Banking Crises as Rare Events

This paper argues that estimates drawn from the extensive literature on the determinants of the probability of banking crises may be significantly biased. This bias stems from the simple fact that banking crises are rare or infrequent events. In fact, once the aforementioned biases are corrected, the probabilistic models commonly used to anticipate the occurrence of a banking crisis see their predictive power increase substantially. This is relevant to the design of early-warning systems, since the enormous costs associated with the occurrence of a banking crisis could be reduced if it is identified in a timely manner.