The thesis
Bollinger-reversion is a textbook mean-reversion play: buy a name when it closes below its lower Bollinger band, sell when it pushes above the upper band. It trades a 24-stock universe of large-cap US names — AAPL, MSFT, NVDA, JPM, JNJ, XOM, CAT and similar blue chips. The premise is that short-term dislocations in liquid, well-covered stocks tend to snap back toward the mean. It's a reasonable prior, and the strategy is currently running live.
The backtest looks good
Over 451 days the backtest returned 17.55% (9.46% CAGR), ending at $11,755.25 from a $10k base. The win rate is a healthy 63.89% across 76 trades, and fees were a trivial $76. On the surface this is exactly what you want from a mean-reversion book: frequent, mostly-winning round trips.
Two numbers temper the enthusiasm. The Sharpe is only 0.66 — modest for a strategy trading this often — and max drawdown reached 20.57%, a real gut-check for a book whose edge is supposed to be stability. Turnover of 1,720% is enormous; this strategy lives or dies on execution and slippage assumptions that a clean backtest flatters.
Validation: the honest part
Here the story turns. Walk-forward validation failed. The encouraging sign is that all four folds were positive. The problem is how they decayed:
| Fold | Return | Sharpe |
|---|---|---|
| 1 | 7.67% | 1.78 |
| 2 | 1.42% | 0.25 |
| 3 | 0.77% | 0.20 |
| 4 | 0.40% | 0.14 |
That is a near-monotonic collapse. The out-of-sample tail returns just 0.4% at a Sharpe of 0.14 — essentially flat. The deflated Sharpe ratio of 0.342 (after adjusting for 6 trials) says the apparent edge largely survives on the strength of the earliest window. The probabilistic Sharpe of 0.814 is less reassuring once you read it alongside the DSR.
Live activity echoes the warning
Recent scheduled runs are heavy on rejections: 2026-10-22 executed 0 of 1, 10-20 and 10-19 each rejected 2, 10-16 rejected 1. Only 10-15 and 10-21 actually traded (NVDA and CAT buys, MCD and NKE sells among them). More tellingly, the live paper account now reads a total of $9,955.16 on $2,666.76 cash — modestly below its starting capital. The backtest's 17.55% is nowhere in sight live.
Verdict
The strengths are real: a defensible thesis, consistent win rate, every fold positive, negligible fees. But the risks dominate. The out-of-sample decay, failed validation gate, 20%+ drawdown, punishing turnover, and a live account treading water all point the same direction — the edge was strongest in the oldest data and has been fading since. Bollinger-reversion earns its live slot as an experiment, not as capital we'd scale. Watch fold-4-like conditions, not the headline 17.55%.