Pennant: A Shape That Breaks Both Ways
A pennant is a sharp move followed by a brief triangle whose boundaries converge, traded on a break in the direction of the original move. Converging shapes resolve in both directions, and the ones that resolve the other way are given a different name, which flatters the pattern's record.
How it works
A sharp directional move, then a brief pause in which the highs come down and the lows come up. The two boundaries converge toward a point, and the pattern is traded on a break in the direction of the original move.
The distinction from a flag is the boundaries. A flag’s are roughly parallel, so its break price barely moves. A pennant’s converge, so the break price falls every bar on the upper side and rises every bar on the lower one.
Brevity is part of the definition. A pennant is a short pause — conventionally one to three weeks on a daily chart. A converging triangle that takes three months is a symmetrical triangle, a different pattern with no directional claim attached.
The naming problem
A converging triangle is a shape that must resolve, and it can resolve either way. When it resolves in the direction of the prior move, it is a pennant and the pattern worked. When it resolves the other way, it is usually described as a reversal — and it stops being counted as a pennant at all.
That is a selection effect built into the vocabulary, and it is worth naming plainly. If the failures are renamed rather than recorded, the pattern’s apparent reliability is a property of the naming convention rather than of markets.
The fix is to count both. Every converging triangle after a sharp move, recorded before the resolution, with the direction it eventually broke. That is a few hours of work on a chart and it is the only version of the question that can be answered.
The apex is a deadline. As the boundaries converge, the space price can occupy shrinks toward nothing, so a break becomes arithmetically inevitable. That is not a prediction about the market — it is a consequence of drawing two converging lines, and it would be true of any shape drawn that way.
In practice: what the break is worth
On this site’s shared 576-bar history there are 39 closes above a 20-bar high, and 38% of them had a higher close ten bars later.
The base rate over the same horizon, for any bar at all, is 54%. Breaking a recent high did worse than doing nothing on this data — and 85% of those breakouts closed back below the level they broke within ten bars.
The series is synthetic and has no trends driven by anything, which is the point. These figures are the null model: what a breakout looks like with no mechanism behind it. A pennant method has to beat those numbers on a real instrument before the shape is contributing anything the underlying breakout was not already doing.
Volume contracting through the pennant and expanding on the break is the classical confirmation, and it is the only element not derived from price. A break on shrinking participation is the pattern’s own warning sign.
The stop is the opposite boundary, and that boundary is moving. Late in the formation the two sides are close together, so the stop is tight — which sounds good and means the position is being placed where a small counter-move removes it. Early in the formation the stop is wide. The same pattern therefore offers a completely different trade depending on when you take it.
Aggregate the chart and the triangle disappears. Ten daily bars become two weekly bars, and two bars cannot converge. The formation is a property of the resolution.
A gap out of a tightening apex is common — the pattern compresses into a catalyst — and it removes the entry. The cleanest resolutions are frequently the ones nobody could take.
Each attempt costs 2% of a typical bar’s range in round-trip costs on this history, and a converging shape near its apex produces several marginal breaks in quick succession.
And both boundaries are yours. Which highs and which lows define them is a choice, and a slightly different choice puts the break at a different price on a different bar.
What a pennant is not
It is not a small symmetrical triangle with a guarantee attached. It is a small symmetrical triangle with a directional expectation attached, and the expectation comes from the move before it rather than from the shape.
It is not a flag. Converging versus parallel changes where the entry is and how the stop behaves over time.
It is not defined numerically. How sharp the prior move, how many bars the triangle, how much convergence — none of it is specified.
And it is not evidence of compression in the market. It is evidence that the highs you selected are falling and the lows you selected are rising, which is a statement about your line choices.
When it fails
The main failure is the break in the wrong direction, which is not recorded as a pennant failure because the shape gets renamed. Any honest evaluation has to log the formation before it resolves.
The second failure is the apex overrun. Price reaches the point where the lines meet and simply continues sideways, at which point there are no boundaries left and no pattern.
A third is the late entry. Taking the break in the last few bars before the apex gives the tightest stop and the highest chance of being removed by ordinary noise, because the shape has compressed to less than the instrument’s usual bar range.
A fourth is repeated marginal breaks. A converging shape produces several small violations of each boundary as it tightens. Trading each one is several round trips for one eventual move.
And a fifth is the missing volume check. Contracting participation into the apex is what the classical version requires; without it the triangle is simply a quiet stretch of chart with two lines on it.
The original data
39 closes above a 20-bar high on this site’s shared 576-bar history: 38% higher ten bars later against a
54% base rate, and 85% back below the broken level within ten bars. The 10-bar and 55-bar lookbacks are
recorded with them in research/series-measurements.json, produced by site/measure_series.py.
The measurement this pattern specifically needs is a direction count, and it is one you have to make yourself. Log every converging triangle that follows a sharp move — before it resolves — and record which way it broke. If continuation and reversal come out near even, the pennant is a coin flip with a name on one side, and every published claim about it is a report on the half that kept the name. That count takes an afternoon on a year of charts, and no article can do it for you because the recording has to happen before the outcome is known.
Related
Symmetrical triangle is the same shape without the directional claim. Bull flag is the parallel-boundary version, with the breakout arithmetic in full. And breakout is what the entry is once the drawing is set aside.
Pennants are where I learned to be suspicious of names. A converging triangle after a big move breaks upward and it is a pennant that worked; it breaks downward and suddenly it was a reversal pattern all along. Same shape, two names, and only one of them gets counted.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.