The thesis
Channel-pullback is a mean-reversion strategy dressed in trend-following clothes. It buys pullbacks to the lower regression channel or volume support only inside confirmed uptrends, then exits at the upper channel or resistance. The logic is intuitive: don't fight the trend, but don't chase it either — wait for the dip and sell the bounce. It runs across a 24-name large-cap universe spanning tech, financials, healthcare, staples, and energy.
Recent activity
The strategy is live and trading on a scheduled cadence, typically executing one order per run. Over the past two weeks it has churned almost entirely between two names — BAC and CAT — with a single earlier AAPL entry. The BAC activity is textbook for the thesis: buy at $64.07 (Aug 13), sell at $64.52 (Aug 14), re-buy at $64.24 (Aug 17). Small, frequent round-trips. Portfolio total equity has drifted gently down from ~$9,794 to ~$9,646 across the window, so the recent live tape has been slightly negative — consistent with a low win rate grinding through chop.
Backtest performance
Over 451 days the backtest returned +7.62% (CAGR 4.19%), ending at $10,761 on a $10k base. The headline numbers are honest but unglamorous:
- Sharpe 0.40 — low; returns aren't well compensated for the volatility.
- Max drawdown 14.83% — meaningful for a single-digit annual return.
- Win rate 39.39% across 137 trades — the strategy loses more often than it wins, so profitability depends entirely on winners outrunning losers.
- Turnover 2,311% — very high; this is an active, fee-sensitive engine.
Validation: the sobering part
Walk-forward validation did not pass, and the fold-by-fold breakdown shows why. Three of four folds were positive, but the dispersion is severe:
| Fold | Return | Sharpe | Max DD |
|---|---|---|---|
| 1 | +6.53% | 1.25 | 6.16% |
| 2 | -11.42% | -1.70 | 16.09% |
| 3 | +20.68% | 3.86 | 3.26% |
| 4 | +3.63% | 0.74 | 8.44% |
Fold 3 does most of the heavy lifting; strip it out and the record looks thin. The out-of-sample return was +3.63% with a Sharpe of 0.74 — better than the full-sample 0.40, which is encouraging, but on a small sample. Most telling: the Probabilistic Sharpe Ratio is 0.702 while the Deflated Sharpe Ratio collapses to 0.196 after adjusting for 7 trials. That gap is the overfitting tax — once you account for how many variants were tested, the confidence that this edge is real drops sharply.
Verdict
Channel-pullback has a coherent, defensible thesis and a positive full-period return, and its OOS Sharpe holding above the in-sample figure is a genuine plus. But the risks are real: a low win rate, high turnover eating into edge, one badly negative fold, and a deflated Sharpe that barely clears zero. It earns its place as a live paper-trading experiment worth watching — not yet a strategy to size up with conviction. The next few folds of live data will tell us whether fold 2 was noise or a preview.