Research on Multivariate Financial Time-Series Forecasting Considering Endogenous Variables

Abstract

The ensemble learning method is outstanding in the prediction accuracy of composite stock index and foreign exchange time-series. Still, it is slightly worse in the prediction accuracy of multivariate financial time-series when considering endogenous variables. This paper proposes a new method to identify similar data patterns and cluster them into groups by Fuzzy C-Means (FCM). The results of an example prove that the improved FCM clustering algorithm and the improved Elman network can be used to classify and forecast the stock data of Shanghai and Shenzhen A-stock markets.