Using alternative models and a combining technique in credit rating forecasting: An empirical study

Abstract

Credit rating forecasting has long time been very important for bond classification and loan analysis. In particular, under the Basel II environment, regulators in Taiwan have requested the banks to estimate the default probability of the loan based on its credit classification. A proper forecasting procedure for credit rating of the loan is crucially important in abiding the rule. Credit rating is an ordinal scale from which the credit category of a firm can be ranked from high to low, but the scale of the difference between them is unknown. To model the ordinal outcomes, this study first constitutes an attempt utilizing the ordered logit and the ordered probit models, respectively. Then, we use ordered logit combining method to weigh different techniques' probability measures as described in Kamstra and Kennedy (International Journal of Forecasting 14, 83–93, 1998) to form the combining model. The samples consist of firms in the TSE and the OTC market and are divided into three industries for analysis. We consider financial variables, market variables, as well as macroeconomic variables and estimate their parameters for out–of–sample tests. By means of cumulative accuracy profile, the receiver operating characteristics, and McFadden R2, we measure the goodness–of–fit and the accuracy of each prediction model. The performance evaluations are conducted to compare the forecasting results, and we find that combining technique does improve the predictive power.