Abstract:
Empirical research in climate economics often relies on panel regressions of different outcomes on disaster damages. Interpreting these regressions requires an assumption that error terms are uncorrelated across counties and time, which climate science research suggests is unlikely to hold. We introduce a methodology to identify spatial and temporal clusters in natural disaster damages datasets, and show that accounting for clustering affects observed economic effects of disasters. Specifically, counties tend to experience 0.45% more disaster damage for every 1% increase in damage across other intra-cluster counties. Moreover, accounting for clustering makes some hazard types, such as droughts, appear more damaging.