Predicting Stock Price Movements Based on Different Categories of News Articles

Publications of financial news articles impact the decisions made by investors and, therefore, change the market state. It makes them an important source of data for financial predictions. Forecasting models based on information derived from news have been recently developed and researched. However, the advantages of combining different categories of news articles have not been investigated. This research paper studies how the results of financial forecasting can be improved when news articles with different levels of relevance to the target stock are used simultaneously. Integration of information extracted from five categories of news articles partitioned by sectors and industries is performed using the multiple kernel learning technique for predicting price movements. News articles are divided into the five categories of relevance to a targeted stock, its sub industry, industry, group industry and sector while separate kernels are employed to analyze each one. The experimental results show that the simultaneous usage of five news categories improves the prediction performance in comparison with methods based on a lower number of news categories.