Predicting Stock Price Movement Using Financial News Sentiment

A central question in financial economics concerns the degree of informational efficiency. Violations of informational efficiency represent capital miss-allocations and potentially profitable trading opportunities. Market efficiency analyses have evolved to incorporate increasingly rich public information and innovative statistical methods to analyze this information. We propose an Automatic Crawling and Prediction System (ACPS) to 1) automatically crawl online media, 2) extract useful information from a rich set of financial news, and 3) predict future stock price movements. ACPS consists of a feature selection pipeline to select an optimal set of predictive features and a sentiment analysis model to measure sentence-level news sentiment. Generated features and news sentiment data are further processed via an ensemble model based on several machine learning and deep learning algorithms to generate forecasts. Results demonstrate the robustness of our proposed model in predicting the directional movement of daily stock prices. Specifically, the model consistently outperforms existing methods on single stock prediction and it performs well across all S&P 500 stocks. Our results indicate the potential value of rich text analysis and ensemble learning methods in a real-time trading context.