Thesis
Bollinger Reversion is a classic mean-reversion play: buy when price closes below the lower Bollinger Band — signalling statistically oversold conditions — and sell when it recovers above the upper band. The universe is deliberately blue-chip, spanning 24 large-cap names across tech (AAPL, MSFT, NVDA, GOOGL), financials (JPM, BAC, V, MA), healthcare (JNJ, UNH, PFE, ABBV), consumer staples and discretionary, and energy. The thesis bets that institutional-quality names revert to trend rather than breaking down — a reasonable prior for large-caps, though not a guarantee.
Backtest Performance
Over 451 calendar days, the strategy executed 76 trades and returned 17.55% (9.46% CAGR) with a Sharpe of 0.66 and a win rate of 63.89%. A peak-to-trough drawdown of 20.57% is the headline risk figure — meaningful for a strategy trading household-name stocks. Turnover came in at 1,720%, which is high for a band-based system; fees (flat $1 per trade, 76 total) are modest, but slippage on real fills could bite harder than the backtest assumes.
On its face, the numbers are solid. A 63.9% win rate suggests the reversion signal fires reliably when it fires at all.
Cross-Validation: The Warning Sign
The walk-forward cross-validation across four folds is where the picture darkens:
| Fold | Period | Return | Sharpe | Max DD |
|---|---|---|---|---|
| 1 | Aug 2024 – Jan 2025 | 7.67% | 1.78 | 7.06% |
| 2 | Jan 2025 – Jul 2025 | 1.42% | 0.25 | 20.56% |
| 3 | Jul 2025 – Dec 2025 | 0.77% | 0.20 | 6.66% |
| 4 | Dec 2025 – May 2026 | 0.40% | 0.14 | 11.19% |
The trend is unambiguous: returns and Sharpe compress in every successive fold. Fold 1 was exceptional (Sharpe 1.78, shallow drawdown). Every subsequent period delivered less. The out-of-sample return sits at 0.40% with a Sharpe of 0.14 — barely distinguishable from cash. The Probabilistic Sharpe Ratio (PSR) of 0.81 is passable, but the Deflated Sharpe Ratio (DSR) of 0.34 accounts for the 6-trial multiple-testing penalty and is the more honest number. The validation gate correctly flags this as not passed.
Recent Live Activity
The live paper book reinforces the cross-validation signal. The last executed trades occurred in early June — a round-trip on PG (bought at 140.03, sold at 148.40 nine days later) and buys in GOOGL, WMT, COST, and DIS. Since then, six consecutive daily runs from June 26 through July 3 have produced zero executions, with a combined 13 rejected signals. The portfolio sat at ~$9,894 as of July 3 against a $10,000 starting balance, meaning the live book is slightly underwater.
This inactivity could reflect one of two things: the market simply hasn't produced enough extreme band touches in this universe, or the signal is too selective to generate consistent alpha in the current regime.
Strengths and Risks
Strengths: The strategy is transparent, rules-based, and operates in liquid names where fill quality is predictable. A 63.9% win rate and all four folds positive (even if modest) show the core signal is not random.
Risks: The fold-by-fold return compression is a textbook pattern for curve-fitting decay — the Bollinger parameters may have been tuned, implicitly or explicitly, to the 2024 volatility regime. The 20.57% max drawdown is also disproportionate for a strategy earning sub-10% CAGR; the risk/reward ratio is unattractive at current calibration. Extended periods of market strength suppress band breaches, leaving the strategy in cash and underperforming a passive index.
Verdict
Bollinger Reversion earns credit for a clean, legible thesis and a positive historical record. But the validation data warrants caution before deploying real capital: the live edge is fading, the DSR is below meaningful thresholds, and the recent paper book is treading water. Watch for parameter recalibration or regime filters before treating this as production-ready.