45623 THE IMPACT OF MACROECONOMIC POLICIES ON POVERTY AND INCOME DISTRIBUTION Macro­Micro Evaluation Techniques and Tools François Bourguignon Maurizio Bussolo Luiz A. Pereira da Silva Editors THE IMPACT OF MACROECONOMIC POLICIES ON POVERTY AND INCOME DISTRIBUTION Macro-Micro Evaluation Techniques and Tools THE IMPACT OF MACROECONOMIC POLICIES ON POVERTY AND INCOME DISTRIBUTION Macro-Micro Evaluation Techniques and Tools François Bourguignon Maurizio Bussolo Luiz A. Pereira da Silva Editors A copublication of Palgrave Macmillan and the World Bank © 2008 The International Bank for Reconstruction and Development / The World Bank 1818 H Street, NW Washington, DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org E-mail: feedback@worldbank.org All rights reserved 1 2 3 4 11 10 09 08 A copublication of The World Bank and Palgrave Macmillan. Palgrave Macmillan Houndmills, Basingstoke, Hampshire RG21 6XS and 175 Fifth Avenue, New York, NY 10010 Companies and representatives throughout the world Palgrave Macmillan is the global academic imprint of the Palgrave Macmillan division of St. Martin's Press, LLC and of Palgrave Macmillan Ltd. Macmillan® is a registered trademark in the United States, United Kingdom, and other countries. Palgrave® is a registered trademark in the European Union and other countries. This volume is a product of the staff of the International Bank for Reconstruction and Development /The World Bank. The findings, interpretations, and conclusions expressed in this volume do not necessarily reflect the views of the Executive Directors of The World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgement on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this publication is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The International Bank for Reconstruction and Development / The World Bank encourages dissemination of its work and will normally grant permission to reproduce portions of the work promptly. For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance Center Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; telephone: 978-750-8400; fax: 978-750-4470; Internet: www.copyright.com. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, The World Bank, 1818 H Street, NW, Washington, DC 20433, USA; fax: 202-522-2422; e-mail: pubrights@worldbank.org. ISBN: 978-0-8213-5778-1 (soft cover) and 978-0-8213-7268-5 (hard cover) eISBN: 978-0-8213-5779-8 DOI: 10.1596/978-0-8213-5778-1 (soft cover) and 10.1596/978-0-8213-7268-5 (hard cover) Library of Congress Cataloging-in-Publication Data The impact of macroeconomic policies on poverty and income distribution : macro-micro evaluation techniques and tools / edited by François Bourguignon, Maurizio Bussolo, and Luiz Pereira da Silva. p. cm. Includes bibliographical references and index. ISBN 978-0-8213-5778-1 -- ISBN 978-0-8213-5779-8 (electronic) 1. Economic assistance--Evaluation. 2. Poverty. 3. Income distribution. 4. Economic assistance--Developing countries--Evaluation. 5. Developing countries--Economic policy--Case studies. I. Bourguignon, François. II. Silva, Luiz A. Pereira da. III. Bussolo, Maurizio, 1964- HC60.I4147 2008 339.2'2--dc22 2007040478 Contents Preface xi Acknowledgments xiii Contributors xv Abbreviations xvii 1 Introduction: Evaluating the Impact of Macroeconomic Policies on Poverty and Income Distribution 1 François Bourguignon, Maurizio Bussolo, and Luiz A. Pereira da Silva Part I. Top-Down Approach with Micro Accounting 2 Winners and Losers from Trade Reform in Morocco 27 Martin Ravallion and Michael Lokshin 3 Trade Options for Latin America: A Poverty Assessment Using a Top-Down Macro-Micro Modeling Framework 61 Maurizio Bussolo, Jann Lay, Denis Medvedev, and Dominique van der Mensbrugghe Part II. Top-Down Approach with Behavioral Micro Simulations 4 Examining the Social Impact of the Indonesian Financial Crisis Using a Macro-Micro Model 93 Anne-Sophie Robilliard, François Bourguignon, and Sherman Robinson 5 Can the Distributional Impacts of Macroeconomic Shocks Be Predicted? A Comparison of Top-Down Macro-Micro Models with Historical Data for Brazil 119 Francisco H. G. Ferreira, Phillippe G. Leite, Luiz A. Pereira da Silva, and Paulo Picchetti v vi CONTENTS Part III. Macro-Micro Integrated Techniques 6 Distributional Effects of Trade Reform: An Integrated Macro-Micro Model Applied to the Philippines 177 François Bourguignon and Luc Savard 7 Simulating Targeted Policies with Macro Impacts: Poverty Alleviation Policies in Madagascar 213 Denis Cogneau and Anne-Sophie Robilliard 8 Wealth-Constrained Occupational Choice and the Impact of Financial Reforms on the Distribution of Income and Macro Growth 247 Xavier Giné and Robert M. Townsend Part IV. Macro Approach with Disaggregated Public Spending 9 Aid, Service Delivery, and the Millennium Development Goals in an Economywide Framework 283 François Bourguignon, Carolina Díaz-Bonilla, and Hans Lofgren 10 Conclusion: Remaining Important Issues in Macro-Micro Modeling 317 François Bourguignon, Maurizio Bussolo, and Luiz A. Pereira da Silva Index 325 Box 3.1 Consistency Issues 67 Figures 1.1 Schematic Representation of the Top-Down Modeling Approach 12 2.1 Impacts of Trade Reform Policies on Poverty in Morocco 40 2.2 Frequency Distributions of Gains and Losses for Trade Policies 1 and 4 42 2.3 Absolute and Proportionate Gains for Policies 1 and 4 43 CONTENTS vii 2.4 Production and Consumption Decomposition of the Welfare Impacts for Policy 4 44 2.5 Net Producers of Cereals in the Distribution of Total Consumption per Person in Rural Areas of Morocco 45 5.1 A Simplified Overview of the Top-Down Macro-Micro Framework 121 5.2 An Overview of the Main Blocks of the Macro Model 130 5.3 Comparison between Actually Observed Changes and Experiment 1, Using Representative Household Groups 155 5.4 Comparison between Actually Observed Changes and Experiment 1, Using Representative Household Groups, and Experiment 2, Using Pure Micro Simulation Model 156 5.5 Comparison between Actually Observed Changes and Experiment 1, Using Representative Household Groups, Experiment 2, Using Pure Micro Simulation Model, and Experiment 3, Using Full Macro-Micro Linking Model 157 6.1 Iterative Resolution of the Integrated Multihousehold CGE Model 182 6.2 Comparative Growth Incidence Curves for Total Population: IMH_FX_w1 versus IMH_FL_w 202 6.3 Comparative Growth Incidence Curves for Total Population: IMH_FX_w1 versus MSS_FX_w1 205 7.1 Fully Integrated Macro-Micro Model Structure 223 7.2 Benefit Incidence of an Agricultural Subsidy under Various Specifications 233 7.3 Benefit Incidence of a Total Factor Productivity Shock in the Agricultural Sector 234 8.1 Occupational Choice Map 255 8.2 Occupational Choice Maps, Socio-Economic Survey Data and Townsend-Thai Data 258 8.3 Foreign Capital Inflows and Financial Liberalization 262 8.4 Intermediated Model, Townsend-Thai Data, 1976­96 264 8.5 Intermediated Model, Socio-Economic Survey Data, 1976­96 265 8.6 Welfare Comparison, Townsend-Thai Data, 1979 269 viii CONTENTS 8.7 Welfare Comparison, Socio-Economic Survey Data, 1996 272 8.8 Access to Capital and Foreign Capital Inflows, Socio-Economic Survey Data and Townsend-Thai Data 274 9.1 Foreign Aid per Capita 303 9.2 Real Wages of Labor with Secondary Education 305 9.3 Present Value of Foreign Aid 306 9.4 Trade-Offs between Human Development and Poverty Reduction 308 Tables 2.1 Predicted Price Changes Due to Agricultural Trade Reform in Morocco 34 2.2 Consumption Shares and Welfare Impacts through Consumption 36 2.3 Percentage Gains from Each Policy: Production Component 37 2.4 Household Impacts of Four Trade Reforms 38 2.5 Mean Gains from Policy 4, by Region 41 2.6 Decomposition of the Impact on Inequality 46 2.7 Summary Statistics on Explanatory Variables in the Regression Analysis 48 2.8 Regression of per Capita Gain/Loss on Selected Household Characteristics 49 2.9 Urban-Rural Split of Regressions for per Capita Gains 51 3.1 LINKAGE Model: Regional and Sectoral Groups 65 3.2 Trade Protection by Origin, Destination, and Sector 70 3.3 Economic Structure for Brazil, Chile, Colombia, and Mexico 74 3.4 Household Incomes by Source, Segment, and Poverty Status 76 3.5 Sectoral Adjustments 78 3.6 Price (Factors, Consumption Aggregates) and Real Income Changes 80 3.7 Initial Poverty Levels and Percentage Changes Resulting from Trade Reforms 82 3.8 Income Elasticity of Poverty Headcount 84 4.1 Evolution of Poverty in Indonesia, 1996­99 107 CONTENTS ix 4.2 Evolution of Occupational Choices and Wages by Segment, 1997­98 108 4.3 Historical Simulation Results 109 4.4 Simulation Results: Macro Aggregates 111 4.5 Simulation Results: Per Capita Income, Inequality, and Poverty Indicators 112 5.1 An Overview of the Three Experiments Conducted 122 5.2 Standard Multipliers of the Macro Model Compared with Other Macro Models 132 5.3 Some Results of the Macro Model, Historical Simulation for 1999 133 5.4 Major Results of the Public Sector and Financial Sector Modules, Historical Simulation for 1999 134 5.5 Aggregate Results from the Macro Model, Occupations 137 5.6 Aggregate Results from the Macro Model, Earnings 140 5.7 Detailed Results from the Top-Down Macro-Micro Models, Occupations by Skill and Sector 148 5.8 Aggregate Results from the Top-Down Macro-Micro Models, Earnings 151 5.9 Paired t Test 158 5A.1 Log Earnings Regression 162 5A.2 Occupational Structure Multinomial Logit Model: Marginal Effects, Rural 164 5A.3 Occupational Structure Multinomial Logit Model: Marginal Effects, Urban 167 6.1 Definition of Model Specification Used 194 6.2 Macro Results 195 6.3 Structural Effects of the Trade Reform, Output (Value Added) Change by Sector 197 6.4 Effects on Poverty (FGT Poverty Indexes) for the Whole Population and by Education Groups 199 6A.1 Labor Supply Model Estimation Results 206 6B.1 Education Code Definition 207 6C.1 Notations 207 7.1 Minimum Yearly Wages, 1990­96 226 7.2 Distribution of Beneficiary Households across Quintiles 227 7.3 Macroeconomic Impact of Alternative Policies 229 7.4 Employment Impact of Alternative Policies 230 x CONTENTS 7.5 Social Impact of Alternative Policies, General Equilibrium Results 231 7B.1 Results from Estimation and Micro Calibration 238 8.1 Maximum Likelihood Estimation Results 259 8.2 Welfare Gains and Losses, Intermediated and Nonintermediated Economic Wealth Distribution 271 9.1 Determinants of MDG Achievements 291 9.2 Impacts on MDG Indicators 298 9.3 Impacts on Macroeconomic Indicators 299 9.4 Impacts on Government Current Expenditures 300 9.5 Impacts on Government Investment Expenditures 301 9.6 Impacts on Government Revenues 302 9.7 Impacts on Labor and Capital 304 Preface This book assembles methodologies and techniques to evaluate the poverty impact of macroeconomic policies. It takes as a departure point a companion volume, The Impact of Economic Policies on Poverty and Income Distribution: Evaluation Techniques and Tools edited by François Bourguignon and Luiz A. Pereira da Silva, pub- lished in 2003. That volume was primarily a review of microeco- nomic techniques aimed at assessing policies that are directly concerned with the welfare of poor households or individuals-- such as changing the level of cash transfers to the poorest house- holds, increasing price subsidies for basic consumer goods, and the like. In addition, the second part of that earlier publication intro- duced basic techniques to deal with the poverty impact of macro- economic policies that by definition are not targeted and affect the whole population. However, as Nicholas Stern stated in the fore- word to the 2003 volume: "[M]ore research is needed to improve the integration of macroeconomic models and the models of house- hold behavior as captured in household surveys. Such an integration is obviously crucial when the distributional incidence and macro- economic effects of key policies are being studied." Five years later, this book presents the research to which he alluded. It deals with evaluating the impact of macroeconomic policies on poverty and income distribution using cutting-edge approaches. Policy makers are increasingly becoming aware that despite a positive effect on the average income of their citizens, many macro policies can sometimes produce such a deterioration in the welfare of specific groups that the policies can become socially undesirable and politically unsustainable in terms of the long-run growth objec- tives for a given economy and society. Similarly, poverty reduction policies designed to target specific individuals and/or households may end up producing macroeconomic (mostly fiscal) consequences. Thus, the selection and implementation of economic policies require a careful assessment of their effects both on aggregate economywide variables--such as employment, inflation, or real GDP growth-- and on income distribution and poverty. Modern micro simulation techniques, which use microeconomic data sets to simulate the xi xii PREFACE policy impact on all individuals in a sample that is statistically rep- resentative of an entire population, are the most promising tool for providing that careful assessment. This volume presents a comprehensive array of macro-micro modeling frameworks. It begins by highlighting the limitation of macroeconomic models that use representative household groups to link macroeconomic policies and microeconomic data. It then moves to more complex, top-down modeling frameworks, which combine (top) macro models and (down) micro simulation models that, in turn, can be simple micro accounting models or behavioral micro models. The book also explores integrated models, in which the macro and micro parts are either linked by iterative feedback loops or solved simultaneously as a single model. By providing clear access to these techniques, by documenting their analytical underpinnings, their data requirements, and their range of applicability, and even by highlighting some of their limitations, this book provides a unique compendium for practitioners, policy makers, and anyone interested in economic development. Acknowledgments This book continued down the path opened by the 2003 companion volume, The Impact of Economic Policies on Poverty and Income Distribution: Evaluation Techniques and Tools. Thus, we owe a debt of gratitude to Nicholas Stern and Stanley Fischer, who at the time of its publication, were, respectively, the chief economist of the World Bank and first deputy managing director of the International Monetary Fund. Their initiative inspired the process of reviewing techniques for evaluating the poverty and distributional impact of various policies available for development. During its four-year lifespan, this project has benefited from the support and advice of many people, including those who contributed comments during the various seminars and conferences at which the authors presented their papers. For their remarks, suggestions, and peer review, we thank in particular Flavio Cunha, Shantayanan Devarajan, Alan Gelb, Coralie Gevers, James Heckman, Thomas Hertel, Jeffrey Lewis, Catherine Pattillo, Guido Porto, and Hans Timmer. Our greatest debt, obviously, is to the authors of the eight con- tributed chapters. This book is really the outcome of their original work. Their names and current affiliations are listed in the contrib- utors section, and we truly thank them for their own relevant pieces, for the comments they provided on their colleagues' work, and for their endurance during the long process of producing this volume. Special thanks go to Nadia Fernanda Piffaretti, Jean Gray Ponchamni, Aban Daruwala, and Roula I. Yazigi, who provided superb administrative support and assistance during various critical phases of this project. We are very grateful to Kim Kelley for her excellent editorial assistance; to Janet Sasser for managing the editing, typesetting, proofreading, cover design, and indexing of the project and for her dedication and professionalism in assisting the editors at the crucial final stages of the production of this book; and to Santiago Pombo- Bejarano for his enduring support during the long life of this project and for his enthusiastic commitment to converting the manuscript into a finished volume. xiii Contributors Editors François Bourguignon Professor of economics and director of the Paris School of Economics; former chief economist and senior vice president at the World Bank, Washington, DC Maurizio Bussolo Senior economist in the Development Prospects Group at the World Bank, Washington, DC Luiz A. Pereira da Silva Chief economist for Brazil's Ministry of Planning and Budget; Deputy Finance Minister for International Affairs, Brazil Other Contributing Authors Denis Cogneau Senior research fellow at the Institut de Recherche pour le Développement (IRD) and the Développement, Institutions et Analyses de Long terme (DIAL), Paris Carolina Diaz-Bonilla Economist in the Latin America and Caribbean Region Poverty Sector at the World Bank, Washington, DC Francisco H. G. Ferreira Lead economist in the Development Research Group at the World Bank, Washington, DC Xavier Giné Economist in the Development Research Group at the World Bank, Washington, DC Jann Lay Senior economist and head of the Poverty Reduction, Equity, and Development Research Area at the Kiel Institute for the World Economy, Germany xv xvi CONTRIBUTORS Phillippe G. Leite Consultant, Development Economics World Development Report team at the World Bank, Washington, DC Hans Lofgren Senior economist in the Development Prospects Group at the World Bank, Washington, DC Michael Lokshin Senior economist in the Development Research Group at the World Bank, Washington, DC Denis Medvedev Economist in the Development Prospects Group at the World Bank, Washington, DC Paulo Picchetti Associate professor in the Department of Economics at the Fundação Getulio Vargas, São Paolo, Brazil Martin Ravallion Director of the Development Research Group at the World Bank, Washington, DC Anne-Sophie Robilliard Research fellow at the IRD and DIAL, Paris Sherman Robinson Professor of economics, University of Sussex, Brighton, UK Luc Savard Associate professor in the Department of Economics at the Université de Sherbrooke, Quebec, Canada Robert M. Townsend Charles E. Merriam Distinguished Service Professor in the Department of Economics at the University of Chicago Dominique van der Lead economist in the Development Mensbrugghe Prospects Group at the World Bank, Washington, DC Abbreviations BoP balance of payments BU bottom-up (part) CES constant elasticity of substitution CET constant elasticity of transformation CGE computable general equilibrium CPI consumer price index DIAL Développement, Institutions et Analyses de Long terme ERR exchange rate regime FIES Family Income and Expenditure Survey FP fixed-point (algorithm) FTAA Free Trade Area of the Americas FULLIB full trade liberalization GDP gross domestic product GNP gross national product GTAP General Trade Analysis Project HD human development IBGE Instituto Brasileiro de Geografia e Estatística IFLS Indonesian Family Life Survey IFPRI International Food Policy Research Institute ILO International Labour Organization IMF International Monetary Fund IMH integrated multihousehold (approach, model) IS-LM investment savings and liquidity preferences LAC Latin America and the Caribbean Region LAV linking aggregated variable LEB Lloyd-Ellis and Bernhard (2000) model LES linear expenditure system MAMS Maquette for MDG Simulations MDG Millennium Development Goal xvii xviii ABBREVIATIONS MLD mean log deviation MSS micro simulation sequential (approach, method) NAFTA North American Free Trade Agreement PNAD Pesquisa Nacional por Amostra de Domicílios [National Household Survey] PV present value RH representative household RHG representative household group RNF rural nontradable formal RNI rural nontradable informal RTF rural tradable formal SAM social accounting matrix SES Socio-Economic Survey SUSENAS Survei Sosial Ekonomi Nasional [National Socio-Economic Household Survey] TD top-down (part) UNDP United Nations Development Programme UNF urban nontradable formal UNI urban nontradable informal UTF urban tradable formal WPI wholesale price index 1 Introduction: Evaluating the Impact of Macroeconomic Policies on Poverty and Income Distribution François Bourguignon, Maurizio Bussolo, and Luiz A. Pereira da Silva Economists have long been interested in measuring the effects of economic policies on poverty and on the distribution of welfare among individuals and households. Devising satisfactory methods for accurate evaluations has proven to be a difficult task. Progress in economic analysis and the growing availability of microeconomic household data have improved the situation. At the same time, how- ever, calls for rigorous assessment have intensified. Partly because of the fierce debate on the social effects of globalization, economic policy objectives and social demands increasingly have focused on poverty reduction and distribution outcomes. In this context, development strategies as well as recurrent eco- nomic policy choices are being scrutinized ex ante--that is, before they are actually implemented--or assessed ex post--that is, after their execution. The range of policy issues subject to these evalua- tions is broad and includes the following: · Public expenditures. What is the poverty impact of specific shifts in public spending? How are the poor affected by changes in 1 2 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA the delivery of public services, especially in the cases of health and education services? · Tax policy. Do poor people bear a disproportionate burden of taxation or do they really benefit from subsidies designed to assist them? · Structural reforms. How can trade liberalization, domestic markets liberalization, privatization, labor market reforms, and decentralization, among other reforms, help the poor? · Macroeconomic policies. More specifically, what is the poverty impact of changes in the fiscal stance, or in monetary and exchange rate policies? What is the most effective macroeconomic policy set- ting to foster investment and productivity and to achieve long-term growth that is beneficial to all? To answer these questions, different methodologies have been devised that can be roughly classified in two groups: (1) microeco- nomic techniques, based mostly on incidence analyses and econo- metric evaluation approaches in partial equilibrium settings; and (2) macro-micro techniques, which, with different degrees of integra- tion, combine macro and micro modeling frameworks, usually in a general equilibrium context. The first set of techniques has been extensively reviewed in the companion volume to this book, The Impact of Economic Policies on Poverty and Income Distribution: Evaluation Techniques and Tools (Bourguignon and Pereira da Silva 2003). This group of tech- niques, well rooted in the public finance literature, has been applied primarily to analyze issues of incidence of tax and public spending (see the questions above related to public expenditures and tax pol- icy). Most of that earlier volume was devoted to case studies that illustrated microeconomic incidence methodologies. Various chap- ters showed numerous policy applications­­changes of indirect taxes, health and education public services, redistributive commu- nity programs­­and exemplified different methodological perspec- tives, namely, simpler accounting incidence analyses were juxta- posed against more complex behavioral approaches. Accounting approaches compute only first-round effects and disregard second- round effects attributable to behavioral reactions. Behavioral inci- dence analyses explicitly include those reactions. For example, An individual may decide to work less than otherwise to avoid losing her eligibility for a means-tested transfer, parents may decide to send their children to school to take advantage of free school lunches, or they may pay more attention to their children's health if a public dispensary is built in the neighbor- hood (Bourguignon and Pereira da Silva 2003: 9). INTRODUCTION 3 Other methodological challenges covered by case studies in the first volume included the following: comparing ex ante and ex post approaches; assessing the average versus marginal effects of a pol- icy; combining quantitative with qualitative approaches; and evalu- ating policies with some important geographic dimension (location of infrastructure projects such as roads, irrigation projects, etc.) using poverty maps. The second set of techniques--the integrated macro-micro techniques--was also introduced in the companion volume (Bour- guignon and Pereira da Silva 2003). The methodologies and case studies included in the second half of that volume were rather simple and did not include the cutting-edge approaches developed in the lit- erature. This more recent and more sophisticated group of techniques is the focus of the present volume. In fact, the application of the vari- ants of a single modeling framework--a macro model linked with a household-level micro model--is the unifying methodological theme of this volume. It is important to emphasize that a macro-micro approach enables different questions to be asked about the poverty and distribution consequences related to policy changes, and answer- ing these questions is a main motivation for this book. First, a macro- micro approach allows assessing the micro effects of macroeconomic policy changes and investigating the second round effects of policy changes. The pure microeconomic techniques described above can- not consider the poverty impacts of choosing, implementing, or alter- ing macroeconomic policies such as the trade regime, tariffs and non- tariff barriers (NTBs), the exchange rate, interest rates, the policy mix of fiscal and monetary policies, the composition of public spend- ing, and the labor market regulation, among other policies (see the questions on page 2 about structural reforms and macroeconomic policy). Second, even micro policies--that is, policies targeted to spe- cific population groups--when scaled up are likely to have macro consequences. Micro techniques of the type described above may measure the overall financial cost of a specific intervention, such as increasing education coverage through a conditional cash transfer program; however, they stop short of "feeding" this cost to a macro model and thus they cannot gauge what kind of macro (fiscal or growth, for instance) repercussions such a program may have. A final common thread connects and motivates the various contri- butions of this volume. This thread is represented by an attempt to measure the complete set of micro, macro, first and second round effects of economic policies by using more than one data set. The stud- ies in this volume show that great gains can be made by using many data sets. Considering standard macro data sets, such as those from a central bank or the national income accounts, together with micro 4 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA data sets, such as those from household surveys, labor force surveys, population censuses, and community-level surveys, provides analysts better opportunities to "look beyond the averages" in the analysis of the growth-inequality-poverty nexus (Ravallion 2001, 2006). In pol- icy-relevant terms, this basically means a better chance to identify specific interventions that can complement growth-oriented develop- ment policies. As shown by the contributions in this volume, looking simultaneously at macro and micro data offers advantages but also presents great challenges. It is well known, for example, that mean consumption from the national accounts and mean consumption from the survey data are different in levels and tend to diverge in growth rates. More specifically, the debate over India's fast growth rate (mea- sured from the national accounts) and slow poverty reduction (measured from the surveys) is not just an academic debate but also a policy debate: sorting it out may have important repercussions on economic policy decisions. In Deaton's words: [T]he reformers argue that the survey data are wrong, and the anti-reformers argue that the national accounts data are wrong. [. . . whereas] both of them are in bad shape [and] an enormous amount of work needs to be done on reconciling national accounts and on reconciling them with survey data (2001).1 The current volume, in the same way as the 2003 Bourguignon and Pereira da Silva edited volume, is organized along methodological lines. A common macro-micro modeling framework is used across all chapters, but its variants highlight methodological choices dictated by the specific question analyzed and data quality and availability. These choices include the following: (1) the types of macro and micro mod- els; (2) the extent of integration between the macro and micro models; (3) the degree of behavioral response, especially at the micro (house- hold) level; and (4) the time frame of the analysis (ex ante or ex post). Before presenting a brief survey of these methodological choices and to place the subsequent chapters in a common broad perspec- tive, the next section of this chapter summarizes the role of poverty and distribution in the literature on development. This chapter con- cludes with a brief summary of lessons learned. The Relationship between Macroeconomic Policy and Distribution in the Development Literature The central theme of this volume--the impact of macroeconomic policies on household welfare--has now been generally recognized as a key development question and has received extensive attention INTRODUCTION 5 in the theoretical and empirical literature. However, this focus on the distributional consequences of macro policies is very much a recent phenomenon, having entered the development literature from fiscal policy incidence analysis in high-income countries. Starting in the 1940s, four distinct phases emerged in the evolution of eco- nomic thinking about the importance of distributional outcomes, with only the last phase devoting significant attention to impact analysis at the household level. During the dawn of development theory and practice, growth and industrialization were the main objectives. Achieving these goals--largely through a mechanical, trickle-down effect--would then bring about development and poverty eradication (Rosenstein- Rodan 1943). This literature did not ignore the distributional con- sequences of growth: the well-known inverted U curve (Kuznets 1955) and surplus labor model (Lewis 1954) both acknowledged that inequality may initially increase as per capita income rises and labor migrates into a modern, industrial, high-income sector. These negative distributional consequences were considered transitory and were an intrinsic part of the growth process, however, and thus not to be actively managed by policy interventions. Instead, because higher growth would eventually result in less poverty and inequal- ity, the policy advice of this strand of literature was to focus squarely on growth. In the second phase of the evolution of the development paradigm--during the 1960s and 1970s--concerns over income dis- tribution and poverty intensified. In 1968, World Bank President Robert McNamara announced poverty reduction as an explicit insti- tutional goal during his first Bank annual meeting speech. This goal marked an important shift in development thinking: growth alone, without improvements in the lives of millions of poor people in the developing world, was no longer sufficient (McNamara, as cited in Birsdall and Londono 1997). On the research front, economists showed that a nation's welfare depends on both the size of national income and its distribution (Sen 1973). An influential 1974 World Bank report, Redistribution with Growth, concluded that growth and distributional goals "cannot be viewed independently . . . [but] should be expressed dynamically in terms of desired rates of growth of income of different groups" (Chenery and others 1974). In par- ticular, the report sought to explicitly incorporate distributional outcomes into measures of social welfare by suggesting the use of the weighted sum of income growth of population subgroups (for example, income deciles) instead of aggregate GNP or GNP per capita metrics. During the third phase--lasting from the mid-1970s to the early 1990s--the development literature reached a broad consensus that 6 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA with adequate policies (not necessarily of the redistributive kind) there should be no conflict between accelerated growth and equitable distribution. During this time, models of incidence analy- sis of public expenditure began to enter the development literature (Meerman 1979; Selowsky 1979; Ahmad and Stern 1991). The scope of policy advice, however, was largely determined by the ele- ments of the Washington Consensus, which did not incorporate explicit distributional objectives. According to Kanbur (2000), this outcome was prompted by several distinct developments. First, a large body of empirical work failed to confirm the U-shaped rela- tionship between levels of per capita income and inequality pro- posed by Kuznets (1955). The lack of data prevents testing the hypothesis as a time-series phenomenon, while a cross-sectional relationship is difficult to identify once controls for countries with high historical inequality (for example, countries in Latin America) are added. Second, many studies documented the antigrowth and anti- distributional consequences of the strongly distortionary policies-- including exchange rate overvaluation, high trade barriers, and large state-owned enterprises--adopted by many African and Latin American countries. Most of these studies showed that these dis- tortions were both inefficient and inequitable and argued that pol- icy reform in these areas would have resulted not only in more growth but also in less inequality and less poverty. The third devel- opment is represented by the observations of the East Asian "miracle" of strong growth with equity, which showed that growth could benefit everybody equally, although equitable initial distribu- tion of assets was acknowledged as a key determinant of favorable outcomes. In many ways, the final phase represents a return to the themes of the second phase, albeit with much more sophisticated analytical tools. The recognition of a multifaceted relationship between growth and distribution came about because of the skepticism toward the results of the Washington Consensus policies, and from the discovery that inequality was trending upward in East Asia. A number of coun- tries in Latin America and Sub-Saharan Africa discovered that their performance in terms of growth and income distribution remained disappointing, even after implementing most of the market-friendly reforms. At the same time, living standards of people at the bottom of the distribution became a central issue in the international policy arena, particularly with the adoption of the 2000 United Nations Millennium Declaration. These developments, combined with a real- ization that little was known about the distributional dynamics of growth, served to draw distributional concerns "back from the cold" (Atkinson 1997). INTRODUCTION 7 The renewed focus on the relationship between macro (growth) and micro (distribution) issues has led economists to realize that "growth is quite a blunt instrument against poverty unless that growth comes with falling inequality" (Ravallion 2004). The avail- ability of detailed household surveys and new analytical tools (many of which are described in this volume) has enabled researchers to move beyond concepts of aggregate inequality and focus on the effects of macro policies on specific household groups. Insights gen- erated by this analysis can help in understanding the outcomes of past reforms, design of compensatory mechanisms, and anticipation of challenges inherent in future policy decisions. Macro-Micro Modeling as a Tool for Poverty and Inequality Analysis To highlight the differences and the specificity of the methods aimed at assessing the poverty and income distribution effects of macro- economic policies, it is useful to first consider the parallel literature on the evaluation of microeconomic policies. The choice of the evaluation technique in microeconomic policy analyses depends, to use Blundell and Costa Dias' (2007: 1) word- ing, on three broad concerns: (1) the nature of the question to be answered; (2) the type and quality of data available; and (3) the mechanism by which individuals are allocated to the program or receive the policy (this mechanism is usually called the "assignment rule"). In an ex post evaluation, the policy action has occurred, and researchers and policy makers want to identify whether the imple- mented policy is working the way they thought it should be, and the nature of the question being asked is thus clearly identified. The evaluation procedure has to properly work out a comparison that clearly separates individuals and households that have been subject to the policy (or "treated") and those that have not but are other- wise similar to the treated. The most convincing method of evalua- tion is the social experiment method, because it builds directly the comparison (or control) group by randomly assigning eligible people to receive (or not receive) the treatment. A series of other methods-- for example, natural experiments methods, discontinuity design methods, matching methods, instrumental variables methods, con- trol function methods, and structural econometric methods (for a survey of these methods, see chapter 5 in Bourguignon and Pereira da Silva 2003 or Blundell and Costa Dias 2007)--attempt to mimic the random assignment of the social experiment or use economic 8 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA theory to model the assignment rule (and thus control for the selec- tion bias). The quality of the data (for example, having long panel data sets) and the complexity of the assignment rule normally dic- tate the choice of the evaluation method. In a typical ex ante situation, building the counterfactual is much trickier because the comparison between the situation "before" and "after" the treatment is purely virtual or notional (given that the policy has not been implemented yet). The "theory-free" social experiment method and its close substitutes are not applicable. The dominant approach in ex ante policy evaluations is represented by structural econometric models or, when estimating fully specified structural models is not feasible, simpler reduced-form or nonbe- havioral models are used. In this latter case, the main issue is to gen- erate counterfactuals by simulating hypothetical situations with the implemented policy and without the policy. The simulation is done using information at the individual or household level, and it is thus labeled micro simulation. Contrasted with the evaluation of microeconomic policy, the assessment of the poverty and income distribution effects of macro- economic policies (ex ante or ex post) presents two distinct prob- lems that require devising novel methods. First, the comparison between groups of individuals and households that are treated (or subject to the experiment) and a control group is much harder. In fact, it is almost impossible to isolate a control group for a macro- economic policy because, by definition, all individuals and house- holds are affected by the same policy. For example, a devaluation of the real exchange rate affects the whole economy with multiple consequences for households, firms, banks, government, and so on. Therefore--and this is the second problem--one has to figure out not only a micro but also a macro counterfactual, and the latter usually has to be done in a general equilibrium setting. These two distinct difficulties justify the use of a macro-micro modeling framework--one that takes into account the macro nature of the policy (or the macro consequences of scaled-up micro interventions) and integrates a microeconomic (that is, individual and household) dimension. Macro Models with Representative Household Groups: A First Step in Assessing Macroeconomic Policy Effects on Poverty and Distribution Traditional aggregate macroeconomic models use the simplest and most restrictive assumption that economic policies do not affect the distribution of welfare. This is tantamount to assuming that the INTRODUCTION 9 economy is composed of only one economic agent or that all indi- viduals in a society are identical. This assumption can be relaxed by using the less-stringent hypothesis that economic policies do not affect the distribution of welfare within groups of homogeneous households. This is the idea behind the construction of macroeco- nomic models in which the single consumer or household is disag- gregated into groups of households that share some common char- acteristics, usually in terms of the structure of their sources of income and their consumption preferences. Identifying--for a given economy--a comprehensive set for such groups would result in building a macro model with a set of representative household groups (RHGs). The ending point of the 2003 Bourguignon and Pereira da Silva volume was precisely the description of this type of technique. In particular, Lofgren, Robinson, and El-Said (2003, chapter 15) showed how to construct macroeconomic models with RHGs to carry out poverty and inequality analyses in a general equilibrium framework. This class of models has been labeled in the modeling literature as the disaggregated SAM-CGE/RHG model (social accounting matrix [SAM]­computable general equilibrium [CGE]). The best examples start with Adelman and Robinson (1978) and Dervis, de Melo, and Robinson (1982), and continue with Bourguignon, de Melo, and Morrisson (1991) and Agenor, Izquierdo, and Jensen (2006), among others. In this tradition, the typical CGE is a macroeconomic model that separates the house- hold population of an economy into RHGs. Disaggregation aims to capture the various ways through which economic policies would affect the factor allocation and remuneration across RHGs. The SAM-CGE/RHG approach models the functioning of factor mar- kets at the level of aggregation that is compatible with the factor remuneration for each RHG. Under this approach, only the aggre- gate behavior of these groups--in terms of supply of labor and consumption demand--matters for the general equilibrium of the economy. Strong assumptions must be made: the distribution of relative income within each RHG is policy-neutral, that is, it is not affected by any change in macroeconomic policy; and the demographic weight of households in each RHG is constant. Hence, this approach essentially focuses on changes in the distribution between RHGs. Empirically, however, analyses of micro data show that changes of within-RHG inequality can be as important as changes of between-RHG inequality in explaining the evolution of overall inequality. Think, for example, of a case in which the RHGs are formed according to the employment status of the head of household (for example, "small farmers," 10 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA "unskilled urban workers in the formal sector," and so on) and in which the shock to be simulated results in a strong reduction of unskilled urban workers employed in the formal sector. After the shock, the RHG labeled "unskilled urban workers in the formal sector" is much smaller, but the standard CGE/RHG does not include any mechanisms of adjustment: that is, it does not model which par- ticular household should leave its original group (should it be a poor or a rich household?) nor in which new group it should go. In addition to the criteria used to form RHGs, household-level criteria normally are appropriate only for the head of household, but other members may belong or move to other groups and this cannot be accounted for. These two phenomena are likely and can strongly affect within- RHG inequality, but they are completely ignored by the standard approach. A direct way to deal with these issues is to introduce as many rep- resentative households into the initial CGE/RHG framework as there are in standard household surveys. Indeed, such models do already exist. See, for instance, Cockburn (2006) for an application to a case study of Nepal and see Rutherford, Tarr, and Shepotylo (2007) for a study of Russia. The household sector in these models includes (a few thousand) heterogeneous individual households reflecting those observed in available household surveys; however, as explained below, a restrictive set of conditions need to be assumed to model the behavior of these households. Is this extension of RHGs to "real" households the way forward? Does it mean that the full integration of household surveys with macro modeling is practically achieved and increasingly will be used as computers and model-solving software become more powerful? The answer to the first question is most certainly yes. Integrating household surveys into macro modeling, whether relying on CGE or other types of models, is undoubtedly the direction to follow to assess the impact of macro policies at the micro level. Most of the chapters in the present volume are about this integration. The answer to the second question must be formed more cautiously. The mere exten- sion of the CGE/RHG approach to individual households taken from household surveys raises methodological issues that cannot be ignored. The most important difficulty in answering this question relates to estimating heterogeneous economic behavior at the household level. Consider, for instance, the issue of modeling the consumption behavior of households. Within the standard CGE/RHG frame- work, it is necessary to specify the way each household in the model uses its income, or how the budget coefficients of a particular household depend on the income of that household and on the vec- tor of prices. Starting from the data of a household survey, the INTRODUCTION 11 behavior of a specific RHG is generally estimated on the basis of observed differences among "real" households belonging to that same RHG. In other words, variability across household incomes and budget shares is used to estimate the way in which the budget share of the representative household of a group changes with income. This variability requires postulating some functional form, which in turn permits inferring price elasticities derived from income effect. Is such an approach possible with "individual households," namely, in the case when the number of RHGs in the CGE is the same as that for the households in a survey? Yes, but with tighter assumptions. Of course, there is no way to "estimate" a consump- tion model for a household on a single observation except under two alternative stringent assumptions: (1) observed household budget shares are assumed to be constant--implying a "linear" model with unit income and own-price elasticities of consumption; and (2) observed household budget shares correspond to a com- mon behavioral model where shares depend on income (and prices) with fixed differences between individual shares and what the model implies. As with the RHG approach, the common consumption behavior can be estimated through standard econometric techniques applied to the whole sample of household variations after some assumption has been made on the functional form to use. Residuals in that procedure stand from some unexplained divergence of indi- vidual households from the common model. It turns out that these assumptions are far from being innocuous and may influence both the macroeconomic properties of the whole framework and the sim- ulated microeconomic consequences of a macro policy or shock. Assumption 1 would imply that a proportional increase of income of the whole population of households does not modify the aggre- gate propensity to consume, not a very satisfactory property par- ticularly when aggregated at the macro level. Assumption 2 would imply that the simplicity and homogeneity of such a "common behavioral model" for individual behavior regarding important characteristics of individual decisions would yield very limited responses to policy changes at the macro level. We would be back to the aggregate properties of the macro model itself losing the effort of disaggregation that meant to model hetereogeneity at the start. A Top-Down Modeling Approach for Integrating Household-Level Data into Macro Models The CGE/RHG approach is an important first step, but it has inher- ent limits in terms of modeling the heterogeneity of individuals and 12 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA households. So how is the impact of macroeconomic policies on het- erogeneous individuals measured? A solution to avoid the problems mentioned in the previous section is to separate the macro and micro parts of the modeling framework. The degree of separation--and the potential alterna- tive ways of linking or integrating (in a different way from the RHG approach)--of these two parts, which are briefly described in the remainder of this introductory chapter, is the distinguishing feature of the various contributions to this volume. The top-down modeling approach works in a sequential two- step fashion: (1) a macro (top) model is solved and (2) its solution in terms of a vector of aggregate prices, wages, and employment variables--the linking aggregate variables (LAVs)--is used to (a) shock a micro-household-level data set or (b) target the aggregate solution values of a micro (bottom) model (see figure 1.1). In case a, the micro simulation is quite simple and broadly corresponds to the micro accounting incidence analysis mentioned above: households (and individuals) do not respond to the price shocks (coming from the top model) by changing the quantities of factor services they supply or the quantities of goods they demand or sell. In case b, the microeconometric model includes behavioral responses used to simulate changes in individual wages, self-employment incomes, employment status, and so on. These individual changes are simu- lated in a way that is consistent with the aggregation of the set of LAVs generated by the macro model. When the micro accounting simulation or the micro behavioral simulation is completed, a full Figure 1.1 Schematic Representation of the Top-Down Modeling Approach "Top" Level: Macro General equilibrium macroeconomic model (with sectoral disaggregation to model factor markets) Linking aggregate variables (LAVs) "Bottom" Level: Micro Individual/household microeconomic data set/ microeconometric model Source: Authors' depiction. INTRODUCTION 13 counterfactual distribution of household income is produced and the (macro) policy change can be evaluated. The details of these two top-down approaches are described in the next two subsections. TOP-DOWN MICRO ACCOUNTING MODELS This volume features two examples of top-down macro-micro accounting modeling: Ravallion and Lokshin (chapter 2); and Bussolo, Lay, Medvedev, and van der Mensbrugghe (chapter 3). In chapter 2, Ravallion and Lokshin use Morocco's national survey of living standards to measure the short-term welfare impacts of depro- tecting cereals (the country's main food staple). The authors find small impacts on mean consumption and inequality in the aggregate and contrary to past claims, find that the rural poor are worse off on average after deprotection. In chapter 3, Bussolo, Lay, Medvedev, and van der Mensbrugghe link a global CGE model with household surveys of Brazil, Chile, Colombia, and Mexico to estimate the first- round impacts on the poor of regional and multilateral trade liber- alization scenarios. Their results show that because of different initial positions in terms of economic structure, poverty levels, and trade protection, the poverty effects are quite dissimilar across the four countries studied. Furthermore, for the countries analyzed, the distributional effects of trade are more important than the growth effects (that is, the increase in average incomes following trade reform). Together, the two chapters illustrate the advantages of micro accounting techniques (among others, their ability to cap- ture the largest impacts of reform and their ease of implementation) and highlight an important limitation of this kind of analysis (that is, the fact that the results are likely to be valid only in the short and medium terms). In these micro accounting models, micro data sets are linked to disaggregated macro models by directly applying changes in prices and wages that result from the solution values of the macroeconomic model. For example, sector-specific vectors of macro simulated prices and wages are used to construct a coun- terfactual income for each individual or household, using simple multiplication or replacement techniques: the actual price and wage rate that explain the components of income for each individ- ual are replaced by the simulated values. The assumption of no behavior responses has been a major criti- cism of these micro accounting models, but under certain not overly restrictive conditions, it can be demonstrated that these models are fully consistent with microeconomic behavior. In fact, they esti- mate first-round effects, which are a good approximation of total welfare effects in situations in which the price (and wage) changes 14 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA are small and markets are competitive. In other words, behavior responses can be safely ignored when evaluating individual welfare change when the macro policy shock causes only marginal changes in the budget constraints faced by agents and when no agents are rationed or do not operate in a perfect market. Using the well- known utility theory of consumer behavior and relying on the enve- lope theorem (or Sheppard's lemma or Roy's theorem), a formal demonstration that micro accounting is consistent with behavior is provided in chapter 2. The main conclusion is that the change in the welfare income metric caused by a change in the price of a specific good is equal to the change in the cost of consuming that good because of the variation in its price (with a constant quantity con- sumed). This conclusion can be generalized to cases in which the agent produces and sells certain goods, including factor services. Apart from cases in which changes are not marginal and markets are not perfectly competitive and therefore behavioral responses can- not be ignored, an important drawback of this approach is given by the unidirectional link between the macro and micro parts of the mod- eling framework. This means that distributional changes at the micro level do not provide any feedback to the aggregate variables at the macro level: these are determined exclusively by the macro model. Additional examples of this approach are found in Chen and Ravallion (2004), who analyze the distributional effects of China's accession to the World Trade Organization; Friedman and Levinsohn (2002), who consider the impact of the Indonesian crisis on poverty; and other examples listed in the survey on poverty and trade by Hertel and Reimer (2005). TOP-DOWN MICRO SIMULATION MODELS The second way to conduct top-down macro-micro modeling is to link the macro model to a micro simulation module. This volume features two examples of this modeling framework: Robilliard, Bourguignon, and Robinson in chapter 4; and Ferreira, Leite, Pereira da Silva, and Picchetti in chapter 5. In chapter 4, Robilliard, Bourguignon, and Robinson link a CGE model with a micro simula- tion model to quantify the effects of the 1997 Indonesian financial crisis on poverty and inequality. Their framework allows for decom- position of the effects of the financial crisis, and their results show that while rural and urban incomes converged in the aftermath of the crisis, overall inequality increased because of the divergence of incomes within these sectors. Thus, the negative income effects of the crisis were augmented by a worsening in the distribution of income. In chapter 5, Ferreira, Leite, Pereira da Silva, and Picchetti use a top- down macro-micro model of the Brazilian economy to examine the INTRODUCTION 15 impacts of the 1998­99 currency crisis in Brazil on the occupational structure of the labor force and on the distribution of incomes. The authors test the ex ante predictive performance of the model by com- paring its simulated results using the 1998 household survey with the actual ex post household survey data observed in 1999. They find that the top-down macro-micro econometric model, while still inac- curate on many dimensions, can actually predict the broad pattern of the incidence of changes in household incomes reasonably well, and much better than the alternative approaches. This chapter thus offers some validation for this macro-micro approach. The key difference between the simpler accounting approach and micro simulation is that this approach can be used when the enve- lope theorem is not applicable--for example, when the policy simu- lated modifies the labor participation decision and/or when there are market imperfections such as rationing. In these circumstances, considering behavioral responses at the micro level becomes essen- tial. These responses are normally simulated by using a structural or reduced-form econometric model, which is initially estimated (or calibrated) from the cross-section data of the household survey. As mentioned, this type of micro model can handle market imperfec- tions. For example, imperfections can be introduced for the labor markets in the macro (CGE) model: wages can be assumed to be rigid in the formal sector in connection with some form of rationing, whereby individuals are not allowed free entry into the formal mar- ket segment. This rationing mechanism needs to be replicated or simulated at the micro level, and this can be done using the estimated micro model. Basically, this micro model identifies which individuals switch from jobs in the formal sector to self-employment or to inac- tivity, or vice versa. In other words, these macro-micro models recon- cile the disequilibrium captured by the macro model variables--where prices, wages, and employment will incorporate the effect of market imperfections--and the heterogeneity of individual behavior. Some individuals are more likely to be responsible for the changes observed at the macro sectoral level. Econometric models of occupational choice by household mem- bers allow this allocation to be performed, accounting for individual heterogeneity while using the relevant variables from the macroeco- nomic general equilibrium model to build the counterfactual distri- bution. These econometric models essentially consist of multilogit models of occupational choices that are conditional on individual and household characteristics. The micro simulation includes mod- ifying a subset of parameters of the multilogit model to generate aggregate levels of employment by occupational type, skill, gender, and so on, which exactly match the results coming from the CGE 16 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA macroeconomic model. Practically, the procedure uses the intercepts in multilogit models to match micro simulated employment and the results of the macro model. The intercept is adjusted for each indi- vidual in a given group so that the average of the group matches (actually, it converges through all the resulting changes in occupa- tional choices in the group) the average of the same group in the macroeconomic model. This is analogous with "grossing up" a small sample of individuals or households. The procedure remains, how- ever, top down in the sense that there is no feedback between the micro and the macro levels, that is, no explicit link or interaction exists between the micro level results and the actual prices in the macro model. For additional examples of how this technique is used, see chapters 4 and 5, as well as Bussolo, Lay, and van der Mensbrugghe (2006) and Hertel and Winters (2006). What does this procedure add to the understanding of the impact of macroeconomic policies? Taking into account individual hetero- geneity in modeling occupational choices certainly adds accuracy and nuance. The evaluation of the impact of economic policies shows some counterintuitive results when this procedure is used. This is shown when the counterfactual produced by the micro sim- ulation approach is compared with that of the CGE/RHG approach (as in chapter 4) and with the actual distribution (as in chapter 5). The counterfactual distributions obtained under the assumption that distribution of income within RHG (defined by the occupation of household head) is constant provide different results than the dis- tributions obtained with the top-down micro simulation framework shown earlier. The latter is closer to actual distributions and thus allows a better grasp of the impact of macroeconomic policies on specific groups and segments of the distribution: it appears that it constitutes a more accurate and better tool. In particular, the coun- terfactuals are radically different in an important dimension, namely, the percentiles of the distribution most affected by the experiment. This is a crucial piece of information to design well- targeted compensatory or supportive responses to a given macro- economic reform. Toward an Integrated Macro-Micro Model: Feedback Loops from Bottom to Top Under both the micro accounting and micro simulation top-down models, the results from the LAVs are "injected" into the micro data set that either takes them as givens or adjusts to them. After the changes are computed, however, the aggregated result of, say, the sum of consumption for all households in the micro data set can INTRODUCTION 17 be different from the result of the aggregate private consumption calculated by the macro model. In other words, there is no feed- back from the micro to the macro parts of the modeling framework. Bourguignon and Savard (chapter 6) address this issue by devising feedback loops between the two layers of the framework. In chap- ter 6, Bourguignon and Savard assess the distributional effects of trade reform in the Philippines, and their results illustrate the bias inherent to methods that ignore feedback effects from the micro to the macro and the assumptions that markets--particularly labor-- are fully competitive at the micro level. The main difference in their model is that the one-step sequential process from the top macro model to the bottom micro model is repeated iteratively and in a bidirectional way; that is, after the first shock, a subset of the LAVs is recalculated by aggregation from the micro data and transmitted to the macro model. This then adjusts again to these new values and once more transmits the new solution to the micro model. The process continues iteratively until conver- gence is reached. Rutherford, Tarr, and Shepotylo (2007) devised such an iterative algorithm in a CGE/RHG model with thousands of households (see above) and found that, in their case study for Russia and with perfectly competitive markets, most of the micro and macro impacts of an across-the-board trade liberalization were adequately accounted for by the first sequential step. In chapter 6, Bourguignon and Savard propose a different and simpler approach that is applica- ble to imperfectly competitive environments. This iterative approach has several advantages over a method that would solve simultaneously for all individual and aggregate equilibrium conditions, as in the CGE/RHG model category with fully disaggregated household groups. First, the macro and the micro parts of the model do not have to be fully consistent in terms of con- sumption or income aggregates. In many cases, the underestimated aggregate consumption from a household survey does not need to be adjusted to the national accounts generally used in CGE model- ing. As Bourguignon and Savard put it, "No correction is necessary for consistency with national account data if it is assumed that the proportion of underestimation is independent from the price of other goods and unit wages" (p. 185 of this volume). A second advantage is that no limit needs to be imposed on the level of disag- gregation in terms of production sectors and number of households to be included in the model. A third advantage is that, with respect to other approaches, fewer restrictions apply to the choice of functional forms for the consumption and labor supply behavior of households. In particular, there is no need to choose functional forms with good aggregation properties. 18 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA The Fully Integrated Micro-Based CGE Approach Quite naturally, one wonders whether or not the convergence process described above truly puts microeconomic consistency into the behavior of the macro aggregates. If the ultimate objective is to get a fully consistent macro-micro framework or, in other words, if the goal is to build the poverty impact of macro policies from the strongest basis of micro observations, then a fully integrated micro- based CGE should be the preferred method (Heckman, Lochner, and Taber 1998; Browning, Hansen, and Heckman 1999; Townsend 2002; Townsend and Ueda 2006). Why not aim to build macroeconomic behavior from all individ- uals and households in a sample survey? This route is taken in this volume by Cogneau and Robilliard (chapter 7) and Giné and Townsend (chapter 8). In chapter 7, Cogneau and Robilliard develop a macro-micro simulation framework to study the effects of targeted transfer schemes on income distribution and monetary poverty in Madagascar. Their results show that the general equilibrium effects of transfer mechanisms may change the distribution of the benefits between rural and urban households, an effect not accounted for by micro accounting or top-down modeling approaches. In chapter 8, Giné and Townsend apply a general equilibrium occupational choice model to two sectors in Thailand between 1976 and 1996. The authors show that without an expansion of the financial intermedia- tion sector, Thailand would have evolved differently, namely, with a much lower growth rate, a higher residual subsistence sector, and nonincreasing wages but lower inequality. The financial liberaliza- tion resulted in welfare gains and losses to different subsets of the population with a limited impact of foreign capital on growth or the distribution of observed income. Intuitively, it seems logical to circumvent the problems of stan- dard CGE/RHG models using a modeling strategy focusing on each household, but some problems remain. One is the difficulty of cali- brating structural behavioral models for individual households with the type of micro data set that is available. The rudimentary way through which some key structural behavior--such as consumption and investment--is modeled at the household level poses a problem for the properties of the overall model when the whole sample (thousands) of households is aggregated. Would the aggregate behavior of all individuals and households for private consumption or investment "react" to macroeconomic policies with the same "known macroeconomic textbook properties" as the observed aggregate variables in national accounts? For example, the macro- economic literature suggests that aggregate private consumption is INTRODUCTION 19 sensitive to income, inflation, and interest rates; however, if it is not possible­­because of the lack of data at the individual level--to esti- mate one or all of these elasticities, what would be the overall prop- erties of the macroeconomic model constructed with the aggregation of these insufficiently modeled household behaviors? In addition, econometric problems result from the difference between estimations done in cross-section with the last available household survey and those done with time-series for larger groups, or with panels. The other question­­when the modeling of key structural behavioral is limited­­is where does the heterogeneity come from? One interpre- tation is that heterogeneity can be a standard residual in the regres- sion equation across households, which is written to explain the behavior at hand, for example, private consumption. Another inter- pretation is to accept heterogeneity as a "heterogeneous behavioral coefficient" that can be added to the coefficients used to explain pri- vate consumption. But then the identification problem remains for this coefficient, because, as an example and thinking of the con- sumption function of a given household, the two interpretations of heterogeneity given above are observationally equivalent--up to heteroskedasticity. Yet they have different implications in terms of aggregate behavior. Macro-Micro Modeling: A Summary of Lessons Learned An alternative seems to be emerging from the overview of the mod- els and approaches included in this volume and briefly described above. On the one hand, the micro accounting approach uses too restrictive an assumption of constant distribution of income within RHGs, irrespective of the type of macroeconomic policy being examined and of the specific characteristics of the markets being affected. The top-down approach that models household behavior and uses LAVs improves the accuracy of the counterfactual but restricts the instruments of micro simulation to a limited number of LAVs. A complex CGE at the top can be designed, but the trans- mission of the heterogeneity will be limited to a few dimensions given by the LAVs (for example, occupational choices). On the other hand, the extension of the CGE/RHG approach to the thousands of households in a given survey seems promising as a way to bypass the problems listed above, but this approach has difficulties, too: only simple behavior can be properly modeled at the household level­­given available data sets­­and then properly "aggregated" at a macroeconomic level in the sense that results are consistent with macroeconomic theory. 20 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA The alternative can be summarized as follows. The evaluation of the poverty and distributional impact of macroeconomic policies theoretically needs to be conducted with a fully disaggregated gen- eral equilibrium model--including as RHGs a sufficiently large and representative sample of the population (possibly tens of thousands); a variety of goods and services produced and consumed in the econ- omy (also thousands); an adequate representation of equilibrium in all major markets; and in particular, consistent modeling of the consumption, investment, production, and savings decisions made by the tens of thousands of RHGs. Although it is theoretically pos- sible to solve such a large model, this volume points to the two major directions found so far to simplify this original­­and first best--approach. These two routes should work with a reduced num- ber of RHGs or should solve the full model by successive iterations, working with a top-bottom or a bottom-up approach. Even by solving the problems noted above, more complex issues remain. In particular, the models described above have nothing to say about the nonincome dimensions of poverty: health, education, and access to infrastructure, among others. Which is the proper macro-micro model that can assess public policies aimed at improv- ing these dimensions? This may seem to be a simple question, but so far, spending on education, health, or cash transfers to households has no direct productive effect in standard CGE or macroecono- metric modeling. These tools treat expenditures on physical or human capital identically, even if they have different long-term effects on an economy's growth potential. It is possible to analyze the distributional effect of these expenditures using a micro simula- tion framework if some behavior is introduced--for example, the demand for schooling or health services. But two difficulties arise: (1) the actual effects on distribution will appear only in the long run (when the young become adults and enter into productive activi- ties); and (2) while these education and health policies are supposed to generate future general equilibrium effects at the macro level (by changing the earnings structure and the growth rate of output), these effects also depend on how the demand side of the economy will evolve (for example, an "excessive" amount of spending on higher education might depress the returns to this kind of "invest- ment"). In this volume, Bourguignon, Diaz-Bonilla, and Lofgren (chapter 9) attempt to construct sectoral (Millennium Development Goal [MDG]-related) demand functions for these services working in a dynamic general equilibrium framework. Although micro data are not directly used to estimate these demand functions, this work could be done and could support the functional form chosen. The attempt provides two messages for policy makers: (1) that dynamic general equilibrium effects of social spending are critical to analyze INTRODUCTION 21 the allocation of resources to reach the MDGs and (2) that social expenditures are a composite good that produces cross-sectoral or cross-MDG externalities that need to be taken into account (that is, spending on basic infrastructure such as water and sanitation, edu- cation, and health in an appropriate proportion are important to achieve better overall development outcomes). Other methodological issues remain: (1) the importance of mod- eling heterogeneity of production and investment decisions by firms and (2) micro simulation techniques largely remain comparisons of two cross-sections of households and dynamic modeling and the proper treatment of growth is needed for a better understanding of the links between micro and macro phenomena. These and other issues for further research are briefly discussed in the concluding remarks of this volume. Note 1. Deaton, Angus (2001). Intervention in a panel discussion on the Inter- national Monetary Fund­sponsored conference on "Macroeconomic Poli- cies and Poverty Reduction," April 13, 2001, Washington, D.C.; transcript available at http://www.imf.org/external/np/tr/2001/tr010413.htm. References Adelman, Irma, and Sherman Robinson. 1978. Income Distribution Policy in Developing Countries: A Case Study of Korea. New York: Oxford University Press. Agenor, Pierre-Richard, Alejandro Izquierdo, and Henning Tarp Jensen, eds. 2006. Adjustment Policies, Poverty and Unemployment: The IMMPA Framework. Malden, MA: Blackwell Publishing. Ahmad, Ehtisham, and Nicholas Stern. 1991. The Theory and Practice of Tax Reform in Developing Countries. New York: Cambridge University Press. Atkinson, Alexander B. 1997. "On the Measurement of Poverty." Econo- metrica 55 (4): 749­64. Birsdall, Nancy, and Juan Luis Londono. 1997. "Asset Inequality Matters: An Assessment of the World Bank's Approach to Poverty Reduction." American Economic Review 87 (2): 32­37. Blundell, Richard, and Monica Costa Dias. 2007. "Alternative Approaches to Evaluation in Empirical Microeconomics." Institute for Fiscal Studies. Available at www.ifs.org.uk. An earlier version appeared in Portuguese Economic Journal 1 (2002): 91­115. 22 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA Bourguignon, François, Jaime de Melo, and Christian Morrisson. 1991. "Poverty and Income Distribution during Adjustment: Issues and Evidence from the OECD Project." World Development 19 (1): 1485­508. Bourguignon, François, and Luiz A. Pereira da Silva, eds. 2003. The Impact of Economic Policies on Poverty and Income Distribution: Evaluation Techniques and Tools. Washington, DC: World Bank; Oxford and New York: Oxford University Press. Browning, M., L. P. Hansen, and J. Heckman. 1999. "Micro Data and Gen- eral Equilibrium Models." In Handbook of Macroeconomics, eds. Taylor and Woodford, vol. 1. Amsterdam: North-Holland. Bussolo, Maurizio, Jann Lay, and Dominique van der Mensbrugghe. 2006. "Structural Change and Poverty Reduction in Brazil: The Impact of the Doha Round." Policy Research Working Paper No. 3833, World Bank, Washington, DC. Chen, S., and M. Ravallion. 2004. "Household Welfare Impacts of China's Accession to the WTO." In China and the World Economy: Policy and Poverty after China's Accession to the WTO, eds. D. Bhattasali, S. Li, and W. Martin. London and New York: Oxford University Press. Chenery, Hollis, M. Ahluwalia, C. Bell, J. Duloy, and R. Jolly. 1974. Redis- tribution with Growth: Policies to Improve Income Distribution in Developing Countries in the Context of Economic Growth. London: Oxford University Press. Cockburn, John. 2006. "Trade Liberalisation and Poverty in Nepal: Com- putable General Equilibrium Micro Simulation Analysis." In Globalisa- tion and Poverty: Channels and Policy Responses, eds. M. Bussolo and J. I. Round, 172­95. London and New York: Routledge/Warwick Studies in Globalisation, Routledge. Dervis, K., J. de Melo, and S. Robinson. 1982. General Equilibrium Models for Development Policy. New York: Cambridge University Press. Friedman, Jed, and James Levinsohn. 2002. "The Distributional Impacts of Indonesia's Financial Crisis on Household Welfare: A `Rapid Response' Methodology." World Bank Economic Review 16 (3): 397­423. Heckman, J., L. Lochner, and C. Taber. 1998. "Explaining Rising Wage Inequality: Explorations with a Dynamic General Equilibrium Model of Labor Earnings with Heterogeneous Agents." Review of Economics Dynamics 1 (1): 1­58. Hertel, T. W., and J. J. Reimer. 2005. "Predicting the Poverty Impacts of Trade Reform." Journal of International Trade and Economic Develop- ment 14 (4): 377­405. Hertel, Thomas W., and L. Alan Winters, eds. 2006. Poverty and the WTO: Impacts of the Doha Development Agenda. New York: Palgrave MacMillan. Kanbur, Ravi. 2000. "Income Distribution and Development." In Hand- book of Income Distribution, eds. Alexander B. Atkinson and Francois Bourguignon. Amsterdam: North-Holland. INTRODUCTION 23 Kuznets, Simon. 1955. "Economic Growth and Income Inequality." Amer- ican Economic Review 45 (1): 1­28. Lewis, W. A. 1954. "Economic Development with Unlimited Supplies of Labor." Manchester School of Economics and Social Studies 22: 139­81. Lofgren, Hans, Sherman Robinson, and Moataz El-Said. 2003."Poverty and Inequality Analysis in a General Equilibrium Framework: The Represen- tative Household Approach." In The Impact of Economic Policies on Poverty and Income Distribution: Evaluation Techniques and Tools, ed. François Bourguignon and Luiz A. Pereira da Silva, 325­37. Washington, DC: World Bank and Oxford University Press. Meerman, Jacob. 1979. Public Expenditure in Malaysia: Who Benefits and Why? New York: Oxford University Press. Ravallion, Martin. 2001. "Growth, Inequality and Poverty: Looking Beyond Averages" World Development 29 (11): 1803­15. ------. 2004. "Pro-Poor Growth: A Primer." World Bank Policy Research Working Paper No. 3242. World Bank, Washington, DC. ------. 2006. "Looking beyond Averages in the Trade and Poverty Debate." World Development 34 (8): 1374­92. Rosenstein-Rodan, Paul. 1943. "Problems of Industrialization in Southern and Eastern Europe." Economic Journal 53: 202­11. Rutherford Thomas, David Tarr, and Oleksandr Shepotylo. 2007. "The Impact on Russia of WTO Accession and the Doha Agenda: The Importance of Liberalization of Barriers against Foreign Direct Invest- ment in Services for Growth and Poverty Reduction." In The WTO and Poverty and Inequality, ed. L. Alan Winters. Cheltenham, UK: Edgar Elgar Publishing. Selowsky, Marcelo. 1979. Who Benefits from Government Expenditures? A Case Study of Colombia. New York: Oxford University Press. Sen, Amartya. 1973. On Economic Inequality. Oxford: Clarendon Press. Townsend, Robert. 2002. "Safety Nets and Financial Institutions in the Asian Crisis: The Allocation of Within-Country Risk." International Monetary Fund. Prepared for the IMF Conference on Macroeconomic Policies and Poverty Reduction, March 14­15, Washington, DC. Avail- able at http://cier.uchicago.edu/papers/papers.htm. Townsend, Robert, and Kenichi Ueda. 2006. "Financial Deepening, Inequality, and Growth: A Model-Based Quantitative Evaluation." Review of Economic Studies 73 (1): 251­93. PART I Top-Down Approach with Micro Accounting 2 Winners and Losers from Trade Reform in Morocco Martin Ravallion and Michael Lokshin As a water-scarce country, Morocco does not have a natural ad- vantage in its production of water-intensive crops such as most cereals--including wheat, which is used to produce the country's main food staples. In the past, the desire for aggregate self-sufficiency in the production of food staples has led to government efforts to foster domestic cereal production--even though cereals can be imported more cheaply. Since the 1980s, cereal producers have been protected by import tariffs as high as 100 percent. There have been concerns that the consequent reallocation of resources has hurt consumers and constrained the growth of pro- duction and trade. Reform to the current incentive system for cere- als has emerged as an important issue on the policy agenda for Morocco (World Bank 2003). The major obstacles to reform stem from concerns about the impacts on household welfare, particularly for the poor. Little careful research has been conducted to identify who will gain and who will lose from such reforms. Nonetheless, much debate about the equity implications has ensued. It is widely agreed that urban consumers will gain from lower cereal prices. More contentious are the impacts in (generally poorer) rural areas. Defenders of the existing protection system have argued that the rural economy will suffer from large welfare losses thanks to trade reform. Critics have argued against this view, claim- ing that the bulk of the rural poor tend to be net consumers and thus lose out from the higher prices because of trade protection. 27 28 RAVALLION AND LOKSHIN These critics argue that the rural poor are likely to gain from the reform, while it will be the well-off in rural areas, who tend to be net producers, who will lose (see, for example, Abdelkhalek 2002 and World Bank 2003). This chapter studies the household welfare impacts of the relative price changes induced by specific trade policy reform scenarios for cereals in Morocco. Past analyses of the welfare impacts have been highly aggregated, focusing on just one or a few categories of house- holds. The estimates presented here consider the impacts across 5,000 sampled households in the Morocco Living Standards Survey for 1998­99. This allows a detailed picture of the welfare impacts to emerge, thus enabling a more informed discussion of the social protection policy response to trade liberalization. Past approaches to studying the welfare impacts of specific trade reforms have tended to be either partial equilibrium analyses, in which the welfare impacts of the direct price changes caused by tar- iff changes are measured at the household level, or general equilib- rium analyses, in which second-round responses are captured in a theoretically consistent way but with considerable aggregation across household types. In general terms, the economics involved in both approaches is well known. And both approaches have found numerous applications.1 These two approaches are combined. In particular, the price changes induced by the trade policy change are simulated from a general equilibrium analysis done for a joint government of Morocco and World Bank study. The present study takes the methods and results of that analysis as given and carries them to the Moroccan Living Standards Survey. This approach respects the richness of detail available from a modern integrated household survey, making it pos- sible to go well beyond the highly aggregative types of analysis often found. Not only are the expected impacts measured across the dis- tribution of initial levels of living but how they vary by other char- acteristics, such as location, are also considered. This chapter is thus able to provide a reasonably detailed map of the predicted welfare impacts by location and socioeconomic characteristics. In studying the distributional impacts of trade reform, the chapter makes a distinction between the "vertical impact" and the "horizon- tal impact." The former concerns the way the mean impacts vary with the level of prereform income. How does the reform affect people at different prereform income levels? The horizontal impact relates to the disparities in impact between people at the same prereform level of income. As argued in Ravallion (2004), many past discussions of the distributional impacts of trade and other economywide reforms tended to focus more on the vertical impacts, WINNERS AND LOSERS FROM TRADE REFORM IN MOROCCO 29 analogous to standard practices in studying the "benefit incidence" of tax and spending policies. As demonstrated in this chapter, how- ever, that focus may well miss an important component of a policy's distributional impact arising from the horizontal dispersion of impacts at given prereform income levels. The study shows how the impact of a policy on a standard inequality measure can be straight- forwardly decomposed into its vertical and horizontal components. The former tells how much of the change in total inequality can be accounted for by the way in which mean impacts that are condi- tional on prereform income vary with the latter. If there is no differ- ence in the proportionate impact by level of income, then the vertical component is zero. The horizontal component tells the contribution of the deviations in impacts from their conditional means. Only when the impact of the reform is predicted perfectly by prereform income will the horizontal component be zero. The chapter studies the rela- tive importance of these two components of the predicted distribu- tional impact of trade reform in Morocco. The following section discusses the approach in general terms; the annex to the chapter provides more detail. The detailed results are presented under the heading on the measured impacts of reform, followed by a review of the main findings. Using Micro Data to Measure and Explain the Welfare Impacts of Reform This study uses preexisting estimates of the household-level welfare impacts of the price changes generated by a computable general equilibrium (CGE) model. The CGE analysis generates a set of price changes that embody both the direct price effects of the trade policy change and the indirect effects on the prices of both traded and non- traded goods, once all markets respond to the reform. Standard methods of first-order welfare analysis are used to measure the gains and losses at the household level. This approach is sequential rather than integrated. In other words, there is no feedback from the empir- ical analysis of welfare impacts to the CGE analysis. The alternative approach is to fully integrate the CGE analysis with the household- level data (see, for example, Cockburn 2006). The study's focus is very much on the short-term welfare impacts of trade policy changes. In keeping with the limitations of the gen- eral equilibrium analysis on which it draws, the approach does not capture the dynamic effects of trade reform through labor market adjustment and technological innovation. Nor does it capture poten- tial gains to the environment.2 30 RAVALLION AND LOKSHIN The specifics of the approach to estimating welfare impacts at the household level are outlined in the annex; only the salient features are summarized here.3 Each household has preferences over con- sumption and work effort represented by a standard utility func- tion. The household is assumed to be free to choose its preferred combinations of consumptions and labor supplies, subject to its budget constraint. The household also owns a production activity generating a profit that depends on a vector of supply prices and wages. The indirect utility function of the household (giving the maximum utility attainable given the constraints) also depends on prices and wages. The predicted price impacts from the CGE model are taken as given for the analysis of household-level impacts. In measuring these impacts, the authors are constrained by the data, which do not include prices and wages. This limitation does not pertain to calcu- lating a first-order approximation to the welfare impact in a neigh- borhood of the household's optimum consumption and labor sup- ply decisions (as outlined in the annex). This calculation provides a measure of the monetary value of the gain for household i, denoted gi. This value is obtained by adding the proportionate changes in all prices weighted by their corresponding expenditure or revenue shares. The weight for the proportionate change in the jth selling price is the revenue (selling value) from household production activ- ities in sector j; similarly, the consumption expenditure shares are the (negative) weights for demand price changes, and earnings pro- vide the weights for changes in the wage rates. The difference between revenue and expenditure gives (to a first-order approxima- tion) the welfare impact of an equi-proportionate increase in the price of a given commodity. In the specific model used, real wage rates are fixed. The likely implications of relaxing this assumption are discussed in the final section. By using the calculus in deriving the welfare gains, gi, for I 1, . . . , n, the authors implicitly assume small changes in prices. Relaxing this requires more information on the structure of the demand and supply system (see, for example, Ravallion and van de Walle 1991). This relaxation would entail considerable further effort, and given the aforementioned problem of incomplete price and wage data, the reliability of the results would be questionable. For the same reason, the authors have little choice but to largely ignore geo- graphic differences in the prices faced or in the extent to which bor- der price changes are passed on locally. Having estimated the impacts at the household level, the authors can study how they vary with prereform welfare and what impact the reform has on poverty and inequality. Let yi denote the prereform WINNERS AND LOSERS FROM TRADE REFORM IN MOROCCO 31 welfare per person in household i while yi* yi gi is its postreform value, where gi is the gain to household i. (Ideally, yi will be an exact money metric of utility, although in practice, it is expected to be an approximation given omitted prices or characteristics.) The distribution of postreform welfare levels is yi*, y2*, . . . y*n. By com- paring standard summary measures of poverty or inequality for this distribution with those for the prereform distribution, y1, y2, . . . yn, overall impacts can be assessed. It is interesting to see how the gains vary with prereform welfare. Is it the poor who tend to gain from these reforms, is it middle- income groups, or is it the rich? Importantly, however, the assign- ment of impacts to the prereform distribution is unlikely to be a degenerate distribution, with no distribution of its own. There will almost certainly be a dispersion in impact at the given prereform welfare levels. This dispersion will arise from (observable and unob- servable) heterogeneity in characteristics and prices. It can also arise from errors in the welfare measure. Averaging across the distribution of impacts at given prereform welfare, one can calculate the condi- tional mean impact given by the following: (2.1) gci Ei (giy yi), where the expectation is formed over the conditional distributions of impacts. By including a subscript i in the expectations operator in equation (2.1), one can allow the possibility that the horizontal dis- persion in impacts is not identically distributed. In the empirical implementation, equation (2.1) is estimated using a nonparametric regression. Taking these observations a step further, one can think of the overall impact on inequality as having both vertical and horizontal components.4 This is straightforward for the mean log deviation (MLD)--an inequality measure known to have a number of desir- able features.5 The MLD defined on the distribution of postreform welfares y*, y*, . . . y*is given by the following: 1 2 n 1 n (2.2) I* ln(y* y*), n i i 1 n where y* y*/n is mean postreform welfare. Similarly, i i 1 1 n (2.3) I ln(y yi) n i 1 is the prereform MLD. In equations (2.2) and (2.3), it is assumed that yi 0 and yi* 0 for all i. Thus, I* I is the change in inequality 32 RAVALLION AND LOKSHIN attributable to the reform. The proposed decomposition of the over- all change in inequality can then be written as follows: 1 n 1 g/y (2.4) I* I ln n 1 i 1 gi/yi 1 n 1 g/y 1 n 1 ln ln gci/yi . n 1 n 1 i 1 gci/yi i 1 gi / yi vertical component horizontal component The vertical component is the contribution to the change in total inequality (I* I) of the way in which mean impacts vary with pre- reform welfare levels. If there is no difference in the proportionate impact by level of welfare (gi /yic g/y for all i), then the vertical component is zero. The horizontal component is the contribution of the deviations in impacts from their conditional means. If the impact of the reform is predicted perfectly by prereform welfare (gi gci for all i), then the horizontal component is zero. This decomposition is largely of descriptive interest. There are no immediate policy impli- cations; however, finding that the horizontal component is large could well motivate greater effort by policy makers to understand what characteristics of households are associated with the differ- ences in impacts found empirically. To help in that task, one can go a step further and try to explain the differences in impacts in terms of observable characteristics of poten- tial relevance to social protection policies. The way the problem of measuring welfare impacts has been formulated allows utility and profit functions to vary between households at given prices. To explain the heterogeneity in measured welfare impacts, one can suppose instead that these functions vary with observed household character- istics. The characteristics that influence preferences over consumption (x1 ) are allowed to differ from those that influence the profits from i personal production activities (x2 ). The gain from the price changes i induced by trade reform depends on the consumption, labor supply, and household production choices, which depend in turn on prices and characteristics, x1 and x2 . For example, households with a higher i i number of children will naturally spend more on food, so if the rela- tive price of food changes, then the welfare impacts will be correlated with this aspect of household demographics. Similarly, differences in tastes may be associated with various life-cycle stages and education levels. Also, systematic covariates of the composition of welfare are likely. Generically, the gain can now be written as follows: (2.5) gi g(psi, pdi , wi, x1 , x2 ). i i WINNERS AND LOSERS FROM TRADE REFORM IN MOROCCO 33 Given that the household-specific wages and prices are not directly observable, further assumptions must be made. In explaining the variation across households in the predicted gains from trade reform, assume that (1) the wage rates are a function of prices and characteristics as wi w(pdi , psi, x1 , x2 ), and (2) differences in prices i i faced can be adequately captured by a complete set of regional dummy variables. Under these assumptions, and linearizing equa- tion (2.5) with an additive innovation error term, the following regression model can be used to represent the gains: (2.6) gi x x , 1 1i 2 2i k Dki i k where Dki 1 if household i lives in county k, and Dki 0 otherwise, and is the error term. i Measured Welfare Impacts of Trade Reform in Morocco Predicted Price Changes and the Survey Data The price changes (implied by trade reform) used in this analysis were generated by a CGE model that was commissioned by a joint working group of the Ministry of Agriculture of the government of Morocco and the World Bank, as documented in Doukkali (2003). The model was constructed to realistically represent the functioning of the Moroccan economy in 1997­98. The model was explicitly designed to assess the aggregate impacts of deprotecting cereals in Morocco. In addition to allowing for interactions between agriculture and the rest of the economy (represented by six sectors), the model is quite detailed in its representation of the agricultural sector. It allows for 16 differ- ent crops or groups of crops, 3 different livestock activities, 13 major agro-industrial activities, and 6 agro-ecological regions. Within each region, the model distinguishes between rain-fed agriculture and four types of irrigated agriculture. The model includes two types of labor, both with fixed real wage rates. Four policy simulations are undertaken. The simulations then differ in the extent of the tariff reductions for cereals--namely, 10 percent (Policy 1), 30 percent (Policy 2), 50 percent (Policy 3), and 100 percent (Policy 4). In all cases, the government's existing open-market opera- tions, which attempt to keep down consumer prices by selling sub- sidized cereals, are removed.6 The loss of revenue from a 50 percent tariff cut approximately equals the savings on subsidies. Table 2.1 gives the predicted price changes for various trade liberalization scenarios, based on Doukkali (2003).7 As expected, 34 Table 2.1 Predicted Price Changes Due to Agricultural Trade Reform in Morocco (percentage change in prices) Consumers Producers Sector Policy 1 Policy 2 Policy 3 Policy 4 Policy 1 Policy 2 Policy 3 Policy 4 Cereals and cereals products 3.062 7.786 12.811 26.691 2.858 7.193 11.744 24.107 Fresh vegetables 0.714 0.884 1.051 1.128 0.580 0.767 0.871 0.756 Fruits 0.637 0.681 0.683 0.139 0.429 0.301 0.104 0.843 Dairy products and eggs 0.472 0.414 0.257 0.751 0.505 0.487 0.333 0.637 Meat (red and poultry) 0.320 0.109 0.332 1.896 0.306 0.078 0.357 1.936 Sugar 0.200 0.100 0.400 1.300 0.368 0.378 0.354 0.094 Edible oils 0.671 1.064 1.405 2.225 0.632 0.998 1.336 2.061 Fresh and processed fish 0.000 0.696 1.300 2.996 0.000 0.600 1.300 2.881 Other agriculture and processed foods 0.369 0.402 0.421 0.635 0.268 1.294 2.475 5.388 Services 0.142 0.500 0.758 1.460 0.056 0.500 0.844 1.708 Energy, electricity, and water 0.060 0.540 1.140 2.580 0.051 0.549 1.149 2.597 Other industries 0.000 0.600 1.200 2.800 0.000 0.600 1.200 2.793 Source: Authors' calculations based on Doukkali's CGE analysis (2003). Note: The tariff cuts on imported cereals are 10, 30, 50, and 100 percent for Policies 1, 2, 3, and 4, respectively. WINNERS AND LOSERS FROM TRADE REFORM IN MOROCCO 35 the largest price impact is for cereals, although there are some non- negligible spillovers into other markets, reflecting substitutions in consumption and production and welfare effects on demand. Some of these spillover effects are compensatory--for example, some pro- ducer prices rise with the deprotection of cereals. The survey data set used here is the Enquête National sur le Niveau de Vie Ménages (ENNVM) for 1998, produced by the gov- ernment of Morocco's Department of Statistics, which kindly pro- vided the data set for this study. This is a comprehensive, multipur- pose survey that follows the practices of the World Bank's Living Standards Measurement Study.8 The ENNVM includes a sample of 5,117 households (of which 2,154 are rural) spanning 14 of Morocco's 16 regions; the low-density southernmost region--the former Spanish Sahara--is excluded. The sample is clustered and stratified by region and urban/rural areas. The survey does not include households without a fixed residence (sans abris). The sur- vey allows calculation of a comprehensive consumption aggregate (including imputed values for consumption from own production). The consumption numbers calculated by the Department of Statis- tics were used as the money metric of welfare. Ideally, this would be deflated by a geographic cost-of-living index, but no such index was available given the aforementioned lack of geographic price data. Implied Welfare Impacts at the Household Level Tables 2.2 and 2.3 give the budget and income shares at mean points and the mean welfare impacts, broken down by commodity based on the ENNVM; table 2.2 is for consumption, while table 2.3 is for production. (All consumption numbers include imputed values for consumption in kind.) Notice how different the consumption pat- terns are between urban and rural areas; for example, rural house- holds have twice the budget share for cereals as urban households. Strikingly, the 1.7 percent gain to urban consumers as a whole is largely offset by the general equilibrium effects achieved through other price changes (table 2.2). The income obtained directly from production accounts for about one-quarter of consumption; the rest is labor earnings, transfers, and savings. In rural areas the share is considerably higher, at 87 percent, and about one-third of this is from cereals.9 Table 2.4 summarizes the results on the implied welfare impacts. These results indicate that the partial trade reforms have only a small positive impact on the national poverty rate, as given by the percentage of the population living below the official poverty lines for urban and rural areas used by the government of Morocco's 36 RAVALLION AND LOKSHIN Table 2.2 Consumption Shares and Welfare Impacts through Consumption Consumption Indicator shares Policy 1 Policy 2 Policy 3 Policy 4 National Cereals 0.084 0.257 0.654 1.076 2.242 Fresh vegetables 0.042 0.030 0.037 0.044 0.047 Fruits 0.022 0.014 0.015 0.015 0.003 Dairy products and eggs 0.032 0.015 0.013 0.008 0.024 Meat (red and poultry) 0.112 0.036 0.012 0.037 0.213 Sugar 0.015 0.003 0.002 0.006 0.019 Edible oils 0.032 0.021 0.034 0.044 0.070 Fresh and processed fish 0.013 0.000 0.009 0.017 0.038 Agriculture and processed foods 0.101 0.037 0.040 0.042 0.064 Services 0.066 0.009 0.033 0.050 0.097 Energy, electricity, water 0.148 0.009 0.080 0.169 0.382 Other industries 0.333 0.000 0.200 0.400 0.933 Total 1.000 0.413 0.482 0.551 0.719 Urban Cereals 0.066 0.203 0.517 0.851 1.773 Fresh vegetables 0.037 0.026 0.033 0.039 0.042 Fruits 0.022 0.014 0.015 0.015 0.003 Dairy products and eggs 0.034 0.016 0.014 0.009 0.026 Meat (red and poultry) 0.107 0.034 0.012 0.035 0.203 Sugar 0.011 0.002 0.001 0.004 0.014 Edible oils 0.024 0.016 0.026 0.034 0.054 Fresh and processed fish 0.014 0.000 0.010 0.018 0.041 Agriculture and processed foods 0.096 0.035 0.039 0.040 0.061 Services 0.067 0.010 0.033 0.051 0.097 Energy, electricity, water 0.155 0.009 0.084 0.176 0.399 Other industries 0.368 0.000 0.221 0.441 1.030 Total 1.000 0.348 0.307 0.262 0.123 Rural Cereals 0.136 0.415 1.056 1.738 3.622 Fresh vegetables 0.055 0.039 0.049 0.058 0.062 Fruits 0.021 0.014 0.015 0.015 0.003 Dairy products and eggs 0.028 0.013 0.011 0.007 0.021 Meat (red and poultry) 0.128 0.041 0.014 0.043 0.243 Sugar 0.028 0.006 0.003 0.011 0.036 Edible oils 0.053 0.036 0.056 0.075 0.118 Fresh and processed fish 0.010 0.000 0.007 0.013 0.029 Agriculture and processed foods 0.115 0.042 0.046 0.048 0.073 Services 0.066 0.009 0.033 0.050 0.097 Energy, electricity, water 0.129 0.008 0.070 0.147 0.332 Other industries 0.232 0.000 0.139 0.278 0.650 Total 1.000 0.604 0.996 1.399 2.471 Source: Authors' estimations. WINNERS AND LOSERS FROM TRADE REFORM IN MOROCCO 37 Table 2.3 Percentage Gains from Each Policy: Production Component Production as a share of total Indicator consumption Policy 1 Policy 2 Policy 3 Policy 4 National Cereals 0.089 0.271 0.690 1.135 2.365 Fresh vegetables 0.053 0.038 0.047 0.056 0.060 Fruits 0.041 0.026 0.028 0.028 0.006 Dairy products and eggs 0.051 0.024 0.021 0.013 0.039 Meat (red and poultry) 0.000 0.000 0.000 0.000 0.000 Sugar 0.000 0.000 0.000 0.000 0.000 Edible oils 0.025 0.017 0.027 0.035 0.056 Fresh and processed fish 0.000 0.000 0.000 0.000 0.000 Agriculture and processed foods 0.002 0.001 0.001 0.001 0.001 Services 0.000 0.000 0.000 0.000 0.000 Energy, electricity, water 0.000 0.000 0.000 0.000 0.000 Other industries 0.000 0.000 0.000 0.000 0.000 Total 0.262 0.377 0.814 1.269 2.450 Urban Cereals 0.010 0.031 0.079 0.130 0.272 Fresh vegetables 0.008 0.006 0.007 0.009 0.009 Fruits 0.016 0.011 0.011 0.011 0.002 Dairy products and eggs 0.007 0.003 0.003 0.002 0.005 Meat (red and poultry) 0.000 0.000 0.000 0.000 0.000 Sugar 0.000 0.000 0.000 0.000 0.000 Edible oils 0.013 0.009 0.014 0.018 0.029 Fresh and processed fish 0.000 0.000 0.000 0.000 0.000 Agriculture and processed foods 0.000 0.000 0.000 0.000 0.000 Services 0.000 0.000 0.000 0.000 0.000 Energy, electricity, water 0.000 0.000 0.000 0.000 0.000 Other industries 0.000 0.000 0.000 0.000 0.000 Total 0.054 0.059 0.114 0.170 0.307 Rural Cereals 0.319 0.978 2.487 4.091 8.524 Fresh vegetables 0.186 0.133 0.165 0.195 0.210 Fruits 0.113 0.072 0.077 0.077 0.016 Dairy products and eggs 0.183 0.086 0.076 0.047 0.138 Meat (red and poultry) 0.000 0.000 0.000 0.000 0.000 Sugar 0.000 0.000 0.000 0.000 0.000 Edible oils 0.061 0.041 0.065 0.086 0.136 Fresh and processed fish 0.000 0.000 0.000 0.000 0.000 Agriculture and processed foods 0.008 0.003 0.003 0.004 0.005 Services 0.000 0.000 0.000 0.000 0.000 Energy, electricity, water 0.000 0.000 0.000 0.000 0.000 Other industries 0.000 0.000 0.000 0.000 0.000 Total 0.870 1.313 2.872 4.500 8.753 Source: Authors' estimations. 38 RAVALLION AND LOKSHIN Table 2.4 Household Impacts of Four Trade Reforms Indicator Baseline Policy 1 Policy 2 Policy 3 Policy 4 National Poverty rate (%) 19.61 20.01 20.33 21.04 22.13 Mean log deviation ( 100) 28.50 28.92 29.00 29.14 29.17 Gini index ( 100) 38.50 38.70 38.90 39.10 39.50 Per capita gain 0 6.52 23.97 54.82 133.81 Mean percentage gain: price changes weighted by mean shares 0 0.06 0.51 0.97 2.14 Mean percentage gain: weighted by ratios of means (tables 2.2 and 2.3) 0 0.04 0.33 0.72 1.73 Production gain 0 32.08 69.01 106.31 201.02 Consumption gain 0 38.60 45.05 51.49 67.21 Consumption per capita 9,350.91 9,357.43 9,326.95 9,296.10 9,217.10 Urban Poverty rate (percent) 12.19 12.05 11.96 12.05 11.76 MLD ( 100) 25.49 25.41 25.32 25.23 24.93 Gini index ( 100) 36.60 36.50 36.50 36.40 36.20 Per capita gain 0 35.52 24.80 13.75 16.49 Mean percentage gain: price changes weighted by mean shares 0 0.36 0.37 0.39 0.44 Mean percentage gain: weighted by ratios of means (tables 2.2 and 2.3) 0 0.29 0.19 0.09 0.18 Production gain 0 6.31 12.10 17.79 31.30 Consumption gain 0 41.83 36.90 31.54 14.81 Consumption per capita 12,031.20 12,066.72 12,056.00 12,044.95 12,014.71 Rural Poverty rate (percent) 28.28 29.31 30.10 31.54 34.25 MLD ( 100) 17.47 17.82 17.82 17.93 17.76 Gini index ( 100) 31.20 31.30 31.50 31.80 32.80 Per capita gain 0 33.53 91.32 149.51 295.85 Mean percentage gain: price changes weighted by mean shares 0 0.63 1.74 2.85 5.71 Mean percentage gain: weighted by ratios of means (tables 2.2 and 2.3) 0 0.71 1.88 3.10 6.28 Production gain 0 67.67 147.61 228.56 435.42 Consumption gain 0 34.14 56.29 79.05 139.57 Consumption per capita 5,649.03 5,615.50 5,557.71 5,499.52 5,353.19 Source: Authors' estimations. Note: All monetary units are Moroccan dirhams per year. Mean log deviation (MLD) is calculated only over the set of households for which consumption is posi- tive. The mean percentage gains weighted by mean shares are simply the means across the sample of the percentage gains at household level. The second mean per- centage gain is weighted by shares at the means points based on tables 2.2 and 2.3. WINNERS AND LOSERS FROM TRADE REFORM IN MOROCCO 39 statistics office.10 However, a larger impact emerges when complete deprotection is simulated (Policy 4). The national poverty rate then rises from 20 percent to 22 percent. All four reforms entail a decrease in urban poverty (though less than 0.4 percentage point) and an increase in rural poverty. The impacts over the whole distri- bution are examined later. Turning to the impacts on inequality, in table 2.4 one finds that the trade reforms yield a small increase in inequality, with the Gini index rising from 0.385 in the base case to 0.395 with a complete deprotection of cereals (Policy 4). Impacts are smaller for the partial reforms (Policies 1­3). The overall per capita gain is positive for the smaller tariff reduction (Policy 1) but becomes negative for Policies 2, 3, and 4. As expected, there is a net gain to consumers and a net loss to producers, though the amounts involved are small overall. There are small net gains in the urban sector for Policies 1­3. Larger impacts are found in rural areas, as expected. The mean percentage loss from complete deprotection is a (nonnegligible) 5.7 percent in rural areas. Table 2.4 presents the results derived for the impact on poverty as estimated using the government's official poverty lines. It is important to test robustness to alternative poverty lines. For this purpose, the poverty incidence curve, which is the cumulative dis- tribution function up to a reasonable maximum poverty line, is used. The results are given in figure 2.1; for readability, the figure focuses on Policies 1 and 4. The curves for Policies 2 and 3 fall between these two policies. From this figure, it is clear that there is an increase in overall poverty from complete deprotection; this is robust to the poverty line and poverty measure used (within a broad class of measures; see Atkinson 1987). The impact on poverty is found almost entirely in rural areas; indeed, there is virtually no impact on urban poverty. In rural areas, however, the results in figure 2.1 suggest a sizable impact on poverty from complete deprotection. The mean loss as a propor- tion of consumption for the poorest 15 percent in rural areas is about 10 percent. There is an increase in the proportion of the rural population living below 2,000 dirhams per person per year, from 6.2 percent to 9.9 percent; the proportion living below 3,000 dirhams rises from 22.2 percent to 26.3 percent. For the country as a whole, the poverty rate for the former poverty line (2,000 dirhams) rises from 2.8 percent to 4.4 percent under Policy 4, while it rises from 11.4 percent to 13.1 percent for the 3,000 dirhams line. The finding of adverse impacts on the rural poor contradicts claims made by some observers who argue that the rural poor tend to be net con- sumers of cereals, the commodity that incurs the largest price 40 RAVALLION AND LOKSHIN Figure 2.1 Impacts of Trade Reform Policies on Poverty in Morocco (Moroccan dirhams) a. Total b. Urban 0.6 0.6 line line 0.5 0.5 poverty 0.4 poverty 0.4 the the 0.3 0.3 below 0.2 below 0.2 people 0.1 people 0.1 % % 0 0 0 1,000 2,000 3,000 4,000 5,000 0 1,000 2,000 3,000 4,000 5,000 annual per capita consumption annual per capita consumption c. Rural 0.6 line 0.5 poverty 0.4 the 0.3 below 0.2 people 0.1 % 0 0 1,000 2,000 3,000 4,000 5,000 annual per capita consumption Baseline Policy 1 Policy 4 Source: Authors' calculations. Note: Total population of sites surveyed, urban and rural. decrease with this trade reform (table 2.1). This point is discussed in more detail in the section on the welfare impacts of these policy reforms. Table 2.5 presents the mean impacts of Policy 4 by region, split between urban and rural areas. Impacts in urban areas are small in all regions, with the highest net gain as a percentage of consumption being 1.3 percent in Tanger-Tetouan, closely followed by Tensift Al Haouz and Fes-Boulemane. The rural areas with largest mean losses WINNERS AND LOSERS FROM TRADE REFORM IN MOROCCO 41 Table 2.5 Mean Gains from Policy 4, by Region Poorest 15% of rural Region Total Urban Rural households Oued Ed-Dahab-Lagouira 0.20 0.20 n.a. n.a. Laayoune-Boujdour-Sakia El Hamra 0.34 0.34 n.a. n.a. Guelmime Es-Semara 0.96 0.72 3.47 0.58 Souss-Massa-Daraa 1.31 0.42 2.4 3.09 Gharb-Chrarda-Beni Hssen 2.16 0.02 3.86 0.10 Chaouia-Ouardigha 4.18 0.32 8.31 10.11 Tensift Al Haouz 0.87 1.12 2.17 0.31 Oriental 0.87 0.38 2.78 0.25 G.Casablanca 0.48 0.41 2.41 n.a. Rabat-Salé-Zemmour-Zaer 0.59 0.33 4.98 0.23 Doukala Abda 3.13 0.76 5.92 3.93 Tadla Azilal 6.93 0.71 11.04 0.95 Meknes Tafil 4.89 0.19 11.35 8.48 Fes-Boulemane 2.4 1.05 11.52 13.43 Taza-Al Hoceima-Taounate 4.47 0.32 5.78 8.39 Tanger-Tetouan 2.94 1.31 9.4 22.03 Total 2.14 0.45 5.71 10.39 Source: Authors' estimations. Note: Means formed over the household-level percentage gains (equivalent to weight- ing proportionate price changes by mean shares); n.a. not applicable. from deprotection of cereals are Tasla Azilal, Meknes Tafil, Fes- Boulemane, and Tanger-Tetouan. Table 2.5 gives mean impacts for the poorest 15 percent in rural areas (in terms of consumption per person). When the focus is on the rural poor defined in this way, the region incurring the largest mean loss for rural households is Tanger- Tetouan, followed by Fes-Boulemane and Chaouia-Ouardigha. The contrast between the small net gains to the urban sector and net losses to the rural poor is most marked in Tanger-Tetouan. To begin exploring the heterogeneity in welfare impacts, fig- ure 2.2 illustrates the cumulative frequency distributions of the gains and losses. The figure is simplified by again focusing on Policies 1 and 4. With complete deprotection (Policy 4), about 9 percent of the households incur losses greater than 500 dirhams per year (about 5 percent of overall mean consumption), while about 5 per- cent lose more than 1,000 dirhams per year. As expected, rural areas have a "thicker tail" of negative gains. About 16 percent of rural households lose more than 500 dirhams and 10 percent lose more than 1,000. In figure 2.3, the mean gains are plotted against percentiles of con- sumption per capita for Policies 1 and 4. Both absolute gains/losses and gains/losses as a percentage of the household's consumption are 42 RAVALLION AND LOKSHIN Figure 2.2 Frequency Distributions of Gains and Losses for Trade Policies 1 and 4 (Moroccan dirhams) a. Total, cumulative density b. Total, probability density 1.0 0.020 0.9 0.8 0.015 0.7 density 0.6 density 0.5 0.010 0.4 0.3 0.005 cumulative 0.2 probability 0.1 0 0 3,000 2,000 1,000 0 1,000 600 400 200 0 200 400 absolute gain per capita absolute gain per capita c. Urban d. Rural 1.0 1.0 0.9 0.9 0.8 0.8 0.7 0.7 density 0.6 density 0.6 0.5 0.5 0.4 0.4 0.3 0.3 cumulative 0.2 cumulative 0.2 0.1 0.1 0 0 3,000 2,000 1,000 0 1,000 3,000 2,000 1,000 0 1,000 absolute gain per capita absolute gain per capita Policy 1 Policy 4 Source: Authors' calculations. shown. For Policy 1, the mean absolute gain has a tendency to rise as one moves from the poorest percentile to the richest, though the gra- dient is small. The mean proportionate gain is quite flat. For Policy 4, mean absolute impacts also rise up to the richest decile or so, but then fall. Proportionate gains follow the same pattern, though (again) the gradient seems small. What is most striking from figure 2.3 is the wide spread of impacts, particularly downward (indicating losers from the reform). The vari- ance in absolute impacts is especially large at the upper end of the consumption distribution, though the dispersion in proportionate WINNERS AND LOSERS FROM TRADE REFORM IN MOROCCO 43 Figure 2.3 Absolute and Proportionate Gains for Policies 1 and 4 a. Policy 1 1,500 30 1,000 20 capita 500 10 per capita 0 0 500 per 10 1,000 gain/loss 1,500 20 2,000 gain/loss 30 2,500 % 40 absolute 3,000 50 0 20 40 60 80 100 0 20 40 60 80 100 per capita consumption per capita consumption percentiles percentiles b. Policy 4 1,500 30 1,000 20 capita 500 10 per capita 0 0 500 per 10 1,000 gain/loss 1,500 20 2,000 gain/loss 30 2,500 % 40 absolute 3,000 50 0 20 40 60 80 100 0 20 40 60 80 100 per capita consumption per capita consumption percentiles percentiles Source: Authors' calculations. impacts tends to be greater at the other end of the distribution, among the poorest. Figure 2.4 provides a split between producers and consumers for Policy 4. As expected, to the extent that there is much impact on producers, they tend to lose, although not more so for poor pro- ducers than rich ones. For consumption, there tends to be more gainers and a higher variance in impact as one moves up the con- sumption distribution. However, the downward dispersion in total welfare impacts shown in figure 2.3 is due more to the conditional variance in impacts through production than through consumption. There are two especially striking findings in figures 2.3 and 2.4. First, notice the sizable losses on the production side among the poor. 44 RAVALLION AND LOKSHIN Figure 2.4 Production and Consumption Decomposition of the Welfare Impacts for Policy 4 a. Policy 4, production b. Policy 4, consumption 1,500 1,500 1,000 1,000 capita capita 500 500 per 0 per 0 500 500 1,000 1,000 gain/loss 1,500 gain/loss 1,500 2,000 2,000 2,500 2,500 absolute 3,000 absolute 3,000 0 20 40 60 80 100 0 20 40 60 80 100 per capita consumption per capita consumption percentiles percentiles Source: Authors' calculations. Granted, some large losses are evident for the high-income groups. But the claims that the poor do not lose as producers are clearly false. Furthermore, the poor are often not seeing compensatory gains as consumers. Second, it is notable that the results in figures 2.3 and 2.4 indi- cate that the mean gains vary little with mean consumption. Focusing on the "poor" versus the "rich" is of little interest in characterizing gainers and losers from this reform. The diversity in impacts tends to be "horizontal" in the distribution of income--meaning that larger differences in impacts are found at given consumption levels rather than between different levels of consumption. These two findings are now examined in greater detail. Who Are the Net Producers of Cereals in Morocco? In the population as a whole, 16 percent of households are net pro- ducers (value of cereals production exceeds consumption). These households are worse off from the fall in cereal prices because of deprotection. In rural areas, this proportion is 36 percent. However, the survey data do not support the claim that the rural poor in Morocco are (on average) net consumers of cereals. Figure 2.5 illustrates how producers and net producers are spread across WINNERS AND LOSERS FROM TRADE REFORM IN MOROCCO 45 Figure 2.5 Net Producers of Cereals in the Distribution of Total Consumption per Person in Rural Areas of Morocco 1.00 1.00 0.75 0.75 cereals of producers 0.50 0.50 cereal producers % 0.25 net 0.25 % 0 0 0 20 40 60 80 100 0 20 40 60 80 100 consumption per capita consumption per capita percentiles percentiles 1.00 cereals of 0.75 consumption 0.50 total 0.25 consumption of 0 minus 0.25 proportion a as 0.50 production 0 20 40 60 80 100 consumption per capita percentiles Source: Authors' calculations. the distribution of total household consumption per person in rural Morocco. Both the scatter of points and the conditional means esti- mated using the local regression method are given.11 The first (top left) panel in figure 2.5 shows the proportion of producers, followed by the proportion of net producers (for whom production exceeds consumption of cereals in value terms). Finally, the net production is presented in value terms. In each case, the horizontal axis gives the percentile of the distribution of consumption from poorest to richest. 46 RAVALLION AND LOKSHIN As shown in this figure, a majority of the rural poor produce cereals. Naturally, much of this is for home consumption. However, even if the focus is solely on net producers, one finds that more than one-third of those in the poorest quintile tend to produce more than they consume. Furthermore, the mean net production in value terms tends to be positive for the poor; in rural areas, the losses to poor producers from falling cereal prices outweigh the gains to poor consumers. More than any single feature of the survey data, it is this fact that lies at the heart of the finding that the rural poor lose from the deprotection reform. Vertical versus Horizontal Impacts on Inequality To measure the relative importance of the vertical versus horizontal differences in impact, one can use the decomposition method out- lined in the section on micro data near the beginning of this chapter. This decomposition requires an estimate of the conditional mean E(gy), that is, the regression function of g on y. In the study, this was estimated using the nonparametric local regression method of Cleveland (1979). Table 2.6 gives the results of this decomposition for each of the four policy reforms examined. For the small partial reform under Policy 1, the vertical component dominates, accounting for 73 percent of the impact on inequality. As one moves to the bigger reforms, however, the horizontal component becomes relatively large. Indeed, one finds that 119.8 percent of the impact of Policy 4 on inequality is attribut- able to the horizontal component, while ­19.8 percent is due to the vertical component. So the vertical component was inequality reduc- ing for Policy 4, even though overall inequality rose (table 2.6). A high degree of horizontal inequality is clear in measured impacts at given mean consumption. Some of this is undoubtedly measurement error, which may well become more important for larger reforms. But some is attributable to observable covariates of consumption and production behavior, as discussed earlier. In trying Table 2.6 Decomposition of the Impact on Inequality Indicator Policy 1 Policy 2 Policy 3 Policy 4 Vertical component 72.69 57.57 38.77 19.77 Horizontal component 27.31 42.43 61.23 119.77 Total 100.00 100.00 100.00 100.00 Source: Authors' estimations. Note: The decomposition is implemented only on the sample of households for which both the baseline consumption and the postreform consumption are positive. WINNERS AND LOSERS FROM TRADE REFORM IN MOROCCO 47 to explain this variance in welfare impacts, the characteristics con- sidered include region of residence, whether the household lives in an urban area, the household's size and demographic composition, the age and age-squared of the household head, and education and dummy variables describing key aspects of the occupation and prin- cipal sector of employment. Table 2.7 gives summary statistics on the variables to be used in the regressions. Although there are endo- geneity concerns about these variables, those concerns are thought to be minor in this context, especially when weighed against the concerns about omitted variable bias in estimates that exclude these characteristics. Under the usual assumption that the error term is orthogonal to these regressors, one can estimate equation (2.6) by ordinary least squares (OLS). The results are given in table 2.8. These results are averages across the impacts of these characteristics on the consumption and pro- duction choices that determine the welfare impact of given price and wage changes. This makes interpretation difficult. These regressions are viewed for their descriptive interest and their ability to isolate covariates of potential relevance in thinking about compensatory policy responses. Focusing first on the results for Policy 4, larger losses from full deprotection of cereals are associated with families that (1) live in rural areas; (2) are relatively smaller (the turning point in the U-shaped relationship is at a household size of about one); (3) have more wage earners, higher education, work in commerce and transport, for example; and (4) live in Chaouia-Ouardigha, Rabat, Tadla Azilal, and Meknes Tafil. Recall that these effects stem from the way household characteristics influence net trading positions in terms of the commodities for which prices change. It appears that larger families tend to consume more cereals and thus gain more from the lower cereal prices. Results are similar for partial deprotection, though education becomes insignificant for Policy 1. Table 2.9 presents an urban-rural breakdown of the regressions for Policies 1 and 4. There are a couple of notable differences. (Again, the focus is on Policy 4 in the interest of brevity.) There are significant positive effects of having more children and teenagers on the gains from trade reform in rural areas, presumably because such families are more likely to be cereal consumers. The education effect at higher levels of schooling is much more pronounced in urban areas. The effect of working in the transport and commerce sector is more statistically significant in urban areas, though this effect is still sizable in rural areas. The regional effects are more sta- tistically significant in urban areas than in rural areas. There are 48 RAVALLION AND LOKSHIN Table 2.7 Summary Statistics on Explanatory Variables in the Regression Analysis Standard Variable Mean deviation Household characteristics Urban 0.580 binary Log household size 1.645 0.550 Log squared household size 3.009 1.621 Female head of household 0.170 binary Currently unemployed 0.248 binary Number of wage earners 5.912 2.878 Share of children 0­6 years of age 0.140 0.162 Share of children 7­17 years of age 0.221 0.204 Share of elderly 60 0.120 binary Characteristics of household head Age of household head 0.505 0.143 Age of household head squared 0.275 0.155 Illiterate head of household 0.582 binary Primary school not completed 0.100 binary Primary school completed 0.164 binary Lower secondary school completed 0.058 binary Upper secondary school completed 0.059 binary University completed 0.036 binary Industry Not employed 0.240 binary Manufacturing/construction 0.004 binary Commerce/transportation/communication/administration 0.273 binary Social services 0.085 binary Other services 0.064 binary Public servants 0.125 binary Unemployed or laid-off worker 0.012 binary Housewife or student 0.037 binary Young child 0.009 binary Old or retired 0.074 binary Sick 0.068 binary Other inactives 0.010 binary Region Oued Ed-Dahab-Lagouira 0.012 binary Laayoune-Boujdour-Sakia El Hamra 0.014 binary Guelmime Es-Semara 0.023 binary Souss-Massa-Daraa 0.094 binary Gharb-Chrarda-Beni Hssen 0.058 binary Chaouia-Ouardigha 0.054 binary Tensift Al Haouz 0.100 binary Oriental 0.065 binary G.Casablanca 0.124 binary Rabat-Salé-Zemmour-Zaer 0.081 binary Doukala Abda 0.067 binary Tadla Azilal 0.047 binary Meknes Tafil 0.072 binary Fes-Boulemane 0.051 binary Taza-Al Hoceima-Taounate 0.058 binary Source: Authors' estimations. Table 2.8 Regression of per Capita Gain/Loss on Selected Household Characteristics Policy 1 Policy 2 Policy 3 Policy 4 Standard Standard Standard Standard Variable Coefficient error Coefficient error Coefficient error Coefficient error Household characteristics Urban 26.139*** 6.275 44.850*** 12.948 64.218** 20.068 113.714** 39.213 Log household size 57.242** 19.583 78.454* 40.407 100.548 62.626 157.373 122.376 Log squared household size 77.337*** 16.806 167.523*** 34.678 260.865*** 53.746 508.026*** 105.023 Female-headed household 2.502 7.431 4.072 15.333 5.605 23.765 9.161 46.438 Currently unemployed 10.018* 5.909 23.344* 12.192 36.428* 18.896 67.997* 36.924 Number of wage earners in household 44.722*** 7.019 101.428*** 14.484 159.842*** 22.448 313.541*** 43.865 Share of children ages 0­6 32.783* 17.72 89.774* 36.564 145.705* 56.67 277.637* 110.736 Share of children ages 7­17 25.070* 14.155 69.367* 29.206 113.738* 45.266 221.518* 88.453 Share of elderly 60 21.3 15.584 23.551 32.155 24.389 49.837 24.334 97.385 Characteristics of household head Age of the head 38.511 108.759 151.473 224.41 272.681 347.809 624.596 679.642 Age of the head squared 44.097 102.579 142.598 211.658 246.231 328.045 543.07 641.022 Household head is literate only 8.871 7.983 23.441 16.472 38.257 25.53 76.735 49.888 Incomplete primary educationa Primary school completed 14.013* 6.757 40.623** 13.942 68.220** 21.608 141.296*** 42.224 Lower secondary school 12.98 10.4 61.634** 21.458 112.583*** 33.258 250.335*** 64.989 Upper secondary school 12.462 10.775 70.619** 22.233 130.320*** 34.458 286.333*** 67.333 University 2.575 13.527 95.376*** 27.912 197.887*** 43.26 476.077*** 84.533 Industry Not working or agriculturea Manufacturing/construction 3.71 36.465 0.277 75.242 4.541 116.616 21.281 227.874 Commerce/transportation/ communication/administration 59.926*** 8.198 122.454*** 16.915 185.113*** 26.216 341.751*** 51.228 Social services 4.424 10.036 17.18 20.707 30.536 32.094 66.804 62.714 49 Other services 0.2 11.251 9.572 23.214 19.812 35.98 47.874 70.306 Public servants 2.385 8.936 6.785 18.439 10.912 28.579 20.23 55.844 Unemployed or laid-off worker 6.627 21.518 27.715 44.399 49.65 68.813 107.951 134.465 (Continued on the following page) Table 2.8 (Continued) 50 Policy 1 Policy 2 Policy 3 Policy 4 Standard Standard Standard Standard Variable Coefficient error Coefficient error Coefficient error Coefficient error Housewife or student 2.26 13.49 13.788 27.835 25.401 43.141 55.785 84.301 Young child 7.629 24.5 3.891 50.553 16.336 78.352 51.207 153.104 Old or retired 6.913 11.039 23.527 22.778 40.651 35.303 86.8 68.984 Sick 3.143 10.96 22.092 22.614 42.489 35.049 100.065 68.488 Other inactives 9.955 22.723 1.817 46.885 15.364 72.667 56.497 141.995 Region Oued Ed-Dahab-Lagouira 19.216 22.51 6.738 46.446 34.818 71.986 111.388 140.665 Laayoune-Boujdour-Sakia El Hamra 1.502 21.067 20.145 43.47 40.764 67.374 98.323 131.652 Guelmime Es-Semara 9.666 16.639 11.901 34.333 12.774 53.212 12.391 103.979 Souss-Massa-Daraa 7.645 10.868 5.611 22.425 22.766 34.756 85.2 67.916 Gharb-Chrarda-Beni Hssen 10.087 12.229 7.485 25.232 3.592 39.107 10.494 76.418 Chaouia-Ouardigha 19.542 12.507 49.255* 25.807 81.319* 39.998 169.114* 78.159 Tensift Al Haouz 2.964 10.696 14.527 22.071 27.258 34.207 65.274 66.842 Oriental 14.038 11.928 19.198 24.612 23.918 38.145 31.056 74.539 G.Casablanca 3.322 10.429 15.762 21.518 28.418 33.35 60.086 65.169 Rabat-Salé-Zemmour-Zaer 15.439 11.326 33.817 23.371 52.199 36.222 97.061 70.78 Doukala Abda 13.169 11.76 23.668 24.265 34.315 37.607 59.462 73.487 Tadla Azilal 55.774*** 13.093 114.700*** 27.016 174.099*** 41.872 320.810*** 81.821 Meknes Tafil 37.594** 11.54 74.192** 23.812 111.929** 36.906 209.391** 72.117 Fes-Boulemane 10.249 12.726 15.356 26.259 20.651 40.699 33.326 79.528 Taza-Al Hoceima-Taounate 5.613 12.367 2.43 25.517 2.415 39.549 21.329 77.281 Tanger-Tetouana Constant 144.096*** 34.638 247.104*** 71.472 354.469** 110.773 642.381** 216.458 R2 0.175 0.080 0.062 0.057 Source: Authors' estimations. Note: *** and ** denote significance at the 5 percent and 10 percent levels, respectively. a. Reference category. Table 2.9 Urban-Rural Split of Regressions for per Capita Gains Urban Rural Policy 1 Policy 4 Policy 1 Policy 4 Standard Standard Standard Standard Variable Coefficient error Coefficient error Coefficient error Coefficient error Household characteristics Log household size 32.840* 16.071 45.705 83.159 89.255* 45.084 527.017* 294.353 Log squared household size 40.492* 17.841 217.663* 92.32 79.415* 32.524 555.880** 212.348 Female-headed household 2.696 6.018 15.603 31.139 11.984 16.902 27.785 110.356 If unemployed present 2.138 4.668 25.238 24.154 11.086 14.482 35.299 94.551 Number of wage earners 23.972** 8.39 143.745*** 43.414 45.101*** 12.237 321.182*** 79.894 Share of children 0­6 15.648 15.206 25.903 78.686 95.815** 36.544 609.370* 238.601 Share of children 7­17 10.44 11.986 34.073 62.023 81.378** 29.771 622.563** 194.376 Share of elderly 60 17.696 13.328 4.67 68.967 35.448 32.512 167.42 212.274 Characteristics of household head Age of the head 26.02 96.18 513.051 497.696 82.081 216.7 1.00E 03 1414.846 Age of the head squared 33.769 91.377 263.429 472.842 103.772 202.766 1129.226 1323.868 Household head is literate only 10.567 6.965 90.700* 36.042 8.718 16.11 75.293 105.182 Incomplete primary educationa Primary school completed 0.157 5.566 44.272 28.804 31.613* 14.794 270.881** 96.589 Low secondary school 6.416 7.632 119.177** 39.494 73.971* 31.399 655.218** 205.005 Upper secondary school 5.731 7.551 249.358*** 39.074 10.925 49.861 46.655 325.547 University 9.241 9.282 433.456*** 48.03 20.185 83.244 18.883 543.507 Industry Not working or agriculturea Manufacturing/construction 4.779 25.641 7.254 132.684 56.769 124.939 366.598 815.737 Commerce/transportation/ communication/administration 96.116*** 10.172 444.047*** 52.634 43.789** 15.445 257.349* 100.843 Social services 1.428 7.574 6.102 39.191 27.61 28.965 247.156 189.116 51 Other services 4.7 9.133 6.023 47.259 21.228 25.434 161.257 166.061 Public servants 2.611 6.884 19.401 35.621 8.742 23.042 57.723 150.44 Unemployed or laid-off worker 1.702 15.213 36.377 78.72 60.148 73.543 457.084 480.167 (Continued on the following page) Table 2.9 (Continued) 52 Urban Rural Policy 1 Policy 4 Policy 1 Policy 4 Standard Standard Standard Standard Variable Coefficient error Coefficient error Coefficient error Coefficient error Housewife or student 4.019 10.145 12.554 52.498 20.295 36.207 110.127 236.4 Young child 2.268 16.343 129.322 84.567 107.247 152.23 720.704 993.92 Old or retired 1.108 8.138 48.765 42.112 25.588 34.261 154.32 223.691 Sick 1.847 8.176 63.019 42.308 5.864 30.489 148.543 199.063 Other inactives 12.094 16.532 23.685 85.547 22.652 67.323 250.306 439.559 Region Oued Ed-Dahab-Lagouira 21.2 15.068 135.288* 77.973 n.a. n.a. n.a. n.a. Laayoune-Boujdour-Sakia El Hamra 2.496 14.153 129.348* 73.236 n.a. n.a. n.a. n.a. Guelmime Es-Semara 7.558 13.813 50.41 71.475 23.284 35.563 165.753 232.195 Souss-Massa-Daraa 1.425 10.023 54.723 51.863 8.417 21.371 211.302 139.535 Gharb-Chrarda-Beni Hssen 44.733*** 11.143 204.020*** 57.663 17.31 23.762 208.808 155.141 Chaouia-Ouardigha 15.625 11.08 89.734 57.333 19.527 25.012 201.804 163.304 Tensift Al Haouz 8.763 9.759 37.2 50.5 8.732 21.097 147.015 137.74 Oriental 18.776* 9.806 96.129* 50.74 0.357 25.851 99.206 168.782 G.Casablanca 9.23 7.849 112.350** 40.617 5.551 49.268 79.412 321.673 Rabat-Salé-Zemmour-Zaer 13.825 8.683 118.444** 44.931 36.873 30.677 142.714 200.295 Doukala Abda 14.916 10.867 80.126 56.232 8.244 22.773 3.679 148.687 Tadla Azilal 50.624*** 12.423 213.855*** 64.285 51.570* 24.832 324.785* 162.13 Meknes Tafil 22.753* 9.622 126.779* 49.79 56.111* 24.782 311.079* 161.8 Fes-Boulemane 11.946 9.954 38.193 51.509 2.002 30.661 5.31 200.186 Taza-Al Hoceima-Taounate 20.264 13.982 161.597* 72.352 16.747 22.229 80.917 145.137 Tanger-Tetouana Constant 135.395*** 30.386 463.951** 157.234 162.613* 72.909 959.343* 476.029 R2 0.46 0.08 0.062 0.067 Source: Authors' estimations. Note: *** and ** denote significance at the 5 percent and 10 percent levels, respectively. a. Reference category. WINNERS AND LOSERS FROM TRADE REFORM IN MOROCCO 53 still sizable regional differences in mean impacts in table 2.9, but they are statistically less significant than those found in table 2.8. In fact, the quantitative magnitudes of the regional differences are just as large for the rural areas in table 2.9 as for urban plus rural areas in table 2.8. The results in tables 2.8 and 2.9 are conditional geographic effects (conditional on the values taken by other covariates in the regres- sions). As in table 2.5, there are pronounced (unconditional) geo- graphic differences in mean impacts in rural areas across different regions. Whether one draws policy lessons more from the condi- tional or unconditional effects depends on the type of policy being used. If the policy is simply one of regional targeting, then the uncon- ditional geographic effects in table 2.5 will be more relevant. How- ever, finer targeting by household characteristics, in combination with regional targeting, will call for the type of results presented in tables 2.8 and 2.9. The share of the variance in gains that is accountable to these covariates is generally less than 10 percent. Values of R2 of this size are common in regressions run on large cross-sectional data sets, although it remains true that a large share of the variance in impacts is not accountable to these covariates. (The exception to the low R2 is for Policy 1, for which almost half of the variance in gains across urban households is explained.) A sizable degree of measurement error in the gains must be expected--stemming from measurement error in the underlying consumption and production data. No doubt, there are also important idiosyncratic factors in household-specific tastes or production choices. These regressions try to explain the variance in the gains from the reform. It is interesting to see how one might better explain the inci- dence of losses from reform among the poor. This is arguably of greater relevance to compensatory policies, which presumably would focus on those among the poor who lose from such reforms. To test how well the same set of regressors could explain who was a "poor" loser from the reforms, a dummy variable was constructed that takes the value unity if a rural household incurred a negative loss and was "poor." To ensure a sufficient number of observations taking the value unity, the poverty line was set higher than the offi- cial line--that is, at a per person consumption of 5,000 dirhams per year (rather than at the official line of about 3,000). This was con- fined to rural areas because that is where the losses are concentrated. In the case of full deprotection (Policy 4), about 14 percent of the variance in this measure can be explained by the set of regressors in table 2.9; for Policy 1, the share is 20 percent.12 A number of 54 RAVALLION AND LOKSHIN covariates can identify likely losers among the poor, but a large share of the variance is left unexplained. Another way to assess how effectively this set of covariates can explain the incidence of a net loss from reform among the poor is by comparing the actual value of the dummy variable described earlier with its predicted values from the model, using a cutoff probability of 0.5. For Policy 4,472 households out of 2,100 both were poor and incurred a loss because of the reform. Of these households, the model could correctly predict that this was the case only for 18 per- cent (86 households). For Policy 1, the model prediction was correct for 27 percent of the 463 households that were both poor and made worse off by the reform. Most forms of indicator targeting--whereby transfers are contin- gent on readily observed variables like location--would be based on variables similar to those used in these regressions; indeed, targeted policies use fewer dimensions. This suggests that indicator targeting is of only limited effectiveness in reaching those in greatest need. Self-targeting mechanisms that create incentives for people to cor- rectly reveal their status (such as using work requirements) may be better able to reach these people. Two Caveats Although the results presented here are suggestive, two limitations of this analysis should be noted. The first concern stems from the fact that the Doukkali (2003) model assumed fixed wage rates. Sensitiv- ity to alternative labor market assumptions should be checked, but one can speculate on the likely impacts of allowing real wages to adjust to the reforms. Here it can be argued that the export-oriented cash crops that will replace cereals will tend to be more labor inten- sive than cereals. Thus, higher aggregate demand for the relatively unskilled labor used in agriculture would be expected, and hence higher real wages for relatively poorer groups would be realized. This wage increase will undoubtedly go some way toward compen- sating the rural poor--and may even tilt the vertical distributional impacts in favor of the poor. The second concern is that dynamic gains from greater trade openness may not be captured by the model used to generate the rel- ative price impacts; for example, trade may facilitate learning about new agricultural technologies and innovation that brings longer- term gains in farm productivity. These effects may be better revealed by studying time-series evidence combined with cross-country comparisons. WINNERS AND LOSERS FROM TRADE REFORM IN MOROCCO 55 Conclusion The welfare impacts of deprotection in developing countries have been much debated. Some people have argued that external trade liberalizations are beneficial to the poor, whereas others argue that the benefits will be captured more by the nonpoor. Expected impacts on domestic prices have figured prominently in these debates. This chapter has studied the welfare impacts at the household level of the changes in commodity prices attributed to a proposed trade reform, namely, Morocco's deprotection of its cereals sector. This deprotection would entail a sharp reduction in tariffs, with implications for the domestic structure of prices and hence house- hold welfare. The analysis presented here draws out the implica- tions for household welfare of the previous estimates of the price impacts of reform undertaken for a joint government of Morocco and World Bank study. The estimates of price impacts are entirely external to (and predate) this analysis. Here some standard methods of first-order welfare analysis are used to measure the gains and losses at the household level using a large sample survey. In future work using this methodology, there may well be more scope for feedback from the household-level analysis to the CGE modeling used to derive price impacts. In a number of respects, this detailed household-level analysis throws into question past claims about the likely welfare impacts of this trade reform. In the aggregate, one finds a small negative impact on mean household consumption and a small increase in inequality. There is a sizable, and at least partly explicable, variance in impacts across households. Rural families tend to lose; urban households tend to gain. Some provinces experience larger impacts than others, with the highest negative impacts found in rural households in Tasla Azilal, Meknes Tafil, Fes-Boulemane, and Tanger-Tetouan. Mean impacts for rural households in these regions are 10 percent or more of consumption, indicating that there are sizable welfare losses among the poor in these specific regions. The adverse impact on rural poverty stems, in large part, from the fact that the losses to the net producers of cereals outweigh the gains to the net consumers among the poor. Thus, on balance, rural poverty rises. This contradicts the generalizations that have been made in the past--that the rural poor in Morocco tend to be net consumers of grain and hence gainers from trade reform. Yes, a majority of them are net consumers, but on balance, the welfare impacts on the rural poor are negative. 56 RAVALLION AND LOKSHIN These results lead to questions about the high level of aggregation common in past claims about welfare impacts of trade reform. Diverse impacts are found at given prereform consumption levels. This "horizontal" dispersion becomes more marked as the extent of reform (measured by the size of the tariff cut) increases. Indeed, it is estimated that all of the impact of complete deprotection of cereals on inequality is horizontal rather than vertical; in this study, the vertical impact on inequality was actually inequality reducing. For a modest reform of a 10 percent cut in tariffs, the vertical component dominates, although a large horizontal component is evident. It is clear from these results that in understanding the social impacts of this reform, one should not look solely at income poverty and income inequality as conventionally measured; instead, one needs to look at impacts along "horizontal" dimensions, at given income levels. This chapter has identified specific types of households whose consumption and production behavior makes them particularly vul- nerable. These results are suggestive of the targeting priorities for compensatory programs. The fact that this analysis also finds a large share of unexplained variance in impacts points to the limitations of targeting based on readily observable indicators, suggesting that self-targeting mechanisms may be needed. Annex: Measuring Welfare Impacts at the Household Level The approach used in this study is relatively standard, but it is still worth explaining the specifics on how the analysis in this chapter is done. Each household is assumed to have preferences over consump- tion and work effort represented by the utility function ui(qi ,Li), d where qdi is a vector of the quantities of commodities demanded by household i and Li is a vector of labor supplies by activity, including supply to the household's own production activities.13 The house- hold is assumed to be free to choose its preferred combinations qdi and Li subject to its budget constraint. The production activity owned by the household generates a profit i(pi) s max[piqi s s ci(qi)], where pi is a vector of supply prices, and ci(qi) is the s s s household-specific cost function.14 The indirect utility function of household i is given by: (2A.1) i[psi, pdi , wi] max [ui(qi , Li)pdi qdi d d wi Li i(psi)], (qi , Li) where pdi is the price vector for consumption and wi is the vector of wage rates. WINNERS AND LOSERS FROM TRADE REFORM IN MOROCCO 57 Taking the differential of expression (2A.1) and using the enve- lope property (whereby the welfare impacts in a neighborhood of an optimum can be evaluated by treating the quantity choices as given), the gain to household i (denoted gi) is given by the money metric of the change in utility: m (2A.2) gi dui s d n pij qij s s dpij pij qij d d dpij wkLiks dwk , i j=1 k 1 wk where is the marginal utility of income for household i--the mul- i tiplier on the budget constraint in equation (2A.2)--and Lsik is the household's "external" labor supply to activity k. (Notice that gains in earnings from labor used in own production are exactly matched by the higher cost of this input to own production.) The propor- tionate changes in prices are weighted by their corresponding expen- diture shares; the weight for the proportionate change in the jth sell- ing price is psijqsij, the revenue (selling value) from household production activities in sector j. Similarly, pijqij is the (negative) d d weight for demand price changes, and wk Lsik is the weight for changes in the wage rate for activity k. The term pijqij s s pdijpdij gives (to a first-order approximation) the welfare impact of an equi- proportionate increase in the price of commodity j. Equation (2A.2) is the key formula used to calculate the welfare impacts at household level, given the predicted price changes. Having estimated the impacts at household level, one can study how they vary with prereform welfare and determine what impact the reform has on poverty and inequality. One can also try to explain the differences in impacts in terms of observable characteristics of potential relevance to social protection policies. The formulation of the problem of measuring welfare impacts presented earlier allows utility and profit functions to vary between households at given prices. To explain the heterogeneity in measured welfare impacts, suppose instead that these functions vary with observed household characteristics. The indirect utility function becomes as follows: (2A.3) (psi, pdi , wi) (psi, pdi , wi, x1 , x2 ) i i i max[u(qdi , Li, x1 )pdi qdi ], i wi Li i where (2A.4) i (psi, x2 ) i max[psiqsi c(qsi, x2 )]. i Note that the characteristics that influence preferences over con- sumption (x1 ) are allowed to differ from those that influence i 58 RAVALLION AND LOKSHIN the profits from own production activities (x2 ). The gain from the i price changes induced by trade reform, as given by equation (2A.3), depends on the consumption, labor supply, and production choices of the household, which depend in turn on prices and char- acteristics, x1 and x2 . Generically, the gain can now be written as i i equation (2.5). Notes The authors are grateful to Touhami Abdelkhalek, Maurizio Bussolo, Jennie Litvack, Sherman Robinson, Hans Timmer, and seminar participants at the World Bank for comments on an earlier version of this chapter, and to Rachid Doukkali for help in using the results of his computable general equilibrium (CGE) analysis. This chapter was originally written as a back- ground paper to the report "Kingdom of Morocco: Poverty Report: Strengthening Policy by Identifying the Geographic Dimension of Poverty," Report No. 28223-MOR (World Bank 2004). 1. See the surveys by McColloch, Winters, and Cirera (2001) and Hertel and Reimer (2005). A number of chapters in part 2 of Bourguignon and Pereira da Silva (2003) describe alternative approaches using general equilibrium models. 2. Although it is not a subject of the analysis in this chapter, arguments are also made about adverse environmental impacts arising from the expan- sion of protected cereal production into marginal areas. It is claimed that scarce water resources have also been diverted into soft wheat production. For further discussion, see World Bank (2003). 3. There are many antecedents of this study's approach in the litera- tures on both tax reform and trade reform, but there are surprisingly few applications to point to in the ex ante assessment of actual reform propos- als. For another example, see Chen and Ravallion (2004). Hertel and Reimer (2005) provide a useful overview of the strengths and weaknesses of alternative approaches to assessing the welfare impacts of trade policies, including references to empirical examples for developing countries. 4. Antecedents to this type of decomposition can be found in the liter- ature on horizontal equity in taxation. In the context of assessing a tax system, Auerbach and Hassett (2002) show how changes in an index of social welfare can be decomposed into terms reflecting changes in the level and distribution of income, the burden and progressivity of the tax system, and a measure of the change in horizontal equity. 5. For further discussion of the MLD, see Bourguignon (1979) and Cowell (2000). MLD is a member of the general entropy class of inequality measures. 6. In addition to administering the tariffs on imported soft wheat, the government of Morocco buys, mills, and sells around 1 million tons of soft WINNERS AND LOSERS FROM TRADE REFORM IN MOROCCO 59 wheat in the form of low-grade flour that is sold on the open market to help consumers. 7. Rachid Doukkali kindly provided price predictions from the CGE model mapped into the categories of consumption and production iden- tified in the survey. The production revenues were calculated from the survey data by matching these consumption categories to the variables containing information about household production of the correspond- ing goods. 8. The survey's design and content are similar in most respects to the 1991 Living Standards Measurement Study (LSMS) for Morocco docu- mented in the World Bank's LSMS Web site at http://www.worldbank. org/lsms/. 9. Notice that income from the sale of meat is not recorded in these data. The most plausible explanation is that Moroccan farmers sell livestock to butchers or slaughterhouses (abattoirs) rather than selling meat as such. Fol- lowing conventional survey processing practices, livestock is treated as an asset, so that the proceeds from the sale of livestock are not treated as income. This is questionable. As a test, the main calculations were reworked using the survey data on the transaction in livestock and adding net sales into income. This made a negligible difference to the results. Further details are available from the authors. 10. These have been updated using the consumer price index. The poverty lines were 3,922 dirhams per year in urban areas and 3,037 in rural areas. See World Bank (2001) for details. 11. See Cleveland (1979). This is often referred to as LOWESS (locally weighted scatterplot smoothing). The authors used the LOWESS program in STATA. 12. The R2 for OLS regressions are 0.139 and 0.191 for Policy 4 and Policy 1, respectively. Using instead a probit model to correct for the non- linearity, the pseudo-R2s are 0.135 and 0.196. 13. The standard assumptions are made: that goods have positive marginal utilities while labor supplies have negative marginal utilities. 14. One can readily include input prices in this cost function; see Chen and Ravallion (2004) for a more general formulation. In the present context, this makes no difference to the subsequent analysis, so factor prices are subsumed in the cost function to simplify notation. References Abdelkhalek, Touhami. 2002. "De I'impact de la Libéralization du Marche Céréalier Marocain: Enseignements à Partir d'un Modèle de Comporte- ment des Ménages Rureaux." Critique Economique 7: 105­14. Atkinson, Anthony B. 1987. "On the Measurement of Poverty." Econo- metrica 55: 749­64. 60 RAVALLION AND LOKSHIN Auerbach, Alan J., and Kevin A. Hassett. 2002. "A New Measure of Horizontal Equity." American Economic Review 92 (4): 1116­125. Bourguignon, François. 1979. "Decomposable Inequality Measures." Econometrica 47: 901­20. Bourguignon, François, and Luiz A. Pereira da Silva, eds. 2003. The Impact of Economic Policies on Poverty and Income Distribution: Evaluation Techniques and Tools. Washington, DC: World Bank and Oxford University Press. Chen, Shaohua, and Martin Ravallion. 2004. "Welfare Impacts of Morocco's Accession to the WTO." World Bank Economic Review 18 (1): 29­58. Cleveland, William S. 1979. "Robust Locally Weighted Regression and Smoothing Scatter Plots." Journal of the American Statistical Associa- tion 74: 829­36. Cockburn, John. 2006. "Trade Liberalisation and Poverty in Nepal: A Computable General Equilibrium Micro Simulation Analysis." In Glob- alization and Poverty: Channels and Policies, eds. Maurizio Bussolo and Jeffery Round. London: Routledge. Cowell, Frank. 2000. "Measurement of Inequality." In Handbook of Income Distribution, eds. A. B. Atkinson and F. Bourguignon. Amsterdam: North-Holland. Doukkali, Rachid. 2003. "Etude de Effets de la Libéralisation des Céréales: Resultats des Simulations à L'Aide d'un Modèle Equilibre Général Calculable." Joint Report of the Ministry of Agriculture of Morocco and the World Bank, Rabat and Washington, DC. Hertel, Thomas W., and Jeffrey J. Reimer. 2005. "Predicting the Poverty Impacts of Trade Reform." Journal of International Trade & Economic Development 14 (December 4): 377­405. McCulloch, Neil, and L. Alan Winters, with Xavier Cirera. 2001. Trade Liberalization and Poverty: A Handbook. London: Centre for Economic Policy Research and Department for International Development (UK). Ravallion, Martin. 2001. "Growth, Inequality, and Poverty: Looking Beyond Averages." World Development 29 (11): 1803­815. ------. 2004. "Competing Concepts of Inequality in the Globalization Debate." Brookings Trade Forum, 2004, Brookings Institution Press, Washington, DC. Ravallion, Martin, and Dominique van de Walle. 1991. "The Impact of Food Pricing Reforms on Poverty: A Demand Consistent Welfare Analysis for Indonesia." Journal of Policy Modeling 13: 281­300. World Bank. 2001. Kingdom of Morocco: Poverty Update. Washington, DC: World Bank. ------. 2003. Kingdom of Morocco: Findings and Recommendations of the Cereals Working Group: A Critical Review. Washington, DC: World Bank. 3 Trade Options for Latin America: A Poverty Assessment Using a Top-Down Macro-Micro Modeling Framework Maurizio Bussolo, Jann Lay, Denis Medvedev, and Dominique van der Mensbrugghe During the past two decades, policy advice given to developing countries has emphasized greater market openness and better inte- gration into the global economy. This advice is based on two major assumptions: (1) that outward-oriented economies are more effi- cient, are less prone to resource waste, and hence grow faster; and (2) that faster income growth is beneficial for rich and poor alike, thereby contributing to poverty reduction in the developing world. Both assumptions have been challenged in the recent empirical literature. In particular, research on the second assumption has illus- trated that the effects of globalization in general, and trade liberaliza- tion in particular, on poverty are uncertain­­at least in the short to medium run. Instead, the emerging consensus seems to be that the distributional and poverty impacts of trade liberalization depend critically on the structure of initial protection, the pattern of liberal- ization, and a number of country characteristics, in particular the func- tioning of the labor market and the sectoral and skill composition of the workforce (see, for example, Winters, McCulloch, and McKay 2004; Harrison 2005; Hertel and Winters 2005). This uncertainty has 61 62 BUSSOLO, LAY, MEDVEDEV, AND VAN DER MENSBRUGGHE brought the issues of possible poverty impacts and the distribution of gains and losses both between and within countries to the center stage of negotiations on multilateral and regional trade reform. The collapse of the Doha Round of global trade talks can be interpreted as a conse- quence of the uncertainty regarding these distributional effects. The empirical literature has given rise to several approaches for assessing the ex ante poverty effects of a trade shock, most often by using some form of numerical simulation model (see, for example, Ianchovichina, Nicita, and Soloaga 2000; Harrison and others 2003; Hertel and Winters 2005). This chapter describes one such approach--a framework that links a global computable general equilibrium (CGE) model with household survey data--and relies on it to estimate the effects of multilateral and regional trade reforms on poverty in four major Latin American economies: Brazil, Chile, Colombia, and Mexico. This effort has two main objectives. First, the chapter demonstrates that a simple macro-micro framework-- despite a limited set of linkage variables and no behavioral responses by individuals in the survey--is highly superior to alternative meth- ods based on growth-poverty elasticities and to the earlier CGE- based analyses that used the representative household group (RHG) assumption. Second, the chapter assesses the distributional impact of trade reform for a group of countries in which the links between trade liberalization and poverty have been the subject of a large debate. Because earlier trade reform in the region has failed to bring about sizable poverty reduction, the question of whether future lib- eralization is likely to generate significant inroads in the fight against poverty remains relevant.1 The remainder of the chapter is organized as follows. The next section presents the macro and micro aspects of the chapter's methodological framework. This discussion is followed by an expla- nation of the general equilibrium results of the policy shocks and links them to the poverty outcomes of trade reform. The final sec- tion concludes by focusing on the policy implications of the chap- ter's findings and directions for future research. The Macro-Micro Framework: Linking a CGE Model to Household Surveys This section provides the details of the methodological approach of the chapter by focusing on three main areas (divided into three sub- sections). First, it discusses the pros and cons of the analytical and empirical framework adopted for this exercise. Second, it summarizes the main features of the macro (CGE) model. Third, it introduces the micro module and the household surveys used in the analysis. TRADE OPTIONS FOR LATIN AMERICA 63 Advantages and Drawbacks The methodological approach of this chapter can be best described as a two-step process. In the first step, a CGE model is used to create two trade reform scenarios and to evaluate the related factor and commodity price changes. In the second step, the general equilib- rium price changes are mapped to adjustments of real incomes of individual households in a micro simulation model. Thus, the pro- cedure accounts for two main transmission channels from trade reform to poverty: (1) movements in real prices of different factors and (2) changes in relative prices of different groups of consumption goods (McCulloch, Winters, and Cirera 2001). Variants of the method described here have been used in other case studies, and it is possible to formalize those variants in commonly used terms of welfare analysis (see Chen and Ravallion 2004; Hertel and others 2004; and chapter 2 in this volume). These formal presentations illustrate the major advantages and shortcomings of the current framework, which are only briefly summarized here. On the macro side, a CGE model has the advantage of being grounded in established trade and general equilibrium theories, of embedding enough data details so that they can be used to simulate realistic trade policy reforms, and of generating price effects that can be directly and unequivocally linked to these reforms. At the same time, CGE models are often criticized for imposing strong assumptions about the structure of an economy (for example, spe- cific functional forms and closure rules) and for the results being largely determined by base year conditions and the chosen values for various elasticities (which, even when econometrically estimated, are susceptible to Lucas' 1976 critique). The micro modeling approach of this chapter is often referred to as micro accounting. By generating a new counterfactual income distribution for the simulated trade scenarios, the micro module allows for a detailed analysis of the poverty and inequality changes induced by the aggregate trade policy shocks. Households and indi- viduals in the micro module are not allowed to change their optimal choices (in terms of demand for goods and factor supplies) and therefore the welfare impacts can be seen only as first-order approx- imations. Hence, the method should not be used for large shocks or for medium- to long-term analysis, that is, when such behavioral changes cannot be ignored. The advantages of the current method include its simplicity as well as the ability to evaluate the trade-induced poverty effects on specific groups of households. For example, as illustrated in this chapter, impact analyses can be conducted separately for rural and urban areas. Other criteria, such as farm and nonfarm households 64 BUSSOLO, LAY, MEDVEDEV, AND VAN DER MENSBRUGGHE or ethnic groups, may be chosen as well; the only requirement is that the production and consumption behavior of these groups are somehow correlated to the factors and commodities in the CGE model. Furthermore, the survey data allow for detailed distribu- tional analysis, such as an assessment of the importance of redistri- bution compared with growth effects. Finally, it is relatively easy to extend a CGE model by linking it to a household survey: it requires only some data handling and some straightforward estimation work. The LINKAGE General Equilibrium Model The CGE model used in this chapter is the World Bank's LINKAGE model, a relatively standard CGE model with many neoclassical fea- tures (for the full model specification, see van der Mensbrugghe 2005). It is based on the Global Trade Analysis Project (GTAP) Release 6.0 data set with a 2001 base year.2 All markets, including factor markets, clear through flexible prices, and the model exhibits constant returns to scale and perfect competition. The model is global, with a full accounting of bilateral trade flows, and its comparative static (as opposed to recursive dynamic) version has been imple- mented for the simulations described in this chapter. In each country, a single representative household earns income from skilled and unskilled labor, capital, and land. International factor mobility is not included, and with the exception of labor, intersectoral factor mobil- ity within countries has been limited, so model results should be inter- preted as short-term impacts. Labor markets are perfectly segmented by skill level, and unskilled labor is further segmented into farm and nonfarm activities. Therefore, unskilled workers are perfectly mobile within agriculture and nonagriculture, but these workers cannot switch employment between these segments. Skilled workers are per- fectly mobile throughout the national economies. For this application, the GTAP data have been aggregated to 18 countries/regions with an emphasis on the countries in the Western Hemisphere (table 3.1). Canada and the United States have been aggregated together, and most of the major countries in Latin America are identified separately. The remaining high-income countries are aggregated into Western Europe and Asia and Pacific, and the remain- ing developing countries fall into one of four broad regions: (1) East and South Asia, (2) Middle East and North Africa, (3) Europe and Central Asia, and (4) Sub-Saharan Africa (aggregated with a small residual). The sectoral concordance focuses on some of the major protected commodities, including agricultural and food products, textiles, clothing and footwear, metals, and motor vehicles and parts. TRADE OPTIONS FOR LATIN AMERICA 65 Table 3.1 LINKAGE Model: Regional and Sectoral Groups Country or group Sector Argentina Cereals Brazil Vegetables and fruits Chile Oil seeds Colombia Sugar Mexico Other crops Peru Livestock Uruguay Other natural resources República Bolivariana de Venezuela Fossil fuels Central America Cattle meats Caribbean Dairy products Rest of South America Other food, beverages, East Asia and Pacific and tobacco Europe and Central Asia Textiles Middle East and North Africa Wearing apparel South Asia Leather products Sub-Saharan Africa and the rest of world Basic manufactures Canada and United States Other manufacturing Western Europe and the European Metals Free Trade Area Motor vehicles and parts Rest of high-income countries Other equipment Electric and gas utilities Construction Services Source: Authors' compilation. Although services are highly protected in most markets, the levels of protection are hard to measure, and the GTAP data set has little infor- mation in this area--thus, the service sectors are highly aggregated. The Micro Accounting Framework The LINKAGE model described in the previous section captures some of the most important transmission channels from trade policy to poverty through changes in relative factor and goods prices. Welfare effects can be assessed only for the single representative household in each country; therefore, they provide little information on the distri- butional consequences of policy reform. This could be remedied by fitting a parametric distribution of changes in welfare (Adelman and Robinson 1978), using a household survey to increase the number of representative households or even to incorporate all sample house- holds into the CGE model (Rutherford, Tarr, and Shepotylo 2005), or by mapping the CGE changes in factor returns and commodity prices to the endowments and consumption patterns of each household in 66 BUSSOLO, LAY, MEDVEDEV, AND VAN DER MENSBRUGGHE the survey. The latter approach--a less computationally intensive alternative to the inclusion of the entire survey into the CGE model-- is the empirical strategy of this chapter. The micro data are connected to the LINKAGE results through the following variables: the average real wage in each of the four labor market segments (skilled/unskilled and agriculture/nonagriculture), the average nonagricultural real capital rent, the average agricultural (combined) capital and land rent, and the relative prices of food and nonfood commodities. The changes in real incomes of households are then calculated by applying the changes in factor returns to that household's labor and capital endowments, and deflating these with a cost-of-living index. This index is a weighted average of the new food and nonfood prices, with the weights calculated as each household's share of food consumption in the total consumption bundle.3 A final household income component--composed of all other sources of income, including pensions, public transfers, remittances (internal as well as international), and when available, autoconsumption--is assumed to be constant in real terms.4 The micro simulation module is implemented with the following household surveys: 2001 Pesquisa Nacional por Amostra de Domicílios (Brazil), 2000 Encuesta de Caracterización Socioe- conómica Nacional (Chile), 1997 Encuesta de Calidad de Vida (Colombia), and 2001 Encuesta Nacional Ingresos y Gastos de los Hogares (Mexico). These four countries were chosen to highlight var- ious aspects of poverty, inequality, and trade policy in Latin America.5 Together, these countries account for more than 60 percent of the population and almost 70 percent of GDP in Latin America, and therefore paint a fairly representative picture of the region. Some preliminary work on the surveys' raw data is needed to ensure a close match to the LINKAGE results; a perfect match is quite diffi- cult to achieve (see box 3.1 for more details). Labor incomes for the actively employed population (those 12 years of age and older) enter the analysis at the individual level. Workers are classified as skilled and unskilled based on level of education; if the information on level of education is not satisfactory, occupational variables are used instead. For wage workers, the entire income from employment is considered either skilled or unskilled (agricultural or nonagricultural) labor income. In contrast, income reported by the self-employed is assumed to have both a labor and a capital (plus land for agriculture) compo- nent. To separate these two components, a wage was imputed for each self-employed individual. This imputation is based on a wage equa- tion that is estimated separately for wage workers in agriculture and nonagricultural sectors, further differentiated by skill levels. The equations are simple Mincerian wage equations with log wage TRADE OPTIONS FOR LATIN AMERICA 67 earnings explained by education and age, the respective squared terms, as well as regional and sectoral dummies. The estimated coefficients of these wage equations are then used to impute a (hedonic) wage for the self-employed. The difference between reported income from self- employment and imputed wage is assumed to represent the capital component of self-employment income. For agricultural activities, the difference should be interpreted as mixed-factor income (from land and capital).6 In addition to capital income of self-employed individu- als, total household capital income includes all dividends, interest, and property rental income earned by household members. Box 3.1 Consistency Issues Combining macro and micro models implies working with different types of data sources, including national accounts and primary surveys, which are notoriously inconsistent. The scope of these inconsistencies is illustrated by Deaton (2004), who reports survey income to be, on aver- age, less than 60 percent of GDP. He discusses the reasons for these dis- crepancies and points to differences in definitions and differences in meeting those definitions, for example, in measuring production. In general, national accounts, in contrast to surveys, are more likely to capture larger transactions than smaller ones. Because these small trans- actions reflect the living standards of the poor, Deaton (2004) concludes that poverty can be measured only using household surveys. One of the challenges of the current exercise is to link macro- and micro-based data sets. In principle, one should think of the macro data as aggregations of the micro data (at least for the vari- ables that concern the household sector). Yet, in light of findings by Deaton and others (for example, Robilliard and Robinson 2003; Round 2003) this typically is not the case. Indeed, for the four countries considered in this chapter, large discrepancies are found between the (mainly) national-accounts-based social accounting matrices (SAM) and the household surveys. The extent of the prob- lem is highlighted by the factor shares from the SAM relative to those calculated from the surveys presented in the box table. Of course, the larger the initial discrepancies between these macro and aggregated micro variables, the larger the deviations between macro and micro results after a simulation. If factor shares between the survey and the SAM differ significantly, passing real factor prices from the CGE model to the household survey will provide real household income growth rates that are different from the CGE results. So what should be done with these discrepancies? Robilliard and Robinson (2003) propose to reconcile the two data sources by adjust- ing the weights in the survey data. Alternatively, factor markets and the (Box continues on the following page.) 68 BUSSOLO, LAY, MEDVEDEV, AND VAN DER MENSBRUGGHE household sector in the SAM could be rendered consistent with the household survey with regard to the aggregate link variables. The framework proposed in this chapter does not enforce consistency; however, this does not imply that no adjustments are made to the sur- vey data and the SAM. Yet these adjustments are partial and imply a number of discrete decisions by the analyst to reduce discrepancies to a "tolerable" level. What is tolerable remains subject to expert judg- ment, but so does the decision to put more trust in either national accounts or household survey data. In fact, there may be good reasons to prefer the survey data in some instances and national accounts data in others­­following the general guideline of surveys being the better source for data on small transac- tions. Eventually, the analyst's choices depend on the type of economic transactions prevalent in the country under consideration, the design of the survey, and the comparative quality of the two data sources. Finally, the problem is mitigated by the fact that SAMs increasingly incorporate information from household surveys­­at least with regard to the combination of CGE models with microeconomic data. Discrepancies between Household Surveys and CGE Model Data Item Brazil Chile Colombia Mexico Value added (percentage of total) in the base year SAM Unskilled labor 36 29 43 23 Skilled labor 21 12 18 10 Capital and land 43 59 38 68 Value added (percentage of total) in the household survey Unskilled labor 69 74 74 78 Skilled labor 19 11 11 12 Capital and land 13 14 15 9 Source: Authors' calculations. Note: SAM social accounting matrices. Poverty Effects of the Free Trade Area of the Americas and Multilateral Trade Liberalization This section investigates the poverty and income distribution effects of two different trade liberalization scenarios for Brazil, Chile, Colombia, and Mexico. The simulations are as follows: · A Free Trade Area of the Americas (FTAA) scenario, where tariffs and export subsidies among the Western Hemisphere coun- tries are eliminated. TRADE OPTIONS FOR LATIN AMERICA 69 · A full trade liberalization (FULLIB) scenario, where tariffs and export subsidies are eliminated for all countries. Domestic support (factors and indirect subsidies) is also eliminated in all countries. Thus, this scenario is a benchmark that represents the best results (in terms of maximizing efficiency gains) a country might expect to achieve. Each scenario has been modeled separately in a comparative static framework. Before implementing the liberalization scenarios, a series of presimulation shocks are imposed on the model to provide a better starting point for the solution and reflect the global trading environment more carefully. These shocks include, for example, the phase-out of the Multi-Fiber Arrangement quotas as well as China's accession to the World Trade Organization. Having obtained a solu- tion for this "presimulation," the model uses it as a starting point for each of the liberalization scenarios. At the end of the solution period, all model variables are reinitialized to the presimulation starting point, and a new liberalization scenario is solved. The analysis is carried out in three parts. First, this section dis- cusses the initial conditions in each of the four countries at both the macro and the micro levels. Second, these initial conditions, together with the nature of policy shocks, are used to explain the macro out- comes of trade reform. Third, the macro outcomes are mapped to the household surveys in accordance with the methodology des- cribed in the previous section. Economic Structure, Composition of Tariffs, and Household Income Sources In addition to the policy shock, the level and sectoral variability of the initial protection as well as the structural features of each econ- omy are key determinants of results in the LINKAGE model. Thus, this section begins with data in table 3.2 that show the import- weighted average tariffs by sector and by origin and destination markets. In general, tariffs levied against Western Hemisphere trading part- ners are somewhat lower than tariffs levied against non­Western Hemisphere exporters, mostly because of preferences granted under the region's many preferential trade agreements. This is particularly evident in the case of Mexico, whose import-weighted tariffs on mer- chandise trade with Western Hemisphere partners are slightly higher than a tenth of its tariffs on trade with countries outside the region. This also suggests that preferential liberalization in Mexico (mainly the North American Free Trade Agreement [NAFTA]) has been tak- ing place behind relatively high external barriers. Similarly, Brazil's 70 Table 3.2 Trade Protection by Origin, Destination, and Sector (import-weighted 2001 tariffs) Exporter Latin Canada and Not in the Western America and the United Western Sector/importer Hemisphere the Caribbean Brazil Chile Colombia Mexico States Hemisphere All sectors (excluding services) Western Hemisphere 1.9 2.4 5.0 5.4 2.1 0.6 1.6 3.9 Latin America and the Caribbean 5.2 6.7 7.7 9.4 2.8 10.8 4.6 11.6 Brazil 8.3 6.1 -- 8.0 10.3 16.4 9.8 10.7 Chile 6.4 6.3 7.0 -- 7.0 0.0 6.7 6.5 Colombia 8.5 7.4 13.4 14.2 -- 11.5 9.3 12.1 Mexico 1.8 11.4 16.5 2.6 2.8 -- 1.2 14.1 Canada and the United States 0.6 1.1 2.5 1.3 1.5 0.0 0.2 2.5 Not in the Western Hemisphere 7.3 11.4 14.2 3.3 7.7 5.8 6.3 3.9 Agriculture and food Western Hemisphere 5.3 3.9 6.5 6.4 2.2 1.4 6.5 6.6 Latin America and the Caribbean 8.4 6.5 6.4 14.1 3.3 14.0 10.0 16.5 Brazil 2.2 1.3 -- 13.0 18.6 15.9 10.8 12.1 Chile 6.9 6.9 7.0 -- 7.0 0.0 7.0 6.7 Colombia 9.7 6.6 14.8 17.1 0 16.4 13.6 13.9 Mexico 9.3 16.5 19.6 9.7 11.5 -- 8.6 24.5 Canada and the United States 3.2 2.2 6.5 1.4 1.6 0.3 4.1 4.4 Not in the Western Hemisphere 27.3 24.9 27.6 10.8 17.4 23.0 29.2 10.2 Agriculture Western Hemisphere 4.4 3.2 7.7 5.5 1.7 0.5 5.6 4.3 Latin America and the Caribbean 8.3 5.0 7.7 14.4 2.3 10.0 10.3 9.8 Brazil 0.8 0.5 -- 10.7 10.5 10.4 6.2 9.5 Chile 7.0 6.9 7.0 -- 7.0 0.0 7.0 6.7 Colombia 10.5 5.9 8.9 14.7 0 10.8 12.7 10.0 Mexico 10.7 12.7 17.9 16.8 3.2 -- 10.5 11.9 Canada and the United States 1.7 2.4 7.7 1.0 1.5 0.2 0.8 2.9 Not in the Western Hemisphere 32.4 31.6 29.4 11.1 20.6 13.0 33.0 12.5 Processed foods Western Hemisphere 6.1 4.7 5.5 7.2 3.4 2.4 7.2 7.4 Latin America and the Caribbean 8.6 7.6 5.9 13.9 3.7 15.2 9.6 19.3 Brazil 5.2 3.4 -- 15.3 20.5 17.1 14.4 13.8 Chile 6.8 6.8 7.0 -- 7.0 0.0 7.0 6.7 Colombia 8.8 6.9 17.4 19.3 -- 18.7 16.9 17.9 Mexico 7.7 19.4 25.5 2.9 13.0 -- 6.3 29.4 Canada and the United States 4.5 1.9 4.9 1.8 2.4 0.4 5.9 4.9 Not in the Western Hemisphere 21.5 17.8 25.4 10.6 3.4 31.4 24.6 9.0 Mining and natural resources Western Hemisphere 0.9 1.2 2.0 2.6 0.4 0.1 0.4 0.5 Latin America and the Caribbean 3.3 3.7 3.0 4.7 2.3 1.9 2.4 2.1 Brazil 0.7 0.7 -- 3.4 0.4 2.4 0.7 0.3 Chile 7.0 7.0 7.0 -- 6.7 0.0 6.7 6.7 Colombia 4.7 3.6 5.2 5.0 -- 4.0 6.8 7.2 Mexico 1.1 5.4 12.2 0.0 1.1 -- 0.6 9.4 Canada and the United States 0.1 0.2 1.5 0.5 0.2 0.0 0.0 0.2 71 Not in the Western Hemisphere 3.7 4.9 1.5 0.6 0.1 8.0 1.8 1.6 (Continued on the following page) 72 Table 3.2 (Continued) Exporter Latin Canada and Not in the Western America and the United Western Sector/importer Hemisphere the Caribbean Brazil Chile Colombia Mexico States Hemisphere Manufacturing Western Hemisphere 1.9 2.6 5.1 5.8 3.2 0.7 1.5 4.1 Latin America and the Caribbean 5.1 7.5 7.9 9.3 2.9 11.7 4.3 12.2 Brazil 9.9 9.2 -- 9.4 14.5 16.6 10.1 12.3 Chile 6.3 5.9 7.0 -- 7.0 0.0 6.7 6.5 Colombia 8.3 7.6 13.7 14.3 -- 11.6 8.9 12.2 Mexico 1.4 11.7 16.5 0.5 3.0 -- 0.7 14.2 Canada and the United States 0.6 1.1 2.3 1.5 4.0 0.0 0.2 2.7 Not in the Western Hemisphere 4.9 7.1 11.1 3.3 2.0 4.7 4.6 3.8 Source: Authors' calculations from the GTAP data. Note: -- not available. TRADE OPTIONS FOR LATIN AMERICA 73 tariffs on imports from other Latin American countries are relatively low, reflecting its participation in MERCOSUR (the Southern Com- mon Market). It is also noteworthy that for Brazil tariffs on agricul- ture and food commodities are much lower than on manufactured goods, while the opposite is true for Mexico. The tariff structure in Chile and Colombia is largely uniform across different sectors, and the level of duties in the former country is, on average, lower than in other Latin American countries. The tariff patterns shown in table 3.2 are also reflected in the structural features of the four economies, which are summarized in table 3.3. For example, exports and imports figure most promi- nently in the GDP of Chile and Mexico, the two countries with the lower average tariffs. Table 3.3 provides important sector detail by summarizing import intensities (measured as the ratio of sectoral imports over sectoral GDP), export intensities (ratios of exports to GDP), and factor intensities (measured as the factor's percentage contribution to total sectoral value added) for each production sec- tor in the LINKAGE model. According to the data in table 3.3, in Brazil the highest dependency on imports is in the capital goods sectors (which have elevated protection rates) and fossil fuels. There- fore, these sectors are likely to be most affected by import competi- tion following trade reform. Conversely, Brazil's export strength is concentrated in export crops, food processing, natural resources, and some manufacturing. As expected, export-oriented sectors (within agriculture) require intensive use of land and unskilled labor, and light manufacturing requires unskilled labor. Apart from the service sectors, the protected import competing sectors are major employ- ers of skilled labor--in conjunction with unskilled workers. The poverty consequences of trade reform are determined to a large extent by the factor endowments and consumption patterns of households in the neighborhood of the poverty line. This information is summarized in table 3.4, which shows the contributions of labor, capital, and transfers to total household income for rural and urban households above and below the poverty line. Several important pat- terns are captured in this table. First, although the rural/urban classi- fication is in no way synonymous with the farm/nonfarm distinction, the majority of incomes in rural areas are earned through farm activ- ities, whereas urban dwellers rely mainly on nonfarm earnings. Sec- ond, poor households derive a large share of their income from unskilled labor and have almost no skill endowments. Third, no clear pattern in the distribution of transfer income emerges by regions or income levels, most likely because of the differences in definitions across countries. In some cases, transfer income may be mainly auto- consumption, more likely to be observed among the rural poor, while 74 BUSSOLO, LAY, MEDVEDEV, AND VAN DER MENSBRUGGHE Table 3.3 Economic Structure for Brazil, Chile, Colombia, and Mexico (percent) Brazil Chile Imports/ Exports/ Un- Imports/ Exports/ Un- Sector GDP GDP GDP skilled Skilled K L GDP GDP GDP skilled Skilled K L Cereals 1 53 30 24 1 74 1 26 15 46 1 53 Vegetables and fruits 0 38 42 23 1 76 5 1 55 46 1 53 Oil seeds 1 5 92 23 1 76 0 308 68 47 0 53 Sugar 1 1 63 35 4 61 0 14 0 43 2 56 Other crops 1 7 52 23 1 76 0 51 101 46 1 53 Livestock 2 1 3 23 1 76 2 2 3 46 1 53 Other natural resources 1 21 152 43 5 52 5 2 91 24 4 73 Fossil fuels 1 142 42 27 5 68 0 1,427 141 13 3 84 Cattle meat 0 5 72 37 6 56 0 98 14 45 8 47 Dairy products 0 10 2 28 5 68 0 15 23 29 4 67 Other food 2 12 63 38 6 55 6 16 91 32 6 62 Textiles 1 41 36 34 5 61 1 82 18 39 6 55 Wearing apparel 1 9 11 79 13 8 1 133 15 43 7 50 Leather products 0 20 165 53 9 39 0 251 43 45 7 48 Basic manufactures 4 20 42 58 10 32 6 43 87 31 5 64 Other manufacturing 4 72 29 38 6 56 3 148 74 36 9 55 Metals 1 64 179 27 4 68 4 22 227 27 5 68 Motor vehicles and parts 1 119 145 28 5 67 0 455 51 32 7 60 Other equipment 4 131 62 40 7 53 1 980 88 44 12 44 Electric and gas utilities 3 16 0 42 13 45 4 1 0 19 9 73 Construction 9 0 0 19 4 76 6 0 0 44 8 48 Services 63 5 3 38 30 33 52 11 9 24 19 57 Total 100 17 16 36 21 43 100 34 37 29 12 59 Agriculture 5 11 39 25 1 74 9 8 37 46 1 53 Mining 2 97 83 33 5 62 6 71 94 23 4 73 Light manufacturing 5 15 56 42 7 51 9 42 67 34 6 60 Other manufacturing 14 78 58 43 7 50 15 134 123 32 7 61 Services 75 5 3 36 26 38 62 10 8 26 17 58 Source: Authors' calculations from the Global Trade Analysis Project data. Note: Unskilled and Skilled labor categories by education level; K L payments to capital and land. TRADE OPTIONS FOR LATIN AMERICA 75 Colombia Mexico Imports/ Exports/ Un- Imports/ Exports/ Un- GDP GDP GDP skilled Skilled K L GDP GDP GDP skilled Skilled K L 1 53 0 50 2 48 1 48 3 46 1 53 3 6 22 46 1 53 1 8 37 46 1 53 0 40 1 46 1 53 0 438 11 46 1 53 1 4 33 33 4 63 1 2 4 27 2 71 2 8 77 46 1 53 1 14 8 46 1 53 4 1 2 46 1 53 1 13 9 46 1 53 1 7 2 74 6 20 2 7 5 21 2 76 4 2 134 18 3 79 2 26 86 8 2 90 1 5 3 74 12 14 1 24 1 16 3 81 1 15 11 32 5 63 0 50 4 14 2 83 4 23 21 45 7 47 6 11 11 14 3 83 1 126 75 56 8 36 2 58 51 26 4 70 1 12 85 61 9 30 1 32 101 23 4 73 0 53 88 72 11 17 1 26 18 29 5 66 3 47 53 49 8 43 6 42 33 19 4 77 3 124 67 38 8 54 5 82 36 19 5 77 1 111 80 56 9 35 2 53 30 22 4 74 0 472 190 65 13 22 2 129 197 24 6 70 1 636 100 62 15 23 7 149 201 25 7 68 2 0 1 18 8 73 0 3 1 17 8 75 5 0 0 41 7 52 4 0 2 54 10 35 61 6 4 44 27 30 54 5 4 20 14 66 100 19 18 43 18 38 100 28 31 23 10 68 12 10 24 46 1 53 5 20 15 44 1 55 4 2 110 28 4 69 4 19 55 13 2 85 7 29 35 51 8 42 11 23 26 18 3 79 8 158 70 47 9 44 21 96 105 22 5 73 69 5 4 43 25 33 59 5 4 22 14 64 Table 3.4 Household Incomes by Source, Segment, and Poverty Status 76 (percentage of total) Farm Nonfarm Country/household Skilled Unskilled Capital Skilled Unskilled Capital Transfer Food Brazil Total 0 9 2 15 46 8 20 30 Poor-urban 0 12 1 4 64 7 12 28 Poor-rural 1 63 8 1 16 4 7 40 Nonpoor-urban 0 2 1 21 49 9 18 28 Nonpoor-rural 3 35 11 5 21 3 23 40 Chile Total 0 18 3 8 34 7 31 38 Poor-urban 0 9 0 3 52 2 35 32 Poor-rural 0 41 1 1 22 0 36 46 Nonpoor-urban 0 3 1 18 43 12 23 32 Nonpoor-rural 1 36 7 3 24 4 25 46 Colombia Total 0 14 4 8 39 7 27 46 Poor-urban 0 4 1 4 63 8 21 45 Poor-rural 0 36 9 1 18 1 35 57 Nonpoor-urban 0 1 1 20 46 14 18 30 Nonpoor-rural 0 23 12 10 26 6 22 48 Mexico Total 0 10 2 8 41 4 35 37 Poor-urban 0 6 0 3 60 2 29 41 Poor-rural 0 32 1 1 20 1 45 48 Nonpoor-urban 0 1 0 18 45 7 29 27 Nonpoor-rural 0 16 5 5 32 5 37 36 Source: Authors' calculations from household surveys. Note: Transfers include autoconsumption; and the last column, Food, represents the share of food items in the consumption basket. TRADE OPTIONS FOR LATIN AMERICA 77 in other cases, it may be mainly pensions, found primarily among the better-off urban residents. The last column of table 3.4 shows the share of income that each household group spends on food products. For Brazil and Chile, these shares are the same for poor and nonpoor households because of the lack of detailed expenditure data. Data from Colombia and Mexico, however, show that poor households devote a much larger share of their income to food expenditures. Furthermore, in all four countries, food expenditures are significantly higher among rural households because of lower incomes in these rural areas. This suggests that the poverty outcomes of trade reform, especially in the rural areas, are likely to be particularly sensitive to changes in food prices. Macroeconomic Results of Trade Reform Because the model is solved in a comparative static mode and factor mobility is restrained, the adjustment process after the trade liberal- ization shock is carried out almost entirely through price changes. To trace the links between these price changes and the liberalization scenario, this section begins by examining the pattern of sectoral adjustment in table 3.5. First, consider the FTAA scenario for Brazil, Chile, and Mexico. For Western Hemisphere partners, Brazil's tariffs are significantly higher against imports of manufactured goods, whereas the con- verse is true for Mexico, and Chile's tariffs are largely uniform across sectors. Tariff reductions result in increased import inflows and a reallocation of resources away from sectors that face more intense competition from foreign producers. Consequently, in Brazil, manufacturing imports rise the most; in Mexico, the agricultural sector experiences the largest growth in import volumes. In Chile, the effect is largely the same across agriculture and manufacturing. This pattern of increased import inflows is also borne out in the behavior of prices of domestic goods sold locally. In Brazil, the largest price declines are observed in manufacturing sectors; in Mexico, the prices of agricultural commodities register the greatest changes. This behav- ior testifies to the market share losses experienced by domestic pro- ducers as a result of tariff reform. Turning to the exports side, notice that the sectors with the largest increase in import volumes are often also the sectors with the greatest increase in exports volumes. This reflects greater incentives to export through lower domestic producer prices. The export response varies across sectors and generally is linked to the structure of each coun- try's comparative advantage, revealed partly in the sectoral exports- to-GDP ratio in table 3.3. Thus, for Brazil, which exports more than 78 BUSSOLO, LAY, MEDVEDEV, AND VAN DER MENSBRUGGHE Table 3.5 Sectoral Adjustments (percentage change with respect to initial levels) Exports Imports Production volume volume volume Country/sector FTAA FULLIB FTAA FULLIB FTAA FULLIB Brazil Total 12.4 21.5 11.6 25.5 0.1 0.2 Agriculture 2.1 17.8 2.7 46.2 0.2 3.1 Mining 2.9 6.0 2.0 7.8 0.1 1.1 Light manufacturing 9.3 40.9 18.9 69.2 0.5 1.0 Other manufacturing 21.6 26.0 16.0 30.4 0.3 2.1 Services 1.6 5.4 0.8 4.0 0.1 0.2 Chile Total 5.9 7.2 7.9 10.1 0.1 0.0 Agriculture 3.5 3.4 17.8 6.7 0.1 0.8 Mining 4.1 5.3 7.3 9.2 0.3 0.3 Light manufacturing 13.6 17.1 16.4 19.6 0.7 0.5 Other manufacturing 8.0 7.4 7.8 10.6 0.3 1.0 Services 1.9 2.2 1.7 2.9 0.1 0.2 Colombia Total 8.5 10.7 7.0 11.6 0.1 0.4 Agriculture 9.5 39.6 19.3 36.7 0.1 0.6 Mining 1.8 1.1 13.3 24.9 0.5 0.1 Light manufacturing 25.5 12.2 25.2 35.0 0.7 1.9 Other manufacturing 7.8 5.6 5.4 8.9 1.3 3.8 Services 6.2 8.4 3.4 3.3 0.0 0.4 Mexico Total 3.3 11.8 3.8 14.8 0.0 0.0 Agriculture 17.0 32.9 18.9 37.1 0.2 0.2 Mining 0.8 1.7 3.1 9.0 0.1 0.9 Light manufacturing 6.2 69.8 10.3 52.0 0.0 1.6 Other manufacturing 3.0 6.6 2.8 10.8 0.3 0.7 Services 0.2 3.0 0.1 2.5 0.1 0.0 Source: Authors' calculations. Note: FTAA Free Trade Area of the Americas; FULLIB full trade liberalization. half of its production of other manufactures, the increase in exports volume more than offsets the decrease in domestic prices, leading to an overall expansion in sectoral production. Conversely, Mexico, which exports a relatively low share of its production of agricultural goods, is unable to compensate for the fall in domestic demand with rising export sales and consequently, experiences a contrac- tion in that sector. In general, the sectoral adjustments brought about by FTAA reform are quite modest. This is likely because of the already low tariffs on most imports from Western Hemisphere trading partners. TRADE OPTIONS FOR LATIN AMERICA 79 The pattern of adjustment in the FULLIB scenario differs some- what from the changes likely to take place with trade reform under the FTAA. Relative to the latter scenario, Brazil, Colombia, and to a lesser extent, Chile orient their production structure much more toward agriculture and away from heavy manufactur- ing. This pattern reflects the comparative advantage of these coun- tries in producing agricultural goods and the high levels of protection their agricultural exports face among non­Western Hemisphere trading partners. For Mexico, the FTAA scenario fur- ther reinforces the regional bias of the country's production struc- ture and therefore global trade reform implies a quite different adjustment pattern for exports, imports, and domestic output. For example, in contrast to the FTAA, full liberalization is likely to result in a much larger increase in exports and production of processed food, but a contraction in the output of textiles and leather products (largely because of competition from East Asian countries). The adjustments in main factor and consumption prices, as well as real consumption changes, are summarized in table 3.6. In all simula- tions, Brazil and Chile experience an increase in the payments to unskilled farm workers and a decrease in the skilled/unskilled wage gap for the agricultural sector. In Colombia, unskilled agricultural labor experiences gains only in the case of multilateral liberalization; in Mexico, unskilled agricultural workers lose in all cases. These results are consistent with the structure of individual simulations, and differences across countries reflect their particular patterns of protec- tion and comparative advantage. For example, Brazil's comparative advantage lies in agricultural products and it tends to protect its man- ufacturing sectors more than the others. Consequently, trade reform leads to an expansion in production of agricultural goods, and because labor is not able to move between agricultural and nonagri- cultural activities, wages in the farm sectors increase significantly. Conversely, domestic prices decline in the previously protected man- ufacturing sectors, which reduces production and wages in that sec- tor. The same effect takes place in Mexico, only with the sectoral roles reversed--because Mexico's agriculture is relatively more pro- tected, farm wages suffer relative to nonfarm earnings. The case of Mexico is particularly interesting because, in addition to the afore- mentioned price effect, another phenomenon takes place--that is, preference erosion. Within each of the liberalization scenarios above, Mexico loses the significant margin of preference it enjoys as a NAFTA member with respect to other Latin American countries. This is one of the reasons why Mexico's nonfarm wages do not experience a significant increase--it already enjoys virtually tariff-free access to 80 Table 3.6 Price (Factors, Consumption Aggregates) and Real Income Changes (percentage change) Brazil Chile Colombia Mexico Indicator FTAA FULLIB FTAA FULLIB FTAA FULLIB FTAA FULLIB Nonfarm skilled wage 0.30 0.52 1.32 1.47 1.53 2.11 0.04 0.64 Nonfarm unskilled wage 0.45 0.30 1.82 1.29 1.41 2.97 0.20 0.72 Nonfarm capital rent 1.57 4.42 1.98 2.02 0.46 0.96 0.32 1.84 Farm skilled wage 0.30 0.52 1.32 1.47 1.53 2.11 0.04 0.64 Farm unskilled wage 3.13 51.44 3.34 20.21 2.34 19.76 6.40 3.26 Farm capital and land rent 1.70 21.52 1.62 5.87 0.61 5.63 1.16 1.27 Nonfarm skill/unskilled wage gap 0.15 0.82 0.49 0.18 0.12 0.88 0.15 0.08 Farm skill/unskilled wage gap 2.75 33.63 1.95 15.59 0.83 18.26 6.88 4.03 Food prices 0.55 10.97 1.06 4.84 2.75 2.10 1.54 6.51 Nonfood prices 0.09 0.33 0.23 0.64 1.70 2.91 0.08 0.76 Consumer price index (CPI) 0.16 1.28 0.44 0.71 1.99 1.76 0.42 0.76 Real consumption 0.11 1.24 0.39 0.51 0.30 0.23 0.09 0.23 Source: Authors' calculations. Note: FTAA Free Trade Area of the Americas; FULLIB full trade liberalization. TRADE OPTIONS FOR LATIN AMERICA 81 its biggest market, the United States and Canada, and further regional or global liberalization serves only to open the North American mar- kets for its competitors. The bottom rows of table 3.6 report the percentage changes in food and nonfood prices, the consumer price index (CPI), and real consumption. With rising farm wages in Brazil and Chile, food prices increase in both FTAA and FULLIB scenarios (despite down- ward pressure from cheaper imports) and drive the increase in the overall CPI. In Colombia and Mexico, food prices fall in the FTAA simulation as factor returns in agriculture decline. This lowers domestic production costs and combined with increased access to cheaper imports, contributes to the decrease in the overall CPI. Con- versely, farm factor prices increase in these countries in the FULLIB scenario, which is sufficient to change the sign of changes in the aggregate CPI for Mexico, but not for Colombia. With the exception of Colombia under the FTAA scenario, all coun- tries experience positive changes in consumption volumes. Because the FULLIB scenario includes larger tariff cuts and represents the "first- best" world trade scenario, consumption gains in this simulation are larger than under the FTAA and are always positive. The aggregate gains in either scenario are rather small, which can be attributed to several model features, including the following: limited factor mobil- ity, no changes in capital accumulation, and a fixed fiscal closure, where tariff revenue losses are compensated by increases in direct taxes on household income. These aggregate gains, however, are much more indicative of changes at the top of the income distribution than at the bottom, because richer households have a larger weight in the expen- diture pattern of the single representative household in the LINKAGE model. To determine the distribution of these gains in the population-- and consequently, their repercussions for poverty--the next section translates the changes in macro aggregates into welfare effects at the household level. Poverty and Income Distribution Results of Trade Reform The initial poverty conditions of the four countries under analysis, as well as the estimated poverty effects of the two liberalization scenarios, are shown in table 3.7. The initial poverty conditions in Brazil, Chile, Colombia, and Mexico are fairly typical for developing countries. Poverty especially affects rural areas, and the rural poor are more likely to be further away from the poverty lines than the urban poor (as shown by the poverty gap and squared poverty gap statistics, which account for inequality among the poor, with squared poverty gap being more sensitive to changes in inequality).7 82 Table 3.7 Initial Poverty Levels and Percentage Changes Resulting from Trade Reforms Brazil Chile Colombia Mexico Indicator H PG P2 H PG P2 H PG P2 H PG P2 Initial poverty level (percent) All 23.5 9.6 5.3 21.0 7.3 3.8 60.7 32.6 21.8 50.2 20.7 11.2 Urban 19.5 7.6 4.0 20.4 7.1 3.7 53.7 27.1 17.6 43.7 16.6 8.5 Rural 44.7 20.3 11.8 24.6 8.5 4.4 79.1 46.7 32.7 69.3 32.7 19.2 FTAA (percentage change from initial levels) All 0.64 1.32 1.70 1.82 2.04 1.85 0.21 0.51 0.54 0.32 0.47 1.04 Urban 0.21 0.85 1.07 1.56 1.92 1.73 0.16 0.66 0.72 0.61 0.48 0.25 Rural 1.64 2.24 2.82 3.14 2.68 2.45 0.29 0.28 0.30 0.24 1.88 2.71 FULLIB (percentage change from initial levels) All 5.97 10.77 13.77 4.42 4.53 4.01 0.12 1.18 1.78 1.17 2.05 2.72 Urban 0.69 2.74 4.08 2.64 3.06 2.70 1.12 0.95 0.93 1.14 1.66 2.15 Rural 18.05 26.46 30.99 13.40 11.92 10.59 2.31 4.38 5.55 1.23 2.62 3.46 Source: Authors' calculations. Note: FTAA Free Trade Area of the Americas; FULLIB full trade liberalization; H the poverty headcount; PG the poverty gap; P2 the squared poverty gap: FGT(0), FGT(1), and FGT(2), respectively. All are estimated at the national poverty line for each country. TRADE OPTIONS FOR LATIN AMERICA 83 In Brazil and Chile, aggregate poverty declines across all scenarios. The reduction in rural and urban headcounts is in line with the factor price changes in table 3.6 and the structure of household earnings shown in table 3.4. Therefore, the rural poor benefit much more from trade reform than the urban poor. The strong decrease in the squared poverty gap shows that the poorest of the poor gain the most under both scenarios--and much more under FULLIB than under the FTAA. In Colombia, the aggregate poverty effects of all simulations are quite small, but again, the results are consistent with factor and consump- tion price changes given by the macro model (in all cases, however, the decline in rural poverty is significant).8 In Mexico, poverty decreases under the FTAA, but the reduction is entirely accounted for by a nationwide distributional shift caused by a widening rural-urban gap. The same pattern takes place under the FULLIB scenario, with the urban poor losing less than the rural poor. The most interesting dynamics are observed when comparing the impact of multilateral and regional liberalization across coun- tries. For Brazil and Chile, multilateral liberalization is unequivo- cally superior to regional scenarios, and the order of magnitude of poverty reductions is proportional to the scale of tariff reductions. In Colombia, the difference between scenarios can be explained by the virtually unchanging rural-urban gap under FTAA and a major closing of this gap under the full trade liberalization scenario. As before, this is consistent with the factor price changes in table 3.6 and the endowments of poor households shown in table 3.4. For Mexico, only the regional liberalization scenarios are poverty reducing, but multilateral liberalization actually increases poverty. The reason for this result is the previously mentioned preference erosion--with regional liberalization, Mexico only loses its prefer- ence margin relative to other Latin American and Caribbean coun- tries. With multilateral liberalization, however, it is now forced to compete on equal grounds with all U.S. and Canadian trading part- ners. Perhaps the most persuasive way to highlight the poverty and dis- tributional effects of trade liberalization is to compare the poverty results obtained by a distribution-neutral growth with those gener- ated by the micro-accounting methodology of this chapter. This com- parison is summarized in table 3.8 in terms of poverty elasticities. The growth elasticity in table 3.8 is calculated by applying the same (distribution-neutral) growth rate to the incomes of every household in the survey. Thus, this elasticity represents the percentage change in the poverty headcount that corresponds to a 1 percentage point increase in the growth rate in the average per capita income in a given country. The trade elasticity, on the other hand, considers changes in 84 BUSSOLO, LAY, MEDVEDEV, AND VAN DER MENSBRUGGHE Table 3.8 Income Elasticity of Poverty Headcount Growth Trade elasticity scenario Country elasticity FTAA FULLIB Brazil 1.0 1.5 12.7 Chile 1.6 1.8 4.1 Colombia 0.5 0.3 0.7 Mexico 1.0 0.3 Source: Authors' calculations. Note: FTAA Free Trade Area of the Americas; FULLIB full trade liberaliza- tion. Mexico's full liberalization trade elasticity is not reported because average growth rates are negative. the shape of the income distribution and is calculated using actual heterogeneous income growth rates for each household.9 Comparing the two sets of elasticity illustrates three major points: (1) both trade and growth elasticities vary across countries because of differences in initial conditions, such as income inequality; (2) trade elasticities are almost always larger than growth elasticities; (3) trade elasticities are higher when reform is more extensive. The initial level of inequality, that is, the shape of the initial income distribution, and the level of the poverty line determine how many individuals escape poverty with a 1 percent increase in aver- age incomes. A common but important feature of the elasticities is that they are all rather small, mainly because of the fact that Latin America is a region with high levels of inequality. Cross-country empirical evidence convincingly shows that, all other things being equal, inequality reduces the growth elasticity of poverty reduction (see Ravallion 2001; Bourguignon 2002). Colombia, which shows the smallest growth elasticity in table 3.8, is the country with the highest initial level of inequality. This means that many of its poor people are still quite distant from the poverty line, and a lot of growth is needed to lift them out of poverty.10 In fact, summary sta- tistics of the income distribution around the poverty line such as the poverty gap and the squared poverty gap are much more relevant for determining the extent of poverty reduction than aggregate inequality measures. The Gini coefficient of Chile is six points below that of Colombia; but even more important, the former's poverty gap and squared poverty gap are much lower than the latter's gaps (table 3.7). This means that the growth of average incomes can be fairly effective in reducing poverty in Chile. Intermediate situations are observed for Mexico and Brazil. If trade liberalization raises the prices of factors owned by the poor or reduces the prices of the goods consumed by them more than the average, the growth elasticity will underestimate the pro-poor TRADE OPTIONS FOR LATIN AMERICA 85 potential of reform. The results of this chapter show that this is the case for Latin America: trade liberalization induces pro-poor growth in most of the scenarios examined here, as evidenced by the fact that trade elasticities are generally higher than growth elasticities. Trade reform generates large poverty reductions for the initially poorer households, namely, those earning large shares of their incomes from agricultural sectors. Much lower trade-induced reductions are observed for households that depend on nonagricultural incomes. These poverty reductions show that trade reform, if implemented suc- cessfully, can have a significant pro-poor distributional effect in addi- tion to the positive effect that it has on average incomes. The differ- ence between the growth elasticity in the first column of table 3.8 and the trade elasticities in the right-hand columns represents the equaliz- ing effect of trade reform--that is, the fact that trade liberalization benefits the poor more than the rich (although the growth and the inequality effects are not strictly additive). Table 3.8 also shows that the degree of poverty reduction is positively correlated with the scope of trade liberalization, with deeper reforms generally bringing about larger reductions in poverty. This larger reduction occurs because the poor gain not only when their own countries liberalize, but also when other nations reduce their trade barriers. Thus, the scope and design of trade reforms determine their distributional outcomes. Conclusion This chapter has illustrated an application of a relatively simple macro-micro model to the analysis of trade reforms and poverty in Latin America. Several important lessons can be learned from this exercise. First, the chapter provides empirical evidence that trade lib- eralization, at both the regional and multilateral levels, can be poverty reducing in Latin America. The strength and even sign of the impact depends critically on the design of reforms and the initial situation of a country. Second, the chapter clearly shows that a policy approach that considers only the effects of growth on poverty is misleading when analyzing trade reform shocks. In most cases, the aggregate poverty changes that are calculated based on such an approach are substantially different from the results obtained under a full survey approach that accounts for changes in the distribution. Third, the chapter has illustrated the advantages and flexibility of macro-micro models, including their ability to focus on specific groups of house- holds by income level, location, or any other characteristic. The empirical analysis in this chapter has shown that the poverty impact of trade reform can vary greatly depending on the type of lib- eralization and the initial conditions of a country. The results point to 86 BUSSOLO, LAY, MEDVEDEV, AND VAN DER MENSBRUGGHE large declines in the poverty headcount in Brazil and Chile following both the FTAA and a full global trade reform. In both cases, rural poverty declines the most, and some of the largest income gains are observed among the poorest of the poor. In Colombia, the poverty reduction potential of trade reform is much more modest, partially because of conflicting poverty trends in rural and urban areas. Although both urban poverty and rural poverty are likely to decline marginally following the implementation of the FTAA, rural poverty falls but urban poverty rises as a result of full trade reform. Finally, in the case of Mexico, rural poverty could rise under both reform sce- narios, although the increase in rural poverty would be offset by the decline in urban poverty under the FTAA. The results of this chapter convincingly show that the differences in rural and urban real income growth induce important distribu- tional shifts that must be considered when judging the poverty impact of trade reform. As shown in the calculations of growth and trade elasticities, the distributional consequences of reform can either reinforce or counteract changes in average incomes brought about by trade liberalization. Alternatively, even if the overall impact on poverty is not too large, its dispersion across households (because of their heterogeneity in terms of factor endowments and consumption patterns) is significant, and recognizing this distribu- tion may help when designing compensatory policies. Thus, macro- micro approaches should be a preferred vehicle for comprehensive analysis of the poverty effects of policy because of their ability to take into account both the endowments of the poor and the relative factor price changes induced by trade liberalization. The particular macro-micro methodology used in this chapter offers a number of important advantages. First, the current approach represents a reasonable approximation to changes in household wel- fare in the short term. Second, the method is not too computationally intensive and does not require extensive efforts to reconcile the incon- sistencies between the macro and micro data. Third, the methodology allows the researcher to exploit the full heterogeneity of the house- hold survey and focus on groups that are not explicitly represented in the macro model, such as rural and urban households. Conversely, the empirical strategy of this chapter has some potential drawbacks. For example, the limited detail of the macro model allows prices to be determined only at the national level, rather than regional or even local levels. In other words, the pass- through effects of trade shocks to prices are the same in border areas and remote regions. In addition, trade liberalization usually triggers more than just factor price changes, especially in the longer run. In particular, liberalization is likely to induce different types of labor market switching. People migrate from rural to urban areas, TRADE OPTIONS FOR LATIN AMERICA 87 and they move between sectors or different occupational categories (from self-employment into wage-employment or vice versa). The simple household survey-based approach used in this chapter does not allow for labor market switching. And this drawback becomes more serious as longer time periods are considered--and as indi- viduals become more mobile across geographic areas, sectors, and occupations. Other long-run implications of trade reform--for example, potential effects on capital accumulation, productivity, and stability of the macro environment--are not considered in this analysis. Notes The authors thank Hans Timmer for his comments on this chapter and Abhijeet Dwivedi for his excellent research assistance. An earlier version of this chapter was presented as a paper at the Second CEPII-IDB Conference: Economic Implications of the Doha Development Agenda for Latin America and the Caribbean, October 6­7, 2003, Washington, DC. 1. Although contrary to policy makers' expectations, postreform growth and poverty reduction have been disappointing, a number of stud- ies caution that without reforms the situation would have been even worse. See, for example, ECLAC (1996), Burki and Perry (1997), Easterly, Loayza, and Montiel (1997), IADB (1997), and Lora and Barrera (1997). 2. The Global Trade Analysis Project (GTAP) database and model are disseminated by Center for Global Trade Analysis of Purdue University. See http://www.gtap.org and Hertel (1999). 3. An important assumption in the micro simulations is that price changes are equal across the whole country. This is not always the case and pass-through effects can be different according to the distance from the bor- der, the types of traded goods, and other factors that can be household spe- cific. For an example of an analysis that takes into account geographic price differentials in Mexico, see Nicita (2005). 4. For some of the surveys analyzed in this chapter, these transfers can be finely disaggregated up to the point at which it is possible to identify spe- cific government transfer policies (such as food stamps, health care reim- bursements, and other social transfer expenditures). 5. For instance, although in Chile the poverty incidence is almost the same in both rural and urban areas, rural poverty in Brazil is twice the urban rate, and the same ratio is 1.5 in Mexico and Colombia. The distribution of income is much more equal in Chile and Mexico than it is in Colombia. Brazil tends to protect its manufacturing sector, but Mexico's tariffs are heavily biased toward agriculture, and Chile has a uniform tariff structure. 6. For some individuals, this procedure yields negative differences that are set to zero. The proportion of self-employed with an imputed wage 88 BUSSOLO, LAY, MEDVEDEV, AND VAN DER MENSBRUGGHE higher than their reported self-employment earnings is significant for Brazil and Colombia but quite low for Chile and Mexico. 7. The rural and urban poverty measurement obtained in this way may be misleading. Price levels and the corresponding purchasing powers in rural and urban sectors of the economy can be quite different; so instead of a single international level applied to the whole population, zone-specific poverty lines could be used to guarantee more accurate estimates. 8. Because the initial level of poverty in Colombia is much higher than in the other countries under consideration, a smaller percentage decrease in the poverty headcount represents a larger reduction in terms of number of people. 9. As for the case of the growth elasticity, to calculate the trade elas- ticity, the growth rate of the average per capita income is normalized to 1 percentage point, although different households experience different growth rates of their incomes. 10. The initial inequality, measured by the Gini coefficient, for each country is as follows: Brazil 59, Chile 57, Colombia 63, and Mexico 53. References Adelman, I., and S. Robinson, 1978. Income Distribution Policy in Develop- ing Countries: A Case Study of Korea. New York: Oxford University Press. Bourguignon, François. 2002. "The Growth Elasticity of Poverty Reduc- tion: Explaining Heterogeneity across Countries and Time Periods." In Growth and Inequality, eds. T. Eicher and S. Turnovski. Cambridge, MA: MIT Press. Burki, J., and G. Perry. 1997. The Long March: A Reform Agenda for Latin America and the Caribbean in the Next Decade. Washington, DC: World Bank. Chen, Shaohua, and Martin Ravallion. 2004. "Welfare Impacts of China's Accession to the World Trade Organization." World Bank Economic Review 18: 29­57. Deaton, A. 2004. "Measuring Poverty in a Growing World (or Measuring Growth in a Poor World)." Review of Economics and Statistics 87 (1): 1­19. Easterly, W., N. Loayza, and P. Montiel. 1997. "Has Latin America's Post-Reform Growth Been Disappointing?" Journal of International Economics 43: 287­311. ECLAC (Economic Commission for Latin America and the Caribbean). 1996. "Economic Panorama of Latin America, 1996." United Nations, Santiago, Chile. Encuesta de Calidad de Vida. 1997. Departamento Administrativo Nacional de Estadística. Bogotá, Colombia. TRADE OPTIONS FOR LATIN AMERICA 89 Encuesta de Caracterización Socioeconómica Nacional. 2000. Ministerio de Hacienda, Gobierno de Chile. Santiago, Chile. Encuesta Nacional Ingresos y Gastos de los Hogares. 2001. Instituto Nacional de Estadística, Geografia e Informática. Aquascalientes, Mexico. Harrison, A. 2005. "Globalization and Poverty: An NBER Study." Univer- sity of California at Berkeley and NBER (Draft). National Bureau of Economic Research. Harrison, G. W., T. F. Rutherford, D. Tarr, and A. Gurgel. 2003. "Regional, Multilateral, and Unilateral Trade Policies of MERCOSUR for Growth and Poverty Reduction in Brazil." Working Paper No. 3051. World Bank, Washington, DC. Hertel, T. W. 1999. Global Trade Analysis: Modeling and Applications. New York: Cambridge University Press. Hertel, Thomas W., Maros Ivanic, Paul Preckel, and John Cranfield. 2004. "The Earnings Effects of Multilateral Trade Liberalization: Implications for Poverty in Developing Countries." GTAP Working Paper No. 16 Revised. Center for Global Trade Analysis, Global Trade Analysis Project, Purdue University. Hertel, T., and L. A. Winters. 2005. Poverty and the WTO. Impacts of the Doha Development Agenda. Washington, DC: Palgrave and World Bank. IADB (Inter-American Development Bank). 1997. "Latin America after a Decade of Reforms: Economic and Social Progress in Latin America, 1997 Report." IADB, Washington, DC. Ianchovichina, Elena, Alessandro Nicita, and Isidro Soloaga. 2000. "Trade Reform and Poverty: The Case of Mexico." The World Economy 25 (7): 945­73. Löfgren, H., R. Lee Harris, and S. Robinson. 2002. "A Standard Com- putable General Equilibrium (CGE) Model in GAMS." Microcomputers in Policy Research No. 5. International Food Policy Research Institute, Washington, DC. Lora, E., and F. Barrera. 1997. "A Decade of Structural Reforms in Latin America: Growth, Productivity and Investment Are Not What They Used to Be." Working Paper No. 350. Inter-American Development Bank, Office of the Chief Economist, Washington, DC. Lucas, Robert E., Jr. 1976. "Econometric Policy Evaluation: A Critique." Carnegie-Rochester Conference Series on Public Policy 1: 19­46. McCulloch, N., L. A. Winters, and X. Cirera. 2001. "Trade Liberalization and Poverty: A Handbook." U.K. Department for International Devel- opment and the Centre for Economic Policy Research, London. Nicita, A. 2005. "Multilateral Trade Liberalization and Mexican House- holds: The Effect of the Doha Development Agenda." Policy Research Working Paper No. 3707. World Bank, Washington, DC. Pesquisa Nacional por Amostra de Domicílios. 2001. Instituto Brasileiro de Geografia e Estatística. Rio de Janeiro, Brazil. 90 BUSSOLO, LAY, MEDVEDEV, AND VAN DER MENSBRUGGHE Ravallion, M. 2001. "Growth, Inequality, and Poverty: Looking beyond the Averages." World Development 29 (11): 1803­815. Robilliard, A. S., and S. Robinson. 2003. "Reconciling Household Surveys and National Accounts Data Using a Cross-Entropy Estimation Method." Review of Income and Wealth 49 (3): 395­406. Round, J. I. 2003. "Constructing SAMs for Development Policy Analysis: Lessons Learned and Challenges Ahead." Economic Systems Research 15 (2): 161­83. Rutherford, Thomas, David Tarr, and Oleksandr Shepotylo. 2005. "Poverty Effects of Russia's WTO Accession: Modeling `Real' Households and Endogenous Productivity Effects." Policy Research Working Paper Series No. 3473. World Bank, Washington, DC. van der Mensbrugghe, D. 2005. "Prototype Model for Real Computable Gen- eral Equilibrium Model for the State Development Planning Commission, P. R. China." http://siteresources.worldbank.org/INTPROSPECTS/ Resources/334934-1100792545130/LinkageTech Note.pdf. Winters, L. A., N. McCulloch, and A. McKay. 2004. "Trade Liberalization and Poverty: The Evidence So Far." Journal of Economic Literature 42 (1): 72­115. PART II Top-Down Approach with Behavioral Micro Simulations 4 Examining the Social Impact of the Indonesian Financial Crisis Using a Macro-Micro Model Anne-Sophie Robilliard, François Bourguignon, and Sherman Robinson Determining the social cost of a macroeconomic crisis like the one that struck Indonesia in 1997 is not an easy task. One year after the crisis, the World Bank (1998) argued that if real gross domestic product (GDP) declined by 12 percent in 1998, then the incidence of poverty in Indonesia could affect up to 14.1 percent of the popula- tion in 1999--compared with a level of 10.1 percent in mid-1997. Other estimates released at about the same time were more pessimistic. Indonesia's Central Board of Statistics (CBS 1998) predicted a fourfold increase of the poverty headcount (rising from 11.3 percent in 1996 to 39.9 percent by mid-1999), whereas the International Labour Organization (ILO 1998) predicted a sixfold increase (of up to 66.3 percent) by the end of 1999.1 Ex post esti- mates were much lower than these dramatic predictions. In a study based on data collected in Indonesia's National Labor Force Surveys (Survei Angkatan Kerja Nasional, or SAKERNAS) from August 1997 through 1998, Manning (2000) found that the "traditional" features of the Indonesian labor markets helped cushion the economic shock of the crisis. Finally, more recent estimates published by the World Bank (Suryahadi and others 2000), based on a comparison of the poverty level between two National Social Economics Surveys (Survei Sosial Ekonomi Nasional, or SUSENAS), 93 94 ROBILLIARD, BOURGUIGNON, AND ROBINSON show that the poverty headcount rose from 9.7 percent to 16.3 per- cent between 1996 and 1999. These various estimates illustrate the basic methodological ambiguity in predicting either what will happen to the poor just after an economic crisis strikes or in deciphering what did happen ex post (after the fact, based on actual data). In both cases, an explicit counterfactual scenario is needed. In the first case, the scenario must show departures from the precrisis evolution of the economy. In the second case, it must permit assessing what would have happened without the crisis and help disentangle the effects of the crisis from other exogenous shocks that are present in the data--such as the climatic effects of the El Niño drought in the case of Indonesia. This counterfactual scenario may be simple. For instance, it is natural to assume that decreases in household income or consumption depend on the economic activity of the social groups being considered. A scenario would thus consist of a set of predic- tions about the rate of growth of either the various sectors of the economy or the aggregate income of the various factors of produc- tion. The early rough estimates of the effect of the Indonesian crisis on poverty were based on this type of approach. But the divergence between those estimates suggests that establishing even a simple coun- terfactual scenario of this type is not easy--and requires more than a rough model of the economy. The use of more rigorous multisector models would probably yield more consensual predictions for the economy as a whole and for its various sectors and factors of production. It is not clear, however, that this would also result in satisfactory predictions for the distribution of income and poverty. Associating household incomes with sector activity or factor remuneration rates is, in effect, equivalent to defining representative household groups (RHGs) that derive income from a predetermined combination of factors. Models that incorporate several sectors and several RHGs with some exogenous distribution within those groups have been used for some time now--see, for example, Dervis¸, de Melo, and Robinson (1982) and the survey by Adelman and Robinson (1989). Whether these models are used to analyze either structural reforms like trade regimes or short-run macroeconomic issues--as in Bourguignon, Branson, and de Melo (1992)--this approach is problematic, as well. In particular, by ignoring changes in the distribution of income within RHGs, these models may ignore major sources of change in the distribution of economic welfare and poverty. In most studies of changes in inequality over time,2 it is indeed shown that changes in the relative income and weight of a few groups of households with identical selected characteristics leave a sizable unexplained SOCIAL IMPACT OF A FINANCIAL CRISIS 95 residual. Focusing on the inequality between representative groups (as multisector, multihousehold models presently do) may thus lead to a biased view of the impact of macro or structural policies on the distribution of income. A simple example may explain the nature of the problem. A majority of households in Indonesia generate income from various sources: (1) salaried employment of some members in the formal sector, (2) wage work in the informal sector of others, and (3) self- employment of yet another group. If RHGs are defined, as is typically done, by the sector of activity and employment status of the household heads (small farmers, urban unskilled workers in the formal sector, and so forth), it may not be too much of a problem to account for this multiplicity of income sources. Thus, the change in the inequality between the groups of small farmers and urban unskilled workers in the formal sector may account for the fact that both groups have different secondary sources of income--because of differences in household composition, labor supply behavior, and the occupation of secondary members. Two difficulties arise, however. First, say that a macroeconomic crisis or a trade reform modifies the number of unskilled urban workers employed in the formal sector. What should be done with the number of households with households heads in that occupation? Should it be modified? If so, from which groups must new households in that RHG be taken, or to which groups should they be allocated? First, could this oper- ation be completed based on the assumption that the distribution of income within all RHGs remains the same? Second, assuming that changes in occupation affect only secondary members and not household heads (so that RHGs are unchanged), is it reasonable to assume that all households in a group are affected in the same way by this change in the activity of some of their group members? A secondary member may move out of the formal sector and back into family self-employment, but this may happen only in a subgroup of households within a given representative group, which may seri- ously affect the distribution within this group. It is phenomena of this kind that may help explain changes in the "within" component of inequality decomposition exercises. But these changes are ignored in multisector, multihousehold group models. This chapter presents a new approach that can be used to quan- tify the effects of macroeconomic shocks on poverty and inequality by overcoming difficulties such as these.3 This new approach combines a micro simulation model with a standard multisector computable general equilibrium (CGE) model. The two models are used in a sequential fashion to simulate the full distributional impact of a financial crisis and generate meaningful counterfactual 96 ROBILLIARD, BOURGUIGNON, AND ROBINSON scenarios.4 The CGE model is based on a standard social accounting matrix (SAM) and is intended to capture both the structural features of an economy and the general equilibrium effects of the macroeconomic constraints that arise from macro shocks. The micro simulation model is based on a subsample of the 1996 SUSENAS survey and simulates income generation mechanisms for approximately 10,000 Indonesian households. The two models are treated separately. The macro (or CGE) model communicates with the micro model by generating a vector of prices, wages, and aggregate employment variables that correspond to a given shock or policy. The micro model is then used to generate changes in individ- ual wages, self-employment incomes, and employment status in a way that is consistent with the set of macro variables fed by the macro model. When this is done, the full distribution of real household income corresponding to the simulated shock or policy may be evaluated. This framework is designed to capture important channels through which a financial crisis of the type that struck Indonesia in 1997 may affect household incomes. Its main focus is the structure and functioning of labor markets, but this approach also captures part of the expenditure-side story by taking into account any increases in the relative price of food. The following section shows the structure of the micro simulation module and explains how it is linked to the CGE part of the model. The general features of the CGE model are then discussed, followed by scenarios, simulation results, and conclusions. The Micro Simulation Model This section briefly describes the specification of the household income generation model used for micro simulation and then focuses on how consistency is achieved between micro simulation and the predictions of the CGE model. A more detailed discussion of the specification and econometric estimates of the various equations of the household income generation model and simulation methodol- ogy can be found in Alatas and Bourguignon (2005).5 In the notations used in the remainder of chapter 4, the house- hold income generation model for household m with working-age members km consists of the following set of equations: (4.1) Log wmi i g(mi) xmi g(mi) mi 1, . . . km (4.2) Log ym f(m) Zm f(m) f(m) Nm m SOCIAL IMPACT OF A FINANCIAL CRISIS 97 1 km (4.3) Ym wmiIWmi ym Ind(Nm 0) y0 Pm m i 1 K (4.4) Pm smk pk k 1 (4.5) IWmi Ind awh(mi) zmibwh (mi) uwmi Sup 0, ash (mi) zmibsh(mi) usmi km (4.6) Nm Ind ash(mi) zmibh s (mi) umi s i 1 Sup 0, ahw uw (mi) zmibw h(mi) mi . Equation (4.1) expresses the (log) earnings of member i of house- hold m as a function of that member's personal characteristics, x. The latter include age, education level, and geographic region. The residual term, , describes the effects of unobserved earning mi determinants. This earning function is defined separately on various "segments" of the labor market defined by gender, skill level (less than secondary or more than primary education), and area (urban/rural). Thus, g(mi) is an index function that indicates the labor market segment to which member i in household m belongs. Equation (4.2) is the (net) income function associated with self- employment, or small entrepreneurial activity, which includes the opportunity cost of household labor and profit. This function is defined at the household level and depends both on the number of household members actually involved in that activity, Nm, and on some household characteristics, Zm. These characteristics include area of residence, the age and schooling of the household head, and land size for farmers. The residual term, , summarizes the effects m of unobserved determinants of self-employment income. A different function is used depending on whether the household is involved in farm or nonfarm activity. This is exogenous and is defined by whether or not the household has access to land, as represented by the index function f(m). Equation (4.3) is an accounting identity that defines total household real income, Ym, as the sum of wage income of its members, profit from self-employment, and (exogenous) nonlabor income, y0 . In this equation, the notation IWmi stands for a dummy m variable that is equal to unity if member i is a wage worker and zero otherwise. Thus wages are summed over only those household 98 ROBILLIARD, BOURGUIGNON, AND ROBINSON members actually engaged in wage work. Likewise, income from self-employment has to be taken into account only if at least one member of the household is engaged in self-employment activity (Nm 0). Total income is then deflated by a household-specific consumer price index (CPI), Pm, which is derived from the observed budget shares, smk, of household m and the price, pk, of the various consumption goods, k, in the model--equation (4.4). Equations (4.5) and (4.6) represent the occupational choices made by household members. This choice is discrete. Each individ- ual must choose from three alternatives: being inactive, a wage worker, or self-employed. This choice is represented within a discrete utility-maximizing framework. The utility associated with the first alternative (inactivity) is arbitrarily set to zero, whereas the utility of being a wage worker or self-employed is a linear function of a set of individual and household characteristics, zmi. The intercept of these functions has a component, aw or as, that is common to all individuals, and an idiosyncratic term, umi, which represents unobserved determinants of occupational choices. The coefficients of individual characteristics, zmi, bw, or bs, are common to all individuals. However, they may differ across demographic groups indexed by h(mi). For instance, occupational choice behavior, as described by coefficients aw, as, bw, and bs, may be different for household heads, spouses, and male or female children. The constants may also be demography-specific. Given this specification, an individual will prefer wage work if the utility associated with that activity is higher than that associated with the two other activities. This is the meaning of equation (4.5). Likewise, the number of self-employed workers in a household is the number of individuals for whom the utility of self-employment is higher than that of the two alternatives, as represented in equation (4.6).6 The model is now complete. Overall, it defines the total real income of a household as a nonlinear function of the observed characteristics of household members (xmi and zmi), some character- istics of the household (Zm), its budget shares (sm), and unobserved characteristics ( , mi m , uw , and usmi). This function depends on five mi sets of parameters: (1) for the earning functions ( gand g), for each labor market segment, g; (2) for the self-employment income functions ( , f f, and f); (3) for the farm or nonfarm sector, f; (4) for the utility of the alternative occupational choices (aw, bw, ah, and s h h bh), for the various demographic groups, h; and (5) for the vector of s prices, p. As is shown later, it is through several of these parameters that the results of the CGE part of the model may be transmitted to the micro module. SOCIAL IMPACT OF A FINANCIAL CRISIS 99 The micro simulation model gives a rather complete description of household income generation mechanisms by focusing on both earn- ing and occupational choice determinants. However, a number of assumptions about the functioning of the labor market are incorporated in this specification. The fact that labor supply is considered to be a discrete choice between either inactivity or full- time work for wages (or for self-employment income) within the household calls for two sets of remarks. First, the assumption that individuals are inactive or work full time is essentially justified by the fact that no information on the number of hours worked is available in the micro data source used to estimate the benchmark set of the model's coefficients. As a practical matter, this implies that estimated individual earning functions--equation (4.1)--and profit functions-- equation (4.2)--may incorporate some labor supply dimension. Sec- ond, distinguishing between wage work and self-employment is implicitly equivalent to assuming that the Indonesian labor market is imperfectly competitive. If this were not the case, then returns to labor would be the same in both types of occupation; and self-employment income would be different from wage income only because it would incorporate the returns to nonlabor assets being used. The specifica- tion that has been selected is justified, in part, by the fact that assets used in self-employment are not observed, so one cannot distinguish between self-employment income derived from labor and that derived from other assets. But it is also justified by the fact that the labor market may be segmented (in the sense that labor returns are not equalized across wage work and self-employment). There may be various reasons for this. On the one hand, there may be rationing in the wage labor market. People unable to find jobs as wage workers move into self-employment, which is a kind of shelter. On the other hand, there may be externalities that make working within and outside the household imperfect substitutes. These two interpretations are consistent with the way in which the labor market is represented in the CGE part of the model.7 It is now time to consider how the link is made between the CGE part and the micro part of the model--and how the effects of macroeconomic shocks and policies are simulated on each household represented in the database. The principle behind these simulations is quite simple. It associates macroeconomic shocks and policies simulated in the CGE part of the model with changes in the set of coefficients of the household income generation model-- equations (4.1)­(4.6). With a new set of coefficients ( , g g, f, , f f, aw, bw, ah, bh) and the observed and unobserved individual and house- s s h h hold characteristics (xmi, zmi, Zm, sm, , w s mi m, umi, umi), these equations allow one to compute the occupational status of all household 100 ROBILLIARD, BOURGUIGNON, AND ROBINSON members, their earnings, their self-employment income, and finally, the total real income of their household. But this association must be done in a consistent way. Consistency with the equilibrium of aggre- gate markets in the CGE model requires that (1) changes in average earnings (with respect to the benchmark in the micro simulation) must be equal to changes in wage rates in the CGE model for each segment of the market for wage labor; (2) changes in self-employment income in the micro simulation must be equal to changes in informal sector income per worker in the CGE model; (3) changes in the number of wage workers and those self-employed by labor market segment in the micro simulation model must match those same changes in the CGE model; and (4) changes in the consumption price vector, p, must be consistent with the CGE model. The link between the CGE part of the model and the micro part is obtained through the resolution of the following system of equations: Ind awh * w u^w (mi) zmib^h (mi) mi m i,g(mi) G Sup 0, ah s * s s (mi) zmib^h(mi) u^ mi E*G Ind ash* s s (mi) zmib^h (mi) u^mi m i,g(mi) G Sup(0, awh * u^ w (mi) zmib^ w h(mi) mi S*G Exp * ^ * w u^ w G xmi G ^mi Ind awh (mi) zmi b^h(mi) mi m i,g(mi) G Sup 0, ah s* wG* (mi) zmi b^h s (mi) u^mi s Exp * ^ ^ N^ + 0 F Zm F F m ^ m Ind Nm IF*, m i,f(m) F with N^ m Ind ah s* (mi) zmib^h s (mi) u^mi s i Sup 0, ah w* (mi) zmib^ w h(mi) u^wmi , where the unknowns are g*, f*, aw*, and ah. This system of equa- s* h tions has as many equations as unknowns and has a unique solution that can be obtained through standard Gauss-Newton techniques.8 Once the solution is obtained, it is a simple matter to compute the new income of each household in the sample, according to the model in equations (4.1)­(4.6), with the new set of coefficients g* , f* , aw*, h and ah, and then to analyze the modification that this implies for the s* overall distribution of income. SOCIAL IMPACT OF A FINANCIAL CRISIS 101 The justification for using the intercepts is that it implies a "neutrality" of the changes being made with respect to individual or household characteristics. For example, changing the intercepts of the log earning equations generates a proportional change of all earnings in a labor market segment, regardless of individual characteristics outside those that define the segments (skill, gender, and geographic area). The same is true of the change in the intercept of the log self- employment income functions. A similar argument applies to the cri- teria associated with the various occupational choices. Indeed, it is easily shown that changing the intercepts of the multilogit model implies the following neutrality property: the relative change in the ex ante probability that an individual has some occupation depends only on the initial ex ante probabilities of the various occupational choices, rather than on individual characteristics. In the Indonesian case, the number of variables that allow the micro and the macro parts of the overall model to communicate, that is, the vector (E* , S*G, w*G, I*F, q*), is equal to 26 plus the num- G ber of consumption goods used in defining the household-specific CPI deflator. The labor market has eight segments. The employment requirements for each segment in the formal (wage work) and infor- mal (self-employment) sectors (E* and S*) lead to 16 restrictions. In G G addition, there are eight wage rates in the formal sector (w*G) and two levels of self-employment income (I*F) in the formal and the informal sectors. Thus, simulated changes in the distribution of income implied by the CGE part of the model are obtained through a procedure that allows numerous degrees of freedom. Two elements must be added to describe the full scope of the model. First, the household-specific price index, Pm, is based on the disaggregation of expenditure into only two goods, food and nonfood. This disaggregation is the most relevant one for the analy- sis of the consequences of the Indonesian financial crisis. Second, other incomes, y0 , are considered as exogenous (in real terms) in m all simulations. They include housing and land rents, dividends, royalties, imputed rents from self-occupied housing, and transfers from other households and institutions. It would have been possible to endogenize some of these items in the CGE model, but this was not done. The CGE Model The CGE model presented in this chapter is based on a 1995 SAM. The SAM has been disaggregated using cross-entropy estima- tion methods (Robinson, Cattaneo, and El-Said 2001) to include 102 ROBILLIARD, BOURGUIGNON, AND ROBINSON 38 sectors, 14 goods, 14 factors of production (8 labor categories and 6 types of capital), and 10 household types, as well as the usual accounts for aggregate agents (firms, government, rest of the world, savings-investment). The CGE model starts from the standard neoclassical specification in Dervis¸, de Melo, and Robinson (1982) but also incorporates the disaggregation of production sectors into formal and informal activities and associated labor market imperfections. Markets for goods, factors, and foreign exchange are assumed to respond to changing demand and supply conditions, which are, in turn, affected by government policies, the external environment, and other exogenous influences. The model is Walrasian in that it determines only relative prices and other endogenous real variables in the economy. Financial mechanisms are modeled implicitly, and only their real effect is accounted for in a simplified way. Sectoral product prices, factor prices, and the real exchange rate are defined relative to the producer price index of goods for domestic use, which serves as the numeraire. The exchange rate represents the relative price of tradable goods with regard to nontraded goods (in units of domestic currency per unit of foreign currency). Activities and Commodities Indonesia's economy is dualistic, and the model captures this by distinguishing between formal and informal "activities" in each sector. Both subsectors differ in the type of factors they use--a distinction that allows for treating formal and informal factor markets differently. Informal and formal sectors are further differentiated by the fact that formal sectors are assumed to rely on foreign credit to operate, whereas informal sectors do not. For all activities, the production technology is represented by a set of nested constant elasticity of substitution (CES) value added functions and fixed (Leontief) intermediate input coefficients. On the demand side, imperfect substitutability is assumed between formal and informal products. Thus, consumers demand an aggregate of the formal and informal products. Domestic prices of commodities are flexible, varying to clear markets in a competitive setting where individual suppliers and demanders are price-takers. Following Armington (1969), the model assumes imperfect substitutability, for each good, between the domestic commodity (which itself results from a combination of formal and informal activities) and imports. What is demanded is a composite good, which is a CES aggregation of imports and domestically produced goods. For export commodities, the allocation of domestic output SOCIAL IMPACT OF A FINANCIAL CRISIS 103 between exports and domestic sales is determined on the assumption that domestic producers maximize profits subject to imperfect transformability between these two alternatives. The composite production good is a constant elasticity of transformation (CET) aggregation of sectoral exports and domestically consumed products. These assumptions of imperfect substitutability and trans- formability grant the domestic price system some degree of auton- omy from international prices and serve to dampen export and import responses to changes in the producer environment. Such treatment of exports and imports provides a continuum of trad- ability and allows two-way trade at the sector level--which reflects what is observed empirically at the level of aggregation of the model. Factors of Production Eight labor categories are included in the Indonesia CGE model: urban male unskilled, urban male skilled, urban female unskilled, urban female skilled, rural male unskilled, rural male skilled, rural female unskilled, and rural female skilled. The designations male and female, as well as skilled and unskilled labor, are assumed to be imperfect substitutes in the production activity of urban or rural sectors. In addition, labor markets are assumed to be segmented between formal and informal sectors. In the formal sector labor markets, imperfect competition mechanisms are assumed to result in some increasing wage-employment curve; and real wages are defined by the intersection of that curve and competitive labor demand. Informal sector labor is equivalent to self-employment, and wages in that sector are set to absorb any labor not employed in the formal sectors. Wages adjust to clear all labor markets in the informal sectors, whereas employment adjusts in the formal sectors. Land appears as a factor of production in the agricultural sectors. Only one type of land is considered in the model. It is competitively allocated among the different crop sectors so that marginal value added is equalized across activities. Capital markets are segmented into six categories: owner-occupied housing, other unincorporated rural capital, other unincorporated urban capital, domestic private incorporated capital, public capital, and foreign capital. Given the short-term perspective of the model, it is assumed that capital is fixed in each activity. The model also incorporates working capital requirements by all sectors. Sectors demand domestic working capital in proportion 104 ROBILLIARD, BOURGUIGNON, AND ROBINSON to their demands for domestically produced intermediate inputs. They also demand working capital denominated in foreign exchange in proportion to their demands for imported intermediate inputs. Informal sectors are assumed not to require any imported intermediate inputs. Working capital is treated as a factor input that is strictly complementary to physical capital. The model incorporates a nested production function in all sectors, with aggregate "capital" consisting of an aggregation of physical capital, domestic working capital, and foreign working capital (foreign exchange). Both types (domestic and foreign) of working capital are assumed to be required in fixed proportions to physical capital. When the supplies of aggregate domestic and foreign working capital are reduced (as an effect of the financial crisis), they are assumed to be competitively allocated across sectors, so that their marginal revenue product is the same. Because physical capital is fixed, this causes capacity underutilization in some sectors. The effect of this treatment is to make aggregate output sensitive to any reduction in the supply of working capital. With cuts in work- ing capital, the utilization of physical capital will also decline.9 The sector impact depends on a sector's dependence on intermediate inputs, both domestic and imported. Households The disaggregation of households in the CGE model is not central to this discussion, because changes in factor prices are passed on directly to the micro simulation model without use of the RHGs used in the original SAM. Consumption demand by households at the CGE level is determined by the linear expenditure system (LES), in which the marginal budget share is fixed and each commodity has a minimum consumption (subsistence) level. Macro Closure Rules Equilibrium in a CGE model is defined by a set of constraints that need to be satisfied by the economic system but are not directly considered in the decisions of micro agents (Robinson 1989). Aside from the supply-demand balances in product and factor markets, three macroeconomic balances are specified in the Indonesia CGE model: (1) the fiscal balance, with government savings equal to the difference between government revenue and spending; (2) the external trade balance (in goods and nonfactor services), which implicitly equates the supply and demand for foreign exchange SOCIAL IMPACT OF A FINANCIAL CRISIS 105 (flows, not stocks, because the model has no assets or asset markets); and (3) savings-investment balance. Practically, a balanced macro closure is used, in which aggregate investment and government spending are assumed to be in a fixed proportion to total absorp- tion. Any shock affecting total absorption is thus assumed to be shared proportionately among government spending, aggregate investment, and aggregate private consumption. While simple, this closure effectively assumes a successful structural adjustment program in which a macro shock is assumed not to cause particular actors (government, consumers, and industry) to bear a dispropor- tionate share of the adjustment burden. Scenarios and Simulations As mentioned earlier, both parts of the model are handled separately, with the macro level communicating with the micro part through a vector of "linking variables" (for prices, wages, and aggregate employment). The overall structure is top-down in that there is no feedback from the micro model back to the macro CGE model. This top-down sequential structure allows running various kinds of experiments. In the first set of experiments (labeled "historical simulation"), historical changes in the linking variables are derived from price statistics and labor market surveys taken during and after the crisis and fed directly into the micro model, without any use of the macro CGE model. Thus, this historical simulation is essentially meant to test the capacity of the micro model to generate income distribution predictions on the basis of a few observed macro indicators. In the second set of simulations (labeled "policy simula- tions"), the value of linking variables is taken from the results of the CGE model. These simulations are used to decompose the historical shock into various elementary components. Time Horizon The question of time horizon requires comment. The financial crisis that struck Indonesia during the summer of 1997, and the resulting turmoil, spanned approximately 20 months--extending until March 1999, when the first signs of output recovery were recorded.10 Given the equilibrium nature of the macro framework and of the linking variables between the macro and micro models discussed in this chapter, the crisis is not tracked month by month. Instead, the impact of the shock is analyzed using comparative statics. The deviations from base values used as historical references are thus computed for a 106 ROBILLIARD, BOURGUIGNON, AND ROBINSON period extending from July­August 1997 to September­October 1998. The latest date corresponds to the peak of the crisis with respect to both macroeconomic indicators (Azis, Azis, and Thorbecke 2001) and poverty indicators (Suryahadi and others 2000). The analysis of this short-term shock in a CGE framework is made possible by imposing a number of rigidities in the specification of factor markets, as shown earlier. The base year for the macro model is the 1995 SAM, with both the consumption structure and the factor disaggregation based on the 1996 SUSENAS. The sample used for the micro simulation is a subsample of the 1996 SUSENAS. Some inconsistency could arise between the macro and the micro parts of the model because they do not refer to the same year. In fact, due to the sequential nature of the framework used in this chapter, full consistency is not required between the macro and the micro sides of the model. Indeed, all of the analysis using this model may be performed in terms of deviations from benchmarks that may not fit perfectly together.11 Historical Changes in Poverty As mentioned earlier, diverse estimates have been published on the before-and-after impact of the Indonesian financial crisis on poverty and income distribution. The results reported by Suryahadi and others (2000) are used as a reference to analyze the historical change in poverty and income distribution. These authors used various household surveys to compute changes in real income over the period from 1996 to 1999. Although poverty rates derived from SUSENAS data would be consistent with the household sample used in the model presented in this chapter, changes derived from the Indonesian Family Life Survey (IFLS), adjusted to achieve consis- tency with other estimates (Suryahadi and others 2000), were used as a general benchmark. This choice is justified, on the one hand, by the fact that the SUSENAS (conducted every three years) does not allow isolation of the crisis period and, on the other hand, by the fact that the second wave of the IFLS was specifically designed to help determine how the crisis affected welfare (Frankenberg, Thomas, and Beegle 1999). Based on IFLS estimates adjusted by Suryahadi and others (2000), poverty incidence is shown to have increased by 164 percent between September 1997 and October 1998.12 Because the IFLS results reported by Suryahadi and others (2000) do not distinguish between the urban and the rural sectors, the present authors report estimates based on both the 1996 and 1999 SUSENASs--to compare how urban and rural households SOCIAL IMPACT OF A FINANCIAL CRISIS 107 Table 4.1 Evolution of Poverty in Indonesia, 1996­99 Percentage Households/indicator 1996 1999 change All Headcount index (P0) 9.75 16.27 66.8 Poverty gap index (P1) 1.55 2.79 80.2 Poverty severity index (P2) 0.39 0.75 91.9 Urban Headcount index (P0) 3.82 9.63 152.3 Poverty gap index (P1) 0.53 1.51 183.0 Poverty severity index (P2) 0.12 0.37 201.6 Rural Headcount index (P0) 13.10 20.56 56.9 Poverty gap index (P1) 2.12 3.61 70.5 Poverty severity index (P2) 0.54 0.99 83.6 Sources: SUSENAS 1996 and 1999, cited by Suryahadi and others (2000). fared over the period (table 4.1). The overall increase in poverty appears to be much smaller than the one that Suryahadi and oth- ers (2000) obtained using IFLS data. This result is consistent with the difference in the time coverage of both sources, because poverty decreased with the recovery after October 1998. The data in table 4.1 show that poverty increased more in the urban sector than in the rural sector. Nevertheless, poverty remains higher in the rural sector because of the initial disadvantage of that sector. The strong increases in the poverty gap indicator (P1) and the poverty severity index (P2) also show that from 1996 to 1999, the situation deteriorated more for the poorest of the poor. Historical Experiment The first experiment, called "historical," uses historical vectors of the linking variables (prices, wages, and aggregate employment changes) to feed into the micro model. Changes in the last two sets of variables, shown in table 4.2, are derived from the comparison of two SAKERNASs (for 1997 and 1998). Consumer price changes (not reported) are taken from reports by Badan Pusat Statistik (BPS). SAKERNASs do not indicate changes in self-employment incomes. The authors assume that these are equal to changes in wages; but because of the effect of increases in relative output prices, this assumption is probably unsatisfactory in the case of rural self-employment incomes. A comparison of the 1997 and 1998 employment surveys shows a dramatic drop in real wages 108 ROBILLIARD, BOURGUIGNON, AND ROBINSON Table 4.2 Evolution of Occupational Choices and Wages by Segment, 1997­98 Wage Self- Nominal Real Segment Inactive worker employed wage Wage Urban male unskilled 0.9 6.5 5.7 8.2 40.8 Urban male skilled 11.9 12.7 9.9 5.3 42.3 Urban female unskilled 2.6 5.1 5.9 21.8 33.4 Urban female skilled 5.9 15.5 2.3 10.3 39.6 Rural male unskilled 1.8 13.6 5.1 27.9 30.0 Rural male skilled 2.5 13.3 9.3 16.8 36.1 Rural female unskilled 5.5 0.0 7.5 47.3 19.4 Rural female skilled 2.7 14.3 3.4 12.2 38.6 All segments 0.3 10.2 5.8 11.7 38.9 Sources: SAKERNAS 1997 and 1998; authors' calculations. Note: Numbers in the first three columns are percentage changes in proportions. Real wage is equal to nominal wage deflated by consumer price index base year 1996 100. and an important shift out of wage work and into self-employment over the period. It also suggests that overall inactivity did not increase significantly. The picture differs slightly, however, across labor types. The movement out of wage work and into self-employment activi- ties is observed for all but two categories, urban and rural unskilled females. Concerning the employment rate, although stable overall, it decreases for all skilled categories but increases for all unskilled categories.13 Table 4.3 shows the results on poverty and inequality derived from the micro model (under the preceding assumptions). They show a 238.6 percent increase in poverty, higher than the historical change of 164 percent reported by Suryahadi and others (2000) based on the comparison of the 1997 and 1998 IFLS. This overestimation can be explained by the simulation, which ignores the fact that self- employment incomes decreased less than real wages. The poverty increase appears to be fueled by the dramatic income shock--a 40.4 percent drop in mean per capita income. Results also show an increase in inequality driven by the increase of within-sector inequal- ity: although rural and urban mean per capita incomes converge (that is, the fall in per capita income in the urban sector is bigger than in the rural sector, 44.8 percent and 26.5 percent, respectively), the decrease in between-sector inequality does not compensate for the increases within the urban and rural sectors. In terms of the rural- urban divide, the results appear consistent with the historical record shown in table 4.1, although those data refer to a distinct time period. SOCIAL IMPACT OF A FINANCIAL CRISIS 109 Table 4.3 Historical Simulation Results All Urban Rural households households households Income and relative Percentage Percentage Percentage price changes Base change Base change Base change Per capita incomea (Rp, thousandsb) 121.1 40.4 171.0 44.3 90.6 35.9 Entropy index 0 ( 100) 35.5 2.7 38.7 10.2 25.6 9.0 Entropy index 1 ( 100) 49.3 0.9 53.9 8.7 33.1 4.9 Gini index (%) 45.6 0.2 47.5 3.9 38.7 2.9 Headcount index (P0) 9.2 238.6 4.0 432.9 12.4 200.4 Poverty gap index (P1) 2.2 340.5 1.0 528.8 2.9 299.0 Poverty severity index (P2) 0.9 408.8 0.4 648.5 1.2 355.9 Source: Results from the authors' micro simulation module, using historical changes in prices, wages, and occupational choices by segment (see table 4.2). Self-employment income is assumed to drop by the same magnitude as male unskilled wage, that is, 40 percent in the urban sector and 30 percent in the rural sector. Note: Base values are used for the Base column and percentage change for other simulations. a. Per capita income is total monthly income. b. Rp rupiah, Indonesia's official currency. The poverty increase in the urban sector is much higher than in the rural sector, but poverty remains higher in the rural sector. These different results show the capacity of the micro simulation framework to generate plausible income distribution predictions on the basis of a few observed macro indicators. CGE Experiments In the following experiments, the vector of linking variables fed into the micro simulation is derived from the results of the CGE model. The set of experiments presented attempts to reproduce and decom- pose the effect of the crisis within the framework of the CGE model. The base CGE scenario seeks to reproduce the evolution of the Indonesian economy between 1997 and 1998 in terms of changes in employment, wages, and macroeconomic aggregates. The most important external shocks during that period are the financial cri- sis and the extended drought caused by El Niño. The drought is simulated through a negative 5 percent shock on the total produc- tivity factor in agricultural sectors. A 25 percent increase in the marketing cost of food is assumed. This increase reflects the fact 110 ROBILLIARD, BOURGUIGNON, AND ROBINSON that traders, more than producers, are expected to benefit from the food price increase. The financial crisis is simulated through a combination of different shocks. It is assumed that the need to adjust the current account led to a real devaluation that is simu- lated through a 30 percent decrease in the exogenous foreign sav- ing flows to the economy (SIMDEV scenario). As a result of the devaluation, all sectors experienced a "credit crunch," simulated through a cut in the supply of working capital. As shown earlier, two types of working capital are considered. In a first stage, the impact of a 25 percent cut in the availability of foreign working capital is examined in combination with the real devaluation described above (the DEVCCF scenario). In a second stage, the impact of a 20 percent cut in the availability of domestic credit is considered (the FINCRI scenario). Because the domestic credit crunch shock is viewed as stemming from the foreign credit crunch, it is simulated in combination with the two previous components of the financial crisis. The resulting simulation can then be ana- lyzed as mimicking a "pure" financial crisis shock, without any other historical shock. The effect of the El Niño drought is first simulated alone (SIMELN scenario) and then in combination with the financial crisis, thus yielding something that should be close to what actually happened in Indonesia between 1997 and 1998 (the SIMALL scenario). Table 4.4 shows how different elements of the crisis contributed to the total negative real GDP shock. The historical simulation cap- tures the main changes observed over the period: a 14.4 percent drop in GDP, a fall in imports and a surge of exports, an increase in the relative price of food commodities, and a drop in real wages. Combining the different shocks shows that the credit crunch is the major force explaining the collapse of GDP, while the drought com- bined with increases in the marketing cost of food appears to be the main driving force behind increases in the relative price of food commodities. In terms of the impact of the macro shocks on poverty and income distribution, the results in table 4.5 show that the modeling exercise yields a 143.4 percent increase in the poverty headcount ratio when all components of the crisis are taken into account (SIMALL). This surge in poverty appears to be fueled by the drop in the average income per capita and by an important increase in inequality indi- cators. Both the financial crisis and the El Niño drought contribute to the negative income impact and the increase in inequality. In terms of the rural-urban divide, the CGE experiments pre- sented in this chapter capture (to some extent) the differences in per capita income changes shown in the historical simulation. This Table 4.4 Simulation Results: Macro Aggregates Indicator BASE SIMELN SIMDEV DEVCCF FINCRI SIMALL GDP at factor costs (Rp, thousands of billionsa) 535.6 0.5 0.9 10.7 14.1 14.4 Exports (Rp, thousands of billions) 122.7 0.4 28.8 19.4 15.4 13.1 Imports (Rp, thousands of billions) 126.8 0.3 19.2 28.4 32.2 34.4 Exchange rate 1.0 5.1 31.8 27.3 27.2 24.3 Food/nonfood terms of trade 1.0 27.3 15.4 4.2 3.3 21.0 Incorporated capital incomeb 1.0 13.2 7.8 43.0 32.2 19.7 Agricultural self- employment incomec 1.6 5.9 8.2 8.0 18.5 23.4 Nonagricultural self- employment income 4.5 19.9 4.1 16.1 16.1 30.7 Skilled labor wagec 4.9 17.5 12.8 37.5 42.2 50.9 Unskilled labor wageb 2.7 14.5 12.6 32.3 35.5 43.0 Source: Results from the authors' CGE module. Note: Base values for BASE column and percentage change for other simulations; SIMELN El Niño drought; SIMDEV real devaluation; DEVCCF real devaluation foreign credit crunch; FINCRI real devaluation foreign credit crunch domestic credit crunch; SIMALL real devaluation foreign credit crunch domestic credit crunch El Niño drought; GDP gross domestic product. 111 a. Rp rupiah, Indonesia's official currency. b. Incorporated capital income includes private, public, and foreign capital income. c. Self-employment and wage incomes are equal to value added divided by quantity of labor units in the social accounting matrix. 112 Table 4.5 Simulation Results: Per Capita Income, Inequality, and Poverty Indicators Indicator BASE SIMELN SIMDEV DEVCCF FINCRI SIMALL All areas Per capita incomea (Rp, thousandsb) 121.1 12.4 5.1 16.3 19.5 27.9 Entropy index 0 ( 100) 35.5 2.9 1.4 3.0 1.3 5.2 Entropy index 1 ( 100) 49.2 4.5 1.7 2.9 1.0 6.7 Gini index (%) 45.5 1.3 0.3 1.8 0.2 2.2 Headcount index (P0) 9.2 49.7 17.1 51.6 80.2 143.4 Poverty gap index (P1) 2.2 54.6 25.2 61.9 101.0 182.0 Poverty severity index (P2) 0.9 54.1 31.3 68.2 111.6 197.5 Urban Per capita incomea (Rp, thousandsb) 170.9 14.0 7.7 23.8 25.3 33.5 Entropy index 0 ( 100) 38.7 6.8 5.4 4.5 8.0 15.0 Entropy index 1 ( 100) 53.9 7.9 5.3 4.5 6.8 15.1 Gini index (%) 47.5 3.3 2.4 2.1 3.5 6.9 Headcount index (P0) 4.0 70.4 44.1 130.8 167.1 301.4 Poverty gap index (P1) 1.1 70.6 55.4 135.1 186.6 324.8 Poverty severity index (P2) 0.4 72.0 67.3 146.9 216.6 353.4 Rural Per capita incomea (Rp, thousandsb) 90.6 10.5 2.0 7.7 12.8 21.5 Entropy index 0 ( 100) 25.6 4.1 3.7 5.1 8.8 12.0 Entropy index 1 ( 100) 33.1 5.2 3.6 5.5 9.8 14.6 Gini index (%) 38.7 1.8 1.3 1.9 3.7 5.1 Headcount index (P0) 12.4 45.6 11.8 36.0 63.1 112.2 Poverty gap index (P1) 2.9 51.0 18.6 45.7 82.1 150.5 Poverty severity index (P2) 1.2 50.1 23.4 50.8 88.5 163.1 Source: Results from the authors' micro simulation module using changes in prices, wages, and occupational choices by segment generated by the computable general equilibrium module. Note: Base values for BASE column and percentage change for other simulations; SIMELN El Niño drought; SIMDEV real devaluation; DEVCCF real devaluation foreign credit crunch; FINCRI real devaluation foreign credit crunch domestic credit crunch; SIMALL real devaluation foreign credit crunch domestic credit crunch El Niño drought. a. Per capita income is total monthly income. b. Rp rupiah, Indonesia's official currency. 113 114 ROBILLIARD, BOURGUIGNON, AND ROBINSON divide is apparent in terms of poverty changes, because urban poverty increases by 301.4 percent and rural poverty increases by only 112.2 percent. This can be explained by differential income shocks in the urban and rural sectors. Results also show that the inequality indicators increase in both sectors. Conclusion The income changes generated by the new macro-micro framework introduced in this chapter (drawn from a sample of households in an Indonesian household survey) are consistent, once they have been aggregated, with the predictions of a multisector CGE-like macro model. Chapter 4 shows that this framework captures important channels through which the 1997 financial crisis affected household incomes in Indonesia. This result is obtained through an explicit representation of the actual combination of different income sources within households and how this combination may change--through desired or undesired modifications in the occupational status of household members. Compared with standard CGE, or before-and-after analysis, the framework developed in this chapter allows for an original analy- sis of the distributional effects of a financial crisis like the one that struck Indonesia in 1997. At the macro level, the analysis shows that the credit crunch was an important force behind the collapse of GDP in Indonesia, while the devaluation (combined with increases in the marketing cost of food) appears to be the primary driving force behind increases in the relative prices of food with respect to nonfood commodities. At the micro level, heterogeneity of households (with respect to factor endowments), consumption behavior, and occupational choices, whether free or forced, prove to be important in explaining the poverty and distribution effect of the crisis. These are pure simulations intended to be consistent with what was observed in aggregate terms in Indonesia--and cannot be com- pared with actual data at the microeconomic level. Under these con- ditions, it is difficult to say that one simulation or methodology is better than another. The appeal of the framework developed in this chapter is that it accounts for realistic shocks on household eco- nomic conditions, especially with regard to the occupational status of household members. That it does so in a way that is selective, across household types, is also appealing--as suggested by the casual observation of household conditions during crisis periods. The main problem, however, is that this selectivity is essentially introduced by SOCIAL IMPACT OF A FINANCIAL CRISIS 115 translating observed cross-sectional differences in household income generation behavior into the time dimension. In other words, the simulation methodology presented in chapter 4 relies on the stan- dard assumption in economics that a household that faces specific conditions of crisis in a future labor market will behave like a house- hold that is observed under those same current conditions. Deter- mining whether this is justified could be accomplished only with panel data--and so is left for future work. Notes The authors are grateful to Indonesia's Badan Pusat Statistik (BPS). They also thank Vivi Alatas for valuable help with the data and programming; and thank Benu Bidani, Dave Coady, Gaurav Datt, Tamar Manuelyan Atinc, Emmanuel Skoufias, and Jaime de Melo for comments and helpful discus- sions on the text. Other useful comments were made by seminar partici- pants at the International Food Policy Research Institute (IFPRI), the World Bank, The Institute for Economic and Social Research (LPEM) at the University of Indonesia, the University of Nottingham in the United King- dom, the Annual World Bank Conference on Development Economics in Europe (ABCDE-Europe 2002), and Développement Institutions et Analy- ses de Long terme (DIAL) in Paris. 1. Results from these International Labour Organization and Central Bureau of Statistics reports are taken from Booth (1998). 2. Starting with Mookherjee and Shorroks's (1982) study of the United Kingdom. 3. A detailed comparison of the approach used in this chapter with the representative household group approach is presented in a companion paper (see Bourguignon, Robilliard, and Robinson 2005). 4. A tighter integration of the micro and macro models has been attempted within a simpler framework by Cogneau (2001) and Cogneau and Robilliard (2001) and applied to Madagascar (as discussed in chapter 7 of this volume). For a general discussion of the link between CGE model- ing and micro-unit household data, see Plumb (2001). 5. A more general discussion of the model can be found in Bourguignon, Ferreira, and Lustig (1998) and Bourguignon, Fournier, and Gurgand (2001). 6. The model also considers the possibility that a person may have con- current income from both wage work and self-employment. This is taken as an additional alternative in the discrete choice model--equation (4.5). A dummy variable controls for this in the earning equation (4.1), and this per- son is assumed to count for half of a worker in the definition of Nm. To sim- plify presentation, the authors do not insist on this aspect of the data (or the model). See Alatas and Bourguignon (2005). 116 ROBILLIARD, BOURGUIGNON, AND ROBINSON 7. This rationing interpretation of the functioning of the labor market leads to reinterpreting the "utility" function--defined in equations (4.5) and (4.6)--as a combination of both utility aspects and the way in which the rationing scheme depends on individual characteristics. 8. For the Jacobian used in the Gauss-Newton method to make sense in the present framework, the number of households and the dispersion of their characteristics must be sufficiently high. If this were not the case, then the discontinuity implicit in the Ind( ) functions would create problems. 9. This representation of the output effect of the crisis fits the analysis made by Stiglitz. See, for example, Furman and Stiglitz (1998). 10. Azis, Iwan J., Erina E. Azis, and Erik Thorbecke. 2001. "Modeling the Socio-Economic Impact of the Financial Crisis: The Case of Indonesia." Ithaca, NY: Cornell University Press. (photocopy) 11. In particular, no attempt was made to reconcile the household survey data with the national accounts data. 12. To be consistent with the latest available estimates of the poverty headcount for 1996, the percentage changes reported by Suryahadi and others (2000) between 1996 and 1997 are applied to the base value com- puted by Pradhan and others (2000). This generates an estimate of the poverty headcount of 9.7 percent in 1997. The present authors then chose an income poverty line that would generate the same headcount for the sample and used that poverty line as the reference value. 13. Because the SAKERNAS does not permit deriving the evolution of self-employment income for agricultural and nonagricultural activities, in this historical simulation, self-employment incomes were assumed to decrease in real terms by the same magnitude as unskilled male wages in the urban and rural sectors. References Adelman, Irma, and Sherman Robinson. 1989. "Income Distribution and Development." In Handbook of Development Economics, eds. Hollis Chenery and T. N. Srinivasan, Vol. II, 949­1003. Amsterdam: Elsevier Science Publishers. Alatas, V., and François Bourguignon. 2005. "The Evolution of Income Distribution during Indonesia's Fast Growth, 1980­96." In The Micro- economics of Income Distribution Dynamics in East Asia and Latin America, eds. François Bourguignon, Francisco H. G. Ferreira, and Nora Lustig. Washington, DC: Oxford University Press and World Bank. Armington, Paul S. 1969. "The Geographic Pattern of Trade and the Effects of Price Changes." International Monetary Fund Staff Papers 16 (2): 179­201. SOCIAL IMPACT OF A FINANCIAL CRISIS 117 Booth, Anne E. 1998. "The Impact of the Crisis on Poverty and Equity." ASEAN Economic Bulletin 15 (3): 353­62. Bourguignon, François, W. Branson, and Jaime de Melo. 1992. "Adjust- ment and Income Distribution: A Micro-Macro Model for Simulation Analysis." Journal of Development Economics 38 (1): 17­40. Bourguignon, François, Francisco H. G. Ferreira, and Nora Lustig. 1998. "The Microeconomics of Income Distribution Dynamics." A Research Proposal. The Inter-American Development Bank and the World Bank, Washington, DC. Bourguignon, François, Martin Fournier, and Marc Gurgand. 2001. "Fast Development with a Stable Income Distribution: Taiwan, 1979­1994." Review of Income and Wealth 47 (2): 139­63. Bourguignon, François, Anne-Sophie Robilliard, and Sherman Robinson. 2005. "Representative versus Real Households in the Macroeconomic Modeling of Inequality." In Frontiers in Applied General Equilibrium Modeling: In Honor of Herbert Scarf, eds. Timothy J. Kehoe, T. N. Srinivasan, and John Whalley. Cambridge, UK, and New York: Cambridge University Press. Central Bureau of Statistics (CBS). 1998. "Perhitungan Jumlah Penduduk Miskin dengan GNP Per Kapita Riil." Jakarta, Indonesia: CBS. Cogneau, Denis. 2001. "Formation du revenu, segmentation et discrimina- tion sur le marché du travail d'une ville en développement: Antananarivo fin de siècle." DIAL Working Paper 2001-18. Paris: Développement Institutions et Analyses de Long terme. Cogneau, Denis, and Anne-Sophie Robilliard. 2001. "Growth, Distribu- tion, and Poverty in Madagascar: Learning from a Micro-simulation Model in a General Equilibrium Framework." Working Paper 61. Inter- national Food Policy Research Institute (IFPRI), Washington, DC. Dervis¸, Kermal, Jaime de Melo, and Sherman Robinson. 1982. General Equilibrium Models for Development Policy. New York: Cambridge University Press. Frankenberg, Elizabeth, Duncan Thomas, and Kathleen Beegle. 1999. "The Real Costs of Indonesia's Economic Crisis: Preliminary Findings from the Indonesian Family Life Surveys." RAND Document DRU-2064- NIA/NICHD. RAND Corporation, Santa Monica, CA. Furman, Jason, and Joseph E. Stiglitz. 1998. "Economic Crises: Evidence and Insights from East Asia." Brookings Papers on Economic Activity 1998 (2): 1­114. International Labour Organization (ILO). 1998. Employment Challenges for the Indonesian Economic Crisis. Jakarta, Indonesia: ILO Regional Office for Asia and the Pacific and the United Nations Development Program. Manning, Chris. 2000. "Labour Market Adjustment to Indonesia's Economic Crisis: Context, Trends, and Implications." Bulletin of Indonesian Eco- nomic Studies 36 (1): 105­36. 118 ROBILLIARD, BOURGUIGNON, AND ROBINSON Mookherjee, Dilip, and Anthony F. Shorrocks. 1982. "A Decomposition Analysis of the Trend in U.K. Income Inequality." Economic Journal 92 (368): 886­902. Plumb, Michael. 2001. "Empirical Tax Modeling: An Applied General Equilibrium Model for the U.K. Incorporating Micro-unit Household Data and Imperfect Competition." PhD Thesis, Nuffield College, University of Oxford. Pradhan, M., A. Suryahadi, S. Sumarto, and L. Pritchett. 2000. "Measure- ments of Poverty in Indonesia: 1996, 1999, and Beyond." SMERU Working Paper. Social Monitoring & Early Response Unit Research Institute, Jakarta, Indonesia. Robinson, Sherman. 1989. "Multisectoral Models." In Handbook of Devel- opment Economics, eds. Hollis Chenery and T. N. Srinivasan, Vol. II, 885­947. Amsterdam: Elsevier Science Publishers. Robinson, Sherman, A. Cattaneo, and M. El-Said. 2001. "Updating and Estimating a Social Accounting Matrix Using Cross Entropy Methods." Economic Systems Research 13 (1): 47­64. Suryahadi, A., S. Sumarto, Y. Suharso, and L. Pritchett. 2000. "The Evolution of Poverty during the Crisis in Indonesia, 1996­99." SMERU Working Paper. Social Monitoring & Early Response Unit Research Institute, Jakarta, Indonesia. World Bank. 1998. Indonesia in Crisis: A Macroeconomic Update. Washington, DC: World Bank. 5 Can the Distributional Impacts of Macroeconomic Shocks Be Predicted? A Comparison of Top-Down Macro-Micro Models with Historical Data for Brazil Francisco H. G. Ferreira, Phillippe G. Leite, Luiz A. Pereira da Silva, and Paulo Picchetti This chapter analyzes the predictive performance of a top-down macro-micro simulation model in reproducing the impact on indi- vidual and household income of a large macroeconomic shock, such as an exchange rate and currency crisis. It compares model simula- tion results with actual distributions. Currency and financial crises such as those experienced in Mexico during 1996, in various coun- tries in East Asia between 1997 and 1999, in Russia during 1998, and in Brazil during 1999 can have devastating effects on both gov- ernment budgets and private sector balance sheets. But such macro- economic shocks do not affect all households alike. Occupational structures across the labor market respond to changes in relative prices and to new expenditure aggregates. The distributions of earn- ings generated in those labor markets also respond to these changes, and thus the distributions of household income per capita--even 119 120 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI when aggregated measures of poverty and inequality appear to be minimal or not affected. There is now an established typology of the common elements that precede and cause financial crises in emerging markets,1 and there is a growing body of literature documenting the impact of dif- ferent types of crises and shocks on poverty and inequality.2 Much less progress has been made on understanding the actual transmis- sion mechanisms through which aggregate shocks affect individual incomes and occupations across the economy in a way that would help policy makers.3 Given its general equilibrium nature, this problem has traditionally been approached through computable general equilibrium (CGE) models, in which all individuals and households in an economy are lumped together into a much smaller number of representative household groups (RHGs). See, for exam- ple, chapter 1 in this volume or Adelman and Robinson (1988). Although the literature applying CGEs to developing countries has generated a number of useful insights, the use of CGEs in address- ing distributional questions has been particularly problematic. As discussed in the introduction to this volume, CGE/RHG models are limited because changes in individual occupations and earnings can be very heterogeneous--even within the sectors of economic activ- ity and skill levels traditionally used to construct the RHGs. In chap- ter 4, Robilliard, Bourguignon, and Robinson show, for the case of Indonesia in 1998­99, that poverty and distributional effects simu- lated using RHGs can be different from those effects simulated on disaggregated real households. There are two novelties in this chapter when compared with chapter 4. First, this analysis uses a different type of macro model on "top." The macro model used is based on a set of investment savings and liquidity preference money supply (IS-LM) equations estimated econometrically on time-series data--not a typical CGE model calibrated with ad hoc parameters. As many parameters as possible are obtained, using time-series national accounts and aggre- gated household survey data from Brazil for 1981­2000. This "top" is then linked to a microeconomic simulation model of household income formation, estimated on cross-section data from a house- hold survey, at the "bottom." Second, and most important, the counterfactual model simulation results for 1999 are compared with the actual changes revealed by the 1999 household survey data, thus providing the first rigorous test (known to the authors) on the per- formance of a top-down macro-micro model against real data. This is possible for Brazil given the annual frequency and comparability of the household survey data (the Pesquisa Nacional por Amostra de Domicílios or PNAD). CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 121 More specifically and going further, this chapter compares three types of model predictions with the observed (ex post) impact of the crisis. This model operates on two levels: the macro general equilib- rium model of the economy on top and the reduced-form household income determination model on the bottom. Linking the two are the key linking aggregated variables (LAVs) that represent the price, wage, and employment vectors generated by the macro model. The framework described in figure 5.1 can be used to construct three types of experiments to assess the predictive performance of this top-down model. The first experiment is designed to define the counterfactual income distribution that would arise from an RHG approach. To distinguish errors from the RHG assumption from those arising from the macro model, this analysis uses on top (as LAVs that should come out of the top macro model) the historically observed--or actual, rather than predicted--LAVs. Average actual values (for each RHG) of the LAVs are imputed to individual households in the 1998 data set, as chapters 1, 2, and 3 in this volume would do in what might be called a micro accounting approach. In the second experiment, the same observed LAVs are used, but the disaggregated micro simu- lation model is now used to simulate changes to the income of indi- vidual households. In the third experiment, the results of simulations using the macro model (LAVs) are combined with the simulations derived from the micro model. The three experiments can be summa- rized as shown in table 5.1. When compared with the actual 1999 distribution, the results of these three experiments help to identify Figure 5.1 A Simplified Overview of the Top-Down Macro-Micro Framework Top Level: Macro General equilibrium macroeconomic model with sectoral disaggregation to model factor markets Linking aggregated variables (LAVs) Bottom Level: Micro Household occupational choice model Household income determination model Source: Authors' depiction. 122 Table 5.1 An Overview of the Three Experiments Conducted Linking aggregated Experiment Top-level macro model variables (LAVs) Bottom-level micro model 1. Representative No macro simulation LAVs: actually observed No micro simulation: household group changes of average each individual receives (RHG) income and employment the average income and for each RHG employment change of the RHG he/she belongs to 2. Pure micro simulation No macro simulation or No LAVs Pure micro simulation: using the household "perfect disaggregated micro model runs so that income micro simulation macro model" using its average results for each model actually observed average RHG converge to the actual income and employment observed average income change of the economy's and employment change RHGs of the economy's RHGs 3. Full macro-micro Macro simulation: macro LAVs: simulated Micro simulation: micro linking model model runs to replicate changes of average model runs so that its average the 1999 financial crisis income and employment results for each RHG for each RHG converge to the simulated average income and employment change of the model's RHGs Source: Authors' depiction. CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 123 possible sources of discrepancy for each level of the framework (the top macro and the bottom micro). This exercise admittedly contains an array of perils and pitfalls. Perhaps most important, the parameters in reduced-form macro models usually conflate "deep" taste and technology parameters with policy parameters, and are thus subject to the Lucas critique that estimates obtained under a certain policy regime may no longer be valid under another. But there are other issues: national accounts data used to estimate the macro model may be at odds with the aggregated picture arising from the household survey data used to estimate the micro model; assumptions about labor market closures are inevitably oversimplifications of a much more complex reality and are likely to involve a search-driven equilibrium unemployment rate; and so forth. This chapter pursues this approach, despite these serious data and methodological limitations, because of the sheer importance of the question. The ability to predict, with some degree of confidence, the direction and magnitude of the impacts of large macroeconomic events (shocks or policy changes) on the incomes and occupations of individuals across the income distribution would be a major asset to policy making in a number of countries, particularly those unfortunate enough to have the combined characteristics of being both volatile-- orshockprone--andpoor.Althoughthischapterdoesnotfullyachieve that objective, comparisons of the model predictions with the actual data, and the decomposition of the errors into elements attributable to each of the macro and micro components, may be useful to other applied researchers working on this important question. The chapter is organized as follows. The following section briefly describes the event to be modeled--the 1998­99 currency crisis (with the floating and devaluation of the Brazilian real)--and the structure of the macro model used for Brazil. The procedure to gen- erate LAVs and the precise scenario of the currency crisis that is sim- ulated are then presented, followed by a discussion of the micro model of income determination at the household level and the results of the micro simulation based on them. The chapter concludes with a discussion of micro model accuracy, the simulations, and concluding remarks. The 1998­99 Currency Crisis and the Macro Model The macro model outlined in this section was designed to simulate the macro shock corresponding to the macroeconomic policy pack- age implemented in Brazil in 1998­99 responding to an exchange 124 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI rate and currency crisis. It consisted chiefly of the abandonment of Brazil's crawling-peg exchange rate regime (ERR) and related fiscal adjustment measures. During most of the first period of the four-year Real Plan (July 1994­January 1999), Brazil maintained a crawling-peg ERR. After the Asian crises in 1997­98, the crisis in Brazil began around the third and fourth quarters of 1998, with pressure on the pegged exchange rate coming from capital outflows. The pressure contin- ued during the first quarter of 1999, after the floating of the Brazil- ian real. The policy response included--among other less salient policies--changes in the following variables: · The "float" of the currency on January 15, 1999, whose aver- age annual parity with the U.S. dollar went up from R$1.161 (annual 1998 average) to US$1, to R$1.816 (annual 1999 average, corresponding to a 56.4 percent nominal devaluation). · A temporary rise in the central bank policy rate (the Banco Cen- tral or BACEN's Selic) from October 1998 until May 1999. The monthly rate was raised from 1.47 percent in August 1998 to 3.33 per- cent in March 1999 (corresponding to annualized rates of almost 50 percent). However, thanks to the rapid resolution of the crisis, the annual average base nominal rate in 1999 actually ended up lower than in 1998. In nominal terms, the Selic was set in 1997, 1998, and 1999 at 24.8 percent, 28.8 percent, and 25.6 percent, respectively, corresponding to real average rates of 16.1 percent, 26.6 percent, and 4.7 percent, respectively. · A renegotiation of the terms of Brazil's Stand-By Arrangement (SBA) with the International Monetary Fund (IMF) to strengthen the credibility of the policy framework, and hence tighten the fiscal stance corresponding to a reduction of the consolidated public sector borrowing requirements (PSBRs), from R$68 billion to R$56 billion (7.5 percent of gross domestic product [GDP] down to 5.8 percent of GDP, that is, a cut of R$12 billion, or 1.7 percent of GDP). · The implementation of an inflation target anchor in 1999 to replace the exchange rate anchor for inflation expectations; the provision of hedge-to-market participants (through the issuance of government foreign exchange­indexed domestic bonds); and undis- closed occasional interventions on the spot market, drawing on international reserves, within the limits agreed on under the new SBA arrangement with the IMF. The goal of this analysis is to model this event at the macro level in the simplest possible way, consistent with the objective of generating LAVs for wage, price, and employment variables that can be applied to the microeconomic simulation model using household data. At the CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 125 bottom or household level, using the household survey database, the model will transmit those average changes across households, with a view to predicting occupational and distributional impacts. At the top of the framework, a conventional IS-LM macroecono- metric general equilibrium model is used, but with a disaggregated labor market and a financial sector. The model is estimated on time- series data, and some equations are specified with a dynamic speci- fication (lags, for example) that allows for some dynamic in the solution of the model. Data availability issues imposed some con- straints on the choice of estimators.4 In spite of these inevitable constraints, several equations are estimated using two-stage least squares when endogeneity of the regressors was considered to be particularly likely. The parameters of the model are estimated on 1981­2000 annual data, both from the national accounts and from a time-series of averages from the PNAD household surveys, which have been fielded annually by the Brazilian census bureau (Instituto Brasileiro de Geografia e Estatística, IBGE) since 1976.5 The basic layout of the macro model is a disaggregated but still standard IS-LM framework (as in the Klein-Goldberger model, MPS6 in the United States, or DMS and METRIC7 in France; also see Artus, Deleau, and Malgrange 1986).8 The functioning of these large mod- els is complex but can be reduced to the interaction of three basic modules. First, a real economy module determines production, com- ponents of aggregate demand (such as private consumption and investment), and factor demand (discussed in the following section). Then, a wage-prices module determines the aggregate price level, wage rates, and labor market characteristics. And finally, a financial and monetary module determines the interest rate and equilibria in asset markets. Because of its modular structure, the macro model can function under various configurations (for example, by assuming that some of the modules can be "frozen" and thus actual or exoge- nous values for its components can be used instead). Or instead of modeling a variety of financial assets equilibria, the financial asset market can be reduced to just one local currency market. The key transmission mechanism in the macro model--between the real economy and the financial markets--comes from the linking of real private consumption and investment to the endogenous domes- tic interest rate, which is determined by equilibrium in the financial sector. In particular, (1) real private consumption is a standard func- tion of disposable income, the general price index, and the real deposit interest rate (to account for a "wealth" or portfolio effect); and (2) real private investment is decomposed into its building and con- struction versus machinery components. Both components follow a standard specification, including aggregate demand (an accelerator) 126 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI and the price of capital, decomposed into the real exchange rate (given the importance of imported equipments) and the domestic working capital interest rate. In a nutshell, a higher base policy (central bank) interest rate would achieve the following: increase real domestic inter- est rates, lower private consumption and investment, lower the cur- rent account deficit, increase demand for domestic financial assets, and thus put pressure for an appreciation of the nominal exchange rate after an adjustment of the current account of the balance of pay- ments (BoP). The workings of the model respond to the standard sta- bilization package implemented during most, if not all, of the exchange rate and currency crises of the 1990s. In addition, the BoP is modeled in a fairly detailed way, with real services, real imports, and exports of goods disaggregated into major types of commodities and services. The general specification for all these items makes each of them dependent (respectively for debit and credit components) on domestic and external demand and relative prices, that is, the real exchange rate. The current account balance is constructed by accounting identities. Capital movements follow uncovered interest parity conditions and are assumed to depend on only the interest rate differential, the expected depreciation of the exchange rate, and country risk. His- torical simulations in the current version of the macro model are based on an exogenous nominal exchange rate that is compatible with the two regimes that recently prevailed in Brazil.9 Details of the macro model, including the exact specification of each equa- tion in each module and the estimation results, are not presented here, but they can be found in Pereira da Silva, Picchetti, and Samy de Castro (2004). Factor Markets in the Real Economy Model AGGREGATE SUPPLY AND DEMAND MODELING STRATEGIES The main motivation behind the breakup of supply into different sectors is the ability of the macro model to differentiate the effects of macro and external shocks on different types of products--and on the workers producing them. Accordingly, the supply side in the model is divided into six sectors: · urban tradable formal (UTF) · urban nontradable formal (UNF) · urban nontradable informal (UNI) · rural tradable formal (RTF) · rural nontradable formal (RNF) · rural nontradable informal (RNI) CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 127 For each of these sectors, production is modeled as value added, and factor demand functions are derived from factor price equals mar- ginal product conditions. FACTOR MARKETS Factor demand functions determine the demand for capital and for labor by skill level. To relate the employment and earnings predic- tions of this level of the model to the household survey data used in the micro simulation stage, the classification of workers by skill level had to be made in terms of observed characteristics for the individuals. Skill level was defined according to years of formal education as reported in the PNAD. Low-skill workers have between 0 and 4 years of formal education, whereas intermediate-skill work- ers have between 5 and 11 years and high-skill workers have more than 11 years. The demands for these different types of labor and for capital are derived by equating factor prices to the marginal products from the production functions for each of the six sectors, which are represented by a three-level nested constant elasticity of substitution (CES) model. The motivation for this approach is to provide for flexibility in the rates of substitution between capital and labor, and between the dif- ferent types of labor. The first level of the CES allows for substitution between capital and a composite measure of labor. In the second level, this composite labor can be decomposed between skilled and unskilled jobs. The third level accounts for the fact that unskilled jobs can be performed by either low-skill or intermediate-skill workers, whereas skilled jobs can be performed by either intermediate-skill or high-skill workers. Therefore, as in Fernandes and Meneses-Filho (2001), it is assumed that there is substitution between all types of labor, except between high-skill and low-skill labor. The production function for each one of the six sectors can then be represented as follows: (5.1) y K (1 )La 1 , where (5.2) La LQ (1 )LU 1 (5.3) LQ LiQ e (1 )Lhe 1e (5.4) LU Ll (1 )LiU 1 ; K capital; La composite labor; LQ composite labor for skilled jobs, can be performed either by Li (intermediate-skill workers) or by Lh (high-skill workers); LU composite labor for unskilled 128 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI jobs, can be performed either by Li (intermediate-skill workers) or by Ll (low-skill workers). 1 ka 1 1 QU 1 (5.5) Elasticities of substitution: 1 ih 1 1 . li 1 Having defined factor demand functions based on these sectoral pro- duction functions, the authors turn to factor supply functions. These functions are separately specified by skill level (in the case of labor) and by economic sector (for both labor and capital). Labor supply is assumed to be perfectly inelastic for each skill group and to corre- spond to the economically active population (EAP) in those groups. Factor market equilibrium conditions are obtained by simultaneously solving equation pairs Ldij (w) Ui Lij s for each wij f (Ui, . . .) labor type i and sector j. In each of these pairs, the second equation is a wage curve, which relates equilibrium wage levels to skill- specific unemployment levels. This setup generates 48 equations for 30 unknowns: 24 endogenous factor prices and 6 endogenous skill- specific unemployment rates.10 A detailed degree of disaggregation is a requirement for a model that purports to focus on occupational and distributional consequences of shocks. The solution to the system generates most of the LAVs required for transmission to the micro model. Specifically, it generates 18 wage rates (three labor types in six sectors) and 21 occupation rates (six employment levels, and one unemployment level for each of three types of workers). The only missing LAVs now are con- sumption price aggregates. To obtain those, however, it is necessary to move from the factor markets to the product markets, and then incorporate the financial markets to derive an endogenous set of interest rates from the IS-LM equilibrium. The first step, while moving to the product markets, is to recognize that this modeling strategy implies that the number and definition of final demand sectors are different from those of production sectors. This assumption creates the issue of reconciling the output and price variables on both sides of the model, which is done through a conversion matrix. CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 129 CONVERSION MATRIX The approach discussed in this chapter follows Fisher, Klein, and Shinkai (1965). On the one hand, the output demanded by the final demand sectors must be distributed over the production sectors; on the other hand, the prices generated by the price-formation equa- tions in the production sectors must be aggregated to obtain prices for the final demand sectors. In this model, data are available on total output by formal production sector, derived from national accounts. But in the case of the informal sector, there is no readily available statistic on production. Therefore, value added for these sectors was estimated based on reported incomes from informal workers in the household surveys. The output conversion matrix was estimated with the natural restrictions (that nontradable sectors do not export, for example), and the resulting weights were used to convert sector prices into gross national product deflators. Figure 5.2 illustrates this macro model. The key elements are the transmissions from the standard IS-LM macro aggregates (depicted on the right-hand side of the figure) into the disaggregated labor market (on the left-hand side). The demand for each labor factor-- broken down by the three skills of the six sectors defined earlier (UTF, UNF, UNI, RTF, RNF, and RNI)--modeled in the framework comes from the appropriate disaggregation of output from the stan- dard IS-LM macro model, taking into account the role of private disposable (after tax) incomes, government financing needs (tax rev- enue versus debt and other payments), demand elsewhere in the world (the BoP), and central bank policies (setting the base BACEN Selic rate) that affects both the government's financing needs and private aggregates (consumption and investment). These demands (sectoral outputs) are depicted in figure 5.2 by arrows that extend from the aggregation matrix to the sectoral boxes depicted above them. The LAVs of the macro model are then produced as explained and depicted by the arrows that link the sectoral boxes and the "Labor market" box. The LM Curve and the Financial Sector Like employment levels and factor prices, product prices and interest rates are also endogenous in the macro model and related to the IS-LM framework. A financial sector was modeled with several agents and markets, roughly following Bourguignon, Branson, and de Melo (1989). Specifically modeling a financial sector was necessary to refine the transmission of financial crises--external and domestic--to the rest of the model and, in particular, to the disaggregated sectoral demand for labor and to the real macro variables (real private con- sumption and investment). Indeed, during financial crises, many 130 Figure 5.2 An Overview of the Main Blocks of the Macro Model Labor Labor Taxes market income UTF UNF UNI RTF RNF RNI Taxes Transfers Disposable Loans income RTF_K Government UNI_K RNF_Y Payments RNF_K UNF_Y UNI_Y RTF_Y RNI_K Financial sector UNF_K UTF_K UTF_Y RNI_Y Loans Central Government consumption bank Aggregation Exports matrix Imports Private consumption Private consumption Payments Reserves Foreign sector Source: Authors' depiction. Note: RNF rural nontradable formal; RNI rural nontradable informal; RTF rural tradable formal; UNF urban nontradable formal; UNI urban nontradable informal; UTF urban tradable formal. CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 131 traditional policy instruments lose their ability to affect the behavior of households, firms, investors, and banks, both domestic and foreign. This exercise attempts to capture some of these episodes and their characteristics--despite the fact that the model's simulation has an annual periodicity, while most of the manifestations of these crises are infra-annual. Finally, this modeling strategy is modular, allowing the financial sector's part of the framework to be activated fully or in part. In particular, two domestic interest rates in local currency are endogenous in the model: the domestic deposit rate for household deposits, and the domestic borrowing rate for firms (working capi- tal interest rate). Two other interest rates are exogenous: one national (the Selic, Brazil's Central Bank policy rate) and one for- eign (a short-term London interbank-offered interest rate in U.S. dollars). Brazil experienced an abrupt change in its ERR during 1999, moving from an exchange rate anchor to a floating exchange rate with inflation targeting and a corresponding change in empha- sis for the Selic. This regime change was integrated for simulations in which the inflation-targeting objective prevails. Once the base policy rate is set, the structure of interest rates (domestic borrowing and deposit rates) is determined by modeling the spreads (see Car- doso 2003; Favero and Giavazzi 2002). Spreads over the base real policy rate will determine the real deposit rate given the supply of bank deposits by households and their first-layer choice between local and foreign currency. Spreads over the deposit rate will deter- mine the real working capital rate given the supply of new credit by commercial banks after their first-layer choice for government domestic bonds and the demand for new credit by firms. This part of the model is important to capture the short-term effect of hikes in the country's base policy rate that can result from either (1) the need under a pegged exchange rate to defend the regime by matching the rise in the country-risk premium and the expected devaluation, or (2) the need under a floating ERR with inflation targeting to estab- lish the credibility of the anti-inflation stance of the central bank. Estimation and Standard Results of the Macro Model As mentioned earlier, the macro model comprises equations esti- mated by ordinary least squares on time-series data.11 The perfor- mance of the macro model (for example, multipliers and deviations from a base in the case of standard simulation exercises of fiscal and monetary shocks) is comparable to that of the macro models of industrial economies (France and the United States), but its invest- ment multiplier is much weaker. These results are summarized in table 5.2. 132 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI Table 5.2 Standard Multipliers of the Macro Model Compared with Other Macro Models ( IM / G) Model Y / G C / G I / G ( X / G) PPSC model (Brazil) 1.31 0.27 0.29 0.25 French models DMS 1.10 0.31 0.35 0.55 Metric 1.38 0.31 0.66 0.60 PITI 1.54 0.22 0.82 0.48 DECA 1.26 0.37 0.58 0.69 U.S. models Brookings 2.79 1.11 0.82 0.14 HC 1.74 0.31 0.53 0.10 Sources: Artus and Muet 1980; Pereira da Silva, Picchetti, and Samy de Castro 2004. With regard to the counterfactual simulation of the 1999 cur- rency crisis, the macro model fares reasonably well in a historical simulation mode. Tables 5.3 and 5.4 present partial results (1995­99), extracted from a dynamic historical simulation for 1986­99. The run captures the major consequences of the crisis, such as the slowdown in real private consumption and the fall in real disposable income and private investment, which explain the modest real GDP growth rate (0.8 percent) in 1999. The major com- ponents of the external sector balances are reasonably well captured by the simulation. The Brazilian currency crisis is milder than the large output contraction that characterized other financial crises experienced elsewhere during the late 1990s. The historical simulation also captures the stabilization period under the Real Plan (mid-1994 to 1998) in its real and price/monetary manifestations. The major feature of the period--consumer price index (CPI) inflation brought down from more than 2,200 percent per year in 1994 to 8.5 percent in 1997--is portrayed by the consistent fall in inflation measured by several price indexes (gen- eral price index, INFL_GPIF; wholesale price index, INFL_WPI; and GDP deflator, INFL_DEF_AGG_Y). Deflation, however, did not ignite sustainable aggregate growth that remained erratic as captured by real GDP growth, total gross fixed capital formation (FBK_TOTAL_REAL_GROWTH), and private consumption real growth (HHS_CONS_REAL_GROWTH). In fact, deflation and the real appreciation of the Brazilian real (seen in the upward trend of the real exchange rate, RER_DEV) until late 1998 produced an increasing deterioration in both the trade (BOP_TB) and the current account (BOP_CA) balances. Despite a growing ratio of tax revenues Table 5.3 Some Results of the Macro Model, Historical Simulation for 1999 Indicator 1995 1996 1997 1998 1999 Real GDP growth (percent) Actuals 4.2 2.7 3.3 0.1 0.8 Baseline 4.7 1.9 0.3 0.7 0.5 FBK_TOTAL_REAL_GROWTH (percent) Actuals 7.3 1.2 9.3 0.3 7.2 Baseline 5.2 3.6 0.2 2.5 6.1 HHS_CONS_REAL_GROWTH (percent) Actuals 4.8 5.6 3.3 1.5 0.5 Baseline 5.2 5.0 0.9 0.2 0.8 XBSZN (R$ thousands) Actuals 49,916,655 54,430,127 65,356,311 67,862,415 100,135,527 Baseline 47,834,680 57,002,630 65,265,190 70,423,570 103,217,900 MBSZN (R$ thousands) Actuals 61,314,054 69,310,584 86,000,488 87,768,795 115,153,991 Baseline 59,486,390 68,346,620 85,603,290 84,773,800 123,693,100 BOP_CA (US$ millions) Actuals 18,382 23,442 30,555 33,435 25,335 Baseline 18,173 20,762 29,613 26,205 29,205 BOP_TB (US$ millions) Actuals 3,464 5,539 6,856 6,594 1,199 Baseline 2,660 1,045 7,317 2,091 3,608 AGG_NH_L (units of workers) Actuals 4,076,944 4,298,040 4,590,574 4,924,643 4,975,319 Baseline 4,073,141 4,279,046 4,544,963 4,918,456 5,080,301 AGG_NI_L (units of workers) Actuals 22,762,114 24,027,848 24,756,877 25,971,134 27,121,571 Baseline 22,245,900 24,340,140 25,658,480 25,778,510 26,534,570 AGG_NL_L (units of workers) Actuals 27,075,090 24,592,552 24,886,158 24,110,653 23,777,927 Baseline 26,260,380 24,462,200 25,116,370 24,350,920 23,894,390 133 Source: Authors' estimates based on Pereira da Silva, Picchetti, and Samy de Castro (2004). Note: FBK_TOTAL_REAL_GROWTH growth of total gross fixed capital formation; HHS_CONS_REAL_GROWTH growth of private consumption real growth; XBSZN exports of goods and nonfactor services; MBSZN imports of goods and nonfactor services; BOP_CA current account; BOP_TB trade account; AGG_NH_L total skilled labor force; AGG_NI_L total semi-skilled labor force; AGG_NL_L total unskilled labor force. Currencies: R$ Brazilian reais; US$ United States dollars. 134 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI Table 5.4 Major Results of the Public Sector and Financial Sector Modules, Historical Simulation for 1999 Indicator 1995 1996 1997 1998 1999 AGG_TAX_TRIB_REAL (percentage of GDP) Actuals 18.65 17.96 18.19 18.48 20.37 Baseline 17.08 18.27 20.18 19.67 19.50 AGG_TAX_INSS_REAL (percentage of GDP) Actuals 5.92 7.13 6.01 7.37 7.34 Baseline 6.04 7.22 6.19 7.46 7.40 AGG_TAX_OTH_REAL (percentage of GDP) Actuals 3.92 3.50 4.37 3.46 3.33 Baseline 4.00 3.54 4.50 3.50 3.36 CARGA (percentage of GDP) Actuals 28.44 28.63 28.58 29.33 31.07 Baseline 27.12 29.03 30.86 30.63 30.26 FIN_CG_INTPAY_Y (percentage of GDP) Actuals 2.90 2.93 2.36 5.95 9.13 Baseline 4.40 3.13 2.27 5.96 7.33 FIN_PS_PRIM_Y (percentage of GDP) Actuals 0.27 0.09 0.95 0.01 3.19 Baseline 1.59 0.31 0.24 0.67 2.31 FIN_PS_INTPAY_REAL (percentage of GDP) Actuals 96,325 80,037 74,040 115,172 181,344 Baseline 119,859 83,422 69,630 111,567 146,627 FIN_GG_DBT_DOM (R$ millions) Actuals 136,904 206,068 261,842 317,212 394,441 Baseline 168,963 182,763 238,856 320,860 383,905 FIN_CG_DBT_DOM_Y (percentage of GDP) Actuals 9.84 14.40 16.79 20.92 22.07 Baseline 13.34 10.75 15.80 22.24 21.93 INFL_WPI (percent) Actuals 58.77 6.33 8.13 3.55 16.58 Baseline 66.53 8.32 8.66 5.89 15.25 INFL_DEF_AGG_Y (percent) Actuals 77.55 17.41 8.25 4.85 5.70 Baseline 76.91 17.68 7.83 5.03 6.36 CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 135 Table 5.4 (Continued) Indicator 1995 1996 1997 1998 1999 RER_DEV (percent) Actuals 7.22 5.43 0.82 1.37 35.27 Baseline 11.55 3.49 1.30 0.87 36.82 Real interest rate, certificates of deposit Actuals 32.40 15.60 15.60 25.30 4.40 Baseline 29.69 15.39 15.01 23.97 5.33 WC_REAL (average percentage points/month) Actuals 75.21 33.46 28.39 32.06 12.21 Baseline 75.21 33.46 28.39 32.06 12.21 SELIC_REAL (average percentage points/month) Actuals 33.40 16.50 16.10 26.60 4.70 Baseline 33.37 16.53 16.09 26.63 4.68 Source: Authors' estimates based on Pereira da Silva, Picchetti, and Samy de Castro (2004). Note: AGG_TAX_TRIB_REAL total tax burden; AGG_TAX_ INSS_REAL social security taxes; AGG_TAX_OTH_REAL other taxes; CARGA total tax burden; FIN_CG_INTPAY_Y central government interest payments; FIN_PS_PRIM_Y public sector primary fiscal result; FIN_PS_INTPAY_REAL public sector interest payments; FIN_GG_DEBT_DOM general government domestic debt; FIN_CG_DEBT_DOM_Y central government domestic debt; INF_WPI annual inflation, wholesale prices; INF_DEFL_AGG_Y annual inflation, GDP deflator; RER_DEV real exchange rate, depreciation; WC_REAL real interest on working capital; SELIC_REAL real central bank base rate. to GDP (CARGA), public sector fiscal primary surplus as a percentage of GDP (FIN_PS_PRIM_Y) was clearly insufficient until a turnaround in policy in 1999, which aimed to stabilize the government's domestic debt-to-GDP ratio (FIN_CG_DBT_DOM_Y). Following the standard models on currency crises, the risk of a change in market perception of the sustainability of the pegged real was clearly growing by the end of 1997 to mid-1998, particularly after the East Asian crises. The macro model also depicts the 1999 financial crisis reasonably well in historical simulation mode. Stocks, issuance, and holdings of the key financial asset (government domestic bonds, FIN_CG_ DBT_DOM) increase, and interest payments (FIN_PS_INTPAY_ REAL) jump. The model captures adequately the increases in domestic prices (general price index, the CPI, and the WPI) brought by the pass- through effect of the depreciation of the Brazilian real that follows its floating in January 1999. The change in the ERR resulted in an aver- age 56.4 percent depreciation of the average annual nominal exchange rate that, given the pass-through on domestic prices, translated into a 136 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI 35 percent real devaluation of the index (RER_DEV). The model cap- tures the fall of the domestic real interest rates that accompanied the surge in domestic prices after the crisis. Finally, the model overshoots slightly its imports projection--although the expected corrections in both the trade and current account balances are picked up. Generating the LAVs to Link the Macro and Micro Models As indicated earlier, the factor markets module of the macroeconomic model generates 20 LAVs for occupational status (three employment levels, by sector, and one unemployment level, for five area/skill com- binations), and 15 LAVs for incomes (the earnings rates in each sec- tor, in each of the five area/skill combinations).12 In addition, there are 6 LAVs for changes in the output prices of the six sectors. There are 41 LAVs in total. The LAVs were generated by area (urban and rural); by skills--low (0 to 4 years of schooling), intermediate (5 to 11 years of schooling), and high (12 or more years of schooling); and by occupational sector (tradable, nontradable, and informal). Tables 5.5 and 5.6 show the estimates produced using the macro model runs for 1999, the actual values observed at the PNAD 1998 and 1999 data, and the errors produced by the macro model. The storyline for the 1998­99 crisis is well known: the financial crisis resulted in an overall decline in urban employment across the country. Unemployment grew in both urban and rural areas and for all skill levels, but more markedly for high-skill workers. Informality also grew, particularly in urban areas. Formal employment fell across all skill groups in urban areas, and more markedly in nontradable sectors, as one would expect. In rural areas, however, the currency depreciation produced a positive output response leading to an increase in employment in the tradable sectors for all skill groups. Table 5.5 presents the occupational structure of the Brazilian population, aggregated by these three sectors and three skill groups, for both urban and rural areas. Column A shows absolute numbers and proportions for 1998, and column B shows the same informa- tion as actually observed in 1999, and calculates the actual changes between the two years, which is sometimes referred to as the "true LAVs." Column C presents the corresponding prediction results from the macro module for 1999. The entries in this column are counterfactual occupational numbers and shares, as predicted by the model, when calibrated to simulate the crisis, on the basis of 1998 data. It includes the "model LAVs," that is, the predicted change in employment shares in each category. The last column in the table subtracts the actual LAVs (in column B) from the predicted Table 5.5 Aggregate Results from the Macro Model, Occupations Errors 1998 actual 1999 actuals 1999 simulated by the of the from PNAD from PNAD macro model only macro model (A) (B) (C) (D) Percentage Percentage change Percentage Percentage of actually in each Units of Units of observed Units Percentage category Absolute Location/skill of workers of workers changes of of workers (model error level/employment workers by category workers by category (true LAVs) workers by category LAVs) (percent) Urban sector 48,809,911 50,317,141 51,620,283 Low skill 17,372,833 54.6 17,259,832 18,043,135 56.0 Unemployed 1,497,575 4.7 1,606,782 5.1 8.49 1,623,210 5.0 6.93 1.56 Formal tradable sector 2,184,630 6.9 2,071,504 6.6 4.08 2,112,696 6.6 4.58 0.51 Formal nontradable sector 3,338,557 10.5 3,206,221 10.2 2.76 3,098,839 9.6 8.34 5.58 Informal sector 10,352,071 32.5 10,375,325 33.0 1.38 11,208,390 34.8 6.87 5.49 Intermediate skill 26,632,953 66.7 28,153,740 28,290,953 67.4 Unemployed 3,703,688 9.3 4,245,037 10.2 9.49 4,265,261 10.2 9.62 0.13 Formal tradable sector 4,345,438 10.9 4,475,094 10.7 1.65 4,556,787 10.9 0.22 1.44 Formal nontradable sector 7,809,610 19.6 7,923,915 18.9 3.12 7,872,205 18.8 4.06 0.94 Informal sector 10,774,217 27.0 11,509,694 27.5 2.00 11,596,700 27.6 2.44 0.44 High skill 4,804,125 79.2 4,903,569 5,286,195 79.3 Unemployed 321,052 5.3 380,467 6.1 14.56 381,562 5.7 8.26 6.29 137 Formal tradable sector 709,379 11.7 723,085 11.5 1.54 782,972 11.8 0.53 2.07 (Continued on the following page) 138 Table 5.5 (Continued) Errors 1998 actual 1999 actual 1999 simulated by the of the from PNAD from PNAD macro model only macro model (A) (B) (C) (D) Percentage Percentage change Percentage Percentage of actually in each Units of Units of observed Units Percentage category Absolute Location/skill of workers of workers changes of of workers (model error level/employment workers by category workers by category (true LAVs) workers by category LAVs) (percent) Formal nontradable sector 2,274,160 37.5 2,217,602 35.3 5.76 2,323,764 34.9 6.92 1.15 Informal sector 1,499,534 24.7 1,582,415 25.2 2.02 1,797,897 27.0 9.25 7.23 Rural sector 10,049,477 10,267,135 10,415,081 Low skill 7,522,219 68.8 7,484,557 7,649,800 71.0 Unemployed 174,659 1.6 176,238 1.7 3.75 180,065 1.7 3.10 0.65 Formal tradable sector 958,768 8.8 984,502 9.3 5.94 942,946 8.8 1.65 7.59 Formal nontradable sector 365,199 3.3 347,372 3.3 1.80 340,327 3.2 6.81 5.01 Informal sector 6,023,593 55.1 5,976,445 56.4 2.36 6,186,462 57.4 2.70 0.34 Intermediate high skill 2,527,258 66.5 2,782,578 2,765,281 67.4 Unemployed 233,247 6.1 276,675 6.7 9.45 279,455 6.8 19.81 10.36 Formal tradable sector 480,512 12.6 538,314 13.1 3.40 533,627 13.0 11.05 7.65 Formal nontradable sector 424,539 11.2 470,756 11.4 2.33 399,930 9.8 5.80 8.12 Informal sector 1,388,960 36.6 1,496,833 36.4 0.55 1,552,270 37.9 11.76 12.30 Total urban and rural 58,859,388 60,584,276 62,035,364 Source: Authors' estimates based on IBGE (1998; 1999). Note: LAV linking aggregated variable; PNAD Pesquisa Nacional por Amostra de Domicílios (National Household Survey). CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 139 LAVs (in column C), and thus measures the absolute errors of the macro model in predicting occupational change. On the whole, the model gets the directions of change right: there are only four errors of direction, corresponding to 20 percent of the simulations. Three of these errors occurred in rural areas, where overall confidence on the underlying data is lower. In terms of pre- cision, however, the macro module performs rather poorly. Ten pre- dictions (50 percent) are off by 5 percentage points or more, in absolute terms. In relative terms, the errors are large, indeed, and on seven occasions are greater than 100 percent (with respect to the actual changes). Table 5.6 presents the results in terms of changes in nominal earn- ings (labor incomes). As expected, output contraction in urban areas translated not only into falls in employment (as seen in table 5.5), but also into falling wages (even in nominal terms). Interestingly, this was the case for all categories, except workers with low or inter- mediate skills in the formal nontradable sector. In rural areas, a much more mixed picture emerged. Interestingly, there were large actual rises in the wages of all workers in the formal nontradable sector. For low-skill workers, this was a rise of 32 percent in nomi- nal terms, which was well predicted by the model. Conversely, wages in the rural formal tradable sector fell marginally. As in table 5.5, the performance of the macro model can be judged in table 5.6 by comparing the predicted changes in wages for each worker category (the model LAVs in column E) with the changes actually observed (the "true LAVs" in column D). Absolute errors are again presented in column F. Fortunately, the performance of the macro model is better for earnings than for occupations. None of the 15 counterfactual changes in earnings for household groups reported went in the opposite direction to the changes actually observed for those groups. There were, however, four significant errors in the magnitude of change (those of more than 5 percentage points) in the urban sector. The nominal monthly wages of workers with low-level skills working in the formal tradable sector were projected to grow by 13.9 percent; instead, they grew by only 5 percent. For intermediate-skill workers in the formal tradable sector, a fall of 14 percent was projected, but these wages fell by only 4 percent. For high-skill workers working in the formal tradable sector, a fall of 7 percent was expected, yet wages fell by only 0.7 percent. Finally, for workers with intermediate- and high-level skills in the formal tradable rural sector, growth of 29.7 percent was projected, but their nominal wages grew by only 12.5 percent. The macro model systematically tends to predict larger declines in wages than the ones actually experienced by workers. In 140 Table 5.6 Aggregate Results from the Macro Model, Earnings Linking aggregate variables Wage (nonzero earnings) in nominal R$ (LAVs) in percentage change for per month each category for 1998­99 1999 Percentage of simulated by actually observed LAVs macro 1998 actual 1999 actual the macro changes (true model Error Location/skill from PNAD from PNAD model only LAVs) (percent) (percent) level/employment (A) (B) (C) (D) (E) (F) Urban sector Low skill Formal tradable 454.67 450.81 449.94 0.85 1.04 0.19 Formal nontradable 385.27 404.02 439.01 4.87 13.95 9.08 Informal 264.53 259.82 258.76 1.78 2.18 0.40 Average for the category 316.34 314.77 317.38 0.49 0.33 0.82 Intermediate skill Formal tradable 627.25 605.30 541.31 3.50 13.70 10.20 Formal nontradable 546.28 547.12 548.47 0.15 0.40 0.25 Informal 398.91 388.51 385.44 2.61 3.38 0.77 Average for the category 492.46 481.73 468.42 2.18 4.88 2.70 High skill Formal tradable 2,011.96 1,997.40 1,869.99 0.72 7.06 6.33 Formal nontradable 1,759.46 1,682.85 1,678.20 4.35 4.62 0.26 Informal 1,391.10 1,327.12 1,315.27 4.60 5.45 0.85 Average for the category 1,676.54 1,608.90 1,575.78 4.03 6.01 1.98 Rural sector Low skill Formal tradable 341.14 337.88 322.60 0.96 5.44 4.48 Formal nontradable 252.03 333.78 333.50 32.44 32.33 0.11 Informal 164.69 172.02 171.47 4.45 4.12 0.34 Average for the category 192.83 202.52 197.93 5.02 2.64 2.38 Intermediate high skill Formal tradable 551.13 529.68 507.13 3.89 7.98 4.09 Formal nontradable 527.82 593.85 684.36 12.51 29.66 17.15 Informal 275.19 273.70 266.69 0.54 3.09 2.55 Average for the category 380.28 389.10 385.50 2.32 1.37 0.95 Source: Authors' estimates based on IBGE (1998; 1999). Note: LAV linking aggregated variable; PNAD Pesquisa Nacional por Amostra de Domicílios (National Household Survey); R$ Brazilian reais. 141 142 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI rural areas, the underprediction affects only intermediate- and high- skill workers. Nevertheless, in one-third of these cases, prediction errors were quite low in absolute terms, and less than 20 percent in relative terms. Although the overall performance of the macro model in predict- ing short-term changes in the occupational structure of Brazil's population between 1998 and 1999 was disappointing, the per- formance in terms of earnings changes was better. Predictions of changes in earnings that are accurate in direction and less than 20 per- cent off in magnitude may begin to be of some use for policy makers seeking to assess which groups may be in greater need of social pro- tection during a crisis episode. Nevertheless, aggregate predictions, defined in terms of groups (such as intermediate-skill workers in the formal nontradable sector), may not provide effective policy handles for the design of safety nets. This is the advantage of combining the macro model with the micro module--that is, to allocate the average changes predicted for each representative group of households to actual individuals in the household survey sample. This is an appro- priate time to turn to the definition of the micro model. The Micro Model The occupational responses to a devaluation such as the one considered in this chapter may differ between men and women within the same area and skill groups, or indeed across women with different numbers of children. It may also differ across workers with the same levels of education but different age and experience profiles. Changes in earnings may be different depending on whether the informal nontradable sector job is in manufacturing in a union- ized sector, or in own account service provision in an urban slum. To capture some of the sources of heterogeneity across the diverse population of individuals and households lumped into these groups of agents, a simple reduced-form model of household income deter- mination is estimated, which is based on Bourguignon, Ferreira, and Lustig (2005) and Ferreira and Barros (1999). Once the model has been estimated,13 it can be used to simulate individual and house- hold responses to the sectoral mean changes (in employment proba- bilities and in earnings) predicted by the macro module, while respecting the conditional distribution of wages and employment on observed individual characteristics. The model, similar to the one in chapter 4, consists of three simple blocks. Because the goal is to obtain a measure of welfare, the first block simply defines the household's income per capita, CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 143 aggregating it across its components. The second block seeks to estimate a descriptive relationship between individual earnings and some of its observed determinants, while the third block esti- mates a relationship between occupational choice and some of its key correlates. The first block, which is given by equation (5.6), simply defines household per capita income, by adding all labor incomes across occupations (indexed by s) and household members (indexed by i). The sum of nonlabor incomes of the individuals in the household is represented by y0 . The size of the household is denoted nh. Iis is an h indicator variable that takes a value of one if household member i works in sector s and zero otherwise. At the simulation stage, nonlabor incomes and public sector wages are assumed to remain constant in real terms (meaning they are deflated to 1999 using the CPI computed from September 1998 to September 1999,14 which is equal to 1.0598). nh 1 3 (5.6) yh Iis wsih y0 . nh h i 1 s 1 The second block of equations is represented by a set of standard Mincerian earnings regressions: (5.7) log wih gs xih gs ih. This equation relates the earnings (w) of an individual i in household h to his or her observed (x) and unobserved () characteristics in the standard manner. The model is estimated separately across occu- pations (denoted by s) and across area/skill household groups (g).15 The population was partitioned into the same groups used in the macro model. There are three occupation sectors s (formal tradable, formal nontradable, informal). Household groups g are defined by urban or rural locations, and along education dimensions: low (0 to 4 years of schooling), intermediate (5 to 11 years of schooling), and high (12 or more years of schooling). As before in the macro model, the individuals who live in rural areas with intermediate or high skills are aggregated. In each of these groups, the vector x includes the fol- lowing characteristics: intercept, education (and its square), experi- ence (and its square), occupation, race, Brazilian geographic regions, and dummy variables for gender and metropolitan areas. For purposes of this chapter, these regressions are interpreted merely as descriptions of multivariate correlations. The coefficients are not interpreted causally, as they are likely to be biased because of selectivity and the correlation between unobserved ability and some of the regressors. The key assumption made, and which 144 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI allowed the earnings equations to be used for the micro simulations, was that any such selection and endogeneity biases are stable between 1998 and 1999. The occupational choice model is defined in the last block. The constrained choice of occupation by the worker as a function of his or her household and individual characteristics is represented as follows: (5.8) Ij s), s I(zih s ih zih j ih j where I is an indicator function, which takes the value one if the inequality within the bracket holds, and zero otherwise. z is a vector of observed individual and household characteristics, and captures unobserved individual-level determinants of occupational choice. Elements of z include education; labor market experience; gender; race; occupation, education, experience, and race of the household head; Brazilian geographic regions; a dummy variable if the house- hold is in a metropolitan area; housing status; and a categorical variable for other incomes. This occupational choice model may be estimated empirically by means of a discrete choice model such as a multinomial logit, for which the probability of choosing the category s (inactivity, unemployment, work in the informal sector, work in the formal tradable sector, and work in the formal nontradable sector) is modeled as follows: ezihs (5.9) Pr(j s) . ezih j j Six such models with identical specifications were estimated: one for household heads, another for spouses, and a third for other house- hold members (each of these in both rural and urban areas). Each individual makes a choice according to whether the criterion within the bracket is higher for that sector than for any of the other four. The parameter vector is specific to each occupation and can be interpreted in two ways: either as a vector of the marginal utilities of each characteristic in Z, in the occupation s; or as a descriptive parameter of the distribution of observed occupations, conditional on the elements of Z. The occupational choice model is written in the reduced form--that is, it does not include the wage rate of indi- viduals or family members in the vector Z of explanatory variables. Marginal effects calculated from the estimation results for all six models are reported in the annex, at the end of this chapter. The model, equations (5.7) through (5.9), is estimated on house- hold-level data from the PNAD, which is fielded annually (except in census years) by the Brazilian Census Bureau, IBGE. This chapter CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 145 uses the unit-record data for the 1998 survey, which had a sample size of 88,356 households (and 333,074 individuals), and for the 1999 survey, which had a sample size of 91,523 households (and 340,986 individuals). The PNAD is the staple household survey for analysis of the Brazilian income distribution. It is representative for both urban and rural areas in all five Brazilian geographic regions, except in the north, where for cost-related reasons, rural areas are fielded only in the state of Tocantins. Income data from the PNAD do, however, suffer from considerable measurement error. The PNAD questionnaires, although much improved during the 1990s, still contain insufficient detail on capital incomes, production for own consumption, and incomes-in-kind. As a result, some evidence suggests that some of the incomes are underreported, particularly in rural areas--and this problem is more severe at both tails of the dis- tribution.16 In what follows, rural incomes are included for the sake of completeness of coverage. Caution is urged, however, because the income levels reported here are likely to reflect substantial measure- ment error. In addition, the labor earnings estimations are restricted to the sample for people who are 15 to 80 years old. After estimating the model, the authors use equations (5.7) through (5.9), with the estimated coefficients reported in the annex and with the individual residual terms from the estimation, and equation (5.6) to simulate the effects of the 1998­99 Brazilian crises on the distri- bution of household per capita incomes, poverty, and inequality. Then the counterfactual distribution thus constructed is compared with the original distribution taken from 1999 PNAD data. Formally, the micro simulations consist of finding the solution of the following system of 21 equations based on the 1998 PNAD data: (5.10) Igs zih^ s ^ j s g s ih zih j ih j fs . . . . . . s,g (5.11) Exp ^ g g . . . . . . . . g. gs xih g ih s s i g s g Equation (5.10) corresponds to the first six equations of the system: one for household heads, one for spouses, and one for other house- hold members, by rural and urban areas. Equation (5.11) corresponds to the remaining 15 equations: the earnings regressions were separately estimated for each group (occupational choice, skills, and area). This system would be overidentified if more than one element in each vector were allowed to vary. Therefore, exactly 21 unknowns are solved for: 6 and 15 terms. 0 The authors' interpretation follows. The first six equations-- represented by equation (5.10)--require that the intercept term of the multinomial logit for occupation s (relative to inactivity) be such that 146 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI the fraction of the population who belong to household group g and choose to work in occupation s is equal to the share of the population (who belong to that household group) that is predicted by the "top" macro model to be employed in occupation s, fs . g 17The remaining equations--represented by (5.11)--require that the intercept term of the earnings equation estimated for household group g be such that the mean of the real wage in the counterfactual distribution be equal to the sector/group wage predicted by the factor markets module of the macro model, g. s The system is fully simultaneous, and it is solved numerically by the application of a Newton-Raphson algorithm, which essentially alters values of the 12 "unknown" parameters progressively to minimize the sum of squared differences between the left- and the right-hand sides of equations (5.10) and (5.11). This procedure is analogous to the one used by Bourguignon, Robilliard, and Robinson (2005). As in that analysis, the authors offer no formal existence or uniqueness proofs for the equilibrium of the system, and their algorithm does converge to reach a seemingly plausible equilibrium. Once the system of equations represented by equations (5.10) and (5.11) converges to a solution, the solution values for the 0 and vectors are substituted into equations (5.7) and (5.9). Equa- tion (5.9) will determine the new distribution of occupations in the population, which is consistent with the macroeconomic changes simulated by the macro model. Taking these counterfactual individ- ual occupations into account, equation (5.7) determines the new predicted earnings for each employed worker. Equation (5.6) aggregates the new earnings distribution, generating the final coun- terfactual distribution of household incomes.18 These simulated distributions are therefore consistent (by construction) with both the actual conditional earnings distributions and the conditional occupational distribution observed in 1998, and with the predic- tions of the macro model for the effects of the devaluation on the Brazilian economy. In what follows, these distributions (referred to as the counterfactual 1999 distributions) are compared with the actual distributions observed in 1999. Results from the Complete Top-Down Macro-Micro Simulation: Employment and Earnings Rates The main results for the occupational simulations are presented in table 5.7, for urban and rural areas by skill category and by occu- pation sector. Employment changes were simulated to target the new distribution of employment across all sectors, but not by the exact CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 147 number of individuals in each segment. Of the six columns (A to F) in this table, the first four (A to D) contain the observed data and the results of the model's simulations; columns E and F analyze the errors of the procedure and decompose them into macro and micro error components. Column A, "1998 actual from PNAD," presents actual 1998 employment numbers and the distribution of workers by skill cate- gory and occupational sector. Column B, "1999 simulated by the macro model only," provides the counterfactual 1999 (absolute and relative) employment numbers predicted by the macro model only, and the proportional changes (the LAVs) implied by these numbers, with respect to the actual 1998 data. These are the same LAVs that were presented earlier, in column C of table 5.6. Column C, "1999 simulated by the macro-micro model," presents the corresponding counterfactual employment numbers and LAVs predicted by the full top-down macro-micro model. Column D, "1999 actual from PNAD," presents the real employment numbers from the 1999 PNAD and the proportional changes (the "true" LAVs) with respect to the actual 1998 figures (as in column B of table 5.6). Column E, "Errors of the macro-micro simulation," analyzes dif- ferences between the top-down macro-micro simulations and the actual changes, in absolute terms. Errors of sign in the direction of change and at over- or underpredictions above a threshold of 5 per- centage points are reported. Column F, "Total error of the macro- micro simulation," reports the absolute errors in worker shares and then decomposes them into two categories: those attributable to prediction errors from the macro model, and those coming from the micro simulation model. Somewhat surprisingly, the performance of the top-down macro- micro model is far superior to that of the macro model alone. For occupations in the urban sector, the absolute error is less than 5 per- centage points for all but three categories: the formal nontradable sector for workers with low skills, the formal nontradable sector for workers with high skills, and the informal sector with high skills. For occupations in the rural sector, the top-down macro-micro model also performed adequately, except for rural households in the formal nontradable sector and for workers with intermediate and high skills. Overall, out of the 20 occupational LAVs, the top-down macro- micro model makes six errors with regard to the observed data (the "true LAVs") that are significant (or about 30 percent of the results). Two of these were errors in direction only, one was an error of both magnitude and direction, and three were errors of magnitude only. As column F indicates, the bulk of these errors can be attributed to Table 5.7 Detailed Results from the Top-Down Macro-Micro Models, Occupations by Skill and Sector 1998 actual from 1999 simulated by the macro 1999 simulated by the PNAD model only micro-macro model (A) (B) (C) Percentage Percentage change in change each in each category Percentage Percentage category Percentage predicted by Location/skill Units of of workers Units of of workers (model Units of of workers macro-micro level/employment workers by category workers by category LAVs) workers by category model Urban sector 48,809,911 51,620,283 49,119,235 Low skill 17,372,833 54.6 18,043,135 56.0 17,739,441 55.9 Unemployed 1,497,575 4.7 1,623,210 5.0 6.93 1,602,373 5.1 7.22 Formal tradable 2,184,630 6.9 2,112,696 6.6 4.58 2,081,482 6.6 4.51 sector Formal 3,338,557 10.5 3,098,839 9.6 8.34 3,046,427 9.6 8.58 nontradable sector Informal sector 10,352,071 32.5 11,208,390 34.8 6.87 11,009,159 34.7 6.55 Intermediate skill 26,632,953 66.7 28,290,953 67.4 26,746,944 67.3 Unemployed 3,703,688 9.3 4,265,261 10.2 9.62 4,060,079 10.2 10.14 Formal tradable 4,345,438 10.9 4,556,787 10.9 0.22 4,280,151 10.8 1.10 sector Formal 7,809,610 19.6 7,872,205 18.8 4.06 7,411,419 18.6 4.65 nontradable sector Informal sector 10,774,217 27.0 11,596,700 27.6 2.44 10,995,295 27.7 2.52 High skill 4,804,125 79.2 5,286,195 79.3 4,632,850 78.8 Unemployed 321,052 5.3 381,562 5.7 8.26 347,442 5.9 11.72 Formal tradable 709,379 11.7 782,972 11.8 0.53 683,461 11.6 0.60 sector Formal 2,274,180 37.5 2,323,764 34.9 6.92 2,015,842 34.3 8.57 nontradable sector Informal sector 1,499,534 24.7 1,797,897 27.0 9.25 1,586,105 27.0 9.15 Rural sector 10,049,477 10,415,081 10,123,593 Low skill 7,522,219 68.8 7,649,800 71.0 7,595,335 70.5 Unemployed 174,659 1.6 180,065 1.7 3.10 182,628 1.7 6.25 Formal tradable 958,768 8.8 942,946 8.8 1.65 957,336 8.9 1.48 sector Formal 365,199 3.3 340,327 3.2 6.81 343,156 3.2 4.49 nontradable sector Informal sector 6,023,593 55.1 6,186,462 57.4 2.70 6,112,215 56.7 1.47 Intermediate 2,527,258 66.5 2,765,281 67.4 2,528,258 67.1 high skill Unemployed 233,247 6.1 279,455 6.8 19.81 259,220 6.9 12.05 Formal tradable 480,512 12.6 533,627 13.0 11.05 489,727 13.0 2.85 sector Formal 424,539 11.2 399,930 9.8 5.80 361,605 9.6 14.06 nontradable sector Informal sector 1,388,960 36.6 1,552,270 37.9 11.76 1,417,706 37.6 2.98 Total urban 58,859,388 62,035,364 59,242,828 and rural Source: Authors' esimates based on IBGE (1998; 1999). Note: LAV linking aggregated variable; PNAD Pesquisa Nacional por Amostra de Domicílios. (National Household Survey) 148 Total error of the macro-micro 1999 actual from PNAD Errors of the macro-micro simulation simulation (in percentage points) (D) (E) (F) Total error (in units of workers) Difference Percentage difference between Percentage of the between percentage of the error units change error coming Actually Sign and predicted by predicted by coming from Percentage observed absolute Absolute macro-micro macro-micro from the the micro Units of of workers changes error over Sign error model and model and macro simulation workers by category (true LAVs) 5% error (over 5%) actual actual model model 50,317,141 17,259,832 1,606,782 5.1 8.49 (4,409) 1.27 84.35 15.65 2,071,504 6.6 4.08 9,978 0.44 87.60 12.40 3,206,221 10.2 2.76 1 (159,794) 5.82 95.94 4.06 10,375,325 33.0 1.38 1 633,834 5.16 94.39 5.61 28,153,740 4,245,037 10.2 9.49 (184,958) 0.65 19.95 80.05 4,475,094 10.7 1.65 (194,943) 0.55 61.86 38.14 7,923,915 18.9 3.12 (512,496) 1.53 61.50 38.50 11,509,694 27.5 2.00 (514,399) 0.52 85.07 14.93 4,903,569 380,467 6.1 14.56 (33,025) 2.84 64.54 35.46 723,085 11.5 1.54 (39,624) 0.94 64.70 35.30 2,217,602 35.3 5.76 (201,760) 2.80 41.06 58.94 1,582,415 25.2 2.02 1 3,690 7.13 98.58 1.42 10,267,135 7,484,557 110,778 2.56 81.72 18.28 176,238 1.7 3.75 6,390 2.50 29.10 70.90 984,502 9.3 5.94 (27,166) 4.45 79.44 20.56 347,372 3.3 1.80 (4,216) 2.69 79.78 20.22 5,976,445 56.4 2.36 135,770 0.69 60.65 39.35 2,782,578 276,675 6.7 9.45 (17,455) 2.61 59.80 40.20 538,314 13.1 3.40 (48,587) 0.55 78.97 21.03 470,756 11.4 2.33 1 1 1 (109,151) 16.38 91.59 8.41 1,496,833 36.4 0.55 1 (79,127) 3.53 87.34 12.66 60,584,276 1 2 4 149 150 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI the macro part of the model. This is clearly the major obstacle for these types of procedures, but the link with a macro model of some sort is nevertheless essential for simulating counterfactual economy- wide policies. The model seems to capture a good deal of the occupational effect of the 1999 crisis on the occupational structure in Brazil. The shock led to the following key changes (change in actually observed data/change predicted by the top-down macro-micro model): · A significant increase ( 13 percent/ 13 percent) in unemploy- ment in both rural and urban areas · A particularly large rise in unemployment for workers with intermediate- and high-skill levels in urban areas ( 9.5 percent/ 10 percent and 15 percent/ 12 percent, respectively) · An increase in the level of informality in both rural and urban areas ( 1 percent/ 4 percent and 4 percent/9 percent, respectively) · A growth of informality in particular in urban areas for work- ers with intermediate and high levels of skills ( 2 percent/ 2.5 per- cent and 2 percent/ 9 percent, respectively). These four general characteristics are picked up fairly well by the top-down macro-micro model, with the exception of an overpre- diction for the increase in urban high-skill informality. Overall, it seems that the micro simulation stage of the procedure contributes to a considerable reduction in the prediction errors in occupations that plagued the macro stage, as reported in table 5.5. Table 5.7 suggests that the predictions of the combined model do seem to cap- ture the main general effects of the financial crisis on the Brazilian labor market.19 Naturally, these changes in occupational status were accompanied by changes in earnings. The top-down macro-micro model also runs the counterfactual simulation for earnings. The Main Results for the Counterfactual Structure of Earnings The predictive performance of the top-down macro-micro model simulations for earnings (nonzero nominal monthly wages) is pre- sented in table 5.8. Nominal monthly wages (in Brazilian reais, R$) are presented in columns A, B, E, and F. The actual wages for 1998 and 1999 are in columns A and E, respectively. Columns B and F list the nominal monthly wages simulated by the authors' macro compo- nent alone, and by the top-down macro-micro model, respectively. For Brazil as a whole, the model seems to slightly underestimate absolute earnings levels in 1999. The errors reported in table 5.8 are Table 5.8 Aggregate Results from the Top-Down Macro-Micro Models, Earnings Linking aggregate variables (LAVs) in percentage Wage (nonzero change Wage (nonzero earnings) for each earnings) in nominal category for in nominal R$ per month 1998­99 R$ per month Total error of the macro-micro simulation Difference between 1999 1999 percentage Percentage simulated Actually simulated change Percentage of error 1998 by the observed 1999 by the predicted of error coming actual macro changes actual micro- by macro- coming from Sign and from model Model (true from macro In nominal micro from micro- absolute Absolute Location/skill PNAD only LAVs LAVs) PNAD model R$ per model macro simulation error Sign error level/employment (A) (B) (C) (D) (E) (F) month and actual model model over 5% error (over 5%) Urban sector Low skill Formal tradable 454.67 449.94 1.04 0.85 450.81 450.10 0.71 0.16 84.70 15.30 0 0 0 Formal nontradable 385.27 439.01 13.95 4.87 404.02 438.88 34.86 8.63 99.64 0.36 0 0 1 Informal 264.53 258.76 2.18 1.78 259.82 259.78 0.04 0.02 51.01 48.99 0 0 0 Average for the category 316.34 317.38 0.33 0.49 314.77 318.42 3.65 1.16 71.32 28.68 0 1 0 Intermediate skill Formal tradable 627.25 541.31 13.70 3.50 605.30 540.31 64.99 10.74 98.47 1.53 0 0 1 Formal nontradable 546.28 548.47 0.40 0.15 547.12 543.40 3.71 0.68 21.13 78.87 0 0 0 Informal 398.91 385.44 3.38 2.61 388.51 385.87 2.64 0.68 87.69 12.31 0 0 0 151 Average for the category 492.46 468.42 4.88 2.18 481.73 466.54 15.19 3.15 87.63 12.37 0 0 0 (Continued on the following page) Table 5.8 (Continued) Linking 152 aggregate variables (LAVs) in percentage Wage (nonzero change Wage (nonzero earnings) for each earnings) in nominal category for in nominal R$ per month 1998­99 R$ per month Total error of the macro-micro simulation Difference between 1999 1999 percentage Percentage simulated Actually simulated change Percentage of error 1998 by the observed 1999 by the predicted of error coming actual macro changes actual micro- by macro- coming from Sign and from model Model (true from macro In nominal micro from micro- absolute Absolute Location/skill PNAD only LAVs LAVs) PNAD model R$ per model macro simulation error Sign error level/employment (A) (B) (C) (D) (E) (F) month and actual model model over 5% error (over 5%) High skill Formal tradable 2,011.96 1,869.99 7.06 0.72 1,997.40 1,876.85 120.55 6.04 94.90 5.10 0 0 1 Formal nontradable 1,759.46 1,678.20 4.62 4.35 1,682.85 1,674.90 7.95 0.47 58.48 41.52 0 0 0 Informal 1,391.10 1,315.27 5.45 4.60 1,327.12 1,321.97 5.15 0.39 63.90 36.10 0 0 0 Average for the category 1,676.54 1,575.78 6.01 4.03 1,608.90 1,576.74 32.16 2.00 97.18 2.32 0 0 0 Rural sector Low skill Formal tradable 341.14 322.60 5.44 0.96 337.88 323.41 14.48 4.28 94.96 5.04 0 0 0 Formal nontradable 252.03 333.50 32.33 32.44 333.78 334.18 0.40 0.12 29.46 70.54 0 0 0 Informal 164.69 171.47 4.12 4.45 172.02 173.62 1.60 0.93 20.41 79.59 0 0 0 Average for the category 192.83 197.93 2.64 5.02 202.52 201.16 1.36 0.67 58.66 41.34 0 0 0 Intermediate high skill Formal tradable 551.13 507.13 7.98 3.89 529.68 502.08 27.60 5.21 81.71 18.29 0 0 1 Formal nontradable 527.82 684.36 29.66 12.51 593.85 650.29 56.44 9.50 72.65 27.35 0 0 1 Informal 275.19 266.69 3.09 0.54 273.70 267.56 6.13 2.24 88.92 11.08 0 0 0 Average for the category 380.28 385.50 1.37 2.32 389.10 379.77 9.33 2.40 38.59 61.41 0 0 0 Source: Authors' estimates based on IBGE (1998; 1999). Note: LAL linking aggregated variable; PNAD Pesquisa Nacional por Amostra de Domicllios (National Household Survey); R$ Brazilian reais. CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 153 mostly small and driven by the urban areas, which account for 80 percent of the population. The model tends to systematically pre- dict larger declines in wages than were in fact observed. In the rural areas, the underprediction affects only intermediate- and high- skilled workers. For the six basic groupings that underpin table 5.8 (the three skill levels in urban and rural areas), the model predic- tions really missed the target in only 5 out of 15 LAVs, for a success record of about 66 percent. Overall, the top-down macro-micro model can be said to capture a great deal of the actually observed changes in earnings in Brazil from 1998 to 1999. The shock led to the key changes outlined below, including the percentage change in actually observed data and (shown in parentheses) the percentage change predicted by the top-down macro-micro model: · Mean earnings fell for all three urban categories of workers: by 0.49 percent ( 0.33 percent) for workers with low-skill level, by 2.18 percent ( 4.88 percent) for workers with intermediate- skill level, and by 4.03 percent ( 6.01 percent) for workers with high-skill level. · The picture is more mixed in rural areas. There, the only winners among low-skill workers were those employed in the for- mal nontradable and the informal sectors (and this is well pre- dicted by the model). The main losers ( 3.89 percent) among intermediate- and high-skill workers were those in the formal tradable sector (and this is overpredicted by the model, 7.98 per- cent). The main winners ( 12.51 percent) among intermediate- and high-skill workers were those in the formal tradable sector--which the model overpredicted by 29.66 percent. Results from the Top-Down Macro-Micro Model: Household Incomes from Three Experiments Having thus described the results of the full top-down macro-micro simulation in terms of group means for occupation and earnings rates, the natural next step is to look at the predicted impacts on the disaggregated distribution of household income per capita. After all, had one been interested only in the changes in mean earnings for workers in each of those area/skill/sector groupings, the applied macro model might have sufficed. The whole point of integrating the macro model with a micro simulation module is to better account for heterogeneities within those groups. This section presents the disaggregated simulation results for household incomes and compare them with the actually observed 154 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI changes. In fact, to compare the performance of the top-down macro-micro model with alternative modeling strategies, the authors actually conduct three experiments. Experiment 1 mimics a "tradi- tional" RHG approach--that is, the average effects of the 1999 shock are applied uniformly to all individuals belonging to the same representative group of households. However, instead of using the macro model's simulated results, the changes actually observed (the "true LAVs" from tables 5.5 and 5.6) are used, as if the macro model were capable of generating perfect predictions. This first experiment corresponds to the RHGs approach used by most macro CGE models. The LAVs are the actually observed changes of average income and employment for each RHG. There is no micro simulation: each individual receives the average income and employment change of his or her RHG. Experiment 2 still uses the observed changes in earnings and employment levels (the "true LAVs"), but now, instead of imputing the LAVs uniformly to all members of a household group, the authors allow the microeconomic model to allocate them--by find- ing the solution to system equations (5.10) and (5.11), which take heterogeneity in personal characteristics (observed and unobserved) into account. This second experiment corresponds to a pure simula- tion using the micro simulation model. The micro model runs so that its average results for each RHG converge to the actually observed average income and employment change of the economy's RHGs. This experiment tests the predictive capability of the micro simulation model. Finally, Experiment 3 combines both previous approaches (that is, the simulated results of the macro model with the functioning of the micro simulation model). However, this time the LAVs generated by the macro economic model are used instead of the observed LAVs, so that the third experiment corresponds to the full top-down macro-micro linking model. The macrosimulation consists of run- ning the macro model to replicate the 1999 financial crisis. The run generates LAVs consisting of simulated changes of average earnings and employment levels (as well as prices) for each RHG. Then the micro simulation model runs so that its average results for each RHG converge to the simulated average income and employment change of the model's RHGs. This experiment tests the predictive capability of the full top-down macro-micro linking model. The results for the distributions of household income per capita are used to construct three incidence curves for changes in nominal incomes. The authors compare the results of each of these three experiments with the actually observed changes in the distribution of household per capita income for Brazil between 1998 and 1999. The comparison is presented graphically in figures 5.3, 5.4, and 5.5. CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 155 COMPARING THE MAIN RESULTS FOR THE OVERALL DISTRIBUTION FOR THE THREE EXPERIMENTS Figures 5.3, 5.4, and 5.5 present the income incidence curves result- ing from the 1999 financial crisis on the distributions of household incomes in Brazil. In all cases, the curves plot the difference in loga- rithms of the mean incomes in each hundredth of the distribution. For instance, the difference in logs between the actual 1999 and the actual 1998 incomes for each percentile of the distribution is repre- sented by the thick black line. That line constitutes the "bench- mark" against which the curves of the three experiments will be assessed. Actual data show that the 1999 financial crisis was inequality decreasing. Apart from the first decile (where changes are often affected by the change in the proportion of households report- ing zero incomes), the upper deciles of the distribution suffer much larger losses (real falls of 4 to 5 percent) than the first deciles of the distribution, whose real losses are limited to about 1 to 2 percent. Figure 5.3 presents the comparison between the actual 1998­99 change in incomes for each percentile of the distribution and the incidence curve of the RHG experiment (Experiment 1). The model under this type of experiment correctly predicted the fall in real incomes for the entire distribution, which can be seen by the line that Figure 5.3 Comparison between Actually Observed Changes and Experiment 1, Using Representative Household Groups 8 6 4 (percent) 2 0 difference 2 log 4 6 0 10 20 30 40 50 60 70 80 90 100 percentile actual Experiment 1 (RHG) Source: Authors' estimates based on IBGE (1998; 1999). Note: RHG representative household group. The x axis represents the percent- age change between 1999 and 1998 in nominal income (R$, reais) per month for each percentile of the distribution in Brazil. 156 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI represents inflation (at 6 percent) during the period. Beyond that, the model substantially underestimates the rises in nominal earnings in all segments of the distribution. In other words, it consistently overestimates the real wage losses during the crisis. Figure 5.4 adds to the previous figure the incidence curve of the pure micro simulation experiment (Experiment 2). This experi- ment performs much better than the first one: the distance between the predicted curve and the real change is much lower now, for the entire distribution and, in particular, between the 50th and 90th deciles. Nevertheless, errors do remain, especially in the bot- tom half of the distribution and for the richest 5 percent of the population. Figure 5.5 finally adds to the previous figure the incidence curve of the full top-down macro-micro simulation experiment (Experi- ment 3). The top-down model performs better than the RHG simu- lations from Experiment 1, but not as well as Experiment 2. Because Experiment 2 was conducted using "true LAVs," its errors (the dif- ferences between its incidence curve and the thick line for the actual Figure 5.4 Comparison between Actually Observed Changes and Experiment 1, Using Representative Household Groups, and Experiment 2, Using Pure Micro Simulation Model 8 6 4 (percent) 2 0 difference 2 log 4 6 0 10 20 30 40 50 60 70 80 90 100 percentile actual Experiment 1 (RHG) Experiment 2 (pure micro simulation) Source: Authors' estimates based on IBGE (1998; 1999). Note: RHG representative household group. The x axis represents the percent- age change between 1999 and 1998 in nominal income (R$, reais) per month for each percentile of the distribution in Brazil. CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 157 Figure 5.5 Comparison between Actually Observed Changes and Experiment 1, Using Representative Household Groups, Experiment 2, Using Pure Micro Simulation Model, and Experiment 3, Using Full Macro-Micro Linking Model 8 6 4 (percent) 2 0 difference 2 log 4 6 0 10 20 30 40 50 60 70 80 90 100 percentile actual Experiment 1 (RHG) Experiment 2 (pure micro simulation) Experiment 3 (full macro-micro linking) Source: Authors' estimates based on IBGE (1998; 1999). Note: RHG representative household group. The x axis represents the percent- age change between 1999 and 1998 in nominal income (R$, reais) per month for each percentile of the distribution in Brazil. changes in figures 5.3 to 5.5) are due entirely to prediction errors from the micro simulation model. The additional distance between the incidence curves from Experiments 3 and 2 corresponds to addi- tional errors arising from the macro module. The important overall message to take from figure 5.5--which in a sense graphically summarizes the main results of the chapter--is that while top-down macro-micro models such as these are not capa- ble of perfectly predicting the incidence profile of a macroeconomic phenomenon such as the 1999 Brazilian currency crisis, they never- theless perform reasonably well in predicting both the direction of changes in earnings and the broad pattern of their incidence along the income distribution. In particular, top-down models such as these perform much better than standard RHG approaches, even when macro errors in RHG approaches are eliminated (as was shown here by the use of true LAVs for Experiment 1). All three incidence curves drawn on figures 5.3, 5.4, and 5.5 are derived, in one way or another, from household survey data. The 158 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI line for actual changes simply represents a line of differences between centile means across two surveys. The other three lines are predictions obtained by adding different amounts of income to those same means. In any case, these curves are clearly a graphical representation of collections of sample statistics and, therefore, differences across them contain--in addition to modeling (predic- tion) errors--an element of sampling error. Strictly speaking, there- fore, the preceding statements about centile differences should be subject to statistical tests for significance. The simplest suitable test is the paired t test. In this exercise, the same variable (income) is measured in different ways on the same condition. Assuming that the incomes were generated from the same random sample, it is easy to test the means. Treating the difference of the two vari- ables as a random sample from a normal distribution, the test is given by the following: H0: 0 0 actual exp i H1: actual exp i (x ) n t 0 tn s 1 n s xi x2 n 1 . i 1 Table 5.9 presents the results of the paired t test for statistical significance for each of the three experiments. At the 5 percent level of confidence, one can reject the null hypothesis H0 that the mean of the logarithms of the actual incomes and those simulated under Experiment 1 are equal. One cannot, however, reject the hypotheses that the lines representing Experiments 2 and 3--the micro model based on true LAVs and the fully top-down macro- micro model, respectively--are equal to the line for the actually observed changes. Table 5.9 Paired t Test Hypothesis test t P t Result H0: 2.7403 0.0073 actual exp 1 H0 rejected H0: 0.3919 0.6959 actual exp 2 H0 accepted H0: 0.3753 0.7082 actual exp 3 H0 accepted Source: Authors' estimates based on PNAD/IBGE 1998­99. CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 159 The interpretation of these test results is that while the differences between the RHG simulations and the actual changes were too large to be attributable to sampling errors alone, the differences between the predictions from the other two experiments and reality were small enough that they may be attributed only to sampling error. Broadly similar results were also found using two alternative test formulations: the Welch test for samples from two different popula- tion distributions, and the Smirnov-Kolmogorov nonparametric test for distributional differences. In all three cases, the p-values for the null hypothesis under Experiments 2 and 3 were higher than under Experiment 1. Conclusion This chapter has outlined a top-down macro-micro model of the Brazilian economy that investigates the link between macroeco- nomic shocks and the distributions of employment, earnings, and household incomes. The approach estimates a macro model based on time-series data and a micro model based on household-level cross-section data. The macro model generates three sets of LAVs: employment and unemployment levels per household group and sector, wage levels per household group and sector, and consumer price levels per sector. These linking variables are then used to recalibrate parameters in the earnings and occupational models at the microeconomic level, and thus to simulate changes in the distribution of earnings and incomes at the household level. This approach, adapted from Bourguignon, Robilliard, and Robinson (2005) was applied to an investigation of the employ- ment, earnings, and income distribution effects of the 1998­99 devaluation of the Brazilian real. Unlike previous studies, the authors took advantage of the benefit of hindsight and compared the counterfactual distributions generated by their model for 1999 with the distributions actually observed in 1999. The shock observed--together with the standard policy response of tightening both the monetary and fiscal stances to ensure price stability after the currency floated--was expected to be rather negative. However, the massive devaluation in Brazil (nominal 56 percent) did not result, as it did in East Asia, in a collapse of the financial sector with devastating effects on the credit market and (eventually) on the real economy.20 Increases in poverty were corre- spondingly smaller in Brazil than in Indonesia, Thailand, or, for that matter, Argentina. Nevertheless, real incomes in Brazil fell across the entire distribution, and aggregated poverty measures rose 160 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI accordingly. The headcount index rose from 28.1 percent to 29.2 per- cent. Because income falls were greatest for the richest households, inequality fell for most commonly used measures. The Gini coefficient fell from 0.593 to 0.587. The main effect on the distribution of occupations was a substantial increase in unemployment levels across the board, but predominantly in urban areas and for more skilled workers. In urban areas the informal sector registered small increases in employment ( 3.5 percent), whereas the formal sector retrenched by 0.5 to 1 percent, regardless of skill level or the tradable nature of the goods. In rural areas, the picture was mixed. There was a pronounced move from employment in the informal and formal nontradable sectors toward the formal tradable sector (the sector that benefited from the real devaluation). The actual effects on the distribution of earnings were reasonably muted, at least in urban areas. Real wages fell for most groups but rose substantially for workers in the rural nontradable sector. Household incomes fell across the distribution--but less so for the poor than for the rich. The changes were thus generally equalizing in the sense that skilled workers had greater declines than those with fewer years of schooling. The predictive performance of the top-down macro-micro model was uneven. Comparing occupational and earnings predictions from the macro model alone with the observed changes (aggregated from the observed 1999 PNAD for the same household groups) yielded at best a mixed picture. As shown in the section on generating LAVs, the model made a number of mistakes even in the direction of employment changes, and the errors were generally large in magni- tude. The performance for the earnings LAVs was better, but not stellar either. In this case, however, at least there were no errors of direction, and only about one-third of the predictions were off by 5 or more percentage points. When the macro and micro levels were combined, however, so that the LAV predictions were not uniformly attributed to households in the corresponding groups but instead allocated in ways that respected the correlations present in the household data, performance improved substantively. In the section on results, errors in occupational predictions were shown to be smaller for the macro-micro model than they had been for the macro model alone. The same pattern held for earnings. Indeed, looking at the distrib- ution of incomes in a truly disaggregated manner, as was done in the section on results, reveals a somewhat less damning verdict on the top-down macro-micro modeling exercise. While the top- down model failed to replicate the incidence of changes in incomes CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 161 along the distribution perfectly, it did get both the direction and the basic pattern of incidence right. In fact, prediction errors in the top-down model were statistically indistinguishable from the sam- pling errors inherent in comparing two separate PNAD samples (1998 and 1999). Importantly, the top-down model performed much better than the simple RHG approach would have, even under the assumption that the latter would get all of the macro changes exactly right. All in all, the authors definitely do not claim that this approach has delivered the ability to predict the distributional outcomes of macroeconomic shocks or policy packages with anything near perfect accuracy. They also recognize that both the macro and the micro modeling are data- and computation-intensive tools, and that large macroeconometric models are not the most elegant tools in the professional toolkit. Nevertheless, they do find evidence that the top-down approach delivers a capacity to predict the distributional impacts of a macro shock in a manner that is both broadly accept- able and considerably superior to existing alternative approaches, such as the RHG approach, for which the crucial element in this improvement over RHG approaches seems to be the use of the microeconomic simulations. (Chapter continues on the following page.) Annex: Main Equations of the Micro Simulation Model 162 Table 5A.1 Log Earnings Regression Low Intermediate High Low Intermediate high Indicator Coefficient P-value Coefficient P-value Coefficient P-value Coefficient P-value Coefficient P-value Urban formal tradable Rural formal tradable R2 0.27 0.40 0.36 0.22 0.42 #obs 4,449 9,004 1,378 1,910 967 Education 0.069 0.01 0.051 0.06 0.505 0.09 0.195 0.00 0.033 0.40 Education2 0.008 0.14 0.011 0.00 0.024 0.02 0.021 0.01 0.006 0.00 Experience 0.037 0.00 0.059 0.00 0.079 0.00 0.021 0.00 0.040 0.00 Experience2 0.000 0.00 0.001 0.00 0.001 0.00 0.000 0.07 0.000 0.03 Race: white 0.189 0.00 0.168 0.00 0.380 0.00 0.186 0.00 0.198 0.00 North 0.064 0.13 0.124 0.00 0.038 0.76 0.123 0.25 0.062 0.67 Northeast 0.326 0.00 0.348 0.00 0.099 0.10 0.197 0.00 0.153 0.01 South 0.073 0.00 0.075 0.00 0.060 0.21 0.045 0.26 0.037 0.46 Center-West 0.066 0.05 0.085 0.00 0.205 0.01 0.199 0.00 0.215 0.00 Metropolitan area 0.134 0.00 0.087 0.00 0.235 0.00 0.093 0.06 0.119 0.01 Gender: male 0.407 0.00 0.389 0.00 0.370 0.00 0.280 0.00 0.392 0.00 Intercept 4.322 0.00 4.534 0.00 7.872 0.00 4.381 0.00 4.142 0.00 Urban formal nontradable Rural formal nontradable R2 0.24 0.35 0.38 0.29 0.47 #obs 7,557 17,959 4,866 803 897 Education 0.018 0.31 0.065 0.00 0.101 0.50 0.094 0.03 0.046 0.27 Education2 0.008 0.06 0.012 0.00 0.010 0.05 0.014 0.17 0.010 0.00 Experience 0.033 0.00 0.054 0.00 0.067 0.00 0.015 0.01 0.045 0.00 Experience2 0.000 0.00 0.001 0.00 0.001 0.00 0.000 0.15 0.001 0.00 Race: white 0.148 0.00 0.195 0.00 0.283 0.00 0.085 0.02 0.301 0.00 Region North 0.098 0.00 0.082 0.00 0.017 0.76 0.175 0.12 0.087 0.58 Northeast 0.287 0.00 0.297 0.00 0.217 0.00 0.238 0.00 0.144 0.01 South 0.014 0.44 0.010 0.42 0.038 0.18 0.098 0.03 0.003 0.96 Center-West 0.055 0.02 0.040 0.01 0.107 0.00 0.121 0.04 0.209 0.00 Metropolitan area 0.117 0.00 0.102 0.00 0.285 0.00 0.131 0.00 0.255 0.00 Gender: male 0.490 0.00 0.417 0.00 0.469 0.00 0.456 0.00 0.434 0.00 Intercept 4.492 0.00 4.518 0.00 4.936 0.00 4.612 0.00 4.274 0.00 Urban informal Rural informal R2 0.29 0.36 0.35 0.25 0.33 #obs 22,630 24,459 3,353 11,136 2,740 Education 0.037 0.00 0.049 0.02 0.692 0.00 0.072 0.00 0.044 0.16 Education2 0.013 0.00 0.010 0.00 0.033 0.00 0.006 0.17 0.010 0.00 Experience 0.054 0.00 0.067 0.00 0.066 0.00 0.032 0.00 0.048 0.00 Experience2 0.001 0.00 0.001 0.00 0.001 0.00 0.000 0.00 0.001 0.00 Race: white 0.142 0.00 0.179 0.00 0.182 0.00 0.143 0.00 0.140 0.00 Region North 0.169 0.00 0.142 0.00 0.137 0.04 0.158 0.00 0.061 0.46 Northeast 0.439 0.00 0.391 0.00 0.277 0.00 0.375 0.00 0.354 0.00 South 0.122 0.00 0.030 0.03 0.078 0.04 0.024 0.33 0.089 0.03 Center-West 0.037 0.04 0.007 0.66 0.239 0.00 0.153 0.00 0.100 0.02 Metropolitan area 0.223 0.00 0.133 0.00 0.238 0.00 0.290 0.00 0.272 0.00 Gender: male 0.646 0.00 0.608 0.00 0.489 0.00 0.578 0.00 0.503 0.00 163 Intercept 3.736 0.00 4.090 0.00 8.598 0.00 3.726 0.00 4.050 0.00 Sources: IBGE 1998 and authors' calculation. 164 Table 5A.2 Occupational Structure Multinomial Logit Model: Marginal Effects, Rural Indicator Inactive Unemployed Formal tradable Formal nontradable Informal Heads Probability 0.05 0.01 0.13 0.04 0.77 Gender: male 0.12 0.01 0.11 0.02 0.04 Education 0.00 0.00 0.01 0.01 0.02 Education2 0.00 0.00 0.00 0.00 0.00 Experience 0.00 0.00 0.00 0.00 0.00 Experience2 0.00 0.00 0.00 0.00 0.00 Race: white 0.00 0.00 0.01 0.00 0.01 Status of house occupation (tenant or owner) 0.01 0.00 0.11 0.00 0.10 Other incomes 0.00 0.00 0.00 0.00 0.00 Age (years) 0­9 0.00 0.00 0.01 0.00 0.01 10­18 0.00 0.00 0.02 0.01 0.02 19­64 0.01 0.00 0.00 0.00 0.01 65 0.00 0.00 0.00 0.00 0.01 Region North 0.03 0.00 0.07 0.01 0.11 Northeast 0.02 0.01 0.11 0.03 0.16 South 0.02 0.00 0.04 0.02 0.08 Center-West 0.04 0.01 0.03 0.03 0.10 Metropolitan area 0.05 0.01 0.03 0.06 0.14 Spouse Probability 0.67 0.02 0.01 0.04 0.26 Gender: male 0.42 0.01 0.03 0.08 0.30 Education 0.01 0.00 0.00 0.01 0.00 Education2 0.00 0.00 0.00 0.00 0.00 Experience 0.03 0.00 0.00 0.00 0.02 Experience2 0.00 0.00 0.00 0.00 0.00 Head's education 0.00 0.00 0.00 0.00 0.00 Head's experience 0.00 0.00 0.00 0.00 0.00 Dummy if head is formal tradable 0.05 0.00 0.02 0.01 0.08 Dummy if head is formal nontradable 0.05 0.00 0.01 0.03 0.01 Head's race 0.00 0.00 0.00 0.00 0.00 Race: white 0.03 0.01 0.00 0.00 0.03 Status of house occupation (tenant or owner) 0.01 0.01 0.00 0.00 0.01 Other incomes 0.00 0.00 0.00 0.00 0.00 Age (years) 0­9 0.00 0.00 0.00 0.00 0.00 10­18 0.01 0.01 0.00 0.01 0.02 19­64 0.03 0.00 0.00 0.00 0.03 65 0.02 0.00 0.00 0.01 0.00 Region North 0.06 0.01 0.03 0.02 0.12 Northeast 0.06 0.02 0.01 0.00 0.09 South 0.12 0.00 0.01 0.01 0.10 Center-West 0.05 0.01 0.00 0.01 0.06 Metropolitan area 0.03 0.01 0.00 0.02 0.06 Others Probability 0.43 0.06 0.04 0.03 0.45 Gender: male 0.47 0.00 0.06 0.00 0.41 165 Education 0.03 0.01 0.00 0.01 0.01 Education2 0.00 0.00 0.00 0.00 0.00 Experience 0.04 0.00 0.00 0.01 0.03 (Continued on the following page) 166 Table 5A.2 (Continued) Indicator Inactive Unemployed Formal tradable Formal nontradable Informal Others Experience2 0.00 0.00 0.00 0.00 0.00 Head's education 0.02 0.00 0.00 0.00 0.02 Head's experience 0.00 0.00 0.00 0.00 0.00 Dummy if head is formal tradable 0.05 0.01 0.05 0.00 0.11 Dummy if head is formal nontradable 0.02 0.02 0.00 0.02 0.06 Head's race 0.00 0.00 0.00 0.00 0.00 Race: white 0.03 0.01 0.00 0.01 0.03 Status of house occupation (tenant or owner) 0.04 0.01 0.02 0.01 0.01 Other incomes 0.00 0.00 0.00 0.00 0.00 Age (years) 0­9 0.00 0.00 0.00 0.00 0.01 10­18 0.02 0.01 0.00 0.01 0.00 19­64 0.01 0.00 0.00 0.00 0.02 65 0.02 0.00 0.00 0.00 0.02 Region North 0.08 0.04 0.04 0.01 0.01 Northeast 0.07 0.02 0.04 0.01 0.00 South 0.08 0.02 0.02 0.00 0.08 Center-West 0.00 0.02 0.01 0.01 0.05 Metropolitan area 0.13 0.04 0.01 0.03 0.21 Sources: IBGE 1998 and authors' calculation. Table 5A.3 Occupational Structure Multinomial Logit Model: Marginal Effects, Urban Indicator Inactive Unemployed Formal tradable Formal nontradable Informal Heads Probability 0.15 0.04 0.13 0.22 0.47 Gender: male 0.22 0.02 0.15 0.02 0.08 Education 0.01 0.00 0.01 0.00 0.02 Education2 0.00 0.00 0.00 0.00 0.00 Experience 0.01 0.00 0.00 0.00 0.00 Experience2 0.00 0.00 0.00 0.00 0.00 Race: white 0.01 0.01 0.00 0.00 0.01 Status of house occupation (tenant or owner) 0.01 0.01 0.01 0.00 0.01 Other incomes 0.00 0.00 0.00 0.00 0.00 Age (years) 0­9 0.01 0.01 0.00 0.01 0.02 10­18 0.03 0.01 0.00 0.01 0.02 19­64 0.02 0.00 0.01 0.00 0.01 65 0.01 0.00 0.00 0.01 0.02 Region North 0.03 0.01 0.09 0.06 0.18 Northeast 0.01 0.01 0.07 0.04 0.12 South 0.04 0.00 0.01 0.01 0.02 Center-West 0.03 0.00 0.07 0.02 0.08 Metropolitan area 0.01 0.02 0.00 0.05 0.07 Spouse Probability 0.60 0.05 0.02 0.10 0.23 167 Gender: male 0.50 0.04 0.06 0.14 0.26 (Continued on the following page) 168 Table 5A.3 (Continued) Indicator Inactive Unemployed Formal tradable Formal nontradable Informal Spouse Education 0.01 0.00 0.00 0.00 0.01 Education2 0.00 0.00 0.00 0.00 0.00 Experience 0.02 0.00 0.00 0.00 0.01 Experience2 0.00 0.00 0.00 0.00 0.00 Head's education 0.01 0.00 0.00 0.00 0.01 Head's experience 0.00 0.00 0.00 0.00 0.00 Dummy if head is formal tradable 0.05 0.00 0.02 0.00 0.08 Dummy if head is formal nontradable 0.00 0.00 0.00 0.05 0.05 Head's race 0.00 0.00 0.00 0.00 0.00 Race: white 0.05 0.01 0.00 0.01 0.03 Status of house occupation (tenant or owner) 0.01 0.01 0.00 0.00 0.00 Other incomes 0.00 0.00 0.00 0.00 0.00 Age (years) 0­9 0.00 0.00 0.00 0.01 0.01 10­18 0.05 0.00 0.00 0.02 0.03 19­64 0.02 0.00 0.00 0.00 0.01 65 0.05 0.00 0.01 0.03 0.02 Region North 0.02 0.01 0.01 0.03 0.04 Northeast 0.02 0.01 0.01 0.02 0.03 South 0.08 0.01 0.01 0.03 0.03 Center-West 0.01 0.00 0.02 0.00 0.03 Metropolitan area 0.01 0.02 0.01 0.01 0.02 Others Probability 0.41 0.13 0.05 0.13 0.28 Gender: male 0.21 0.01 0.05 0.02 0.14 Education 0.04 0.01 0.01 0.02 0.00 Education2 0.00 0.00 0.00 0.00 0.00 Experience 0.05 0.00 0.01 0.02 0.02 Experience2 0.00 0.00 0.00 0.00 0.00 Head's education 0.03 0.00 0.00 0.01 0.02 Head's experience 0.00 0.00 0.00 0.00 0.00 Dummy if head is formal tradable 0.01 0.00 0.04 0.00 0.05 Dummy if head is formal nontradable 0.01 0.00 0.00 0.04 0.04 Head's race 0.00 0.00 0.00 0.00 0.00 Race: white 0.02 0.01 0.00 0.01 0.01 Status of house occupation (tenant or owner) 0.01 0.01 0.00 0.01 0.01 Other incomes 0.00 0.00 0.00 0.00 0.00 Age (years) 0­9 0.02 0.00 0.01 0.01 0.00 10­18 0.02 0.00 0.00 0.01 0.01 19­64 0.01 0.00 0.01 0.01 0.01 65 0.00 0.00 0.00 0.01 0.01 Region North 0.11 0.02 0.06 0.08 0.04 Northeast 0.12 0.02 0.06 0.07 0.02 South 0.02 0.01 0.01 0.01 0.01 Center-West 0.01 0.01 0.04 0.00 0.05 169 Metropolitan area 0.01 0.04 0.01 0.02 0.06 Sources: IBGE 1998 and authors' calculation. 170 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI Notes The authors thank François Bourguignon and Anne-Sophie Robilliard for their guidance and inspiration and Armando Castelar Pinheiro, Aart Kraay, and Alexandre Samy de Castro for useful comments on a previous version of this chapter. 1. Most if not all financial crises in emerging markets (a) occurred after significant financial liberalization under rigid exchange rate regimes; (b) were preceded by massive capital inflows that allowed the accumulation of significant unhedged foreign currency liabilities by domestic agents that became illiquid or insolvent when these capital flows suddenly reversed; and (c) tended to cause contagion and spread to other countries. The liter- ature has proposed interpretations of the origin and spread of the crises ranging from a "fundamentalist" view (that the crises resulted from weak macroeconomic and financial fundamentals) to a "financial panic" view (that the crises were self-fulfilling due to investor behavior unrelated to economic conditions. For a survey, see Pereira da Silva (2001). 2. See, for example, Lokshin and Ravallion (2000) on the case of Russia, and Kakwani (1998) on Thailand. Baldacci, de Mello, and Inchauste (2002) use cross-country analysis to show that financial crises tend to have a negative impact on the income distribution and to increase poverty. There is also comparative work on the impact of financial crises in Asia and Latin America on labor markets and household incomes; see, for example, Fallon and Lucas (2002). These analyses conclude that employment fell by much less than production in crisis-hit countries, but that there were considerable changes in employment status, location, and sectoral composition. They also show that cuts in real wages (resulting from real depreciation of the currency) were accompanied by small rises in unemployment, and that families smoothed their incomes through increased participation and private transfers. 3. Policy makers may want to compare the likely distributional impacts of alternative stabilization strategies, such as a tighter monetary policy with regard to a tighter fiscal policy--or may like to know whether the negative impact of a devaluation on the balance sheets of (predominantly urban) firms indebted in hard currencies might be offset by income gains in the rural tradable sector. In designing safety nets to cope with the crisis, policy makers may wonder which sectors would be most hurt by declines in the demand for labor, and whether those sectors are likely to respond predom- inantly through lower wages or through higher unemployment. 4. Data on wages and employment for the disaggregated labor market could be obtained only through the household surveys, which start in the late 1970s and are available yearly. Likewise, data for the financial sector are not available with consistent methodologies for a time span that allows a large sample. CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 171 5. PNAD data are annual except for census years and a few other exceptions, such as 1994. 6. MPS was named for the initials of the institutions that helped develop it: Massachusetts Institute of Technology, the University of Penn- sylvania, and the Social Science Research Council. 7. Both the DMS and METRIC were macroeconometric models devel- oped in France in the late 1970s and 1980s. DMS stands for the French words Dynamic multisectoriel, that is, multisectoral dynamic; METRIC stands for Modèle économétrique trimestriel de la conjuncture, that is, quarterly econometric model of the economic situation. 8. Macroeconometric models usually have an ad hoc treatment of expectations (naïve, adaptive mechanisms). They are also subject to the Lucas critique; parameters in most equations are not invariant to a change in regime. The critique forced macro models not only to give more empha- sis to theory, long-run relationships, and the supply side, but to specify and estimate dynamic adjustments more robustly. Models tried to incorporate rational expectations or model-consistent expectations to address the Lucas critique. The estimation strategy in the authors' "top" macro model addresses only some of these issues. Most of the behavioral equations are estimated by ordinary least squares, some with an error-correction mecha- nism. Attempts to construct macro models based purely on a bottom-to-top aggregation of microeconomic behavior are under way (see Townsend and Ueda 2001); for an example, see chapter 8 of this volume. 9. In most macro frameworks there are two possible ERRs. Under a fixed regime, the central bank intervenes in the foreign exchange market to maintain a fixed parity or a crawling peg path with respect to a specific for- eign currency target (the U.S. dollar, for example). Thus, the change in the demand for money by households is affected by this foreign component (an exogenous element in the supply of money). Under a flexible regime, the central bank does not intervene, and the BoP equation determines the freely floating exchange rate. 10. The 30 unknowns are four factor prices for each of six sectors (24) and six unemployment levels (one for each skill type, and separately for urban and rural areas). Capital is assumed to be fully utilized in all sectors. In prac- tice, however, insufficient numbers of observations for high-skill workers living in rural areas required that skill type be grouped with intermediate-skill work- ers in rural areas. Since the price of capital is not an LAV either, this reduces the number of LAVs in practice to three factor prices in five sectors (15) and five unemployment levels. To these 20 LAVs obtained from the solution of these systems of equations, 15 employment levels will be added. LAVs are later discussed in this chapter in a section on LAV links. 11. Details in Pereira da Silva, Pichetti, and Samy de Castro (2004). In the present version, the simulations describe essentially movements between long-term solutions in levels. However, one of the main forthcoming 172 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI extensions involves enhancement of the dynamics of the solutions by estimat- ing movements of the variables in differences, through an error-correction mechanism. The idea, in line with the basic motivation for the proposed model, is to gain further insight into the paths of adjustment of the endoge- nous variables in response to shocks, allowing the analysis of trade-offs not only in terms of macro stability versus social indicators, but also in terms of the time periods involved. 12. Recall that there are only five area/skill combinations because the intermediate- and high-skill workers in rural areas are being considered together due to an insufficient number of observations. 13. While the macro model is estimated on aggregated time-series data, the micro model is estimated on a single cross-section of the household sur- vey (PNAD 1998). In both cases, identical definitions of skills and sectors are used, to guarantee a consistent mapping of individuals into groups. 14. September is the reference month of the survey. 15. For simplicity, the corresponding g and s subscripts are dropped from the variables (w, x, and ) in equation (5.7). 16. See Ferreira, Lanjouw, and Neri (2003) and Elbers and others (2001) for an assessment of these measurement problems that is based on compar- isons between the PNAD and an alternative Brazilian household survey, the Pesquisa de Padrões de Vida (PPV). 17. One difficulty is that the macro module allows for changes in the relative skill composition of labor demand when constructing the employ- ment LAVs, but this micro simulation does not allow for changes in the education level of the worker. This may be economically realistic for the short term, but it implies that there are six actual unknowns for potentially 18 exogenous variables (fs targets). In the simulations, the adjustment g occurs through the number of people left over for unemployment and inac- tivity from each skill category, which corresponds to a quantity closure to the labor market. 18. These counterfactual distributions assume that a number of features of the population and economy remained constant at their 1998 levels. These include the spatial, racial, gender, and education composition of the population; the distribution of nonlabor incomes; and the internal compo- sition of the households. 19. Whether or not much consolation should be derived from this improvement will depend on how much of the improvement is attributable to mechanical factors behind the convergence of the Newton-Raphson algorithm. 20. The real GDP growth rates in Brazil for 1997, 1998, and 1999 were, respectively, 3.27 percent, 1.32 percent, and 0.81 percent--still in positive territory--as opposed to the dramatic changes from positive 6 to 8 percent real growth down to 5 percent to 15 percent in the same period in East Asian crisis-hit countries such as Thailand and Indonesia. CAN DISTRIBUTIONAL IMPACTS OF SHOCKS BE PREDICTED? 173 References Adelman, Irma, and Sherman Robinson. 1988. "Macroeconomic Adjust- ment and Income Distribution: Alternative Models Applied to Two Economies." Journal of Development Economics 29 (1): 23­44. Artus, Patrick, Michel Deleau, and Pierre Malgrange. 1986. Modélisation Macroéconomique. Paris: Economica. Artus, P., and P. A. Muet. 1980. "Une Etude Comparative des Proprietes Dynamiques de Dix Modeles Americains et Cinq Modeles Français." Revue Economique 31 (1). Baldacci, Emanuele, Luiz de Mello, and Maria G. Inchauste. 2002. "Finan- cial Crises, Poverty, and Income Distribution." IMF Working Paper 02/4. International Monetary Fund, Washington, DC. Bourguignon, Francois, W. Branson, and Jaime de Melo. 1989. "Macro- economic Adjustment and Income Distribution: A Macro-Micro Simu- lation Model." OECD Technical Paper 1. Organisation for Economic Co-operation and Development, Paris. Bourguignon, François, Francisco H. G. Ferreira, and Nora Lustig. 2005. The Microeconomics of Income Distribution Dynamics in East Asia and Latin America. Washington, DC: World Bank; New York: Oxford University Press. Bourguignon, François, Anne-Sophie Robilliard, and Sherman Robinson. 2005. "Representative versus Real Households in the Macroeconomic Modeling of Inequality." In Frontiers in Applied General Equilibrium Modeling: In Honor of Herbert Scarf, eds. Timothy J. Kehoe, T. N. Srinivasan, and John Whalley. Cambridge, UK, New York: Cambridge University Press. Cardoso, Eliana. 2003. "Seigniorage, Reserve Requirements, and Bank Spreads in Brazil." In Taxation of Financial Intermediation: Theory and Practice for Emerging Economies, ed. Patrick Honohan, 241­68. Washington, DC: World Bank and Oxford University Press. Elbers, C., J. O. Lanjouw, Peter Lanjouw, and Philippe G. Leite. 2001. "Poverty and Inequality in Brazil: New Estimates from Combined PPV- PNAD Data." World Bank, Development Economics Research Group, Washington, DC. Fallon, Peter, and Robert E. B. Lucas. 2002. "The Impact of Financial Crises on Labor Markets, Household Incomes, and Poverty: A Review of Evidence." World Bank Research Observer 17 (1): 21­45. Favero, Carlo A., and Francesco Giavazzi. 2002. "Why Are Brazil's Interest Rates So High?" SSRN Electronic Paper Collection. University of Bocconi Innocenzo Gasparini Institute for Economic Research and IGIER, Univer- sity of Bocconi, Milan, Italy. Fernandes, Reynaldo, and Naércio A. Meneses-Filho. 2002. "Escolaridade e Demanda Relativa por Trabalho: Uma Avaliação para o Brasil nas 174 FERREIRA, LEITE, PEREIRA DA SILVA, AND PICCHETTI Décadas de 80 e 90." In Chahad J. P. Zeetano and N. Meneses-Filho (eds.), Mercado Do Trabalho No Brasil. Sao Paulo: Ltr Editora. Ferreira, Francisco H. G., and Ricardo Paes de Barros. 1999. "The Slippery Slope: Explaining the Increase in Extreme Poverty in Urban Brazil, 1976­1996." Brazilian Review of Econometrics 19 (2): 211­96. Ferreira, Francisco H. G., Peter Lanjouw, and Marcelo Neri. 2003. "A Robust Poverty Profile for Brazil Using Multiple Data Sources." Revista Brasileira de Economia 57 (1): 59­92. Fisher, Franklin, L. R. Klein, and Y. Shinkai. 1965. "Price and Output Aggregation in the Brookings Econometric Model." In The Brookings Quarterly Econometric Model of the U.S., ed. J. Duesenberry, 653­79. Washington, DC: Brookings Institution. IBGE (Instituto Brasileiro de Geografia e Estatística). 1998. Pesquisa nacional por amostra de domicílios (PNAD). Rio de Janeiro. ______. 1999. Pesquisa nacional de domicílios (PNAD). Rio de Janeiro. Lokshin, Michael, and Martin Ravallion. 2000. "Welfare Impacts of the 1998 Financial Crisis in Russia and the Response of the Public Safety Net." The Economics of Transition 8 (2): 269­95. Kakwani, Nanak. 1998. "Impact of Economic Crisis on Employment, Unemployment, and Real Income." Bangkok, Thailand: National Economic and Social Development Board. Pereira da Silva, Luiz A. 2001. "Boom and Bust in East-Asia: How the East- Asia Miracle Produced the Financial Bubbles That Ended in the 1997­98 Crises." World Bank, Development Economics Research Group, Washington, DC. Pereira da Silva, Luiz A., Paulo Picchetti, and Alexandre Samy de Castro. 2004. "A Macroeconometric Financial Model for Brazil Estimated with the PNAD." World Bank Development Economics Research Group, Washington, DC. Townsend, Robert M., and Kenichi Ueda. 2001. "Transitional Growth with Increasing Inequality and Financial Deepening." IMF Working Paper WP/01/108, International Monetary Fund, Research Department, Washington, DC. PART III Macro-Micro Integrated Techniques 6 Distributional Effects of Trade Reform: An Integrated Macro-Micro Model Applied to the Philippines François Bourguignon and Luc Savard Analyzing the micro impact of policy reforms is essential to under- stand their impact on poverty and more generally on income distri- bution, and therefore their social acceptability. When reforms are shown to be beneficial for society as a whole but not to particular groups, such an analysis gives policy makers information on the measures to be taken to compensate losers and on the cost of these measures. The Philippines' government is faced with numerous pol- icy choices that are all the more difficult because of concerns voiced by various pressure groups about the impact of these policies on vulnerable groups. An important policy choice has to do with uni- lateral trade liberalization. This policy likely will have beneficial aggregate effects in the medium run when markets will have fully adjusted. Domestic agents, however, likely will benefit unevenly from this reform, some of them running the risk of being net losers. Moreover, the presence of market imperfections may substantially modify the overall gain of the reform and increase individual losses. The identification of the distributional effects of a policy reform, and in particular of the losers and the way they could possibly be compensated, is difficult. The reason for this difficulty is the need to jointly evaluate two types of effect: (1) the aggregate effects of the 177 178 BOURGUIGNON AND SAVARD policy reform, something that is generally done through conven- tional macro modeling­­with more or less sectoral disaggregation; and (2) the heterogeneous effects of the reform on individual agents, an analysis that requires an essentially microeconomic perspective. Recent years have witnessed a flourishing body of literature about this macro-micro nexus in modeling the poverty and distributional impact of macro policies, including trade reforms. The integration of the macro and the micro perspectives remains somewhat imper- fect or rather cumbersome, however. The present chapter proposes an alternative approach and applies it to the issue of trade liberal- ization in the Philippines. The chapter is organized as follows. The first section discusses the various methodological approaches to the macro-micro link in the analysis of policy reforms and presents the original approach applied in the chapter. The second section presents the original fea- tures of the application of that methodology to the Philippines and, in particular, the way it accounts for labor market imperfections. The third section shows the results of simulating the distributional effects of across-the-board trade liberalization in the Philippines. It compares the results obtained from the methodology used in this chapter with those derived from alternative approaches. An Iterative Top-Down Approach to the Macro-Micro Link This section starts by briefly reviewing the existing methodologies that link the macro and micro parts of a modeling framework, high- lighting their advantages and limitations. It then proposes an alter- native approach and describes its implementation. Existing Approaches to the Modeling of the Macro-Micro Link Three main approaches are being used in the literature to link macro reforms to changes in income distribution and poverty within the framework of economywide computable general equilibrium (CGE) models. The first and most common is the representative household (RH) approach. The population of households is partitioned into groups, and each group is represented by a virtual household assumed to behave as the mean of the group. Income distribution within groups is taken as exogenous, so this approach considers only between-group sources of variations in the distribution of income and poverty. This is a severe drawback given the importance DISTRIBUTIONAL EFFECTS OF TRADE REFORM 179 of the within-group components in existing empirical evidence on the sources of change in income distribution. The second approach may be referred to as the CGE micro simula- tion sequential (MSS) method. This is essentially a top-down approach. At the top, a conventional CGE model­­with or without representative households­­is used to simulate policy reforms and estimate changes in prices and factor rewards resulting from the reform. These changes are then fed into a conventional household survey database to yield estimates of the change in the income and expenditures of individual households under the assumption of no behavioral response. It is known that with perfect markets and mar- ginal changes in the price system, the difference between these two amounts yields a money metric measure of the change in the individ- ual welfare of households. Applying these changes to initial incomes derives the change in the distribution of real income within the population and in poverty. This simple approach to the micro conse- quences of macro policy reforms combines conventional CGE model- ing and micro simulation and is being used increasingly--see, for instance, chapters 2 and 3 in this volume by Lokshin and Ravallion and by Bussolo, Lay, Medvedev, and van der Mensbrugghe; Coady and Harris (2001); King and Handa (2003); Vos and De Jong (2003); and Chen and Ravallion (2004). A more complicated approach that considers labor force participation behavior, labor market imperfec- tions, and possibly nonmarginal price changes has been explored by Bourguignon, Robilliard, and Robinson (2005); and Ferreira, Leite, Pereira da Silva, and Picchetti (chapter 5 in this volume). An obvious critique of the MSS approach is the lack of feedback from the micro side of that methodology (the micro simulation based on household surveys) to the macro side (that is, the CGE model). Household behavioral responses to price changes may well be ignored when computing marginal changes in welfare at the micro level under the assumption of perfect markets. As noted by Hertel and Reimer (2004) or Bourguignon and Spadaro (2006), however, these responses are not necessarily negligible at the macro level, and the approxima- tion may be grossly incorrect in the case of market imperfections or nonmarginal changes in the price system. This top-down approach is also inappropriate when policy changes are specified at the household level, for instance, with cash transfer programs. A third approach that is being explored in recent work handles the micro and the macro parts of the modeling in a fully integrated way, rather than sequentially as with the MSS approach or through intermediate aggregation as with the RH approach. Practically, this approach is simply an extension of the latter. It includes as many "representative households" as there are actual households in the 180 BOURGUIGNON AND SAVARD household survey that would be used with the MSS approach. In the prototype model by Decaluwé, Dumont, and Savard (1999), each household is characterized by its share in total factor endow- ments in the economy (as computed from the income part of the household survey), its saving rate, and the allocation of its con- sumption budget among the various goods and services appearing in the CGE model. All these shares and rates are fixed, and the model solves for the complete equilibrium of the economy, includ- ing saving and consumption demand for each household in the sam- ple. Cogneau and Robilliard (2001) applied this type of integrated approach of micro-macro modeling to the Malagasian economy, based on a set of 2,000 households observed in a household survey and under the assumption of a dualist labor market. In comparison with the other approaches, this integrated multi- household(IMH)approachappearsastheonlyonebasedonarigorous theoretical framework that considers all the observed heterogeneity of the population of households. Yet it raises several difficulties at the implementation stage. First, reconciliation between the aggregate data in the macro part of the model and micro data coming from the house- hold survey can be problematic--especially concerning the definition of aggregate goods and services to be used in both the micro (house- hold expenditure) and macro (sectoral production) sides of the model; see Rutherford, Tarr, and Shepotylo (2005) or Cockburn (2006). Sec- ond, the numerical resolution can be challenging. Boccanfuso, Cabral, and Savard (2005) were able to handle an integrated model including around 3,500 Senegalese households, but Rutherford, Tarr, and Shepotylo (2005) found it almost impossible when including 50,000 households in their analysis of the effect of Russia's accession to the World Trade Organization. Finally, detailed microeconomic behavior or micro consequences of market imperfections can be difficult to model in this context. For instance, the introduction of involuntary unemployment in the modeling of the labor market requires specifying rationing schemes at the individual level that somehow imply exter- nalities among individuals or households. This feature may be difficult to handle within a standard CGE framework. Cogneau and Robilliard (2001) provide an example of such a model that includes externalities among households. This example, however, seems to have been pro- vided at the cost of an oversimplified macro framework. It is likely that advances in computing power will soon make it possible to include a much larger number of households in a CGE framework. It is less clear whether that will permit solving the other difficulties. It is thus important to explore other approaches to micro- macro modeling that permit the full integration of standard CGE modeling and a detailed representation of a large population of indi- vidual households. DISTRIBUTIONAL EFFECTS OF TRADE REFORM 181 The MSS method may be seen as a first iteration in the IMH approach. Introducing some feedback from the micro simulation level into the CGE model and then applying the whole MSS again seems like a natural iterative way of integrating the micro and macro analy- ses of policy reforms. Rutherford, Tarr, and Shepotylo (2005) devised such an iterative algorithm and found that in the case of Russia and with perfectly competitive markets, most of the micro and macro impacts of an across-the-board trade liberalization were satisfactorily accounted for by the first MSS step. This chapter proposes a different and simpler approach that can be applied to imperfectly competitive environments. It examines whether the same practical conclusion can be obtained in the case of trade liberalization in the Philippines in the presence of strong imperfections of the labor market. An Iterative MSS Solution to the IMH Model An elementary Walrasian representation of the economy is used here to present and discuss the iterative method proposed in this chapter to solve an IMH model. The model actually used for the Philippines' application in the second part of the chapter is more elaborated. Let Ch (yh, , p) be the consumption function of good i ( 1, . . ., I) ,i h by household h, where yh is the income of household h ( 1, . . ., H), is a set of demographic characteristics, and p is the vector made of h the prices of goods and services. Let Lh,n( , h h, w, p, Rh) be the sup- ply of labor of type n ( 1, . . ., N) by household h, where is the h vector of the specific productivities of household h in the various types of labor (say, skilled-unskilled), w is the corresponding vector of unit wages, and Rh is nonlabor income. Finally, let Yi(ki, w, p), Pi(ki, w, p), and Li (ki, w, p) be, respectively, the supply, the profit, and the vector d of labor demands of the sector producing good i, with ki standing for the fixed factors of production. The competitive equilibrium of that economy is given by the solu- tion in (p, w) of the following system of equations: H Ch (yh, , p) i 1, . . ., I ,i h Yi(ki, w, p) h 1 H I L ( ., d n 1, . . ., N hn hn h h, w, p, Rh) Lin (ki, w, p) h 1 i 1 (6.1) N yh , hnwnLhn( h. h, w, p, Rh) Rh; n 1 I Rh h 1, . . ., H, hiPi(ki, p, w) i 1 where is the share of the profits of sector i that goes to household h. hi 182 BOURGUIGNON AND SAVARD This simple model includes the main features of an IMH model. What makes it somewhat difficult to solve is that the number of H households, and therefore the total number of equations, may be extremely large. Things would be much simpler if it were possible to group the H households into a much smaller number of aggregate households for which the consumption and labor supply would depend only on their aggregate characteristics, including productiv- ity and nonlabor income. Then the solution of the system of equa- tions (6.1) could be cast in terms of these RHs and the income and consumption of each individual household could be assumed to be proportionate to that of the RHs in the group to which it belongs. In effect, such an approach combines the RH and the MSS approach into the solution of the IMH model. Introducing some heterogene- ity among households, however, or making the consumption func- tions and labor supply functions nonlinear with respect to income, is enough to make aggregation and the preceding simplification invalid.1 A simple algorithm for solving an equilibrium system like (6.1) is the familiar fixed point. The structure of the algorithm is shown in figure 6.1. Formally, let wm, pm be the vector of unit wages and prices at iter- ation m, and let m, m be the corresponding vectors of aggregate household demand for goods and labor supply at those prices and Figure 6.1 Iterative Resolution of the Integrated Multihousehold CGE Model Top module: CGE Exogenous (consumption C, labor supply L) Endogenous and output to HHMS (p, w) Loop until C, L Bottom module: household micro simulation Exogenous (p, w) Endogenous (Yh, Ch, Lh) Output to CGE (total consumption C, total labor supply L) Source: Authors' representation. Note: CGE computable general equilibrium. DISTRIBUTIONAL EFFECTS OF TRADE REFORM 183 wages, as given by the left-hand side of the first two sets of equa- tions in (6.1): H m m , pm) i 1, . . ., I i Ch (yh , ,i h h 1 H m L ( , m n 1, . . ., N n hn hn h. h , wm, pm, Rh ) h 1 (6.2) N with yhm , m m hnwn Lhn( m h. h , wm, pm, Rh ) Rh n 1 I and Rhm h 1, . . ., H . hi.Pi(ki, pm, wm) i 1 Simplifying this set of equations, the preceding definition may be rewritten as a bottom-up (BU) equation: m (wm, pm) (6.3) m (pm, wm) . Equation (6.2) corresponds to the MSS approach to the micro impact of macro policies. It simulates the impact of changes in the price system (p, w) on individual incomes. Aggregating the micro responses to those changes provides the BU part of the iterative algorithm proposed in equation (6.3). This step feeds household responses to price changes back to the top of the CGE part of the algorithm, a feedback that is missing in the standard noniterative MSS approach. Aggregate demand and labor supply being consid- ered as exogenous in iteration m 1, aggregate equilibrium condi- tions in equation (6.1) can now yield new values for the price system. This particular step of the iteration writes now as a top-down (TD) equation: Yi(ki, wm , pm ) 1 1 m i 1, . . ., I i (6.4) I Lin (ki, wm , pm ) d 1 1 m n 1, . . ., N , n i 1 which yields a new vector of prices and wages (pm , wm ). This 1 1 new vector is then sent down to micro simulation at the household level for a new iteration. This system of equations thus provides the TD part of the algorithm that solves the integrated household model. 184 BOURGUIGNON AND SAVARD Writing the solution of the preceding system as follows: (wm , pm ) 1 1 ( m, m ) , it is now possible to put the two parts, BU and TD, of the algorithm together, leading to the following fixed-point (FP) algorithm: (6.5) (wm , pm ) 1 1 [ (wm, pm), (wm, pm)] . Convergence of the algorithm is obtained when the distance between two successive iterations on the vectors (pm, wm) is below some arbi- trary small threshold.2 With a single market, the preceding algorithm is the familiar cob- web model. Some conditions must hold for this algorithm to con- verge, namely, the elasticity of the demand side of the market must be smaller than the elasticity of the supply side near the equilibrium. In a multimarket framework, the condition for multimarket equi- librium stability is that the matrix (J I), where J is the Jacobian of the system of FP equations (6.5) and I is the unit matrix, is definite negative. An interesting property of this FP algorithm is that its first itera- tion essentially corresponds to the MSS approach. Starting from some initial equilibrium situation, suppose that a shock hits the economy on the supply side. As a first approximation, the MSS approach is equivalent to supposing that aggregate demand and labor supply are not modified. Then the TD solution of equation (6.4) gives the resulting shocks in the price system. The BU part of the algorithm identifies the effect of this shock on the income and wel- fare of each household in the sample being used. The MSS approach would stop there, ignoring the possible feedback of the initial shock on aggregate demand and labor supply. An interesting question is whether considering feedback effects of the household sector on the economy, as is done in the iterative procedure proposed here, even- tually leads to different estimates from those obtained at the first iteration. It is possible to get closer to final effects using a single iteration or the MSS approach by introducing some behavioral response on the demand side of the goods markets and on the supply side of the labor markets into the CGE model. This can be done by introducing in that model an RH whose behavior has some similarity with the aggregate behavior of the sample of individual households.3 The first iteration of the algorithm would thus rely on a CGE model that has a single aggregate household with a demand system. Aggregate income and price elasticities of the demand for goods and services DISTRIBUTIONAL EFFECTS OF TRADE REFORM 185 and the supply of labor would have to be close to the elasticity obtained from aggregating individual households' behavior. The iterative top-down, bottom-up resolution method shown above--which applies to much richer representations of the econ- omy than equation (6.1)--has several advantages over a method that would solve simultaneously for all individual and aggregate equilibrium conditions. First, there is no obligation to make the macro and the micro parts of the model fully consistent in terms of consumption or income aggregates. Some rule would be needed that permits converting the aggregates obtained from the BU part of the algorithm into the aggregates used in the CGE part of the model. For instance, the household survey may be underestimating the aggregate consumption of a particular good as given by the national accounts generally used in CGE modeling. No correction is neces- sary for consistency with national account data if it is assumed that the proportion of underestimation is independent from the price of other goods and unit wages. In other words, income and expendi- ture data in household income and expenditure surveys can be used as they are. A second advantage is that there is no limit to the level of disaggregation in terms of production sectors and number of households to be included in the model. This issue is discussed in Chen and Ravallion (2004) and Rutherford, Tarr, and Shepotylo (2005). The third advantage is that the flexibility of the functional forms used to model the consumption and labor supply behavior of households is greater than in other approaches. In particular, there is no need to choose functional forms with good aggregation prop- erties. Finally, as can be seen in the following section, it is possible to introduce labor market imperfections without major difficulty and to explicitly consider the externality among households when rationing occurs on one side of the market. Labor Market Imperfections Like many other developing countries, the Philippines is character- ized by a dual labor market with a formal sector in which most employees are wage workers, most often under a labor contract, and an informal sector dominated by self-employment and family business (Riveros 1993). Taking this dualism into account is impor- tant because workers with the same characteristics are not remu- nerated at the same rate in the two sectors, and the allocation of workers between the two sectors has a direct impact on poverty and the distribution of income. This feature is introduced in the present micro-macro model using the well-known specification first presented 186 BOURGUIGNON AND SAVARD by Roy (1951), revisited by Heckman and Sedlacek (1985), and further enriched by Magnac (1991), in which workers decide which sector, if any, they want to join depending on the wages they are offered or anticipate. Another feature of the Philippine economy seems to be the wage rigidity in the formal sector of the economy.4 This implies that some rationing takes place in the formal sector of the economy that forces workers who would have preferred a job in that sector to work in the informal sector or to be inactive. This feature of the labor mar- ket is taken into account by following closely the micro framework proposed by Magnac (1991). Formally, the representation of the labor market in the present model of the Philippine economy is as follows. Assume in a first stage there is only one type of labor but that productivity, , varies h across individuals h--who momentarily will be assumed to coincide with household heads. Productivity is assumed to depend on indi- vidual characteristics according to a Mincerian model type. Then the wage, j, of individual h in the segment j ( 1 for formal, 2 for h informal) of the labor market is given by the following: (6.6) j with log j h h wj h Hh. ujh , where wj is the general level of earnings in segment j, as given by the solution of the aggregate CGE model; Hh stands for the human cap- ital characteristics of individual h (essentially education, age, and gender); jis a vector of coefficients specific of segment j; and ujh is a residual term for the effect of unobserved characteristics on indi- vidual productivity in segment j. The key assumption is that the elasticity of individual labor productivity with respect to human capital characteristics is segment specific. Participation decisions are taken by comparing the potential wages in the various market segments to a reservation wage, 0, h given by the following: (6.7) ln 0 0 h Hh Zh u0h , where 0stands for the semi-elasticities of the reservations wage with respect to the observable characteristics of workers, is for the semi-elasticities with respect to household characteristics, and Zh and u0h summarize the effect of unobserved variables. The reserva- tion wage, 0, is not directly observed and must be inferred from the h observed participation behavior of individuals in the sample. With these three distinct wage variables, it is possible to represent the decision process of an individual who has to choose among three alternatives: being inactive, working in sector 1, or working in sec- tor 2. To take into account the possible imperfection of the labor market and entry restrictions in the formal sector, it is convenient to DISTRIBUTIONAL EFFECTS OF TRADE REFORM 187 introduce a cost of entry in that sector. A simple assumption is that this cost is proportional to the formal wage, so that it represents something like the waiting time to get a job in the formal sector. Accordingly, the net (logarithm of the) gain in the formal segment of the labor market can be defined as follows: (6.8) ln 1 h uch , where uch stands for the logarithm of the proportion of earnings in the formal sector that is actually received by the worker--after tak- ing into account the cost of entry. The labor market may be said to be perfectly competitive when this cost is nil, which implies that uch is nil instead of negative. Taking into account the cost of entry into the formal sector, the employment decision process can be described by the following set of conditions: formal employment: ln 1 uch ln 2 , and ln 1 uch ln 0 ; h (6.9) informal employment: ln 2 ln 1 uch, and ln 2 ln 0 ; h inactive: ln 0 ln 1 h uch, and ln 0 ln 2 . h Observing the sector of employment of individuals (and their earnings in that sector) and making some simplifying assumptions on the distribution of the unobserved terms, u..h, it is possible to esti- mate the parameters of equations (6.6) and (6.7). Under the assump- tion of a normal distribution of the unobserved terms, Magnac (1991) provides a two-step estimation method of the Heckman type that starts with a bivariate probit on the sectors of employment and participation. It is not possible, however, to estimate precisely the unobserved components, u..h, of the various earning equations. Only one unobserved term may be directly derived from the estimation for employed individuals. It is the residual of the earning equation in the segment of the market where they are employed. For the other equations, these unobserved terms are drawn randomly in the appropriate conditional distributions. For instance, the estimated û1h, ûch, and û0h terms for an individual employed in the informal sec- tor must be drawn according to the following rule: û1h N(0, ) N(0, ) N(0, ) 1 û0h 0 ûch c (6.10) û1h ûch lnw2 lnw1 Hh(^2 ^1) û2h û0h lnw2 Hh(^ 2 ^0) Zh^ û2h . 188 BOURGUIGNON AND SAVARD For the sake of brevity and given the scope of the present chap- ter, the detail of the estimation procedure and the results obtained in the case of the Philippines are not provided. A simplified version of Magnac's (1991) method was applied to household heads with the additional assumption that all employed individuals in a household were in the same segment of the labor market as the head.5 This micro specification of the labor market implied several departures from the simple Walrasian model discussed in the previ- ous section. At the top level, formal and informal are considered as two different types of labor input in the production process of the various sectors of the aggregate CGE model. A basic labor market imperfection is introduced by postulating a fixed real wage (w1/P) in the formal segment of the labor market,6 which results in some rationing, whereas the informal labor market is supposed to clear through a flexible wage w2. Numerous CGE models actually repre- sented the labor market in that way (see, for instance, Fortin, Marceau, and Savard 1997; Decaluwé, Martens, and Savard 2001; Agénor and El Aynaoui 2005). Labor supply in the two markets is taken as exogenous in the first iteration, but results from the micro simulation are used to endogenize the labor supply in the subsequent iterations. At the micro simulation level, the general level of wages obtained in the top part of the algorithm is used to scale up or down the potential earnings of individuals in the various segments of the labor market with respect to initial estimates of individual earnings. The reservation wage is reasonably assumed to be scaled up or down using a consumer price index (P). These potential earnings can then be used to revise the employment choices of households. Some rationing will possibly take place if employment in the formal sector is below the spontaneous supply of workers. The rationing scheme is analyzed in further detail below. The consequences of such a functioning of the labor market have already been analyzed in the CGE literature, particularly in the RH context (see, for instance, Thomas and Vallée 1996; Fortin, Marceau, and Savard 1997; Savard and Adjovi 1998; Devarajan, Ghanem, and Thierfelder 1999; Agénor, Izquierdo, and Fofack 2003). Evaluating the aggregate and distributional impact of this imperfection of the labor market requires more care in a fully dis- aggregated representation of the population of households. Evaluating the aggregate supply of labor in the two segments of the labor market at iteration m 1 in the micro simulation module can be done by counting the number of people in the various cases defined by equation (6.9) with the prices and wages of iteration m. Thus, the total labor supply L.1 m 1 to the formal sector is given by DISTRIBUTIONAL EFFECTS OF TRADE REFORM 189 the cardinal of the following: {ln 1,m ln 2,m, and ln 1,m ln 0,m } , h uch h h uch h which also writes as follows: (6.11) w1m ln w2m Hh(^1 ^2) û1h ûch û2h 0 and Lm 1 Card , .1 w1 m lnPm Hh(^1 ^0) Zh û1h ûch û0h 0 where Card{C} stands for the cardinal of the set of individuals defined by the conditions C. Likewise, the labor supply, Lm , to the 1 .2 informal sector is given by the following: (6.12) w1m ln w2m Hh(^ 1 ^ 2) û1h ûch û2h 0 and Lm 1 Card . .2 w2 m ln Pm Hh(^ 2 ^ 0) Zh û2h û0h 0 Other people are inactive. Now consider the case in which the formal labor supply exceeds the demand at the fixed real wage w1 m Pm. Then some people who want to work in that segment of the market will not find a job there. Likewise, it may be the case that the demand exceeds the supply, in which case the formal sector will have to attract workers who were initially inactive or employed in the informal sector. How is this adjustment implemented in the model? The entry cost in the formal sector is used to adjust the labor sup- ply in the formal labor market to match the actual demand. If the demand of formal labor initially exceeds the supply of workers, then the cost of entry in the formal sector is reduced in the same proportion for all individuals, so that some individuals will move from the informal to the formal sector and others will switch from inactivity to formal employment. In the opposite case of excess sup- ply in the formal labor market, the adjustment takes place by increasing the cost of entry. Fewer people are then willing to work in the formal sector, some of them preferring the informal sector and others becoming (or remaining) inactive. Modifying the entry cost is like changing ûch by the same amount for all individuals. Equilibrating the formal labor market through entry cost thus requires determining the amount m 1 by which all the ûch terms 190 BOURGUIGNON AND SAVARD must be modified for the constrained labor supply to the formal sector Lm 1 ( m 1 ) to be equal to the demand, Ldm. From equation .1 1 (6.11), it can be seen that m 2 is given by the solution of the fol- lowing equation: (6.13) Lm 1( m 1 ) Ldm .1 1 w1 m ln w2 m 1 m Hh(^ 1 ^ 2) û1h ûch û2h 0 and Card . w1 m ln m 1 Pm Hh(^ 1 ^ 0) Zh û1h ûch û0h 0 This modification of the cost of entry in the formal sector also modifies the labor supply to the informal segment of the market, which now writes as follows: m 1 (6.14) L.2 ( m 1 ) w1m ln w2 m 1 m Hh(^ 1 ^ 2) û1h ûch û2h 0 and Card . w2 m ln Pm Hh(^ 2 ^ 0) Zh û1h ûch û0h 0 Taking into account the imperfection of the labor market is thus equivalent to replacing the labor supplies coming from the BU part in the initial algorithm by the constrained labor supplies, Lm ( 1 m 1). .j In effect, the equilibrium on the formal labor segment always holds--at the prices of iteration m--so that, practically, the introduc- tion of a fixed real wage in that segment of the market is equivalent to replacing a quantity variable of the initial model by a shadow price, that is, the cost of entry into the formal labor market. It is also possible to dispense with this shadow price and imple- ment more directly the rationing process. Consider the case of an excess supply in the formal labor market. The preceding mechanism is equivalent to expelling N L.1 m 1 L1 dm individuals from the notional labor supply in the formal labor market. To see which indi- viduals will be expelled, it is sufficient to rank all people in L.1 m 1 given by equation (6.11) according to the following criterion, Gh: Gh Inf Hh(^ 1 ^ 2) û1h ûch û2h, (6.15) Hh(^ 1 ^ 0) Zh ^ û1h ûch û0h, which happens to be independent of the price system and thus can be established once for all individuals when estimating the labor DISTRIBUTIONAL EFFECTS OF TRADE REFORM 191 supply model in equation (6.9). By definition of the notional labor supply, there are Lm 1 .1 individuals such that the criterion Gh is pos- itive. Rationing employment in the formal segment of the labor mar- ket to only Ldm individuals is equivalent to selecting the Ldm indi- 1 1 viduals with the highest Gh. Put another way, it is equivalent to expelling from the notional labor supply the N individuals with the lowest score Gh. Whether those individuals will go to the informal sector or to inactivity depends on which one of the two terms on the right-hand side of equation (6.14) is binding. The opposite proce- dure can be applied to the case in which there is excess demand in the formal segment of the labor market. The preceding procedure has to be applied independently for the various types (n) of labor, depending on whether there is a poten- tially binding minimum wage for both skilled and unskilled work- ers. Practically, however, the minimum wage is assumed to be bind- ing only for skilled workers. In other words, the market for unskilled workers is assumed to clear through the wage scale factor w2. The description of the way the segmentation of the labor market and the rigidity of wages are taken into account in the present IMH model of the Philippine economy is now complete, except for a last detail. The framework that has just been presented is based on the heterogeneity of individual productivities, . This heterogeneity h. implies that, at the aggregate level, there is a relationship between the number of people being employed in one segment, j, of the labor market and their productivity. As can be seen from the ranking given by equation (6.14), people who leave the formal labor market in case of a contraction are not taken randomly in the initial popula- tion of employees in the formal sector. The same is true of those who would join if the formal sector were expanding, and conse- quently, it is true of the informal segment of the labor market. This means that the average productivity of workers in the two segments of the labor market depends on the number of people working there. This endogeneity of the labor productivity must be taken into account in the CGE part of the model. Thus, for each type of labor, the BU part of the algorithm must return at each iteration not only the (constrained) total labor supply in the two segments j 1, 2 of the labor market, Lm ( 1 m 1 ), but also the mean productivity of .j the two groups of workers, ^m ( 1 m 1 ), which is a function of the .j cost of entry into the formal segment. The original algorithm has now been generalized to noncompet- itive mechanisms that affect microeconomic agents in a way that depends on their comparative individual characteristics. These mechanisms are likely to matter in determining the distributional impact of a policy reform. The fact that this can be done in a rather 192 BOURGUIGNON AND SAVARD simple way within the present sequential algorithm may be an important advantage in comparison with the simultaneous resolu- tion of a fully integrated micro-macro CGE model. Application to a Trade Reform in the Philippines This section discusses the results obtained in applying the above described IMH model to the study of the distributional effects of a trade reform in the Philippines. Important details of the actual spec- ification of the model are discussed before focusing on the results of a few simulations. The CGE model used in this chapter is based on the EXTER model (externally open model) of a standard small developing econ- omy provided by Decaluwé, Martens, and Savard (2001), with extensions that take into account the dualism of the labor market. The CGE model is disaggregated into 20 sectors and includes 873 equations. The bottom part of the overall micro-macro model is based on a sample of 39,520 households. The main data sources used in that exercise are the 1997 Family Income and Expenditure Survey (FIES), the Labor Force Survey (LFS) for 1997 to 1998, and the 1990 social accounting matrix (SAM). The FIES and LFS were used first to estimate the structural econometric labor supply model described in equation (6.9) and then were used in the micro simulation module. Both surveys are based on the same master sample, and 98 percent of the households are found in the FIES and LFS. The FIES and SAM were used for the macro CGE module. The main data manipulation required was the conversion of the FIES nomenclature into the national accounts nomenclature found in the SAM. This conversion was relatively easy and straightforward, because the level of aggregation was quite high in the FIES. It is not necessary to have perfect consistency between the income and expenditure accounts at the micro and macro levels, because the effects are transmitted from the aggregate results of the micro module to the macro CGE model through per- centage variations. This way of linking the micro module and the CGE model also avoids the adjustment of the structure of house- holds' expenditure observed in the micro data. The household micro simulation module relies on a representa- tion of the spending and labor supply behavior of the household. Household consumption is modeled with a linear expenditure demand system (LES) based on total consumption expenditures. The calibration method proposed by Dervis, de Melo, and Robinson (1982), with all households having the same income elasticity for all goods, as well as the same Frisch parameter, is used. The resulting DISTRIBUTIONAL EFFECTS OF TRADE REFORM 193 demand for consumption goods is made consistent with observed spending in the FIES through household-specific additive shift para- meters.7 Total expenditure is derived from total income after savings and income taxes. Both the savings rate and the income tax rate are taken to be fixed and household specific.8 All transfers received and given are exogenous. On the income side, capital endowments are supposed to be pro- portional to the level of capital income observed in the 1997 FIES. Labor incomes are given by the model discussed in the preceding section. Based on information provided in the FIES and LFS, work- ers were classified as employed in the formal or informal sector depending on their occupations as specified in the survey.9 At the macro level, the main features of the CGE model are as fol- lows.10 Producers in the various sectors of the economy are assumed to maximize profits in a fully competitive environment subject to a Cobb-Douglas production function for effective labor and capital and to a Leontief function for intermediate inputs. In each sector, it is assumed that formal and informal firms produce the same aggregate good and that they can be aggregated in a single representative firm, employing simultaneously formal and informal labor.11 The aggre- gate labor input in the Cobb-Douglas production function is sup- posed to be the cost-minimizing combination of formal and informal labor under the assumption of a constant elasticity of substitution (CES) between them. In all sectors, the model distinguishes between skilled and unskilled labor. Both types of labor are fully mobile across sectors, but capital is assumed to be fixed, which generates branch- specific returns to capital. This assumption is consistent with a medium-term perspective on the effect of trade liberalization. In terms of trade, the Philippines is assumed to be a small open economy. The demand of imported goods is derived from Arming- ton's (1969) specification of a CES between domestic and foreign goods. Likewise, domestic production is allocated to the domestic or foreign markets (exports) through a standard constant elasticity of transformation (CET) function. On the consumer side, the income of the RH is composed of earn- ings from skilled and unskilled labor, capital payments, dividends, and transfers from other agents (households and remittances from abroad). As consumption is determined by the micro module, the aggregate saving rate of households is implicitly allowed to vary. The government levies an income tax (on households and firms), taxes on goods and services, and import duties and transfers from the rest of the world. Its expenditures include various subsidies, the production of public services, and public investments. As for closure rules, total investment is exogenous and current gov- ernment expenditures are scaled up or down to balance investment 194 BOURGUIGNON AND SAVARD and savings. The exchange rate is endogenous and adjusts to meet an exogenous current account constraint. Finally, the gross domestic product (GDP) deflator is used as a numéraire. These closure rules are equivalent to assuming that the burden of the adjustment to any reform is born by households, either directly through their income or expenditures or indirectly through current government expenditures. The policy simulation reported in this chapter consists of an across-the-board reduction in import duties of 30 percent. This is a rather conventional exercise in the analysis of the effects of trade reforms. The Philippines is a rather open economy--the average tar- iff rate is around 12 percent--and thus no large effect is expected from such a reform. Because initial protection is heterogeneous across sectors, such a reform entails some restructuring of the econ- omy, which should have some effect on the price system and on the distribution of welfare. The simulation is performed under different specifications of the overall micro-macro model and assumptions about the functioning of the labor market (see table 6.1). In a first specification, the labor market is supposed to be fully competitive with market-clearing wages in both the formal and informal segments. In the second spec- ification, the real wage is assumed to be fixed in the formal sector, and adjustments are supposed to take place in the way described in the preceding section. The objective of the third specification is essentially methodological. Only the first iteration of the algorithm described above was performed, which is equivalent to the MSS approach. The objective of the comparison of the second and third specifications is to get some idea about the consequences of ignoring the feedback effects from the micro simulation to the macro part of model and to get some idea about the overall precision of the MSS approach. To maximize the precision of the MSS approach, the first iteration at the CGE level was performed with a single RH that Table 6.1 Definition of Model Specification Used Definition of specification Acronym Integrated multihousehold model with flexible wages IMH_FL_w Integrated multihousehold model with fixed formal wage IMH_FX_w1 Micro simulation sequential approach with flexible wages MSS_FL_w Micro simulation sequential approach with fixed formal wage MSS_FX_w1 Source: Authors' classification. DISTRIBUTIONAL EFFECTS OF TRADE REFORM 195 approximated the aggregate consumption and labor supply behav- ior implied by the micro part of the algorithm. Aggregate Effects The direct effect of liberalizing trade is to reduce the price of imports and, therefore, to increase the domestic demand for the most protected goods. Given the fixed current account balance, the real exchange rate has to go up, thus reducing imports and increasing exports to balance out the current account. More important, the government income is negatively affected by the drop in tariffs. As total investment is exoge- nous, government consumption must be cut down to balance savings and fixed investments. It is this direct revenue effect of the trade reform that produces the most important general equilibrium effects in the model. Other effects are due to the shifting of part of consumption from nontradable to tradable goods caused by change in relative prices (numerical results are shown in the first column of table 6.2). The strong reduction in government expenditure puts downward pressure on the labor market as civil servants are laid off. In the first and third specifications, real formal wages are flexible and they are Table 6.2 Macro Results IMH_FL_w IMH_FX_w1 MSS_FL_w MSS_FX_w1 (percentage (percentage (percentage (percentage Variable Base change) change) change) change) Gross domestic product 104,510.0 0.04 0.69 0.05 1.27 Real household income 86,476.0 1.45 1.13 1.37 0.64 Household real consumption 72,607.0 1.00 1.40 1.35 1.03 Formal wage (index) 1.0 3.18 0 3.86 0 Informal wage (vs. formal) 0.5 0.61 1.25 0.67 0.46 Government income 20,367.0 8.39 8.43 8.32 8.84 Real public spending 16,818.0 6.63 11.34 9.48 13.02 Real investment 23,684.0 2.25 2.26 2.02 2.17 Firm income 26,172.0 0.60 0.55 0.74 0.14 Firm savings 7,810.0 1.04 0.95 1.29 0.24 Employment rate 0.8316.0 0 0.66 0 2.03 Exchange rate (index) 1.0 0.23 0.30 0.17 0.27 Source: Authors' calculations. Note: The Base column units are billion pesos, except as otherwise specified. 196 BOURGUIGNON AND SAVARD driven down by this drop in labor demand. This reduces the overall cost of labor and pushes employment up in other sectors, including informal employment. As the public sector is more intensive in formal labor, the end result is a drop in real formal wages and a slight increase in informal wages. This result is partly due to the fact that nontradable sectors, which gain relative to the tradable sectors from the change in relative prices, are on average more intensive in informal labor. On the supply side, these changes in relative wages induce some workers to move from the formal to the informal segments of the labor market and induce other workers to become inactive, in accordance with the labor supply model discussed above. The change in relative wages is not big enough to produce substantial changes in output. Thus, GDP decreases only slightly. On the demand side, the drop in government expenditures is compensated by an increase in real invest- ment, which is caused by the drop in the relative prices of imports and, to a lesser extent, by the drop in household consumption. Things are somewhat different when wages are assumed to be rigid in the formal sector (see table 6.2, second column). As before, formal workers are laid off by the government sector, but as formal wages do not fall, these workers do not find jobs in the other sectors and move to inactivity or to informal work at a lower productivity. This excess supply of formal labor in turn leads to a slight drop in informal wages--unlike in the preceding specification. Interestingly enough, this movement of labor comes with some changes in the average pro- ductivity of labor in the two sectors. Average productivity, of both skilled and unskilled workers, increases by 1.03 percent in the formal sector and decreases by 0.61 percent in the informal sector. Overall, GDP falls by 0.69 percent in the IMH model, while 0.7 percent of the labor force goes to inactivity. On the demand side, the price rigidity induced by wage rigidity reduces the substitution of public spending by investment or household consumption. In effect, households are more severely affected by the trade reform, and their real income falls by substantially more than with flexible wages. Because of the lack of response of domestic prices in sectors employing predominantly for- mal workers, the devaluation is more pronounced, amounting to 50 percent more than in the full-employment simulation. Sectoral Effects Changes in the structure of production shown in table 6.3 are easily interpreted in light of the preceding arguments. These changes result from the combination of four types of effects: (1) the contraction of public spending; (2) changes in labor costs as analyzed above; (3) changes caused by the drop in tariffs--lower input prices but DISTRIBUTIONAL EFFECTS OF TRADE REFORM 197 Table 6.3 Structural Effects of the Trade Reform, Output (Value Added) Change by Sector Base IMH_FL_w IMH_FX_w1 MSS_FL_w MSS_FX_w1 (pesos, (percentage (percentage (percentage (percentage Sector billion) change) change) change) change) Palaya and corn 5,198 0.29 0.41 0.42 0.13 Fruit and vegetable 4,211 0.19 0.51 0.28 0.26 Coconut 1,790 0.38 0.63 0.47 0.21 Livestock 4,474 0.42 0.68 0.57 0.37 Fishing 3,997 0.27 0.81 0.37 0.28 Other agriculture 1,846 0.26 0.92 0.20 0.08 Logging and timber 857 0.54 0.76 0.65 0.40 Mining 1,604 1.50 1.59 1.62 1.16 Manufacturing 13,112 0.83 0.72 0.96 0.30 Rice manufacturing 2,023 0.46 0.56 0.61 0.27 Meat industry 2,081 0.57 0.78 0.77 0.41 Food manufacturing 3,696 0.03 0.39 0.03 0.06 Electricity, gas, and water 2,341 0.04 0.52 0.04 1.05 Construction 6,848 1.15 1.49 1.26 1.24 Commerce 15,150 0.64 0.68 0.77 0.28 Transport and commerce 5,206 0.44 0.46 0.55 0 Finance 3,580 0.02 0.81 0.05 1.33 Real estate 7,314 0.87 0.49 1.16 0.24 Services 6,960 0.91 0.31 0.98 0.39 Public services 12,223 6.06 10.40 7.32 12.24 Source: Authors' calculations. a. Palay is the term used in the Philippines to designate the rice grain in its husk that cannot be consumed directly. also enhanced competition of imports; and (4) changes caused by the devaluation, which dampens the effect of the tariff cut and favors export sectors. The contraction of the public sector is more pro- nounced with rigid wages because the effect of lower revenues caused by the fall in tariffs is not compensated for by a drop in labor cost. It can be seen that the difference is quite substantial--the drop is 60 percent more pronounced with rigid formal wages. Through backward links, the drop in public spending has a negative effect on various sectors. This effect is uncompensated by other effects in the finance and utility sectors when wages are rigid. Output thus falls in both sectors, but when wages are flexible, it is compensated by the drop in labor cost. 198 BOURGUIGNON AND SAVARD The flexibility of labor costs tends to compensate for the reduction of protection in sectors exposed to foreign competition, except when those sectors are relatively more intensive in informal labor, the cost of which was seen to increase in the flexible wage specification. This is the case in the other agriculture and food manufacturing sectors, the output of which tends to decline. This fall is not observed in the case of rigid formal wages, because the cost of informal labor tends to fall, thus compensating for the enhanced exposition of these sectors to foreign competition. Favorable consequences of the trade reform and the accompanying devaluation of the domestic currency can be seen on export-oriented sectors like mining, logging, and some agricultural subsectors. The drop in tariffs is favorable in sectors with import-intensive inputs like construction. In all these cases, the direction and intensity of output changes are comparable across the flexible and rigid formal wage specifications, with the slight differences in output changes being mostly attributable to the differences in labor costs. Overall, it is interesting to see that assumptions made about the flexibility or rigidity of formal wages do make a difference in both the aggregate and the structural effects of trade reform. The con- traction of the public sector is more pronounced in the case of rigid wages, which entails a contraction of GDP and a contraction in sec- tors that depend on public demand. More than this rather mechan- ical effect occurs, however. Differences in the changes in labor costs and the asymmetry in the changes in the cost of informal labor are responsible for additional structural effects. All of these effects would be magnified if the trade reform had been more ambitious, for example, with the total elimination of tar- iffs. In that case, however, the assumption that most of the adjust- ment would be borne by recurrent public expenditures would have been untenable. Part of this assumption should have been applied to investment, with the effects being difficult to analyze in an essen- tially static framework. Poverty Effects The main objective of the methodology used in this chapter is to derive the impact of policy reforms defined at the macro level on the distrib- ution of income and, in particular, on poverty. Table 6.4 shows the impact of the trade reform analyzed in this section on poverty accord- ing to the three specifications that have been used. Poverty is summa- rized by the three usual Foster-Greer-Thorbecke (1984) indicators. FGT0 stands for the poverty headcount, which is the proportion of people below the poverty line. FGT1 is the depth of poverty, which measures the amount of money that should be transferred to the poor DISTRIBUTIONAL EFFECTS OF TRADE REFORM 199 Table 6.4 Effects on Poverty (FGT Poverty Indexes) for the Whole Population and by Education Groups Poverty IMH_FL_w IMH_FX_w1 MSS_FL_w MSS_FX_w1 index Groups Base (percent) (percent) (percent) (percent) FGT0 All 0.311 2.19 1.46 2.21 1.79 FGT1 All 0.096 2.82 1.67 2.90 2.25 FGT2 All 0.040 3.63 1.75 3.54 2.68 0 0.564 1.33 1.48 1.47 1.55 1 0.501 1.91 1.38 2.15 1.58 2 0.384 2.02 0.81 1.79 1.25 FGT0 3 0.317 2.78 2.11 1.84 2.45 4 0.184 3.93 3.08 3.97 3.28 5 0.092 2.78 0.34 3.38 2.06 6 0.021 3.91 1.96 4.63 3.42 0 0.185 2.47 2.47 2.13 2.46 1 0.168 2.47 1.75 2.59 2.15 2 0.116 2.97 1.56 3.09 2.17 FGT1 3 0.090 3.23 1.79 3.26 2.57 4 0.048 3.31 1.68 3.64 2.29 5 0.022 3.48 1.56 3.85 2.51 6 0.005 3.78 0.03 4.76 3.42 0 0.080 3.36 3.07 2.87 3.06 1 0.075 3.26 2.12 3.28 2.68 2 0.048 3.86 1.60 3.74 2.54 FGT2 3 0.035 3.94 2.07 3.87 3.30 4 0.018 4.04 0.62 4.24 2.11 5 0.007 4.52 6.37 4.62 2.19 6 0.002 4.04 3.35 4.44 3.73 Source: Authors' calculations. Note: FGT Foster-Greer-Thorbecke (1984) indicators; FGT0 poverty headcount; FGT1 depth of poverty; FGT2 severity of poverty. Education groups are defined in annex 6B. to eliminate poverty under perfect targeting. It may be expressed as the product of the headcount and the poverty gap, or the relative distance at which poor people are from the poverty line. Severity of poverty, FGT2, corresponds to the same concept but uses the square of the dis- tance from the poverty line, and thus gives more weight to extreme poverty. Because micro-macro modeling allows for taking into account the whole distribution, it would be possible to use any other poverty indicator available in the literature. A more comprehensive represen- tation of the change in the distribution caused by the trade reform is shown below. At this stage, the analysis concentrates only on these three poverty measures in the whole population or in groups defined by the education of household heads. Before getting into the detail of table 6.4, it is worth insisting on the various forces behind changes in the three poverty indexes. There are two sources of changes. The first is purely distributional. It arises 200 BOURGUIGNON AND SAVARD because the income of different households changes in different ways and proportions caused by the trade reform and its general equilib- rium effects on household incomes. The second is found in the changes that take place in the structure of prices and that modify the real income of households, possibly in different ways depending on their consumption basket. Two possibilities can be used to take this into account. The first one corrects all changes in (nominal) household incomes by a price index that is based on their consump- tion, maintaining the poverty line (nominally) fixed. The second approach modifies the poverty line only, using a price index meant to fit the average consumption basket of the poor. This second approach is pursued in the calculations summarized in table 6.4, with the price index being based on the minimum basket of consumption goods used in the LES representation of household consumption behavior in the micro part of the micro-macro algorithm.12 This effect of the changes in the final price of goods on the poverty line--that is, the price effect--is undoubtedly negative. The trade lib- eralization directly reduces the price of imported goods and of those domestic goods that compete with them, and indirectly reduces the price of goods that use imported inputs. Other things equal, poverty would thus fall, this being true for all poverty indexes. But one has to consider the income effect or, in other words, the way the nominal income of households is modified through the general equilibrium effects. With flexible wages, it may be expected that the income effect will reduce poverty further. It is true that formal labor incomes are lower because of the trade reform, but the opposite is true of infor- mal labor incomes. Moreover, it was seen that changes in the volume and the structure of employment were minimal. As informal labor incomes are likely to be of greater importance among the poor, it may thus be expected that the income effect also contributes to reduc- ing poverty when wages are flexible. This pattern is precisely what can be seen in the fourth column of table 6.4. For the whole population, poverty is falling because of the trade reform, and the higher the severity of the poverty index, the higher it falls. This shows that the price effect and the income effect are at work. It can be proven that, by itself, the price effect should reduce FGT1 and FGT2 more or less in the same proportion.13 The fact that FGT2 falls by proportionally more than FGT1 means that the relative income of the poorest households is increasing. The same pattern may be observed for the various education groups, except for the most educated ones--for which changes in table 6.4 may not be relevant because of low initial poverty. But the income effect is likely not to be important in those groups for which most labor income is likely to come from the formal segment of the labor market. DISTRIBUTIONAL EFFECTS OF TRADE REFORM 201 The examination of the changes in poverty indexes by education group with the flexible wage specification seems to suggest that the higher the education level of household heads, the stronger the impact of the trade reform on poverty. This correlation must be taken with much care. If, indeed, most of the drop in poverty with flexible wages comes from the lowering of the poverty line, then there may be some- thing purely mechanical, evident in the fact that the change in poverty indexes increases with the education level. For instance, the elasticity of the poverty headcount (FGT0) with respect to the poverty line, z, is given by the ratio zf(z) F(z), where f( ) is the density of the distribu- tion of income and F( ) is the cumulative function. It is conceivable that this ratio changes in a systematic way with the level of poverty across education groups. Checking this would require a rather detailed estimate of the density of the distribution of income in the various groups. More interesting is the fact that different patterns in poverty changes are obtained under the assumption of rigid wages in the formal sector and the rationing scheme that is supposed to apply to the labor market in that case. At the aggregate level, first, it can be seen that the drop in poverty is much less important than with flex- ible wages. Moreover, the difference tends to be bigger when mov- ing from headcount to depth of poverty and then to severity of poverty. Not only are fewer people lifted out of poverty, but welfare improvements are also lower for the poorest. As the initial effect of the drop in tariffs on the poverty line is com- parable in the two specifications, the explanation of that difference must be found in different income effects. Indeed, the fact that infor- mal labor incomes decrease with rigid formal wages and that some workers withdraw from the labor force has an impact on poverty that goes in the direction opposite of the decline in the poverty line and that is opposite to the income effect seen with flexible formal wages. The same phenomenon is behind the disappearance of the pattern that was present in the drop in poverty by education groups with flex- ible wages. Opposite income effects affect all groups--not only the low-education groups that are more affected by the drop in informal labor incomes, but also the higher-income groups in which some peo- ple are forced into informal work or into inactivity. At this stage, the interest of modeling explicitly the effect of the trade reform on the distribution of income at the individual level appears quite clearly. The phenomena just described could not be considered properly using a traditional RH approach. The impact of such a reform on poverty depends on individual household cir- cumstances: how much of the initial income comes from informal labor, how many people are forced out of their jobs in the formal 202 BOURGUIGNON AND SAVARD sector, and so on. An aggregate analysis would, at best, be able to give information on the effect of lower prices on the poverty line, and possibly make a tiny change in the mean income of some house- hold groups. Changes in the Overall Distribution Rather than focusing on poverty, the macro-micro framework used in this chapter allows for an analysis of the change in the whole distrib- ution of income within the population. In what follows, this change is described by the growth incidence curve that shows the changes in real income by percentile of households before and after trade reform. Figure 6.2 illustrates the results of the specification with a flexible (IMH_FL_w1) and a fixed real wage in the formal sector. Figure 6.2 indeed shows substantial differences in the distribu- tional impact of the reform with these two specifications. The top of the distribution, between percentiles 88 and 97, are the relative losers Figure 6.2 Comparative Growth Incidence Curves for Total Population: IMH_FX_w1 versus IMH_FL_w 0.0350 0.0300 0.0250 income 0.0200 real in 0.0150 0.0100 change 0.0050 0 percentage 0.0050 0.0100 0 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 percentile average change IMH_FX_w1 approach income growth curve IMH_FX_w1 average change IMH_FL_w approach income growth curve IMH_FL_w Source: Authors' calculations. DISTRIBUTIONAL EFFECTS OF TRADE REFORM 203 (although net gainers) in the specification; flexible wage rates and the top two percentiles are relative winners. Conversely, the growth incidence curve is rather flat with the fixed wage rate specification except for the very bottom percentiles, which are both relative and absolute losers. The explanation of that difference is both simple and interesting. With the flexible wage rates, it was seen in table 6.2 that the trade reform had a negative effect on the wage rates in the formal sector of the economy. Households that derive income almost exclusively from labor in the formal sector belong mostly to the 87 to 97 per- centiles of the distribution, whereas other households with members in the formal sector derive income from other sources as well, such as land or capital. Likewise, households in the top three percentiles include business owners in the export and informal sectors. These effects largely disappear with the fixed wage specifica- tion, because costs and profits are essentially rigid. In that case, what drives the negative part of the growth incidence curve in the first percentiles is the fact that, because of the rigidity in the labor market, some households lose their job and end up inactive. In effect, households in the bottom percentiles are being replaced by households that were initially at a higher rank in the distribution but that are affected by job losses. In other words, the relative drop in formal wages with the flexible wage specification is replaced by formal employment contraction, with some workers being absorbed by the informal sector--moving down in the income ranking but with limited impact on average income--and other workers losing jobs. Comparison of Micro Simulation Sequential and the Complete Algorithm It is now time to examine the third and fourth specifications appear- ing in tables 6.2­6.4, which are simply the first TD iterations of the micro-macro model rather than its complete resolution. The issue is to determine whether this simpler micro simulation approach used by several authors is a satisfactory approximation of the overall effects of the simulated reform. The answer to that question depends on the type of result and the specification that is being examined. With the flexible wage specification, gaps between the first iteration and the complete res- olution are quite limited. The first iteration only exaggerates the contraction of public spending, because it misses the household consumption feedback in the complete algorithm, even though the macro model tries to mimic the aggregate consumption behavior 204 BOURGUIGNON AND SAVARD of the whole population of households. The same is true with the rigid wage specification except for the fact that, because of this rigidity, the gap in public spending feeds into an equivalent gap in GDP and employment, exaggerating the negative effect of the reform. The same remark applies to sectoral changes (table 6.3). Gaps are limited in the flexible wage specification, with the first iteration underestimating systematically final changes in the complete model. In the fixed wage specification, differences are bigger in size and in the direction of the effect. Discrepancies can be seen between the first iteration and the complete model that come essen- tially from the overestimation of the drop in real public spending in the former model. This bias generates negative biases in all sec- tors that provide inputs to the public sector (such as utilities, finance, and services). Poverty figures show the same pattern. Aggregate results are comparable with the flexible wage specification, and they tend to overestimate the drop in poverty with the flexible wage specifica- tion. Poverty effects in the flexible wage case essentially are due to price changes that have been seen to be rather satisfactorily approximated by the first iteration. Results are more surprising in the rigid wage case. Because it tends to overestimate the drop in GDP and employment, one would have expected the initial nega- tive impact on poverty to be smaller than at equilibrium. The explanation of that correction is to be related to sectoral biases for the pure TD rigid wage specification. This result is fully confirmed by the comparison of the growth incidence curves of the MSS and complete IMH rigid specifications (figure 6.3). The first iteration simply misses the increase in poverty because of job losses at the very bottom of the distribution. This does not mean that no job loss occurs in the first iteration, but rather that these losses do not take place in the same part of the distribution because they are not located in the same sectors. It is thus at the sectoral level that the MSS approach results dif- fer the most from the results of the IMH approach. This is counter- intuitive, and larger deviations for the aggregate variables were expected given the incomplete feedback effects of the MSS approach. Results would have been quite different if the initial resolution of the CGE had been undertaken with a single household whose behav- ior had little to do with the aggregation of individual household behavior in the micro database. At the same time, this relative impre- cision of the first-round micro effects points to the interest of resolv- ing the whole model, or at least to iterate between the macro and the micro parts of the full micro-macro model. DISTRIBUTIONAL EFFECTS OF TRADE REFORM 205 Figure 6.3 Comparative Growth Incidence Curves for Total Population: IMH_FX_w1 versus MSS_FX_w1 0.0125 0.0100 0.0075 income 0.0050 real in 0.0025 0 change 0.0025 0.0050 percentage 0.0075 0.0100 0 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 percentile average change IMH_FX_w1 approach income growth curve IMH_FX_w1 approach average change MSS_FX_w1 approach income growth curve MSS_FX_w1 approach Source: Authors' calculations. Conclusion This chapter tries to take into account the microeconomic conse- quences of a macro policy through the integration of a micro data- base of households within a conventional CGE framework. This was done in two ways: (1) through a conventional TD approach with an aggregate CGE model feeding a micro simulation module (the MSS model) without any kind of feedback at the macro level; and (2) through iterations between those two modules to obtain the solution of an IMH model. In the simulation of trade liberalization policies undertaken in this chapter, differences between the two approaches were found to be important in the presence of rigidities in the labor market, which led to some rationing situations. In that case, the MSS approach tended to overestimate the negative impact of the reform on GDP and employment and, consequently, to underestimate its effect on reduc- ing poverty. In comparison, differences between the two approaches appeared to be minor in the presence of flexible wages, as most micro 206 BOURGUIGNON AND SAVARD effects were essentially channeled through price-induced changes in real income, rather than quantity adjustment at the micro level. It must be stressed, however, that this result was obtained because the CGE model included a rather satisfactory approximation of the aggre- gate consumption behavior of the population of households, which itself required several simulations at the micro level. This experimentation with IMH models thus suggests that the standard TD micro simulation approach to the distributional impact of policies is satisfactory as long as no purely quantitative adjustment is assumed to take place at the micro level. This fits intuition. More work is needed to experiment and measure whether the difference in results is relatively robust to change in behavioral hypotheses, macro closure rules, or to different policy scenarios. Conversely, it is inter- esting to know that simple iterative techniques can be used to solve integrated models with more complete micro-based market adjust- ment mechanisms. Annex 6A Table 6A.1 Labor Supply Model Estimation Results Regressor Coefficient Standard error t-statistic Prob |t| Probit Constant 1.61683 0.46963 3.44281 0.00029 Education 0.14937 0.00932 16.02265 0 Age 0.10990 0.02984 3.68280 0.00012 Age2 0.00121 0.00030 3.99504 0.00003 Experience 0.02414 0.00976 2.47298 0.00671 Sex of head 0.02718 0.05456 0.49819 0.30918 Family size 0.06281 0.00779 8.06703 0 Two-stage Heckman selection model estimations--formal market Constant 4.15523 0.55819 7.44413 0 Education 0.22921 0.03320 6.90336 0 Age 0.06746 0.02143 3.14754 0.00084 Age2 0.00064 0.00025 2.59636 0.00476 Experience 0.01057 0.01798 0.58781 0.27837 Sex of head 0.26829 0.08243 3.25484 0.00058 Family size 0.00591 0.04418 0.13380 0.44679 1 0.90843 0.25598 3.54883 0.00020 Two-stage Heckman selection model estimations--informal market Constant 3.25639 0.48463 6.71934 0 Education 0.12500 0.03129 3.99533 0.00003 Age 0.05280 0.01901 2.77727 0.00275 Age2 0.00055 0.00055 2.52059 0.00588 Experience 0.01826 0.01756 1.03933 0.14935 Sex of head 0.11637 0.09675 1.20278 0.11456 Family size 0.04344 0.03247 1.33789 0.10650 2 1.65604 0.25121 6.59213 0 DISTRIBUTIONAL EFFECTS OF TRADE REFORM 207 Annex 6B Table 6B.1 Education Code Definition Education code Level of education 1 Elementary undergraduate 2 Elementary graduate 3 One to three years of high school 4 High school graduate 5 College undergraduate 6 At least college graduate 0 Not reported or no grade Annex 6C Table 6C.1 Notations Variable Description pi Market price in sector i wk Wage rate by type of labor k ki Capital of sector i Pi Profit of sector i Specific productivity endowment by household h hi Yi Supply of sector i yh Income of household h Chi Consumption of good j by household h Ldi Labor demand by sector i Lh,k Supply of labor k by household h Household h specific characteristics h Household h endowment of capital of sector i hi Notes The authors thank D. Boccanfuso, J. Davies, and R. Medhora for very helpful advice and comments, as well as participants at the World Institute for Devel- opment Economics Research of the United Nations University (UNU-WIDER) conference in Helsinki, March 2003, for their valuable comments. 1. It is known from the literature on aggregation that properties for aggregating labor supply are more demanding than those for aggregating consumption functions--see Deaton and Muellbauer (1980) and Muell- bauer (1981). Note that introducing nonlinearity in labor supply through participation conditions is enough to make aggregation impossible. 2. The FP algorithm could be written in the space of goods and labor rather than in the space of prices. 208 BOURGUIGNON AND SAVARD 3. This is the methodology proposed by Rutherford, Tarr, and Shepotylo (2005), but at every step of the algorithm rather than only at the first step. 4. The National Wage and Productivity Commission of the Department of Labour and Employment of the Philippines establishes a grid for mini- mum wages in formal sector activities. The wage grid is extremely complex because it is region specific, and each region has a multidimensional table specifying the size of the firms, sector of activity, and location characteristics, among other factors. Complete tables can be found on the commission's Web page at http://www.nwpc.dole.gov.ph/. 5. Details of the procedure and a discussion of results can be found in Savard (2006, chapter 3). The simplification with respect to Magnac (1991) consisted of running a univariate probit on participation at the first step, with the identifying assumption that participation depended on the difference between an average of formal and informal earnings and the reservation wage. Then ordinary least squares were run on the earnings within the two segments of the labor market with the standard Heckman correction for selec- tivity. Finally, arbitrary assumptions were made about the standard deviation of the cost of entry and the unobserved term in the reservation wage. 6. See note 5. 7. These conditions seem to make possible the perfect aggregation of individual demands. Yet full aggregation does not hold because of specific individual savings and tax rates. The authors could have chosen to use Frisch parameters and LES parameters differentiated across households could have been chosen. This choice was raising some calibration difficul- ties, however. Moreover, it seemed desirable for households to have the same nondiscretionary spending, so that poverty analysis could be based on that minimum basket of goods. The LES parameters in this application are drawn from Pollak and Wales (1969) after establishing some correspon- dence rule for the definition of goods. 8. In fact, the saving rate is based on disposable income minus the nondis- cretionary income--that is, the cost of the minimum basket of consumption goods--assuming that the heterogeneous savings and tax rates are enough to rule out perfect aggregation of household consumption behavior. 9. The information on the type of work performed by the worker is detailed with decomposition into 200 types of work categories. Given the rich set of information, it was not too difficult to classify the workers as for- mal and informal workers. 10. The complete equation listing can be provided upon a request to the authors. 11. This approach is similar to what is proposed by Agénor, Izquierdo, and Fofack (2003) and Boccanfuso, Cabral, and Savard (2005). 12. For a discussion about the advantages and inconveniences of the two approaches, see Ravallion (1998). DISTRIBUTIONAL EFFECTS OF TRADE REFORM 209 z( FGT z) 13. It may be easily proven that FGT FGT 1 . 1 FGT , where z is the poverty line and 0. The initial poverty indexes suggest that is approximately the same for 1 and 2. References Agénor, Pierre-Richard, and Karim El Aynaoui. 2005. "Politiques du marché du travail et chômage au Maroc: une analyse quantitative." Revue d'économie du développement 19 (1): 5­51. Agénor, Pierre-Richard, Alejandro Izquierdo, and Hippolythe Fofack. 2003. "IMMPA: A Quantitative Macroeconomic Framework for the Analysis of Poverty Reduction Strategies." World Bank, Washington, DC. Armington, Paul. 1969. "A Theory of Demand for Products Distinguished by Place of Production." IMF Staff Paper No. 16, 159­76, International Monetary Fund, Washington, DC. Boccanfuso, Dorothee, Francois Joseph Cabral, and Luc Savard. 2005. "Une Analyse d'Impacts de la Libéralisation de la Filière Arachide au Sénégal : une Application EGC Multi-ménages." [An Impact Analysis of the Groundnut Sector Liberalization in Senegal. An Application of a Multi-Household CGE Model]/Perspective Afrique 1 (1): 32­58. Bourguignon, François, Anne-Sophie Robilliard, and Sherman Robinson. 2005. "Representative versus Real Households in the Macroeconomic Modelling of Inequality." In Frontiers in Applied General Equilibrium Modelling, eds. T. J. Kehoe, T. N. Srinivasan, and J. Whalley. Cambridge: Cambridge University Press. Bourguignon, François, and Amedeo Spadaro. 2006. "Microsimulation as a Tool for Evaluating Redistribution Policies." Journal of Economic Inequality 4 (1): 77­106. Chen, Shaohua, and Martin Ravallion. 2004. "Welfare Impacts of China's Accession to the World Trade Organization." The World Bank Eco- nomic Review 18 (1): 29­57. Coady, David, and Rebecca Harris. 2001. "A Regional General Equilib- rium Analysis of the Welfare Impact of Cash Transfers: An Analysis of Progresa in Mexico." TMD Discussion Paper No. 76, International Food Policy Research Institute, Washington, DC. Cockburn, John. 2006. "Trade Liberalisation and Poverty in Nepal: Com- putable General Equilibrium Micro Simulation Analysis." In Globaliza- tion and Poverty. Channels and Policy Responses, eds. M. Bussolo and J. Round. London and New York: Routledge. 210 BOURGUIGNON AND SAVARD Cogneau, Denis, and Anne-Sophie Robilliard. 2001. "Growth, Distribu- tion, and Poverty in Madagascar: Learning from a Microsimulation Model in a General Equilibrium Framework." DIAL Working Paper 2001/19, and IFPRI-TMD Discussion Paper 61, Développement Institu- tions et Analyses de Long terme, Paris; and International Food and Research Policy Institute, Washington, DC. Deaton, Angus, and John Muellbauer. 1980. Economic and Consumer Behaviour. Cambridge: Cambridge University Press. Decaluwé, Bernard, Jean-Christophe Dumont, and Luc Savard. 1999. "How to Measure Poverty and Inequality in General Equilibrium Frame- work." Cahier de recherche, Centre de Recherche en Économie et Finance Appliquées (CREFA), Université Laval, no. 99-20, Québec. Decaluwé, Bernard, André Martens, and Luc Savard. 2001. La politique économique du développement. Montréal: Presse de l'Université de Montréal. Dervis, Kemal, Jaime de Melo, and Sherman Robinson. 1982. General Equilibrium Models for Development Policy. London: Cambridge University Press. Devarajan, Shantayanan, Hafex Ghanem, and Karen Thierfelder. 1999. "Labor Market Regulations, Trade Liberalization and the Distribution of Income in Bangladesh." Journal of Policy Reform 3 (1): 1­28. Fortin, Bernard, Nicolas Marceau, and Luc Savard. 1997. "Taxation, Wage Controls and the Informal Sector." Journal of Public Economics 66 (2): 293­312. Foster, James, Joel Greer, and Erik Thorbecke. 1984. "A Class of Decom- posable Poverty Measures." Econometrica 52 (3): 761­66. Grandmont, Jean-Michel. 1987. "Distributions of Preferences and the Law of Demand." Econometrica 55: 155­61. Heckman, James, and Gilherme Sedlacek. 1985. "Heterogeneity, Aggrega- tion and Market Wage Functions: An Empirical Model of Self-selection in the Labor Market." Journal of Political Economy 93: 1077­125. Hertel, Thomas, and Jeffrey Reimer. 2004. "Predicting the Poverty Impacts of Trade Reform." Policy Research Working Paper No. 3444. World Bank, Washington, DC. King, Damien, and Sudhanshu Handa. 2003. "The Welfare Effects of Bal- ance of Payments Reforms: A Macro-Micro Simulation of the Cost of Rent-Seeking." The Journal of Development Studies 39 (3): 101­28. Magnac, Thierry. 1991. "Segmented or Competitive Labor Markets?" Econometrica 59 (1): 165­87. Muellbauer, John. 1981. "Linear Aggregation in Neoclassical Labour Supply." Review of Economic Studies XLVIII: 21­36. Pollak, Robert A., and Terrence J. Wales. 1969. "Estimation of the Linear Expenditure System." Econometrica 37 (4): 611­28. DISTRIBUTIONAL EFFECTS OF TRADE REFORM 211 Ravallion, Martin. 1998. "Poverty Line in Theory and in Practice." Living Standards Measurement Study Research Paper No. 133. World Bank, Washington, DC. Riveros, Luis A. 1993. "Equity Impact and the Effectiveness of Adjustment Policies with a Segmented Labor Market: The Case of the Philippines." Journal of Economic Development 18 (2): 81­105. Roy, A. D. 1951. "Some Thoughts on the Distribution of Earnings." Oxford Economic Papers 3 (2): 135­46. Rutherford, Thomas, David Tarr, and Oleksandr Shepotylo. 2005. "Poverty Effects of Russia's WTO Accession: Modeling `Real' Household and Endogenous Productivity Effects." World Bank Policy Research Work- ing Paper No. 3473. World Bank, Washington, DC. Savard, Luc. 2006. "Analyse de pauvreté et distribution des revenus dans la cadre de la modélisation en équilibre général calculable." PhD thesis Paris School of Economics--École des Hautes Études en Sciences Sociales. Savard, Luc, and Epiphane Adjovi. 1998. "Externalités de la santé et de l'éducation et bien-être: Un MEGC appliqué au Bénin." L'Actualité Économique/Revue d'Analyse Économique 74 (3): 523­60. Thomas, Mark, and Luc Vallée. 1996. "Labor Market Segmentation in Cameroonian Manufacturing." Journal of Development Studies 32 (6): 876­98. Vos, Rob, and Niek De Jong. 2003. "Trade Liberalization and Poverty in Ecuador: A CGE Macro-Microsimulation Analysis." Economic System Analysis 15 (2): 211­32. 7 Simulating Targeted Policies with Macro Impacts: Poverty Alleviation Policies in Madagascar Denis Cogneau and Anne-Sophie Robilliard The modeling technique presented in this chapter integrates a static micro simulation module of labor supply or income and consump- tion demand, which is based on cross-sectional survey data with a static (computable general equilibrium [CGE]­type) macro mod- ule. This simulation model is designed to study the short- to medium-term impact of policies that select individuals within groups and have economywide implications. The model is applied to the case of large targeted poverty alleviation policies in Mada- gascar. The model builds on a structural microeconometric model of occupational choices and labor income that is estimated on a standard cross-sectional microeconomic data set derived from a "multitopic household survey" (see Scott 2003). The motivation for building and using this kind of tool is discussed in the first sec- tion. This discussion is followed by the micro simulation module and its econometric estimation and presentation of the integration of the macro and micro modules. The chapter concludes with sim- ulations and results. 213 214 COGNEAU AND ROBILLIARD A Structural Microeconometric Model of Income Generation This model is a member of the family of applied macro-micro tools that attempt to account for within-group heterogeneity when simu- lating the counterfactual distributive effect of a given policy or economic shock. In contrast with other approaches of the same fam- ily, this tool places greater weight on the microeconometric side of the model; as a consequence, its macroeconomic and multimarket framework is less sophisticated. The tool employs a structural microeconometric modeling of occupational choices and labor income formation. Advances in microeconometrics allow the consideration of complex production, labor supply, and consumption behaviors of heterogeneous house- holds and individuals confronted with transaction costs, information asymmetries, and employment rationing--that is, various kinds of "market imperfections." Cogneau and Robilliard (2007) consider the nonrecursive behavior of Malagasy agricultural households in the absence of a market for agricultural labor, which prevents the equalization of the productivity of agricultural labor between house- holds. Structural microeconometric estimation may also explicitly consider the market structure that constrains the decisions of various agents. For example, Cogneau (1999, 2001) estimates a labor income and occupational choice model for Madagascar's capital city of Antananarivo under various assumptions on the segmentation (dualism) of the urban labor market. A synthesis of that earlier work follows. A Simplified Macro Module Augmentation The structural nature of the microeconomic module paves the way for a consistent connection to a macro module: agents react to prices and other signals that are determined at the macro level. Because even simple microeconomic models do not lead to perfect aggrega- tion, outcomes from micro decisions must be summed up and mea- sured against each other and against other macro aggregates. To achieve a consistent macro-micro equilibrium, some macro vari- ables (such as prices) vary--until all aggregates arising from the micro components (such as the supply of categories of labor, the consumption demand, or total wage earnings) are equal to the cor- responding macro aggregates (such as the demand for categories of labor, the domestic supply of consumption goods, and the wage bill). The macro module includes the determination of these latter macro SIMULATING TARGETED POLICIES WITH MACRO IMPACTS 215 aggregates and the specification of macro closures for each macro- accounting identity. The module built here is a simple three-market CGE model. Study of Targeted Policies with Macro Impacts Because of identification and algorithm limitations, structural microeconometric modeling usually precludes a high level of disag- gregation of market segments or sectors. As a result, this approach is less suited for either the study of subtle intersectoral reallocations of supply and demand or fine modifications of the price and earnings schedule.1 Simulating short-term targeted policies with macro impacts might be the true comparative advantage of this type of model. This notion is explored in this chapter. In this context, "targeted policies" refers to policies that aim to reach specific categories of the population, most often among the poor, through various targeting devices. These devices include not only labor market interventions like wage poli- cies and workfare programs or job creation linked to foreign direct investment, but also land reforms and product market interventions like marketing boards. The first problem is to evaluate the efficiency of the targeting device. When the targeting is imperfect and depends on self-selection of individuals, a structural microeconometric model may be most useful. For instance, this model can be used to determine how many people will choose the new wage offer from a workfare program or from an export processing zone. Another problem is to assess the overall distributional impact of such policies within and outside the target population, particularly when its magnitude is big enough to have a macroeconomic impact. Under such circumstances, it may be helpful to apply a general equilibrium model with a clearly defined macro closure. For instance, the integrated macro-micro modeling framework described in this chapter can be used to answer the following questions: · How many people will benefit from an increase in the mini- mum wage, and how will this increase be transmitted to other segments in the labor market through a raise in the informal labor earnings? · What are the respective impacts of a job-creation policy and of a wage policy in a developing country urban labor market?2 · How much will a food price subsidy that is operated through a marketing board benefit small farmers, and how much will it benefit the urban poor through a relative food price reduction? 216 COGNEAU AND ROBILLIARD · How much of the workforce will a workfare program attract, and what will be the consequences on the production and prices of other sectors and, hence, on the overall income distribution? The Microeconometric Model of Income Generation This section first presents a canonical version of the model and then discusses the basic identification and micro calibration issues. These are followed by some extensions. The Labor Income Model The labor income model presented here follows Roy's model (1951), as formalized by Heckman and Sedlacek (1985), and is character- ized by Neal and Rosen (1998) as the most convincing model to explain labor income distribution. In this model, each "individual" pertains to a given family or household whose composition and location are exogenously deter- mined. Working-age individuals (those 15 years and older) have three types of work opportunities: family work, self-employed work, and wage work. Family work includes all kinds of activities super- vised by the household head or the spouse, such as family help in agricultural activities, as well as domestic work, nonmarket labor, and various forms of declared "inactivity." Self-employed work cor- responds to informal independent market activities. Wage work includes all other kinds of work performed by (mainly) civil ser- vants and large-firm workers. To self-employed work (J 1) and wage work (J 2), first assign two potential earnings functions. Individual potential earnings, wji, are the product of a task price, (J 1, 2), and of a fixed idio- j syncratic amount of efficient labor that depends on observable char- acteristics, Xi (education, labor market experience, and geographic dummies), as well as unobservable skills, tji: (7.1) ln w1 ln i 1 Xi 1 t1i (7.2) ln w2 ln i 2 Xi 2 t2 . i Returns to characteristics are differentiated by sector and by gender. j To family work, associate an unobserved individual value that depends on both individual and household characteristics: (7.3) ln w0 ~ i (X0 , Z0 ) i h 0 t0 , i where w0 may be seen as a reservation wage. Vector X0 contains the ~ i same variables as Xi (education, labor market experience, and SIMULATING TARGETED POLICIES WITH MACRO IMPACTS 217 geographic dummies) plus a variable indicating the father's occupa- tion. Vector Z0 includes the demographic structure of the house- h hold and the household's nonlabor income. Given these elements, the choice of an occupation J can be expressed as follows: (7.4) J k iff wki max(w0 , w1 , w2 ) ~ for k 0, 1, 2 . i i i This simple form of the Roy occupational model assumes that the labor market is not imperfect or segmented; in other words, there is no job rationing.3 In the presence of segmentation, the selection condition in equation (7.4) does not hold in many cases. Some indi- viduals would prefer to work in a given segment but cannot find an available job. Without any loss of generality,4 one may introduce one segmentation variable defined as the relative cost of entry between wage work and self-employment: (7.5) ln w2 ~ ln ~ ~ . i 2 X2i 2 ~t 2i Finally, comparing the respective values attributed to the three labor opportunities, workers allocate their labor according to their individ- ual comparative advantage. The selection condition in equation (7.4) is replaced by the following: w2 i is observed in family work iff w0 ~ i i w1 and w0 ~ i i w2 ~ i w2 (7.6) i is observed in self-employment iff w1 i i w0 and w1 ~ i i w2 ~ i w2 ~ w2 i is observed in wage work iff i w0 and i w1 . w2 ~ i w2 ~ i i i Econometric Identification and Micro Calibration The segmented model contains the simpler "competitive" Roy model as a particular constrained case (Magnac 1991). For econometric identification, one must assume independence of the residuals (t0, t1, t2, t2~) between individuals--as well as joint nor- ~ mality for the (t0, t1, t2, t2) vector: (7.7) (t0, t1, t2, t2) N(0, ) . ~ Under these assumptions, the occupational choice and labor income model represented by the expressions in equations (7.1) (7.3) and the series of selection conditions in equation (7.6) may then be esti- mated by maximum likelihood; one obtains a bivariate tobit, as in Magnac (1991). The coefficients of self-employment benefits and 218 COGNEAU AND ROBILLIARD wages are exactly identified, but only some parameters of the family work value (or reservation wage) and of the relative cost of entry are identified, as shown later. Likewise, only some elements of the underlying covariance structure between unobservables can be identified. Moreover, observed earnings are measured with errors or include a transient component (j 1, 2) that must be taken into account. These j unobservable components of earnings do not enter into labor sup- ply decisions of (risk-neutral) individuals. One may then assume for estimation: (7.8) (t0, t1, t2, t2, ~ , 1 2) N(0, *). may be identified: Under these assumptions, eight variance~t2or correlationvar( parameters corr (t1 t0, t2 t0), ), j tj j var(t1 ~t 0) k , var(t2 ~t 2 ~t 0) 1 corr (t1 1 , t1 t0), 2 corr (t2 1, 2. While 2, t2 ~t2 t0), and j corr (tj j , t2 ~t2 t1) for j all the parameters of potential earnings in self-employment and wage ~ work are identified, only the contrasts 1 0 and 2 2 0 are identified. (t1 t0) (t1 ^t2 t0) For purposes of simulation, one needs to recover the parameters ~ and ~ ~ 0 2 for w0 and w2, respectively, and the whole covariance structure, *. Therefore, proceed to a micro calibration, assuming that measurement errors and transient components are white noises (uncorrelated with others). One might then "guesstimate" three kinds of parameters: (1) the variance of measurement errors, (2) the correlation ( 12 ) between t1 and t2, and (3) the standard error of (t2 ~ t2 t1). A linear system of equations is then solved, with the econometrically estimated parameters and the guessti- mated parameters as givens and remaining structural parameters as unknowns. Check that the resulting matrix * is semi-definite positive. Then draw for each individual a whole set of unobservables (t0, t1, t2, t2, ~ , ) within the multidimensional normal distribution with 1 2 the covariance structure * and constrain the draws to respect the occupational selection conditions in equation (7.6). For instance, for an individual who is observed in the informal sector, start from the observed t1 and draw all other unobservable components 1 conditionally on it, constraining the draws to respect w1 w0 and w1 ~ w2. One finally obtains the set (w0, w1, w2, w2) ~ ~ w2 ~ for each individual at base "prices" ( , , ~ ) (1, 1, 1).5 1 2 2 SIMULATING TARGETED POLICIES WITH MACRO IMPACTS 219 An Extension for Nonhead Household Members Here assume a hierarchical decision-making process within the household. The household head makes his or her decision first, with- out taking into account the choices of other household members; the household head's spouse then makes his or her decision; and finally, the other working-age, secondary members make their deci- sions. The latter decisions are simultaneous. In making their choices, the other nonhead members do not consider the consequences of the decision on other household members. Accordingly, in the case of nonhead members, a variable indicat- ing the link to the household head (spouse/child/other) is added to the Z0 vector. In the case of spouses, Z0 also includes the head's occupational choices and earnings. In the case of nonspouse sec- ondary household members, it includes both the head's and the spouse's occupational choice and earnings.6 Farm Income and Reservation Wage in Farm Households Many household members are typically involved in farm activities.7 To farming households, associate a reduced farm profit function derived from a Cobb-Douglas technology with homogeneous labor: (7.9) ln 0h ln p0 ln Lh Zh u0 . h The variable Lh stands for the number of members working on the farm, while the vector Zh includes the amount of land and capital, the household head's education and age, a dummy variable in the case of a female head, and geographic dummies. Assume that the farm head always works on the farm (at least on a part-time basis; see the part-time extension in annex 7A). As a result, only nonhead members may choose whether or not to par- ticipate in farm work. Moreover, w0 is assumed to depend on the ~ individual's contribution to farm profits. Estimate this contribution as , holding fixed the decisions of other household members 0h and the farm global factor productivity u0: (7.10) ln 0h ln p0 ln(Lh i Lh ) i Zh u0 , h where Lh 1 if i is actually working on the 1 Lh and Lh 1 Lh farm in h, and Lh 1 Lh 1 and Lh 1 Lh, alternatively. Here again, the labor decision model is hierarchical between the head of the household and nonhead members, and simultaneous among nonhead members. Then write the value of family work as follows: (7.11) ln w0 ~ ln i (X0 , Z0 ) i h 0 0i t0 , i 220 COGNEAU AND ROBILLIARD where stands for the (nonunitary) elasticity of the value of fam- ily work in agricultural households to the price of agricultural products. For estimation, assume that u0, the idiosyncratic total factor productivity of the household, is independent from (t0, t1, t2, t2) ~ for all household members.8 This allows one to follow a limited information approach. In a first step, estimate the reduced profit function (7.10) and then derive an estimate for the individual potential contribution to farm production (7.10); in a second step, estimate the reservation wage equation (7.11), including this latter variable and retaining the wage functions estimated for nonagri- cultural households.9 Again, make separate estimations for each gender, excluding the farm heads whose occupational choice is not modeled. Results of Estimation and of Micro Calibration The microeconometric model is estimated on a household sample provided by the Enquête Permanente auprès des Ménages (EPM) survey for 1993­94, with approximately 4,500 households and 12,800 individuals ages 15 years and older. The part-time extension presented in annex 7A is estimated. Annex 7B gives the results of the micro calibration procedure for all the coefficients of the four basic variables of the structural microeconomic model: w0 , w1 , w2 , ~ i i i and w2 .10 Here the authors comment only on the results that are of ~ i importance for the subsequent simulations. In the farm profit function estimates (not shown), the number of family workers comes out with a coefficient that is consistent with usual orders of magnitude: a doubling of the workforce leads to an increase of about 20 percent in agricultural profits. The amounts of arable land and of capital also come out with a decreasing marginal productivity and a similar impact on profits. Age and education of the farm head also come out with a positive and significant coefficient. The returns to education are rather close in the self-employment and wage sectors. Returns to labor market experience (or to job tenure) are higher in the wage sector. Self-employment benefits are 25 percent lower in the rural areas. Costs of entry in the wage sec- tor vary negatively with education and experience and, not surpris- ingly, are 20 to 25 percent higher in the rural areas. The reservation wage in nonfarm households is positively related to education, the effect of which lies in between the returns to educa- tion in the informal sector and the "discounted" returns (monetary returns less cost of entry) in the wage sector. This wage is lower in both the rural area and the Antananarivo faritany (region), which SIMULATING TARGETED POLICIES WITH MACRO IMPACTS 221 translates into higher participation rates in those areas. Almost by definition, household heads are less often inactive and nonlabor income increases the propensity to stay inactive. The demographic structure of the household and the hierarchical decisions of other members play only a minor role in the decision to participate in the labor market. In farm households, educated people prefer to work outside the farm, whether in self-employment or wage jobs (lower family-work value). When the farm head already works part time in nonagricul- tural activities, other household members have a higher propensity to do the same. Activity is more diversified out of agriculture in the Antananarivo faritany. The estimate of the effect of the marginal productivity of labor has a negative effect on the farm-work value. This effect could stem from the fact that resource-endowed agricul- tural households, with more land or more capital and hence a higher labor productivity, are more prone (or have more opportunities) to diversify their activities. It should, however, be stressed that this diversification of activity is not frequent among agricultural house- holds. Only 13 percent of the total agricultural households' labor force works outside the farm at least part time, the bulk of which (10 percent) work in part-time informal activities. Diversification is higher for household heads (20 percent work outside of the farm) than for other members (only 10 percent work outside). This absence of real opportunities for diversification of activities among agricultural households, especially the poorest, is one of the most important features of the distribution of income in Madagascar, and it strongly constrains the short-run impact of agricultural price and workfare policies that are examined in the remainder of this chap- ter. This feature also explains why the authors could not obtain an acceptable estimate for the elasticity ( , see the previous section) of the farm-work value with respect to the agricultural price, p0. The remainder of this chapter fixes this elasticity to one, as in other sec- tors (see annex 7B). Before turning to the features of the macro-micro integration, it is worth pointing out that this structural microeconomic framework has welfare implications that are only partially taken into account in this chapter. As far as occupational choices are concerned, agents are indeed assumed to derive utility not only from monetary income (whether it comes from labor or other sources), but also from job- specific attributes and from leisure. Nonmonetary arguments of utility are ignored in this analysis, which focuses on the distribution of household real income, that is, the sum of 0, w1, w2, and other nonlabor income deflated by a household-specific cost of living index (see annex 7C). These arguments are indeed reflected in the 222 COGNEAU AND ROBILLIARD two variables w0 and w2, where w0 stands for the utility of leisure or ~ ~ ~ family work (including the relative cost of entry in self-employment)-- recalling that in the case of farm households, it includes the profit from farm activities--and w2 stands for the relative disutility of working in w2 ~ the wage sector. A "full-income" concept would sum up w0, w1, and ~ ~ over w2 individuals within each household. In a first step, however, the authors prefer to use the microeconomic model as a tool only for generating counterfactual income distributions, even at the expense of theoretical consistency from the standpoint of welfare. This choice is led by the fact that w0 and w2 are purely unobserved vari- ~ ~ ables that at the end of this micro calibration procedure, also come out with a high variance (see annex 7C). This variance is why the authors felt that the reliability of full-income counterfactuals was still to be explored, and thus left it for further research. Macro-Micro Integration Once micro calibration has been achieved, the segmented occupa- tional choice and labor income model is ready for simulation. If the size of the economic shocks or policies under study is small enough, there is no need to consider macro-level interactions. The micro- econometric model can be simulated alone, under the assumption that the variation of goods prices and of factor returns is negligible.11 Conversely, if the size of the shocks or policies under study is large enough, macro-micro links must be considered. The database for the macro module comes from a social account- ing matrix (SAM) built for the year 1995 (Razafindrakoto and Roubaud 1997). To achieve consistency between micro and macro data, household statistical weights of the 1993­94 EPM were recom- puted to comply with the income structure of the 1995 SAM. The reweighting procedure relies on a cross-entropy estimation (Robilliard and Robinson 2003). Figure 7.1 presents the global structure of the macro-micro inte- gration. Equations in the micro module describe the behavior of indi- viduals and households in terms of their labor supply and consump- tion demand. At the micro level, all income sources, stemming from individual occupational choices and household-level endowments in capital, are added up in a household income generation equation. Household expenditure is computed as the disposable income after taxes and savings have been subtracted. Consumption demands for the different goods are then derived based on household-specific budget shares (see annex 7C). These household-level consumption SIMULATING TARGETED POLICIES WITH MACRO IMPACTS 223 Figure 7.1 Fully Integrated Macro-Micro Model Structure Macro Module (CGE-type model) Equations - Macro closures * Savings--investment balance * Government budget balance * Current account balance - Factor demand - Goods supply Output - Macro aggregates - Production and prices Upward-link variables: Downward-link - Labor supply variables: Iterations - Prices - Demand for goods - Wages Micro Module (household survey) Equations - Structural wage and labor supply model - Consumption demand system - Income generation equation Output - Full income distribution Source: Authors' creation. Note: CGE computable general equilibrium. demands are added up and confronted to goods supply. Relevant prices are adjusted by a tâtonnement process so that market equilib- rium is achieved. The same applies to labor market equilibrium, with labor supply defined as the sum of the individual occupational choices and wages defined as the adjustment variable. More specifically, the three task prices ( , , ~ 1 2 2) and the agricultural price p0, introduced above, are the variables that link the micro module to the macro mod- ule. The variables p0 and are endogenously determined on the 1 goods market equilibrium for agricultural and informal goods, respec- tively. The variable is exogenous and may be used to simulate a 2 uniform wage increase in the wage sector, and ~ varies endogenously 2 to match labor supply with labor demand in the wage sector. 224 COGNEAU AND ROBILLIARD Only three sectors are considered in this model. The agricultural sector produces a tradable good and is a family-based sector, with total production equal to the sum of household-level productions.12 The informal sector produces a nontradable good and is an individual- based sector, with total production equal to the sum of individual- level value added augmented by intermediate consumption. Finally, the formal sector produces a tradable good, and total domestic formal production is fixed. Both agricultural and formal goods are imperfect substitutes for exports, the formal good is a perfect sub- stitute for imports, and the agricultural good is an imperfect substi- tute for imports. Following common specifications for this class of models, imperfect substitution is captured through constant elasticity of transformation (CET) functions at the production level and through constant elasticity of substitution (CES) functions at the consumption level.13 At the macro level, closure rules for three constraints need to be specified for the model to be "closed." They are the (current) gov- ernment balance, the savings-investment balance, and the external balance (the current account of the balance of payments, which includes the trade balance). These three constraints may be expressed as follows: (7.12) GSAV GINC pqf QG (7.13) FSAV mps Yh GSAV pqcQINVc EXR h c (7.14) pwmcQMc pwecQEc FSAV , c c where GSAV is government savings; GINC is government income; QG is government consumption; FSAV is foreign savings; mps is marginal propensity to save; Yh is household h net income; QINVc is investment demand for good c; EXR is the exchange rate; QMc and QEc are, respectively, import and export quantities of good c; pqc is the consumption price of good c; and pwmc and pwec are, respectively, import and export world prices of good c. Assume that both government and foreign savings are flexible, and that government consumption, the exchange rate, and total investment are fixed.14 By these closure rules, assume that any large poverty reduction policy, such as those simulated later, will actually be financed by an increase in foreign savings (or, equiva- lently, by a reduction in current debt service). Whether this assumption is sustainable in the long term remains an open question. This choice of closure was mainly led by the desire to compare the direct and general equilibrium impact of policies without clouding this impact with those stemming from various SIMULATING TARGETED POLICIES WITH MACRO IMPACTS 225 government revenue-increasing mechanisms, such as flexible direct or indirect tax rates. Scenarios and Simulations This chapter explores three simulations with the objective of improving the situation of the poor: a direct subsidy on agricultural prices, a workfare program, and an untargeted transfer program.15 These policies are compared in terms of both macroeconomic impact and their impact on poverty and income distribution. All experiments are designed so that their ex post costs are equal (in real terms). Description of the Scenarios The first simulation looks at the impact of a direct subsidy on agri- cultural production prices. The subsidy is set at 10 percent and is introduced as a negative tax on producer prices, thus creating a 10 percent gap between producer and consumer prices. Such a policy could be achieved by the intervention of a marketing board on agricultural goods markets, which would buy at high prices (from producers) and sell at 10 percent less (to consumers). The second experiment simulates the implementation of a work- fare scheme. Workfare programs, whereby participants must work to obtain benefits, have been used widely to fight poverty, usually in times of crises caused by macroeconomic or agroclimatic shocks (Ravallion 1999). The workfare scheme studied is assumed to be highly labor intensive. The government buys at a fixed rate the ser- vices of labor to build or rehabilitate roads and other infrastructures. Given the occupational choice model described in the previous sec- tion, the workfare scheme designed in this experiment can be summarized by two characteristics: the workfare wage level and the corresponding workload. A part-time workfare scheme was designed whereby participating individuals are allowed to continue working (in part) in their original occupation. Whether individuals choose to participate in the workfare program depends on the level of the workfare wage and on their formal, informal, and reservation wages (see the selection rule presented in annex 7A). As discussed, the level of the workfare wage is fixed ex ante so that the ex post cost of the scheme matches the cost of the agricultural price subsidy. The resulting yearly wage is 257,625 Malagasy francs (FMG), which translates into FMG 515,250 in full-time equivalent. Table 7.1 shows official minimum wages in different sectors from 1990 to 1996. This 226 COGNEAU AND ROBILLIARD Table 7.1 Minimum Yearly Wages, 1990­96 (1993 Malagasy francs) Sector 1990 1991 1992 1993 1994 1995 1996 Agriculture 576,015 614,821 543,323 494,400 554,188 603,866 557,053 Nonagriculture 566,458 604,270 533,960 485,880 537,589 592,923 547,576 Public 774,965 811,706 716,329 651,828 665,334 719,102 653,844 Source: Ministry of Finance and Ministry of Civil Service and Labor, Antana- narivo, Madagascar, www.mefb.gov.mg. database has been scaled to match structural and demographic fea- tures of the year 1995. Consequently, the meaningful figures are in the 1995 column (shown in bold in table 7.1). They show that this simulation workfare wage is relatively close to official minimum wages and represents 87 percent of the minimum wage in nonagri- cultural sectors. Given this workfare wage level, a total of 908,470 workers--corresponding to 12.7 percent of the labor force--choose to participate in the workfare scheme. The third and last simulation is a uniform, untargeted, per capita transfer program. Again, the amount paid is computed so that the aggregate ex post cost of the program matches the cost of the previous programs. The resulting amount is FMG 17,887 per capita, which will be added to household nonlabor income (and has the corresponding microeconomic effects of an increase in the value of inactivity in nonfarm households). All three programs share a high budgetary cost equivalent of almost 5 percent of gross domestic product (GDP). They should therefore have large macroeconomic impacts as well as the intended distributional micro impacts. Targeting Issues A central issue related to the poverty and income distribution impacts of all three simulations is the targeting properties of each scheme. Obviously, the uniform, untargeted transfer per capita is distributed evenly across quintiles of income, but this is not the case for the agri- cultural subsidy and workfare simulations. To explore this issue, table 7.2 presents the distribution of individuals in beneficiary house- holds across quintiles of per capita income for these two simulations. Not surprisingly, the agricultural subsidy appears to have good targeting properties in terms of the distribution of beneficiary house- holds. But this result does not hold when one considers the distrib- ution of the program cost: while 83.9 percent of individuals in the first quintile are in a household that benefits from the agricultural Table 7.2 Distribution of Beneficiary Households across Quintiles Agricultural subsidy Workfare scheme Share of Share of Beneficiary Row program cost Beneficiary Row program Quintile households (percent) (percent) households (percent) cost (percent) 1st 2,184,281 83.9 7.2 821,244 31.5 16.4 2nd 2,073,064 79.5 12.6 842,288 32.3 17.1 3rd 2,073,970 79.4 18.5 919,525 35.2 23.9 4th 1,752,415 67.2 24.0 829,048 31.8 22.0 5th 1,036,771 39.7 37.7 715,063 27.4 20.6 Total or average 9,120,501 69.9 100.0 4,127,168 31.6 100.0 Source: Authors' estimations. Note: Quintiles are computed using per capita income in the base year. Row percentage figures are shares of beneficiary population by quintile. 227 228 COGNEAU AND ROBILLIARD subsidy, only 7.2 percent of the total program cost accrues to these households, and the largest share of the cost (37.7 percent) accrues to the last quintile. This result is related to the fact that the price subsidy is proportional to agricultural output and thus, by con- struction, is regressive in terms of program cost allocation. When compared with the agricultural subsidy, the workfare scheme appears to be less progressive in terms of the distribution of individuals in beneficiary households, because they are distributed evenly across quintiles. Because the benefits accruing to households are not proportional to their incomes, however, the distribution of the program cost is actually less regressive than in the agricultural subsidy experiment. The targeting performance of the workfare scheme is nevertheless disappointing as it fails to reach a large number of workers in poor households. This is explained by the fact that the reservation value (w0) estimated and calibrated from actual ~ data not only reflects preferences for family work but also includes a cost-of-entry component in outside informal activities. Estimated parameters indicate, for instance, that activity is more diversified out of agriculture in households living in the Antananarivo faritany or in urban areas, as well as more diversified in land-rich house- holds. As a result, individuals from poor agricultural households dwelling in remote areas are given large reservation values, which reflect large costs of access to all markets, including the labor mar- ket. This cost of access prevents some agricultural workers from seizing the workfare job opportunities. In other words, because the workfare scheme fails to take these costs into account, it is implic- itly targeted toward urban areas. As a result, it has a large impact on urban poverty (see the following section, "Simulation Results"). Simulation Results Table 7.3 shows various price and macro aggregate changes as a result of the three programs. Macro aggregate changes are presented in real terms. One common point across all three experiments is the increase in the relative price of the agricultural goods. In particular, even the subsidy simulation leads to a 4.7 percent increase in the price of agri- cultural goods for consumers (relative to the consumer price index). This result stems from large income effects that raise the demand for agricultural products. The workfare program has the strongest impact of all on the agricultural prices (8.2 percent increase against 4.7 and 5.6 percent in the other simulations), because it also leads to a decrease in the labor available for agriculture (see table 7.4). Results also show that the macroeconomic impact of all three policies is small and positive in terms of GDP.16 As mentioned earlier, all experiments were designed to equalize their ex post cost. Because SIMULATING TARGETED POLICIES WITH MACRO IMPACTS 229 Table 7.3 Macroeconomic Impact of Alternative Policies Untargeted Part-time uniform Agricultural workfare per capita Indicator BASE subsidy program transfer Agricultural price 1.0 4.7 8.2 5.6 Informal price 1.0 1.8 3.5 1.7 Formal price 1.0 3.0 5.3 3.5 Consumption price index 1.0 0 0 0 GDP at market prices 4,713.5 1.0 1.6 1.0 Absorption 4,975.2 4.4 5.0 4.4 Private consumption 4,274.5 5.2 5.9 5.2 Investment 467.2 0.7 1.2 0.7 Government consumption 233.6 0 0 0 Exports 1,144.3 0.5 1.3 0.2 Imports 1,406.0 12.0 13.2 12.0 GDP at factor cost 4,424.0 0.3 1.0 0.3 Agricultural value added 1,429.1 0.4 1.1 0.3 Informal value added 413.6 1.2 1.0 1.7 Formal value added 2,581.2 0.2 2.2 0.1 Cost (FMG, billion) 227.5 228.4 226.8 Cost (percentage of base GDP) 4.8 4.8 4.8 Source: Authors' estimations. Note: FMG Malagasy francs; GDP gross domestic product. Base values are reported in the first column, and percentage changes are reported in the following columns. Cost figures are ex post. program costs are entirely distributed to the households, all three simulations have the same impact on private consumption. The employment impact is presented in table 7.4. The top part of the table shows the number of workers by occupational choice, while the lower part presents aggregate values of the sectoral allocation of labor. Results show that the subsidy simulation leads to a mild increase in total employment. In terms of sectoral employment, labor appears to be reallocated from the informal ( 5.9 percent) to the agricultural sector ( 1.5 percent). As expected, the workfare scheme has a strong impact on urban underemployment, with the number of inactive workers decreasing by almost 18 percent. It also leads to important reallocations of labor out of the agricultural ( 3.9 per- cent) and informal sectors ( 12.6 percent) into workfare. As a result, the total active population increases by 3.3 percent. Given its design, the workfare program obviously drives transitions out of full-time work and into part-time work. The uniform transfer scheme has a milder impact on the structure of employment. 230 COGNEAU AND ROBILLIARD Table 7.4 Employment Impact of Alternative Policies Untargeted Part-time uniform Agricultural workfare per capita Indicator BASE subsidy program transfer Full-time agricultural workers 4,248.9 2.8 5.3 1.6 Full-time informal workers 324.6 4.4 42.2 3.2 Full-time formal workers 527.0 0.4 2.9 0.4 Part-time workers 874.9 12.7 67.9 7.0 Full-time inactive workers 1,144.3 2.1 17.7 1.5 Agricultural labor 4,536.3 1.5 3.9 0.8 Informal labor 687.0 5.9 12.6 2.9 Formal labor (including workfare) 602.1 0.2 76.1 0.3 Total active workers 5,825.4 0.5 3.3 0.3 Total labor force 7,119.7 0 0 0 Source: Authors' estimations. Note: Base values are reported in the first column, and percentage changes are reported in the following columns. Part-time workers category includes either part- time formal or informal work with inactivity, part-time formal or informal work with agricultural activity, as well as part-time inactivity, agricultural activity, formal or informal work with workfare in the case of the workfare scheme simulation. Total active workers and sectoral labor are in full-time equivalent, with full-time workers counting for 1.0 and part-time workers counting for 0.5. Table 7.5 shows results in terms of poverty and income distribu- tion for all households, in both urban and rural areas. Changes in three indicators of inequality are presented: the Gini index and two entropy indexes. All indicators show that the agricultural price subsidy simulation leads to an improvement in the distribution of income at the national level. A closer look into each area suggests that the decrease in over- all inequality is driven both by the convergence in urban and rural per capita incomes and by the decrease in inequality in the urban area. The introduction of a subsidy on agricultural production leaves the inequality within the rural area almost unchanged (the Gini index slightly increases by 0.3 percent), while inequality in the urban area only slightly decreases. As mentioned earlier, the small increase in rural inequality stems from the targeting property of the subsidy, whereby agricultural households with higher incomes benefit more (in absolute terms) than do smaller agricultural households. As a result, changes in poverty indicators are mainly driven by changes in per capita income. SIMULATING TARGETED POLICIES WITH MACRO IMPACTS 231 Table 7.5 Social Impact of Alternative Policies, General Equilibrium Results Untargeted Part-time uniform Agricultural workfare per capita All households BASE subsidy program transfer Per capita income 352.7 4.4 4.6 5.0 General entropy index 0 45.2 2.5 7.6 11.2 General entropy index 1 59.0 3.0 6.8 8.2 Gini index 51.1 1.3 3.6 4.8 Poverty incidence 59.0 5.0 6.6 6.2 Poverty gap 24.9 8.2 13.5 16.3 Poverty severity 13.4 10.0 17.4 24.2 Urban households Per capita income 631.1 0.8 3.1 2.7 General entropy index 0 48.1 1.1 7.2 6.2 General entropy index 1 62.8 0.9 5.5 4.5 Gini index 52.7 0.5 3.3 2.6 Poverty incidence 30.5 1.0 11.1 7.0 Poverty gap 10.3 3.2 24.0 19.7 Poverty severity 4.5 5.1 29.5 28.2 Rural households Per capita income 260.7 7.3 5.7 6.8 General entropy index 0 33.2 0.8 8.4 14.3 General entropy index 1 39.7 0.6 8.2 11.3 Gini index 44.0 0.3 4.1 6.4 Poverty incidence 68.4 5.6 5.9 6.1 Poverty gap 29.7 8.8 12.3 16.0 Poverty severity 16.4 10.4 16.3 23.8 Source: Authors' estimations. Note: Base values are reported for the first column, and percentage changes are reported in the following columns. In terms of poverty reduction, the workfare scheme has a stronger impact than the subsidy program: the poverty headcount is reduced by 6.6 percent, while the subsidy program reduces it by 5 percent. Workfare also has a stronger effect on income distribution, with a 3.6 percent decrease in the Gini index (compared with a 1.3 percent decrease with the subsidy program) and a 17.4 percent decrease in the poverty severity indicator (compared with a 10 percent decrease with the subsidy program). This strong decrease in inequality is explained both by the convergence of average per capita incomes between urban and rural areas and by the decrease of inequality within both areas. The workfare scheme has by far the strongest impact on inequality and poverty in urban areas. Thanks to the workfare scheme, poverty incidence in urban areas decreases by 232 COGNEAU AND ROBILLIARD more than 11 percent, whereas it decreases only slightly in the case of the agricultural subsidy and is reduced by 7 percent with the uni- form transfer. Although the GDP impact of the untargeted transfer program is mild, both the poverty and income distribution impacts are significant: the program reduces the poverty headcount by 6.2 percent and the Gini index by 4.8 percent, and its impact on poverty severity is the highest among the three experiments. These results again show that the workfare scheme does not achieve much better targeting than the untargeted transfer program and does not satisfactorily reach the poorest of the poor. In sum, the two targeted programs that have been examined here indeed have large impacts on monetary poverty alleviation, even once general equilibrium effects are taken into account. Given the large budgetary amounts that are transferred to households, this does not come as a surprise. Apart from scaling and financing issues, however, the simulations reveal that there is room to improve the quality of targeting. Indeed, a general subsidy to agricultural pro- ducers does not appear to be an adequate scheme for reaching the poorest farmers, because it fails to do better than a uniform per capita transfer or even a workfare scheme--even in rural areas. A general workfare program offering part-time job opportunities paid at about the minimum wage reaches somewhat disappointing results, especially in rural areas. Costs of access to the labor market prevent individuals living in remote areas or in poor autarkic agri- cultural households from seizing the workfare opportunities. The workfare scheme performance is relatively good in urban areas, where it draws a lot of people out of inactivity or out of informal underemployment, but it falls short in rural areas, where it is out- performed by the untargeted transfer. All three schemes have been designed to have the same ex post budgetary cost in terms of the total amount of transfer received by households. They all, however, have specific implementation costs that should be taken into account when comparing their relative effi- ciency. For instance, the implementation of an agricultural subsidy would call for the reconstruction of a marketing board, which raises many institutional issues and might imply high administrative costs. Likewise, the implementation of a workfare scheme has more costs than pure wage costs, no matter how labor intensive it is: organiza- tional and administrative costs, advertisement costs, and input costs (see Ravallion 1999). In this case, however, some of these additional costs are internalized by individuals who give up the workfare job offers when these are located too far from their household. Finally, even the untargeted transfer scheme would entail an additional cost of bringing the cash to the households, even in remote areas. SIMULATING TARGETED POLICIES WITH MACRO IMPACTS 233 Comparing Micro Accounting Ex Ante and Ex Post Results This section turns to a more methodological question and compares the simulation results of three specifications of the model. The first version corresponds to the results of a micro accounting exercise in which neither behavior nor general equilibrium effects would be taken into account. The second version still does not account for general equilibrium effects but allows individuals and households to respond to the shock. The final version accounts for both micro behaviors and general equilibrium effects. It corresponds to the version used above for the analysis of poverty reduction policies. Two types of shocks are examined: the 10 percent agricultural price subsidy analyzed previously and a 20 percent total factor produc- tivity shock in the agricultural sector. The results of both simula- tions are presented in figures 7.2 and 7.3. These figures show the Lorenz curve (built on income per capita) together with the con- centration curves of the benefits of the two shocks under the three specifications of the model. In figure 7.2, the micro accounting and ex ante curves track closely. Both indicate that the incidence of the subsidy program is progressive. The ex post curve does not reverse that conclusion Figure 7.2 Benefit Incidence of an Agricultural Subsidy under Various Specifications 1.0 capita per 0.8 0.6 income of 0.4 share 0.2 cumulative 0 0 0.2 0.4 0.6 0.8 1.0 cumulative share of population Lorenz curve micro accounting ex ante ex post Source: Authors' estimations. 234 COGNEAU AND ROBILLIARD Figure 7.3 Benefit Incidence of a Total Factor Productivity Shock in the Agricultural Sector 1.0 capita per 0.8 0.6 income of 0.4 share 0.2 cumulative 0 0 0.2 0.4 0.6 0.8 1.0 cumulative share of population Lorenz curve ex ante ex post Source: Authors' estimations. but appears closer to the 45-degree line, indicating that the program is more progressive than ex ante simulations would predict. This is reversed in the second simulation (figure 7.3), where results indicate that taking into account general equi- librium actually leads to the conclusion that the shock is less progressive than micro accounting or ex ante simulations would predict.17 In the case of Madagascar and of shocks that affect the relative price of the agricultural good, general equilibrium effects will mainly change the distribution of the benefits between rural and urban households. Given the big difference in mean incomes between urban and rural households, it does not come as a surprise than any shock that leads to an increase in the relative price of the agricultural good will "redistribute" the benefits of the program toward rural households, thus making it more progressive ex post than ex ante. Symmetrically, any shock that leads to a decrease in the relative price of the agricultural good (such as a productivity shock), will "redistribute" the benefits of the program toward urban households, thus making it more progressive ex post than ex ante. The two experiments presented here show that it is not possible to reach a conclusion on a systematic bias in terms of poverty or inequality changes when ignoring general equilibrium effects. SIMULATING TARGETED POLICIES WITH MACRO IMPACTS 235 Conclusion This chapter has presented the basic motivations for the construc- tion of an integrated static macro-micro model for a low-income economy. It has outlined the main features of such a model in terms of microeconometric specifications and macro closures. Finally, it has explored the use of this kind of model for the simulation of tar- geted transfer schemes dedicated to poverty alleviation. These types of transfer schemes might be implemented either following a macro- economic shock or as permanent safety nets. For purposes of illus- tration, three large-scale transfer schemes have been simulated and compared: (1) a price subsidy to agricultural producers, (2) a gen- eral workfare program proposing part-time job opportunities paid at about the minimum wage, and (3) a uniform unconditional and untargeted transfer provided to each individual regardless of their age and job situation. The macro-micro model yields interesting results on the counterfactual impacts of each program on the over- all distribution of income, by taking into account both microeco- nomic targeting issues and macroeconomic general equilibrium effects. Considerations about the financing of the programs and about their technical implementation costs could supplement the simulations to build realistic, efficient, and sustainable poverty alle- viation schemes. To conclude, it may be useful to review briefly the comparative advantages and disadvantages of the integrated macro-micro approach. The authors first argued that the approach is well suited to incorporating current advances in the microeconometrics of household behaviors and market structures in developing countries. The illustrations presented show the usefulness of a thorough mod- eling of labor supply behavior in the context of highly segmented markets. However, much remains to be done to improve the model- ing of behavior in agricultural households where collective produc- tion in family farms does not fit this "individualistic" framework as well. (For an alternative, see Cogneau and Robilliard 2007.) More- over, it should be emphasized that structural estimation based on cross-sectional data may either overstate or understate the true reac- tion of poor households with respect to labor incentives. This type of estimation would benefit from the availability of dynamic panel data or from experimental knowledge on the response of poor households to various programs (Duflo 2004). Second, the authors argued that integrated tools might be desired for the sake of macro-micro consistency, as far as "aggregation issues" and "interlinked welfare issues" are concerned. It should, however, be stressed that such consistency in the modeling of household welfare 236 COGNEAU AND ROBILLIARD (labor supply, earnings, consumption) is obtained at the expense of sectoral disaggregation and dynamic considerations. Depending on the policy problem at stake, trade-offs must be solved inside a trian- gle made of "household heterogeneity," "sectoral detail," and "intertemporal issues." The authors therefore argued that the static integrated tool might be better suited for analyzing the distributional aspects of general development strategies, on the one hand, and for evaluating the impact of short- to medium-term targeted programs with macro impacts, on the other. Through the applications implemented in this chapter, the authors hope to have shown that integrated macro-micro modeling could be useful in the design of these latter programs. The design of other structural policies, such as minimum-wage increases or foreign- investment-led jobs creation, could also benefit from this type of approach. Annex 7A: A Part-Time Extension To account for individuals who wish to pursue outside part-time activities when they also work for the family, one must introduce a "part-time" variable in the wages and benefits equations that accounts for the variability of hours worked: ln w1 ln i 1 X1 i 1 1T1i t1i ln w2 ln i 2 X2 i 2 2 T2 i t2 , i with 0, and 1 2 0 and where T1 (and T2 ) is a dummy variable i i indicating whether the individual works part-time.18 One may then redefine full-time incomes as follows: (A7.1) ln w1i ln w1 i 1T1i (A7.2) ln w2i ln w2 i 2 T2 . i Finally, assume that when reservation value is close enough to either full-time wage or self-employment benefits, individuals choose to work (simultaneously or successively) inside and outside the family. The listing of selection rules then becomes i chooses full-time family work iff ~ w2 w0 (1 and w0 ~ (1 i i a)w1 i i a) ~w2i SIMULATING TARGETED POLICIES WITH MACRO IMPACTS 237 i chooses family work and self-employment iff w2 (1 a)w1 w0 ~ (1 and i i i a)w1i w1i w2 ~ i i chooses family work and wage work iff w2i w2 w2 (1 a) ~ w0 ~ i i w2 (1 and i a) ~ i w2 i w2 ~ w1 i i i chooses full-time self-employment iff w2 (1 a)w1 w0 ~ and i i i w1 i w2 ~ i i chooses full-time wage work iff w2i w2 (1 a) ~ w0 ~ i w2 and i i w2 ~ w1 . i i For econometric estimation, the likelihood of the model is rewritten according to this new selection rule. The workfare program that is subsequently simulated introduces a new kind of part-time job offer that is paid at a rate w3. In this case, once the former selection rule has been run, add the following rules: if i had chosen full-time family work, i takes the workfare offer iff 2w3(1 a) w0 ~ i if i had chosen self-employment, i takes the workfare offer iff w3 w1 [1 exp( )] i 1 if i had chosen wage work, i takes the job offer iff w3 w2 [1 exp( )] . i 2 The two last conditions apply whether i chooses a full-time or part- time option. In the case of a part-time choice, and if the relevant condition holds, this worker relinquishes part-time family work in favor of workfare. In any case, this worker ends up working part- time in self-employment or wage work, and the balance of time in the workfare program. If the first condition holds, the work is then part time in the family, with the remaining time spent in the work- fare program. 238 COGNEAU AND ROBILLIARD Annex 7B: Estimation and Micro Calibration Table 7B.1 Results from Estimation and Micro Calibration ~ Variable 1 2 2 0 Men Nonfarm households Number of years of education (/10) 0.9841 0.8995 1.3384 0.9429 Number of years of experience (/10) 0.3232 0.5100 0.7785 0.1533 Number of years of experience squared (/1,000) 0.3879 0.6276 1.1335 0.0315 Region of Antananarivo ( 1) 0.0561 0.0689 0.1520 0.2380 Rural area ( 1) 0.2504 0.0891 0.4053 0.0039 Father in the informal sector ( 1) 0 0 0.1901 0.1696 Father in the formal sector ( 1) 0 0 0.0695 0.0485 Household head in the informal sector ( 1) 0 0 0.4913 0.0505 Household head in the formal sector ( 1) 0 0 0.4124 0.2920 Spouse in the informal sector ( 1) 0 0 0.2107 0.1436 Spouse in the formal sector ( 1) 0 0 0.7386 0.0473 Number of children ages 0 to 9 years old 0 0 0 0.0641 Number of males ages 10 to 14 years old 0 0 0 0.0160 Number of males ages 15 to 69 years old 0 0 0 0.0261 Number of females ages 10 to 14 years old 0 0 0 0.0150 Number of females ages 15 to 69 years old 0 0 0 0.0134 Number of adults ages 70 years and older 0 0 0 0.1188 Household head ( 1) 0 0 0 0.7030 Spouse of the head ( 1) 0 0 0 0.5665 Child of the head ( 1) 0 0 0 0.1395 Nonlabor income 0 0 0 0.6465 Household head wage income 0 0 0 0.7335 Spouse wage income 0 0 0 0.2203 Part-time correction 0.80 0.43 0 0 Constant 3.8386 3.3470 1.6192 4.8406 Part-time thresholda 0.28 Standard errors (diagonal) and correlation of unobservables t1 t2 ~t2 t0 t1 0.9740 0.5000* 0.5020 0.8580 t~2 0.5940 0.0180 0.5110 t2 1.8330 0.6220 t0 1.4780 SIMULATING TARGETED POLICIES WITH MACRO IMPACTS 239 ~ Variable 1 2 2 0 Men Farm households Number of years of education (/10) 0.9841 0.8995 1.8892 0.8106 Number of years of experience (/10) 0.3232 0.5100 1.4856 0.1862 Number of years of experience squared (/1,000) 0.3879 0.6276 3.8351 0.6089 Region of Antananarivo ( 1) 0.0561 0.0689 0.5764 0.8237 Rural area ( 1) 0.2504 0.0891 0.8868 0.3592 Household head in the informal sector ( 1) 0 0 6.6620 0.6282 Household head in the formal sector ( 1) 0 0 1.5937 0.0513 Number of children ages 0 to 9 years old 0 0 0 0.0611 Number of males ages 10 to 14 years old 0 0 0 0.0738 Number of males ages 15 to 69 years old 0 0 0 0.0598 Number of females ages 10 to 14 years old 0 0 0 0.0824 Number of females ages 15 to 69 years old 0 0 0 0.0252 Number of adults ages 70 years and older 0 0 0 0.1822 Spouse of the head ( 1) 0 0 0 0.1442 Child of the head ( 1) 0 0 0 0.1180 Nonlabor income 0 0 0 0.3589 Marginal productivity of agricultural labor 0 0 0 1* Part-time correction 0.81 0.44 0 0 Constant 3.8386 3.3470 4.1720 6.1760 Part-time thresholda 0.45 Standard errors (diagonal) and correlation of unobservables t1 t2 ~t2 t0 t1 0.9740 0.3000* 0.0280 0.8930 t~2 0.5940 0.1040 0.5940 t2 2.0450 0.4120 t0 1.7070 (Continued on next page) 240 COGNEAU AND ROBILLIARD Table 7B.1 (Continued) ~ Variable 1 2 2 0 Women Nonfarm households Number of years of education (/10) 1.0697 1.3439 1.6047 0.8535 Number of years of experience (/10) 0.2387 0.5243 0.4425 0.1333 Number of years of experience squared (/1,000) 0.2571 0.6476 0.7708 0.3569 Region of Antananarivo ( 1) 0.2530 0.0541 0.2790 0.3937 Rural area ( 1) 0.2494 0.0135 0.3500 0.1241 Father in the informal sector ( 1) 0 0 0.2720 0.1344 Father in the formal sector ( 1) 0 0 0.2139 0.2915 Household head in the informal sector ( 1) 0 0 0.3239 0.1247 Household head in the formal sector ( 1) 0 0 0.1145 0.2320 Spouse in the informal sector ( 1) 0 0 0.3565 0.2135 Spouse in the formal sector ( 1) 0 0 0.4257 0.1466 Number of children ages 0 to 9 years old 0 0 0 0.0038 Number of males ages 10 to 14 years old 0 0 0 0.0675 Number of males ages 15 to 69 years old 0 0 0 0.0046 Number of females ages 10 to 14 years old 0 0 0 0.0187 Number of females ages 15 to 69 years old 0 0 0 0.0124 Number of adults ages 70 years and older 0 0 0 0.1094 Household head ( 1) 0 0 0 0.3917 Spouse of the head ( 1) 0 0 0 0.0583 Child of the head ( 1) 0 0 0 0.2056 Nonlabor income 0 0 0 1.1474 Household head wage income 0 0 0 0.0574 Spouse wage income 0 0 0 0.7528 Part-time correction 0.83 0.20 0 0 Constant 3.5774 2.4824 2.0197 4.7003 Part-time thresholda 0.29 Standard errors (diagonal) and correlation of unobservables t1 t2 ~t2 t0 t1 0.9750 0.5000* 0.4960 0.8760 0.5590 0.0890 0.3300 ~ t2 t2 1.8810 0.6530 t0 1.4520 SIMULATING TARGETED POLICIES WITH MACRO IMPACTS 241 ~ Variable 1 2 2 0 Women Farm households Number of years of education (/10) 1.0697 1.3439 2.3181 0.7551 Number of years of experience (/10) 0.2387 0.5243 0.5943 0.2368 Number of years of experience squared (/1,000) 0.2571 0.6476 0.4778 0.5286 Region of Antananarivo ( 1) 0.2530 0.0541 0.1836 0.5821 Rural area ( 1) 0.2494 0.0135 0.5646 0.3172 Household head in the informal sector ( 1) 0 0 0.1660 0.6827 Household head in the formal sector ( 1) 0 0 0.8654 0.1080 Number of children ages 0 to 9 years old 0 0 0 0.0074 Number of males ages 10 to 14 years old 0 0 0 0.1417 Number of males ages 15 to 69 years old 0 0 0 0.0070 Number of females ages 10 to 14 years old 0 0 0 0.0670 Number of females ages 15 to 69 years old 0 0 0 0.0311 Number of adults ages 70 years and older 0 0 0 0.1026 Spouse of the head ( 1) 0 0 0 0.0994 Child of the head ( 1) 0 0 0 0.1417 Nonlabor income 0 0 0 0.2221 Marginal productivity of agricultural labor 0 0 0 1* Part-time correction 0.84 0.21 0 0 Constant 3.5774 2.4824 3.9111 6.3980 Part-time thresholda 0.55 Standard errors (diagonal) and correlation of unobservables t1 t2 ~t2 t0 t1 0.9750 0.3000* 0.0900 0.8340 t~2 0.5590 0.5100 0.0950 t2 1.9590 0.4470 t0 1.7470 Sources: Enguête Permanente auprès des Ménages 1993 survey and authors' calculations. Note: Coefficients in roman type (first two columns) are econometrically estimated. In contrast, the three coefficients with an asterisk (*) are pure guesses. Other guessed coefficients not shown in the table include the two measurement errors variances (which are assumed null) and the (t2 ~ t2 t1) standard error (at 2 and 1.5 in nonfarm and farm households, respectively). Coefficients in italics (last two columns) result from a "micro calibration" using both econometric estimates and guessed coefficients. See section titled "Econometric Iden- tification and Micro Calibration" for more details. a. For the definition of part-time corrections and threshold, see annex 7A. 242 COGNEAU AND ROBILLIARD Annex 7C: A Simple Expenditure System with Heterogeneous Preferences The macro-micro model tries to make use of the wealth of data available--not only for labor supply and income generation but also for consumption. However, data limitations prevent going too far in that direction. To avoid microeconometric complications, sav- ings and consumption choices are first assumed as separable from labor supply decisions. Second, saving rates derived from the data come out as unreliable; therefore, a fixed saving rate common to all households (and equal to 0.052 in the application) is assumed: (C7.1) Ch (1 s)Yh . Household disposable income Yh is equal to the sum of agricultural benefits (including autoconsumption of goods produced by the household), self-employment benefits and wage earnings, nonlabor income stemming from capital income, and transfers. Ch stands for household h total consumption expenditures. Third, total consumption is then split between the three compos- ite goods of the model (agricultural, informal, and formal) through idiosyncratic budget shares derived from the data: j,h (C7.2) Cj 0, 1, 2 and 1 . ,h j,h Ch with j j,h j 0,1,2 This specification corresponds to the simplest Cobb-Douglas homo- thetic utility function for consumption. Notes The authors thank the National Institute of Statistics of Madagascar for providing the data presented in this chapter. Special thanks also go to Mireille Razafindrakoto, Francois Roubaud, and members of the MADIO project in Antananarivo for helpful discussions about their research and the Malagasy economy. The authors are grateful to Francois Bourguignon, Jesko Hentschel, Phillippe Leite, Dominique van der Mensbrugghe, Luiz Pereira da Silva, and Abdelkhalek Touhami for discussions about earlier versions of this work. 1. When supplemented with a dynamic demographic module, this approach can be relatively well suited to exploring demo-economic issues like the distributive impact of the HIV/AIDS epidemics (Cogneau and Grimm forthcoming) or general poverty reduction strategies like the long- term impact of education policies (Grimm 2004, 2005). SIMULATING TARGETED POLICIES WITH MACRO IMPACTS 243 2. Cogneau (1999, 2001) shows that a macro-micro model of the dis- tribution of income can simulate the historical decrease in poverty observed in the city of Antananarivo during the 1995­99 period, thanks to job cre- ation and minimum-wage increases in the formal sector. 3. Moreover, this simple form assumes that individuals compare self- employment and wage-work opportunities only in terms of earnings; in other words, they do not bring differential nonpecuniary benefits. See Cogneau (2001). 4. The reservation value w0 includes the cost of entry into the informal ~ activities. 5. In econometric estimation, the X vectors include a constant. 6. For estimation, the authors still assume independence for (t1, t2, ~t2, t0) between individuals, even among members of the same household. 7. It also might be the case for some nonagricultural occupations. In light of the Malagasy case and data, however, the authors choose to treat nonagricultural self-employment as a purely individual occupation. These data suggest that the great majority of self-employed workers in nonagri- cultural sectors are running very small, most often individual, businesses. 8. This latter assumption should allow for a direct identification of the effect in w0, through the effect of u0 . However, as ~ is presumably 0 h 0 affected by large measurement errors, the authors exclude "available land" from the variables in w0, taking it as an instrument for the identification of the effect of . 0 9. This latter option is rather innocuous for potential earnings outside the farm, as only a small number of individuals declare out-of-farm earn- ings in agricultural households. 10. More detailed econometric results are available from the authors upon request. 11. Even with small policies, this assumption of no price variation may be violated if there is a strong spatial segmentation of markets. In this lat- ter case, local price variations may matter. 12. Production functions parameters are estimated (see section "Econo- metric Identification and Micro Calibration") and technical coefficients are taken from the survey. Although all technical coefficients are scaled up so that the sum of intermediate consumption equals national accounts aggre- gate, they remain household specific for the agricultural production. 13. For the calibration of the agricultural CET function, the share of exports on total production is idiosyncratic and taken from survey data. 14. By Walras's law, one of the system constraints is redundant. System constraints include markets as well as macro balances. In this model, the redundant equation is the external balance equation (7.13). 15. Previously, the authors showed that neither a devaluation of 20 percent nor a fourfold increase in agricultural tariffs could achieve a sig- nificant reduction in poverty and inequality indicators (Cogneau, Grimm, and Robilliard 2003). 244 COGNEAU AND ROBILLIARD 16. The GDP aggregate does not include the value of goods, services, or infrastructure produced by the workfare program. 17. Under the current version of this algorithm, the authors are not able to distinguish micro accounting from ex ante results in the case of the pro- ductivity shock because the shock amounts to changing a technical para- meter that does not affect household behaviors in the first round. 18. The authors thank François Bourguignon for a fruitful discussion about this extension. References Cogneau, Denis. 1999. "La formation du revenu des ménages à Antana- narivo: une microsimulation en équilibre général pour la fin du siècle." Economie de Madagascar 4: 131­55. . 2001. "Formation du revenu, segmentation, et discrimination sur le marché du travail d'une ville en développement: Antananarivo fin de siècle." DIAL Working Paper 2001/18, Développement Institutions et Analyses de Long terme, Paris. Cogneau, Denis, and Michael Grimm. Forthcoming. "The Impact of AIDS Mortality on the Income Distribution in Côte d'Ivoire." Journal of African Economies. Cogneau Denis, Michael Grimm, and Anne-Sophie Robilliard. 2003. "Evaluating Poverty Reduction Policies: The Contribution of Microsim- ulation Techniques." In The New International Strategies for Poverty Reduction, eds. J.-P. Cling, M. Razafindrakoto, and F. Roubaud. London: Routledge. Cogneau, Denis, and Anne-Sophie Robilliard. 2007. "Growth, Distribu- tion, and Poverty in Madagascar: Learning from a Microsimulation Model in a General Equilibrium Framework." In Microsimulation as a Tool for the Evaluation of Public Policies: Methods and Applications, ed. A. Spadero. Madrid: Fundacion BBVA. Duflo, Esther. 2004. "Scaling up and Evaluation." Annual World Bank Conference on Development Economics 2004: Accelerating Develop- ment, Vol. 1, Report No. 30228. Washington, DC: World Bank. Grimm, Michael. 2004. "A Decomposition of Inequality and Poverty Changes in the Context of Macroeconomic Adjustment: A Microsimu- lation Study for Côte d'Ivoire." In Growth, Inequality, and Poverty: Prospects for Pro-Poor Economic Development, eds. A. F. Shorrocks and R. van der Hoeven. Oxford: Oxford University Press. . 2005. "Educational Policies and Poverty Reduction in Côte d'Ivoire." Journal of Policy Modeling 27 (2): 231­47. Heckman, James, and Guilherme Sedlacek. 1985. "Heterogeneity, Aggrega- tion, and Market Wages Functions: An Empirical Model of Self-Selection in the Labor Market." Journal of Political Economy 93 (6): 1077­125. PSIMULATING TARGETED POLICIES WITH MACRO IMPACTS 245 Magnac, Thierry. 1991. "Segmented or Competitive Labor Markets?" Econometrica 59 (1): 165­87. Neal, Derek, and Sherman Rosen. 1998. "Theories of the Distribution of Labor Earnings." In Handbook of Income Distribution, eds. A. B. Atkinson and F. Bourguignon. Amsterdam: North-Holland. Ravallion, Martin. 1999. "Appraising Workfare." The World Bank Research Observer 14 (1): 31­48. Razafindrakoto, Mireille, and Francois Roubaud. 1997. "Une matrice de Comptabilité Sociale pour Madagascar." Projet Madio. INSTAT-DIAL- IRD, Study 9744/E, Antananarivo, Madagascar. Robilliard, Anne-Sophie, and Sherman Robinson. 2003. "Reconciling Household Surveys and National Accounts Data Using Cross-Entropy Estimation." Review of Income and Wealth 49 (3): 395­406. Roy, A. 1951. "Some Thoughts on the Distribution of Earnings." Oxford Economic Papers 3 (June): 135­46. Scott, Kinnon. 2003. "Generating Relevant Household-Level Data: Multi- topic Household Surveys." In The Impact of Economic Policies on Poverty and Income Distribution: Evaluation Techniques and Tools, eds. F. Bourguignon and L. A. Pereira da Silva. Washington, DC: World Bank and Oxford University Press. 8 Wealth-Constrained Occupational Choice and the Impact of Financial Reforms on the Distribution of Income and Macro Growth Xavier Giné and Robert M. Townsend How should the financial system of a country be evaluated? How should policy makers determine the appropriate policy response to observed inequality or lack of economic growth? The answers to these questions--put forth by finance ministers, financial sector specialists, academics, and other practitioners and policy experts-- generally follow one of two approaches. The first takes the view that the world economy is optimal and that nothing needs to be done because government intervention could lead to a less optimal situation. The second approach is proactive and supports the notion that governments should promote the regulation or liberalization of markets. The appropriate economic policy advice depends, of course, on the context; but this chapter suggests that policy questions like these are best analyzed with the help of an algorithm that com- bines theory and data. This chapter provides an example of how an algorithm of this type works, based on earlier published work in Giné and Townsend (2004). In this example, economic theory is used to refine the logic that can be applied to a few key observed 247 248 GINÉ AND TOWNSEND facts, and microeconomic data are then used again--to validate the model. In practice, then, one iterates from theory to data and back again. Model-based advice can be viewed as just another opinion to consider in conjunction with other policy advice; however, model- based opinions emanate from a specified set of assumptions and rules that must be consistent with certain scientific norms within the economics profession. That is, a model requires reduced-form or behavioral equations that often are based on rational or quasi- rational economic behavior. But a model also requires these equa- tions to be consistent with one another, as in explicit definitions of constrained optimality or concepts of equilibrium. More to the point, a model has implications for cross-sectional relationships or the evolution of economic variables over time, and so it can be validated or refuted when confronted with data. Thus, this chapter proposes that policy recommendations be gen- erated based on somewhat realistic, estimated versions of the reality of a given economy. The logic of the model is made explicit, so this model has the advantage of guiding policy choices because researchers can trace a particular policy recommendation to a given set of assump- tions or rules. The question addressed in chapter 8, along with the proposed algorithm, is that of the potential costs and benefits of financial sec- tor reforms. A major policy concern related to general structural reforms is the idea that benefits will not trickle down. There is con- cern that the poor will be neglected and that inequality will increase. Similarly, globalization and capital inflows are often claimed to be associated with growth--although the effect of economic growth on poverty is still a much-debated topic.1 Not all possible forms of liberalization are (or could be) con- sidered in this chapter. The focus is rather on reforms that (1) increase outreach on the extensive domestic margin (less restricted licensing requirements for foreign and domestic financial institutions, for example), (2) reduce excess capitalization require- ments, and (3) enhance the ability to open new branches. These types of reforms are captured in the model, although crudely, by characterizing them as domestic reforms that allow deposit mobi- lization and access to credit at market-clearing interest rates for a segment of the population that otherwise would not have formal sector savings or credit. The theory used here to address the costs and benefits of policy reform is a relatively simple general equilibrium model with credit constraints. Specifically, the authors chose from the literature and in this chapter, extend the Lloyd-Ellis and Bernhardt (2000) model (the WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 249 LEB model), which features wealth-constrained entry into business and wealth-constrained investment for entrepreneurs. This model has several advantages for purposes of this chapter. It allows for ex ante variation in ability. It has a general (approximated) production technology that allows labor share to vary, and the household occu- pational choice has a closed-form solution. Finally, the model fea- tures a dual economy that captures several widely observed aspects of the development process. These include industrialization with persistent income differentials, a slow decline in the subsistence sec- tor, and an eventual increase in wages--all of which contribute to growth with changing inequality. The authors extend this occupational choice model that does not include intermediation to an intermediated sector that allows bor- rowing and lending at a market-clearing interest rate. The interme- diated sector is expanded exogenously at the observed rate in the data, given initial participation and the initial observed distribution of wealth. Even though endogenous financial deepening may be pre- ferred (see Greenwood and Jovanovic 1990; Townsend and Ueda 2006), exogenous financial deepening has a peculiar, distinct advan- tage in this model because it can be varied (either to mimic the data and its upturns and downturns or to keep it flat) and provides a counterfactual experiment. One can thus gauge the consequences of these various experiments and compare them. In short, general equi- librium policy analysis can be applied following the seminal work of Heckman, Lochner, and Taber (1998), despite endogenous prices and an evolving endogenous distribution of wealth in a model where preferences do not aggregate. In this chapter, the explicit structure of the model is used as given in the occupational choices and investment decisions of households to estimate certain parameters of the model using a variety of distinct microeconomic data sets. But not all parameters of the model can be estimated via maximum likelihood. The rest are calibrated to match the growth rate and observed changes in inequality, labor share, sav- ings, and the number of entrepreneurs reported in the data. This model can be applied to any economy that has experienced a financial liberalization for which the relevant data are available. The prime example in this chapter is Thailand, from 1976 to 1996.2 This is a good country to study for a number of reasons. First, Thailand is often portrayed as an example of an emerging market with high income growth and increasing inequality. The gross domestic product (GDP) growth from 1981 to 1995 was 8 percent per year, and the Gini measure of inequality increased from 0.42 in 1976 to 0.50 in 1996. Second, there is evidence that Thailand had a relatively restrictive credit system but also liberalized it during this 250 GINÉ AND TOWNSEND period (for details, see Klinhowhan 1999; Okuda and Mieno 1999). Third, Jeong (1999) shows that the increase in the number of house- holds with access to formal intermediation did contribute to growth in per capita income during this period. Finally, Thailand experi- enced a relatively large increase in capital inflows from the late 1980s to the mid-1990s. This structural, estimated version of an actual economy can then be compared to what would have happened if there had been no expansion in the size of the intermediated sector. Without financial liberalization at estimated parameter values from both data sets, the model predicts a dramatically lower growth rate, a higher residual subsistence sector, nonincreasing wages, and lower and decreasing inequality. Thus, financial liberalization appears to be the engine of growth it is sometimes claimed to be, at least in the context of Thailand. However, growth and liberalization do have uneven conse- quences, as the critics insist. The distribution of welfare gains and losses in these experiments is not uniform, because there are various effects that depend on wealth and talent. With financial liberaliza- tion, savings earn interest, although the wealthy tend to benefit most. But credit is available to facilitate occupation shifts and to finance setup costs and investment, and quantitatively, there is a striking conclusion. The primary winners from financial liberaliza- tion are either talented but low-wealth, would-be entrepreneurs who without credit cannot go into business at all or entrepreneurs with little capital. Modal gains range from 17 percent to 34 percent of the observed overall average of Thai household income. But there are losers as well. Liberalization induces an increase in wages in later years, and even though this benefits workers, other things being equal, it hurts entrepreneurs because they face a higher wage bill. The estimated welfare loss in both data sets is roughly the same order of magnitude as the observed average income of firm owners overall. This fact suggests a plausible political economy rationale for (observed) financial sector repressions. Finally, the estimated structure of the model is used to conduct a policy experiment. The economy is "opened up" to the observed foreign capital inflows. These contribute to increased growth, inequality, and the number of entrepreneurs, but they do so only slightly. Otherwise, the macroeconomic and distributional conse- quences are quite similar to those of the closed economy with liber- alization. Indeed, if the expansion is changed to grow linearly (rather than as observed in the data), the model cannot replicate the high growth rates observed in Thailand during the late 1980s and early 1990s, despite large capital inflows at that time. WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 251 The steps of the algorithm detailed in the following sections may be applied in a variety of countries. The theory is first presented in detail, followed by a description of the core theory as given in an occupational choice map (for an autarky sector and for a credit sec- tor). The maximum likelihood estimation of most major parameters of the model is then presented, followed by calibration of the remaining parameters, which matches the macro, aggregate data. Simulations are then presented at the estimated and calibrated val- ues for each of two data sets, followed by two extension applica- tions to village economies and other aspects of income distribution. The chapter closes with various measures of the welfare gains and losses associated with the financial liberalization, a discussion of international capital inflows, and conclusions. Specify the Environment: The General Equilibrium Model The LEB model begins with a standard production function, map- ping a capital input k and a labor input l at the beginning of the period into output q at the end of the period. In the original LEB model,3 and in the numerical simulations presented in this section, this function is taken to be quadratic. In particular, it takes the following form: 1 1 (8.1) q f(k, l) k k2 kl l l2. 2 2 The quadratic function in equation (8.1) can be viewed as an approximation for virtually any production function and has been used in applied work (see Griffin, Montgomery, and Rister 1987, and the references cited in that work). This function also facilitates the derivation of closed-form solutions and allows labor share to vary over time. Each firm has a beginning-of-period setup or fixed cost x, and this setup cost is drawn at random from a known cumulative distri- bution H(x, m) with 0 x 1. (8.2) H(x, m) mx2 (1 m)x, m [ 1, 1]. This distribution, shown in equation (8.2), is parameterized by the number m. If m 0, the distribution is uniform; if m 0, the distribution is skewed toward high setup cost x, and the converse arises when m 0. This setup cost is supposed to vary inversely with talent--that is, it takes both talent and an initial investment to start a business--but the higher the level of talent, the lower the 252 GINÉ AND TOWNSEND setup cost. More generally, the cumulative distribution H(x, m) is a crude way to capture and allow estimation of the distribution of tal- ent in the population and is not an unusual specification in the indus- trial organization literature (for example, Das, Roberts, and Tybout 1998; Veracierto 1998). Unobserved talent is, of course, a key to education choice and other discrete choice models.4 Cost x is expressed in the same units as wealth and thus has the same units as a utility differential. Every agent is born with an inheritance or ini- tial wealth b. Alternatively, the agent starts the period with wealth saved from the previous period. The distribution of inheritances in the population at date t is given by Gt(b) : Bt [0, 1], where Bt R is the changing support of the distribution at date t. The time argu- ment t makes explicit the evolution of Bt and Gt over time. The beginning-of-period wealth b and the cost x are the only sources of heterogeneity among the population. These variables are modeled as independent of one another in the specification used in this exam- ple and allow the existence of a unique steady state.5 All units of labor can be hired at a common wage, w, to be deter- mined in equilibrium. (There is no variation in skills for wage work, although this can be added.) There is a storage technology that car- ries goods from the beginning to the end of the period one-to-one, so the effective interest rate is zero. This assumption puts a lower bound on the gross interest rate in the corresponding economy with credit and limits the input that k firms wish to use in the production of output q, even in the economy without credit. Firms operate in cities, and the associated entrepreneurs and workers incur a com- mon cost of living measured by the parameter . Alternatively, captures a fixed cost of leaving agriculture and assumes that firms can operate in rural areas. Under this alternative, setup costs and x may vary explicitly with distance to a district center or city. The choice problem of the entrepreneur is as follows: (b, x, w) max f(k, l) wl k (8.3) k,l s. t. k [0, b x], l 0, where (b, x, w) denotes the profits of the firm with initial wealth b, without subtracting the setup cost x, given wage w. Because credit markets have not yet been introduced, capital input k cannot exceed the initial wealth b less the setup cost x--as in expression (8.3). This is the key finance constraint of the model and may or may not be binding, depending on x, b, and w. More generally, some firms may operate, but if wealth b is low relative to cost x, they may be con- strained in capital input use k--that is, for constrained firms, wealth b limits input k. Otherwise unconstrained firms are all alike and have identical incomes before netting out the cost x. One can WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 253 distinguish firms by sector, but the added heterogeneity, though more realistic, complicates the calculation. The capital input k can be zero but not negative. Even though all agents are born with an inherited nonnegative initial wealth b, not everyone need be a firm. There is also a subsis- tence sector in which agents earn . In the original LEB model, everyone is placed in this subsistence sector initially, at a degenerate steady state distribution of wealth. For various subsequent periods, labor can be hired from this subsistence sector, at subsistence plus cost of living, thus w . After everyone has left this sector (as a laborer or as an entrepreneur), the equilibrium wage will rise. The simulations impose an initial distribution of wealth as estimated in the data and allow the parameter to increase at an exogenous imposed rate of , which also increases the wage, but could be gr gr estimated to be zero. For a household with a given initial wealth-cost pair (b, x) and wage w, the choice of occupation reduces to an essentially static problem of maximizing end-of-period wealth W(b, x, w) given in equation (8.4): b if a subsistence worker, (8.4) W(b, x, w) w b if a wage earner, (b, x, w) x b if a firm. At the end of the period, all agents take this wealth as a given and decide how much to consume, C, and how much to bequeath, B, to their heirs; that is-- maxC U(C, B) ,B (8.5) s. t. C B W . In the original LEB model and in the simulations presented in this chapter, the utility function is Cobb-Douglas; that is-- (8.6) U(C, B) C1 B . This functional form yields consumption and bequest decision rules given by constant fractions 1 and of the end-of-period wealth, respectively. Again, indirect utility would be linear in wealth. Parameter denotes the bequest motive.6 The key to both static and dynamic features of the model is a par- tition of the equilibrium occupational choice in (b, x) space into three regions: unconstrained firms, constrained firms, and either workers or subsistence workers. These regions are determined by the equilibrium wage w. One can represent these regions as (b, x) combinations yielding the occupational choices of agents of the 254 GINÉ AND TOWNSEND model, using the exogenous distribution of costs H(x, m) at each period along with the endogenous and evolving distribution Gt(b) of wealth b. The population of the economy is normalized so that the fractions of constrained firms, unconstrained firms, workers, and subsistence workers add to unity. An equilibrium at any date t given the beginning-of-period wealth distribution Gt(b) is a wage wt, such that given wt, every agent with wealth-cost pair (b, x) chooses occupation and savings to maximize equations (8.4) and (8.5), respectively. The wage wt clears the labor market in the sense that the number of workers, subsistence workers, and firms adds to unity. As is made clear in the following section, existence and uniqueness are ensured. Because of the myopic nature of the bequest motive, the explicit reference to date t often can be dropped. Characterizing Household Choice: The Occupation Partition Although the model is essentially dynamic to explain growth and inequality in transition to a steady state, the occupational choice that every household faces on every date is static. In the noninter- mediated sector, this choice depends on the individual beginning-of- period wealth b, cost x, and the economywide wage w. In the inter- mediated sector, it depends only on the cost x and economywide prices w and R. The occupational choice in both sectors is now described in greater detail. Choices in the Nonintermediated Sector For an individual with beginning-of-period wealth b facing an equilibrium wage w, there are two critical skill levels: xe(b, w) and xu(b, w), as shown in figure 8.1. An individual whose cost level x is higher than xe(b, w) becomes a worker; if the level is lower, the indi- vidual becomes an entrepreneur. Finally, if x is lower than xu(b, w), the individual becomes an unconstrained entrepreneur whose cost of capital is calculated using the implicit zero interest rate of the storage technology (see figure 8.1). The exact specification can be used to derive these curves and cutoffs. For a complete set of equa- tions, see Giné and Townsend (2004). To define x as the maximum fixed cost, such that for any x x*, the agent will never be an entrepreneur. More formally, and sup- pressing the dependence of profits on the wage w, x* is such that (8.7) x* u w, where u max (k, w), k WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 255 Figure 8.1 Occupational Choice Map subsisters and workers x* xe(b,w) cost x~ setup xu(b,w) constrained entrepreneurs unconstrained entrepreneurs b^ b* wealth Source: Authors' creation. if x x*, the maximum income as an entrepreneur will always be less than w; therefore, the agent is always better off becoming a worker. Denote by b* the wealth level of an entrepreneur with cost x* so that this individual is just unconstrained. That is, b* x* ku, where ku ^ arg max (k, w). Finally, note that for b less than b, the potentially bindingk constraint is financing the setup costs x, even with k 0. Choices in the Intermediated Sector A major feature of the baseline model is the credit constraint in equa- tion (8.3) that is associated with the absence of a capital market. For example, a talented person (low fixed cost) may be unable to become an entrepreneur because that person cannot raise the necessary funds to buy capital. Likewise, some firms cannot capitalize at the level they would choose if they could borrow at the implicit interest rate of the storage technology. The most obvious variation to the baseline model is to introduce credit to a market and allow the fraction of the popu- lation in this market to increase over time. This is what is meant by a financial liberalization in the context of this chapter.7 256 GINÉ AND TOWNSEND The authors consider an economy with two sectors of a given size at date t, with one sector open to borrowing and lending. Agents born in this sector can deposit their beginning-of-period wealth in the financial intermediary and earn gross interest R on this amount. If they decide to become entrepreneurs, they can borrow at the inter- est rate R to finance their fixed cost and capital investment. The borrowing and lending rate is assumed to be the same for all of those in the financial, intermediated sector. Again, in this context, liberalization does not mean a reduction in the interest rate spread; it instead means an expansion of access on the extensive margin. Labor (unlike capital) is assumed to be mobile, so there is a unique wage rate w for the entire economy, common to both sectors. In the intermediated sector, gross profits do not depend on wealth or setup costs. Because all entrepreneurs operate the same technol- ogy and face the same factor prices w and R, they will all operate at the same scale and demand the same (unconstrained) amount of capital and labor, regardless of their setup cost or wealth. The deci- sion to become an entrepreneur, a worker, or a subsistence worker is dictated by the value of the fixed cost. Indeed, given factor prices w and R, there is a value of x~(w, R) at which an agent would be indifferent to the two options. Anyone with a setup cost greater than x~(w, R) will be a worker, and vice versa. The thick dotted line in figure 8.1 represents the threshold fixed cost x~(w, R). Figure 8.1 thus shows the overlap of the relevant occupational map in each sector. Thick solid curves represent the nonintermedi- ated sector, and a thick dotted line represents the intermediated sec- tor, thereby partitioning the (b, x) space into different regions that (as explained later) will experience a differentiated welfare impact from a financial liberalization. As in a standard two-sector neoclassical model, the factor prices R and w can be found, solving the credit and labor market-clearing conditions.8 Estimating Some Key Parameters from Cross-Section Data This section explores the idea that the occupational choice is static, to estimate some of the parameters of the model. If the initial wealth b and the wage w are observable, while x is not, then the likelihood that an individual will be an entrepreneur in the nonintermediated sector can be determined entirely as in the occupation partition dia- gram, from the curve xe(b, w) and the exogenous distribution of tal- ent H(x, m). That is, the probability that an individual household with WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 257 initial wealth b will be an entrepreneur is given by H(xe(b, w), m), with the likelihood that cost x is less than or equal to xe(b, w). The resid- ual probability 1 H(xe(b, w), m) corresponds to the likelihood that the individual household will be a wage earner. The fixed cost x enters additively into the entrepreneur's problem defined at wealth b. Thus, setup costs can be large or small relative to wealth (depending on how the 1997 Thai baht is converted into LEB model units).9 The authors therefore searched over various scaling factors s to map wealth data into the model units. In a related move, they pinned down the subsistence level in the model by using the estimated scale s to convert to LEB model units the baht earnings of those in Thai subsistence agriculture. Now let denote the vector of parameters of the model related to the production function and scaling factor, that is, ( , , , , , s). Suppose that there is a sample of n households with yi as a zero-one indicator variable for the observed entrepreneurship choice of house- hold i. Then with the notation xe(bi , w) for the point on the xe(b, w) curve for household i with wealth bi, at parameter vector with wage w, the explicit log likelihood of the entrepreneurship choice for the n households can be written as follows: 1 n Ln( , m) yi ln H[xe(bi , w), m] (8.8) n i 1 (1 yi)ln{1 H[xe(bi , w), m]} . In equation (8.8), the parameters to be searched are again the produc- tion parameters ( , , , , ), the scaling factor s, and the skewness m of H(·, m). Intuitively, however, the production parameters in vector can- not be identified from a pure cross-section of data at a given point in time. Essentially, only three parameters, not five, are deter- ^ mined. The regions represented in figure 8.1 are determined by b, b*, x*. However, one can fully identify the production parameters by exploiting the variation in the wages over time observed in the data, as two of these cutoffs move with wage.10 The derivatives of the likelihood in equation (8.8) can be deter- mined analytically, and then with the given observations of a database, standard maximization routines can be used to search numerically for the maximum.11 The standard errors of the estimated parameters can be computed by bootstrap methods. And comparable estimates can be made for other countries. Dynamic simulations are sensitive to the scale parameter s. High values increase the likelihood function but give the initial economy so much wealth that initially there is high consumption, low saving, and negligible growth. In 258 GINÉ AND TOWNSEND practice then, the scale parameter can be calibrated as part of the algorithm that follows. Likewise, numerical maximization routines are improved in accuracy if there is a priori information about x*, and so x* may be included in the following calibration exercise. (Likeli- hood estimation routines are available from the authors.) The primary database featured in these calculations is the widely used and highly regarded Socio-Economic Survey12 (SES) conducted by the National Statistical Office in Bangkok, Thailand. The sample is nationally representative and includes eight repeated cross-sections collected between 1976 and 1996. As in the complementary work of Jeong and Townsend (2003), the calculations in this chapter are restricted to relatively young households (with members ages 20­29), whose current assets might be regarded as somewhat exogenous to their recent choice of occupation. This sample is also restricted to households with no recorded transaction with a financial institution in the month before the interview (a crude measure for lack of finan- cial access), as assumed in the LEB model. However, the SES does not record direct measures of wealth. From the ownership of various household assets, the value of the house, and other rental assets, Jeong (1999) estimates a measure of wealth based on principal components analysis that essentially estimates a latent variable that can best explain the overall variation in home ownership and other household assets (for details, see Jeong 1999; Jeong and Townsend 2003). Observations for the first available years are used (1976 and 1981) to obtain full identification because the household wages var- ied over these two periods (see figure 8.2). The sample consists of a Figure 8.2 Occupational Choice Maps, Socio-Economic Survey Data and Townsend-Thai Data a. Socio-Economic Survey data b. Townsend-Thai data 1.0 1.0 Northeast 1976 0.8 0.8 1981 x x 0.6 0.6 cost cost Central 0.4 0.4 setup setup 0.2 0.2 0 0 0 0.5 1.0 1.5 2.0 2.5 3.0 0 4 8 12 16 20 inheritance b inheritance b Source: National Statistical Office of Thailand, Bangkok (http://www.nso.go.th/); authors' data. WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 259 total of 24,433 observations--with 9,028 observations from 1976 and 15,405 from 1981. The second data set is a specialized but substantial cross-sectional survey of 2,880 households conducted during May 1997 (Townsend- Thai data).13 Because the LEB model is designed to explain the behav- ior of those agents without access to credit, the sample presented here is restricted to households that reported having no relationship with any credit institution (formal or informal), which is another strength of the survey.14 A disadvantage of the second data set is that, as a single cross-section, there is no temporal variation in wages. Thus, the production parameters are identified by dividing the obser- vations into two subsamples of households in the northeast and cen- tral regions, which exploits regional variation in the wages.15 The final sample consists of a total of 1,272 households, with 707 house- holds from the northeast region and 565 households from the central region. Table 8.1 reports the estimated parameters and the standard errors. The parameter for both data sets was found by multiplying an estimate of the subsistence level from the data by the scaling factor estimated. For the SES data, the authors used the mean income of farmers in 1976, which amounted to 19,274 Thai baht. Table 8.1 Maximum Likelihood Estimation Results Socio-Economic Survey data Townsend-Thai data Standard Standard Indicator Coefficient error Coefficient error Scaling factor sa 1.4236 0.00881 1.4338 0.03978 Subsistence level 0.02744 0.00119 0.01538 0.00408 Fixed-cost distribution m 0.5933 0.05801 0.00559 0.17056 Technology 0.54561 0.06711 0.97545 0.00191 0.39064 0.09028 0.0033 0.00013 0.03384 0.00364 0.00966 0.00692 0.1021 0.02484 0.00432 0.00157 0.2582 0.03523 0.12905 0.04146 Number of observations 24,433 1,272 Log-likelihood 8,233.92 616.92 Source: Authors' computations. Note: Socio-Economic Surveys were conducted by the National Statistical Office in Bangkok, Thailand. a. The parameter value and standard error reported are multiplied by a factor of 106. 260 GINÉ AND TOWNSEND Analogously, they used the average income of workers in the north- east region without access to credit (as reported in the Townsend- Thai data), or 10,727 baht. The wages for the two time periods in the model units at the estimated scaling factor s were w76 0.048 and w81 0.053 for the SES data set, and wNE 0.016 and wC 0.037 for the two regions in the Townsend-Thai data set. The max- imized value of the likelihood function obtained using the SES data was 8,233.92, whereas the Townsend-Thai data set yielded a value of 616.92. Calibration of Primary Parameters against Dynamic Paths The cost of living and the "dynamic" parameters, namely, the sav- ings rate and the subsistence income growth rate must still be gr determined (as well as other parameters if it is not clear that these have been reliably estimated in the maximum likelihood estimation routines). One way to determine these parameters is calibration: look for the best , , and (and other combinations) according gr to some metric relating the dynamic data to be matched with the simulated data. The Data The Thai data used here are the growth of GDP, the Thai national savings rate, labor share, the fraction of entrepreneurs, and the Gini coefficient. (Giné and Townsend 2004 describe the data in more detail.) The data show an initially high net growth rate of roughly 8 per- cent in the first three years, which then fell to a more modest 4 per- cent up through 1986. The period 1986­94 displayed a relatively high and sustained average growth rate of 8.43 percent, and within that period, from 1987 to 1989, the net growth rate was 8.83 per- cent. During this same period, the Thai economy GDP growth rate was the highest in the world (at 10.3 percent). These high growth periods have attracted much attention. Labor share was relatively stable at 0.40 and rising (after 1990) to 0.45 by 1995. A trend from the 1990­95 data was used to extrapolate labor share for 1996. Savings as a percentage of national income were roughly 22 percent from the initial period to 1985. Savings then increased to 33 percent after 1986, in the higher growth period. Although typical of Asia, WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 261 these numbers are relatively high. The fraction of entrepreneurs was remarkably steady, but slightly increasing, from l4 percent to l8 per- cent. The Gini coefficient stood at 0.42 in the 1976 SES and increased more or less steadily to 0.53 in 1992. Inequality decreased slightly in both the 1994 and 1996 rounds, to 0.50. This downward trend mirrors the rise in the labor share during the same period, and both may be explained by the increase in the wage rate. This level of inequality is relatively high, especially for Asia, and rivals many countries in Latin America. Other measures of inequality (Lorenz curve, for example) display similar orders of magnitude within Thai- land over time and relative to other countries. (For a more detailed explanation, see Jeong 1999.) The fraction of population with access to credit was estimated at 6 percent in 1976 and increased to 26 percent by 1996. The data also reveal that as a measure of financial deepening, access to credit grew slowly in the beginning and more sharply later (from 1986 onward). The authors recognize that this measure of intermediation is at best limited and not what they would like to have, and it seems likely that these levels are inaccurate. Issues in the Calibration Method FINANCIAL LIBERALIZATION Begin with the standard benchmark occupational choice model, shutting down credit altogether. Then consider an alternative inter- mediated economy, with two sectors, one open to credit and saving and the other remaining nonintermediated. Only labor is mobile, hence a unique wage rate, whereas capital cannot move to the other sector. In other words, a worker residing in the nonintermediated sector may find a job in the credit sector but will not be able to deposit wealth in the financial intermediary. The relative size16 of each sector is assumed to be exogenous and changing over time (given by the fraction of people with access to credit as shown in figure 8.3). This is the key measure of liberalization. INITIAL WEALTH DISTRIBUTION The initial 1976 economywide distribution of wealth is relevant for dynamic simulations.17 As mentioned earlier, Jeong (1999) con- structs a measure of wealth from the SES data using observations on both household assets and the value of owner-occupied housing units. 262 GINÉ AND TOWNSEND Figure 8.3 Foreign Capital Inflows and Financial Liberalization 0.30 0.25 0.20 GDP of 0.15 fraction 0.10 0.05 0 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 year foreign capital inflows as a fraction of GDP fraction of population with access to intermediation Source: Authors' creation. THE METRIC Any calibration exercise requires a metric to assess how well the model matches the data. The authors used the normalized sum of the period-by-period squared deviations of the predictions of the model from the actual Thai data.18 The deviations are normalized in the five variables by dividing them by their corresponding mean from the Thai data. More formally, 5 1996 zst sim zst ec 2 (8.9) C wst , s 1 t 1976 zs where zs denotes the variable s, t denotes time, and wst is the weight given to the variable s in year t.19 Finally, sim and ec denote "simu- lated" and "Thai economy," respectively, and denotes the vari- zs able zs mean from the Thai data. WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 263 The search covers the cost of living , subsistence-level growth rate , and the bequest motive parameter using a grid of 203 gr points or combinations of parameters.20 All of the statistics except for the savings rate have natural counter- parts in the model. "Savings" is considered to be the fraction of end-of- period wealth bequeathed to the next generation. The savings rate is then computed by dividing this measure of savings by net income. The simulation routines are available from the authors. But roughly speaking, given the parameters and initial distribution of wealth--a guessed wage and interest rate--the routines determine household occupational choices and intermediated sector borrow- ing and lending choices. Then, if the labor market and credit market do not clear, the interest and wages are adjusted using a bisection algorithm. Having found the market-clearing prices, equilibrium wealth at the end of period is determined--as is wealth at the begin- ning of the subsequent period. The process begins again. Simulated and Actual Economies This section illustrates the simulation results using the calibrated and estimated parameters. Simulations with Parameters from the Townsend-Thai Data The simulation featured here is generated from the economy with no access at all to intermediation at the Townsend-Thai parameters and displays similar characteristics to the simulation that uses the SES data parameters (figures 8.4 and 8.5). Essentially, the Thai economy fails to grow except toward the end of the 1976­96 period, and then only at the rate of exogenous technological change, . Now consider gr the intermediated economy at these parameter values. If all variables are equally weighted each year, the calibrated parameters21 are 0.004, 0.267, and 0.006. The corresponding graphs gr are presented in figure 8.4. The model does well here at explaining the levels and changes in all variables. Particularly striking is the growth rate of income, which although somewhat low in levels, tracks the Thai growth experience. The model also does remarkably well in matching labor share and the Gini measure of inequality. It underpredicts the fraction of entrepreneurs, however, although it is able to replicate a positive trend. As usual, the model features a flatter savings rate-- although it does well at matching the last subperiod (1988 to 1996). Economywide growth is driven primarily by growth in the interme- diated sector, where the bulk of the economy's entrepreneurs lie in 264 GINÉ AND TOWNSEND Figure 8.4 Intermediated Model, Townsend-Thai Data, 1976­96 a. Income growth b. Labor share 14 1.0 12 0.8 10 rate 8 output 0.6 6 of 4 0.4 growth 2 share0.2 0 2 0 1978198019821984198619881990199219941996 1978198019821984198619881990199219941996 year year c. Gini coefficient d. Savings rate 1.0 0.6 0.8 0.5 0.4 0.6 income of 0.3 0.4 coefficient 0.2 0.2 fraction 0.1 0 0 1978198019821984198619881990199219941996 1978198019821984198619881990199219941996 year year e. Fraction of entrepreneurs 0.4 0.2 fraction 0 1978198019821984198619881990199219941996 year Thai economy simulation at estimated parameters mean simulation confidence intervals Source: Authors' creation. WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 265 Figure 8.5 Intermediated Model, Socio-Economic Survey Data, 1976­96 a. Income growth b. Labor share 15 1.0 10 0.8 rate 5 output 0.6 of 0 0.4 growth 5 share0.2 10 0 1978198019821984198619881990199219941996 1978198019821984198619881990199219941996 year year c. Gini coefficient d. Savings rate 1.0 0.6 0.8 0.5 0.4 0.6 income of 0.3 0.4 coefficient 0.2 0.2 fraction 0.1 0 0 1978198019821984198619881990199219941996 1978198019821984198619881990199219941996 year year e. Fraction of entrepreneurs 0.4 0.2 fraction 0 1978198019821984198619881990199219941996 year Thai economy simulation at estimated parameters mean simulation confidence intervals Source: Authors' creation. 266 GINÉ AND TOWNSEND addition to a relatively high number of workers from both the inter- mediated and nonintermediated sectors.22 Village Economies The same algorithm can be applied to more aggregated data. Felkner and Townsend (2004) have village-level data from a census administered by Thailand's Community Development Department (a ministry of the interior agency) to male village leaders biannu- ally from 1986 to 1996. Using a principal components analysis of the fraction of households in a village owning tractors, motorcy- cles, pickup trucks, and toilets, the authors estimated the initial distribution of cross-village wealth. Also, according to the village leaders, the history of credit access from formal providers is known (with regard to the Bank for Agriculture and Agricultural Cooper- atives and commercial banks). So the increasing weight of the inter- mediated sector is known for the 1986­96 period, and the model is thus simulated (at trial parameter values) to predict wages and interest rates, among others. Then, knowing the history of credit intervention in a village, predictions are made for the fraction of households in each village that should be represented as firms (those engaged exclusively in trade and handicrafts, for example). Finally, using mean square error criteria (as in the calibration sec- tion), most of the key parameters (except for the end-of-sample year 1996) are reestimated in an effort to match the observed entre- preneurial rate to the rate predicted by the data. In particular, if the parameter m of the cost distribution is allowed to vary in the obvi- ous way with distance (so that villages far from main roads are more likely to have high costs), then the predictions of the model are reasonably good, and correlations of prediction errors with observables and with clusters of urbanization are small. Inequality Decompositions and the Dynamics of Change It is interesting to explore more fully how well the LEB occupational choice model can track growth and the change in inequality. Jeong and Townsend (2003) reconfirm that the model does reasonably well with growth, by tracking movements in GDP (even though the model has no aggregate shocks), particularly the expansion of incomes in the late 1980s. The model also does quite well with a Theil measure of income inequality, and though that measure is lower than in the data, the model picks up the decrease in inequality observed during the 1990s in Thailand, when wages began to increase. But the model underestimates the actual number of entrepreneurs, overpredicts the WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 267 income differential over wage earners, exaggerates the impact of occupation shifts on increases in inequality, and underestimates the inequality within occupations. Making access to the financial sector endogenous helps to remedy some of these anomalies. Micro Impact: The Distribution of Gains A goal of this analysis is to identify a measure of the welfare impact of the observed financial sector liberalization. And because there can be general equilibrium effects in the model from this liberaliza- tion, the appropriate welfare comparison must be made clear. The analysis that follows compares the economy with the exogenously expanding intermediated sector to the corresponding economy with- out an intermediated sector, at the same parameter values. The cri- terion will be end-of-period wealth--or what households in the model seek to maximize. For a given period, then, a household will be characterized by its wealth b and beginning-of-period cost x; and it will be asked how much end-of-period wealth would increase (or decrease) if that household were in the intermediated sector in the liberalized economy, as compared with the same household in the economy without intermediation, a restricted economy.23 If, in fact, the wage is the same in the liberalized and restricted economies, then this is also the obvious, traditional partial equi- librium experiment that entails a simple comparison of matched pairs, each person with the same wealth-cost (b, x) combination but residing in two different sectors of a given economy, one with intermediated sector treatment and one without. The wage is the same with and without intermediation in both SES and Townsend- Thai simulations before 1990, when the subsistence sector is not depleted. If the wage is different across the two economies, this cross-sector comparison does not measure the net welfare impact of the financial sector liberalization. Instead, it measures end-of-period welfare dif- ferences across sectors of a given economy that has experienced price changes due to liberalization. To be more specific, those in the nonintermediated sector of the liberalized economy will experience the impact of the liberalization through wage changes. Workers in the nonintermediated sector may benefit from wage increases, but entrepreneurs in the nonintermediated sector suffer losses, because they face a higher wage. There is, of course, a similar price impact for those in the intermediated sector, but there is a credit effect there as well. And such wage effects are present using the parameters esti- mated from both data sets after 1990. 268 GINÉ AND TOWNSEND More to the point, differences in estimates for a given economy provide an inaccurate assessment of welfare changes if liberalization influences the wage. In this case, differences in the estimator of income of laborers would identify only changes in income from sav- ings, because both sectors face a common wage. Analogously, losses due to wage changes would not be captured in a comparison of entrepreneurial profits across both sectors.24 Implicit in this discussion is another problem that has no obvious remedy, given the model presented in this chapter. Although house- holds in the model maximize end-of-period wealth, they pass on a fraction of that wealth to their heirs. Thus, the end-of-period wealth effects of the liberalization are passed on to subsequent generations. The problem is that there is no obvious summary device; house- holds do not maximize discounted expected utility, as in Green- wood and Jovanovic (1990) and Townsend and Ueda (2006), for example. So the authors do not attempt to circumvent the problem in this chapter but rather present the more static welfare analysis for various separate periods. They first look at the liberalized economy in 1979, three years after the initial policy start-up (in 1976), using the overall best fit: the Townsend-Thai data economy with financial sector liberalization. As noted earlier, the wage has not yet increased as a result of the liber- alization. Its value is 0.0198 in the liberalized and restricted bench- mark economies. And the interest rate in the intermediated sector of the liberalized economy is very high, at 93 percent. This reflects the high marginal product of capital in an economy with a relatively low distribution of wealth. Figure 8.6a displays the corresponding end-of-period wealth per- centage changes in the same (b, x) space. Because the wage is the same in both sectors, agents will only benefit from being in the credit sector, not only because they can freely borrow at the prevailing rate if they decide to become entrepreneurs but also because they can deposit their wealth and earn interest on it. The wealth gain result- ing from interest rate earnings can best be seen by fixing x and mov- ing along the b axis, noting the rise. If, however, one looks at the highest wealth, b 0.5 edge, the wealth changes that correspond to changing setup costs x can be tracked. Moving forward from the rear of the diagram in figure 8.6a, at high x the wealth increment is shown to be constant; however, these households had workers in both economies, so setup costs x were never incurred. So the wealth increment drops for households with entrepreneurs in the no-credit economy that invested some of their wealth in the setup costs x. Those with high x gain the most, WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 269 Figure 8.6 Welfare Comparison, Townsend-Thai Data, 1979 b. WNC 0.02, WC 0.02, r 1.93 1.0 0.8 W in NC, C 0.6 x a. Percentage change in wealth 0.4 E in NC, W in C 0.2 W in NC, E in C 2,500 E in NC, C 0 0.1 0.2 0.3 0.4 2,000 b c. Wealth distribution of no-credit economy wealth 1,500 0.2 in x 0.1 1,000 change % 0 500 0.1 0.2 0.3 0.4 b d. Wealth distribution of credit economy 1.0 0.2 0.5 x 0.1 x 0.4 0.3 0 0.1 0.2 0 b 0.1 0.2 0.3 0.4 b workers in the no-credit economy and intermediated sector of the liberalized economy entrepreneurs in the no-credit economy and workers in the intermediated sector of the liberalized economy workers in the no-credit economy and entrepreneurs in the intermediated sector of the liberalized economy entrepreneurs in the no-credit economy and intermediated sector of the liberalized economy Source: Authors' creation. by quitting that investment and becoming workers in the inter- mediated sector. Thus, the percentage of wealth increment drops as x decreases. A trough is reached, however, when the household head decides to remain an entrepreneur. Lower setup costs benefit entrepreneurs in the intermediated sector more than in the corre- sponding no-credit economy, however, because the residual funds can be invested at interest. Hence, the back edge rises as x decreases further. 270 GINÉ AND TOWNSEND Figure 8.6b displays the corresponding occupation partition, but now with variables representing given beginning-of-period (b, x) combinations for the corresponding occupation of a house- hold in both the no-credit economy and the credit sector of the intermediated economy. The darker shading in figure 8.6b denotes households with (b, x) combinations that do not change their occupation as a result of the liberalization--that is, they are entre- preneurs (E) in the no-credit (NC) economy and in the intermedi- ated sector of the liberalized (C) economy, or workers (W) in both instances. The lighter shading denotes households that switch: these include low-wealth, low-cost agents who had been workers but became entrepreneurs; and high-wealth, high-cost agents who had been entrepreneurs but became workers. As explained earlier, figure 8.6b represents the overlap of the occu- pational maps in both sectors. For the credit sector, the key para- meter is x~(w, R), whereas for the no-credit sector, it is the curve ^ xe(b, w). The most dramatic welfare gains, however, are experienced by agents who are compelled to be workers in the no-credit economy but become entrepreneurs in the intermediated sector. Although their setup cost was relatively low, their wealth was not enough to finance it. They were constrained on the extensive margin. When credit barriers are removed, however, these entrepreneurs benefit the most. The sharp vertical rise corresponds to those on the margin of becoming entrepreneurs in the no-credit economy. Intuitively, because of their low x, these workers would have earned the high- est profits if they could have become entrepreneurs. Credit in the intermediated sector allows that option. A problem with this analysis, however, is that it may compute welfare gains for households with (b, x) combinations that do not actually exist in either the liberalized economy or the no-credit economy. In other words, those households have zero probability to exist under the endogenous distribution of wealth. To remedy this, figure 8.6c and 8.6d display the wealth distribution of the no-credit economy and the wealth distribution of the credit economy (over both sectors), respectively, in 1979. The upper part of table 8.2 displays the welfare gains from liber- alization in 1979 for both weighting distributions. The mean gains correspond to roughly 1.5 times and twice the average household yearly 1979 income25 using the intermediated economy wealth dis- tribution and the nonintermediated economy wealth distribution, respectively, as weighting functions. The modal gains are signifi- cantly lower, at roughly 17 or 19 percent of the 1979 average house- hold yearly income. WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 271 Table 8.2 Welfare Gains and Losses, Intermediated and Nonintermediated Economic Wealth Distribution Intermediated economic Nonintermediated economic wealth distribution wealth distribution 1997 Thai U.S. Percentage 1997 Thai U.S. Percentage Indicator baht dollars of income baht dollars of income From Townsend-Thai data (1979) Welfare gains Mean 82,376 3,295 200.93 61,582 2,463 150.21 Median 22,839 914 55.71 3,676 147 8.97 Mode 7,779 311 18.97 6,961 278 16.98 Percentage of population 100 100 From Socio-Economic Survey data (1996) Welfare gains Mean 76,840 3,074 100.54 83,444 3,338 109.18 Median 25,408 1,016 33.24 20,645 826 27.01 Mode 25,655 1,026 33.57 18,591 744 24.32 Percentage of population 86 95 Welfare losses Mean 117,051 4,682 107.59 115,861 4,634 106.50 Median 113,705 4,548 104.51 112,097 4,484 103.04 Mode 117,486 4,699 107.99 118,119 4,725 108.57 Percentage of population 14 5 Source: Authors' calculations. Note: SES Socio-Economic Surveys conducted by the National Statistical Office in Bangkok, Thailand. A contrast is offered by the welfare comparison from the simula- tion using the "best fit" estimated maximum likelihood estimation parameters from the SES data in 1996. The wage is 0.05 in the non- intermediated economy and runs to 0.08 in the intermediated one. Thus, agents who remain workers in the credit sector are better off in no small part because they earn a higher wage, and those who remain entrepreneurs in both economies end up losing somewhat because they face higher labor costs (although they may gain because of the savings rate). The interest rate in the intermediated sector has fallen to 9 percent. These welfare gains and losses are reported in the lower part of table 8.2. Using the intermediated economy wealth distribution as weighting function, the model predicts that 86 percent of the pop- ulation benefits from the financial sector liberalization, and even 272 GINÉ AND TOWNSEND more benefit (95 percent) if the nonintermediated wealth distribu- tion is used. The modal welfare gains of those who gain corre- spond to roughly 34 percent (when the intermediated wealth dis- tribution is used as weighting function) and 24 percent (nonintermediated wealth distribution) of the 1996 average house- hold yearly income. The mean losses, for those who are worse off, amount to 1.08 times or 1.06 times the average household yearly income for the sample of entrepreneurs for intermediated and non- intermediated wealth distribution, respectively. Thus, it seems that Figure 8.7 Welfare Comparison, Socio-Economic Survey Data, 1996 b. WNC 0.05, WC 0.08, r 1.09 1.0 0.8 W in NC, C 0.6 x 0.4 a. Percentage change in wealth W in NC, E in C 0.2 E in NC, C 0 0.05 0.1 0.15 0.2 0.25 0.3 b 900 c. Wealth distribution of no-credit economy 800 0.2 700 x 0.1 600 wealth in 500 0 400 0.05 0.1 0.15 0.2 0.25 0.3 b change 300 d. Wealth distribution of credit economy % 200 0.2 100 0.3 0 x 0.1 0.2 1.0 0.1 b 0 0.5 0.05 0.1 0.15 0.2 0.25 0.3 x 0 b workers in the no-credit economy and intermediated sector of the liberalized economy workers in the no-credit economy and entrepreneurs in the intermediated sector of the liberalized economy entrepreneurs in the no-credit economy and intermediated sector of the liberalized economy Source: Authors' creation. WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 273 a fraction of the population loses a great deal from the liberaliza- tion policy (see also figure 8.7). Macro Impact and Policy Evaluation Figure 8.3 (see page 256) displays in a solid line capital inflows as a fraction of GDP. The data are from the Bank of Thailand as reported in Alba, Hernandez, and Klingebiel (1999). From 1976 to 1986, private capital inflows to Thailand remained relatively low at an average of 1.05 percent of GDP. From 1986 to 1988, however, they increased rapidly to 10 percent of GDP and remained at that aver- age level until 1996. This enhanced capital availability was funneled through the financial sector and thus is modeled here as additional capital for households with access to the financial market (those residing in the credit sector). The authors ran this extended (open-economy) ver- sion of the model at the estimated and calibrated parameters and compared it with the previous closed-credit economy model at the same estimated and calibrated parameter values from the two data sets.26 Although not shown, capital inflows contributed to a larger number of entrepreneurs and larger firm size, especially in the late 1980s and early 1990s. Since the marginal product of labor increases with capital utilization, more labor is demanded, and thus the frac- tion of subsistence workers is depleted earlier. Labor share rises and inequality decreases, both relative to the actual path and relative to the earlier simulation. The interest rate tends to be lower with cap- ital inflows. Nevertheless, the welfare changes are small--indeed, almost negligible. Because the surge in capital inflows coincides with the phase of high growth of per capita GDP, it has often been portrayed as an important factor in that high growth. To disentangle the extent to which the phase of high growth was due to increased participation in the credit market versus additional capital availability from capi- tal account liberalization, the authors simulated the economy at the estimated and calibrated parameter values, allowing for interna- tional capital inflows but using a linearized credit participation from 6 percent to 26 percent (that is, a 1 percent increase per year for each of the 20 years). As shown in figure 8.8, this version of the model fails to match the upturn in GDP growth as compared with the benchmark credit economy. Thus, it seems from the model that capital inflows per se were not the cause of the high growth that Thailand experienced in the late 1980s. 274 GINÉ AND TOWNSEND Figure 8.8 Access to Capital and Foreign Capital Inflows, Socio-Economic Survey Data and Townsend-Thai Data a. Socio-Economic Survey data b. Townsend-Thai data 12 14 10 12 8 10 6 rate 4 rate 8 2 6 0 growth 4 2 growth 2 4 6 0 8 2 1976 1980 1984 1988 1992 1996 1976 1980 1984 1988 1992 1996 year year Thai economy closed economy economy with capital flows and linear Source: Authors' creation. Conclusion This chapter describes how an algorithm that combines theory and data can be used to provide useful information about the relative impacts of financial liberalization. Although the magnitude of the welfare gains and losses depends on the assumptions of the model, and the plausibility of these depends on the context, the advantage of this approach is that it can be used by researchers to trace the effects of particular recommen- dations to given sets of assumptions or rules. Thus, because the logic of the model is made explicit, it becomes easier to improve the model to better characterize the economy. For example, in related work, Paulson and Townsend (2001) use the Townsend-Thai data to estimate (using maximum likelihood methods) not only the LEB model featured in this chapter, but also the collateral-based lending model of Evans and Jovanovic (1989) (the EJ model), and also the incentive-based lending of Aghion and Bolton (1997) and Lehnert (1998) (the ABL model). Observed rela- tionships of entrepreneurship, investment, and access to credit as functions of wealth and talent suggest that the ABL model best fits the data, but the EJ model fits well for those with relatively low lev- els of wealth and those in the northeast region of Thailand, and the WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 275 LEB model discussed in this chapter is a close contender. This find- ing suggests that a calculation of the welfare gains and losses to financial intermediation based on these other models would be worthwhile, even though the average and modal estimates presented in this chapter should not be rejected out of hand. It seems plausi- ble, however, that the dramatic gains found near the loci when wealth can only cover the setup costs x (or the 45-degree line) would be vulnerable to alternative specifications. The growth and inequality literature that relies on these under- pinning models presupposes (as in the LEB model discussed in this analysis) an overlapping generations model with a bequest motive or a simplistic, myopic solution to the household savings problem. More work is needed to make the models dynamic and to couple households with firms and model the intertemporal decision prob- lems confronted by firms. But given the preliminary results pre- sented in this chapter, that work appears to be warranted. Notes 1. See, for example, Gallup, Radelet, and Warner (1998) and Dollar and Kraay (2002) for evidence that growth helps reduce poverty. For con- cerns about these approaches, see Ravallion (2001, 2002). 2. The focus here is on this 20-year transitional period, not on the finan- cial crisis of 1997. The authors' own view is that one needs to understand the growth that preceded the crisis before analyzing the crisis itself. 3. The authors used the functional forms contained in the 1998 working paper, although the published 2000 version contains slight modifications. 4. In extended models, this would be the analog to the distribution of human capital, although the education investment decision obviously is not modeled in this chapter. Related extensions allow wealth and talent to be correlated. 5. The authors also estimate the Lloyd-Ellis and Bernhard (LEB) model for various stratifications of wealth (above and below the median, for exam- ple) to see how parameter m varies with wealth. This way, wealth and tal- ent are allowed to be correlated. Even though the point estimates of m vary significantly, simulations with the different estimates of m are roughly sim- ilar. If correlation between wealth and ability were allowed, there could be poverty traps, as in Banerjee and Newman (1993). The authors do recognize that in practice, wealth and ability may be correlated. In related work, Paulson and Townsend (2001) estimate (with the same data used here in chapter 8) a version of the Evans and Jovanovic (1989) model that allows the mean of unobserved ability to be a linear function of wealth and edu- cation. Evans and Jovanovic find the magnitude of both coefficients to be 276 GINÉ AND TOWNSEND small. Karaivanov (2003) allows a correlation between parameter m and wealth for a series of exogenously specified incomplete market models. Buera (2003) develops a dynamic model that endogenizes the relationship between wealth and talent. 6. More general monotonic transformations of the utility function U(C, B) are feasible, allowing utility to be monotonically increasing but concave in wealth. In any event, the overall utility maximization problem is converted into a simple end-of-period wealth maximization problem. More realistically, the model can be interpreted as having an exogenously imposed myopic savings rate , which is later calibrated against the data. Attention can then be focused on the nontrivial endogenous evolution of the wealth distribution. 7. The model is, at best, a first step in distinguishing between agents with and without access to credit. In this context, it is assumed that inter- mediation is perfect for a fraction of the population and nonexistent for the rest. Allocation of a household into the intermediated sector is an exoge- nous treatment. 8. Note that, in particular, net aggregate deposits in the financial inter- mediary can be expressed as total wealth deposited in the intermediated sector minus credit demanded for capital and fixed costs. For low levels of aggregate wealth, the deposits will constrain credit and the net will be zero. However, note that net aggregate deposits can be strictly positive if there is enough capital accumulation, in which case the savings and the storage technology are equally productive, and both yield a gross return of R 1. 9. The relative magnitude of the fixed costs will drop over time (as wealth evolves). 10. The appendix in Giné and Townsend (2004) details how the coeffi- cients are estimated and how the production parameters are recovered. 11. In particular, the "fmincon" routine of the MATLAB Optimization Toolbox software was used, starting from a variety of predetermined guesses. 12. See Jeong (1999) for details about the Socio-Economic Survey (SES); for an application of the SES data, see Schultz (1997) or Deaton and Paxson (2000). 13. Robert M. Townsend is the principal investigator for this survey. See Townsend and others (1997) and Paulson and Townsend (2001) for details. 14. These households could have borrowed from friends and relatives, although the bulk of the borrowing through this source consists of con- sumption loans rather than business investments. 15. Unfortunately, estimating a model that features a unique wage by exploiting the geographic variation in the wage observed in the data is a contradiction. Costly migration could be introduced, but that explicit approach is not taken in these calculations. Some confidence might, WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 277 however, be drawn from the fact that these are secondary data--and that its estimates are compared to those from the SES data set, with its temporal variation in wages consistent with the estimated model. 16. The intermediated sector, with its distribution of wealth, is assumed to be scaled up period-by-period according to the exogenous credit expan- sion. Alternatively, a sample from the no-credit sector distribution of wealth could have been selected (with the corresponding fraction to the exogenous expansion), but the increase is small and would have made little difference in the numerical computations. 17. This estimated measure of wealth is likely to differ (in scale and units) from the wealth reported in the Townsend-Thai data; these calculations allow for a different scaling factor to convert SES wealth into the model units. In other words, the authors employ two scaling factors to calibrate the model, using the parameters estimated with the Townsend-Thai data. One is esti- mated with maximum likelihood techniques and converts wealth and incomes reported in the data, whereas the other is calibrated and converts the SES wealth measure used to generate the economywide initial distribution. 18. In computing the growth rate, one observation is lost, so the time index in the formula given in equation (8.9) is from 1977 to 1996 for the growth rate statistic. 19. All variables are weighted equally to focus on a particular period; more weight may be given to those years. Analogously, all the weights may be set to one variable to assess how well the model is able to replicate that variable alone. All weights are renormalized so that they add up to unity. 20. As mentioned earlier, when the Townsend-Thai data are used, a grid of 20 scaling factors is searched for the initial distribution of wealth. 21. The scaling factor chosen for the initial distribution is 15 percent of the one used to convert wealth when using the maximum likelihood estimation. 22. It is important to conduct a robustness check to see whether the results are sensitive to estimated calibrated parameters. In these calcula- tions, robustness is checked in two ways. First, parameters are changed one at a time and checked to see whether the new simulation differs significantly from the benchmark simulation. Alternatively, one could see how sensitive the model is to changes in all of the estimated parameters in one simulation. For details on this approach, see Jeong and Townsend (2003) and Giné and Townsend (2004). 23. If the comparison had been conducted using an intermediated econ- omy (in which the intermediated sector is fixed at 6 percent), then when agents living in the credit sectors of both economies were compared, the welfare gains and losses that arose (because of interest rate levels) were larger in the economy that did not experience liberalization. Therefore, wealthier but less talented workers in the benchmark economy may be 278 GINÉ AND TOWNSEND better off without liberalization, because they already earn a higher interest rate income. 24. A cross-country comparison would be more accurate if one could control for the underlying environment, but countrywide aggregates would conceal the underlying gains and losses in the population. 25. The 1979 average household yearly income is estimated from the SES data. Actual SES data are not available for 1979, so it is interpolated using the average annual growth rate of SES data between 1976 and 1981. 26. In addition, the cost-of-living , subsistence-level growth rate , gr and the bequest motive parameter for this open-economy version were recalibrated and found to have even fewer differences compared with the closed-economy model. In particular, the calibrated bequest motive para- meter is lower for both data sets in the open-economy version, so that the depletion of subsistence workers occurs at a slower rate. References Aghion, Philippe, and P. Bolton. 1997. "A Theory of Trickle-Down Growth and Development." Review of Economic Studies 64 (2): 151­72. Alba, P., L. Hernandez, and D. Klingebiel. 1999. "Financial Liberalization and the Capital Account: Thailand 1988­1997." World Bank, Washington, DC. Banerjee, A., and A. Newman. 1993."Occupational Choice and the Process of Development." Journal of Political Economy 101 (2): 274­98. Buera, F. J. 2003. "A Dynamic Model of Entrepreneurship with Borrowing Constraints." University of Chicago. http://economics.uchicago.edu/ download/buera_applications2003.pdf. Das, M., M. Roberts, and J. Tybout. 1998. "Experience, Expectations and Export Dynamics." http://citeseer.ist.psu.edu/642470.html. Deaton, Angus, and C. Paxson. 2000. "Growth and Saving among Individu- als and Households." Review of Economics and Statistics 82 (2): 212­25. Dollar, David, and Art Kraay. 2002. "Growth Is Good for the Poor." Journal of Economic Growth 7 (3): 195­225. Evans, D., and B. Jovanovic. 1989. "An Estimated Model of Entrepreneur- ial Choice under Liquidity Constraints." Journal of Political Economy 97: 808­27. Felkner, J., and Robert M. Townsend. 2004. "The Wealth of Villages: An Application of GIS and Spatial Statistics to Two Structural Economic Models." Paper presented at the Center for Humanities, Arts, and Social Sciences EPIC workshop, "Spatial Thinking in the Social Sciences and Humanities," on December 18 and 19, 2006. http://www.chass.uiuc.edu/ events_gis_conference_dec_2006.html. WEALTH-CONSTRAINED OCCUPATIONAL CHOICE 279 Gallup, J. L., S. Radelet, and A. Warner. 1998. "Economic Growth and the Income of the Poor." Harvard Institute for International Development. Giné, Xavier, and Robert M. Townsend. 2004. "Evaluation of Financial Liberalization: A General Equilibrium Model with Constrained Occupa- tion Choice." Journal of Development Economics 74: 269­307. Greenwood, J., and B. Jovanovic. 1990. "Financial Development, Growth, and the Distribution of Income." Journal of Development Economy 98: 1076­107. Griffin, R., J. Montgomery, and M. Rister. 1987. "Selecting Functional Form in Production Function Analysis." Western Journal of Agricultural Economics 12 (2): 216­27. Heckman, James, L. Lochner, and C. Taber. 1998. "Explaining Rising Wage Inequality: Explorations with a Dynamic General Equilibrium Model of Labor Earnings with Heterogeneous Agents." Review of Economic Dynamics 1: 1­58. Jeong, H. 1999. "Education and Credit: Sources of Growth with Increasing Inequality in Thailand." PhD thesis, University of Chicago, Population Research Center 98-12. http://ideas.repec.org/p/fth/chiprc/98-12.html. Jeong, Hyeok, and Robert M. Townsend. 2003. "Growth and Inequality: Model Evaluation Based on an Estimation-Calibration Strategy." USC Institute of Economic Policy Research Paper No. 05.10. Available at SSRN: http://ssrn.com/abstract=661283. Karaivanov, A. 2003. "Financial Contracts and Occupational Choice." PhD thesis, Department of Economics, University of Chicago. Klinhowhan, U. 1999. "Monetary Transmission Mechanism in Thailand." Master's thesis, Thammasat University, Bangkok, Thailand. Lehnert, A. 1998. "Asset Pooling, Credit Rationing, and Growth." Working Paper 1998­52, Finance and Economics Discussion Series, Federal Reserve Board, Washington, DC. Lewis, A. 1954. "Economic Development with Unlimited Supplies of Labor." Manchester School of Economic Studies 28: 139­91. Lloyd-Ellis, H., and D. Bernhardt. 2000. "Enterprise, Inequality, and Eco- nomic Development." Review of Economic Studies 67 (1): 147­68. Okuda, H., and F. Mieno. 1999. "What Happened to Thai Commer- cial Banks in Pre-Asian Crisis Period: Microeconomic Analysis of Thai Banking Industry." Hitotsubashi Journal of Economics 40 (2): 97­121. Paulson, Anna, and Robert M. Townsend. 2001. "The Nature of Financial Constraints: Distinguishing the Micro Underpinnings of Macro Models." http://cier.uchicago.edu/papers/Paulson/PaulsonTownsend1.pdf. ------. 2004. "Entrepreneurship and Financial Constraints in Thailand." Journal of Corporate Finance 10: 229­62. Ravallion, Martin. 2001. "Growth, Inequality, and Poverty: Looking Beyond Averages." World Development 29 (11): 1803­15. 280 GINÉ AND TOWNSEND ------. 2002. "The Debate on Globalization, Poverty, and Inequality: Why Measurement Matters." Policy Research Working Paper No. 3038, World Bank, Washington, DC. Schultz, T. P. 1997. "Diminishing Returns to Scale in Family Planning Expenditures: Thailand 1976­81." Yale University, New Haven, CT. http://www.econ.yale.edu/~pschultz/thai.pdf. Townsend, Robert M., with Anna Paulson, Sombat Sakuntasathien, Tae Jeong Lee, and Michael Binford. 1997. Questionnaire Design and Data Collection for National Institute of Child Health and Human Develop- ment grant, "Risk, Insurance and the Family," and National Science Foundation grants. Townsend, Robert M., and Kenichi Ueda. 2006. "Financial Deepening, Inequality, and Growth: A Model-Based Quantitative Evaluation." Review of Economic Studies 73 (1): 251­93. Veracierto, M. 1998. "Plant Level Irreversible Investment and Equilibrium Business Cycles." Working Paper Series WP-98-1, Federal Reserve Bank of Chicago. (Revised) http://ideas.repec.org/p/fip/fedhwp/wp-98-1.html. PART IV Macro Approach with Disaggregated Public Spending 9 Aid, Service Delivery, and the Millennium Development Goals in an Economywide Framework François Bourguignon, Carolina Díaz-Bonilla, and Hans Lofgren The United Nations (UN) Millennium Summit in 2000 witnessed the historic adoption of the Millennium Development Goals (MDGs) by the global community. These goals committed the international com- munity to achieving by 2015 an ambitious vision of development that encompasses not only higher incomes, but also broader human development (HD) goals related to health, education, and access to water and sanitation. Two years later, in Monterrey, the interna- tional community met again to address the challenge of financing the MDGs, where it was recognized that a substantial increase in official development assistance (ODA) would be required to reach the MDGs. But it was also recognized that donors and recipients had parallel responsibilities. Aid alone would not be enough­­donor commitment to provide more resources needed to be matched by recipient policies and programs that would ensure that incremental assistance was well used. Over the last few years, some encouraging signs have demonstrated progress toward these goals. Many developing countries have acceler- ated growth and reduced poverty by reforming domestic institutions and integrating their economies into global markets. But this progress has been uneven. Many countries, particularly in Sub-Saharan Africa, 283 284 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN lag behind in growth and remain off track in terms of achieving the goal of halving poverty by 2015. Regarding the nonpoverty MDGs, the picture is also mixed. Although some countries have made impres- sive advances toward health and education objectives, others­­even some with a strong growth performance­­lag behind. This mixed performance raises important questions. First, faster growth is a facilitating factor for MDG achievements on the HD goals. Thus, slow growers might find it difficult to make progress on the nonpoverty MDG fronts. It is also the case, however, that improved health and education standards can increase productivity and accelerate growth at a later stage. These improvements can bring positive synergies when service access improves simultane- ously in different areas (health, education, water, and sanitation), which ultimately may increase the efficiency of service delivery or reduce its cost. To the extent that growth and higher incomes can generate increased funding for services and raise service demand, investing more today in nonpoverty MDGs could trigger a virtuous circle of growth and human development. Second, growth and HD service delivery may conflict. The need to finance HD investments may crowd out investments and growth in other parts of the economy, making the allocation of government resources between human development and public infrastructure (including roads, power, and irrigation) a critical policy issue. Another potential source of conflict is that relative costs of govern- ment services will rise if productivity growth within the government is slower than in the private economy. These tensions work in the other direction as well. Increased employment of teachers, health personnel, and other skilled workers may drive up wages and reduce the number of skilled workers available for private sector employ- ment, at least in the short and medium terms. Various approaches have been used to plan and monitor progress toward achieving the MDGs and to evaluate the additional (or total) public resources, including foreign aid, needed to meet them. Clemens, Kenny, and Moss (2004) and Reddy and Heuty (2004) surveyed a large number of studies that forecast and cost MDGs. As emphasized by Vandemoortele and Roy (2004), however, data avail- ability and simplifying analytical assumption severely affected the quality of quantitative estimates of all these studies. Four major sets of limitations affect studies on MDGs' achieve- ment. First, many sectoral studies have focused on individual MDGs, but even those studies that consider multiple goals often fail to properly account for the interdependencies that exist among dif- ferent MDGs and among policies designed to reach them. Second, MDG-related policies interact with the rest of the economy (namely, AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 285 the private sector) by altering prices of specific factors (such as skilled labor) and their overall supply. Third, intertemporal equilibrium con- sistency is seldom checked. Financing needs, debt accumulation, and the intertemporal sustainability of fiscal policies need to be integrated in a complete study on strategies to achieve the MDGs. Fourth, as stressed by Devarajan, Miller, and Swanson (2002), the policy and institutional environment is as important a component of success in achieving the MDGs as the availability of public resources or finan- cial assistance. Keeping these potential limitations in mind, the authors briefly report on some recent studies and approaches in use at the United Nations and the World Bank. The United Nations Development Programme (UNDP) Human Development Report (2005) addresses most of the MDGs, project- ing trends for individual countries, aggregated to regions and glob- ally. Policies and links between MDGs are not considered. The authors point out that it is problematic to make projections on a goal-by-goal basis, given strong links between different MDGs. Nev- ertheless, their approach may be adequate for their purpose­­that is, to highlight the fact that if current trends continue, most coun- tries will fail to achieve most MDGs. The report, however, is not designed for analysis of MDG strategies. With the more ambitious objective of helping countries design Poverty Reduction Strategy Papers (PRSPs), Christiaensen, Scott, and Wodon (2002) developed SimSIP (Simulations for Social Indicators and Poverty), a set of tools that address different aspects of strategy analysis (also applicable to MDGs), including target setting and assessments of costs and fiscal sustainability. The Excel-based tools are user-friendly and analytically sim- ple.1 The objective of the target-setting module is to assess the realism of targets related to poverty, health, education, and basic access to water and sanitation. The module provides alternative specifications for forecasting these indicators on the basis of econometrically estimated equations that include gross domestic product (GDP), urbanization, and time as arguments. The cost analysis, which covers the same set of indicators (except for poverty), is based on fairly detailed modules that consider input needs and assessments about future wage changes. The module for analysis of fiscal sustainability compares the estimated costs of achieving targets to available public funding (based on assump- tions about GDP growth, tax collection, and the level of sustain- able deficits). The target-setting module is a useful tool for assessing target realism; however, the fiscal sustainability component is relatively weak. This weakness reflects the fact that a set of inde- pendent tools cannot capture interdependencies between GDP 286 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN growth, MDG targets, program costs (including wage changes), and alternative financing approaches. The different publications of the UN Millennium Project report represent a more detailed sectoral approach (see, for example, UN Millennium Project 2005). Its main feature is to estimate and add up the costs of specific interventions in areas such as education, health, and public infrastructure. As implemented, this is essentially a fixed-coefficient­fixed-price planning exercise. This approach-- rich in detail--has typically ignored or simplified the synergies across the MDGs and, more important, the interactions with the broader economy. Agénor and others (2005) apply a novel approach to MDG analysis to Niger.2 Recognizing the need for an economywide per- spective, they combine a macro model with an MDG module. Government spending has different repercussions depending on whether it is identified as being used for education, health, or infrastructure. The MDG module is used for postcalculations of MDGs and other social indicators, including poverty, malnutri- tion, literacy rate, infant mortality, life expectancy, and access to safe water. The determinants of most of these indicators are esti- mated using cross-section, cross-country data for a sample of developing countries (when feasible, limited to Sub-Saharan Africa), allowing for links between MDGs. A key strength of this approach is that it requires relatively little data and draws on econometrically estimated parameters. The macro model is highly aggregated, however, presenting two limitations: (1) it has only one production sector (meaning that one dollar of additional gov- ernment demand for investment in infrastructure has the same direct effect on production and imports as one dollar of addi- tional government demand for education), and (2) it does not include intermediate inputs, factor markets, or factor wages (rents). These limitations restrict the model's ability to analyze key aspects of MDG strategies, such as the labor market reper- cussions of scaled-up government services and Dutch disease effects (characterized by an appreciating real exchange rate, a shift of resources toward nontradable goods, and lower export growth). Its high level of government and labor market aggrega- tion (only "educated labor" is used in production) makes it more difficult to draw on for Public Expenditure Reviews (PERs) and in other contexts for fiscal analysis. Nevertheless, the model can pro- vide useful macro insights for strategy analysis and may be devel- oped further to address some of these limitations. The links between growth, service delivery, and MDG achieve- ments outlined above demonstrate that a more sophisticated AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 287 framework is needed. The analysis must consider macroeconomic factors and trade-offs between objectives. For example, the prospect of significant increases in foreign aid (for most countries in Sub-Saharan Africa, external assistance required to meet the MDGs in 2015 may require more than a doubling of aid flows) leads to concerns over the possibility of Dutch disease. A related critical issue is the pace at which large, aid-financed programs should be scaled up. Rapid initial expansion may drive up costs more quickly and could be more expensive in real present-value terms. Conversely, given time lags, especially in education, expanding investment too slowly may make it impossible to achieve the MDGs by 2015. A coherent analytical framework is needed to capture macro-micro links, Dutch disease effects, and timing issues. The need to evaluate and provide policy advice on such trade- offs­­across sectors and over time­­has led to innovative research efforts. This chapter presents Maquette for MDG Simulations (MAMS), which was produced within a research program on the MDGs conducted at the World Bank. MAMS is an economywide framework designed to analyze the interactions between the delivery of HD services (health, education, water, and sanitation), the MDGs, growth, and foreign aid. The framework is equally applicable to the analysis of the same set of policy issues in the context of Poverty Reduction Strategies (PRSs). MAMS belongs to the class of dynamic general equilibrium models, but it has been substantially augmented to capture key processes that generate MDG outcomes as well as feedbacks to the rest of the economy. MAMS does not replace detailed sectoral studies, but instead complements and draws on the research that underpins sector strate- gies for achieving the MDGs. Without sector studies to provide a strong empirical basis, the analysis of MDG strategies in an econo- mywide framework loses much of its power. By fully embedding such strategies in a comprehensive economywide framework, MAMS fills a gap in the toolkit that is available to policy analysts. Especially for low-income countries, the policy challenges related to the MDGs cannot be well understood unless sector issues are viewed in the context of constraints in the macro environment and in labor markets. This chapter is divided into two sections. The next section pre- sents the model structure, emphasizing the features that distinguish MAMS from other computable general equilibrium (CGE) models, particularly the feedbacks from and links between different MDG goals and the rest of the economy. This discussion is followed by a set of simulations that illustrates how MAMS captures some of the 288 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN MDG issues discussed in this introduction. The conclusion outlines a possible future research agenda. Model Structure and Key Mechanisms A key premise of the model is that government spending and MDG outcomes are linked in a dynamic way, with several outside influ- ences. But that relationship is not a simple, invariable one for three essential reasons: · The returns to scale of government spending vary with the level of service delivery. At low levels, increasing returns may prevail as network effects, learning effects, and synergies are predominant. At high levels of service delivery, government spending may suffer from decreasing returns to scale. Water supply, health care, and education can be provided relatively easily in densely populated areas, but doing so becomes increasingly expensive as coverage expands to remote areas. When mortality rates are already low, it becomes increasingly difficult to reduce these rates further. Similarly, if com- pletion rates in education are already high, it is difficult to ensure that the last percentages of children complete the program. · Effectiveness of government spending depends on many vari- ables. For example, spending on education becomes more effective if health conditions improve (reducing absenteeism at schools), pub- lic infrastructure improves (facilitating access to schools), income levels rise (and parents are less inclined to keep children at home), or skill premiums increase (triggering a greater incentive to finish formal education). In general terms this means that spending on ser- vices becomes more effective if demand conditions for those services are more favorable. · Costs of service delivery change with macroeconomic conditions. The services are often skill intensive and in many cases also capital intensive. The more intense the MDG effort, the stronger the impact on costs as skilled labor becomes scarcer and financial conditions become tighter. From a general budgetary perspective, the impacts on costs are even larger because changes in macroeconomic conditions affect not only MDG spending, but also other, non-MDG government spending (as well as the competitiveness of the private sector). The first two aspects (changing returns to scale and impact of demand variables) are captured in the "MDG production functions" introduced in MAMS. The last aspect (macroeconomic interactions) is captured as the MDG production functions are incorporated in a dynamic CGE framework that also includes detailed fiscal accounts. AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 289 The dynamic framework not only reflects the key macroeconomic interactions, but also allows planning to target the MDGs in 2015 and to incorporate autonomous baseline forecasts. The Production of the MDGs MAMS focuses on the subset of MDGs that is most costly and has the greatest interaction with the rest of the economy: universal primary school completion (MDG 2, measured by the net primary completion rate), reduced under-five and maternal mortality rates (MDGs 4 and 5), halting and reducing the incidence of HIV/AIDS and other major dis- eases (part of MDG 6), and increased access to improved water sources and sanitation (part of MDG 7). Achievements in terms of poverty reduction (MDG 1) are also monitored.3 Because of their impact on overall growth and, through that, on poverty, investments in public infrastructure are explicitly taken into account. This consid- eration also allows the modeling of the positive influence of infra- structure on the effectiveness of spending on social sectors. The modeling of the production of a typical MDG (except for MDG 2­­primary school completion for all­­which is discussed later in the context of the education sector) consists of two blocks of equa- tions: the first models the production of MDG-related services; the second defines MDG outcomes as a function of service delivery and other determinants.4 In the first block, the production of MDG- related services, substitution possibilities among the three broad cate- gories of inputs (labor, capital goods, and intermediate products) are assumed to be negligible. Assuming fixed input-output coefficients, the inputs required for a level Q of service delivery are as follows: L Q L (9.1) K Q k INT Q , INT where L is the labor requirement (for example, teachers or nurses), K is the capital requirement (for example, classrooms or hospital beds), and INT represents intermediate inputs (for example, text- books or medicine). Aggregate labor L results from the combination of three different kinds of labor: those with less than completed secondary education (N), those with completed secondary education (S), and those with completed tertiary education (T). The elasticity of substitution between the different forms of labor is assumed to be constant, and the government is assumed to use the most cost-effective combination of different labor types. The demand for specific education categories 290 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN thus depends on education premiums. Under the assumption of con- stant substitution elasticity, , the demand is given by the following: W N L N WN W (9.2) S L S WS W T L , T WT where WN, WS, and WT are the respective wages for workers with less than secondary, completed secondary, and completed tertiary schooling, and W is the average wage across all workers­­and the unit cost of aggregate labor, L. The positive coefficients, , describe the structure of the labor demand by education category for given unit costs of the various categories. The capital stock is built up over time through investments and deteriorates at a constant depreciation rate ( ). (9.3) It Kt (1 1 )Kt . The investment in the current period (t) is chosen such that the required capital stock in the next period, as given by the capital demand equation (9.1), is achieved. Government capital spending on MDGs will be large when service delivery is expanding and will reduce to replacement investment when the level of service delivery is constant. Intermediate purchases include domestically produced products and imported products, with the two linked through a constant elas- ticity of substitution demand function. As for labor, cost reduction by the government implies that the demand for domestic (INTd) and imported (INTm) intermediate inputs takes the following form: P INTd INT d (9.4) Pd P INTm INT , m Pm where P is the unit price of the aggregate intermediate input (INT); Pd and Pm are the price of the domestic and imported goods, respec- tively; and ( 0) is the elasticity of substitution. As before, is a positive coefficient. The second block of equations defines MDG achievements, relat- ing service delivery and other determinants to MDG indicators (for MDGs 4, 5, 7a, and 7b). The changing returns to scale are repre- sented by a logistic curve, showing increasing returns to scale at low AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 291 levels of development indicators and decreasing returns to scale at high levels of development indicators. (9.5) MDGk extk k 1 e k kZk , where MDGk is the indicator used to monitor MDG k; Zk is an intermediate variable that summarizes the influence of the determi- nants of MDG performance; extk is the extreme (maximum or minimum) level of the indicator (for example, 1 or 100 percent for completion rate); shows the responsiveness of the indicator to k changes in Zk; determines whether increasing or decreasing k returns prevail at the starting point; and is used to replicate the k initial MDG value and the slope of the function, which is positive if declines in the MDG indicator denote an improvement (mortality rate) and negative in the reverse situation (for example, rates of access to safe water). The intermediate variable, Zk, is defined by the following Cobb-Douglas relationship: n (9.6) Zk Qk k . Dik . ik i=1 Table 9.1 lists the arguments (for example, service levels, Qk, and other determinants, Dik) that defined Zk in the Ethiopia application. Simulation results are discussed at the end of this chapter. These Table 9.1 Determinants of MDG Achievements Other determinants Per capita Per capita real service household Public Other MDG delivery consumption infrastructure MDGs 4 X X X 7a, 7b 5 X X X 7a, 7b 7a X X X 7b X X X Source: MAMS version for Ethiopia was developed by the authors. Note: MDG Millennium Development Goal. The MDGs referred to in this table are defined as follows: MDG 4: reduce by two-thirds the mortality rate among children under five; MDG 5: reduce by three-quarters the maternal mortality ratio; MDG 7a: halve the proportion of people without sustainable access to safe drinking water; and MDG 7b: halve the proportion of people without sustainable access to sanitation services. The target year is 2015 and the reference year is 1990. The ser- vices related to these MDGs are health (disaggregated by technology) and water- sanitation services. Other determinants should be added if they are important in the context of a particular country study; if any of the determinants listed in the table are unimportant, then they can be omitted (or given an elasticity of zero). MDG 2 is covered in the following discussion of education. 292 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN variables are identified by sectoral studies underpinned by econo- metric analysis. They include other MDGs­­better access to water and sanitation may improve health outcomes (MDGs 4 and 5)­­as well as infrastructure or consumption per capita. For example, a higher level of consumption per capita may influence health achieve- ments positively. Pregnant women who are better fed face reduced health risks for themselves and for their babies. Among the "other determinants," per capita household consumption and other MDGs represent demand-side factors, whereas public infrastructure facili- tates both demand and supply. To implement the first block of MDG equations, data are required on government spending by function (one or more health sectors, water and sanitation, other public infrastructure, and other govern- ment) and type of outlay (current versus capital). Current outlays must be disaggregated into payments to different types of labor (wages) and intermediate inputs.5 This information, complemented by elasticities of factor substitution, is similar to what is required for other (nongovernment) sectors in a standard CGE model and can easily be built into the model's social accounting matrix (SAM). In parallel with data on payments to labor, information is also needed on the number of people employed. The information needed for this block typically can be found in sectoral studies and databases of governments, international organizations, and other research institutions. For the second block of MDG equations, which translates govern- ment services into MDG indicators, information is needed on (1) base- year values and 2015 targets for MDG indicators; (2) extreme values for MDG indicators; (3) a set of elasticities of MDG indicators with respect to the relevant determinants (with one version provided in table 9.1);6 (4) the position of the initial situation (in terms of MDGk or Zk) relative to the inflection point (at which the function switches from increasing to decreasing returns to scale); and (5) a scenario indicating one set of 2015 values for the arguments of equa- tion (9.6) under which the MDG in question is achieved. It is relatively straightforward to collect the base-year values and 2015 targets. With respect to the extreme values for MDG indicators, function (extk) can be determined by pure logic (for example, the maximum share of the population with access to a service is 1) or international experience (the minimum observed maternal mortality rate across countries). For a set of elasticities of MDG indicators, it is possible to draw on a growing body of econometric research, in particular in the areas of health and education. Although sometimes contradictory, the findings of these studies provide broad support for inclusion of the determi- nants referred to in table 9.1.7 Econometric estimates of basic MDG AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 293 elasticities are hampered by the fact that it is difficult or impossible to observe the full functional form­­at least among countries for which it can be asserted that MDG outcomes are generated by the same processes. These outcomes are difficult to achieve because they are concentrated within a limited range that is far from MDG targets and extreme outcome values. Given this fact, econometric analysis must be complemented with other approaches to be able to fully parame- terize the MDG production functions. Sectoral studies of MDG strate- gies and discussions with experts make it possible to determine the position of the initial situation and a scenario indicating one set of 2015 values. Using this information, one can infer from the logistic function the rate at which marginal returns decline and ensure that MAMS is consistent with sectoral studies. In sum, if data are avail- able for these five scenarios, it is possible to calibrate the , , , k k k and parameters.8 k The treatment of education is more complex than that of health and other MDGs. The model gives a complete account of the sec- tor, dividing it into different cycles (or levels): primary, secondary, and tertiary. The primary cycle is needed because it is linked to MDG 2. The higher cycles are needed to link education to the labor market, provide a complete picture of the dynamic fiscal consequences of achieving MDG 2, and expand the education sys- tem. In each grade in each cycle, a student may pass, drop out, or repeat the grade next year. Students who pass may proceed to a higher grade within the cycle or graduate from the cycle. In the lat- ter case, they may continue to the next cycle or exit from the school system. The two-block structure and the functional forms, described above for the other MDGs, also apply to education. However, the second block--equations (9.5) and (9.6)--is applied to two types of behavioral outcomes in all cycles: entry rates (to the first grade of any cycle, out of the qualified population9) and passing rates (from each grade within a cycle). More specifically, in the logistic functions, equation (9.5), the left-hand-side variables are the shares of students that pass their current grade (one vari- able per cycle) and the shares, out of the relevant population, that start the first year (also one variable per cycle). The extreme value for all of these variables is one. Other behavioral rates are com- puted on the basis of the share variables that are defined by the logistic. Rates of repetition and dropout are scaled up or down on the basis of changes in passing rates. The students who pass are split into graduates from the cycle and passers within the cycle, which assumes that as entry and passing rates improve, the stu- dents who pass eventually become evenly distributed across the grades within the cycle. 294 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN Each logistic equation (9.5) is associated with a Cobb-Douglas equation--compare equation (9.6)--where the relevant Z variable is defined. The arguments determining Zs in education may be similar to those appearing in table 9.1. In the Ethiopia application, the argu- ments determining the educational Zs include education services per student enrolled, per capita household consumption, public infra- structure, and health performance (proxied by the value for MDG 4). As noted, apart from the service argument, these variables all influ- ence the demand side; public infrastructure may also facilitate service supply. The education equations include an additional demand-side argument, wage incentives (measured by the relative wage gain stu- dents would enjoy if, instead of entering the labor market with their current education achievement, they would study enough to climb one notch in the labor market). As the indicator for MDG 2­­universal primary school comple- tion (every child should complete a primary cycle of education)­­the authors use the net (on-time) completion rate, that is, the share of the population in the relevant age cohort that graduates from the primary cycle in the right year. It is computed on the basis of rele- vant entry and graduation rates. For example, for a four-year pri- mary cycle, the value for MDG 2 in year t is the product of the entry rate in t-3 and the graduation rates in t-3, t-2, t-1, and t. Rising completion rates in the primary cycle tend to increase the number of students in subsequent cycles, raising demands on services if quality is to be maintained. With a time lag, education expansion increases the supply of skilled labor in the economy. The data requirements for education and its MDG are more extensive than for the noneducation MDGs. In addition to the infor- mation that is needed to cover the production of services (which is identical), it is necessary to know base-year rates and elasticities for a wider range of outcomes and enrollment numbers in each cycle. General Equilibrium and the Dynamics of MDG Attainment The MDG production functions are integrated in a standard, open- economy CGE model in the tradition that goes back to Dervi¸s, de Melo, and Robinson (1982). The simultaneous determination of MDG achievement, supply and demand of private goods and ser- vices, and factor market equilibrium is a key feature of MAMS. Because MAMS is a general equilibrium model, it accounts for numerous important interactions between the pursuit of the MDGs and economic evolution. Two important such interactions are the economywide impact of additional public spending caused by the MDGs and the impact of AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 295 MDGs on growth. Additional government services needed to reach the MDGs require additional resources--for example, labor, inter- mediate inputs, and investment funding--that compete with other demands in competitive labor, goods and services, and, possibly, loanable funds markets. This may generate substantial wage hikes for skilled labor given the combination of a small supply (especially in low-income countries) and rapid demand expansion. Conversely, as (the bulk of) school graduates enter the labor force as skilled labor, MAMS captures the positive impact of education on the growth potential of the economy. In the loanable funds market of the model, investments in capital for MDG services compete with other investments for available sav- ings. The outcome depends on the mechanisms through which the economy achieves balance between savings and investment. If MDG-related additional public spending is partly financed by foreign resources (grants or loans), the impact on domestic private consumption and investment may be limited or even positive. How- ever, larger inflows of foreign aid tend to generate Dutch disease effects. In the medium to long run, the most important determinant of the size of such effects is the import share of the additional spend- ing that these inflows finance--if it is low, Dutch disease effects tend to be strong.10 In the model, the appreciation of the real exchange rate caused by the inflow of foreign currency provides the incentives required for suppliers to export a smaller share of their output and for demanders to switch from domestic outputs to imports. The resulting increase in the trade deficit is covered by the inflow of for- eign currency brought about by aid. As a complement to foreign resources, MDG strategies are, at least in part, financed with domes- tic resources, either taxes or borrowing. In the model, selected tax rates may be adjusted endogenously to meet targets for government savings or foreign aid. Alternatively, tax rates may adjust in response to changes in fiscal solvency indicators (like the ratio between government debt and GDP), ensuring that these indicators remain unchanged. Of course, the cost of higher taxes is reduced private savings and consumption spending, with a negative impact on growth and on efforts to reduce poverty. The fact that MAMS is a dynamic model makes it possible to take into account that many of the links between MDGs, factor markets, and growth operate with significant time lags. The expan- sion of MDG services may follow different time paths, approaching target levels at constant growth rates or doing so with different degrees of front- or backloading. These lags are particularly impor- tant in modeling progress in education and its impact on the econ- omy. Indeed, the model accounts for the growth and change in the 296 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN age structure of the population, the multiyear duration of the vari- ous education cycles, and the time lags between expansion in the number of students and graduates at low levels of education and changes in the skill structure of the labor force. For example, improved primary school completion rates affect the skill structure of the labor force with considerable delays. The dynamic structure of the model is mostly recursive. The bulk of endogenous decisions of economic agents depends on the past and the present, not the future. However, some features may be nonrecur- sive. For instance, the government's current investment decisions are driven by future decisions on service provision (in health, education, and other areas). In this context, a multiyear simultaneous model solu- tion is preferable to the usual recursive algorithm. Quite important, this approach makes it possible to simulate highly relevant scenarios under which the government endogenously selects growth patterns for government services that are subject to the constraint that certain MDGs be achieved by 2015, while also considering the roles of other determinants of MDG performance. In this case, the government is assumed to have perfect foresight: its decisions in early periods depend on future decisions and the future evolution of the economy. The model structure has been designed to address four broad groups of issues, each of them crucial to the interaction between growth, aid, and MDGs: · The model describes the mechanisms through which service deliv- ery and other determinants of MDG achievements interact, capturing the roles of the demand and supply sides of MDG services. · The model analyzes competition for scarce resources (labor, investment funding, and other goods and services) between MDG ser- vices and other sectors, as well as the role of MDG services in adding to the resources of the economy via the labor market and by promot- ing long-run growth in incomes and investments. · The model captures the impact of alternative foreign aid scenarios on the production of tradable goods (Dutch disease phenomena) and its role in adding to the pool of savings, thereby mitigating resource com- petition between MDG services and other sectors. · The model may be solved simultaneously for the full planning horizon, permitting it to produce future scenarios and analyze the impact of the sequencing of large programs. MDG Strategy Simulations for Ethiopia The preceding discussion shows how MAMS is designed to address key aspects of MDG strategies. This section illustrates some of the features of MAMS through a set of simulations of the evolution of AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 297 the Ethiopian economy. Among other things, these simulations address the following questions: What effects do selected MDG strategies have on MDG indicators, economic growth, exports, the labor market, and the roles of the government and the private sec- tor in the economy? How much does it cost to achieve the MDGs? What roles may synergies among MDGs or between MDGs and the economy have in reducing these costs? How are the effects of MDG strategies influenced by the availability of more or less foreign aid? What kinds of trade-offs may Ethiopian policy makers have to face given limited foreign aid and domestic resource constraints?11 To answer these questions and illustrate key model features, the authors designed a set of simulations with MAMS applied to an Ethiopian database. The first simulation (Base) corresponds to a simple extrapolation of current trends and is used as a benchmark for comparison with other scenarios. In this first simulation, MDGs are not reached by 2015. Conversely, the second scenario is designed to reach the MDGs, with foreign aid filling any financ- ing gap. This simulation, entitled MDG-Base, provides a first indi- cation of the effects of pursuing an MDG strategy, including its costs and the need for foreign resources. Two variants on this sce- nario explore the impact of less foreign aid combined with heavier reliance on domestic financing through direct taxes (MDG-Mix) and improved government productivity (MDG-Gprd). Finally, to explore trade-offs, the authors analyzed the impact of scenarios with less foreign aid and less government spending either on human development or on infrastructure (with MDG-HDcut and MDG-Infcut, referred to in the tables, as two examples). In case resources to reach all the MDGs were not available, this permits calculating what is the cost of reaching a specific MDG in terms of the others. Selected results from these simulations are presented in tables 9.2­9.7 and figures 9.1­9.4. Under MDG-Base, the authors imposed full achievement of the education, health, and water-sanitation MDG targets. The evolu- tion of MDG 1 is monitored using a simple constant-elasticity rela- tionship between the headcount poverty rate and real GDP per capita.12 Foreign aid in grant form is assumed to fill any financing gap.13 This scenario constitutes a strong and extended economic shock. Tables 9.4 and 9.5 show that achievement of these targets requires rapid expansion in the provision of the MDG-related gov- ernment services and therefore rapid expansion of current and cap- ital public spending. In water and sanitation, current public expen- ditures increase at an average annual growth rate of 21 percent, whereas capital spending increases at a rate of 40 percent. Both in primary education and infrastructure, the current and capital growth rates are, respectively, 15 percent and 24 percent. Of course, 298 Table 9.2 Impacts on MDG Indicators MDG- MDG- MDG- MDG- MDG- Base Base Mix Gprd Infcut HDcut Target MDG indicatora 2005 Rate in 2015 1. Headcount poverty rate 33.8 27.8 18.7 22.6 18.6 21.3 18.6 19.2 (percent) 2. First cycle primary net 29.1 48.1 99.9 99.9 99.9 99.9 93.8 100.0 completion rate (percent) 4. Under-five mortality rate 156.2 110.5 68.0 68.0 68.0 67.9 79.0 68.0 (per 1,000 live births) 5. Maternal mortality rate 580.0 387.2 217.5 217.5 217.5 217.2 260.1 217.5 (per 100,000 live births) 7a. Access to safe drinking 24.4 26.4 62.5 62.5 62.5 62.5 59.5 62.5 water (percent) 7b. Access to improved 12.0 14.1 54.0 54.0 54.0 54.0 50.6 54.0 sanitation (percent) Source: World Bank staff simulations with the MAMS model. Note: MDG Millennium Development Goal; Base business-as-usual scenario; MDG-Base core MDG scenario; MDG-Mix MDG scenario with a smaller increase in foreign aid; MDG-Gprd MDG scenario with increased government productivity; MDG-Infcut MDG scenario with reduced spending on infrastructure (human development focus); MDG-HDcut MDG scenario with reduced spending on human development (growth focus). a. The 1990 values are as follows: 38.4 (MDG 1); 24.0 (MDG 2); 204.0 (MDG 4); 870 (MDG 5); 25.0 (MDG 7a); 8.0 (MDG 7b). The targeted changes relative to the 1990 value are as follows: 50 percent reduction (MDG 1); reach 100 percent in 2015 (MDG 2); two-thirds reduction (MDG 4); three- fourths reduction (MDG 5); 50 percent reduction in share without (MDG 7a); and 50 percent reduction in share without (MDG 7b). AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 299 Table 9.3 Impacts on Macroeconomic Indicators (levels in 2005, average annual percent compound growth rate, 2006­15) 2005 MDG- MDG- MDG- MDG- MDG- (US$ Base Base Mix Gprd Infcut HDcut Indicator millions) Real annual growth 2006­15 (percent) Absorptiona 10,153 3.5 8.5 6.5 7.8 7.6 7.9 GDP at market prices 8,528 3.5 5.7 4.8 5.9 5.2 5.7 Private consumption 6,734 3.1 5.4 1.8 5.0 4.7 5.5 Government consumption 1,458 4.0 8.4 8.6 8.4 8.7 6.8 Private investment 942 4.4 8.5 3.0 7.6 7.2 8.3 Government investment 1,019 4.0 20.0 20.3 18.2 17.9 18.4 Exports 1,283 3.7 1.0 1.1 1.0 0.7 0.6 Imports 2,908 3.4 12.8 9.1 10.7 11.2 11.4 GDP at factor cost (total) 7,704 3.6 5.5 4.7 5.5 5.0 5.5 GDP at factor cost (private sector) 7,101 3.5 5.2 4.3 5.3 4.6 5.4 GDP at factor cost (government) 603 4.0 8.7 8.8 8.6 8.9 7.3 Real exchange rate (index) 1.0 0.2 3.4 0.9 1.8 2.9 2.5 2005 (percent) Percentage of GDP in 2015 Absorptiona 119.1 118.6 141.7 137.4 137.6 138.7 138.5 Private consumption 79.0 76.0 72.1 58.6 72.9 71.2 74.3 Government consumption 17.1 17.9 22.1 24.4 19.0 23.9 19.0 Private investment 11.0 12.1 12.0 8.9 12.1 11.5 12.5 Government investment 12.0 12.6 35.5 45.5 33.6 32.0 32.7 Exports 15.0 15.6 5.7 9.7 8.1 6.4 7.3 Imports 34.1 34.2 47.4 47.1 45.7 45.1 45.9 Source: World Bank staff simulations with the MAMS model. Note: MDG-Base core MDG scenario; MDG-Mix MDG scenario with a smaller increase in foreign aid; MDG-Gprd MDG scenario with increased government pro- ductivity; MDG-Infcut MDG scenario with reduced spending on infrastructure (human development focus); MDG-HDcut MDG scenario with reduced spending on human development (growth focus). a. Absorption is the sum of private and public consumption and investment. these rates of growth are also those of inputs in those services, for example, numbers of teachers and classrooms in primary education. As a result of this acceleration in public spending for the MDGs, the GDP share of the government (measured by the sum of government consumption and investment) is almost doubled, increasing from 29.1 percent in 2005 to 57.6 percent in 2015--see table 9.3. 300 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN Table 9.4 Impacts on Government Current Expenditures 2005 MDG- MDG- MDG- MDG- MDG- (US$ Base Base Mix Gprd Infcut HDcut Expenditure millions) Real annual growth 2006­15 (percent) 1st cycle primary education 95.1 4.0 15.6 15.9 15.7 16.4 9.3 2nd cycle primary education 68.0 4.0 12.9 12.9 12.9 12.9 12.9 Secondary education 55.9 4.0 11.2 11.2 11.2 11.2 11.2 Tertiary education 51.3 4.0 13.1 13.1 13.1 13.1 13.1 Low-tech health 22.0 4.0 14.7 15.5 14.7 16.0 8.3 Medium-tech health 31.0 4.0 11.9 12.6 11.9 13.0 6.8 High-tech health 110.7 4.0 15.5 16.4 15.5 16.9 8.8 Water and sanitation 16.3 4.0 21.4 21.9 21.5 21.9 20.4 Public infrastructure 17.0 4.0 15.4 15.4 15.4 12.3 15.4 Other government 990.8 4.0 4.0 4.0 4.0 4.0 4.0 2005 (percent) Percentage of GDP in 2015 1st cycle primary education 1.1 1.1 2.7 2.8 2.2 3.0 1.5 2nd cycle primary education 0.8 0.8 1.5 1.6 1.2 1.6 1.5 Secondary education 0.7 0.8 1.2 1.2 0.9 1.2 1.1 Tertiary education 0.6 0.7 1.2 1.3 1.0 1.3 1.2 Low-tech health 0.3 0.3 0.6 0.7 0.5 0.7 0.3 Medium-tech health 0.4 0.4 0.6 0.7 0.5 0.7 0.4 High-tech health 1.3 1.4 3.0 3.7 2.7 3.6 1.7 Water and sanitation 0.2 0.2 0.8 0.9 0.7 0.9 0.7 Public infrastructure 0.2 0.2 0.5 0.5 0.4 0.4 0.5 Other government 11.6 12.2 10.0 10.9 8.8 10.6 10.1 Domestic interest payments 0.2 0.7 0.5 0.6 0.5 0.5 0.5 Foreign interest payments 0.8 1.1 0.6 0.9 0.7 0.7 0.7 Total recurrent public spending 18.2 19.6 23.2 25.9 20.3 25.2 20.2 Source: World Bank staff simulations with the MAMS model. Note: MDG-Base core MDG scenario; MDG-Mix MDG scenario with a smaller increase in foreign aid; MDG-Gprd MDG scenario with increased govern- ment productivity; MDG-Infcut MDG scenario with reduced spending on infra- structure (human development focus); MDG-HDcut MDG scenario with reduced spending on human development (growth focus). Compared with the Base scenario, annual real GDP growth under MDG-Base accelerates strongly for government activities (from 4 per- cent to 8.7 percent) and more moderately but yet substantially for the private sector (from 3.5 percent to 5.2 percent; see table 9.3). In com- parison with the Base simulation, the present value (PV) of total AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 301 Table 9.5 Impacts on Government Investment Expenditures MDG- MDG- MDG- MDG- MDG- 2005 Base Base Mix Gprd Infcut HDcut (US$ Expenditure millions) Real annual growth 2006­15 (percent) 1st cycle primary education 18.6 4.0 24.1 24.7 24.3 25.6 12.4 2nd cycle primary education 13.6 4.0 24.4 24.4 24.4 24.4 24.4 Secondary education 26.6 4.0 21.1 21.1 21.1 21.1 21.1 Tertiary education 36.7 4.0 24.9 24.9 24.9 24.9 24.9 Low-tech health 16.3 4.0 28.0 29.5 28.0 30.4 14.9 Medium-tech health 23.9 4.0 22.6 23.8 22.6 24.6 11.3 High-tech health 45.3 4.0 29.6 31.1 29.5 32.1 15.9 Water and sanitation 15.4 4.0 40.4 41.1 40.5 41.1 38.6 Public infrastructure 378.4 4.0 24.6 24.6 24.6 18.9 24.6 Other government 444.3 4.0 4.0 4.0 4.0 4.0 4.0 2005 (percent) Percentage of GDP in 2015 1st cycle primary education 0.2 0.2 0.9 1.2 0.9 1.1 0.4 2nd cycle primary education 0.2 0.2 0.7 0.9 0.6 0.7 0.7 Secondary education 0.3 0.3 1.0 1.3 1.0 1.1 1.1 Tertiary education 0.4 0.5 1.9 2.4 1.8 2.1 2.0 Low-tech health 0.2 0.2 1.1 1.5 1.0 1.4 0.4 Medium-tech health 0.3 0.3 1.0 1.4 1.0 1.3 0.4 High-tech health 0.5 0.6 3.4 4.8 3.2 4.4 1.2 Water and sanitation 0.2 0.2 2.6 3.4 2.5 2.9 2.4 Public infrastructure 4.4 4.7 19.2 24.0 18.2 13.0 20.3 Other government 5.2 5.5 3.7 4.6 3.5 4.0 3.9 Total public capital spending 12.0 12.6 35.5 45.5 33.6 32.0 32.7 Source: World Bank staff simulations with the MAMS model. Note: MDG-Base core MDG scenario; MDG Mix MDG scenario with a smaller increase in foreign aid; MDG-Gprd MDG scenario with increased govern- ment productivity; MDG-Infcut MDG scenario with reduced spending on infra- structure (human development focus); MDG-HDcut MDG scenario with reduced spending on human development (growth focus). foreign aid over the 2006­15 period is more than quadrupled, reach- ing US$31 billion. In the final year, 2015, foreign aid is US$81 per capita or 37 percent of GDP (table 9.6). This huge inflow of foreign aid causes the onset of Dutch disease, which manifests itself in an appreciation of the real exchange rate, depressed exports, and larger 302 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN Table 9.6 Impacts on Government Revenues (as share of GDP) MDG- MDG- MDG- MDG- MDG- Base Base Mix Gprd Infcut HDcut 2005 Revenue (percent) Percentage of GDP in 2015 Direct taxes 6.3 6.0 5.8 24.8 5.8 7.9 3.6 Import taxes 6.4 6.2 7.0 6.3 6.8 6.9 6.9 Other indirect taxes 3.3 6.2 5.9 6.0 6.1 5.9 6.0 Central bank borrowing 1.2 1.4 1.1 1.2 1.1 1.1 1.1 Other domestic borrowing 2.0 2.5 1.9 2.2 2.0 2.0 2.0 Foreign borrowing 5.8 4.2 2.4 3.4 2.8 2.7 2.7 Foreign grants 5.1 5.7 34.6 27.4 29.2 30.7 30.7 Net other capital inflows and errors 0 0 0 0 0 0 0 Total 30.1 32.2 58.7 71.4 53.8 57.2 53.0 Memorandum items Total public expenditures 30.1 32.2 58.7 71.4 53.8 57.2 53.0 Foreign aid per capita (US$)a 16.2 18.5 80.8 51.4 61.9 67.5 67.5 Present value (PV) of foreign aid (2005 US$ billions) 6.9 31.4 20.2 26.5 26.7 26.7 Source: World Bank staff simulations with the MAMS model. Note: MDG-Base core MDG scenario; MDG-Mix MDG scenario with a smaller increase in foreign aid; MDG-Gprd MDG scenario with increased govern- ment productivity; MDG-Infcut MDG scenario with reduced spending on infra- structure (human development focus); MDG-HDcut MDG scenario with reduced spending on human development (growth focus). a. Foreign aid per capita includes an allowance for aid outside the government bud- get. In per capita terms, aid in the government budget was around US$11 in 2005. imports, thus allowing the economy to fully use the foreign currency inflow that comes with foreign aid. Figure 9.1 shows the expansion of foreign aid per capita in the various scenarios. It can be seen that it increases monotonically, except for a decline in 2011. The decline reflects two factors. First, the period of big investments--in schools and teacher training--to support rapid expansion in primary education comes to an end, reducing government spending needs. Second, the model captures an Ethiopia-specific threshold effect based on expert assessments. Private sector productivity is boosted because the public infrastruc- ture capital stock exceeds a threshold above which productivity- enhancing network effects are triggered in the private sector. AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 303 Figure 9.1 Foreign Aid per Capita 90 80 70 60 US$ 50 2005 40 30 20 10 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 Base MDG-Base MDG-Mix MDG-Gprd Source: World Bank staff simulations with the MAMS model. Note: MDG-Base core MDG scenario; MDG-Mix MDG scenario with a smaller increase in foreign aid; MDG-Gprd MDG scenario with increased government productivity. Part of the huge increase in public spending in pursuit of MDGs is due to changes in unit costs caused by the increase in the demand of several types of goods and services. Of particular importance are wage developments, which depend on what happens in the educa- tion system (influencing supply) and government services (driving demand changes). Labor supply growth by workers with little or no education (the bulk of the labor force) declines, given that an increasing share of the children--by 2012, close to all--pass pri- mary school, with many continuing beyond this level (see table 9.7). As a result, GDP growth is affected negatively in this first stage. For the more educated (but much smaller) segments of the labor force, supply growth accelerates gradually as more students graduate from higher cycles. Demand for more educated labor in government services grows quickly throughout most of the simulation period, especially up to 2012 (the year in which everyone in the primary- school cohort has to start the cycle and, after this, manage to successfully proceed through the different grades, graduating in 2015). Conversely, demand growth for this type of labor in the private sector is relatively steady. The combined impact of these demand- and supply-side changes is relatively rapid wage growth for the least educated throughout the period (albeit starting and remaining at a low level). For the two more educated groups, wages grow rapidly until around 2012 and, after this, start to decline as 304 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN Table 9.7 Impacts on Labor and Capital (levels in 2005, average annual percent compound growth rate, 2006­15) MDG- MDG- MDG- MDG- MDG- 2005 birr Base Base Mix Gprd Infcut HDcut Wages and return per year to capital (thousands) Nominal annual growth 2006­15 (percent) Labor ( secondary education) 0.8 0.4 3.9 2.4 3.4 3.2 3.4 Labor (secondary education) 2.1 0.7 1.6 0.2 0.6 1.1 1.1 Labor (tertiary education) 9.6 2.2 3.4 2.2 1.8 3.3 2.7 Private capital 2.7 0.4 0.4 1.1 0.4 0.1 0.6 2005 birr per year Factor quantities (millions) Real annual growth 2006­15 (percent) Labor ( secondary education) 29.8 3.5 1.9 1.9 1.9 1.9 2.2 Labor (secondary education) 2.3 3.9 5.0 4.9 5.0 4.9 4.8 Labor (tertiary education) 0.2 1.9 4.4 4.2 4.3 4.2 4.4 Private capitala 76.7 3.5 4.9 2.9 4.7 4.6 4.8 ICOR 3.7 6.8 7.6 6.7 6.5 6.2 Source: World Bank staff simulations with the MAMS model. Note: ICOR incremental capital-output ratio; MDG-Base core MDG scenario; MDG-Mix MDG scenario with a smaller increase in foreign aid; MDG-Gprd MDG scenario with increased government productivity; MDG-Infcut MDG scenario with reduced spending on infrastructure (human development focus); MDG-HDcut MDG scenario with reduced spending on human development (growth focus). a. Private capital units billions of constant 2002 birr. the supply of skilled workers starts to accelerate. Comparing MDG-Base to Base, private sector employment expansion for more educated labor is minor (given competition from the government), whereas its employment contraction for the least educated labor type is similar to that of the government. Private sector GDP growth under MDG-Base is boosted by more rapid productivity growth. Figure 9.2 shows the evolution of wages for the segment that has completed secondary but not tertiary education, for MDG- Base and other simulations. In this analysis, the costs of achieving the MDGs are influenced by the fact that MDG achievements do not depend only on the supply of relevant services, but also on progress in terms of a set of other determinants: other MDGs, availability of public infrastruc- ture, household consumption per capita, and wage premia (influ- encing education decisions). To assess the role that such "synergies" AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 305 Figure 9.2 Real Wages of Labor with Secondary Education 2.7 2.6 2.5 2.4 2.3 (thousands) 2.2 birr 2.1 2002 2.0 1.9 1.8 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 Base MDG-Base MDG-Mix MDG-Gprd Source: World Bank staff simulations with the MAMS model. Note: MDG-Base core MDG scenario; MDG-Mix MDG scenario with a smaller increase in foreign aid; MDG-Gprd MDG scenario with increased government productivity. can play in influencing costs, the authors compared the costs for MDG-Base with the costs of primary education, health, and water sanitation in three separate scenarios (where MDGs were achieved in each of these areas separately).14 The results--not reported here-- indicate that the present value of total costs in these three areas is 22 percent higher when the MDGs are pursued separately compared with the costs for MDG-Base, where they are pursued simultane- ously. The differences are primarily due to savings in health. This result suggests that bottom-up costing exercises that do not consider the economywide context of MDG strategies may be misleading, often overestimating the costs. The scenario MDG-Base looks unfeasible. It is unlikely that donors will be willing to provide foreign aid in the required amounts. Moreover, such a huge expansion of foreign aid and the government GDP share most likely would generate severe gover- nance problems.15 Given this, alternatives need to be considered. The next scenario, MDG-Mix, considers one alternative. It has been constructed to address the following question: in a setting with less foreign aid, what would be the consequences of pursuing the same MDG targets (in health, education, and water sanitation) and main- taining the same real growth in other areas of government spending (including infrastructure)? MDG-Mix is identical to MDG-Base 306 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN Figure 9.3 Present Value of Foreign Aid 35 30 25 (billion) 20 US$ 15 2005 10 5 0 Base MDG-Base MDG-Mix MDG-Gprd Source: World Bank staff simulations with the MAMS model. Note: MDG-Base core MDG scenario; MDG-Mix MDG scenario with a smaller increase in foreign aid; MDG-Gprd MDG scenario with increased government productivity. except for the fact that the increase in foreign grant aid relative to the base scenario is only half as large; in per capita terms, foreign aid reaches US$51 in 2015 (see table 9.6). The PV of total foreign aid in 2006­15 falls from US$31.4 billion to US$20.2 billion (see figure 9.3 and table 9.6). As a result of reduced foreign aid, the appreciation of the real exchange rate is less pronounced, whereas export growth increases and import growth slows down. Direct tax collection adjusts to ensure that government receipts are sufficient to cover government spending. The direct tax increase is huge, going from a share of GDP of 6.3 percent in 2005 to close to 25 percent in 2015. Such an increase has a strong dampening impact on growth in household factor incomes, consumption, savings, and invest- ments, resulting in slower growth in the private capital stock and private GDP, the latter falling from 5.2 percent under MDG-Base to 4.3 percent under MDG-Mix. Government demand (the sum of gov- ernment consumption and investment) reaches almost 70 percent of GDP, exceeding the highest share in the world.16 As a result of slower GDP growth, the MDG target for poverty reduction is not met. Compared with MDG-Base, more rapid growth is needed in government spending in education, health, and water-sanitation ser- vices to achieve the MDG targets. This spending is required because of slower growth in per capita household consumption, that is, a source of weaker synergy effects, influencing the demand side for different government services. AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 307 Although the scenario MDG-Mix has a more realistic outcome for foreign aid, it has the drawbacks of reducing private and overall GDP growth, achieving only a subset of the MDGs (MDG 1 is far from being reached) and generating an even larger government share in GDP. To explore the potential for government productivity in facili- tating progress toward the MDGs, the authors constructed a sec- ond alternative scenario (MDG-Gprd) that has more rapid govern- ment productivity growth but otherwise is identical to MDG-Base. Under MDG-Gprd, the productivity of government labor and inter- mediate input use improves by an additional 1.5 percent per year, and government investment efficiency grows at the same annual rate.17 Compared with MDG-Base, the results include noteworthy declines in foreign aid needs (to US$26.5 billion; US$61.9 per capita in 2015) and declines in the GDP share for the government (to 52.6 percent). The deterioration in terms of poverty reduction, private consumption growth, and GDP growth is minor. In an addi- tional simulation, not reported elsewhere in this chapter, the authors let the productivity improvement of the government be doubled, to 3 percent per year. The result is a further strengthening of these outcomes: the PV of aid declines to US$22.7 billion and the govern- ment GDP share in 2015 falls to 45.7 percent, without any signifi- cant impact on poverty reduction. Although these scenarios high- light the importance of improving government efficiency, such efficiency gains may be particularly difficult to bring about in the context of rapid government expansion. These simulations exemplify the type of questions that MAMS can address under scenarios that achieve MDG targets. They sug- gest that in the face of constraints (on foreign aid, domestic resources, and the scope for productivity improvement), the gov- ernment may have to confront difficult trade-offs, adjusting down- ward the MDG targets it strives to achieve by 2015. If these targets are not adjusted, taxes and government spending may become exces- sive or unreasonable relative to the total size of the economy, with a negative impact on private sector development and household consumption. The remaining scenarios analyze trade-offs between spending on infrastructure and human development in a setting with reductions in foreign aid relative to MDG-Base. Under the scenario MDG- HDcut, the government receives 85 percent of the aid under MDG- Base. It maintains its spending on infrastructure and cuts spending on HD MDGs (here defined to include primary education, health, and water sanitation).18 Compared with MDG-Base, GDP growth is virtually unchanged. The required reduction in domestic final 308 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN demand (driven by the fact that with less foreign aid, the country has to live with a smaller trade deficit) is spread quite evenly across pri- vate and government consumption and investment. The key result is that, for the HD MDGs, the country achieves 91.6 percent of the increase required to meet the MDGs; for the poverty objective, how- ever, 102.6 percent of the required drop is achieved (that is, a slight overachievement). Conversely, if the country maintains a 100 percent achievement rate for its HD MDGs while cutting spending on infra- structure (the scenario MDG-Infcut), 89.3 percent of the required fall-in-poverty MDG is achieved. The driving force behind this out- come is that, during the simulation period, spending on infrastruc- ture has a considerably stronger impact on GDP growth than spend- ing in the HD area--annual GDP growth for MDG-HDcut is close to MDG-Base. For MDG-Infcut, a significant slowdown occurs (by 0.5 percent in annual growth), partly because of the lag in the effect of human development on growth. Figure 9.4 provides a broader perspective on trade-offs between human development and poverty reduction in the face of foreign aid constraints. It summarizes trade- offs for a larger set of simulations with alternative cuts in foreign aid--the simulations along each curve have identical levels of foreign aid in the final year and, in PV terms, for the period 2006­15. Figure 9.4 Trade-Offs between Human Development and Poverty Reduction 100 PV aid 100% PV aid 90% 95 (%) PV aid 85% 90 target HD 85 PV aid 80% of share 80 PV aid 75% 75 70 80 90 100 share of poverty target (%) Source: World Bank staff simulations with the MAMS model. Note: HD human development; PV present value. Along a given curve, the present value of aid is kept constant. The point at the upper right corner of the graph (PV aid 100%) corresponds to the core MDG simulation (MDG-Base). The results for MDG-Base are as follows: the HD target is achieved to 100 percent; the poverty target is achieved to 102.8 percent; and the PV of aid 2006­15 US$31.4 billion. AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 309 Conclusion This chapter has described the design and application of the MAMS model, intended to focus on strategies and trade-offs related to efforts to achieve the MDGs over the next decade. The scenarios presented exemplify the use of MAMS. Other examples of issues that MAMS can address include the effects of front- or backloaded increases in government MDG services, reallocation of government spending from unproductive areas, foreign debt forgiveness, and alternative allocations of resources between government and the private sector in the context of a fixed foreign aid envelope. In settings less focused on full achievement of the MDGs (for example, growth analysis), MAMS provides results for MDG indicators along with more stan- dard economic indicators, making it possible to maintain a focus on poverty and human development. This tool can simulate the effects of the level and, more important, the structure of public spending, using a level of disaggregation rarely found in economywide model- ing--see tables 9.4­9.6. One virtue of the MAMS framework is that it provides a compre- hensive perspective on the MDG challenge through its representation of the entire economy. The analysis highlights the fact that changes in wages and exchange rates influence domestic relative prices and the domestic purchasing power of foreign aid, thereby invalidating sim- plistic costing and aid-forecasting exercises. Unlike such exercises and strict sectoral approaches, MAMS facilitates an examination of how the different goals complement one another, while at times competing for resources. Moreover, by focusing explicitly on the goals them- selves, rather than simply on more resources, MAMS supports efforts to move away from traditional reliance on measuring "inputs" (such as teachers hired or foreign aid received) to measuring "outcomes" (the goals themselves). This, in turn, encourages greater attention to the consideration of the appropriate sequencing of resources, priori- ties, and policies to reach the MDG targets. Currently, the MAMS framework is being applied to 6 countries in Africa and 18 countries in Latin America (in a project managed by the UNDP). This broad application suggests that it is a valuable tool for strategy analysis in a wide range of countries, not only in low-income countries, for which it was initially designed and for which the economywide interactions between development, exter- nal aid, and the MDGs are the strongest. The MAMS framework has particular operational appeal for the World Bank. The government of Ethiopia has drawn on results from MAMS in its MDG strategy document (FDRE 2005a, 2005b). The World Bank is drawing on MAMS simulations in its country-level 310 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN dialogue on Ethiopia's PRS as well as in ongoing studies on aid, labor, and population. Similarly, the International Monetary Fund (IMF) uses results from MAMS in the formulation of its MDG scenarios for Ethiopia (IMF 2006). MAMS has also provided inputs to several recent World Bank and IMF documents on MDGs, aid, fiscal policy, and growth (World Bank and IMF 2004, 2006; Patillo, Gupta, and Carey 2005; World Bank 2005b). In dialogues with the government of Ethiopia and other partners inside and outside the World Bank, the authors have found that issues related to labor and education, synergies, long-run macro issues (including growth and financing), and trade-offs between human development and infrastructure are of particular interest. They have also learned that it is important to view model-based analysis and the development of a multipurpose database as part of an ongoing process based on multiple tools--in this case sub- stituting a simpler macro framework for MAMS when the micro foundations of scenarios were not available to the degree of detail required for MAMS. As more countries move ahead with ambitious PRSs that are built around accelerating progress toward achieving the MDGs, availabil- ity of an operational tool to integrate detailed sector analysis within an economywide framework to capture the interactions between and trade-offs among MDG-related and other expenditures is invaluable. Properly used, MAMS can enrich the dialogue among the different partners in the development community to establish coherent long- term strategies for achievement of the MDGs. Notes The authors acknowledge the initial inspiration of Luiz Pereira da Silva and significant contributions from Maurizio Bussolo, Ahmed Kamaly, Jeff Lewis, Hans Timmer, and Dominique van der Mensbrugghe, as well as the assistance provided by Denis Medvedev and Shuo Tan. Comments and suggestions have been provided by Ishac Diwan, Enrique Ganuza (United Nations), Pablo Gottret, Wafik Grais, Jee Peng, Sherman Robinson (University of Sussex), Agnès Soucat, and Rob Vos (United Nations). The views and findings in this paper are those of the authors and not necessarily those of the World Bank, its executive board, or member country governments. 1. The SimSIP tools and related documentation can be downloaded from www.worldbank.org. 2. Agénor, Bayraktar, and El Aynaoui (2005) include more detail on the macro model. 3. MAMS is compatible with any standard treatment of economy- wide modeling of poverty, including representative household approaches, AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 311 micro simulation, and more simple relationships based on a constant elas- ticity of the poverty rate with respect to GDP or household consumption per capita. 4. This presentation is simplified, highlighting key mechanisms. For a detailed technical documentation of MAMS, see Lofgren and Díaz-Bonilla (2006). 5. The national accounts rarely ascribe value added to government capital--by accounting conventions, only labor creates value added in the government sector--making it impossible to derive coefficients of the MDG production functions from value-added shares as is standard practice in the calibration of private sector production functions. The assumption of Leontief production functions made earlier is justified by that difficulty, as well as by the lack of information about the substitutability between capital and labor in these service sectors. 6. For each argument, these "full" elasticities are the product of two elasticities: the elasticity of the MDG indicator with respect to Zk--equation (9.5)--and the elasticity of Zk with respect to the argument in the constant- elasticity function ( and )--equation (9.6). k ik 7. For examples of the literature on health that support the statements in this paragraph, see Glewwe and Jacoby (1995), Lavy and others (1996), Anand and Bärnighausen (2004), and Baldacci and others (2004). Similarly, the authors' statements on education draw on Anand and Ravallion (1993), Deolalikar (1998), Case and Deaton (1998), Mingat and Tan (1998), Baldacci and others (2004), and World Bank (2005a). For more details, see Kamaly (2006). 8. A simultaneous-equation model can be solved to generate the values of , , and that permit the logistic function to (1) replicate base-year k k k MDGk; (2) have an inflection point at a specified distance relative to the ini- tial Zk; and (3) exactly achieve the MDG for the value of Zk, which is defined by the specified MDG scenario. The preceding scenario assumes that the user relies on exogenous values and k ik(elasticities of Zk with respect to Qk and Dik). Alternatively, if the user wants to impose the "full" base-year elasticities of MDGk with respect to Qk and Dik, then the model has to be extended in two ways: (1) in one set of new equations, these elas- ticities are imposed and, at the same time, the parameters and are k ik endogenized; and (2) in a second set of equations, Zk is defined as a con- stant-elasticity function of Qk, Dik, , and . It is then no longer possible k ik to impose a value a priori for Zk because its value depends on and , k ik which now are endogenous. The prespecified scenario is only one out of an infinite number of scenarios that generate the same MDGk value in 2015. For example, in simulations targeting the MDGs, the actual need for ser- vices, Qk, will vary depending on the evolution of the other arguments, Dik, in equation (9.6). 9. For the first grade of primary school, the qualified population includes everyone in the relevant age cohort (often those who are six years 312 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN old, but this may vary across countries). For the first grades of secondary and tertiary education, those qualified include those who graduated from the preceding cycle in the previous year. In addition, any cycle can have additional entrants (most important, slightly older students who start pri- mary school but also potentially other entrants from outside the school sys- tem, such as migrants from abroad). 10. For an analysis of Dutch disease effects of foreign aid, see Heller (2005). 11. Different aspects of Ethiopia's MDG strategy are explored in Lofgren and Díaz-Bonilla (2005) and in Sundberg and Lofgren (2006). 12. GDP per capita was preferred to household consumption per capita given that GDP is much less influenced by the level of foreign aid in a given year, providing a better indicator of the long-run capacity of the economy to sustain a flow of household consumption. 13. For the sake of clarity, note that for the BASE scenario (and, unless otherwise noted, all other scenarios), the following variables clear the three macro balances: (1) the government balance--foreign grants; (2) the balance of payments--the real exchange rate; and (3) the saving-investment balance-- private investment. The domestic consumer price index is the model numéraire. (The limited changes in these rules for other scenarios are indi- cated below.) 14. In other words, the cost of primary education was defined as the cost of government spending in this area when only MDG 2 was targeted, the cost of health on the basis of government health spending in a simulation in which only MDGs 4 and 5 were targeted, and the cost of water and sanita- tion on the basis of government costs in this area when MDGs 7a and 7b were targeted. 15. Given the large trade deficit, which permits absorption (total domes- tic final demand) to reach 142 percent of GDP, there is a big difference between the government share in absorption (around 41 percent) and its share in GDP (around 58 percent). The same observation applies to the other MDG scenarios. 16. In 2002, the most recent year with a comprehensive data set, the largest GDP share for the sum of government consumption and investment in any developing country was 65.5 percent (for Eritrea). Few countries exceeded 40 percent. Note, moreover, that this percentage does not include major redis- tribution schemes (like pay-as-you-go pension systems or health insurance) as in the countries with the highest share of public spending over GDP. 17. For MDG-Base, the rates of total factor productivity growth are 1.1 percent for the government (only for labor) and private health services and 1.9 percent for the rest of the private sector. 18. In terms of macro closure rules, for the scenarios analyzing spend- ing trade-offs, the government budget is cleared by adjustments in a selected spending area, not by adjustments in foreign grants. AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 313 References Agénor, Pierre-Richard, Nihal Bayraktar, and Karim El Aynaoui. 2005. "Roads out of Poverty? Assessing the Links between Aid, Public Invest- ment, Growth, and Poverty Reduction." Policy Research Working Paper No. 3490, World Bank, Washington, DC. Agénor, Pierre-Richard, Nihal Bayraktar, Emmanuel Pinto Moreira, and Karim El Aynaoui. 2005. "Achieving the Millennium Development Goals in Sub-Saharan Africa: A Macroeconomic Monitoring Framework." Policy Research Working Paper No. 3750, World Bank, Washington, DC. Anand, Sudhir, and Till Bärnighausen. 2004. "Human Resources and Health Outcomes: Cross-country Econometric Study." The Lancet 364: 1603­609. Anand, Sudhir, and Martin Ravallion. 1993. "Human Development in Poor Countries: On the Role of Public Services." Journal of Economic Perspectives 7 (1): 135­50. Baldacci, Emanuele, Benedict Clements, Sanjeev Gupta, and Qiang Cui. 2004. "Social Spending, Human Capital, and Growth in Developing Countries: Implications for Achieving the MDGs." IMF Working Paper No. 04/217, International Monetary Fund, Washington, DC. Case, Anne C., and Angus Deaton. 1998. "School Quality and Educational Outcomes in South Africa." Paper 184, Princeton, Woodrow Wilson School, Development Studies, New Jersey. Christiaensen, Luc, Chris Scott, and Quentin Wodon. 2002. "Development Targets and Costs." In A Sourcebook for Poverty Reduction Strategies, ed. Jeni Klugman. Vol. 1. Washington, DC: World Bank. Clemens, Michael A., Charles J. Kenny, and Todd J. Moss. 2004. "The Trouble with the MDGs: Confronting Expectations of Aid and Develop- ment Success." Working Paper No. 40, Center for Global Development, Washington, DC. Deolalikar, Anil B. 1998. "Increasing School Quantity vs. Quality in Kenya: Impact on Children from Low- and High-Income Households." Journal of Policy Reform 2 (3): 223­46. Dervis¸, Kemal, Jaime de Melo, and Sherman Robinson. 1982. General Equilibrium Models for Development Policy. New York: Cambridge University Press. Devarajan, Shantayanan, Margaret J. Miller, and Eric V. Swanson. 2002. "Goals for Development: History, Prospects, and Costs." Policy Research Working Paper No. 2819, World Bank, Washington, DC. FDRE (Federal Democratic Republic of Ethiopia). 2005a. Ethiopia: Building on Progress: A Plan for Accelerated and Sustained Develop- ment to End Poverty (PASDEP) (2005/06­2009/10). Addis Ababa: Ministry of Finance and Economic Development. 314 BOURGUIGNON, DÍAZ-BONILLA, AND LOFGREN ------. 2005b. Ethiopia: The Millennium Development Goals (MDGs) Needs Assessment Synthesis Report. Addis Ababa: Ministry of Finance and Economic Development. Glewwe, Paul, and Hanan Jacoby. 1995. "An Economic Analysis of Delayed Primary School Enrollment and Childhood Malnutrition in a Low Income Country." Review of Economics and Statistics 77 (1): 156­69. Heller, Peter S. 2005. "`Pity the Finance Minister': Issues in Managing a Substantial Scaling Up of Aid Flows." Working Paper No. 05/180, Inter- national Monetary Fund, Washington, DC. IMF (International Monetary Fund). 2006. "The Federal Democratic Repub- lic of Ethiopia: Selected Issues and Statistical Appendix." IMF Country Report No. 06/122, International Monetary Fund, Washington, DC. Kamaly, Ahmed. 2006. An Econometric Analysis of the Determinants of Health and Education Outcomes in Sub-Saharan Africa. Washington, DC: World Bank. Lavy, Victor, John Strauss, Duncan Thomas, and Philippe de Vreyer. 1996. "Quality of Health Care, Survival and Health Outcomes in Ghana." Journal of Health Economics 15 (3): 333­57. Lofgren, Hans, and Carolina Díaz-Bonilla. 2005. "An Ethiopian Strategy for Achieving the Millennium Development Goals: Simulations with the MAMS Model." Draft paper, Development Prospects Group, World Bank, Washington, DC. ------. 2006. "MAMS: An Economywide Model for Analysis of MDG Country Strategies." Technical documentation, Development Prospects Group, World Bank, Washington, DC. Mingat, Alain, and Jee-Peng Tan. 1998. "The Mechanics of Progress in Education." Policy Research Working Paper No. 2015, World Bank, Washington, DC. Patillo, Catherine, Sanjeev Gupta, and Kevin Carey. 2005. "Sustaining Growth Accelerations and Pro-Poor Growth in Africa." IMF Working Paper No. WP/05/195, International Monetary Fund, Washington, DC. Reddy, Sanjay, and Antoine Heuty. 2004. "Achieving the MDGs: A Critique and a Strategy." Harvard Center for Population and Development Studies Working Paper Series 14 (3). Sundberg, Mark, and Hans Lofgren. 2006. "Absorptive Capacity and Achiev- ing the MDGs: The Case of Ethiopia." Chapter 6 in The Macroeco- nomic Management of Foreign Aid: Opportunities and Pitfalls, eds. Peter Isard, Leslie Lipschitz, Alexandros Mourmouras, and Boriana Yontcheva. Washington, DC: International Monetary Fund. United Nations Development Programme. 2005. Human Development Report 2005: International Cooperation at a Crossroads: Aid, Trade and Security in an Unequal World. New York: Oxford University Press. AID, SERVICE DELIVERY, AND MILLENNIUM DEVELOPMENT GOALS 315 United Nations Millennium Project. 2005. Investing in Development: A Practical Plan to Achieve the Millennium Development Goals. New York: United Nations Development Program, Earthscan. Vandemoortele, Jan, and Rathin Roy. 2004. "Making Sense of MDG Costing." Poverty Group, United Nations Development Pro- gramme, New York. http://www.undp.org/poverty/docs/prm/ MakingsenseofMDGcosting-August.pdf. World Bank. 2005a. "Education in Ethiopia: Strengthening the Foundation for Sustainable Progress." Human Development Department (AFTH3), February 28, World Bank, Washington, DC. ------. 2005b. Global Monitoring Report: Millennium Development Goals: From Consensus to Momentum. Washington, DC: World Bank. World Bank and the International Monetary Fund. 2004. "Aid Effective- ness and Financing Modalities." Background Paper for the October 2 Development Committee Meeting, Washington, DC. ------. 2006. "Fiscal Policy for Growth and Development: An Interim Report." Background Paper for the April 23 Development Committee Meeting, Washington, DC. 10 Conclusion: Remaining Important Issues in Macro-Micro Modeling François Bourguignon, Maurizio Bussolo, and Luiz A. Pereira da Silva The benefits of a stable macroeconomic environment are undisputed and, for many developing countries, the main challenges for macro policies have shifted from a stabilization phase--where key objectives included balanced budgets, moderate to low inflation, and sustain- able positions for the current account and government debt--to a poststabilization phase. In this new phase, governments are engaged in an effort to improve the efficiency and quality of public spending, taxation, and economic management. Policy makers are still con- cerned with macro stability, but they also need to consider, for exam- ple, the fiscal implications of scaling up pro-poor interventions such as conditional cash transfers; or reconcile the tension between the need of additional external resources (through foreign aid, interna- tional remittances, or commodities exports) and the ensuing pressure on the real exchange rate; or regulate imperfect markets while avoid- ing excessive red tape and governance issues. Accelerating growth and, at the same time, fighting poverty and unequal access to oppor- tunities are the main goals of this new phase. Evaluating the policies of this new phase in terms of their contribution toward the attain- ment of these goals is a difficult exercise, but demand for these evalu- ations from policy makers, practitioners, and researchers is rapidly increasing. This volume is a response to this growing demand. 317 318 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA The studies in this volume provide clear illustrations of the impor- tant advantages that come from adopting a macro-micro integrated approach to evaluate the impact of macroeconomic policies on poverty and income distribution and the macro effects of micro poli- cies directed toward reducing poverty and inequality. In fact, this integrated approach is the only one that can overcome such crucial difficulties as creating macro counterfactuals or accounting for the macro effects of scaling up micro interventions. It is also the only one that permits determining with some precision who are the win- ners and losers in a reform at the macroeconomic level and what are the macro and second round effects of micro policies directed toward specific agents or sectors. In a broad perspective, this macro- micro approach also represents the "natural" methodological approach to assess the contributions of growth and distribution to the development process. The preceding chapters warned that linking a macro model with a household-level micro model, irrespective of the degree of integra- tion between the macro and micro parts, could be challenging. The authors of the studies in this book demonstrated various ways to overcome this challenge, but major difficulties as well as under- researched areas remain. These concluding remarks highlight some of these challenges. The first difficulty any empirical method faces is data quality. In the context of a macro-micro modeling framework, this issue is com- pounded by data reconciliation. Most developing countries now have the technical capacity to gather and document national accounts sta- tistics, and these statistics--along with ancillary data from central banks, customs authorities, and other agencies--usually provide a fair, if not always accurate and timely, macro picture for the economy. Many topics, such as better measurement of employment by skill level or capital stocks, still require attention. Alongside these developments in macro data availability and quality, collection of micro data, mainly in the format of household surveys, has become more and more common for many countries. Efforts by national agencies and interna- tional organizations have improved the quality and thematic coverage of these surveys. An outstanding issue is that of generating more panel data sets--that is, linking consecutive surveys so that households and individuals can be followed through time. Panel data would be espe- cially useful in estimating intertemporal behavior and thus would help with building more detailed dynamic models. Panel data could also facilitate the validation of many of the macro-micro techniques described in this volume,1 even though most of these techniques do not aim at mimicking panel data (in the sense that these techniques do not focus on tracing the history of specific households). CONCLUSION 319 Notwithstanding these positive developments, a systematic attempt to reconcile micro and macro data is missing in most if not all countries. Obvious differences in definition aside, measurements for the same aggregate from two sources should be reconciled. As mentioned in the introduction, the availability of macro and micro data that are in synch not only is a requirement for the construction of a consistent quantitative model, but also can directly benefit policy making. The well-known case of private consumption in India (also cited in the introduction) is exemplary in this sense: consump- tion growth and poverty reduction rates calculated from the surveys appear to be much slower than the same rates estimated from national accounts. And so supporters of additional market-friendly reforms of the Indian economy appeal to the positive results from the national accounts, whereas opponents of the reforms use the sluggish poverty reduction shown in the surveys as a proof against the recent or further liberalizations. The second major challenge is better modeling of growth or, more generally, the dynamics of economic systems. Dynamic macro-micro modeling largely remains comparisons of two cross-sections of households in different states of the economy at two points in time, under the implicit assumption that macro dynamics are somehow independent from distribution or heterogeneity parameters at the micro level. A proper treatment of growth is required to better understand the links between micro and macro phenomena. A brief digression on the "aggregation problem" is useful here. An aggregation problem exists whenever the aggregate agents' behavior, such as aggregate private demand, cannot be "treated as if it were the outcome of the decision of a single maximizing consumer" (Deaton and Muellbauer 1980: 148). When aggregation conditions do not hold, macro models or models with representative agents do not necessarily tell the whole story and, in particular, miss out on some important interactions between distribution and growth. It turns out that aggregation conditions tend to be quite stringent and, as Deaton and Muellbauer (1980: 149) observe, this has "tempted many economists to sweep the whole problem under the carpet or to dismiss it as of no importance." The intuitively appealing way of writing off the aggregation issue often consists of assuming that the heterogeneity in circumstances of individual agents cancels out. But again, adopting this view severely limits not only the possibility of assessing poverty/distribution impacts of macro policies or apprais- ing macro consequences of micro interventions, but it also hinders the proper modeling of growth. Macro literature on endogenous growth has repeatedly emphasized the key role of nonlinearity and market imperfections. Notably, these same two features are sufficient 320 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA to break the aggregation conditions.2 In these aggregation problem- atic situations, a macro-micro modeling approach can be helpful. And its usefulness in terms of improving an understanding of the growth process is definitely an underresearched area. The recent literature on the inequality of opportunities (see Roemer 1998; Roemer et al. 2003; Bourguignon, Ferreira, and Menendez 2007) has shown that unequal initial distribution or unequal access to education, health, and other human development factors leads to inequality traps in which investment opportunities are missed and institutional arrangements tend to be biased to main- tain a status quo that favors those with more influence. These, in turn, result in lower growth. These direct links between greater equity and higher efficiency allow conceiving "efficient redistribu- tions" and, more generally, to overcome the old idea of a dichotomy between distribution and efficiency and therefore growth. Empirical models where these ideas could be tested have to overcome relevant obstacles; in particular they have to deal with the (very) long run, namely, with the large time lags between the achievements of a more equal distribution of opportunities and its effects on growth. Chap- ter 8 in this volume, as well as Heckman, Lochner, and Taber (1998), is a good example of this promising research on growth and distri- bution complementarities. More has to be done and care must be taken to stay close to the data. A second area of microeconomic research that promises interest- ing results in terms of linking growth and distribution is that of modeling firms' behavior. The importance of modeling heterogene- ity of households has been clearly shown by the studies in this vol- ume, but the importance of modeling heterogeneity of production and investment decisions by firms is perhaps as essential, especially when the focus of the analysis is on the determinants of productivity and growth. A simple example can clarify the issues. Many studies (for example, Tybout and Westbrook 1995; Nickell 1996; Pavcnik 2002; Lopez-Cordova 2003; De Hoyos and Iacovone 2006; Fernandes 2007) have identified a positive relationship of competi- tion, mainly in the form of increased penetration of foreign sup- pliers in cases of trade liberalization and economic performance. This effect may be different across heterogeneous firms, however, with good firms (that is, those closer to the technological frontier) benefiting disproportionately and bad firms being affected nega- tively. The heterogeneity across firms is enormous and so is their behavior. Solely owned firms with no employees (that is, the self- employed) are quite different from large corporations, and within the large group of small and medium enterprises (SMEs), significant variation exists. Besides, the same enterprise may change its behavior CONCLUSION 321 because of its age, sectoral shocks, macroeconomic pressures, and other factors. A single micro simulation model cannot capture this complexity. The proper modeling of entry, growth, survival, or exit of firms; the effects of macro policies thereon; and the aggregate macro results of firms' behavior, as well as a combination of approaches, are needed. This combination may include some varia- tion of the macro-micro integrated framework presented in this vol- ume, clearly adapted to deal with firms, but it may also include other methods. Realistically, however, even if it would be nice to have such a firm-focused macro-micro simulation tool, developing it will take years. As in the case of households, the issue is one of "aggregation" and its related features of nonlinearity and market failures. Perhaps an intermediate step, before getting to the final complete model, may be possible and should be attempted. This intermediate model could, for example, include the following characteristics. If empirical obser- vation shows that SMEs behave differently from large firms, then an intermediate model could include two representative firms, with possibly some market power for the large firm. Continuing with the example, the faster churning through entry and exit observed for SMEs could be modeled by larger adjustment parameters in the familiar cost-of-investment model. The real issue is whether enough is known about the evolution of the structure of individual firms to summarize it through a few representative aggregate firms. And it is true that having a dynamic micro simulation model based on a sample of firms would be the first best; but again, at this stage this looks unrealistic, and an intermediate step seems an acceptable second best. Counterparts to these two frontiers of microeconomic research-- modeling inequality of opportunity and heterogeneous firms­­exist at the macro level. The amount and the nature of public spending-- for example, spending more on education may be an obvious mechanism to remedy inequality in the distribution of opportunities-- have macro effects on growth. These effects have been taken into account by the endogenous growth models developed in the macro lit- erature of the 1990s, but the reliability and policy relevance of these types of studies have been questioned. Apart from the generally scarce robustness of the empirical results (Gemmell 2007), most of the models are in reduced form (and partial equilibrium), so tracing the direct effects of policy interventions on agents' behavior is not possible. An example of a structural model in which the growth and the general equilibrium effects of public expenditure programs are accounted for is given by the MAMS model described in chapter 9 of this volume (Bourguignon, Diaz-Bonilla, and Lofgren), but more needs to be done. 322 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA Another area of fruitful macro-micro research includes the development of structural models that identify clear channels through which the quality of governance and, more generally, poli- cies aimed at improving the investment climate can affect actual investment levels, productivity gains, and ultimately growth. There is a strong link with what was mentioned above on the literature on firms' behavior. Research developments may allow assessments of how heterogeneous firms react to macro policies and shocks, and perhaps even how firms are affected by policies that change the institutional environment. Large and small firms in the same sector may react differently and, for instance, take opposite deci- sions in terms of investment plans. Appropriately aggregating these micro results and linking them back to a macro model where other general equilibrium effects can be accounted for may be a useful step forward. Advancing research in macro-micro modeling can be highly rele- vant for policy. Development theories increasingly insist on the importance of institutional arrangements, but it is known that institu- tions are endogenous to the development process and that their qual- ity often depends on which coalitions control economic and political power. Reforms that may be necessary to improve the institutions and thus accelerate growth may negatively affect these coalitions and may thus be opposed. A better understanding of the political econ- omy of institutional change seems thus quite important to single out and successfully implement politically feasible reforms. Analytical tools such as those presented in this volume may be very helpful in this political economy analysis. After all, central results of a macro- micro framework consist of identifying winners and losers of a reform and assessing what compensation must be given for a reform to be undertaken. Therefore, in a broad sense, macro-micro modeling can not only disentangle the complex mutual interactions between distribution and growth, but it can also be viewed as a relevant tool in the polit- ical economy analysis of macro and micro policies and perhaps even of institutional change. Notes 1. Validation exercises are also possible with a sequence of cross-section data sets, as shown in chapter 5. 2. A simple example of nonlinearity is given by the labor participation decision. Not all individuals participate in the labor market, but participa- tion is expected to increase as wages rise. At the micro level this can be CONCLUSION 323 obtained in two ways: either an individual increases his or her working hours, or someone who was not working enters the labor market. Modeling both of these effects becomes impossible if aggregate labor supply is treated as com- ing from a representative worker. Similar discontinuities can be observed when workers move from informal to formal employment. Examples of imperfect markets also abound. These range from the cases of monopolistic power to situations of asymmetric information affecting markets of all types: goods, factors, credit, education, and the like. References Barro, R. 1990. "Government Spending in a Simple Model of Endogenous Growth." Journal of Political Economy 98: 103­17. Bourguignon, F., F. H. G. Ferreira, and M. Menendez. 2007. "Inequality of Opportunity in Brazil." Review of Income and Wealth 53 (4): 585­618. De Hoyos, R., and L. Iacovone. 2006. "Impact of NAFTA on Economic Performance: A Firm-Level Analysis of the Trade and Productivity Channels." Working paper, Inter-American Development Bank, Washington, DC. Deaton, Angus, and John Muellbauer. 1980. Economic and Consumer Behaviour. Cambridge, U.K.: Cambridge University Press. Fernandes, A. M. 2007. "Trade Policy, Trade Volumes and Plant-Level Productivity in Colombian Manufacturing Industries." Journal of the European Economic Association 71 (1): 52­71. Gemmell, N. 2007. "The Composition of Public Expenditure and Economic Growth: A Report to the World Bank." Unpublished manuscript, World Bank, Washington, DC. Heckman, J., L. Lochner, and C. Taber. 1998. "Explaining Rising Wage Inequality: Explorations with a Dynamic General Equilibrium Model of Labor Earnings with Heterogeneous Agents." Review of Economics Dynamics 1: 1­58. Lopez-Cordova, E. 2003. "NAFTA and Manufacturing Productivity in Mexico." Economia: Journal of the Latin American and Caribbean Economic Association 4 (1): 55­88. Nickell, S. J. 1996. "Competition and Corporate Performance." Journal of Political Economy 104 (4): 724­46. Pavcnik, N. 2002. "Trade Liberalization, Exit and Productivity Improve- ments: Evidence from Chilean Plants." Review of Economic Studies 69 (1): 245­76. Roemer, John E. 1998. Equality of Opportunity. Cambridge, MA: Harvard University Press. Roemer, J. E., R. Aaberge, U. Colombino, J. Fritzell, S. P. Jenkins, I. Marx, M. Page, E. Pommer J. Ruiz-Castillo, M. J. S. Segundo, T. Traanes, 324 BOURGUIGNON, BUSSOLO, AND PEREIRA DA SILVA G. Wagner, and I. Zubiri. 2003. "To What Extent Do Fiscal Regimes Equalize Opportunities for Income Acquisition among Citizens?" Journal of Public Economics 87: 539­65. Tybout, J. R., and M. D. Westbrook. 1995. "Trade Liberalization and the Dimensions of Efficiency Change in Mexican Manufacturing Indus- tries." Journal of International Economics 39 (1­2): 53­78. World Bank. 2005. World Development Report 2006: Equity and Devel- opment. Washington, DC, and New York: World Bank and Oxford University Press. Index Figures and tables are indicated by f and t, respectively. accounting incidence analysis Bernhardt, D., 248­49. see also LEB application, 2 (Lloyd­Ellis­Bernhardt) model in top-down modeling, 12 Blundell, R., 7 Agénor, P-R., 286 Bolton, P., 278 aggregation problem, 18­19, bottom-up modeling, in iterative 319­20 micro simulation sequential Aghion, P., 278 solution to integrated multi- average effects, 3 household model, 183, 184 Bourguignon, F., xv, 1­23, 93­118, behavior responses 177­211, 283­315, 317­24 firm-level, 320­21, 322 Brazil, 13, 14­15 Indonesian CGE modeling, 115 currency crisis. see Brazil's in iterative micro simulation currency crisis (1998­99), top- sequential solution to down macro-micro modeling of integrated multi-household economic structure, 74­75t model, 184­85 household income and in micro accounting modeling, consumption patterns, 13­14, 15 73­77, 76t micro accounting results of trade flows, 73 poverty reduction simulations, trade policies, 69­73 233­34 trade policy reform micro simulation modeling outcomes, 78t limitations, 179 see also Latin American trade opportunities for improving reform effects macro-micro research, Brazil's currency crisis (1998­99), 320­21 top-down macro-micro purpose of incidence analysis, 2 modeling of representative household group aggregated earnings results, modeling, 11 140­41t, 151­52t requirements for modeling capital market, 127 policy, 248 constant elasticity of in top-down macro-micro substitution, 127 modeling, 12­13, 14, 19 deflation in stabilization see also occupational choices period, 132 325 326 INDEX Brazil's currency crisis (1998­99) capital markets (Continued) Indonesia CGE model, 103­4, 110 distribution among representative inflows to Thailand, 250, 262f, household groups, 121 273, 274f economy-to-market transmission occupational choice in financial mechanism, 125 liberalization, 255­56 exchange rate policy, 123­24, top-down macro-micro modeling 125­26, 131, 135­36 of Brazil's currency crisis, 127 financial sector modeling, case studies, methodological 129­31 techniques, 2­3 goals, 119­20, 123, 124­25 Chile, 13 historical simulation, 133t, economic structure, 74­75t 134­35t, 135­36 household income and household income distributions, consumption patterns, 153­59, 160­61 73­77, 76t household survey data, 144­45 trade flows, 73 income changes among trade policies, 73 individual households, 121 see also Latin American trade inflation target anchor, 124 reform effects interest rates, 125­26 China's accession to WTO, Latin investment savings and liquidity American trade reform and, 69 preference money supply, 120, closure rules 125, 129 aggregate gains and, 81 labor market, 127­28, 136­39, computable general equilibrium 146­47, 150, 160 modeling, 104­5 limitations, 123 fully integrated macro-micro linking aggregated variables, model, 224­25 121, 136­42, 154, 159 integrated multi-household macro model design, 120, 121, modeling of distributional 123­31, 130f, 132t outcomes of trade reform, micro simulation design, 120, 193­94 121, 142­46, 162­64t top-down macro-micro occupational outcomes, modeling, 121f 137­38t, 148­49t Cogneau, D, xv, 213­45 occupational structure model, Colombia, 13 164­69tt economic structure, 74­75t output conversion matrix, 129 household income and con- policy responses, 123­24 sumption patterns, 73­77, 76t poverty changes, 159­60 trade flows, 73 predictive performance, 120, trade policies, 73 121­23, 132, 135­36, 139­42, see also Latin American trade 147­53, 155­59, 160­61 reform effects results of macro model, computable general equilibrium 131­36 modeling supply side sectors, 126­27 advantages, 63 wage changes, 139­42, balanced macro closure in, 150­53, 160 104­5 Bussolo, M., xv, 1­23, 61­90, of costs of Indonesian financial 317­24 crisis, 95­96, 101­5 INDEX 327 elasticities of substitution and Moroccan trade policy effects, transformation, 102­3 35, 36t, 44f, 45f equilibrium in, 104 patterns in Latin America, 77 financial crisis modeling, 120 structural microeconometric limitations, 63, 120 model of income generation, micro simulation modeling 214, 242 linkage, 14­16, 99­101 Costa Dias, M., 7 with micro simulation sequential cost-of-living index method, 179 LEB model calibration, 260 for modeling trade policy reform in LINKAGE modeling of trade outcomes, 62 reform effects, 66 Moroccan trade policy reforms, credit access 29­30 absence of, in Thai simulations, strategies for linking macro- 263­66 micro models, 178­81 distribution of welfare gains in trade reform effects, 28 Thai liberalization, 250 see also LINKAGE model LEB model, 249, 263­66 constant elasticity of modeling costs and benefits of substitution, 224 policy reforms, 248­49 integrated multi-household modeling occupational choice modeling of distributional and, 254­56 outcomes of trade Thai economic development, reform, 193 249­50, 261 in MAMS, 291, 292 currency crises, generally modeling labor and capital distribution of effects, 119­20 demand, 127 see also Brazil's currency crisis modeling production goods in (1998­99), top-down macro- Indonesia, 102­3 micro modeling of three-level nested model, 127 constant elasticity of data sets transformation, 224 Brazilian household survey, integrated multi-household 144­45 modeling of distributional consistency issues, 67­68, 106, outcomes of trade reform, 193 222, 319 modeling production goods in household surveys, 67, 68 Indonesia, 103 Indonesian financial crisis, 106 consumption, household integrated multihousehold data sets, 4 modeling of Philippines trade fully integrated macro-micro reform, 192 modeling, 179­80, 222­23 LINKAGE modeling of Latin Indonesia CGE model, 104 American trade reform effects, iterative micro simulation 64­65, 66 sequential solution to Madagascar study, 220, 222 integrated multi-household for MAMS, 292 model, 185, 192­93 for micro-macro analysis, 3­4, Latin American trade reform 67, 318­19 effects, 81 national accounts, 67 modeling challenges, 10­11, opportunities for 18­19 improvement, 318 328 INDEX data sets (Continued) outcomes of Thai liberalization social accounting matrices, 67­68 reforms, 250, 267­73, for structural microeconometric 271t, 272f modeling, 235 outcomes of untargeted transfer for Thai reform simulations, program in Madagascar, 232 258­60 outcomes of workfare program Deaton, A., 4, 67 in Madagascar, 231 development theory Philippines' trade policy outcomes, conceptual evolution, 5­6, 61­62 177, 202­3, 202f, 205f, 206 human development and rationale for macro-micro economic growth, 284, 307­8 analysis, 4­8, 20, 177­78 LEB model, 249 strategies for linking macro- political economy of institutional micro models, 178­81 change, 322 subsistence sector poststabilization macroeconomic modeling, 253 policies, 317 targeting properties of poverty- progress toward Millennium reduction schemes, 226­28 Development Goals, 284 Thai economic development, trade liberalization advocacy, 61 250, 261 Washington Consensus, 6 top-down macro-micro modeling Díaz-Bonilla, C., xv, 283­315 of Brazil's currency crisis, 145, distribution of welfare 153­59 capital inflow effects, 250 traditional macroeconomic credit access and, 250 modeling, 8­9 currency crisis effects, 119­20 use of representative household differences in rural­urban groups to study, 9, 94­95, 120, outcomes of trade 178­79 liberalization, 35­40, 44­46, vertical vs. horizontal, 28­29, 47­53, 55 46­54, 56 economic growth and, 320 see also poverty levels effects of Moroccan trade dynamic modeling, 288­89, reforms, 27­28, 35­44, 42f, 294­96, 319 43f, 55­56 evolution of development theory, econometric modeling, 8 5­6, 61­62 micro accounting modeling and, inequality traps, 320 15­16 Latin American trade reform educational attainment effects, 81­86, 82t employment and wage LINKAGE modeling of trade outcomes, 303­4 reform effects on, 65­67 integrated multi-household micro accounting modeling, 63 modeling of Philippines' trade modeling trade liberalization in reforms, 201 Morocco, 29­33 MAMS focus, 289, 293­94 nonincome factors, 20­21 outcomes of Moroccan trade obstacles to Moroccan trade policy reforms, 47 reform, 27­28 structural microeconometric outcomes of Indonesian financial modeling, 220­21 crisis, 106­7, 108­9, 110­14, trade reform transmission to 112­13t poverty outcomes, 200 INDEX 329 entrepreneurs ex post evaluation, 1 distribution of welfare gains in poverty reduction strategies in Thai liberalization, 250, Madagascar, 233­34 267­69, 270 purpose, 7 intermediation access and EXTER model, 192 occupational choice, 254­56 modeling setup costs, 251­53 factor markets Ethiopia, MAMS simulations for, aggregate supply and demand 309­10 modeling, 126­27 extrapolation of current formal and informal trends, 297 differentiation, 102 focus, 296­97 modeling with representative MDG attainment with foreign household groups, 9 aid, 297­305 short-term CGE modeling, 106 MDG attainment with less top-down macro-micro modeling foreign aid, 297, 305­8 of Brazil's currency crisis, modeling scenarios, 297 127­28, 136, 146 evaluating policy impacts top-down modeling advantages of macro-micro approach, 12f approach, 4 farm households choice of microeconomic diversification of activities, 221 technique, 7­8 income modeling, 219­21 modeling rationale, profits modeling, 220 247­48, 317 feedback loops, 16­17, 179 perceptions of role of Ferreira, F. H. G., xv, 119­74 government and, 247 financial sector rationale, 1, 177­78 access modeling, 248­49 requirements for mathematical constraints in LEB model, models, 248 252­53 scope of poverty reduction distribution of welfare gains from issues, 1­2 Thai liberalization, 267­73 techniques, 2­4 evolution in Thailand, 249­50 Evans, D., 278 LEB model, 249 ex ante evaluation, 1, 62 occupational choice in poverty reduction strategies in liberalization, 255­56 Madagascar, 233­34 top-down macro-micro modeling research methods, 8 of Brazil's currency crisis, exchange rate 129­31 limits of microeconomic foreign aid modeling, 3 challenges to modeling macroeconomic modeling Millennium Development challenges, 8 Goals, 287 MAMS modeling, 306, 309 debt forgiveness, 309 modeling Indonesian financial MAMS modeling, 295, crisis, 102 297­305 top-down macro-micro modeling Foster-Greer-Thorbecke indicators, of Brazil's currency crisis, 198­99 123­24, 125­26, 131, Free Trade Area of the Americas, 135­36 68, 77­78, 79, 81, 86 330 INDEX fully integrated macro-micro Indonesian CGE modeling, modeling, 18­19, 179­80 114­15 occupational choice and labor integrated multi-household income model, 222­25 modeling of Philippines' trade schematic, 182f, 223f reforms, 200 with structural iterative micro simulation microeconometric model of sequential solution to income generation, 216­22 integrated multi-household targeted policy simulations in model, 185, 192­93 poverty reduction in labor market factors in Madagascar, 225­35 modeling, 99 see also iterative micro simulation LINKAGE modeling of trade sequential solution to integrated reform effects, 66 multi-household model linking with CGE model, 99­101 micro simulation modeling, geographic analysis, 3 96­99, 179 Giné, X., xv, 247­80 modeling challenges, 18­19 globalization, recommendations outcomes of Moroccan trade for developing countries, 61 policy reforms, 47, 49­50t Global Trade Analysis Project, 64­65 patterns in Latin America, government productivity, 307, 309 73­77 research needs, 321 per capita transfer to reduce poverty, 226, 229, 232 HIV/AIDS, 289 representative household group household characteristics modeling, 9 CGE modeling of poverty shortcomings of modeling with effects, 63­64 representative household outcomes of Moroccan trade groups, 95 policy reforms, 47­53 structural microeconometric size, 47 model, 214­22 top-down macro-micro modeling top-down macro-micro modeling of Brazil's currency crisis, of Brazil's currency crisis, 143­44 153­59, 160­61 wage earners, 47 trade reform transmission to see also consumption, poverty outcomes, 200 household; income, household transmission mechanisms in human development financial crises, 120 economic growth and, 284, 308 see also distribution of welfare; poverty reduction trade off, wages 307­8 Indonesia, 14 Human Development Report capital markets, 103 (UN), 285 financial crisis (1997). see Indonesian financial crisis Impact of Economic Policies on labor market, 103 Poverty and Income Distribution, 2 Indonesian financial crisis income, household behavioral variables, 115 changes in Brazilian currency CGE model, 101­5 crisis, 121, 142­43, 145 data sets, 106 farm household, 219­20, 221 drought effects, 109­10 INDEX 331 evolution of occupation iterative micro simulation choices, 108t sequential solution to integrated goals of counterfactual multi-household model analysis, 94 advantages, 185, 191­92 historical simulation, 105, 107­9 aggregate effects of trade reform, income distribution and poverty 195­96, 203­4 outcomes, 93­94, 106­7, 107t, behavioral responses in, 184­85 108­9, 108t, 110­14, 112­13t bottom-up equation, 183, 184 labor market outcomes, 107­8 competitive equilibrium macro aggregate outcomes, 111t under, 181 micro simulation model, 96­101 data sources, 192 modeling methodology, 95­106 distributional effects of trade policy simulations, 105 reform, 192­95, 202­6 rural­urban differences in fixed-point algorithm, 182, 184 outcomes, 106­7, 109t, 110­14 household aggregation, 182 significance of modeling, 114­15 labor market imperfections in, social costs of, challenges in 185­92 estimating, 93­94 method, 181­85 time horizon for modeling, micro simulation sequential 105­6 modeling versus, 203­4, 205­6 See also Indonesia multimarket equilibrium industrialization, evolution of stability, 184 development theory, 5 poverty effects of trade reform, infrastructure development, 289 198­202, 204 economic growth and, 308 rationale, 181 inheritance, 252, 268 sectoral effects of trade reform, initial conditions 196­98, 204 CGE modeling, 63 top-down equation, 183­84 modeling Latin American trade reform outcomes, 69­77, 81, Jovanovic, B., 278 82t, 84, 85­86 as trade policy reform outcome labor market factor, 61 capital inflow effects, 273 integrated multi-household cost of entry, 186­87, modeling. see fully integrated 189­90, 217 macro-micro modeling; iterative demand modeling, 189 micro simulation sequential distribution of talent, 251­52 solution to integrated multi- education outcomes, 303­4 household model formal/informal, 103, 185­86, interest rates 193, 224 LEB model, 252 fully integrated macro-micro top-down macro-micro modeling modeling, 223, 224 of Brazil's currency crisis, Indonesia CGE model, 103 125­26, 131, 136 Indonesian financial crisis International Monetary Fund, 124 effects, 107­8 investment savings and liquidity integrated multi-household preference money supply, 120, 129 modeling of distributional modeling distribution effects of outcomes of Philippines' trade Brazilian currency crisis, 125 reforms, 185, 186, 194­96 332 INDEX labor market (Continued) LEB (Lloyd-Ellis­Bernhardt) model LINKAGE modeling of trade aggregation for village reform effects on poverty, 64, economies, 266 86­87 applications, 248­49 micro simulation modeling, 15 basic elements, 251­54 micro simulation modeling of calibration, 249, 260­63 household income, 99 data sets and parameter modeling imperfections in, estimation for modeling Thai 185­92 reforms, 256­60 outcomes of poverty reduction distribution of welfare gains simulations in Madagascar, 229 from financial sector part-time employment, 236­37 liberalization, 267­73 structural microeconometric financial constraints, 252­53 model of income generation, firm setup costs, 251­53 214­16 interest rates, 252 supply modeling, 188­91, 206t intermediation extension, 249 top-down macro-micro modeling modeling distribution of of Brazil's currency crisis, talent, 252 127­28, 136­39, 146­47, occupational choice modeling, 150, 160 249, 253­56 as trade policy reform outcome outcomes of Thai liberalization factor, 61 policies, 250 workfare program to reduce predictive performance, poverty, 225­26, 228, 229 274­75 see also occupational choices; production function, 251 wages simulation with no access to land market, 103 intermediation, 263­66 Latin American trade reform subsistence sector, 253 effects Lehnert, A., 278 data set for modeling, 64­65, 66 Leite, P. G., xvi, 119­74 Free Trade Area of the Americas LINKAGE model and, 68, 77­78, 79, 81, 86 advantages, 86 full trade liberalization scenario, data sets, 64­65 69, 79­81, 86 micro accounting framework, income elasticity and, 83­85 65­67 initial conditions as outcome modeling poverty and factor, 69­77, 84 distribution effects of Latin macroeconomic results, 77­81 American trade reform, 64­65 modeling methodology, 62­68, shortcomings, 86­87 86­87 linking aggregate variables, 12, poverty and income distribution, 16­17 81­86, 82t, 84t factor market modeling, 128 prices and wages, 79­81, 80t top-down macro-micro modeling protectionist policies, of currency crisis, 121, 136­42, international comparison, 154, 159 69­73, 70­72t Lloyd-Ellis, H., 248­49. seealso sectoral adjustments, 78t LEB (Lloyd­Ellis­Bernhardt) model significance of findings, 85­86 Lofgren, H., xvi, 283­315 Lay, J., xv, 61­90 Lokshin, M., xvi, 27­60 INDEX 333 macroeconomic modeling variant applications, 4 to assess distribution of welfare, vertical vs. horizontal impacts, 8­9 28­29 for assessing Millennium Madagascar, poverty in, 18 Development Goals, 286­87. agricultural price subsidy see also Maquette for MDG simulation, 225, 228, 229, 231, Simulations 232, 233f rationale, 317 data sources, 220, 222, 235 with representative household goals of policy modeling, 215­16 groups, 8­11 household income transfer macroeconomic policy simulation, 226, 227t, 229, 232 distribution of welfare implementation costs of poverty linkage, 4­7 reduction strategies, 232 poststabilization strategies, 317 micro accounting results of scope of, 3 poverty reduction simulations, macro-micro techniques, 2, 3 233­34 advantages of, for policy modeling methodology for evaluation, 317 policy analysis, 213, 214­15, aggregation problem, 319­20 216­25, 235­36 analysis of nonincome outcomes of policy simulations, dimensions of poverty, 20­21 228­35, 229t, 230t, 231t challenges, 4, 318­19 targeted policy simulations, data sets for, 3­4, 67­68, 225­38 318­19 wage patterns, 225­26, 226t dynamic modeling, 319 workfare program simulation, evolution of distribution 225­26, 229, 231­32 analysis, 4­7 Magnac, T., 187­88 fully integrated approach, MAMS. see Maquette for MDG 18­19, 179­80 Simulations micro accounting modeling, Maquette for MDG Simulations 13­14, 19 (MAMS), 287­88 modeling firm behavior, advantages, 309, 310 320­21, 322 applications, 287, 296, 309­10 opportunities for improving, conceptual basis, 288 319­22 data sources, 292 purpose, 3 delayed effects modeling, rationale for poverty research 295­96 with, 86 dynamic framework, 288­89, rationale for researching 294­96 distributional effects of policy educational focus, 289, 293­94 with, 177­78 focus, 289 with representative household government productivity groups, 8­11, 94­95, 178­79 analysis, 307 research needs, 21 implications for government for studying poverty alleviation spending, 307­8 policies in Madagascar, 213, nonrecursive features, 296 222­25 production functions, 288, top-down approach, 11­16, 19 289­90, 292­94 use of feedback loops, 16­17 public spending impacts, 294­95 334 INDEX Maquette for MDG Simulations in macro-micro modeling of (MAMS) (Continued) currency crisis, 120, 142­46 structure and implementation, shortcomings, 179 289­94 see also iterative micro see also Ethiopia, MAMS simulation sequential solution simulations for to integrated multi-household marginal effects, 3 model market imperfections, micro Millennium Development Goals simulation modeling, 15 costs of achieving, 304­5 McNamara, Robert, 5 economic growth and, 284 MDG. see Millennium financing, 283 Development Goals goals, 283 Medvedev, D., xvi, 61­90 government spending outcomes, Mexico, 13 288, 294­95 economic structure, 74­75t human development objectives, household income and 208, 283, 284 consumption patterns, 73­77 macroeconomic modeling tools, household income patterns, 76t 20­21, 286­87. see also trade flows, 73 Maquette for MDG trade policies, 69 Simulations see also Latin American trade monitoring, 284 reform effects performance evaluation micro accounting modeling challenges, 284­85, 286­87 advantages, 13­14, 63 performance evaluation features, 13­14 strategies, 285­86, 291t limitations, 13, 63 progress, 283­84 in LINKAGE modeling of trade minimum wage, 191 reform effects, 65­67 Madagascar, 225­26 poverty reduction strategies in Morocco's trade policy, 13 Madagascar, 233­34 consumption patterns and, 36t, microeconomic modeling, 2, 3 44t, 45f applications, 3 differences in rural­urban choice of evaluative technique, outcomes, 35­40, 44­46, 7­8 47­53, 55 limitations, 3 distributive effects of reform, Moroccan trade policy reforms, 27­28, 35­44, 55­56 29­33 horizontal and vertical effects of see also micro accounting reform, 28­29, 46­54, 56 modeling; micro simulation household characteristics as modeling outcome variables, 47­53, micro simulation modeling, 14­16 49­50t definition and purpose, 8, household impacts of reforms 14­15, 16, 179 in, 38t features, 15­16, 179 import tariffs on cereals, 27 household income generation, methodology for modeling 96­99 reform effects, 28­29, 56­58 iterative integrated multi- microeconomic analysis, 29­33 household approach versus, obstacles to reform, 27 203­4, 205­6 price change impacts, 33­35, 34t INDEX 335 production patterns and, sectoral effects of trade 37t, 44t liberalization, 196­98, 197t reform rationale, 27 trade reform modeling shortcomings in modeling methodology, 181­85 reform effects, 54 Picchetti, P., xvi, 119­74 wage change effects on modeling poverty levels outcomes, 54 assessing progress toward welfare impacts, 35­44 Millennium Development mortality, MAMS focus, 289 Goals, 285­86 Multi-Fiber Arrangement, 69 educational attainment multilogit modeling, 15­16, 101 effects, 201 effects of Moroccan trade occupational choices reforms, 35­39, 40f distribution of welfare gains evolution of development theory, from financial sector 61­62 liberalization in Thailand, 270 evolution of research farm household modeling, approaches, 5­6 219­20, 221 Foster-Greer-Thorbecke Indonesian financial crisis and, indicators, 198­99 98, 108t human development and, 307­8 in intermediated sector, implementation costs of poverty 255­56, 264f reduction strategies, 232 in LEB model, 249, income change effects of trade 253­56 reform and, 200, 201­2 modeling Brazilian financial integrated multi-household criss, 144 modeling of Philippines' nonhead household trade reforms, 198­202, members, 219 199t, 204 in nonintermediated sector, Latin American trade reform 254­55, 255f effects, 81­86, 82t, 84t micro accounting modeling, 63 panel data, 318 Millennium Development Goals, partial equilibrium modeling, 28 283­84 part-time work, 236­37 modeling policy outcomes in Pereira da Silva, L. A., xv, 1­23, Madagascar, 213 119­74, 317­24 outcomes of Indonesian financial Philippines crisis, 93­94, 106­7, 108­9, aggregate effects of trade 110­14 liberalization, 195­96, 195t price change effects of trade current trade policy, 194 reform and, 63, 200 distributional outcomes of trade rationale for evaluating policy liberalization, 192­95, 202­3, impacts, 1 202f, 205f, 206 rationale for macro-micro labor market, 185, 186, 206t analysis, 7­8, 20, 177 modeling distributional scope of policy issues, 1­2 outcomes of trade strategies for linking macro- liberalization, 177 micro models, 178­81 poverty outcomes of trade targeted policy simulations in liberalization, 198­202, 199t Madagascar, 225­36 336 INDEX poverty levels (Continued) rural­urban differences top-down macro-micro modeling income patterns in Latin of Brazil's currency crisis, 159­60 America, 73­77 trade liberalization effects, 85 Indonesian financial crisis see also distribution of welfare; effects, 106­7, 109t, 110­14 Madagascar, poverty in Moroccan trade reform poverty maps, 3 outcomes, 35­40, 44­46, Poverty Reduction Strategies, 287 47­54, 51­52t, 55 Poverty Reduction Strategy outcomes of agricultural price Papers, 285 subsidies in Madagascar, 234 price changes outcomes of workfare program Indonesian financial crisis in Madagascar, 232 modeling, 109­10, 114 integrated multi-household SAM-CGE­RHG model, 9 modeling of Philippines' trade Savard, L., xvi, 177­211 reforms, 200 savings behavior Latin American trade reform fully integrated macro-micro effects, 79, 81 modeling, 179­80 Moroccan trade liberalization LEB model, 249 outcomes, 33­35, 34t MAMS modeling, 295 outcomes of poverty reduction no access to intermediation in simulations in Madagascar, 228 Thai simulation, 263­66 trade reform transmission to Thai economic development, poverty outcomes, 63, 200 260­61 productivity second round effects, 2, 3, 318 government, 307 self-employment labor, 186, 191 income modeling, 67, 97, 98, 99 public spending structural microeconometric evolution of development theory, 6 modeling of income generation, general equilibrium effects, 20­21 216­18 incidence analysis, 6 setup costs, 251­53, 268­69 poverty impact, 1­2 Simulations for Social Indicators and Poverty, 285 Ravallion, M., xvi, 27­60 social accounting matrices, Redistribution with Growth, 5 67­68, 96 representative household groups Indonesian financial crisis income changes in Brazilian modeling, 101­2 currency crisis, 121 in MAMS, 292 limitations in CGE-modeling, for studying poverty alleviation 62, 120 policies in Madagascar, 222 macroeconomic modeling with, social accounting 8­11, 178­79 matrix­computable general shortcomings of modeling with, equilibrium model, 9 94­95 social experiment research, 7­8 Robilliard, A. S., xvi, 93­118, structural microeconometric 213­45 modeling Robinson, S., xvi, 93­118 advantages, 235 Roy, A., 216, 217 aggregation for, 214­15 INDEX 337 for analysis of targeted policies, time-lagged effects, 295­96, 320 215­16 top-down modeling, 19, 105 applications, 235­36 basic features, 12­13 data sources, 235 in iterative micro simulation expenditure system with sequential solution to heterogeneous preferences, 242 integrated multi-household farm income, 219­20, 221 model, 183­84 of income generation, 214­15, limitations, 19 216­22 micro accounting in, 13­14 limitations, 215, 235 micro simulation modeling, macro-level interactions, 222­25 14­16 micro calibration, 220­21, micro simulation sequential 238­41t method, 179 part-time employment, 236­37 rationale, 11­12 subsidies, agricultural schematic, 12f, 121f distribution of benefits, 226­27 see also Brazil's currency crisis implementation costs, 232 (1998­99), top-down macro- micro accounting results of micro modeling of poverty reduction simulations, Townsend, R. M., xvi, 247­80 233­34 trade policy poverty reduction strategy in aggregate effects of Madagascar, 225, 226­27, 231, liberalization, 195­96 232, 233­34 current Philippines', 194 targeting accuracy, 232 determinants of reform subsistence sector modeling, 253 outcomes, 61 distributive effects of tariffs liberalization, 55­56, 177, Moroccan cereal imports, 27 192­95, 202­3, 206 patterns in Latin American, evolution of development theory, 69­73 61­62 Thailand, 18 government budget, capital inflows, 250, 273 liberalization effects on, 195­96 credit access, 250, 261 poverty outcomes in Philippines' data sets and parameter reforms, 198­202 estimation for reform protectionist policies in Latin simulations, 256­60 America, 69­73, 70­72t distribution of welfare gains, sectoral effects of liberalization, 250, 261, 267­73 196­98 economic development, 249­50, see also Latin American trade 260­61 reform effects; Morocco's trade household characteristics, policy; tariffs 258­59 liberalization outcomes, 250 utility theory, 14 methodology for modeling liberalization effects, 251­63 van der Mensbrugghe, D., xvi, savings behavior, 260­61 61­90 village economies, 266 variants of single modeling wages, 250 framework, 3, 4 338 INDEX wages rigidity in formal sector of distribution of welfare gains in Philippines' economy, 186 Thai liberalization, 250, structural microeconometric 267­68, 271 model, 216­18 education outcomes, 303­4 top-down macro-micro modeling Indonesian financial crisis of Brazil's currency crisis, effects, 107­8 139­42, 150­53, 160 integrated multi-household see also minimum wage modeling of distributional Walrassian modeling, 188 outcomes of Philippines' trade water supply and access, 289 reforms, 195­96 wealth constraints, modeling. see iterative micro simulation LEB (Lloyd-Ellis­Bernhardt) model sequential solution to workfare programs, 225­26 integrated multi-household distribution of benefits, 228 model, 186, 188, 191 implementation costs, 232 Latin American trade reform outcomes in Madagascar poverty effects, 79­81 reduction simulations, 229, LEB model, 252 231­32 LINKAGE modeling of trade rural­urban differences in reform effects on poverty, 66­67 outcome, 232 micro simulation modeling World Bank, 309­10 of household income, 97­98 ECO-AUDIT Environmental Benefits Statement The World Bank is committed to preserving Saved: endangered forests and natural resources. · 15 trees TheOfficeofthePublisherhaschosentoprint · 11 million Btu of The Impact of Macroeconomic Policies on total energy Poverty and Income Distribution on recycled · 1,357 lb. of net paper with 30 percent postconsumer fiber in greenhouse gases accordance with the recommended standards · 5,631 gal. of waste for paper usage set by the Green Press Initia- water tive, a nonprofit program supporting pub- · 723 lb. of solid lishers in using fiber that is not sourced from waste endangered forests. For more information, visit www.greenpressinitiative.org. EQUITY AND DEVELOPMENT SERIES M acro-level policies, whatever their impact on macro-level aggregates, can have significant distributional consequences. Polities and policy makers are well aware of this, but their economic advisers, especially those in international agencies, seemed in the past to have been less well attuned to these realities. This volume shows that the landscape is changing. The chapters were written by the leaders in their field and showcase techniques and applications that allow the distri- butional consequences of policy reform to be analyzed systematically. There can no longer be any technical excuse for totally ignoring the distributional impacts of macro-level policy instruments. --Ravi Kanbur, T. H. Lee Professor of World Affairs and Professor of Economics, Cornell University This book uses a single modeling framework--a macro model linked with a household-level micro model-- to examine the consequences on poverty and income distribution of changes to the trade regime, tariffs and nontariff barriers, the exchange rate, interest rates, the mix of fiscal and monetary policies, the composition of public spending, and labor market regulation. It also examines the macroeconomic consequences of scal- ing up micro-level programs, such as a conditional cash transfer program. The book represents the state of the art in using models to understand the impact of policies on poverty alleviation, and is a must-read for both policy makers and students interested in poverty, income distribution, and growth. --Raghuram G. Rajan, Eric Gleacher Distinguished Service Professor of Finance, Graduate School of Business, University of Chicago This book represents a significant advance in development policy analysis. It demonstrates how macro- economic modeling can be married to micro data sets to produce more meaningful analyses of economy- wide shocks or policy changes. It has important lessons to offer to both macro and micro development economists: it shows the former how to use better data and the latter how to broaden their analyses. --Dani Rodrik, Professor of International Political Economy, John F. Kennedy School of Government, Harvard University The interested reader will find in this volume state-of-the-art research and highly valuable, policy-relevant knowledge on the crucial issue of growth and distribution. --Ernesto Zedillo, Director, Yale Center for the Study of Globalization; Chairman of the Board, Global Development Network; Member, Commission on Growth and Development; Former President of Mexico ISBN 978-0-8213-5778-1 SKU 15778