Enhanced credit default models for heterogeneous SME segments

Considering the attention placed on SMEs in the new Basel Capital Accord, we propose a set of Bayesian and classical longitudinal models to predict SME default probability, taking unobservable firm and business sector heterogeneities as well as analysts’ recommendations into account. We compare this set of models in terms of forecasting performances, both in-sample and out-of-sample. Furthermore, we propose a novel financial loss function to measure the costs of an incorrect classification, including both the missed profits and the losses given defaults sustained by the bank. As for the in-sample results, we found evidence that our proposed longitudinal models outperformed a simple pooled logit model. Besides, Bayesian models performed even better than classical models. As for the out-of-sample performances, the models were much closer, both in terms of key performance indicators and financial loss functions, and the pooled logit model could not be outperformed