The thesis
bollinger-reversion runs one of the oldest ideas in technical trading: buy when price closes below the lower Bollinger band, sell when it pushes above the upper band. The bet is that extreme moves overshoot and snap back to the mean. It trades a broad, liquid universe of 24 US large caps — the usual suspects across tech, financials, healthcare, staples and energy (AAPL, MSFT, JPM, JNJ, XOM and friends) — which keeps the mechanics simple and the fills clean.
Backtest: solid on paper
Over 451 days the strategy returned 17.55%, ending at $11,755 on a $10k base, for a 9.46% CAGR. The win rate is a healthy 63.89% across 76 trades, and total fees of just $76 barely dent the result. That said, two numbers deserve a raised eyebrow. The Sharpe of 0.66 is modest for a strategy with a 20.57% max drawdown — you are taking meaningful equity-curve pain for a fairly ordinary risk-adjusted return. And turnover of 1,720% is high; this is an active strategy whose edge has to survive a lot of churn.
Validation: this is where it breaks
The walk-forward validation did not pass, and the reason is instructive. All four folds are positive, which sounds reassuring — but the returns decay monotonically as the test window moves forward:
| Fold | Window | Return | Sharpe |
|---|---|---|---|
| 1 | 2024-08 → 2025-01 | 7.67% | 1.78 |
| 2 | 2025-01 → 2025-07 | 1.42% | 0.25 |
| 3 | 2025-07 → 2025-12 | 0.77% | 0.20 |
| 4 | 2025-12 → 2026-05 | 0.40% | 0.14 |
The headline 17.55% is front-loaded into the earliest fold. Out of sample, the strategy returns just 0.4% at a Sharpe of 0.14 — statistically indistinguishable from flat. The deflated Sharpe ratio (DSR 0.342) confirms it: after accounting for the 6 trials run, the edge does not clear a significance bar, even though the probabilistic Sharpe (PSR 0.814) looks superficially fine. This is a classic signature of a decaying or over-fit signal, not a durable one.
Live activity: mostly waiting
Recent scheduled runs reinforce the caution. From 17–24 September, most days executed zero trades with a single rejected order — the strategy is finding few valid band breaches and gating the rest. The account total sits around $9,647 as of 24 September, below the $10k line, so live paper performance has been slightly negative. The last real activity was 18 and 22 September (buys in NKE and MCD, sells in GOOGL and WMT).
Verdict
bollinger-reversion is a clean, cheap, high-win-rate strategy with a genuinely positive backtest — but the validation tells the honest story. The edge is concentrated in one early window and fades to noise thereafter, and the live account is currently underwater. Treat it as a monitoring candidate, not a conviction allocation, until out-of-sample performance shows signs of re-firing.