Barbon & Buraschi — Gamma Fragility (Swiss Institute of Banking and Finance WP 2020/05, Nov 2020)

T2 — a university working paper, not peer-reviewed: Working Papers on Finance No. 2020/05, Swiss Institute of Banking and Finance (S/BF-HSG), 5 November 2020; first version January 2019, also on SSRN as 3725454. andrea-barbon (University of St. Gallen) and andrea-buraschi (Chair of Finance, Imperial College Business School).

The reason to hold it next to pervasive-impact-of-option-trading is that it asks the same question without the private data — which is the position anyone building a dealer-gamma signal today is actually in.

The proxy, and the assumption inside it

Gammas and open interest come from OptionMetrics (optionmetrics) Ivy DB, 1996–2017, for the 300 underlyings with the largest average dollar open interest, index and single-name; intraday returns and volume from TAQ. Dollar gamma per contract is summed across strikes and maturities and scaled by the underlying's average daily dollar volume.

Then comes the step that separates this from the CBOE-data papers. The authors write that "regulatory restrictions limit the amount of information available about the specific identity of derivative holders" — so who holds which side is not observable, and they must assume it. They build two proxies: calls minus puts (dealers long calls, short puts) and calls plus puts (dealers long both), and they choose the first by citing information-in-option-volume, whose CBOE data shows non-market-makers buying index puts and selling index calls on average. So the public-data gamma measure is anchored to a private-data finding, and for single-name options the same paper says the imbalance "varies over time and in the cross section" — the assumption is weakest exactly where the unusual-options-activity screener lives.

Findings

• Volatility. Absolute daily return regressed on the previous day's gamma imbalance gives a negative, strongly significant slope — the same sign as the CBOE-data result. The effect is gone after about two trading days, and it is larger for the least liquid underlyings, which is the paper's central claim: the feedback needs limited liquidity to exist at all, since in a frictionless market options are redundant.
• Intraday autocorrelation. Negative gamma imbalance goes with intraday momentum, positive with reversal, and the coefficients peak at a 30-minute sampling horizon — read as roughly how often dealers rebalance. The authors state plainly that the economic magnitude here is limited: one standard deviation of gamma imbalance shifts autocorrelation by about 1% of a standard deviation.
• Flash crashes. Using drift-burst detection to date the events, crashes are more likely and larger when ex-ante gamma imbalance is negative — economically significant in the early sample, weaker after circuit breakers arrived in 2010.
• Not just toxic flow. Gamma imbalance keeps its significance after controlling for VPIN, the order-flow-toxicity measure of Easley, López de Prado and O'Hara — the same marcos-lopez-de-prado whose trial-count work sits behind backtest-overfitting here.

What it is worth as evidence

It is the weakest-tier of the three options sources in this spoke and the one closest to what a practitioner can rebuild, and those two facts are the same fact. Everything here is estimated from public open interest under a sign assumption, whereas pervasive-impact-of-option-trading observed the positions. That the two agree on sign and on the liquidity condition is the useful part; the honest reading is that a public gamma-exposure number is a proxy for a quantity its own authors could not see.

Related

dealer-gamma-exposure · options-greeks · unusual-options-activity · pervasive-impact-of-option-trading · information-in-option-volume · option-pricing · andrea-barbon · andrea-buraschi · synthesis