[Topic]: Financial Big Data Asset Price Forecasting: From Artificial Intelligence perspective

[Time] March 29, 2019 (Friday) from 3-5 p.m

[Location] Meeting Room 1620 in the back main building

[Speaker] Jiang Fuwei, Associate Professor, School of Finance, Central University of Finance and Economics

Abstract

This paper is the first to use a comprehensive machine learning approach to forecast asset prices based on big data in the Chinese stock market. We use hundreds of predictive variables related to company characteristics to build a large data set, and use a variety of machine learning methods to analyze the parameters of listed companies and asset return prediction. These machine learning methods include principal component regression, partial least squares, ridge regression, lasso regression, elastic networks, support vector regression, random forest, and neural networks. The results show that financial big data containing company characteristics can effectively predict future asset prices. Compared with methods based on linear assumptions, nonlinear or nonparametric methods (support vector regression, random forest, neural networks) have smaller mean square prediction errors, and the predictive ability is less affected by a single variable. Among them, support vector regression and one layer neural network method have the strongest out-of-sample prediction ability. The forecasting errors of different methods will change with the economic cycle and market state. For different company characteristics, the trade friction index and industry mean index play a significant role in improving the prediction accuracy. In the study of cross-sectional stock returns, we find that the long-short hedge portfolios constructed by most methods have annualized returns of more than 15% and Sharpe ratio of more than 0.70. Among them, support vector regression and one-layer neural network method can predict cross-sectional stock returns most effectively, and their excess returns measured by the latest five-factor model are significant at the 1% level. The results also show that because of the various noises contained in financial big data, the more complex the effective prediction model is, the better.

Key words: Machine learning; Financial big data; Asset price prediction; Chinese stock market

About the speaker

Jiang Fuwei is an associate professor at the School of Finance, Central University of Finance and Economics, and a "Longma Scholar" young scholar. He has a PhD in finance from the Lee Kong Chian School of Business, Singapore Management University. His research interests include asset pricing (return forecasting, market anomalies, investment management), behavioral finance (investment sentiment, manager sentiment, limited arbitrage), financial big data and artificial intelligence (machine learning, deep learning, text analysis), and Chinese capital market. He has presided over many projects of National Natural Science Foundation of China, Beijing Natural Science Foundation, etc. His research results are published in Journal of Financial Economics, Review of Financial Studies, Journal of Banking and Finance, Journal of International Money and Finance and other International authoritative journals.