The thesis
bollinger-reversion is a classic mean-reversion play: buy when price slips below the lower Bollinger band, sell when it pushes above the upper band. It trades a 24-name large-cap universe spanning tech (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare (JNJ, UNH, PFE, ABBV) and staples/industrials (PG, KO, WMT, COST, CAT, HON). The logic is intuitive — fade the extremes and collect the snap-back — and the strategy is a built-in, hand-written approach that cleared review at risk 0.
Recent activity
The strategy is live and trading on schedule. Over the last week it has been a steady buyer: 8 shares of GOOGL at $283.35 and 3 of CAT at $647.16 on 2026-10-29, 4 of MSFT at $514.53 on 2026-11-02, and 7 of AAPL at $329.82 on 2026-11-03. Every scheduled run executed cleanly with zero rejections.
That said, the live book is modestly underwater. Account total has drifted from $9,957.81 on 2026-10-29 to $9,830.18 on 2026-11-03 — a small decline on a $10k base. It is early days, and a handful of fresh long positions sitting on the lower band is exactly what the thesis prescribes, but the direction so far is down rather than up.
Backtest performance
On paper the numbers look solid. Across 451 days the strategy returned 17.55% (final equity $11,755.25), a 9.46% CAGR, with a 63.89% win rate over 76 trades. Fees were a trivial $76 with no FX cost. The weak spots are a Sharpe of just 0.66 and a 20.57% max drawdown — meaning the headline return came with real volatility and a stomach-churning peak-to-trough dip. Turnover of 1,720% also confirms this is an active, high-churn system.
Validation: the warning light
Here is where the story turns. The validation gate failed. Walk-forward testing across four folds is superficially encouraging — all four were positive — but the pattern is damning. Fold 1 (Aug 2024–Jan 2025) did almost all the work: +7.67% at a 1.78 Sharpe. Every fold since has decayed: +1.42%, +0.77%, and finally +0.40% in the most recent out-of-sample window, where Sharpe collapsed to 0.14.
The statistical gate agrees. Against 6 trials, the Deflated Sharpe Ratio is just 0.342 — well short of the confidence needed to call the edge real rather than a product of selection. The probabilistic Sharpe (0.814) is friendlier, but the out-of-sample return of 0.4% is the figure that matters, and it is barely distinguishable from zero.
Verdict
bollinger-reversion is a reasonable, cheap-to-run idea with a genuine win-rate edge and a clean execution record. But the evidence points to a front-loaded backtest: strong early, fading hard, and failing validation precisely because its recent out-of-sample performance is flat. Combined with a 20% drawdown and a live book slightly in the red, this is a strategy to keep on a short leash — watch whether the recent buys revert as advertised, and treat any paper gains with scepticism until the out-of-sample window earns back its keep.