Glossary term
Model Risk
Subject: Risk Management, Operational and Model Risk
Also known as: Model Risk Management (MRM)
Model risk is the risk of loss or wrong decisions because a model is flawed, built on poor data, or used outside its intended purpose. The idea covers both errors in a model and misuse of it.
Explanation
Firms rely on models for pricing, credit decisions, capital estimates, fraud detection and forecasting. A model can fail through bad design, faulty data, coding mistakes or untested assumptions. It can also be used for a purpose it was never built for or kept in use after conditions change. Because outputs often look precise, errors are easy to trust. The risk is managed by keeping a list of models, validating them independently, setting limits on use, monitoring performance and documenting assumptions, which together make up model risk management. Artificial intelligence has widened the issue, since complex models are harder to explain. A typical supervisory review begins with the model inventory, then checks whether each model has an owner, a validation date and a stated limit of use.
Example
Mortgage pricing at a bank rests on a model. It was built for fixed-rate loans, but a product team starts using it for variable-rate loans without telling anyone. The model understates the risk and the bank prices the loans 0.3 percentage points too low. The gap is found only when profits come in below plan, and the bank has to restate its pricing.
Not to be confused with
- Model validation
- Validation is one control that tests a model. Model risk is the exposure that validation, monitoring and governance try to reduce.