Price-Pattern Recognition Using a Local Polynomial
Regression

Abstract

Technical analysis, also known as “charting,” has received con - siderable attention for many decades. Much of the work of technical analysts has relied on the ability to recognize patterns when displayed pictorially. Some researchers have investigated the use of nonparametric methods to identify price patterns. Subjective bandwidth selection and boundary problems are two main issues when applying the Nadaraya–Watson kernel estimator to pattern recognition. To fill the gap, we propose a complete data-driven technical analysis algorithm with the application of a nonparametric local linear estimator. We incorporate trading volume together with trading price to define the patterns. Empirical implementation on S&P 500 Index stocks indicates that the proposed algorithm is very informative and promising.