The Thesis
Bollinger-reversion executes one of the oldest ideas in technical trading: buy oversold, sell overbought — defined here by price crossing outside the Bollinger Bands. When a name from its 24-stock large-cap universe (AAPL, MSFT, NVDA, JPM, DIS, and 19 peers across tech, financials, healthcare, consumer staples, and energy) closes below the lower band, the strategy enters long. It exits above the upper band. No leverage, no shorting — a clean, rules-based long-only mean-reversion loop.
Backtest Highlights
Over 451 days and 76 paper trades, the strategy returned 17.55% (9.46% annualised), with a win rate of 63.89% — meaning nearly two in three trades closed profitable. Those are headline numbers worth paying attention to.
The caveat is in the risk metrics. A 20.57% maximum drawdown is significant for a strategy trading conservative household names; it implies a period where the portfolio shed more than a fifth of its value before recovering. The Sharpe ratio of 0.66 is acceptable but not compelling — it sits squarely in "worth investigating further" territory rather than "deploy with confidence."
Turnover ran high at 1,720% annualised, which in a paper environment costs a flat $1 per trade but would translate to meaningful friction in live execution.
Walk-Forward Validation: The Warning Sign
The strategy's validation run tells a more cautious story. Across four sequential out-of-sample folds, performance degrades sharply and consistently:
| Fold | Period | Return | Sharpe | Max Drawdown |
|---|---|---|---|---|
| 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% |
Fold 1 looks excellent. Folds 2–4 look like a strategy that has largely exhausted its edge. The aggregate out-of-sample return is 0.40% with an OOS Sharpe of 0.14 — essentially flat on a risk-adjusted basis.
The probabilistic Sharpe ratio (PSR) of 0.814 and deflated Sharpe ratio (DSR) of 0.342 reflect this: after adjusting for multiple trials and non-normal return distributions, the Headmars validation framework correctly failed this strategy. A DSR below 0.5 generally indicates the in-sample Sharpe is more likely luck than a persistent edge.
Recent Live Activity
In paper-trading mode, the strategy has been quiet. The past week logged mostly zero executions per daily run, with 1–2 rejected signals each session — signals that met the band-touch trigger but were blocked (likely by cash or position limits). The most recent executed trades were a DIS buy at $99.18 on June 10 and a PG sell at $148.40 on June 9, closing a position opened June 1 at $140.03 for a roughly 6% gain on that leg. GOOGL, WMT, and COST entries from late May remain open.
Portfolio total has drifted from $10,101 on June 10 to $10,281 on June 16 — a quiet week, consistent with a market where large-cap names haven't stretched far enough from their bands to generate signals.
Verdict
Strengths: Positive across all four folds, high win rate, clear and auditable logic, no exotic instruments.
Risks: Performance decay is steep and consistent — not a noisy wobble but a directional fade. The strategy may have been fitted to a 2024 volatility regime that hasn't repeated. The high drawdown in Fold 2 (20.56%) also demonstrates meaningful tail risk in trending or volatile markets where mean-reversion assumptions break down.
The validation gate was right to hold. Bollinger-reversion is a productive research baseline — but it needs either parameter refinement, a regime filter, or a longer OOS track record before it earns a capital allocation.