Performance
ⓘ How this book is constructed
Does the swarm actually make money? A mechanical long-short book — every scored lean followed (+1 bullish, −1 bearish), equal-weight, daily-rebalanced — daily-marked as a fully-funded, total-return curve against passive benchmarks, for both the out-of-sample backtest and the growing live ledger.
performance: as of 2026-10-14
Invested +99.6% of reference days (idle cash earns 0) · net of 10 bps/side · total-return (dividend + split adjusted).
Out-of-sample window 2025-07-03 → 2026-06-29.
| Series | Total return | Annualized | Sharpe | Max drawdown | Hit rate | N |
|---|---|---|---|---|---|---|
| Strategy (long-short EW) | +46.1% | +47.0% | 1.86 | -9.6% | 65% | 292 |
| EW universe (buy and hold) | +70.3% | +71.7% | 2.90 | -9.5% | — | — |
| NVDA (buy and hold) | +22.5% | +22.9% | 0.76 | -20.2% | — | — |
| SOXX | +153.6% | +157.5% | 2.56 | -15.8% | — | — |
| QQQ | +30.8% | +31.4% | 1.60 | -12.0% | — | — |
| Eli Lilly (buy and hold) | +58.7% | +59.8% | 1.41 | -23.2% | — | — |
| Boeing (buy and hold) | -0.6% | -0.6% | 0.14 | -25.0% | — | — |
| Broadcom (buy and hold) | +36.3% | +37.0% | 0.91 | -28.7% | — | — |
| Amazon (buy and hold) | +7.5% | +7.6% | 0.39 | -21.7% | — | — |
| Alphabet (buy and hold) | +97.5% | +99.7% | 2.47 | -20.4% | — | — |
How to read this
- Mechanical strategy — every scored lean is followed, no discretion, no cherry-picking. The rule is fixed and disclosed.
- Total-return, apples-to-apples — the strategy and all four benchmarks use dividend + split-adjusted closes on the same NYSE reference calendar and window, with the same forward-fill for missing (foreign-holiday) days. Net of 10 bps/side transaction cost.
- Fully-funded — idle cash earns 0 on days nothing is open (see the invested fraction), so Sharpe is apples-to-apples with the benchmarks.
- Annualized return and Sharpe are withheld until the curve spans ~6 months — annualizing a short sample is meaningless even when N is large.
- Out-of-sample backtest — weights are fit on an earlier window and only applied out-of-sample; it can only replay price-derivable signals on today's surviving company universe (survivorship bias biases it up). Full-swarm accuracy accrues live.