Model validation metrics
ROC-AUC, Gini, KS, precision, recall, PR-AUC, calibration, Brier score, and log loss.
Reference methods
Validation and monitoring
Use a metric with a defined population, horizon, assumptions, and owner. Review discrimination, calibration, stability, and outcomes together rather than relying on a single score.
Portfolio monitoring metrics
PSI/CSI, feature drift, missingness, outcomes, migrations, arrears, recoveries, LGD, EAD, and ECL.
Metric index
26 covered metrics with formulas, scope, assumptions, segmentation, limitations, and threshold caveats.
- ROC-AUC
- Gini coefficient
- Kolmogorov–Smirnov (KS)
- Precision (positive predictive value)
- Recall (sensitivity)
- Precision–recall AUC
- Observed-to-expected (O/E) defaults
- Calibration curve
- Calibration intercept and slope
- Brier score
- Log loss
- Population Stability Index (PSI)
- Characteristic Stability Index (CSI)
- Feature drift and data-quality change
- Missingness rate
- Observed bad/default rate
- Rating migration matrix
- Delinquency roll rates
- Vintage and cohort curves
- Arrears and delinquency indicators
- Cure and re-default
- Prepayment and exit rates
- Recovery amounts and timing
- LGD outcome monitoring
- EAD and CCF outcome monitoring
- Observed versus expected ECL