Reading candlesticks, and counting your own patterns
The answer
A candle is a summary of one auction: where it opened, how far it travelled in each direction, and where it settled. The named patterns built from candles are claims about what that auction means, and almost none of those claims have been tested by the person telling you about them, or by you.
Why this costs you money
You learn 20 candlestick patterns. You can name them on sight. Then you trade them, and here is what happens.
You take a bullish engulfing signal. It works. You take another. It fails. You take a third. It fails. You take a fourth and it works beautifully, and you remember that one. Over 6 months you take 30 of these, roughly half work, and your account is down because the losers were larger than the winners and every trade cost you a spread.
You conclude that you need to filter the pattern better, or add an indicator, or find a more advanced pattern. So you learn 20 more.
The problem was never the pattern. The problem is that you never counted. You have a memory of the trades, and memory keeps the vivid ones. Your fourth trade, the one that worked beautifully, is doing the work of 10 in your head.
Here is the cost in one sentence. A pattern that works 50% of the time is not worthless and it is not a signal either. It is a coin, and the only way to know which one you have is to count instances on a chart, which takes 10 minutes and which almost no retail trader has ever done.
How it works
The anatomy, and what it means mechanically
Every candle has 3 parts.
The body runs from the open to the close. Its length is the net result of the session. Its colour or fill tells you the direction.
The upper wick runs from the top of the body to the high. It marks a price that was reached and not held.
The lower wick runs from the bottom of the body to the low. Same thing, downward.
Now the mechanical reading, which is the part worth having. A wick is not a mood. It is a record of a failed price.
A long upper wick means: buyers pushed price to that level, and by the end of the session nobody was willing to keep paying it.
That is a statement about supply appearing at a price. It has content. It is also far weaker than most people treat it, because you do not know when in the session it happened. A long upper wick on a daily candle could mean price spiked at 9:20 and drifted down all day, or that it rose steadily and collapsed in the last 10 minutes. Those are completely different events and they produce the same candle.
This is the central limitation of candle reading and it is rarely stated: a candle discards the order of what happened inside it.
The 5 patterns worth knowing
Ignore the dictionaries with 80 entries. These 5 cover the ideas, and the rest are variations.
Doji. Open and close are almost equal, so the body is very small. Both wicks may be long. The claim is indecision: buyers and sellers finished the session in the same place they started.
Hammer. A small body near the top of the range, a long lower wick, little or no upper wick, appearing after a decline. The claim is that sellers pushed price down and were absorbed.
Shooting star. The mirror image. Small body near the bottom of the range, long upper wick, appearing after an advance. The claim is that buyers pushed price up and failed.
Bullish engulfing. Two candles. A down candle, then an up candle whose body completely covers the previous body. The claim is a change of control.
Bearish engulfing. The same, reversed.
Marubozu is worth 1 line: a candle with a large body and no wicks at all, meaning price moved in one direction all session. In India, check for a price band before you believe it.
The definition problem, which is why nobody has good numbers
Ask 5 sources how long a hammer's lower wick must be. You will get "at least twice the body", "at least 3 times the body", "at least two thirds of the total range" and "use your judgement".
Ask how small a doji's body must be. You will get "zero", "very small", "less than 5% of the range" and, again, judgement.
This is not a small problem. A pattern you cannot define in numbers cannot be counted, and a pattern that cannot be counted cannot be shown to work or to fail. It can only be discussed.
This is exactly why the most serious academic work on chart patterns, by Andrew Lo, Harry Mamaysky and Jiang Wang in the Journal of Finance in 2000, had to begin by inventing mathematical definitions using a smoothing method. They could not test the patterns until they had pinned them down, because the practitioner literature had never pinned them down.
So the first thing you do with any pattern, before you trade it, is write your own numerical definition. Not because your definition is correct. Because a fixed definition is the only thing that lets you count.
Context, and the honest problem with it
Every book says context matters. A hammer after a long decline means something; a hammer in the middle of a sideways range means nothing.
That is almost certainly true. It also creates a problem the books do not mention.
Once you require the pattern to appear in a specific context — after at least 5 declining sessions, at a level touched twice before, on above-average volume — the number of instances in 2 years of daily data collapses from 40 to about 4. And 4 instances tells you nothing at all. You cannot measure a 55% win rate from 4 trials.
This is the real reason candlestick patterns are hard to evaluate. Without context they are too common to be informative. With context they are too rare to be measurable. Anybody who claims to have solved this should be asked for their sample size.
Confirmation costs you the move
One more mechanical point. An engulfing pattern is not complete until the second candle closes. A hammer is not confirmed until the following session opens higher.
So by the time the pattern exists, the move it is describing has already happened. You are entering after the information, not before it. That is not fatal. It is simply a cost, and it should be subtracted from whatever edge you think the pattern has.
What it tells you, and what it does not
A candle tells you where price was rejected. That is real information about where supply or demand appeared, and it is why levels marked by long wicks sometimes matter later.
It does not tell you who rejected it or how much size was behind the rejection. A long upper wick on low volume is a different event from the same wick on 5 times average volume, and the candle shape does not distinguish them.
It does not tell you the sequence inside the session.
And a named pattern does not tell you a probability. When a book says a hammer is a reversal signal, it is describing a claim, not reporting a measurement. You are about to fix that on your own chart.
The decision rule
A candlestick pattern is a reason to look, never a reason to act.
Act only when 3 things line up:
- The pattern is at a level that mattered before — a prior high, a prior
low, a level the price has turned at.
- The volume on the pattern candle is above average, so the rejection had
size behind it.
- You have already decided where you are wrong, which for a hammer is
below its low and for a shooting star is above its high.
If a pattern appears and you cannot say which level it is at, you have found a shape in the middle of nothing. Leave it.
Try this now
This is the most valuable 10 minutes in this cluster. A trader who counts their own pattern outcomes learns more than a year of reading pattern books will teach them.
- Choose 1 pattern. Use bullish engulfing or hammer for your first count.
- Write your definition in numbers, before you look at any chart. For a hammer, for example: lower wick at least 2 times the body; upper wick less than half the body; the candle comes after at least 3 sessions where the close was lower than the previous close. Write it down. Do not change it later. If you change it after seeing results, you are data mining yourself.
- Open the daily chart of 1 liquid stock or index you follow. Set the range to 2 years.
- Find every instance that meets your written definition. Every one, not the good-looking ones. Write the dates in a list. You will usually find between 5 and 20.
- For each date, write down 2 numbers: the close on the pattern day, and the close 5 sessions later. Mark W if the price moved in the direction the pattern is supposed to predict, and L if it did not.
- Count your W and L totals.
- Now find the base rate, which is the step everybody skips. Over those same 2 years, how often was the close 5 sessions later higher than the close today? You can approximate this: if the stock rose over the 2 years, the base rate for "higher in 5 days" is above 50%, often around 52% to 55%.
What you should see. Most readers, on most stocks, get somewhere between 40% and 60% wins. That range feels like it might be an edge and it is not, because you have only 5 to 20 samples and the base rate is already near 50%.
Two specific things usually surprise people.
How few instances there are. You expected the pattern to be everywhere. Once you apply a written definition, it is not.
How different your count is from your memory. If you have traded this pattern before, compare the count against what you believed your win rate was. The gap between those 2 numbers is the most useful thing you will learn this week.
Now repeat the whole count on a second stock. If the win rate is 60% on one and 40% on the other, you have measured noise, and that is a genuine finding, not a failure of the exercise.
Three real cases
1. The Nifty 50, 3 and 4 June 2024 (India) — a textbook pattern that failed immediately On 3 June 2024 the Nifty rose sharply on exit poll expectations, producing a large up candle. On 4 June, when the general election results arrived, it fell heavily, producing a large down candle whose body covered the previous day's body completely. That is a bearish engulfing on the daily chart of the country's main index, on enormous volume, at an all-time high. It is about as textbook as the pattern ever gets.
Within weeks the Nifty had recovered and gone on to new highs. A trader who acted on the pattern was wrong quickly. Keep this case, because it is the honest counter to every pattern example you will be shown, which will always be one that worked.
2. Marshall, Young and Rose, 2006 (United States) — the pattern test nobody quotes Ben Marshall, Martin Young and Lawrence Rose published a study in the Journal of Banking & Finance testing candlestick trading strategies on Dow Jones Industrial Average component stocks over roughly a decade. They tested a large set of single and multiple candlestick patterns and found no evidence that they produced value.
Related work reached similar conclusions on US data. Some studies on Asian markets have found positive results for certain patterns. The overall picture is weak and inconsistent, which is exactly what you would expect from an effect that is small or absent.
3. Munehisa Homma and the origin story — the attribution nobody checks Almost every candlestick book opens with the same story. Munehisa Homma, an 18th-century rice trader at the Dojima exchange in Osaka, is said to have made an enormous fortune using candlestick methods and to have written the rules that became the Sakata methods.
Treat this carefully. Homma was a real and successful rice trader. Whether the candlestick charting technique as it is taught today descends directly from his methods is much less certain, and historians of Japanese markets have questioned parts of the standard account. What is documented is that Steve Nison introduced the technique to Western traders in Japanese Candlestick Charting Techniques in 1991, and that Western use of candlesticks essentially dates from that book.
The lesson is not that the technique is worthless because its origin story is embellished. It is that a compelling origin story is used, constantly, in place of evidence. When a method's main support is a legend about a man 250 years ago, the method has not been tested.
The question that resolves it
A novice asks: what pattern is this?
An expert asks: what did somebody have to do, at what price, to make this shape appear?
A long lower wick exists because sellers hit bids down to a level and then buyers took everything offered. Somebody bought size. If you can name that transaction, the candle has content. If the wick is 4 ticks long on 200 shares in an illiquid stock, nobody did anything, and no amount of pattern naming will change that.
What would make this wrong
If candlestick patterns carried no information, then a properly defined pattern would show a win rate indistinguishable from the base rate, on every market and every period. Some studies find close to that. Some find small effects. The Lo, Mamaysky and Wang result found that patterns changed the conditional distribution of returns, which means the shapes are not pure noise, while explicitly declining to claim profitability.
The honest limits are 3.
First, this article's sceptical position could be too strong. Patterns tested mechanically on daily data are not what a skilled discretionary trader does. A trader using candles as one input among several, on instruments and timeframes that studies do not cover, may be doing something the tests do not measure.
Second, the sample size problem cuts both ways. If a pattern requires context and context makes it rare, then the absence of statistical evidence is partly an absence of data, not a demonstration of absence.
Third, your own count will not settle the question either. Fifteen instances on one stock is a small sample and you should treat your result as a first look, not a proof. What the count reliably does is correct your memory, and that is worth more than a statistic.
In India
Three Indian features change what a candle is before you read it.
The close is an average. The official NSE closing price is a weighted average of the last 30 minutes of trading, not the last trade. Every daily body in India is drawn to an averaged number. A late surge in the last 5 minutes appears in the candle as a smaller move than it was. This slightly compresses bodies and slightly lengthens wicks compared with a last-trade close.
Price bands clip wicks. A stock at its lower band produces a candle with a flat bottom and no lower wick, because the price was not allowed to fall further. That candle looks like a strong close off the lows on some days and like a marubozu on others, and both readings are wrong. Always check whether the day's move equals the band exactly.
Expiry sessions distort index candles. Indian index options volume concentrates enormously into the weekly and monthly expiry sessions. On those days a large amount of index movement is driven by option settlement and hedging flows near particular strike prices, rather than by an ordinary auction between buyers and sellers. Intraday candlestick patterns on Nifty expiry days are not comparable to patterns on other days, and mixing them into one count will give you a number that means nothing.
In the United States
The candle depends on which session your platform draws. Regular hours run 9:30 to 16:00 Eastern. With extended hours off, which is the default on most platforms, a company reporting results after the close produces a gap between 2 days. With extended hours on, the same event appears inside the candles as a continuous move.
Test this once. A pattern that exists in one setting can vanish in the other, on the same day, on the same stock. This is not a rare edge case in the United States. It happens 4 times a year on every company that reports.
Gaps carry different meaning. Because pre-market trading exists, most US opening gaps have real transacted prices inside them. The opening candle after a gap is therefore the continuation of a session that already happened, not the start of a new discovery process.
Limit Up-Limit Down means US candles are rarely truncated. A US flat top is much more likely to be a genuine rejection than an Indian one, which is a real advantage for pattern reading in the US and a real hazard when Indian traders read American material.
Half days. Around some holidays the US session closes at 13:00 Eastern. Those days produce small candles with low volume, and they should be excluded from any count you run, in the same way Indian expiry days should be.
Where they differ, and what that tells you
The same pattern definition, applied in India and in the United States, is applied to 2 different objects.
In India, the daily candle covers the whole tradable day, but its close is an average and its extremes can be clipped by a rule.
In the United States, the daily candle has a clean auction close and unclipped extremes, but by default it covers only 6.5 of the roughly 16 hours in which the stock actually trades.
What that tells you is that "the same pattern" is not the same measurement. An engulfing candle in India engulfs a real full-session body. An engulfing candle in the United States may engulf only the regular-hours portion of a day whose important move happened at 16:05 the previous evening.
The practical consequence is specific. When you run the counting exercise, run it on your own market, on your own data settings, and do not import a win rate from an American book or a YouTube channel. The number they measured, if they measured anything, was measured on a differently shaped object.
Carry this
- A wick records a price that was refused. A body records where it settled.
- A candle throws away the order of events inside the session.
- Write your pattern definition in numbers before you look, then count every instance.
- Always compare your win rate against the base rate. Without that, the number means nothing.