Extracting Sentiment of Selected Twitter Accounts and Considering Its Relationship with the S&P 500 Index

Abstract

Twitter is a source of streaming data. In this thesis, we examine whether and to what extent we can find a relationship between the sentiment of selected Twitter accounts and the S&P 500 index. This thesis uses data from 18 most-followed Twitter accounts and 20 accounts of those who tweet about financial markets in 50 months from January 2017 to March 2021. The sample period encompasses about 1.1 million uncleaned tweets from most-followed accounts and 0.6 million tweets from traders’ accounts. We find that the Granger causality between the most-followed accounts sentiment and S&P suggests that while the most-followed accounts sentiment Granger causes the S&P 500, the S&P 500 Granger causes the traders sentiment. Also, we find a significant long-run effect of the net positivity first difference on the S&P 500 index first difference, which is intensified after replacing the most-followed accounts sentiment with the traders’ sentiment. Our results show that using an error correction time series model; it is possible to explain 62 to 64 percent of the variation in the first difference of the S&P 500 index by the first difference of the net positivity index and the lagged values of two indices. Finally, we examine the possibility of the predictability power of the sentiment index added to a model consisting of topic probabilities as explanatory variables on the S&P 500 index.