Credit Portfolio Risk - Modelling, Estimation and Backtesting
- Publisher: Universität Mannheim
This thesis studies an asset value-based approach for the valuation of credit portfolio risk including the estimation and the backtesting of the model’s forecasting ability. On the basis of the credit valuation model proposed by Black and Cox (1976) an asset value-based factor model for the forecasting of credit portfolio risk is suggested. The model includes credit losses from the default of obligors as well as negative changes of credits’ mark-to-model values. Using market prices of defaultable corporate bonds a quasi-maximum likelihood estimation of the state-space models of systematic factors and asset values of risk classes is performed. The backtesting procedure assesses the forecasting ability of the credit portfolio model using three zones of credit portfolio loss that are defined analogously to the regulatory backtesting of internal market risk models. Probability distributions of credit portfolio loss are simulated and the backtesting zones of portfolio loss are compared for different credit portfolios and different variations of the backtesting procedure, the model structure and the parameters. The results show that for typical levels of confidence and typical time horizons the Credit-VaR exceeds banks’ capital considerably. Furthermore, backtesting zones of credit portfolio loss that fit to banks’ capital levels result in a level of test significance that is distinctly lower than in the backtesting of market risk models. It is concluded that the revised capital requirements set by the Basel Committee do not prevent credit portfolio losses exceeding banks’ capital with the statistical confidence presumed by supervisory authorities.