Research · Candlesticks
The short answer
The honest answer changed on 2026-09-06: the measured run was computed on a superseded engine and has been withdrawn rather than published with a caveat. What each page still carries is the rule, its thresholds and the conditions that invalidate it.
Nobody on this site can currently tell you, and the reason is worth more than a number would be.
Every one of the 107 candlestick formations is published here as an executable rule with its thresholds and the conditions that invalidate it. The measured outcomes — hit rates, benchmarks, intervals — are withheld, because the run that produced them was computed on a superseded engine and included the still-forming bar. It was withdrawn rather than published with a caveat.
For a candlestick formation, "does it work" decomposes into three separate claims, and most published answers conflate them:
Testing 107 rules at the conventional 5% threshold produces roughly five apparently significant results from noise alone. Any study of candlestick patterns that does not correct for this — Benjamini-Hochberg or equivalent — is reporting its false positives as findings.
This is the single most common defect in the published literature on the subject, and it is why a table of "the best performing patterns" from an uncorrected study is close to worthless: it is a list of whichever rules got lucky.
Worth knowing
When a corrected result does arrive here, expect it to be less exciting than the uncorrected tables elsewhere. That gap is the correction doing its job.
The genuinely useful part of a formation page is not the hit rate. It is the invalidation — the conditions under which the detection is wrong. A hammer is geometrically identical to a hanging man; only the prior trend separates them, which is why the prior-slope threshold is a parameter rather than a hidden assumption.
Knowing that two formations you thought were distinct fire on the same bar changes how you read a chart, whether or not either has a measurable edge. Start with the hammer and follow its "easily mistaken for" links.
Candlestick charting is usually traced to Japanese rice trading in the 18th and 19th centuries, and reached Western markets largely through Steve Nison's work in the 1990s. The formations carry names and interpretations from that lineage.
What changed is not the shapes but what can be done with them. A formation described in prose — "a small body after a decline, showing sellers failing" — cannot be counted, so for most of the tradition's life there was no way to ask how often it preceded what it claimed to.
Writing the same formation as a condition on open, high, low and close makes it countable. That is the entire difference between a description and a testable claim, and it is why the thresholds on each page matter more than the name.
Most arguments about whether candlesticks work conflate two questions that have different answers.
| Question | Status here |
|---|---|
| Can the shape be identified consistently? | Yes — 125 executable rules, two implementations agreeing on 81,301 detections |
| Does the shape precede what tradition says? | Withheld — the measured run was computed on a superseded engine |
A trader who says "candlesticks work for me" is usually making a claim about the second, supported by evidence about neither. A study reporting that they do not work is usually answering the second under one specific set of thresholds for the first — which is why the thresholds have to be published alongside the result.
Treat a formation as a description of a market state rather than as a signal. A hammer says price was driven well below where it closed, after a decline. That is a fact about the bar, and it is true whether or not the formation carries an edge.
What it does not say is what happens next. Until that is measured with a benchmark and a correction, any use of it as an entry trigger is a bet on the tradition rather than on evidence — which is a defensible thing to do, provided it is named as such.