Factor models are frequently used for financial modeling. In credit risk, they are a main tool of capturing dynamic dependence, and their estimation and calibration poses a difficult problem. We approach the estimation via a nonlinear filtering approach. The filtering problem is non-standard as observations consist of diffusive and point-process observations. The EM-algorithm allows to dynamically estimate the background factor process.
An illustration with data from the iTraxx over a period of five years shows a surprisingly good fit throughout the whole period.