The thesis
Donchian-breakout is a classic trend-following rule: buy when price makes a fresh 20-day high, exit when it prints a 20-day low. It trades a 24-name universe of US large caps spanning tech (AAPL, MSFT, NVDA), payments (V, MA), staples (PG, KO, WMT), healthcare (JNJ, UNH), and energy (XOM, CVX). The logic is simple and well-understood: cut losers quickly at the lower channel, let winners ride the upper channel. That design shows up cleanly in the stats — a 38.46% win rate across 108 trades. Fewer than four trades in ten make money; the system lives or dies on whether the winners are big enough to pay for the many small stop-outs.
Recent activity
The live paper book is quiet and slightly underwater. Across the six most recent scheduled runs (18–25 September), most executed zero trades and several logged rejected orders — on 25 September, 0 executed and 2 rejected, with total equity at $9,911.22 against roughly $3,513 in cash. The one burst of action was 22 September: four fills, including buys of WMT (1 @ $109.83) and AAPL (4 @ $340.31) alongside sells of V and MA. The repeated rejections suggest signals are firing but being blocked — most plausibly by cash or position-sizing limits — while the portfolio drifts just below its $10k starting base.
Backtest and validation
The headline backtest turned $10,000 into $10,695 over 451 days: +6.95% total, a 3.83% CAGR, and a modest Sharpe of 0.34. Two numbers deserve caution. First, the maximum drawdown of 21.73% is large relative to the return earned — an uncomfortable pain-to-gain ratio. Second, turnover of 2,083% signals heavy churn, though fees stayed contained at $108.
The walk-forward picture is more encouraging: 3 of 4 folds positive, with the two most recent folds strong (fold 3: +14.09%, Sharpe 2.72; fold 4: +11.35%, Sharpe 1.86). The blemish is fold 2 (Jan–Jul 2025): −7.46% at a −1.07 Sharpe — the kind of choppy, whipsaw regime that punishes breakout systems.
Despite that, validation failed. The reason is statistical discipline, not recent performance: with 6 trials behind the search, the deflated Sharpe ratio (DSR) is just 0.198 — after adjusting for how many variants were tried, confidence that the true edge is above zero is weak. The probabilistic Sharpe (PSR 0.674) is respectable but not enough to clear the multiple-testing hurdle.
Verdict
There is a real trend-following signal here, and the last two folds are impressive. But the full-period Sharpe is thin, the drawdown is steep, and the anti-overfitting gate is right to withhold approval. Watch whether live execution can convert breakout signals into fills — and whether it can hold up in the next fold-2-style chop.