An option pricing model is a mathematical method for estimating the fair, or theoretical, value of an option. It takes a small set of inputs describing the contract and the market and returns a price, together with measures of how that price responds to changes in each input. Traders, market makers, risk managers and accountants use such models to quote options, to hedge them and to value positions that do not trade often.
Modern models rest on an argument published in 1973 by Fischer Black, Myron Scholes and Robert Merton. An option can be replicated by a continuously adjusted holding of the underlying asset and cash, so its value must equal the cost of that replicating portfolio, or arbitrage would be possible. The resulting Black-Scholes model remains the reference point, and most later models relax one or more of its assumptions.
Inputs
Nearly every model for an option on a stock uses the same six inputs.
- Price of the underlying. The current market price of the asset.
- Strike price. The strike price fixed in the contract.
- Time to expiry. Expressed as a fraction of a year.
- Volatility. The expected variability of the underlying's returns over the option's life. It is the only input that cannot be observed and must be estimated or inferred.
- Interest rate. The risk-free rate for the period.
- Dividends. Any dividend or other income expected on the underlying before expiry.
The type of option, call or put, and its exercise style also matter. The expected return of the underlying does not appear, a consequence of the replication argument.
The Black-Scholes formula as an example
For a European call on a stock that pays no dividend, the model gives:
with
- is the value of the call, the price of the underlying and the strike.
- is the annual risk-free rate, the time to expiry in years and the annual volatility.
- is the cumulative standard normal distribution, and .
Take , , , and . Then and , with and . The discounted strike is . The call is worth .
Main families of models
- Closed-form models. Formulas in the Black-Scholes tradition give an answer directly. Variants exist for options on futures, on currencies and on assets paying a continuous yield.
- Lattice models. The binomial option pricing model of Cox, Ross and Rubinstein (1979) lets the price move up or down in small steps and works backward through the resulting tree. Because it can test at each step whether exercise is worthwhile, it is the standard method for an American option.
- Simulation. Monte Carlo simulation generates many random price paths and averages the discounted payoffs. It suits an exotic option whose payoff depends on the path taken.
- Finite-difference methods. These solve the pricing equation numerically on a grid of prices and times.
- Extended dynamics. Local-volatility, stochastic-volatility and jump models replace the assumption of constant volatility with a richer stochastic process, in order to match the prices seen across strikes and expiries.
How models are used
In liquid markets option prices are set by supply and demand, and the model is used in reverse. Given the market price, it is solved for the volatility that reproduces it, the implied volatility, which has become the usual way of quoting and comparing options. Models also produce the option Greeks, which tell a dealer how much of the underlying to hold for delta hedging and how exposed a book is to time and volatility. For contracts with no market price, such as employee stock options or bespoke derivatives, the model value is the basis of accounting and collateral.
Limitations
Every model simplifies. The Black-Scholes assumptions of constant volatility, continuous trading without costs and a lognormal distribution of prices do not hold in practice. Markets gap, volatility changes and extreme moves are more frequent than the model implies. The evidence is the volatility smile: options on the same underlying trade at different implied volatilities depending on strike, which would not occur if the model were exact.
A model's output is only as good as its volatility input, and two reasonable estimates can give materially different values. Firms therefore treat model choice and calibration as a source of risk in itself, subject to model risk management and independent validation.