I saw this line in some recent research,
GEX high: short next-day volatility, meaning bet on a quiet day (for example, sell options or trade a range).
GEX is gamma exposure, it estimates how much gamma the option dealers hold. When dealers hold a lot of positive gamma, they hedge by selling into rallies and buying into dips, so in theory the market moves less. It sounds good, and many people use it like a rule.
The problem is, high GEX usually comes together with low VIX. When VIX is low, options are already cheap, so “tomorrow is quiet” is not news, the market already priced it. So the real question is,
Does high GEX predict a smaller move than the options market already expects?
Why not my own data
I have recorded my own GEX every morning since August, but it is only 32 clean sessions. With 32 sessions I can only see a very strong effect. To see a small one I need about 680 sessions, roughly 3 years of recording. 😬
So I used free data instead,
- SqueezeMetrics daily GEX history, from 2011 to now, DIX.csv
- VIX, VIX9D and VIX1D daily closes from Cboe
I used VIX9D as the main control, not VIX. VIX looks 30 days ahead, which is too long for a next-day question. VIX9D looks 9 days ahead, VIX1D looks 1 day ahead but only starts from May 2022.
How I tested it
- Wrote the rules down and committed them before reading any data after 2018 (pre-registration, basically no moving the goalposts after seeing the result)
- Built and debugged the script only on 2012 to 2017, it reproduced the earlier numbers
- Ran 2018 to 2026 once, 2,196 days, no second try
The model is simple: next-day move size against the VIX9D forecast, plus the size of recent moves (1, 5 and 22 days), plus GEX ranked against its own past year.
Result
| Read | Forecast used | Days | GEX effect | One-sided p | Pass? |
|---|---|---|---|---|---|
| Main | VIX9D | 2,196 | −0.049 | 0.025 | Yes |
| 0DTE era, 2022 on | VIX1D | 1,097 | −0.043 | 0.063 | No |
| Check | VIX | 2,196 | −0.087 | 0.0002 | Yes |
| High GEX flag | VIX9D | 2,196 | −0.071 | 0.020 | Yes |
It passed, barely.
In plain numbers, the average next-day move from 2018 was 0.79%. On high-GEX days (top fifth of its past year), the next day moved about 0.07 points less than the forecast said, so about 9% smaller, around 5 SPX points. GEX adds only 0.2% to what the model can explain (R² from 0.340 to 0.342).
The part I find more interesting, the shorter the forecast, the weaker GEX looks.
- Against VIX (30 days), strong
- Against VIX9D (9 days), just passes
- Against VIX1D (1 day), does not pass
It looks like short-dated options already know most of what GEX knows.
What others found
- Barbon and Buraschi, Gamma Fragility, same direction and a similar size, and they did control for implied volatility
- Amaya, Garcia-Ares, Pearson and Vasquez, 0DTE Index Options and Market Volatility, used real dealer positions, their words for the effect are “not large”
- The SqueezeMetrics white paper, the popular source of this idea, sorts on GEX and on VIX separately, it never holds VIX fixed
So “high GEX, quiet day” is true, but mostly because high GEX means low VIX, and low VIX means quiet day.
What is next
This test is only about volatility, not money. To know if “sell options on high-GEX days” is a trade, I need to compare it with the straddle price, after costs. Straddle history for 2018 onward is paid data (Cboe DataShop, ORATS, ThetaData), so for now my own snapshots keep recording, one session a day.