Classic Patterns and the Evidence Behind Them

Every named pattern is a supply-and-demand claim

The catalogue of patterns is long, and it is worth noticing that the good ones all encode the same three situations. A pattern describes either an *accumulation*, where a large buyer is absorbing supply inside a range and the exit is the moment the supply is exhausted; an *exhaustion*, where the buyers who drove a move are finished and the last of them are trapped; or a *failed auction*, where price goes looking for liquidity beyond a level, finds none, and returns. A head and shoulders is exhaustion: the second shoulder is a failure to make a new high with the same force, and the neckline break is the point at which the participants who bought the right shoulder accept they were wrong. A flag is accumulation: a sharp advance, a shallow drift back on declining volume, and a continuation once the pause has digested the sellers. An ascending triangle is the same thing with a flat ceiling, where a specific seller at a specific price is being eaten by successive buyers. Once the pattern is stated that way, its strengths and weaknesses become visible. The strength is that the shape is a *consequence* of real order flow — a triangle with a flat top is not decorative, it is evidence that somebody was selling at that price and that the buyer eventually cleared them. The weakness is that the same shape can arise from other causes entirely: a stock can drift sideways in a narrowing wedge because of a scheduled index change, an option expiry, or simply because nobody has an opinion. So the pattern narrows the possibilities rather than identifying one, which is exactly why the base rates cluster in the fifties and sixties. Research that tested technical patterns statistically rather than anecdotally found that some formations carry genuine incremental information about the distribution of future returns and others do not, and that the information is small relative to the noise around it. The practical consequence is that a pattern is a *trigger* attached to a thesis, not the thesis. The tradeable content of a head and shoulders is the level it defines and the invalidation above the right shoulder: the pattern tells you where you are wrong, and that is the part with a number on it. The measured move — the height of the pattern projected from the break — is a convention that survives because it is easy to compute and widely watched, not because there is a mechanism that delivers it; and the honest reading is that the target is a place to reassess rather than a place to be filled. The pattern, priced — Pattern height (the head less the neckline): $9.40 on a $104 stock — about 9% ← · Measured move: Projected from the break: a target, not a mechanism · Invalidation: Above the right shoulder — a level with a number on it, which is the pattern’s real gift · Base rate: Roughly 55–65%, weaker when it forms against a strong trend ← The three families worth remembering: accumulation (a buyer absorbing supply, continuation), exhaustion (the last buyers trapped, reversal) and failed auction (price tried and was refused). Names are labels for those three, and the label is not a claim about reliability.

What separates a pattern that works from one that does not

Four things, and three of them are checkable before the trade. Location: the same triangle is a continuation setup when it forms at a prior high after an advance and a coin flip when it forms halfway through a four-month range with no prior trend to continue. Volume: the break has to arrive with participation, and a break on 0.6× average volume is a drift that is likely to be given back, which is the filter a confirming retest supplies. The pattern’s own geometry: a flag that retraces 20% of the previous advance is a different object from one that gives back 80% of it, and a head and shoulders whose right shoulder is higher than its head is not a head and shoulders. And the fourth is the market: the same pattern in a trending index and in a choppy one are different propositions, which is the regime check from the previous lesson arriving once more. The most expensive pattern mistake is not choosing the wrong name; it is letting the name replace the arithmetic. Once a learner has labelled the shape, the label acquires authority — measured moves become targets, base rates become expectations, and the invalidation level that the pattern actually defined gets ignored in favour of “the pattern is still valid”. The defence is to write the trade down in the fields a pattern can actually supply: the level, the trigger, the invalidation, the target and the size. If the setup cannot produce those five fields, the name is decoration. There is one more honest observation about the whole catalogue: patterns are abundant enough to be found retrospectively and rare enough to be unreliable prospectively. A chart of two years contains dozens of formations, and a learner who knows twenty names will find one that explains any move after the fact. The test that keeps this honest is to write the pattern down *before* the outcome, with the level it must clear and the price at which it fails, and then track the outcomes. That is simply journaling applied to chart shapes, and it is the only way to find out whether a particular pattern works in the markets you actually trade — which will differ from the book’s sample, and often by more than the pattern’s edge. • Location is the largest single determinant of whether a shape means anything. • A break without volume is a drift: the retest is the filter that separates the two. • Geometry is a filter too — a flag that gives back 80% of the move is not a flag. • The pattern’s real gift is the invalidation level, because that is the part with a number on it. • Measured moves are reassessment points rather than delivery mechanisms. • Track the names you use, in writing, before the outcome: the book’s sample is not your market. A specific and costly confusion: treating a pattern’s measured move as a target while treating the pattern as infallible. The measured move is computed from the pattern’s height, so it inherits the pattern’s error bar — which is exactly the thing a fifties-to-sixties base rate describes. A plan that risks 1R to capture a measured move three times that size is fine arithmetic; a plan that does so while refusing to honour the invalidation is a large position in an unmeasured idea.

The clock inside the pattern

A pattern has two dimensions and most instruction covers only one. The shape is the obvious part — the two peaks, the triangle, the flag — and the other is **duration**, which is how long the formation takes to build and, more importantly, what that duration implies about the outcome. The distinction that makes duration useful is between a *price* correction and a *time* correction. A trend that pauses by falling back is absorbing supply through price; a trend that pauses by going sideways for six weeks is absorbing supply by letting sellers meet buyers at roughly one level. The second transfers the same shares without the price moving, and it leaves a different footprint: volume tends to build in the formation, the range has a centre of gravity, and the eventual breakout starts from a base rather than from a retracement. Both can precede continuation, but the measurement is different — a time correction’s move is often measured from the base it built, which is why the width of the base matters more than the depth of a pullback. Duration also sets the expectation for how long the result should take, and this is where it becomes practical rather than descriptive. A three-week flag that resolves does so in a rhythm set by the three weeks that built it; a nine-month base that resolves produces a move whose timescale is measured in months. Holding through a resolution with the wrong expected duration is how a pattern trade is abandoned one week before it works, and it is a failure of expectation rather than of analysis. The third use is the one that connects the pattern to this subject’s scepticism. Most named formations, measured across a large sample, have a modest edge and a wide dispersion, and the ones that do better tend to share a property that is not in the shape: they occur in the direction of the longer trend, they occur after a period of unusually low volatility, and the breakout arrives with relative volume. Duration is the variable that is least dependent on a judgement call — the shape you think you see can be argued about, but the number of bars in the base cannot — which is why it is the one worth writing down in a plan. So the way to use a pattern without relying on its name is to grade it on the things that can be counted: how long it took, how quiet it became, how much volume accumulated, and whether the breakout happened on relative volume. A long, quiet base that breaks on volume is a different proposition from a two-day wobble that has been given a name, and the two are drawn identically in a book of patterns. • Duration is the countable dimension: the shape can be argued about, the bar count cannot. • A time correction absorbs supply sideways; a price correction absorbs it in the retracement. • The base’s width sets the target and its length sets the expected timescale. • Grade the pattern on duration, quietness, accumulated volume and the breakout’s relative volume. A pattern base with declining volume and a tight closing range is doing something specific: it is transferring shares from impatient holders to patient ones without moving the price. That transfer is the mechanism, and it is easier to see in a volume profile than in a sketch.

Sources

• Measured moves, failure rates and pattern reliability dataBulkowski, “Encyclopedia of Chart Patterns”
• Head and shoulders, triangles and flags as supply-and-demand structuresMurphy, “Technical Analysis of the Financial Markets”
• Why chart patterns are weak in isolation and stronger with volume and locationLo, Mamaysky & Wang, “Foundations of Technical Analysis” (Journal of Finance)