COMPARATIVE PERFORMANCE OF MACHINE LEARNING MODELS INCORPORATING MACROECONOMIC INDICATORS FOR CREDIT RISK EARLY WARNING

Abstract

Credit risk management remains a critical concern for financial institutions, particularly during economic downturns when default rates surge unexpectedly. This research evaluates the comparative performance of six machine learning models—Logistic Regression, Random Forest, Gradient Boosting, Support Vector Machine, Neural Networks, and XGBoost—in predicting credit default by incorporating macroeconomic indicators alongside traditional borrower characteristics. Using a dataset of 45,000 loan accounts from 2015-2023, the study examines how macroeconomic variables including GDP growth, unemployment rate, inflation, and interest rates enhance predictive accuracy of credit risk early warning systems. Results demonstrate that ensemble methods, particularly XGBoost and Gradient Boosting, achieve superior performance with AUC scores of 0.89 and 0.87 respectively when macroeconomic indicators are integrated. The inclusion of macroeconomic factors improves prediction accuracy by 8-12% compared to models using only borrower-specific variables. Neural networks show strong performance in capturing non-linear relationships between economic conditions and default probability. The findings provide practical insights for financial institutions seeking to develop robust early warning systems that anticipate credit risk deterioration during economic stress periods.