The Thesis
bollinger-reversion is a classic mean-reversion play: buy when price closes below the lower Bollinger band, sell when it pushes above the upper band. It runs on a 24-name large-cap universe spanning tech (AAPL, MSFT, NVDA), financials (JPM, V, MA), staples (PG, KO, WMT, COST) and healthcare (JNJ, UNH). The logic is intuitive — stretched prices snap back — and staples-heavy exposure fits that mandate, since defensive names tend to oscillate around a mean rather than trend hard.
Backtest Performance
On paper, the numbers look inviting. Over 451 days the strategy returned 17.55% (9.46% CAGR), with a 63.89% win rate across 76 trades. That hit rate is genuinely strong and is the strategy's headline strength. But the supporting stats temper the enthusiasm: a Sharpe of 0.66 is mediocre, the max drawdown of 20.57% is uncomfortably deep for a defensive-leaning book, and turnover of 1,721% signals heavy churn — this is not a set-and-forget approach, and fees (though modest here at $76) will scale with activity.
Validation — Where It Breaks Down
Here is the problem: walk-forward validation did not pass. All four out-of-sample folds were positive, which sounds encouraging until you read the trend. Fold returns decay monotonically — 7.67%, 1.42%, 0.77%, 0.40% — and fold Sharpe collapses in lockstep: 1.78 → 0.25 → 0.20 → 0.14. The most recent fold (Dec 2025–May 2026) earned a near-flat 0.40% at a Sharpe of 0.14. The edge isn't just weak; it's fading in real time.
The robustness statistics agree. While the Probabilistic Sharpe Ratio of 0.814 says the Sharpe is likely above zero, the Deflated Sharpe Ratio of 0.342 — which penalizes for the 6 trials run — sits well below the 0.5 confidence threshold. In plain terms: after accounting for how many variants were tested, there's a better-than-even chance this Sharpe is a product of selection, not skill. That is precisely why the gate blocked it.
Live Activity — Stalled
The live tape confirms the concern. The last executed trades date to early June 2026 (buys in DIS, GOOGL, PG; a WMT and COST entry). Since late July, every scheduled run tells the same story: 0 executed, 1–2 rejected, cash pinned at $1,001.98 while total equity drifts between ~$9,640 and ~$9,936 — below the $10k starting line. The engine wants to trade and can't fill; rejected orders warrant an operational look (sizing, buying power, or price limits).
Verdict
bollinger-reversion is a well-behaved idea with a high win rate and a clean, defensible thesis — but the evidence points to an edge that worked in 2024, weakened through 2025, and is now flat. The failed validation, sub-threshold DSR, and stalled live book all argue for keeping this one in paper mode. Watch the rejection cause first, then revisit only if a fresh out-of-sample window shows the Sharpe stabilizing rather than decaying.