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 trades a 24-name large-cap universe spanning tech, financials, healthcare, staples and energy (AAPL, MSFT, NVDA, JPM, JNJ, XOM and friends). The bet is that short-term dislocations snap back — a sensible premise in liquid, well-behaved blue chips.
Backtest performance
On paper the headline reads well. Over 451 days the strategy turned $10,000 into $11,755.25 — a 17.55% total return, 9.46% CAGR — with a 63.89% win rate across 76 trades. That hit rate is genuinely high.
But the quality numbers are more sober. The Sharpe ratio is 0.66, and the max drawdown is 20.57% — a hard 1:5 return-to-pain ratio. The gap between a 64% win rate and a middling Sharpe is a familiar mean-reversion signature: many small winners offset by occasional large losers when a dislocation keeps dislocating. Turnover is also heavy at 1,720%, so the modest $76 in fees would scale uncomfortably at larger size.
Why validation fails
This is where the story turns cautionary. Walk-forward testing splits the history into four folds, and all four are positive — which sounds reassuring until you watch the trend: 7.67%, then 1.42%, then 0.77%, then 0.40%. The out-of-sample fold delivers just 0.40% at a Sharpe of 0.14. The entire 17.55% is essentially earned in fold one and never repeated.
The statistics confirm the suspicion. The probabilistic Sharpe ratio (PSR) is a healthy 0.814, but the deflated Sharpe ratio (DSR) — which penalises for the six trials run — collapses to 0.342. That deflation is why the validation gate returns passed: false: once we account for how many variants were tested, the edge is not convincingly distinguishable from luck.
Recent live activity
Live paper trading echoes the stall. Across the last six scheduled runs the strategy executed almost nothing — mostly "0 executed" with one or two rejections — and portfolio value has drifted sideways around $9,850–$10,015, with roughly $2,667 (27%) sitting in cash. The last real activity was 21 October: a 3-share CAT buy and a 9-share MCD sell. Earlier rotations (NVDA, JPM, JNJ in, NKE, COST, WMT out) show the engine working as designed; it simply isn't finding many band breaches lately.
Verdict
Strengths: a real positive return, a high win rate, four positive folds and a solid PSR. Risks: a steep performance decay, a flat out-of-sample fold, a low DSR, a 20%+ drawdown and high turnover. The honest read is that bollinger-reversion looks like an edge that was strongest early and has faded — promising enough to keep observing live, not yet validated enough to trust with conviction.