The Thesis
bollinger-reversion runs one of the oldest ideas in technical trading: price tends to snap back to its mean. The rule is deliberately simple — buy when a name closes below its lower Bollinger band, sell when it pushes above the upper band. It trades a broad, liquid universe of 24 large-cap US names spanning tech, financials, healthcare, staples and energy (AAPL, MSFT, NVDA, JPM, JNJ, XOM and more). It is currently live.
Headline Numbers Look Good
On paper, the backtest is attractive. Over 451 days the strategy returned 17.55% (9.46% CAGR), finishing at $11,755 on a modeled book, with a 63.89% win rate across 76 trades. A win rate near two-thirds is exactly what you'd hope to see from a mean-reversion engine — lots of small, high-probability snap-backs.
But the risk-adjusted picture is more sober. The Sharpe of 0.66 is modest, and a 20.57% max drawdown is a heavy price for a 17.55% return. Turnover of 1,720% is also striking: this is a high-churn strategy, and while modeled fees were only $76, that level of activity leaves it exposed to slippage and cost drag that a paper backtest tends to flatter.
Validation Tells the Real Story
Here is where the strategy earns its skepticism. Walk-forward validation failed. All four folds were positive — a genuine strength — but the trend is unmistakable decay:
- Fold 1 (Aug 2024–Jan 2025): +7.67%, Sharpe 1.78
- Fold 2: +1.42%, Sharpe 0.25
- Fold 3: +0.77%, Sharpe 0.20
- Fold 4 (Dec 2025–May 2026): +0.40%, Sharpe 0.14
The out-of-sample return is just 0.4% with a Sharpe of 0.14 — statistically indistinguishable from noise. The Deflated Sharpe Ratio of 0.342 (below the 0.5 threshold, across 6 trials) confirms it: once we adjust for how many variants were tried, the edge does not clear the bar. The Probabilistic Sharpe Ratio of 0.814 is the lone bright spot, but it can't rescue a strategy whose live-era folds are flat.
Live Behaviour Is Stalling
The recent tape reinforces the concern. The last executed trades were in early June (buys in DIS, GOOGL, PG, WMT, COST; a PG sell). Since then, every scheduled run has executed zero trades and rejected one or two candidates — Jul 30 through Aug 6. The paper book has clawed back from roughly $9,651 to $9,989 over that stretch, but it remains below the round $10,000 mark, and the engine is effectively idling with $1,001.98 in cash.
Verdict
bollinger-reversion is a clean, interpretable strategy with a plausible edge in its early history — but that edge appears to be fading in real time, not just in cross-validation. The failing DSR, the fold-by-fold decay, and a live run of rejected orders all point the same direction. It's worth keeping live as a low-stakes observation, but the data does not yet justify scaling it up.