Do speculative non-paying shares traded more than usual beat quieter ones overnight relative to the day, unlike paying shares?
Study 78 · Changed a rule or plan · recorded 31 October 2026 · answer: Partly · full paper with figures
The question
Each week we rank shares priced at five or more with no payout in the prior year on four speculative traits, and split the most speculative fifth by that week's trading volume against its usual level. Does the busier half's overnight return less its daytime return rank higher in the fifth the next week, and not among paying shares?
Rules fixed before the result
Fixed before any outcome: the answer is read on the plausible range of the weekly rank difference against zero and a size of 0.012, widened by 1.1 so that resampled earlier years kept each side's false-alarm rate at or under 0.05. Yes needed that range above zero and reaching 0.012, the difference standing once balanced on earlier traits, and the paying shares' range wholly below 0.012. It required 200 weeks and on average 100 shares a week in the fifth.
Data and method
Shares priced at five or more at each week's last close, with no payout in the prior year, were ranked on four speculative traits; the top fifth was split at the median of each share's volume over the week's last five sessions against its usual volume. Each share's overnight return less its daytime return over the next week was ranked within the fifth and the halves' average ranks compared, allowing for correlation between weeks up to a quarter apart.
The answer
Partly. Over 260 weeks, 2018 to 2022, the busier half ranked 0.0095 higher than the quieter half (plausible range 0.0017 to 0.0173): above zero, reaching the 0.012 set beforehand only at its upper end. The figure is measured from the official opening price, and an official opening price recorded about 3.4 basis points too high on every trading day of the following week, for the busier half only, could by itself produce the same figure; no quote data are held to tell them apart. The return gap was 0.38 percentage points a week (0.09 to 0.68), mostly lower daytime returns than the quieter half's (-0.0104 in rank, -0.0189 to -0.0018); the overnight part was unclear (0.0043, -0.0040 to 0.0126). Balanced on four earlier traits it stayed (0.0125, 0.0047 to 0.0203); among paying shares it was unresolved (0.0073, -0.0032 to 0.0178).
Main estimate
0.009 fraction (95% range 0.002 to 0.017), from 260 observations
What it changes
By the rule set in advance, the project proposes, for its records only, that later studies of speculative shares record each share's trading volume in the sorting week as a measured condition; as the paying shares' comparison was not settled, the split is recorded as not shown to be confined to shares paying nothing. The plan gave this kind of holding a chance of 0.27, at low confidence.
What remains uncertain
Earlier studies used these years, so this is a check on development data, not a fresh test, and an earlier sort of all shares on the same trading volume was already known. An opening price recorded too high after a busy week would give the same result, and no quote data separate the two. Only 141 weeks held enough paying shares in the most speculative fifth of all shares, under the 150 needed, so the two groups were not compared.
How it was made
Run took 69 minutes; independent code review: approve; the recorded calculation reruns to the same result.
Fingerprints
plan 48b9b16ef42a64588abe869dc768f7c092908a9ef2f4fdbb717637c6cc240bda result a2c65a63c49499eb1a2f3e5e1a1ee28ac6a98e911f817668aa859d3917db9a9d
Exploratory work on development data; not a performance claim.