The thesis
bollinger-reversion is a textbook mean-reversion strategy: buy when price closes below the lower Bollinger band, sell when it pushes above the upper band. It trades a universe of 24 US large caps spanning tech, financials, healthcare, staples, and energy — names like AAPL, JPM, JNJ, COST, and XOM. The strategy is currently live in paper trading.
Recent activity
The bot runs on a daily schedule, and the last two weeks look like classic reversion plumbing. On 7 Oct it bought 7 shares of JPM at $328.78; the prior day it trimmed 2 shares of COST at $934.52. Earlier trades fit the pattern cleanly — buying beaten-down NKE (70 shares at $35.64) and MCD, while selling into strength in GOOGL ($351.91) and CAT ($847.03).
Execution has been light and mostly clean: recent scheduled runs executed 0–2 orders each, with a handful of rejections (one on 1 Oct, one on 30 Sep, one on 2 Oct). The paper account sits around $9,619 total equity as of 7 Oct — modestly underwater against the implied $10k start, with cash fluctuating between roughly $490 and $2,800 as positions cycle.
Backtest performance
Over 451 days the backtest returned 17.55% (final equity $11,755), a 9.46% CAGR with a 63.89% win rate across 76 trades. That hit rate is the strategy's headline strength — reversion setups tend to win often, even if each win is small. The costs are tame too: $76 in fees, no FX drag.
The warning signs are in the shape of the risk. Max drawdown reached 20.57%, and turnover is a punishing 1,720% — this strategy churns, which raises its sensitivity to slippage and fees that a backtest can understate. A Sharpe of 0.66 is mediocre for how actively it trades.
Validation: the hard part
This is where enthusiasm should cool. The walk-forward validation fails its gate. All four folds are positive, which sounds reassuring — but the per-fold returns decay monotonically: 7.67% → 1.42% → 0.77% → 0.40%, and fold Sharpe collapses from 1.78 to 0.14. The out-of-sample return is just 0.40% with an OOS Sharpe of 0.14, a world away from the full-sample 0.66.
The deflated metrics confirm it. PSR of 0.814 is borderline, but the Deflated Sharpe Ratio is 0.342 across 6 trials — well below any level that would let us trust the edge isn't a product of selection. In plain terms: the backtest edge looks increasingly like it lived in the first fold and faded as markets moved on.
Verdict
bollinger-reversion is a coherent, cheap-to-run strategy with a genuinely high win rate, and its recent trades are disciplined. But the evidence points to overfitting risk: the edge degrades out-of-sample and the validation gate says no. It earns its place as a live paper experiment to observe — not as a candidate for real capital until the OOS picture improves.