Backtest evidence

Does a high win rate on a Williams %R fade translate into profit?

Williams %R Mean Reversion: 240 Settings Tested

Finding

The win rate was high and the result was negative. Across the 120 hourly long settings in 2025 the median win rate was 59.0% (range 48.57% to 67.05%), and 64.1% on four-hour bars, yet only 10 of the 120 hourly settings finished positive. The textbook period 14, buy below −80 and close above −20, lost 1,082.8 pips in 2025 and 569.3 in 2024. Losses were larger than wins: in 2024 that hourly setting won 68.42% of trades, but its average win was 44.86 pips against an average loss of −110.75, a factor of 2.5. Only the four-hour long side kept many settings profitable in both years: 89 of 120 were profitable in 2025, 71 in 2024, and 40 in both.

Key results

The 120 hourly long settings in 2025
median win rate 59.0%; 10 finished positive
Textbook period 14, −80 / −20, hourly
−1,082.8 pips in 2025, −569.3 in 2024
Average win and loss, hourly textbook setting in 2024
68.42% win rate; +44.86 average win, −110.75 average loss

Scope

  • USDJPY, 2025-01-01 to 2025-12-31, with 2024 run identically for comparison
  • 15-minute, hourly and four-hour bars; 6,226 hourly bars in 2025, 24,903 on 15-minute, 1,610 on four-hour
  • 6 periods (7 to 40) x 5 entry levels x 4 exit levels x 2 directions = 240 settings, per timeframe and per year
  • 0.3 pip spread, 0.1 lots, zero slippage

Method

Measured with Formiq's backtester. The long side buys when %R drops below an entry level (−95 to −70) and closes when it rises above an exit level (−50 to −10); the short side mirrors it toward zero. No stop or target in the main sweep.

Limitations

  • One pair, two years.
  • Williams %R varies little between tools, though conventions differ on whether a level comparison is inclusive. These figures come from Formiq's implementation.
  • Entries and exits both fill at bar closes.
  • The spread is fixed at 0.3 pips and slippage is zero.
  • Four-hour settings trade 9 to 92 times a year; win rates at that count are wide numbers.