Gap Fill Rate

do opening gaps fill intraday — and does gap size predict whether they do?

An opening gap is the distance between the prior session’s close and today’s open. This study asks: does price return to the prior close on the same day? We measure gap size in ATR units— the gap divided by the instrument’s own 14-day average true range — so the buckets are comparable across names. A 0.5-ATR gap means the same thing on SPY (low vol) as on TSLA (high vol); a raw-% axis would quietly mix “huge gap for SPY” with “nothing gap for TSLA.” Across 12,306 gap events on 15 liquid US large-caps and ETFs (Jan 2022 – May 2026), the answer depends entirely on the gap’s size relative to the stock’s own range.

The fill-rate curve

0.10–0.25 ATR

74.7%

n=4,776 · CI95 [73.5%, 75.9%] · avg gap 0.17 ATR (~48 bps)

0.25–0.50 ATR

50.9%

n=4,675 · CI95 [49.4%, 52.3%] · avg gap 0.36 ATR (~100 bps)

0.50–1.00 ATR

29.6%

n=2,324 · CI95 [27.7%, 31.4%] · avg gap 0.67 ATR (~181 bps)

≥1.00 ATR

12.4%

n=531 · CI95 [9.8%, 15.3%] · avg gap 2.31 ATR (~750 bps)

50% reference = coin-flip / pure random. Bars teal = reversion zone · grey = random zone · red = momentum zone. The crossover through 50% sits cleanly at ~0.5 ATR.

Why measure in ATR, not percent

When this study was first run in raw basis points, the crossover from reversion to momentum was smeared across a wide “random zone” (≈75–149 bps) — because a 100-bps gap is a major event for SPY but routine noise for TSLA, and the bps buckets blended the two. Normalizing each gap by its own instrument’s ATR removes that volatility-mix confound. The result is sharper: the reversion zone strengthens (75% vs 69%) and the 50% crossover collapses to a single clean threshold near 0.5 ATR. The same edge, measured in the right units.

Reversion zone · <0.25 ATR

Small gaps are noise. 75% fill same day (CI 95% clears 50% comfortably). These are routine overnight positioning shifts — small relative to the day’s normal range — that get absorbed in the first 30–60 minutes. Fading them, or assuming they fill before acting on a setup, is structurally sound.

Random zone · 0.25–0.50 ATR

Medium gaps are coin flips. 50.9% fill rate — CI straddles 50%, indistinguishable from random. This is the knife-edge: enough size to mean something, not enough to trade directionally. Neither a fade nor a follow has structural edge here, in any instrument.

Momentum zone · >0.50 ATR

Large gaps are catalysts. Only 30% fill at 0.5–1.0 ATR; just 12% at ≥1.0 ATR. Fading these loses ~7-in-8 times intraday at the extreme. This is Brooks’ “strong move with follow-through” — when the open exceeds the stock’s own daily range, the open IS the signal, and the market continues.

Per-bucket detail

Gap size (ATR)	N gaps	Fill rate	CI 95%	Up-gap	Down-gap	Avg gap	Zone
0.10–0.25 ATR	4,776	74.7%	[73.5%, 75.9%]	72.6%	77.1%	0.17 ATR	reversion
0.25–0.50 ATR	4,675	50.9%	[49.4%, 52.3%]	50.8%	50.9%	0.36 ATR	random
0.50–1.00 ATR	2,324	29.6%	[27.7%, 31.4%]	30.2%	28.8%	0.67 ATR	momentum
≥1.00 ATR	531	12.4%	[9.8%, 15.3%]	12.4%	12.5%	2.31 ATR	momentum

Up-gaps vs. down-gaps — near-perfect symmetry

In every size bucket, the fill rate for up-gaps and down-gaps differs by less than 5 percentage points — mostly well inside the CI. The market has no directional asymmetry in gap fill behavior: fading a gap down is statistically indistinguishable from fading a gap up of the same ATR-size. This rules out the common intuition that “stocks recover faster from gap-downs than gap-ups.”

0.10–0.25 ATR

↑ 72.6% · ↓ 77.1%

Δ = 4.5%

0.25–0.50 ATR

↑ 50.8% · ↓ 50.9%

Δ = 0.1%

0.50–1.00 ATR

↑ 30.2% · ↓ 28.8%

Δ = 1.3%

≥1.00 ATR

↑ 12.4% · ↓ 12.5%

Δ = 0.1%

What this means for trading

Use one rule across every instrument: 0.5 ATR. Below it, gaps fill; above it, they extend. Because the axis is volatility-normalized, the same 0.5-ATR threshold works on SPY, TSLA, and everything in between — no per-name tuning.
Don’t fade gaps larger than the daily range.At ≥1.0 ATR, only 12% fill — fading is a 7-in-8 loser intraday. The gap-and-go setup is statistically dominant at this size. Follow, don’t fade.
The “small gap fills” assumption is justified — when small is measured right.A sub-0.25-ATR gap closes 75% of the time. When building an intraday setup, a small opening gap (relative to the stock’s range) is not a structural obstacle to filling your target.
The 0.25–0.5 ATR band is dead money.No directional edge. A gap fade triggered on raw points or percent will blend this random band with the clean reversion zone; an ATR-gated fade (<0.25 ATR) or an ATR-gated momentum follow (>0.5 ATR) will outperform.

How this fits the other studies

First-Hour Fade: That edge already sizes everything in ATR (0.30 ATR target, 0.75 ATR stop). This study uses the same volatility ruler, and its finding — small local extremes get absorbed — is the same force the fade harvests at an 87% win rate.
Gap Fade (K and K2):Those backtests gated on raw basis points and produced a “NULL” on the strict verdict because the bps gate blends instruments of different volatility. This study shows the edge is real but only visible once the gap is measured against each instrument’s own range.
Prior-Day Extremes: The same mean-reversion force that fills sub-0.5-ATR gaps also drives the prior-day H/L retest to fail — short-term positioning pressure gets absorbed before it can sustain a breakout.

Pre-registered verdict

✗H12(a) — all buckets > 50%: FAILED. Only the <0.25 ATR bucket is clearly above 50%. The 0.25–0.5 ATR zone straddles 50%; the >0.5 ATR zones are decisively BELOW 50%.

✓H12(b) — monotone decrease with size: PASSED. 75% → 51% → 30% → 12% — strictly decreasing across all four ATR buckets.

✓H12(c) — up/down symmetry within 10 pp: PASSED. Max directional difference is 4.5 pp (<0.25 ATR bucket); all others <1.5 pp.

H12(a) failure is the more interestingresult: the hypothesis that “gaps tend to fill” is only true for gaps small relative to the instrument’s range. The actual finding — that fill rate inverts cleanly through 50% at ~0.5 ATR — is structurally richer than the original claim, and the ATR ruler makes it a single threshold rather than an instrument-dependent one.

Honest caveats

Fill is detected using the daily L (for up-gaps) and daily H (for down-gaps), which includes the closing auction. An intraday-path study (1-min bars) might show a slightly lower fill rate. This is a conservative proxy — daily H/L overstates the reachable range slightly.
ATR is the 14-day Wilder ATR computed strictly on bars before the gap day, so there is no lookahead in the size normalization.
Universe is 15 mega/large-caps and ETFs, all US. The ATR normalization is designed to make the result generalise across instruments, but it has not been validated on small-caps or non-US markets where gap mechanisms differ.
Earnings gaps are included and cluster in the ≥1.0 ATR bucket — structurally different from routine macro gaps. Excluding earnings days would likely lift the large-gap fill rate modestly, not enough to change the narrative.

Methodology

12,306 gap events (≥0.10 ATR) on 15 names (2022-01-03 – 2026-05-30) via Massive/Polygon daily bars. Gap size = |open − prev close| ÷ Wilder-ATR(14), ATR strictly pre-gap. Fill detection via daily OHLC. Bootstrap CI (n=5,000, seed=17). Engine: scripts/ml/backtest_gap_fill_rate.py. Run 2026-06-14.