Candlestick patterns are the most widely taught idea in retail trading. Every course covers them, every charting platform can detect them, and most traders can name a dozen from memory. A hammer means reversal. Engulfing means the other side has taken control. Doji means indecision.
The academic literature has tested these claims repeatedly across multiple markets and decades of data. The results are considerably less flattering than the courses suggest, and understanding why is more useful than memorizing another twenty formations.
What the patterns claim
A candlestick encodes four numbers — open, high, low, close — into a shape. Patterns are rules about one candle or a short sequence of them, and each carries a directional prediction.
A hammer is a candle with a small body and a long lower shadow appearing after a decline, said to indicate that sellers pushed price down and were overwhelmed. A bullish engulfing pattern is a down candle followed by an up candle whose body covers it entirely. A gravestone doji has open and close near the low with a long upper shadow.
The underlying claim in every case is the same: this shape carries information about what happens next, beyond what is already in the price.
What the research found
This is a testable claim, and it has been tested extensively. The broad finding across the literature is that most candlestick reversal patterns do not produce statistically significant returns.
A 2017 study in SAGE Open examining candlestick charting in the Stock Exchange of Thailand found that most reversal patterns failed to generate statistically significant mean returns, and that filtering signals through common technical indicators such as %D, RSI, or MFI generally did not improve profitability. Even the patterns that did show statistically significant returns carried high variance, meaning the edge was small relative to the risk of following it.
Research on the Swedish OMXS30 constituents over 2007 to 2015 found no short-term predictive power. Work on the Japanese market — the origin of candlestick charting — similarly concluded the patterns showed neither predictive power nor profitability.
The evidence is not uniformly negative. Some studies have found certain bullish reversal patterns profitable in the Taiwanese market, and a systematic review of the literature notes that results vary by market and period. But the reasonable summary of the body of work is that candlestick patterns lack reliable predictive power across most markets and timeframes, and that where an effect appears it tends to be market-specific, small, and unstable.
The honest reading is not that candlesticks are fraudulent. It is that the shape of a candle is far weaker evidence than the way it is taught implies, and that anyone relying on it as their primary edge is relying on something the data does not support.
A note on these sources: several date from the 2010s. That is a limitation worth stating, though it cuts less than it might elsewhere, because the claim being tested is timeless and the more recent literature has not overturned the general finding.
Why the results are so weak
The definitions are not fixed
There is no authoritative specification of a hammer. How small must the body be relative to the total range? How long must the lower shadow be — twice the body, three times? Must the upper shadow be absent, or merely short? Different textbooks, platforms, and scanners answer differently.
This means two traders can look at the same chart and disagree about whether a pattern exists, and two studies can test "the hammer" and be testing different things. It also means a trader reviewing their own history will find patterns retroactively, because the definition is elastic enough to accommodate what they are looking for.
The pattern is a small part of the information
A hammer at a level that has held three times, on volume four times normal, in a stock with a catalyst, during the first hour, is a different proposition from an identical-looking hammer in the middle of a range at midday on no volume.
The pattern is the same. Practically everything that matters is different. Studies testing patterns in isolation are testing the weakest possible version of the idea, which is also exactly the version most commonly taught to beginners.
Multiple comparisons
There are well over a hundred named candlestick patterns. If you test a hundred patterns at conventional significance thresholds, several will appear significant by chance alone. This is a structural feature of testing many hypotheses, not a flaw in any particular study, and it means isolated positive findings should be treated cautiously — including positive findings in your own data.
Costs eat small edges
Suppose a pattern genuinely shifts the probability of an up-move from 50% to 53%. That is a real edge. It is also smaller than the round-trip cost of trading it in many instruments once spread and commission are included. An edge that exists in the data and disappears after execution costs is not tradeable, and academic returns are frequently reported before those costs.
Where candlesticks retain genuine value
Dismissing them entirely would be as unsupported as the courses that oversell them. Three uses survive scrutiny.
As a compact summary of what happened. A candle communicates open, high, low, and close at a glance, and a long lower shadow does genuinely tell you that price traded materially lower and did not stay there. That is a fact about what occurred, not a prediction, and it is useful.
As a risk-definition tool. This is arguably the strongest practical use. The extreme of a candle gives an objective, non-arbitrary place to put a stop. A trader who enters after a hammer and stops below its low has a defined, mechanical risk level derived from actual traded prices. The pattern may not predict anything, and the stop placement is still sound. This connects directly to calculating risk-to-reward, which requires a defensible stop location to mean anything.
As a trigger inside a thesis you already have. This is the reframe that matters. If you have a reason to be interested in a stock — a level, a catalyst, a trend — then using a candle to time the entry is reasonable. The candle is not the edge; it is the trigger for an edge that exists for other reasons.
The failure mode is inverting that: seeing a pattern first and constructing a reason afterward.
How to test them on your own trades
Published research tells you about average behavior across a market. It does not tell you whether your execution of a pattern, in the instruments you trade, at the times you trade them, is profitable. That question is answerable, and your own trade log is the only place it can be answered.
The method is straightforward. Write down your definition first, in specific numeric terms, before reviewing any trades — body no more than one third of total range, lower shadow at least twice the body, and so on. Committing to the definition in advance is what stops the elastic-definition problem from contaminating your own review.
Tag the pattern at entry, not afterward. Tagging retrospectively is how you find patterns in trades that worked and overlook them in trades that did not.
Record the context alongside it — location relative to a level, relative volume, session, trend direction. This is what lets you distinguish "hammers do not work" from "hammers do not work for me when I take them mid-range on average volume," which are entirely different conclusions with entirely different responses.
Compare against your baseline, not against zero. The question is not whether your pattern trades are profitable. It is whether they are more profitable than your other trades. A pattern that wins 48% when the rest of your trading wins 55% is costing you.
Be careful about how much you conclude from a handful of trades — the multiple-comparisons problem applies to your own data too, and enthusiasm for a pattern you like is a strong source of bias. The general caution about reading too much into small samples applies here as directly as it does to reading your win rate.
A worked example
A trader believes bullish engulfing patterns work for them. Over four months they tag every one they take, with context, and end up with 34 trades.
The aggregate: 44% win rate, average win $180, average loss $150, net roughly break-even before costs. Their overall trading across the same period ran a 52% win rate with similar average win and loss. On the face of it, the pattern is underperforming their baseline.
Segmenting by context changes the picture. Of the 34 trades, 19 were taken at a prior support level and 15 were taken mid-range. The 19 at support won 58% with an average win of $210. The 15 mid-range won 27%.
The conclusion is not "engulfing patterns work" or "engulfing patterns do not work." It is that this trader has a location filter they were not applying, and the actionable rule is to stop taking the setup mid-range. Note also that 15 and 19 are small samples and this is a hypothesis to keep testing, not a settled fact — which is precisely why the tagging continues rather than stopping once an answer appears.
Common mistakes
Treating the pattern as the reason for the trade. The most common and most expensive error. The pattern should confirm a thesis, not supply one.
Tagging patterns after the outcome is known. Guarantees a flattering and useless dataset.
Using a platform's automatic pattern detection without checking its definition. Scanners apply specific numeric rules that may not match what you think you are trading.
Ignoring context in the log. Without location, volume, and session recorded, a negative result tells you nothing about what to change.
Abandoning after six losses. Six trades is noise. Traders routinely discard workable approaches and retain unworkable ones on samples this size.
FAQ
Do candlestick patterns work?
The weight of peer-reviewed evidence says they have no reliable predictive power on their own across most markets and timeframes, with occasional market-specific exceptions. They remain useful as risk-definition and entry-timing tools within a thesis that exists for other reasons.
Why do so many traders say patterns work for them?
Several reasons at once. Patterns are usually combined with context that does the real work, definitions are elastic enough to fit outcomes retrospectively, and memory favors the instances that worked. None of this requires anyone to be dishonest.
Are chart patterns like head and shoulders different from candlestick patterns?
They are a different class — multi-session structures rather than one or two candles — and they have their own mixed literature. The methodological problems are similar: definitions vary, context dominates, and retrospective identification is easy.
If patterns are unreliable, what should I use instead?
The question assumes the pattern was doing the work. Most consistently profitable discretionary approaches rest on a specific situation — a catalyst, a level, a liquidity condition — with the pattern used only to time entry and define risk. The situation is the edge.
How many trades do I need before I can judge a pattern in my own data?
More than most traders use, and the honest answer depends on how large the effect is. Small samples will produce apparent effects in both directions that will not persist.
Conclusion
The research does not support using candlestick patterns as a standalone signal, and the way they are typically taught — memorize the shapes, act on them — is close to the version the evidence most clearly rejects.
What survives is narrower and more useful. Candles summarize what happened, they provide objective stop levels, and they can time an entry into a trade you already had a reason to take. Whether your own use of them beats your baseline is an empirical question about your data, and the only way to answer it is to define the pattern in advance, tag it at entry with its context, and compare honestly.
TheSpeculatorsJournal supports custom strategy tags and pre-trade plan notes, so patterns can be tagged at entry with their context and reviewed as their own segment against the rest of your trading. You can try it free for 7 days — Basic is $19/month and Pro is $29/month after that.
This article is for educational purposes only and is not financial advice. TheSpeculatorsJournal does not recommend specific securities, entries, or exits, and past performance of any setup does not guarantee future results. Always do your own research and consider consulting a licensed financial professional before making trading decisions.