A High-Frequency Stock Price Prediction Method Based on Mode Decomposition and Deep Learning

Abstract

The modeling and prediction of stock prices is the core work in securities investment, and it is of enormous significance to reducing decision-making risks and improving investment returns. Existing research mainly focuses on mid or low-frequency stock price prediction, which is challenging to apply to intraday high-frequency trading scenarios. Meanwhile, the model accuracy face limitation due to the neglect of the influence of random noise and the refinement of the price sequence law. This paper proposes a high-frequency stock price prediction method based on mode decomposition and deep learning to improve intraday stock price prediction accuracy. Firstly, this method stabilizes the stock price series through empirical mode decomposition to tackle the issue of random noise interference. Then the convolutional neural network is introduced to extract the high-dimensional data features hidden in the stock price series by using multiple convolution kernels. Furthermore, the gated recurrent unit is used to process time-sequential data and to predict the stock prices at the minute level. The experimental result indicates that the proposed high-frequency stock price prediction method can achieve a significant forecasting effect, and its accuracy outperforms the existing methods.