that using only trend values as expected. LFP ig- nores an important attribute of the Argentine labor market-that is, LFP is systematically influenced by cyclical factors. When cyclical factors are incorporated, the "surprise" component of the increase in labor supply in the 1990s diminishes significantly. The pattern of unemployment in Argentina indicates that it is a wide- spread phenomenon, and not restricted to a few clearly identified worker groups. The relatively uniform increase in unemployment rates across worker categories points to falling labor demand as the main cause of growing unemployment, and these patterns do not lend themselves easily to effective labor market interventions by the government. With wide- spread unemployment, special incentives for hiring workers with certain attributes (for example, relaxed hiring rules for women and younger work- ers, and incentives adopted by the government to encourage employers to hire certain types of workers), will be less effective than reform of labor market regulations that allow real wages to respond to economic shocks, and growth-enhancing policies that raise the demand for labor. Compre- hensive labor reform would allow the government to concentrate on the SOME GUIDELINES FOR THE NEW DECADE: SUMMARY AND CONCLUSIONS 227 task of instituting policies for raising long-term employment growth, in- stead of being distracted by temporary but sharp increases in unemploy- ment, which are politically difficult to ignore. Growth, Inequality, and Uncertainty in Chile The combination of an open economy and a flexible labor market is be- lieved by some to be the cause of growing socioeconomic ills in Chile, in- cluding income inequality. It is argued that international trade has led to widening differentials in earnings between Chile's rich and poor, and labor market reforms have prevented the disadvantaged from stemming this tide. The deregulation of labor markets under the military regime has been followed by a period of flexibility since 1987 (Cortazar 1997). Some observers have alleged that the new jobs created in the 1990s are "low- paying, low-quality, short-term jobs" (Leiva 1996). "Low quality" is often used to connote work in which job stability is not a norm: concerns of in- creased precariousness of employment were commonplace even before the Asian and Russian financial crises, and are likely to occupy even more at- tention today. Increasing inequality and uncertainty of earnings can slow growth and may induce reversal of earlier reforms, especially the reform of labor legis- lation. Chapters 6 and 7 address two sets of issues: first, whether the labor market has changed so that those with human capital (for example, high levels of education) are being rewarded proportionately more, and whether the structure of Chile's labor market implies that returns to human capital are lower for poorer workers; and second, whether the combination of a flexible labor market and increased openness has resulted in greater uncer- tainty of earnings, and whether these recent developments warrant signif- icant policy changes. Chapter 6 uses quantile regression techniques to analyze whether the rewards to human capital differ for workers in different parts of the wage distribution, and uses simple statistical techniques to quantify the benefits of improved education policies for poorer sections of society. The main findings are: During 1987-92, average labor earnings increased by about 30 per- cent for the top and bottom quintile groups in Chile, and by about 20 percent for the middle groups. During 1992-95, average earnings fell for the top and bottom quintile groups, but rose modestly for the middle classes. 228 CRAFTING LABOR POLICY * Wage inequality, as measured by the Gini coefficient and the spread between the 90th and 10th percentile groups, rose between 1960 and 1987-88, but has declined sizably since then. This change in direc- tion has been missed or underemphasized by earlier studies on the subject. * Rates of return to education behave in a similar way, rising from 1960 until 1987-88 and declining since then. * These findings suggest that the Chilean government's emphasis on education as an instrument for combating inequality is appropriate, though this is tempered by the finding that rates of return to educa- tion are lower for lower quantile groups. * Simulations suggest that further improvements in access to schooling by the poor will have modest effects on earnings inequality, and this only at higher levels of education (grade 9 and above). Improvements in the quality of education appear to be twice as effective as im- provements in access to schooling, and the returns to improvements in education quality are obtained largely at lower levels of schooling (0 to 8 years). Chapter 7 examines the validity of claims that, although employment growth during 1990-97 outstripped labor force growth, this may have been cold comfort for workers because employment has also grown more precarious in Chile. While precariousness normally refers to the uncer- tainty surrounding employment, the more meaningful measure of the un- certainty in the labor market is the precariousness of earnings, that is, of a worker being without labor income and related benefits. This depends less on how likely the loss of a job is (the incidence of unemployment), and more on how long a worker is likely to be without work (the duration of unemployment), best estimated as the probability of finding a job. In ei- ther case, because precariousness refers to the probability of a change in employment status, the relevant indicators measuring precariousness are flows, not stocks. The authors use a labor flows model developed by Haindl (1985) to convert reported data into magnitudes that are better suited for address- ing the question of precariousness of employment and earnings. The re- sults contradict the belief that employment and labor earnings have be- come more precarious in Chile: * In the years since 1987, the Chilean labor market has become an increasingly reliable source of income: expected job tenure has in- SOME GUIDELINES FOR THE NEW DECADE: SUMMARY AND CONCLUSIONS 229 creased, unemployment rates and duration have fallen significantly, and long-term unemployment has become practically nonexistent. * The approach followed during the last decade has paid dividends: poverty and wage inequality fell impressively since 1987, and direct intervention in the labor market to reduce poverty or inequality is neither necessary nor effective, and may prove counterproductive. * Given the average tenure and unemployment duration in the econ- omy before the international currency crises slowed growth, the Chilean mandated severance payment system seemed adequate to provide support for most formal sector workers if they become unemployed. * Changes in labor legislation to make dismissals harder are likely to prove counterproductive, and further increases in minimum wages are as likely to worsen poverty and inequality as to improve them. Improvements in the quality of vocational secondary education are more likely to benefit the poor, but these have proved difficult and expensive to attain in public schools. Since 1998, unemployment rates have climbed to above 10 percent in Chile, and the clamor for a formal unemployment insurance system has increased. In late 1999 and early 2000, aided by economic recovery, em- ployment growth picked up again and unemployment fell considerably. Changes in labor laws that would have introduced rigidities were aborted in mid-1999 due to pressure from employers, but the new government that took office in 2000 will face pressure to introduce changes that edge labor markets away from competitiveness and toward intermediate- centralized structures. Lessons Learned It is difficult to draw general lessons in the area of labor-policy by study- ing the experience of just three countries over less than a decade but, with care, some conclusions are possible. Labor Policy Issues are Country Specific The first general lesson is, paradoxically, that generalizations about labor policy are likely to be pointless or misleading. The findings in this book suggest that the economic and political history of a country is a critical de- terminant of the relative importance of legislation on wages and employ- 230 CRAFTING LABOR POLICY ment, so that the subset of binding laws is country specific. For example, the book provides evidence that a large number of workers are paid exactly the legal minimum wage even in Brazil's unregulated sector, and adjust- ments in this wage are matched by salary adjustments. It is possible-and indeed it has been confirmed-that in some other countries minimum wages are not binding even in the regulated sector. It would be foolhardy to assert either that minimum wage legislation is not important for eco- nomic outcomes or that it is important for all or most countries. Similarly, it would be misleading to use the demand elasticity results for, say, Ar- gentina, to argue that reductions in labor costs would have the same em- ployment effects in Brazil or Chile or Mexico; at best, these estimates can serve as a guide for policymakers in Argentina. Again, the finding that sus- tained growth in Chile has not been associated with increasing earnings inequality does not guarantee similar results for other countries; it merely refutes some of the claims by people who question the desirability of growth-oriented labor policies that were instituted by the Chilean govern- ment during the 1980s. And the finding that many of Brazil's federal gov- ernment employees are grossly overcompensated may not necessarily apply to other countries. Implementable labor policies have to be designed by studying and un- derstanding labor markets country by country, not by compiling and an- alyzing cross-country data to obtain a policy verdict that is correct on av- erage, or by extrapolating results found for one country to others that happen to share some attributes.6 This has some obvious implications for what type of research would be most useful in promoting better labor poli- cies in developing countries. Labor Outcomes Depend Only Partly on Labor Policies The second general lesson is rather obvious, but bears repeating since it is so often forgotten: Labor reform is neither necessary nor sufficient for im- proving labor outcomes. As the experience of these countries shows, a suc- cessful stabilization unaccompanied by changes in labor policy will in gen- eral improve labor outcomes, though it will also unmask microeconomic imbalances (public-private compensation differentials in Brazil, relative prices of labor and capital in Argentina, and the gaps between the rich and poor in Chile). Similarly, fiscal adjustment could also improve employ- ment and earnings outcomes, illustrating that actions such as putting gov- ernment finances in order can improve outcomes in the private labor mar- ket, even if no labor reforms take place. And moving from a period of SOME GUIDELINES FOR THE NEW DECADE: SUMMARY AND CONCLUSIONS 23I adjustment to sustained economic growth will improve earnings and em- ployment outcomes, even if there are no accompanying improvements in labor policies. However, this does not rule out that, all things being equal, outcomes would be better still if appropriate labor reforms are made, though it is harder to make this case persuasively. Making Labor Policies Better Is Dificult During Good Times The first corollary of the above-that labor outcomes can improve even without labor reform, or worsen after labor reform-is that it is difficult to carry out such reforms during periods of general economic improve- ment. In Argentina, for example, labor policies did not rise to the top of the reform agenda during 1990-94, even though unemployment was in- creasing steadily, because of generally improving wage and employment conditions as a result of successful stabilization. When the Tequila Crisis hit Argentina, unemployment skyrocketed and labor reform came to the forefront of discussions, only to recede again when economic growth re- sumed as fiscal and financial reforms advanced. Again, in Brazil, labor re- forms only briefly dominated the political landscape when unemployment rates rose to historic levels in mid-1998 after Brazil was rocked by the Russian financial crisis and the economy slid into a recession. When the economic slowdown proved to be less severe than anticipated and unem- ployment rates fell marginally, labor policy reform was moved off the Brazilian government's list of priorities. Finally, a decade of sustained growth and improving employment and earnings outcomes in Chile had the perverse effect of prompting the reversal of labor reforms that may have made these outcomes possible in the first place. Unsubstantiated (but vocal) allegations that the improvements in average earnings and employ- ment had masked deteriorated distribution and increased uncertainty of labor earnings have made the Chilean government serio-usly consider re- versing some of the policies adopted in the 1980s. Quantifying Key Magnitudes Can Facilitate Labor Reform The second corollary of the finding that labor outcomes depend only partly on labor policies is that quantification of the likely effects of policy reform can help to advance the agenda. Rhetoric on labor policy is as abundant as relevant numbers are rare. Labor reform is always difficult and, when attempted-reforms are usually piecemeal. In every country, there are well-entrenched labor interests such as labor unions, political 232 CRAFTING LABOR POLICY risks for reformers are high, and proponents of reform-generally em- ployers or e6onomists-are often ineffective in convincing policymakers of the benefits of taking these risks. Under these circumstances, to help policymakers focus their efforts and explain them to the electorate, it helps to know whether labor policy changes are necessary and, if so, which as- pects of labor policy are binding anid which are relatively irrelevant, and what are the likely benefits of such reforms in terms of improved labor market outcomes. Drazen (2000) discusses "stage one" reforms such as stabilization, the benefits of which are widely distributed, but also refers to "stage two" re- forms such as labor market restructuring, where there are obvious winners and losers. Drazen notes that one reason for delay or nonadoption of re- form may be the inability of policymakers to communicate the benefits of reforms to the electorate, and suggests that convincing "left-wing" policy- makers about the benefits of "right-wing" reforms such as labor market re- structuring may be a good reform strategy since they may be more likely to achieve success than right-wing policymakers. This line of reasoning suggests that quantification of the benefits of reform may be even more crucial for such stage two reforms than for macroeconomic reforms, which is one of the themes of this book. In this book, we quantify the growth in employment that labor policy reform in Argentina could have generated. For Brazil, again using a quan- titative approach, the book illustrates an approach to determine aspects of labor legislation-both design and implementation related-that are the most important for outcomes and those that may be relatively nonbind- ing. Quantifying the concept of precariousness leads to the recommenda- tion that Chile may be better off leaving labor policies unaltered, and looking to other policy measures such as improved education quality to narrow the income gap between the rich and the poor. Optimism for the Next Decade In the early 1990s it would not have been an exaggeration to characterize the reform .of labor policies as a "forgotten reform" (Edwards and Lustig 1997). Most countries had by then put in place successful macroeconomic stabilization programs, and many had already launched financial sector re- forms. By the middle of the decade labor policy changes could reasonably be upgraded from forgotten to the "unfinished agenda" (Guasch 1999). We are optimistic that labor reform will continue during the next decade, and even be upgraded from largely unfinished to merely tardy in some SOME GUIDELINES FOR THE NEW DECADE: SUMMARY AND CONCLUSIONS 233 countries of the region. Recall that the 1980s saw much of Latin America attempting macroeconomic reform with little success. We think that some of these experiments bore fruit in the 1990s, because most countries in the region stabilized, so the decade of the 1980s was not as much of a loss as many believe. Being microeconomic in nature, labor reform will be even more difficult because it generates clearly identifiable winners and losers. But we hope that a decade from now, the 1990s will be regarded as the time when the seeds of durable labor reform were sown in Latin America. Notes 1. Together, the population of Argentina, Brazil, and Chile is about 215 mil- lion, or about 40 percent of the regional total of 500 million; Argentina had 36 million, Brazil 164 million, and Chile 15 million. Of a total labor force of 206 million in Latin America in 1997, the sample countries had about 94 million, or 45 percent of the total (Argentina with 14 million, Brazil with 74 million, and Chile with 6 million). In 1998, these countries had US$1,150 billion-or more than 50 percent-of the regional GDP of about US$2,000 billion; Argentina had US$300 billion, Brazil US$775 billion, and Chile US$75 billion. Naturally, therefore, the income levels in these countries were somewhat atypical of the re- gion: in 1998, Argentina (with US$8,600), Brazil (with US$4,700), and Chile (with US$5,000) each had a GNP per capita higher than the regional average of US$3,900. 2. Under the Brazilian Constitution, upon eligibility government workers get 100 percent of their final salary as pensions; contribution rates average about 10 percent, but eligibility rules are notoriously generous and are often applied in a lax manner It is not an exaggeration to describe pensions as a continuation of salary payments upon retirement. The rules for determining pensions in the pri- vate sector are somewhat more in line with international standards for pay-as- you-go social security, but again are considerably more generous because of easy eligibility and the absence of a minimum age at which social security benefits commence. Brazil has been engaged in pension reform for the last three years (see World Bank 2001 for details). 3. As it turned out, few reforms were actually implemented. 4. For example, Chapter 3 reasons that, even in the unregulated sector, changes in mandated minimum wages were used by workers as a signal for revising nom- inal wages during a period of high inflation rates, and continue to be used in this manner even in today's low-inflation conditions. 5. See Maloney (1998) for an analysis of informality using data from Mexico. 6. Cross-country analysis such as that in Forteza and Rama (2000), where the authors use data for more than 100 countries to conjecture that minimum wages and mandatory benefits have a marginal impact only, but the size and strength of 234 CRAFTING LABOR POLICY organized labor appear to be crucial for outcomes, should therefore be no more than a starting.point for policy discussions. References Amadeo, Edward J., and Jose Marcio Camargo. 1997. "Brazil: Regulation and Flexibility in the Labor Market." In S. Edwards and N. Lustig; eds., Labor Markets in Latin America: Combining Social Protection with Market Flexibility. Washington, D.C.: The Brookings Institution Press. Bour, Juan Luis. 1995. "Mercado de trabajo y Productividad en la Argentina." Draft. Latin American Economic Research Foundation. Buenos Aires. Cortazar, Rene. 1997. "Chile: The Evolution and Reform of the Labor Market." In S. Edwards and N. Lustig, eds., Labor Markets in Latin America: Combin- ing Social Protection with Market Flexibility. Washington, D.C.: The Brookings Institution Press. Drazen, Allan. 2000. Political Economy in Macroeconomics. Princeton, N.J.: Prince- ton University Press. Edwards, Sebastian, and Nora Lustig, eds. 1997. Labor Markets in Latin America: Combining Social Protection with Market Flexibility. Washington, D.C.: The Brookings. Institution Press. Forteza, Alvaro, and Martin Rama. 2000. "Labor Market 'Rigidity' and the Suc- cess of Economic Reforms Across More than One Hundred Countries." Paper prepared for the research project on The Impact of Labor Market Policies and Institutions on Economic Performance. World Bank, RPO 680-96. Washington, D.C. Guasch, Jose Luis. 1999. "Labor Market Reform and Job Creation: The Unfin- ished Agenda in Latin American and Caribbean Countries." Directions in De- velopment Series. Washington, D.C.: World Bank. Haindl, Erik. 1985. "Un modelo para la determinaci6n de flujos y parametros dinamicas en el mercado de trabajo." Estudios de Economia 12 (1) (primer trimestre). Leiva, Fernando. 1996. "Flexible Labor Markets, Poverty and Social Disintegra- tion in Chile, 1990-1994: The Limitations of World Bank Policies." Paper presented at the World Bank, OXFAM draft. Washington, D.C. Maloney; Williarn F. 1998. Efficiency Wages, Unions, and the Structure of the Mexican Labor Market-Three Essays. Latin America and Caribbean Region, Poverty Reduction and Economic Management Department. World Bank, Washington, D.C. World Bank. 2000.' Argentina-Labor Market in the New Millenium. Confiden- tial Report No. 19996-AR. Washington, D.C. . 2001. "Brazil: Critical Issues in Social Security." A World Bank Coun- try Study. Washington, D.C. Technical Annexes A. Technical Annex to Chapter 2: The Public-Private Wage Gap B. Technical Annex to Chapter 5: Wage and Output Elasticities of Labor Demand C. Technical Annex to Chapter 6: The Quantile Regression D. Technical Annex to Chapter 7: HaindI's Labor Flows Model 235 A Technical Annex to Chapter 2: The Public-Private Wage Gap This Annex presents a brief methodological discussion of how both the wage gap between the public and private sectors, and the controlled wage gap, are measured in Chapter 2. Wage Gap Concepts In the empirical analysis of wage differentials, the use of wages, logwages, arithmetic means, and geometric means is common. In this section we briefly review the relationships between these concepts and how they are used to obtain measures for the public-private earnings and wage gap. We begin by reviewing the advantages and disadvantages of using dif- ferences in logwages compared with absolute or relative differences in wages. Let wa and wb be the average wages in the public and private sectors, respectively. Using this notation, we can define the absolute wage difference (GO), two alternative versions for the relative wage difference (Gia and Glb), and the logwage difference (G2) as follows: Go = wa - Wb (A- 1) GI. Wa, wb (A-2) Wa Gib Wa -Wb (A-3) Wb 237 238 CRAFTING LABOR POLICY G2 = ln(w,a) - ln(wb) = ln(wa lwb) (A-4) The absolute wage difference, G., has the disadvantage compared to the relative difference in wages, G,a and G,b; and the difference in log- wages, G2, that it is sensitive to the unit of measurement. The relative differences in wages, Gla and G l, have the major disadvantage relative to differences in logwages, G2, that their calculation requires selection of a baseline wage as reference and that the measure is sensitive to this choice. Consequently, one has to constantly refer to the chosen baseline wage, which makes the use of relative wage differences cumbersome in a detailed analysis of wage differentials. For example, if the average wage in the pub- lic and private sectors is 50 and 20 monetary units, respectively, then the wage gap can be expressed in relative terms as 150 percent of the private sector average wage or 60 percent of the public sector average wage. The logwage difference G2 overcomes this issue and hence eliminates the need to keep track of the wage used as reference. In our example above, the log- wage difference between the public and private sectors is 0.92. The major disadvantage of G2 can be seen in the fact that it is more dif- ficult to interpret. One way to facilitate its interpretation is to notice that, for small variations, the logwage difference is an approximation for the rel- ative change in the level of wages. For example, if the wages are 50 and 51, the relative wage variations using 50 or 51 as the baseline references are 0.0200 and 0.0196, respectively. In this case the logwage difference is 0.0198. In the general case, if wa>wb then a -Wb 2>In Wa > Wa,-Wb Wa wb Wb or G,. ˇ 2 2 Gg Calculation of Controlled Wage Gap Let w denote the wage and let p be an indicator for the public sector, which assumes 0 for the private sector and 1 for the public sector. Denote by h the regression function of logwages on gender (g), race (r), schooling (e), age (a), tenure (t), andp, then E[ln(w)/g, r, e, a, t, p = h(g,r, e,a,t,p) (A-5) TECHNICAL ANNEXES 239 Now, the specification of our regressions can be seen as a series of hy- potheses about the functional form of h. Our basic assumption is that the regression function is separable on gender, race, schooling, age, and ten- ure, but not necessary on the indicator for the public sector, that is, h(g, r, e, a, t,p) = fi(g,p) + f2(r,p) + f3(e,p) + f4(a,p) + f5(t,p) (A-6) Nonseparability in p implies that the wage advantage of workers in the public sector varies with their characteristics. If the regression function is separable in p, then the wage advantage is the same for all public sector workers independent of their individual characteristics. To simplify our empirical analysis, we assume that the dependency on gender, race, schooling, and tenure is linear, and that the dependency on age is quadratic. This allows us to express the logwage regression function as lig, r, e, a, t,p) = a0(p) + a,(p).g-+ a2(p).r-+ a3(p).e (A-7) + a4(p).a + b4(p).a2 + a5(p).t We will refer to this expression as the "general model." To estimate the "general model" we regress logwages on gender, race, schooling, age, and tenure for the public and private sectors, separately. In addition, we estimate and analyze a simplified version of this model, which we will refer to as the "basic model." It is given by h(g, r, e, a, t, p) = a + ao.p + a,.g + a2.r f a3.e + a4.a + b4.a2 + a5.t (A-8) The basic model imposes separability of p, which, given the above as- sumption, implies that all coefficients except the intercept are common to both sectors. It is estimated by regressing logwages on gender, race, school- ing, age, tenure, and an indicator for public emplbyment in a sample pool- ing together the public and private labor force. In the case of the "general model," the average logwage in the public sector is aO(l) + a1(l).p(g/1) + a2(1).pt(r/1) + a3(1)41(ell) + a4(1).k(a/l) + b4(1).k(a211) + a5(1).g(t/1) where, g(x/l1) denotes the average of the characteristic x in the public sec- tot Based on the same model the average logwage that would prevail in the private sector if the labor force in the sector had the same characteris- tics as those of the public sector is given by a((O) + a1(O).t(g/l) + a2(0).4(r/1) + a3(0).kI(ell) + a4(0).4(a/l) + b4(0).pg(a2/1) + a5(0).4(t/l) 240 CRAF1'ING LABOR POLICY Therefore, the estimate of the logwage gap between workers with identi- cal observed characteristics is given by (ao(l)-ao(O)) + [a,(l)'-a(O)].j.t(g/l) + [a2(1)-a2(0)].-9(r/l) + [a3(1)-a3(O)] Rt(ell) + [a4(1)-a4(0)].pu(a/1) + [b4(1)-b4(0)].-.t(a2/l) + [a5(1)-a5(0)]4t(t/1) In the case of the "basic model" the average logwage in the public sec- tor is given by CC + a0 + a,.g(ggl) + a2.g(r/1) + a34t(e/1) + a4.t(a/1) + b4.j.t(a2/1) + a54t(t/1) and the average logwage that would prevail in the private sector if the labor force in the sector had the same characteristics as those of the pub- lic sector labor force is given by a + aj.pt(g/l) + a2.pt(r/l) + a3gt(e/l) + a4.4t(a/i) + b44t(a2/1) + a54d(t/1) Consequently, the logwage gap between workers with identical observed characteristics is given simply by ao. B eTechnical Annex to Chapter 5: Wage and Output Elasticities of Labor Demand Cobb-Douglas Technology The Cobb-Douglas production function is Y = A. KP. L'-P (B-I) where Y denotes output, K capital, and L labor. ,B is the share of capital. A indicates the level of technology. Using B.1, through cost minimization we can derive the cost function C = C(w, r, Y) = Z. wl-P.rP.Y (B-2) where Z is a constant, and w and r are the prices of labor and capital. Using Shephard's lemma, the labor demand equation is obtained: Ld = x.w(-HP).rI3.Y (B-3) otl is a constant. The Cobb-Douglas technology restricts the elasticity of substitution between L and K, y, to equal 1. The own-wage elasticity of demand can be estimated in two ways: (i) Taking logs of B-3 InLd = InA + (-P)lnw + ,Blnr + InY (B-4) The own-wage elasticity of demand corresponds to the coefficient on Inw, that is (-,B). 241 242 CRAFTING LABOR POLICY (ii) Estimating the production function directly, using a log-log form of equation B-1: InY=lnA+ j3lnK+(I-[)lnL. (B-5) Constant Elasticity of Substitution (CES) Technology The CES production and cost function are, respectively Y= [aLP+ (1-ox).KP]"1P (B-6) C = Y[aO.wiG + (1-a)".0-011/0-0) (B-7) where a = /[Il-p]; a is the elasticity of substitution and is a constant, but not necessarily 1. The CES is sufficiently general that ay is free to fluctu- ate between 0 and c, so one can infer its size. Having obtained an esti- mate of a, the own-wage elasticity 5LL can then be inferred by using EL = -1-SL]G. (B-8) Using Shephard's lemma, the labor demand equation can be derived, and taking logs we obtain In L = a" - 61nw + lnY (B-9) Alternatively, we can estimate a by using what Hamermesh (1993) calls the estimate of the marginal productivity condition: In L = a" - canw + ELylnY (B-10) This produces estimates not only on G, but also on 6LY* Together with in- formation on SLp the estimate of a generates ELL' as can be seen in B-8. Estimation of Single Factor Demand Equations For easy estimation, the Shephard condition can be written in logarithmic form: lnL = o' + 6LL Inw + ELK Inr + SLy lny (B- 1) This yields the own-wage elasticity ELL' the cross-elasticity of demand for labor 8LK, and the employment-output elasticity. Generalized Leontief Technology Input demand functions using well-established forms that are second- order approximations to arbitrary cost or production functions-such as TECHNICAL ANNEXES 243 the generalized Leontief and the translog forms-can also be estimated. The advantage of these functional forms in the two-factor case over the CES function lies in the fact that Y and the elasticities are not restricted to be constant, but are allowed to vary with the values of the factor inputs or prices. Imposing constant returns to scale, the generalized Leontief cost func- tion becomes C = Y[alw + 2a12w05r05-+ a22r] (B-12) Using Shephard's lemma, the labor demand equation is Ld = a, 1Y + al2Y(w/r)-05 (B-13) Dividing by Y this yields L/Y = a,, + a12(w/r)0-5 (B-14) Proceding similarly for capital provides KY = a22 + a12(w/r)-05 (B-15) To estimate this model, we use Zellner's seemingly unrelated estimator (SURE) with the cross-equation restriction imposed. We employ the three-stage least squares estimation procedure to correct for simultaneous equation bias. Transcendental Logarithmic (Translog) Technology The translog cost function is written as InC = In ao + a, Inw + O.5a2 [Inw]2 + a3 Inw.lnr + 0.5b1 [Inr]2+ b)nr + 0.5b, [Inr]2 + b3.lnY + b4 Inr.InY + a4lnw.lnY + 0.5b5 [Iny]2 (B-16) The labor demand function-actually the share equation-derived from the above translog cost function, is given by wL/C = a, + a)nwv + agnr + a4InY (B-17) When the full model and the cross-equation restrictions are specified, there is one redundant equation. In our case, with only two inputs of data available, only one equation, the share of labor, is specified. The translog cost function as specified above is a highly general, nonhomothetic form. In the general case, nonhomotheticity implies that output is not separable from factor prices. Instead, the effect of factor prices depends on the scale 244 CRAFITING LABOR POLlCY of output. This is reflected in the terms b4lnY.lnr and a4lnY.lnw in B-16. A test for homotheticity is whether a4 and b4 are equal to 0. Since the cost function is homogeneous of degree one in prices a4+ b4 = 0, which implies the case that here we simply have to test that a4 = 0. Homogeneity of a constant degree in output occurs if, in addition to the homotheticity re- strictions, b5 = 0; in this case the degree of homogeneity equals 1Ib3. Con- stant returns to scale of the dual production function occurs when, in ad- dition to the above restrictions, b3 = 1. Finally, the translog function reduces to the CRS Cobb-Douglas when, in addition to the above restric- tions, a3=0 Estimating Returns to Scale In the presence of scale effects (B.8) becomes CL= -{1-SL]a - Sr (1-18) where T1 is the elasticity of product demand. To estimate returns to scale, it is usually necessary to add the translog cost function to the share equation to be estimated, since b3 and b5 appear only in the cost fimction. Christensen and Greene (1976) note that the op- timal procedure is to jointly estimate the cost function and the cost share equation as a multivariate regression system. This has the effect of adding additional degrees of freedom and only a few (b3 and b) unrestricted re- gression coefficients. Following Hanoch (1975), returns to scale At are com- puted as the inverse of the elasticity of cost with respect to output. That is p. = I/Ecy = average cost/marginal cost (B-19) For the translog cost function Ecy = b3 + b4 lnr + a4lnw + b}n Y (8-20) Hence, with constant returns to scale, both ECY and opare equal to 1. If there are increasing returns (p > 1), then 8acy < 1. A more natural way to represent returns to scale is to define ee = (1-E-cy). This results in a posi- tive number for increasing returns to scale, and negative numbers other- wvise. Note that unless -CY is constant for all values of Y and independent of factor prices (that is, horiiothetic), it will not equal the elasticity of labor demnancd with respect to output. In the general nonhomothetic case with the cost function C = C6%v, r, Y), there is no exact relationship between ECY and ELY. With homotheticitw the nvo are equal. TECHNICAL ANNEXES 245 We estimate the joint system using iterated three-stage least squares im- posing both the homogeneity and cross-equation restrictions. This is iden- tical to the SURE procedure, but with the fitted values of the endogenous variables regressed on all the instruments and exogenous variables. The disturbance covariance matrix is estimated using the original variables, not the fitted ones. References Hamermesh, Daniel S. 1993. Labor Demand. Princeton and Chichester, U.K.: Princeton University Press. Hanoch, Giora. 1975. "Production and Demand Models with Direct or Indirect Implicit Activity." Econometrica 43 (3): 395-419. C cechnical Annex to Chapter 6: The Quantile Regression The standard econometric tool used to estimate the Mincerian wage equa- tion has been ordinary least squares (OLS). This technique implies the minimization of the sum of the squared errors under the assumptions that (a) the model is the true model, (b) that the expected value of the errors is zero, (c) that the matrix of explanatory variables (x) is a nonstochastic matrix of full column rank, and (iv) that the vector of errors is indepen- dent with zero mean. This formulation implies that the predicted value of y given x is equal to the mean of y (conditional on the x values). The rep- resentativeness of the predicted value of y given x depends on the variance ofy given x. This leads to the fact that when disturbances are not normal, or when there is an important proportion of outliers, the traditional mean regression loses robustness. One robust way to estimate the relationship between two variables is the use of the least absolute deviation (LAD) technique. In this case, instead of minimizing the sum of the squared er- rors, we minimize the sum of the absolute value of the error. In other words, OLS minimizes n n 5= 2i B72 =)(Yi _xi3)2 (C-I) i=1 i=1 and LAD minimizes V>LAD = i Y;| =| i xg (C-2) 2=1 j=1 247 248 CRAFTING LABOR POLICY In this case the predicted value ofy conditional on x corresponds to the median of y conditional on x; that is, the value that would leave 50 per- cent of the errors above and 50 percent of the errors below. One obvious extension of this concept is to adjust the different hyperplanes so that 0 percent of the errors will be negative and (1-0) percent will be positive. Achieving this only requires weighting differently positive and negative errors in equation C-2. This is what gives the origin to the quantile re- gression method (QRM). Note that with the QRM, in principle, we may have as many fitted curves as we want (just varying 0), but in practice the number of curves is limited by the numbers of observations available. The purpose of the classical least squares estimation is to determine the conditional mean of a random variabley, E(y!x), given some explanatory variables x, usually under some assumptions about the functional form of E(ylx), for instance, linearity. The QRM enables us to pose such a question at any quantile of the conditional distribution. A real-valued ran- dom variable y is fully characterized by its distribution finction F(y) = P(Y< y). Given F(y), we can for any 0 E (0,1) define 0th quantile of Y given by Qy (0) = inf(y i 91 I F(y) 2 01. The quantile function, Qy (0) as a function of 0, completely describes the distribution of the random vari- able y. Hence, the estimation of conditional quantile functions allows us to obtain a more complete picture about the dependence of the condi- tional distribution of y on x. In other words, this means that we have a possibility of investigating the influence of explanatory variables on the shape of the distribution, and we also have a possibility of investigating the conditional impact of the regressors on the dependent variables at differ- ent "layers" of the distribution. To formalize the presentation of the QRM, let us assume that we have a sample of size n on wages (y) and some vector of individual characteris- tics. Let yi (i=1. , n) be the wage of individual i and let x. be a known vector of individual characteristics. Let us assume that the Oth quantile of the conditional distribution of y, given xi is linear, that is y, = x,; + ue1 with Qe (yi|xi) = x/, i = 1,2 .........,n (C-3) where x' is a k X I vector of covariates with xii = 1, and P3 for all i, and is an unknown k X 1 parameter vector whose estimation, for different val- ues of 0 [0 < 0 < 11 is of interest to us. The term Q0(yi|x1) denotes the conditional quantile of yi given x;. In this specification uoi is defined by u= = - x 'P,w, from where it follows that Q6(u6il xi) = 0. Therefore, the TECHNICAL ANNEXES 249 Oth quantile regression based on a sample (yi, x) i = 1, 2 ...., n, is a vec- tor P. that minimizes: YfRM 0 0 Y. |i X, P0|+ (1-0) I -Xi-, ASI I/Y;QPXi4' 2X Xilyi (X:6+(- X (C-4) One typical misconception in the interpretation of the QRM is that comparing different estimates of P is equivalent to comparing the returns of the rich (for 0 = 0.90) versus the returns for the poor (for instance for 0 = 0.10). This is not the case. What we are comparing is the distribution of the residuals after compensating for education and experience. In other words, those individuals who are high in the distribution are those who earn above what would be expected given their individual characteristics, which may include both rich and poor people. The robustness of the method comes from the fact that the estimates rely only on the sign (positive or negative) of each observation (that is, only on whether the observation is above or below the hyperplane), but the actual magnitude of the error is irrelevant. The QRM method can be easily explained using graphs. The following figure (taken from Montenegro 1998) shows three different distributions of the errors. (a) (b) (C) 250 CRAFTING LABOR POLICY In panel (a) we observe a homoscedastic distribution. In this case the set of QRM estimates is basically a set of parallel lines. Note that the slopes of the different curves adjusted for different Os will be the same, but the position parameters for each curve (that is, the parameter that shows the intersection of the curve with the y-axis will vary). In the panel (b) we again have a homoscedastic distribution, but with a smaller variance. In this case, again, the different slopes of the different curves adjusted for dif- ferent Os will be the same; the position of the parameters for each curve will vary, however, although less than in the case of panel (a), given that the distribution has a smaller variance (in an extreme case, if the variance of the distribution is zero, all the position parameters estimated will be the same). Finally, panel (c) shows a distribution of y given x that exhibits het- eroscedasticity. In this case the slopes of the different curves increase as we move from the lower quantile estimates to the upper quantile estimates. Note that the estimation of C-4 is not a problem that may be estimated using standard minimization techniques, because those require that the functions involved be of a continuous nature (and the absolute value func- tion is not). Linear programming techniques can be used to find the vec- tor P. that minimizes equation C-4, as it is shown in Koenker and Bassett (1978, 1982). The solution does not have an explicit form and does not have some common properties attributed to OLS standard estimation. Nevertheless, it can be shown to fit into the generalized method of mo- ments framework. From this well-developed asymptotic theory we can de- rive statistical tests and perform statistical analysis on the parameters (Koenker and Bassett 1982). Using the first-order condition along with the GMM it is possible to demonstrate that the parameters estimated will be consistent and asymp- totically normal (Koenker and Bassett 1978, 1982; Powell 1984; Buchin- sky 1998). If lim[X'X)/n]-V with Xi,= 1 for all i, and the error distri- bution, F, has strictly positive density at the Oth quantile [that is, f(F-1(O))>0], then f is asymptotically normal. That is, 4n(I30 & /30)AN(0,g2(6,F) VI1) (C-5) where ,, = ,B +(F-'(e), 0 , , Q)'and g2(,F) = e(1-e)1/f2(-'F9)). From this it follows that a natural statistical test for Ho: MP = m is A = j-2(Ml - m_)(M(X'X)>1M')(Mf - m) (C-6) where g is a consistent estimator of g. The bootstrapping method is used to calculate the standard errors in the empirical test. The reason is that when the residuals are heteroscedas- TECHNICAL ANNEXES 251 tic, the formula given in Koenker and Bassett (1982) appears to grossly underestimate the true standard errors. Note that the precision of the pa- rameters estimated depends on the magnitude of the density for each quantile. Therefore, the quantiles at the "border" of the distribution are more difficult to estimate because the density of the "dots" is low, and so the test has less power. The QRIM just discussed is not exempt from criticism. One criticism is related to the arbitrariness in the election of the proportion 0. This criti- cism is not relevant to our study: our main purpose is to characterize the distribution of the wage structure, so we use several conditional quantiles. Another problem associated with the QRM is that the final solution may not be unique. But uniqueness can always be achieved by selecting an ap- propriate design or by using an arbitrary rule to select from any set of mul- tiple solutions (this problem is exactly similar to selecting the median among a sample that has an even number of observations). References Buchinsky, Moshe. 1998. "Recent Advances in Quantile Regressions Models: A Practical Guideline for Empirical Research." Journal of Human Resources 33 (1): 88-126. Koenker, R, and G. Bassett. 1978. "Regression Quanriles." Econometrica 46: 33-50. .1982. "Test of Linear Hypothesis and I 1-Estimation." Econometrica 50: 1577-83. Montenegro, Claudio E. 1998. "The Structure of Wages in Chile 1960-1996: An Application of Quantile Regression." Estudios de Economica 25 (1): 71-98. Powell, James. 1984. "Least Absolute Deviation Estimation of the Censored Re- gression Model." Journal of Econometrics 25: 303-25. D eTechnical Annex to Chapter 7: Haindl's Labor Flows Model The main purpose of Haindl's labor flows model is to derive flow variables from stock variables such as employment, unemployment, and labor force reported in the University of Chile's Household Surveys for Santiago. Definition of Variables Let us first define the following variables: E = Employment level (in thousands of persons) F = Labor force (in thousands of persons) D = Unemployment level (in thousands of persons) D' = Unemployed that are first-time job seekers (in thousands of persons) D2 = Unemployed persons that have been looking for a job for less than one quarter (in thousands of persons) D 3 = Unemployed persons that have been looking for a job more than one quarter (in thousands of persons) d = Number of persons that have been fired, have retired, or have left the labor force (in thousands of persons) f = Number of persons that have entered the labor force during the period (in thousands of persons per quarter) f = Number of persons that entered the labor force for the first time (in thousands of persons per quarter) f- = Number of persons that have entered the labor force that have worked before (in thousands of persons per quarter) c = Number of persons that are hired in the current period (in thou- sands of persons per quarter) 253 254 CRAFTING LABOR POLICY cl = Flow of persons hired among those looking for a job for the first time (in thousands of persons per quarter) c2 = Flow of persons hired among those that had a job and have been looking for a new job one quarter or less (in thousands of persons per quarter) C3 = Flow of persons hired among those that had a job and have been looking for a new job more than one quarter (in thousands of per- sons per quarter) iC = Probability of finding a job in a quarter P = Probability of finding a job in a month and a half R = Average employment duration (in months) T = Expected unemployment duration (in months) We add the subscript "t" to the previous variables to indicate the re- spective quarter. The previous variables are related in the following ways. Identities D , =D,+D +D, (D-1) (D-2) Ct =CI+ 2 + 3 (D-3) c=t +c +c D, = Ft - E, (D-4) Relationships Between Flows and Stocks Ft = Ft_1 + ft (D-5) Et = Et-, +c, -d, (D-6) D,= D_1 +f +dt -ct (D-7) I = I I (D-8) 12 = 2]+d -c7 (D-9) D10 =D2 ±+D3 3-e (D-10) Closing the System The previous equations represent identities and relationships that deter- mine the movements of the stocks variables based on the flow variables. TECHNICAL ANNEXES 255 These relationships are deterministic and by definition are true in each pe- riod. However, they are not sufficient to obtain the flow variables from the stock variables. One way to close the system of equations is to relate the flow of people hired in a period to the probability of finding a job. To do that we assume that the probability of two different people finding a job in a certain period of time is independent of how long they have been looking for a job. The simplification permits closing the system and ob- taining the flow variables based on the available information. If we esti- mate the flow of people hired in a period based on their expected value we have c = P, - f, + (D-1 1) r2 = p(2 +dt) (D-12) c = ; (D7I +d 1) (D- 1 3) The probability of finding a job for people who were unemployed in the previous period is different from the probability of finding a job for people that are first-time job seekers. If we use a quarter as the period of reference, then the search period of those that were unemployed at the be- ginning of the period is a quarter. If we assume that the flow of people that are entering the labor force for the first time follow the uniform distribu- tion, the average search period for them is equal to half the period (one and a half months). This permits us to relate both probabilities as follows: (1 _Jp)2 = or (D-14) Obtaining Flows from Stocks We can now solve the system. Using equations (D-10) and (D-13) we get: D3 Jr=1- D D2 l + D 3 (D-15) From equation (D-14) we get: PI = i - a - nt )112 (D- 1 6) From equations (D-2) and (D-15) we get: _I = tt + ft2 = Ft - Et-, (D-17) 256 CRAFTING LABOR POLICY Now using equations (D-8) and (D-1 1) we obtain the flow of people entering the labor force for the first time: gl_(D,' - Dt,_, (I1- 7 t (-PI) (D-18) Using (D-17) and (D-18) we get: _t = fr-tt-l (D-19) Combining equations (D-8), (D-9), (D-11), (D-12), (D-18), and (D-19) we can obtain the flow of people that have been fired, that have re- tired, or that have dropped out of the labor force in the period: d, = Dt _ f2 (D-20) Finally we can obtain the flow of people hired in the period using equa- tions (D-6) and (D-20): Ct = Et - Et-, +dt (D-2 1) The average tenure can be defined as the time it would take to fire all the people that are working at the beginning of the period, based on the assumption that the flow of people that are fired, retired, or just drop out of the labor force does not change and that the flow of newly hired peo- ple is zero. This can be expressed in the following equation: R, = 3E,_, (D-22) dt The expected time that a person is going to be unemployed can be es- timated from the probability of finding a job in a given period. Let P* be the probability that a person finds a job in a particular period t. In this case, the probability that the person will be unemployed for n months be- fore finding a job, Pn, (that is, will find work in time n + 1) is given by: p: = Pt 1 p' )n (-23) The expected unemployment time is then E(n)= XnP,n = XnP]'(1 -J)n (D-24) n=O n=O TECHNICAL ANNEXES 257 Which implies that =E(n) = (D-25) In other words, to calculate the expected unemployment time, T7, it is necessary to estimate the probability that a person would find a job in any given month in quarter t. This can be done relating the probability P*, with the probability of finding a job in period t, 7it. If the period is a quar- ter, then the relationship is the following: (I1- 7r ) = (1- PtY )3(D-26) From this we can get PI = 1- (1-;rt) (D-27) Finally, combining equation (D-25) and (D-27): Tt= (1 ti3 (D-28) 1-(- 7r t)113' Contributors Edward J. Amadeo is the Secretary of Economic Policy in the Ministry of Finance of the government of Brazil. His research has included topics in macroeconomics and labor markets. His contribution to the book was written when he was Professor of Economics in the Pontifical Catholic University of Rio de Janeiro. Since that time, he has served as the Minister of Labor for Brazil. He has a Ph.D. in economics from Harvard University. Ricardo Paes de Barros is Director for Social Programs at the Brazilian Institute for Applied Economic Research (IPEA) in Rio de Janeiro, and has been a Research Economist for Inequality, Education, Poverty, and Labor Market Issues there since 1979. He has written extensively on issues related to poverty, labor markets, and education in Brazil. He has taught at the University of Chicago and Yale University. He has a Ph.D. in eco- nomics from the University of Chicago. Dorte Domeland is a doctoral student in Economics at the Universitat Pompeu Fabra in Barcelona, Spain. While contributing to this book, she was a consultant to the World Bank. She has a master's degree in eco- nomics from Universitat Pompeu Fabra and a bachelor's degree in eco- nomics from Eberhards-Karls-Universitat, Tiibingen, Germany. Miguel Nathan Foguel is a Researcher at the Brazilian Institute for Ap- plied Economic Research. Since contributing to this book, he has served as advisor to Brazil's Minister of Labor. He has a master's degree in eco- 259 260 CRAFT1ING LABOR POLICY nomics from the Catholic University of Rio de Janeiro, where he has also taught. Indermit S. Gill is the Lead Economist for Human Development in the World Bank's Latin America and Caribbean Region. This book was writ- ten while he was a Senior Economist in the Brazil Country Management Unit based in Brasilia. Before joining the World Bank in 1993, he taught at the State University of New York at Buffalo. He has a Ph.D. in eco- nomics from the University of Chicago. Jose Luis Guasch is Sector Manager for Private Sector Development in the World Bank's Latin America and Caribbean Region. His recent work includes a book on labor market reform in Latin America. He is Professor of Economics at the University of California at San Diego, on leave since 1998. He has a Ph.D. in economics from Stanford University. Erik Haindl is Director of the Economics Institute at Gabriela Mistral University in Santiago, Chile. He has written on labor and poverty issues in Latin America. He has a master's degree in economics from the Uni- versity of Chicago. Rosane Silva Pinto de Mendonca has been a Researcher at the Brazilian Institute for Applied Economic Research in Rio de Janeiro since 1987, and Assistant Professor in the Economics Department of the Federal Uni- versity Fluminense. She has a master's degree in economics from the Cath- olic University of Rio de Janeiro, and a Ph.D. in economics from the Fed- eral University of Rio de Janeiro. Claudio E. Montenegro is an Economist with the Development Research Group in the World Bank. He worked on this book as a consultant to the Bank. He has held several positions at the University of Chile in San- tiago, the Inter-American Development Bank, and the World Bank. He has a masters degree in economics from the University of Maryland at College Park and a masters degree in statistics from The George Washing- ton University. Marcelo Neri teaches at the Getulio Vargas Foundation in Rio de Janeiro. During the time he contributed to this book, he was a researcher at the Brazilian Institute for Applied Economic Research and Professor of Eco- nomics at the Federal University of Rio de Janeiro. He has written exten- CONTRIBUTORS 261 sively on labor issues in Brazil. He has a Ph.D. in economics from Prince- ton University. Carola Pessino is Professor of Labor Economics at CEMA in Buenos Aires. Since contributing to this book, she has served as the Secretary for Social and Fiscal Affairs in Argentina's Federal Ministry for Finance and the Economy. She has taught at Duke University and Yale University. She has a Ph.D. in economics from the University of Chicago. Claudio N. Sapelli is Associate Professor of Economics at the Pontifical Catholic University in Santiago, Chile. His recent research has been in the areas of education and health economics. Before taking up his current ap- pointment, he worked at the World Bank. He has a Ph.D. in economics from the University of Chicago. Index Note: f indicates figures, n indicates notes, and t indicates tables. Ability, wage inequality analysis Apprenticeships, in Chile, 209 model and, 162 Argentina. See also Labor demand in Acemoglu, Daron, 28n19 Argentina Added worker effect adjustment, labor demand, and in Argentina, 102 unemployment in, 223-27 in Argentina vs. U.S. during demographics of labor force, recessions, 118 income and GDP for, 233nl definition of, 98 growth vs. job creation in female LFP in Argentina and, 1990s for, 1 106, 117-18, 123 informal sector employment in, labor force participation 28n22 decisions and, 112 labor force participation by in United States, 126nl 1 gender in, 104-7, lO5t Administrative capacity, for unem- labor legislation changes in, 5, ployment insurance system in 27 Chile, 206 labor supply dynamics in, 27 Administrative Reform Bill, Brazil, 39 macroeconomic challenges in Administrative requirements costs, for 1990s of, 217, 218-19 unemployment compensation, 21 older workers' unemployment Amadeo, Edward, 94n3, 222 in, 15 Annual data, extrapolation for use of, payoffs for improvements to 137 labor market function, 97 263 264 INDEX Argentina (continued) demographics of labor force, population, labor force, and income and GDP for, unemployment in, 102-4 233n1 privatization in, 24 effective work week before and projections of labor reform after new constitution, 92t effects in, 98 employment structure in severance pay mandates in, 12, 1980s, 70-71 14t, 17t formal vs. informal employ- social security contributions in, ment in, 221-22 22 Fundo do Amparo ao temporary contracts' rules Trabalhador (FAT), 23, 80 changes in, 6, 7-9 growth vs. job creation in unemployment compensation 1990s for, 1 in, 21, 22-23, 24, 28n21 hours worked regulations in, unemployment in, 2, 107, 90-92 108t, 109-10, lilt informal sector employment in, unemployment of men in, 26-27 labor force participation job creation and unemploy- and, 114-15 ment in, 70 Asian crisis labor code and changes in, Chilean labor market reforms 73-80 and, 221 labor income from 1980s to labor market reform in Brazil 1990s, 72-73 and, 223 labor legislation changes in, Autocorrelation, cyclical effects of 5-6, 26, 34 unemployment calculation and, labor legislation reforms pro- 126n8 posed and implemented in Aviso pr6vio law ("previous notifica- 1990s, 80-82, 233n3 tion" of firing) law, Brazil, 76 labor turnover and job quality in 1990s, 71-72, 72t Bahia, Monica, 10-Il macroeconomic challenges in Barbados, unemployment 1990s of, 217, 218-19 compensation in, 22 measures of conformity with Bassett, G., 251 labor and social security Bentolila, Samuel, 150 laws, 93t Berndt, Ernst R., 147, 153nl 1 privatization in, 24 Blanchard, Oliver Jean, 195 public sector wage premium, Bolivia, severance pay mandates in, 48t, 49t 17t Real Plan, labor stabilization Brazil. See a/so FGTS; Public sector and, 33-34, 77, 219 efficiency, Brazilian sectoral earnings ratios in, 73 carteira de trabalho (work card) severance pay mandates in, 10, for formal sector, 26 12, 14t, 17t, 75-77 INDEX 265 social security contributions in, education and employment in, 22, 34-35 27 stabilization, public employment employment stability in, 27-28 and informality in, 219-23 growth, inequality and uncer- temporary contracts' rules in, 6 tainty in, 227-29 unemployment compensation macroeconomic challenges in changes in, 20t 1980s of, 217-19 unemployment compensation Servicio Nacional de Capa- in, 21, 22, 23-24, 28n23, citaci6n y Empleo, 208-9 75-77 severance pay mandates in, 10, wage and nonwage labor costs, 12, 14t, 16, 17t 79t unemployment compensation wage indexation regimes, changes in, 20t 1980-2000, 78t unemployment compensation wage laws in, 77 in, 21, 22 Brazilian National Statistical Institute, union and collective bargaining PNAD and, 47 reforms in, 28n5 China, world market and unskilled CAGED (Cadastro Geral de workers of, 160 Empregados e Desempregados, Christensen, L., 244 Brazil), 71-72 Clark, K. B., 112 Camargo, Jose Marcio, 222 CLTistas, Consolidacao das Leis do Capital Trabalho and, 65n3, 73-74 calculation of price of, 132 Cobb-Douglas representation of tech- costs in Argentina, data for nology, 133 calculation of, 135-36 calculation of, 241-42 Caribbean countries, labor regulations own-wage elasticity and output in, 28n4 elasticity estimates using, CASEN household surveys, Chile, 138-39, 139t 161, 163-64, 184n9, 212n4 Collective bargaining description of, 183n4 in Argentina, labor force partic- unemployment rates among rich ipation and, 118 and poor analysis, 202-4 in Brazil in 1990s, 81 wage inequality analysis using, Colombia 168 compensation funds in, 16, Chile. See also Flexible labor markets; 18-19 Wage gap labor legislation changes in, 6 compensation funds in, 16, severance pay mandates in, 10, 18-19 11, 14t, 17t demographics of labor force, in- social security contributions in, come and GDP for, 233nl 22 earnings and employment out- temporary contracts' rules comes in, 155-57 changes in, 6, 7, 9 z66 INDEX Colombia (continued) GDP, productivity, and unemployment compensation unemployment growth after, changes in, 20t 130-31 unemployment during 1990s, 2 labor force participation cyclical Compensation funds, 16, 18-19 and trend factors and, withdrawals from, 28n18 117-18 Conflict and litigation, severance pay unemployment and, 224 mandates and, 11 Corruption, employment and, 28n3 Consolidated Labor Code (CLT), Corseuil, Carlos, 10-11 Brazil, 73-74 Cortazar, Rene, 16 Constant elasticity of substitution Cost minimization, as factor in labor technology, 133 demand, 132 cost function approach to, 140t Cowan, Kevin, 201 econometric calculation of, 242 Cox Edwards, Alejandra, 205, 212n3 estimates of, 151 Cross-substitution effect own-wage elasticity and output income effect and, 125n3 elasticity estimates using, labor supply and, 106 139-42 Cyclical effects production function approach of unemployment, 112, 113t, to, 142t 126n7 Constant returns to scale (CRS) of unemployment, in econometric estimation of, Argentina, 123-24 244-45 wage inequality in Chile, 172 as factor in labor demand, 132 Constitutions de Ferranti, David, 2, 18, 208 Brazilian, reform of, 74 De Gregorio, Jose, 201 Brazilian (1988), labor-related Delfino, Jose A., 139, 148-49 clauses in, 67, 74-75, 233n2 Diamond, Peter, 195 Brazilian (1988), on public Discouraged worker effect servant layoffs, 38, 60 in Argentina, 102 Brazilian labor law definition of, 98 contradictions with, 81 female LFP in Argentina and, Consumer price index, annual 106 percent change in, for labor de- labor force participation mand estimates in Argentina, 138 decisions and, 112 Controlled public-private sector wage in U.S. vs. Argentina during gap, Brazil, 51-53 recessions, 118 sector and class of workers and, Dispersion, natural measure of, for 55-56 wage inequality study, 167 worker differences and, 53-55, Domestic servants, in Chile, wage 53t inequality study and, 164 Convertibility Plan, Argentina Dominican Republic, severance pay (1991), 129 mandates in, 10, 14t description of, 126n12 Drazen, Allan, 232 INDEX 267 Economic Commission for Latin Edwards, Sebastian, 205, 212n3 America (ECLAC), 217 Elasticity of substitution Ecuador definition of, 152n3 compensation funds in, 19 estimates of labor demand in temporary contracts' rules Argentina, 151 changes in, 6, 7 Ellas, Victor, 139 unemployment compensation Employment Law, Argentina (1991), in, 22 130 Education Employment levels. See also Job during 1990s, 2 instability in Chile, for poor children, in Argentina, labor force growth 175-76 vs. growth in, 129-30 in Chile, impact of policies to in Argentina, population reduce wage inequalities growth and, 104 with, 179t job security legislation and, 4-6 in Chile, mean and median in Latin America during 1990s, estimates of wage inequality 217 and, 173f Employment precariousness in Chile, mean years of school- in Chile, 195-98, 216, 228-29 ing by decile group, 169, in Chile, poor workers and, 171, 171t 201-204, 202t, 203t in Chile, of parents by child's in Latin America, 217 school type, 177t model for measurement of, in Chile, quantile estimates of 192-93, 194f wage inequality and, 174f Encuesta de Caracteriaci6n Socio- in Chile, rates of return to, economica Nacional (CASEN) 155-56, 161, 184nlO, 187, household surveys, 161, 163-64, 228 in Chile, wage inequality and, 184n9, 212n4 169, 171-73, 172t, 175, description of, 183n4 227-28 wage inequality analysis using, in Chile, wage inequality for 168 blue-collar workers and, Equal employment opportunity laws, 173, 175, 175f, 189 in Chile, 180 in Chile, wage inequality for Estatuto Docente, Chile, 184n12 white-collar workers and, education reforms and, 182 173, 174f 188 Exchange rates, in Argentina, 99, in Chile, wage inequality pat- 152n6 terns and trends and, 164 Experience rating, 206, 213n6 public-private pensions differ- ences in Brazil and, 59-60 Fallon, Peter R., 28nl3 public-private sector compensa- Family workers, severance payments tion differentials in Brazil and, 28nl4 and, 38 Ferreira, Francisco, 159, 164, 166, severance pay mandates and, 13 183n4, 183n5 268 INDEX FGTS (Fundo de Garantia do GDP (gross domestic product) Tempo de Servi:o, Brazil), 19, in Argentina, data for 30n23, 35 calculation of, 135 dismissal costs and, 76-77 and capital costs in Argentina, reform of, 81 data for calculation of, severance rules and unemploy- 135-36 ment compensarion under, as main determinant of labor, 75 131 withdrawals from, 94n9 Gender. See also Women Fixed-term contracts. See Temporary and controlled public-private contracts sector wage gap, Brazil, 54 Flexible labor markets, in Chile. See public-private pensions also Employment precariousness differences in Brazil and, employment and earnings and, 59-60 191 public-private sector compen- expected tenure analysis, 195-98 sation differentials in Brazil income inequality and, 160, 226 and, 38 issues, methodology, and data reducing public employment in for, 192-94, 212nl Brazil and, 61-62 precariousness and poor work- Generalized Leontief representation ers analysis, 201-204, 202t, of technology, 133-34, 138 203t calculation of, 242-43 unemployment duration analy- estimates of, 144, 145t sis, 198-99, 199t, 200f 201 Giacchino, Leonardo, 112 unemployment insurance sys- Gill, Indermit S., 127n16 tem and, 203-6, 207t, 208 Gini coefficient, for wage inequality Formal sector employment study, 166-68, 184n8 in Brazil, distribution analysis Greene, William H., 244 of, 85-87, 88f Guasch, Jose Luis, 7, 127n 16, 205 in Brazil, distribution of hours GuatemaLa worked among, 91 labor legislation changes in, 6 in Brazil, labor legislation severance pay mandates in, 10 effects on, 83-85 Guyana, severance pay mandates in, in Brazil in 1980s, 70 10 minimum wage and wage distribution for, 89-90, 89f Habit formation in labor markets, severance pay mandates and, labor codes and, 84-85 12-13 Haindl, Erik, 191, 228 signed work cards and, 83, labor flows model, 253-57 94n14 Hamermesh, Daniel S., 136, 147, Forteza, Alvaro, 233n6 150, 242 Fundo do Amparo ao TrabaLhador Hanoch, Giora, 244 (FAT), Brazil, 23, 80 Heckman, James J., 13, 28n10, 112 INDEX 269 Hiring practices, affirmative, INDEC-EPH household surveys, legislation for wage inequality Argentina, 134-35 reduction and, 180 changes in sample sizes of, Hopenhayn, Hugo, 7, 130 126nl3 Hours worked predictive power of LFP and, in Brazil, as benchmark for 119 "fair" remuneration, 93 Indice Nacional de Precos ao in Brazil, effective work week Consumidor (INPC) deflator, 43 before and after new Inflation constitution, 92t in Argentina, female LFP and, in Brazil, maximum without 121-22 overtime assessment of, 84, in Brazil, variations in, 84 85, 94nl3 labor legislation and, 221 in Brazil, regulation of, labor market issues and, 217 90-92 legally-specified minimum in Brazil, weekly maximums salaries in Brazil and, 68 for, 82 wage adjustments and, 4 enforcement of limits on, Informal sector employment public sector efficiency in during 1990s, 2, 25 Brazil and, 64 in Argentina, 28n22 labor demand as total of, in Brazil, 34, 68, 222-23 152nS in Brazil, distribution analysis and maximum limits in Brazil, of, 85-87, 88f distribution of, 91-92 in Brazil, distribution of hours Hungary, job retraining program worked among, 91 evaluation in, 210 in Brazil, labor legislation effects on, 83-85, 233n4 Illiteracy, workforce in Brazil, labor turnover and labor legislation and, 221 job quality in, 71-72, 72t legally-specified minimum in Brazil, measures of confor- salaries in Brazil and, 68 mity with labor and social Income, by labor in Brazil from security laws, 93t 1980s to 1990s, 72-73, 94n6 in Brazil, minimum wages in, Income effect, labor price relative to 85-86 capital in Brazil, one minimum wage in Argentina, 106 earners in, 86-87 cross-substitution effect and, in Brazil in 1980s, 70 125n3 in Chile, unemployment Income inequality. See also Flexible compensation and, 208 labor markets; Wage gap as choice not default option, in Chile, trends in, 161 94n2 prolabor bias to labor laws in labor legislation and, 221 Brazil and, 69, 222 in Latin America, 217 270 INDEX Informal sector employment Killingsworth, Mark R., 112 (continued) Koenker, R., 251 lighthouse effect (efeito-farol) Kugler, Adriana, 9, 11, 13, 19 of regulations on, 87-88, 93 in Mexico, 233n5 Labor contracts, in Argentina, govern- minimum wage and wage dis- ment alternatives for, 130, 152nl tribution for, 89-90, 89f Labor costs severance pay mandates and, 12 in Argentina, data for severance payments and, 28n14 calculation of, 135 small firms and, 28nl, 28n12 labor supply vs., 137-38 unemployment compensation Labor courts, in Brazil and, 21, 206 policy-setting powers for, 80, Instituto Nacional de Seguridade 82, 94 Social (INSS), Brazil, 57 use of, 95nl2 public-private pension gap Labor demand in Argentina, 224-25 calculation using, 58-60 basic methodology for subsystems of, 65n7 calculation of, 132-38 Instrumental variables, constant Cobb-Douglas technology for elasticity of substitution and, 141 estimates of, 138-39, 139t Inter-American Development Bank constant elasticity of (IDB), 2, 217 substitution technology for International Labour Organization estimates of, 139-42 (ILO), 2, 9-10, 217 data sources for estimation of, 134-35 Job instability, Chilean economic econometric specification and reforms and, 156 results, 138-49 Job search assistance USA), 210 elasticity of substitution Job security legislation estimates of, 151 during 1980s, 28n7 human capital vs. labor and, about, 4-6, 28n6 152n6 in Chile, flexible labor markets issues affecting calculation of, and, 191 136-38 compensation finds, 18-19 labor force participation and, distortionary, political process 127n16 and, 28n8 output data sources for, 135 labor demand and, 28n13 output elasticity estimates of, severance payments, 10-18 150 temporary contracts, 6-10 policy implications, 149-51 Job training programs, evaluation of, single factor demand equations 209-210, 213n7 for estimates of, 143-44 Jorgenson, D. W, 153nll technical change and, 147-49 technology representations of, Khaled, Mohamed S., 147, 153nl 1 133-34 INDEX 27I transcendental logarithmic informal sector and, 221 technology for estimates of, on job security, 4-6 144-47 Labor market reforms wage elasticity estimates of, in Argentina, 99 149-50 in Argentina, output elasticity years needed to reduce vs. wage elasticity and, 97 unemployment to 8%, 151 t in Brazil, focus on relative Labor flows model, Haindl's, 253-57 benefits in, 220-21 Labor force definition of, 3 in Argentina, growth in, 129 during economic improvement in Argentina, variations in, 98 cycles, 231 in Brazil, income of, 72-73 hazards of generalizations entrants in Chile, employment about, 229-30 flows and, 195, 197t labor outcomes and, 230-31 flows, 194f in Latin America, 215-16 growth during 1990s of, 2, 217 in Latin America, cross-country influences on, 101 studies of, 233n6 turnover and job quality in political economy of a country 1990s, 71-72 and, 94nl Labor force participation (LFP), 101 quantification of likely effects in Argentina, policy implica- of, 231-32 tions of, 124-25, 124f regional context for, 216-19 225-26 Labor markets in Chile, 162 in Argentina, summary of, Labor policies. See Labor market 122-24 reforms data sources for, 125n2 Latinobarometro survey, 2-3 labor demand and, 127nl6 Leiva, Fernando, 160, 191 methodology for determinants Length of Service Program, of of, 110, 112, 114 Instituto Nacional de Seguridade predictive power of models of, Social (INSS), Brazil, 57 119, 120t Leontief representation of technology, trend and cyclical effects, generalized, 133-34 114-18, 115t, 116t econometric calculation of, Labor legislation. See also Labor 242-43 market reforms estimates of, 144 in Argentina, 5, 27, 216 Litchfield, Julie, 159, 164, 166, in Brazil, effects on informal 183n4,183n5 and formal workers, 83-85 Lopez, Ramon, 181 Brazilian reforms proposed and Lora, Eduardo, 2, 25 implemented in 1990s, Low-skilled workers, severance pay 80-82,95nll mandates and, 18 informal sector, women workers Lucas, Robert, 29nl3 and, 3 Lundberg, Shelley, 106, 126n II 272 INDEX MacIsaac, Donna, 11, 12, 16, 28n17 in Chile, average salaries rise Maloney, William F., 16,83,87, 233n5 and, 184nll sample of, 94n15 legislation for wage inequality Marcel, Mario, 201 reduction and, 180, 181 Mirquez, Gustavo, 2, 11 Ministerio do Trabalho, Brazil, 94n7 Mazza, Jacqueline, 23 Modalidades promovidas (Argentine Mean regression, wage inequality temporary contracts), 8 analysis model and, 163 Monetary rates, in Argentina, 99 Men. See also Gender; Women Montenegro, Claudio E., 13, 28nlO, in Argentina, labor force 166, 167, 173, 181, 183, 196, participation of, 114-15, 197, 212n2 117,123,126n9 Moral hazards, of unemployment in Chile, average unemploy- compensation, 21, 206 ment rates for, 202-3, 202t in Chile, expected tenure and Nicaragua, severance pay mandates age of, 196-97 in, 10, 14t, 17t Mexico Nonpaid workers, in Brazil in 1980s, growth vs. job creation in 70 1990s for, 1 Nufiez, Jairo, 87 privatization in, 24 sample of, 94n15 unemployment compensation in, 22 Oaxaca decompositions, for income Migrants to Argentina, population inequality analysis, 161, 183 growth and, 103-4, 125nl Ohta, 153nlO Mincer, Jacob, human capital earn- Old Age Program, of Instituto ings function of, 162, 178, 183nl Nacional de Seguridade Social Minimum wages (INSS), Brazil, 57 in Brazil, 82 Older workers in Brazil, as benchmark for severance pay mandates and, 15 "fair" remuneration, 93, 222 temporary contracts in in Brazil, by metropolitan Argentina for, 8 region, wages exactly equal Ordinary least squares (OLS). See to one, 86-87, 86t Quantile regressions in Brazil, changes in, wage Organisation for Economic Coopera- adjustments and, 90t tion and Development (OECD) in Brazil, legally-specified, 68 female LFP rates in, 106, 126n5 in Brazil, regulation of, 85-90 social security contributions in, in Brazil, wage distribution of 22 formal and informal sector Output elasticity of aggregate workers and, 87-90 employment in Brazil, wage earners paid in Argentina, 151, 224-25 above, 85-86, 85t in Argentina, employment in Brazil, wages exactly equal to growth from economic one, 87, 87t expansion and, 97 INDEX 273 difficulties with using, 136-37 Pension contributions World Bank comparative data public sector efficiency in Brazil on, 152n2 and, 64-65 Output elasticity of labor demand. by public workers in Brazil, 65n8 See also GDP Pension funds, in Chile, mandatory, in Argentina, 129, 131, 136, 157 224-25 Pensions, public-private differences in in Argentina, estimates of, 150 Brazil, 56-60, 59t, 63, 64 econometric calculation of, Performance monitoring, public 241-45 sector efficiency in Brazil and, 64 Overtime pay, PME data and, 91 Peru Own-wage elasticity of labor demand compensation funds in, 16, in Argentina, 131, 151, 224-25 18-19 in Argentina, estimates of, dismissals and severance pay- 136-38,149-50 ments in, 28nl6 calculation of, 132 labor legislation changes in, 6 World Bank comparative data older workers in informal on, 152n2 employment in, 15 privatization in, 25 Paes de Barros, Ricardo, 10-11, 49 severance pay mandates in, 5, Pages, Carmen, 13, 25, 28nI0, 181, 10, 11, 12, 14-1St, 17t 196,197,212n2 temporary contracts' rules Panama changes in, 6, 7, 9 compensation funds in, 16, unemployment compensation 18-19 changes in, 20t dismissals and severance unemployment during 1990s, 2 payments in, 28n16 wage earners in, 28nI5, 28n17 labor legislation changes in, 6 Pessino, Carola, 105-6, 112, 126n6, severance pay mandates in, 14t, 127nl6,152n6 17t PIS/PASEP (microenterprise unemployment compensation support), in Brazil, 80, 94nlO changes in, 20t PME (Pesquisa Mensual do Paraguay, unemployment during Emprego, monthly household 1990s, 2 surveys), 70 Paredes, Ricardo, 209 minimum wage and wage Payday, in Brazil, 68, 222 distribution for formal and Payroll expenses, Brazil, 38, 64, informal sector workers, 79-80. See also Social security 87-89,89f progressive scale for, 94n8 scope of. 94n4 quality of programs funded by, on structure of employment, 71 67-68,81,94 PNAD (Pesquisa Nacional por PDVs (programas de demissao volun- Amostra de Domicilios, annual taria), Brazil, 39, 60-63 household surveys), Brazil Pencavel, John, 112 on earnings, 1981-95, 43 274 INDEX PNAD, Brazil (continued) Public-private wage gap, uncontrolled on employment, 1981-95, 40 data for, 47 on public-private sector heterogeneity in public sector compensation, 38 and, 49-50 scope of, 94n4 in Latin America, 217 on structure of employment, overall, 47-49, 65n4 70-71 regional differences in, 50-51 on tenure at current job, 65n6 Public sector efficiency, Brazilian on uncontrolled public-private about, 37-40, 65nl, 216, 219 wage gap, 47 annual growth in employment, wage bill numbers and, 45 average earnings, and wage Polachek, S., 183nl bill, 46f Political violence, employment and, controlled public-private wage 28n3 gap, 51-53 Poor workers earnings, 1981-95, 43-44, 43t in Chile, educational earnings premium, by sector and opportunities and, 175-76 class of worker, 1995, 56t in Chile, government- employment in Brazil, sponsored training and, 1981-95, 40-43 208-11, 213n7 public-private compensation in Chile, unemployment and, differentials, 38-40, 219-20 201-4, 202t, 203t public-private pension differ- in Chile, wage inequality and, ences in, 56-60, 63, 219-20 159-61 reducing public employment employment and, 28n3 and, 61t severance pay mandates and, 16 sectoral shares in employment. Population size 1981-95, 41t in Argentina, 102-4, 103t share of workers by class of labor force and, 101 worker distinction, 1995, Private sector employees 42t in Argentina, labor force uncontrolled public-private growth and, 130 wage gap, 47-51 in Brazil, 65n5 voluntary severance programs Probit analysis, of unemployment in (PDVs) and, 60-63 Argentina, 110, 126n6 wage bill, 1992-95, 44-47 Production labor, as factor in labor wage bill and share of employ- demand, 132 ment of different employ- Profit maximization, calculation of, ment sectors, 45f 132 worker differences and, 53-55 Programas de demissao voluntaria Public sector employment (PDVs), Brazil, 39, 60-63 in Brazil in 1980s, 70 Public debt, Argentina's default on, earnings and, 24-25 99 in Latin America, 217 INDEX 2z75 Quantile regressions Savings accounts, individual, calculation of, 247-51 unemployment insurance systems for income inequality analysis, and, 206, 208 161, 174f Scholarships, for job training in Chile, 209 Race Seguro de Desemprego program, ahd controlled public-private Brazilian unemployment sector wage gap, Brazil, 54 compensation program, 77 public-private sector compensa- Self-employment tion differentials in Brazil in Argentina, 152n4 and, 38 in Brazil, informal, 222 RAIS (Relac,o Annual de Informac6es in Brazil in 1980s, 70 Sociais data, Brazil), 71-72 in Chile, wage inequality study Rama, Martin, 11, 12, 16, 28n17, and,s164 233n6 labor legislation effects on, 83 Ramos, Lauro A., 28n23 Self-insurance. See Compensation Real exchange rate fufnds female LFP in Argentina and, Service sector, in Brazil, turnover in implications of, 126nl5 formal, 71, 94n5 i of fr , lo dmn Servicio Nacional de Capacitaci6n y logs om, Cor laCor ieman- estimates in Argentina, 137 Empleo (SENCE, Chile), 208-9 Real GDP index, for labor demand Severance estimates in Argentina, 137 in Argentina, mandates for, Real GDP per capita, for labor de- 130 mand estimates in Argentina, 137 in Brazil, 75-77 Real wage growth, in Latin America, in Brazil, total cost of, 76t 217 reform of Brazilian laws on, 81 Recessions, employment reduction World Bank comparative data and, 195 on, 152n2 Redundancy programs, Brazil, 39 Severance payments, 10-18 Regime Geral da Previdencia Social in Chile, earnings inequality (RGPS, National Security System), and, 156 Brazil, 57, 65 deferred compensation versus, Regime Juridico Uinico (RJU, public 205-6 service pension program), Brazil, dismissal causes and changes in, 57-58 17t Returns to scale. See Constant returns groups not covered by, 28nl4 to scale (CRS) job stability and, 28n16 Revenga, Ana, 150 job turnover and, 28nll Riveros, Luis, 180 laws on, unemployment Ruiz-Tagle, Jaime A., 183, 183n7 compensation and, 22 tenure-related, permanent Saavedra, Jaime, 5 workers and, 4 276 INDEX Shephard's condition SURE procedure estimation of single factor estimating returns of scale and, demand equations and, 242 245 labor demand estimation and, generalized Leontief technology 143-44 and, 243 Shimer, Robert, 28nl9 translog representation of Siebert, W S., 183nl technology and, 153n8 Simultaneity bias, cyclical effects of unemployment calculation and, Tax burden, social security and, 22, 126n7 28n20 Single factor demand equations, 1 53n7 Tax credits, for job training in Chile, econometric estimation of, 242 209 for labor demand estimates, Technical change 143-44,143t translog estimations of labor Sistema de Medici6n de la Calidad de demand and, 147-49, 148t la Educaci6n (Educational Quality unskilled workers, income Measurement System) (SIMCE), inequality and, 160 Chile, 176 Teenagers, labor force participation in Small enterprises. See also Argentina, 105-6 Self-employment cyclical and trend factors of, range of quality in, 28n2 116-17 severance payments and, 28nl2 Temporary contracts, 6-10, 7t, 28n9 Socialist policies, prolabor bias to in Brazil, reform of, 82 labor laws in Brazil and, 69 limitations on, permanent Social security. See also Payroll workers and, 4 expenses, Brazil Tenured employment reform in Brazil, 39, 223 in Chile, recessions and, RGPS (private pension 195-96, 196f 212n2 program), Brazil, 57, 65 and controlled public-private RJU (public service pension sector wage gap, Brazil, program), Brazil, 57-58 53-55, 53t, 64 supplemental, in Brazil, 59, precariousness of employment 65n9 flows in Chile and, 195-98 tax burden and, 22, 28n20 for public vs. private workers in Solimano, Andres, 201 Brazil, 65n6 Spread (econometric trend measure- severance pay and, 4 ment), for wage inequality study, severance pay mandates and, 166,167 13, 14-1St, 15 Substitution effect. See also of short duration, severance Constant elasticity of substitution payments and, 28n14 technology Tequila Crisis, Argentina (1995) labor price relative to capital, in GDP decline, employment Argentina, 106, 129, 224 decline, and unemployment Summers, L. H., 112 increase during, 130 INDEX 277 unemployment and, 103 flows vs. stocks analysis of, unemployment compensation 192-93, 194f and, 23 increased labor force Total factor productivity growth, participation and, 98 147-48, 153nl1 in Latin America, 217 Trade liberalization, in Argentina, in Latin America during 1990s, labor force and wages and, 129 217 Transcendental logarithmic (translog) and LFP over business cycle, 102 representation of technology, of of men in Argentina, labor labor demand, 134, 138 force participation and, econometric calculation of, 114-15 243-44 Unemployment compensation, 19-24 estimates of, 144-47, 146t in Chile, flexible labor markets nonneutral technical change and, 191, 203-6, 207t, 208 and, 147-49, 148t deferred, severance payments SURE procedure and, 153n8 versus, 205-6 Trend factors, in unemployment, legislated changes in, 20t 112, 113t severance pay mandates as, 16 in Argentina, 123 Union reprcscntation, in Brazil, Two-factor demand equations. See firm-level conditions and, 82 Transcendental logarithmic Unions, on group-specific, fixed-term (translog) representation of labor contracts in Argentina, 130 technology United States, labor force participa- tion cyclical and trend factors in, Unemployment 118 during 1990s, 2 University of Chile Households in Argentina, 99, 101, 223-24 Survey, 161, 163-64 in Argentina, after Convertibil- flexible labor markets analysis iry Plan (1991), 130-31 using, 192 in Argentina, labor force wage inequality analysis using, participation and, 118, 123, 166-68, 168t 124-25, 124f Urban population Greater Buenos in Argentina, labor market Aires, for labor demand estimates reform effectiveness and, in Argentina, 137 226-27 Uruguay in Argentina, trends in LFP growth vs. job creation in and, 107, 108t, 109-10, 1990s for, 1 111 t older workers' unemployment in Brazil, 67, 223 in, 15 in Chile, duration analysis of, social security contributions in, 198-99, 199t, 200f 201, 22 21 2n3, 21 2n4 unemployment compensation in Chile, poorest households in, 22 and, 201-4, 202t, 203t unemployment during 1990s, 2 278 INDEX Venezuela, Republica Bolivariana de in Chile, quantifying impact of compensation funds in, 16, policies to reduce, 177-80, 18-19 179t severance pay mandates in, 10, in Chile, returns to education 1St, 17t and, 169, 171-73, 172t, unemployment compensation 173f 175, 228 changes in, 20t in Chile, trends in hourly, 186 unemployment compensation econometric concepts of, in, 22 237-38 unemployment during 1990s, 2 education effects on, 155-56 Voluntary severance programs, Brazil, labor market issues in, 161-62 39, 60-63 in Latin America, 217 model and methodology for Wage Adjustment Law, Brazil (1965), analysis of, 162-63 77 postprimary education access Wage elasticity. See also Output and quality and, 181-82 elasticity of aggregate employment; Wage levels. See also Public sector Output elasticity of labor demand efficiency econometric calculation of during 1990s, 2 labor demand and, 241-45 in Argentina, labor force of employment in Argentina, participation and, 118 labor policy reforms and, 97 in Brazil, 223 of labor demand in Argentina, in Brazil, distribution analysis estimates of, 149-50, 224-25 of, 85-87, 88f in the United States, 151-52 in Brazil, public-private sector Wage gap (inequality) differentials in, 38 calculation of controlled, labor legislation effects on, 238-40 84, 85 in Chile, about, 159-61, 216 wage indexation regimes in in Chile, data for analysis of, Brazil, 1980-2000, 78t 163-64, 183n3, 183n4, Wealthy workers, severance pay 183n5, 183n6, 183n7 mandates and, 16 in Chile, implications of Weekly work hours. See Hours policies to reduce, 176-77 worked in Chile, individualized sever- Wholesale price index, in Argentina, ance accounts and, 157 data for calculation of, 135-36 in Chile, market intervention Willis, R., 183nl and, 180-81 Women. See also Gender; Teenagers in Chile, measures of, 166-68, in Argentina, unemployment 168t rates and, 23 in Chile, patterns and trends in Chile, average unemploy- in, 164, 165t, 166, 169f ment rates for, 202-3, 203t, 170f 228 212n5 INDEX 279 in Chile, expected tenure and labor force participation in LFP of, 196, 197, 198t other countries, 106-7, in Chile, wage inequality study 126n4 and, 164 severance pay mandates and, labor force growth during 13, 16, 18 1990s and, 2 unemployment rates for, 6 labor force participation and, Wood, Adrian, 160 101 Workers, classes of labor force participation public-private sector compen- decisions by, 112 sation differentials and, 41 labor force participation tenure, and controlled public- determinants for, 119-22, private sector wage gap, 122t, 126nl4 Brazil, 53-55 labor force participation in Workforce illiteracy, legally-specified Argentina, 102, 104-5, minimum salaries in Brazil and, 68 105t, 106, 122-23, 126n5, World Bank, 152n2, 217, 225 225, 226 World Trade Organization labor force participation in on informal sector Argentina, added worker employment, 28n1 effect and, 106, 117-18, informal sector worker 123, 126nlO definition of, 94n14 labor force participation in Argentina, cyclical and trend Young people factors of, 1 14-15, 115t, severance pay mandates and, 116t, 117, 118 13,16,18 labor force participation in temporary contracts in Argentina, husband's em- Argentina for, 8 ployment status and, 122t unemployment rates for, 6 labor force participation in Latin America by, 217 Zellner's estimator (SURE), 243 tr~~ ~~ - --h ., ;'L L 3 C IZ0 - r I o NUeqp; )jD2qpaP :112qW3 DIjueqOjJ0.wMMMAA I2UJeiui pi 7 ,__Z_SI ____ 1 1 -_ vs t AA Do '0u9oA u'seH v- )INVS alGOM 3HIl A > # < > E F t24ˇ .S.r.- siuapni ,puC JdQJP.3 li sJaeew V'iI0 ~i 01 aJalui 40 eq pInots .1ooq Ni'q uu 2uia i Id2L;jai dpq .isunri palu ap ,Iow le'4 aL4J1 S JSeld2lb 61JIphpIJC: PLJc Qoo7npoj eI U-u jo 1p}Dpoq Jlq2peJ JIP IjIiL! I elwqap Aoijod iOQej ag OJ sispq atil se *4ua'nflte a%olDaIqo aiawowjd oi'sjua'ui iAueq palosuods JOAt U O pese: -I 'AJI,Od J20027. .&u 44P'fl -jII ' uiu6wAojdwe pup 5ijc~,u 'Ul e kt'0'CdLuA e'a .ids.d..a' iallip e- IrUod-eu '.;; F.O->";; ;.- pUF Afddrn- ioqej UO buisroj, :Arl-eds-ad *oel uoaup ienQi- 4, SS F t JAS . '-'$z\ES *;-¢'V6''- , nn'' ' qiinc" ipee LEOJddp sJ04rqF, al4 sQ65 L. aJPI --q) eauI. e)UaUv pn u- F /swiolai joqP .f' uzlP e n l u ww O)J?peQJql ;141 EP''- aoid ieidetp Ajr2v_noon uv SnI 'M un5SFIlIlu .J4 - <, - r .-*''' ' ~ s7e > i'4;rJsal; paUllu. i2q1l rrn5'JO-nd ioqel Awap;uiw2etjoirq-i- pLU J O ;-t I t se > >>t +; ah ;-4fis *X;U- rzeig euuue6jw ur r ,Se,Jc,uYi .OD qeg tuIpC&3-siOllq'J,lu0) 21 C 4 2 4,; t9 r:1,; ,;. - 'htw''; '., ' ,,e-E,s .... .... _ _1I .s J J2UuPW .4ilrnu b 2 ,ii suoI ..ant k);jod, J0q2_ uP dW i sue o s enb"aq:ai- WI, a U''r P)IJaWve LrI 2111 ,rJ I..j: r 0i- P'4'UIlulu '4I SL2 01 peJJal-iJ lj3q e i Wqm LJOjaj1j LC'IW ,Jqejf)uEApeCt ;I (Ajce_ joo we d :1 pJo;\fl C V ________F,; - 1', x , h . ;Jf't 4s SX '- '; .