Stocks that have outperformed their peers tend to keep doing so for a while, through underreaction and herding (Jegadeesh and Titman). The known weakness is the crash: momentum collapses at the turn from bear market to bull, when last year's losers rebound hardest (Daniel and Moskowitz). The bet here was to keep the momentum premium and avoid the crash by trading only while SPY is in an uptrend and VIX is low.
One more defensive condition was added: skip the very top of short-term momentum, to avoid buying blow-off tops.
How it was tested
Not pre-registered. Signal: 60-day momentum rank above 90, SPY in an uptrend, VIX below 25, and a 20-day momentum rank between 60 and 90. US stocks at $15 or more with at least $10M of 20-day dollar volume. Buy at the next open, exit after 10 sessions, with a 2×ATR stop and a 4×ATR target. Validation was a held-out test: train on 2014-2019, test on 2020-2025, judged on raw Sharpe, the lab's method in May 2026. The rule allowed the out-of-sample Sharpe to degrade, but by no more than half.
Where it died
On 2026-05-09 the held-out half did not degrade, it reversed. In-sample, 20,073 trades gave a Sharpe of +0.260; out of sample, 18,702 trades gave −0.985. It was rejected on the spot. It was never re-measured with Jensen alpha, so the record holds only raw Sharpe ratios with no benchmark adjustment.
The record
Thesis
Buy the top decile of 60-day cross-sectional momentum, but only while SPY is in an uptrend and VIX is low, to sidestep the momentum crashes that happen at bear-to-bull transitions. Momentum works on average through underreaction and herding; the regime filter is meant to cut its worst episodes.
Jegadeesh, N. & Titman, S. (1993). Returns to Buying Winners and Selling Losers. Journal of Finance.
Daniel, K. & Moskowitz, T. (2016). Momentum Crashes. Journal of Financial Economics.
Asness, C., Moskowitz, T. & Pedersen, L. (2013). Value and Momentum Everywhere. Journal of Finance.
Specification
signal
60-day momentum rank > 90; SPY uptrend; VIX < 25; 20-day momentum rank between 60 and 90 (deliberately excluding the very top, to avoid buying blow-off tops)
Raw Sharpe ratios, no benchmark adjustment — the lab's method in May 2026. Never re-measured with Jensen alpha.
Why it died
The out-of-sample half did not just weaken, it changed sign: Sharpe +0.260 in 2014-2019, −0.985 in 2020-2025, with win rate falling from 48.6% to 44.4%. The in-sample edge was small to begin with. The suspected cause is the filter that excluded the top decile of recent momentum: in 2020-2024 those were the mega-cap leaders that kept rising for years, so the strategy bought the laggards.
What it taught
A filter that looks obviously prudent — don't buy the blow-off top — can invert an edge in a regime where leadership is concentrated rather than rotating. Defensive conditions are parameters too, and they overfit like any other.