Sooner or later somebody forwards you a neat table. Bullish engulfing: 74% success. Morning star: 78%. Hanging man: 59%. It has the look of research, the numbers are oddly precise, and it settles an argument in a WhatsApp group in about four seconds.
Ask three questions of any such table and it usually collapses. Success measured how? Over what holding period? With the stop placed where? None of those is a hostile question — they are the questions that decide whether the number means anything at all.
Why a published success rate means very little
- "Success" is undefined. Up the next day? Up 2% at any point in ten sessions? Those produce wildly different numbers from identical data.
- There is no exit. A pattern with no stop and no target cannot have an outcome. Change the stop by half a percent and the "success rate" moves several points.
- The universe is unstated. Largecaps or smallcaps, index or single stocks, which years — a rule measured across a decade-long bull market is measuring the bull market.
- The definition is fuzzy. How long must the lower wick be for a hammer to count — twice the body, or three times? Two people applying the same pattern name to the same chart routinely disagree.
- Costs are ignored. Brokerage, STT, exchange charges, stamp duty, GST and the spread all come out of the winners and are added to the losers.
Counting your own base rate
The good news is that you can settle this for yourself, offline, with a chart and a notebook, and the exercise teaches more than the answer does. What you are building is a base rate — how often this thing happens across all comparable cases, rather than in the cases you happen to remember.
- 1Write the definition before you look at anything
Mechanically, with numbers. "Lower wick at least twice the body; upper wick no more than a quarter of the body; occurring after at least three consecutive lower closes." If you cannot write it down, you cannot test it — and you will unconsciously widen it whenever a near-miss would have worked.
- 2Fix the universe and the period
One index — say the constituents of a broad NSE index — on the daily timeframe, over one full calendar year that includes both rising and falling stretches. Writing this down stops you from quietly wandering into charts that suit your case.
- 3Fix the outcome rule
Something with no judgement in it: "the trade is a win if price closes above the pattern candle’s high within five sessions before trading below its low; otherwise it is a loss." Now every occurrence has an unambiguous answer.
- 4Find every occurrence — not the good ones
Scroll each chart from January to December and mark every instance that meets your written definition, including the embarrassing ones. This step is where almost everybody cheats, and it is the only step that matters.
- 5Tally, then look at the losers
Count wins and losses, and then go back through the failures looking for something they share — a market falling that week, no volume, a level overhead, a smallcap that barely trades. That is where your actual filter comes from.
The number that matters more than the hit rate
Hit rate is the number everyone quotes and the least useful one on its own, because it says nothing about the size of the wins relative to the losses. A pattern that is right 40% of the time while winning three rupees for every one it risks makes money comfortably. A pattern that is right 65% of the time while winning half a rupee for every one it risks loses money steadily.
- Win rate
- Share of occurrences that met your written win condition
- Average win
- Mean gain on the winners, after costs
- Average loss
- Mean loss on the losers, after costs — usually close to your stop distance
Example: A rule that wins 40% of the time, gaining 3 units when right and losing 1 unit when wrong, has an expectancy of (0.40 × 3) − (0.60 × 1) = +0.60 units per trade. A rule that wins 65% of the time, gaining 0.5 and losing 1, has (0.65 × 0.5) − (0.35 × 1) = −0.025. The second one feels far better and is quietly losing money.
The two biases that make this genuinely hard
| Bias | How it shows up here | The defence |
|---|---|---|
| Hindsight bias | You scroll back through a chart and every hammer that preceded a rally leaps out at you. The four that preceded nothing are invisible, because nothing is not memorable. | Mark occurrences going forward through the chart, left to right, before you can see what happened next. Cover the right side of the screen if you have to. |
| Confirmation bias | You already believe the pattern works, so a candle with a wick 1.7× the body counts today and would not have counted yesterday. | The written, numeric definition. It is the only thing standing between you and a definition that quietly adjusts to fit the outcome. |
So what is the honest answer?
Broadly, this: candlestick patterns are a good language and a poor signal generator. As a language, they are excellent — "long lower wick on heavy volume at a level that held twice before" is a compact and accurate description of something that genuinely happened, and describing it well is most of reading a chart.
As a standalone trigger, taken anywhere on any chart, they add little that survives costs. Their two durable contributions are that they mark an invalidation point — the candle’s extreme, beyond which your reading was simply wrong — and that they add weight to a level you had already identified for other reasons. That is a real, modest, usable edge, and it is smaller than any forwarded table will tell you.
Candlestick patterns, assessed honestly
3 questions. Answers are revealed once you submit all of them.
1.Pattern A is right 45% of the time, gaining 2.5 units when right and losing 1 when wrong. Pattern B is right 70% of the time, gaining 0.4 units and losing 1. Which is better?
2.Why is scrolling back through a chart to find examples of a pattern a poor way to test it?
3.What is the most durable practical contribution of a candlestick pattern?
Kisi ne bheja: "bullish engulfing 74% kaam karta hai." Poochho — kaam karne ka matlab kya, kitne din mein, stop kahan tha, kharcha kaata ya nahi? Jawaab kisi ke paas nahi hota. Ek copy lo, ek saal ka chart lo, aur har engulfing ginn lo — jo yaad hain sirf woh nahi, saare. Chalis ke baad aapko apna jawaab mil jaayega, aur woh kisi forward kiye hue number se zyada sachcha hoga.
- A pattern is not a strategy — only a pattern plus an entry, stop, exit, universe and size can have a success rate.
- Published reliability tables leave "success", the exit, the universe and the costs undefined.
- Test one pattern yourself: written numeric definition, fixed universe and period, fixed outcome rule, every occurrence counted.
- Record reward to risk alongside win or loss — hit rate alone cannot tell you whether a rule makes money.
- Candles are a good language and a weak trigger; their durable value is an objective invalidation point at a level you already marked.
Mark it done to track your progress through the curriculum.
Common questions
Short, direct answers to what people ask about this topic.
- do candlestick patterns actually work
- There is no single answer, because a pattern on its own cannot have a success rate — a shape with no entry, no stop, no exit rule, no universe and no position size has no outcome to measure. The forwarded tables quoting oddly precise figures collapse the moment you ask how success was defined, over what holding period and with the stop placed where. The answerable version of the question is whether one fully specified system, tested on your own market and your own costs, does better than the alternative.
- a pattern combined with an entry a stop an exit rule and a position size is called a
- A strategy — and only a strategy can be measured. The pattern is a description of a shape; the entry, the stop, the exit rule, the universe and the size are what turn it into something with an outcome. Once you have specified all five you are no longer answering “does the hammer work”, you are answering whether one particular system worked.
- why are published candlestick success rates misleading
- Because five things are usually left unstated: how success was defined, where the stop and the exit sat, which stocks and which years were covered, how strictly the pattern was defined, and whether costs were deducted. Brokerage, STT, exchange charges, stamp duty, GST and the spread come out of every winner and are added to every loser, and moving a stop by half a percent can shift a quoted figure by several points. A number with none of that attached is decoration, not evidence.
- how do I test whether a candlestick pattern works for myself
- Write the definition down in numbers before you look at a single chart, fix the universe and the period, fix an outcome rule with no judgement in it, then find every occurrence — including the ones that embarrass you. Counting only the instances you happen to remember produces a flattering number, and that is the step almost everybody skips. Then go back through the failures looking for something they share; that is where a real filter comes from.
- how many occurrences do you need to test a chart pattern
- Far more than most people count — a few dozen cannot separate a genuine edge from luck, and a serious statistical test needs many hundreds. What a run of forty occurrences will settle in an afternoon is whether you can identify the pattern consistently and whether your version of it is anywhere near as common or as clean as the textbook implied. For most people learning this, that is the more valuable finding.