A DFS Model for Forecasting Stock Price

Abstract

Currently, forecasting stock price is the hotter topic for achieving the smallest lost in investment. However, the previous stock price forecasting model practically cannot satisfy the requirement of accuracy. To raise the forecast accuracy, a decomposition-forecast- synthesis (DFS) model is proposed by this paper, based on the analysis of the characteristics of the stock price time series, combined with the established single stock price prediction model, for instance, time series model, grey prediction model, neural network prediction model, etc. DFS model decomposes a stock price time series into three components, including time tend component, quasi periodic component and random component. For each component, an adaptive prediction model is adopted to predict, afterwards, the synthesis of three component is to acquire stock price forecast sequence. The wavelet analysis, combination forecast method, Fourier Transform, fitting analysis, and conventional time series models, for instance, the ARMA (Autoregressive-moving-average) method and exponent smoothness method are adopted in the DFS model. For evaluation the forecasting performance of proposed model, the daily stock prices of SAIC Motor from December 2, 2013 to March 2, 2016 are used as experimental dataset and the Mean Square Error (MSE) and Mean Square Coefficient of Variation (MCV) as evaluation criterion.