Distributionally robust optimization approaches to credit risk management of corporate loan portfolios

Empirical divergence-based distributionally robust optimization (DRO) offers a novel approach to managing credit risk in financial institutions by accounting for data uncertainty and model misspecification. This study examines two specific applications of DRO: loss forecasting for predicting the significant increase in credit risk (SICR) status of loans under the International Financial Reporting Standard 9 expected credit loss provisioning framework; and risk limit management of corporate loans. Our findings indicate that DRO methods improve model robustness by explicitly addressing distributional uncertainty in potential future scenarios. By considering worst-case scenarios within an ambiguity set, DRO enables financial institutions to make more informed modeling decisions that are aligned with regulatory requirements, ultimately leading to more reliable risk management practices.