Do stocks that run up into an earnings report reverse afterwards?
Do US stocks that ran up (or down) sharply into an earnings report reverse after it, enough to pay trading costs, and does the past-year-winner version reproduce a published 2010 reversal?
Not pre-registered: the rules were recorded retrospectively on 2026-11-21, after every result was seen, so nothing here is a pre-registered test. On 37,010 US earnings events from November 2021 to October 2026, stocks that had run up 15% or more in the 21 sessions before the report did not reverse: they rose a further 52bp on the reaction day (fade-signed -52bp, week-clustered t -2.37), and the post-reaction fade after tier costs was small and insignificant at 5, 10 and 20 sessions (+11.5bp, t 0.40; +50.5bp, t 1.46; +38.2bp, t 0.71). The mirror run-down signal (a 15% fall into the report, faded long) was also insignificant: reaction day -34bp (t -1.02) and +54.5bp at 5, +57.0bp at 10, +47.7bp at 20 (t 0.81, 1.00, 0.67). One of the 110 cells in the study's own bar met its candidate rule — the trade sector at 20 sessions, n=250, +237.5bp, t 2.58 — but it did not clear the study's multiple-testing bar of |t| 3.51, so it is a candidate for a new test, not a finding. The strongest cell, late-cycle at 20 sessions (n=1,286, +268.7bp, t 4.98), has no second half and is one regime spell. A first past-year-winner check (n=3,036) was invalid as run because its ranking window contained the five days being measured; ranked before that window, the pre-report gain fell to +0.58% gross, or +0.36% net after costs (t 0.92), and the published post-report reversal did not appear (+0.35% raw over the reaction plus five sessions, t 0.80). The follow-up tests on 2016-2021 earnings failed: the late-cycle version had no events to test (n=0, recorded as a fail), and the trade-sector version made -141.0bp (t -1.04). No lead survives. The descriptive verdict records what was seen; no pass bar was met.
The report's label is exploratory because the study sits outside the staged campaign and was not pre-registered: its rules were recorded retrospectively on 2026-11-21, after every result was seen. The descriptive verdict records what was seen, not a pass: one of the 110 cells in the study's bar met its candidate-pocket rule (the trade sector at 20 sessions, n=250, +237.5bp, t 2.58) but did not clear the bar's |t| 3.51; the strongest cell (late-cycle, 20 sessions, n=1,286, +268.7bp, t 4.98) has no second half and is one regime spell; the corrected past-year-winner check failed its replication bar (+0.36% net, t 0.92); and the two follow-up tests on 2016-2021 earnings failed or could not be run. 111 round-1 cells were charged to the trial ledger; family_n is 110, the number the study's own multiple-testing bar covered, not a pre-registered family.
Backtested by QS Research
Pre-registration
No pre-registration. The record is retrospective, written on 2026-11-21 after every result was seen. The study's two cell lists were exploratory freezes committed before each round ran (the first before the first slice, the second after round 1 and before round 2); they are not a pre-registration and no result is a pre-registered test. As run: the signal is the 21-session run-up to the session before the report (the close on the day before the report divided by the close 22 sessions earlier, minus one); a run-up of +15% or more is faded short and a run-down of -15% or less is faded long. Outcomes are the raw reaction-day move and the excess over the S&P 500 from the reaction-day close at 5, 10 and 20 sessions, net of a round-trip cost by 20-day dollar-volume tier (40, 20, 10 and 5 basis points); statistics are n, mean, t clustered by ISO week and halves split at the median event date. A cell is a candidate pocket if n is at least 100, t is at least 2 and both halves are above zero. The study examined 108 slice cells and two reaction-day lines, and computed a Bonferroni bar of |t| 3.51 over those 110 cells; the benchmark replication cell was charged to the trial ledger but was outside that bar, which is why 111 cells are charged and family_n is 110. A past-year-winner benchmark measured the five sessions before and after the report for the top decile of the past 12 months' return; its first version was invalid as run because the ranking window contained the measured pre-window, and the corrected version ranked on the 12 months ending at the session before the five pre-report sessions. Three follow-up trials were frozen after round 1 and before any round-2 run: the corrected benchmark (fail unless the pre-report net mean is above zero with t at least 2 and both halves positive), and the late-cycle and trade-sector 20-session fades on 2016-2021 events only (each needs n at least 100, a positive net mean, t at least 2 and both halves positive; an untestable trial is a fail). The three follow-up trials are judged at their own bars and are not in the 110-cell family. The trial ledger holds 114 rows: the 111 round-1 cells and the three follow-up trials.
Registered 2026-11-21Trials · 111 trials · 0 passed · judged at N = 110 trials; 2965 trials run in total
| Trial | Verdict | |
|---|---|---|
| crowd_long (|+5|all|all | RESULT | counted |
| crowd_long (|+5|liquidity|T1 $5-25M | RESULT | counted |
| crowd_long (|+5|liquidity|T2 $25-100M | RESULT | counted |
| crowd_long (|+5|liquidity|T3 $100M-1B | RESULT | counted |
| crowd_long (|+5|liquidity|T4 >$1B | RESULT | counted |
| crowd_long (|+5|regime|expansion | RESULT | counted |
| crowd_long (|+5|regime|late-cycle | RESULT | counted |
| crowd_long (|+5|regime|recession | RESULT | counted |
| crowd_long (|+5|regime|recovery | RESULT | counted |
| crowd_long (|+5|sector|construction | RESULT | counted |
| crowd_long (|+5|sector|finance/real estate | RESULT | counted |
| crowd_long (|+5|sector|manufacturing | RESULT | counted |
| crowd_long (|+5|sector|mining/energy | RESULT | counted |
| crowd_long (|+5|sector|other | RESULT | counted |
| crowd_long (|+5|sector|services/tech | RESULT | counted |
| crowd_long (|+5|sector|trade | RESULT | counted |
| crowd_long (|+5|sector|transport/utilities | RESULT | counted |
| crowd_long (|+5|sector|unmapped | RESULT | counted |
| crowd_long (|+10|all|all | RESULT | counted |
| crowd_long (|+10|liquidity|T1 $5-25M | RESULT | counted |
| crowd_long (|+10|liquidity|T2 $25-100M | RESULT | counted |
| crowd_long (|+10|liquidity|T3 $100M-1B | RESULT | counted |
| crowd_long (|+10|liquidity|T4 >$1B | RESULT | counted |
| crowd_long (|+10|regime|expansion | RESULT | counted |
| crowd_long (|+10|regime|late-cycle | RESULT | counted |
| crowd_long (|+10|regime|recession | RESULT | counted |
| crowd_long (|+10|regime|recovery | RESULT | counted |
| crowd_long (|+10|sector|construction | RESULT | counted |
| crowd_long (|+10|sector|finance/real estate | RESULT | counted |
| crowd_long (|+10|sector|manufacturing | RESULT | counted |
| crowd_long (|+10|sector|mining/energy | RESULT | counted |
| crowd_long (|+10|sector|other | RESULT | counted |
| crowd_long (|+10|sector|services/tech | RESULT | counted |
| crowd_long (|+10|sector|trade | RESULT | counted |
| crowd_long (|+10|sector|transport/utilities | RESULT | counted |
| crowd_long (|+10|sector|unmapped | RESULT | counted |
| crowd_long (|+20|all|all | RESULT | counted |
| crowd_long (|+20|liquidity|T1 $5-25M | RESULT | counted |
| crowd_long (|+20|liquidity|T2 $25-100M | RESULT | counted |
| crowd_long (|+20|liquidity|T3 $100M-1B | RESULT | counted |
| crowd_long (|+20|liquidity|T4 >$1B | RESULT | counted |
| crowd_long (|+20|regime|expansion | RESULT | counted |
| crowd_long (|+20|regime|late-cycle | RESULT | counted |
| crowd_long (|+20|regime|recession | RESULT | counted |
| crowd_long (|+20|regime|recovery | RESULT | counted |
| crowd_long (|+20|sector|construction | RESULT | counted |
| crowd_long (|+20|sector|finance/real estate | RESULT | counted |
| crowd_long (|+20|sector|manufacturing | RESULT | counted |
| crowd_long (|+20|sector|mining/energy | RESULT | counted |
| crowd_long (|+20|sector|other | RESULT | counted |
| crowd_long (|+20|sector|services/tech | RESULT | counted |
| crowd_long (|+20|sector|trade | RESULT | counted |
| crowd_long (|+20|sector|transport/utilities | RESULT | counted |
| crowd_long (|+20|sector|unmapped | RESULT | counted |
| crowd_short |+5|all|all | RESULT | counted |
| crowd_short |+5|liquidity|T1 $5-25M | RESULT | counted |
| crowd_short |+5|liquidity|T2 $25-100M | RESULT | counted |
| crowd_short |+5|liquidity|T3 $100M-1B | RESULT | counted |
| crowd_short |+5|liquidity|T4 >$1B | RESULT | counted |
| crowd_short |+5|regime|expansion | RESULT | counted |
| crowd_short |+5|regime|late-cycle | RESULT | counted |
| crowd_short |+5|regime|recession | RESULT | counted |
| crowd_short |+5|regime|recovery | RESULT | counted |
| crowd_short |+5|sector|construction | RESULT | counted |
| crowd_short |+5|sector|finance/real estate | RESULT | counted |
| crowd_short |+5|sector|manufacturing | RESULT | counted |
| crowd_short |+5|sector|mining/energy | RESULT | counted |
| crowd_short |+5|sector|other | RESULT | counted |
| crowd_short |+5|sector|services/tech | RESULT | counted |
| crowd_short |+5|sector|trade | RESULT | counted |
| crowd_short |+5|sector|transport/utilities | RESULT | counted |
| crowd_short |+5|sector|unmapped | RESULT | counted |
| crowd_short |+10|all|all | RESULT | counted |
| crowd_short |+10|liquidity|T1 $5-25M | RESULT | counted |
| crowd_short |+10|liquidity|T2 $25-100M | RESULT | counted |
| crowd_short |+10|liquidity|T3 $100M-1B | RESULT | counted |
| crowd_short |+10|liquidity|T4 >$1B | RESULT | counted |
| crowd_short |+10|regime|expansion | RESULT | counted |
| crowd_short |+10|regime|late-cycle | RESULT | counted |
| crowd_short |+10|regime|recession | RESULT | counted |
| crowd_short |+10|regime|recovery | RESULT | counted |
| crowd_short |+10|sector|construction | RESULT | counted |
| crowd_short |+10|sector|finance/real estate | RESULT | counted |
| crowd_short |+10|sector|manufacturing | RESULT | counted |
| crowd_short |+10|sector|mining/energy | RESULT | counted |
| crowd_short |+10|sector|other | RESULT | counted |
| crowd_short |+10|sector|services/tech | RESULT | counted |
| crowd_short |+10|sector|trade | RESULT | counted |
| crowd_short |+10|sector|transport/utilities | RESULT | counted |
| crowd_short |+10|sector|unmapped | RESULT | counted |
| crowd_short |+20|all|all | RESULT | counted |
| crowd_short |+20|liquidity|T1 $5-25M | RESULT | counted |
| crowd_short |+20|liquidity|T2 $25-100M | RESULT | counted |
| crowd_short |+20|liquidity|T3 $100M-1B | RESULT | counted |
| crowd_short |+20|liquidity|T4 >$1B | RESULT | counted |
| crowd_short |+20|regime|expansion | RESULT | counted |
| crowd_short |+20|regime|late-cycle | RESULT | counted |
| crowd_short |+20|regime|recession | RESULT | counted |
| crowd_short |+20|regime|recovery | RESULT | counted |
| crowd_short |+20|sector|construction | RESULT | counted |
| crowd_short |+20|sector|finance/real estate | RESULT | counted |
| crowd_short |+20|sector|manufacturing | RESULT | counted |
| crowd_short |+20|sector|mining/energy | RESULT | counted |
| crowd_short |+20|sector|other | RESULT | counted |
| crowd_short |+20|sector|services/tech | RESULT | counted |
| crowd_short |+20|sector|trade | RESULT | counted |
| crowd_short |+20|sector|transport/utilities | RESULT | counted |
| crowd_short |+20|sector|unmapped | RESULT | counted |
| crowd_long (|reaction_day | RESULT | counted |
| crowd_short |reaction_day | RESULT | counted |
| aboody_winners_top_decile | RESULT | counted |
| pre-5 excess, gross | RESULT | not counted |
| pre-5 excess, net of tier cost | FAIL | not counted |
| reaction + 5 after, raw | RESULT | not counted |
| reaction + 5 after, net short | RESULT | not counted |
| b: crowd_long, late-cycle, +20 | FAIL | not counted |
| c: crowd_long, trade sector (SIC 50-59), +20 | FAIL | not counted |
Limitations
- Not pre-registered: the record is retrospective, written on 2026-11-21 after every result was seen. The two pre-run cell lists are exploratory freezes, not a pre-registration, and no result is a pre-registered test.
- Today's listed companies only: companies that later delisted are missing (survivorship), and ticker renames split histories.
- One era and mostly one regime: reaction days from 2021-11-23 to 2026-10-27. The late-cycle cell is a single regime spell with no second half, and the expansion cells have no first half.
- Costs are frozen tier round trips (40, 20, 10 and 5 basis points by 20-day dollar volume), not measured fills, spread or borrow; the reaction-day line is before costs.
- The study's multiple-testing bar covers 110 cells (108 slices and 2 reaction-day lines), but 111 cells were charged: the benchmark cell was charged and kept out of the bar. 24 of the 108 slice cells have fewer than 100 events. The analysis printed the 3.51 bar 'for context only', and the write-up then used it to judge the one candidate pocket.
- The first past-year-winner result is invalid as run: the ranking window contains the five pre-report sessions being measured, which biases the pre-report number upward. The corrected cell ranks before that window and is a new trial, not a re-cut; it ranks the 29,889 events with a usable 258-session ranking (2,989 winners), and the other 7,121 are not ranked.
- The post-report comparison with the 2010 paper is not exactly like-for-like: the study's 'reaction plus 5 after' covers six sessions (the reaction day and the next five), while the paper's -1.86% covers the reaction day plus four sessions (days 0..+4). The study's five-sessions-after figure is -0.17% (t -0.64), and the reaction-plus-five figure is +0.35% (t 0.80); neither is significantly different from zero.
- The 30,192 2016-2021 out-of-sample events were built outside the engine; the builder and event file are not committed, so that sample cannot be rebuilt from the published record. The overlap check reproduced 7,032 of 7,034 engine events, but the builder also has 1,510 events the engine does not (536 inside the engine's observable window), and like-for-like values are not identical: the largest differences are 1.16% on the reaction-day move, 0.96% on the 21-session run-up and 1.49% on the 20-session excess.
- No holdout was reserved. The 2016-2021 test is a new out-of-sample extension decided after round 1, not a sealed holdout, and no one opened a sealed period. Its late-cycle version could not be tested because no month of 2016-2021 is labelled late-cycle (n=0); it is recorded as a fail by the n-at-least-100 rule, which is not evidence against the effect. The trade-sector version reverses sign out of sample: n=264, -141.0bp, t -1.04, halves +270.2/-267.2.
- The report's provenance records no pinned data snapshot and no release commit. The study's own record names a panel snapshot key and an engine base commit with a dump-only patch, but neither is a content hash in the report, so the numbers cannot be replayed byte-for-byte from the published provenance. The engine's event dump is not committed and the analysis scripts were removed from the folder (they remain in git history); the result files, not the scripts, are the binding record.
The numbers: every trial's checks, metrics, curve and yearly results, and the provenance
1 published version · current b400fd3a795beaa8, published 2026-11-21