DEFAULT CORRELATION IN REDUCED-FORM MODELS

Fan Yu

Reduced-form models have proven to be a useful tool for analyzing the dynamics of credit spreads. However, some have recently questioned their ability to match the level of empirical default correlation. The key concern appears to be the assumption that defaults are independent conditional on the state variables driving the default intensity. In this paper, I use a “thought experiment” as well as numerical examples calibrated to recent studies to show that the model-implied default correlation can be quite sensitive to the common factor structure imposed on the default intensity. Therefore, the “inability” of reduced-form models to generate sufficient default correlation has more to do with a restrictive common factor structure than the assumption of conditional independence.