Measuring portfolio credit risk: modelling versus calibration errors

A model-based assessment of credit risk is subject to both specification and calibration errors. Focusing on a well known credit risk model, we propose a methodology for quantifying the relative importance of alternative sources of such errors and apply this methodology to a large data set. We find that flawed calibration of the model can substantially affect the measured level of portfolio credit risk. By contrast, a model misspecification generally has a limited impact, especially for large, well diversified portfolios.