55%
Kolmogorov–Smirnov separation in validation
Probability of default model
Logistic regression, weight of evidence and information value, vintage and roll-rate analysis
A logistic-regression model of credit card default. Vintage curves set a 6‑month performance window; roll-rate analysis tracked accounts within it.
- Narrowed the candidate variables with break-and-heal binning, weight of evidence and information value, stepwise regression and cluster analysis.
- Set warning and breach thresholds for ongoing monitoring.
- KS
- 55%
- PSI
- 13%
- MAPE
- 10%

