What is Dispersion Trading?
Dispersion trading is long single-name volatility, short index volatility. It expresses a view that realised correlation across the index will be lower than implied — the "correlation risk premium."
Definition
In a dispersion trade, you sell index vol (via variance swap, straddle, or futures option) and buy a basket of single-name vols across the index constituents, weighted to be approximately vega-neutral. The trade pays when realised correlation between the constituents is lower than the level implied by the relative pricing of index-vol to single-name-vol. Equivalently: it pays when single names move independently around an index that does not move. Dispersion is one of the longest-standing edges on a vol desk — implied correlation has historically averaged well above realised correlation, especially in non-crisis regimes — but it is also dangerous: in a true correlation event (March 2020 style), everything goes to 1 and the trade loses violently.
Why it matters & how it's calculated
The textbook formulation uses variance swaps: σ²_index ≈ Σᵢ wᵢ² · σ²ᵢ + 2 · Σᵢ<ⱼ wᵢ wⱼ ρᵢⱼ σᵢ σⱼ. The first sum is the "weighted single-name variance" — what you would get if all correlations were zero. The cross-term is the correlation contribution. Implied correlation = (σ²_index − Σ wᵢ² σ²ᵢ) / (2 Σᵢ<ⱼ wᵢ wⱼ σᵢ σⱼ). Trade: be long the single-name variance basket, short the index variance, sized so the cross-term gives you positive expected P&L if realised correlation is below implied. In practice, listed options (not variance swaps) are used, with constant rehedging to maintain the vega-neutral construction. Dispersion has its own Greeks: it is short correlation, long idiosyncratic vol, and roughly flat to overall vol level if hedged. The "correlation risk premium" — implied minus realised correlation — is the source of expected P&L, and it has historically averaged 10-20 correlation points. The trade is structurally short the tail: every correlation crash event is a drawdown, with 2008, March 2020, and August 2024 all being painful reminders.
Worked example
SPX implied vol 15%, weighted single-name vol 22%. Naive: σ²_index < Σ wᵢ² σ²ᵢ × ρ. If the math gives implied correlation = 0.45, but historical realised correlation has averaged 0.30, the trade is: short SPX variance, long basket of top-50 SPX single-name variance, vega-weighted. Expected P&L if realised correlation comes in at 0.30: the difference of 0.15 correlation points scaled by the index vega. Risk: a crash where correlation goes to 0.90 — single-name vols rise some, but index vol rises a LOT more, and the short index leg dominates the loss.
Related concepts
Learn this in depth on a live desk
These are the building blocks the desk uses every day. The 180+ hour Whop course shows you how they fit together on a real book — and the book walks you through the framework end-to-end.