No: on ES and NQ futures about 60 percent of regular-session gaps fill by the same day's close, and TradingStats counts 60.3 percent for NQ over 2,791 trading days from 2015 to 2025. The catch is size: on NQ the smallest gaps fill 77.8 percent of the time and the largest 8.2 percent, and a driftless random walk produces a similar size pattern. A gap (the distance between one regular-session close and the next open) is not a magnet, and a fill rate is no edge until it beats chance and pays for the gaps that run.

Key takeaway

Gaps do not always fill. TradingStats counts 60.3 percent of NQ regular-session gaps filling by the close over 2,791 days, from 77.8 percent for the smallest to 8.2 percent for the largest, and our random-walk computation gives 81.3 and 43.3 percent for the two smallest size buckets, so most of the fill rate is distance against volatility. A gap fade only pays if your fills beat both that chance rate and the breakeven win rate your stop sets.

How often do gaps actually fill on ES and NQ?

Over one regular session, roughly 6 in 10 index futures gaps fill. Over an open-ended horizon nearly all SPY gaps do, while one peer-reviewed study of large gaps in nine markets found up to 80 percent unfilled after five days, so the answer depends on horizon, definition and size.

The longest futures sample is the TradingStats NQ gap study (February 2026), which measures the RTH (regular trading hours, 9:30 a.m. to 4:00 p.m. ET) open against the previous RTH close: "The baseline: 60.3% of NQ gaps fill completely by the close." An independent ten-year count by Beyond Candlesticks lands on 60 percent for ES and 61 percent for NQ. The table shows the main published figures; each uses its own definition, so they cannot be merged.

SourceMarket and periodHorizonFill rate
TradingStats (2026)NQ, 2015 to 2025, 2,791 daysSame RTH session60.3 percent
Beyond Candlesticks (2026)ES, NQ, RTY, gold, ten years to September 2026Same sessionES 60, NQ 61, RTY 63, gold 49 percent; gaps of "meaningful size": ES 40.6, NQ 42.5, RTY 44.8, gold 33.1 percent
edgeful (2026)ES, six months to April 2026Not stated64 percent up, 63 percent down
StatOasis (2026)SPY, February 1993 to June 2026, 8,397 gapsNo time limit98 percent down, 94 percent up
Caporale and Plastun (2017)Nine markets, large gaps only, 2000 to 2015Five daysUp to 80 percent not filled

Do 100 percent of gaps fill by the close?

No; that claim comes from a misread chart. The TradingStats fill-timing study plots a line reaching 100 percent at the 15:15 CT close, but it only counts days on which the gap filled: 803 ES days and 839 NQ days from January 2020 to April 2025. By 9:00 CT, 30 minutes after the 8:30 CT open, 51.43 percent of those ES fills and 60.91 percent of the NQ fills had happened.

The study says "The first 30 minutes account for over half of all gap fills". Chance front-loads near touches too: a random walk makes 78.1 percent of its same-session touches of a level 0.1 session sigma away within the first 30 of 390 minutes, and 45.8 percent at 0.25 sigma, so an early fill says nothing special about the open.

What counts as a gap on ES when Globex trades nearly 24 hours?

On a normal weekday an ES gap is a session definition, not a hole in trading: it is the distance between one RTH close and the next RTH open, and the contract traded through that range overnight. Only the weekend, holidays and the daily one-hour halt leave untraded price.

According to NinjaTrader's ES contract page, "E-mini S&P 500 futures trade nearly 24 hours a day, five days a week, from Sunday 6:00 pm ET through Friday 5:00 pm ET", with a 60-minute break from 5:00 p.m. ET. So the split between RTH and the overnight session decides what counts as a gap, and the Sunday reopen and holiday closures are where true no-trade gaps come from.

Picture one weekday as a timeline: the cash close at 4:00 p.m. ET, the futures RTH close some studies use at 4:15 p.m. ET (15:15 CT), the 5:00 to 6:00 p.m. ET halt, a night of Globex trading, and the 9:30 a.m. ET (8:30 CT) open. Draw the median ES overnight gap of 11.5 points, as Beyond Candlesticks measures it, between the close and the open. The overnight path connecting them has already traded every price inside it.

ONE ES WEEKDAY: WHERE THE GAP COMES FROM cash close 4:00 p.m. ET futures RTH close 4:15 p.m. ET (15:15 CT) RTH open 9:30 a.m. ET (8:30 CT) HALT GLOBEX TRADES OVERNIGHT 4:00 p.m. close 4:15 p.m. close settlement RTH GAP 11.5 points median ES every price inside the gap already traded overnight 4:00 p.m. 6:00 p.m. midnight 6:00 a.m. 9:30 a.m. time ET, to scale halt 5:00 to 6:00 p.m. ET Illustrative price path and close levels. Median ES gap: Beyond Candlesticks. Times: NinjaTrader ES contract page.
On a weekday the RTH gap was traded through overnight, so there is no vacuum to fill, and three different closes give three different gap sizes.

Studies also pick different closes. TradingStats uses 4:00 p.m. ET in its NQ study and 15:15 CT in its timing study, and MPM Markets measures from the prior settlement price, so one morning can give three gap sizes. No vendor publishes how it treats holiday half-days, the Sunday reopen or contract rolls, and a roll can draw a gap nobody could trade; see how futures rollover works and how back-adjusted data is built.

Do gaps fill because the market abhors a vacuum?

No evidence supports it. Caporale and Plastun, in their Investment Analysts Journal paper on price gaps, test the saying "the market abhors a vacuum and all gaps will be filled" and conclude that "in most cases there is no significant evidence of anomalous price behaviour associated with the emergence of gaps". On a weekday ES chart there is no vacuum anyway, and as with stops resting past round numbers, a level does not need to attract price for price to reach it.

Why do small gaps fill and large gaps rarely do?

Small gaps fill more often because the chance of touching any level falls steeply with its distance measured in volatility. A random walk with no memory, no magnet and no orders waiting at yesterday's close shows the same pattern.

The tool is the reflection principle from MIT OpenCourseWare lecture notes: for a driftless Brownian motion, the chance of touching a level within a period is twice the chance of ending beyond it. For a trader, P = 2 x (1 - Phi(d)), where d is the gap divided by the session sigma (the standard deviation of open-to-close moves) and Phi is the standard normal distribution.

Distance to level (session sigma)Touch probability, formulaSimulated, 1-minute steps
0.2580.3 percent78.0 percent
0.561.7 percent59.8 percent
0.67450.0 percentnot simulated
131.7 percent30.4 percent
1.513.4 percent12.7 percent
24.6 percent4.3 percent

Computed by Phoenix Technologies; method: P = 2 x (1 - Phi(d)), checked against 100,000 simulated sessions of 390 one-minute steps (seed 20261119), which run up to about 2 points lower because minute bars miss touches between samples. A level a quarter of a session's volatility away gets touched about 4 times in 5 by chance alone.

TradingStats calls its size result "the single most important finding in the entire dataset": "The gap fill percentage drops nearly 10x from tiny to large gaps." Picture the chance curve falling from 80.3 percent at 0.25 sigma to 4.6 percent at 2, with the vendor's four NQ size buckets as dots. Converted at 1 ATR (average true range, here a 14-period ATR on daily bars) = 1.596 session sigma, the dots at 77.8, 42.0, 25.6 and 8.2 percent sit at 0.24, 0.80, 1.52 and 2.55 sigma: the first two close to the curve, the two large-gap dots above it.

TOUCH ODDS FALL WITH DISTANCE 0%25%50%75%100% 00.250.511.522.5 distance to the level, in session sigma Chance of touching a level within one session 50.0% at 0.674: coin flip random walk 2 x (1 - Phi(d)) 80.3% 61.7% 31.7% 13.4% 4.6% 77.8% at 0.24 42.0% at 0.80 25.6% at 1.52 8.2% at 2.55 TradingStats NQ size buckets two small near the curve, two large above it Bucket positions use 1 ATR = 1.596 session sigma. Curve: reflection principle, computed by Phoenix Technologies.
The two small-gap vendor fill rates sit almost on the chance curve: small gaps fill because they are close, not because they are pulled.

For ES, take an illustrative session sigma of 50 points; no public source publishes it, so substitute your own. The median 11.5-point gap, $575 on one contract at $50 a point (see ES tick and point values), fills 81.8 percent of the time by chance. A 30-point gap drops to 54.9 percent, a 50-point gap to 31.7 percent, and the coin-flip size is 33.7 points.

This is the base-rate test we apply to ICT fair value gaps, a three-candle pattern and a different object from a session gap.

Do published gap fill rates beat a random walk?

For the small and mid-sized gaps that make up most days, no: the TradingStats NQ fill rates sit 1.3 to 3.5 points below a random walk at the same size. Only the two large buckets beat chance, by a margin a volatility mismatch can explain.

The table sets each ATR bucket beside the random-walk rate, and in the last column assumes the gap day is 1.5 times as volatile as the trailing ATR implies, plausible because big gaps arrive on volatile days. No public ES figure pins that multiple down.

Gap size (daily ATR)Approx. NQ pointsVendor fill rateRandom-walk baseVendor minus baseBase if 1.5x volatile
Under 0.3Under 10077.8 percent81.3 percent-3.587.4 percent
0.3 to 0.7100 to 25042.0 percent43.3 percent-1.359.8 percent
0.7 to 1.2250 to 40025.6 percent14.0 percent+11.631.8 percent
Over 1.2400 and up8.2 percent1.6 percent+6.69.8 percent

Computed by Phoenix Technologies from the TradingStats buckets; method: 1 ATR = sqrt(8/pi) = 1.5958 session sigma (the expected daily range of a random walk), P averaged evenly across each bucket, open bucket capped at 2.0 ATR (illustrative). The size effect vendors present as a finding is mostly distance against volatility.

Picture four pairs of bars on one baseline, vendor beside chance, with a whisker for the 1.5x case. The apparent edge is the +11.6 and +6.6 points in the large buckets, yet at 1.5x volatility the chance bars reach 31.8 and 9.8 percent, above the vendor's 25.6 and 8.2; at 1.25x they read 23.3 and 5.0 percent.

VENDOR FILL RATES VS RANDOM-WALK BASE 0%25%50%75%100% vendor fill rate (TradingStats NQ) random-walk base base if the gap day is 1.5x as volatile VENDOR MINUS BASE 87.4 -3.5 77.8 81.3 under 0.3 59.8 -1.3 42.0 43.3 0.3 to 0.7 31.8 +11.6 25.6 14.0 0.7 to 1.2 9.8 +6.6 8.2 1.6 over 1.2 gap size, in daily ATR small gaps: below chance large gaps: above, until volatility rises 1 ATR = 1.5958 session sigma; P averaged across each bucket; open bucket capped at 2.0 ATR (illustrative). Computed by Phoenix Technologies.
The only part of the fill rate that could be an edge is the difference above chance, and in the large buckets it disappears if gap days are 1.5 times as volatile as the trailing ATR.

TradingStats does not say whether its ATR uses Globex or RTH bars, and we found no public ES study that compares fill rates with a random baseline. Its open-location split is the size effect again: opens inside the prior day's range filled 70.4 percent, above it 47.1 percent, below it 44.1 percent.

Beyond Candlesticks puts the consequence plainly: "The moment a gap is big enough that you would actually want to trade it, the base rate is against you."

Do gaps fill eventually, and do gap downs fill more than gap ups?

Most do, on an index that drifts upward, and a random walk predicts it. StatOasis counts 98 percent of SPY down-gaps and 94 percent of up-gaps filled at some point between February 1993 and June 2026, with an intraday touch counting.

The MIT notes give the reason: a driftless Brownian motion reaches any level eventually with probability 1, while with drift the furthest move against it is finite. In a rising index a gap down's fill level lies above price, in the direction of the drift, which fits down-gaps filling slightly more often (98 against 94 percent).

Over one session the direction evidence is mixed: TradingStats has NQ gap downs at 62.2 percent and ups at 58.8, while edgeful's six-month ES window (fill horizon not stated) shows 64 for ups and 63 for downs. StatOasis sums it up: "Direction is close to noise, and size is what carries the signal."

StatOasis finds that SPY gaps over 2 percent filled 83 percent of the time, at a median of two daily bars, and MPM Markets sees most large futures gaps fill within about ten sessions. Caporale and Plastun, studying only the largest opens (about 5 to 6 percent of days), report that "the evidence suggests that up to 80% of gaps are not filled within five days." A fill three sessions later does not help a trade stopped out on day one.

Is a high gap fill rate an edge?

No. A fill rate tells you how often you win, not what you lose when the gap runs, and with a wide-stop fade bracket like the one below, a pure random walk already wins more often than the published small-gap fill rates.

Take an illustrative ES fade: a 4-point target at the prior close, a 20-point stop, one contract, 10 trades.

  1. Eight gaps fill: 8 x 4 = +32 points, or $1,600 at $50 a point.
  2. Two gaps run to the stop: 2 x 20 = -40 points, or $2,000.
  3. Net: 32 - 40 = -8 points, -0.8 a trade, or -$400.
  4. Add one tick (0.25 points) of commission and slippage per trade: -10.5 points.
  5. Breakeven fill rate: 20 / (4 + 20) = 83.33 percent, or (20 + 0.25) / 24 = 84.38 percent with costs.

Picture it as a ladder: eight short green steps of +4 climbing to +32, then two long red drops of -20 ending at -8 points, below the start. An 80 percent fill rate, near the 77.8 percent small-gap figure, still loses.

80 PERCENT FILLS, STILL NEGATIVE -100+10+20+30 1 2 3 4 5 6 7 8 9 10 trade number Cumulative points, one ES contract (illustrative) -8 -10.5 +4 a fill 8 fills = +32 +$1,600 -20 -20 2 stops = -40 points, -$2,000 net -8 points, -$400 after 0.25 point cost per trade: -10.5 FILL RATE NEEDED 83.33% breakeven: 20 / (4 + 20) 84.38% with cost this run: 80% 83.33% short of breakeven, so it loses A RANDOM WALK hits +4 before -20: 83.33% expectancy zero Illustrative bracket: 4-point target at the prior close, 20-point stop, $50 a point, 10 trades.
A fill rate of 80 percent loses with this bracket because chance alone would win 83.33 percent of the time.

Karl Sigman's Columbia University notes on Brownian motion show the chance of hitting +a before -b is b/(a+b) with an expected value of zero, so a random walk hits +4 before -20 exactly 83.33 percent of the time and makes nothing. The breakeven win rate for any risk-reward follows the same formula, and it explains high win rates that still net zero.

Real sessions add a third exit, the close. In our seeded simulation (200,000 random-walk sessions, target 0.25 sigma, stop 1.25 sigma, flat at the close), 77.7 percent of trades fill, 13.5 percent stop out and 8.9 percent time out at an average of -0.497 sigma. Booking each target at the limit price and each stop and time-out where price actually was, the mean is -0.0223 sigma a trade before costs, with a standard error of 0.0012: chance yields a vendor-grade fill rate and a small loss even before commission.

TradingStats reports NQ drawdowns of 44.2 points at the median and 203.0 points at the 90th percentile, defined as how far price moves against the fill direction before filling. It does not say whether gaps that never filled are included; if not, the trades that hurt most are missing.

Grant, Wolf and Yu's 15-year study of S&P 500 futures (1987 to 2002) found highly significant reversals after large opening moves, but "the significance of intraday price reversals is sharply reduced when gross trading results are adjusted by a bid-ask proxy for transactions costs". MPM Markets tested eight gap strategy implementations on six futures markets: "None of the eight implementations produced a statistically reliable edge after costs." The wider record is in what research says about technical analysis.

Is a half-gap target better than a full gap fill?

Not by default. A half-gap target fills more often for the same geometric reason a small gap does, and it pays half as much per win, so the breakeven fill rate rises with it.

edgeful reports half gaps lifting ES gap downs from 63 to 87 percent and gap ups from 64 to 72 percent over the six months before its April 2026 post, about 125 sessions, and notes that "in theory, a smaller target gets hit more often." TradingStats' longer NQ sample shows the gradient: a quarter of the gap filled on 86.6 percent of days, half on 76.1 percent, three quarters on 68.0 percent, all of it on 60.3 percent.

Under pure chance, halving the distance adds 15.2 points for a 0.4-sigma gap (68.9 to 84.1 percent) and 30.0 points for a 1-sigma gap (31.7 to 61.7 percent). edgeful's uplifts of 24 points for downs and 8 for ups are what geometry alone predicts, or less. Rerun the ladder with a 2-point target and the same 20-point stop: breakeven climbs to 20 / 22 = 90.91 percent (92.05 with one tick of cost), and 8 fills in 10 now net -24 points.

When is a gap fade worth testing on your own data?

Only when your out-of-sample fill rate, for gaps of a given size, beats both the random-walk base rate for that size and your bracket's breakeven win rate after costs. Clearing one without the other is no trade.

  1. Fix the definition: which close (4:00 p.m. ET, 4:15 p.m. ET or settlement), which open, and how your continuous series handles rolls.
  2. Measure your instrument's session sigma from its own RTH open-to-close moves and express each gap as d = gap / sigma.
  3. Look up chance: 2 x (1 - Phi(d)), for example 61.7 percent at 0.5 sigma and 31.7 percent at 1.
  4. Compute breakeven: (stop + cost) / (target + stop), for example 84.38 percent for 4 points against 20 with one tick of cost.
  5. Test on data the rule was not built on, with all three exits (target, stop, session close); see walk-forward testing and overfitting.
  6. Judge expectancy and profit factor at your win rate, never fill rate alone.

Picture a grid with three rows of gap size (under 0.5 sigma, 0.5 to 1 sigma, over 1 sigma) and three columns of evidence. Below the base rate or below breakeven: no trade. Above both on fewer than about 100 out-of-sample gaps: a small-size test. Above both on a large out-of-sample sample: tradeable, at a size your stop can carry.

WHEN A GAP FADE IS WORTH TESTING Your out-of-sample fill rate, against the higher of chance and breakeven more evidence below base rate or below breakeven above both, fewer than about 100 out-of-sample gaps above both, large out-of-sample sample GAP SIZE CHANCE OF FILL under 0.5 sigma chance above 61.7% no trade small-size test tradeable 0.5 to 1 sigma chance 61.7% to 31.7% no trade small-size test tradeable over 1 sigma chance below 31.7% no trade small-size test tradeable 0 100% breakeven = (stop + cost) / (target + stop), e.g. 83.33% for 4 vs 20 before costs Chance: 61.7% at 0.5 sigma, 31.7% at 1 sigma (random walk). Test with target, stop and session-close exits.
A gap rule earns real size only when it beats chance and its own breakeven on data it was not built on.

For most traders the honest result is no trade. A gap you do not fade still tells you how far the session opens from yesterday's close: context, not a signal.

Go deeper

Frequently asked questions

Do Sunday gaps fill in futures?

We found no public ES study that isolates gaps at the Sunday reopen, so we cannot quote a fill rate for them. They differ from weekday gaps because no trading happened in between, which makes them real gaps rather than a session definition. In FX and commodities, Caporale and Plastun found that gaps appear mainly on Mondays.

Do gaps fill the same way in gold and Russell futures?

Gold fills less often than the equity index contracts. Beyond Candlesticks counts 49 percent of gold gaps filling the same session against 63 percent for RTY, and 33.1 against 44.8 percent for gaps of what they call "meaningful size". Their page does not define the size threshold, so treat the comparison as rough.

Is a fair value gap the same as a price gap?

No. A fair value gap is a three-candle pattern inside continuous trading, while a session gap is the distance between one regular-session close and the next open. Both can be tested against a random baseline, but their statistics are not interchangeable.

What exactly counts as a gap fill?

Studies such as StatOasis count a single touch of the prior close as a fill. StatOasis defines it as "price traded back to the prior close at some point (an intraday touch counts)." A touch is not the same as a filled limit order at that price, so live results can trail the published rate.

How much is a typical ES gap worth in dollars?

The median ES overnight gap in the Beyond Candlesticks data is 11.5 points, which is $575 on one ES contract at $50 a point. Your own median depends on which close you measure from and over which years.

How many gaps do you need before trusting a fill rate?

More than six months of data gives you: a six-month window holds about 125 sessions, against 2,791 days in the TradingStats NQ sample. In our decision grid, fewer than about 100 out-of-sample gaps only justifies a small-size test.

Is a gap fade the same as an opening reversal?

Not quite. Grant, Wolf and Yu studied reversals after large opening moves in S&P 500 futures from 1987 to 2002, which means price moving back toward the prior close, not necessarily touching it. Their reversals were highly significant before costs and much weaker after a bid-ask cost proxy.

Do gaps fill faster on NQ than on ES?

Slightly, among the gaps that do fill. In the TradingStats timing data, 60.91 percent of NQ fills had happened by 9:00 CT against 51.43 percent for ES. Those shares only cover days that filled, so they say nothing about how often each contract fills.

Sources

  1. MIT OpenCourseWare (2013), 6.265/15.070J Advanced Stochastic Processes, Lecture 7: The Reflection Principle. The Distribution of the Maximum. Brownian motion with drift
  2. Investment Analysts Journal (Caporale and Plastun), Brunel University Research Archive copy (2017), Price gaps: Another market anomaly?
  3. Journal of Banking and Finance (Grant, Wolf and Yu), Montclair State University (2005), Intraday price reversals in the US stock index futures market: A 15-year study
  4. TradingStats (2026), Gap Fill Strategy: 2,791 Days of NQ Data (2015-2025)
  5. TradingStats (2026), When Do Gaps Fill? ES & NQ Gap Fill Timing by 5-Minute Intervals
  6. MPM Markets (2026), Testing the "Gaps Always Fill" Myth Across Six Futures Markets
  7. StatOasis (2026), S&P 500 Opening Gaps: Direction Is Noise, Size Is the Signal
  8. Beyond Candlesticks (2026), Who Is Willing To Trade In The Gap
  9. edgeful (2026), half gaps vs full gaps: how one setting changes your gap fill rate
  10. NinjaTrader (2026), E-mini S&P 500 Futures (ES): Contract Specs and How to Trade
  11. Columbia University (Karl Sigman), IEOR 6712 course notes (2014), Notes on Brownian Motion