Early-Warning Stress Testing for Consumer Loans Based on Dynamic Macroeconomic Scenario Generation

Abstract

This study proposes a dynamic stress-testing framework for consumer-finance portfolios by generating forward-looking macroeconomic scenarios using a vector autoregression model combined with Monte Carlo simulation. The scenarios are fed into a panel logistic-regression engine trained on 4.2 million loan accounts from 2015-2024. Using 1,000 simulated macro paths, the model quantifies stressed default probabilities up to 12 months ahead. Results show that under the 5% most adverse scenarios, 90-day delinquency rates rise by 28.9-46.2%, depending on product type. The model achieves an ROC-AUC of 0.82 in out-of-sample stress periods and captures 72.5% of actual default surges during the 2020 downturn. This approach offers a systematic tool for scenario-driven risk assessment in consumer lending.