V.13:08 (352-358): Pattern Recognition and T-Bond Futures by Scott W. Barrie
Pattern recognition is used to standardize and categorize market behavior into quantifiable market movements. After a collection of patterns have been
identified, the market movement following each pattern can be measured. This post-pattern description can be used as a forecasting tool. In this article, Merrill waves are used as the basis for the patterns and forecasting.
The goal of market analysis is to answer two simple questions: What is the direction of prices, and how much time will elapse before the direction changes? Pattern recognition is one avenue that an analyst may take to answer these questions. A major problem in designing a pattern recognition system lies in the fact that not all patterns are static in terms of time and price. Some simple chart patterns, such as a key reversal, are static in terms of time and price; for example, an operational definition of a key reversal could be "The high today is greater than yesterday's high, and the close today is less than yesterday's close." This pattern is static in terms of time (only two days to complete) and static in terms of price (today's high is either higher than yesterday's high or it's not, and the same can
be said for the close). Candlestick patterns, another set of chart patterns, in general lend themselves to simple static definitions.
Many chart patterns are much more complex; for instance, consider the head-and-shoulders pattern. This pattern is not static in terms of price or time. A head-and-shoulders pattern can take from several days to months to form, and the relative positions of the left and right shoulder can be different, and even double- and triple-shoulder formations are permissible. All possible combinations of a simple head-and-shoulders pattern is beyond any simple static definition, but there are ways to reduce the complexities of patterns. The easiest way to overcome these difficulties is to incorporate filtering into the pattern recognition process.
In my previous article, I discussed using filters of x % as a basis for building a reference point to predict future market behavior. This is done by grouping together all moves of a certain percentage magnitude and taking average behavior after such moves. This is a simple form of pattern recognition. The pattern, a filtered wave , is composed of the inception point (starting point) and a point that is x % above or below the inception point, referred to as the