The thesis
bollinger-reversion runs a classic mean-reversion play: buy when price closes below the lower Bollinger band, sell when it pushes above the upper band. It operates 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). It is currently flagged live.
The logic is intuitive and cheap to reason about — it bets that stretched moves snap back. Total fees over the backtest were just $76 with zero FX cost, so this is a low-friction idea even at a punchy 1,720% turnover.
Backtest headline vs. the fine print
On paper the numbers flatter: 17.55% total return over 451 days, a 9.46% CAGR, and a 63.89% win rate across 76 trades. Winning nearly two of every three trades is genuinely attractive for a systematic rule.
But the risk profile is less kind. The Sharpe of 0.66 is mediocre, and a 20.57% max drawdown means investors had to stomach real pain to collect that return. High win rate plus deep drawdown is a familiar mean-reversion signature: many small wins punctuated by a few ugly losses when a band break keeps breaking.
Validation says: not so fast
This is where the story turns. Walk-forward validation failed. All four folds were positive, which sounds reassuring — until you read the trajectory. Fold 1 returned 7.67% at a strong 1.78 Sharpe; folds 2 through 4 delivered 1.42%, 0.77%, and just 0.40% respectively, with Sharpe collapsing to 0.14 in the most recent window. The out-of-sample return is essentially flat.
The deflated statistics tell the same tale. A PSR of 0.814 looks decent in isolation, but the Deflated Sharpe Ratio of 0.342 — which penalizes for the 6 trials run — sits well below any comfort threshold. In plain terms: the backtest edge concentrates in the earliest period and fades as we approach the present. That is the hallmark of a rule that worked in one regime and is being overfit forward.
Live behaviour: idle
Recent activity reinforces the caution. The last executed trades were in early June (buys in DIS, GOOGL, PG, WMT, COST; a PG sell). Since late July, every scheduled run reports 0 executed, 1–2 rejected, with cash pinned at $1,001.98 and total equity drifting in the $9,650–$9,960 range — below the $10k starting mark. The strategy simply is not finding qualifying setups, and the ones it does find are being rejected.
Verdict
bollinger-reversion is a clean, low-cost idea with a respectable historical win rate, but the evidence points to a decaying edge rather than a durable one. The failed walk-forward, near-zero out-of-sample return, and weeks of idle rejected runs argue against scaling it. It earns a place on the watchlist — not the capital — until it can demonstrate an edge in the recent fold, not just the oldest one.