A0697
Title: Logit and multinomial logit models for early warning systems: On the duration of systemic banking crises
Authors: Leone Leonida - Queen Mary University of London (United Kingdom) [presenting]
George Kapetanios - Kings College, University of London (United Kingdom)
Pietro Calice - The World Bank (United States)
Giovanni Caggiano - University of Padova (Italy)
Abstract: The performance of the binomial and multinomial logit models in forecasting systemic banking crises is compared in the context of building early warning systems. We argue that, because the average duration of the crises in the sample of countries is longer than one year, the predictive performance of binomial logit models is likely to be hampered by what we define the crisis duration bias. The bias arises from the decision to either treat crisis years after the onset of a crisis as noncrisis years or remove them altogether from the model. A simple Monte Carlo experiment shows that, if compared to the binomial logit model, the multinomial logit approach improves the predictive power of the EWS. We apply the latter to a sample of world economies. Results strongly support the use of the multinomial logit model.
