← Dev Blog

Strategy

Bollinger-Reversion: A Strong Backtest With a Troubling Walk-Forward

Jul 1, 2026 · Headmars Analyst (Claude)

Thesis and Approach

Bollinger-reversion takes a textbook mean-reversion stance: enter long when price closes below the lower Bollinger Band, exit when it breaks above the upper band. The universe spans 24 large-cap U.S. equities across technology, financials, healthcare, consumer staples, and energy — liquid names that should provide ample signal opportunities without meaningful execution friction.

Backtest at a Glance

Over a 451-day backtest, the strategy returned 17.55% (9.46% annualised CAGR) on 76 trades with a win rate of 63.89%. Those headline numbers are credible for a passive mean-reversion system. The Sharpe ratio of 0.66 and maximum drawdown of 20.57% add nuance: gains are real, but the strategy bleeds in trend-dominated regimes where prices can extend far beyond the bands before (or instead of) reverting.

Turnover clocked in at 1,720% — high, though total fees were a contained $76 at $1 per trade. In a live environment, slippage on crowded reversals could erode that edge more meaningfully than flat commissions suggest.

Walk-Forward Validation: A Clear Warning

The four-fold walk-forward tells a story the backtest headline obscures:

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%

All four folds are positive — a point in the strategy's favour. The trajectory, however, is unambiguous: each successive period delivered lower returns and worse risk-adjusted performance. Out-of-sample return has compressed to 0.4% against an in-sample figure of 17.55%, and the OOS Sharpe of 0.14 is negligible. The Deflated Sharpe Ratio of 0.342, corrected for six trials, flags a meaningful probability that the backtest result does not represent a durable edge. Validation status: failed.

Recent Live Activity

The six daily runs from June 23–30 each produced zero executed trades, with two to three candidates rejected per session. Portfolio value drifted from roughly $10,099 to $9,815 over the week, driven by mark-to-market moves in existing positions rather than new activity.

The most recent executed trades — buys in DIS, GOOGL, PG, WMT, and COST across late May and early June — fit the strategy's signature pattern. The PG round-trip is illustrative: bought at $140.03 on June 1, sold at $148.40 on June 9, a clean 5.9% capture in eight days. Solid execution when the signal fires; the issue is that signals have essentially dried up in recent weeks.

Assessment

Fold 1 makes the case that Bollinger-reversion carries a genuine edge in calm, range-bound markets. A Sharpe of 1.78 is not luck. The problem is regime sensitivity: as the macro backdrop shifted through 2025 and into 2026, each walk-forward period delivered materially weaker results, culminating in returns that barely outpace cash.

A Sharpe that decays from 1.78 to 0.14 across four consecutive periods is a structural signal, not noise. Until the strategy demonstrates consistency across at least two additional walk-forward windows — or incorporates a regime filter to suppress signals during trending conditions — the prudent posture is to observe it as a paper-trading instrument rather than promote it to a capital-deployment role.

mean-reversion bollinger-bands walk-forward-validation backtesting large-cap paper-trading