Multi-Category Events Driven Stock Price Trends Prediction

Previous studies have shown that news events are relevant to stock price movements. We propose a novel multi-category events driven stock price trends prediction model, which uses the historical price data of China's A-share market and the stock-related news of listed companies as training set. We extract several categories of news events that are related to investment, including performance pre-release event, executive increasing or decreasing of shareholdings event, stock dividends event, employee stock option plan event, etc. Both SVM models and neural networks are used with event features. Empirical experimentation shows that our method outperforms classic bag-of-words features model, and neural networks outperforms SVM-based model.