Small caps that close strong: does holding overnight pay?
After a $1-$5 US stock rises 10% or more on three times its usual volume and closes in the top 20% of its range, does buying or shorting at the close pay after costs at the next morning's open?
No. None of the 4 pre-registered trials passed. Holding overnight earned 1.1% gross close to open but 0.2% after costs, with a lower bound below zero (41% winners); on +30% days it lost 0.1%. Fading the strong close lost 2.5% a trade. The flat-by-the-close default stands. The holdout stays sealed.
Backtested by QS Research
Pre-registration
Four trials fixed before any return was read: long and short from the official close to the next official open, on all events and on +30% days. 2,705 events from 2016 to October 2026, split 60/20/20 by date; discovery is judged at alpha/4 with week-clustered bounds and a deflated Sharpe. Costs are the spread by liquidity tier both sides and, for shorts, borrow at 100% a year. One amendment before any return: the data pins, rule readings and stage boundaries.
Registered 2026-11-14Amendments
- 2026-11-14 Data pins, rule readings, the 2,705-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 · 4 trials · 0 passed · judged at N = 4 trials; 2028 trials run in total
| Trial | Verdict | |
|---|---|---|
| scon-long-all | FAIL | counted |
| scon-short-all | FAIL | counted |
| scon-long-big | FAIL | counted |
| scon-short-big | FAIL | counted |
Limitations
- Daily bars only: the official close and open auctions; fills of size on $1-$5 names are not guaranteed.
- The overnight session itself is not traded.
- Every short pays a 100% a year borrow floor and assumes shares could be located.
- No news or filing filter.
- Ticker changes and re-used symbols on the event day are not checked.
- January 2016 is thinly screened: the 2015 daily files carry about 300 symbols, so the first events fall in February 2016.
- 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 843530e067d75c61, published 2026-11-15