Research Seminar in Economics

June 3, 2026

Extreme weather events pose a growing risk to global production. The growing body ofliterature on the impacts of natural disasters, such as floods, on households, firms, andeconomic development largely relies on event study designs exploiting publicly availableclimate data. This paper documents the challenges and pitfalls associated with thisapproach using a commonly used database of floods. To do so, we study the responsesand adaptations to flood exposure of firms in the automotive industry. We show that,in our setting, (i) floods are often missing or spatially / temporarily misallocated

in thedata; (ii) canonical Difference-in-Differences methods are sensitive to the design and leadto qualitatively different conclusions. We explore the mechanisms behind this dispersionand provide guidance on how to apply the methods to these datasets to deliver credible results.