The thesis
donchian-breakout is one of the oldest ideas in systematic trading dressed in modern clothing: buy when price closes above its 20-day high, exit when it closes below its 20-day low. It trades a universe of 24 US large caps — the mega-cap tech names (AAPL, MSFT, NVDA), financials (JPM, V, MA), staples (PG, KO, WMT, COST) and industrials (CAT, HON). The premise is that sustained trends pay for the many false starts that trend-following inevitably generates.
Backtest in brief
Over 451 days the strategy returned 6.95% (final equity $10,695 on a $10k base), a 3.83% CAGR, with a Sharpe of just 0.34 and a painful 21.73% max drawdown. The win rate is 38.46% across 108 trades — low, but that is the expected signature of breakout systems, where a handful of large winners subsidise a majority of small losers. Turnover is a hefty 2,082%, though fees stayed negligible at $108 and there was no FX cost. The headline problem is the reward-to-pain ratio: a sub-7% return earned while enduring a drawdown north of 20% is not a trade most allocators would take.
Validation: failed
This is where the update turns cautionary. The strategy did not pass walk-forward validation. Three of four folds were positive, and the out-of-sample segments looked genuinely strong — fold 3 returned 14.09% (Sharpe 2.72) and fold 4 returned 11.35% (Sharpe 1.86). But fold 2 was ugly: −7.46% with a −1.07 Sharpe across the first half of 2025. More telling are the robustness statistics. The Probabilistic Sharpe Ratio sits at 0.674, but the Deflated Sharpe Ratio is only 0.198 after accounting for 6 trials. In plain terms: once we penalise for how many variants were tried, the evidence that this edge is real rather than lucky is weak. The validation gate is doing its job here.
Recent live activity
Live paper trading reinforces the caution. The six most recent scheduled runs (16–23 October) all report the same pattern — 0 executed, 2–3 rejected — with cash pinned at $2,241.07 and total equity drifting between roughly $9,734 and $9,819. In other words, the live book is currently below its $10k start, and the engine keeps declining proposed orders, likely for sizing or signal-quality reasons worth investigating. The last executed trades were mid-October: a WMT buy and an NVDA sell on the 15th, an MSFT buy and CAT sell on the 13th.
Verdict
Donchian-breakout is a clean, interpretable trend-follower with respectable out-of-sample folds, and its low win rate is a feature, not a bug. But the low deflated Sharpe, the 20%+ drawdown, one badly negative fold, and a live book that is both underwater and persistently rejecting orders add up to a strategy that has not yet earned capital. It stays live for observation, not conviction — exactly the kind of candidate the validation layer exists to flag before it costs real money.