Credit Risk Model Monitoring: What to Monitor After Validation
Credit risk model monitoring starts after validation, not before it.
Even a well-validated model can deteriorate over time as portfolio characteristics, borrower behavior, economic conditions and data quality change. That is why banks need a structured model monitoring framework to identify problems early.
What Should Banks Monitor?
A strong credit model performance monitoring process should track both statistical performance and business behavior.
Key areas usually include:
1. Model Performance
For PD models, banks should monitor whether predicted default rates remain aligned with actual defaults.
Useful measures may include:
- Actual vs predicted default rates
- Gini or ROC/AUC
- KS statistics
- Calibration by rating grade
- Actual-to-expected ratios
For LGD and EAD models, predicted values should be compared with realized recoveries and actual exposures at default.
2. Population Stability
A model may still calculate correctly but become less reliable if the portfolio changes significantly.
Population Stability Index and segment-level trend analysis can help identify model drift credit risk.
For example, a portfolio originally dominated by low-risk customers may gradually shift toward higher-risk borrowers. If the model was developed on a very different population, performance may deteriorate.
3. Data Quality
Monitoring should also check whether model inputs remain complete and consistent.
This may include:
- Missing values
- Rating gaps
- Incorrect default dates
- Duplicate records
- Changes in source systems
- Unusual movements in balances or limits
Poor data quality can create apparent model deterioration even when the methodology itself is still sound.
4. Overrides and Expert Judgement
Banks should monitor how frequently model outputs are overridden.
A rising level of manual adjustment may indicate that the model is no longer capturing portfolio behavior properly.
For IFRS 9, this also includes monitoring management overlays, scenario adjustments and staging overrides.
5. Forward-Looking Performance
For ECL models, monitoring should assess whether macroeconomic relationships remain reasonable.
This includes reviewing:
- Macroeconomic variable relevance
- Scenario behavior
- Scenario weights
- PIT adjustments
- ECL sensitivity to changing economic conditions
Ongoing Model Validation
Ongoing model validation does not always mean performing a full independent validation every month.
Instead, regular monitoring should identify whether performance has breached predefined thresholds and whether recalibration, redevelopment or deeper validation is required.
This creates an early-warning system for model deterioration.
Credit Risk Model Governance
Strong credit risk model governance should define:
- Monitoring frequency
- Performance thresholds
- Escalation rules
- Model ownership
- Validation responsibilities
- Recalibration triggers
- Documentation requirements
Without clear governance, monitoring can become a reporting exercise rather than a risk-control process.
Key Takeaway
Credit risk model monitoring is about checking whether a model remains reliable after implementation.
The best monitoring frameworks combine performance testing, stability analysis, data-quality checks, override monitoring and forward-looking assessment.
The objective is simple: detect deterioration early enough to take action before the model becomes unreliable.