Buy stocks that have risen strongly over the last 60 days, but calmly: low daily volatility, comfortably above their 200-day average, and with an RSI that shows strength without being overbought. The intuition is that an orderly trend in a "quality" name is more likely to continue for another two weeks than a wild one. No mechanism, counterparty or reference was written down when the rule was found.
momentum / low volatility
Quality momentum long
The idea
How it was tested
Not discovered by pre-registration. Signal: 60-day return above 10%, ATR(14) below 3% of price, more than 5% above the 200-day average, RSI(14) between 50 and 70. US stocks at $20 or more with at least $20M of 20-day dollar volume. Buy at the next open, exit after 10 sessions, 2×ATR stop, 5×ATR target.
It went through a held-out test on 2026-04-30, judged on raw Sharpe as was the method of the time, and was rejected on 2026-05-10 by a yearly audit. It was later re-measured as a frozen configuration with the rest of the canonical set, at neutral sizing: $5,000, 1% risk per trade, 10 slots.
Where it died
Every re-measurement with Jensen alpha has found it negative. On 2026-09-08 the full history gave t −1.64 with beta 0.695. The 2026-10-17 gap-fill engine fix pushed it to −3.08. On 2026-10-30 the audit's calendar-clustered standard error, on the same trades and the same alpha, moved it to −2.448: less extreme, still significantly negative. The current record is −0.15% per trade over 11,869 trades, beta 0.669. Gate 8 was not evaluated.
The record
Thesis
Stocks with strong 60-day momentum, low volatility and a healthy uptrend (above the 200-day average, RSI between 50 and 70) keep rising over the next two weeks: trend plus "quality" filtering. No mechanism or reference was written down when it was found.
Specification
- signal
- 60-day return > 10%; ATR(14) < 3% of price; > 5% above MA200; 50 < RSI(14) < 70
- universe
- US stocks, price ≥ $20, 20d dollar volume ≥ $20M
- entry
- next open
- exit
- 10 sessions, 2×ATR stop, 5×ATR target
- sizing
- neutral default for the re-measurement — $5,000, 1% risk per trade, 10 slots
- caveat
- Not discovered by pre-registration. Rejected in May on a yearly audit, then re-measured as a frozen configuration with the rest of the canonical set.
Gates
Gate 7 — Jensen alphafail
After regressing out market exposure, is there alpha left?
t ≥ 1.65
t (calendar SE)-2.45alpha / trade %-0.15beta+0.67trades11,869
Gate 8 — Capture-preserved alphanot evaluated
Does the subset of trades a real portfolio can actually take keep the alpha?
Not evaluated once Gate 7 fails.
Why it died
Significantly negative Jensen alpha with beta about 0.67: the selected names deliver less than their market exposure pays. Its positive raw Sharpe (0.74 in-sample, 0.52 out-of-sample) was market drift. The May rejection came from a 2022 tail (Sharpe −2.43) that two macro regime filters made worse, not better (−5.07 and −6.68).
What it taught
Long momentum with extra "quality" conditions is the most consistent loser in the lab's long book. The gap-fill engine fix hurt it more than most (t −1.85 → −3.08): long setups that gap down through their stop were being filled at the stop instead of at the open.
Verdict history
- 2026-04-30Held-out test: in-sample Sharpe 0.74, out-of-sample 0.52
- 2026-05-10Rejected: yearly tail −2.43 in 2022; VIX and SPY-trend guards worsen 2022 to −5.07 / −6.68 Unstable
- 2026-09-08Re-measured with Jensen alpha on full history: t −1.64, beta 0.695 No alpha
- 2026-10-17Gap-fill engine fix: t −1.85 → −3.08 No alpha
- 2026-10-30Standard error by calendar (audit): Gate 7 t -3.079 → -2.448, same trades and alpha