Macroeconomic-Sensitive Credit Risk Forecasting for Consumer Finance Using Regime-Switching Models

This study develops a regime-switching credit-risk forecasting model for consumer-finance portfolios that dynamically adapts to changing macroeconomic conditions. Using monthly data from 2016–2024 covering 3.8 million loan accounts and 27 macro indicators, the model identifies two latent economic regimes—expansion and stress—via a Markov-switching structure. A hybrid system integrating gradient boosting with regime-specific probability calibration is used to predict 90-day delinquency. Results show that incorporating regime states reduces the mean absolute forecasting error by 22.7% compared with non-regime models, and improves early-warning lead time by 3.4 months on average. Stress testing under simulated recession conditions indicates a potential 31.5% increase in portfolio-wide default probability. The findings demonstrate the effectiveness of regime-aware models for credit-risk management in volatile environments.