Home › Learn › Lesson 32

Triangles — ascending, descending, symmetrical, and how to tell a pennant apart

By TradingPrimer · Published 4 Sep 2026

Every guide gives triangles the same three-line summary: ascending is bullish, descending is bearish, symmetrical could go either way. Our own course slides say something different, and they say it in writing on the charts themselves — the ascending slide says the breakout direction depends heavily on the trend already in force and the descending slide says which way this pattern breaks depends heavily on the trend in force, and all three shapes are filed under the heading bilateral. That is not the course hedging. This lesson shows that the arithmetic agrees with the slides and not with the textbook: the same triangle pays exactly the same whichever way it breaks, which means the shape cannot be telling you the direction. Then it prices the two things those slides do mention and nobody ever costs — the four-touch rule, and the backtest.

Schematic only — deliberately drawn without a price scale, because nothing in it is meant to be measured. The four dots are the point of the picture — two on each edge — though the order in which they arrive is not the order the worked example below uses. Every number in this lesson lives on the charts further down, which are drawn from the prices themselves.

What actually makes a shape a triangle — and how many touches does the course require?

Two boundaries that converge, and four touches — at least two on each edge. The convergence is what makes it a triangle rather than a flag; the four touches are what make it a pattern rather than a doodle.

Lesson 29 already set out how the converging shapes divide up by the slope of their two boundaries, and there is no reason to redraw that table here. The one-line version: if a boundary is flat you have a triangle, and which boundary is flat gives it its name. Flat ceiling with rising lows is an ascending triangle. Flat floor with falling highs is a descending triangle. Neither flat, leaning towards each other, is a symmetrical triangle. Our course files all three together under one heading — bilateral patterns (continuation / reversal) — and Section 4 explains why that heading is the honest one.

The qualification rule is the part that never makes it into the summaries, and it is written directly on the course charts for the ascending and descending cases: to qualify, price must touch each edge at least twice. Four touches means three swings between them, and those three swings are the reason the rule matters more than it looks.

Two lines drawn through two points are not a pattern; they are a pair of points with lines attached. Any two highs define a ceiling and any two lows define a rising floor, so a chart with three swing points on it can always be made to look like a triangle by whoever wants to see one. The fourth touch is the first piece of information you did not choose — the first time the market gets to agree or disagree with the lines you already drew.

Read the left panel in order. Price makes a low at $3,040, runs up to a ceiling at $3,280, comes back to the rising floor at $3,136, and returns to the ceiling. That fourth touch is the moment the shape is allowed to be called a triangle. The three swings are +$240, −$144 and +$144, and the right panel shows what those swings did to the gap between the two lines: $240 at the first touch, then $180, then $144, then $108.

That last number is the one that decides the trade, because the gap between the two boundaries is where your stop has to fit. This is the quantity Lesson 29 calls c when written as a fraction of the opening width, and everything in the next section is about a fact that lesson did not need: the qualification rule and the width are the same measurement. You cannot wait for the fourth touch without also waiting for the pattern to narrow, and you cannot have a narrow pattern that is also unqualified.

Why is the four-touch rule really a stopwatch?

Because the fourth touch does not arrive at a fixed place. It arrives when three swings have been completed, and how far into the triangle’s life that is depends entirely on how fast the shape closes compared with how fast price moves inside it. One number captures the ratio, and it is measurable in seconds.

On the chart above, the mouth is $240, the apex is 20 candles away, and we will take the average candle range as $48 — a stated assumption for four-hour ETH around $3,200, and the number you should replace with your own reading. So N = 20 × 48 ÷ 240 = 4. Price has room for four mouth-crossings; the qualification rule wants three of them.

Three of four does not mean three-quarters of the way to the apex, because the crossings get shorter as the shape closes. Two things have to be said about how they shorten, and only one of them is obvious.

The obvious one is that the gap narrows. The one that is easy to get wrong is that the two directions do not narrow it equally. A swing that runs up to a flat ceiling is racing a stationary target: the gap shrinks only by however much the rising floor climbs in the meantime. A swing that falls back to the rising floor is racing a target coming the other way, so it arrives sooner and the gap shrinks less. Writing a = 1 ÷ N for the closing rate, the two rules are:

• swing towards the flat edge: the width becomes W × (1 − a)
• swing towards the sloping edge: the width becomes W ÷ (1 + a)

Three swings starting from a low on the sloping edge therefore leave a fraction c = (1 − a)² ÷ (1 + a) of the opening width. At N = 4 that is exactly 0.45 — the fourth touch lands at 55% of the way to the apex, on candle 11 of 20, with $108 of width left. Those are the numbers on the chart above, and they came out of a step-by-step simulation of price walking between two moving lines before they were checked against the formula.

Now put that width through the reward-to-risk identity Lesson 29 established for every converging pattern, R:R = 1 ÷ (c + β), taking the stop buffer as half an average candle — $24, so β = 0.10. The result is a table of what the course’s own qualification rule hands you at the earliest moment it lets you act.

Three numbers fall out of that table, and all three are checkable on any chart you like.

The 2R line sits at N = 3.53. With a half-candle buffer, a triangle that converges any slower than that cannot be a two-to-one trade at its first legal entry, no matter how tidy it looks. Not because you did anything wrong — because the rule let you in too early.

The classical break window is a range of N. Lesson 31 quotes the traditional teaching that the break usually comes somewhere between half and three-quarters of the way to the apex. Run that band backwards through the formula and it corresponds to N between 2.46 and 4.56. Inside that range, the fourth touch and the expected break arrive at roughly the same time — which is a reasonable account of why a rule about touches ever became a rule at all. It qualifies you just as the window opens.

Outside that range the two rules fall out of step, in opposite ways. Above N = 4.56 the shape becomes legal well before the window opens: you are allowed to trade it and it is not yet worth trading, so the honest move is to keep waiting and let Lesson 31’s table tell you what the waiting is worth. Below N = 2.46 the problem reverses and gets worse. A triangle that converges that quickly only collects its fourth touch after the classical window has shut — at N = 2, not until 83% of the way to the apex, by which point the width left is 0.17 of what it started as. The number in the last column looks wonderful. It is the number attached to a stop that is now smaller than the noise it was supposed to survive.

Which is worth stating plainly, because it is the least intuitive thing in this lesson: a fast-converging triangle can be legal only at the point where it has stopped being tradeable. The rule and the shape can disagree, and when they do it is the rule that is right and the chart that is flattering you.

Why does the same triangle pay exactly the same in both directions?

Because both directions use the same two lines. The distance from one boundary to the other is what your stop has to span whichever way price leaves, and the measured move is the same height whichever way it is projected. Nothing in the geometry prefers one side.

Take the qualified triangle from Section 1 at its fourth touch, with the ceiling at $3,280, the rising floor at $3,172, and $108 between them. Add the same $24 buffer to both cases and measure the target as the pattern height, $240, projected from whichever boundary broke. Both rows assume a fill at the boundary itself, which is the clean way to compare the two directions; Section 6 prices the two fills you can actually get.

Identical. Not approximately, and not as a quirk of these particular numbers: risk is the width plus one half-candle buffer in both rows, and reward is the pattern height in both rows, so the ratio cannot come out different. It is the same for a descending triangle, and the same for a symmetrical one.

The consequence is stronger than it first appears. If the geometry paid better one way, you could argue for a direction on price alone. It does not, so whatever edge a triangle has, it is not an edge the shape can express. An advantage would have to live in how often each break turns out to be real — and frequency is exactly the thing this site does not put a number on, because we have not measured it. Every percentage in this lesson that describes a trade’s odds is a break-even threshold — the strike rate a trade would need in order to be worth taking, never a claim about the strike rate it gets. The other percentages here measure geometry: how far along a pattern is, how much width is left, how much wider one stop is than another.

So the honest reading of the arithmetic is a shrug about direction. Which is precisely what our course slides do: all three triangles filed as bilateral, and the ascending and descending charts each carrying their own version of one sentence — that which way the pattern breaks depends heavily on the trend already in force.

If the arithmetic is silent, what decides which way a triangle breaks?

The trend that was already running when the triangle started. And there is a mechanical reason a triangle has to borrow its direction from outside, which the wedges do not: a triangle contains two opposite explanations at once.

Lesson 29 set out the two readings a converging boundary can support. A flat boundary is a stack of resting orders that can be eaten through, so it eventually gives way. A sloping boundary has no stack to consume; instead the side doing the chasing is the one that gives out first. Both readings are mechanisms rather than measurements — reasoning about why a convention exists, not evidence that it works — and Lesson 29 was careful to say so.

Notice what happens when you apply both to one shape. A rising wedge has two sloping edges, so only the second reading applies, and it points one way: down. A falling wedge, the mirror, points up. That is why wedges get to be directional patterns.

An ascending triangle has one of each. The flat ceiling is a stack, and the stack argument says the break comes upward through it. The rising floor is a chase, and the chase argument says the buyers doing the chasing will tire and price will fall through their own steepening support. Two mechanisms, one shape, opposite conclusions. A descending triangle is the same collision in reverse: the flat floor is a stack of resting bids that sellers can exhaust, arguing for a break downwards, while the falling ceiling is sellers marking down less aggressively each time, arguing that the selling is what is running out.

This is the point the filing system has been trying to tell you. Wedges are directional because only one mechanism is present; triangles are bilateral because both are, and they disagree. The tie-break is which side is larger, and the readable proxy for that is the trend already in force on a higher frame — which is what Lesson 21 is for, and why Lesson 20 comes before any of this.

Two things this does not license. It does not make the classical shorthand wrong: an ascending triangle inside an uptrend is still a buy setup, and Lesson 29 uses that shorthand correctly when it is separating a wedge from a triangle by slope. The shorthand is what happens when the trend and the flat edge agree, which is the common case. What the slides add is the condition: the shape names the trade, the trend decides whether the trade exists. It is worth being precise about where that leaves the stack argument, because Lesson 29 states it in causal form — resting orders get eaten, which is why an ascending triangle breaks up through its ceiling. That reasoning is sound and this lesson keeps it. What it adds is that the same triangle also contains a chase, so the stack argument is one of two present rather than the only one, and it wins when there is a trend behind it. Read that way the two lessons do not disagree; Lesson 29 is describing the case where the flat edge has the trend on its side, which is the case it needed in order to separate a wedge from a triangle. And it does not mean a flat edge is always a real level. Where the “flat” price is simply somewhere nobody happened to trade, there is no stack to exhaust and the whole mechanism evaporates. A flat edge that lands on an old high or an old low the market already respected is a different proposition from one that lands on empty air — a distinction Lesson 33 takes further.

Where should the stop actually go?

There are two defensible answers, they are $36 apart on the chart above, and the gap between them is worth 5.7 percentage points of break-even win rate. Most guides give you one of them without mentioning the other exists.

The first is the one every converging-pattern lesson uses: just past the opposite boundary as it stands at the moment of the break, plus a buffer. At the fourth touch the rising floor reads $3,172, so the stop is $3,148 and the risk is $132.

The second is structural: just past the last swing the market actually made. That low was $3,136, three candles earlier, so the stop is $3,112 and the risk is $168.

The gap exists because the sloping boundary keeps climbing after price has left it. The line at the moment of the break sits above the low that created it — by $36 here, and by more the longer the triangle has run since its last touch. So the two stops are not variants of one idea. They answer different questions.

Stop A is 21.4% tighter, and that is a real saving, not a rounding difference. But it is placed on a line that no candle has ever traded at — a construction of yours that moves every candle. Stop B is placed at a price where buyers demonstrably showed up. You are paying 5.7 points of break-even for the difference between a drawing and an event.

Neither is the right answer in general, and the useful discipline is not to pick a winner but to pick one and keep picking it. A performance log that mixes the two is measuring two strategies and calling them one, which is the same trap Lesson 31 flags for entries. What does tilt the choice is how far the last touch is behind you: on a triangle that has just turned at the floor the two stops nearly coincide and A is close to free, while on one that has run a long way since its last touch the line has drifted well above any evidence and B starts to earn its extra width.

What is the backtest worth — and why does the course bother to mention it?

Because it is the only entry in the whole pattern that reaches the ceiling. The course charts for the ascending and descending triangle both carry the note that this pattern very often backtests after the breakout, and the pricing explains why that detail earned space on a slide.

Compare the two ways in. Chasing the break means buying a close beyond the level — call it half an average candle above, so $3,304 — while your stop still belongs back under the pattern at $3,148. Waiting for the retest means letting price come back to $3,280, the level it just broke, and putting the stop one average candle below it at $3,232, because the level is now the thing being defended.

Chasing costs you twice over: a worse price on the way in, and a stop that is still measured from the far side of a pattern you have already left. The retest fixes both at once. It buys the level rather than a candle above it, and it moves the stop from “the other boundary” to “one candle below the line that is now supposed to hold”.

Look at what that second number is. Risk $48, reward $240, and $240 ÷ $48 = 5.00R — which is the pattern height divided by the average candle range, the exact ceiling Lesson 31 sets for what any converging pattern can really pay. The retest is not a good entry in the ordinary sense. It is the ceiling, reached with the pattern still intact and the direction already demonstrated, instead of by grinding towards an apex where the same ratio arrives attached to a stop that noise crosses on a quiet afternoon.

The obvious objection is that the retest does not always come, and that is right — it is the whole cost of the rule. We have not measured how often it does, and will not put a number on it. What can be worked out is how often it would need to happen, which is a different question and an answerable one. If both entries have the same chance of the direction being right — call it p — then waiting matches chasing when the retest shows up a fraction q = (1.3846p − (1 − p)) ÷ (5p − (1 − p)) of the time.

Across every value of p in that table — a range chosen to bracket the plausible, not measured — the retest has to turn up somewhere between one time in twenty-three and one time in five to be the better rule. It can miss four breakouts out of five and still win. That is a large enough margin to survive being wrong about the assumption behind it — and the assumption is worth naming, because it is doing work: it takes the two entries to be equally likely to be right about direction. If retests happen mainly on the weaker breaks, the true figure is worse than the table; if a level that gets retested and holds is a level that meant something, it is better. Either way, the margin is not small.

Note also what this comparison is not. It is not the cost of waiting for a confirming close, which is a different measurement the course tracks separately across lessons. Waiting for a retest changes the stop as well as the entry, so the two figures are not on the same ruler and should not be read against each other.

Why does the symmetrical triangle behave differently from the other two?

Because it has no flat edge, and every good thing in Sections 4 to 6 came from the flat edge.

The course slides treat the symmetrical triangle as a different animal in two specific ways, and it is easy to miss because the difference is in what they omit. The ascending and descending slides carry the four-touch rule and the note about backtesting. The symmetrical slide carries neither. What it says instead is that the pattern makes successive lower highs and higher lows, and that price usually reaches the convergence point before the breakout.

Put that behaviour through this lesson’s arithmetic and it is not a neutral observation. Reaching the convergence point means c heading towards zero, which drives the width left below the buffer and eventually below one candle — the region Lesson 31 and Lesson 29 both identify as the place where the numbers stop describing a trade. And the symmetrical triangle has no second route out of it. There is no horizontal level, so:

• nothing to retest. The boundary that broke is a sloping line, and by the time price comes back the line has moved. The 5.00R entry in Section 6 does not exist here.
• no stack to exhaust. Both mechanisms in Section 4 are the chasing kind, and they point at each other.
• nothing other traders can all see at the same price. A horizontal level is visible to everyone with a chart; a sloping one is visible to everyone who drew it the same way you did.

So the shape most people think of as the archetypal triangle is the one with the fewest usable properties — and that is a better explanation of why it prices worse than the usual complaint that it is directionally ambiguous. All three triangles are directionally ambiguous; Section 3 proved it with arithmetic. What separates them is that two of them come with a horizontal price, and one of them does not.

Which gives a single line to carry away: the flat edge is the tradeable part of a triangle, and a triangle without one is a range with a deadline. Range is not an insult — Lesson 20 is entirely about trading them — but it is a different job with different rules, and it is the job you are actually doing when you draw a symmetrical triangle and wait.

How do you tell a pennant from a symmetrical triangle without arguing about it?

Work out N. If a converging shape has walking room below about 2, it cannot collect four touches while it still has any width to speak of, so it is not a triangle. If there is a sharp run immediately in front of it, it is a pennant; if there is not, it is not yet anything.

Lesson 31 gives the full set of tests for this — the flagpole requirement, the formation time, the direction of the break, and the fact that volume will not separate them — along with what mislabelling one as the other costs. That is the section to read, and this one only adds a number to a single line of it: the course’s observation that a pennant forms in far less time than a symmetrical triangle.

“Far less time” is a comparison without a scale, and N supplies one. Forming quickly, for a converging shape, means converging quickly relative to how fast price is moving — which is a small N. Run the table from Section 2 downwards: at N = 2 the fourth touch does not arrive until 83% of the way to the apex, and below that it effectively never arrives before the shape runs out of room. So the two rules the course states separately turn out to be one rule seen from two sides. A pennant is a shape that could not satisfy the four-touch test even if you were patient.

That also explains something about the slides themselves. The qualification rule is printed on the ascending and descending charts and is absent from the pennant material — not an oversight, but the only consistent thing to do, since applying it to pennants would disqualify all of them.

In practice the test runs in the same few seconds as everything else here. Count the candles to where your two lines meet, multiply by your average candle range, divide by the width the shape opened with. Under 2, stop trying to make it a triangle and go and look at what came immediately before it.

When is everything above wrong?

The pricing in Sections 3, 5 and 6 is arithmetic and holds wherever the inputs hold. The stopwatch in Section 2 is a model, and models have conditions.

It assumes price moves at about one average candle range per candle inside the pattern, in a straight line between the edges. That is a reasonable average and a poor description of any particular week. A burst of volatility inside the triangle runs the clock fast and the fourth touch arrives earlier and wider than the table says; a dead patch runs it slow. If you want the honest version, measure N again after each touch rather than once at the start.

It assumes the swings actually reach the far edge. A swing that stalls halfway does not count as a touch and does not advance the count, but it does use up time — so shapes with lots of shallow internal noise qualify later than the model suggests. That is a bias in one direction, and it is worth knowing which.

It assumes your two lines are the right two lines. If moving a boundary by one plausible wick changes N from 3 to 5, the number is not telling you anything. Two touches per edge is the minimum for a reason, and a third touch on each is materially better evidence than a fourth candle of patience.

The buffer is a choice, not a constant. Everything here uses half an average candle, β = 0.10. A full-candle buffer doubles β and the 1.82R above becomes 1.54R — risk $156 instead of $132, break-even 39.4%. Nothing about the structure changes; the whole table just gets worse, and it should, because you have decided to survive more noise.

Thin markets break the retest case first. The $48 risk in Section 6 is one candle wide, so anything that adds a fixed cost — a wide spread, a slipped fill — eats a disproportionate share of it. The tighter the entry, the more the frictions matter, and the retest is the tightest entry in the lesson.

And the pattern can simply expire. If price crawls all the way to the apex without breaking out, there is nothing left to break out of and no version of the arithmetic that still applies. The shape is finished; read what is left as a range and start again. A break that fails and reverses back inside is a separate subject with its own rules, and it gets its own lesson later in the path.

What are the most common mistakes here?

• Trading a three-touch shape. Three points can be fitted with any two lines you like. The fourth touch is the first one the market chose rather than you.
• Calling an ascending triangle bullish and stopping there. The geometry pays the same both ways — Section 3. The direction comes from the trend already running, which is a separate piece of work on a higher frame.
• Taking the opposite side when trend and shape disagree. Disagreement is a reason to skip, not a signal in reverse. Both mechanisms are still present; you have only established that neither is clearly bigger.
• Not knowing which stop you use. Under the line and under the last low are 5.7 break-even points apart. Alternating between them by feel makes your own results unreadable.
• Chasing a break because the retest might not come. It has to appear about one time in five to beat chasing outright. Deciding in advance costs nothing; deciding while price runs away costs the difference between 1.38R and 5.00R.
• Waiting for the apex on a fast-converging shape. Below N = 2.46 the pattern qualifies only once the stop no longer fits inside it. The multiplier goes up and the trade goes away.
• Treating a flat edge as automatically meaningful. A horizontal line through two highs in a stretch of empty chart is a coincidence with a ruler through it.
• Sizing off the R:R instead of off the risk. A 5.00R entry with a $48 stop is not permission to take a larger position — the position comes from the account, per Lesson 42.

What else do people ask about triangles?

PRACTICE CORNER

Do this on your own charts before the next section of the path, because it takes one number and settles two arguments at once. Open any four-hour chart, find a converging shape and count: how many candles to where your two lines meet, what the average candle range is, and how wide the shape was when it opened. Multiply the first two, divide by the third, and you have N. Under 2, it is a pennant or nothing. Between 2.5 and 3.5, its fourth touch and the classical break window arrive together. Over 4.6, you are allowed to trade it long before it is worth trading. Then mark the fourth touch and write down both stops — under the line, and under the last real low — and note which one you would actually have used.

Doing that on live charts needs a platform where you can draw two boundaries and read an average candle range off the same screen. These are the three exchanges this site uses for its own worked examples; all three have chart tools that will do it, and you can size the exercise at nothing at all.

We may earn a commission if you open an account through these links, at no cost to you. It does not change what is written above.

Binance→ OKX→ Bybit→

Educational content only — not financial advice, and not a trade recommendation. Every figure on this page comes from one worked model built for this lesson and each can be reproduced from the numbers given: an ascending triangle on a four-hour chart with flat resistance at $3,280 and a lower edge rising from $3,040, so it opens W₀ = H = $240 wide and meets its apex 20 candles later; an average candle range of $48 and a stop buffer of half that, $24. From those, N = 20 × 48 ÷ 240 = 4, the fourth touch falls on candle 11 with $108 of width left, and R:R = 1 ÷ (c + β) gives 1.82R. The swing schedule was produced by simulating price walking between the two lines at one average candle range per candle and then checked against the closed form c = (1 − a)² ÷ (1 + a); the two agree to four decimal places. Break-even win rate is 1 ÷ (R + 1) throughout. The $48 candle range, the $24 buffer and the one-candle-per-candle walking speed are stated modelling assumptions, not measurements of any particular market; substitute your own. Every percentage on this page that describes a trade’s odds is a break-even threshold — the win rate a trade would need in order to be worth taking; the rest measure geometry, such as how far along a pattern is. No historical hit rate or pattern reliability percentage is quoted anywhere, because none was measured. The measured-move target rule, the four-touch qualification rule, the note that these patterns often backtest and the observation that a symmetrical triangle usually reaches its convergence point are all taken from the chart-pattern section of the TradingPrimer slide course, including the labels written on the course charts themselves. The order-book readings in Section 4 are mechanisms offered to explain a convention, not evidence for it. Sources: that course material, plus the arithmetic above.