Backtest evidence

Which Bollinger deviation should be used, and do wider bands perform better?

Bollinger Bands: 1, 2 or 3 Standard Deviations? Backtest Results

Finding

The breakout favoured 1σ, and wider was not stronger. On hourly USDJPY every period from 10 to 50 was profitable in both 2024 and 2025 at 0.5σ through 1.25σ; at 1σ the seven-period mean was +1,568.8 pips in 2024 and +1,081.9 in 2025. At 1.75σ the 2025 mean reached +1,850.2 pips but fell to +590.2 in 2024. Beyond 3σ the samples are too small to read: the hourly breakout averaged 21.7 trades a year at 3σ, 6.3 at 3.5σ and 2.3 at 4σ in 2025. The fade had no stable deviation — of 105 hourly settings only 23 fade-long and one fade-short were profitable in both years. Containment on 6,226 hourly bars at period 20 was 45.0% inside 1σ, 87.1% inside 2σ and 99.1% inside 3σ, against the normal-distribution figures of 68%, 95% and 99.7%.

Key results

Deviations profitable in both years, hourly breakout
every period at 0.5σ to 1.25σ; the 1σ seven-period mean was +1,568.8 / +1,081.9 pips
Hourly breakout trades a year in 2025
21.7 at 3σ, 6.3 at 3.5σ, 2.3 at 4σ
Closes inside the bands, period 20 hourly
45.0% inside 1σ, 87.1% inside 2σ, 99.1% inside 3σ

Scope

  • USDJPY. 1 Jan to 31 Dec 2025, with 2024 and the two halves of 2025 as comparisons
  • 15-minute, hourly and four-hour bars; 24,903 M15 bars in 2025, 6,226 H1 and 1,610 H4
  • 7 periods (10 to 50) x 15 deviations (0.5σ to 4σ in 0.25 steps) x 3 trading rules = 315 settings, on 3 timeframes and 4 windows: 3,780 runs
  • 0.3 pip spread, 0.1 lots, zero slippage

Method

Measured with Formiq's production backtester. The fade enters at the close of a bar whose low or high touched the band and exits at the close of a bar touching the opposite 1σ band; the exit is fixed at 1σ so that only the entry deviation varies. The breakout enters on a close beyond the band and reverses on a close beyond the opposite one.

Limitations

  • One pair, USDJPY, and two independent calendar years.
  • Selecting the maximum from a grid of periods and deviations carries multiple-comparison uplift. The confirming year is not untouched future data.
  • Beyond 3σ the trade counts are small and win rates and profit factors are wide. An average of 2.3 trades a year at 4σ cannot be read with the confidence of 1σ.
  • Touches use highs and lows; breakouts use closes. Counting a wick piercing as a breakout changes the result.
  • Fills use bar closes and the spread is fixed throughout.
  • The fade grid fixes the exit at the opposite 1σ band so that only the entry deviation varies; a test that also optimises the exit answers a different question.