Stochastic Volatility Model: How Options Smile Gets Priced

Key Takeaways

• A stochastic volatility model adds a second correlated SDE for variance, giving a two-factor framework Black-Scholes lacks.
• In the Heston model, a negative rho near -0.7 for equity indices directly generates the observed put skew.
• Calibrating an SV model to only one maturity produces parameters that break when extrapolated to other tenors.
• Options desks use SV parameters to price and hedge across strikes and maturities that a flat-vol model cannot fit.

What It Is

A stochastic volatility (SV) model is a continuous-time framework in which the asset price and its instantaneous variance each follow a stochastic differential equation, coupled through a correlation parameter. The most cited specification is the Heston 1993 model, which gives a closed-form characteristic function for European option prices under stochastic variance.

The difference from GARCH is architectural. GARCH is discrete-time and has one shock per period that drives both returns and variance. Stochastic volatility models have two separate shocks, usually correlated, so returns and variance can surprise independently.

The Intuition

Black-Scholes assumes a single constant volatility. Implied volatility surfaces in real markets are not flat: out-of-the-money puts trade richer than out-of-the-money calls on equity indices, and long-dated options price differently than short-dated ones. That structure, called the volatility smile or skew, is evidence that the market treats volatility as uncertain and asymmetric.

Stochastic volatility models encode that uncertainty directly. Volatility can drift, spike, and mean-revert on its own timetable. Option prices derived from these models line up much better with traded quotes across strikes and maturities.

How It Works

The Heston model specifies two coupled SDEs, one for the stock price S_t and one for the instantaneous variance v_t:

dS_t = mu * S_t * dt + sqrt(v_t) * S_t * dW1_t
dv_t = kappa * (theta - v_t) * dt + xi * sqrt(v_t) * dW2_t
corr(dW1_t, dW2_t) = rho

The variance follows a CIR (Cox-Ingersoll-Ross) process:

• kappa is the speed of mean reversion of variance.
• theta is the long-run variance level.
• xi is the volatility of volatility.
• rho is the correlation between the price shock and the variance shock.

A negative rho, typically around -0.7 for equity indices, captures the leverage effect. The Feller condition 2 * kappa * theta >= xi^2 keeps the variance process strictly positive.

Other members of the family

• SABR (Hagan, Kumar, Lesniewski, Woodward 2002) is widely used for interest-rate options. It models forward rates with a stochastic volatility that is itself driven by a geometric Brownian motion. SABR produces an analytic approximation for the implied volatility smile rather than a price directly.
• 3/2 model inverts the variance dynamics and gives different smile behavior in the wings.
• Jump-diffusion extensions (Bates, SVJ) add Poisson jumps to the price process on top of stochastic variance to match very short-dated wings.

Worked Example

Consider calibrating Heston to a liquid index option surface. A typical calibration on SPX might give: v_0 = 0.04 (initial variance, 20 percent vol), theta = 0.04, kappa = 2.0, xi = 0.4, rho = -0.7.

Under this setup, instantaneous volatility starts at 20 percent. The half-life of variance shocks is ln(2) / kappa, about 0.35 years or roughly 4 months. A negative rho means that when the market drops, the variance process receives a positive shock, so implied puts trade richer than calls: the classic equity skew.

Pricing a 1-year at-the-money call under these parameters gives a slightly lower price than Black-Scholes with a flat 20 percent vol, because the smile adds value in the wings and removes a little at the money. The important test is not a single strike but the full surface fit against market quotes.

Common Mistakes

1. Treating SV and GARCH as interchangeable. They model the same phenomenon from different angles. GARCH fits daily return time series and forecasts realized variance. Stochastic volatility fits option prices and produces a risk-neutral surface. Using GARCH parameters to price options or SV parameters to forecast realized vol crosses the measures incorrectly.

2. Ignoring the correlation parameter. Setting rho = 0 is tempting for simplicity but eliminates the skew. For equity index work, rho must be negative and material. For FX and rates, the sign can differ.

3. Calibrating to a single maturity. A good calibration fits the surface across strikes and maturities simultaneously. Fitting only one maturity can produce parameters that explode when extrapolated.

4. Over-trusting closed-form solutions. Heston has a "closed form" in the sense that it gives an integral representation via characteristic functions. The integral still has to be evaluated numerically, and naive implementations suffer from branch cuts in the complex logarithm. Use a tested pricer.

Frequently Asked Questions

Q: What is a stochastic volatility model in simple terms? It is a pricing framework where volatility is not a fixed number but a separate random variable with its own dynamics, so the model can generate the range of implied volatilities observed across option strikes and maturities.

Q: How does a stochastic volatility model affect investment decisions? Options desks use it to price and hedge across the full volatility surface, buying or selling variance swaps and forward volatility agreements based on where the model says current vol is cheap or expensive relative to calibrated parameters.

Q: What is a real-world example of a stochastic volatility model? When calibrating the Heston model to SPX options, practitioners routinely fit rho near minus 0.7, which directly reproduces the steep put skew that makes downside options more expensive than equivalent upside calls.

Q: How can investors use stochastic volatility models? Long-volatility investors use Heston or SABR calibrations to identify when the market's variance of variance is low, meaning cheap vol-of-vol, and buy variance swaps or long-gamma positions before the next spike.

Q: How is a stochastic volatility model different from GARCH? GARCH is a discrete-time model fit to historical returns and forecasts realized variance. Stochastic volatility models are continuous-time and fit to option prices under the risk-neutral measure, so they produce implied vol surfaces, not realized vol forecasts.

Sources

1. Heston, S.L. (1993). "A Closed-Form Solution for Options with Stochastic Volatility with Applications to Bond and Currency Options." Review of Financial Studies 6(2), 327-343. https://www.ma.imperial.ac.uk/~ajacquie/IC_Num_Methods/IC_Num_Methods_Docs/Literature/Heston.pdf
2. Hagan, P., Kumar, D., Lesniewski, A., Woodward, D. (2002). "Managing Smile Risk." Wilmott Magazine. http://www.frouah.com/finance%20notes/The%20SABR%20Model.pdf
3. Jacobson, H. "A Brief History of Volatility Models." https://volquant.medium.com/a-brief-history-of-volatility-models-cc0bbefe8b90

Disclaimer

This article is educational content only and is not financial advice. Nothing here is a recommendation to buy, sell, or hold any security. Consult a licensed advisor before making investment decisions.