Gap down below 20dma — what happened next
The answer
≥300 effective observations and the same sign in both halves of the period. Enough to report, but not enough to lead with. How grades are decided.
The interval includes zero. On this sample the pattern is not distinguishable from the market.
This pattern showed no clear gap
This pattern was written down on 2026-09-05 as part of a set chosen in advance. The measured gap versus the matched market control was −0.5pp. Its 95% interval runs from −1.5 to +0.5pp. The interval includes zero. On this sample, across 20,564 effective observations, the pattern could not be told apart from the market. Every pattern written down in advance gets a page, whatever the result.
In plain English. The stock opened at least 2% below yesterday's close, on a day when it was already trading below its 20-day average price.
open[t] / close[t-1] - 1 <= -0.02 AND close[t-1] < sma(close, 20)[t-1]
Written down on 5 September 2026, before it was measured. The headline uses the middle setting chosen in advance: a gap of at least 2% and a 20-day average price. See the registration log.
The sample this rests on
The sample
| Design | Same-day correlation | Design effect | Effective observations |
|---|---|---|---|
| Raw rate, no control | 0.317 | 4.05 | 7,168 |
| Paired against the control (published) | 0.043 | 1.41 | 20,564 |
The public daily bhavcopy has no sector data for past dates. So the sector-concentration check could not be run, and there is no sector-matched version. A pattern that is really a whole sector moving together would not be caught by this sample. What that means.
Every horizon, with its control
| Trading days | Observed | Matched control | Gap | 95% interval | Median change | Excess median | Break-even at 2% |
|---|---|---|---|---|---|---|---|
| 1 | 47.7% | 47.4% | +0.3pp | −0.8 to +1.6pp | 0.00% | +0.11% | 55.6% |
| 3 | 48.6% | 48.7% | −0.1pp | −1.2 to +1.3pp | −0.14% | −0.13% | 55.6% |
| 5 | 49.2% | 49.8% | −0.5pp | −1.5 to +0.5pp | −0.10% | −0.15% | 55.6% |
| 10 | 49.0% | 50.2% | −1.2pp | −2.7 to 0.0pp | −0.18% | −0.24% | 55.6% |
| 20 | 50.5% | 51.9% | −1.3pp | −3.0 to +0.1pp | +0.18% | −0.52% | 55.6% |
The distribution is the answer
| Days | p10 | p25 | Median | p75 | p90 | Trimmed mean | Best point reached | Worst point reached |
|---|---|---|---|---|---|---|---|---|
| 1 | −3.8% | −1.8% | 0.00% | +1.9% | +4.4% | +0.22% | +2.1% | −1.9% |
| 3 | −9.0% | −3.8% | −0.14% | +3.5% | +7.8% | −0.24% | +3.7% | −3.6% |
| 5 | −11.1% | −4.8% | −0.10% | +4.7% | +10.0% | −0.31% | +4.9% | −4.7% |
| 10 | −14.8% | −6.3% | −0.18% | +6.2% | +13.3% | −0.64% | +6.8% | −6.4% |
| 20 | −17.1% | −7.8% | +0.18% | +8.8% | +19.0% | +0.46% | +9.6% | −8.8% |
Year by year, and by market state
| Year | 2011 | 2012 | 2013 | 2014 | 2015 | 2016 | 2017 | 2018 | 2019 | 2020 | 2021 | 2022 | 2023 | 2024 | 2025 | 2026 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Gap | −3.6pp | −1.3pp | −4.4pp | −7.9pp | +1.2pp | −1.1pp | −4.9pp | +0.9pp | +0.8pp | −1.4pp | +1.8pp | +0.7pp | −7.3pp | +0.1pp | −0.0pp | −0.4pp |
| Effective n | 300 | 646 | 586 | 380 | 913 | 1,083 | 480 | 2,036 | 740 | 2,129 | 1,164 | 2,367 | 834 | 1,572 | 2,115 | 1,500 |
| Market vs its 200-day average | Volatility | Gap | Effective n |
|---|---|---|---|
| Above | lowest third | +37.2pp | 6 |
| Above | middle third | +3.0pp | 46 |
| Above | highest third | −2.4pp | 159 |
| Below | lowest third | −1.9pp | 4,065 |
| Below | middle third | −0.6pp | 6,867 |
| Below | highest third | −0.0pp | 8,471 |
Robustness
Does it survive being pushed?
This is one of 25 pre-registered variants of this pattern. It is also one of 1,125 tests on this site. The figures below are adjusted for that.
| First half of the period | −0.7pp | Split at the median event date. |
|---|---|---|
| Second half of the period | −0.3pp | |
| Same sign in both halves | yes | An effect present in only one half is a regime, not a base rate. |
| Years with the same sign | 10 of 16 | |
| Non-overlapping subsample | −0.2pp | One event per security per window, so no two observations share days. |
| After published charges | −0.3pp | A move counts only if it cleared the round-trip charge. |
| Censoring worst case | −0.7pp | Every censored observation counted as a loss. |
| Median across the parameter grid | −0.6pp | The headline sits close to the grid median. |
| Entering one session later | −1.1pp | A multi-session pattern survives it. An artefact of the entry bar does not. |
| Placebo: same securities, unrelated dates | +0.9pp | Should be near zero. If it is not, the method itself is creating a gap. |
| Sector-matched variant | not available | Sector classification is not reconstructible point-in-time from the public bhavcopy. |
| Multiple-testing adjusted p | 0.2384 | Does not survive the false-discovery adjustment. |
| gap threshold | 16 | 18 | 20 | 22 | 24 |
|---|---|---|---|---|---|
| 0.016 | −0.1pp | −0.1pp | −0.2pp | −0.3pp | −0.4pp |
| 0.018 | −0.5pp | −0.4pp | −0.6pp | −0.7pp | −0.7pp |
| 0.02 | −0.4pp | −0.3pp | −0.5pp | −0.7pp | −0.8pp |
| 0.022 | −0.6pp | −0.4pp | −0.6pp | −0.8pp | −0.9pp |
| 0.024 | −0.6pp | −0.6pp | −0.7pp | −0.9pp | −1.0pp |
A real pattern is a plateau across neighbouring settings. A single bright cell in a field of noise is an artefact of one threshold. The outlined cell is the pre-declared centre. The headline is taken from that cell whether or not it is the strongest.
The quality checks
| Check | Required | This sample | Result |
|---|---|---|---|
| Effective sample size | >= 100 | 20564 | passed |
| Distinct event dates | >= 60 | 2736 | passed |
| Distinct securities | >= 25 | 1969 | passed |
| Calendar span | >= 5 years | 14.8 years | passed |
| Largest share in one quarter | <= 20.0% | 11.3% | passed |
| Largest share from one security | <= 5.0% | 0.3% | passed |
| Censoring rate | <= 2.0% | 0.4% | passed |
| Events on relaxation rung 3-4 | <= 20.0% | 5.3% | passed |
What this does not tell you
- It does not tell you what happens next. A base rate describes a sample of past events. The future is not drawn from that sample.
- It does not identify any security. These figures are aggregates across hundreds of securities over many years. No individual security's behaviour can be inferred from them.
- It does not account for your costs, your position size or your taxes. It also does not account for your ability to sit through the losses. The break-even column shows the frequency required to cover published charges alone. It excludes slippage, impact, and everything specific to you.
- A gap versus the control is not an edge. It is an observation about a sample, measured with uncertainty. That observation may not persist.
- The interval matters more than the number. Where the interval includes zero, this sample does not distinguish the pattern from the market.
- Sample size is not what it looks like. Events on the same day are not independent observations. Use the effective sample size shown above.
Questions about this page
What is the matched control, and why does every number have one?
For every event, comparable securities were drawn from the same date — same price band, same realised-volatility band, same traded-value band — that did not themselves meet the condition. The Indian market rose in roughly two of every three rolling twenty-day windows over this period, so any long-side pattern shows a frequency near 60% and means nothing on its own. The publishable number is the gap between the two, not either one.
Why is the effective sample smaller than the number of events?
Events on the same day are not independent observations. They share one market move. Effective sample size accounts for that link. Each interval and check uses it instead of the raw count.
Does this tell me what will happen next time?
No. These figures count past events in a sample, with uncertainty beside each result. They do not predict the next event or recommend a security.
Would this have covered costs?
The break-even column shows the frequency a symmetric 2% target and stop would need just to cover published charges. On this page that level is 55.6%, against an observed 49.2%. It excludes slippage, impact and anything specific to you, all of which move it the wrong way.
Why show a result that shows nothing?
Because the pattern was written down before it was measured, and every pattern on that list gets a page, whatever the result. Showing only what worked would hide how often nothing was found.
Other registered patterns
Gap up above 20dma
The stock opened at least 2% above yesterday's close, on a day when it was already trading above its 20-day average price.
Open →Close crosses above 20dma
The stock closed back above its 20-day average price after having spent at least five of the previous ten sessions below it.
Open →Early session of the month
The third trading session of a calendar month, for a stock trading above its hundred-day average price.
Open →Market wide down day
A session on which the median eligible security fell at least 2%, for a stock trading above its hundred-day average price.
Open →What the terms mean · Work out the frequency your own bracket needs
Next stepHow this figure was produced
The point-in-time universe, the matched control drawn from the same dates, the interval, and the quality checks a sample must pass before any rate is shown.
Read the methodologySource. NSE daily bhavcopy. Data through 2026-08-06. Stock list: point-in-time, as it stood on each date. Measured on 2026-09-05. How it is measured · What the terms mean · Download these figures, under the CC BY 4.0 licence.