Research — SPX dealer gamma, measured on the whole archive
Measurements on the session archive — each with its sample, its method and how to reproduce it
Learn explains the terms and the mechanics; these pages put one question at a time to the whole archive and report what it says. Every post states the sessions it was measured on, the caveats that bound it, and enough of the method to rebuild it from the public session files. Posts are dated because the archive grows. None of it is a signal, and none of it is investment advice.
How fast do dealers hedge SPX 0DTE flow? A footprint study with a placebo
Every dealer-positioning story assumes dealers hedge what customers force on them. Nobody has published how fast that happens in SPX 0DTE. Measured on 1,088 sessions: the minute's customer delta imbalance predicts the next two to three minutes of index returns and then nothing - and the same magnitudes with randomized signs predict nothing at all, so the signal lives in the signing, not in the volume. About 0.04 basis points per standard deviation: a measurement of the hedging machine, not a trading edge.
What is SPX option flow actually made of? Floor, auctions and the retail slice, 2022-2026
Every print on the SPX tape carries a condition code saying which pipe it came through - electronic book, complex order book, price-improvement auction, legged execution, or the physical trading floor. Tabulated over five years: the floor holds a steady fifth of the near book (matching the one recent academic estimate exactly), the electronic complex book has grown to 45%, and the auction share - which a free exchange truth file shows to be 97.8% customer-initiated and entirely small-lot, i.e. retail - has doubled.
How accurately can a complex option trade be signed? A first grading against exchange truth
Around forty percent of SPX option prints are legs of multi-leg packages, and the classic quote rule is provably invalid on them - an exchange allocates the package's net price across legs by convention. The terminal signs these prints at the package level. Cboe's free trade-by-trade sample carries the true side of every sampled leg, so for once the question has an answer: 80.4% of customer legs signed correctly, 87.9% excluding a category the exchange itself cannot attribute - against 75.1% for the single-leg rule on the same legs. As far as we can find, the first published accuracy number for any complex-order signing scheme.
How much does the volatility model move a gamma terminal's numbers?
Every published GEX construction prices gamma the same way: Black-Scholes from per-contract mid IVs. The literature offers upgrades - fitted smiles, Gram-Charlier corrections for 0DTE - and no paper measures what switching does to the aggregate. Measured on the production book over 14 sessions including two disaster days: the zero-gamma flip carries a 2-3 point model systematic, net gamma at spot 5-13% - and both upgrades make frame-to-frame stability worse. What the flip needed instead was an error bar. It has one now.
How accurate is the SPX expected move? A calibration over 1,088 sessions and 82,000 readings
Every 0DTE dashboard quotes an expected move; nobody has published how often the close actually lands inside it. Measured on 82,000 five-minute readings over 1,088 sessions: 72.4% inside one expected move against a theoretical 68.3% - the quote is conservative by about four points of coverage, uniformly along the curve, at every hour of the session and in every year since 2022. One observation the naive story does not predict: coverage is highest when dealers are short gamma.
Does 0DTE trading raise index volatility? Three published answers, one tape, 1,083 sessions
Three papers give three answers: 0DTE trading raises volatility, does nothing, or dampens it. On 1,083 SPX sessions of the mature regime the raises-volatility association measures +0.6% of mean volatility per sigma of 0DTE share against a published +9%, the share does not lead volatility at all, and the one robust fact points the other way: after a volatile session, 0DTE flow retreats twice as hard as the rest of the book.
Is intraday momentum still alive? Two published effects, retested on 1,085 sessions of the 0DTE era
Gao, Han, Li and Zhou (2018) and Baltussen, Da, Lammers and Martens (2021) both found the last half hour continues the day's direction. On 1,085 SPX sessions of the 0DTE era the unconditional slope is 0.006 with a t of 0.6 — flat in every year. The one exception: on the 15% of days the tape-signed dealer book is short at 15:30, continuation reappears with the right sign in all five years — reported with its caveats, which are real.
Replicating Vilkov (2026) on 0DTE trading rules: exact on his data, a cost line off by two orders of magnitude, and confirmed by the author
Grigory Vilkov's 0DTE trading rules, run twice: his published code on his published panels, and the same rules on our own SPXW tape. The first half reproduces his tables exactly — and shows the half-spread charged at one hundredth of its size, which takes every unconditional structure negative. We sent that to the author before publishing; he confirmed it, reran the private sample behind the paper, and the headline conditional result reverses sign. The 46-session half cannot test anything, which is itself the finding.
Where SPXW 0DTE prints actually land in the quote: a Muravyev & Pearson (2020) replication that does not transfer
Muravyev & Pearson (2020) show that option executions cluster inside the NBBO, so the real cost is well below the quoted spread. On 13,147,178 SPXW 0DTE prints across 38 sessions it does not transfer: 92.56% print exactly on the bid or the ask, the effective half-spread is 0.9619 of the quoted one per print and 1.0697 per contract, and the reason is mechanical — the quote is one tick wide on 71.91% of prints.
Do SPX option returns change sign from day to night? Replicating Muravyev & Ni (2020) on 1,082 sessions
The paper says delta-hedged index options lose 1.0%/day overnight and gain 0.3%/day intraday. On our SPX tape the split is not there: the pooled specification prints one, with the sign reversed, and two diagnostics show that reversal is an artefact of penny wings and the 16:00 closing mark. What does survive is the paper's premise — intraday variance per hour is 5.69 times the overnight rate — and the finding that the quotes already price it.
Does open interest pin the index? Golez & Jackwerth (2012) and Elms (2026) replicated on 1,072 SPX sessions
Two published claims, one question: does the open interest sitting at a strike pull the index toward it, or push its range wider? Replicated on 1,072 SPX sessions with SPX open interest rather than a SPY proxy. The pinning is absent to two thousandths of a point, four of Elms's five tests replicate, and his one significant result reverses sign.
Does a volatility state rescue short 0DTE premium? A rejection on 1,073 sessions
Almeida, Freire and Hizmeri (2025) report an after-cost Sharpe of 0.404-0.470 in a low realised-volatility state against 0.004-0.082 in a high one. Conditioning plain short 0DTE premium on a trailing-RV state over 1,073 SPX sessions gives a Low-minus-High difference between -0.013 and -0.051 per trade in all nine structure-by-cost cells — the wrong sign — and the sample is powered to have seen the claimed effect if it were there.
Which book is the zero gamma? Two feeds, one word, three different levels
Two dealer-gamma services published a zero-gamma level for the same session and the numbers sat 30 points apart. Joining them minute by minute shows the gap is not an error in either: the same word names a call-minus-put open-interest book, a cumulative-volume book and a flow-signed measured book, at different expiry scopes.
Does dealer gamma compress the range? True in points, gone in sigma, 1,066 sessions
The most repeated claim in dealer-positioning commentary, put to the whole archive: read gamma at 10:00 ET and measure the range to the close. In index points the claim holds — 71.8 against 46.3, a 25-point gap. Divided by the volatility the tape was already making it disappears. Low gamma and high volatility are one state.
Is the dealer-gamma sign just the leverage effect? Two books say yes, one says no
Every dealer-gamma screen rests on one sentence: short gamma amplifies moves. It is true in the data, and on the two public conventions it is the leverage effect wearing a gamma label — spot below the flip means the market fell, and falling markets are volatile. The tape-signed book is the one whose sign survives the control.
Seven ways to measure dealer gamma, and only two of them see anything
A companion to the leverage post. One outcome, one sample, one set of controls, and seven measures of the same idea. Four of them see nothing at all; the two that survive are both tape-signed. And the same book on everything EXCEPT the expiring contracts is a null — so whatever this is, it is a 0DTE effect.
How fast does a dealer-gamma level go stale? 1,077 sessions of drift, and what re-reading buys
The terminal recomputes the flip every minute, so the level a reader saw at 10:00 is not the level at 15:00. Measured over 1,077 sessions: the flip drifts a median 50.9 points by the close, its direction relative to spot inverts on 23.5% of days — and a level four hours out of date holds the rest of the session as well as a freshly computed one.
What monthly expiration does to the book: 53 opex sessions against 1,024 others
Monthly expiration is supposed to pin price and gut the book. Measured on 53 opex sessions against 1,024 others, most of that is absent: the point range is unchanged, the hold band is unchanged, the book at 10:00 is unchanged. Two things survive — opex runs about 15% quieter than the volatility the tape was already making, and the contract count falls 3.3% overnight instead of the wipe-out folklore describes.
Does 25-delta skew mark intraday tops and bottoms? 506 sessions, with a matched control
A widely repeated practitioner claim says the 25-delta risk reversal marks near-term tops and bottoms — puts bid means fear means a bottom. Tested on 506 archived sessions, the puts-bid half comes out with the wrong sign and the calls-bid half comes out empty. What survives a control for the move that preceded the signal is an advance that stops, worth 2.8 bp at 30 minutes and uncosted.
Trading the sign of a public gamma feed: 63 sessions, 64 trades, −0.062 R per trade
A widely followed public dealer-gamma feed publishes a 0DTE net-gamma series; a live intraday rule goes long when it crosses above +100k and short when it crosses below −100k. We rebuilt the series from our own per-second book, reproduced the entry logic, and ran it over 63 sessions: 64 trades, −0.062 R per trade, and a randomly timed entry with the same targets and stops does better.
Does price at the cumulative-gamma trough predict a larger move? 63 sessions, 22,680 minutes
A popular dealer-positioning idea says the strike where the running sum of per-strike gamma bottoms is where the next move is largest. Tested on its exact definition over 63 sessions: −0.08 pt on the forward 30-minute range, t −0.06. The pilot that motivated the test was six sessions — and they were the six most recent in the archive.
What a 0DTE straddle round trip costs, and what actually takes the edge
A round trip in the ATM 0DTE SPX straddle, priced at the honest touch on 61 sessions: $34.72 to $36.30 per 1-lot, 7.8-21.8% of the average move, 1.5% of premium. The forecast the trade is built on is strong and the spread is small — and the trade still dies, on a leg 3.5 times larger that a delta hedge cannot remove.
Does the dealer-gamma state rescue short 0DTE premium? A null on 34 sessions
The most common retail use of a dealer-gamma dashboard is to time short 0DTE premium: sell when dealers are long gamma and the tape will pin. Tested on 34 sessions across five structures, three entry times and every threshold, the gamma state does not beat selling unconditionally on the same days and does not beat a random selection of the same size. What the sample cannot settle is the absolute level of short 0DTE premium, and the confidence interval says so.
What the 0DTE literature actually claims: 25 papers, sorted by whether a tick tape can test them
A map, not a measurement: 25 published papers on US index options and intraday index dynamics, each with its central claim in one clause and a verdict on whether it can be checked against a tick SPXW tape with a signed dealer book. Ten clear the bar, ten clear it partly, five do not — and four are behind a paywall that refused every automated fetch, so their numbers are flagged as unverified.
How much of the SPX options book is new each day? Open-interest change across 1,081 sessions
Per-contract open interest and volume, every SPX expiry within a month, 1,081 sessions: the book grows by ~40% of each day's volume, far expiries are open-and-hold while near ones churn, and on its own expiry day 0DTE trades about six times its opening open interest — eight times in 2026. What an OI-based dealer book can and cannot see.
How much of the SPX options tape is dealer-to-dealer? A free daily floor from OCC data
OCC publishes who traded each day (customer, firm, market maker) as sides, not trades. One line of algebra turns that into a lower bound on the share of SPXW trades with a market maker on both sides: positive on 95% of days, median 5.3%, rising year over year. A free daily noise floor for any tape-based dealer book.
How far does SPX travel before the close? An empirical touch surface over 1,086 sessions
The probability that SPX reaches a level a given number of sigmas away with a given number of minutes left — measured on 1,086 sessions of minute data, no model, no lookahead. A baseline for judging any intraday level.
How often do SPX dealer-gamma levels hold? 1,086 sessions, read at 10:00, no lookahead
Hold rate, break-at-close, comeback and overshoot for the zero-gamma flip, the hold band and the prior day's range, each read at 10:00 ET and judged against the rest of the session — and how much of the holding is just distance.
Does dealer state move the touch probability? A pre-registered test of four hypotheses on 987 sessions
Four hypotheses about the hold band, the gamma regime, dealer delta and 25-delta skew, written down before any conditional number was computed, scored against the empirical touch surface on 987 development sessions with a 95-session holdout. Three are null; the fourth is real, robust — and the opposite sign of what was registered.