Sixteen strategies tested on seven years of daily bars. Fifteen failed. The one that survived beat the S&P out-of-sample on return, risk-adjusted return and drawdown — and it did so while 95 of its holdings were delisted out from under it — some acquired, some failed — each charged out at its last traded price.
Rank every liquid US stock each month by 12-month momentum minus 6-month volatility, hold the top decile, size each position by inverse volatility. Long-only, no shorts, no leverage, cash earns 0%.
| Book | CAGR | Sharpe | Max DD |
|---|---|---|---|
| momvol, inverse-vol | 26.0% | 1.12 | -23.8% |
| momvol, equal | 28.8% | 1.08 | -24.0% |
| mom12, equal (naive) | 28.4% | 0.89 | -25.9% |
| SPY buy & hold | 26.7% | 1.41 | -19.4% |
| Book | CAGR | Sharpe | Max DD |
|---|---|---|---|
| momvol, inverse-vol | 18.1% | 0.92 | -16.7% |
| momvol, equal | 18.6% | 0.80 | -19.6% |
| mom12, equal (naive) | 12.3% | 0.54 | -22.5% |
| SPY buy & hold | 11.6% | 0.77 | -20.2% |
It lost in-sample and won out-of-sample — the opposite of what an overfitted strategy does. Naive momentum (mom12) decayed exactly as a fragile result should, Sharpe 0.89 → 0.54. The risk-controlled version held.
Adding 4,930 delisted companies cost it only 1.5 points of out-of-sample CAGR. The bias was real and it was not the story.
| Year | momvol | SPY | Edge |
|---|---|---|---|
| 2019 | +3.4% | +18.5% | -15.1 |
| 2020 | +32.1% | +18.3% | +13.8 |
| 2021 | +30.5% | +28.7% | +1.8 |
| 2022 | -1.3% | -18.2% | +16.9 |
| 2023 | +8.3% | +26.2% | -17.9 |
| 2024 | +29.1% | +24.9% | +4.2 |
| 2025 | +29.5% | +17.7% | +11.8 |
| 2026 part year | +20.1% | +9.0% | +11.1 |
Beats the index in six of eight years, the last of which is seven months, not twelve — both columns are measured over the same months. 2022: −1.3% against SPY's −18.2%. It gives that back in narrow mega-cap rallies (2019, 2023) — textbook low-volatility behaviour, which is why the mechanism is believable rather than coincidental.
13,160 position-periods across 2,073 symbols. An equity curve can hide anything; this cannot. Search a ticker, sort any column, or filter to the 95 holdings that were delisted out from under it.
53.7%
mean +1.94%, median +0.94% per position
FCEL +396% · EQC -91%
single position-periods, not compounded
95
exited at last traded price, loss charged in full
| Entry | Symbol | Entry px | Exit px | Return | Weight |
|---|
Trades run back to 2006, but coverage before 2019 is thin and 2015, 2016 and 2018 are missing entirely — too few symbols passed the liquidity screen in those months for a decile to mean anything. The headline results above cover 2019-07 → 2026-07 only. Earlier rows are shown for completeness, not as evidence.
| Year | Positions | Win rate | Mean return |
|---|---|---|---|
| 2006 | 53 | 50.9% | +0.23% |
| 2007 | 77 | 44.2% | +0.72% |
| 2008 | 89 | 39.3% | -2.27% |
| 2009 | 83 | 62.7% | +1.25% |
| 2010 | 94 | 57.4% | +1.10% |
| 2011 | 103 | 44.7% | +0.02% |
| 2012 | 105 | 63.8% | +2.97% |
| 2013 | 121 | 58.7% | +2.42% |
| 2014 | 176 | 54.0% | +0.95% |
| 2017 | 86 | 75.6% | +7.35% |
| 2019 | 767 | 54.1% | +1.00% |
| 2020 | 1,464 | 57.9% | +4.60% |
| 2021 | 1,786 | 55.9% | +0.66% |
| 2022 | 1,709 | 51.6% | +0.54% |
| 2023 | 1,641 | 50.8% | +0.93% |
| 2024 | 1,776 | 55.1% | +2.61% |
| 2025 | 1,969 | 53.5% | +3.12% |
| 2026 | 1,061 | 48.8% | +2.06% |
Ranked by the share of symbols where the strategy beat simply holding the stock. The mark at the middle of each bar is 50% — break-even with doing nothing.
With eleven attempts, the best would be expected near 53% by luck alone. The best observed was 28%. Several of them turned a profit on a majority of the symbols they traded while still losing to simply holding those same symbols — which is why “% profitable” is a vanity metric.
Also killed: the opening-range breakout (−13.1%, its 2R target filling on 1 of 73 trades) and the heavy-volume catalyst thesis, refuted when a random filter keeping the same fraction of days matched it and beat it per trade.