Stock Price Forecasting Using Support Vector Regression Based on Network Behavior Data

Stock market research based on network behavior data has become one of the focuses in behavioral finance. In this paper, we firstly construct a proxy variable of investors' attention based on comments data collected from the Snowball Finance, which is a popular online financial community in China. Then we analyze the lead-lag relationship between investors' attention and the stock price of all A-shares listed in both the Shanghai Stock Exchange and Shenzhen Stock Exchange, using Thermal Optimal Path (TOP) method. And we find that investors' attention and stock price have a dynamic relationship, which differs from stock to stock. In terms of quantity, only a small number of stocks have a relationship where investors' attention changes ahead of the stock price. Those two facts may account for conflicting conclusions drew by different studies. Further on, this paper establishes two support vector regression models, comparing the predictive capability of investors' attention to the selected two kinds of stocks. The results show that adding investors' attention to models can only enhance the prediction precision of the "leading stocks", while it has little effect to the "lagging stocks".