Estimating Probabilities of Default with Support Vector Machines

This paper proposes a rating methodology that is based on a non-linear classification method, the support vector machine, and a non-parametric technique for mapping rating scores into probabilities of default. We give an introduction to underlying statistical models and represent the results of testing our approach on Deutsche Bundesbank data. In particular we discuss the selection of variables and give a comparison with more traditional approaches such as discriminant analysis and the logit regression. The results demonstrate that the SVM has clear advantages over these methods for all variables tested.