Thesis
momentum-code runs a simple, legible idea: buy the top positive movers in a 24-name large-cap universe — spanning tech (AAPL, MSFT, NVDA), financials (JPM, V, MA), staples (PG, KO, WMT) and energy (XOM, CVX) — with a cap on each position. It is a classic cross-sectional momentum tilt with no shorting and a hard per-name ceiling to limit concentration. The strategy is currently marked live.
Backtest performance
Over 451 days the backtest returns 19.76%, growing a $10k book to $11,976 for a 10.6% CAGR. Risk-adjusted, the picture is more modest: a Sharpe of 0.65 against a 20.48% max drawdown means investors absorbed a one-fifth peak-to-trough decline for that return. Turnover is high at 92.7%, though fees are trivial ($5 total) at this size. One figure to flag: the reported win rate is 0% across 5 trades, which almost certainly reflects a metric measured on open, still-held positions rather than a literal string of losers — worth reconciling before it misleads.
Validation — the important part
This is where the strategy earns its scrutiny. Across four walk-forward folds, all four were positive (15.38%, 3.27%, 12.18%, 13.71%), with strong out-of-sample fold Sharpes as high as 2.47. The aggregate out-of-sample return is 13.71% at a 1.93 Sharpe, and the Probabilistic Sharpe Ratio sits at a respectable 0.811.
And yet our gate reports validation: failed. The reason is the Deflated Sharpe Ratio of 0.338. Once the Sharpe is adjusted for the 6 trials behind this result, the confidence that the edge is real — rather than the best of several attempts — drops below our threshold. This is exactly the discipline the platform is built for: four green folds are encouraging, but with only five trades per fold the sample is thin, and multiple-testing deflation is the honest correction. The strength (consistent positive folds) and the risk (a fragile, low-trade-count signal that may not survive selection bias) live in the same table.
Recent activity
Production tells a quieter story. The last executed trades were a cluster on 31 May–1 June: buys in MSFT, HON, BAC, then NVDA and XOM. Since then, every scheduled run from 8–15 September reports 0 executed and 2–3 rejected, with cash pinned at $608.79 and total equity drifting between roughly $10,320 and $10,540. The most plausible read: the book is nearly fully invested, and residual cash is too small to clear the position cap and open new momentum names, so signals are generated but nothing fills.
Verdict
momentum-code is a clean, interpretable strategy with encouraging fold consistency, but two cautions stand out: it did not clear the deflated-Sharpe robustness bar, and it is effectively inert in production — holding, not trading, for over three months. Before leaning on it, resolve the win-rate metric, revisit cash/position-cap sizing so the engine can actually act, and treat the OOS numbers as promising rather than proven.