The thesis
donchian-breakout is about as pure a trend-following rule as they come: buy a 20-day high, exit on a 20-day low. It runs across a 24-name universe of US large caps — mega-cap tech (AAPL, MSFT, NVDA), payments (V, MA), healthcare (JNJ, UNH, ABBV) and staples (KO, PG, WMT). The premise is old and well-documented: let breakouts run, cut losers quickly, and accept that most trades will be small losses in exchange for a handful of large winners.
Backtest and validation
The headline backtest covers 451 days: +6.95% total return, a 3.83% CAGR, and a Sharpe of just 0.34. The win rate is 38.46% across 108 trades — exactly the profile you'd expect from breakout trading, where being right less than half the time is normal. Two figures deserve a hard look. The 21.73% max drawdown is steep for a single-digit return, and turnover of 2,082% means the book churns roughly twenty times over, so the modest $108 in fees flatters a strategy that is genuinely trade-heavy.
Walk-forward validation is where the picture sharpens — and sours. Three of four folds were positive, with the two most recent windows returning +14.09% (Sharpe 2.72) and +11.35% (Sharpe 1.86). Out-of-sample results genuinely look better than the full-sample blend. But fold 2 (Jan–Jul 2025) lost 7.46% at a Sharpe of −1.07, and the deflated Sharpe ratio lands at 0.198 against a probabilistic Sharpe of 0.674. Adjusted for the six trials behind this candidate, the edge is not distinguishable from noise. Validation status: failed. That verdict is the honest one, and we're keeping it visible rather than cherry-picking the strong recent folds.
Recent activity
Live behaviour lately has been notably inert. The scheduled runs from Aug 19–26 all report 0 executed with one or two orders rejected each day, cash parked at $1,510.88, and total equity oscillating in a tight band around $10,100–$10,260. That's a breakout system finding no fresh 20-day highs it can act on — either the universe is range-bound or position/cash constraints are blocking entries. The last real fills were in mid-August (buying V and MA, trimming AAPL and UNH) and a cluster back in June.
The balance
Strengths: the rule is simple, mechanical, and its recent out-of-sample folds are legitimately strong, with contained drawdowns in the 3–6% range. Risks: the full-sample Sharpe is thin, the drawdown-to-return ratio is unattractive, turnover is high, and — most importantly — the deflated statistics say the apparent edge may be a multiple-testing artefact. A live book that currently can't find a trade only underscores that this is a regime-dependent strategy: it will do nothing for stretches, then live or die on a few breakouts. We'd want to see it clear the validation gate, not just post two good folds, before leaning on it.