Trading Volatility Without an Opinion on Direction: the Variance Swap

Suppose you have a view. Not on where the stock market is going, you have no idea about that, but on how turbulent it is going to be. You think the next three months will be far choppier than the market expects. How do you bet on that?

The obvious answer is to buy options. Options gain value when volatility rises, so buy a straddle: a call and a put at the same strike.

The obvious answer is wrong, and the reason it is wrong is the whole motivation for this paper. A straddle gives you volatility exposure only while the market stays near the strike. Let the index rally 15 percent and your straddle is now deep in the money on the call side, behaving almost like a plain long position in the index. Your bet on volatility has quietly turned into a bet on direction, without you doing anything. Your volatility exposure, your vega, has bled away exactly when you did not want it to.

So the trader's honest complaint is: I want to bet on volatility and I cannot get a clean instrument. The 1999 Goldman Sachs research note by Kresimir Demeterfi, Emanuel Derman, Michael Kamal and Joseph Zou answered that complaint, and the answer turned out to be one of the most consequential results in modern derivatives.

The problem: every option's volatility exposure depends on the spot price

The technical version of the complaint: an option's vega is not constant. It is largest when the option is near the money and it collapses when the option moves far in or out of the money. So any single option, or any small collection of them, gives you a volatility exposure that changes as the market moves.

Could you build a portfolio of options whose total volatility exposure is the same regardless of where the index is? A portfolio that does not care about direction at all?

The key idea via analogy: stacking tents to make a flat roof

Picture each option's vega profile as a tent: a hump, peaked at the strike, tapering away on both sides.

If you buy one option, you have one tent. Your exposure is high in the middle and low at the edges.

Now buy options at many different strikes. You have many tents, side by side. If you weight them all equally, the tents in the middle of the range overlap and pile up, giving you a hump in the middle: still not flat.

The question is what weights make the tents stack into a perfectly flat roof. And the answer, which Demeterfi and colleagues derived and which is the paper's central result, is astonishingly specific:

Weight each option in inverse proportion to the square of its strike price.

Buy a strip of options across every strike, put and call, weighted by one over strike squared, and the resulting portfolio has a constant vega, no matter where the index goes. You have manufactured pure, undiluted volatility exposure.

That weighting scheme is not arbitrary and it is not a fit. It falls out of the mathematics of how a hedged option position accumulates profit and loss. And it means that low strikes get much heavier weights than high ones, which has a real consequence we will come back to.

What the portfolio actually pays you

Here is the second half of the result, and it is the part that makes the instrument tradeable.

If you buy that one-over-strike-squared strip of options and delta hedge it continuously, your accumulated profit turns out to be proportional to the difference between the realised variance of the index over the period and a fixed number known at the outset. In other words, this portfolio is a synthetic contract that pays you the difference between how volatile the market actually was and how volatile the options market said it would be.

That contract is a variance swap, and its fair strike, the number you are agreeing to at the start, is computable from today's option prices. No model required. No assumption about how volatility behaves. Just a weighted sum of observable option prices, which is what makes the result so powerful: it is a replication, not a valuation.

Two further points that matter enormously in practice.

It is variance, not volatility. The clean replication works for variance (volatility squared), not for volatility itself. Volatility swaps, which pay off in volatility rather than variance, cannot be replicated exactly, and the difference between the two (the convexity adjustment) is a genuine, tradeable risk that desks have to manage. The paper is careful and explicit about this, and it is the reason the market standardised on variance swaps.

Variance swaps are convex. Because they pay off in squared volatility, a buyer gains more from a volatility spike than they lose from an equivalent drop. That convexity is why variance swaps are prized as crisis hedges, and also why sellers of variance can be destroyed by a single event, as several funds discovered in February 2018 when a volatility spike wiped out short-volatility products in a matter of hours.

Why it mattered

• It created the volatility asset class. After this, volatility was not just a parameter inside a pricing model. It was a thing you could buy and sell directly, with a clean payoff and a model-free fair value. Volatility hedge funds, dispersion trading, and volatility arbitrage as a discipline all rest on this.
• It is the mathematical basis of the VIX. In 2003, the CBOE changed how it calculates the VIX index. The old method was based on Black-Scholes implied volatilities of a few near-the-money options. The new method is precisely the Demeterfi-Derman-Kamal-Zou replication: a weighted strip of out-of-the-money options across all strikes, weighted by one over strike squared. The most-watched fear gauge in the world is a direct implementation of this paper. That alone would justify its reputation.
• It made the variance risk premium measurable. Because the variance swap strike is the market's price of future variance, and realised variance is observable after the fact, you can measure the gap. It is persistently positive: implied variance is systematically higher than what actually gets realised. That gap is the payment investors make for volatility insurance, and it is the source of income for essentially every short-volatility strategy in existence.
• It is model-free, and that is rare. Almost everything in derivatives pricing depends on assuming a model. This replication does not. It holds whatever process the stock follows, provided the price path is continuous.

The honest limitations

• The replication needs options at every strike, and the market does not have them. The theory requires a continuum of strikes from zero to infinity, weighted by one over strike squared, which puts enormous weight on very low strikes, precisely where options are illiquid or simply not quoted. In practice you truncate the strip, and the truncation introduces an error that is small in calm markets and large in a crash. The VIX has exactly this issue.
• It assumes the price path is continuous, and jumps break it. This is the most important caveat, and it is deeply ironic. The replication argument requires the index to move smoothly. If it gaps, the delta hedge fails and the replication is wrong, in the direction that hurts the seller. So the variance swap, an instrument whose main appeal is crash protection, is precisely mis-replicated in a crash. Sellers of variance are short a risk their hedge does not cover.
• The convexity is a trap for sellers. A variance swap pays in squared volatility. If volatility triples, the buyer makes roughly nine times. Short variance positions have a payoff profile that looks wonderful for years and then loses everything at once. This is not a theoretical concern: it happened in 2008, and again in February 2018, when a single day's volatility spike destroyed several short-volatility funds.
• The strike is a price, not a forecast. The variance swap strike sits systematically above the variance that actually gets realised, because it includes a risk premium. Reading it as the market's honest expectation of future volatility overstates that expectation, usually substantially.
• Discrete sampling introduces error. Realised variance in a real contract is computed from daily closing prices, not from a continuous path, and the two differ.

The one-line takeaway

Demeterfi, Derman, Kamal and Zou showed that a strip of options weighted by one over the strike squared has constant volatility exposure regardless of where the market goes, and that delta hedging that strip synthesises a pure bet on realised variance, a model-free replication that created the volatility asset class, defined the modern VIX, and comes with a fatal asterisk: the whole argument assumes the market does not gap, which is exactly what it does when you need the hedge most.