Pre-registered statistical studies of the forces that mechanically move markets — dealer hedging, options flow, volatility pricing. Each study states its hypothesis, its trigger, and the result that would kill it, before it runs. Validated or refuted, the outcome is published either way.
Research corpus801 sessions captured · 787 research-grade · 1 study validated— 1 validated so far; the runner publishes verdicts as the corpus grows
Evidence, before we test it
79 of these are popular market-structure claims with weak public evidence — narratives our competitors sell as fact. We expect the data to refute many of them, and we’ll publish those refutations as prominently as the survivors.
191 of 195 studies require Terminal + AI. Featured studies are free to read in full; Terminal + AI members get the complete library — methodology, statistics, and every future result.
Options dealers who are net short gamma must hedge in the **direction** of the move — sell into weakness, buy into strength — which mechanically **amplifies** realized range. Dealers net long gamma hedge **against** the move (buy weakness, sell strength), **damping** it. Under the naive dealer convention (dealers long calls, short puts) the sign of aggregate gamma exposure (GEX) at a given spot tells you which regime the tape is in.
Sessions whose 9:35am net GEX sits in the bottom decile of the trailing 60 sessions realize a wider RTH range (range_pct and range_vs_em) than top-decile sessions, and both tails differ from the all-session baseline.
Dealers short gamma must hedge in the direction of the move (sell into weakness, buy into strength), amplifying realized range; dealers long gamma hedge against the move, damping it. This is the least deniable mechanical claim in index options — and the premise of the entire product.
Prior evidence
Barbon & Buraschi document gamma-driven fragility in equity markets; SqueezeMetrics' GEX whitepaper popularized the dealer-gamma/realized-vol relationship; the damping/amplifying asymmetry is consistent across practitioner research. Strongest prior in the catalog. Our test adds: OUR construction (naive convention, Black-76 on forward, specs/03), OUR symbols, intraday horizons, honest baselines.
What would kill it
Top- and bottom-decile range_pct distributions are statistically indistinguishable (95% CI on median difference includes 0) in both split halves of the evaluation window.
Status Validated — open the paper for the complete publication record. Current statistics are available only through the governed engine result identity; this index embeds no frozen result statistic.
Consecutive non-overlapping post-09:35 30-minute SPX returns agree in sign more often in negative-net-GEX sessions than in positive-net-GEX sessions, with negative gamma above and positive gamma below the unconditional agreement rate.
After spot's first downward cross of the gamma flip in a session (≥0.05% beyond the level), forward 30–120 min drift is more negative and range wider than the time-matched baseline; upward reclaims show the mirrored (damping-side) profile. Down-cross and reclaim are tracked as separate result rows.
The signed distance between SPX and the authenticated repriced aggregate zero-gamma level at the opening decision time adds out-of-sample information about remaining-session realized variance beyond net-gamma sign and magnitude, opening volatility, implied volatility, recent realized volatility, calendar, and price-path controls.
Crossing the estimated gamma-flip level intraday produces no abnormal same-hour drift or range beyond what the CONTINUOUS net-GEX measure already explains — i.e. the threshold is not informative once the level is known.
The predictive power of end-of-day net GEX for next-session realized volatility is approximately zero after controlling for VIX and ATM implied vol — most of the raw GEX-vol relationship is the vol regime itself.
The strength of the net-GEX -> intraday-range relation is monotonically greater in QQQ than in SPX/SPY, because the mechanical hedging impact scales inversely with underlying liquidity.
An intraday touch of the vendor-style 'call wall' (largest positive-gamma call strike above spot) is followed by rejection (negative forward 30-min return) more often than a distance-matched baseline — and this does NOT survive controlling for distance-to-VWAP and net GEX.
An intraday touch of the 'put wall' (largest put-gamma strike below spot) is followed by a bounce (positive forward 30-min return) more often than a distance-matched baseline, and does not survive VWAP/GEX controls.
The straddle expected move overprices realized range MORE in positive-gamma regimes — realized volatility is suppressed below what the market paid for — so positive-gamma sessions show a lower range-to-EM ratio and a higher expected-move band hit rate than negative-gamma sessions.
Conditional on spot being a given distance from the largest absolute-gamma strike ('gravity' strike) at midday, price drifts TOWARD that strike by the close more often than chance.
On monthly OPEX days, last-hour SPY realized range is LOWER when a single near-money strike holds a large share of total near-money gamma than when gamma is dispersed.
On days when 0DTE volume is an unusually large share of total SPX option volume, the naive-convention GEX loses next-session range-predictive power relative to its own historical baseline.
Dealer gamma whose SIGN is inferred from aggressor-classified signed option flow predicts next-session intraday vol better than gamma signed by the naive long-call/short-put convention on raw OI.
The intraday return autocorrelation gap between negative-gamma and positive-gamma sessions is larger in high-VIX regimes than in low-VIX regimes -- i.e. dealer-hedging amplification is state-dependent, strongest when both gamma is negative and vol is elevated.
A call/put wall is respected (price reverses on approach) more often when the gamma at that strike is highly CONCENTRATED (top-decile share of total absolute gamma) than when the same wall is diffuse -- concentration, not the wall label alone, carries the hedging force.
A large intraday DECLINE in net GEX from the open to midday (top-decile negative dGEX) predicts a wider final-two-hours realized range than the session-open GEX level alone implies -- the RATE of gamma erosion adds forecast power over the level.
Price does NOT close nearer to the call/put wall than to a random same-distance placebo strike more often than chance -- the popular "gamma wall is a magnet" claim does not survive a placebo control. (Pre-registered skeptical/refutation study.)
Crossing DOWN through the gamma flip (positive->negative regime) produces stronger same-direction follow-through than crossing UP (negative->positive), an asymmetry: the amplifying transition is more actionable than the stabilizing one.
The negative-gamma range-amplification effect is concentrated in the final trading hour, when 0DTE gamma density peaks, rather than spread evenly across the session.
Intraday price does NOT gravitate toward the daily gamma-flip level more than toward a matched placebo level; the popular 'price is drawn to the flip' claim fails for the index.
A strike singled out by raw open-interest concentration but LOW gamma does not act as intraday support/resistance beyond a matched placebo; 'big OI = brick wall' has no content once gamma is accounted for.
The index does NOT migrate toward the max-pain strike into expiry more than a matched placebo, and if anything ANTI-pins (moves away) before index-option expiration.
In positive-gamma sessions, conditional on spot sitting a non-trivial distance from the |GEX|-weighted strike centroid at midday, price drifts TOWARD that centroid by the close more than a distance-matched placebo level does.
In positive-gamma sessions, conditional on spot sitting a non-trivial distance from the positive-GEX-weighted strike centroid (long-gamma strikes only) at midday, price drifts TOWARD that centroid by the close more than a distance-matched placebo — and more than the incumbent |GEX| centroid does.
In positive-gamma sessions, conditional on spot sitting a non-trivial distance from the single largest positive-GEX strike at midday, price drifts TOWARD that strike by the close more than a distance-matched placebo.
In positive-gamma sessions, conditional on spot sitting a non-trivial distance from the strike of maximum ABSOLUTE GEX at midday, price drifts TOWARD that strike by the close more than a distance-matched placebo.
In positive-gamma sessions, conditional on spot sitting a non-trivial distance from the OI max-pain strike at midday, price drifts TOWARD max-pain by the close more than a distance-matched placebo.
In positive-gamma sessions, conditional on spot sitting a non-trivial distance from the positive-GEX centroid restricted to the [put_wall, call_wall] corridor at midday, price drifts TOWARD that centroid by the close more than a distance-matched placebo.
In positive-gamma sessions, conditional on spot sitting a non-trivial distance from the distance-decayed |GEX| centroid (each strike weighted by exp(-|K-spot|/EM)) at midday, price drifts TOWARD that level by the close more than a distance-matched placebo.
In positive-gamma sessions, conditional on spot sitting a non-trivial distance from the spot-ladder zero (the hypothetical spot at which aggregate dealer gamma reprices to zero) at midday, price drifts TOWARD that level by the close more than a distance-matched placebo.
Sessions whose 9:35am FLOW-IMPLIED net GEX (Estimate B) sits in the bottom decile of the trailing 60 sessions realize a wider RTH range (range_pct and range_vs_em) than top-decile sessions, and both tails differ from the all-session baseline — the exact GEX-01 claim, but read from Estimate B instead of the shipped naive-OI Estimate A.
On the SAME sessions and the SAME validated GEX-01 endpoints (9:35 net-GEX tail -> realized RTH range, same-day; and the net-day range robustness arm), the flow-implied estimate (B) classifies the realized-range vol-state AT LEAST AS WELL AS the naive-OI estimate (A), with the pre-registered directional prior that B is BETTER. Measured by a PAIRED per-session difference in classification skill (delta_skill = skill_B - skill_A), block-bootstrapped by session. Two-sided and falsifiable both ways: the null is delta_skill = 0 (B no better than A); delta_skill < 0 means B is worse.
Partition sessions by the joint sign of the two 9:35 estimates into four cells: agree-positive (A>0,B>0), agree-negative (A<0,B<0), and the two DISAGREEMENT cells (A>0,B<0) and (A<0,B>0). In the disagreement cells, same-day intraday realized variance tracks the FLOW-IMPLIED sign (B) more than the naive-OI sign (A): B-negative disagreement sessions realize WIDER intraday variance than B-positive disagreement sessions, by more than A's sign orders the same sessions. This is the direct test of whether the fresher estimate governs 0DTE hedging when OI and flow point opposite ways.
The divergence is WP-2's precomputed gamma_map.flow_implied.divergence.net_gex_divergence (= B - A = net_gex_flow_implied - net_gex). To make it cross-session comparable the registered series is its trailing-60-session z-score, div_t = z(net_gex_divergence_t), point-in-time (the panel-native net_gex_divergence_frac = (B-A)/|A| is the pre-registered robustness alternative). Sessions whose 9:35 |div| is in the top decile of the trailing 60 sessions — flow-implied positioning has moved far ahead of open interest — realize (i) a WIDER forward RTH range than the day-of-week/time-of-day matched baseline, and (ii) a forward drift SIGNED toward B's implied regime relative to A (div>0, i.e. B more long-gamma than A, damps; div<0 amplifies). Falsifiable on both the range and drift legs.
Adding the exact 09:35 net-GEX state to a strictly prior, expected-move-aware forecast improves untouched prediction of the remaining-session realized range beyond the same forecast without GEX; otherwise GEX-01 is a volatility-state description already priced by the option market.
GEX-01 (validated) establishes that net-GEX sign predicts realized range in absolute terms; VRP-01 establishes that the 0DTE straddle EM already prices most of that range. The open question that decides monetizability: does the straddle SCALE for the gamma regime (range ÷ EM ≈ 1 on both signs — the signal is real for vol but not an options edge) or does negative-gamma range OUTRUN the implied move (range ÷ EM > 1 — a tradeable surprise)? The load-bearing separation is the atm_iv control: EM is a monotone transform of atm_iv, so a real gamma effect must survive residualizing range ÷ EM on atm_iv, distinguishing "gamma predicts range beyond IV" from "gamma IS IV".
Prior evidence
(b) mechanically plausible, unproven. GEX-01 is validated in raw bps; VRP-01 is thin (native premium +6.3 bps, p=0.078). No published test of whether the dealer-gamma range premium survives EM-normalization at our granularity — this study is the graph bridge between the two mechanisms.
What would kill it
Refute incremental GEX information if the untouched paired forecast-skill lower bound is not positive after the declared family correction, if either time-shifted or matched-random GEX placebo performs as well, or if the sign is not stable across powered eras. Mark inconclusive, never refuted, when authenticated point-in-time inputs or paired support are inadequate. This leaves GEX-01's absolute same-day range classification untouched.
Status Inconclusive — open the paper for the complete publication record. Current statistics are available only through the governed engine result identity; this index embeds no frozen result statistic.
On the SAME sessions and the SAME validated GEX-01 endpoint (9:35 net-GEX decile tail -> realized RTH range, same-day), the 0DTE FLOW-IMPLIED estimate (B) classifies the realized-range vol-state AT LEAST AS WELL AS the shipped naive-OI estimate (A), with the pre-registered directional prior that B is BETTER. Measured by a PAIRED per-session difference in tail-separation skill (delta_skill = skill_B - skill_A), block-bootstrapped by session. Two-sided and falsifiable both ways: the null is delta_skill = 0 (B no better than A); delta_skill < 0 means B is worse.
Across the admitted 0DTE sessions, the signed 0DTE FLOW-IMPLIED net GEX (B) rank-orders the session's realized range (range_vs_em) AT LEAST AS WELL AS the shipped naive-OI net GEX (A), with the directional prior that B is BETTER. Measured by a PAIRED per-session difference in Spearman rank correlation with realized range: delta_rho = rho_B - rho_A, block-bootstrapped by session. Two-sided and falsifiable both ways: null is delta_rho = 0 (B no better than A); delta_rho < 0 means B orders realized range worse than the shipped OI read.
At a regular-session five-minute decision time, the contemporaneous signed gamma-ladder topology improves calibrated 5/15/30/60-minute and close classification of one-way drift, two-leg reversal, and chop beyond price, volatility, liquidity, scalar net-GEX, and static-ladder controls.
Frozen dynamic gamma-wall state may add stable calibrated out-of-sample skill for the mutually exclusive first outcome after an SPX wall approach—clean target hit, rejection, break-and-hold, or dwell/chop—beyond distance, extremity, approach, volatility, time, expiry, flow, round-number, and matched-placebo controls at 15, 30, and 60 minutes.
Persistent coherent residual D1 vectors may add stable calibrated out-of-sample information for SPX return, first passage, excursion, and path state at 5/15/30/60 minutes and close beyond momentum, mechanical repricing, current gamma shape, distance, volatility, flow, and placebos.
Residual temporal acceleration, spatial curvature, and mixed strike-time fronts may add stable calibrated out-of-sample information for SPX direction, first passage, excursion, and path regime beyond the complete frozen GEX-42 D1 model, price/volatility controls, and placebos.
Dynamic per-strike concentration, persistence, and change may add stable calibrated out-of-sample information for separate hit, approach, dwell, rejection, pass-through, and chop risk sets beyond distance, price/volatility context, static concentration/walls, and matched placebo strikes.
Validated opening gamma-ladder geometry may select a one-lot SPXW 0DTE put credit vertical, call credit vertical, or abstention with incremental calibrated predictive skill and positive net untouched-holdout expectancy beyond option-implied, volatility, price, liquidity, scalar-GEX, calendar/event, and matched-placebo controls; attraction, barrier, and null maps compete equally.
Continuous call-versus-put wall GEX imbalance may change fade, continuation, or unresolved path probabilities after a first upward SPX call-wall approach, with regime-dependent effects that must add stable untouched-holdout value beyond volatility, price-path, distance, liquidity, scalar-GEX, calendar, and matched-placebo controls; fade, continuation, and null have equal prior standing.
A strike with unusually positive point-in-time gamma-magnitude flux may contain incremental 15-minute information about SPX movement toward that frozen strike or about subsequent realized path magnitude after ordinary momentum, volatility, distance, static gamma, and regime controls; structural repricing and flow-implied positioning are separate arms and the null has equal standing.
After SPX accepts beyond the frozen repriced aggregate zero-gamma level, 5/15/30-minute continuation and realized path magnitude may exceed matched price-only breakout baselines; any incremental effect may be stronger when model zero lies near the contemporaneous local per-strike GEX sign seam and when the repriced curve is steep, stable, and confirmed by the flow-implied estimate.
Mechanism, trigger definition, falsification criteria and all future results are visible to Terminal + AI members. View publication →
II.Vanna & charm8 studies
Mechanism Plausible — unproven
**Vanna** (∂delta/∂vol) and **charm** (∂delta/∂time) describe how a dealer's delta hedge drifts even when spot is still — as implied vol changes (vanna) or simply as time passes (charm). A dealer holding a hedged book must trade to stay neutral as these second-order greeks move the delta, producing flow that is a function of the vol path and the clock rather than of spot.
In positive-gamma sessions, conditional on spot sitting a non-trivial distance from the |charm|-weighted strike centroid at 15:00 ET, price drifts TOWARD that centroid by the close more than a distance-matched placebo. Evaluated in the FINAL HOUR only (charm decay is a close effect).
SPX average return over the 5 sessions into monthly OPEX is NOT reliably positive after controlling for the equity risk premium and the contemporaneous VIX change.
The week AFTER monthly/quarterly OPEX shows higher realized volatility than the week before, conditional on a large amount of near-money gamma having rolled off at expiration.
On days with a large VIX DECLINE (top-decile), positive-vanna dealer hedging adds upward drift BEYOND the mechanical price-vol comovement -- i.e. after residualizing on the size of the VIX move itself, a vanna-attributable drift remains.
On vol-crush days there is NO residual upward directional drift attributable to vanna once the drift is residualized on the VIX move itself; the 'vanna rally' is price-vol comovement, not a separate forward edge.
Mechanism, trigger definition, falsification criteria and all future results are visible to Terminal + AI members. View publication →
III.Options flow15 studies
Mechanism Plausible — unproven
When customers aggress options (lift offers / hit bids), the dealer takes the other side and is left with a position to hedge. Net options flow (NOF) — aggressor-signed, delta-adjusted premium — is our measure of that customer intent. Two distinct channels can make it predictive:
After an onset of unusually large 15-minute all-tracked-expiry SPX net put premium sold, 60-minute SPX drift exceeds a same-time-of-day and strictly-prior opening-volatility-matched non-trigger baseline by at least 2.5 basis points on the untouched evaluation cohort.
Aggressor-classified put selling may reveal informed beliefs, volatility supply, or willingness to absorb downside insurance. Dealer hedging is not mechanically one-way: the initial hedge and subsequent hedge path depend on inventory, moneyness, gamma, spot, volatility, and time. FLOW-01 therefore tests whether the flow contains incremental information; it does not assume automatic future dealer buying or treat a large print as directional proof.
Prior evidence
Pan and Poteshman (2006) report predictive information in option volume for future stock prices, and Easley, O'Hara and Srinivas (1998) develop and test an informed option-volume channel. Those studies support a broad information hypothesis, not this exact intraday SPX premium-percentile trigger. The earlier NOF S9 exercise had only about 38 research-grade sessions and explicitly classified every output as pipeline validation rather than a finding.
What would kill it
Refute the directional claim only when the evaluation is powered, the session-block 95% interval for primary 60-minute median excess excludes zero on the wrong side or fails the positive 2.5-bp economic floor, either chronological evaluation half is nonpositive, or the circular-time-shift placebo reproduces the effect. An interval including zero, fewer than 126 authenticated evaluation sessions, fewer than 20 complete evaluation weeks, or inadequate event support is inconclusive rather than refuted. Any identity or chronology failure blocks execution and never becomes a market finding.
Awaiting live data — no result yet. Will measure drift_pct at 30 min · 60 min vs a regime-matched baseline, and publish validated or refuted either way. · a data source for this study is not yet captured.
15-minute windows in which net call premium BOUGHT (aggressor-classified) exceeds the 95th percentile of the trailing 60 sessions are followed by positive excess drift at 30/60/120 minutes relative to the time-of-day- and regime-matched unconditional baseline.
Top-decile 5-minute delta-signed SPY option flow (HIRO/NOF-style) predicts positive SPY returns over the next 5-15 minutes, before decaying and reversing.
The predictive content of delta-signed flow is concentrated in the first 30-60 minutes after a flow spike and reverses by end of day (a microstructure impact-then-reversion signature).
Block/sweep-tagged bullish option prints ('unusual options activity') do NOT predict positive next-day SPY returns after controlling for contemporaneous signed flow.
15-min windows where net put premium BOUGHT (aggressor-classified) exceeds the 95th percentile of the trailing 60 sessions are NOT followed by negative excess drift beyond the matched baseline - i.e. 'protection buying' does not predict weakness.
15-min windows where net call premium SOLD (aggressor) exceeds the 95th percentile of trailing 60 sessions are followed by negative excess drift at 30/60/120 min vs the matched baseline.
Signed dealer-hedging delta flow at window t does NOT predict forward returns at t+30/60 min beyond the CONTEMPORANEOUS flow-return relationship once VIX/regime-controlled - i.e. HIRO does not lead price as claimed.
Windows where |signed net options flow| reaches the >=97.5th percentile of trailing 60 sessions are followed by REVERSAL (mean-reverting excess return opposite the flow) at 30/60 min, not continuation.
A dealer-gamma estimate built from MEASURED trade sign (Lee-Ready on the NBBO-at-trade options tape) predicts next-hour realized volatility better than the standard sign-ASSUMED (long-call/short-put) GEX, especially when the two disagree.
A day's SIGNED options net-delta flow (aggressor-classified from the tape) predicts the next session's drift beyond what the public put/call ratio or unsigned volume predicts.
Extreme 15:30–15:50 ET aggressor-signed option delta flow with upper-quartile, strictly-prior-normalized price nonresponse changes the flow-signed next-session return relative to high-flow/responded, price-matched normal-flow, and ordinary matched sessions after price, volatility, gamma, liquidity, calendar, and event controls; the two-sided sign distinguishes continuation from reversal.
Historically unusual gross call-buy, call-sell, put-buy, or put-sell flow may precede aligned, reversed, or null SPX paths, and event-time dealer gamma may amplify or damp that transfer; any claim must add stable untouched-holdout value beyond pre-event price, volatility, liquidity, calendar, scope, quality, and circular-shift controls.
Confirmed absorption-response-v1 events within 0.10 fixed day-EM of a canonical structural level may have a different two-sided 15-minute flow-signed return and rejection/breach path than matched non-event times near the same level type and the same events at least 0.25 EM from every canonical level, after volatility, gamma, momentum, intensity, calendar, and quality controls.
Mechanism, trigger definition, falsification criteria and all future results are visible to Terminal + AI members. View publication →
IV.Vol risk premium19 studies
Mechanism Established literature
Implied volatility systematically exceeds subsequently realized volatility — sellers of options are paid a premium for bearing variance risk. The volatility risk premium (VRP = implied − realized) is one of the most robust, best-documented effects in derivatives, and it has a direct product form: the day's expected move (EM), priced from the ATM straddle, is on average *wider* than the range the market actually travels.
On an untouched chronological SPX evaluation window, the prior-close fixed straddle expected move exceeds the subsequent session's absolute close move by a positive median fraction of spot, and that positive premium is sign-stable across both evaluation halves.
Systematic demand for index-option protection can keep the one-session straddle-implied move above the move subsequently delivered. That spread compensates option sellers for crash and volatility-of-volatility risk; it is not itself a trading return or an option-selling claim.
Prior evidence
The index variance-risk premium is established in the literature, but the exact Ivolution construction is not: a D-1 close SPXW straddle, D-1 underlying anchor, and next-session close or full-range outcome differ from standard variance-swap and longer-tenor evidence. The exact effect, calibration, and regime stability therefore remain unproven.
What would kill it
With adequate authenticated support, refute the overpricing claim if the untouched evaluation interval for median em_minus_abs_close_bps includes zero or either chronological evaluation half has a nonpositive median. Conditioning cells may restrict the scope of a surviving aggregate claim but cannot rescue a failed primary or overturn it merely because a cell is underpowered.
Awaiting live data — no result yet. Will measure em_minus_abs_close_bps, closed_within_em_rate, range_vs_em at close vs a regime-matched baseline, and publish validated or refuted either way. · a data source for this study is not yet captured.
When the variance risk premium (VIX-squared minus trailing realized variance) is in its top quintile, SPX 1-3 month forward excess returns exceed the bottom-quintile baseline.
After the prior completed session's VX front slope ranks in its bottom decile, SPX realizes a wider 1–5 session range than its matched baseline; after a top-decile slope, SPX realizes a narrower range, conditional on the prior VIX-level tercile.
When the VIX-futures curve is in steep contango (front-to-second-month roll in the top decile), forward returns to being long SPX / short front vol are positive over the following month, with a fat left tail when the curve inverts.
Steepening front-end put skew (25-delta put IV minus ATM IV rising) predicts lower forward SPX returns / higher realized downside over the next 1-4 weeks.
The IV-implied 1-day expected move is well-calibrated: realized absolute move is within the expected-move band on approximately 68% of sessions (a calibration test, not an edge).
Conditional on estimated dealer 0DTE gamma being NEGATIVE (rare), intraday SPX range is HIGHER than baseline — isolating the tail/amplification channel that coexists with the average-dampening of VRP-07.
Sessions where the IV-RV variance-risk-premium sits in the top quintile of trailing 60 sessions are followed by forward realized range UNDERSHOOTING the implied by more than the matched baseline over the next 1/3/5 sessions.
Sessions in the top quintile of vol-of-vol (dispersion of intraday ATM-IV changes) over trailing 20 sessions are followed by larger forward realized range than the matched baseline over 1/3 sessions.
Sessions where the 25-delta put skew STEEPENS by >=80th pctl of its trailing 60-session change distribution show greater forward downside asymmetry than the matched baseline over 1/3 sessions.
When the variance risk premium is in its top tercile, intraday touches of the ±EM boundary revert toward the session open at a materially higher rate than the matched baseline.
High early intraday EM-utilization predicts rest-of-day continuation in negative-gamma regimes and exhaustion/reversion in positive-gamma regimes; unconditionally it predicts neither.
A high CBOE-SKEW-style index skew reading does NOT predict lower forward SPX returns or higher drawdown probability beyond the VIX level; index SKEW is not a crash timer.
Conditioning on high IV-rank adds NO forward-vol/return predictive power beyond the variance risk premium itself; the 'IVR>50 → sell premium' rule is repackaged VRP, not an independent signal.
A calibrated opening model can improve the SPX official-close distribution over inside/outside-up/outside-down versus strictly prior option-implied, volatility, price, liquidity, calendar, and event baselines; prior outside streaks may mean-revert, persist, or add no information, and gamma/news contributions must survive nested ablation before attribution.
At least one preregistered opening-known condition may identify separately positive net hold-to-settlement expectancy in a one-lot SPXW 0DTE long straddle or debit iron condor after executable natural fills, slippage, fees, tail risk, and capacity; gamma must add stable untouched- holdout value beyond option-implied, volatility, price, liquidity, calendar, and placebo controls.
The exact replicated ES MST-48 physical-volatility forecast available at 10:00 ET improves out-of-sample prediction of SPX 10:00–15:59 realized variance by more than one percent MSE after a strictly prior mapping from the contemporaneous SPX 0DTE ATM straddle is already included.
Mechanism, trigger definition, falsification criteria and all future results are visible to Terminal + AI members. View publication →
V.Dark pool5 studies
Mechanism Narrative — weak evidence
Off-exchange (TRF-reported) prints and dark-pool volume are read by many traders as institutional footprints — volume shelves that mark accumulation/distribution, or an aggregate off-exchange buy pressure (DIX-style) that leads price. We compute derived aggregations only (binned VWAP shelves, significant-print flags) — never a raw tape replica (licensing + compliance, rule 3).
Mechanism, trigger definition, falsification criteria and all future results are visible to Terminal + AI members. View publication →
VI.Microstructure63 studies
Mechanism Plausible — unproven
Intraday price is shaped by mechanical market events: the closing auction (MOC) and its published imbalance, the opening-range formation, and reversion to session VWAP. These are structural features of how the session is organized rather than dealer-positioning effects, and they can interact with the gamma regime (an imbalance into a short-gamma close hits a market with no damping).
The exact same-contract ES return from the prior RTH close through 10:00 predicts the final half-hour return; its strictly prior 20th/80th-percentile tails earn positive after-cost P&L in both directions and add information beyond opening-only and stale-signal controls.
Deviations of SPY above VWAP predict modest NEGATIVE forward 15-30 min returns (short-side mean reversion), while opening-range-breakout LONG signals show no significant forward edge.
A strong first-30-minute directional return (top/bottom quintile of trailing 60 sessions) continues into the session in NEGATIVE-net-gamma regimes but NOT in positive-net-gamma regimes - testing whether the weak opening-range/momentum signal is rescued by gamma conditioning.
Sessions where the published market-on-close imbalance (3:50 ET) is in the top/bottom decile of trailing 60 sessions show same-direction excess drift into the 4:00 close beyond the matched baseline.
The sign of the rest-of-day return (after the first half hour) predicts the sign of the last-30-minute return, and the effect concentrates in negative-gamma regimes and on high-vol days.
Breakouts of a narrow initial balance (first-hour range) continue at a higher rate than the matched baseline, and the edge concentrates in negative-gamma regimes; a symmetric SHORT leg is included so the test cannot be inflated by market drift.
A completed ES RTH profile whose top or bottom two ticks each contain at least two distinct 30-minute TPO periods is revisited in the next complete RTH session more often than both an equidistant mirror and a distance-standardized completed/excess extreme, and a next-open trade toward that poor level is profitable after adverse fills and fees in both directions.
Touches of a session anchor level (an options-notional or time-weighted analogue of VWAP, since true share-VWAP needs volume we lack) revert no more than touches of a fixed random anchor; VWAP/anchored-level support is folklore once placebo-controlled.
The exact 0.618-0.65 retracement zone of a causally completed intraday ES leg produces a ratio-specific and executable continuation bounce, separately for long and short legs, beyond 100 same-width non-Fibonacci retracement zones.
In the 09:33-09:53 ET window the DJIA (DIA) does NOT lead SPY/QQQ/NVDA at any human-tradeable latency: prior-minute DIA returns carry no material, direction-asymmetric predictive content for the laggard's next-minute return, and the as-narrated V-bounce rule earns no positive net return.
Sessions closing in the bottom decile of internal bar strength (IBS = (close−low)/(high−low), i.e. a close near the day's low) are followed by a higher next-day close-to-close return than the unconditional next-day baseline.
A negative close-to-close session (down day) is followed by a higher next-day close-to-close return than the unconditional baseline — the index exhibits short-horizon (daily) return reversal.
The sign of the first-30-minute return predicts the last-30-minute (into-close) return in the same direction — an intraday momentum continuation the sign-following payoff captures above zero.
ES sessions whose absolute same-contract overnight gap reaches the strictly prior 252-eligible-session 90th percentile continue in the gap direction from the 09:30 open through the 10:29 bar close after a 3-bp round-turn cost. NQ is a frozen cross-index replication.
Directly fading a Surge displacement candle with a body of at least 25 NQ points at a symmetric 10-point bracket produces positive net expectancy after costs and improves on a matched non-Surge large-candle baseline.
Directly fading a BODY-25 Surge event with a 10-point NQ target and 20-point stop produces positive cost-adjusted expectancy and improves on matched non-Surge large-candle backfill.
When NQ opens with an absolute same-contract gap at or above the strictly prior 252-eligible-session 90th percentile, fading the gap from the exact 09:30 open through the 10:29 bar close produces a positive return after a 3-bp round-trip cost. ES is the frozen falsification instrument.
Within the settled percent-scaled NQ Surge displacement pool, above-median signal-bar volume or below-median volume per percent-range, measured against strictly prior 60-session distributions, separates +1.52R bracket outcomes from their complements and same-side random controls.
The exact MST-19 above-median signal-volume cell transfers from NQ to ES under the unchanged percent-scaled Surge rule and +1.52R bracket, establishing a portable entry-specific effect rather than an instrument-local or generic tape-state association.
A recent same-direction single-use strength-one break of structure improves +1.52R resolution inside the exact percent-scaled Surge displacement pool and transfers from NQ to ES.
The exact MST-19 above-median signal-volume cell retains a positive selected-minus-comparator +1.52R win-rate delta when entry is moved from the completed signal close to the next consecutive one-minute bar's open and remains positive under one- and two-tick adverse-entry stresses.
After an unusually narrow ES first-hour initial balance, the first completed close breakout predicts positive direction-signed return from the next consecutive one-minute open to the RTH close after fixed adverse execution and fees.
An ES close unusually far above its strictly prior session-to-date VWAP proxy predicts positive net short return and positive paired excess from the next consecutive one-minute open through 30 minutes.
After a causal same-clock five-minute ES price shock, abnormal volume adds a nonzero executable 30-minute directional effect versus price-shock controls, and the ES-discovered continuation or reversal sign replicates unchanged on NQ.
The first causal ES one-tick breach of an immediately prior same-contract RTH extreme that closes back inside the prior range predicts positive net 30-minute inward-fade return, and the identical rule replicates unchanged on NQ.
The first decisive ES regular-hours break of the fully completed 18:00-09:29 ET overnight range carries positive net direction-following return over the next 60 minutes, with untouched NQ replication required before candidate status.
The sign of the same-contract ES return from the prior 15:59 ET close to the current 09:30 open predicts positive net direction-following return during 15:30-16:00 in the post-publication period, with untouched NQ transfer required before candidate status.
An ES session whose 09:30-10:00 realized volatility is at or above its strictly trailing 252-session 75th percentile has a remaining-session high-volatility target rate more than ten percentage points above the unconditional threshold-ready rate, with untouched NQ replication required before candidate status.
Among ES sessions already in the strictly trailing high-opening-RV state, downside-dominated opening semivariance raises the remaining-session high-volatility target rate by more than eight percentage points versus upside-dominated opening semivariance, survives total-RV matching, and transfers unchanged to NQ.
Among sessions already in the strictly trailing high-opening-RV state, openings whose six five-minute squared returns are unusually concentrated in one interval have a remaining-session high-volatility target rate at least eight percentage points below other high-opening-RV sessions, survive total-RV matching, and transfer unchanged from ES to sealed NQ.
At 10:00 ET, sessions whose 09:30-09:59 front-contract volume is unusually high have a higher 10:00-15:59 high-realized-volatility target rate than other sessions after outcome-blind matching on causal opening realized volatility, and the effect transfers unchanged from ES to sealed NQ.
Among literature-exact prior-close-through-09:59 composite-tail sessions, opening volume at or above its strictly prior 252-session 75th percentile identifies more profitable 15:30-open to 15:59-close continuation than other tail sessions after outcome-blind matching on composite-tail surprise, and the interaction transfers unchanged from ES to sealed NQ.
Among literature-exact prior-close-through-09:59 composite-tail sessions, causal 09:30-09:59 realized volatility at or above its strictly prior 252-session 75th percentile identifies more profitable 15:30-open to 15:59-close continuation than other tail sessions after outcome-blind matching on composite-tail surprise, and transfers unchanged from ES to sealed NQ.
Among sessions already in the strictly trailing high-opening-RV state, openings whose finite-sample-normalized realized bipower variation is unusually low relative to opening realized variation have a remaining-session high-volatility target rate at least eight percentage points below other high-opening-RV sessions, survive total-RV matching, and transfer unchanged from ES to sealed NQ.
Among sessions already in the strictly trailing high-opening-RV state, sessions whose exact 18:00-09:29 overnight realized volatility is also unusually high have a remaining-session high-volatility target rate at least eight percentage points above other high-opening-RV sessions, survive opening-RV matching, and transfer unchanged from ES to sealed NQ.
At 10:00 ET, a rolling NQ forecast using exact five-minute overnight and opening realized volatility reduces out-of-sample remaining-session log-variance forecast loss by at least one percent relative to the otherwise identical exact-opening-RV-only model.
MST-37's unchanged opening-plus-overnight-RV challenger reduces ES one-step-ahead remaining-session log-variance forecast loss relative to the unchanged exact-opening-RV-only baseline.
At 10:00 ET, exact overnight and opening realized variance reduce rolling out-of-sample loss for the normalized six-hour remaining-RTH variance curve relative to its exact prior-504 mean curve.
At 10:00 ET, adding strictly prior daily, weekly, and monthly realized-volatility memory reduces rolling out-of-sample remaining-RTH variance forecast loss beyond exact overnight and opening RV.
At 10:00 ET, adding the strictly prior downside-versus-upside composition of realized variance reduces rolling out-of-sample remaining-RTH forecast loss beyond exact MST-40 total-HAR state.
At 10:00 ET, adding strictly prior jump share of realized variance reduces rolling out-of-sample remaining-RTH forecast loss beyond exact MST-40 total-HAR state.
At 10:00 ET, a strictly prior realized-quarticity adjustment to the daily HAR coefficient reduces rolling out-of-sample remaining-RTH forecast loss beyond exact MST-40.
At 10:00 ET, adding the completed opening half-hour's range-based variance reduces rolling out-of-sample remaining-RTH forecast loss beyond exact MST-40 total HAR-RV.
At 10:00 ET, adding the completed opening half-hour's Garman-Klass variance reduces rolling out-of-sample remaining-RTH forecast loss beyond exact MST-44 opening-range RV.
At 10:00 ET, strictly prior negative RTH returns carried at daily, weekly, and monthly horizons reduce rolling OOS remaining-RTH forecast loss beyond exact MST-44 opening-range + total HAR-RV.
At 10:00 ET, strictly prior whole-RTH realized-range variance at daily, weekly, and monthly horizons reduces rolling OOS remaining-RTH forecast loss beyond exact MST-44 opening-range plus total HAR-RV.
At 10:00 ET, the completed current overnight realized range reduces rolling OOS remaining-RTH forecast loss beyond exact MST-48, including its overnight squared-return RV and prior range HAR.
The exact replicated MST-48 prior-session realized-range HAR model reduces genuinely prospective joint ES/NQ remaining-RTH forecast loss beyond exact MST-44.
At 10:00 ET, strictly prior whole-RTH Rogers-Satchell variance at daily, weekly, and monthly horizons reduces rolling OOS remaining-RTH forecast loss beyond exact MST-48.
At 10:00 ET, the literature-exact 66-session bias-corrected realized range reduces rolling OOS remaining-RTH forecast loss relative to exact MST-48 raw range-HAR without adding parameters.
Exact MST-23 causal initial-balance breakouts occurring when the exact MST-48 10:00 forecast is in its strictly prior 252-forecast top quartile have positive net close-horizon drift and exceed outcome-blind matched non-high-forecast breakouts after frozen costs.
On untouched NQ, exact MST-23 narrow-initial-balance breakouts in the exact MST-53 high forecast state earn more than +5 net basis points when faded and outperform outcome-blind matched other-state fades by more than +5 basis points after frozen costs.
Exact retained MST-48 prior-session range-HAR forecasts reduce QLIKE loss relative to their exact MST-44 baseline on both already-published ES and NQ replication arms.
The unchanged replicated overnight-RV forecast reduces genuinely prospective QLIKE, MSE, and MAE loss beyond the exact opening-RV baseline on both ES and NQ.
A single jointly confirmed high overnight/opening-range state improves exact MST-48 remaining-RTH variance forecasts under paired QLIKE, MSE, and MAE loss.
At 10:00 ET, exact NQ opening realized variance reduces rolling OOS ES 10:00-15:59 forecast loss beyond exact MST-48 range-HAR when discovery consumes only a pre-outcome opening artifact.
The first completed RTH crossing of a 100-point ES/NQ level has greater net signed 30-minute continuation than an outcome-blind matched first crossing of a 25/50/75 placebo level.
Strictly trailing asymmetric 90% residual intervals around the exact MST-40 HAR challenger have calibrated tails, lower proper interval score, and narrower width than intervals around the exact overnight-plus-opening baseline.
A single frozen adaptive conformal update around the exact MST-40 HAR challenger has calibrated tails, lower proper interval score, and narrower width than the identically adapted exact overnight-plus-opening baseline on the previously unopened NQ interval population.
Mechanism, trigger definition, falsification criteria and all future results are visible to Terminal + AI members. View publication →
VII.Composites12 studies
Mechanism Plausible — unproven
Single signals are often weak or regime-dependent; their *interaction* can be stronger and more honest. A cross-signal composite conditions one force on another — e.g. large put selling **while** dealers are in a positive-gamma (damping) regime versus a negative-gamma (amplifying) regime — on the hypothesis that the same flow means different things depending on the market's hedging state.
The FLOW-01 effect (excess forward drift after 95th-percentile put selling), if present, differs materially by dealer gamma regime — mechanically expected to be larger in negative-gamma sessions, where hedge flow amplifies rather than dampens.
A composite positioning score (net gamma + signed flow + skew + short-horizon momentum) predicts next-session direction better than any single component.
The median forward path of the top-N analogue days returned by find_similar_days predicts the actual forward path (lower median absolute path error over the next 60/120 min) better than the unconditional median path over the same history.
Any signal in the library that shows a raw edge retains economic significance ONLY IF it survives BOTH a VIX/ATM-IV/realized-vol control AND realistic transaction costs — the decisive filter.
The forward excess drift after large call buying (FLOW-02) is conditional on dealer gamma sign: stronger and more persistent in negative-net-gamma sessions than positive-net-gamma sessions.
The forward excess drift after strong signed options flow is conditional on the variance-risk-premium state: weaker when vol is rich (moves already priced) and stronger when vol is cheap.
When put selling, call buying, and positive signed flow align in the same window, the combined forward excess drift is NOT materially greater than the single best marginal signal after controlling for the shared vol regime - i.e. stacking is largely redundant, not additive.
As intraday support/resistance, a gamma-weighted 'value area' outranks an OI-weighted one, which outranks a time-at-price (TPO) value area, which outranks a random-level placebo.
Fading a ±EM boundary touch works (reverts) in positive-gamma / high-VRP calm regimes and fails (breaks) in negative-gamma regimes — the composite of VRP-12 and the gamma sign beats either marginal.
At 15:30 and 15:45 ET, point-in-time option-surface, dealer-hedging, price, flow, liquidity, volatility, and close-context features can identify an executable long-premium SPXW 0DTE position with positive net expectancy and higher 2x/4x/7x/10x barrier probabilities than matched, option-implied, price/vol/tape, marginal-feature, and no-trade baselines.
Mechanism, trigger definition, falsification criteria and all future results are visible to Terminal + AI members. View publication →
VIII.Calendar18 studies
Mechanism Plausible — unproven
The options and equity calendar imposes structure that recurs on a schedule: monthly/quarterly OPEX and the positioning unwind around it, index rebalances, AM- vs PM-settlement mechanics for SPX, and day-of-week patterns in 0DTE volume. These are not forces that a single trade creates — they are properties of *when* in the cycle the market is.
For optionable single stocks, the monthly-OPEX close falls within a small band of the largest-OI strike more often than an unconditional strike-proximity baseline.
In the last three trading days of a quarter, index drift is negatively related to the quarter's equity return (a mean-reverting rebalance signature): strong-up quarters see rebalance-driven selling pressure into quarter-end, and vice versa.
Realized range is SUPPRESSED during monthly OPEX week and EXPANDED in the week immediately after, as the large expiring positive-gamma position pins the tape and then rolls off -- a paired within-cycle compression/expansion, not just an average difference.
On monthly SPX expiration days, AM-settled contracts (settled at the open via SET) leave weaker final-hour pinning than PM-settled expirations, because the large AM gamma is already gone by the afternoon.
The overnight return into monthly (3rd-Friday) SOQ-settled expirations is positively biased and partially reverses by midday, scaling with the magnitude of dealer charm, on those days and not on matched non-expiration days.
The monthly option-expiration week shows an abnormal index return relative to matched non-OPEX weeks, and the sign/size depends on the prevailing dealer-gamma regime entering the week.
The index's positive drift is concentrated in the OVERNIGHT (prior-close → open) return, with the intraday (open → close) return near zero or negative — the documented overnight/intraday split.
Turn-of-month sessions (the last trading day of a month plus the first three of the next) carry a higher close-to-close return than the rest-of-month baseline.
Index returns do NOT follow the claimed FOMC biweekly cycle (higher returns in even weeks of the FOMC calendar); the popular even-week-outperformance pattern fails for the index in our era.
Quad-witching sessions (quarterly simultaneous expiry of index futures, index options, stock futures, and stock options) carry a lower close-to-close return than ordinary monthly-OPEX Fridays.
The historically positive ES return from the exact 02:00 bar open through the 02:59 bar close persists after 2021, clears a 1-bp round-trip cost, remains positive at 3 bps, and localizes against three frozen paired adjacent hours. NQ is the frozen replication.
Mechanism, trigger definition, falsification criteria and all future results are visible to Terminal + AI members. View publication →
IX.Systematic8 studies
Mechanism Plausible — unproven
Rules derived **purely from price, volatility, and macro time series** — moving-average trend filters, time-series momentum, real-time volatility scaling, cross-asset spread rules, and macro-liquidity signals. Unlike the rest of the library, these carry **no options-microstructure mechanism**: no dealer is forced to hedge, no expiry unwinds, no order-flow imbalance prints. What (if anything) makes them work is **risk-shaping** (changing *when* you hold equity risk) or a **documented risk premium**, not a market-structure force a single participant creates.
Holding the index only while it is above its 200-day simple moving average (flat otherwise) reduces maximum drawdown relative to random schedules with the same time in market — a risk-shaping claim, not a return-improvement claim.
Months following a positive trailing-12-month index return earn more than the T-bill hurdle (the time-series-momentum premium), vs months following a negative trailing-12-month return.
Scaling index exposure by inverse realized variance (more exposure when vol is low, less when high), computed in real time, improves the risk-adjusted return (Sharpe) vs constant exposure.
Index return does NOT follow the claimed 52-week-high anchoring pattern; neither the "near the high underperforms" anchoring form nor its dip-buying inverse is a structural, regime-robust edge.
The QQQ-SPY relative-value spread carries NO tradable daily momentum or reversion edge net of costs (neither continuation of spread moves nor reversion to a spread mean pays).
Weekly changes in Fed net liquidity (reserves minus TGA minus RRP, or the popular proxy) do NOT predict forward weekly index returns once trend is controlled.
Among point-in-time S&P 500 constituents, a market-neutral cross-sectional portfolio formed from inverse-price rank and short-term reversal only for stocks in a stock-specific positive-drift state earns a positive cost-adjusted next-session return relative to both the ungated composite and matched placebo gates.