The volatility of an equity index is not a free parameter. Given constituent volatilities and weights, it depends on one additional quantity: the average pairwise correlation. Start from the portfolio variance identity, σ²_I = Σw_i²σ_i² + Σ_i≠j w_iw_jσ_iσ_jρ_ij, and assume a common correlation. You can then back out an implied correlation from index and single-stock option prices. That number is observable every day, and over long samples it has told a consistent story.

Across decades of data on the S&P 500 and Euro Stoxx 50, implied correlation has usually exceeded the correlation later realized, often by a wide margin. Index options are expensive relative to options on the stocks that make up the index. The gap is not a calibration artifact. It persists under model-free variance measures, across sampling windows, and after reasonable allowances for transaction costs. Option markets are effectively charging a premium for one specific risk: the risk that diversification fails.

Dispersion trading tries to harvest that gap. The trader sells index volatility, buys constituent volatility, and lets the difference between implied and realized correlation accrue. The concept fits on a napkin; the execution does not. The trade carries residual exposure to the level of volatility. Index and single-name markets have very different liquidity. And the payoff is sharply negatively skewed. Its economics therefore depend on how carefully the position is constructed and how honestly its tail is priced.

Driessen, Maenhout and Vilkov showed something with important consequences for option pricing. Index options carry a large, persistent variance risk premium. Individual equity options carry a much smaller one, and in some samples it is statistically indistinguishable from zero. If constituent variance is priced roughly fairly and index variance is not, the wedge must sit in the cross-term. In other words, the index variance risk premium is largely a correlation risk premium.

The economic foundation is that correlation is countercyclical. Pairwise correlations rise in drawdowns, exactly when diversification is most valuable and least available. A claim that pays off when correlation rises therefore insures against the collapse of diversification itself. In stochastic discount factor terms, realized correlation covaries positively with marginal utility. Long-correlation payoffs arrive in bad states, so they trade above their expected value. The negative expected return to buying correlation is the price of that insurance.

Order flow reinforces this. Pension funds, insurers and asset managers hedge with index options because index options are liquid and match aggregate portfolio exposure. There are few natural sellers. Single-name options, by contrast, attract systematic supply from covered-call writers and yield-enhancement programs, which pushes their implied volatilities down. Gârleanu, Pedersen and Poteshman formalized why this matters. When dealers face inventory constraints, end-user demand moves option prices away from risk-neutral fair value. The result is a structural wedge: buyers dominate index volatility while sellers dominate single-stock volatility.

Measurement requires care. The standard estimator is ρ_impl = (σ²_I − Σw_i²σ_i²) / Σ_i≠j w_iw_jσ_iσ_j, computed from model-free implied variances built from option strips. It rests on three simplifying choices. It assumes equicorrelation, it often uses a subset of constituents, and it is sensitive to which strikes enter the calculation. Implied correlation also has its own skew. Downside strikes embed higher correlation than at-the-money options, because the market expects stocks to fall together more readily than they rise together.

The disciplined interpretation, in the spirit of Fischer Black, is that a persistent return in options markets usually signals a risk the model has omitted, not a free lunch. A seller of correlation is underwriting the diversification of the market portfolio. Steady positive average returns from that activity are consistent with fully rational pricing. They compensate for losses that arrive precisely when every other part of an institutional book is also under stress.

Takeaway

The correlation premium is the market's price for insuring diversification itself. Selling it is not arbitrage; it is underwriting a crash-correlated liability.

The cleanest instrument is a correlation swap, which pays notional × (ρ_realized − K). It isolates correlation directly. In practice, however, it trades over the counter, is illiquid, and is priced with wide dealer markups that absorb much of the premium. Most practitioners therefore build correlation exposure synthetically. They sell index volatility and buy constituent volatility, either through variance swaps or through delta-hedged options.

In variance-swap form, the trader is short index variance and long constituent variance, weighted by index weights. The payoff is approximately Σw_iσ_i² − σ²_I. With many names and equicorrelation, this reduces to roughly (1 − ρ) times average constituent variance. So the trade is long dispersion, but its payoff scales with the volatility level. It is not a pure correlation bet. Losses in a high-correlation regime are amplified precisely because volatility is also high in that regime.

In options form, the trader buys straddles on constituents and sells index straddles, delta-hedging each leg with stock and futures. Sizing is the central design decision. A vega-neutral book sets index vega equal to aggregate constituent vega. Because single-name implied volatility exceeds index implied volatility, the long leg then bleeds more theta than the short leg collects, so the trader pays carry while waiting for dispersion. A theta-neutral book removes the carry but leaves the trader net short vega. Each weighting scheme is a choice about which residual exposure to hold.

Implementation compounds these issues. Desks typically trade only the largest constituents, enough to span most of the index variance, which introduces basis risk against the residual names. Single-name bid-ask spreads are often several times wider than index spreads, so frequent rebalancing erodes the edge. As spot moves, straddles drift away from the money and their vegas diverge, which forces periodic re-striking. Earnings dates inside the tenor add idiosyncratic jumps that favor the long leg, though those jumps are already partly reflected in single-name prices.

The choice of vehicle determines the error structure. A delta-hedged option earns approximately ½∫Γ_tS_t²(σ_r² − σ_i²)dt. Its P&L is therefore weighted by dollar gamma along the realized path, and a correct correlation view can still lose money if the large moves happen far from the strike. Variance swaps remove that path dependence. But single-name variance swaps have been capped and thinly traded since 2008. Execution is ultimately a decision about which approximation errors one is prepared to tolerate.

Takeaway

A dispersion trade is never just a correlation bet; it is a correlation bet wrapped in volatility-level, path, and liquidity exposures, and the weighting scheme decides which of those you keep.

Short correlation is economically short a crisis. In October 2008, August 2011 and March 2020, realized correlation among large-cap constituents surged as a single macro factor came to dominate returns. Index realized volatility exceeded even elevated implied levels. In those episodes, losses on the short index leg dwarfed gains on the long single-name leg. The return distribution resembles that of a short out-of-the-money index put: frequent modest gains interrupted by rare, severe losses.

A one-factor model clarifies the mechanism. Let r_i = β_if + ε_i. Pairwise correlation is then approximately β²σ_f² / (β²σ_f² + σ_ε²). When factor volatility jumps, correlation rises toward one mechanically, even if idiosyncratic volatility does not change. Correlation spikes are factor-volatility spikes. This framing points directly to the hedge. The exposure to protect against is convexity in index volatility, and liquid instruments exist for exactly that.

The main hedging tools follow from this. Out-of-the-money index puts or VIX calls can be financed partly from the premium the trade collects. Selling capped rather than uncapped index variance limits the loss on the short leg. Corridor variance on the downside region offers a more targeted alternative. Each hedge consumes premium. The real decision is therefore not whether to hedge but how much of the expected return to surrender for a truncated left tail.

Greeks are local, while correlation jumps are global. Sizing should therefore come from a scenario grid that varies correlation and volatility level jointly. A typical stress case pushes correlation toward 0.9 and doubles index volatility, then incorporates realistic exit costs. In stress, single-name option liquidity evaporates. The long leg cannot be monetized near mid, even as margin calls on the short index leg arrive immediately. A position that survives this grid with acceptable capital impact is sized correctly; one that only looks good on vega is not.

Conditioning improves the trade-off. The spread between implied and realized correlation varies through time, and the term structure of implied correlation contains information: inversion tends to accompany acute stress. Entering when implied correlation is low and markets are calm captures a thinner premium and carries greater spike risk. Scaling exposure to the spread and to the term structure produces a conditional strategy. That strategy is more robust than permanently harvesting a premium whose losses cluster in time.

Takeaway

Treat a correlation spike as a jump in factor volatility rather than an exotic event, and the hedge becomes tractable: you are buying convexity on the market factor with part of the premium you collect.

The implied correlation premium is among the best-documented regularities in derivatives markets. It rests on two foundations: diversification fails in bad states, and order flow is asymmetric, with structural buyers of index protection and structural sellers of single-name volatility.

Dispersion trades can harvest the premium, but the quantity they deliver is rarely pure correlation. Weighting conventions, the choice of vehicle, constituent selection and rebalancing frequency each introduce exposures that can dominate the intended one. Careful construction is the strategy, not an operational detail.

For institutional practice, the premium is better understood as compensation for underwriting systemic risk than as an anomaly to exploit. Its expected return is real, and so is its left tail. The managers who earn it durably are those who price the crash scenario explicitly, size positions against joint stresses in correlation and volatility, and accept that hedging is the cost of staying in the trade.