Testing discriminant methodologies on general and industry specific Italian company models

This paper aims at a better understanding of the effectiveness and efficiency of methodologies that estimate default risk, for both general and industry specific models. Analysis is based on regional Italian companies (Tuscany). The main research question is “how is the probability of default efficiently estimated according to the Basel II New Accord of Capital (NAC)?”. This implies testing for the best methodology to discriminate between “in bonis” and “in default” companies within the new regulated European framework. Other tested propositions concern model generalization: where Industry specific models are compared to the general model. Results based on a parity sample of 300+300 Tuscan companies show that Multivariate Discriminant Analysis (DA-GEE) and Logit Regression (LR) provide better estimation of Default Risk than can Partial Least Squared Regression Discriminant Analysis (PLS-DA). Using PLS-DA, 42% of these companies fall into the overlapping area (grey zone); using DA-GEE and LR, respectively, only 31% and 29% of companies are misclassified. Industry specific models offer a more efficient classification of the default if compared to the general model, but percentages of error are not significantly lower.