The Thesis
Bollinger-reversion is about as classical as systematic trading gets: buy when price slips below the lower Bollinger band, sell when it pushes above the upper band. The strategy trades a broad, liquid universe of 24 US large caps spanning tech, financials, healthcare, staples, and energy. The implicit bet is that these names are range-bound enough that extreme moves snap back — a reasonable prior for mega-cap staples like PG, KO, and WMT, a shakier one for momentum-prone names like NVDA.
Backtest Performance
The headline numbers are presentable. Over 451 days the strategy returned 17.55% (9.46% CAGR), ending at $11,755.25 on a notional $10k. The win rate is a healthy 63.89% across 76 trades, consistent with a mean-reversion profile that clips many small winners. Fees were a negligible $76 and FX cost was zero.
The caveats sit right next to those figures. The Sharpe is only 0.66 — modest for a strategy with a 64% hit rate, which tells you the losers are larger than the winners. Max drawdown reached 20.57%, and turnover is an eye-watering 1,720%, meaning the book churns its capital roughly 17 times a year. That pace survives in a backtest with trivial fees; it is far less forgiving against real spreads and slippage.
The Validation Problem
Here is where enthusiasm should cool. Walk-forward validation fails. On paper the four folds all print positive returns, which looks reassuring — until you read them in order:
| 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 monotonic decay. Nearly all the edge lives in the first fold (Aug 2024–Jan 2025); the most recent out-of-sample window returned just 0.40% at a Sharpe of 0.14. The deflated Sharpe ratio of 0.342 — which penalises the 6 trials run — and the gap between a flattering PSR of 0.814 and that DSR confirm the pattern: the full-sample result is more a story about one good regime than a durable, repeatable edge.
Live Behaviour
The live book echoes the warning. Across the last six scheduled runs (24 Sep–1 Oct), the strategy executed zero trades while rejecting one on most days, sitting on $2,398.39 cash and a total value drifting between roughly $9,600 and $9,734 — below the $10k mark. The last real activity was in late September (buying MCD and NKE, selling WMT and GOOGL). A signal engine that keeps proposing trades the risk layer rejects is either starved of qualifying setups or fighting its own constraints.
Verdict
Bollinger-reversion is a clean, interpretable strategy with a genuine backtest and a strong win rate — but its edge is decaying out-of-sample, its validation gate is red, and its live equity is underwater. It earns a place on the watchlist, not fresh capital. The honest read: this looks like a regime that worked in 2024, not a strategy that works.