Home › Learn › Lesson 27

Three-candle patterns — morning star, evening star and inside bars

By TradingPrimer · Published 3 Sep 2026

Three-candle patterns are where candlestick teaching gets confident. Two candles were a scuffle; three look like a story with a beginning, a middle and an end, and stories are persuasive. This lesson takes that confidence apart in a specific way. It shows that the classical drawings of the morning star and the abandoned baby require something a crypto chart cannot produce, that one of the two definitions of an inside bar in circulation describes almost half of all candle pairs, and — the part worth the read — that the same price path chopped into candles on a different schedule stops being a morning star five times out of six. The third candle is not the expensive part. The clock is.

Illustrative example with hypothetical numbers, drawn to the price scale on the right. Note that the three candles touch: there are no gaps between them, and that is not an artistic choice.

What is a three-candle pattern actually claiming?

That a trend ran, stalled, and then got taken back — and that you can see all three phases inside a window three candles wide. The claim is not about any one candle. It is about the shape of the sequence.

Lesson 24 read a single candle as the record of a fight inside one interval. Lesson 26 extended that across one boundary: candle 1 wins, candle 2 gives ground back, and the whole family reduced to a measurement of how much got given back. Three candles adds a distinct middle state, and that middle state is the entire point. Our course material states the morning star in three lines and the middle line is the one carrying the weight:

A small body means the interval opened and closed in nearly the same place. Lesson 25 showed what that does and does not tell you: it says the two sides finished level, and nothing about who is winning. Put in sequence, then, the three candles read as selling → nobody in control → buying. That middle beat is what separates this from a two-candle reversal, where the handover happens with no pause at all.

Two boundaries before the mechanics, both of which the rest of the lesson keeps returning to.

The first is the precondition. “Appears at the bottom of a downtrend” is the first line of the checklist, not a stylistic note. Our course notes put the reason in terms of force: a pattern is worth something only when there is a force for it to reverse, and the force pushing back has to prove itself against the one that was already there. Three candles in the middle of a quiet range are three candles. They are not a reversal of anything, because nothing was being done. Diagnosing trend or range first is not preparation for the pattern — it is half the pattern.

The second is sharper, and it comes from the same course notes' list of habits that empty accounts. One of them is named flatly: reading the signal off one to three candles instead of off the whole move. A morning star is exactly three candles. So the honest frame for this entire lesson is that a three-candle pattern is a timing device inside a read you already had — never the reason for the trade. Section 5 gives that warning a number, and the number is worse than the warning.

Why does the textbook morning star look nothing like yours?

Because the textbook drawing has gaps in it and your chart cannot make gaps. This is not a small cosmetic difference; it removes a whole tier of named patterns from the crypto vocabulary.

Read the description in our own course material carefully, and notice the clause in the middle:

The Morning Star shows a market in a downtrend: the second candle keeps falling and makes a new low below the first, so selling pressure is real. But on the third and last candle of the pattern, the open jumps clean past the range of the second — the heavy selling no longer has the strength to drag the market lower, and control passes from the bears to the bulls.

“The open jumps clean past the range of the second” is a gap. And the classical form has two of them: candle 2 drops away from candle 1's body, then candle 3 opens away from candle 2's body. The visual drama of the pattern is entirely in those two jumps — they are what makes it look like the market physically let go.

Lesson 26 established why crypto cannot do this, and the argument is worth restating in one line because everything below rests on it: on a continuously traded pair, a candle's open is the previous candle's close. The same print. There is no interval between them for price to travel across. In the 400,000-candle series used throughout this lesson, the number of candles whose open differed from the previous close was exactly 0 out of 399,998, and that is a fact about arithmetic, not about the sample.

Which means we can rank the classical candlestick vocabulary by something nobody usually ranks it by: how many gaps its definition requires. That number decides, before any discussion of reliability, whether the pattern is available to you at all.

Now look again at how our course material is written. The prose keeps the classical gap, as any candlestick book does. The operative checklist — the three lines you actually trade — drops it and keeps only downtrend, small middle body, third candle breaks the move. Lesson 26 found the identical editorial choice in the dark cloud cover entry. Twice in one section is not an oversight. It is somebody who has traded a continuous market writing down the version that can happen.

The exception is worth knowing so you are not confused when you meet it: instruments that close do gap. CME bitcoin futures gap over weekends, thin altcoin pairs gap on illiquid hours, and any market gaps after an exchange outage. If you trade those, the classical form is back on the menu. On Binance spot BTCUSDT, it never was.

So what happened to the abandoned baby?

It collapsed into the morning star. Not “became rare” — collapsed, in the sense that removing its defining conditions leaves you holding a different named pattern.

The abandoned baby is the strictest entry in the whole candlestick vocabulary. Our course material states it as: a run of heavy selling, then a baby candle that opens below candle 1 and never touches the range of candles 1 or 3, then candle 3 opening well above candle 1. That is two gaps measured wick to wick, not body to body — the middle candle has to float completely free, which is where the name comes from.

Counted the same way: 0 out of 399,997 three-candle windows, in both directions. Again the reason is arithmetic. If candle 2's open equals candle 1's close, and candle 1's close is inside candle 1's range by definition, then candle 2's range always touches candle 1's range. Always. There is no market condition that changes this, no volatility level that produces it, no timeframe on which it becomes possible.

Take the two gaps away and the abandoned baby is not a rarer pattern — it is a different pattern with a different name.

Now strip the two gaps and read what remains: a long red candle, a tiny-bodied candle at the low, a long green candle. That is the morning star, line for line. So the practical rule is blunt and it saves you from a specific kind of self-flattery:

This matters more than a naming quibble because rarity is what people use to justify size. “This one is exceptional, so I will take a bigger position” is a sentence that should be supported by something in the market, not by the label you chose.

How small does the middle candle have to be?

Nobody says. The checklist says “a small body” and stops, and the answer you pick changes how many patterns exist by a factor of nearly five.

This is Lesson 25's argument, not a new one — that lesson showed a threshold without a number is not a definition, and that the doji rate swings by more than ten times depending on which unpublished number a reader assumes. The morning star's middle candle has exactly the same hole in it. So we measured it the same way, with the same method and the same random seed so the figures in the three lessons sit on one scale.

Method, stated so it can be checked: a driftless random walk of 400,000 candles, each assembled from 60 sub-steps so the open, high, low and close come from a real price path rather than being invented; volatility 0.4% per candle; seed 20260903; each candle's open set to the previous candle's close so the series is genuinely continuous. Morning star scored as: candle 1 red with a body in the top quartile, candle 2's body no larger than the threshold, candle 3 green and closing past the midpoint of candle 1's body. There is no trend, no news and nobody trading, so everything counted is coincidence. That is the point — it is a floor, not a win rate.

Between the strictest and the loosest sane reading — 10% and 50% — the pattern is 4.84 times more or less common. Two traders working from the same three-line checklist are not looking at the same market, and neither of them is wrong, because the checklist never told them.

But here is the part that came out the opposite of what we expected, and it is the reason to keep the condition rather than dismiss it. Drop the small-body requirement completely and the rate goes to 3.334%; keep it at 30% and it drops to 0.913%. The middle candle's small body throws away 72.6% of the candidates. It is not decoration and it is not implied by the other two conditions. Of every four sequences that look like down-then-up with a proper recovery, three are eliminated by that one line, and the survivors are the ones where the market genuinely paused rather than simply turning around fast.

Set against the frequency table, one more thing falls out. At 0.913%, a morning star on a 4-hour chart shows up about once every nine days counting both directions; on the daily, once every 55 days; on a 15-minute chart, nearly twice a day. Trends do not turn twice a day. That is the same supply argument Lesson 26 made for the engulfing, and it is the mechanism underneath the course's advice to run these patterns on the higher frames — not that candles are more honest up there, but that coincidences are scarcer.

What happens if the bars had started twenty minutes later?

Most of your morning stars stop existing. This is the finding worth taking away from the lesson, and it is easier to state than to accept: the pattern is partly a property of your chart's schedule rather than of the market.

Start with the thing nobody puts in a candlestick book. A candle is not an event. It is a bucket. Somebody decided that a new bucket starts on the hour, and everything traded until the next hour goes in it. That decision is a convention of the charting software and the exchange's clock. The price path underneath does not know about it.

So we ran the obvious experiment, which the sub-step construction of the simulation makes possible. Take the identical price path. Build candles from it twice: once with boundaries on the hour, once with boundaries shifted by a third of a candle. Nothing about the market changed — not one trade moved. Then count how many patterns found on the first chart are still the same pattern on the second.

Read the fourth column first and the fifth column second, because the fourth on its own points the wrong way. 85.6% of morning stars are gone after a shift that changed nothing about the market. Five out of six of the patterns you would have traded were assembled by the clock.

The hammer looks better at 22.3%, but it is not better. A hammer occurs on 17.48% of candles anyway, so if you picked a candle at random on the shifted chart you would find a hammer roughly 17% of the time — the survival rate is only 1.28 times that. A single-candle shape is very nearly a pure product of where the bucket boundaries fell. The morning star's 14.4% against a 0.94% base rate is 15.36 times chance, which is the highest number in the table.

One price path, two chart schedules. The market did not move between these two panels — only the decision about where a candle starts.

There is a second, cleaner version of the same problem that needs no simulation at all, just counting. If your chart aggregates k lower-timeframe candles into one, a window of n candles survives intact only when it does not straddle a boundary, which happens with probability (k − n + 1) ÷ k, and never when n exceeds k.

Half of all three-candle patterns on the hourly chart are cut in half by a 4-hour boundary, and a five-candle pattern on the hourly can never fit inside one 4-hour candle at all. Lesson 21 priced what happens when you take your stop from one frame and your target from another. This is a different problem sitting next to it: when you go looking for the higher frame to “confirm” your three-candle pattern, geometry has already decided that half the time there is nothing up there to look at.

The uncomfortable part is that you never get to run this test yourself. Your platform will not offer you the hourly chart starting at twenty past. So you never see the version of your chart in which the pattern was not there, and every pattern you do see arrives looking equally solid.

Does that make candlestick patterns meaningless?

No, and the same numbers that look damning are the ones that say so. It is worth being precise here, because both the dismissive answer and the credulous answer are wrong in the same way — they read one column of the table and stop.

A 15.36× lift over chance is a large effect. It means that when a morning star forms, something about the underlying path really does make it more likely to still be a morning star on a differently-scheduled chart. There is a shape down there. What the 14.4% says is that the name does not stick to it reliably. You are trying to grip a real feature of the market with a definition that depends on an arbitrary boundary, and the grip slips five times out of six.

That has one clean practical consequence, and it is the whole reason the section on inside bars comes later in this lesson rather than earlier:

This is also where the pattern family divides. Lesson 28's double top takes weeks to build and leaves behind a neckline — a price anybody can mark. A morning star takes three candles and leaves behind nothing except the memory of three candles. Both are called reversal patterns. Only one of them gives you something to put a stop against next week.

And it puts a number on the course note about reading signals off one to three candles. That warning is usually delivered as advice about discipline. It is also a measurement: 85.6% of the objects you would be reading are artefacts of the bucket boundaries.

What does the third candle actually cost?

Less than you would guess, and much less than the clock costs you. About 7.9 percentage points of break-even win rate, which makes it the cheapest confirmation this site has priced.

The mechanism is the one Lesson 24 established: reward-to-risk is D ÷ h − 1, where h is the height of the pattern from its extreme to your entry and D is the distance from that same extreme to your target. That identity does not care how many candles made the shape. What changes with the candle count is h, and it changes for a purely structural reason: your stop goes under the lowest point of the whole window, and a wider window has a lower low.

Measured on the same series, taking the lowest low of an n-candle window and entering at the close of the last candle:

The stop grows a little faster than the square root of the candle count, steadily. Now hold the target fixed — the same level, 1,500 points away from the pattern's low on a 60,000 chart — and let only the pattern's candle count vary:

Put that next to the running tally this site has been building. The price of waiting for confirmation, in break-even win-rate points: +12.3 for a confirming candle after a single-candle pattern (Lesson 24), +18.3 for a confirmed break of structure (Lesson 17), +24.5 for the late entry in a momentum move (Lesson 22), +29.4 for a confirmed double top (Lesson 28), +33.3 for waiting on three indicators to agree (Lesson 23). The third candle's +7.9 is the smallest entry on the list. If you are going to buy confirmation somewhere, this is the cheapest counter in the shop.

Where the money actually goes

Take the worked example from the chart at the top, with every level on a round grid line. Downtrend into the pattern. Candle 1 red, closing 61,000. Candle 2 small-bodied, low at 60,000. Candle 3 green, closing 63,000 — past the midpoint of candle 1's body, so the pattern qualifies. Stop under 60,000, target at the next resistance, 69,000.

One thousand points of extra enthusiasm on the third candle costs more break-even win rate than the entire decision to use a three-candle pattern instead of a one-candle one. And notice which one feels better at the time: the bigger, more convincing third candle. This is the same trap Lesson 26 found in the engulfing — the version that looks most decisive is the one you pay most for — and it repeats here because the geometry repeats. The size of the final candle both widens the stop and eats the distance to the target, so it hits you twice.

That gives you a four-second test, in the same shape as Lesson 24's:

What is an inside bar, and which definition did you learn?

An inside bar is a small candle that sits within the previous candle's range. There are two versions of that sentence in circulation, they differ by 4.43 times in how often they occur, and both appear in our own course material a few lines apart.

Here are the two, word for word from the source. The description of the pattern says the small candle's open and close both sit inside the range of the long-bodied candle before it, and adds that its wicks may sit outside without breaking the pattern. The checklist for a sequence of them says, ideally, every baby candle sits inside the mother candle's range — which is the textbook version, range within range.

Before counting them, notice what the continuity rule from section 2 does to the first definition. If candle 2 opens exactly where candle 1 closed, and a close is always inside its own candle's range, then candle 2's open is inside candle 1's range automatically — measured across the whole series, 100.0000% of the time. Half the definition does no work at all. (Lesson 26 did this same reduction to the engulfing rule; the technique is that lesson's, the result here is new.) What is left is one comparison: did candle 2 close inside candle 1's range?

The loose reading fires on nearly half of all candle pairs in a market with nobody in it. At that rate it is not a pattern; it is the default condition of a chart. If you learned the inside bar that way, the thing you have been marking up is roughly “the market did not do very much for one candle”, which is true about half the time and carries no information by itself. The textbook version at 10.383% is a real, if common, event.

One result here is worth pausing on because it runs against the instinct. Filtering for a long-bodied mother candle does not make inside bars rarer — it makes them more common, 10.383% to 13.696%. Of course it does: a taller mother is easier to sit inside. So the mother-size filter is not a scarcity filter, it is a meaning filter. It does not reduce how often you see the shape; it changes what the shape is telling you when you do. Those are different jobs and it is easy to credit a filter with the wrong one.

Which leads to the way to hold the pattern in your head:

Why is waiting inside a mother candle cheaper?

Because the stop is pinned to a candle that has already finished forming. This is the one structural advantage in this lesson, and it is the reason the inside bar earns its place next to the star patterns rather than in a footnote.

Everywhere else on this site, waiting costs money in a specific way: the level you are stopping against keeps moving away from you while you wait. In a three-candle window, every additional candle can print a new low, so the stop widens and the entry drifts towards the target at the same time. That is the whole content of the table in section 7.

The inside-bar sequence breaks that link. Our course material describes it exactly: the mother candle, then two, three or more baby candles ideally staying inside her range, then a breakout candle that must close beyond the close of the mother candle. The stop goes beyond the mother's extreme. Once the mother has closed, that price is fixed. The babies cannot move it, because by definition they stay inside it.

The stop line does not move while the baby candles form. That is the structural difference between this setup and every star pattern in the lesson.

Measured, and stated honestly rather than as a slogan. Going from one baby candle to four or more:

So waiting is 1.84 times cheaper when the stop is anchored to the mother. It is not free — the breakout candle itself still has to travel, and a bigger breakout candle still costs you exactly as section 7 describes. But the consolidation is close to free, which is unusual enough to be worth building a habit around.

It also lines up with what our course notes say about stop placement more generally: the stop should correspond to the structure of the timeframe you are actually trading. A stop under a three-candle window is a stop against the smallest structure that exists — three candles of noise. A stop under a decisive mother candle is a stop against something that at least required somebody to do something. Both are on your chart. Only one is a level.

PRACTICE CORNER

Twenty minutes, and it will change which of these patterns you take. Open a 4-hour chart of any liquid pair on the exchange you already use and scroll back six months. First, find three morning stars or evening stars — at the 30% reading they turn up about once every nine days, so six months is plenty. For each one, write down three numbers: the low of the three-candle window, the close of candle 3, and the next obvious level in the direction of the trade. Then apply the four-second test: is the entry more than a third of the way from the window's low to that level? Most will fail it, and the ones that fail worst will be the ones with the most impressive third candle. Second, and this is the part that sticks: switch the same chart to the 1-hour and look at the same moment in time. Half the time the three candles you were admiring will have been re-cut by the boundaries into something with no name at all. That is the 85.6% from section 5, happening on your own screen.

Affiliate disclosure: the links below are partner links. We may earn a commission at no cost to you. It does not change what this lesson says. Full disclosure.

Binance→ OKX→ Bybit→

How do you run the check, in order?

Order matters, because three of these five steps can end the trade before you spend any attention on the candles. Run them top to bottom and stop at the first failure.

1. Is there a trend to reverse? The first line of the checklist, not the last. If the last twenty candles are a range, a morning star inside it is three candles and no claim. Diagnose this first.
2. Where is the level, and is it mine or the pattern's? Find the next obvious support or resistance in the direction of the trade before you look at the shape. If the only reason you have a target is that a pattern appeared, you have no target.
3. Four-second test. From the window's extreme to that level, is the entry inside the first third? If not, stop here. This kills most candidates and costs nothing.
4. Now check the shape. Candle 1 decisive. Candle 2's body genuinely small — pick your number and write it down, because the checklist will not. Candle 3 closing past the midpoint of candle 1's body. And no gaps, because there will not be any.
5. Set the stop against the best level available, not the window. If there is a swing low a little below the pattern, use it. Anchoring to three candles of noise is the cheapest-looking and least durable choice on the chart.

For an inside-bar sequence the order is the same, with one substitution: at step 5 the mother candle's extreme is the level, and the waiting between mother and breakout is the one part of this lesson that does not charge you much.

When is everything above wrong?

Four conditions, and the first two are the ones most likely to apply to you.

If your instrument closes, the gap patterns are real again. Everything in sections 2 and 3 depends on continuous trading. CME bitcoin futures gap over the weekend; equities gap every night; a thin altcoin pair on a quiet Sunday can gap on a single large order. On those charts the classical morning star and even the abandoned baby exist, and the drama the textbook describes is genuine drama. The argument here is about 24/7 spot and perpetual pairs, and it is only about them.

Every simulated number is a floor, not a forecast. The 0.913%, the 45.972%, the 14.4% survival — all of them come from a market with no participants, no trend and no memory. They tell you how often a shape appears by coincidence, which is the right benchmark for asking whether a shape is self-evidently meaningful. They tell you nothing about whether a pattern wins, because there is nobody in that market to win against. Anyone quoting a candlestick “success rate” owes you their sample, their exit rule and their definition of the middle candle's body, and almost nobody supplies all three.

A random walk has no momentum, which cuts both ways here. Real reversals cluster at levels, and real ranges compress before they expand, so both the frequency figures and the clock-shift survival rate would move in a real series. The direction of the clock-shift result is safe — boundary dependence is a geometric fact, not a statistical one — but if you want the exact number for BTCUSDT on your own timeframe, measure it rather than trusting ours.

The thresholds in this lesson are ours, not the industry's. Thirty per cent for a small body, the top quartile for a decisive candle, the midpoint of candle 1 for a valid third candle: we picked those so the arithmetic could be checked, and said so each time. Someone using 15% and a two-thirds recovery will get different frequencies and a different-looking market, and will not be wrong. That is precisely the problem Lesson 25 identified and it has not gone away.

What are the most common mistakes here?

• Drawing the gaps. People sketch the textbook morning star with its two jumps, then look for it on a chart that cannot produce them, and quietly conclude they are bad at spotting patterns. There is nothing to spot. Learn the no-gap version, which is the one our own checklist teaches.
• Treating an “abandoned baby” as a rare event that justifies extra size. On a continuous chart it is a morning star with a more exciting name, at ordinary morning-star frequency.
• Never fixing a number for the middle candle. Without one you will unconsciously loosen it when you want a trade and tighten it when you are reviewing your losses. Write the number down before you need it.
• Letting the pattern supply the stop. The window low is the most convenient level on the screen and the least durable. Section 6 is the reason: it will not be there when the chart is re-cut.
• Reading a bigger third candle as a stronger signal. It is a more expensive one — +11.1 points of break-even win rate for one grid line in the worked example.
• Marking every quiet candle as an inside bar. Under the loose definition that is 45.972% of candle pairs. Use range-inside-range, and require the mother to have done something.
• Concluding from section 5 that candlesticks are noise. The 15.36× lift says otherwise. The finding is about the label's grip, not about the market's structure.

What else do people ask about three-candle patterns?

Sources: our own course material on Japanese candlestick patterns (Part 6), which supplies the definitions and the entry, stop and target rules quoted here; and an original simulation written for this lesson — a driftless random walk of 400,000 candles built from 60 sub-steps each, volatility 0.4% per candle, seed 20260903, with every candle's open set to the previous candle's close. The same method and seed were used in Lessons 25 and 26, so the frequencies in the three lessons are directly comparable; the script reproduces Lesson 26's published engulfing rate of 12.521% (12.522% here) and mean candle range of 0.581% as a check on itself. All simulated figures are floors produced by coincidence in a market with no participants, not win rates. Published 3 September 2026.