Testing the "Gaps Always Fill" Myth Across Six Futures Markets

1. Introduction

Few statements appear in trading material as often as "gaps always fill." It is repeated in books, in chatrooms, and in social media threads, usually without conditions and almost always without evidence. The implication is operationally specific: if price opens away from the prior settlement, it will return to that prior settlement, and a trader can act on that expectation.

This study takes the claim at face value and reframes it as a measurable question: how likely is a futures gap to fill, and how quickly does it fill? Using approximately seven years of session-level data across six liquid futures contracts, we measure same-session fill rates, multi-session fill rates, the dependence on gap size, and whether any of the observed behaviour can be converted into a statistically reliable trading edge after costs.

The objective is not to prove or disprove a market opinion. The objective is to replace an assumption with a measurement.

2. Definitions

An opening gap is defined as a difference between the prior session's settlement price and the current session's opening print larger than a small instrument-specific threshold chosen to filter quote noise. A gap is classified as filled at horizon N if, at any point during the next N trading sessions, price trades through the prior settlement.

Gap size is normalised by instrument using the ten-day average true range so that a one-unit gap in Nasdaq-100 is comparable to a one-unit gap in Copper. Gaps are bucketed into four size bands: small, medium, large, and extreme. Fill probabilities are reported at multiple horizons — same session, one session, three sessions, five sessions, and ten sessions.

3. Markets and Data

Six front-month continuous futures contracts were examined: Nasdaq-100 (NQ), S&P 500 (ES), Gold (GC), Silver (SI), Crude Oil (CL), and Copper (HG). The sample window covers approximately seven years of regular trading hour sessions, with rollovers handled on volume-based crossover and prices back-adjusted to remove roll discontinuities.

Sessions containing exchange holidays, abbreviated trading hours, or major contract specification changes were excluded. The final sample contains tens of thousands of session observations per market.

4. Same-Session Fill Rates

The headline same-session fill rate is meaningfully below the colloquial assumption of near-certainty. Equity index futures show the highest same-session fill rates, consistent with their tighter intraday range distributions. Energy and metals show lower same-session fill rates, with Crude Oil and Silver showing the widest dispersion of outcomes.

Conditioning on gap size reveals a clean monotonic relationship: small gaps fill quickly and almost always during the same session; medium gaps fill the majority of the time; large gaps fill a minority of the time within the session; extreme gaps almost never fill the same day. The pattern holds in every market tested.

Reporting only the same-day number is misleading. A large gap that does not fill in the session has not failed the gap-fill hypothesis — it has only failed it on the shortest horizon.

5. Short-Term Persistence

Large gaps display short-term persistence. After a large opening gap, price more often continues in the direction of the gap during the first session, and frequently extends further during the following one to three sessions, before fill probability begins to dominate. This is the opposite of what a naïve same-day fade rule assumes.

Persistence is consistent with an informational interpretation of gaps. A large opening gap typically reflects new information that arrived outside session hours, and the market continues to price that information for some time before participants who anchor to the prior settlement re-engage.

6. Multi-Session Fill Behaviour

Extending the horizon changes the picture substantially. Fill probability increases steadily with the number of sessions allowed. By the end of the first week, a majority of large gaps in every market in the sample have filled. By approximately the tenth trading session, the cumulative fill rate of large gaps is high across all six markets — typically the large majority of cases — though it never reaches certainty within any finite horizon.

The two-phase behaviour — short-term persistence followed by mean reversion — is the central empirical finding of this study. Same-day fades fight the persistence phase; patient fades benefit from the mean-reversion phase but must survive variable holding periods and adverse excursion.

7. Why Gap Size Matters

Gap size is not just a descriptive variable. It is informational. A large opening gap typically reflects new information that arrived outside session hours — an earnings release, an inventory print, a geopolitical event, a central bank statement. The market is repricing, not mispricing.

Small gaps, in contrast, often reflect overnight noise — liquidity-thin trading in adjacent sessions, ordinary order flow imbalance, or carry-over from the previous close. The prior settlement remains a credible anchor, and same-session fill is the base case.

8. Eight Strategy Implementations

To test whether the observed gap-fill behaviour can be converted into a tradeable edge, eight strategy implementations were evaluated across the same dataset and the same six markets. The variants spanned same-session fades, delayed fades that waited one to three sessions before entering, persistence-following trend variants for the first session, multi-session swing fades with horizons up to ten sessions, and combinations filtered by gap size, volatility, and direction.

Every implementation was evaluated with realistic transaction costs and slippage, with no parameter optimisation beyond the structural choice of the variant. Each variant was tested per instrument and pooled across instruments.

None of the eight implementations produced a statistically reliable edge after costs. Some variants showed high win rates but small average wins relative to losing tail outcomes; others produced positive raw expectancy that did not survive cost assumptions; others were unstable across the six markets. The behaviour is real, but it is not large enough or clean enough to be a strategy on its own.

This is the central practical finding: a pattern is not an edge.

9. Limitations

The seven-year window is long enough to span multiple volatility regimes but is not the full history of any contract. Results are sample-specific and would shift modestly under different windows.

The eight strategy variants are not the universe of refinements practitioners overlay — session context, intermarket signals, discretionary judgement, and product-specific structure are all outside the scope of this study. The point was to test the stand-alone claim and the most natural rules built directly on it, not to certify that no rule anywhere could work.

Fill is defined at the prior settlement only. Partial fill and alternative anchor definitions were not the subject of this study.

10. Discussion

The data are consistent with a behavioural rather than a structural interpretation of "gaps always fill." Participants act on the assumption, and that participation produces some of the observed mean reversion at longer horizons. The same-day version of the claim is too strong; the multi-session version is broadly supported, with the important qualifier that early-session behaviour often runs against it.

From a research standpoint, the negative tradability result is useful. A simple, widely repeated rule does not survive disciplined testing in any of the eight forms in which it was implemented. Negative results narrow the search space and reduce the cost of believing things that are not true.

11. Conclusion

Across six futures markets and seven years of data, gap-fill behaviour is real but two-phase: large gaps often persist in the short term and tend to fill on longer horizons, with most filling within approximately ten trading sessions. Same-day fill rates decline sharply with gap size, and eight strategy implementations built on the observed behaviour did not produce a statistically reliable edge after costs. The colloquial claim is an oversimplification of a conditional, multi-horizon behaviour, and the observed pattern alone is not enough to constitute a strategy.

Future work will extend the framework to overnight versus regular-hours definitions, alternative anchor points, and gap behaviour conditioned on prior-session structure.