A New Convolution Neural Network Model for Stock Price Prediction

Abstract

The stock market is a highly nonlinear dynamic system, not only stock prices have a certain tendency, but also it is influenced by many factors such as political, economic and psychological factors. With the flourishing development of deep learning technique, a well-designed neural network can accomplish feature learning tasks more effectively. For the task of feature extraction and price movement prediction of financial time series, a novel convolutional neural network framework is proposed to enhance the accuracy of prediction. The proposed method is named stock sequence array convolutional neural network (SSACNN). It constructed a sequence array for the historical data and applies this array to be an input image for the proposed CNN framework. There are five Taiwanese stocks as a testing benchmark in the experimental results. SSACNN compared with previous algorithms, the performance of movement prediction is improved obviously.