Gamma exposure: signal or description?

Aggregate gamma exposure has become a standard object in index options commentary. The construction is simple, the data is public, and the resulting number is presented as though it forecasts realised volatility and price behaviour. Whether it does is an empirical question. What is worth separating first is that the number is, in its own construction, a description rather than a prediction, and the arguments made for it frequently rely on confusing the two.

What is actually being computed

The quantity is the sum, across strikes and expiries, of option gamma multiplied by open interest multiplied by contract multiplier and by spot squared, signed according to an assumption about which side of each contract the dealer community holds.

Every term except the last is mechanical. Gamma comes from a pricing model given an implied volatility. Open interest is published. The multiplier is a contract specification. The signed aggregation is where the modelling actually happens, and it is the part least often examined.

The sign convention is the whole model

The standard convention assumes dealers are short calls and long puts, on the reasoning that customers buy calls for upside and buy puts for protection, leaving dealers on the other side. Under it, a positive aggregate implies dealers hedge against the move, buying weakness and selling strength, which suppresses realised volatility. A negative aggregate implies the reverse and amplifies it.

That mechanism is sound. The assumption underneath it is a heuristic, and it fails in identifiable ways. Institutional overwriting programmes sell calls in size, putting dealers long calls. Put spread collars and risk reversals invert the put side. Dispersion and volatility relative value trades leave dealers positioned by strike in ways no aggregate convention captures. On any given expiry a meaningful fraction of open interest sits on the opposite side of the assumption, and nothing in the public data says which fraction.

So the headline number is a function of an unobserved variable. It is not wrong to compute it, but the uncertainty on the sign assumption is the dominant uncertainty in the estimate, and it is almost never reported. A figure quoted to three significant digits, resting on a binary assumption about counterparty positioning, is presenting a precision it does not have.

Open interest and volume answer different questions

Two variants circulate and they are not interchangeable.

Open interest weighting measures positions that exist. It is a stock, it updates once daily after clearing, and it is stale intraday.

Volume weighting measures contracts that traded. It is a flow, available intraday, and it does not distinguish opening from closing transactions. A contract traded to close a position adds to volume while removing exposure, so a volume-weighted aggregate counts position reduction as position increase.

These disagree most exactly when they matter most, which is during heavy trading, because that is when closing flow is largest. Comparative work on which weighting carries more information consistently favours open interest, and the reason is structural rather than incidental: volume weighting is measuring something other than positioning while being interpreted as though it measures positioning.

Description and prediction are different claims

A descriptive statistic summarises current state. A predictive signal contains information about future state not already reflected in price. Gamma exposure is unambiguously the first. Whether it is also the second requires evidence that the estimate improves a forecast beyond what is available from the price and volatility surface it was computed from.

That bar is higher than it looks, because the inputs are not independent of what is being predicted. Gamma is computed from implied volatility. Implied volatility is a forecast of realised volatility. A finding that gamma exposure predicts realised volatility may be recovering the information already in the surface, transformed. The correct test conditions on the surface first and asks what the positioning estimate adds.

The second difficulty is that the mechanism, if real, is self-limiting. If dealer hedging in a negative-gamma regime amplifies moves, and that is known and public, participants position for it. The effect that remains is the part not already arbitraged, which is smaller than the raw relationship suggests and unstable over time in a way a full-sample backtest will average away.

What is defensible

Several claims survive scrutiny and are worth separating from the ones that do not.

Large open interest concentrations at specific strikes are real and observable, and hedging flow around them is a genuine mechanism. Realised volatility does differ between high-positive and high-negative aggregate regimes in most samples. Expiry effects around large concentrations are documented and mechanical.

What does not follow is that a threshold on the aggregate produces a tradeable rule. The regime relationship is a conditional description of a volatility environment, not a directional forecast, and the distance between those is where most strategies built on this fail. The specific case of concentration levels treated as barriers is taken up in why gamma walls do not produce edges.

How to treat it

Treat the aggregate as a conditioning variable rather than a signal. It describes an environment, and environments are useful for sizing, for choosing between strategies, and for setting expectations about realised volatility. They are not entries.

Report the estimate with its assumptions attached: which sign convention, which weighting, which expiries, and what fraction of open interest sits in strikes where the convention is least reliable. Compute it both ways and look at the disagreement, since the disagreement is itself informative about how much the number should be trusted on that day.

And test it against the surface rather than against price alone. The question is never whether gamma exposure correlates with subsequent volatility. It is whether it adds anything to what the options market was already saying, which is a harder question and the only one whose answer is worth acting on.