Model Risk and Validation
Two pricing models can fit the same instruments and still disagree about the value or hedge of an instrument that does not trade. A validation review tests that disagreement with independent prices, simple payoff bounds and stress cases. An option priced within a defensible range can still have a hedge that is sensitive to the model assumptions.
How it works
Model risk arises when different assumptions can agree with calibration data and disagree on a decision. Document the target payoff, which quotes were used for fitting, and which were withheld for testing. Model-free bounds, alternative specifications, and risk sensitivities each answer a distinct validation question.
Worked example
Our report constructs different terminal distributions that reprice the same observed options. With a hypothetical ceiling of 140 on the terminal share price, an attainable valuation range is USD 4.80–5.00. Raising the ceiling to 150 widens the range to USD 4.80–5.70. Two fitted models also prescribe different stock hedge sizes under the same stated shock.
Validation and limitations
Choose an independent benchmark that tests the intended use. Verify what the model predicts when a key calibration or support assumption changes, and identify the remaining uncertainties when several models satisfy the available checks. Record whether the decision concerns a mark, a hedge or an intended use beyond the calibration sample.
Further reading
- Model Uncertainty and Its Impact on the Pricing of Derivative Instruments
Rama Cont · PDF
- Revised Guidance on Model Risk Management
Federal Reserve, OCC and FDIC · U.S. supervisory guidance