The thesis
Channel-pullback is a classic mean-reversion-in-a-trend idea: buy pullbacks to the lower regression channel or volume support inside confirmed uptrends, then exit at the upper channel or resistance. It trades a 24-name large-cap universe spanning tech (AAPL, MSFT, NVDA), financials (JPM, V, MA), staples (PG, KO, COST) and industrials (CAT, HON). The logic is intuitive — you are trying to buy dips that the prevailing trend is likely to repair — but it lives or dies on whether the "confirmed uptrend" filter is real signal or hindsight.
Recent activity
The agent is actively trading paper capital of roughly $10k. Over the last week it logged steady scheduled runs: 2 executed on 2026-07-14, 6 on 2026-07-13, and 2 each on the surrounding days, with only a single rejection (2026-07-07). Total equity has drifted in a tight band — $10,333 on 2026-07-08 down to $10,020 on 2026-07-14 — so recent live performance is essentially flat.
The trade tape shows the strategy in motion: buys of PG (13 @ $146.44) and AAPL (6 @ $315.56) on the 14th, and a busy 13th that rotated out of GOOGL, NVDA, WMT, DIS and COST while adding CAT. Note it bought DIS at $96.10 on 2026-07-10 and sold 20 shares at $96.17 three days later — a near-scratch round trip that illustrates how thin the per-trade edge can be.
Backtest and validation
Over 451 days the backtest returned 7.62% (final equity $10,761.52), a 4.19% CAGR, with a 0.4 Sharpe and a 14.83% max drawdown. Two numbers deserve scrutiny. First, the 39.39% win rate across 137 trades means the strategy is wrong more often than not and relies on winners outrunning losers. Second, turnover of 2,311% is very high for the return produced — a lot of churn for a mid-single-digit result, and fees ($137) scale directly with that activity.
Most importantly, formal validation failed. Walk-forward across four folds was genuinely mixed: folds 1, 3 and 4 were positive (+6.53%, +20.68%, +3.63%), but fold 2 (2025-01-22 to 2025-07-05) lost -11.42% with a -1.7 Sharpe and a 16% drawdown. Out-of-sample return was a thin 3.63% at a 0.74 Sharpe. The deciding figures are the probabilistic ones: a PSR of 0.702 is acceptable, but the deflated Sharpe (DSR) of 0.196 — which penalises for the 7 trials run — falls short. In plain terms, once you account for how many variants were tested, the edge is not statistically convincing.
Verdict
Strengths: a coherent, well-understood thesis; a liquid universe; and a strong showing in the most recent fold. Risks: a sub-coin-flip win rate, punishing turnover, a real double-digit losing regime in early 2025, and a failed deflated-Sharpe test that flags overfitting risk. Channel-pullback is worth keeping live on paper to gather forward evidence, but it has not earned auto-deployment. Watch whether the recent flat live results resolve toward the fold-3 optimism or the fold-2 warning.