Crisis Duration Bias: A Multinomial Logit Approach for Islamic Banking

In this study an early warning system (EWS) model is constructed for predicting the banking crisis (BC) of Islamic banks (IBs). To this aim, multinomial logit regression is employed regarding the levels of the explanatory variables are different in their tranquil and crisis/post-crisis levels which is called as the crisis duration bias (CDB). The analysis is conducted with three time windows following Bussiere and Fratzscher (2006) and Caggiano et al. (2016). Accordingly, the IBs in each country experiences three states as tranquil period (Yi,t=0), a pre-fragility period (Yi,t=1) and post-fragility period (Yi,t=2). The results address CDB problem and both macroeconomic and banking sector variables are crucial in determining the early warning indicators of the fragility of IBs. This study also draws attention to predictive performances of binomial logit approach and multinomial logit approach. According to the results, the overall predictive performance of the multinomial logit regression reveals superior results than binomial logit regression.