📘 Qualitative Overlay Guide for PD Modeling
1. What is a Qualitative Overlay?
A qualitative overlay (sometimes called a model adjustment, post-model adjustment, or post-management adjustment) is when you add judgment on top of a purely quantitative model’s output.
- The model gives you an estimated PD (say 1.8%).
- But based on new information or expert knowledge that wasn’t captured in the data, you adjust it (say, to 3%).
- Purpose: to ensure the model’s output is fit for decision-making, especially when the data or model doesn’t reflect current or forward-looking risks.
2. When to Apply an Overlay
You don’t apply overlays casually; they’re used in specific situations where the model is known to have blind spots. Common cases:
- Data Gaps or Limitations
- Example: A new loan product has no historical defaults, so the PD model may underestimate risk.
- Overlay: Add a conservative buffer until data matures.
- Macroeconomic or Sector Shocks
- Example: Oil & gas companies in 2020 suddenly faced huge default risk not reflected in prior data.
- Overlay: Increase PDs for energy sector loans.
- Forward-Looking Risk Factors
- Example: A corporate borrower’s management is under investigation, which is not in historical financial ratios.
- Overlay: Increase PD to reflect governance risk.
- Model Weaknesses Identified
- Example: The logistic regression model underpredicts defaults for very small firms.
- Overlay: Apply a higher PD floor for small-firm exposures.
3. How to Apply an Overlay
There are structured ways to apply overlays so they’re defensible, documented, and consistent:
- Define the Trigger / Rationale
- Document why the overlay is needed (e.g., “COVID-19 impact on hospitality sector”).
- Determine the Method
Several approaches can be used:
- Additive Adjustment: +X% to predicted PD.
- Multiplicative Adjustment: Scale PD by a factor (e.g., ×1.5).
- Floor / Cap: Set minimum PD (e.g., no corporate loan can have PD < 0.5%).
- Segment-Specific Adjustment: Apply overlays to certain industries, geographies, or product types.
- Calibrate the Size of the Overlay
- Use external data, expert panels, stress scenarios, or regulator guidance.
- Example: If sector default rates rose from 2% → 4%, apply a 2% absolute overlay.
- Governance and Review
- Every overlay should be approved, documented, and backtested.
- Regulators expect overlays to be temporary, not permanent substitutes for poor models.
4. Example: PD Overlay in a Loan Portfolio
- Baseline Model: Logistic regression predicts PD = 1.5% for Hotel Co.
- Issue: COVID-19 has pushed observed hotel default rates closer to 6%.
- Overlay: Apply multiplicative factor ×3 → Adjusted PD = 4.5%.
- Documentation: “Overlay applied to hospitality sector loans due to pandemic shock, reviewed quarterly.”
✅ Key Takeaway:
A qualitative overlay is a controlled, documented adjustment to a quantitative model’s outputs, used when the model cannot fully capture new risks, data limitations, or forward-looking conditions. Overlays ensure the model remains fit-for-purpose without replacing the model itself.
Qualitative Overlay Checklist – Example
1. Overlay Identification
- Model Name: Corporate Loan PD Model (Logistic Regression)
- Metric impacted: Probability of Default (PD)
- Segment / Portfolio affected: Hospitality sector corporate loans (hotels, restaurants, travel services)
2. Rationale for Overlay
- Reason: COVID-19 caused a sudden, unprecedented decline in hospitality revenues, not reflected in historical training data.
- Evidence:
- Industry reports (Moody’s, S&P) showing sector-wide defaults increasing to 6% in 2020 (vs 2% historical).
- Bank’s internal monitoring: sharp increase in payment deferrals and covenant breaches in hospitality borrowers.
- Regulatory guidance encouraging forward-looking adjustments during COVID-19.
3. Overlay Methodology
- Type of adjustment: Multiplicative (× factor)
- Formula / approach: Adjusted PD = Base PD × 3
- Example calculation:
- Hotel Co. predicted PD (model) = 1.5%
- Overlay factor = ×3
- Adjusted PD = 4.5%
4. Overlay Size & Calibration
- Size of adjustment: 3× base PD
- Basis for calibration: External rating agency reports showing ~3x increase in sector default rates.
- Sensitivity analysis: Yes – tested overlay multipliers of 2×, 3×, 4×. The 3× factor aligned best with external observed defaults.
- Justification: Conservatism balanced with alignment to observed sector data.
5. Governance
- Approving authority: Credit Risk Committee
- Date of approval: April 15, 2020
- Expected duration: 12 months (subject to quarterly review)
- Conditions for removal: Hospitality sector default rates return to pre-COVID levels (<3%), or sufficient new model recalibration performed.
6. Review & Monitoring
- Monitoring frequency: Quarterly
- Metrics to track:
- Actual vs predicted defaults in hospitality sector.
- Sector delinquency trends.
- Macro indicators (tourism activity, travel restrictions).
- Next review date: July 2020
- Responsible team: Portfolio Risk Analytics Team
✅ Result:
The overlay ensured the bank’s PD estimates reflected real-world heightened risk, avoided underestimation of capital needs, and stayed defensible to regulators by being documented, calibrated, and temporary.
Perfect — let’s extend the example into a portfolio-level overlay framework, where multiple overlays can be applied in a structured, transparent way. Think of it as a layered adjustment process on top of your quantitative PD model.
Portfolio-Level Overlay Framework
1. Organizing Overlays
Banks usually categorize overlays into types, then apply them at the portfolio, segment, or borrower level.
Common categories:
- Macroeconomic overlays (forward-looking shocks)
- Sector / Industry overlays (sector-specific risks)
- Borrower-level overlays (idiosyncratic risks)
- Data / Model limitation overlays (new products, thin data, structural weaknesses)
2. Example: Portfolio-Level PD Overlays
A. Macroeconomic Overlay
- Trigger: Recession scenario, GDP drop 3%.
- Method: Additive +0.5% PD across the entire portfolio.
- Rationale: Broad deterioration in credit quality, beyond historical training sample.
B. Sector Overlay (Hospitality Sector)
- Trigger: COVID-19 sector shock.
- Method: Multiplicative ×3 factor applied to hospitality borrowers.
- Rationale: Observed defaults surged; model underestimates risk.
C. Borrower-Level Overlay
- Trigger: Borrower flagged for governance risk (CEO fraud case).
- Method: Override model PD from 1.2% → 5%.
- Rationale: Forward-looking risk not captured in quantitative predictors.
D. New Product Overlay
- Trigger: Launch of a new SME lending program, no default history.
- Method: PD floor at 2% until enough performance data is available.
- Rationale: Conservatism due to absence of training data.
3. Governance Structure
- Each overlay documented separately using the checklist template.
- Portfolio overlays aggregated to see overall impact on risk-weighted assets (RWA) and expected losses.
- Quarterly review to add, modify, or retire overlays.
4. Example Portfolio Impact Table
Overlay Type |
Segment Affected |
Method |
Avg PD Before |
Avg PD After |
Impact on Portfolio EL* |
Macroeconomic |
Entire portfolio |
+0.5% additive |
2.1% |
2.6% |
+\$15M |
Sector (Hospitality) |
Hotels & Restaurants |
×3 multiplier |
1.5% |
4.5% |
+\$22M |
Borrower-Level |
Borrower XYZ Corp. |
Direct override |
1.2% |
5.0% |
+\$3M |
New Product |
SME lending program |
2% PD floor |
0.8% |
2.0% |
+\$7M |
*EL = Expected Loss
✅ Key Takeaway:
A portfolio-level overlay framework:
- Ensures overlays are consistent, transparent, and not ad hoc.
- Allows management and regulators to see both individual justifications and aggregate portfolio impact.
- Keeps overlays temporary while new data or recalibration is developed.