Backtest evidence

Does pairing RSI with a Bollinger band beat RSI on its own?

Does Combining RSI and Bollinger Bands Help? 720 Backtests

Finding

No. Pairing discards part of the RSI pool, so it is measured against discarding the same number of bars at random. Of 108 pairings (three timeframes, two years, two sides, three RSI levels, three deviations), 21 land outside the 95% range of random draws and 19 of those land below it. Comparing the 18 hourly pairings with the RSI level alone, the win rate rose in five, all on the long side, by 0.02 to 3.67 points; all nine short pairings fell, by up to 10.39 points. The band condition keeps bars at the same RSI reading whose five-bar run-up is 1.80 to 2.82 times larger in ATR and whose own range is 1.43 to 2.03 times larger.

Key results

Pairings outside the random range
21 of 108. 19 below, 2 above
Outside in both years the same way
4 of 54, all hourly short and all below
Pairings that beat the RSI level's win rate
5 of 18 hourly, all long, by 0.02 to 3.67 points
Short side win-rate change
All nine fell, by up to 10.39 points
What the band keeps
Run-up 1.80 to 2.82x larger, bar 1.43 to 2.03x larger, mean RSI within 2.1 points
Profitable in both years
7 of 15 hourly long, 2 on the 4-hour short. Nothing in the other four cells
Cases where pairing rescued a losing level
2 (hourly RSI 35 with 2σ and 2.5σ), both inside the random range

Scope

  • Instrument: USD/JPY
  • Period: 2024-01-01 to 2024-12-31 and 2025-01-01 to 2025-12-31
  • Timeframes: 15-minute, 1-hour, 4-hour
  • Conditions: 3 RSI alone, 3 band alone, 9 paired, per side

Method

RSI period 14 and Bollinger period 20. Longs use RSI below 25, 30 and 35 with a lower-band touch; shorts use RSI above 75, 70 and 65 with an upper-band touch. The control draws the pairing's own number of bars from the RSI pool a thousand times and takes the 2.5th and 97.5th percentiles of the resulting win rates, from a fixed seed. Trades run through Formiq's backtester, entering at the close of the signal bar and exiting ten bars later, at a fixed 0.3 pip spread, 0.1 lots and no slippage; the exit is identical for all three so that only the entries are compared. 15 conditions x 2 sides x 3 timeframes x 4 windows x with and without cost = 720 runs. Every figure is checked in research/combo-verify.run.ts.

Limitations

  • One pair over two years.
  • Fills are at the close with a fixed spread and no slippage, so live results will differ.
  • The exit is fixed at ten bars, so a rule using stops and targets will read differently.
  • The random range is built for the win rate; the distribution of net pips is a separate question.