A Bayesian-Inspired Approach to Improving the Precision of Synthetic Panel Estimates of Poverty Dynamics

This paper proposes a Bayesian-inspired refinement to the synthetic panel method introduced by Dang et al. (2014), yielding point estimates of poverty transitions and other income mobility metrics. It uses bootstrap samples of household income (and consumption expenditure) correlation coefficients observed in available panel surveys to proxy for correlations over time in countries where no panel surveys are available. This allows more precise estimation of poverty dynamics based on repeated cross-sectional surveys. We present results in the form of a band of two standard deviations (both plus and minus) around the point estimate. A validation analysis using actual Tanzanian panel data shows that, with few exceptions, these bands overlap with the true estimates’ confidence intervals. Subsequent application to Malaysian poverty dynamics over the period 2004-2022 illustrates how the method can be employed. The resulting four standard deviation bands are much narrower than intervals based on the original, conservative, synthetic panel bound estimates, in particular for conditional probabilities, such as the chance of escaping poverty. Hence, the results of this approach yield more relevant policy information.