docs/research/free-look-battery-2026-08-27.md
Superseded discovery record: The complete 28 August campaign record replaces this first fifteen-look snapshot.
Decision. The first cross-sectional battery found one lead worth pre-registering: liquid, long-only 12−1 momentum. Five ideas earned confirm-cards from 15 initial looks; four were then refuted, which is the screening process doing its job. (Sources: free-looks/LEAGUE_TABLE.txt; free-looks/killcards/*.killcard.md.)
- Positive case: momentum survived timing, cost, permutation and liquidity checks strongly enough to justify a one-shot lockbox test—not trading. (Source:
free-looks/killcards/momentum-positive.killcard.md.) - Rigorous verdict: the other apparent winners came from same-auction look-ahead, entry at an already observed close, dividend-adjustment/cost bookkeeping, or a misread volume-shock portfolio. (Sources: the four refuted confirm-cards under
free-looks/killcards/.) - Output: the battery produced one pre-registration: momentum 12−1 long-only v1. (Source:
docs/research/prereg-momentum-12-1-long-only-v1-2026-08-27.md.)
What the battery is
Cheap, comparable screening before scarce validation data is touched.
- Window: screen only from 1998-01-01 through 2021-12-31; keep 2022-01-01 through 2026-08-21 untouched as the one-shot lockbox. (Source:
free-looks/SPEC.md, amendment A1.) - Trade clock: form the signal with data known by close t, fill at adjusted open t+1, and exit at the next rebalance open; use last observed close for a delisted holding. (Source:
free-looks/SPEC.md, amendment A2.) - Primary endpoint: long-only top decile minus the equal-weight universe; long/short is diagnostic because borrow and locate are unknown. (Source:
free-looks/SPEC.md, amendment A2.) - Universe: price at least $5 and trailing 20-session ADV at least $1m; no category filter. (Source:
free-looks/SPEC.md, amendments A2–A3.) - Costs: charge actual turnover at 10 bp per side for current-metadata Large/Mid and 25 bp per side otherwise, then repeat at 2× costs. (Source:
free-looks/SPEC.md, amendment A4.) - Statistics: one primary net annual-excess result per look, overlap-based Newey–West lag, plus 0.5× and 2× lag sensitivity. (Source:
free-looks/SPEC.md, amendment A4.) - No tuning: each look and at most one declared variant entered the ledger before evaluation. (Sources:
free-looks/SPEC.md, amendment A5;free-looks/TRIAL_LEDGER.jsonl.)
The 15 free looks
Positive case first; the verdict column records what survived the cheap screen. Headline figures are the initial primary net annual excess and Newey–West t, before confirm-card corrections.
| Idea | Proposed payer | Screen verdict | Headline evidence |
|---|---|---|---|
| 12−1 momentum | Slow information diffusion; benchmark and mandate constraints | Weak → confirmed survivor | +3.78%/yr; t=1.10. (Source: free-looks/momentum-12-1.md.) |
| One-month reversal | Forced liquidity demand | Dead | −4.56%/yr; t=−1.45. (Source: free-looks/reversal-1m.md.) |
| Weekly Small/Micro reversal | Liquidity provision | Dead | −13.63%/yr; t=−3.72. (Source: free-looks/reversal-1w-small.md.) |
| Overnight premium | Retail/institutional timing asymmetry and margin constraints | Promising → refuted | +25.52%/yr; t=12.46. (Source: free-looks/overnight-premium.md.) |
| Low volatility | Leverage-constrained investors overpay for lottery-like high-volatility names | Dead | −0.77%/yr; t=−0.25. (Source: free-looks/low-volatility.md.) |
| MAX lottery | Retail overpays after extreme daily gains | Dead | −1.20%/yr; t=−0.43. (Source: free-looks/max-lottery.md.) |
| 52-week high | Anchoring and underreaction around the prior high | Dead | −3.91%/yr; t=−1.57. (Source: free-looks/52w-high.md.) |
| Turn of month | Payroll and pension flows | Dead | −11.32%/yr; t=−3.04. (Source: free-looks/turn-of-month.md.) |
| Ex-dividend | Dividend-capture demand and tax clienteles | Promising → refuted | +24.39%/yr; t=8.65. (Source: free-looks/ex-dividend.md.) |
| Splits/spinoffs | Attention underreaction and mandate-driven child selling | Weak | +3.05%/yr; t=0.72. (Source: free-looks/splits-spinoffs.md.) |
| Volume shock | Attention and information shocks | Dead → confirm-card on reversed framing | −15.01%/yr; t=−11.11. (Source: free-looks/volume-shock.md.) |
| Intermediate vs long-term | Intermediate winners against long-term losers | Dead | −3.49%/yr; t=−2.12. (Source: free-looks/intermediate-vs-longterm.md.) |
| Residual momentum | Firm-specific information after removing the broad market component | Weak | +1.12%/yr; t=0.56. (Source: free-looks/residual-momentum.md.) |
| Gap fade | Liquidity providers fade adjusted overnight gaps | Promising → refuted | +96.60%/yr; t=43.17. (Source: free-looks/gap-fade.md.) |
| Industry mean reversion | Idiosyncratic liquidity demand against industry peers | Dead | −1.07%/yr; t=−0.54. (Source: free-looks/industry-mean-reversion.md.) |
Source conflict: free-looks/LEAGUE_TABLE.txt prints residual momentum at −0.005%, while the detailed report gives +1.12%/yr and t=0.56; this register uses the detailed report and does not conceal the mismatch. (Sources: free-looks/LEAGUE_TABLE.txt; free-looks/residual-momentum.md.)
The 5 confirm-cards
Confirm-cards asked whether each exciting lead could earn the right to proceed.
- Gap fade — refuted: the original ranked the official open and booked that same open-to-close return. Honest next-open execution changed +96.60%/yr (t=43.17) to −56.17%/yr (t=−25.90). (Source:
free-looks/killcards/gap-fade.killcard.md.) - Overnight premium — refuted: the original entered at the close it had just observed, made about 39.9 round trips/month, and credited benchmark costs. The frozen next-open result was −4.20%/yr (t=−7.38). (Source:
free-looks/killcards/overnight-premium.killcard.md.) - Ex-dividend — refuted as framed: the source has no point-in-time announcement timestamp; the original close-adjusted exit booked the dividend and its benchmark treatment credited costs. The corrected +8.25%/yr (t=2.92) became −3.86%/yr at 2× costs. (Source:
free-looks/killcards/ex-dividend.killcard.md.) - Volume shock — refuted: the original tested signed high-shock D10, not the quiet sleeve or exclusion filter. The filtered book showed +9.88%/yr absolute, but only −1.73%/yr improvement (t=−13.98). (Source:
free-looks/killcards/volume-shock-reversed.killcard.md.) - Momentum — conditional pre-register: the liquid $10-price/$10m-ADV sleeve earned 15.0%/yr net, Sharpe 0.49; its causal volatility overlay earned 6.75%/yr, Sharpe 0.55, with −27.6% drawdown. The verdict was pre-register, not trade. (Source:
free-looks/killcards/momentum-positive.killcard.md.)
Trial accounting
Count every look, confirmation and registration.
- Free-look ledger: raw
N=15, exactly trials 1–15 in the JSONL ledger. (Source:free-looks/TRIAL_LEDGER.jsonl.) - Full decision ledger: raw
N=21= 15 free looks + 5 confirm-cards + 1 pre-registration. (Source:docs/research/prereg-momentum-12-1-long-only-v1-2026-08-27.md, §§4 and 6.) - Clustered effective count:
N=21by a deliberately conservative no-discount assumption; no empirical correlation-clustering receipt exists. (Source:docs/research/prereg-momentum-12-1-long-only-v1-2026-08-27.md, §§4 and 6.)
Caveats that survive the verdict
This is a candidate, not a promise.
- Point-in-time metadata: size, category and industry fields are current, not historical; they are diagnostic or cost-bucket inputs, never eligibility evidence. (Source:
free-looks/SPEC.md, amendment A3.) - Right-tail dependence: removing momentum's best 14 months leaves 0.41%/yr and Sharpe 0.02. (Source:
free-looks/killcards/momentum-positive.killcard.md.) - Capacity: the confirm-card's 1%-of-ADV binding sleeve is $14.9m median and $9.5m at p10; market impact remains modelled, not observed. (Source:
free-looks/killcards/momentum-positive.killcard.md.) - Delistings: last-close substitutions are imperfect because complete delisting proceeds are unavailable; momentum alone reports 373 last-close top exits. (Sources:
free-looks/SPEC.md, amendment A2;free-looks/killcards/momentum-positive.killcard.md.) - Lockbox: no result here uses 2022-01-01 onward; the reserved data remained untouched when the decision was made. (Sources:
free-looks/SPEC.md, amendment A1; all five confirm-cards.)
Final verdict: pre-register momentum once; carry every trial into its deflated-statistics penalty; do not reopen, repair or trade on the strength of the screening window.
On this page 5 sections
LAB documentation describes design intent and research safeguards. It does not provide trading instructions or operational access.