The thesis
bollinger-reversion runs one of the oldest ideas in technical trading: buy weakness, sell strength. It enters when price closes below the lower Bollinger band and exits above the upper band, betting that stretched moves snap back toward the mean. The universe is 24 large-cap US names — the megacap tech complex (AAPL, MSFT, NVDA, GOOGL), payments (V, MA), staples (PG, KO, WMT, COST), and healthcare (JNJ, UNH, PFE). These are liquid, well-behaved instruments, which is exactly where band-based reversion tends to behave itself. The strategy is live.
What the backtest says
On paper, the numbers look inviting. Over 451 days the strategy returned 17.55% (9.46% CAGR), turned $10,000 into $11,755, and won 63.89% of its 76 trades. A win rate that high is characteristic of mean reversion — many small wins punctuated by occasional larger losses. The catch shows up in the risk column: a 20.57% max drawdown against a Sharpe of just 0.66, and eye-watering 1,720% turnover. Fees were a modest $76, but that churn is a structural drag and a fragility signal.
Where it breaks: validation
This is the section that matters, and it is not flattering. The walk-forward validation failed. Superficially all four folds were positive, but the returns decay monotonically: 7.67% → 1.42% → 0.77% → 0.40%, with fold Sharpe collapsing from 1.78 to 0.14. Nearly all of the headline profit was earned in the first fold (Aug 2024–Jan 2025); everything since has been noise around zero. The out-of-sample return is 0.40% at a 0.14 Sharpe — effectively flat.
The deflated statistics confirm the story. The Probabilistic Sharpe Ratio is a healthy 0.814, but the Deflated Sharpe Ratio is only 0.342 after accounting for 6 trials. A DSR below the conventional 0.5 bar means we cannot reject the hypothesis that the backtest edge is a product of selection rather than skill. The most recent fold also carried an 11.19% drawdown for that 0.40% of return — poor compensation for the risk taken.
Live activity
Live behaviour matches a strategy whose signal has gone quiet. Recent scheduled runs mostly execute nothing: the six latest sessions logged just two executions and five rejections, with the book sitting on $2,398 of idle cash and a total value around $9,734 — marginally below the $10,000 line. The last real activity was 22 September (a MCD buy and WMT sell); before that, trades are weeks apart. Reversion simply is not being triggered often in this tape.
Verdict
Strengths are real: a coherent, well-understood thesis, a clean liquid universe, and a high hit rate. But the risks dominate. The edge is front-loaded, fades out of sample, and does not survive deflation for multiple trials — the classic overfitting signature. High turnover magnifies that fragility. This strategy earned its live slot on backtest optics, not validated skill. I would keep it on a short leash, treat the flat live P&L as the honest read, and require a genuine out-of-sample recovery before adding capital.