Candlestick charts

Osaka rice, three centuries before your screen

The setting is a warehouse. In Osaka, feudal lords stored the rice of their domains and issued coupons against it, each coupon a claim on future delivery. Those coupons changed hands at the Dojima Rice Exchange, running from the end of the seventeenth century and formally recognised by the shogunate in 1730. It is often described as the first organised futures market in the world. The detail that matters here sits elsewhere: what traded was a contract rather than a sack, so a price had to be followed through time, and written down.

This is where Munehisa Homma appears, a rice merchant from Sakata in the province of Dewa, born in 1724, died in 1803. He is credited with a body of rules known as the Sakata rules, five methods built on groups of three: three mountains, three rivers, three gaps, three soldiers, three methods. In fairness, the attribution is only partly his. The texts carrying his name were copied and reworked for two centuries, and candlestick charting as you know it settled later than that. What is documented is the exchange, the corpus, and the idea holding them together.

That idea is the one part of the whole inheritance that has not aged a day. A price does not only reflect the supply of rice and the demand for it: it reflects the state of a contest between the people trading it. Homma kept records of harvests and weather, but he also recorded how the crowd of merchants behaved. Three centuries later the sentence still stands: a candle encodes a balance of force over a span of time, not a value.

The West came to it late. Steve Nison introduced candlestick charting to Western markets with Japanese Candlestick Charting Techniques, published in 1991. In three decades it became the default display of almost every charting package. That success carries a price still being paid: the catalogue of names was imported along with the tool, and the catalogue is by far the weakest part of the inheritance.

What a line chart throws away

A line chart joins one price per period, almost always the close. A candle shows four: open, high, low, close. The ratio is one to four, and the difference is not decoration: three pieces of information per period are quietly deleted by the line.

What the line loses is the path. A session that drops hard and then takes back everything it lost ends where it started. The line draws a flat point there, and nothing signals that the market went looking for a much lower price and handed it back within hours. The candle keeps the trace of that descent as a wick, which is to say as a price reached and then refused.

One example makes it concrete. Five sessions in a row, each one reaching for the same low price and coming back. On a line: five roughly aligned points, nothing to report, an uneventful week. As candles: five wicks stopping at the same place, in other words five refusals at the same price. Those are not the same information, and the second one is what you decide on.

⚠️ A limit worth knowing before the first pattern name: a candle does not tell you the order in which things happened. It does not say whether the high came before the low. Two sessions drawn identically can tell two opposite stories, and no software can settle it from the candle alone. You have to drop a timeframe to find out, which is already an admission that the candle is a summary.

Four prices, two pieces of information

The exact anatomy, once and for all. The open is the first price traded in the period, the close the last one. The body is the rectangle joining them. The high and the low are the extremes reached between those two moments, and the wicks, also called shadows, are the thin lines running from the body out to those extremes.

Colour is not a fifth data point. It is only the sign of the difference between close and open: green when the close sits above, red when it sits below. Changing your palette changes nothing about what the candle holds, and a beginner who thinks he is reading a signal in the colour is simply reading the same information twice.

Here are the two pieces of information a candle really carries. Wicks say where price tried to go: those are prices that were reached and then given back, so places where somebody refused to follow. The body says who won the period: the direction and the distance between the first and the last price traded. Everything else in the vocabulary, including the most exotic names, follows from those two readings.

⚠️ A candle is never read in absolute terms. “Long wick” means long compared with the previous twenty candles on the same chart, never long in points. The same distance in points is enormous on a quiet market and strictly invisible during an economic release. Every candle reading is relative to its immediate neighbourhood, otherwise it is not a reading, it is a measurement taken with an unmarked ruler.

Three readings that replace the catalogue

Three shapes cover most of what the named patterns cover. Refusal: a small body at one end, a long wick at the other. Price reached for a level and did not keep it. Conviction: a wide body, almost no wick. The period was held end to end in the same direction. Indecision: a tiny body between two wicks. The two sides cancelled each other out.

Now compare what that classification replaces. The open-source TA-Lib library, whose source is public and therefore countable without argument, exposes sixty-one candlestick recognition functions. Open the list and file every entry into one of the three boxes: hammer, hanging man, shooting star and inverted hammer are refusals; the big bodies and their variants are convictions; the doji and all its variants are indecision. Almost nothing is left outside the three boxes.

The classification is not a lazy shortcut, it fixes a real defect in the catalogue. Two patterns carrying two different names can differ by a few percent of body-to-wick ratio, and the threshold separating them moves from one source to the next. Take the definition of the same hammer from three authors, count the occurrences on the same data, and you will get three different counts. Three readings describing a balance of force are sturdier than sixty definitions resting on an arbitrary threshold.

The practical gain is immediate: there is nothing left to memorise. You look at a candle and you name it in three words at most. Refusal from below. Buying conviction. Indecision. If the sentence does not come on its own, the candle is saying nothing clear, which is also an answer, and one with the advantage of costing nothing.

The place matters more than the shape

A refusal candle shows up hundreds of times a year on any instrument. If its mere presence carried an edge, nothing would be left of that edge by now: the shape is visible to everyone, for free, and detectable by any program in three lines of code. What is not equally available is the context it lands in.

The same indecision in open space says nothing at all, because nobody knows what it is undecided about. The same one, at a zone marked before it arrived, says something precise: at this place, falling stopped being easy. Context carries the signal, the pattern only timestamps it.

The order of the chain never changes, and the pattern is its last link: regime first, then the zone, then the timeframe, and only then the trigger. Learning pattern names without the first three links is learning to tell the time without owning a watch. It is also why a beginner who knows eighty names does not trade better than a beginner who knows none: he learned the end of the chain.

The honest test. Hide the pattern name. Ask yourself whether you would take the trade knowing only the context, meaning the regime, the zone, and exactly where price sits. If the answer is yes, the pattern is a comfort that helps you press the button, and that has genuine psychological value. If the answer is no, no pattern will rescue a missing context, and what you are about to take is an entry on a name.

Wait for the close

A live candle is not a candle, it is a provisional state. Its body keeps moving until the last second, and its extremes can only widen, never tighten. The perfect hammer you are watching forty minutes before an hourly candle ends can close as a full bearish body, and the pattern you entered on will never have existed.

What that costs can be measured, and measured on your own records rather than on belief. Add a two-value column to your journal: entry taken before or after the close of the signal candle. After fifty trades, split the two populations and look. One precaution decides how much the record is worth: the column gets filled at the moment you click, never in the evening. Filled in afterwards, it records what you remember, which is precisely the thing the record set out to correct.

The trade-off is real, and pretending otherwise would be dishonest. Waiting for the close costs you the distance between the point where you could have entered and the closing price. That distance is measurable too: record it every time, average it, and you have the exact price of your patience. Compare that price with the cost of entries taken on patterns that never formed. Both numbers come out of the same journal, and that is the only trade-off worth weighing, because you weighed it with your own data.

⛔ One trap empties the question of meaning if you ignore it: the close of a daily candle depends on your broker's server time. Two brokers whose servers sit two hours apart display two different daily candles on the same market, with hammers appearing on different days. The pattern is not in the market, it is in the cut. Check the cut-off time of your chart before building anything at all on daily candles.

A candle only exists at the height you look from

A daily candle is made of two hundred and eighty-eight five-minute candles on a market open around the clock. Its long lower wick, the one you call a refusal, is an entire fall followed by an entire recovery, with its own refusals, its own convictions and its own indecision inside it.

Merging keeps the extremes and destroys everything else. The high and the low of the day are those of the two hundred and eighty-eight candles taken together, so they survive intact, to the point. The bodies cancel one another out: the daily body holds only the gap between the first price and the last, keeping nothing of how many round trips it took to get from one to the other. So the daily hammer you see is not an event, it is a summary, and dropping one step down never confirms it: it dissolves it into several hours of travel, a good half of which runs the other way.

The other consequence is how arbitrary the cut is. Shift the candle boundary by half an hour and some patterns vanish, replaced by others that were not there. Nothing changed in the market, only the grid moved. A signal that does not survive a shift of the grid is an artefact of the grid, not a property of the market.

The cure is the one from the technical analysis lesson, and it does not need rewriting here: one decision timeframe, chosen and kept. What belongs to candles specifically is how you check that you are keeping it. Write the timeframe next to every pattern you record in your journal, at the same time as the pattern. The day an entry carries a timeframe lower than your decision one, you did not find a pattern, you changed grid until you manufactured one.

The two and three candle patterns worth the trouble

Three of them deserve to have their names remembered, and all three for the same reason: each describes a mechanism, not a silhouette. A mechanism can be verified, a silhouette can only be argued about.

The engulfing first, because it says nothing its two candles do not already say once merged. Merge them: the open is the open of the first, the close is the close of the second, the extremes are the extremes of both. Out comes one candle with a small body and a long wick, in other words a refusal.

⚠️ You will read everywhere that an engulfing is the candle of the timeframe above. That holds under a condition nobody states: both candles have to fall in the same slot of the upper grid. An engulfing straddling two even hours matches no two-hour candle your chart will ever draw. Merge three engulfings by hand on your hourly chart, then open the two-hour: you will find roughly one in two of them, and the ones you cannot find read exactly the same way as the ones you can.

The inside bar next, the candle whose extremes fit entirely inside those of the candle before it. It says one precise thing, verifiable at a glance: the period produced no new extreme. That is the cleanest definition of a pause anyone can write, and it serves to spot a contraction without predicting the direction it resolves in.

The run of full bodies in the same direction last, what the Sakata rules called three soldiers. It describes displacement, a series of periods in which one side held without pausing for breath. It is worth adding that by the time you can name it, most of the displacement is often behind you, which makes it a good descriptor and a poor trigger.

Why the others are not worth the trouble. They are separated by thresholds no two sources set the same way, they carry names suggesting an outcome nobody has measured, and above all they count badly. A pattern for which three authors give three definitions cannot enter a statistical record, so you will never know whether it works for you, on your instrument, on your timeframe. A reading you cannot count is not a reading, it is decoration.

Practice: re-read your entries, before and after the close

This exercise takes an hour and requires no new knowledge. It answers a question almost nobody asks, one that bears on what you actually did rather than on what you believe you do: how many of my entries rested on a candle that never existed?

Pull your last fifty entries with their timestamps and the timeframe you decided on. For each one, answer yes or no to a single question: at the moment I clicked, was the signal candle closed? You now have two piles, and you have already learned something just from looking at their relative size.

Now take the pile of entries made before the close and reopen the chart at the right date. For each one, look at the candle once finished and ask whether the pattern you saw was still there. Count the ones that had vanished. That number is your first figure. Set against the size of the pile, it gives you the share of your entries built on an imaginary pattern, and it is a share nobody knows until they have counted it.

Second figure, the price of patience. For those same entries, measure the distance between your actual entry price and the close of the signal candle. Average it. That is exactly what waiting would have cost you, expressed in your instrument's own unit. You now hold both sides of the trade-off, in numbers, and they belong to you.

Third pass, the one that teaches the most. Go back to the entries taken after the close and, for each, write one line of context: the regime at the time, and whether or not price was in contact with a zone marked in advance. Cross that line with the outcome of the trade. If your best entries are the ones where the context was written before the pattern appeared, you have just verified the central sentence of this lesson on your own data, and you no longer need to believe me.

⚠️ Fifty trades is not many. Read the direction of the gap, not the decimal, and run the exercise again next quarter. A gap that turns up on two independent records says something about the way you trade. The same gap seen once may be nothing more than the trace of ten trades taken in the same week, on the same instrument, in the same frame of mind.