Text Representation for Direction Prediction of Share Market

Financial news articles are textual sources that report on economic events which affect the stock market. News articles published when the stock market is open, influence the movement of the stock market indices. An implicit relationship exists between different economic events that eventually influence the stock market movement. This paper studies the relationship between news articles and the direction of close price movement of the stock market. It explores the effectiveness of different statistical and deep learning-based classifiers that use price features or text features as input to predict the direction of close price movement. Further, this paper proposes a parallel CNN model which combines text representation and price representation to predict the direction of close price movement. The proposed model gives high accuracy, and the normalized cross-correlation scores show that the predicted and actual time series are strongly correlated. These metrics indicate that economic events in news articles were effectively harnessed to predict the close price movement.