Volatility Arbitrage
The trade whose own name is the first thing it gets wrong.
Quick answer: Volatility arbitrage is buying an option believed cheap on volatility, or selling one believed expensive, and delta-hedging it so that direction is stripped out and only the difference between realised and implied volatility is left — a positive-expectancy bet with fat-tailed losses, not the riskless profit the word 'arbitrage' implies.
In simple words
Start with the honest disclaimer, because the name is doing dishonest work. A true arbitrage is a riskless profit locked in at zero cost, and this is not that. Volatility arbitrage means you think an option's implied volatility is wrong — too high or too low compared with how much you believe the underlying will actually move — so you buy it or sell it and then delta-hedge to remove the up-or-down bet. What is left is not a certainty. It is a bet that realised volatility will differ from implied volatility in the direction you predicted, and if you are wrong about that, you lose. Calling it arbitrage borrows the reassurance of a riskless trade for something that is nothing of the kind.
Here is the mechanism in one picture. You buy a 30-day NIFTY option because you think 12.8% implied volatility is too low and NIFTY will really swing more than that. On its own, that option is a directional bet — it needs NIFTY to move your way. So you delta-hedge it with futures, adjusting as NIFTY moves, which removes the direction. Now, if NIFTY does swing more than 12.8% annualised, your re-hedging buys low and sells high often enough to more than pay for the option; if it swings less, the option decays faster than the hedging earns and you lose. In theory the profit is the gap between realised and implied variance. In practice the hedge is discrete, costs money, leaks value, and is computed from a model that is wrong — so the clean theory and the messy outcome are two different things.
Realised versus implied: where the P&L comes from
You are paid the gap, if the path lets you
The accumulated hedging profit and loss of a delta-hedged long option as realised volatility varies against the implied volatility paid.
Professional explanation
Why 'arbitrage' is the wrong word, said early
A genuine arbitrage is a set of trades that costs nothing to put on and carries no possibility of loss — a violation of the law of one price that the market has momentarily left lying around. Volatility arbitrage is none of that. It costs money to put on, it can lose, and it depends on a forecast. The name survives because the trade removes the obvious risk, direction, and what remains feels like a pure relative-value play between two volatility numbers. But removing the visible risk is not the same as removing risk. What is left after the delta hedge is a bet that realised volatility will land on the side of implied volatility you predicted, executed through a hedge that is discrete and costly, priced by a model that is wrong. That is a position with positive expectancy if your edge is real, and a loss distribution with a long, fat left tail. Describing it with a word that means 'riskless profit' is the single most dangerous thing about the whole subject.
The profit-and-loss identity, term by term
For a delta-hedged option, the accumulated profit and loss over the life of the trade is, to a good approximation, the integral of one-half times gamma times spot squared times the difference between realised and implied variance: ∫ ½·Γ·S²·(σ²_realised − σ²_implied)·dt. Read it left to right. The bracket (σ²_realised − σ²_implied) is the edge you are betting on — positive if the market moves more than you paid for, negative if it moves less. The Γ·S² term is the weight: it says the edge only counts where gamma is large, and gamma is largest when spot is near the strike. The integral over dt says you collect the edge continuously through time, weighted by that gamma at every instant. The crucial and uncomfortable reading is that the sign of the bracket is not enough. You can be exactly right that realised volatility will exceed implied volatility and still lose, because the market may realise all that movement while spot is far from your strike, where Γ·S² is small, so your position was barely exposed to the very movement it correctly predicted.
Path dependence: right about volatility, wrong about where
This is the failure mode the P&L identity makes precise, and it deserves its own treatment because it surprises people who understand everything else. Gamma is not uniform across price — it is concentrated near the strike and decays away from it. So a long-gamma volatility position harvests the realised-versus-implied difference most effectively when the underlying oscillates around the strike, and barely at all when the underlying trends away from it. Two scenarios with identical realised volatility over the same window can therefore produce opposite results: one where the market chops around your strike and the position banks scalp after scalp, and one where the market makes the same total movement but as a single directional drift away from the strike, where your gamma had already collapsed. The volatility forecast was correct in both. Only the path differed. This is why volatility arbitrage is a relative-value trade with genuine risk rather than a coin that pays whenever your volatility view is right.
The hedge leaks, and the model is wrong
The clean P&L identity assumes continuous, costless hedging under the correct model. Peel away each assumption and value leaks out. Because the hedge is discrete, you capture only a noisy sample of the realised variance, and the sampling error is itself a source of profit and loss that has nothing to do with your edge. Because the hedge costs a spread, every re-hedge pays a toll, and a thin volatility edge can be entirely consumed by hedging costs if you rebalance too eagerly. Because the delta you hedge on comes from a model that assumes constant volatility and no jumps, it is the wrong delta, so the direction is not fully removed and a residual directional exposure contaminates the supposedly pure volatility bet. Add the three leaks together and the honest description of the result is not 'realised minus implied variance' but 'realised minus implied variance, minus a random hedging error, minus transaction costs, plus whatever the model's wrongness leaves behind'. The first term is the reason to do the trade; the rest is why it is not an arbitrage.
The same lesson at three scales: LTCM, 2008, 2020
Volatility-selling and relative-value volatility strategies share a characteristic profit-and-loss shape: long stretches of small, steady gains that look like skill, punctuated by rare, enormous losses when the market moves more than anyone priced and correlations that were assumed independent all move together. Long-Term Capital Management in 1998 held relative-value positions that were individually sensible and collectively catastrophic when liquidity vanished and every spread widened at once. The 2008 crisis did the same thing to a generation of short-volatility and correlation books. February 2018 and March 2020 did it again to short-volatility products directly, some of which lost most of their value in a session. These are not four different stories. They are one story — that the edge in selling or arbitraging volatility is an insurance premium, and insurance underwriters are paid steadily until the claim arrives, at which point they pay for years of premiums at once — told at four scales. The lesson does not get less true because it is famous.
Gamma scalping is how the volatility difference is harvested
A long-gamma position re-hedged at intervals: each re-hedge buys the underlying lower and sells it higher, banking small gains as the price oscillates.
Formula
The delta-hedged option P&L as a variance bet
P&L ≈ ∫ ½ · Γ · S² · (σ²_realised − σ²_implied) · dt
The profit and loss of a continuously delta-hedged option is, to a good approximation, the accumulated difference between realised and implied variance, weighted at every instant by the dollar gamma. The weighting is why the trade is path-dependent: the edge only counts where gamma is large, which is near the strike. A long option is long this expression; a short option is short it.
- P&LAccumulated profit and loss of the delta-hedged option over its life, per unit of the underlying.
- ΓGamma — the rate at which the option's delta changes with spot; largest near the strike and near expiry, and the weight that makes the trade path-dependent.
- SSpot price of the underlying; S² converts gamma into a dollar-gamma weight, so that a given percentage move counts more at higher prices.
- σ²_realisedThe variance the underlying actually realises over the interval — the realised volatility squared, annualised on 252 trading days.
- σ²_impliedThe variance implied by the volatility paid when the option was bought or received when it was sold — the implied volatility squared.
- dtThe increment of time over which the difference is accumulated; the integral sums the weighted edge across the life of the trade.
How a volatility-arbitrage trade is structured, conceptually
- Form a view on realised volatility — how much you believe the underlying will actually move over the option's life — independently of the option's price.
- Compare that forecast with the option's implied volatility; a trade exists only if the two differ by more than transaction costs and your own uncertainty.
- Buy the option if you believe implied volatility is too low, or sell it if you believe it is too high, sizing the position to the confidence in the forecast, not to the apparent gap.
- Delta-hedge the option in the futures to remove the directional exposure, so that the residual profit and loss depends on volatility rather than on where the underlying goes.
- Choose a re-hedging discipline that balances capturing realised variance against paying transaction costs, recognising that hedging too often can consume a thin edge.
- Monitor the position's gamma exposure and where spot sits relative to the strike, because the edge is only harvested where gamma is large.
- Accept before entering that the loss distribution is fat-tailed and path-dependent, and size so that a single adverse path cannot end the book — the risk that killed the famous cases was leverage, not the idea.
Practical example
NIFTY worked example
NIFTY is at 24,000 and the 30-day at-the-money option carries an implied volatility of 12.8%, pricing an annualised variance of 0.128² ≈ 0.0164. You forecast that NIFTY will actually realise closer to 16% over the month, a variance of 0.16² = 0.0256. If you are right, the bracket in the P&L identity, σ²_realised − σ²_implied, is about 0.0256 − 0.0164 = 0.0092 — positive, so a delta-hedged long option should accumulate a gain as your re-hedging scalps the extra movement. But notice what the trade actually requires. NIFTY must not only move 16% worth; it must do so while trading near 24,000, where the gamma of your option is largest. If NIFTY realises exactly 16% but does it by trending smoothly to 25,500 and sitting there, your gamma collapsed as spot left the strike, the scalps you needed never happened, and you can lose despite a correct volatility forecast. The number 0.0092 is the reason to consider the trade. The path is the reason it is not a certainty, and no amount of being right about volatility removes it.
BANKNIFTY worked example
BANKNIFTY teaches the cost-and-leak side of the same identity. Suppose BANKNIFTY at 52,000 shows a 30-day implied volatility of 15% and you believe it will realise 19%. The variance edge is 0.19² − 0.15² = 0.0361 − 0.0225 = 0.0136, larger than the NIFTY example, which is tempting. But BANKNIFTY moves further and faster, so harvesting that edge requires more frequent re-hedging, and its futures spread is wider than NIFTY's, so each hedge costs more. The gross edge of 0.0136 is real; the net edge is that number minus a hedging bill that grows with how often you trade and a sampling error that grows with how rarely you do. It is entirely possible to forecast BANKNIFTY's realised volatility correctly, capture the variance difference in theory, and hand most of it back through transaction costs — a different way to lose than the NIFTY path-dependence example, and just as instructive. The gross variance gap is never the trade; the gross gap minus the leaks is.
Lot sizes used above (NIFTY 75, BANKNIFTY 30) are those in force at the time of writing; NSE revises them periodically. Figures exclude brokerage, STT, exchange charges, stamp duty and GST. Examples are teaching scenarios built on round numbers — they are not historical quotes, not backtests and not trade calls.
Advantages & limitations
What it is good for
- It separates a volatility view from a directional view. By delta-hedging, the trade lets someone who has an opinion about how much the market will move express it without also having to be right about which way, which is a genuinely different and sometimes tractable question.
- Its edge has an economic source. The tendency of implied volatility to exceed subsequently realised volatility is a risk premium paid to those who underwrite market uncertainty, so the expected edge is not a market inefficiency that arbitrage should erase but a compensation for bearing a real, fat-tailed risk.
- The profit and loss is decomposable. The realised-minus-implied variance identity lets a desk attribute results to volatility, to hedging error and to transaction costs separately, so the strategy can be studied and improved rather than treated as a black box.
- It is expressible with liquid instruments. In India a NIFTY volatility view can be put on with exchange-traded options and hedged with the NIFTY future, both deep markets, so the concept is not confined to over-the-counter variance swaps unavailable to most participants.
- It disciplines the forecast. Because the trade only exists when the volatility forecast differs from implied volatility by more than costs, it forces the trader to quantify a view and compare it against the market's price rather than trading on a vague sense that options are cheap or dear.
Where it breaks down
- It is not arbitrage and can lose, because the residual after hedging is a forecast-dependent bet on realised versus implied volatility, not a locked-in price discrepancy.
- It is path-dependent. Being right that realised volatility will exceed implied volatility does not guarantee a gain, because the edge is only harvested where gamma is large, and the market may realise its movement far from the strike.
- The hedge leaks. Discrete rebalancing samples the realised variance noisily, transaction costs erode the edge on every adjustment, and a thin volatility difference can be entirely consumed by the cost of harvesting it.
- The delta is model-dependent. Hedging on a delta from a constant-volatility model leaves a residual directional exposure under skew, so the supposedly pure volatility bet is contaminated by a direction the trade meant to remove.
- Its loss distribution is fat-tailed and correlated across positions. The strategy produces long runs of small gains and rare large losses that arrive together, so a book that looks diversified in calm markets can be a single bet in a crisis.
- It has capacity constraints. The volatility risk premium is finite, so as more capital chases it the edge per unit of risk shrinks, and the same crowding that compresses the premium makes the eventual unwind more violent.
Common mistakes
- Taking the word 'arbitrage' literally and sizing the position as if it were riskless. The name describes the intent to trade a mispricing, not a guarantee, and leverage chosen on that misreading is what turns a manageable loss into a terminal one.
- Trading the gross variance gap and ignoring the leaks. A 0.01 variance edge is not the profit; the profit is that edge minus hedging error minus transaction costs, and a trader who banks the gross number is counting money the hedge will give back.
- Confusing being right about volatility with being paid for it. The P&L is weighted by gamma near the strike, so a correct forecast realised on a trending path away from the strike can still lose, and blaming the loss on bad luck misses that the path was always part of the bet.
- Re-hedging on the wrong delta. Using a flat-volatility delta in a skewed index leaves a directional residual, so the position is not the pure volatility trade the trader believes, and the leftover direction can dominate the volatility edge.
- Selling the volatility premium with too much leverage because the losing days are rare. The rarity is the trap: long quiet runs build false confidence and larger size, and the one clustered move then pays back years of premium at once.
- Assuming diversification across many volatility trades removes the tail. In a crisis the correlations the diversification relied on move toward one, so the independent-looking positions become a single concentrated loss precisely when it hurts most.
Professional usage
Dedicated volatility-arbitrage desks run this as their core business: they maintain a forecast of realised volatility for each underlying, compare it against the implied volatility surface, buy or sell the options they judge mispriced, and delta-hedge continuously so the residual profit and loss tracks the realised-versus-implied difference rather than direction. The research effort goes not into the trade structure, which is standard, but into the forecast and into the hedging discipline — how often to rebalance, on which delta, and how to size so that the fat left tail cannot end the fund. The discipline that separates the desks that survive from the ones that become case studies is almost entirely about leverage and about respecting the path-dependence and the tail, not about the elegance of the volatility view.
Variance-swap and volatility-swap markets exist precisely because the delta-hedging approximation is imperfect: a variance swap pays the realised variance directly, removing the path-dependence and the hedging error that contaminate the option-plus-hedge version, so an institution that wants clean exposure to the realised-minus-implied difference can buy it as a single instrument. Those products are largely over-the-counter and institutional, and they are not generally available to Indian retail participants, which is one reason the delta-hedged-option version — with all its leaks — remains the accessible form of the trade on the NIFTY and BANKNIFTY chains.
Key takeaways
- Volatility arbitrage means buying or selling an option judged mispriced on volatility and delta-hedging it, so that only the realised-versus-implied volatility difference remains — it is not a riskless arbitrage despite the name.
- The delta-hedged P&L is approximately ∫ ½·Γ·S²·(σ²_realised − σ²_implied)·dt: the bracket is the edge and the dollar gamma is the weight, which makes the trade path-dependent.
- You can be right that realised volatility will exceed implied volatility and still lose, because gamma is largest near the strike and the market may realise its movement elsewhere.
- The hedge is discrete, costly and computed from a wrong model, so the clean identity leaks into a noisier real result of edge minus hedging error minus transaction costs.
- The strategy's profit shape — steady small gains, rare enormous losses — is the same one that ended LTCM and repeated in 2008 and 2020; the danger is leverage against a fat left tail, not the idea itself.
- The edge is a volatility risk premium with finite capacity, and the crowding that compresses it makes the eventual unwind more violent.
The most useful thing to remember about volatility arbitrage is contained in its own name, read as a warning rather than a promise. The trade removes the risk you can see — direction — and leaves you with a bet that realised volatility will land where you predicted, harvested through a hedge that leaks and priced by a model that is wrong. Done with a real edge and honest leverage it has positive expectancy; done with the false confidence the word 'arbitrage' encourages it becomes the trade in every famous blow-up. The edge is an insurance premium, and insurers are paid steadily until the day they pay everything back. Nothing on this page is a recommendation to sell that insurance.
Frequently asked questions
What is volatility arbitrage?
Is volatility arbitrage really arbitrage?
How does a volatility-arbitrage trade make or lose money?
Why can I be right about volatility and still lose?
What is the role of delta hedging in volatility arbitrage?
What does the term ½·Γ·S² mean in the P&L?
What is gamma scalping?
Why does the hedge leak value?
What is the volatility risk premium and how does it relate?
Why do LTCM, 2008 and 2020 keep being cited together?
Can volatility arbitrage be done by retail traders in India?
What is a variance swap and why is it cleaner?
How is the position sized safely?
Does re-hedging more often improve a volatility-arbitrage trade?
What is the difference between a long and a short volatility-arbitrage position?
Why is the strategy described as positive-expectancy but dangerous?
Does volatility arbitrage remove all directional risk?
What forecast does a volatility-arbitrage trader actually need?
Is selling volatility a reliable income strategy?
What are the capacity constraints of volatility arbitrage?
How does skew affect a volatility-arbitrage trade?
Voice search & related questions
Natural-language questions people ask about volatility arbitrage.
Is volatility arbitrage actually risk-free?
How do you trade implied against realised volatility?
Why did LTCM blow up if their trades were arbitrage?
Can I make money selling options when volatility is high?
What does path dependence mean for a volatility trade?
Why do people delta-hedge a volatility bet instead of just buying the option?
Is gamma scalping the same as volatility arbitrage?
Sources & references
- Emanuel Derman & Iraj Kani — Riding on a Smile (1994)
- Roger Lowenstein — When Genius Failed: The Rise and Fall of Long-Term Capital Management
- Peter Carr & Dilip Madan — Towards a Theory of Volatility Trading
- NSE — Index options (NIFTY, BANKNIFTY) product information
Last reviewed 10 July 2026. Educational content only — not investment advice.