Systematic forecast errors from extreme weather events in credit risk models

Extreme weather events generate systematic forecast errors in credit risk models. Using proprietary loan-level data covering approximately 250,000 corporate loans over 2012–2023, we find that exposure to extreme weather significantly predicts realized default rates after conditioning on model-implied probabilities. Effects are stronger for early-year exposures and reflect persistent regional vulnerability rather than transitory fluctuations. These results suggest that omitted climate risk factors contribute to systematic model miscalibration, robust across specifications. More broadly, when relevant weather events are excluded, forecast errors may become systematic rather than idiosyncratic, with implications for model design and the efficient allocation of credit.