The claim that gaps always fill is the most durable one in technical trading, and as usually stated
it is unfalsifiable, because no horizon is attached to it. Given enough time a stock will touch almost any
price, so the claim is either trivially true or empty depending on how long you are prepared to
wait.
Attach a horizon and it becomes answerable. We have 181 sessions of daily bars for every US symbol;
after screening for a prior close of at least $1 and $1,000,000 of turnover, 144,286 gaps of 2% or
more. Regular-session highs and lows only, which we checked against minute bars first, since a daily
bar that included pre-market prints would inflate every fill rate here.
1. Same-day fill is the minority, at every size
| Up gap | Events | Same day |
|---|
| 2-5% | 55,923 | 39.2% |
| 5-10% | 13,691 | 28.8% |
| 10-20% | 4,067 | 22.4% |
| 20%+ | 1,774 | 12.3% |
| Down gap | Events | Same day |
|---|
| 2-5% | 53,483 | 42.0% |
| 5-10% | 11,088 | 29.6% |
| 10-20% | 3,202 | 17.1% |
| 20%+ | 1,058 | 3.9% |
Even the smallest gaps in this set, a median of 2.8%, close the distance on the same session under
half the time. By 20% the up gaps are down to one in eight, and the down gaps to one in twenty-five.
Whatever the rule means, it does not mean today.
2. Over sixty sessions it is still not always
| Cumulative fill | 1 | 5 | 20 | 60 |
|---|
| up 2-5% | 39.2% | 69.2% | 82.4% | 87.5% |
| up 5-10% | 28.8% | 61.2% | 76.9% | 84.2% |
| up 10-20% | 22.4% | 54.4% | 70.6% | 78.0% |
| up 20%+ | 12.3% | 35.0% | 51.8% | 62.9% |
| down 2-5% | 42.0% | 70.6% | 83.4% | 90.4% |
| down 5-10% | 29.6% | 56.6% | 72.1% | 80.7% |
| down 10-20% | 17.1% | 38.9% | 54.2% | 65.1% |
| down 20%+ | 3.9% | 15.8% | 27.0% | 37.9% |
Three months is a long horizon for a claim people make about an intraday chart, and even there the
large gaps have not filled. After 60 sessions, 37.9% of 20%+ down gaps had filled, so most were still open.
Notice also where the curve flattens. For small gaps most of the eventual filling is done within five
sessions; the move from 20 to 60 sessions adds five to seven points. The large gaps are still climbing
at the end of the window, which means their true rate is somewhat higher than shown, and also that
they are taking months rather than days.
3. The asymmetry is the finding
Small gaps behave almost identically in both directions. 2-5% up fills 87.5% within 60 sessions and
2-5% down fills 90.4%. Nothing to see.
The directions separate as the gap grows, and by 20% they have separated completely.
| 20%+ gaps | Same day | Within 60 |
|---|
| Up | 12.3% | 62.9% |
| Down | 3.9% | 37.9% |
A large up gap is 1.7 times more likely to be filled within three months than a large down gap. Same
day, it is roughly three times more likely.
That direction is worth sitting with, because the naive expectation runs the other way. A stock opening
31.3% above yesterday's close, the median of that bucket, is expensive and might be expected to give
it back; one opening 27.4% below is cheap and might be expected to bounce. The data says the opposite, and the reason is structural rather than
behavioural. A large down gap is usually the market repricing something permanent, and once a company
is worth less it has no obligation to visit its old price again. A large up gap frequently is
enthusiasm, and enthusiasm decays.
We have measured other parts of this same low end. Stocks that
fall below $1 are working through a
delisting clock, and the ones that lose it
stop existing. A 20%+ down gap
is often the first session of that story, which is exactly why the price does not come back.
4. Why the folklore survives
Because it is mostly true about the gaps people mostly see.
Gaps of 2-5% are 109,406 of the 144,286 events here, 75.8% of the total. Those fill 87.5% and 90.4%
of the time within three months. If your experience of gaps is the ordinary ones, the rule looks
excellent, and you will carry it into the rare cases where it is worst.
The rule is not wrong so much as it is quoted at the wrong end of the distribution.
What this does not tell you
It does not say a gap is likely to fill on any particular day, and it is not a reason to take the
other side of one. Fill here means the previous close was touched at some point, by any amount, for
any length of time. A price that is touched once intraday and rejected counts as filled on this
measure and would not have been a workable exit.
It also says nothing about what happens in between. A 20%+ down gap that eventually fills after dozens of
sessions may have halved again first, and this measurement would not see it.
Limitations
Of 2,173,336 symbol-days screened, 148,802 failed the $1 prior-close test and 909,386 the
$1,000,000 turnover test. A further 970,862 had a gap smaller than 2% and were not counted as gap
events.
325 events were censored: the symbol stopped appearing in the daily bars before filling and before
the 60 sessions elapsed. They are excluded from the numerator but remain in the denominator, which
biases the fill rates down very slightly. The effect is largest in the 20%+ down bucket, where 41 of
1,058 events are censored, and it runs in the same direction as the finding rather than against it.
The 60-session horizon truncates the largest gaps, whose curves have not flattened. Their true
eventual fill rate is higher than reported here; the shape of the comparison between up and down is
not affected, since both are truncated identically.
Fills are detected from daily highs and lows, so a gap filled and re-opened within the same session
is recorded once. We did not measure how long the price stayed at the filled level, which is the
number that would matter to anyone trading it.