Low-float pre-market gap-ups with no SEC filing: chase or fade?
When a US stock with under 20M shares outstanding is up 30% or more at 09:25 ET with no SEC current report overnight, does buying or shorting at the open pay after costs at the close, the next close or five sessions later?
No. None of the 9 pre-registered trials passed. Buying the gap lost money in every counted year (-13% to -27% a trade after costs). Shorting at the open and covering one or five sessions later made money on average after costs (+7% and +12% a trade, positive lower bounds, about 78% winners), but failed the deflated Sharpe test (0.10 against 0.80): a heavy tail of squeezes keeps the result from standing out from luck given the trials already run. Shorting and covering the same day was a coin flip. The holdout stays sealed.
Backtested by QS Research
Pre-registration
Nine trials fixed before any return was read: long and short at the 09:30 open, exiting at the close, the next close or the fifth close, plus the short on names with a dilution flag. 1,319 events from 2016 to October 2026, split 60/20/20 by date; discovery is judged at alpha/9 with week-clustered bounds and a deflated Sharpe. Costs are the spread by liquidity tier, the opening minute's range and, for shorts, borrow at 100% a year. Two amendments before any return: the news screen uses SEC current reports only, because the headline archive is too thin before 2022, and the data pins and stage boundaries were recorded.
Registered 2026-11-09Amendments
- 2026-11-10 The equity-line phrase test: a whole-word match, and only a termination notice ends the flag. approved: Approved 2026-11-10, before any data was pinned
- 2026-11-14 No news means no SEC current report; headlines are reported as a slice, not screened. approved: Decided 2026-11-14 from coverage counts, before any return
- 2026-11-14 Data pins, the 1,319-event table and the stage boundaries, recorded before the run. approved: Recorded before the counted run
A holdout sample is sealed: its results stay unread until the study is final.
Trials · 9 trials · 0 passed · judged at N = 9 trials; 2018 trials run in total
| Trial | Verdict | |
|---|---|---|
| lf-long-d0 | FAIL | counted |
| lf-long-d1 | FAIL | counted |
| lf-long-d5 | FAIL | counted |
| lf-short-d0 | FAIL | counted |
| lf-short-d1 | FAIL | counted |
| lf-short-d5 | FAIL | counted |
| lf-short-dil-d0 | FAIL | counted |
| lf-short-dil-d1 | FAIL | counted |
| lf-short-dil-d5 | FAIL | counted |
Limitations
- Every short pays a 100% a year borrow floor and assumes shares could be located; most of these names are hard to borrow.
- No news means no SEC current report; wire headlines and social media are not screened.
- Float is shares outstanding from SEC cover pages, not free float, and goes stale between filings.
- Ticker changes and re-used symbols on the event day are not checked.
- Halts are inferred from gaps in the minute bars.
- One regime: discovery is 2016 to mid-2022, including the 2020-2021 retail wave.
The numbers: every trial's checks, metrics, curve and yearly results, and the provenance
1 published version · current 49e6189e30944214, published 2026-11-15