Machine learning approaches to credit risk: Evaluating Turkish participation and conventional banks

Abstract

This study investigates the impact of competition on credit risk in the Turkish banking system, focusing on Islamic (participation) and conventional banking under the same regulatory conditions regarding credit risk. The credit risk model is trained using CatBoost, extreme gradient boosting (XGBoost), random forest, and LightGBM algorithms, and the results are analyzed using Tree SHAP (SHapley Additive exPlanation) algorithms with swarmplots. The empirical analysis covers 33 conventional and 6 Islamic banks in T & uuml;rkiye, using annual data between 2009 and 2022. The findings reveal that (1) credit risk is relatively higher at participation banks than conventional banks; (2) competition increases credit risk; (3) loan size is a key determinant of credit risk; (4) profitability increases credit risk; and (5) economic growth reduces credit risk. This study recommends some policy measures, such as increasing the economy of scale at Islamic banks and implementing specific regulations at participation banks to reduce risk.