The Day Correlation Went to One
On 5 February 2018, the VIX index rose 116% in a single session, its largest one-day percentage jump on record, and the S&P 500 fell 4.1%. Short volatility products imploded, most famously the XIV exchange-traded note, which lost more than 90% of its value after the close and was terminated by Credit Suisse days later. Less publicised was what happened inside correlation books: as the index fell, its constituents stopped behaving like 500 different companies and started behaving like one. Stocks that normally offset each other fell together, and strategies built on the difference between index volatility and single-stock volatility took losses that had been accumulating premium for years.
That difference is the subject of this article. Dispersion trading is the business of trading index volatility against the volatility of the index's constituents, and it exists because of a persistent quirk: index options, on average, trade expensive relative to the options on the stocks inside the index.
The quirk has a name, the correlation risk premium, and a logic. What links the volatility of an index to the volatilities of its members is correlation, and investors will pay up for protection against the state of the world where correlation spikes, because that state is a crash. Dispersion traders are the ones selling that insurance.
The Core Idea
An index is a weighted basket, so index variance is built from two ingredients: how much the members move, and how much they move together. In simplified form, with average constituent volatility sigma_s and average pairwise correlation rho:
sigma_index^2 = rho * sigma_s^2 (approximately)
Hold single-stock volatilities fixed and crank correlation up, and index volatility rises; crank it down and the members' moves cancel in the basket and index volatility falls. Correlation is the bridge between the two volatility markets.
Now compare what the two options markets imply. Take the implied volatility of index options and the implied volatilities of the constituents, and back out the correlation that would make them consistent. That number, implied correlation, has historically sat well above the correlation subsequently realised. Studies of S&P 500 options across the 1996 to 2020 era typically find implied correlation exceeding realised by roughly 10 percentage points on average, and CBOE publishes tradeable benchmarks of it (the COR3M index) precisely because the gap is a market in itself.
Rearranged as a trade: index implied volatility is rich relative to constituent implied volatility. Sell the rich one, buy the cheap one, and you are short implied correlation, collecting the premium in normal times and paying out when correlation spikes.
Implied Correlation, Made Concrete
A worked number makes the definition stick. Suppose one-month index implied volatility is 18% and the weighted average implied volatility of the constituents is 28%. Using the simplified identity above:
rho_implied = sigma_index^2 / sigma_s^2 = (0.18)^2 / (0.28)^2 = 0.0324 / 0.0784 = 0.41
The options market is pricing the average stock pair to move with correlation 0.41 over the next month. If stocks typically realise pairwise correlation nearer 0.30, index options are overpriced relative to single-name options, and the dispersion seller expects to collect the difference. (The full formula weights every pair by index weights and both volatilities; the ratio version above drops the small sum-of-squared-weights correction but preserves the intuition and gets within a point or two for broad indices.)
The same arithmetic in code, for a small basket:
import numpy as np weights = np.array([0.30, 0.25, 0.25, 0.20]) single_vols = np.array([0.32, 0.27, 0.25, 0.30]) # constituent implieds index_vol = 0.18 # index implied basket_var = np.sum((weights * single_vols) ** 2) cross_terms = (np.sum(weights * single_vols) ** 2) - basket_var implied_corr = (index_vol**2 - basket_var) / cross_terms print(f"Implied correlation: {implied_corr:.2f}")
When this number is high relative to your forecast of realised correlation, the dispersion trade is attractive. When it is low, there is nothing to sell.
Building the Trade
The classic structure has two legs, always volatility-hedged so the position is about correlation rather than direction.
Short index volatility. Sell index options, traditionally at-the-money straddles that are delta hedged, or, more cleanly, sell an index variance swap. This leg collects the rich index implied.
Long single-name volatility. Buy options (or variance swaps) on a representative basket of constituents, weighted to match the index. This leg pays away the cheaper single-name implied and protects the book against volatility rising everywhere.
Because you are short one volatility and long a basket of others, the net exposure to the general level of volatility can be sized close to zero (vega-neutral dispersion) or deliberately left short (the more aggressive version common when the premium is wide). What cannot be hedged away is the exposure to correlation itself: if stocks move a lot individually while the index stays quiet, the long single-name leg pays off while the short index leg expires cheaply, and the trade wins. If everything moves together, index volatility realises high while diversification within the single-name book does nothing extra, and the trade loses. Short correlation is the residual, and it is the point.
Running the single-name leg is operationally heavy: dozens or hundreds of option positions, each with its own volatility surface, borrow, earnings dates and liquidity. This is why dispersion in size lives at dedicated desks rather than in retail accounts.
Variance Swaps
The cleanest instrument for both legs is the variance swap: an OTC contract that pays the difference between realised variance over the life of the trade and a strike agreed upfront, with no options to manage and no delta hedging required by the holder.
Payoff = variance notional * (realised variance - strike variance)
Quotes come in volatility points, but the contract settles on variance, volatility squared, so conventions matter. Dealers quote a vega notional (pounds of P&L per volatility point at the strike) and convert it to variance notional as vega notional / (2 * strike).
A worked P&L. You sell an index variance swap struck at 20 volatility points (400 variance points) with £100,000 vega notional. Variance notional = 100,000 / (2 * 20) = £2,500.
Scenario one: markets stay calm and realised volatility comes in at 15 (225 variance points). Payoff to you as the seller = 2,500 * (400 - 225) = +£437,500.
Scenario two: a shock arrives and realised volatility prints 30 (900 variance points). Payoff = 2,500 * (400 - 900) = -£1,250,000.
The asymmetry is not an accident of the numbers chosen. Because the contract settles on the square of volatility, losses to the seller grow quadratically as realised volatility rises: being wrong by 10 points on the upside costs nearly three times what being wrong by 5 points does. Sellers of variance are short convexity in volatility itself, which is exactly why variance swap strikes trade above at-the-money implied volatility, and why post-2008 (and again post-2018) many dealers cap single-name variance swap payoffs at 2.5 times the strike.
Who Trades This
Dispersion started as a dealer inventory trade. Structured products desks in the 1990s and 2000s sold retail products embedding single-stock options, leaving banks structurally long single-name volatility and short index volatility hedges; managing that book was proto-dispersion. Today the participants are the volatility trading desks at the major banks, dedicated volatility hedge funds and the vol arms of multi-strategy platforms, plus the options market makers who end up warehousing pieces of everyone else's flow.
It remains an institutional market. The instruments (variance swaps, correlation swaps) are OTC, the single-name leg needs scale, and the risk management burden, hundreds of positions whose collective exposure is one number, correlation, requires serious infrastructure. The retail-accessible echo of the trade is simply understanding that selling volatility premium on indices earns a spread partly composed of this correlation premium, with the same crash-shaped bill attached.
When It Blows Up
The correlation seller's nightmare has a well-documented shape: in a crash, correlation goes to one.
Diversification is a fair-weather property. In February 2018, in October 2008, in March 2020, the average pairwise correlation of S&P 500 stocks that normally sits somewhere around 0.3 lurched toward 0.7 and beyond, and during the worst sessions effectively everything fell together. Each time, the dispersion P&L profile did what short-premium profiles do: years of steady collection, then a payout concentrated into days. February 2018 was especially instructive because the equity move was modest by crisis standards, around a 10% index correction, yet the volatility and correlation complex moved as if it were 2008. The trigger was positioning, the crowded short volatility trade unwinding into itself, not economics.
The structural reason the premium exists is the same reason the loss is so bad: a correlation spike hits when everything else in a portfolio is also losing. The dispersion seller earns a premium precisely for delivering money in the states of the world where money is most valuable to everyone else. Collecting that premium is rational; mistaking it for riskless yield is how volatility desks end careers.
Risks, Sizing and the Honest Caveats
Beyond the headline crash risk, four practical points separate surviving dispersion traders from former ones.
Sizing to the tail, not the average. The P&L distribution is short-premium shaped: limited steady gains, unbounded episodic losses that arrive when liquidity is worst. Position sizes that look conservative against average monthly volatility are reckless against a correlation-to-one scenario, and the variance swap arithmetic above shows losses accelerating quadratically exactly then. Stress the book at realised volatility doubling and correlation at 0.9, and size to survive that, because roughly once a decade that is the realised outcome.
The wings and the caps. Short index variance is short the far out-of-the-money puts embedded in the variance replication, the most crash-sensitive instruments in existence. Capped variance swaps and put-spread-collared structures give away some premium to bound the tail, and after 2018 most sensible programmes accept that cost.
Implementation drag. The single-name leg crosses wider spreads, pays borrow, and suffers around earnings and corporate actions. Realistic frictions consume a meaningful slice of the theoretical correlation premium, and academic estimates of the premium shrink notably once bid-ask costs are applied.
The premium is not constant. Implied correlation itself cycles. Selling dispersion when implied correlation is already at the bottom of its range earns little and risks the same tail; the trade is a relative-value decision, not a permanent income machine. Disciplined desks trade it when the spread to a defensible realised-correlation forecast is wide, and stand flat otherwise.
Dispersion is one of the purest examples of what professional derivatives trading actually is: identifying a risk the market persistently overpays to shed, warehousing it with careful sizing, and respecting the fact that the overpayment exists because sometimes the risk shows up.
Correlation is mean reverting. Careers, unfortunately, are path dependent.
Skip the £25k programme - try the alternative
Master's programmes are slow and expensive. Quantt is a self-paced alternative. Start free with a real lesson and interview practice, then unlock 50+ courses and your personalised plan.
Free to start · No credit card required