Support vector regression for loss given default modeling

Loss given default modeling has been increasingly crucial in credit risk management in the past few years. In this paper, support vector regression techniques are applied to predict loss given default of corporate bonds, where improvements are proposed to increase prediction accuracy by accounting for heterogeneity of bond types. Two different transformation distributions on loss given default - logistic and beta - are also investigated. The empirical study has three major results. First at an aggregated level, a total of 13 methods are compared in terms of out of sample prediction performances. The proposed improved versions of support vector regression techniques outperform other methods significantly given the p-values of pair-wise t-tests. Second, at a segmented by bond seniority level, least square support vector regression also demonstrates significantly better predictive abilities compared with the other statistical models. Third, transformations of loss given default do not reduce the prediction errors. The empirical results show that support vector regression techniques are promising in predicting loss given default.