Yes, measurably. Institutional managers may take up to 45 days after quarter end to file Form 13F, and they use that window very differently: a tenth file within two weeks, the median waits about 39 days, and roughly a fifth arrive on the deadline itself. Any statistic computed across managers before the window closes is therefore built from whichever managers have filed so far.
Across 16 quarters and 85,036 manager-quarters, three standard cross-manager characteristics were flagged in 11 of 11 screened quarters:
| characteristic | mean ρ̂ | cycles flagged |
|---|---|---|
| log position count | −0.1789 | 11 / 11 |
| top-10 share | +0.1474 | 11 / 11 |
| concentration (HHI) | +0.1446 | 11 / 11 |
The noise floor on this same panel is 0.0152 — the largest reading reached by any signal drawn independently of everything else. The real readings are about ten times that.
The arrival times are observed, not reconstructed. Each quarterly data set carries the date SEC received the filing, so nothing about the timing is inferred. Screens that must rebuild arrival order from a carry-forward panel report a lower bound; this one does not.
That active and larger managers delay is documented: Christoffersen, Danesh and Musto find that “more active institutions delay their holdings longer” — where activeness is measured partly by portfolio concentration — with a strategic explanation (protection from front-runners, concealment of voting power).
Our readings reproduce their pattern independently, which is a check this screen had to pass. What is added is the consequence, not the pattern: their paper asks why institutions delay, and never asks what the delay does to a cross-sectional quantity computed before the window closes. This screen supplies that number, and the required completeness that follows from it.
| test | result |
|---|---|
| arrival permuted within quarter (3 reps) | ρ̂ collapses to 0.000–0.007, 0 of 11 flagged in every rep |
| drop the 15,152 deadline-day filers | 0.107–0.134, still flagged |
| quadratic size control | 0.122–0.147, still flagged |
| trailing window 3 and 8 quarters | 0.143–0.186, still flagged |
The shuffle control is the decisive one: permuting arrival order inside each quarter removes the reading entirely, so it is arrival order that carries the information.
A planted-truth positive control on the real filing dates classified 6 of 6 cases correctly — including two where the signal was coupled to portfolio size at up to six times the leak strength and the screen still, correctly, read benign.
pip install pit-release-gate polars
git clone https://github.com/MaxWellApexLab/pit-audit-registry
cd pit-audit-registry/methodology/2026-08_sec-13f
python screen_13f.py # the aggregated panel is committed; no download needed
python robustness_13f.py
Free SEC data, no account, no licence. build_13f_panel.py rebuilds the panel
from sec.gov if you would rather not trust the committed one.
Full report, panel and scripts · Registry · Method paper · Screen implementation