The thesis
momentum-code runs one of the oldest ideas in systematic trading: buy strength. It scans a 24-name large-cap universe — spanning tech (AAPL, MSFT, NVDA), financials (JPM, V, MA), staples (PG, KO, WMT) and energy (XOM, CVX) — and allocates to the top positive movers, with a cap on each position to avoid concentration. Simple, transparent, and easy to reason about. It currently carries live status.
Recent activity
The honest headline is that the engine has stalled. The last executed orders date to late May and early June 2026: buys in MSFT, HON and BAC on 31 May, then XOM and NVDA on 1 June. Since then, nothing has cleared. Every scheduled run from 21–28 September reports 0 executed, 1–3 rejected, with cash pinned at $608.79 and total equity drifting in a narrow $10,315–$10,402 band.
The cause is visible in the numbers: the book is effectively fully invested, leaving too little cash to open new positions when fresh movers appear. A momentum strategy lives on rotation — buying today's leaders and shedding yesterday's — so a portfolio that cannot act on its own signals is momentum in name only right now.
Backtest and validation
The research record is more encouraging. Over 451 days the backtest returned 19.76% (final equity $11,976), a 10.6% CAGR, on just 5 trades and a trivial $5 in fees. More reassuring is the walk-forward evidence: all four folds were positive (15.38%, 3.27%, 12.18%, 13.71%), out-of-sample return reached 13.71% and out-of-sample Sharpe a healthy 1.93. The probabilistic Sharpe ratio of 0.811 suggests the edge is unlikely to be pure noise.
And yet validation did not pass. Two figures explain why. The headline Sharpe is a modest 0.65, and the deflated Sharpe ratio — which penalises the 6 trials run during search — sits at just 0.338. A max drawdown of 20.48% against a ~10.6% CAGR is a poor return-to-pain trade. The gate is doing its job: strong folds are not enough when the risk-adjusted, multiple-testing-corrected picture is thin.
Risks and read
Two caveats deserve flagging. First, the reported win rate is 0 across 5 trades — but with no recorded sells, these are open buys rather than realised losers; treat it as "no closed winners yet," not proof of a losing hand. Second, five trades is a small sample on which to stake conviction.
The verdict is split. The out-of-sample consistency is genuinely appealing, and the thesis is sound. But the failed validation, the heavy drawdown, and — most urgently — the live book's inability to trade point to a strategy that is promising on paper and paralysed in practice. Freeing up cash to restore rotation is the first thing to fix.