The thesis
Donchian-breakout is a classic trend-following system: buy when price sets a new 20-day high, exit when it prints a 20-day low. It runs on a 24-name large-cap universe spanning tech (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare, staples, and energy. The logic is deliberately unglamorous — ride what's breaking out, cut what's breaking down — and it carries the signature trade-off of the style: you accept a low hit rate in exchange for letting winners run.
That profile shows up directly in the numbers. The backtest's win rate is just 38.46% across 108 trades, which is normal, not alarming, for a breakout system. The concern is what the winners paid for.
Backtest and validation
Over 451 days the backtest returned 6.95% (final equity $10,695), a 3.83% CAGR, with a Sharpe of 0.34 and a 21.73% max drawdown. A sub-0.4 Sharpe against a 22% drawdown is a thin reward for the risk taken. Turnover is the other red flag: 2,082%, meaning capital churned roughly twenty times, with $108 of fees — survivable here, but a real drag that scales badly with size.
Walk-forward validation is more nuanced. Three of four folds were positive, and the out-of-sample window is genuinely good: +11.35% at a 1.86 Sharpe. Fold 3 was excellent (+14.09%, Sharpe 2.72, 3.1% drawdown). But fold 2 lost 7.46% at a Sharpe of −1.07 — the strategy clearly struggles in choppy, non-trending regimes, exactly where breakout systems whipsaw.
Crucially, validation did not pass. The probabilistic Sharpe (PSR 0.674) is middling, and the deflated Sharpe ratio (DSR 0.198) is low once you account for 6 trials of selection. That gap between a strong OOS read and a failed gate is the honest headline: the good numbers may owe something to regime luck and multiple-testing optimism.
Recent activity
The live paper book tells a cautious story. Over the last week of scheduled runs, executions were sparse and rejections frequent — the 2026-10-01 run executed 0 and rejected 1, and most sessions rejected more than they filled. Total equity has drifted from roughly $9,932 down to $9,792, so the strategy is modestly underwater live, not yet confirming its OOS promise. Recent fills include buys in NVDA (10 @ $230.43) and CAT (3 @ $825.98) against exits in KO, V, and MA — breakouts bought, laggards trimmed, as designed.
Verdict
The engine works as specified and the out-of-sample behaviour is encouraging, but a failed validation gate, a 0.198 deflated Sharpe, heavy turnover, and a soft live start all argue for keeping this one on a short leash. Promising, unproven.