Do News and Sentiment play a role in Stock Price Prediction?

Despite continuous improvement in the range and quality of machine learning techniques, accurately predicting stock prices still remains as elusive as ever. We approach this problem using a modern autoregressive neural network architecture and incorporate sentiment predictors, which are becoming increasingly available due to advances in text mining techniques. We find that the inclusion of predictors based on counts of the number of news articles and twitter posts can significantly improve the quality of stock price predictions.