Predicting Banking Crises: A Horse Race using Dynamic-Recursive Forecasting

This study conducts a horse race of recently developed econometric methods that pre- dict banking crises and applies a more accurate dynamic-recursive forecasting technique to improve their predictive performance. Our results show that simple pooled models that account for regional heterogeneity and the entire crisis period outperform more complex models and the practice of dropping post-crisis periods. Finally, the dynamic signal ex- traction approach is recommended for policymakers who value avoiding banking crises at all costs, while the binomial logit model is more suitable for less conservative policymakers who consider the economic and social costs of false alarms.