Measuring Twitter Sentiment Impact on Stock Prices of a Company

In recent years, social media services have become one of the most valuable sources of information for conducting social, political and economic researches and for purposes of marketing and investing. In finance social media, especially Twitter, can be used through the lens of behavioral finance: sentiment and emotions of social media users can be measured in real time and can be used as a proxy for a sentiment of financial market participants so that herding effect and panics can be considered when creating investment strategies beforehand. In this paper, we analyze the influence of sentiment in Twitter on the dynamics of stock prices movements for 22 companies from different industries – IT, telecommunications, automobile and other. To analyze the effect, first, the event study was conducted and then predictive models were built (both for individual companies and for groups of companies based on their industry). Predictive models were built on logistic regression and model selection techniques were applied for finding the optimal specifications of the models. As a result, it was discovered that the prediction accuracy increases by 1-2% for individual company if Twitter sentiment is included, and the achieved average accuracy among 22 companies was 83.5%, which is 0.4% higher than accuracy of models without the Twitter sentiment. Besides that, it was concluded that accuracy of predictive models applied on group of companies is consistently increased for companies for IT and Telecommunications industries, while for others accuracy either did not change, or in case of financial industry accuracy even decreased.