An analysis of combining correlation screening with artificial neural network for FITX futures prediction

Abstract

FITX futures provide investors with tools to evade risks and serve as an investor commodity; in other words, speculators and arbitragers make use of the high-level feature of futures to earn enormous returns with a small sum of money. FITX price is easily influenced by various factors, including finance, economy, politics, society, and investors’ mind. In this study, some features of the FITX futures market and the back propagation network (BPN) are used to predict FITX futures. The BPN is adopted to predict the next-day closing prices of FITX futures. In this paper, the variables of the original data regarding price, quantity, and time, as well as the fluctuation estimated by GARCH, were taken as the input variables. The R software was used to establish the BPN. The next-day closing indices of FITX futures were predicted after training, as based on the back propagation model. In addition, the coefficient of correlation between the input variables and the closing prices is used to select the optimal input variable to establish the BPN model. According to the empirical research results, the proposed methods are more accurate in prediction in comparison with the original BPN model.