Dispersion Trading

You already know how to sell variance and harvest the volatility risk premium — collect a premium for insuring against turbulence, and pay out when turbulence arrives. You also know, from portfolio theory, that a basket of imperfectly correlated assets wobbles less than its average ingredient. Dispersion trading is what happens when you bolt those two ideas together and trade the seam between them.

The pitch, in one breath: an index’s implied volatility is almost always lower than the average implied volatility of the stocks inside it, because those stocks don’t all move in lockstep. The size of that gap is a price — the market’s bet on how correlated everything will be. Dispersion traders sell the index’s volatility and buy the constituents’ volatility, which nets out to a clean wager on correlation itself. Most of the time it pays. In a crash it can take your face off. By the end of this lesson you’ll know exactly why both halves of that sentence are true.

Index vol is less than the sum of its parts

Analogy. Picture a choir of fifty slightly out-of-sync singers. If all fifty hit the exact same note at the exact same instant — perfect correlation — the room is deafening, the full sum of every voice. But real singers drift: someone’s a hair early, someone’s flat, someone takes a breath. Those imperfections partly cancel, and the choir as a whole is noticeably quieter than one soloist’s volume multiplied by fifty. An equity index is that choir, and “volume” is volatility. The index can only be as loud as its singers are synchronized.

This is the diversification result you already met in portfolio theory, wearing volatility-trading clothes. Portfolio variance isn’t the average of the parts’ variances — it depends on the correlations between them. The less correlated the constituents, the more their idiosyncratic moves cancel, and the calmer the basket. Push every pairwise correlation to 1 and the cancellation vanishes: the index becomes exactly as volatile as its average member. Push correlations down and the index goes quiet.

Carry that straight into the options market. The same forces that damp realized index moves also compress what the market is willing to pay for index volatility. So:

Index implied volatility < weighted-average single-name implied volatility.

The two are only equal in the extreme, never-quite-real case where every stock is perfectly correlated with every other. In practice the index trades at a discount to its parts, and that discount is the raw material of the whole trade.

The hero visual below makes it tangible. Five constituent stocks each carry their own single-name implied vol (the bars); the dashed line is their weighted average; the lone shorter bar is the index’s implied vol, always sitting below the average. Watch the gap — and the implied correlation the island computes from it — and keep the crisis toggle in mind, because we’ll come back to it.

The math: index variance, correlation, and implied correlation

Time to make “the gap is a price” precise. Start from the full variance identity for an index built from weights $w_i$ on constituents with volatilities $\sigma_i$ and pairwise correlations $\rho_{ij}$:

$\sigma_I^2 = \sum_i w_i^2\sigma_i^2 + \sum_{i\neq j} w_i w_j \sigma_i \sigma_j \rho_{ij}$

The first sum is each stock’s own variance contribution; the second is every pair’s covariance contribution, and that second sum is where correlation lives. Crank every $\rho_{ij}$ to 1 and the whole thing collapses to $\left(\sum_i w_i \sigma_i\right)^2$ — index vol equals the weighted-average single-name vol, the no-diversification ceiling. Drop the correlations below 1 and the index variance falls beneath that ceiling.

That formula has one correlation per pair, which is unmanageable for 500 names. So traders compress all of them into a single average correlation $\rho$ — pretend every pair shares the same $\rho$ — and ask: what value of $\rho$ would justify the index vol the market is actually quoting? That number is the implied correlation — the correlation the options market is pricing in, backed out from the gap between index implied vol and single-name implied vols.

For an equal-weighted basket of $n$ similar names, the algebra simplifies to a clean, much-used proxy:

$\rho \approx \dfrac{\sigma_{\text{index}}^2}{\sigma_{\text{avg}}^2}$

Read it slowly: implied correlation is the index variance divided by the average single-name variance. The big trap is to forget the squares and divide the vols directly — correlation lives in variance (squared) space, so you must square both before dividing.

Worked example. Take the five-stock basket from the island. Each name sits near 31% implied vol, so $\sigma_{\text{avg}} = 0.31$ and $\sigma_{\text{avg}}^2 = 0.0961$. Now price two regimes:

Walk the calm row by hand: $0.183^2 = 0.0335$, and $0.0335 \div 0.0961 \approx 0.35$. The stressed row: $0.286^2 = 0.0818$, and $0.0818 \div 0.0961 \approx 0.85$. Same five stocks, same 31% average — the only thing that moved was the index vol, and it dragged implied correlation from 0.35 all the way to 0.85. The index vol bar and the implied-correlation number are two views of one quantity: how synchronized the market expects the basket to be.

The dispersion trade

Here’s the actual trade. You sell index volatility — via index variance swaps, index straddles, or index options — and simultaneously buy single-name volatility on the constituents, via their variance swaps or straddles. Two legs, opposite signs, same underlying basket.

Why bother with two legs instead of just shorting the rich index vol outright? Because a naked short index-vol position is mostly a bet on the level of volatility — you’d lose if vol simply rose for everyone, correlation or not. The long single-name leg is the hedge: if volatility rises across the board, your single-name longs gain roughly what your index short loses. What doesn’t cancel is the piece that depends on how the two move relative to each other — which is exactly the correlation. Net out the common vol exposure and what’s left is a comparatively clean wager on correlation:

So the punchline: a dispersion trade is short implied correlation. You collected the spread between index vol and single-name vol, and you profit if realized correlation comes in below the implied correlation you sold — that is, if the stocks turn out to be more independent than the market priced. If they end up more synchronized than priced, you lose.

Two practical notes on sizing. Traders don’t just dump equal dollar amounts into each leg; they vega-weight (match the two legs’ sensitivity to a 1-point change in vol) or gamma-weight (match their sensitivity to realized moves) so the structure is roughly vega-neutral at inception — neutral to the overall level of vol, leaving correlation as the live exposure. The exact weighting scheme is a desk-by-desk choice, but the goal is always the same: strip out the level-of-vol bet so the correlation bet stands alone.

Why it usually works — and how it blows up

The good news first. Implied correlation is usually priced too high, which is exactly why selling it pays. The reason is a demand imbalance you’ve half-met already: institutions are relentless buyers of index puts as portfolio crash insurance. That one-directional demand bids up index option prices, which inflates index implied vol, which — by the proxy above — inflates the implied correlation embedded in it. Meanwhile single-name options see far less hedging demand and stay comparatively cheap. The dispersion trader sells the over-bid index vol and buys the under-bid single-name vol, pocketing the spread. It’s a structural premium, a first cousin of the volatility risk premium — you’re being paid to provide something the market chronically over-pays for.

Now the bad news, and it’s the whole game. The premium isn’t free money; it’s compensation for a vicious tail. In a market crash, stocks stop disagreeing. Everything sells off together — correlation spikes toward 1. And you know from the variance identity what that does: with $\rho \to 1$, the diversification gap collapses, and index vol shoots up toward the average single-name vol. The very discount you sold evaporates precisely when the world is on fire.

Toggle the island above from “Calm: low correlation” to “Crisis: high correlation” and watch it happen: the single-name bars don’t move, but the index-vol bar climbs from 18.3% almost up to the 31% average, and implied correlation jumps from 0.35 to 0.85. Now read that as P&L. You were short the index vol that just exploded and long single-name vol that barely moved — your short leg is hemorrhaging while your hedge sits still. The short-correlation book takes a large loss exactly in the crisis, when every other risky thing you own is also bleeding.

Practicalities

A few things separate the clean diagram from a real book:

1. Constituent selection. You can’t trade liquid options on all 500 names — the illiquid tail would eat you alive on bid/ask. Desks pick a liquid subset of constituents and weight it to proxy the index, accepting some basket-mismatch (tracking error) as the price of executable legs.
2. Single-name option costs. The long leg is many separate option positions, each with its own bid/ask spread to cross and its own premium to pay. Those frictions are a real drag — dispersion has to clear a meaningful cost hurdle before it’s net profitable, so a thin edge can be eaten entirely by execution.
3. Vega- vs gamma-weighting. As noted, you size the legs to be roughly vega-neutral (matched to a change in implied vol) or gamma-weighted (matched to realized moves), depending on whether you want the bet to ride on implied or realized correlation. The two aren’t identical, and the choice shapes exactly which “correlation” you’re long or short.
4. What you’re really saying. Strip away the machinery and a dispersion trade is one sentence: single-name vol is cheap relative to index vol. You’re not forecasting the market’s direction or even its overall volatility — you’re claiming the relative pricing of parts versus whole is off, and that the gap will widen (or at least not collapse) in your favor.

When to use it

Put the trade on when implied correlation is historically elevated — when the index vol is unusually rich relative to its constituents, the gap is stretched wide, and the premium for selling correlation is fat. That’s when the market is most over-paying for the synchronization it fears. Conversely, when implied correlation is already low, there’s little premium left to harvest and the asymmetry turns ugly: small reward, same catastrophic tail. And always size for the crash, because the crash is the trade’s defining feature, not its footnote.

Putting it together

A stock index is a choir, and it can only be as loud as its singers are synchronized — so index implied vol sits below the weighted-average single-name implied vol, with the gap governed entirely by correlation. Square the vols and that gap becomes a number: implied correlation, $\rho \approx \sigma_{\text{index}}^2 / \sigma_{\text{avg}}^2$, the synchronization the option market is pricing right now. The dispersion trade sells index vol and buys single-name vol, netting to a clean short-correlation position that wins when stocks realize more independently than priced. It usually wins because index-put demand keeps implied correlation chronically rich — a cousin of the VRP — but it carries the same vicious negative skew: in a crash correlations rush toward 1, the diversification gap collapses, index vol explodes toward the single-name average, and the short-correlation book takes its big loss exactly when everything else is burning. Run it on a liquid weighted subset, vega- or gamma-weight the legs, put it on when implied correlation is elevated, and never forget you’re being paid pennies to stand in front of a steamroller.

Next — we move from selling correlation to the broader question of building a volatility portfolio: combining VRP harvesting, dispersion, and tail hedges so the steady-premium strategies don’t all detonate on the same crash day.