Backtest evidence

Do candlestick patterns behave differently from the bars that formed no pattern?

Candlestick Pattern Win Rates: 17 Patterns Tested

Finding

Barely. Of the 6,216 hourly bars in 2025, 50.35% closed higher ten bars later and 49.63% closed lower. Twenty-two of the 26 pattern directions finished within five points of that control. Across 78 cells (26 directions on three timeframes), the excess was positive in both 2024 and 2025 in 19, and the sign agreed across the two years in 43, against the 39 chance would give. Among the 57 cells with at least 100 signals a year, the correlation between the 2024 excess and the 2025 excess is -0.15. All 723 three-white-soldier signals were also three bullish candles in a row, and the morning star finished below its control in both years.

Key results

Control, hourly 2025, all 6,216 bars
50.35% higher ten bars later, 49.63% lower
Within five points of the control
22 of the 26 pattern directions
Above the control in both 2024 and 2025
19 of 78 cells. None cleared all six cells of three timeframes by two years
Three white soldiers against three bullish candles in a row
100% of the 723 soldier signals were runs of three; 95.38% of the 758 runs were soldiers
Restricting to the textbook run-up before the pattern
Raised the hit rate in 58 of 128 cells
Profitable in both years on a ten-bar hold
2 patterns on the 15-minute chart, 6 hourly, 7 on the 4-hour. None on all three

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: 17 candlestick patterns, 26 pattern directions

Method

All 17 conditions were counted bar by bar with the same code path the backtester runs. The close 1, 5, 10 and 20 bars after each signal is compared with every bar of the same window and timeframe. Trades enter at the close of the signal bar and exit at the close ten bars later, with no stop, target or opposite signal, at a fixed 0.3 pip spread, 0.1 lots and no slippage. Eight of the 17 conditions only ever produce one side and therefore have no opposite signal to close against, which is why the exit is fixed at a bar count. Every figure in the article is checked against the run output in research/candlestick-verify.run.ts.

Limitations

  • One pair over two years. This does not show that candlestick patterns fail in every market.
  • Fills are at the close with a fixed spread and no slippage, so live results will differ.
  • The exit is fixed at the close ten bars later. Stops, targets and higher-timeframe filters were not tested as part of the entry.
  • The dragonfly and gravestone doji occur around 20 times a year on the hourly chart and 1 to 6 times on the 4-hour, so their figures move a long way on one trade.