Fund managers report their holdings at quarter end. A stock that collapsed during the quarter is embarrassing to show, so, the story goes, managers sell it in the last days before the report date, whatever they think it is worth. After the date passes, the pressure lifts and the price should recover. The obligation here is career risk rather than law, which is also the argument for why it might not be arbitraged away.
It was moved up the queue the same day the tax-loss study found a large year-end seasonality and refuted the tax explanation. Window dressing was the obvious alternative.
How it was tested
Pre-registered on 2026-09-06 over 2014-2022, with 2023-2025 not loaded. Signal: bottom decile of 60-day return, roughly one quarter. US stocks above $5 with more than $1M of daily dollar volume. Entry at the next open during the last 3 sessions of each quarter, exit after 15 sessions, one run per quarter-end, four variants and a bar of alpha t 2.24. Controls: the same recipe mid-quarter, far from any report date, and an annual twin ranked by year-to-date return. Four predictions were registered: an effect in at least 3 of 4 quarters, quarter-end beating mid-quarter, quarterly ranking beating annual, and a stronger effect in liquid names.
Where it died
Pooled over 1,042 trades, alpha was −0.153% per trade, t −0.19, and t captured in a portfolio was 0.26. The decisive comparison went the wrong way. The mid-quarter placebo, with no report date nearby, earned +0.438% per trade, beating the quarter-end entry. If the report date drove the selling, the opposite should have happened.
The record
Thesis
Fund managers sell the quarter's embarrassing losers before the holdings report date, so they don't show on the report, whatever they think the stock is worth. The pressure concentrates in the last days of the quarter and reverses once the date passes. The obligation is career risk, not economics — which, if true, is what would keep it from being arbitraged away.
Prior: Prioritised from the queue because the tax-loss study the same day found a large year-end seasonality and refuted its tax explanation. Window dressing was the obvious alternative and was already registered.
Specification
signal
returns_60d (about one quarter) in the bottom decile
universe
US stocks, 20d dollar volume > $1M, price > $5
entry
last 3 sessions of each quarter, next open — run once per quarter-end
exit
15 sessions, 2.0 / 2.5 × ATR stop / target
controls
Mid-quarter placebo (same recipe, no report date nearby) and an annual twin ranked by returns_ytd. Pre-registered predictions: effect in at least 3 of 4 quarters (P1); quarter-end beats the mid-quarter placebo (P2); quarterly ranking beats annual (P3); stronger in liquid names, where funds are (P4, reported only).
held out
2023-2025 not loaded
Gates
Pre-registrationpass
Was the hypothesis, its parameters and its pass/fail rule written down before any data was looked at?
Market-adjusted alpha (legacy)fail
Does the strategy beat SPY over the same holding windows, assuming beta = 1?
t ≥ 2.24 (Bonferroni, 4 quarters)
alpha t-0.19alpha / trade %-0.15trades1,042
Pooled over the four quarter-ends. Run on 2026-09-06, before Gate 7 moved to Jensen alpha; measured with beta assumed = 1 and not re-derived.
Gate 8 — Capture-preserved alphafail
Does the subset of trades a real portfolio can actually take keep the alpha?
t_cap ≥ 1.65 and ratio ≥ 0.50
t captured+0.26capture ratio-1.46
Measured with the portfolio simulator before its 2026-09-19 slot-leak fix; not re-derived.
Why it died
The report date does not matter. If pressure built toward the quarter-end, entering then should beat entering mid-quarter; the mid-quarter placebo won (+0.438% vs −0.153% per trade). The liquidity test pointed the wrong way too: window dressing needs funds in the name, so it should be stronger in liquid stocks; it was stronger in illiquid ones (+0.066% vs −0.375%). The counterparty is not fund managers.
What it taught
A sign-only prediction is too weak. P1 "passed" with three positive quarters at t +0.49, +0.66 and +0.77 — none significant, and three positive signs out of four is what chance produces. Later pre-registrations of this kind require significance, not sign. The one quarter with magnitude was June (−3.59% per trade, t −2.45), matching the tax-loss study's June control from a different ranking; that observation came from controls and was not chased from here.
Verdict history
2026-09-06Pre-registered: one run per quarter-end, placebo and annual twin as controls
2026-09-06Single run — placebo beats quarter-end (P2 fails), pooled alpha t −0.19 No alpha
2026-09-06Mechanism moved to rejected in the registry