Identifying key factors in accounting-based models of credit risk based on a predictive model averaging approach

Accounting-based models in credit risk have been shown to perform well in predicting a firm's ability to meet its financial obligations, even if they include only a limited number of financial ratios measuring different aspects of the firm's financial health. However, there is little agreement on a specific set of ratios to be incorporated in these models in the existing literature. This study provides guidance on the set of accounting ratios to include in such models based on empirical results obtained for rating implied 1-year probabilities of default for a data set of large U.S. corporations. The analysis performed consists of a predictive Bayesian model averaging approach where the models included are restricted in the number of accounting ratios from different categories. The identified model is shown to provide similar predictive performance as more complex models, while retaining interpretability and simplicity.