66237 A WORLD BANK STUDY Knowing, When You Do Not Know S I M U L AT I N G T H E P O V E R T Y A N D DISTRIBUTIONAL IMPACTS OF AN ECONOMIC CRISIS Ambar Narayan and Carolina Sánchez-Páramo Editors W O R L D B A N K S T U D Y Knowing, When You Do Not Know Simulating the Poverty and Distributional Impacts of an Economic Crisis Edited by Ambar Narayan and Carolina Sánchez-Páramo © 2012 International Bank for Reconstruction and Development / International Development Association or The World Bank 1818 H Street, NW Washington, DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org 1 2 3 4 14 13 12 11 World Bank Studies are published to communicate the results of the Bank’s work to the development community with the least possible delay. The manuscript of this paper therefore has not been prepared in accordance with the procedures appropriate to formally-edited texts. This volume is a product of the staff of The World Bank with external contributions. The �ndings, interpretations, and conclusions expressed in this volume do not necessarily reflect the views of The World Bank, its Board of Executive Directors, 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 judg- ment 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 work is subject to copyright. Because The World Bank encourages dissemina- tion of its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full a ribution to the work is given. For permission to reproduce any part of this work for commercial purposes, 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 (print): 978-0-8213-8981-2 ISBN (electronic): 978-0-8213-8954-6 DOI: 10.1596/978-0-8213-8981-2 Library of Congress Cataloging-in-Publication Data Knowing, when you do not know : simulating the poverty and distributional impacts of an economic crisis / edited by Ambar Narayan, Carolina Sánchez-Páramo. p. cm. Includes bibliographical references. ISBN 978-0-8213-8981-2 (alk. paper) -- ISBN 978-0-8213-8954-6 1. Poverty--Mathematical models. 2. Income distribution--Mathematical models. 3. Global Financial Crisis, 2008-2009. I. Narayan, Ambar. II. Sánchez-Páramo, Carolina. HC79.P6K66 2011 339.4’6015195--dc23 2011048299 Cover photo: Job seekers visit a job fair for graduating university students at the Qujiang International Convention and Exhibition Center in Xian of Shaanxi Province, China. © Ge y Images Contents List of Contributors ................................................................................................................ vii Acknowledgments ................................................................................................................viii Foreword ....................................................................................................................................ix 1. Assessing the Poverty and Distributional Impacts of the Financial Crisis with Microsimulations: An Overview of Country Studies ........................................ 1 Financial Crisis and Its Impact on Poverty ...................................................................... 2 Evaluating the Impact of the Crisis................................................................................... 3 The Methodology to Estimate Impacts Ex Ante ............................................................. 4 How the Approach Works ................................................................................................ 5 De�ning the “Impact�: Two Types of Countries ............................................................ 6 Simulating the Impact of Policy Changes ........................................................................ 8 Caveats to the Approach .................................................................................................... 8 Results from Micro-Macro Simulations for Four Countries ......................................... 9 Results for Type I Countries ............................................................................................ 12 Results for Type II Countries ........................................................................................... 19 Simulating the Impact of Policy Responses ................................................................... 25 Conclusion .......................................................................................................................... 26 2. On Economy-Wide Shocks, Models and Politics ....................................................... 30 When Are Second-Round and Systemic Effects a First-Order Concern? .................. 32 Implications ........................................................................................................................ 37 When Can Information Affect Government Behavior? ............................................... 38 Conclusion and a Note on the World Bank ................................................................... 40 3. The Distributional Consequences of the Economic and Financial Crisis of 2008–09: A Comment ........................................................................................................ 42 Backdrop to the Issue of Distributional Consequences .............................................. 42 Thinking about the Problem ............................................................................................ 42 The Guts of the U.S. Story ................................................................................................ 43 Some Basic Generalizations ............................................................................................. 45 The Current Crisis in Perspective.................................................................................... 45 Results of Quick Post Crisis Surveys .............................................................................. 46 Takeaway Observations of Relevance for Policy .......................................................... 48 4. Stress Testing for the Poverty Impacts of the Next Crisis ........................................ 50 Introduction........................................................................................................................ 50 The Need for a Social Protection Assessment Program (SPAP) ................................. 50 iii iv Contents An Overview of the Workshop Papers and Presentations .......................................... 51 An Assessment, and What’s Missing.............................................................................. 52 Conclusion .......................................................................................................................... 55 References.................................................................................................................................. 56 Figures Figure 1.1: Scheme of the modeling process ..........................................................................5 Figure 1.2: Impacts in Type I and Type II countries ..............................................................7 Figure 1.3: Percent change in sectoral employment between benchmark and crisis (2010) ...................................................................................................................................13 Figure 1.4: Percent change in household income between benchmark and crisis scenarios (2010) ..................................................................................................................13 Figure 1.5: Percent change in poverty/inequality indicators between crisis and benchmark ..........................................................................................................................14 Figure 1.6: Percent losses in income for the general population (left) and the crisis-vulnerable (right) ....................................................................................................15 Figure 1.7: Percentage of poor and crisis-vulnerable households in rural areas .............15 Figure 1.8: Percentage of households with low-skilled heads............................................16 Figure 1.9: Growth incidence curves: percent change in income (relative to the benchmark) due to crisis ..................................................................................................17 Figure 1.10: Transitions across deciles of per capita income (percent of households in each decile) ...............................................................................................18 Figure 1.11: Macroeconomic projections for Mexico (2008–09 are actual observations; 2010–11 are forecasts) ...............................................................................19 Figure 1.12: Growth incidence curve for Mexico (2008–09) ................................................22 Figure 1.13: Transitions across deciles in Mexico .................................................................22 Figure 1.14: Growth incidence curves for Mexico (2010–11) ..............................................24 Figure 1.15: Transitions across deciles for the period 2010–11 ...........................................24 Figure 2.1: The welfare impact of food price changes in the 1997 Indonesian crisis across households ..............................................................................................................33 Figure 2.2: The pa ern of change in deposits, by size and ownership of deposits, in the lead up to the Argentine �nancial crisis..............................................................36 Figure 3.1: The top percentile (1%) of U.S. income (incomes above US$368,000 in 2008).....................................................................................................................................43 Figure 3.2: U.S. Lorenz curves for total family income and net worth .............................44 Figure 3.3: Household coping strategies in Turkey (survey responses) ...........................46 Figure 3.4a: Coping strategies in Armenia by asset quintile ..............................................47 Figure 3.4b: Coping strategy survey responses ....................................................................47 Contents v Tables Table 1.1: Macroeconomic impacts for Type I countries .....................................................12 Table 1.2: Employment projections (population of age 15+ yrs) ........................................20 Table 1.3: Projected household income and its sources .......................................................20 Table 1.4: Poverty and inequality with and without crisis (constant poverty line) ........21 List of Contributors T his volume includes contributions by Bilal Habib (World Bank), Ravi Kanbur (Cor- nell University), Danny Leipziger (George Washington University), Ambar Narayan (World Bank), Sergio Olivieri (World Bank), Carolina Sánchez-Páramo (World Bank) and Michael Walton (Harvard Kennedy School). vii Acknowledgments T his book has been edited by Ambar Narayan and Carolina Sánchez-Páramo and in- cludes contributions by Bilal Habib and Sergio Olivieri, together with the editors (Chapter 1), Michael Walton (Chapter 2), Danny Leipziger (Chapter 3), and Ravi Kanbur (Chapter 4). All chapters are based on the presentations and commentaries at the World Bank PREM Network Conference “Distributional Impacts of Macroeconomic Shocks: An International Overview,� organized in May 2011 in Washington, DC. The editors would like to thank conference participants for helpful comments and suggestions. They are also grateful to Ihsan Ajwad, Tamar Atinc, Otaviano Canuto, Lou- ise Cord, Francisco Ferreira, Anna Fru ero, Alan Gelb, Indermit Gill, Jesko Hentschel, Carlos Felipe Jaramillo, Emmanuel Jimenez, Daniela Maro a, David Newhouse, Ricar- do Paes de Barros, Pierella Paci, Martin Ravallion, Ana Revenga, Jaime Saavedra, Sudhir She y, Emmanuel Skou�as, Carolyn Turk, and Ruslan Yemtsov for valuable comments on the microsimulation exercises presented in Chapter 1. Francisco Ferreira and J. Hum- berto Lopez provided valuable comments as peer reviewers when the microsimulation project was launched. The editors are also deeply grateful to Ana Pacheco, Diego Bat- tistón, and Raúl Oscar Sánchez for capable research assistance during the empirical work conducted for multiple countries; Bilal Habib for his assistance in pu ing this volume together; and Christopher Humphrey for helping edit the volume. This work would not have been possible without the advice and support of colleagues in different country teams, including Zahid Hussain, Sanjay Kathuria, Lalita Moorty, Nobuo Yoshida, and Sanjana Zaman for Bangladesh; Jehan Arulpragasam, Rashiel Besana Velarde, Ulrich Lachler, Eric Le Borgne, Andrew Mason, Sheryll Namingit, Laura Pabón, and Mano- har Sharma for the Philippines; Irene Clavijo and Samuel Freije for Mexico; and Benu Bidani, Kaspar Richter, and Victor Sulla for Poland. Finally, the editors are grateful to Stephen Pazdan for coordinating the production of this publication; to Debra Cubi , Lanto Ramanankasina, Nelly Obias, and Jae Shin Yang for their assistance at all stages of the work; and to Paula Suarez Buitron for her support on website and communications. The microsimulation work and this publication received �nancial support from the Trust Fund for Environmentally and Socially Sustainable Development and the Multi- Donor Trust Fund for the Gender Action Plan. viii Foreword E conomists have long sought to predict how macroeconomic shocks will affect in- dividual welfare. Macroeconomic data and forecasts are easily available when cri- ses strike. But policy action requires not only understanding the magnitude of a macro shock, but also identifying which households or individuals are being hurt by (or ben- e�t from) the crisis. Moreover, in many cases, impacts on the ground might be already occurring as macro developments become known, while micro level evidence is still unavailable because of paucity of data. Because of these reasons, a comprehensive real- time understanding of how the aggregate changes will translate to impacts at the micro level remains elusive. This problem is particularly acute when dealing with developing countries where household data is sporadic or out of date. A popular solution is to extrapolate the welfare impact of a shock from the histori- cal response of income or consumption poverty to changes in output, by estimating an “elasticity� of poverty to growth. Although this method provides an estimate for the aggregate poverty impact of a macro shock, it has limited value for analysts and policy makers alike. Aggregate numbers are useful to capture the a ention of policy makers and the international community, but in the absence of any information on who is affect- ed and to what extent, provide li le guidance on what actions need to be taken. To take one example, most targeted anti-poverty programs focus on the existing poor. But when a crisis occurs, any efforts to expand or retool existing programs would require �nd- ing out who is likely to become poor and how much more deprivation is likely to occur among the existing poor as a result of the shock. Moreover, the speci�c characteristics of an output shock, whether it is caused, for example, by a natural disaster, a reduction of external demand, or internal macroeconomic mismanagement, may lead to very differ- ent impacts along the income scale that demand different policy responses. This volume outlines a more comprehensive approach to the problem, showcasing a microsimulation model, developed in response to demand from World Bank staff work- ing in countries and country governments in the wake of the global �nancial crisis of 2008-09. During the growing catastrophe in a few industrialized countries, there was ris- ing concern about how the crisis would affect the developing world and how to respond to it through public policies. World Bank staffs were scrambling to help countries design such policies; this in turn required information on which groups of the population, sec- tors and regions the crisis would likely affect and to what extent. The problem was exacerbated by the fact that the 2008–09 crisis was somewhat unique. Unlike many other recent episodes of economic crisis, it did not originate in the developing world and did not involve high levels of inflation. There was also no consensus on who would be affected, partly because the anatomy of the crisis in each country would have depended on the nature of the economy, the extent of “exposure� to the global turmoil and the extent of integration with the developed world. Would the middle class bear the brunt of the downturn? How much would the impacts differ across countries, in terms of who are affected? Without good high-frequency micro-level data, as is the case in most developing countries, the answers to these questions had to rely on simulations that link macro projections with pre-crisis household data. Using simple ix x Foreword extrapolations from a historical series of income and poverty was not enough. On the other hand, devising a comprehensive macro-micro simulation model, linking pre-crisis household data with Computable General Equilibrium (CGE) models was impractical for most countries given that such models are hard to �nd and calibrate within the lim- ited time available to analysts. Rather, �lling this void required �nding the �ne balance between two competing needs: capturing the complexities that influence how macro impacts are transmi ed to the households and feasibility in terms of time, cost, and data requirements. The void was at least partially �lled by the approach developed by staff from the Poverty Reduction & Equity Department of the World Bank, modifying some of the well-known microsimulation models from the economic literature. Starting with the idea of using simple macroeconomic projections as the “macro linkages� to a microbe- havioral model built from household data, the model was conceptualized, re�ned and tested in a diverse mix of countries: Bangladesh, the Philippines, Mexico, Poland, and Mongolia. The results fed into country policy dialogue and lending operations of Bank teams as well as various reports, research papers and briefs. Although conceived with the �nancial crisis in mind, the model can be applied to different kinds of macroeconomic shocks—essentially any macro event that leads to a change in real output at the national, sectoral and (subject to data availability) sub-na- tional level for a country. The model has already been used to determine the impacts of typhoons in the Philippines and Dzud (harsh winter) in Mongolia. It can also be a plat- form to build extensions to simulate the distributional impacts of other types of macro shocks, as long as they share certain common characteristics. For one, since this model was developed in the context of a crisis that manifested primarily in the form of output shocks, it focuses on the labor market as the main transmission channel, which is ideal for analyzing the impacts of certain types of crises but not for others (such as commodity price shocks). While microsimulation models are powerful analytical tools to assess the impact of macroeconomic shocks, they have important limitations and economists will have to continue to strive to make them more efficient and flexible. At the heart of the problem is the need to strike a delicate balance between expediency and sophistication, while creat- ing a model that is still workable for developing countries with sparse or outdated data, in a context were information is needed in order to take time sensitive policy decisions: Should a safety net program be expanded quickly to urban areas? Should employment subsidies be provided to �rms for hiring young workers? Are farmers in need of tempo- rary support through cheaper credit? An important concern is that of second order impacts. The method hinges on ex- trapolating the future from structural relationships between demographics, income, and employment, estimated from historical data. As a result, it is unable to account for what could happen as people change their behavior in response to (and cope with) a crisis. This may be less serious a problem than it seems—one can argue that these responses are in fact irrelevant, since the simulations are intended to help institute policies that minimize the need for people to change their behavior in the �rst place. Even so, it is an issue that merits careful thought. In order to �nd solutions to methodological and policy debates, economists need to take advantage of periods of relative calm, rather than having to deal with them in Foreword xi the midst of a crisis. Ideally, when a macro shock occurs, economists should be ready to respond with adequate analysis of impacts to inform policy measures by governments. The true measure of “success� of these efforts would be whether the predictions influ- ence the governments to put in place the right kind of policies that minimize the suffer- ing of people. In building such capacity, it is relevant to look ahead to what lies ahead in the im- mediate future for us at the Bank and the development community in general. Building on the experience gained from employing the model to multiple countries and se ings, a user-friendly simulation software is currently being developed by staff from the Pov- erty Reduction & Equity Group and the Development Research Group, as a new module of the ADePT platform developed by the Bank. Probably the �rst ever of its kind, this software would greatly enhance the usability of microsimulation to predict the distribu- tional impact of future shocks. At the same time, we see this to be not an end in itself, but rather a foundation to build on for the future. As the model is used more frequently, we expect to see further improvements, re�nements, and applications emerging organically from interactions with users, which will continue to improve the quality of analysis the Bank can provide to countries during future crises. The volume is organized as follows. Chapter 1 summarizes the methodology un- derlying the microsimulation model to predict distributional impacts of the crisis, along with several case studies that highlight how the model can be used in different country contexts. Chapters 2 to 4 are wri en by experts external to the Bank, two of whom par- ticipated as discussants at a workshop on the microsimulation work organized in May, 2010 at the World Bank headquarters. Chapter 2 comments on the broader implications and shortcomings of applying the technique described in Chapter 1 and the ability or willingness of governments to respond adequately to its results. Chapter 3 draws par- allels between the United States and developing countries to discuss the lessons that can be learned for mitigating the impacts of future crises. Chapter 4 discusses how the microsimulation approach can be sharpened to make it a be er tool for distributional analysis moving forward. Jaime Saavedra-Chanduvi Director, Poverty Reduction & Equity Poverty Reduction & Economic Management Network The World Bank CHAPTER 1 Assessing the Poverty and Distributional Impacts of the Financial Crisis with Microsimulations: An Overview of Country Studies Bilal Habib Ambar Narayan Sergio Olivieri Carolina Sánchez-Páramo T he �nancial crisis started in a few industrialized countries in 2008, but soon led to contractions in growth, employment and household income in both developed and developing countries. In many countries in the developing world, the growth contrac- tion led to increased poverty and lower income and non-income welfare. Global esti- mates by the World Bank suggest that by 2010, the crisis could have resulted in 89 mil- lion more people living below US$1.25 a day and 120 million more below US$2 a day, compared to a scenario of uninterrupted economic expansion without the crisis (Chen and Ravallion, 2009). Along with the impact on poverty, the crisis could have also had signi�cant impacts on the distribution of income and consumption within countries. This chapter summarizes the results and insights from simulation exercises to esti- mate the poverty and distributional impact of the �nancial crisis in selected developing countries.1 These simulations are based on approaches suggested in economic literature, notably Bourguignon et al. (2008) and Ferreira et al. (2008). The four countries studied here—Bangladesh, Mexico, the Philippines, and Poland—belong to different regions of the world, and they are also characterized by vast differences in terms of per capita income and other development characteristics, economic conditions prior to the crisis, and the ex- tent and nature of impact of the crisis.2 These countries all lack real-time data to measure the welfare impacts of the crisis, which necessitated simulations to estimate these impacts. The report has two distinct objectives. The �rst is to illustrate the kind of welfare im- pact that the crisis is likely to have had, depending on a range of country-speci�c factors including the size of the aggregate shock and the channels through which impacts are transmi ed. While four disparate case studies cannot represent the developing world, the insights they provide can serve as guides to the potential impacts of future macro- 1 2 World Bank Study economic shocks. The second objective is to highlight the methodological techniques and innovations needed for a simulation approach that can provide analytically sound and policy-relevant insights for a developing country in the absence of real-time micro- level data or complex general equilibrium models. We hope that this report and its sup- porting country case studies will provide a methodological road map, with its potential uses and caveats clearly spelled out, for those analyzing countries affected by similar shocks in the future, especially when they are confronted with data and time constraints similar to those faced here. The report also draws upon results obtained from other simulation exercises con- ducted by Bank teams in the context of the �nancial crisis (see Aran et al., 2010). These simulations were undertaken by Bank teams that needed to come up with the best ana- lytical �ndings possible, facing data constraints similar to ours. The report does not dis- cuss the results of these exercises in detail, but compares the methodologies adopted and insights obtained with those from our simulations.3 Financial Crisis and Its Impact on Poverty Estimating the welfare impacts of the �nancial crisis within developing countries has proved difficult for a number of reasons. The nature of the crisis has shifted rapidly, both across countries through trade, �nancing (credit and FDI) and remi ances, and within them through �scal policies and adjustments in domestic credit and labor markets. As the contagion spread, it took a different shape within each country, based on the relative importance of each of these factors. Moreover, evidence from previous crises suggests that extrapolations based on historical experience can be imperfect guides to estimating the potential impact of a macroeconomic shock. For example, there is evidence to sug- gest that the output elasticity of wages tends to be larger during downturns than dur- ing booms, and that relative inequality falls about as often as it rises during aggregate contractions (Paci et al., 2008). There is also evidence that the type of macroeconomic shock ma ers for welfare impacts and the pace of recovery: shocks that originate in the �nancial markets, for example, are often believed to result in a slower recovery of em- ployment and household incomes.4 That said, there is some consensus on how the impact of the crisis is likely to have spread among different groups in developing countries. The initial impact is thought to have been felt mostly by the emerging middle class in these countries, since they were more likely to be employed in export-oriented industries and salaried jobs in the services sector—parts of the economy that appear to have borne the brunt of the impacts initial- ly.5 Labor markets in the formal sector were affected by declining demand for domestic goods and exports, falling commodity prices and reduced availability of credit for �rms. Approximately 90 percent of households entering the middle class during 1990–2005 joined the lower tier of consumption (Ravallion, 2009), and thus risked falling back into poverty as a result of large employment or earnings shocks. On the other hand, the im- pact on the poor in developing countries is likely to have been limited initially, due to the very isolation from the global markets (and the formal sector) that had prevented them from exiting poverty in the past. But as the contagion spread from the formal to the informal sector through the backward and forward linkages between different parts of the economy, the poor are likely to have been affected as well. Knowing, When You Do Not Know 3 The poor in developing countries could have been impacted by the crisis through a number of channels. Shocks to employment and wages in the formal sector are likely to have pushed more people into the informal sector, which in turn would have depressed earnings and crowded out employment. Shocks to urban manufacturing and services may also have led to reverse migration from urban to rural areas, increasing the burden on poor rural households and drying up remi ances from workers previously employed in the formal sector. Given that these dynamics would have occurred in country contexts where governments were in �scal stress due to falling revenues—and therefore had li le room to react with appropriate policy measures—the effects could have been deep and long lasting.6 Evaluating the Impact of the Crisis As the discussion so far highlights, the impact of the crisis on income distribution and poverty within each country is likely to have been complex and dynamic. Given these complexities, an analysis of the impacts must, at a minimum, address the following is- sues: (i) which sectors, areas and/or regions are most likely to be impacted and in what way; (ii) how sectoral and regional impacts translate into impacts across the income or consumption distribution; (iii) what are the characteristics of those who will become poor as a result of the crisis; and (iv) what are the implications of how these impacts are distributed for the design of policy responses, particularly those that seek to provide safety nets to the affected. Moreover, in order to usefully inform policy responses to the crisis, the above ques- tions would have to be analyzed ex ante, without the bene�t of micro data that capture actual impacts. In developing countries, real-time data from the crisis period is rarely available. Thus in most cases impacts must be estimated based on data that predate the crisis, rather than delaying the analysis until post- or during-crisis data becomes available—usually too late to inform policy measures that governments need to respond with during a crisis. Even in countries where some real-time data is available from crisis- affected sectors or regions, an ex ante approach using historical data can be useful to simulate future impacts for hypothetical scenarios that are not available from real-time data—for example, to compare different alternative scenarios that may unfold in the fu- ture. This is especially true for countries where the situation is changing rapidly, as was the case for many countries when the �nancial crisis almost overlapped with the food and fuel price crisis of 2008. Finally, the method for assessing poverty and distributional impacts of a crisis must be able to account for multiple channels through which the impacts can be transmi ed to households and individuals, and identify the relative importance of these channels in a given country context. It must do so at the individual and household levels over the entire income/consumption distribution. A number of different approaches have been used in the economic literature and by development institutions, including the World Bank, to estimate ex ante the impact of a crisis on household incomes and poverty. A commonly used approach involves estimating an output elasticity of poverty, in which historical trends of output and poverty are used to determine the responsiveness of poverty rates to growth in output, which is then combined with macroeconomic projections to estimate the impacts of reduction in 4 World Bank Study future growth on poverty. Although this method is easy to apply, it only provides ag- gregate poverty (or at most, sectoral or regional) impacts and very li le information on how the impacts are likely to be distributed among different groups or sub-populations, which is critical for policy design. Other approaches, used in a few middle-income coun- tries, involve using microsimulation methods that combine computable general equilib- rium (CGE) models with the predictions from behavioral regressions built on pre-crisis household data to simulate household-level impacts across the entire income/consump- tion distribution.7 The approach adopted here is best seen as a compromise between “aggregate� ap- proaches that rely on growth-poverty elasticities and complex macro-micro simulation approaches that harness the power of general equilibrium models and household data. The compromise involves combining the behavioral estimations from pre-crisis house- hold data with aggregated macroeconomic projections. This leads to a model that is leaner than the typical macro-micro simulation models, takes less time to compute, and above all is applicable in countries where CGE models are either unavailable, outdated or of poor quality. In contrast to CGE models, aggregate macroeconomic projections— such as those for national, sectoral, or regional GDP and remi ance flows—are available for most countries with which the Bank and/or the International Monetary Fund (IMF) has an ongoing dialogue. Compared to the simple elasticity-based approach outlined above, this approach has the main advantage of being able to generate estimates for individuals and households all along the distribution with and without the crisis, which can be used for detailed poverty and distributional impact analyses. Aran et al. (2010) use an approach that is broadly similar with ours, in terms of combining simulations with household data and macro projections, while differing in how household behavior is modeled and linked to macro projections. The Methodology to Estimate Impacts Ex Ante8 The approach adopted here involves superimposing macroeconomic projections on be- havioral models built on pre-crisis micro data. The model is based on the approaches described in Bourguignon et al. (2008) and Ferreira et al. (2008)—with the important simpli�cation of omi ing the CGE component. Instead, the behavioral models are linked to aggregate and sector-level macroeconomic projections for a speci�c country and year, and the micro-level (individual and household) snapshot of future impacts is extrapo- lated from these projections. The model focuses on labor markets and (international) remi ances as the main channels through which the macro-level shocks impact individuals and households. It allows for shocks (negative or positive) to labor income—modeled as employment shock, earnings shock or a combination of both—and shocks to non-labor income from changes in remi ance flows. Most of the changes in total income are captured, since labor income and remi ances account for a signi�cant proportion of household income in most developing countries. Reasonable assumptions are made about impacts on other sources of income, such as capital income, domestic remi ances, rents, or public trans- fers. These assumptions can reflect speci�c information about impacts on these sources of household income, depending on what information is available for a certain country. Knowing, When You Do Not Know 5 How the Approach Works Using pre-crisis micro data and macro projections for the crisis period, income distribu- tions can be predicted at the individual and household levels for each year for which macroeconomic projections are available. The data requirements are fairly straight- forward. At the macro level, projections are needed on: (i) output and employment by sector and remi ances, (ii) population growth, and (iii) predicted price changes. At the micro level, a household survey is essential, with information on: (i) household income and consumption, (ii) household and individual characteristics, and (iii) individual-level labor force and employment status and earnings. Macroeconomic projections for outputs (sectoral and national) are almost always available, while employment projections are usually unavailable. Thus an important ini- tial step in the simulation exercise is to translate output growth estimates into employ- ment changes at the sector level, for a speci�c crisis year and macroeconomic scenario. This is typically done using sectoral output-employment elasticities (estimated from historical data), which can then be applied to the output growth projections to generate changes in employment by sector. In Mexico, for example, the elasticities for 2010–11 are estimated using information on sectoral GDP and employment changes during 2003–09 and applying a simple correlation model between both series. Once the employment projections are generated, the simulation process is accom- plished in three main steps (Figure 1.1). The �rst step consists of using household data for the latest available pre-crisis year to estimate behavioral models on employment status (whether an individual is employed or not, and the sector of employment) and earnings for individuals, as a function of their individual and household characteristics.9 Figure 1.1: Scheme of the modeling process Baseline (Calibration) Simulation Assessment of impacts Input Micro data Macro projections Price data estimate adjust predict ∆ in LF status (ind) LF status model ∆ real earnings (ind) Income and Output Earnings equation ∆ remittances (HH) consumption Migration/remittances (individuals and HH) Population Results Population growth Income/consumption distributions Rule: Replicate macro Poverty and inequality Rule: Best fit to micro proportional changes at measures data micro level Source: World Bank. 6 World Bank Study The second step consists of replicating the macro-level changes in output and em- ployment (actual or projected) between baseline and crisis years on the household data of the baseline year. This process uses predictions of behavioral models estimated on the pre-crisis household data (as in step 1 above) and generates household and individual- level predictions for employment, earnings and remi ances for speci�c crisis years and macroeconomic scenarios. Since an individual’s labor income depends on his/her em- ployment status and labor earnings, how the output shock in a particular sector is ap- portioned between employment shocks, earnings shocks, and adjustments across sectors depends on how responsive (elastic) employment in that sector is to output changes. It also implies that at the household level, the extent of the impact depends on the size of the aggregate shocks at the sector level and the demographics and characteristics of household members, which influence the labor force status and earnings of household members after the shock. The simulations are also adjusted for population growth, using official population projections (disaggregated by gender and age groups) to fully account for demographic changes that would affect the size and composition of the labor force and, through that, the estimates of per capita household income. This is done by re-weighting households in the baseline data to replicate demographic changes predicted by population projections. To simulate changes in non-labor income, projections of aggregate changes in remit- tances are linked to pre-crisis remi ance information from household data using a simple assignment rule to ensure that the total change in remi ances received by households is equal to the projected change in remi ances from the macro data.10 Some components of non-labor income are assumed to grow at the rate of aggregate GDP for the relevant pe- riod (pro�ts, rents and domestic remi ances), while others are kept constant in real terms at the baseline year level or changed in accordance with speci�c information received for each country (social bene�ts, pensions or other transfers, depending on the country). The steps outlined above yield a projected income distribution corresponding to a speci�c crisis year and macroeconomic scenario. In the case of Bangladesh, where pov- erty is de�ned in terms of consumption, household incomes are transformed into pre- dicted consumption levels using the average propensity to consume for each household from the baseline year. The �nal step consists in using price projections to adjust the poverty line to reflect the difference in food and non-food inflation rates between base- line and crisis years. Since the poverty line is typically anchored to a food basket that en- sures a minimum calorie intake, for countries where food inflation is expected to be sig- ni�cantly different from general inflation between baseline and crisis years, the baseline poverty line would not be enough for a household to meet the basic food requirements in the crisis year.11 For the countries where food prices are a concern of policy makers (Bangladesh, the Philippines, and Mexico), a small upward adjustment is made to the poverty line to ensure that the same food basket remains affordable.12 These poverty lines are then compared with predicted household incomes/consumption to generate poverty estimates for each crisis year and scenario. De�ning the “Impact�: Two Types of Countries The impact of the crisis on poverty indices and income distribution can be measured in two ways: a “before-after� comparison of the change in welfare status in a country between pre-crisis and crisis years, and a “with-without� comparison of the difference Knowing, When You Do Not Know 7 between with-crisis and without-crisis scenarios for the same year. It turns out that a cer- tain type of comparison is more meaningful for each country, depending on the nature of macroeconomic impact the country is likely to suffer from the crisis and the interests of counterparts in Bank teams working in the country and the government. The countries we study are divided into two groups. The �rst, referred to as Type I countries, comprises Bangladesh, the Philippines, and Poland. In these countries, aggre- gate GDP is expected to grow during the crisis, but with a lower trajectory than in the absence of the crisis. As a result, employment and household income are also expected to grow, and the simulation identi�es the “losses� incurred by households from lower output growth due to the crisis. In order to determine the impact of the crisis in Type I countries, macroeconomic projections for two scenarios (“non-crisis� and “crisis�) are used to generate two income distributions for the same crisis year. For a given crisis year, the income distribution and poverty estimates of the crisis scenario is compared with those for the counterfactual “benchmark� scenario to determine the impact of the crisis. Type II countries are de�ned as those that are expected to experience negative GDP growth as a result of the crisis. In general, this description applies to countries—usually middle-income—with close market links to developed countries that have entered into a recession due to the crisis. For a country like Mexico, which has experienced a contrac- tion in GDP, policy makers and the wider development community are more interested in comparing poverty and income distribution before and after the crisis, rather than comparing the crisis scenario with the counterfactual (for the same year) of what would have happened without the crisis. Therefore, for Mexico we compare the projected in- come distributions for 2010 and 2011 with the actual income distribution of 2008. Figure 1.2 shows the difference in how the impact of the crisis is de�ned for Type I and Type II countries. Since the de�nition of “impact� is different for the two types of countries, the results of our simulation should be read differently for the two types of countries as well.13 Figure 1.2: Impacts in Type I and Type II countries Baseline Macro projections Predictions (Pre-crisis) (2010) (2010) “Treatment� I) PE (Crisis) (TY act Imp Impact (TYPE II) Micro data (Household/LF survey) Benchmark Source: World Bank. Note: 1. For Mexico and Egypt we use a before/after crisis comparison to determine “impact.� 2. For Bangladesh, the Philippines, and Poland we construct a “benchmark� case and use a benchmark/ crisis comparison. 8 World Bank Study Simulating the Impact of Policy Changes For Poland and Mexico, the simulations also examine the potential impact of policy changes, speci�cally expanding existing safety net programs. While social transfers were held constant in real terms (at the baseline year level) for all countries in conducting the simulations, additional scenarios incorporating expansion of speci�c social programs were considered for Poland and Mexico, since these countries were either implement- ing (Poland) or considering (Mexico) changes in safety net programs. The results of the policy analyses, summarized in the next section, illustrate the use of microsimulations in projecting the impacts of social policy changes. In the case of Poland, our analysis also provides insights into the relative roles played by two different types of social programs in mitigating poverty and distribution- al effects of the crisis. An unemployment insurance program has acted as an “automatic stabilizer� in Poland that (by design) expands automatically as unemployment increases due to the crisis. Given that it operates in a counter-cyclical manner without any deliber- ate policy shift, the projected impact of the crisis on household incomes and poverty in Poland takes into account the mitigating impact of unemployment insurance. This adds a useful dimension to the simulation results: one can estimate the role that a commonly used stabilizer like unemployment insurance can play in mitigating the crisis impacts, and compare these projections against the simulated effects of the expansion of a target- ed social transfer program. While these �ndings apply to Poland, they offer insights that may be relevant for other countries, especially those with comparable economic charac- teristics and a similar menu of potential policy responses to macroeconomic shocks.14 Caveats to the Approach The caveats to this methodology relate to issues relevant for microsimulation approach- es in general, as well as those that arise from the modi�cations we adopt. Below we note some of the most signi�cant caveats. First, the quality of projections from the model depends on the nature and accuracy of the data underpinning the exercise. While applicable to all simulation approaches, this caveat is particularly important here given our reliance on aggregate macroeconom- ic projections. Moreover, while macroeconomic projections are available for aggregated sectors like industry, services, and agriculture for all four countries, the lack of more disaggregated projections constrains our ability to account for heterogeneity within each sector. A distinction between formal and informal sectors or tradeable and nontradeable sectors within industry or services, for example, is not possible because the projections are not disaggregated in that form.15 Second, the approach relies on behavioral models built on past data that reflect the pre-crisis structure of labor markets, household incomes and their relationships with household and individual characteristics. These structural relationships are assumed to remain constant over the period for which the projections are made, without allowing for any changes over time or between crisis and non-crisis scenarios. The more distant the baseline year is, the more questionable this assumption is likely to be.16 This ca- veat, which applies to microsimulation approaches in general, also links back to the constraints imposed by data availability (see above). Knowing, When You Do Not Know 9 Third, our approach assumes that changes in labor market conditions are propor- tional to the projected change in outputs, based on the estimated past relationship be- tween output and employment. This implicitly assumes stable relationships between output, demand for labor, and labor earnings, which may not hold due to distortions (such as segmentation and downward stickiness of nominal wages) that typically exist in the labor market and are likely to affect adjustments during a crisis.17 Fourth, the model does not allow for geographic factor mobility (labor or capital) across time or scenarios, even as workers experience a change in sector or status of em- ployment. While this assumption is an abstraction from truth, it is likely to ma er only when the results are disaggregated spatially or across rural and urban areas. The model does incorporate changes in domestic remi ances from urban to rural areas, so that the lack of factor mobility does not necessarily imply constant income flows across space. Fifth, our approach is limited in its ability to account for shifts in relative prices be- tween different sectors of the economy as a result of the shock. While the poverty impact of shifts in the price of food relative to other prices is taken into account, other potential sources of price impacts are ignored, such as the general equilibrium effect of a change in the terms of trade between agriculture and other sectors. In the absence of a CGE model, it is nearly impossible to explicitly model for such changes. Sixth, in taking into account the channels through which output shocks are trans- mi ed to households, our approach must predict individual and household-level in- comes. But to provide poverty projections for a country that de�nes poverty in terms of consumption (for example, Bangladesh), household incomes must be converted into consumption, using the (questionable) assumption that a household’s propensity to con- sume is constant over time and across crisis and non-crisis scenarios. In the absence of more information, however, this assumption has at least the advantage of being simple and transparent.18 The �nal limitation is related to the validity of the results of the simulations when compared against the actual changes when they occur, and applies to all ex-ante ap- proaches. The only true validation would be to combine ex-ante and ex-post analysis (see Bourguignon and Ferreira, 2003). Since ex post data for crisis years will not be available in most countries for some time, some uncertainty about the validity of the simulations is bound to remain.19 As we await the availability of such data, a limited validation exer- cise undertaken for one country (Bangladesh) using historical data suggest that with the bene�t of reliable macroeconomic data and output-employment elasticities, the model yields results close to what is observed from actual household data. Results from Micro-Macro Simulations for Four Countries Some common pa erns are visible in all four countries studied in this report. However, the manner in which the crisis is transmi ed to each country differs, which means that the size of the impact is different in each context, as are the pro�les of those who bear the brunt of the contraction. In Bangladesh, the Philippines, and Poland, lower GDP growth is expected to result in the percentage of the population in poverty being 1.2, 1.5, and 0.4 percentage points higher in 2010, respectively, compared to a no-crisis scenario. In Mexico, GDP contracted 10 World Bank Study by nearly 7 percent in 2009 and is projected to return to pre-crisis levels only by 2011. Mexico’s poverty rate is projected to rise by more than 3 percentage points between 2008 and 2009, and even as growth resumes, the recovery in poverty is expected to be slow. As a result, more Mexicans are projected to be poor in 2011 than in 2008. The poverty impact in Mexico for 2009 is comparable to what was projected by Aran et al. (2010) for Latvia, where GDP was expected to contract by 18 percent between 2008 and 2009, increasing the poverty headcount rate from around 14 percent in 2008 to 20 percent in 2009. There is li le or no impact on aggregate measures of inequality in any of the coun- tries. The aggregate numbers, however, do not tell the complete story of how impacts are distributed. More detailed analysis, using a number of analytical devices, shows that some regions and groups within each country are expected to suffer more losses than others, depending on which sectors and income sources are more affected. While the impacts on income and consumption are relatively large among middle- income households, the poor in some of the most affected countries are severely im- pacted as well. Ravallion (2009) has argued that middle-income groups in developing countries are more exposed to a macroeconomic crisis than the rest of the population. Simulations for individual countries partly con�rm this view: between 15 and 20 percent of households in the 40th–80th percentiles of the income distribution in Mexico and the Philippines suffer per capita income losses that push them to a lower income decile, which suggests that households in the middle of the distribution suffered a sizeable im- pact in absolute terms. At the same time, signi�cant impacts are also estimated among poor households relative to their income levels, particularly so in countries where the crisis has been more severe. The poorest 20 percent of Mexican households are projected to suffer an average loss in per capita income of about 8 percent between 2008 and 2009, compared with a 5 percent income loss for the entire population. Similarly, the simulations by Aran et al. (2010) suggest that in Latvia, the bo om 40 percent of the income distribu- tion are hardest hit by the crisis—although the very poorest households in rural areas ap- pear to have been spared somewhat, likely due to their isolation from the formal economy. The impacts on the middle of the distribution can be a ributed primarily to signi�- cant employment shocks to manufacturing and service sectors that employ a large num- ber of workers from middle-income households. Urban households suffer more losses, on average, than do rural households in all countries because the manufacturing sector that bears the brunt of the losses in employment and labor earnings is concentrated in urban areas. In the Philippines, for example, per capita income of urban households is projected to decline by 6 percent, compared to a fall of 3 percent for rural households. Beyond these common pa erns, there is considerable variation among countries in how impacts are distributed within countries because of the differences in transmission channels and the structure of economic sectors. Approximately 90 percent of the aver- age loss in per capita household income in Mexico and the Philippines are a ributable to a fall in labor income, compared with 50 percent of the average loss in Bangladesh, where an expected slowdown in remi ance growth is also important. Losses resulting from lower remi ances tend to be more concentrated among be er-off households than losses in labor income, mainly because the initial (pre-crisis) distribution of remi ances was skewed towards be er-off households. This is an important reason why projected income losses in the Philippines are more evenly distributed across the income distribu- tion than in Bangladesh. Knowing, When You Do Not Know 11 The crisis may have also impacted women and men differently. In a world of highly mobile capital and less mobile labor, often the brunt of adjustment to shocks—whether external or internal in origin—falls on labor earnings and hence directly on households (Paci, Revenga, and Rijkers, 2009). Women and men tend to work in different indus- tries and occupations, which may also result in lower productivity among women on average. It is also commonly believed that women tend to hold more precarious jobs, operate in smaller and less capitalized �rms, or engage in more vulnerable economic activities than men. These factors would suggest that women are more likely to be af- fected by economic shocks. The existing empirical evidence is mixed, however. Using data from the Asian crisis, Hallward-Driemeier, Rijkers, and Waxman (2010) show that female employment was not hit the hardest, primarily due to the fact that women were disproportionately employed in �rms more resilient to the crisis. On the other hand the gender earnings gap increased, particularly in larger �rms, implying that the smaller net employment impact on women came at the cost of a larger reduction in earnings. Our work shows that impacts of the �nancial crisis were quite heterogeneous across countries and sectors, with no common pa erns by gender and no evidence that women were more affected than men. Other evidence suggests that gender impacts vary sig- ni�cantly by country and are hard to generalize (Sabarwal, Sinha and Buvinic, 2010; Newhouse and Cho, 2010; Turk and Mason, 2009). On the one hand, relatively high female participation rates in export-oriented industries that were the initial casualties of the crisis may have led to strong �rst-round negative employment effects. On the other hand, lower female participation rates in pro-cyclical sectors, such as construction, or industries dependent on external demand, such as tourism, implied that the aggregate employment effects for women relative to men were muted once these second-round effects are taken into account. The direction of impacts and their gender differences, as well as the channels through which they occur, are likely to depend on the speci�c na- ture of the crisis at hand, and merit further analysis. The simulations also allow us to examine the characteristics of those who are ex- pected to be in poverty as a direct result of the crisis—information that can be of use to policy makers. We �nd that the “crisis-vulnerable,� namely households who would not have been poor had there been no crisis, tend to have speci�c characteristics that distin- guish them from the “structurally poor� who would have been poor with or without the crisis. On average, the crisis-vulnerable appear to be more educated or skilled than the structurally poor. They are also more likely to be economically active than the structur- ally poor, indicating that the crisis would have had a sizeable effect on the number of “working poor.� In Mexico, the crisis-vulnerable are more likely to be urban, employed in the manufacturing sector prior to the crisis, and belong to households headed by women compared to the chronic poor and non-poor. In other countries, the crisis-vul- nerable fall somewhere between the structurally poor and the non-poor in terms of these characteristics. Social programs can play a role in determining the characteristics of the crisis-vul- nerable in some countries. For example, although most of the impact on household in- come in Poland is due to losses in labor income, our estimates indicate that a large part of this loss would have been compensated by higher unemployment insurance bene�ts, which would have reduced the number of crisis-vulnerable among households with formal sector workers.20 In Turkey, Aran et al. (2010) �nd that even though the largest 12 World Bank Study employment impacts are in the middle of the income distribution, most of the welfare losses come from the informal sector where poorer workers are more likely to be em- ployed. Formal sector workers, by contrast, are protected to some extent by unemploy- ment bene�ts. The simulations also suggest that expansion in targeted social transfers can mitigate some of the impacts on the poorest, depending on the size of the expansion and the size and characteristics of the target population of the program. In Mexico and Poland, for example, expansion of targeted social transfer programs is likely to have bene�ted a sizeable number of the poorest, in spite of the small impact of the expansion on national poverty rates. Below we discuss the results of the simulation in more detail, differentiating be- tween Type I and II countries, as de�ned earlier. Results for Type I Countries The Philippines, Bangladesh, and Poland show signi�cant heterogeneity in terms of the predicted impact of the crisis both at the aggregate level and by sector (Table 1.1). Total GDP in 2010 is predicted to be 4.9 percent lower under the crisis scenario in the Philip- pines, 2.2 percent lower in Bangladesh, and 5.3 percent lower in Poland than what it would have been without the crisis—according to projections made by World Bank coun- try economists in early 2010.21 The manufacturing sector is projected to be the most af- fected in terms of output losses in all three countries, followed by services. International remi ance flows are projected to be 1.5 percent lower under the crisis scenario in Philip- pines, 3.7 percent lower in Poland, and 10.1 percent lower in Bangladesh than what they would have been without the crisis. Table 1.1: Macroeconomic impacts for Type I countries Philippines Bangladesh Poland Total GDP −4.9 −2.2 −5.3 Agriculture −2.6 0.3 5.2 Manufacturing −6.7 −3.9 −10.1 Services −4.6 −2.1 −3.4 Remittances −1.5 −10.1 −3.7 Source: World Bank. Note: 1. Impact calculated as % difference between 2010 projected crisis GDP (or remi ances) and 2010 pro- jected benchmark GDP (or remi ances). A negative number indicates lower GDP (or remi ance) due to crisis. 2. These �gures are based on projections (not actual estimates) of GDP, as of early 2010 by World Bank country economists. Using information on aggregate and sectoral output elasticities of employment for each country, projected changes in GDP can be translated into projected changes in total and sectoral employment (Figure 1.3). Employment losses are consistent with overall outputs shocks, with Poland suffering the most on both macro and micro fronts, and Bangladesh the least. At the sectoral level, however, services account for the bulk of employment losses in all countries but Bangladesh, even though manufacturing is pro- jected to suffer the largest output shock in all three countries. In other words, for two out of the three countries the output elasticity of employment in the service sector is higher than that in the manufacturing sector. Knowing, When You Do Not Know 13 Figure 1.3: Percent change in sectoral employment between benchmark and crisis (2010) Phillippines Bangladesh Poland 0.5 0 Percent change −0.5 −1 −1.5 Agriculture Manufacturing Services −2 Source: World Bank. Not surprisingly, household income is projected to fall in all three countries as a consequence of reductions in both labor and non-labor income (Figure 1.4). The relative importance of shocks to each of these income sources varies by country, however, reflect- ing variation in the role of different transmission mechanisms and policy environments. For example, Poland suffers the lowest shock to total household income, despite sizeable shocks in both labor income and remi ances. This is because of inbuilt stabilization by unemployment insurance, which protects income (by increasing non-labor income) in the face of employment losses. Bangladesh, which is relatively insulated from employ- ment losses, experiences a larger shock to household income than Poland, mainly due to lower remi ances that translate into lower non-labor household income. In the Philip- Figure 1.4: Percent change in household income between benchmark and crisis scenarios (2010) Total HH Income HH Labor Income HH Remittances −1 −3 Percent change −5 −7 Philippines Bangladesh Poland −9 Source: World Bank. 14 World Bank Study pines both labor and non-labor household income are projected to fall as a consequence of the crisis.22 Both Bangladesh and the Philippines lack a large unemployment insur- ance scheme that could act as a stabilizer against loss in labor income. Losses in household income translate to signi�cant increases in both the level and depth of poverty in all three countries compared to the non-crisis scenario (Figure 1.5). The Philippines is projected to experience the largest percentage increase in the poverty headcount and in the poverty gap, which measures the mean distance from the poverty line (that is, the “depth� of poverty), while Poland is shielded from large increases in poverty by the protection provided by unemployment insurance schemes. Although in- come losses are larger in the Philippines, the poverty increase is actually slightly larger in Bangladesh. This could indicate that those impacted in Bangladesh are nearer to the poverty line than their Filipino counterparts. There is, however, li le impact on aggre- gate inequality in any of the countries. Aggregate macroeconomic and average microeconomic impacts on employment, income and poverty hide a signi�cant amount of heterogeneity across households and individuals. We examine this issue in more detail in two ways. First we compare those households that are projected to be poor in the crisis scenario and would not be poor in the hypothetical no-crisis scenario in 2010 (the “crisis vulnerable�) with others in the population. For this purpose we distinguish between structurally poor households— households observed to be poor in the baseline year and projected to remain poor under both the benchmark and crisis scenarios in 2010—and the average household. Second, we examine how shocks are distributed along the entire income distribution, irrespec- tive of whether households are projected to become poor as a consequence of the shock. The most notable characteristic about the crisis-vulnerable is that they suffer income losses ten times greater (in percentage terms) than those of the general population (Fig- ure 1.6). Interestingly, the relative weights of labor and non-labor incomes within the overall income loss are unchanged. The changes are driven primarily by labor income in the Philippines and Poland, and by non-labor income in Bangladesh. The fact that such large income losses are responsible for increasing poverty implies that people are falling Figure 1.5: Percent change in poverty/inequality indicators between crisis and benchmark Poverty Headcount Poverty Gap Gini 8 6 Percent change 4 2 0 −2 Philippines Bangladesh Poland Source: World Bank. Knowing, When You Do Not Know 15 Figure 1.6: Percent losses in income for the general population (left) and the crisis-vulnerable (right) Philippines Bangladesh Poland Philippines Bangladesh Poland 20 20 10 10 Percent losses Percent losses 0 0 −10 −10 −20 −20 −30 −30 −40 −40 −50 −50 Labor Income Non-Labor Income Source: World Bank. into poverty from parts of the distribution that are not close to the poverty line. This is most likely driven by the urban middle class, which is mostly engaged in the hardest-hit sectors of manufacturing and services. In addition, crisis-vulnerable households have “be er� characteristics than structur- ally poor households, but “worse� than those of the average household. Crisis-vulnerable households are likely to be more urban than the structurally poor in all three countries, and more rural than the general population in Poland (Figure 1.7). In the Philippines, 57 percent of crisis-vulnerable households live in rural areas, compared to 74 percent of structurally poor households and 51 percent of the national population. In two out of three countries, crisis-vulnerable households are smaller and with lower dependency ratios than their structurally poor counterparts. In Poland, however, structurally poor households are about the same size as crisis-vulnerable households, and all household types have almost the same (very low) dependency ratios. Figure 1.7: Percentage of poor and crisis-vulnerable households in rural areas 100 80 60 Percent 40 20 0 Philippines Bangladesh Poland Crisis-vulnerable Structurally poor Entire population Source: World Bank. 16 World Bank Study Figure 1.8: Percentage of households with low-skilled heads 100 80 60 Percent 40 20 0 Philippines Bangladesh Poland Crisis-vulnerable Structurally poor Entire population Source: World Bank. The characteristics of crisis-vulnerable household heads are also signi�cantly differ- ent from those of other groups in the population. In all three countries, crisis-vulnerable household heads are relatively more skilled than those of structurally poor households, but less skilled than those of the average household (Figure 1.8). This is consistent with the pa ern of middle-income households suffering the largest impacts (in absolute terms) of the crisis, since they are more likely to be urban and employed in the formal sector (and therefore more skilled) than household heads among the structurally poor. In addition, heads of crisis-vulnerable households are younger on average and more likely to be male than household heads among the general population. We turn now to the distributional impact of the crisis by examining how income losses are allocated across different households in urban and rural areas. For this pur- pose, we map out income losses between the benchmark and crisis scenarios at every percentile in both the urban and rural income distributions using what are called growth incidence curves. We order households according to their pre-crisis per capita household income level (from lowest to highest), group them into income percentiles (as de�ned in the benchmark case), and plot the average percentage loss in per capita household income by percentile. We perform this exercise for all households as well as for rural and urban groups, which allows us to compare percentage income losses across households within the group. Note, however, that comparisons across rural and urban groups are not straightforward because income percentiles are group-speci�c rather than popula- tion-speci�c. For instance, given that on average income levels are higher in urban than in rural areas, a household in the 30th percentile of the urban income distribution is likely to be signi�cantly be er off than a rural household in the 30th percentile. In all three countries, income losses associated with the crisis tend to be larger in urban compared to rural areas (Figure 1.9). This is consistent with employment losses being relatively concentrated in urban areas, where manufacturing and service sector jobs tend to cluster. In addition, there are important differences across countries in dis- Knowing, When You Do Not Know 17 Figure 1.9: Growth incidence curves: percent change in income (relative to the benchmark) due to crisis a. The Philippines b. Bangladesh 3 3 2 2 1 1 0 0 % Change % Change −1 −1 −2 −2 −3 −3 −4 −4 −5 −5 −6 −6 −7 −7 −8 −8 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 Income per capita percentile Expenditure per capita percentile c. Poland 3 2 1 0 −1 % Change −2 −3 −4 −5 −6 −7 −8 0 10 20 30 40 50 60 70 80 90 100 Income per capita percentile Urban Rural Source: World Bank. Note: The horizontal axis shows percentiles of per capita income (consumption in case of Bangladesh) in the benchmark (without crisis) scenario. Vertical axis shows percentage change in per capita income (consumption for Bangladesh) between the benchmark and crisis scenarios for the same year, for every percentile (1 percent) of the distribution. tributional impacts within urban and rural areas, because of difference in the sources of impact across countries. In the Philippines, the main impact is felt on labor income in the manufacturing and services sectors. Since these income sources are important for most urban households, the impact is more or less uniformly distributed across the urban distribution (Figure 1.9a). In Bangladesh the loss in non-labor income due to a decline in international remi ances is an important source of impact. Since remi ances are more skewed towards the be er-off among rural households than among urban households, the impact is skewed towards the middle and upper ends of the distribution for rural households and more uniformly distributed for urban households (Figure 1.9b). In Po- land, the main impact is felt on non-agricultural labor income, which results in the losses being distributed quite uniformly across urban households; at the same time the growth of the agriculture sector in the wake of the crisis absorbs much of the employment loss from other sectors, thereby raising the incomes of poor rural households (Figure 1.9c). Growth incidence curves provide information about income losses incurred by the average household within each percentile of the income distribution, and are thus useful in identifying what income groups suffer relatively larger or smaller losses. However, they can hide a signi�cant amount of heterogeneity in the absolute size of impacts, even 18 World Bank Study among households with very similar pre-crisis per capita household income levels. An alternative way of evaluating the distributional impact of the crisis is to examine the size of the income shock suffered by households in each income group. We do this by focus- ing on income shocks large enough to push households to a lower decile, as de�ned by the benchmark distribution.23 The results con�rm that those in the middle of the distribution are the hardest hit in terms of absolute level of income change (as opposed to percentage change, as with the growth incidence curves) in all three countries (Figure 1.10). Although most households tend to stay in the same decile, approximately 15 percent of all households suffer income losses that push them to a lower decile. Moreover transitions downwards are concentrat- ed in the middle of each of the distribution, indicating that middle-income households are most likely to suffer large income losses due to the crisis. Movements up are gener- ally rare, except in Poland where safety nets (particularly unemployment insurance) protect those at the bo om of the distribution and move them up to a different decile. In sum, in all three Type I countries considered here, the crisis would have resulted in signi�cant output, employment and household income losses, compared to a hypo- thetical no-crisis scenario. As a consequence, the incidence and depth of poverty are projected to increase. Moreover households predicted to become poor as a consequence Figure 1.10: Transitions across deciles of per capita income (percent of households in each decile) a. Philippines b. Bangladesh 100 100 80 80 % values % values 60 60 40 40 20 20 0 0 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 Deciles of per capita income Deciles of per capita income c. Poland 100 80 % values 60 40 20 0 1 2 3 4 5 6 7 8 9 10 Deciles of per capita income Stay in same decile Movement up Movement down Source: World Bank. Knowing, When You Do Not Know 19 of the crisis appear to be different from both structurally poor households and the aver- age household in terms of both their pre-crisis income levels and their characteristics. Finally, households in the middle of the income distribution are projected to bear the brunt of the adjustment in absolute terms. We turn now to the analysis of impacts in countries with projected absolute GDP losses in 2009 and 2010 (Type II countries), using the case of Mexico as an illustration. Results for Type II Countries Mexico suffered signi�cant total and sectoral GDP losses in 2009 and is projected to re- turn to positive albeit modest economic growth in 2010–11 (Figure 1.11). Given these pat- terns, we distinguish between a crisis period (2008–09) and a recovery period (2010–11) in the discussion that follows. Remi ances are expected to rise during the crisis and then return to pre-crisis levels during the recovery. The predicted countercyclical behavior of international remi ances assumes increased efforts by migrants abroad, in the form of higher transfers, to support their families in times of economic distress. As before, we use estimated values of the output elasticity of employment to calcu- late projected employment impacts in 2010 and 2011, using actual employment data for 2009. The size of the labor force is predicted to grow over time as a consequence of popu- lation growth, but labor force participation rates remain relatively constant during 2009- 2011 at around 66 percent (Table 1.2). In contrast, employment levels declined in 2009 and are projected to recover only slightly in 2010 and 2011. The combination of stable participation rates and stagnant aggregate employment levels during the crisis trans- lated into an increase in the unemployment rate from 4.5 percent in 2008 to 6.2 percent in 2009, although this trend subsides as the market absorbs more workers by 2011. Sectoral employment shares remain stable between 2008 and 2011, except for small drops in the shares of agriculture and manufacturing that are absorbed by the services sector. Figure 1.11: Macroeconomic projections for Mexico (2008–09 are actual observations; 2010–11 are forecasts) 8,800 Agriculture Manufacturing 8,600 Others Industries Services 8 Real GDP (billions of pesos) 8,400 6 % change in output for sector 4 8,200 2 8,000 0 −2 7,800 −4 −6 7,600 −8 −10 7,400 −12 2008 2009 2010 2011 2008 2009 2010 2011 Source: World Bank. 20 World Bank Study Table 1.2: Employment projections (population of age 15+ yrs) Actual Projected 2008 2009 2010 2011 Employment status (millions) Inactive 23.3 23.8 24.1 24.7 Active 44.5 45.4 46.1 47.2 Unemployed 2.0 2.8 2.8 2.4 Employed 42.5 42.5 43.3 44.8 Labor force participation rate 65.7 65.6 65.6 65.6 Unemployment rate 4.5 6.2 6.0 5.1 Sectoral shares (% of employed in each sector) Agriculture 12.0 12.1 11.7 11.2 Manufacturing 16.6 16.5 16.4 16.2 Other industries 9.8 9.8 9.8 9.8 Services 61.6 61.6 62.1 62.7 Source: National Labor Force Survey (Encuesta Nacional de Ocupación y Empleo—ENOE) and projections based on National Statistics Institute (Instituto Nacional de Estadística y Geografía—INEGI) 2008. Note: Output-employment elasticities are used to translate changes in output (Figure 1.11) into the above employment projections for 2010–11. The evolution of employment, earnings and remi ances determines to a large extent observed and projected changes in household income, since combined they account for more than 80 percent of total household income. In the crisis period, total household in- come is projected to fall by 6 percent due primarily to an 8 percent drop in labor income and a much smaller (1 percent) drop in non-labor income (Table 1.3).24 As aggregate output and employment recover after 2010, household income is projected to grow at around 2 percent annually, while non-labor income stays relatively stagnant. Both the level and depth of poverty are expected to increase as a result of crisis, and remain higher than 2008 levels even after the crisis subsides. The poverty headcount rate is expected to increase by 3.4 percentage points from 2008 to 2009 due to the crisis (an additional 3.6 million people in poverty), and then decline by around 2 percentage Table 1.3: Projected household income and its sources Actual Projected (peso/mo) (% change from previous year) 2008 2009 2010 2011 Total HH income 11,409 −6.1 1.8 1.7 HH labor income 8,466 −8.2 2.3 2.3 HH non-labor income 2,180 −0.6 0.5 0.0 Remittances 136 −2.5 −2.4 −10.5 Pensions and other private transfers 891 0.7 0.8 0.8 Capital income 215 −8.8 1.6 1.6 Safety nets (Oportunidades, Procampo, etc.) 212 0.4 0.4 0.4 Other (public transfers) 726 0.4 0.4 0.4 Mean earnings per worker 5,364 −6.4 2.3 1.6 Source: INEGI and projections. Note: Total HH income also includes implicit rent. Knowing, When You Do Not Know 21 Table 1.4: Poverty and inequality with and without crisis (constant poverty line) Actual Projected 2008 2009 2010 2011 Poverty Headcount rate 47.5 50.9 49.9 48.8 Poverty gap 20.3 22.4 21.9 21.2 Inequality (per capita income) Gini 0.52 0.52 0.52 0.52 Source: Microsimulations using macro projections and INEGI. points by 2011 (Table 1.4). The poverty gap is expected to show a similar trend. How- ever, the rise in the depth of poverty between 2008 and 2009 is much higher than that in the poverty headcount, relative to the baseline (2008) levels—increases of 11 percent for the poverty gap, compared with a 7 percent increase in the headcount. This suggests that the crisis had a strong impact on those who were already poor, which is not reflected in the projected rise in poverty headcount. As with the Type I countries, aggregate inequal- ity is expected to stay stable through the entire period. We turn next to the household-level impacts of the crisis and the recovery. As be- fore, we look at households that move in and out of poverty, as well as impacts across the entire income distribution. We present results for the crisis and recovery periods separately, so as to be able to distinguish between crisis-vulnerable households (that is, households that are projected to fall into poverty as a consequence of the crisis) and households that are “poverty-exiters� (that is, households that are projected to exit pov- erty in 2010–11). This allows us to compare the pro�les of affected households across groups and time. The Crisis Period, 2008–09 Crisis-vulnerable households are projected to suffer large income losses in 2008–09, with a 24 percent drop in average household income and a 32 percent loss in per capita in- come. Almost all of this loss comes from a 28 percent loss in household labor income. Crisis-vulnerable households are more likely to be urban and slightly more likely to be headed by a woman, compared to structurally poor households and the average house- hold. In addition, the heads of crisis-vulnerable households are more likely to be em- ployed in the manufacturing sector and less likely to be employed in agriculture in 2008 than the heads of other households. As in Type I countries, income losses are higher in urban compared to rural areas across the distribution—the average household in urban areas suffers a larger shock than the average household in rural areas (Figure 1.12). This difference is likely due to the fact that the crisis was most felt in manufacturing and services, both sectors that tra- ditionally locate in urban areas. Per capita household income losses are largest among those at the bo om of the 2008 income distribution (0 to 10th percentiles), with average losses ranging between 4 and 9 percent. Households in the 11th percentile and above suffer losses of about 4 percent, with no signi�cant differences across percentiles. This con�rms that the crisis disproportionally affected those who were already in a dire situation in 2008. 22 World Bank Study Figure 1.12: Growth incidence curve for Mexico (2008–09) 0 −3 −6 % Change −9 −12 −15 −18 0 10 20 30 40 50 60 70 80 90 100 Income per capita percentile Total Urban Rural Source: World Bank. Finally, households in the middle of the distribution are most impacted by the crisis in terms of the absolute size of the income loss (as opposed to percentage change, as with the growth incidence curve). Although approximately 85 percent of all households remain in the same income decile, most of the downward movements are concentrated in the middle of the distribution, namely in the 4th to 7th deciles (Figure 1.13). Between 20 and 22 percent of households in these deciles suffer income losses as a consequence of the crisis that push them to a lower decile. Figure 1.13: Transitions across deciles in Mexico 100 90 80 70 60 Percent 50 40 30 20 10 0 1 2 3 4 5 6 7 8 9 10 Deciles Movement down Movement up Stay in same decile Source: World Bank. Knowing, When You Do Not Know 23 The evidence presented in Figure 1.13 is not inconsistent with the results depicted in Figure 1.12, which suggested that income losses were largest at the bo om of the dis- tribution both in urban and rural areas. First, Figure 1.12 captures percentage changes in income, whereas Figure 1.13 is constructed on the basis of absolute changes in income. This implies that a loss of similar magnitude would (i) represent a much larger decline, in percentage terms, for households at the bo om of the distribution than for households in the middle or at the top, but (ii) at the same time would lead to more transitions across deciles in the middle of the distribution, where income brackets are narrower, than at the extremes. Second, Figure 1.13 focuses on movements across deciles rather than within income deciles. Since, by construction, those in the bo om decile cannot move further down and those in the highest decile cannot move further up, Figure 1.13 “misses� shocks to the bo om and top deciles. Third, as mentioned above, Figure 1.12 captures average impacts while Figure 1.13 also captures how many households received a shock in a particular decile. The Recovery Period, 2010–11 We now turn to the recovery period and perform similar analysis to that described for 2008–09 with a focus on those who exit poverty as positive economic growth resumes, as well as on the distributional impacts of this growth. Households that escape poverty in 2010–11 are projected to experience signi�cant income gains over the period, mostly due to a gain in labor income. “Exiter� households experience a 34 percent increase in average household income and a 48 percent increase in per capita income. Almost all of the growth in income comes from a 42 percent gain in household labor income. Household heads among exiters are more likely to be employed in manufacturing than any other groups. This is not surprising since this sector is predicted to grow rap- idly during this period. Perhaps more surprising is the fact that these individuals are less likely to be employed in services than comparator groups, despite relatively strong growth in the sector. Even though most individuals heading households that exit pov- erty during this period are predicted to be employed in the services or manufacturing sectors, their distribution across the two sectors is more skewed toward manufacturing compared to the sectoral distribution of the crisis-vulnerable. Similarly, the percentage of rural households and households headed by women is slightly higher and lower, re- spectively, among poverty exiters than it was among the crisis-vulnerable. These differences suggest that although those who became poor during the crisis are more likely to exit poverty during the recovery than those who are structurally poor, the two processes are not entirely symmetrical. As a result, some of those who fell into poverty may become structurally poor, and some of those who were structurally poor may be now pulled out of poverty. These dynamics will be a function mainly of relative growth rates and employment creation across sectors and across space. Projected economic growth in 2010–11 is expected to translate into higher income across the entire distributions and in both rural and urban areas (Figure 1.14). Growth at the bo om of the distribution is predicted to be signi�cantly faster, particularly in rural areas, where households in the bo om 0–20th percentiles are projected to experience an average 8 percent increase in per capita household income, compared to 4 percent 24 World Bank Study Figure 1.14: Growth incidence curves for Mexico (2010–11) 12 10 8 % Change 6 4 2 0 0 10 20 30 40 50 60 70 80 90 100 Income per capita percentile Total Urban Rural Source: World Bank. for the rest of the distribution. The bo om of the urban distribution is also predicted to grow faster than the rest, but higher income growth is concentrated on a narrower band (0–10th percentiles) and is lower in magnitude than that observe in rural areas. More generally, projected income growth in urban areas appears to be slightly lower than in rural areas, particularly in the middle of the urban distribution, although differences are not very signi�cant. As was the case above, there is signi�cant churning across income deciles once we move beyond average income changes (Figure 1.15). Interestingly, this churning is more evenly distributed across deciles compared to the transitions in the crisis period. We Figure 1.15: Transitions across deciles for the period 2010–11 100 90 80 70 60 Percent 50 40 30 20 10 0 1 2 3 4 5 6 7 8 9 10 Deciles Movement down Movement up Stay in same decile Source: World Bank. Knowing, When You Do Not Know 25 now observe a signi�cant percentage of households in the bo om deciles experiencing income gains that are large enough to place them in a higher decile, and the highest proportion of upward movers is now concentrated in the low-middle section of the dis- tribution (3–6th deciles as opposed to 4–7th deciles in 2008–09). Labor income seems once more to be the main force behind the observed dynamics. Simulating the Impact of Policy Responses The model presented here can be used to predict the potential impact of policy respons- es, particularly the type of policy measures that are intended to directly compensate affected workers and households for their losses, or protect the poorest against depriva- tion as a result of the crisis. To illustrate the potential use of the model in such scenarios, we highlight examples of simulations conducted for two countries, Poland and Mexico, focusing in both cases on the impact of an increase in social assistance on poverty and income distribution. The results for Poland in the previous section show that unemployment insurance, if administered according to pre-crisis rules, would have had a signi�cant stabilizing ef- fect on household incomes. The combination of lower employment levels and earnings translates into household labor income being 3 and 5 percent lower in 2010 and 2011, respectively, as a result of the crisis, while non-labor income is 3 and 2 percent higher, mainly due to unemployment insurance bene�ts. Since non-labor income accounts for about 30 percent of the average household’s income, its increase offsets some of the losses in labor income. As a result, the poverty headcount rate is projected to be higher by just 0.4 and 0.7 percentage points in 2010 and 2011, respectively, due to crisis. A stabilizing influence of unemployment insurance is also found by Aran et al. (2010) for Latvia, where the poverty rate is simulated to be around 20 percent with unemployment insurance as opposed to 23 percent in its absence. To simulate the impact of increase in family allowance or social transfers in Po- land, two scenarios for non-labor income were computed for each year (2010 and 2011). While both these scenarios included a projected increase in unemployment insurance payments, one of them also included a 20 percent increase in family allowance trans- fers that was a part of the government’s crisis response package. The increase in social transfers is projected to have reduced the poverty headcount rate by a further 0.3 and 0.2 percentage points in 2010 and 2011, respectively. This is due to an increase of 0.1 percent in average household income, which includes larger increases of 0.7 and 0.3 percent for the structurally poor and the crisis-vulnerable, respectively. After accounting for the im- pact of the increase in social transfers, the impact of the crisis on poverty headcount rate amounts to just 0.1 and 0.5 percentage points in 2010 and 2011, respectively. In the case of Mexico, the model is used to simulate the impact of the introduction a broad policy package to protect poor households in the wake of the crisis. The package includes an expansion of existing safety net programs—Oportunidades and the Nutri- tion Assistance Program (Programa de Apoyo Alimentario—PAL)—to increase the number of bene�ciaries by a total of 1 million. The simulations considered two different hypo- thetical targeting scenarios based on existing targeting methods in the country. Both scenarios are projected to reduce the incidence and/or depth of poverty as compared to the status quo, although the size and nature of the impact would vary with the choice of targeting method. Scenario 1 assumes perfect targeting of the poorest, namely that new bene�ciary households would be those with the lowest welfare scores nationally 26 World Bank Study (among non-bene�ciary households), using the proxy means test (PMT) formula used by the program. Scenario 2 assumes that new bene�ciary households are chosen among poor households that are closest to the poverty line nationally, again using the welfare score computed by the PMT used by the program. While neither of these extreme sce- narios would reflect the actual targeting strategy employed by the government, they are useful in providing an upper and a lower bound on the poverty impact of the potential program expansion. Scenario 2 provides an upper bound on the potential change in the poverty headcount that could be generated with the proposed expansion in coverage, while scenario 1 provides a lower bound. The converse is true about changes in the pov- erty gap and the severity of poverty. For the �rst scenario, the chronic poor are projected to bene�t the most from the proposed expansion, particularly in urban areas. The poverty headcount rate is not pro- jected to decline in 2010 and 2011 as a result of the program expansion, which is not surprising. Firstly, 1 million additional bene�ciaries represent a large expansion with re- spect to the existing coverage of the both programs, but only about 2 percent of the pro- jected number of moderate poor in Mexico in 2010. Secondly, since households with the lowest welfare score are selected, new bene�ciaries are likely to be so far away from the moderate and even the extreme poverty lines that the transfers would not be sufficient to take them out of poverty.25 The impact is however larger on the depth (poverty gap) and severity (squared poverty gap) of poverty, since the transfers reduce their distance from the moderate and extreme poverty lines. The second scenario would result in higher impact on poverty headcount—1.8 and 1.7 percentage point reduction in 2010 and 2011 respectively—than for scenario 1, but a lower impact on depth and severity of poverty. Regardless of whether the actual expansion was closer to scenario 1 or 2, new ben- e�ciary households are much more likely to have been chosen among the chronic poor than among other groups (the new poor and the non-poor).26 This is consistent with the official program goal of serving the extreme and chronic poor. Conclusion Using the example of four countries facing the recent �nancial crisis, the approach de- scribed here intends to provide analysts with a tool to project the welfare impacts of mac- roeconomic shocks in the absence of real-time micro data on employment, earnings, or income. The examples also show that focusing only on aggregate poverty and inequality indices, as typically allowed by methods often used by analysts responding to demands for quick simulations, would provide a narrow and partial view of the distributional impact of the crisis in a country. In the Philippines and Bangladesh, for example, even though the impact on aggregate poverty and inequality is low, some groups are esti- mated to suffer a disproportionately high impact. Urban households are projected to suffer relatively larger losses than rural households in the Philippines, and within-urban area losses are higher for the poor and the middle class. In Bangladesh, the estimated im- pact is particularly signi�cant for the top half of the distribution, with the urban middle class being one of the most affected groups. How the impacts of shocks are distributed in a country and what that implies about the characteristics of vulnerable groups can be valuable information for policy makers in a number of different ways. Firstly, the simulation results provide information on the people likely to suffer the most losses, and can therefore inform the design of policy responses. The fact that the Knowing, When You Do Not Know 27 crisis-vulnerable and the structurally poor differ signi�cantly in their characteristics in all countries suggests that expanding existing safety net programs intended for the structurally poor to mitigate the losses of the crisis-vulnerable may not be effective or practical. Instead, interventions that address increased levels of vulnerability and pro- tect households against risk may be required. In some cases, the distributional impacts can also have political economy implications that in turn ma er for policy responses. For example, the substantial losses suffered by the urban middle-income and poor house- holds are important factors to take into account because of the disproportionate influ- ence these groups can have on public perceptions and the political debate in a country. Secondly, as the section above, “Simulating the Impact of Policy Responses,� il- lustrates, the model can also be used to simulate the impact of particular policy inter- ventions, especially those that directly affect household income. While the results of such simulations vary signi�cantly by country, certain lessons can be generalized. In countries where losses in labor income dominate, automatic stabilizers like unemploy- ment insurance (as in Poland and Latvia) can mitigate the poverty impact to a signi�cant degree. While losses in non-labor income due to fall in remi ances cannot be compen- sated by automatic stabilizers, social assistance programs can potentially play a role in mitigating some of these losses. Even in the context of labor income losses, which part of the market is affected by the crisis can be an important consideration for policy makers, since automatic stabilizers would typically apply to workers in the formal sector only. In a country like Mexico, where automatic stabilizers are nonexistent or limited in their coverage of affected workers, expansion in social assistance can be effective in mitigating some of the impacts of the crisis on the poor. Expanding the coverage of the Oportuni- dades and PAL programs in Mexico by a total of 1 million households in 2010 could have reduced the depth of poverty by up to 2 percent or the poverty headcount rate by up to 1.8 percentage points, depending on what targeting strategy was employed to select the new bene�ciaries. Thirdly, the simulation exercise can help identify a list of possible “leading indica- tors� that can be monitored rapidly to gauge the likely welfare impacts of a crisis, in the absence of real-time information on household income/consumption. Although the appropriate choice of indicators would depend on the main transmission channels in a speci�c country, manufacturing employment, wages, remi ance flows, and relative food prices emerge as potential candidates for monitoring in all four countries. Information on the simulated impact of both shocks and potential policy interven- tions proved to be a valuable input into the policy discussion during the crisis, and will continue to be relevant for policy makers and analysts when real-time data is not avail- able. The methodology discussed here, as well as possible re�nements and improve- ments to this approach, is therefore likely to remain relevant. Clearly, the model is ap- plicable to a number of country contexts and for addressing a range of questions related to distributional impacts of an economic shock. Minimum data requirements include up-to-date macroeconomic projections by sector, with and without crisis, and a fairly recent household survey with income data that can be classi�ed by sector. Looking ahead, the analytical tool would bene�t from future efforts to extend or nuance the approach used here, as long as such efforts are tailored to the availability of data and the speci�c country context. For example, the availability of more disaggregat- ed macroeconomic projections would allow for a more detailed distributional analysis, to capture movements within sectors and across types of employment. Availability of 28 World Bank Study be er migration data in household surveys could allow for explicit modeling of migra- tion decisions for countries where remi ances are an important source of household income, as opposed to the somewhat arbitrary rules applied in the simulations here. And availability of surveys conducted during a crisis (such as quarterly labor force or rapid monitoring surveys), which hint at some of the impacts in real time, can be useful in �ne-tuning the model to take into account unanticipated changes in relative prices or in the labor market. Exploring such extensions to the model, along with improving the availability of data that make such extensions possible, would be a worthwhile future area for international institutions like the World Bank to invest in, to inform the advisory and �nancial support they often provide to countries affected by macroeconomic shocks. Notes 1. The full studies for Bangladesh and the Philippines are available as World Bank Policy Research Working Papers (Habib et al. 2010a and 2010b). Some of the early results are also summarized in an Economic Premise published by PREM Network, World Bank (Habib et al. 2010c). Full results for Mexico are presented as a chapter in World Bank (2010, draft); the Poland results are summarized in an annex to a World Bank loan document. 2. In addition to these four countries, simulations have also been conducted for Mongolia (draft pa- per under review) and for the impact of natural disasters (cyclones) in the Philippines. The results of these exercises are not discussed in this chapter. 3. While a number of other simulation exercises are likely to have been conducted at the Bank to simulate welfare impacts of the crisis from pre-crisis data, this report refers primarily to Aran et al. (2010), since their methodology and approach are the closest to ours. Unlike most of the other exercises, their methodology and results are also readily available in the form of a comprehensive, citable source. 4. See Kannan (2009) and Kannan et al. (2009). They cite research done at the IMF on pa erns of recovery from �nancial crises in advanced economies, which shows that recessions associated with �nancial crises tend to be severe and recoveries from such recessions are typically slow. It takes almost three years to return to the pre-recession output level—twice the time it takes to recover from other recessions. 5. There is li le consensus in the development literature about how to de�ne the “middle class�. A number of different de�nitions have been used, such us Blackburn and Bloom (1987), Beach (1989), Levy and Murnane (1992), Jenkins (1995) and Burkhauser et al. (1999). For our purpose, we de�ne the middle class broadly as those above the poverty line but below the richest decile (10%) of the population. In Bangladesh, this is the group between the 40th and 90th percentiles of per capita expenditure or income. 6. See World Bank (2011) for a discussion of channels through which the poor in Europe and Cen- tral Asia were impacted during the �nancial crisis. 7. See Bussolo (2008), Bourguignon et al. (2005), Ganuza et al. (2002) and Ganuza et al. (2006). An- other common approach is PovStat, an Excel-based World Bank simulation package, which uses household survey data and macroeconomic projections as inputs and estimates changes in poverty and inequality indicators. Although it allows for impacts to occur through multiple channels, it focuses exclusively on household heads and therefore does not allow for a full accounting of labor market impacts. Also, it does not offer an easy way to examine changes in non-labor income (such as remi ances), which has important implications for crisis impacts in some countries. 8. For a more detailed description than what is provided here, see Habib et al. (2010a, 2010b). 9. The estimations of employment status (though a multinomial model) are done separately for urban and rural areas, and those of earnings (an OLS model) are done separately for each sector. 10. For some countries the assignment rule is also such that the regional distribution of remi ances in the crisis year remains identical to that in the baseline year. 11. Note that by construction, the projected income/consumption distribution for a crisis year is expressed in terms of prices in the baseline year. This would imply that in order to obtain poverty Knowing, When You Do Not Know 29 estimates that are comparable with the baseline estimates, the baseline poverty line should be kept unchanged in real terms. 12. See Habib et al. (2010a, 2010b) for more details on how the adjustment for food price inflation is applied to the poverty line. 13. To state an obvious point: either type of comparison, Type I or Type II, is possible for any country. The differentiation between the countries is therefore purely driven by what type of com- parison was found to be most useful by the Bank country team and the government of that country. 14. Note that the policy being simulated should ideally have a large enough macroeconomic im- pact to carry through to the micro level. This is usually achieved by countrywide policies, for example, unemployment insurance. 15. Aran et al. (2010), on the other hand, are able to use macro data from Turkey to disaggregate their microeconomic outputs into formal and informal sectors. 16. In the case of Bangladesh, for example, the baseline year is 2005, which is a full four to �ve years away from the prediction years (2009 and 2010). 17. Sectoral movements of labor are modeled as depending only on individual/household char- acteristics (through the multinomial logit model) and population growth. This cannot take into account the kind of structural shifts that have apparently been observed in some countries due to the crisis, such as a reduction in the relative demand for skilled labor. 18. The use of this assumption for Bangladesh would not have affected the results signi�cantly, given the high correlation between household per capita income and consumption in the baseline Bangladesh data. The validation exercise also suggests that the simulation results are reasonable in spite of this naïve assumption. 19. Importantly, even when household data capturing the impact of the crisis becomes available, a true validation of the simulation results will be difficult in case of Type I countries, since the actual household data cannot be used to generate the counterfactual scenario of what household incomes and employment would have been in the absence of the crisis. Even in the case of Type II countries, a before-after ex post comparison may not constitute a validation of the ex ante simulation if the actual policy responses are different from what was included in the simulation. 20. Aran et al. �nd that in Latvia, unemployment insurance would have reduced the crisis poverty rate by as much as 3 percentage points. 21. Note that actual GDP �gures for 2010, which are available by now, may be different from these projections. 22. It is important to note that within a given country, those receiving most of their total household incomes from one or other of the two sources tend to have different characteristics. As a result, these differential impacts in labor and non-labor income have signi�cant distributional implications. 23. We plot the share of households in each decile who: (i) suffer an income loss large enough to shift them to a lower income decile (that is, had the income distribution at benchmark been equal to that in the crisis scenario, these households would have fallen to a lower income decile); (ii) suf- fer a relatively small income loss and therefore their position does not change much (that is, had the income distributions been equal in the two scenarios, these households would have remained in the same income decile); and (iii) suffer a positive shock and therefore are be er off in the crisis scenario (that is, had the income distribution in the crisis been equal to that at benchmark, these households would have climbed to a higher income decile). 24. While macroeconomic data on output and employment for 2009 was available at the time the report was prepared, no post-crisis household-level information existed. The last National House- hold Income and Expenditure Survey (Encuesta Nacional de Ingresos y Gastos de los Hogares—ENIGH) was collected in 2008 and data from the next round scheduled in 2010 has become available as this study is being completed. None of the analysis here is based on actual data during or after the crisis; all results on the impacts of the crisis and the subsequent recovery are from the microsimula- tions using pre-crisis household survey data. 25. The average simulated Oportunidades (PAL) transfer is equal to 76 percent (28 percent) and 17 percent (10 percent) of the extreme and the moderate poverty gaps, respectively. 26. The chronic poor can be understood as either those who were poor before the crisis and remained so or those who were poor after the crisis and have a low probability of exiting poverty by 2011. CHAPTER 2 On Economy-Wide Shocks, Models and Politics Michael Walton W hen a society suffers a big economic shock, what should be done? How does what should be done relate to what is typically actually done in practice? And where there is a difference, how can technical analysis serve to understand and reduce this gap? At �rst sight, the question of what should be done looks simple. Big adverse shocks are surely less subtle in their policy demands than solving questions of long-run growth, industrial policy, or service delivery. The problems are clear, there is political pressure to do something, and the motivation to act is much higher, amongst government agencies and civil society. Or at least the problem should be clear, since the classic shocks—com- modity price changes, sudden stops in �nancial flows, international slowdowns, harvest failures, floods and so on—are, unfortunately, repeated again and again. There should be substantial experience of the consequences of the shocks, and of alternative responses. Then the task of government, and of supporting external agencies, is to adjust policy to provide greatest relief to those hardest hit, and those most likely to be permanently hurt. Both interpretation and policy design is, however, a lot murkier and more compli- cated than this account would suggest. As the 2010 debates in the post-�nancial crisis industrialized world illustrate, there is even substantial professional and political dis- agreement over the basic intertemporal decisions over optimal de�cit and debt man- agement for the whole economy. Beyond aggregate choices the questions are deeply distributional, with respect to short and long-run effects across households, individu- als, and groups. This substantially complicates ma ers. Transmission mechanisms from shocks to different groups and their behavioral responses are often ill-understood, and only weakly integrated into a coherent overall account. Working out what is optimal is not trivial. When we turn to what is typically done in practice, further complications arise. There are two kinds of reason why actual responses can diverge from what is optimal. First, as just indicated, optimal policy may itself be unclear, especially where short-run information on actual effects is unavailable. Systematic information collection in devel- oping countries almost always comes with a lag, sometimes of years (for example, for national household surveys in poorer countries). Second, when many individuals, �rms, or groups are taking a hit, the natural tendency is to act to protect your assets, consump- tion, or other dimension of (household or �rm) well-being. Yet the capacity to act, in terms of information, resources, political influence, or other connections, is likely to be highly heterogeneous, and much more likely to be positively correlated with wealth, sta- 30 Knowing, When You Do Not Know 31 tus, and power. And this will often be exactly the opposite of the simple view on optimal policy suggested above. All this underlies the potential desirability of an approach to analysis and policy interpretation that combines: ■ Practicality and timeliness ■ Enough sophistication to capture the heterogeneity of transmission mechanisms and behavioral responses ■ Political salience. This is a tall order. But it is an important challenge for development practitioners, with precisely the kind of technical and political economy mix that should lie at the heart of development work. The point of departure of this essay is a reaction to a set of practical models for analyzing the poverty and distributional impacts of crises. These lie precisely in this domain. These are described in other contributions to this volume. They were sponsored by, and in many cases undertaken within, the World Bank. However, most of the issues at stake actually have nothing to do with the World Bank itself—since both the questions and the modeling approach are of general concern and practice. The bulk of this essay will assess the approach in this broader context. At the end I briefly discuss the speci�c question of the World Bank’s role. The technical work was inspired by the series of economy-wide shocks that have hit developing countries in the last few years—from big increases in food and oil prices, to the more complex shocks associated with the international �nancial crisis, which af- fected both external demand for goods and international �nancial markets. The concern is with systemic shocks experienced by a country, as opposed to the individual shocks that households suffer owing to idiosyncratic influences, for example from ill-health, job loss, or a business failure in normal economic times. There are two features of the general approach that cut across the particularities of the models. First, they are concerned with mapping economy-wide developments on to the conditions of individuals and households. Second, they seek to be practically, and politically, relevant—that is to provide information to policy makers and interest groups in time to shape policies. These two considerations drive the main design choice of developing models that link macroeconomic or sectoral developments to the characteristics of individuals and households, via existing labor market or household surveys, and then develop a capac- ity to simulate alternative scenarios. This is often called a microsimulation approach. Linking economy-wide developments to households (and �rms) is an obvious require- ment. The use of simulations flows from the typical absence of up-to-date information on outcomes. And the link to households and �rms is essential to assess impacts and responses, and also, if less obviously, to the interpretation of political economy. This approach is in contrast to one that relies primarily on gathering real informa- tion on actual impacts and responses. While the two approaches are obviously comple- mentary, the whole point of undertaking a structured, microsimulation approach is to provide information that is both sufficiently timely to be of practical relevance and hope- fully to be politically salient. 32 World Bank Study The approach has one enormous strength: it allows systematic, illustrative explora- tion of alternative policy choices—or indeed alternative views on the behavior of dif- ferent actors. These can be developed for policy makers as they assess what to do. This work can be undertaken by a ministry’s technical staff or by an external actor, such as a domestic think tank or the World Bank. But the approach also has a potentially large weakness: it may get responses wrong, and deliver misleading results. We are dealing with highly complex systems. As noted above, it is hard enough to model responses to shocks in the macroeconomy. When the task is to include the behavioral responses of the large, heterogeneous range of house- hold and �rm actors in a society, the systemic effects may be of bewildering complexity. The main response to the systemic challenge is to choose to analyze partial equilib- rium effects on part of the system—for example, on household consumption or unem- ployment, and often limit this to �rst-round impact effects. This is good for tractability and transparency: it is relatively easy to understand what is going on, and this is surely be er than a modeling approach that adds more complexity but becomes a black box, obfuscating rather than improving the informational basis of policy and public debate. But the nature of second-round and systemic effects cannot be ignored. This includes the potential behavioral response of one of the most important actors—the government— and so inevitably gets into political economy. The remainder of this comment has two themes. First, that second-round and sys- temic effects can be of major, and sometimes �rst-order, signi�cance to technical and political economy analysis. And second, that it is important to have an explicit view on what might be termed the practical political economy of information—that is on how in- formation, in this case from analysis and modeling, can lead to changes in public action. When Are Second-Round and Systemic Effects a First-Order Concern? An ex ante, �rst-round, empirically informed impact analysis of the effects of shocks on part of the households and �rm system clearly contributes to the interpretation of the consequences of crises. As several studies (including those described in Chapter 1) illustrate, this type of analysis suggests distribution ma ers (in the sense that impacts vary across households), that effects can differ from initial intuitions (so the complexity of the modeling adds something), and that these potentially make a difference for policy choice (see Ajwad et al., 2010, Ferreira et al., 2010, Habib et al., 2010) So far, so good. But are there conditions in which a �rst-round impact analysis is sufficiently incomplete or misleading to make a large difference to both interpretation and policy? There are a number of reasons for thinking this could often be the case. To fully work through this would require a careful comparison between an ex ante model- ing analysis and ex post experiences. That is not done here—though it would surely be valuable to do so in the future. So the argument is suggestive. It is useful to distinguish two kinds of reasons: second-round effects; and things happening in response to a shock outside the part of the system being analyzed. Second-Round Effects Here I give illustrations of three areas associated with household responses in response to changes in market conditions after a shock. Knowing, When You Do Not Know 33 (A) HOUSEHOLD CONSUMPTION The �rst point is obvious: households respond to changes in prices of goods by changing consumption pa erns. Is this sufficiently large and heterogeneous enough to be prac- tically important? There are at least indications that this can be the case. A survey of Turkish households found that almost three quarters shifted consumption pa erns into cheaper items (see Chapter 3). For an illustrative systematic analysis of what this kind of response means in welfare terms, Friedman and Levinsohn (2002) use an ex ante model- ing of the welfare impact of food price changes in the Indonesian crisis using empirically estimated substitution effects and an approximation of welfare effects. This found that the welfare impact—as measured by the compensating variation—compared with no substitution is substantial. Estimated total welfare losses decline by between a third and a half across the household distribution compared with a scenario of no substitution, with larger reductions in urban than rural areas, and in the middle of the rural distribu- tion of expenditure (see Figure 2.1). As this analysis is also ex ante, it is only illustrative. Moreover, since it is based on an econometric analysis of substitution effects, it almost certainly underestimates heteroge- neity across households driven by interactions between household preferences, needs, Figure 2.1: The welfare impact of food price changes in the 1997 Indonesian crisis across households Source: Friedmann and Levinsohn, 2002. 34 World Bank Study and local market conditions. Thicker local markets for consumption goods will allow more opportunities for substitution, and potentially less extreme price effects, especially when thinner local markets lead to local hoarding or exploitation of scarcity by traders with market power. For policy purposes the issue of heterogeneity will often be more important than ge ing be er estimates of the size of effects. In the wake of a crisis, we want to know if adverse effects are “large� or “small�—and the mitigating influence of substitution itself is unlikely to influence this assessment. But ge ing a good enough grip on who is suffering most does ma er, and can be relevant to policy design and modeling strategy. (B) LABOR MARKET RESPONSES A lot of the action in response to shocks happens in the labor market. Here too speci�c mechanisms ma er. Again the structure of the issue is how individuals in households behave in response to potential changes in labor market functioning. How �rms and �rm-households respond to a shock determines the labor market op- portunities available to households. In effectively all developing countries, employment is distributed between formal and informal sectors. Formal �rms may respond by reduc- ing wages, reducing pro�ts, reducing hours worked, or laying off workers, depending on expectations of the length of the shock, the potential to reduce wages and the cost of �ring. This can make a large difference to implications for worker-households. High levels of layoffs both have the sharpest effect on those who lose their work and put ad- ditional pressure on the informal labor market. Furthermore, different labor markets have different closures. This is partly influ- enced by the degree of flexibility of formal work, but of equal or greater importance are options for informal work. In Latin America, for example, the labor markets of Argen- tina and Chile have typically closed with higher levels of open unemployment, while those of Mexico and Peru—and to some extent Brazil—in lower earnings. Elsewhere, in South Africa the labor market primarily closes in higher open unemployment, while in Indonesia and India it closes mainly in shifts into lower paying informal work. Then the response of households ma ers. In Indonesia and Thailand, there was re- ported to be signi�cant reverse migration to rural areas after the 2007/08 crisis, in re- sponse to sharp falls in employment opportunities in urban areas. An informal survey by the Self-Employed Women’s Association of India in 2009 also found many of their members were moving back to villages, especially in Gujarat (that has a high level of connection to export markets). Labor reallocations also occur within households—for example with women or children increasing their labor force participation to compen- sate for lost income from declining wages or work of primary wage-earners. The pa ern of changes in the labor market can again make signi�cant differences to the size of welfare effects, and, more important, these can be heterogeneous across households, depending, in these examples, on connections to rural areas, and the poten- tial to increase household labor supply. (C) BEHAVIOR AFFECTING ASSET CHANGES A third category of response concerns assets. It is both a standard part of economic analysis, and generally observed, that households respond to a shock by changes in their asset position. This can involve drawing down �nancial savings, selling ca le, gold, or other physical assets, or borrowing. Or it can involve changes in human asset accumula- Knowing, When You Do Not Know 35 tion, for example by taking children out of school (to save on fees or put them into work) or postponing health-related spending. The last may be exacerbated for those who lose job-related health insurance. There is evidence that this is signi�cant. As Leipziger reports in Chapter 3, accord- ing to Crisis Response Surveys conducted by the World Bank’s Europe and Central Asia group, 29 percent of Turkish households in the bo om quintile postponed health spend- ing, and 14 percent reduced education spending. In Armenia in the bo om quintile, almost 60 percent of households stopped visiting health centers or purchasing medi- cine (World Bank, 2010a). However, it is not necessarily the case that children shift into work—Schady (2004) found that school a endance actually rose in response to severe economic crises in Peru, which the author interpreted as a response to a fall in work op- portunities for children. Drawing down on assets or postponing investments can be an efficient response to an adverse shock for a household. However there are two issues of concern for interpre- tation and policy. First, there is likely to again be substantial heterogeneity, depending on initial assets, �nancial savings, social connections, forms of �nance for education and health, and so forth. Local market and community conditions ma er, especially as we are concerned here with covariate shocks—prices of some assets (ca le for instance) fall if many households are selling, and opportunities for borrowing or other support also diminish when many are hurt. Second, there is the possibility of persistent effects, which are of particular concern if they affect human capital, whether of children or adults. System-Wide Factors beyond Households Since the immediate concern is with the distributional effects on households, there is a tendency to focus analysis on pa erns of impacts in the household sector. Yet this may miss the big action in loss allocations across groups in response to a shock. This is par- ticularly important where the shock involves the nexus of �nancial, macroeconomic, and corporate sector interactions. Analyzing speci�c distributional effects across the distri- bution from data available in surveys could be second-order relative to the distribution between holders of �nancial assets and others, and �scal action to support the �nancial- corporate sector. Take the characteristic sequence of emerging market crises. As awareness of the impending crisis spreads, holders of �nancial assets start moving out of domestic cur- rencies and the domestic �nancial system. Typically those with larger holdings—and so more information and international �nancial connectivity—move earlier. Figure 2.2 illustrates for the lead up to the Argentine crisis: in December 2000 to March 2001, do- mestic assets below US$100,000 were actually rising, while larger holdings and foreign asset holders were already moving their money out. As expectations of a collapse—and the likely abandonment of the peso-dollar peg—became widespread, shifts out of peso assets spread down the size distribution, but the proportional change in larger and for- eign-owned assets remained much higher. The aggregate effect was of large-scale capital flight, of some US$13 billion between the end of 2000 and the �rst quarter of 2002: the peg was indeed abandoned in January 2002, and the peso lost 75 percent of its value in a few months. The smaller depositors who had shifted into domestic dollar assets (Figure 2.2) meanwhile suffered an effective expropriation, as their deposits were converted into pesos at the old 1:1 exchange rate. Those who got their money out actually experienced capital gains with respect to Argentine assets. 36 World Bank Study Figure 2.2: The pattern of change in deposits, by size and ownership of deposits, in the lead up to the Argentine �nancial crisis a. December 2000 to March 2001 Percent 21.7 Peso deposits 20 10.1 Dollar deposits 10 5.6 6.4 1.3 3.2 0 –0.9 –0.5 –10 –5.9 –20 –14.3 –22.3 –30 –40 –50 –60 –56.8 Up to $5,000 $5,000–$20,000 $20,000– $100,000– Over $500,000 Foreign residents $100,000 $500,000 b. December 2000 to November 2001 Percent 20 16.9 Peso deposits 10 Dollar deposits 0 –10 –3.9 –6.0 –20 –11.3 –11.7 –15.2 –30 –22.1 –23.0 –28.2 –40 –50 –49.4 –60 –70 –62.2 –80 –90 –100 –94.5 Up to $5,000 $5,000–$20,000 $20,000– $100,000– Over $500,000 Foreign residents $100,000 $500,000 Source: Central Bank of Argentina, as calculated and presented in Halac and Schmukler (2004). Note: The �gure shows the cumulative percent change in peso and dollar private time deposits by resi- dence and size of deposit for the periods December 2000 to March 2001 and December 2000 to November 2001. The dollar sign ($) stands for both U.S. dollars and Argentine pesos, as the exchange rate was still one peso per dollar. Now this is only the beginning of the story in �nancial crises. Since the �nancial sys- tem is both central to economic functioning and generally carries substantial political in- fluence, it typically becomes the recipient of large-scale government support, in the form of protection for deposit holders and bailouts. Fiscal and quasi-�scal costs have been estimated at 19 percent of GDP for the Mexican 1994–96 crisis, 26 percent of GDP for the Republic of Korea, and 50 percent of GDP for Indonesia in the wake of the 1997–98 East Asian crisis (Honohan and Klingebiel, 2000). This comes at the cost of higher future taxes or lower expenditures, and so involves an intertemporal transfer. But it also has distri- butional effects across households, since the bene�ciaries are those within the �nancial system (and in some egregious cases even the equity holders of banks) that typically exclude much of the population in a developing country. Moreover, where spending is Knowing, When You Do Not Know 37 on an increasingly progressive path, as it was in Latin America in the 1990s and 2000s, the foregone spending has an additional regressive impact. A �nal issue concerns wage and pro�t shares. There is some evidence that crises are associated with reductions in wage shares (Diwan, 2001). Indeed, Diwan suggests that crises work as a mechanism for resolution of distributional �ghts when mechanisms in normal economic times prove inadequate. A crisis forces a loss allocation process that of- ten hurts labor incomes disproportionately. Most household surveys only capture labor incomes—or the consumption of households whose dominant source of income is from labor or the returns from self-employed activities—and so misses this action. Implications What do these comments imply? Of course the world is more complex than can be cap- tured in any modeling approach. The point is not to seek an ever-more complicated model—that comes at the cost of tractability and, even more important, transparency. However, I think there are three implications for the design and use of models, all of which go in the direction of pragmatic completeness. i. It will often be important to place speci�c modeling of part of the socio-economic system in the broader context of both system functioning and responses. This will frequently have to be somewhat ad hoc, in the sense of linking household impacts of a change in price or employment conditions to what is happening in the overall drivers and responses of the crisis—as illustrated for the �nancial- �scal responses to emerging market �nancial cries. This assumes the goal is to assess overall distributional effects of a crisis, as I believe should be part of any technical input to crisis analysis. Sometimes, of course, the goal may be nar- rower; for example, comparing alternative instruments to reach a more speci�c objective, such as introducing price subsidies versus public works, or calculat- ing the likely impact of speci�c policy interventions. ii. In modeling strategy there may be opportunities to explore particular aspects of heterogeneity and market context through imposition of alternative closures or potential responses. I have not worked through this in detail, but here are here a couple of intuitive examples. Alternative labor market closures could be explored both with respect to the extent to which �rms lay off workers, as opposed to cu ing hours and wages, and in terms of whether laid off workers lead to higher open unemployment or reduced labor incomes in the informal sector. And second-round household responses could be explored, in terms of consumption, household labor allocation, and asset responses. iii. The issue of the actual heterogeneity in impacts and responses should clearly also be explored empirically. In particular this suggests high priority for look- ing at historical episodes, in addition to new quick surveys (see World Bank, 2010a and 2010b). In either case the question is more about exploring any varia- tion around an estimated pa ern, and seeing if this is related to any observ- ables, or not. These considerations ma er because they have a bearing on the interpretation of ef- fects of shocks, with potentially large implications for policy. Each of the areas reviewed has potential consequences for policy. Where there is unknown heterogeneity in house- 38 World Bank Study hold impacts, there is a stronger case for self-targeting instruments—low-wage public works or subsidies to inferior goods for example. If asset drawdowns are expected to be signi�cant, this can imply different instruments, such as school-related grants (as used in the response to the Indonesian 2008 �nancial crisis) or increased subsidies for use of health services. Understanding the nature of labor market closure will also influence the instrument mix, from supporting negotiations to reduce layoffs, to the mix of unemploy- ment support and public works. And heterogeneity in conditions will generally support the need for use of a family of instruments and associated monitoring. For example, temporary increases in conditional cash transfers helps existing recipients, but not those falling into poverty as a consequence of the crisis. When Can Information Affect Government Behavior? When can the results of modeling—or indeed any analysis—affect government behav- ior? This is central to the approach under discussion here, since it is fundamentally mo- tivated by the goal of achieving be er outcomes for society in the wake of shocks. It is not an academic exercise. So we are deeply within the domain of the role of information in real decision-making, as it interacts with “technocratic� and political economy pro- cesses. Practical technical analysis often has an implicit theory of change, and of how information influences this. I think it is important to make this explicit. Consider two kinds of case: where analysis can play a role under conditions of tech- nocratic space, and where interests determine policy but information may influence po- litical economy outcomes. Technocratic Space The implicit, or explicit, theory in much of the work here is of the political salience of well-intentioned technocrats—a view that has pervaded much development policy advice for decades. A strong version of this assumes the social welfare function of the government is aligned with the analyst, but policy makers lack information on how to achieve this. Suppose a government genuinely wants to provide support to those hurt most by a crisis. This will be some function of the interaction between the size of house- hold-speci�c shocks and their initial level of deprivation. Then an empirically informed modeling approach—and even more one that embeds speci�c model-driven analysis with a pragmatic treatment of systemic and second-round effects—can both provide critical information on who should be targeted and on the likely efficacy of different policy instruments. This view of the world may seem far from the reality of real government objectives, influences, and political constraints. However, a weaker requirement is that some seg- ment of policy making has these technocratic characteristics. Perhaps a government is going to bail out rich bankers in any case, but still wants to allocate some resources to compensate middle and poor households, and has a technocratic group with the man- date to design an implementable approach. Even in this technocratic case, I believe the analysis should go beyond the purely technical. In particular it is valuable to make explicit the information-based theory of change, and assess its realism. Here are three aspects of this. i. Speci�c goals. What speci�c goals underpin the technocratic space? If they are really to support a particular support base (laid-off formal workers, existing Knowing, When You Do Not Know 39 patronage networks in poor areas) that may differ from a more general, dis- tributionally weighted, social welfare function. At a more technical level, the above phrase that weights will be “some function of the interaction between the size of household-speci�c shocks and their initial level of deprivation� requires resolution in the context of local social and political priorities. ii. Administrative feasibility. The feasibility, leakage, scope for corruption and likeli- hood of local capture will both vary substantially across programs, and be quite context-speci�c. iii. Political constraints. Even a technical analysis needs to take account of political constraints—indeed, the more these are incorporated, the more useful will the technical analysis be. Technical analysis, and exploration of optimal responses, can feed directly into policy design in cases of technocratic space, but even here supposedly “non-technical� considerations also intrude. Information and the Political Economy of Interests A second category concerns the (more realistic) case in which political and economic in- terests are salient, and differ in the structure of influence from the social welfare function of the analyst. Here the role of information, and modeling or other analysis, is different: it is about changing the political dynamic. And this requires a different theory on the role of information. Two things are likely to be useful here. First, to be relevant, analysis needs to map on to politically relevant interest groups. Doing this systematically can itself be a useful exercise. In rich and poor societies alike, narratives over the bene�ciaries of policy choices often have a weak relationship to real- ity. In the United States there seems to be some obfuscation over who bene�ted from tax cuts in the early 2000s: the discourse of “middle class� tax cuts hides the fact that the pri- mary bene�ciaries were the truly rich. In developing countries formal industrial workers and public school teachers—in Mexico and India for example—are often characterized as being amongst the relatively deprived. That may be true relative to the rich, but not to the bulk of the population. As a �nal example, and as the discussion above illustrates, it is important to bring in analysis of impacts on pro�ts and holders of �nancial assets into an integrated analysis. Second, it is useful to have an explicit account, or theory, of how new information may influence political dynamics. This could work, for example, via the accountability of politicians in a democracy: politicians depend, if via multiple and murky routes, on support from the electorate. Be er information on the likely effects of policies could provide incentives for different behavior—for example, to provide more support for budgets and designs that favor more deprived groups hurt by a crisis. Now this also requires a strategy for making the new information salient, for example via civil society, parliament, and media. The theory and empirics of such channels of influence are an ongoing area of re- search: for example Besley and Burgess (2002) provide an example of an empirical study that �nds that local newspaper density increases the responsiveness of local govern- ments to shocks. The point here is to explore hypotheses on how information can make a difference and explicitly introduce this into practical designs and experiments. Where 40 World Bank Study we are dealing with economy-wide shock this will typically not be via experiments us- ing randomized controls trials (see Banerjee et al., 2009, for an example of the impact of information on voting). But making explicit the hypothesized channels and influence, and systematically tracking changes is still feasible. Conclusion and a Note on the World Bank The systematic analysis of the effects of shocks across households is of huge interest and importance. An ex ante modeling approach, linking historical household information to macroeconomic and sectoral analyses of the structure and dynamics the shock makes a lot of sense, precisely because information on actual impacts typically comes with a lag—and too late for timely policy responses. The methodology outlined in Chapter 1 is a valuable addition to this approach. This note has offered two kinds of comments on the approach. First, I have highlighted the potential importance of both second-round effects and changes in parts of the socioeconomic system that fall outside the domain of the mod- el. These may be large enough to substantially change the interpretation of aggregate loss allocations and the distributional pa ern of vulnerability. How to deal with this will vary from case to case, and can range from incorporation of second-round effects into a model estimated from historical information, exploration of how alternative as- sumptions on �rm behavior or labor market closure affect simulations, or ensuring any analysis takes at least ad hoc account of responses elsewhere—especially in the �nancial- corporate-�scal nexus. This can all make a difference to policy—either in choices over speci�c sectoral designs (for example, in labor market, education or �nancial bailout designs), or in shifting the emphasis to self-targeting mechanisms and an array of instru- ments to reach heterogeneous households. Second, I have argued that technical analysis of the impact of shocks would bene�t from an explicit theory of how new information can make a difference in the speci�c political economy of the case at hand. Sometimes a modeling approach will seamlessly feed into a situation of technocratic space. Even here considerations of administrative feasibility of alternative policy designs, and indeed political constraints, would be of value. More generally, the approach could be useful structured into hypotheses on the political economy of information, hypotheses that can be then assessed against the use of modeling results—even when the intrinsically economy-wide features of aggregate shocks preclude rigorous causal testing. A �nal question concerns the role of the World Bank—since this work was spon- sored by the World Bank. Does this make sense? The general answer is absolutely yes: the World Bank has technical expertise, and can in particular support the internalization of externalities inherent in the development of technical approaches that are applicable to many country se ings, as well as facilitating cross-country learning. These issues form part of the overall question of the political economy of information, and the comment on making implicit theories of how information affects change applies with force to World Bank analysis. Over time, the goal should be to both ensure useful technical approaches are adopted in countries, in governments, think tanks and universities, and to support direct communication across countries. There is, however, one apparent puzzle. The standard advice on the management of shocks is to prepare for them in normal times. Analysis and action after a shock hits Knowing, When You Do Not Know 41 is likely to be late, fuzzy and (for policy) particularly susceptible to immediate political influences. Yet the World Bank has been deeply engaged with the kinds of issues thrown up the recent shocks in many previous episodes of shocks. Furthermore, the technique of linking economy-wide changes to micro distributional analysis has been around for many years, with World Bank research an important contributor to the approach (see Bourguignon et al., 2004). So why didn’t the World Bank itself invest in this kind of approach in normal times, so that analyses could be quickly run out when new shocks arrive? I think the answer lies with internal incentives and pressures. Countries �nd it hard to develop policy instruments in “normal� times that can swiftly, or endogenously, kick in when shocks occur. Similarly, World Bank policy analysis concentrates on non- shock issues outside crisis periods. Of course real choices have to be made in relation to pressing priorities. A potential solution involves exploring technical approaches that are of value both in interpretation and policy design in “normal� times and in response to major shocks. The question of the distributional effects of economy-wide or sectoral developments have particular urgency when there are large, adverse shocks, but are just as important for the analysis of slower-moving developments or the assessment of idiosyncratic shocks. CHAPTER 3 The Distributional Consequences of the Economic and Financial Crisis of 2008–09: A Comment Danny Leipziger Backdrop to the Issue of Distributional Consequences In examining the impact of the recent economic and �nancial crisis on different seg- ments of the population, it is useful to start off with what we think we know, which is that the distributional impacts of crises are generally regressive. Why? Because poor people tend to have less savings. In the crisis of 2008–10, we can easily overlook distribu- tional or microeconomic issues because the macroeconomic dilemmas are so stark and outlook generally so worrisome. But this would be a mistake, since the impacts felt by different parts of society are highly uneven and the measures governments have in place to deal with distress are so imperfect. The overall drop in national income and wealth in many parts of the developed world has been acute. Joblessness has dramatically increased and is slow to recover. The confluence of collapses in the markets for housing and the rest of the real economy com- bined with shocks to equity markets and accompanying banking problems are events not witnessed since the Great Depression. As a result, policy makers have been scram- bling to �nd solutions and to rebuild con�dence. This has put the spotlight on the big picture, perhaps to the detriment of impacts on individual households and �rms. Thinking about the Problem Government policies can be either progressive or regressive, based on who pays and who bene�ts. Some examples of progressive policies are unemployment insurance or targeted social programs; programs that we may think of as regressive include bank bailouts and other �nancial sector interventions that are of a systemic nature. The notion of an entity “too big to fail� being saved by taxpayers sounds somewhat regressive in nature. My main point is that the impact on income distribution of a �nancial crisis depends largely on the preexisting distribution. Everybody loses, but some lose more than others. I am going to focus in the �rst instance on the United States. The reader may legitimately wonder why we should worry about the United States, and what we have to learn from the U.S. experience. I argue that, �rst of all, the United States still constitutes 25 percent of global GDP, and what happens in the United States in terms of consumption, imports 42 Knowing, When You Do Not Know 43 and global investment has a major global impact. Secondly, of course, it gives us clues to what will happen potentially in other countries, particularly those with a burgeoning middle class. Third, a look at the United States can help shape the research agenda and give us some lessons on the incidence of public policy. And �nally, what happens in the United States has an effect on the future face of globalization, and the worse the impact, the less likely we are to return to the pre-crisis trajectory in terms of globalization. The Guts of the U.S. Story Let’s take a look at what has happened to income concentration in the United States over the past century. Based on work by Pike y and Saez (2003), as reported in the Vice President’s Task Force on the Middle Class (February 2010), we have the chance to examine the income of the top one percent of U.S. families over a long period of time. Between 1978 and the pre-crisis year of 2007, the top one percent of families went from commanding 9 percent of total U.S. income to controlling 23.5 percent of income (Figure 3.1). If you want to run for office in the United States and the people voting understand graphs, I would go with this graph above anything else. It reveals a shockingly uneven distribution of income before the crisis. Pu ing the pa ern in some historical perspective, the data reveal that the Gini coef- �cient has worsened over the last 20 years and that real incomes fell even compared to a previous period of the 1960s and 1970s. More remarkably, average incomes fell during the boom period of 2000 to 2008, during which everybody was supposedly doing well. However, it seems that not everybody was doing quite so well. Apropos Figure 3.1, the Figure 3.1: The top percentile (1%) of U.S. income (incomes above US$368,000 in 2008) 25 Share of total income accruing to each group (%) 20 15 10 5 0 13 18 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 98 03 08 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 20 20 Year Source: Pike y and Saez (2003), series updated to 2008. Note: Income is de�ned as market income including capital gains. 44 World Bank Study last time we saw as concentrated a distribution of income in the upper one percent of the population was in 1928, a year before the Great Depression. If you look at it by quintile, you see that everybody was losing except the uppermost quintile. That is also mislead- ing in a way because within that upper 20 percent it is really the top 5 percent that were doing the best, since 95 percent of American wage earners lost ground in the last 20 years. Now let’s shift the microscope from income to wealth. There is no Horatio Alger- style rags to riches story to be found here, because actually the mean-to-median wealth measure, as reported by Kennickell of the Federal Reserve Board (2009), has increased over time. The bo om line is that in terms of net wealth, the picture has turned more uneven. In 2007, the wealthiest one percent of Americans owned 34 percent of family wealth and the next 9 percent owned 37 percent. In other words, the richest decile of the distribution owns 71 percent of national private individual wealth, and that was before the crisis. So when I say that Joe Six Pack runs dry, it basically means that before the crisis you see declines in wealth in deciles �ve through nine, while the top decile gains. You also see that for most of the middle class, let’s say deciles �ve through nine and a half, most of their wealth—55 percent of it in fact—is in housing. Now let’s add to that what we know about the housing market collapse in the United States, where housing prices fell according to the Case-Shiller Index by 20 percent between 2004 and 2007, and another 20 percent in the subsequent two years, 2008–09. This is not good news for working class Americans. These pa erns are clearly seen in Figure 3.2, which shows the Lorenz curves for the United States with respect to income and wealth. Figure 3.2: U.S. Lorenz curves for total family income and net worth Source: Kennickel (2009). Knowing, When You Do Not Know 45 So this is the pre-crisis situation for income and for wealth. If we were to redraw them, particularly on the wealth side, and say that the middle class lost 40 percent of its housing worth and 50 percent of stock markets value, and these two make up the bulk of net worth of the households, except for the upper �ve percent, this would show a much more uneven distribution of wealth post-crisis. This is an easy simulation to do. Based on household balance sheets as reported by Kinneckell, we know that roughly 85 percent of household wealth for the core of the distribution is in housing and �nancial assets, and we know that their balance sheets suffered a major decline. Some Basic Generalizations Although there are signi�cant differences between the United States and other countries, there are some major similarities as well. What are the similarities between looking at OECD countries or the United States and developing or emerging market economies? First, we know from the work of Narayan and Petesch (2007) and others that there is an equivalent movement in and out of poverty in both rich and poor countries. Movement into poverty usually coincides with the occurrence of multiple disadvantageous events; in the case of the United States and other OECD countries, these misfortunes included the housing market, the stock market, and employment collapsing at the same time. We know that public policies in many developed countries are just as poor as in many developing and emerging economies in terms of sheltering these income declines. The one place that seems to be be er at cushioning declines is Europe, where safety nets are more extensive. We also know that public expectations of what people want from government ex- ceed what governments can do, and at the end of the day it is households that have to cope, often with devastating effect. So why are governments so inept at aiding house- holds during a crisis? There are a few possible explanations. One is that governments are �xated on the big picture—on �nancial bailouts, for example—or on protecting the �xed exchange rate regime, as in the case of Greece. Second is that targeting mechanisms may not be sufficiently in place beforehand and that governments are ill-equipped in a crisis to undertake the kind of means testing that is ideally required. The third is that resources are scarce. Tax revenues tend to decline in a big recession. Expenditures, ironically, are often regressive for a lot of goods, even merit goods (see Goldstein and Estache, 2009). Constituencies are not particularly strong for the middle class (see Estache and Leipziger, 2009) and certainly not for the lower end of the distribution. Moreover, even when there are viable public options, governments may be myopic in their decision-mak- ing, in part because �scal accounts are stressed and the political discount rate is very high. The Current Crisis in Perspective What is new in the transmission channel of this particular crisis? One unique factor was that trade �nance was affected, which was unusual. Remi ances have dropped more than one would expect, and we have seen that in some countries, such as Bangladesh, this had a big impact. External �nancing by banks has cut off and credit was affected. Even creditworthy borrowers, such as Korea, lost about half of their normal rollovers at the beginning of the �nancial collapse. Korea is experiencing the V-shaped recovery that all aspire to and is managing well, but it just shows how severe the upheaval was, and 46 World Bank Study how dramatically capital flows dried up. This had serious potential consequences for �rms, and in many emerging economies, public sector banks had to come to the rescue of �rms who could not access their normal international sources of credit. If this happens at the aggregate level, what happens at the household level? That is the focus of Chapter 1. It seems to me that the regional differences do occur with respect to what expenditures get cut �rst, whether it’s consumption or educations or health. But household coping mechanisms involve reduced spending, and the primary develop- ment issue is to trace the long-lasting impact of these coping mechanisms and also their infra-familial pa ern. Of course, the level of development ma ers. Normally, one might say that female labor force participation can be a buffer, and we have seen in previous crises in Argen- tina and other places where labor force participation rates for women suddenly rose. The trouble is in a globalized downturn like the one recently experienced, it’s not a question of joining labor force; it’s a question of who gets sacked �rst! Household decisions on health and education obviously can have long-term effects if they are not mitigated, and assistance is often faulty, particularly in a synchronized, global downturn where �scal resources are scarce and where the focus on the narrowly- de�ned poor may not be the right policy. In this respect, Ravi Kanbur, who has wri en on this, is correct (Kanbur, 2010). I think the leaky bucket analogy is more relevant in a large, systemic crisis than it is normally—be er to imperfectly target than not to act, especially given the linkages between earnings, consumption, and employment. Results of Quick Post Crisis Surveys We can learn quite a bit from the quick response surveys conducted by the Eastern Eu- rope and Central Asia (ECA) Region of the World Bank. In Turkey, for example, most of the response to lower incomes is borne on the food side, where substitution or reduction of foods seems to be the most immediate coping mechanism (Figure 3.3). The peck- ing order in terms of what gets cut seems to be food and then health services and then education, with the adjustments being predictably sharpest for the poorest. Additional insights can be gleaned from the labor market in Turkey, where most of the job losses are in the informal sector, more urban than rural (World Bank, 2010a). Figure 3.3: Household coping strategies in Turkey (survey responses) Percent of households 0 10 20 30 40 50 60 70 80 Substituted to cheaper food items 73 Substituted to cheaper non-food items 65 Decreased the use of entertainment services 61 Decreased food consumption 53 Decreased social outings with friends 50 Stopped buying some non-food items 48 Increased production of food for own consumption 43 Increased the use of public transportation or walking 31 Reduced the use of information services (newspaper, internet) 30 Reduced sports activities 30 Reduced visits to the doctor for preventative medical control 25 Reduced the use of health services 21 Dropped out of courses studying language, computers, etc. 10 Sent a member of the hhd to find seasonal employment 7 Transferred children from private to public school 6 Withdrew/postponed admission to school, college, or kindergarten 6 Source: World Bank, 2010a. Knowing, When You Do Not Know 47 Data from Turkey also suggest an important and somewhat staggering phenomenon, namely, the acquisition of new household debt. Quintiles two and three have debt levels that are equivalent to two-thirds their level of income. This is a huge rise in household indebtedness, which will have a potentially devastating long-term impact. As a coping mechanism, only middle-income households have access to credit of this kind. However, should interest rates rise, which is not unforeseen, the results could be quite devastating. In the case of Armenia (World Bank, 2010b), one sees that health expenditures are cut ahead of food consumption (Figures 3.4a and 3.4b). So, apropos my point regarding the importance of country pa erns, there is some variation even among the ECA cases studied in terms of what gets cut more.1 Figure 3.4a: Coping strategies in Armenia by asset quintile 60 50 Percent of households 40 Decreased food consumption 30 Stopped visiting health centers 20 Stopped buying medicines 10 0 1 2 3 4 5 Source: World Bank (2010b). Figure 3.4b: Coping strategy survey responses Percent of households 0 10 20 30 40 50 60 70 Decreased the use of entertainment services 64.1 Decreased social outings with friends 58.6 Substituted to cheaper food items 58.3 Reduced or stopped visits to healthcare centers 47.2 Reduced or stopped buying medicines 41.2 Decreased food consumption 41.1 Stopped buying some non-food items 40.0 Increased the use of public transportation or walking 32.8 Started working odd jobs 12.6 Increased production of food for own consumption 11.4 Sent a member of household to find seasonal employment 9.9 Started buying second hand items 9.1 Withdrew/postponed admission to school, college, or kindergarten 2.1 Took one or more children out of school 0.6 Source: World Bank (2010b). 48 World Bank Study Coming back to the more advanced economies—as well as for the middle class in emerging and developing countries—the main objective to enable households to ride out the storm revolves around trying to maintain employment. Regre ably, we seem to have relatively few policy instruments available to encourage this. The evidence on �scal incentives in the face of dramatic downturns is not encouraging, and only where work- sharing agreements were already in place was government able to induce the private sector to hold on to employees. In East Asia, the cushioning expenditures of government were very good, but that is because they had very strong �scal accounts going into the crisis. Takeaway Observations of Relevance for Policy There are some lessons that we can take away from the current crisis and a quick exami- nation of public policy responses: First, if governments do not have pro-equity policies in place prior to the crisis, they will likely be unable to prevent a sharp income decline for the non-rich during a crisis. Put differently, a concern for equity must precede the crisis, both because the crisis will exacerbate existing inequality and because the political economy issues involved in pro- tecting the less well off will not receive sufficient support in a crisis where all are losing absolutely. Second, there are some incentive-compatible options available to policy makers, but like the issue of counter-cyclical �scal rules, they are difficult to initiate in the middle of a crisis. This applies to work-sharing agreements, for example, which tended to work pre y well in Germany and other places, but are difficult to start when the government is already in a crisis. Third, economic crises are not gender neutral, with the prevailing rule often being “last in–�rst out.� Some hard-won gains for women can easily be lost in a crisis. Even in the advanced countries, there is some misguided support for the view that male jobs should be protected ahead of female employment. Fourth, the pre-crisis �scal situation largely determines the government’s room to maneuver. If we compare experiences and look at China or Korea, they executed perfect counter-cyclical policies, stimulating domestic demands to substitute for exports. But they were able to do this because they had the �scal capacity. Fifth, it is crucial that public policies assess and integrate long-term costs and ben- e�ts. The costs of interrupted education, joblessness, or loss of home ownership are all very hard to reverse. A comprehensive general equilibrium model incorporating all the costs might induce governments to spend to prevent certain outcomes; frankly, how- ever, that’s not usually the way decisions are made, least of all in a crisis. Next, I would be very careful about the second-round effects, particularly on house- hold indebtedness. I argue this because all the countries that have taken on additional debt will eventually experience higher interest rates—just imagine the impact on house- holds that have indebted themselves as a coping strategy. This could well be a time bomb waiting to affect the middle class in emerging market economies. Seventh, I would say that it is hard to initiate brand new safety nets during a crisis; therefore, it is important to use existing mechanisms, even if they are imperfect. Next, I offer the obvious comment that governments are normally important (see Growth Commission, 2008) but that they are indispensable in a crisis of con�dence in Knowing, When You Do Not Know 49 which many markets are affected. More importantly, it seems that the public expects governments to act as a cushion in a crisis, regardless of ideology beforehand or sterile arguments over the appropriate size of government. Ninth, I would argue that governments should resist the temptation to increase pro- gram coverage that may later prove hard to reverse. I would also be inclined, if policy makers are going to increase safety nets, to issue a second check, just to differentiate it from the �rst, and I wouldn’t put people permanently into the recipient category, be- cause it seems to be difficult to get people off of it. Lastly, it is important to do incidence analysis at all times, and hopefully prior to the crisis. In a systemic and brutal crisis, such as 2008–10, however, if you remember that in a �re, it’s be er to have a leaky bucket than none at all, it may be tolerable to mistarget a bit. If it is necessary to cover some people under the safety net umbrella who may not belong there ideally, the externalities associated with a major crisis may validate this course of action. Notes 1. For more ECA country cases, see World Bank (2011). CHAPTER 4 Stress Testing for the Poverty Impacts of the Next Crisis Ravi Kanbur Introduction In a presentation to the World Bank’s Executive Board in July 2009, at the height of the present global �nancial crisis, I argued that we have to prepare ourselves for the next crisis and its impact on the poor, even though we do not know when the next crisis will strike and what its precise impacts will be (see Kanbur 2009a, Kanbur 2009b, and Kan- bur 2010). I proposed three stages of systematic preparation at the country level. First, a program of assessment, which “stress tests� the social protection system against a range of crises, to identify gaps and weaknesses, analogous to the role that the Financial Sector Assessment Program plays in stress testing the �nancial system. Second, a medium-term program of �nancing improvements in social protection programs, now viewed as a sys- tem, to close gaps and address weaknesses. Third, a pre-quali�ed line of assistance for social protection, which would be available to a country immediately when previously agreed upon crisis triggers are breached. These three steps are somewhat different in nature and require different types of fora for discussion and debate. The third step has to be formulated in the context of overall development �nancing, and indeed a permanent “Crisis Response Window� was discussed in the context of the IDA16 replenishment. The second step comes close to “normal business� for donor agencies, although there are issues about the extent to which funding of precautionary investments, for a crisis sometime in the future, will be a ractive (to donors and recipients) relative to expenditures with immediate impact. However, both of these steps build upon, and depend upon, the �rst step: a solid assess- ment of how the collection of social protection programs as currently constituted would or would not respond to protect the poor in the face of a range of crises. The set of papers presented at the World Bank workshop on simulating the impacts of crisis is an excellent contribution to this effort. In this note I will briefly discuss why the effort is important, and then consider some of the speci�cs of the analysis. The Need for a Social Protection Assessment Program (SPAP) The present global �nancial crisis will pass. But crises are here to stay. By a crisis I mean a macro-level shock over which the country itself has no control, but which affects eco- nomic wellbeing in the country. A crisis worth the label will have a dramatic negative impact on per capita income. But the precise impact on the pa ern of incomes and well- 50 Knowing, When You Do Not Know 51 being will depend on the precise origin and nature of the crisis. Developing countries are likely to face many such sources—natural climatic disasters, sudden emergence and spread of an infectious disease, unrest in a neighboring country with possible flow of refugees, collapse of demand in a particular industry (such as tourism as a result of real or perceived terrorism threats, or sudden shifts in fashion in apparel), or the conse- quences of a global �nancial crisis. Each of these headings in turn covers many speci�c possibilities. For example, the �nancial crisis of 1997 is recognized to have been quite different in nature and impact from the crisis that began in 2007–08. Furthermore, for a given country, particularly a small country, a global crisis can have very different impact depending on the (unpredictable) reactions of its large neighbors or trading partners. Each of the different types of crises will have very different types of impact on the poor in different countries, depending again on the details of economic structure in that country. Added to the uncertainty of crisis type is uncertainty of crisis timing—timing of the onset of the crisis, and timing of when it recedes (the la er depending on uncertain policy responses). I have argued that these twin uncertainties—of crisis type and crisis timing—should frame the development of an ex ante strategy for protecting the poor. We should examine how the current social protection programs, viewed as a system, would respond to the needs created by crises of different types. We should “stress test� the system to identify vulnerabilities and gaps in (i) coverage and (ii) flexibility and speed of response. Such a comprehensive assessment would be a Social Protection Assessment Program (SPAP), corresponding to the Financial Sector Assessment Program (FSAP). I would now like to review the workshop presentations with this objective in mind. An Overview of the Workshop Papers and Presentations The papers and presentations at the World Bank Workshop on Distributional Impacts of the Crisis represent an important contribution to our understanding of crises, their im- pacts on the poor, and how social protection programs can and have responded. Chapter 1 in this volume summarizes the overall approach described there. The �rst step is to model the crisis itself and its macro impacts on the economy. This requires a careful enumeration, elaboration, and discussion of the transmission mecha- nisms. The exercise has to be country speci�c, and will involve a fair amount of judg- ment, for example on assumptions made about near neighbors’ policy responses, which in turn can influence macro outcomes for the country. From these would come various macroeconomic projections—overall and sectoral growth, inflation in general and in its major components such as food and fuel, exchange rate, international remi ances, and so forth Building on these would come projections of employment and unemployment. A key method for translating these macro projections to the household level are esti- mated models of labor force status and Mincerian wage equations, and assumptions that allocate international and domestic remi ances across households. With the above procedure and a base household survey, one can project the dis- tributional impact of the crisis being modeled. Poverty calculations can then be done assuming that the expenditure-to-income ratio stays constant. Impacts across the distri- bution can be shown using growth incidence curves and other methods. Who is likely to become poor, in terms of household characteristics, can thus be identi�ed. 52 World Bank Study The general approach is then applied to a number of countries for scenarios match- ing the impact of the global �nancial crisis. The results are intuitive, but interesting and of relevance for policy. Poverty rises but there is not much of an impact on overall in- equality. The new poor tend to be different from those who were poor before the crisis— in other words, the crisis impacts speci�c sections of the population. And the impacts are particularly concentrated in the middle of the income distribution. An Assessment, and What’s Missing The above is a broad summary of the methodology outlined in Chapter 1. Of course there is much greater detail in the speci�c papers, and there is a rich weaving together of country-speci�c institutional factors. But the overall approach and method should be clear. While one can of course have speci�c comments and criticisms of the individual country studies, or one or more of the components described above, my overall assess- ment is that this is an excellent start in the direction of what I have in mind, namely preparing ourselves for protecting the poor against the next crisis. The authors of the studies and overall summary are of course themselves aware of the limitations—the level of disaggregation depends on availability of macro projec- tions; various structural relationships are assumed to remain unchanged; no factor mo- bility; share of pro�ts remains constant; and so forth. But moving in this direction would require a full-fledged general equilibrium model of the economy, the construction and operation of which would require its own assumptions and make its own data demands. In my view, for the task at hand this project has struck the right balance between �ne detailed modeling of economic relationships and interactions, and the need to get to concrete results with the data available. I would like to suggest, however, that effort should go further developing a number of missing elements that are central to providing a complete picture of impacts on the poor. There is a strong focus on labor market status and labor income. Other sources of income are not treated with the same degree of careful a ention. This may be accept- able in economies where labor income is the dominant source of income for the poor, such as in transition economies of Eastern and Central Europe, but for economies where small holder agriculture is signi�cant, speci�c effort will have to go into modeling the farm and non-farm production activities of the household. I would see this as an issue in the Bangladesh case, for example. It may well be that the �nancial crisis, which was the focus of this project, transmits its impacts mainly through labor markets in the manu- facturing sector. But for crises in general, models should have the capacity to build in transmission mechanisms that affect farm incomes and incomes from non-farm artisanal activities. For example, the impact of climate-driven crises will certainly have this as a major transmission mechanism. I would like to use the above to make two further, more general, points. There is a strategic choice between how general our model should be—whether the base model should be able to capture a range of different types of crises, or whether the model- ing itself should be done crisis by crisis. While a �nal answer to this will have to wait for the results of trial and error, my initial position is that for a capacity to do ex ante assessments of crises as a category, we perhaps need a broader range of transmission mechanisms built in to the modeling, even if we decide to switch some of them off for Knowing, When You Do Not Know 53 speci�c simulations. In any event, what should be resisted is the notion that the same set of transmission mechanisms can be translated across crises. My second point relates to the role of income in this analysis. The sources of house- hold income are crucial in following through the macro transmission mechanisms down to the level of poor households. And yet, it might be argued with some justi�cation that we give more importance in our household surveys to the expenditure side of informa- tion gathering. This flows from our natural concern with measuring poverty, and from the broad consensus that expenditure data are the best for this task. But the impacts of crises, and indeed of any major macro change on poverty, whether short term or long term, passes through income. We need to be able to allocate macro-level shifts in income composition, by sector and by factor, down to the household level. For this we need a de- tailed account of the sources of household income in our household surveys. I would ar- gue that this aspect is underemphasized in household income and expenditure surveys, particularly for agrarian economies. With given survey resources, the balance between emphasis on the expenditure modules and the income, employment, and activity mod- ules is a difficult one to assess. But the line of work in these papers, and the line of work that would be needed to get a handle on the impact of crises on the poor, argues for more weight than is currently given to modules that help to pin down the sources of income. A central aspect of my proposed “stress testing� of the social protection system, in the face of different types of crises, is the need to model how individual protection programs will respond to different impacts of the crisis, transmi ed through different mechanisms that are country and crisis-speci�c. This is done to some extent in the Po- land study (Habib et al, 2010d) for unemployment insurance, but I would like to argue that this aspect is not as well developed in this approach as it should be. What is it that needs to be done? For each country, we �rst of all need an inven- tory of the major social protection programs and, very important, their rules of opera- tion. The most obvious programs for the “employment/unemployment� focus of this approach are of course unemployment bene�t programs, and this is what is done for example in the Poland study. Thus when the crisis is simulated, and its unemployment effects are projected, the income impacts follow from the wage equations, but the net in- come impacts—that is to say net of unemployment bene�ts—need to be simulated based on the rules of unemployment bene�t in that particular country. But unemployment bene�ts are only one of a myriad of programs whose rules will, in principle, affect flows of public resources to households when a crisis hits. Thus, for example, conditional cash transfer programs have speci�c rules of eligibility, sometimes based on assessed household income. If the crisis leads to a dramatic fall in the incomes of some categories of households, and the eligibility assessment happens soon after the crisis hits, then these households will start receiving cash transfers. Although the logic of the program may have li le to do with crisis support—it may instead be to do with incentivizing behaviors such as keeping children in school, or early health checks—in its operation it in effect becomes crisis support, and needs to be incorporated into the analysis. If these programs were small then perhaps they could be ignored. But some of these programs are very large in some countries—Progresa-Oportunidades in Mexico or the National Rural Employment Guarantee Act in India, for example. And even if individual programs are small, if collectively they represent signi�cant injection to poor 54 World Bank Study households, then they need to be incorporated systematically into the analysis of the possible impacts of crises on the poor. The above perspective requires that household surveys capture flows to households from social programs, distinguished by program in some detail. This is yet another di- mension of the importance of having a detailed accounting of the sources of household income—market and non-market. The extent of detail available is likely to vary from country to country. The best is if the actual cash flow from different programs is ac- counted for. If this is not available, but there is information on whether the household participates or not, this could be used to develop assumptions on how the total outlay of the program (which we would get from national level administrative data) could be al- located across households as a baseline—in the same way that international remi ances were allocated in the current project. In any event, what is important is that we do de- velop such a baseline where the monetary bene�ts of major social protection programs are allocated across households. There are two further issues that I want to raise on this theme of social protection programs in crises. First, there are programs that make a contribution in kind, not cash. Free food, or subsidized food from ration shops, is the leading example, but there are others such as excusing school fees or paying for school uniforms, and so forth. There is of course a monetary equivalent of these in-kind transfers, and if these transfers to households change as program rules dictate when the crisis hits, then they should in principle be incorporated into the analysis. The second issue is an important one, but difficult to address in practice. Of course, from the program rules of eligibility we can identify households that would fall into the program after the crisis hits, or specify the change in bene�ts of households already enrolled in the program. But this is the de jure analysis. In reality, whether the program delivers on the bene�ts depends on the budgetary allocations made to it. A good ex- ample of this is an “employment guarantee� scheme. In these programs, employment is supposedly guaranteed at a given wage (usually the local minimum wage) to all who turn up to the public works site. Thus, when the transmission mechanisms translate the crisis into employment losses and sharp wage declines, the number of people show- ing up at the public works site goes up. But will they be hired? The answer depends of course on whether the budget will rise pari passu. If it does not, there will be rationing of employment. For the domestic political economy, this raises the question of whether such budgetary allocations can indeed be guaranteed. In India, the National Rural Em- ployment Guarantee Act provides a justi�able guarantee, an example of the polity trying to tie its own hands behind its back. For analysts, however, the above scenario requires that a judgment has to be made on whether or not budgets will indeed be increased to convert de jure rules into de facto outcomes. Of course, once the machinery for simula- tions is set up, we can simulate the impact of different levels of budgetary provisions, if we are willing to model how the rationing is implemented. All of the above is in relation to identifying needs, gaps, and vulnerabilities as a range of crises materializes. But recall that there are two types of uncertainties—un- certainty of crisis type and uncertainty of crisis timing. The second type of uncertainty also raises issues for policy. In particular, it requires us to focus not so much on the ex ante comprehensiveness of the system of social protection programs, but on their flexibility. And this flexibility has two aspects—rapid response in scaling up programs when the Knowing, When You Do Not Know 55 crisis strikes, and equally rapid scaling down when the crisis passes. In a sense we have already touched on the �rst—the question of budgetary resource availability can be re- lated to the speed with which budgets can be increased. But we have not as yet discussed flexibility in the other direction, and this requires us to discuss what happens when the crisis passes—which microsimulation studies do not analyze. To illustrate the issues that arise, let us compare food and fuel subsidies with em- ployment guarantee schemes. On the scaling up side, the �rst has relatively easy flexibil- ity both in technical terms (for example, simply reintroducing pass-through provisions for global fuel prices into domestic fuel prices) and in terms of political economy (gen- eralized subsidies will have greater political support than employment schemes, which bene�t only the poorest). When the crisis passes, however, the ranking is reversed. As employment and wages pick up after the crisis, public works rolls will presumably sim- ply melt away as workers move to be er opportunities in the rest of the economy. But once generalized subsidies are increased, it is very difficult politically to reduce them— for the very reason that the constituency that bene�ts from them is much larger. The general point I am making is that an assessment of how well a social protection system does in the face of a crisis is incomplete without an assessment of what is left behind when the crisis passes. A full assessment will allow us to be er design ex ante programs that, as a collectivity, are comprehensive and flexible, and flexible moreover in both directions. Conclusion To conclude, I have looked at the current body of microsimulation crisis forecasting (as exempli�ed in Chapter 1) through a particular lens—whether it contributes towards preparing to protect the poor against the next crisis. I believe these studies represent an excellent step in this direction, and any further work should and indeed will have to use them as a starting point. But future work will also need to extend and elaborate the analyses in a number of directions, including different types of crises, sources of market income, sources of non-market income, and the operation of a full range of social protec- tion programs when crises hit and when crises recede. References Aran, M., M. Azam, M. I. Ajwad, and J. Hentschel (2010). “A Methodology Note on the Employment and Welfare Impacts of the 2007–08 Financial Crisis.� Draft. World Bank, Washington, DC. Banerjee, A., D. Green, J. Green, and R. Pande (2010). “Can Voters be Primed to Choose Be er Legislators? Evidence from Voter Campaigns in India.� Processed. World Bank, Washington, DC. Beach, Ch. (1989). “Dollars and Dreams: A Reduced Middle Class? Alternative Explana- tions�, Journal of Human Resources 24: 162–193. Besley, T., and R. Burgess (2002). “The Political Economy of Government Responsive- ness: Theory and Evidence from India.� Quarterly Journal of Economics 117(4): 1415–1451. Blackburn, M., and D. Bloom (1987). “Earnings and Income Inequality in the United States.� Population Development Review 13(4): 575–609. Bourguignon, F., M. Bussolo, and L. Pereira da Silva (2008). “Introduction: Evaluating the Impact of Macroeconomic Policies on Poverty and Income Distribution.� In The Impact of Macroeconomic Policies on Poverty and Income Distribution, eds. Bour- guinon, F., Bussolo, M. and Pereira da Silva, L. Washington, DC: World Bank. Bourguignon, F., F. Ferreira, and N. Lustig (2004). The Microeconomics of Income Distribu- tion Dynamics in East Asia and Latin America. New York: Oxford University Press for the World Bank. Bourguignon, F., A. S. Robilliard, and S. 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. Kehoe, T.J., Srinivasan, T.N. and Whalley, J. Cambridge, UK: Cambridge University Press. Burkhauser, R., A. Crews, A. Daly, and S. Jenkins (1999). “Testing the Signi�cance of Income Distribution Changes Over the 1980s Business Cycle: A Cross-national Comparison.� Journal of Applied Econometrics 14: 253–272. Bussolo, M., J. Lay, D. Medvedev, and D. van der Mensbrugghe (2008). “Trade Options for Latin America: A Poverty Assessment Using a Top-Down Macro-Micro Mod- eling Framework.� In The Impact of Macroeconomic Policies on Poverty and Income Distribution, eds. Bourguinon, F., M. Bussolo, and L. Pereira da Silva. Washington, DC: World Bank. Chen, S., and M. Ravallion (2009). “The Impact of the Global Financial Crisis on the World’s Poorest.� VOXeu note. April 30, 2009. Available at: h p://www.voxeu. org/index.php?q=node/3520. Cho, Y., and D. Newhouse (2010). “How Did the Great Recession Affect Different Types of Workers? Evidence from 17 Middle-Income Countries.� Social Protection and Labor team. World Bank, Washington, DC. 56 Knowing, When You Do Not Know 57 Diwan, I. (2001). “Debt as Sweat: Labor, Financial Crisis, and the Globalization of Capi- tal.� World Bank. Washington, DC. Available at: h p://info.worldbank.org/etools/ docs/voddocs/150/332/diwan.pdf. Estache, A., and D. Leipziger, eds. (2009). Stuck in the Middle: Is Fiscal Policy Failing the Middle Class? Washington, DC: Brookings Institution Press. Ferreira, F., P. Leite, L. Pereira da Silva, and P. Picche i (2008). “Can the Distributional Impacts of Macroeconomic Shocks Be Predicted? A Comparison on Top-Down Macro-Micro Models with Historical Data for Brazil.� In The Impact of Macroeco- nomic Policies on Poverty and Income Distribution, eds. Bourguinon, F., M. Bussolo, and L. Pereira da Silva. Washington, DC: World Bank. Friedman, J., and J. Levinsohn (2002). “The Distributional Impacts of Indonesia’s Finan- cial Crisis on Household Welfare: A ‘Rapid Response’ Methodology.� World Bank Economic Review 16(3): 397–423. Ganuza, E., S. Morley, S. Robinson, V. Piñeiro, and R. Vos (2006). “Are Export Promotion and Trade Liberalization Good for Latin America’s Poor? A Comparative Macro- Micro CGE Analysis.� In Who Gains from Free Trade? Export-led Growth, Income Distribution and Poverty in Latin America, eds. Vos, R., E. Ganuza, S. Morley and S. Robinson. Massachuse s: Edward Elgar Publishing. Ganuza, E., R. Paes de Barros, and R. Vos (2002). “Labour Market Adjustment, Poverty and Inequality during Liberalization.� In Economic Liberalization, Distribution and Poverty: Latin America in the 1990s, eds. Vos, R., L. Taylor, and R. Paes de Barros. Massachuse s: Edward Elgar Publishing. Goldstein, M., and A. Estache (2008). “The Scope and Limits of Subsidies.� In Stuck in the Middle: Is Fiscal Policy Failing the Middle Class?, eds. Estache. A. and D. Leipziger. Washington, DC: World Bank. Growth Commission (2008). The Growth Report: Strategies for Sustained Growth and Inclu- sive Development. Washington, DC: World Bank. Habib, B., A. Narayan, S. Olivieri, and C. Sanchez-Paramo (2010a). “Assessing Ex Ante the Poverty and Distributional Impact of the Global Crisis in a Developing Coun- try: A Microsimulation Approach with Application to Bangladesh.� Policy Re- search Working Paper 5238. World Bank, Washington, DC. ——— (2010b). “Assessing Poverty and Distributional Impacts of the Global Crisis in the Philippines: A Microsimulation Approach.� Policy Research Working Paper 5286. World Bank, Washington, DC. ——— (2010c). “The Impact of the Financial Crisis on Poverty and Income Distribu- tion: Insights from Simulations in Selected Countries.� PREM Economic Premise 7. World Bank, Washington, DC. ——— (2010d). “Employment, Poverty and Distributional Impacts of the Crisis in Po- land: Brief Overview of Results from Microsimulation Exercises.� World Bank, Washington, DC. Halac, M., and S. L. Schmukler (2004). “Distributional Effects of Crises: The Financial Channel.� Economía: Journal of the Latin American and Caribbean Economic Associa- tion 5(1): 1–51. Hallward-Driemeier, M., B. Rijkers, and A. Waxman (2010). “Women in Crisis: Firm level Evidence on the Labor Market Impacts of the Indonesian Crisis.� Mimeo. World Bank, Washington, DC. 58 World Bank Study Jenkins, S. (1995). “Did the Middle Class Shrink During the 1980s? UK Evidence from Kernel Density Estimates.� Economics Le ers 49: 407–413. Kanbur, R. (2009a). “The Crisis, Economic Development Thinking, and Protecting the Poor.� Presentation to the World Bank’s Executive Board, 7 July 2009, Washing- ton, DC. h p://www.kanbur.aem.cornell.edu/papers/WorldBankBoardPresenta- tion7July09.pdf. ——— (2009b). “Systemic Crises and the Social Protection System: Three Proposals for World Bank Action.� Presentation to the World Bank’s Poverty Reduction and Economic Management (PREM) Network, 28 April 2009, Washington, DC. h p:// www.kanbur.aem.cornell.edu/papers/SystemicCrisesAndTheSocialProtection- System.pdf. ——— (2010a). “Macro Crises and Targeting Transfers to the Poor.� In Globalization and Growth: Implications for a Post-Crisis World, eds. M. Spence and D. Lepiziger. Wash- ington, DC: Commission on Growth and Development. ——— (2010b). “Protecting the Poor Against the Next Crisis.� Distinguished Lecture Series, Egyptian Center for Economic Studies, Cairo, Egypt. Kannan, P. (2009). “The Lingering Effects of Financial Crises.� VoxEU website (19 No- vember). h p://www.voxeu.org/index.php?q=node/4232. Kannan, P., A. Sco , and M. Terrones (2009). “From Recession to Recovery: A Long and Hard Road.� VoxEU website (6 May). h p://www.voxeu.org/index. php?q=node/3534. Kennickell, A. (2009). “Ponds and Streams: Wealth and Income in the United States, 1989 to 2007.� Finance and Economics Discussion Series, Federal Reserve Board, Wash- ington, DC. Levy, F., and R. Murnane (1992). “U.S. Earnings Levels and Earnings Inequality: A Re- view of Recent Trends and Proposed Explanations.� Journal of Economic Literature 30(3): 1333–81. Narayan, D., and P. Petesch, eds. (2007). “Moving out of Poverty: Cross-Disciplinary Perspectives on Mobility.� World Bank, Washington, DC. Paci, P., B. Rijkers, and A. Revenga (2009). “Coping with Crisis. Why and How to Protect Employment and Earnings.� Policy Research Working Paper 5094. World Bank, Washington, DC. Paci, P., B. Rijkers, and A. Seinaert (2008). “Crunch Time in the Developing World Too? Policy Options for Dealing with the Potential Poverty and Distributional Im- pact of the Financial Crisis.� Poverty Reduction Group Policy Brief. World Bank, Washington, DC. Pike y, T., and E. Saez (2003). “Income Inequality in the United States: 1913–1998.� Quarterly Journal of Economics 113(1): 1–39 (updated to 2008, available at: h p:// elsa.berkeley.edu/~saez/). Ravallion, M. (2009). “The Developing World’s Bulging (But Vulnerable) Middle Class.� Policy Research Working Paper 4816. World Bank, Washington, DC. Sabarwal, S., N. Sinha, and M. Buvinic (2010). “How do Women Weather Economic Shocks. What We Know.� PREM Economic Premise 46. World Bank, Washington, DC. Spence, M., and D. Leipziger, eds. (2010). Globalization and Growth: Implications for a Post- Crisis World. Washington, DC: Commission on Growth and Development. Knowing, When You Do Not Know 59 Turk, C., and A. Mason (2009). “Impacts of the Economic Crisis in East Asia: Findings from Qualitative Monitoring in Five Countries.� In Poverty and Sustainable Devel- opment in Asia: Impacts and Responses to the Global Economic Crisis, eds. Bauer, A., and M. Thant. Manila, the Philippines: Asian Development Bank. United States Vice President’s Task Force on the Middle Class (2010). Annual Report of the White House Task Force on the Middle Class. Washington, DC. World Bank (2010a). “Household Welfare in the Face of Economic Slowdown: Results from Five Urban City Centers in Turkey.� Report No. 55062-TR, Human Develop- ment Network, Eastern Europe and Central Asia Region. World Bank, Washing- ton, DC. ——— (2010b). “Armenia: The 2008–09 Global Economic Crisis, Policy Responses, and Household Coping Strategies.� World Bank, Washington, DC. ——— (2010c). “Recent Trends and Forecasts of Poverty in Mexico: A Poverty Note.� Latin America and the Caribbean Region. Draft. World Bank, Washington, DC. ——— (2011). “The Jobs Crisis. Household and Government Responses to the Great Recession in Eastern Europe and Central Asia.� World Bank, Washington, DC. ECO-AUDIT Environmental Bene�ts Statement The World Bank is committed to preserving en- In 2010, the printing of dangered forests and natural resources. The Of- this book on recycled paper �ce of the Publisher has chosen to print World saved the following: Bank Studies and Working Papers on recycled • 11 trees* paper with 30 percent postconsumer �ber in • 3 million Btu of total energy accordance with the recommended standards • 1,045 lb. of net greenhouse for paper usage set by the Green Press Initia- gases tive, a nonpro�t program supporting publish- • 5,035 gal. of waste water ers in using �ber that is not sourced from en- • 306 lb. of solid waste dangered forests. For more information, visit * 40 feet in height and 6–8 inches in www.greenpressinitiative.org. diameter K nowing, When You Do Not Know is part of the World Bank Studies series. These papers are published to communicate the results of the Bank’s ongoing research and to stimulate public discussion. Economists have long sought to predict how macroeconomic shocks will affect individual welfare. Macroeconomic data and forecasts are easily available when crises strike. But policy action requires not only understanding the magnitude of a macro shock, but also identifying which households or individuals are being hurt by (or bene�t from) the crisis. One popular solution is to extrapolate the welfare impact of a shock from the historical response of income or consumption poverty to changes in output, by estimating an “elasticity� of poverty to growth. Although this method provides an estimate for the aggregate poverty impact of a macro shock, it has limited value for analysts and policy makers alike. Aggregate numbers are useful to capture the attention of policy makers and the international community, but in the absence of any infor- mation on who is affected and to what extent, they provide little guidance on what actions need to be taken. This World Bank Study outlines a more comprehensive approach to the problem, showcasing a microsimulation model, developed in response to demand from World Bank staff working in countries and country governments in the wake of the global �nancial crisis of 2008–09. Starting with the idea of using simple macroeconomic projections as the “macro linkages� to a microbe- havioral model built from household data, the model was conceptualized, re�ned, and tested in Bangladesh, the Philippines, Mexico, Poland, and Mongolia. The results have fed into country policy dialogue and lending operations of Bank teams, as well as various country-level and global knowledge products of the Bank. In addition to a chapter on the microsimulation approach and its results, this volume includes three chapters written by experts external to the World Bank. In these chapters, the authors discuss a range of relevant issues: the implications of applying microsimulation techniques, how the approach can be extended or improved upon, the ability of governments to respond adequately to the results of such analysis, and the lessons that can be learned from the �nancial crisis of 2008–09 for mitigating the welfare impacts of future crises. World Bank Studies are available individually or on standing order. This World Bank Studies series is also available online through the World Bank e-library (www.worldbank.org/elibrary). ISBN 978-0-8213-8981-2 SKU 18981