Estimating Poverty Transitions from Repeated Cross-Sections: A Statistical Perspective

Studying poverty transitions is challenging due to the limited availability of longitudinal data, leading to a growing interest in methods that estimate transition probabilities via cross-sectional data. These approaches can be grouped into two main strands: a parametric approach, which assumes a distributional model for income and leverages econometric techniques on pseudo-panels, and a semi-parametric approach, which employs matching procedures to construct synthetic panels. A critical step in both methodologies is estimating income correlation across time that is not directly available from cross-sectional data. Although these methods are popular in the economics literature, to the best of our knowledge, they have not been systematically reviewed or assessed from a statistical perspective. In this paper, we examine these methods and their limitations before proposing a novel scenario-based framework with two alternative approaches, each aligned with a distinct strand of literature. Specifically, the parametric proposal incorporates Bayesian models that use scenarios as prior information for income autocorrelation, while the semi-parametric approach is based on a matching-based procedure. The latter includes a tuning parameter that adjusts the number of neighbors to regulate autocorrelation. Finally, we evaluate and compare the discussed methods, including our newly proposed ones, through a Monte Carlo simulation study based on Italian EU-SILC data.