Chart Patterns

Scope. This article describes the pattern vocabulary practitioners use and what the research has found about it. It does not teach pattern trading. No entry, exit, target or confirmation rule appears, no pattern is described as reliable or unreliable, and no backtested result is given. It is the second of the four-article technical cluster introduced in Technical Analysis: What It Is and What the Evidence Says.

The vocabulary is large, old and mostly inherited from the 1930s and 1940s, and it is the most visible part of technical analysis — the part that gets drawn on television.

It is also the part where the identification problem bites hardest, and that problem, rather than any question about whether prices trend, is the reason this material is difficult to evaluate.

The vocabulary practitioners use

Reversal shapes are read as marking the end of a move. The head and shoulders is three peaks with the middle one highest and the two outer ones roughly level, joined by a line drawn across the intervening troughs called the neckline. Double and triple tops and bottoms are two or three turns at roughly the same level. Rounding tops and bottoms are gradual curved transitions rather than sharp turns.

Continuation shapes are read as pauses within a move. Triangles come in three forms — ascending, with a level upper boundary and rising lows; descending, the mirror; and symmetrical, with both boundaries converging. Flags and pennants are short consolidations after a sharp move. Rectangles are sideways ranges between two roughly parallel boundaries. Wedges are converging boundaries that both slope the same way.

Gaps are discontinuities between one period's range and the next, and unlike the shapes above they are an observable fact about the data rather than an interpretation — a gap either exists in the series or it does not. Practitioners classify them by where in a move they occur.

Candlestick formations are a separate Japanese-derived vocabulary describing the relationship between open, high, low and close within one or a few periods, with names such as doji, hammer and engulfing.

Most of these shapes come with a measured objective — a convention for projecting a distance from the pattern, usually its own height, beyond the point where it completes. The convention has no derivation. It is a rule of thumb that entered the literature by repetition, and no article in this portal presents it as anything else.

The identification problem

Every pattern definition contains tolerances that the chartist supplies. How level is level, for two shoulders? How close is close, for a double top? Over what window does a peak count as a peak? Change any of those and the same chart yields a different inventory of patterns, and there is no authority that fixes them.

This is not a debating point raised by critics. It is the stated motivation for the most serious academic attempt to evaluate the subject: a 2000 study observed that the presence of geometric shapes in price charts is often in the eye of the beholder, and set out to remove the subjectivity by defining patterns algorithmically through nonparametric kernel regression before testing them. The subjectivity had to be engineered out before the question could even be asked.

What a series with no mechanism in it produces

The decisive question about any pattern is not whether it appears but how often it would appear anyway. That has a computable answer, and computing it is more informative than any amount of argument.

The test below generates 1,000 independent synthetic price paths of 250 trading days each, as driftless random walks with a daily standard deviation of 1.2%. Nothing in these series carries information about anything: each day's move is independent of every previous day. Swing points are identified as highs or lows over a five-day window on each side. A head and shoulders is counted when five consecutive alternating swing points form peak–trough–peak–trough–peak with the middle peak highest, the outer peaks within 3% of each other and the two troughs within 3% of each other. A double top is counted when two consecutive peaks sit within 2% of each other with an intervening trough at least 3% below.

Why the reference class is everything

The pattern literature is overwhelmingly illustrative, and illustrations are selected. A book showing thirty examples of a shape that preceded a reversal is showing thirty cases drawn from a population whose size is not disclosed. The denominator — how many times the shape appeared and was followed by nothing — is the entire question, and it is the thing an illustration cannot supply.

The same asymmetry operates in memory. A shape that was followed by a large move is memorable; the same shape followed by three weeks of drift is not, and it is not recorded anywhere. This is ordinary confirmation bias operating on a dataset with a very large number of opportunities to find shapes, which is the condition under which it does the most damage.

What survives of all this

Three things can be said without taking a side.

The shapes are descriptions of what already happened. A rectangle means the price went sideways between two levels; a triangle means the range narrowed. As descriptions these are accurate and sometimes useful shorthand — the range has been narrowing for six weeks is a fact about the chart, and it is a different statement from therefore it will break upward.

Some patterns describe order flow rather than geometry. Gaps are observable discontinuities, and the concentration of activity around prior turning points has a documented mechanism in at least one market, set out in Support and Resistance. Where a pattern names a real feature of how orders arrive, it is doing something different from naming a shape.

The definitional looseness is a property of the practice, not an accusation. Practitioners are generally explicit that identification requires judgement and that a pattern is not confirmed until it completes. The consequence is simply that the practice resists testing, which is why the literature took forty years to produce evidence either way.