The thesis
Donchian-breakout is trend-following in its purest, oldest form: buy the 20-day high, sell the 20-day low. The logic is that a new one-month high is evidence a trend has begun, and a new one-month low is evidence it has ended. It trades a diversified large-cap universe of 24 names spanning tech, financials, healthcare, staples and energy — from AAPL and NVDA to KO, XOM and JPM. There is no forecasting here, only reaction, which is both the design's charm and its limitation.
Backtest performance
Over the 451-day backtest the strategy returned 6.95% (a 3.83% CAGR), ending on £10,695 of equity from £10,000. The headline numbers are honest about what trend-following costs you: a Sharpe of just 0.34, a 21.73% maximum drawdown, and a win rate of only 38.46% across 108 trades. Losing more often than you win is expected for breakout systems — the model survives on a handful of large winners paying for many small losers. Less comfortable is the turnover of 2,083%, which churns capital hard and racked up £108 in fees on a £10k book.
Validation: the gate says no
This is where balance matters. Our walk-forward validation fails the strategy, and it's worth understanding why the failure sits alongside some genuinely encouraging out-of-sample numbers. Of four folds, three were positive, and the two most recent were strong: fold 3 returned +14.09% (Sharpe 2.72) and fold 4 +11.35% (Sharpe 1.86). The blended out-of-sample Sharpe of 1.86 looks excellent.
So why the red flag? Fold 2 was ugly — -7.46% at a -1.07 Sharpe — showing the model can bleed for months in choppy, trendless markets. More decisively, the Deflated Sharpe Ratio is 0.198: once we penalise for the 6 configurations trialled, the evidence that this edge is real rather than lucky is weak. The PSR of 0.674 is suggestive but short of conviction. The gate is doing its job: promising recent folds are not the same as a proven strategy.
Recent activity: quietly idle
The live account tells a cautionary story. The last executed trades were on 12–13 August (buying V and MA, selling AAPL and UNH). Since then, six consecutive scheduled runs have executed zero orders — every candidate rejected — with cash pinned at $1,510.88 and total equity drifting between roughly $9,755 and $10,118. No new 20-day highs are clearing the model's filters, so it sits on its hands. For a breakout system in a rangebound tape, that inactivity is arguably correct behaviour, not a bug.
Verdict
Donchian-breakout is a clean, interpretable trend-follower with real recent out-of-sample strength but a failing robustness grade, a punishing drawdown profile and a low deflated Sharpe. Treat it as a candidate under observation, not a conviction holding — the next trending regime will tell us far more than the last backtest did.