Head & Shoulders, Algorithmically (Part 1) by Giorgos Siligardos
New Tech For Old Tech
Ever wanted to use a simple mechanical recognition algorithm
in your favorite software for spotting head & shoulders patterns?
If yes, this two-part article is for you. Here’s the first
If you were asked to represent the term “chart analysis”
somehow, you would probably start with a drawing of
the patriarch of chart formations: the head & shoulders
(H&S). The H&S is usually one of the first formations you’ll
encounter in chart analysis, and it is certainly an unforgettable
one. In this article — the first of two — I will present a way
to algorithmically identify milestone points of a H&S in a
scale-free fashion. Then, in the second article, I will tell you
about the basic conditions these points satisfy so a recognition
algorithm will be achievable.
Pattern recognition & technical analysis
The difficulty in algorithmic pattern recognition for technical
analysis lies mainly in variation and scale problems. They are:
The beauty of a price formation is in the eye of the beholder.
This means that since price moves in zigzags, there are endless
ways for a textbook pattern to manifest itself in a real
price chart. If you try to create a pattern recognition algorithm using a simple swing indicator (that is, a crooked line that
connects peaks to troughs based on a percentage threshold),
you will probably give up when you realize that your eyes,
perception, and probably your charting experience are not
satisfied with the results and that numerous cases must be
taken into account.
Almost all chart formations of classic technical analysis are
scale-free. A formation may take from a few to as many as
hundreds of price bars to develop and it may take price swings
ranging from minuscule to huge, so again the use of a simple
swing indicator is inefficient for decent pattern recognition
and of little practical help most of the time.