Bollinger Bands on EUR/USD 4H: Nine Configurations, Zero Survivors

1. Hypothesis & Parameter Grid

Bollinger Bands consist of a middle band (a simple moving average of closing price) and an upper and lower band set a multiple of standard deviations away from it. The mean-reversion hypothesis: when price closes below the lower band, it has moved statistically far enough from its recent average that it is likely to revert back toward the middle band.

Signal type: Mean reversion, long only. Entry: close price crosses below the lower band. Exit: close crosses above the middle band, or a fixed 5-bar timeout, whichever triggers first.

Grid structure: full factorial, period × std_dev = 9 configurations (hypotheses HYP_0112 through HYP_0120). Direction: long only. Transaction cost: 2.0 pips round-trip.

2. Methodology

Identical methodology to Part 3 (RSI study) and every prior study in this series. This allows direct, apples-to-apples comparison between indicator families.

Data Specification

Signal Logic

Walk-Forward Windows

5-Stage Kill Filter

Same sequential filter as every study in this series. Critically, Stage 3 is evaluated per-window, not on aggregate average drift — a configuration passing 5 of 6 windows but failing one is still killed.

3. Results

Kill Filter Breakdown

All 9 Configurations — Aggregate

All 9 killed at Stage 3 (IS/OOS drift). No hypothesis reached Stage 4.

Performance Summary (All 9 Variants)

For Trade Count, higher is not automatically “better” here — the lowest-trade configuration (90 trades) is flagged in Section 4 as statistically unreliable, not just infrequent. For Mean |Drift|, lower is better; “Best” and “Worst” are reversed accordingly from the other rows.

4. Best Performer: Period=25, std_dev=2.5 (HYP_0120)

The longest period paired with the widest bands in the grid — the most selective configuration tested, and also the highest-ranked by aggregate Sharpe.

Per-Window Breakdown

Windows passing drift check: 2 out of 6 (33.3%) — Windows 3 and 4.

5. Worst Performer: Period=15, std_dev=2.5 (HYP_0114)

The shortest period paired with the widest bands — the only configuration in the entire grid with a negative aggregate Sharpe, and the only window in the sweep to drift by more than 38 percentage points.

Per-Window Breakdown

Windows passing: 2 out of 6. Trade counts of 7–21 per window make several of these Sharpe estimates unreliable regardless of the drift result.

Best vs Worst: Side-by-Side

6. Trade Frequency vs Drift

std_dev directly controls how often the signal fires: narrower bands are touched more often, wider bands more rarely. Conventional wisdom says higher-frequency signals should drift less — more trades means a more stable sample. We checked this directly against the 9-variant set.

Effect of std_dev on Trade Frequency and Drift

Window-Level Pass Rate is the share of the 18 individual windows (3 configurations × 6 windows) in each std_dev group that stayed under the 5pp drift threshold. All 9 configurations are still 100% killed at the hypothesis level, since the kill rule triggers on any single failing window — this table shows where the failing windows concentrate.

Effect of Period on Aggregate Sharpe

Longer periods consistently outperform shorter periods on aggregate Sharpe, regardless of std_dev. This is consistent with slower, smoother bands filtering out noise-driven entries — but as Sections 4 and 7 show, the improvement comes with a much smaller sample and does not translate into a stable per-window result.

7. Regime Dependence

Window 5 (OOS: 2022–2023) — The Outlier

Window 5 is the dominant performance driver across the entire sweep. Every one of the 9 configurations posted its best or near-best result here, and every one of them also posted some of its worst drift here.

The 2022–2023 EUR/USD period followed sharp directional moves tied to post-COVID monetary policy normalization and ECB rate hiking. Mean-reversion signals fired at volatility extremes and profited from the subsequent corrections. This is a regime-specific effect, not a robust edge.

Windows 3 and 4 (OOS: 2018–2019, 2020–2021) — Consistent Failure

2018–2019 (EUR/USD trending lower then choppy) and 2020–2021 (COVID volatility followed by EUR strength) were systematically hostile to mean-reversion entries. Price violated the lower band, continued lower, and did not return to the middle band within 5 bars — the signal entered at the start of extended moves rather than at exhaustion points. 7 of 9 configurations posted a negative Sharpe in at least one of these two windows.

Kill Filter Verdict

9 of 9 hypotheses killed at Stage 3. No hypothesis reached Stage 4 or Stage 5. The dominant kill pattern: win rate drift exceeding 5 percentage points in at least one window was universal across all 9 configurations. The signal consistently learns roughly 10 to 20 percentage points of in-sample noise that does not transfer out-of-sample. Maximum drift observed across the sweep: -38.1pp (period=15, std_dev=2.5, Window 3, 2019). This matches the zero-survivor pattern found across the full 183-hypothesis Phase 2 sweep.

8. Bollinger Bands vs MACD vs SMA

All three studies used identical testing conditions — same pair, same timeframe, same windows, same costs, same kill filter — so results are directly comparable across the series.

9. Why This Matters

For Retail Traders

If you are trading Bollinger Bands mean reversion on MT5 default settings or a course-taught configuration, here is what this data actually supports.

For Strategy Developers

Bollinger Bands mean reversion is not worthless here, it is unfinished. Paths worth testing, none of them validated in this study:

10. Frequently Asked Questions

11. Next in the Series

This is Part 4 of the MT5 Indicator Backtest Series. Every study uses the same methodology, the same dataset, and the same 5-stage kill filter, making results directly comparable across indicator families.