Do Gamma Levels Actually Work? We Tested 100 Days of Ours
- We tested three things our own platform publishes against 100 trading days of our own stored levels. Two of the three did not work.
- Worked: the regime read 15 minutes after the open. Days opening in short gamma ran 16–36% wider than days opening in long gamma.
- Did not work: the call wall held 44% of the times it was touched, the put wall 51%. That is a coin flip.
- Did not work: the gap between index and ETF gamma levels told us nothing usable about direction — tested twice, on two different pairs.
- Wider is not directional. None of this predicts which way price goes.
The question nobody publishes an answer to
Every gamma platform publishes levels. Ours included. What almost nobody publishes is whether those levels did anything — because measuring it honestly means admitting when they didn’t.
A gamma level is a model output, not an observation of the market. So the useful question is not “where is the level” but “does price behave differently around it than it would anyway”. That question has an answer, and the answer can be no.
We are in an unusual position to ask it. Our engine has computed and stored levels every ten seconds of every session since April. That gives us the level and what price did afterwards — the second half being the part that never gets published.
How we tested
100 trading days, 29 April to 7 August 2026, on the six tickers with continuous stored history: SPY, SPX, QQQ, NDX, NVDA and UVXY. Snapshots taken every ten seconds, resampled to one observation per minute, restricted to the regular session.
Significance throughout is a permutation test: we shuffle the labels 20,000 times and count how often chance alone produces a gap as large as the one we measured. Where we report a median it is because a mean can be manufactured by two violent days and a median cannot.
What held up: the opening regime
Fifteen minutes into the session we read whether dealers are net long or short gamma. Then we measure how wide the range runs for the rest of the day. Nothing after 9:45 ET feeds the read — it uses only information a trader would have at that moment.
| Ticker | Long-gamma days | Short-gamma days | Median range, long | Median range, short | Ratio | p-value |
|---|---|---|---|---|---|---|
| QQQ | 28 | 37 | 1.18% | 1.56% | 1.32× | 0.0003 |
| SPX | 44 | 19 | 0.77% | 1.05% | 1.36× | 0.0016 |
| SPY | 25 | 39 | 0.68% | 0.89% | 1.30× | 0.034 |
| NDX | 43 | 13 | 1.33% | 1.53% | 1.16× | 0.064 |
Three of the four clear the usual significance bar, one at roughly 3,000-to-1 odds against chance. The fourth, NDX, misses it — and has the smallest sample of short-gamma days, which is the most likely reason.
What didn’t hold up: the walls
The call wall and the put wall are the levels everyone quotes, ours included. So we asked the simplest possible question about them: when price actually reaches a wall, does the wall hold?
For every session where a ticker traded within 0.1% of a wall, we checked whether it closed on the side it approached from.
| Ticker | Call wall touches | Held | Put wall touches | Held |
|---|---|---|---|---|
| SPY | 46 | 48% | 33 | 45% |
| SPX | 40 | 45% | 35 | 66% |
| QQQ | 38 | 29% | 27 | 48% |
| NDX | 44 | 43% | 39 | 46% |
| NVDA | 31 | 42% | 19 | 53% |
| UVXY | 18 | 72% | 20 | 50% |
| All six | 217 | 44% | 173 | 51% |
That is a coin flip. Across six tickers and a hundred days, a wall being touched told us nothing useful about whether it would hold. The put wall landed at 51% — one percentage point from pure chance.
We are not going to dress this up. A wall is a real feature of the options chain: it marks where gamma is concentrated, and that concentration is a fact you can verify. What we could not show is that touching one predicts what happens next. Anyone quoting you “the wall held 80% of the time” owes you their sample size, their definition of a touch, and their failures.
What didn’t hold up either: index vs ETF divergence
SPX and SPY track the same index, so once you scale them their gamma levels should agree. They frequently don’t — sometimes by more than a hundred SPX points. The gap is real and measurable, and it looked like it ought to mean something.
It doesn’t, at least not for direction. Across 1,110 hours on SPX/SPY and a second full test on NDX/QQQ, the size of the divergence in any given hour told us nothing usable about where price went next. Sorted into quartiles, the share of up-hours came out at 51%, 55%, 50% and 54%. No gradient, no edge.
We ran it twice, on two different pairs, because a first null result is the one you are most tempted to explain away.
The mistake that nearly buried the whole thing
Our engine runs around the clock, so weekends and market holidays sit in the data with a frozen price. Our first pass counted them as trading days.
It made the regime effect look weak — only one ticker of four cleared significance — and we came close to reporting that as the result. After removing them (105 ticker-days discarded of 374) three of four cleared it.
What this does not prove
- One hundred days is a short sample. It covers a single volatility environment. The same study run in a different quarter could disagree with this one.
- Wider is not directional. Knowing a day is likely to run wide tells you nothing about which way it runs. It is an observation about volatility, not a trade.
- This is our model, tested against itself. Platforms compute these levels differently — different expirations, different volatility inputs, different dealer assumptions. These results describe our numbers, not gamma in the abstract.
- We did not test a strategy. No entries, no exits, no costs. Whether anything here survives commissions and slippage is a separate question we have not answered.
Why we published the failures
Two of the three things we tested didn’t work, and one of those two is a level on our own dashboard. We could have published the finding that worked, said nothing about the other two, and it would have read like a stronger product.
We would rather you know which of our numbers we can defend and which we can’t. If a level is a coin flip it belongs in your read as context rather than as a trigger — and the only way you can know that is if someone measures it and tells you.
If you replicate any of this and get a different answer, we want to hear about it. We will publish the correction.
See today’s regime before the range sets
The regime read this study found useful is on the dashboard from the opening bell, for SPY, SPX, QQQ, NDX and 26 more. From $19/mo during beta.
Frequently Asked Questions
Do gamma levels actually work?
It depends which level. Tested against 100 trading days of our own stored data, the dealer regime read 15 minutes after the open was associated with a 16–36% wider daily range, significant on three of four index instruments.
The call wall and put wall were not. They held 44% and 51% of the times price touched them — indistinguishable from chance.
Do call walls and put walls hold?
In this study, no more often than chance. Across 217 call wall touches and 173 put wall touches on six tickers over 100 days, the call wall held 44% of the time and the put wall 51%.
A wall is a real feature of the options chain — it marks where gamma is concentrated — but touching one did not predict what happened next.
Does the gamma flip predict price direction?
No. Nothing in this study predicts direction. The regime finding is about how wide the day’s range runs, not which way it goes.
The divergence between index and ETF gamma levels was tested twice — SPX/SPY over 1,110 hours, then NDX/QQQ — and showed no usable directional edge in either.
How was the study measured?
100 trading days, 29 April to 7 August 2026, on SPY, SPX, QQQ, NDX, NVDA and UVXY. Snapshots every 10 seconds resampled to one per minute, restricted to the regular session.
Weekends, holidays and sessions where price changed on fewer than half the minutes were discarded as non-trading. Significance is a permutation test with 20,000 shuffles.
Why publish results that make your own levels look weak?
Because a level that behaves like a coin flip belongs in your read as context rather than as a trigger, and the only way to know which is which is if someone measures it and says so.