Comparing Logit-based Early Warning Systems: Does the Duration of Systemic Banking Crises Matter?

This paper compares the performance of binomial and multinomial logit models in the contextof building early warning systems (EWS) for systemic banking crises. We test the hypothesisthat the predictive performance of binomial logit models is hampered by what we define asthecrisis duration bias, arising from the decision to either treat crisis years after the onset ofa crisis as non-crisis years or remove them altogether from the sample. In line with our hy-pothesis, results from a large sample of world economies suggest that i) the multinomiallogit outperforms the binomial logit model in predicting systemic banking crises, and ii) thelonger the average duration of the crisis in the sample, the larger the improvement.