AI-Based Credit Risk Modeling: A Comparative Study Between Traditional and AI- Enhanced Approaches
4 viewsAbstract
In this research, it is explored whether machine learning models based on artificial intelligence are more effective than the traditional logistic regression in predicting corporate financial distress. A sample of 20 publicly listed U.S. firms, 10 financially healthy firms and 10 confirmed distressed or bankrupt firms, were compiled based on SEC 10-K filings between 2000 and 2023. Four financial ratios were calculated on each company current ratio, debt-to-equity ratio, asset turnover, and return on assets. An estimation of a conventional logistic regression model was used as a baseline, and an AI-based classification model was estimated which was able to capture non-linear associations between financial variables. The overall classification accuracy, the Area Under the Receiver Operating Characteristic Curve (AUC-ROC), precision, and recall were used to measure model performance. The logistic regression model was 75.0 percent accurate and had an AUC- ROC of 0.780, and the AI-based model 90.0 percent accurate and an AUC-ROC of 0.940. The findings support the conclusion that machine learning methods provide a statistically significant enhancement to the traditional statistical methods in terms of corporate credit risk rating, and in specific, the positive effects in the ability to identify financially distressed companies prior to default.
Keywords
AUC- ROC, Credit Risk, Default Prediction, Financial Distress, Financial Ratios, Logistic Regression, Machine Learning