Abstract

Using Credit Risk Models in Practice

Gunter Löffler

Commerzbank AG

This paper uses Monte Carlo simulations to assess the robustness of credit risk models with respect to estimation error in input parameters. In a Merton-type analysis, estimates of stand-alone default risk and fair spreads turn out to be very sensitive to uncertainty about asset volatility. Noisy volatility estimates can also lead to spurious autocorrelation in estimated default probabilities. We then examine two models of portfolio credit risk, CreditMetrics and an equity-based approach akin to the one promoted by KMV. For both approaches, uncertainty about the probability distribution of negative credit events seems to have the largest impact on the models' reliability. The effects of recovery rate uncertainty increase with the default risk of the credit instruments. They can be more important than noise in correlation estimates, which we also address. Putting all three sources of uncertainty together results in an estimation error which is still smaller than if a long time series (50 years) of returns on a comparable credit portfolio is used to assess risk. The results demonstrate the usefulness of current credit risk models, and help to identify those input parameters which deserve the greatest attention in the process of improving the models' performance.