Head and shoulders, and why three peaks get a name

Three peaks in a row: one, then a higher one, then a third at roughly the height of the first. The line drawn under the two troughs between them is the neckline. Turn the whole thing upside down and you have an inverse head and shoulders, read as the opposite.

The story attached to it

The conventional reading is that the first peak is a normal advance, the head is the advance pushed too far, and the second shoulder is a failed attempt to repeat it. A fall through the neckline is treated as the point where that failure is confirmed. It is a tidy story and it is told about a shape, not about anything the company did.

Where it goes wrong

• Three peaks appear constantly in random data. The human eye finds the shape whether or not anything produced it.
• The pattern has no fixed definition: how close the shoulders must be, how deep the troughs, over what period. Two analysts can disagree about whether one exists at all.
• It is identifiable only after the right shoulder forms, by which point the move it supposedly predicts is partly done.

The Lo, Mamaysky and Wang study that tested chart patterns computationally had to define each one precisely before a machine could find it, which is itself the point: the definitions are not agreed, and a pattern you cannot specify is a pattern you cannot test. Their finding for head and shoulders was the same modest one as for the other patterns, information slightly beyond the price, not a tradable edge.