Twitter based sentiment effect on stock market returns

Sammanfattning: The purpose of this study is to investigate the effect of Twitter based sentiment on stock market returns. We use a uniquely large dataset of 95 million tweets concerning the 102 biggest US companies, S&P 500 index and other market keywords for the year 2017. Correspondingly, the dataset also includes the returns from our sample companies and the S&P 500 index. The methodology used in this study comprises of several steps, but above all, 1) we use two dictionary based sentiment analysis tools for converting the tweets to sentiment scores, and 2) we use an autoregressive distributed lag (ADL) model for the empirical analysis. Our contribution to the literature can be summarized with the following four key findings: a) the newer data confirmed that Twitter sentiment, especially the bullishness index, has predictive power of S&P 500 returns; b) Vader NLTK sentiment analysis outperforms the traditionally used Loughran and McDonald Lexicon-based method; c) some industry returns have higher sensitivity to sentiment, while predictive power was only found for the IT industry, up to 2 days ahead; and d) the conventional opinion that retail investors are more affected by sentiment is not confirmed, on the contrary, we find that sentiment has more predictive power for companies with a high institutional investor share.