Forecasting stock market trends using support vector regression and perceptually important points

Abstract

Intelligent stock trading systems use soft computing techniques in order to make trading decisions in the stock market. However, the fluctuations of the stock price make it difficult for the trading system to discover the underlying trends. In order to enable the trading system for trend prediction, this paper suggests using perceptually important points as a turning point prediction framework. Perceptually important points are utilized as a high-level representation for the stock price time series to decompose the price into several segments of uptrends and downtrends and define a trading signal which is an indicator of the current trend. A support vector regression model is trained on this high-level data to make trading decisions based on predicted trading signal. The performance of the proposed trading system is compared with three other trading systems on five of the top performing stocks in Tehran Stock Exchange, and obtained results show a significant improvement.