The thesis
bollinger-reversion runs one of the oldest ideas in technical trading: buy when price closes below the lower Bollinger band, sell when it pushes above the upper band. It's a pure mean-reversion play, applied to a universe of 24 US large caps spanning tech (AAPL, MSFT, NVDA), financials (JPM, V, MA), staples (PG, KO, WMT, COST) and industrials (CAT, HON). The status is live. The appeal is obvious — the rule is simple, symmetric, and easy to reason about. The risk with any reversion system is equally well known: it sells winners early and catches falling knives, and it only works when the underlying names actually revert rather than trend.
Backtest performance
Over 451 trading days the strategy returned +17.55% (final equity $11,755), a 9.46% CAGR with a 63.89% win rate across 76 trades. That headline looks healthy, but two numbers temper it. The Sharpe of 0.66 is modest — return-per-unit-risk here is unremarkable — and the max drawdown of 20.57% is deep relative to the gains, meaning an investor would have had to stomach a fifth of their capital in the red. Turnover of 1,721% also flags a busy, churn-heavy system; the $76 in fees is small in this paper context but would scale with size.
Validation: the real story
This is where enthusiasm should cool. Validation did not pass. On paper, all four walk-forward folds were positive — a green light in isolation — but the magnitudes tell a decay story:
| Fold | Return | Sharpe |
|---|---|---|
| 1 | 7.67% | 1.78 |
| 2 | 1.42% | 0.25 |
| 3 | 0.77% | 0.20 |
| 4 | 0.40% | 0.14 |
Almost all of the edge lives in the first fold. By the most recent out-of-sample window the strategy returned just 0.40% at a Sharpe of 0.14 — statistically indistinguishable from noise. The deflated Sharpe ratio (DSR 0.342) sits well below the probabilistic Sharpe (PSR 0.814), and with 6 trials behind the search, the gap points squarely at overfitting: the full-sample result flatters a strategy whose live edge is thin.
Recent activity
Live behaviour echoes the validation warning. Of the last six scheduled runs, five executed zero trades — bands simply aren't triggering — and most logged a rejected order. The only action was 22 Sep, when it bought 9 MCD and sold 21 WMT. Earlier fills (NKE, GOOGL, DIS, PG, CAT) show the reversion logic firing on staples and cyclicals alike. Meanwhile the paper account total has hovered around $9,650–$9,780, below its starting line — a sober contrast to the backtest's gains.
Verdict
The thesis is clean and the win rate is genuinely good, but the evidence points to a strategy that was strong in one regime and has faded since. Until out-of-sample returns recover past noise, bollinger-reversion is best treated as a monitored experiment, not a capital allocation.