Social Media and Forecasting Stock Price Change

Abstract

The Stock Market is a big influence on both national and international economies. Stock prices are driven by a number of factors: industry performance, company news and performance, investor confidence, micro and macro economic factors like employment rates, wage rates, etc. Stock pricing trends can be gauged from the factors that drive it as well as from the stock's historical performance. As fluctuations in stock prices become more volatile and unpredictable, forecasting models help reduce some of the randomness involved in investing and financial decision making. Users on social media platforms like twitter, StockTwits, and eToro discuss issues related to the stock market. Can the analysis of posts on StockTwits add value to the existing features of stock price predicting models? An existing model that uses twits as features was extended to include sentiment analysis of the text referenced by the URL in the twits to see if the model accuracy did improve. Initial results indicate that the addition of sentiment analysis of the text referenced by the URL does not improve the performance of the model when all twits for a given day are taken into account since the model only identifies the direction of change and not the degree of change. The stock prediction model achieves 65% accuracy compared to the base case accuracy of 44% and augmenting the model with sentiment analysis did not change the accuracy. The study highlights some interesting observations regarding users on the StockTwits social media platform and proposes the need for a domain specific sentiment analyzer in future work.