Michiel van Dijk - Long-run subnational projections of income and poverty for Ethiopia: A CGE-spatial microsimulation approach

  • Presenting author: Michiel van Dijk (Wageningen Economic Research)

  • Authors: Michiel van Dijk, Marijke Kuiper, Thijs de Lange, Jason Levin-Koopman

  • Session: C01D - Micro-Macro Linkage - Wednesday 9:00-10:30 - Erika-Weinzierl Hall

Introduction

SDG1 - End poverty in all its forms everywhere and SDG10 - reducing inequality within and between countries - are key goals for increasing global welfare and achieve a more equal global society. Better insights on the drivers of income change and how they will affect income distribution and poverty in the future are fundamental for the development of policies that contribute to achieving SDG1 and SDG10. To capture both the macro-drivers of income change and the observed heterogeneity in household income, a combination of macro- and micro-economic simulation approaches is commonly used (Bourguignon et al., 2013). Prominent examples of this approach are the World Bank GIDD (Bussolo et al., 2010) and the IFPRI POVANA (Laborde Debucquet et al., 2018) microsimulation frameworks that combine CGE models with household survey data to assess changes in future income distribution and poverty. A shortcoming of these studies is that they do not account for spatial differences in income and poverty levels (Chi et al., 2022) that are related to differences in access to resources (e.g. cropland) and agglomeration effects (i.e. urbanization). The aim of this study is to fill this gap by demonstrating a spatial dynamic microsimulation model that can provide income, poverty and inequality projections at subnational scale. The model combines household survey data with subnational projections on key drivers, including demographic change, urbanization and structural transformation of the economy to simulate how the distribution of income might change under different socio-economic futures. This approach can be used to assess questions such as: What are the long-run spatial patterns of income distribution and poverty under different socio-economic scenarios. Which regions are most likely to achieve SDG1 and SDG10 and which not? Are there overlaps between poverty and climate change hotspots? To illustrate the model we provide an application to Ethiopia, one of the largest countries in Africa in terms of population size, which is characterized by high poverty levels and deep inequality.

Methods and materials

Spatial microsimulation is a methodology to derive indicators at fine spatial resolution (e.g. districts and municipalities) by combining socio-economic data from a household survey (referred to as the ‘seed’), with aggregated information on population characteristics (referred to as the ‘benchmark’) that is available for small geographical areas from the population census and other sources. An algorithm is used to reweigh the seed and match it with the benchmark information to derive representative small area statistics (Edwards et al., 2013), in this case income and poverty indicators. In dynamic spatial microsimulation approaches, the benchmark is projected forward and new weights are computed for future periods to obtain trends in income distribution under different scenarios (Harding et al., 2011). To ensure consistency between macro- and micro-level drivers of income distribution and account for structural economic change, we updated the results of the spatial microsimulation model with sectoral wage and food price projections from the MAGNET model. MAGNET is a GTAP-based global computable general equilibrium (CGE) model developed by Wageningen Economic Research that is frequently used for food security, climate change and biodiversity analysis (Meijl et al., 2020). Our main source of information on household income (i.e. the seed) was the Ethiopia Socioeconomic Survey (ESS), which is part of the World Bank’s Living Standards Measurement Study - Integrated Surveys on Agriculture (LSMS-ISA) initiative. The ESS is implemented every two years and covers a wide range of topics, including an household income module. We used the latest available version for the year 2015/2016, which is representative at the state level, the 1st order administrative division. We used four benchmark variables: age, sex, urban-rural status and occupation, which together are key drivers of income change, in particular for emerging economies like Ethiopia, of which the long-run future growth trajectory is characterized by structural economic transformation, urbanization and demographic change. We used the narratives and quantified drivers of the Shared-Socioeconomic Pathways (SSPs) (Riahi et al., 2017), to inform the benchmark projections. The SSPs are a set of five global socio-economic scenarios that have been used to inform recent global assessments on climate change, Biodiversity and food and nutrition security. To construct the subnational benchmark projections, we combined several data sources. We used high-resolution population data from WorldPop (worldpop.org) to derive subnational population and demographic data for the base year. These were subsequently projected to 2050 by linking them to spatially explicit demographic and urbanization projections for the SSPs (Jones et al., 2016). To account for structural change in the shifts of jobs from agriculture to manufacturing and services, we collected subnational information on the distribution of five major occupational groups and projected these into the future in line with the SSPs. Finally, we prepared SSP-specific benchmark projections for each of the 60 states in Ethiopia for the period 2018-2050.

Preliminary results

Typical results from our study includes district-level per capita income projections for the period 2018-2050, which can be used to show changes in income distribution over time under different socio-economic scenarios. We combines the results with information on the (inter)national poverty line to estimate the poverty headcount ratio, the standard metric to measure progress towards SDG1. Apart from the total population poverty rate, we prepared population headcount estimates for subgroups, including urban-rural status, age group and occupation. Finally, we combined the national poverty maps with flood and heat stress maps to identify areas that are both characterized by extreme poverty and climate change.