432 – Advances in Graphical Frameworks and Methods Part 2
Discovering and Evaluating Trend Patterns in Financial Time Series
James Shine
James Gentle
George Mason University
Charles Perry
USDA (Retired)
Over the decades of stock price analysis, a number of heuristics have been developed. One such heuristic is the "head and shoulders" pattern; conventional wisdom has it that a large peak surrounded by two smaller peaks, at the end of a long upward trend, predicts a long downward trend afterwards. We developed a program to automatically find such patterns for evaluation. Simulated data indicates that given a sequence of high and low points, approximately 0.4% to 0.6% of these sequences can be classified as head and shoulders, with small differences depending on the underlying distribution. We present results from processing stock prices between 2007 and 2014 and analyze the empirical accuracy of the head and shoulders rule.