The thesis
bollinger-reversion runs one of the oldest ideas in quantitative trading: fade the extremes. It buys when price closes below the lower Bollinger band and sells when it pushes above the upper band, betting that stretched prices snap back toward their moving average. The universe is 24 large-cap US names — mega-cap tech (AAPL, MSFT, NVDA), financials (JPM, V, MA), staples (PG, KO, WMT, COST) and healthcare (JNJ, UNH, PFE). It is a liquid, well-diversified book, and the strategy is currently live.
Recent activity
The last two weeks have been quiet. Scheduled runs on 2026-10-08 through 10-14 mostly executed nothing — a string of "0 executed" days, with one buy on 10-07 and a lone rejection on 10-14. That is characteristic of a band-touch strategy in a calm tape: no signal, no trade. The most recent executions tell the mean-reversion story cleanly — buying JPM (7 @ $328.78), JNJ (9 @ $255.36) and MCD (9 @ $250.63) on the dip side, while trimming COST (2 @ $934.52) and CAT (1 @ $847.03) into strength. The paper account sits around $9,692 total with $493.66 in cash, drifting in the $9,600–$9,790 range over the window — below the $10k round number rather than compounding above it.
Backtest performance
Over 451 days the strategy returned 17.55% (9.46% CAGR), finishing at $11,755 from an implied $10k start. Win rate is a healthy 63.89% across 76 trades, and max drawdown of 20.57% is tolerable for equity. The Sharpe of 0.66, however, is modest — this is a strategy that wins often but whose risk-adjusted return is unremarkable. One flag worth naming: turnover of 1,721%, meaning the book churns roughly 17x. Fees were only $76 in the backtest, but real-world slippage on that churn is a live risk the simulation understates.
Validation: the sobering part
Here the picture turns. Validation failed. All four walk-forward folds were positive — superficially reassuring — but returns decay monotonically: fold 1 delivered 7.67% at a 1.78 Sharpe, then 1.42%, 0.77%, and finally 0.40% at a 0.14 Sharpe in the most recent fold. The out-of-sample return is effectively flat. While the probabilistic Sharpe ratio (0.814) looks acceptable, the deflated Sharpe ratio is just 0.342 across 6 trials — below the threshold for confidence that the edge is real rather than a product of selection.
Verdict
Strengths: high hit rate, sensible diversification, disciplined fade logic. Risks: a Sharpe that is thin to begin with, decaying out-of-sample performance, heavy turnover, and a formal validation failure. The full-sample return flatters a strategy whose recent folds suggest its edge has largely dissipated. Watch, but don't scale.