Backtesting Crypto Gamma on IBIT, MSTR and COIN

Most published crypto positioning research does not state its sample. That is usually because the sample is short, uneven, and would undercut the conclusion if stated plainly. This article states it first.

What follows is the coverage and a research design. It deliberately does not report performance results, because a study worth citing has to be run and pre-registered rather than assembled to fit a narrative. The design below is the one we think fits the data. For an example of the format applied end to end, see the 8-year GEX/DEX/VEX/CHEX study on SPY.

Coverage

Queried 17 August 2026:

Two things in that table should change how you design a study.

Index on trading sessions, not calendar days. MSTR spans 3,511 calendar days and 2,405 sessions with data - roughly 69% of calendar days, which is what the trading calendar leaves once weekends and holidays are out. Any study that indexes on calendar dates rather than sessions will sample non-trading days and fill them with whatever your join does by default, which is usually the previous observation. That manufactures autocorrelation and flatters any momentum result.

IBIT is short. 433 sessions since options listed in November 2024 is enough to spot-check a result found elsewhere, not enough to establish one on its own. Hold it out as the out-of-sample check.

The Listing-Date Trap

Every crypto instrument in this table started trading at a moment chosen by an issuer responding to demand. That is not a random start date, and it interacts badly with regime studies.

• IBIT options listed in November 2024, into an environment of high interest. The sample therefore begins in an unusually active period rather than a representative one.
• COIN listed in April 2021, within weeks of a cycle peak. A study starting at IPO inherits a drawdown that has nothing to do with dealer positioning.
• MSTR's bitcoin strategy began in 2020, so its 2017-2019 history is a different company. The ticker is continuous; the economic entity is not.

The MSTR point is the subtle one. A naive study using all 2,405 sessions is silently pooling a pre-bitcoin software company with a levered bitcoin treasury vehicle. Either split the sample at the strategy change or exclude the earlier period, and say which you did.

A Design That Survives This

The question worth asking is narrow: does dealer gamma regime carry information about subsequent realised volatility? Not returns - volatility. The gamma mechanism is a statement about hedging amplifying or dampening moves, so realised volatility is the outcome the theory actually predicts. Testing it against directional returns tests a claim nobody made.

1. Sample. MSTR from 2020 forward (post strategy change) and COIN from 2021, both indexed on trading sessions. Hold IBIT out entirely as an out-of-sample check given its short window.
2. Signal. Sign of net GEX at the close, plus distance from spot to the gamma flip in percent. Two states, then a finer four-state split once the two-state result is known.
3. Outcome. Next-day realised volatility, Yang-Zhang estimator, compared against a trailing baseline. See why Yang-Zhang rather than close-to-close.
4. Control. Bitcoin's own realised volatility over the same window. Without it you will rediscover that crypto is volatile when crypto is volatile, and attribute it to gamma.
5. Inference. Block bootstrap, not IID resampling. Gamma regimes persist for days, so independent resampling will overstate significance badly.
6. Pre-registration. Fix the two-state hypothesis and the sample before looking at outcomes. The four-state refinement is exploratory and must be labelled as such.

Pitfalls Specific To Crypto

The underlying trades when the options do not. IBIT, MSTR and COIN options trade US equity hours; bitcoin does not stop. A meaningful fraction of the move you are trying to explain happens while the hedging channel is closed. Overnight and intraday returns are different objects here and should not be pooled.

Corporate actions. MSTR has split and issued convertibles repeatedly. Strike-level open interest across a split is not comparable unless adjusted, and an unadjusted series will show spurious jumps in aggregate gamma exactly at the split dates.

Settled versus flow open interest. Historical GEX is computed on settled open interest, which is a start-of-day quantity. It is the correct input for a structural positioning study and the wrong one for an intraday timing study. The distinction is worked through in effective open interest.

Two hedging populations in MSTR. Convertible arbitrage desks hedge MSTR equity alongside options dealers. A gamma study on MSTR is not measuring a clean options-hedging channel, and the convert issuance calendar is a confounder you should control for or acknowledge.

Pulling The History

GET https://historical.flashalpha.com/v1/tickers?symbol=IBIT
GET https://historical.flashalpha.com/v1/exposure/gex/MSTR?at=2026-06-16T15:30:00
GET https://historical.flashalpha.com/v1/exposure/summary/COIN?at=2026-06-16T15:30:00

Historical replay is an Alpha-tier capability. The paths are the live paths on the historical host with an at timestamp, so the code you research with is the code you trade with. Live endpoint and entitlement detail is in the crypto options data API guide.

The crypto gamma archive runs from January 2017 or each instrument's option listing, whichever is later: MSTR offers 2,405 trading sessions back to 2017, COIN 1,335 back to its IPO, IBIT 433 since its options listed. Stating the sample up front is the difference between research and marketing. Index on trading sessions, split MSTR at its 2020 strategy change, hold IBIT out as the out-of-sample check, test against realised volatility rather than returns because that is what the gamma mechanism actually predicts, and use a block bootstrap because regimes persist. We will publish results against this design rather than in place of it.