When Volatility Refuses to Sit Still: Hull and White's Stochastic Volatility

There is a single number sitting at the centre of Black-Scholes, and it is the only input you cannot look up: volatility. The stock price is on a screen. The strike is in the contract. The interest rate is quoted. The time to expiry is on a calendar. But volatility, how much the stock is going to bounce around between now and then, is a guess about the future.

Black-Scholes asks you to make that guess once and then assumes you were right, and that the answer never changes for the entire life of the option.

Nobody has ever believed this. Volatility is visibly not constant. Markets have hectic weeks and dull weeks. Volatility clusters: turbulence begets turbulence, calm begets calm. Engle's ARCH work had already established this rigorously in the time-series literature by 1982.

John Hull and Alan White were among the first to ask what happens to option prices if you take that seriously and let volatility be its own random process, with a mind of its own.

The problem: a random input treated as a constant

If volatility is genuinely random, then when you price an option you face uncertainty on two levels. You do not know where the stock will go. And you do not know how wildly it will travel while getting there.

The naive response is: "so use the average expected volatility." Hull and White showed that this is wrong, and the way it is wrong is instructive.

An option's value is not a straight-line function of volatility. It is curved. For an at-the-money option the curvature is mild, but for options away from the money the value is convex in volatility: it rises faster than linearly. And there is a mathematical fact about curved functions, sometimes called Jensen's inequality, that says the average of a curved function is not the curved function of the average. If volatility is uncertain, the option is worth more than the option priced at the average volatility, because the good scenarios (high vol) help you more than the bad scenarios (low vol) hurt you.

Uncertainty about volatility is therefore itself worth money, and Black-Scholes does not price it.

The key idea via analogy: an engine with a wandering throttle

Picture the stock price as a car whose speed is set by a throttle. Black-Scholes assumes the throttle is welded in place at one setting for the whole journey.

Hull and White unwelded it. In their model, the throttle itself drifts around randomly, sometimes high, sometimes low, driven by its own source of noise. The car's position at the end of the trip is now uncertain for two reasons: the road wobbles, and the speed wobbles.

They then made an assumption that lets them solve the problem cleanly: they supposed that the volatility's randomness is independent of the stock's randomness, that the throttle wobbles for its own reasons, unconnected to which way the car is drifting.

Under that assumption a lovely simplification emerges. Since volatility does not care where the stock goes, you can price the option in two stages. First, imagine you knew the average volatility that was going to be realised over the option's life. Given that, the option is worth exactly its Black-Scholes value. Second, you do not know that average, so take the expected Black-Scholes value over the distribution of possible average volatilities.

So the answer is: the option price is the average Black-Scholes price, averaged across the possible volatility paths. A weighted blend of Black-Scholes prices. That is both computationally usable and conceptually clean.

What comes out of it

The results are exactly what a modern options trader would expect, which is remarkable given they were derived in 1987.

• The distribution of returns gets fat tails. Mixing together normal distributions with different widths produces something with a sharper peak and heavier tails than any single normal. This is a mathematical fact and it is the single best simple explanation of why financial returns look fat-tailed.
• Black-Scholes systematically misprices, and it does so in a shape. Hull and White found that Black-Scholes tends to overprice options in some regions and underprice in others, with the discrepancy growing with time to maturity. Because fat tails make far-from-the-money options relatively more valuable than a bell curve implies, converting the model's prices back into Black-Scholes implied volatilities produces a smile: higher implied volatility for strikes away from the money.
• You get a term structure. Because volatility mean-reverts, short-dated options reflect current volatility, while long-dated ones reflect something closer to the long-run average. That is the volatility term structure that every options desk trades.

Why it mattered

• It made volatility an asset with its own dynamics. Once volatility is a random process with a drift, a volatility of its own, and a mean-reversion speed, it becomes something you can model, forecast, hedge and trade. Vega, the sensitivity of an option to volatility, stops being a nuisance and becomes the central risk on a volatility desk.
• It is the direct ancestor of Heston. Steven Heston's 1993 model is Hull-White's idea with two changes: a specific choice of volatility process that admits a closed-form solution, and crucially, allowing the correlation between the stock and its volatility to be non-zero. Hull and White's independence assumption is exactly what Heston relaxed, and relaxing it is what generates the asymmetric equity skew rather than a symmetric smile. Hull-White gave the field the framework; Heston made it tradeable.
• It explained fat tails without needing jumps. Merton's route to fat tails was to add lightning strikes. Hull and White got there with nothing but a wandering throttle. Both stories are partly true, and the debate over which matters more is still live.
• It published in 1987, the year of the crash. The volatility smile appeared in the equity index market almost immediately afterwards and never left. Hull and White's framework was one of the few things sitting on the shelf ready to describe it.

The honest limitations

• The independence assumption is the model's weak point, and it is a big one. In equity markets, volatility and returns are strongly negatively correlated: when the market falls, volatility explodes. Hull and White's assumption of independence produces a symmetric smile, but the real equity smile is a lopsided skew, with much higher implied volatility for downside strikes. The model cannot generate that shape. This is not a detail, it is the dominant feature of the equity options surface.
• The market becomes incomplete, and the price is no longer unique. With two sources of randomness and only one hedging instrument (the stock), you cannot replicate the option. That means no-arbitrage does not pin down one price. To get a number, you have to make an assumption about how the market prices volatility risk, and Hull and White essentially assumed it carries no premium. Empirically, that is false: investors pay a persistent premium for volatility exposure (the variance risk premium), which is precisely why systematically selling options has historically been profitable.
• You cannot hedge vega with the stock. The whole practical apparatus of a modern volatility desk, hedging vega by trading other options, exists because of this model's central implication. But then you need a model to tell you which other options, and you are back to needing the model to be right.
• Calibration is genuinely hard. Volatility is unobservable. You are estimating the parameters of a process you cannot see, from prices that are noisy.
• Volatility jumps too. Even a random-but-continuous volatility process understates how fast implied volatility can double in a crisis.

The one-line takeaway

Hull and White let volatility wander randomly on its own and showed that the resulting return distribution has fat tails, that Black-Scholes therefore misprices options in a pattern that looks like a smile, and that uncertainty about volatility is itself worth money, laying the framework that Heston would complete six years later by allowing volatility to fall when the market does.