The thesis
donchian-breakout is a textbook trend-following system: buy when price prints a 20-day high, exit when it prints a 20-day low. It runs long-only over a 24-name large-cap universe — the usual megacap tech (AAPL, MSFT, GOOGL, NVDA), payments (V, MA), healthcare, staples, and a few industrials and energy names (CAT, XOM, CVX). The logic is deliberately simple: let breakouts carry you, cut positions that roll over. That simplicity is a feature, not a shortcut — it leaves little room for curve-fitting.
Recent activity
The last executed orders were on 2026-10-15: a buy of 21 WMT at $118.23 and a trim of 10 NVDA at $217.62. Since then the picture is quieter than it looks. Across six scheduled runs from 2026-10-20 to 2026-10-27, the strategy executed zero trades while logging between one and three rejections each session. Cash has sat frozen at $2,241.07 and total equity has drifted around $9,790–$9,830 — modestly below the $10k seed.
That pattern matters. The engine is still generating signals, but something — most likely available capital or position-sizing constraints — is blocking them. A breakout system that cannot act on its breakouts loses its edge, so the rejections deserve a look before anything else.
Backtest and validation
Over 451 days the backtest returned +6.95% (final equity $10,695), a CAGR of 3.83%, with a Sharpe of just 0.34 and a max drawdown of 21.73%. Win rate is 38.46% across 108 trades — low, but entirely normal for trend following, where a minority of large winners is supposed to pay for many small losers. Turnover is eye-watering at 2,082%, though fees stayed contained at $108.
The walk-forward picture is more flattering than the headline Sharpe: 3 of 4 folds positive, with the two most recent folds strong (+14.09% at Sharpe 2.72, then +11.35% at Sharpe 1.86 out-of-sample). Fold 2 is the scar — a −7.46% stretch at Sharpe −1.07 through the first half of 2025.
Despite that, validation did not pass. The Probabilistic Sharpe Ratio sits at 0.674, but the Deflated Sharpe Ratio — which penalises the 6 trials behind this configuration — collapses to 0.198. In plain terms: once we account for how many variants were tried, the full-sample Sharpe of 0.34 is not convincingly different from luck.
The balance
Strengths are real: coherent trend-following logic, positive and improving out-of-sample folds, and tightly controlled costs. But the risks are equally real — a thin full-sample Sharpe, a 20%+ drawdown, dependence on fat-tailed winners, and a validator that correctly refuses to be dazzled by two good recent folds. Keep it live and observed, resolve the order rejections, but don't scale it until the deflated numbers earn it.