Robust Signals for Banking Crises

International audienceWe develop an Early Warning System framework for predicting banking crises in 48 countries from 1977 to 2010. We deal with the problem of model uncertainty and omitted variables bias using Bayesian Model Averaging. Consistent with previous findings, GDP and credit growths, financial liberalization and external total debt are decisive in predicting the occurrence of banking crises. By maximizing the relative usefulness, we find an optimal level of type I and II errors. The robustness analysis shows that our results remain broadly stable when using different income groups of countries