Credit Risk Management Through Robust Generalized Linear Models

In this work, a robust methodology is developed for the classification of a sample of small and medium firms on the basis of their default probability. The importance of this classification procedure is emphasized by the New Basel Capital Accord (Basel II) for the capital adequacy of internationally active banks. The Basel accord introduces the possibility to adopt models of internal rating for the estimation of the default probability of customers' banks. The reference framework of this paper is the class of generalized linear models which allows to classify units avoiding strict assumptions such those required by the linear discriminant analysis. Another advantage of generalized linear models is the possibility to explore different links between the expected value of the dependent variable and the linear predictor. Parameters are estimated using balance ratios and data coming from Centrale dei Rischi for a set of firms which are customers of a medium sized bank of Northern Italy. Finally, we perform a robust analysis of the model estimates through the forward search in order to monitor the influence of outliers on the final classification.