Not pre-registered — see caveat · data snapshot 2026-09-19T090800Z
The idea
The low-volatility anomaly is well documented: calm stocks earn better risk-adjusted returns than CAPM predicts. One explanation is leverage constraints. Investors who cannot borrow reach for high-beta stocks instead, overpricing them and leaving the calm ones cheap (Frazzini and Pedersen). Benchmarked managers are reluctant to hold low-beta names that lag in rallies (Baker, Bradley and Wurgler). This configuration buys calm stocks in an uptrend with medium momentum, and only when markets are calm too.
It is part of the lab's canonical set, early configurations frozen and re-measured each time an instrument changed.
How it was tested
Signal: a 20-day volatility rank below 30, price more than 5% above its 200-day average, a 20-day momentum rank between 50 and 90, VIX below 20, and SPY above its 200-day average and rising. Universe: US stocks priced at $20 or more with at least $30M of daily dollar volume. Buy at the next open, hold 10 sessions, with a 1.5×ATR stop and a 3.5×ATR target.
The VIX and SPY filters were added on 2026-05-10, in two refinement rounds after seeing a bad 2022, and the configuration was rejected on yearly Sharpe anyway. Only the later locked re-measurements of the frozen rules count as verdicts.
Where it died
On 2026-09-08 Jensen alpha said it passed Gate 7 with t +2.56, and Gate 8 said a portfolio could not take it: capture ratio 0.041. Fixing the simulator's slot leak on 2026-09-19 lifted the ratio to 0.161, still a fail.
The engine fix of 2026-10-17 then removed the alpha itself. Stops and targets had been filled at their exact level even when the bar gapped past them; filling them at the open moved Gate 7 from t +2.39 to +1.29. The calendar standard error of 2026-10-30 left it at t +1.029: 0.092% per trade over 6,130 trades, with a beta of −0.055. Almost no market exposure, and not enough alpha to stand on its own.
The record
Thesis
Low-volatility stocks earn better risk-adjusted returns than CAPM predicts (leverage constraints, lottery preference). Restricted to stocks in an uptrend with medium momentum, and only when VIX is low and SPY confirms a bull regime.
Frazzini, A. & Pedersen, L. (2014). Betting Against Beta. Journal of Financial Economics.
Asness, C., Frazzini, A. & Pedersen, L. (2019). Quality Minus Junk. Review of Accounting Studies.
Baker, M., Bradley, B. & Wurgler, J. (2011). Benchmarks as Limits to Arbitrage. Financial Analysts Journal.
Specification
signal
20d volatility rank < 30; price > 5% above MA200; 20d momentum rank between 50 and 90; VIX < 20; SPY above its MA200 and in an uptrend
universe
US stocks, price ≥ $20, 20d dollar volume ≥ $30M
entry
next open
exit
10 sessions, 1.5×ATR stop, 3.5×ATR target
caveat
Not a clean pre-registration. The two regime filters (VIX, SPY) were added in May 2026 in two refinement rounds after seeing a bad 2022 — the kind of iteration the lab later banned. Its verdicts are reported anyway because they came from later, locked re-measurements of this frozen configuration.
Gates
Gate 7 — Jensen alphafail
After regressing out market exposure, is there alpha left?
t ≥ 1.65
t (calendar SE)+1.03alpha / trade %+0.09beta-0.06trades6,130
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 — no alpha in the signal means nothing to capture.
Why it died
For six weeks this was the lab's textbook Gate 8 case: real alpha in the signal (t +2.56), almost none in the trades a portfolio could take (capture ratio 0.04, later 0.16). Then a backtest-engine bug was fixed — stops and targets were filled at their exact level even when the bar gapped past them — and the signal-level alpha itself fell to t +1.29. It now fails one gate earlier.
What it taught
Verdicts move when the instrument is fixed, and the site shows the trail instead of only the current answer. Gap-through stops inflated long mean-reversion and low-vol longs more than anything else in the set.