A Wavelet Based Evaluation of the Head and Shoulders pattern

Daniel Persson & Bo Sandström

Handledare: Jan Holst

Centre for Mathematical Sciences
Mathematical Statistics
Lund Institute of Technology,
Lund University,

Trying to predict stock prices based solely on stock price and volume information is the controversial and mocked financial science called technical analysis, by many financial experts referred to as "Voodoo finance" and a game of chance. Still, numerous of well-established investors such as George Soros do not conduct a trade without consulting a technical analyst. Technical analyses try to forecast future stock prices by finding patterns and geometrical shapes in stock prices. The evaluation of such patterns often lies in the eyes of the beholder and personal preferences play a vital part. In this paper we propose a scientific approach to detect and evaluate the Head and shoulders (HS) pattern on data from the stocks on the Dow Jones Industrial Average and 30 stocks on the Nasdaq-100. We use wavelet based de-noising and an iterative process to select demands and parameters setting up the HS detection filter. The aim of this paper is to determine if there is any predictive power to the HS theory, and investigate the differences the pattern has experienced over time. We conclude that some relevance can be established for the HS theory but a strong time dependency is imminent. Data from 1960s and 1970s reveal a clear applicability regarding the HS pattern, while recent data, ranging from the 1980s up until today show a weak if any relevance of the technical analysis approach.
Keyword: Wavelet, Technical Analysis, Head and Shoulders