An enterprise default discriminant model based on optimal misjudgment loss ratio

Abstract

The enterprise default discriminant analysis provides a basis for decision making on bank loans and corporate bond investments. As such, this study attempts to address the following two questions: how should an optimal feature set with the highest default discriminant accuracy be selected, and how can a default discriminant model with the lowest total classification error rate be established? For these issues, we propose two methods. First, we set up an optimal feature set by identifying the minimum default discriminant error among all feature sets of lasso-logistic regression equations and derive an optimal feature set with a non-zero weight in the equations. Second, we use the discriminant accuracy maximization as criteria in the optimal default discriminant model to derive optimal parameters; thus, we obtain the optimal discriminant model. This method ensures the classification error is minimized and occurrence of Type II errors is lowest. An empirical analysis shows a feature set with 20 features, including asset-liability ratio and total capitalization rate, has the highest discriminant accuracy. Further analysis shows that the default discriminant accuracy is highest when the ratio of Type II error to Type I error is 10.4.