The thesis
The donchian-breakout agent runs one of the oldest ideas in systematic trading: buy the 20-day high, sell the 20-day low. It trades a 24-name universe of US large caps — the megacap tech complex (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare, and staples. The bet is simple and honest: catch sustained trends, ride them, and cut losers when price rolls over through the lower channel.
Recent activity
The last two weeks have been quiet — arguably too quiet. Every scheduled run from August 14 through August 20 executed zero trades while logging one or two rejected orders, with cash pinned at $1,510.88 and total account value drifting between roughly $9,755 and $10,087. The last real fills came on August 12–13: buys in V (5 @ $362.35), MA (3 @ $559.85), and KO earlier, against sells in AAPL (7 @ $301.65) and UNH. A run of rejected orders alongside a thin cash balance suggests the strategy is signalling entries it cannot fund — a capital-allocation friction worth watching rather than a signal failure.
Backtest and validation
Over 451 days the strategy returned +6.95% (3.83% CAGR) on a Sharpe of just 0.34, with a 21.73% max drawdown and a 38.46% win rate across 108 trades. None of that is flattering in isolation — but it is textbook trend following: most trades are small losers, and a minority of big winners carry the book. The 2,083% turnover confirms an active, whipsaw-prone system.
The walk-forward picture is more interesting. Three of four folds were positive, and the two most recent — fold 3 (+14.09%, Sharpe 2.72) and fold 4 (+11.35%, Sharpe 1.86) — were genuinely strong, with shallow drawdowns. The out-of-sample segment posted +11.35% at a 1.86 Sharpe. The ugly stretch was fold 2 (Jan–Jul 2025): -7.46% at Sharpe -1.07, a choppy, trendless market that punished breakouts exactly as theory predicts.
The verdict from the gate
Crucially, validation did not pass. The probabilistic Sharpe (PSR 0.674) is decent, but the deflated Sharpe (DSR 0.198) — which penalises for the 6 trials run — is low, meaning we cannot confidently rule out that the full-sample edge is noise dressed up by selection. That is the right call: the recent folds look great, but one bad regime and a weak deflated score argue for humility.
Balance sheet
Strengths: a coherent, non-overfit thesis; genuinely strong recent out-of-sample behaviour; disciplined exits. Risks: a sub-0.35 headline Sharpe, a 22% drawdown, a low win rate that demands patience, and a failed validation gate. Trend following pays in trends and bleeds in ranges — this agent is honest about which one it is. It stays live, but on a short leash.