The thesis
Bollinger-reversion is a textbook mean-reversion play: buy when price closes below the lower Bollinger band, sell when it pushes above the upper band. The logic is that extreme moves away from a rolling average tend to snap back. It trades a 24-name large-cap universe spanning tech, financials, healthcare, staples and energy — AAPL, MSFT, NVDA, JPM, JNJ, XOM and friends — which gives it enough breadth to find stretched names most weeks.
Headline numbers
On the full backtest the strategy looks respectable: 17.55% total return over 451 days, a 9.46% CAGR, a 63.89% win rate across 76 trades, and a final equity of $11,755 from a $10k start. The catch is the risk profile. Sharpe sits at a modest 0.66, max drawdown reached 20.57%, and turnover is a punchy 1,721% — this is an active book that churns its capital many times over. Fees were trivial ($76 total, no FX cost), so the drag here is risk and churn, not transaction cost.
Where it breaks down
The validation gate failed, and the walk-forward folds explain why. All four out-of-sample folds were positive — a good sign on the surface — but the trend is unmistakable decay:
| Fold | Window | Return | Sharpe |
|---|---|---|---|
| 1 | Aug 2024–Jan 2025 | 7.67% | 1.78 |
| 2 | Jan–Jul 2025 | 1.42% | 0.25 |
| 3 | Jul–Dec 2025 | 0.77% | 0.20 |
| 4 | Dec 2025–May 2026 | 0.40% | 0.14 |
Most of the headline return was earned in the first fold; the edge has been grinding toward zero ever since. The latest out-of-sample window returned just 0.4% at a Sharpe of 0.14 — barely distinguishable from flat. The Probabilistic Sharpe Ratio (0.814) is reassuring, but the Deflated Sharpe Ratio, which penalises for the 6 trials run, drops to 0.342. Once you account for multiple looks at the data, the confidence that this edge is real evaporates.
Recent activity
Live trading has been quiet and a little frustrated. Over the last six scheduled runs the book executed trades on only two days and logged four rejected orders — signals that couldn't fill, likely on sizing or cash constraints. On 28 October it trimmed JPM (7 shares at $339.38) and JNJ (9 at $271.18); on 21 October it bought CAT and sold MCD. Total equity is hovering near breakeven at roughly $10,047, with a large cash balance ($7,472 after the latest sells) that suggests the model isn't finding many bands worth fading right now.
Verdict
The strength is a genuine, intuitive thesis with a solid win rate and controlled costs. The risk is that the backtest flatters a decaying edge: the drawdown is deep for the Sharpe on offer, and the fold-by-fold erosion plus a sub-0.5 DSR is exactly the pattern you'd expect from mild overfitting. It's reasonable to keep it live on a small allocation to gather forward data — but on the current evidence, this is a candidate for scrutiny, not scaling.