Improvements in rating models for the German corporate sector

Group-specific estimations can significantly improve the predictive power of accountingbased rating models. This is shown using a binary logistic regression model applied to the Deutsche Bundesbank's USTAN dataset, which contains 300,000 financial statements provided by German companies for the years 1994 to 2002, i. e. throughout a complete business-cycle. The robustness and the representability of this result is verified through out-of-sample tests and through comparisons with a benchmark model which applies the variables of Moody's RiskCalcTM for Germany. -- Credit Risk ; Credit Rating ; Probability of Default ; Logistic Regression