LAUSR

Predicting stock price based on the features of raw data has been a significant but challenging task for researchers. Various frequency components of the raw stock price series represent characteristics of stock prices in different time scales. Therefore, it makes sense for predicting stock prices to take these frequency components into account. In this paper, a novel hybrid model is proposed to predict stock prices, which combines empirical mode decomposition (EMD), convolutional neural network (CNN) and Long Short-Term Memory (LSTM). For this purpose, the original stock price series are first decomposed into a finite number of intrinsic mode functions (IMFs) under different frequencies by EMD. For each component, a CNN is used to extract the features. Then through a LSTM network, the temporal dependencies of all extracted features are modeled and the final predicted prices are obtained after a linear transformation. Two prediction steps, one day and one week, of Shanghai Stock Exchange Composite Index (SSE) are used to test the proposed model. The experimental results show that the hybrid network can achieve better performances by modeling different frequencies compared with other state-of-the-art models.