Based on the model of Schmid & Zagst (2000) for pricing defaultable bonds, we develop a modified hybrid model which allows for correlations between the different factors. To embed this new model, we compare it to an intensity-based model according to Duffee (1996) and a structural model based on Bakshi et al. (2001). The parameters of the different models are estimated by the Kalman filter methodology as described in Schmid (2002). Based on these estimates we compare different quantities with regard to the performance of the models. First we consider the absolute pricing errors in- and out-of-sample, then we examine a regression test to explain the changes of the yields. Further we discuss the sensitivities of the several factors on the spreads and finally we compare the observable default rates to the filtered firmspecific factors. We can show that our new model provides the best results in comparison.
International Journal of Pure and Applied Mathematics