StockTwits: Comprehensive records of a financial social media platform from 2008 to 2022

Abstract

This paper describes and characterizes the relationship between social media and financial markets, focusing on two key questions: (1) Do users of financial social media perform better than guessing at predicting stock movements? (2) When do users in the aggregate perform better than guessing at the same task? To study these questions, we introduce the first publicly available, comprehensive dataset of posts on a social media platform: StockTwits. StockTwits is a financial social media platform where more than 7 million active users have discussed financial markets and investing strategies across over 550 million posts since 2008. We provide a complete record of all StockTwits posts up to 2022, including the poster's anonymous ID, the text and timestamp of the message, and whether the user tagged their post as optimistic ("bullish") or pessimistic ("bearish"). We study the temporal dynamics of this dataset, analyzing it at both the ticker level and the user level. First, we find that while most users perform approximately at a guessing rate, a significant percentage consistently perform statistically better or worse than guessing, especially at longer timescales. Second, corroborating existing literature, we observe that attention — as measured by the number of messages mentioning a stock — is generally less predictive than sentiment. However, there is meaningful variation across securities and over time. Some stocks are best predicted by attention, others by sentiment, and some stocks by each at different times.