Social Network Financial Sentiment: Constructing Proxies and Testing Returns Predictability on S&P500 Futures Returns

Abstract

This article examines the ability of StockTwits social network sentiment proxies to predict S&P500 Futures. Positive and negative levels and first-difference sentiment proxies were constructed from 59,907,378 tweets. Using the lexicon approach and Loughran-McDonald positive and negative word lists, this study considers the tweets’ informal language and 140-character constraint. It was found that one standard deviation of change in negative word sentiment compared to the previous day predicts lower S&P500 Futures by 3.4 basis points after controlling for past returns and macroeconomic variables. The results are robust to macro announcements, futures turnover, major Asian and European market returns, the day-of-the-week effect, the January effect, and the holiday effect. Investors can easily replicate the methodology to construct the social network sentiment proxies introduced in this study and employ these proxies in their investment strategies. This study hopes to spur more research to construct and improve social network sentiment proxies for various financial markets.