Empirical performance of loss given default prediction models

The global financial crisis highlighted the fact that default and recovery rates of multiple borrowers generally deteriorate jointly during economic downturns. The vast majority of the literature, as well as many industry credit-portfolio risk models, ignore this and analyze default probabilities and recoveries in the event of default separately. As a result, the models project losses that are too low in economic downturns such as the recent financial crisis. Nevertheless, alternatives that incorporate the dependence between probabilities of default and recovery rates have been proposed. This paper is the first of its kind to assess the performance of these structurally different approaches. Four banks using different estimation procedures are compared. We use root mean square errors and relative absolute errors to measure the predictive accuracy of each procedure. The results show that models accounting for the correlation of default and recovery do indeed perform better than models ignoring it.