Do fair value gaps get filled, and does the filling mean anything?
Do Fair Value Gaps Really Fill? 2 Years of USD/JPY Tested
Finding
They get filled, and it means nothing. On USDJPY, 96.2% to 99.4% were revisited across six cells of three timeframes by two years. But a bullish gap's ceiling is the third bar's low, so 'the gap was filled' and 'price came back to that bar's low' are the same sentence. Counting the same thing on bars that opened no gap gives 97.2% to 99.4%, and in all six cells the gap bars were revisited slightly less often. The median wait was two bars in every cell, with 41.9% to 46.1% filled on the very next bar and 62.3% to 64.1% within three. The only setting that mattered in trading was the minimum gap size: on the 15-minute chart, reading the gap as a level to return to, a floor of 0.25x ATR and 0.5x ATR were profitable in 5 of 5 cells, no floor in 3 of 5, and 1x ATR or more in 0 of 5. The middle candle's body made no difference — splitting it into ATR quartiles, price moved the gap's way over the following ten bars on 47.0% to 51.8% of 15-minute cases, 41.9% to 53.7% hourly and 44.3% to 52.8% on four-hour.
Key results
- Share revisited by price
- gaps 96.2% to 99.4%, bars with no gap 97.2% to 99.4%
- Bars until filled
- median 2; 41.9% to 46.1% on the next bar, 62.3% to 64.1% within three
- Profitable cells by minimum size, 15-minute return reading
- 0.25x and 0.5x ATR 5 of 5, no floor 3 of 5, 1x or more 0 of 5
Scope
- USDJPY. 2025 as the main window, 2024 as the check, plus the two halves of 2025
- 15-minute, hourly and four-hour bars
- 6 minimum gap sizes (0, 0.25, 0.5, 1, 1.5, 2 x ATR) x 5 validity windows (5, 10, 20, 50, 100 bars, for the return reading only)
- 0.3 pip spread, 0.1 lots
Method
Measured with Formiq's backtester across two readings — the gap as a level to return to, and entering on its formation — exiting on the opposite signal. The control counts how often price returns to the high or low of bars that opened no gap, by the same method. A return is recorded when price touches the edge of the band.
Limitations
- One pair, USDJPY. Another pair may behave differently; the width of the band in particular follows how far price moves.
- Two years only, 2024 and 2025.
- Exits are on the opposite signal. Placing a stop on the far side of the band, or holding to the last swing, was not measured.
- A return is recorded when price touches the edge of the band. Judging it on the close would give different numbers.
- Using a higher timeframe's gap on a lower chart was not measured; each timeframe is self-contained here.