# Option Model Risk: Assumptions, Inputs, Calibration, and Implementation

Understand why a mathematically correct option model can still misprice or mismeasure risk, and how to validate structure, data, calibration, code, outputs, and use.

Canonical: https://wiki.fcontext.com/options/option-model-risk/
Fact checked: 2026-07-22

> For educational purposes only; not investment advice.

<a id="answer"></a>

## Direct answer

**Option model risk** is the risk of adverse decisions or losses because a model is wrong, is fed inappropriate data, is implemented incorrectly, or is used outside its intended purpose. A formula can be coded exactly and still produce an unsuitable price, Greek, hedge, probability, or stress result.

A model output is conditional on its contract interpretation, assumptions, inputs, calibration, numerical method, and timestamp. It is not an executable market price or a guarantee that the assumed dynamics will occur. Model risk cannot be eliminated by adding decimal places; it must be identified, challenged, bounded, and monitored.

<a id="mechanism"></a>

## Five layers of model risk

### 1. Contract and structural risk

The model may represent the wrong payoff, exercise style, settlement method, multiplier, deliverable, dividend treatment, or path dependence. A European formula cannot directly capture American early exercise. Constant volatility cannot reproduce a full strike-and-maturity volatility surface or specify how that surface moves after a spot shock.

Structural assumptions can omit jumps, stochastic volatility, local volatility, changing rates, borrow constraints, transaction costs, discrete hedging, market impact, or trading halts. A more complex model adds flexibility but also more parameters and failure modes.

### 2. Input and data risk

Spot, forward, rates, dividends, borrow, time, quotes, corporate actions, and contract metadata can be stale, mismatched, or measured with different conventions. Bid, Ask, Mid, Last, and settlement prices answer different questions. Time can use calendar days, trading days, or precise timestamps; rates and volatility can use different compounding and annualization.

### 3. Calibration risk

A calibration may fit liquid options well but extrapolate poorly to illiquid strikes and maturities. Weighting errors, crossed or stale quotes, overfitting, unstable parameters, and arbitrage-violating interpolation can create smooth-looking but unreliable surfaces. A low in-sample error does not prove realistic future dynamics.

### 4. Numerical and implementation risk

Trees, finite differences, and Monte Carlo methods have discretization, convergence, boundary, and random-sampling error. Code can use the wrong sign, unit, contract multiplier, dividend date, interpolation, or exercise condition. Vega quoted per one volatility point versus per 1.00 volatility unit differs by a factor of 100; Theta per day versus per year differs by the day-count scale.

### 5. Use and governance risk

A valuation model may be unsuitable for margin, P&L attribution, execution, or tail stress. Users can ignore limitations, combine Greeks from inconsistent systems, override values without records, or treat a theoretical Mid as liquid cash. Model inventory, ownership, independent validation, change control, documentation, and outcome monitoring are part of the control framework.

<a id="example"></a>

## One contract, two volatility assumptions

Consider an illustrative European Call with stock and strike both `$100`, six months to expiration, continuously compounded rate `4%`, dividend yield `1%`, and no transaction costs.

Using a constant-volatility Black-Scholes calculation:

- at `20%` volatility, model value is approximately `$6.34` per share;
- at `25%` volatility, model value is approximately `$7.73` per share;
- input change alone moves value by about `$1.39`, or `$139` for a standard 100-share contract.

Suppose the live quote is `$6.10 Bid / $6.50 Ask`. The 20% result lies inside the spread; the 25% result is far above it. Neither calculation proves an arbitrage. The market may imply a different volatility, the quote can reflect dividends and American exercise differently, or the model's constant-IV assumption may be inappropriate. Buying at Ask because one model says `$7.73` also assumes the position can later be hedged and exited under compatible dynamics and costs.

A sensitivity table should vary more than IV. Changing dividend yield, exact expiration timestamp, rate curve, or quote side can alter both price and Greeks. For an American equity option, compare a tree or finite-difference result and inspect early-exercise behavior. For a path-dependent product, compare an independently implemented Monte Carlo or another appropriate method.

## Validation tests that answer different questions

- **Unit and regression tests:** does code reproduce known identities and unchanged historical results?
- **Benchmarking:** does an independent implementation or simpler limiting case agree where it should?
- **Convergence tests:** do more tree steps, grid points, or simulation paths stabilize the output?
- **Sensitivity tests:** do signs, magnitudes, and nonlinear changes make economic sense?
- **Static-arbitrage checks:** does the fitted surface respect relevant monotonicity and convexity relationships?
- **Outcome analysis:** how do model marks, hedges, and forecasts compare with executable markets and realized P&L?

Agreement between two systems is weak evidence if both share the same data, library, assumptions, or code lineage.

<a id="risks"></a>

## Model review checklist

- Verify contract terms against authoritative specifications before valuation.
- Write the intended use and explicitly prohibited uses.
- Record model family, version, owner, code release, and validation status.
- Preserve source, timestamp, units, transformations, and quality flags for every input.
- Separate observed market data from inferred or manually overridden values.
- Compare Bid-, Mid-, and Ask-calibrated outputs where liquidity is weak.
- Check rates, dividends, borrow, calendars, time zones, and corporate actions.
- Validate price and Greeks independently; a correct price does not prove correct sensitivities.
- Test extreme spot, IV, time, rate, and dividend inputs without silent numerical failure.
- Examine calibration residuals by strike and maturity, not only one total error.
- Test interpolation and extrapolation outside liquid nodes.
- Compare at least one structurally different appropriate model.
- Add jump, skew-dynamic, liquidity, and discrete-hedging stresses outside the base model.
- Translate model differences into contract and portfolio dollars.
- Use executable quotes when assessing realizable value.
- Monitor overrides, exceptions, stale values, and unexplained P&L.
- Revalidate after material code, data, product, or market-regime changes.
- Size positions for model uncertainty rather than selecting the most favorable output.

<a id="misconceptions"></a>

## Common misconceptions

- “The formula is correct, so the price is correct.” Correct mathematics can represent the wrong market problem.
- “Calibration removes model risk.” It fits selected observations under a chosen structure.
- “More parameters always improve accuracy.” They can increase overfitting and instability.
- “Mid is objective truth.” It is a quote-derived reference, not necessarily executable.
- “Implied volatility is directly observed.” It is backed out through a model and input conventions.
- “Greeks are contract constants.” They depend on model, inputs, units, and timestamp.
- “Two matching systems prove correctness.” Shared assumptions or code can create common error.
- “Backtesting alone validates an option model.” Rare tails, changing regimes, and limited observations remain.
- “A conservative price is a conservative risk measure.” Price, hedge, margin, and stress uses can require different conservatism.
- “Complex models eliminate simplification.” They move simplification into new dynamics and parameters.
- “Model error is only a programmer problem.” Governance, data, interpretation, and use can dominate code error.

<a id="related"></a>

## Related topics

- [Black-Scholes model](/options/black-scholes-model/)
- [Binomial option pricing](/options/binomial-option-pricing/)
- [Monte Carlo option pricing](/options/monte-carlo-option-pricing/)
- [Implied volatility](/options/implied-volatility/)

<a id="sources"></a>

## Primary sources

- [Federal Reserve SR 11-7: Guidance on Model Risk Management](https://www.federalreserve.gov/supervisionreg/srletters/sr1107.htm)
- [OCC: Characteristics and Risks of Standardized Options](https://www.theocc.com/company-information/documents-and-archives/options-disclosure-document)
- [Black and Scholes (1973): The Pricing of Options and Corporate Liabilities](https://doi.org/10.1086/260062)
- [Merton (1973): Theory of Rational Option Pricing](https://doi.org/10.2307/3003143)