Survey-to-Survey Poverty Monitoring under Economic Shocks : The Role of Proxies and Flexible Nonlinear Learners

In many low-income countries, consumption surveys are too infrequent to track poverty during economic shocks. Survey-to-survey imputation can fill this gap, but its reliability depends on whether prediction models estimated in one period remain valid in another. This paper develops a formal identification framework for the transportability of survey-to-survey estimators and shows that failures during crises arise primarily from missing shock-responsive information rather than from insufficient model flexibility. When such information is omitted, both linear models and flexible learners yield biased poverty estimates. When fast-changing proxy variables that track unobserved welfare changes are included, out-of-sample validity can be restored under well-defined conditions. Poverty estimation, however, requires a stronger condition—stability of the full conditional distribution of welfare—beyond that needed for mean welfare estimation. Monte Carlo simulations and evidence from Afghanistan, Uganda, and Rwanda support these predictions. The findings imply that improving questionnaire content is more important than increasing model complexity for timely poverty measurement in the face of shocks.