Modeling Probability of Default of LDP-Portfolios for Commercial Bank Provision Estimation

In accordance with the aim, current thesis provides PD-model for low-default portfolios. Initial data is represented by financial reports for 100 companies of USA industrial and energy sector. Dependent variable consists of historical ratings for each company. Modeling probability of default provides efficient estimation of credit risk components and bank’s provision. Obtained ordered logit models for panel data have relatively low coefficient of determination however these models allow to identify significant influence of selected independent variables on rating. Validation of the models showed relatively high forecasting power for group of companies with high level of equity, model ranges prevailing part of the companies in accordance with applied benchmark. Based on conducted analysis several ways of further modeling were provided. In order to increase forecasting power, initial data has to be improved by usage of new variables highly correlated with the ratings. Single-index method for longitudal data appears to be the best approach for modeling credit rating and estimation of bank’s provision.