Interpretable EBM Enhances Credit Risk Stress Testing
This study develops an interpretable explainable boosting machine (EBM) within a discrete-time survival analysis to predict forward-looking probability of default for a real-world credit-card portfolio. The EBM model achieves predictive accuracy comparable to logistic regression while using one-fifth of the sample and avoiding extensive manual feature engineering, and it also improves macro-sensitivity by addressing EBM extrapolation limitations.
Key Points
- 1Demonstrates EBM-based survival model matches logistic regression accuracy using one-fifth the training sample size.
- 2Improves macro-sensitivity by addressing EBM extrapolation limitations, capturing nonlinear relationships without heavy manual feature engineering.
- 3Offers interpretable shape functions for model validation, aiding regulatory compliance and faster risk-model development.
Scoring Rationale
Practical interpretable-model application across credit risk; limited novelty and single-source demonstration restricts broader, immediate industry adoption.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
