Backtest evidence

Which Fibonacci retracement level actually holds, and is 61.8% special?

Does Fibonacci 61.8% Really Hold? 2 Years of USD/JPY Tested

Finding

No ratio had an effect of its own. Measuring strength as the share of pullbacks that ended at the level, across 6,266 USDJPY swings 45% came first at 4.29%, then 50% at 4.27%, 61.8% at 4.22%, 78.6% at 3.95%, 38.2% at 3.94% and 23.6% at 2.73% — and the leader is not a Fibonacci ratio. The 61.8% level turned 4.22% while 62.0% turned 4.30% and 61.0% turned 4.22%; against the depths within three points either side the five Fibonacci ratios averaged −0.037 points and came out ahead in 31 of 60 cells. What did matter was depth: 23.6% was reached on 95.9% of swings, 38.2% on 87.2%, 50% on 78.8%, 61.8% on 71.4% and 78.6% on 62.3%. In trading, the 256 settings that bought the level won 61.17% of trades for a median of −629.1 pips a year, while the 217 that traded the break won 43.23% for +206.1 pips.

Key results

Share of pullbacks that ended at the level
45% at 4.29%, 50% at 4.27%, 61.8% at 4.22%, 23.6% at 2.73%
Fibonacci ratios against the depths three points either side
−0.037 points on average, ahead in 31 of 60 cells
Share of swings that reached each level
95.9% at 23.6%, 87.2% at 38.2%, 78.8% at 50%, 71.4% at 61.8%, 62.3% at 78.6%

Scope

  • USDJPY, 2025-01-01 to 2025-12-31, with 2024 run identically for comparison
  • M15, H1 and H4; 6,266 swings (M15 4,752 / H1 1,176 / H4 338)
  • Depth grid 10.0% to 90.0% in 0.2% steps = 401 levels. Trading: 9 levels (5 Fibonacci, 4 controls) x 5 swing definitions x 2 triggers = 90 settings per timeframe per window
  • 0.3 pip spread, 0.1 lots, zero slippage

Method

Measured with Formiq's backtester. Swings are detected automatically — a bar beating the five bars on each side — and repeated at 3, 8, 13 and 21 bars as a sensitivity check. The turn rate is the share of swings reaching a depth whose pullback ended within 5 points of it, also computed at 2.5 and 10 points. Depths that are not Fibonacci ratios are mixed into the same grid as the control.

Limitations

  • One pair, USDJPY, and two years. Nothing here guarantees the same behaviour on another pair or another period.
  • Swings are detected, not drawn. A person choosing anchors by hand may pick different waves, and this test cannot measure that choice. Five detection widths were run and the ordering did not change.
  • The 5-point window that counts as 'ended here' is arbitrary. It was also computed at 2.5 and 10 points, and the relationship between levels did not change.
  • The turn rate asks whether the pullback finished at the level. A pause that resumed lower is not counted as a turn.
  • The 401 depths overlap, so neighbouring figures are not independent. The maximum cannot be read as the result of 401 independent trials.
  • Entries and exits use bar closes, and the spread is held at 0.3 pips throughout.