The thesis
donchian-breakout is a textbook trend follower: buy a 20-day high, sell a 20-day low. It runs on a universe of 24 US large-caps — the mega-cap tech names, the banks, the staples, the industrials. The appeal is simplicity and discipline: no forecasting, just ride confirmed momentum and cut positions mechanically when the trend rolls over. It is currently flagged live.
Recent activity
The last six scheduled runs (13–20 October) were quiet and, frankly, frustrating. Executions were sparse — 2, 0, 2, 0, 0, 0 — while rejected orders piled up on most days. Total equity drifted in a narrow $9,760–$9,794 band with roughly $2,241 parked in cash. Worth stating plainly: against a $10,000 paper base, the book is underwater in live trading.
The executed trades tell the strategy's story cleanly. On 30 September it bought 10 NVDA at $230.43; on 15 October it sold the same 10 at $217.62 — a textbook stop-out on a 20-day low, taking the small loss exactly as designed. It rotated into WMT (21 shares at $118.23) and nibbled MSFT and AAPL, while trimming CAT and KO. This is trend following behaving honestly: frequent small losses, waiting for the occasional large winner.
Backtest and validation
Over 451 days and 108 trades the backtest returned 6.95% (final equity $10,695), a CAGR of 3.83%. The headline numbers are mediocre: a Sharpe of 0.34, a win rate of 38.46%, and a stomach-testing max drawdown of 21.73%. Turnover was extreme at 2,083%, though fees stayed negligible at $108.
The walk-forward picture is more interesting. Three of four out-of-sample folds were positive, and the two most recent were genuinely strong — fold 3 returned 14.09% at a 2.72 Sharpe, fold 4 returned 11.35% at 1.86. Fold 2, however, was ugly: −7.46% at a −1.07 Sharpe with a 15.58% drawdown. The out-of-sample aggregate (11.35% return, 1.86 Sharpe) actually flatters the full-sample result.
Despite that, validation did not pass. The probabilistic Sharpe sits at 0.674, but the deflated Sharpe — which penalises for the 6 trials searched — collapses to 0.198. That is the gate doing its job: the recent out-of-sample strength is not statistically distinguishable from luck once we account for how many variants were tried.
The verdict
Strengths: the logic is transparent, losses are cut mechanically, and recent regimes have suited it. Risks: a 22% drawdown, a low win rate that demands patience, and — most importantly — a failed validation gate and a live book that is currently below its starting line. The honest read is a strategy with a plausible edge that we cannot yet prove is real. It stays live on a short leash, not on conviction.