Abstract
Modeling credit default swap (CDS) spreads usually requires more than calibrating a single model. Focusing on the structural model category, CreditGrades as in Finger (2002) and the Das and Sundaram (2000) trinomial tree model are two promising model candidates. Using a broad sample of 54 individual CDSs from the DJ CDX.NA.IG universe during May 14th, 2004 to July 7th, 2006, we find that the trinomial model steadily outperforms the CreditGrades benchmark. However, both models' out-of-sample performance can be relatively poor, with a mean correlation in spread changes below 50% and vast cross-sectional differences in model performance. Interestingly, company specific variables hardly explain these differences. Applying the models to companies with above average R-squared in the spirit of Roll (1988) remarkably improves the results. We conclude that above average correlation with a CDS market proxy indicates above average structural model performance.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have