Almost every gap to the moving average eventually closes. But across 697 episodes on 55 Bybit perpetuals, most close because the average catches up with price — and in more than one case in three, price kept moving in the gap's direction.
A price that has run too far from its moving average eventually meets it again. That is the idea behind the gap fill, and it holds: over nearly three years of data, almost every gap closes. But it hides a costly confusion. "The gap will fill" gets read as "price will come back". Those are not the same thing, and the difference can be measured.
The previous article set out a comparable unit for measuring stretch: the gap to the moving average divided by the asset's own volatility. It showed that this gap does not predict the direction of the next move. It left one question open: when a gap closes, who does the travelling — price, or the average?
A gap between price and its moving average can disappear in two ways, which charts easily blur together.
The two nearly always combine, and the share of each can be computed exactly. Between the day a gap opens and the day it closes, its total reduction is the sum of two terms: the distance price travelled toward the average, and the distance the average travelled toward price. Comparing them is enough.
Measured on 55 Bybit perpetuals among the sixty most traded, with up to 1,000 daily candles per contract — from late December 2023 to September 2026 — for 41,350 observations.
Result: 697 episodes across 54 contracts, 656 of them closed within the period — 387 gaps above the average, 310 below.
The first finding confirms the gap-fill intuition: a large gap nearly always closes in the end. The second tempers it considerably.
| Closed within | Gap above | Gap below | Control |
|---|---|---|---|
| 5 days | 9.0 % | 5.5 % | 56.9 % |
| 10 days | 33.4 % | 15.8 % | — |
| 20 days | 63.7 % | 53.9 % | 85.4 % |
| 60 days | 99.3 % | 96.1 % | — |
| Median time | 14 days | 19 days | 4 days |
The control is an ordinary day on which price sits only slightly away from its average: it crosses back in 4 days at the median. A large gap takes three to five times longer. Fewer than one gap in ten closes within the week; most need two to three weeks.
In other words, betting on a quick fill means betting against the numbers. The fill is likely, but it is not imminent.
This is the heart of the measurement. For every closed gap, the share due to price is separated from the share due to the average:
| Closed gaps | Above | Below |
|---|---|---|
| Episodes | 346 | 310 |
| Closed mainly by price | 45.4 % | 32.9 % |
| Closed mainly by the average | 54.6 % | 67.1 % |
| Equal weight per contract, by price | 43.2 % | 31.8 % |
| Price kept moving in the gap's direction | 38.4 % | 37.4 % |
Most gaps close through time first. Price does come back a little, but the average covers most of the distance. The result does not hinge on a few contracts: giving each one the same weight moves the proportions by two points at most.
The last line says the most. In more than one case in three, price kept moving in the direction of the gap — higher for a gap above, lower for a gap below — and the gap closed anyway, because the average moved faster. A trader who sold "the gap will fill" was right about the gap and lost on the price.
For gaps above the average, the price move between the day the gap opens and the day it closes gives two figures that contradict each other:
This is exactly the trap identified in the stretch article: in this universe, the mean and the median of the same sample point in opposite directions. Anyone reading only the mean would conclude that an upside gap announces further gains. The typical case says the opposite — a moderate pullback — and 61.6 % of gaps close with price below its level on the opening day.
The 4-standard-deviation threshold is a choice. To check that it does not manufacture the result, the measurement was run again with two other thresholds:
| Threshold | Side | Time | By price | Price kept going |
|---|---|---|---|---|
| 80th (3.0) | above | 12 d | 48.8 % | 33.0 % |
| 80th (3.0) | below | 17 d | 38.6 % | 35.2 % |
| 90th (4.0) | above | 14 d | 45.4 % | 38.4 % |
| 90th (4.0) | below | 19 d | 32.9 % | 37.4 % |
| 95th (4.9) | above | 18 d | 33.5 % | 41.9 % |
| 95th (4.9) | below | 19 d | 37.9 % | 31.6 % |
In all six configurations, the share closed mainly by price stays between 33 % and 49 %: never a majority. And price keeps moving in the gap's direction in roughly one case in three, without exception.
One caveat, in fairness: at the 90 % threshold, gaps below the average look more "corrected through time" than gaps above. At 95 %, the order flips. That asymmetry is therefore not a finding: it depends on the setting, and no rule should be drawn from it.
The same measurement on the EMA 200 yields only 86 episodes — too few for reliable percentages. It still points the same way: a majority of gaps closed mainly by the average. What changes is the timescale: a median of 87 days for a gap above and 140 days for a gap below. On the long moving averages, a fill is counted in quarters.
The screener's stretch column reports how far price sits from its moving averages. In the light of this measurement, a large stretch reads as follows:
It is a nuance charts render poorly. In hindsight, one sees the closed gap and remembers a "reversion to the mean". One forgets that one time in three, price never came back.
It does not say where to enter. It describes how gaps close, not the point at which a gap stops widening.
It also ignores what surrounds the average. An average that coincides with an old horizontal floor or ceiling may pull price harder than an isolated average: that is the question of confluence, which will get its own article.
Finally, it covers the most traded contracts with enough history. Recent listings, the majority on Bybit, are absent by construction.
What it does establish is enough to change a common reading. "The gap will fill" is a forecast about the average as much as about price — and in most cases, it is mainly the average that moves.
This is not investment advice. This article sets out technical-analysis concepts for educational purposes. Past performance is no guide to future performance, and trading crypto-assets carries a risk of losing the entire capital committed. Legal information.