Third run: among a day's biggest fallers, do shares closing at a year's low rank lower the next day?
Study 92 · Not supported · recorded 9 November 2026 · answer: Not at the size tested · full paper with figures
The question
Fallers: each day's fifth of shares, priced at five or more in US currency at the previous close with next-day prices, that fell most against the market fund. On days in 2018 to 2022 with five or more closing at a year's low and fifty not, did those at the low rank lower on next-day return against the market fund?
Rules fixed before the result
Fixed before any outcome: twelve thousandths of the ranking as the smallest gap worth reporting, where the rise-side study used ten; a day counted when five or more fallers closed at a year's low, eight as the fallback; and checks that stop the study first. A 95% range above zero reads as a lower rank at the low, one below zero as the reverse, one touching zero under the bar as not at the size tested, others as not settled.
Data and method
Daily we took the fifth of shares falling most against the market fund, adjusted their next-day returns by a straight-line fit for the day's fall and for where the close sat in the day's range, and ranked them among fallers at a similar multiple of their usual volume. The measure is the others' average rank less that of fallers closing at or below their lowest closing price of the past year, with a 95% range allowing for correlated nearby days.
The answer
Not at the size tested. Over 968 days from 2018 to 2022, fallers that closed at a year's low ranked on average 0.003 lower next day than other fallers, on a 0 to 1 scale where 0.012, the bar set in advance, is 1.2 percentile points; the 95% range, from 0.004 higher to 0.009 lower, lies under it, so smaller gaps remain compatible. The test could not read, at the error rate set in advance, which part of the day carried the result, nor whether it goes beyond volume. On the same days the same measure for the biggest risers that closed at a year's high, from an earlier study, was larger by 0.012 (95% range 0.002 to 0.021), but a stricter 99% check did not separate them; that comparison rests on the paired measure, not on the bar.
Main estimate
0.003 fraction (95% range -0.004 to 0.009), from 968 observations
What it changes
This is an earlier study's test run a third time: the first run stopped for want of qualifying days, the second at a check of its own plan that its larger smallest gap had made impossible to pass, and this plan corrected that check and two made-up test cases sized for the old gap, changing nothing else. No trading decision or rise-side note changes. By the owner's condition this was the design's last attempt before older price data arrives.
What remains uncertain
Earlier studies used these years, so this is a check inside our development data, not an independent test. Each day counts once, allowing for shares moving together on a date, and the small average leaned on one year. Side readings were each read alone, and with no true gap at least one shows a pattern in 95.7 percent of runs resampled from 2013 to 2017, so the larger gap among the quietest-volume fallers is a lead, not a finding.
How it was made
Run took 52 minutes; independent code review: approve; the recorded calculation reruns to the same result.
Fingerprints
plan c81f91423b650115b633003eea07a4dcbd1c73beb8c88055cf7e1d689a68869e result dc1817f891ac561e24a59cb974df2c252a7ac05200d58bfc590fd9af03fe765e
Exploratory work on development data; not a performance claim.