2006 Dec 06 - Wed

Finding Local Peaks in Quote/Trade Streams

Just about any book on technical analysis you open will have a number of charts showing the usual peaks and valleys of a instrument's trading range. Many trading strategies are designed around the specific arrangement of peaks and valleys. I thought, up till now, that these peaks and valleys could only be determined through studies through the use of the good old Mark I Eyeball.

However, after reading Bollinger's book entitled Bollinger on Bollinger Bands, in which he discusses computer aided determination of those peaks and valleys, I set out to work on an algorithm to do the same.

The C# PeakMatch Code Segment is my first attempt at peak and valley pattern matching.

It is implemented as a state machine in order to make it easy to determine, during live streams, whether the stream is going up or down. You can use various summary statistics from quotes, trades, or even bars as input values.

The variable dblPatternDelta is the grey zone used for determining when the pattern flips. This variable will need to be adjusted on an instrument by instrument basis. In addition, further tuning is necessary if you wish to capture small nuances or just large swings in the trading value. As such, determining the peak is a problem of lag. Sigh, so much for having a magical realtime signal for determining when the top or bottom of a range. The variable dtPatternPt1 holds the DateTime of the last determined peak. When a change of direction of determined, then that peak/valley attribute is stored in dsPattern.

dsPattern can then be used within a sliding window of pattern analysis. I'll show an implementation of Arthur Merrill's patterns in another article.

Further information on choosing a good value for dblPatternDelta can be found in Bollinger's book.