The thesis
Bollinger-reversion is a textbook mean-reversion play: buy a name when it closes below its lower Bollinger band, sell when it pushes above the upper band. It runs on a universe of 24 US large caps spanning tech (AAPL, MSFT, NVDA, GOOGL), financials (JPM, V, MA), staples (PG, KO, WMT) and energy (XOM, CVX). The implicit bet is that band breaks in liquid, well-covered stocks are overreactions that snap back — a reasonable prior, and one with decades of academic support behind it.
Recent activity
The strategy is live but quiet. Its scheduled runs from 9–13 November mostly executed nothing (0 trades on several days, with one or two rejections), and the book is sitting on $1,055.32 of cash. Total equity on the 13 November run was $9,882.99 — modestly below the $10,000 start, so the live deployment is currently down roughly 1%.
The trade tape shows the mechanic working as designed. In early November it bought AAPL at $329.82 and sold the same 7 shares at $352.27 a day later, and did the equivalent round trip on MSFT ($514.53 in, $557.86 out) — exactly the quick band-to-band reversion the thesis predicts. More recent buys (XOM at $166.26, NVDA at $184.15, CAT, GOOGL) are still open positions.
Backtest vs. validation
The headline numbers look appealing: +17.55% total return over 451 days, a 9.46% CAGR, a 63.89% win rate across 76 trades, and only $76 in fees. But two things temper that. First, a 0.66 Sharpe against a 20.57% max drawdown is a mediocre risk-adjusted profile — you absorbed a one-fifth equity hit for single-digit annualised return. Second, and more important, validation failed.
All four walk-forward folds were positive, which sounds healthy, but their returns decay monotonically: 7.67% → 1.42% → 0.77% → 0.40%, with fold Sharpe collapsing from 1.78 to 0.14. The out-of-sample return is just 0.4% at a 0.14 Sharpe. The probabilistic Sharpe ratio (0.814) is respectable, but the deflated Sharpe ratio — which penalises the 6 trials run — drops to 0.342, below a confidence threshold. In plain terms: most of the backtest's edge lives in the earliest, oldest window and has faded as the sample moves forward.
Verdict and risks
The strengths are real — a coherent thesis, a high win rate, low trading costs, and no fold that actually lost money. But the pattern here is the classic overfitting tell: an impressive full-sample curve propped up by one strong early period, with the recent edge indistinguishable from noise. The turnover of 1,720% also means the economics are sensitive to slippage and spread, not just the $76 of modelled fees. Live underperformance and the near-flat recent folds argue for keeping this on a short leash and small size until it demonstrates an edge in current conditions.