Financial Time Series Forecasting Model Based on EMD and Rolling Grey Model

Over the last few decades, the traditional grey model (GM) has been proved to be able to forecast the financial time series. However, there still remains space for improvement in terms of prediction accuracy, especially when the time series is of high complexity and volatility. In this paper, we leverage the rolling grey model (RGM) and the empirical decomposition (EMD) method to enhance the prediction ability. With the strong ability of decomposing and smoothing non-linear and non-stationary time sequences, the EMD method can be served as the preliminary data processing step for the entire prediction procedure. By attaching more importance to the last several days' data and constantly updating the coefficient of the GM function, the RGM method can better follow the market change and avoid error accumulation. Our proposed approach greatly improves the vanilla GM method and further outperforms the RGM method. In contrast to the RGM method, our improved method reduced 7.7% on ARPE and 7.9% on MAE when predicting the 2010-2019 Shanghai Stock Exchange Index. When forecasting the 2010-2019 China Securities Index, this method also reduced 6.9% on ARPE and 7.3% on MAE.