The thesis
Channel-pullback is a textbook mean-reversion play layered onto a trend filter: buy dips toward the lower regression channel or volume support only in confirmed uptrends, then sell into the upper channel or overhead resistance. It runs across a 24-name large-cap universe spanning tech, financials, healthcare, and staples — AAPL, MSFT, NVDA, JPM, V, JNJ, WMT, XOM and peers. The design intent is sensible: let the trend keep you on the right side, and use the channel geometry to time entries and exits rather than chase breakouts.
Recent activity
The strategy is live and trading on schedule. Its most active recent session was 2026-09-11, when it executed three sells — trimming V at $371.31, AAPL at $332.75, and UNH at $377.29 — leaving cash at $5,730.06 against a total account value of $9,397.67. Preceding days show the same rhythm: a V buy at $366.62 on the 10th, a BAC round-trip (bought at $63.04, sold at $62.79), and a cluster of AAPL trades in late August. Several scheduled runs (2026-09-04, -07, -08) produced zero trades, which is expected when no name sits at a channel extreme.
Worth flagging: the live account total of ~$9,400 sits below a $10,000 baseline. The live book is currently down, even as the backtest shows a gain — a reminder that recent realized behaviour is lagging the historical record.
Backtest and validation
Over 451 days the backtest returned 7.62% (final equity $10,761.52, ~4.19% CAGR) on 137 trades. The headline caveats are the same ones that make me cautious:
- Sharpe of 0.40 is thin — modest reward for the volatility taken.
- Max drawdown of 14.83% against a single-digit CAGR is an uncomfortable ratio.
- Win rate of 39.39% means the edge depends entirely on winners outrunning losers; there is no margin for the average win to shrink.
- Turnover of 2,311% is high, so fees and slippage matter more than the $137 modelled cost suggests.
Walk-forward validation did not pass. Three of four folds were positive, but the spread is stark: fold 3 (2025-07-05 to 2025-12-16) returned +20.68% at a Sharpe of 3.86, while fold 2 (2025-01-22 to 2025-07-05) lost 11.42% at a Sharpe of -1.70 with a 16% drawdown. Out-of-sample return was a soft +3.63% (Sharpe 0.74). The deflated Sharpe ratio of 0.196 — with 7 trials behind it — and a probabilistic Sharpe of 0.702 tell the honest story: once you account for how many variants were tried, the evidence for a durable edge is weak.
Verdict
Channel-pullback has a coherent thesis and one genuinely excellent regime (fold 3), but its performance is regime-dependent, its risk-adjusted return is low, and it flunks the validation gate. The live drawdown reinforces the concern. I'd keep it running as a paper strategy to gather more out-of-sample data, but it does not yet clear the bar for higher conviction.