Dragonfly Doji: Meaning, How to Trade It and What the Data Shows

A dragonfly doji is a candlestick pattern in which the open and the close are nearly identical and sit at the high of the session, with a long lower shadow and no upper shadow, or a very short one. It looks like a "T": sellers pushed the price down during the session, then buyers brought it all the way back by the close. In technical analysis, the dragonfly doji is read as a potential bullish reversal pattern after a downtrend.

This guide explains what the pattern means, how to identify it and how to trade it, and adds what a definition alone cannot: what happened after it on 58,908 daily occurrences across stocks, crypto, forex, ETFs, indices and commodities, compared with random days on the same assets.

What is a dragonfly doji?

The dragonfly doji is one of several doji formations: the session ends where it started, so the body is a thin line. What makes it a dragonfly is where that line sits, at the top of the range. The lower wick is the trace of a sell-off that did not hold.

Steve Nison, who introduced Japanese candlestick charting to Western traders in Japanese Candlestick Charting Techniques (1991), describes it as a doji whose open and close are at the high of the session. Our test bench applies the same idea as a rule a computer can follow without judgment: a doji, a very short upper shadow and a lower shadow of at least half the average range of the previous 10 candles, with sizes measured against recent sessions as in the TA-Lib library, whose own function flags more cases (our rule finds 46% of them). No prior trend is required. That rule is looser than the textbook picture: it accepts a lower shadow of moderate length, where the textbook asks for a long one, and a small gap between the close and the high.

The chart below is not drawn by hand: it is the median shape of the occurrences our bench detected on daily charts across all markets, with the long side of the two-sided trade we test.

What does a dragonfly doji mean?

The session opens, sellers drive the price well below its starting level, then buying pressure absorbs the selling and lifts it back to where it started. Two opposite moves and no momentum left at the end of the session: indecision that turned in favor of buyers.

At the bottom of a bearish trend, the tail shows that buyers rejected lower prices: the classic bullish signal that a trend reversal may be near, ideally at a support zone. At the top of an uptrend, the same formation shows that sellers could push the market sharply down during the session, and it can be read as a warning, like the hanging man.

Our data does not show a clear difference between the two. After a decline, the breakout trade described below reached its 2:1 target 35.7% of the time (+0.07 R per trade); after a rise, 35.3% (+0.05 R). The pattern marks indecision, not a direction, which is why we trade the first side that breaks.

How to identify a dragonfly doji

1. The opening and closing prices are equal or almost equal.
2. They sit at or very near the top of the range.
3. A long lower shadow hangs well below them.
4. No upper shadow, or a very short one.

On the median daily occurrence, the whole range spans 1.94% of the price, or 0.74 times the average true range (ATR): narrower than a typical recent session. The pattern is common: we found one every 144 eligible daily bars.

Dragonfly doji vs. other doji patterns and the hammer

The gravestone doji is the mirror image, with its tail above the body. On daily data, the gravestone doji trade reached its 2:1 target 32.0% of the time against 32.2% at random, for −0.05 R against −0.04 R per trade, a gap that does not pass our statistical test; the dragonfly did a little better, 35.5% against 35.0%, or +0.06 R against +0.04 R, a small gap that does not pass our test either.

The hammer pattern has the same long wick but a visible body, and it requires a downtrend before it. Its setup reached the target 35.4% of the time against 36.1% at random, for an average result of +0.06 R against +0.09 R per trade. The hammer page rates that trade No clear edge: the gap in average result is too small to separate from chance. The takuri sits between the two.

Among the other doji types, the standard doji has its body anywhere in the range, and the long-legged doji has long wicks on both sides; its trade reached the target 33.2% of the time, almost exactly as often as at random.

The Real examples section shows dragonfly doji candles from our bench with their actual outcome. Trade results are in R, the initial risk of the trade.

How to trade the dragonfly doji

Because our test assigns this candle no expected direction, the trade we test is symmetrical: a buy stop just above the high and a sell stop just below the low, valid for 20 bars. Whichever side breaks first opens the trade; the stop goes at the other end of the range, and the target at twice the risk. On daily charts the first break was upward in 68.0% of cases, which proves little: the close sits at the top of the range, so the buy stop is the nearer one, and the gravestone doji, which closes at the bottom, broke upward in only 36.8%.

The setup can also be filtered by a support zone, the RSI or a moving average. We measured two filters: above the 200-day moving average, the trade reached its target 35.0% of the time (+0.04 R), below it 36.3% (+0.08 R); on volume above 1.5 times the 20-day mean, 36.0% (+0.08 R) against 35.4% (+0.05 R). These breakdowns are descriptive and not corrected for the number of tests. We have not tested RSI or support filters.

Is the dragonfly doji reliable?

Not on its own: in our tests, its trade did not beat chance on any market or timeframe. Daily data with all markets together reads no clear edge: the trade did a little better than on random dates, but not by a margin that passes our test. US stocks daily, on data that misses most delisted companies, a survivorship bias we have not sized, reads no clear edge. Forex weekly, the largest raw gap (+0.47 R per trade against +0.10 R on random dates), also reads no clear edge, on only 53 trades. Crypto 4-hour reads no clear edge: the trade still loses money, about as much as on random dates. Commodities daily reads No clear edge (−0.05 R against +0.10 R), a gap within the margin of error. Commodity results come from continuous futures that are not adjusted at contract rolls; the bench cuts each series at the large price jumps those rolls leave. Indices weekly comes closest to a positive result, rated promising (+0.37 R against +0.10 R on 246 trades), a gap that passes a simple test but not our correction. On hourly charts, crypto reads Worse than chance, forex Weak, below chance and all markets together Worse than chance.

Read the table in pairs of columns. On daily data, the trade reached its 2:1 target 35.5% of the time; the same entry, stop and target distances placed on random days of the same assets reached it 35.0% of the time. That gap does not pass our statistical test, once corrected for the thousands of tests we run: +0.06 R per trade against +0.04 R at random, before fees and slippage, is too close to call. On hourly crypto, the trade did worse than chance after correction (−0.11 R); on hourly forex it was also below chance (−0.06 R), but that gap passes a simple test, not our correction. The full table is in the statistics section of this page, and the method in our methodology.

Real cases

The gallery in the Real examples section shows occurrences picked by our bench (best and worst results, textbook shapes, most recent cases and typical draws). The two cases below come from daily crypto charts and are told in more detail. They are recorded sessions, not simulations.

On October 19, 2023, DOT/USDT opened at 3.653, fell to 3.562 and closed at 3.649, with a high at 3.66. The next day it traded above 3.66, triggering the buy stop, with a stop at 3.562 and a target at 3.856, hit on October 21. DOT was 18.3% higher 10 days after the pattern.

On May 3, 2023, XRP/USDT opened at 0.4644, fell to 0.4497 and closed at 0.4632, with a high at 0.4655. On May 5 it broke the high first, triggering the buy stop, with a stop at 0.4497 and a target at 0.4971; on May 7 it fell to 0.4457 and hit the stop, a full 1 R loss. Ten days after the pattern, XRP was 8.5% lower: the break of the high was a false start.

Mistakes, advantages and limitations

• Trading it alone: the daily gap is too small to count and turns negative on hourly data. Trend and support zones are the classical additions; our splits by trend barely changed the result, and we have not tested support zones. Risk management is not optional.
• Entering before the session ends: a tail in the middle of a session can still end near the bottom.
• Advantage: an easy "T" to spot, with a clear stop at the other end of the range.
• Limitation: no direction tested, no proven edge even before costs, and fast losses.

FAQ

Summary

A dragonfly doji opens and closes at or near the high of the session, with a long lower shadow and little or no upper shadow. It is read as a bullish reversal after a downtrend. On 58,832 daily trades, the breakout trade reached its 2:1 target 35.5% of the time, against 35.0% at random: a small gap that does not pass our correction, even before costs. Treat it as a location for a trade, not a reason for one: let the market break one side, set the stop at the other end of the range and keep the risk per trade low. This page is educational and is not investment advice; trading involves a risk of loss, and past statistics are no promise of future results.

About the author and the data. Written by Amaury, founder of PatternsWizard, who is passionate about finance, quantitative trading and data analysis. He built the PatternsWizard test bench, which detects each setup with published rules on full market history: crypto, forex, US stocks, ETFs, indices and commodities. Trades and win rates are compared with random bars on the same assets, excess returns with the rest of the market, with error margins and a correction for multiple testing, and the whole pipeline is checked on random-walk prices, where nothing can be predicted. Statistics on this page were generated on November 5, 2026. The full methodology explains every step.