WCSE

Stock Trends Prediction Using the Feature Pattern Constructed with the Panel Data of Asian Stock Markets

Abstract— The artificial intelligence has been widely used in many different industries for many years. In particular, the deep learning approaches have profoundly improved the traditional banking services and generated a great impact on the financial industry. By using the artificial intelligence technology, the financial advisors will be able to explore various data sources, and the meaningful information can be discovered and analyzed more efficiently. In this study, both the conventional machine learning and current deep learning methods were applied to predict the trend direction of the weighted stock price index of Taiwan based on the panel data from associated Asian stock markets. A novel feature pattern constructed with the spatial and temporal correlation of the stock markets was used as the input to train neural network. The arrangement of two dimensional pattern for input data might facilitate the convolutional neural network to explore the embedded historical trend for stock index. Three different experiments were conducted to evaluate the prediction performance of models. The experimental results showed that the convolutional neural network built with 6 and 12 distinct kernels for convolution layers outperformed the other models, and achieved the highest prediction accuracy. Although the dimensionality of original input variables was increased, the result indicated the transformation of the input variables into a two dimensional feature pattern might be helpful for the deep learning approaches to learn the historical trend pattern.

Index Terms— feature pattern, machine learning, convolutional neural network

Tsung-Nan Chou
Department of Finance, Chaoyang University of Technology, CHINA

ISBN: 978-981-11-3671-9 DOI: 10.18178/wcse.2017.06.17Xsrc="http://www.wcse.org/uploadfile/2019/0823/20190823055609629.png" style="width: 120px; height: 68px;" />[Download]

Cite: Tsung-Nan Chou, "Stock Trends Prediction Using the Feature Pattern Constructed with the Panel Data of Asian Stock Markets," Proceedings of 2017 the 7th International Workshop on Computer Science and Engineering, pp. 1296-1301, Beijing, 25-27 June, 2017.