Evolutionary computation with multi-variates hybrid multi-order fuzzy time series for stock forecasting
Financial time series forecasting has attracted substantial attention among data mining community for many years. However, achieving a reasonable accuracy in forecasting is a difficult task and extremely challenging for the researchers. Investors often use technical indicators to analyze stock market tendency and make decisions. In this paper, a new method called RPRS for stock forecasting is presented based on combing multi-variates hybrid multi-order fuzzy time series with the genetic algorithm. In our approach, technical indicators such as ROC, PSY, RSI and STOD are used as the dependent variables to improve performance. RPRS applies hybrid multi-order fuzzy time series (1-order, 2-order and 3-order) to forecast future stock prices and uses the genetic algorithm to search for a good domain partition. In order to evaluate the performance of RPRS, univariate fuzzy time series models and three classic fuzzy time series models are selected for comparison based on TAIEX, HSI and NASDAQ data. Experiment results show that RPRS performs better than other models.