The thesis
momentum-code runs a simple, legible idea: buy the top positive movers in a 24-name large-cap universe — spanning tech, financials, healthcare, staples, and energy — with a per-position cap to avoid concentration. There is nothing exotic here, which is a feature. A momentum tilt on liquid names is easy to reason about and easy to audit, and the strategy currently sits in live status.
Backtest performance
Over 451 days the strategy returned 19.76%, lifting a $10,000 book to $11,976 for a 10.6% CAGR. The Sharpe of 0.65 is modest, and the 20.48% max drawdown is the number that should give any allocator pause — you had to stomach a fifth of the book underwater to collect that return. Turnover of 92.74% is reasonable for a momentum sleeve, and fees were negligible at $5.
One quirk deserves a flag: the backtest reports a 0% win rate across just 5 trades, yet finishes solidly positive. With so few discrete trades, the win-rate metric is close to meaningless here — the return came from position drift, not from a high hit rate. Treat headline win rate as noise at this sample size.
Validation: the honest part
The walk-forward study is where momentum-code earns respect and a warning. Across four folds it was positive four times — 15.38%, 3.27%, 12.18%, and 13.71% — with the most recent out-of-sample slice returning 13.71% at a 1.93 Sharpe and only a 6.32% drawdown. Fold-level drawdowns (6.12%, 16.55%, 5.21%, 6.32%) are all well below the full-run 20.48%, which hints the worst drawdown was a stitching artifact between periods.
So why does validation read passed: false? Because the gate looks past raw returns. The Probabilistic Sharpe (0.811) is encouraging, but the Deflated Sharpe (0.338) — which penalises the 6 trials it took to find this configuration — is weak. Four positive folds are good; four positive folds that only survive after multiple attempts are exactly the pattern overfitting produces. The gate is right to withhold its blessing.
Live reality: nothing is happening
The most sobering data is the live log. Every scheduled run from August 19 through August 26 executed zero trades and rejected one or two candidates each day. Cash is stuck at $608.79 — too little to open new capped positions — while total equity has drifted down from $10,634 to $10,467 over the week. The last executed trades (XOM, NVDA, BAC, HON, MSFT) were back in late May and early June. In practice the strategy is a fully-invested, frozen book, not an active momentum engine.
Verdict
momentum-code has a clean thesis and genuinely encouraging out-of-sample folds — but it is not validated, its drawdown is steep, and its live incarnation has been unable to act for nearly three months. The next fix is operational, not analytical: free up cash or relax the per-position cap so the schedule can actually rotate into the movers it was built to chase.