The thesis
bollinger-reversion runs a classic mean-reversion play: buy when price dips below the lower Bollinger band, sell when it pushes above the upper band. It trades a 24-name large-cap universe spanning tech, financials, healthcare, staples and energy — AAPL, MSFT, NVDA, JPM, V, JNJ, UNH, PG, COST, XOM, CAT and others. The strategy is currently live.
The logic is intuitive: in range-bound, liquid names, extreme deviations from a moving average tend to snap back. The question is whether that edge holds net of trading friction and across regimes.
Recent activity
The last week has been quiet. Of the six most recent scheduled runs, three executed zero trades, and the others fired just one or two. On 2026-10-07 it bought 7 shares of JPM at $328.78; on 2026-10-06 it sold 2 shares of COST at $934.52; and on 2026-10-02 it bought JNJ (9 @ $255.36) while trimming CAT (1 @ $847.03), with one order rejected that day.
The live book stood at a total value of $9,723.43 as of 2026-10-09, with $493.66 in cash. That sits below the $10,000 starting mark implied by the backtest — a reminder that paper-trading reality is running cooler than the historical curve.
Backtest performance
Over 451 days the strategy returned 17.55%, lifting equity to $11,755.25, a CAGR of 9.46%. The win rate is a healthy 63.89% across 76 trades. On the surface, a strong result.
Two caveats temper it. First, the Sharpe is only 0.66 against a 20.57% max drawdown — the returns are lumpy and the risk-adjusted profile is mediocre. Second, turnover is 1,720%, extraordinarily high churn that generated $76 in fees. A strategy trading this actively is heavily exposed to cost drag and slippage that a clean backtest understates.
Validation: the red flag
This is where the story turns. The strategy fails validation. All four walk-forward folds are positive, which sounds reassuring — but their returns decay almost monotonically: 7.67% → 1.42% → 0.77% → 0.40%, with fold Sharpes collapsing from 1.78 to 0.14. The most recent out-of-sample window (through 2026-05-29) earned just 0.40% at a 0.14 Sharpe.
The deflated metrics confirm the concern. While the probabilistic Sharpe ratio (PSR) is a respectable 0.814, the deflated Sharpe ratio (DSR) is only 0.342 after accounting for 6 trials. A DSR below 0.5 means we cannot confidently reject the hypothesis that the apparent edge is a product of selection rather than skill.
Verdict
bollinger-reversion is a plausible, interpretable strategy with a strong headline backtest and a good win rate. But the evidence that matters most — out-of-sample decay, a weak DSR, high turnover, and a live book tracking below par — argues for caution. The early-period performance looks real; the recent-period performance looks like fading edge. Keep it live on a small allocation and watch whether the next out-of-sample fold breaks the downward trend, but do not scale it until validation passes.