Overlays vs Post-Model Adjustments (PMAs) in Credit Risk Scorecards
Source: www.peaks2tails.com
In the dynamic field of credit risk management, models are essential — but they are rarely perfect. Economic changes, emerging risks, and strategic business needs often require additional adjustments beyond what a model originally predicts. Two important techniques for refining model outputs are Overlays and Post-Model Adjustments (PMAs). Though related, they serve distinct purposes. Let’s explore how they differ and why both are critical for modern risk management.
What Are Post-Model Adjustments (PMAs)?
Post-Model Adjustments (PMAs) are systematic corrections made after a model has produced a score or a probability of default (PD). PMAs aim to fine-tune these outputs without directly altering the final accept/reject decision. Here’s a closer look at their characteristics:
-
Systematic Application: PMAs are applied across an entire group or a segment, not to individuals. For example, adjustments might target all applicants from a specific geographic area or income group.
-
Driven by External Factors: PMAs address influences the model didn’t predict — like sudden economic shifts, regulatory changes, or model deficiencies identified through monitoring.
-
Business Strategy Alignment: They help institutions align model outputs with updated business strategies, market expansions, or risk appetite changes.
-
Score Adjustment, Not Outcome Change: PMAs modify the probability or score used in the decision-making process, but they don’t directly override the model’s decision.
-
Enhancing Model Reliability: PMAs ensure that model results stay relevant in changing conditions without needing a full redevelopment.
In essence, PMAs act like a “fine-tuning knob,” subtly adjusting the model’s signals to better match reality and business objectives.
What Are Overlays?
Overlays, on the other hand, are much more direct — they intervene in the final decision itself. When an overlay is applied, it can change the model’s recommendation (accept or reject) entirely, often regardless of the model’s calculated score. Key features include:
-
Individualized or Policy-Based: Overlays can be applied case-by-case (such as manual expert review) or based on pre-set policies (such as no loans to specific industries).
-
External Information Influence: Decisions are adjusted based on additional insights or data the model didn’t consider.
-
Expert Judgment: Sometimes models miss important context. Human experts step in to apply their experience and judgment to correct or override automated decisions.
-
Total Decision Override: Unlike PMAs, overlays don’t just tweak the score; they can completely change a decision — approving someone the model would have rejected, or vice versa.
-
Handling Exceptions: Overlays are critical for handling rare or exceptional cases where model-based decisions could lead to poor outcomes.
Think of overlays as the institution’s safety valve — a way to ensure the right decisions are made, even when the model’s view is incomplete.
Why Understanding the Difference Matters
While both PMAs and Overlays are forms of model adjustment, they serve different purposes:
| Aspect | Post-Model Adjustments (PMAs) | Overlays |
|---|---|---|
| Target | Score or Probability of Default | Final Decision |
| Application | Systematic across groups | Specific cases or broad policies |
| Driver | External factors, model performance | External data, expert judgment |
| Impact | Refines model output | Overrides model outcome |
| Purpose | Align with strategy, correct model | Handle exceptions, unique scenarios |
A well-governed credit risk framework uses both techniques thoughtfully: PMAs to keep the model tuned to reality, and Overlays to manage special cases that models cannot predict.
Final Thoughts
Credit risk management is part science, part art. Models bring statistical rigor, but human judgment and business strategies must remain at the center.
Using Post-Model Adjustments and Overlays effectively ensures that institutions remain resilient, agile, and fair — especially in a world that’s always changing.
