Dealer Gamma in Practice: Why Most GEX Maps Hug the Spot — and How to Separate Structure from Day Weather
What this document is: a mechanics and measurement study of dealer gamma levels as a trading tool — not a performance study. Statistical validation of the regime hypotheses (running on five years of reconstructed SPY options history) will follow as its own paper. Data basis here: full option chains (OI, IV, per-strike volume) on QQQ/DIA/GLD, live observation of the NQ cash session on 2026-06-10. Not trading advice.
Gamma exposure maps ("GEX") have gone from niche tool to retail mainstream: gamma flip, call wall and put wall now hang on thousands of charts. What almost nobody discusses: most of these maps are methodologically broken — in a specific, measurable way. This study demonstrates the defect, quantifies it on a live trading day, and describes the architecture that fixes it.
The mechanics in 60 seconds
Options market makers keep their books delta-neutral — not out of opinion, but out of obligation. How they must hedge is determined by the sign of their aggregate gamma:
- Long gamma (spot above the flip): dealers sell strength and buy weakness → moves get dampened. Pinning at OI clusters, mean reversion, small candles.
- Short gamma (spot below the flip): dealers sell weakness and buy strength → moves get amplified. Trend days, large candles, gap risk — in both directions (which is why squeeze rallies in short-gamma phases are so violent).
GEX therefore maps the market's only large forced, price-insensitive order flow. That is why the levels carry information at all — and simultaneously the standard against which any concrete computation must be judged.
The defect: the 0DTE dominance problem
Gamma is not evenly distributed: it concentrates at the money and explodes toward expiry (for ATM options it grows roughly with 1/√T). A map that throws every maturity from 0 to 60 days into one sum is therefore arithmetically doomed to be dominated by the nearest expiry — and in the 0DTE era that means: by options that will no longer exist tonight.
The consequences, all measured on our own tool before we rebuilt it:
- The walls hug the spot. Morning of 2026-06-10, NQ: call wall 29,091 with spot at 29,084 — seven points away, 0.02%. That is not structural information; that is the ATM strike of today's expiry.
- The regime label becomes a coin flip. Gold the same morning: flip 4,178.34 vs. spot 4,177.80 — labeled "short gamma" on the basis of a 0.01% distance. A binary label in exactly the place where the map has nothing to say.
- The map decays during the session all by itself. This is the subtlest and most expensive effect — it deserves its own measurement.
Live decomposition: a call wall "migrates" 1,300 points — without a single new position
On 2026-06-10 the NQ fell roughly 1% intraday and implied volatility spiked. We computed the same option inventory at two points in time with identical (old) methodology — and additionally, at the second point, with separated maturity baskets:
| Computation | Flip | Call wall | Put wall |
|---|---|---|---|
| Blended 0–60 DTE, morning (spot 29,084) | 29,078 | 29,091 | 28,968 |
| Blended 0–60 DTE, 17:30 (spot 28,767) | 29,431 | 30,390 | 28,748 |
| Structure basket 7–45 DTE, 17:30 | 30,050 | 30,390 | 28,748 |
| Near basket 0–5 DTE, 17:30 | 29,098 | 29,569 | 28,748 |
Three lessons live in this table:
First: with unchanged methodology and unchanged open interest, the call wall jumped +1,300 points intraday. Nobody repositioned — OI only updates overnight. What changed: the market fell away from today's ATM strikes, their residual gamma collapsed (almost no time left + strikes now far out of the money), and the weighting tipped to the weekly clusters behind them. A blended map measures two different things in the morning and at midday — without telling you.
Second: the "new" structure flip at 30,050 did not appear out of nowhere — it was there all along. It was merely invisible inside the blend, drowned out by the 0DTE share. Separating the baskets creates no new levels; it unmixes existing ones.
Third: the near-basket flip (29,098) is practically identical to the morning blend (29,078). Put differently: the popular GEX map is in truth a 0DTE map that believes it is a structure map.
Consequence 1: three layers instead of one number
| Layer | Maturities | Data source | Character | Validity |
|---|---|---|---|---|
| Structure | 7–45 DTE | overnight OI | the terrain: the week's regime, real walls, the strike shelf | days |
| Near | 0–5 DTE | overnight OI | the morning weather: the day's pinning candidates | until ~midday of the US session |
| 0DTE flow | 0–1 DTE | today's per-strike volume | the live weather: the day's magnets as they actually form | rest of the day |
The third layer addresses the structural blind spot of every OI-based map: 0DTE positions opened today appear in no open interest (OI settles overnight). Today's per-strike option volume is the honest free proxy for that invisible positioning.
Validation on the same live day: after the first US trading hour, the volume-based 0DTE call wall sat at 29,167 — the NQ made its high of the day at 29.17k and sold off. None of the OI maps (all three rows of the table above) had that level. One day is an anecdote, not proof — but it is precisely the behavior the mechanics predict: intraday flow builds the walls that matter intraday.
Consequence 2: freeze, don't chase — the repainting fallacy
If levels move this much on recomputation, one conclusion suggests itself: recompute more often, "to keep the map current." That is the fallacy — and it has a precise refutation:
Intraday, only the weighting refreshes — not the object. Positioning (OI) is fixed from one overnight settlement to the next. A midday recomputation reweights yesterday's identical positions at the new spot and new IV — the result follows the price. But a level that follows price cannot serve as a reference: it is the chart equivalent of a repainting indicator — always plausible live, never predictive.
The proof is again in the live day: the frozen morning call wall (29,091) stood firm before the afternoon rally tested it (high 29.13–29.17k into the morning zone, sold off). A map recomputed at 15:00 would have shown entirely different values — and erased exactly that predictive hit. Only frozen levels can be judged on whether they work at all.
A practical operational note: for US underlyings, every computation between the OI update (~late morning European time) and the US open is identical — before the open, the data chain serves yesterday's closing IVs and spots anyway. There is exactly one meaningful computation time per day (pre-open, then freeze), plus optionally one deliberate volume update of the third layer after the first trading hour. Everything else is noise.
What this means for trading (and what remains to be proven)
The mechanics imply asymmetric readings — a different playbook per regime:
- Below the structure flip: moves get amplified. Breaks deserve trust and time; every rally is suspect until the flip is reclaimed and held. If the put wall breaks, the next target is the strike shelf behind it (second-largest clusters) — not "unlimited": the cascade starves where the gamma profile thins out, or when volatility turns (vanna buy-backs). The reversal signature is measurable: an IV peak plus a reclaim of the broken put wall from below.
- Above the structure flip: moves get dampened. Walls are magnets and fences; targets beat trails, partials at the wall, pinning behavior into expiry.
- In the flip zone (±0.3%): no signal. A binary regime label in this zone is self-deception — the zone belongs explicitly marked as "neutral."
Important for intellectual hygiene: so far these are mechanically motivated hypotheses with anecdotal live confirmation. The statistical test — does the breakout family perform measurably differently on short-gamma days? Does MFE systematically terminate at the walls? Do early session breaks in long gamma work better toward the magnet? — is currently running on five years of reconstructed SPY options history and will be published as its own paper under the same gates as the rest of this series (out-of-sample, costs, multiple-testing correction).
Methodology box
- GEX per strike: Black-Scholes gamma × OI × multiplier × S² × 1%, sign convention dealers long calls / short puts (call GEX +, put GEX −). This convention is a model assumption, not a dealer bank statement — GEX is an estimate of positioning with a known bias.
- Flip = zero crossing of total GEX over a spot grid (−25%/+30%; "no zero crossing inside the window" is itself a signal in crash phases: deep short gamma).
- Time to expiry in trading days/252; ETF proxies (QQQ/DIA/GLD) scaled to index points via the same-day close ratio.
- Expected move (1 day) from ATM IV: S·σ·√(1/252) — the ruler for "how far does a day carry."
Limitations
ETF chains are an incomplete proxy (SPX/NDX index options are missing; institutional complexes like ES futures options likewise). The sign convention is a model; flow data (who buys, who writes) would replace it but costs institutional money. The volume proxy of the third layer does not know the opener's direction. One live day validates mechanics, not profitability — that belongs to the announced validation paper.
Disclaimer: Historical statistics and mechanical analysis are no guarantee of future market behavior. Not investment advice. Trading carries risk of loss up to total loss.