Candlestick Patterns Tested: The Reversal Signal Continued the Move 58% of the Time

Candlestick patterns are the first thing most traders learn and the last thing anyone measures. The pictures are memorable, the names are vivid, and the claim underneath them is specific enough to test: this shape marks a turn. So I coded five of them as rules — bullish and bearish engulfing, hammer, shooting star, morning and evening star, and doji — and ran them across 16,926 EURUSD 5-minute bars from June 3 to August 26, 2026. Same dataset as my MACD crossover backtest and my RSI divergence test, so the numbers are directly comparable.

The entry is the close of the pattern bar. No discretion, no “in the right context”, no drawn levels. If the shape means anything on its own, it has 16,926 chances to show it.

Do reversal candles actually reverse?

This is the only question that matters, and it has a clean answer. For every signal I measured where price was heading in the twelve bars before the pattern and where it was two hours after. If the pattern reverses anything, the two should point opposite ways.

All three lean the same way, and none of them leans toward reversal. On a five-minute EURUSD chart these shapes are, if anything, mild continuation signals. That is not a subtle statistical wobble — it is a 16-point gap on the engulfing bar across five hundred trades.

There is a second, more embarrassing number underneath it. A reversal pattern is supposed to appear at the end of a move. I checked how often each pattern actually did:

Six bullish engulfings out of ten formed while price was already rising. There was no downtrend for them to end. The pin bar is the only one of the three that shows up at a turning point more often than not, and it barely clears a coin flip. Most of what traders call a reversal pattern is a continuation bar with a reversal name.

What did each pattern pay?

Entry at the pattern close, exit 24 bars later — two hours. Non-overlapping, so no signal is counted twice. The last column charges the 1-pip EURUSD spread I measured in my spread and swap costs test.

One positive number before costs, none after. The star is the only pattern in the family with a pulse, and I will come back to it. Note the split inside the engulfing row: the bearish version beat the bullish version by eight percentage points, 48.3% against 40.3%. Both fired an almost identical number of times, so this is not a sample-size story — over this three-month window the short side of the pattern simply worked better than the long side.

Why is the pin bar the weakest of them?

The pin bar has the best reputation and the worst numbers, and the reason shows up when you stop looking at where price finished and look at where it went in between.

The pin bar gave up half again as much ground as it gained. Almost a third of pin-bar trades went ten pips or more the wrong way before the two hours were up — which is exactly the distance a sensible stop sits at, per my stop loss distance test. In practice most of those trades are closed losses, not open ones.

It also gets worse the harder you filter. Demanding a longer wick — the standard advice for a “quality” pin bar — cut the win rate steadily: 43.2% at a wick twice the body, 32.0% at 2.5x, 26.3% at 3x. Nineteen signals left at 3x, five wins. Selectivity did not concentrate an edge here; it concentrated noise.

What about the doji?

The doji is the one that breaks, and it breaks in a way that says something about five-minute charts generally.

Loosening the definition from 5% to 20% of the range added seven bars out of sixteen thousand. That is not a threshold you can tune — the population is entirely made of bars where open and close are the same tick. The median five-minute EURUSD bar is 1.33 pips tall, so “the market could not decide” is the normal condition of this chart, not an event. A pattern that describes a fifth of all bars is a description of the timeframe, not a signal.

Does holding longer help?

For two of the three patterns, no. For the star, it is the whole story.

A clean rise to a peak at four hours and a fall after it. The engulfing bar and the pin bar show nothing like this — engulfing hovers between -0.16 and -0.89 pips at every horizon from 15 minutes to 8 hours, and the pin bar’s worst result is its four-hour one, 33.8% wins and -2.73 pips.

I would not build an account on the star. It is one positive cell in a large table, the eight-hour column already gives most of it back, and the same pattern of a single flattering horizon showed up in rules I have since discarded. But it is the only candlestick result on this dataset that clears the spread at all, and it is worth knowing that it is a four-hour idea being traded on a five-minute chart.

How do these compare to an actual indicator?

The engulfing bar and a 12/26 EMA crossover finished on the same win rate to the decimal place. That is a coincidence of rounding, but the ranking around it is not: a shape you read off the chart performed within noise of the indicators built to replace it, and all four lost. The fuller comparison, including where each one’s signals actually differ, is in the companion piece on candlestick patterns against MACD and moving averages.

Do stops and targets rescue any of it?

I re-ran all three with a 1x ATR stop and a 2x ATR target instead of a timed exit, the structure my fixed against ATR stop test found least bad.

A 2:1 target needs a 33.3% hit rate to break even before costs. All three landed just under it. Not a collapse — a near miss that the spread then turns into a loss, which is the same shape of result my rule against random entry test kept finding. Every trade was resolved; nothing timed out.

What I would actually do with this

• Stop calling them reversal patterns. On this data all three continued the prior move more often than they reversed it, and 60% of engulfing bars formed with no move to reverse in the first place.
• Delete the doji from your checklist on M5. One bar in five qualifies, and every one is an artefact of a 1.33-pip median bar.
• Do not chase a longer wick. Tightening the pin-bar filter from 2x to 3x took the win rate from 43.2% to 26.3%.
• If you keep one, keep the star, and hold it four hours. 52.4% and +0.92 net across 227 trades — the only variant here that survived the spread.
• Price the shape, do not admire it. The engulfing bar landed on the same 44.8% as an EMA crossover. A pattern is just another rule, and it deserves the same arithmetic.

Frequently asked questions

Do candlestick patterns actually work in forex?

Not as standalone entries on this data. Across 16,926 EURUSD 5-minute bars, engulfing bars lost 0.64 pips a trade over 505 trades, pin bars lost 1.60 over 87, and only the morning/evening star was positive before costs at +0.51. After a 1-pip spread, a single variant out of everything I ran finished ahead.

Is the engulfing pattern a reliable reversal signal?

No. In 58.2% of cases price continued the direction it had been travelling before the pattern rather than reversing, and 60.0% of engulfing bars formed while price was already moving the way the pattern pointed — so there was no prior move for them to reverse.

Which candlestick pattern performed best?

The morning and evening star, held four hours: 52.4% wins across 227 trades at +1.92 pips, or +0.92 after spread. It was the only result on the whole dataset that cleared trading costs, and it gave most of it back at an eight-hour hold.

Why is the doji useless on a 5-minute chart?

Because 20.0% of EURUSD 5-minute bars qualify as one, and all of them have a body of exactly zero — 36.1% of bars close on the same tick they opened, since the median bar is only 1.33 pips tall. A pattern that describes a fifth of all bars carries no information.

Does waiting for confirmation improve candlestick signals?

Barely, and not enough. Requiring the next bar to close in the pattern’s direction cut engulfing trades from 505 to 294 and left the win rate essentially unchanged at 44.6%. The star improved from +0.51 to +0.65 pips, still short of the spread.

Are candlestick patterns better than indicators?

They are indistinguishable from them here. The engulfing bar and an EMA 12/26 crossover both won 44.8% of the time on the same bars, and MACD’s signal cross won 41.9%. Details are in my MACD against moving average crossover comparison and my moving average crossover test.

What data was this run on?

16,926 EURUSD 5-minute bars, June 3 to August 26, 2026, entries at the pattern close with no discretionary filtering — the same set behind my stochastic overbought test and my ADX above 25 test.