Improving Accuracy of Insolvency Prediction for German Savings Banks and Credit Cooperatives

This study employs a rating model based on SVM to seek to improve hazard model’s accuracy prediction of default probability for German savings banks and credit cooperatives. For variables, which are chosen using linear and parametric model, SVM yields only moderate (mostly statistically insignificant) improvement. When we replace returns on equity variable by the its first difference divided by its standard variation, we add nonlinearity to the relationship between variables used in the analysis, and thus are able to obtain significant improvement in prediction accuracy. While results demonstrate advantage of SVM rating system over that based on linear hazard model, it remains to automate the procedure of variable selection for obtaining even greater accuracy in default prediction.