A taxonomy that actually conditions on something
Technical-analysis literature partitions gaps into four buckets: common, breakaway, continuation, exhaustion. The labels are useful only because they correspond to four roughly distinguishable conditional distributions of next-day returns. Treat them as buckets, not as essences.
Common gaps
Open within an established trading range, on volume at or below the 20-day average, no identifiable catalyst. Empirically, these revert to the prior close on a horizon of 1–3 sessions in roughly 70–80% of observations (the precise number depends on universe and gap-size threshold; do not anchor on a single figure). The signal-to-noise ratio is low. There is little tradeable structure here unless you are running a high-frequency mean-reversion book with very tight cost assumptions.
Breakaway gaps
Open outside a support/resistance level or out of a multi-week consolidation — a triangle, a flat base, a coiled range — on volume substantially above average (a 2σ or larger excursion from the rolling-20 mean is a reasonable threshold). These are genuine regime-change signals. The conditional probability of fill within 5 sessions is materially lower than for common gaps; the conditional return over the next 5–20 sessions has a positive mean and a fat right tail. A breakaway gap is among the cleaner technical signals available to a discretionary trader, which is a low bar but a real one.
Continuation (runaway) gaps
Occur mid-trend, after a breakaway has already established direction. Volume expands. The gap confirms momentum is accelerating. These rarely fill until the trend itself reverses. From an options-pricing standpoint, continuation gaps are interesting because they tend to compress realised vol (the move is directional, not chaotic) while implied vol on near-dated strikes can stay elevated — a structural HV-IV gap that occasionally creates short-premium edge for someone who knows what they are doing.
Exhaustion gaps
Occur near the end of a trend. Final gap in the direction of the trend, but volume is lower than the previous gap, and the price reverses intraday or within one session. Conditional fill probability over 1–5 sessions is empirically high — in the 80–90% range on most universes I have looked at. The structural problem is that distinguishing an exhaustion gap from a continuation gap in real time is a classification problem with one informative feature: continuation gaps print on expanding volume; exhaustion gaps on declining volume. That is the entire diagnostic. It is not always sufficient.
The fill statistic, used carefully
The popular formulation “all gaps fill eventually” is technically true on an infinite horizon and operationally useless on any horizon a trader cares about. A breakaway gap on a stock that compounds 40% over the next eighteen months will only “fill” if the stock gives back the entire move. Treating that as a near-term mean-reversion signal is how short books blow up.
The honest summary of the empirical fill statistics looks roughly like this:
- Common gaps: fill within 1–5 sessions about 70–80% of the time.
- Breakaway gaps: do not fill on the relevant trading horizon. When they do fill, the breakout has failed, which is itself a tradeable signal in the opposite direction.
- Continuation gaps: fill only when the broader trend reverses. Conditioning on “trend intact”, fill probability over 20 sessions is low.
- Exhaustion gaps: fill within 1–5 sessions about 80–90% of the time.
I would treat all four numbers as point estimates with wide confidence intervals. Run them on your own universe, your own date range, and your own gap-size threshold before you trade them.
Three intraday strategies, with their actual conditioning
Strategy 1: Gap and Go
Setup. Stock gaps up on volume in the upper tail of its 20-day distribution, holds above the opening print through the first 15 minutes, and consolidates in a tight range near the high. Catalyst is identifiable and asymmetric (earnings beat above the analyst-consensus 90th percentile, FDA approval, accretive M&A).
Entry. Long on a break above the 15-minute high.
Stop. Below the 15-minute low or the opening print, whichever is tighter.
Target. Minimum 1:2 risk-to-reward. If risk is $0.50, target is $1.00.
This is a momentum play conditioned on the breakaway-gap distribution. It does not work on common gaps. It does not work on gaps without an identifiable catalyst. Filtering hard is the entire edge.
Strategy 2: Gap Fade
Setup. Stock gaps up but trades below the opening print within the first 30 minutes on declining or moderate volume. No fresh news after the open.
Entry. Short (or exit existing long) on a break below the 15-minute low.
Stop. Above the 15-minute high.
Target. Prior session close (the gap-fill level).
Gap fades are conditioned on the common-gap and exhaustion-gap distributions. They are actively dangerous when applied to breakaway gaps. Misclassifying a breakaway as an exhaustion is the single most expensive mistake in this category, and it is not a rare one.
Strategy 3: Gap and Wait
Setup. You cannot yet classify the gap. The first 30 minutes have not given you enough information.
Entry. After 10:00 AM Eastern, condition on what the tape has done. If the stock is holding above the gap on declining-but-stable volume, enter long. If it has faded materially, wait for the gap-fill level and enter there only if support holds on visible buying.
Of the three, this is the one I have the most respect for, because it explicitly admits that the first half-hour is the noisiest part of the session and that classifying a gap with insufficient data is a recipe for being whipsawed. Patience is underweighted in retail gap trading. The opening auction and the first few minutes after it are a low-information regime with high realised vol; waiting is, in expectation, free.
Pre-market: features worth conditioning on
Most gaps are visible in pre-market (04:00–09:30 ET). The features that matter, in roughly descending order of predictive value:
- Catalyst class. Earnings, guidance revision, FDA, M&A, sector rotation, no-news. Each class has a different conditional distribution. A gap with no identifiable catalyst is, in expectation, a common gap.
- Pre-market volume. Heavy pre-market volume (call it >5% of the 20-day ADV before the open) signals genuine institutional interest. Light pre-market volume on a large gap means the print was set by a thin book and is unlikely to hold.
- Pre-market drift. Trending higher into the open is a different distribution from fading off the pre-market high. Direction of drift in the last 30 minutes pre-open has measurable predictive value for the first 30 minutes post-open.
- Broader market. If S&P 500 futures are 1.5σ or more from their overnight mean, the gap is partly market-driven. Market-driven gaps are noisier and fade more frequently than stock-specific gaps. Decompose the move before trading it.
Mistakes I have either made or watched up close
Treat this less as a list of rules and more as a checklist of where retail gap-trading P&L typically leaks.
- Trading every gap. Most gaps are common. Selectivity is the edge. If your filter does not eliminate roughly 80% of pre-market candidates, the filter is not doing its job.
- Mean-reverting the breakaway. Shorting a legitimate breakaway because “all gaps fill” is the canonical retail blow-up. The fill happens, on average, after the position has been carried through a 30–50% adverse move. The math does not survive that.
- Trading the first five minutes. Spreads are wide. Fills are unreliable. The implied half-spread cost in those five minutes can erase an entire session of edge. Wait for the auction to settle.
- Ignoring relative gap size. A 2% gap on a name with 4% average daily range is noise. A 10% gap on the same name is a signal. Normalise gap size by the 20-day ATR or realised vol before classifying.
- No stop. A gap trade without a defined stop is not a trade. It is a position with an undefined risk profile, which is to say, an unbounded short put on your own discipline.
A note on the classification problem
I want to flag something honestly. Most of what is written about gap trading, including most of what is above, is a taxonomy and a set of heuristics. I have looked at this as a supervised-learning problem — can you train a classifier on (gap size, pre-market volume, catalyst dummy variables, ATR-normalised features, prior-trend features) to distinguish breakaway from exhaustion gaps in real time, before the session opens? The answer in my own work is: marginally, yes, but the edge is smaller than the discretionary literature implies, and it degrades quickly out of sample. The features that matter are the obvious ones; the interactions are noisy; and the labels are themselves contaminated by the path taken after the gap, which is a leakage problem you have to be careful about. Euan Sinclair’s point in Volatility Trading is the relevant one here: a feature that improves classification accuracy from 52% to 55% on a high-cost, low-frequency signal is not necessarily a tradeable feature once you net out execution. I have a backtest on a gap-fade variant that ran for four months in 2023, Sharpe of about 0.6 gross, roughly flat after slippage on the universe I was using. I shut it down.
That is, I think, the honest framing. Gaps are real signals. The conditional distributions are different enough that the four-bucket taxonomy is not nonsense. But the classification step is harder than it reads, and the post-cost edge, even when the classification is correct, is thinner than most descriptions suggest.
Model assumptions, stated for this article: equity gaps in liquid US single names, 20-day rolling baselines for volume and ATR, regular-trading-hours session, half-spread plus a small impact term as a transaction-cost model. Pre-market microstructure is genuinely different from regular-session microstructure; thinly-traded names violate most of the volume conditioning above. Adjust accordingly.
See also: Earnings Season · Momentum Trading: Riding the Trend · Day Trading · Support and Resistance · Understanding Volume
