Do Actions Speak Louder than Words? Evidence from Microblogs

Abstract

This research identifies the determinants of investors' future beliefs by analyzing more than 50 million tweets on thousands of stocks from the microblogging platform StockTwits. To distinguish between the different sources of changes in beliefs, I divide tweets into two categories: beliefs, representing the average sentiment of all investors regarding a particular stock, and actions, representing the actual transactions disclosed by StockTwits' users in their tweets. The results show that investors’ average next-period beliefs are positively impacted by the average sentiment (beliefs and actions) of the previous day. The effect is stronger once the quality of the investment advice is taken into account. Finally, more communication between investors is associated with greater diversity in beliefs as well as higher uncertainty.

Keywords: Sentiment analysis, NLP, Text analysis, Microblogs, Text classification, Herding, Word-of-mouth

JEL Classification: G14, G40