Thesis
Bollinger Reversion is a textbook mean-reversion play: buy when price falls below the lower Bollinger Band, sell when it climbs above the upper band. The universe is deliberately conservative — 24 blue-chip names spanning tech (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare (JNJ, UNH, PFE, ABBV), consumer staples (PG, KO, WMT, COST), and a handful of discretionary and industrial names. The logic is sound: large-caps have deep liquidity and institutional support that tends to drag prices back toward fair value after short-term dislocations.
Backtest Performance
Over 451 days, the strategy completed 76 round-trip trades and returned 17.55% in aggregate, implying a CAGR of roughly 9.46%. The win rate of 63.89% is solid — nearly two in three trades closed in the green. Fees totalled $76 (flat $1 per trade), and there were no FX costs given the all-USD universe.
The headline numbers look reasonable. The concern is in the risk-adjusted figures: a Sharpe of 0.66 is mediocre, and a maximum drawdown of 20.57% means the strategy has, at its worst, lost more than a fifth of equity before recovering. For a rules-based large-cap strategy, that drawdown is wide relative to the return it generates.
Turnover of 1,720% annualised is also elevated — this strategy is trading actively, which amplifies the impact of any slippage or spread costs not captured in the backtest fee model.
Walk-Forward Validation: A Cautionary Story
The validation breakdown across four equal folds is where the strategy's true trajectory emerges:
| Fold | Period | Return | Sharpe | Max DD | Trades |
|---|---|---|---|---|---|
| 1 | Aug 2024 – Jan 2025 | +7.67% | 1.78 | 7.06% | 24 |
| 2 | Jan 2025 – Jul 2025 | +1.42% | 0.25 | 20.56% | 11 |
| 3 | Jul 2025 – Dec 2025 | +0.77% | 0.20 | 6.66% | 18 |
| 4 | Dec 2025 – May 2026 | +0.40% | 0.14 | 11.19% | 20 |
Fold 1 is genuinely impressive — a 1.78 Sharpe with a shallow 7% drawdown. But the decay from there is steep and monotonic. By Fold 4, the strategy is barely generating positive returns (0.40%) with a Sharpe of 0.14 and still absorbing an 11% drawdown. The out-of-sample return of 0.40% against an in-sample return of 17.55% is a classic overfitting signature.
The probabilistic and deflated Sharpe ratios tell the same story: PSR of 0.814 and DSR of 0.342. The DSR in particular — which adjusts for the number of trials and strategy length — suggests there is only about a 34% probability this Sharpe ratio reflects a genuine edge rather than chance. With 6 parameter trials, the multiple-testing penalty is doing real damage. Validation status: failed.
Recent Activity
Live paper trading in late June saw a cluster of buys — DIS at $99.18, PG at $140.03, GOOGL at $377.90, WMT at $115.75, and COST at $956.32 — followed by a PG exit at $148.40, a clean 6% gain on that leg in under ten days. Since early July, however, the strategy has gone quiet: six consecutive daily runs with zero executions and between one and two rejections each session. Total portfolio value has drifted between roughly $9,710 and $9,893, suggesting the market is not offering the extreme band breaks the strategy requires.
Strengths and Risks
Strengths: The strategy's logic is sound and transparent, the universe is liquid, and Fold 1 proves the signal has worked in the right regime. A 64% win rate is structurally encouraging.
Risks: The performance decay across folds is too consistent to dismiss as noise. Mean reversion has historically struggled in trending or momentum-driven markets — the post-2024 large-cap tape may simply not be mean-reverting on the timescales this strategy targets. High turnover amplifies real-world friction costs, and the DSR flags meaningful overfitting risk across the parameter search.
Bollinger Reversion warrants continued observation but should not be promoted to live capital allocation without either a regime filter (e.g., only trade when VIX is elevated) or a parameter re-optimisation that survives a stricter out-of-sample test.