Knowing When You Do Not Know Simulating the Poverty and Distributional Impacts of an Economic Crisis

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 benefit 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 This problem is particularly acute when dealing with developing countries where household data is sporadic or out of date.A popular solution is to extrapolate the welfare impact of a shock from the historical response of income or consumption poverty to changes in output, by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 policymakers alike. Aggregate numbers are useful to capture the attention of policymakers and the international community, but in the absence of any information on who is affected and to what extent, provide little 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 finding 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 specific 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 different 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 working in countries and country governments in the wake of the global financial crisis of 2008-09. During the growing catastrophe in a few industrialized countries, there was rising 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 sectors, 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, different levels 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 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 find and calibrate within the limited time available to analysts. Rather, filling this void required finding the fine balance between two competing needs: capturing the complexities that influence how macro impacts are transmitted to the households and feasibility in terms of time, cost and data requirements.The void was at least partially filled by the approach developed by staff from the Poverty Reduction & Equity Department at 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 micro behavioral model built from household data, the model was conceptualized, refined and tested in a diverse mix of countries: Bangladesh, 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 financial 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-national 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 platform 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 creating 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 firms for hiring young workers? Are farmers in need of temporary support through cheaper credit? An important concern is that of second order impacts. The method hinges on extrapolating 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 first place. Even so, it is an issue that merits careful thought. In order to find solutions to methodological and policy debates, economists need to take advantage of periods of relative calm, rather than having to deal with them in 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 influence the governments to put in place the right kind of policies that minimize the suffering of people.In building such capacity, it is relevant to look ahead to what lies ahead in the immediate 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 settings, user-friendly simulation software is currently being developed by staff from the Poverty Reduction & Equity Group and the Development Research Group, as a ne