Research on Credit Risk Forecasting and Stress Testing for Consumer Finance Portfolios Based on Macroeconomic Scenarios

Credit risk forecasting in consumer finance must fully account for macroeconomic fluctuations ( ) to ensure portfolio stability and meet regulatory requirements. This study constructs default probability prediction models using logistic regression, XGBoost, and LightGBM based on 5 million credit accounts, then conducts stress testing under macroeconomic scenarios including unemployment rate, inflation, and interest rate shocks. Results indicate that within a 24-month rolling window sample, the XGBoost model achieves an AUC of 0.82, surpassing logistic regression by 7.5%. Under severe recession scenarios, the model accurately captures the trend of default rates rising to 6.3% with an error margin below 0.5%. Simultaneously, stress test results provided portfolio-level Value at Risk (VaR) and Expected Shortfall (ES) intervals, enhancing the robustness of capital adequacy assessments. The study demonstrates that integrating machine learning with macroeconomic scenario analysis strengthens systemic risk monitoring, supports consumer protection, and provides quantitative tools for regulatory stress testing.