Dispersion and Correlation Trading: Implied Correlation, COR3M, and How the Trade Works

A dispersion trade sells volatility on a stock index and buys volatility on the stocks inside it, so it profits when the stocks move a lot but cancel each other out, and loses when they all move together. The number that decides which happens is correlation: how closely the index members move with each other. Index option prices contain a market estimate of that number, called implied correlation, and Cboe publishes it as an index.

In short: an index is a portfolio, so its volatility depends on its members’ volatilities and on how correlated they are. If you know the implied volatility of index options and of each member’s options, you can solve for the average correlation the market is pricing. That implied correlation has historically been higher than the correlation stocks actually realized, which is why index options look expensive relative to single-stock options. Dispersion traders try to collect that gap by selling index volatility and buying single-stock volatility. The trade is a short position in correlation, and correlation tends to spike exactly when markets crash.

This article assumes you are comfortable with implied volatility and with how straddles pay off. It describes how professional volatility desks think about the trade. It is not a recommendation, and the costs, data, and margin it needs put it out of reach of most individual accounts.

A common misconception: “index vol is just the average of stock vols”

On a typical day, at-the-money S&P 500 options imply far less volatility than the average S&P 500 stock’s options. A trader who sees an index at 20% and its members averaging 30% might think the index is “cheap”. It is not: the gap is diversification. When one stock falls and another rises, the moves partly cancel inside the index. The only way the index could be as volatile as its average member is if every stock moved in lockstep, a correlation of 1. So the right question is never “is index vol low?” but “what correlation does this index vol imply, and is that correlation likely to happen?”

The arithmetic of index volatility

Harry Markowitz’s 1952 paper “Portfolio Selection” (Journal of Finance) gives the variance of any portfolio. For an index with weights w and member volatilities σ:

Index variance = Σ w_i² σ_i² + 2 Σ Σ w_i w_j σ_i σ_j ρ_ij

The first sum is the variance the index would have if the members were uncorrelated. The second adds in every pair, scaled by its pairwise correlation ρ_ij. Replace every pairwise correlation with a single average ρ and solve:

ρ = (σ_index² - Σ w_i² σ_i²) / (2 Σ Σ w_i w_j σ_i σ_j)

Put implied volatilities into that formula and ρ becomes implied correlation: the single average correlation that makes the index options’ price consistent with the single-stock options’ prices. No option pricing model is involved, only the portfolio identity.

Worked example: implied correlation from five stocks

Take a hypothetical index of five stocks with these weights and 3-month at-the-money implied volatilities:

Stock	Weight	Implied vol
A	30%	32%
B	25%	28%
C	20%	36%
D	15%	24%
E	10%	40%

The weighted average member volatility is 31.4%. Suppose index options trade at 23% implied volatility. Working the formula (in decimals):

Index implied variance: 0.23² = 0.0529
Uncorrelated basket variance: Σ w² σ² = 0.022196
Sum of pairwise weighted products: 2 Σ Σ w_i w_j σ_i σ_j = 0.076400
Implied correlation: (0.0529 - 0.022196) / 0.0764 = 0.402, about 40%

The same identity tells you what the index “should” cost at other correlations, holding the members’ vols fixed:

Average correlation	Index volatility
0	14.9%
0.3	21.2%
0.402 (implied)	23.0%
0.6	26.1%
0.9	30.2%
1.0	31.4% (the weighted average)

A shortcut you will see on desks is ρ ≈ (σ_index / weighted average σ)². Here that gives (23 / 31.4)² = 0.54, well above the exact 0.40. The shortcut ignores the uncorrelated term Σ w² σ², which is large with five stocks and small with fifty, so it is only a rough guide for broad indexes.

How Cboe measures implied correlation

Cboe has published implied correlation since 2009. Its original methodology paper, “CBOE S&P 500 Implied Correlation Index” (2009), described daily values starting in July 2009 for indexes tied to specific January expirations, built from SPX options and the LEAPS of the 50 largest S&P 500 members. Each stock’s at-the-money volatility was interpolated from the put just below and the call just above the stock price, using the Barone-Adesi and Whaley American option model; SPX volatility used the Black model on the forward. The paper’s sample calculation for May 29, 2009 shows an SPX at-the-money implied volatility of 28.17% for December 2009 options.

The current family is described in Cboe’s “Cboe Implied Correlation Index (COR3M)” white paper (2021) and a July 18, 2022 Cboe press release. The main points:

Universe: the top 50 S&P 500 members by float-adjusted market capitalization, chosen from the previous business day’s close, with members ranked 51 to 55 held as a replacement pool in case a member leaves the index. Weights are each stock’s capitalization divided by the total of the 50, so they sum to 100% even though the 50 are only part of the S&P 500 (54.43% of it in the paper’s February 10, 2021 example).
Inputs: constant-maturity implied volatilities of 0.5-delta (at-the-money) options from Cboe Hanweck, for SPX and each of the 50 stocks.
Formula: exactly the average-correlation formula above. The white paper attributes it to Markowitz.
Tenors and skew: COR3M (three months) began publishing in July 2021. The 2022 release added COR1M, COR6M, COR9M, and COR1Y, all from 50-delta volatilities, plus skew versions (COR10D, COR30D, COR70D, COR90D) built from three-month volatilities at other deltas, which estimate correlation in large up or down moves.

The white paper’s worked example uses February 10, 2021, when the SPX 3-month at-the-money implied volatility was 20.23%. Cboe’s published COR3M closed at 33.54 that day.

DSPX: the same idea as a dispersion number

On September 27, 2023, Cboe and S&P Dow Jones Indices launched the Cboe S&P 500 Dispersion Index (DSPX). Their joint release described it as measuring the expected dispersion in the S&P 500 over the next 30 calendar days, calculated from S&P 500 index options and single-stock options with a modified VIX methodology. A companion index, the Cboe S&P 500 Constituent Volatility Index (VIXEQ), launched in November 2024 per Cboe’s October 17, 2024 release, measures the market-cap-weighted 30-day implied volatility of a basket of S&P 500 members and is a direct input to DSPX.

Neither release prints the formula, but the published data are consistent with a simple identity. In our check of Cboe’s daily history (3,116 days from June 2014 to November 9, 2026), DSPX matched √(VIXEQ² - VIX²) with a median absolute difference of 0.06 points. On November 9, 2026: VIXEQ 43.90, VIX 19.92, and √(43.90² - 19.92²) = 39.12, against a published DSPX of 39.09. In words, DSPX is the part of single-stock variance that does not show up in the index.

What implied correlation has done

Cboe’s daily COR3M history (our calculation from Cboe’s published file, as of November 9, 2026) shows how far the level moves. Values before July 2021 are Cboe backtests, not live publications.

Year	Average	Low	High
2008	49.2	34.7	75.2
2011	65.1	45.3	90.8
2017	27.8	16.2	42.0
2020	46.9	25.4	90.3
2022	44.9	25.9	53.8
2024	15.5	7.6	35.7
2025	19.3	11.1	46.7
2026 (to Nov 9)	13.6	7.2	34.5

Three patterns stand out. Correlation is highest in macro crises: the all-time high of 90.8 was December 8, 2011, during the euro-area debt crisis, and March 16, 2020 printed 90.3. It jumps in sell-offs: on April 8, 2025, during the tariff shock, COR3M was 46.7 with the VIX at 52.3. And it has been structurally low since 2023, when a handful of very large technology stocks drove the index and moved on their own news. On July 3, 2024, COR3M was 7.6, and the series made a new low of 7.2 on July 10, 2026. As of November 9, 2026, it stood at 14.8.

The evidence: correlation risk is priced

The core academic paper is Joost Driessen, Pascal Maenhout, and Grigory Vilkov, “The Price of Correlation Risk: Evidence from Equity Options,” Journal of Finance, 2009 (volume 64, pages 1377-1406). Using S&P 100 index options and options on all its members from 1996 to 2003, the working-paper version reports an average implied correlation of 46.7% against an average realized correlation of 28.7%. In their words, index puts were priced “as if correlations between stocks in the index are on average 63% higher than seems historically the case.”

Their other findings matter as much as the gap:

Single-stock options did not carry a variance risk premium in their sample; index options did. Implied variance systematically exceeded realized variance for the index but not for individual stocks. Their model attributes the entire index variance risk premium to the price of correlation risk.
The economic story is insurance. Index options are expensive, in the authors’ account, because they hedge correlation increases: the moments when diversification disappears are the moments investors most want protection.
It is hard to harvest. The published abstract says a strategy exploiting priced correlation risk earns a high alpha without frictions, but “cannot be exploited with realistic trading frictions,” which the authors read as a limits-to-arbitrage explanation.

Cboe’s own 2021 research note on implied correlation makes a similar descriptive point: a smoothed 63-day realized correlation of the top 50 stocks was “in most cases slightly below” COR3M. The note also documents a quarterly seasonal pattern in 1-month implied correlation around earnings seasons, when single-stock implied volatilities rise.

How a dispersion trade is built

A long dispersion trade (sometimes called short correlation) has two legs:

Sell index volatility: short at-the-money index straddles, or a short index variance swap.
Buy single-stock volatility: long straddles or variance swaps on the members, usually the largest and most liquid ones.

Cboe’s 2009 paper describes the classic version as selling at-the-money index straddles and buying at-the-money straddles on the components, on the logic that when implied correlation is high, index premiums are rich relative to single-stock premiums. Short dispersion is the reverse and is a way to buy crash-time correlation.

Both legs are delta hedged, so what is left is exposure to volatility and correlation. The mechanics of that hedging, and why a hedged option’s profit depends on realized versus implied volatility, are covered in delta hedging and gamma scalping.

Straddles or variance swaps

Straddles are listed and cheap to trade, but their profit is path dependent: a delta-hedged straddle earns gamma only near its strike, so a stock that moves far away and stays there stops paying. Variance swaps, which pay realized variance minus a fixed strike, remove most of that path dependence and are the cleaner instrument for the arithmetic. Their replication and the reason variance is the natural unit are in variance swaps and volatility desks. The worked example below uses variance swaps for that reason.

Sizing: what “neutral” means

A desk has to choose how much of each leg to hold. Common choices:

Vega neutral: total single-stock vega equals index vega, split by index weight. Simple, but it leaves exposure to the general level of volatility.
Theta neutral: size so the daily time decay of the two legs offsets. Because the member options are pricier, this means less single-stock vega than index vega.
Correlation weighted: size the single-stock leg so a uniform change in member volatilities is roughly offset by the index’s response at the current correlation, leaving a purer correlation bet.

No choice removes every exposure. The example shows why.

Worked example: a vega-neutral dispersion trade

Use the five-stock index above. The desk sells a 3-month index variance swap struck at 23 with $10,000 vega notional (it gains or loses about $10,000 per volatility point for small moves), and buys member variance swaps struck at each stock’s implied volatility, with vega notionals of $10,000 split by weight ($3,000, $2,500, $2,000, $1,500, $1,000). A variance swap’s variance notional is vega notional divided by twice the strike, so the index leg is $10,000 / (2 x 23) = $217.39 per variance point, and its payoff is -217.39 x (realized vol² - 23²).

Case 1: member vols realize exactly at implied; only correlation differs.

Realized correlation	Realized index vol	Single-stock leg	Index leg	Net P&L
0.10	17.27%	$0	+$50,139	+$50,139
0.20	19.36%	$0	+$33,530	+$33,530
0.30	21.24%	$0	+$16,922	+$16,922
0.402 (implied)	23.00%	$0	$0	$0
0.50	24.58%	$0	-$16,296	-$16,296
0.60	26.08%	$0	-$32,904	-$32,904
0.80	28.86%	$0	-$66,122	-$66,122

This is the pure correlation bet: positive if realized correlation comes in below the 40% implied, negative above it.

Case 2: volatility levels move too.

Scenario	Realized index vol	Single-stock leg	Index leg	Net P&L
Member vols +5 points, correlation at implied	26.66%	+$54,076	-$39,463	+$14,613
Member vols -5 points, correlation 0.20	16.29%	-$45,924	+$57,309	+$11,385
Member vols +20 points, correlation 0.20	31.66%	+$265,218	-$102,887	+$162,331
Member vols +20 points, correlation 0.80 (crash)	47.24%	+$265,218	-$370,226	-$105,008

Two lessons. First, “vega neutral” was not volatility neutral: with correlation unchanged, a rise in member vols made money, because variance payoffs are convex and the index leg’s variance notional is not matched to the stocks’. Second, the losing scenario is the one that matters: a crash where every stock’s volatility jumps and they all fall together. The short index leg then loses more than the long single-stock leg makes. That is the correlation risk Driessen, Maenhout, and Vilkov argue investors pay to avoid, and it is what a long dispersion book is paid to carry.

What practitioners worry about

Which stocks to buy. Buying options on all 500 members is impractical. Desks hold a subset, typically the largest names, which creates basis risk: the index can move on stocks the desk does not own.
Costs on the long leg. Single-stock options have wider bid-ask spreads than index options, and the long leg has many of them. That is the friction Driessen, Maenhout, and Vilkov found large enough to absorb the premium.
Earnings. Single-stock implied volatility rises into earnings, which Cboe’s note shows lifting 1-month implied correlation seasonally. A dispersion book that buys stock volatility just before earnings pays for the event; one that sells it pays out if the move is large. The event arithmetic is in earnings and the implied move.
Skew. Index skew is much steeper than single-stock skew. The 2022 Cboe skew indexes exist because correlation in a large down move is priced higher than at-the-money. A book that is correlation neutral at the money can be short correlation in the tails. See volatility surface and skew.
Exercise style. SPX options are European, but US single-stock options are American, so the long leg carries early exercise and dividend considerations that index options do not.
Concentration. When a few mega-cap stocks dominate the index, the “average correlation” is driven by how those few interact. Low implied correlation in 2024 to 2026 partly reflects that structure, not only diversification.

What can go wrong

Correlation goes to one in a crisis. COR3M’s highest readings (90.8 in December 2011, 90.3 in March 2020) and its jump in April 2025 are precisely the conditions in which long dispersion loses. Short correlation is, in effect, short a crash.
Volatility on the index leg is not capped. A short index variance swap pays the square of realized volatility, so its losses accelerate in a crash. Straddle versions are short index gamma and can need large delta hedges at bad prices.
Margin when it hurts. Short index options and variance swaps draw large margin calls in exactly the scenario where the strategy loses, and the long single-stock leg’s gains may sit with different counterparties.
The premium can be gone. Implied correlation near its lows (7.2 in July 2026) leaves little room for realized correlation to come in lower, and a lot of room for it to rise.
Model and data risk. Implied correlation depends on which volatilities are used (at-the-money, delta, tenor) and on the weights. Two desks can compute different numbers from the same market.

The short version

Index volatility is member volatility plus correlation. Solving the portfolio-variance identity for the average correlation turns index and single-stock option prices into an implied correlation, which Cboe publishes as COR3M and its relatives, and which DSPX restates as expected dispersion. Academic evidence, led by Driessen, Maenhout, and Vilkov (2009), shows implied correlation has historically exceeded realized, which is why index options carry a volatility premium that single-stock options mostly do not. Dispersion trades try to earn that gap by selling index volatility and buying stock volatility, but they are short correlation, and correlation rises fastest in the crashes that make such positions most expensive to hold.