【Abstract】As the largest source of public opinion, social media are believed to capture the “wisdom of the crowd”. Using information extracted from social media to predict social and economic activities—for example, stock market behavior— has become an important research topic. We study the relationship between daily stock return and social media sentiment using a dataset spanning five years from StockTwits. Contrary to several past studies, we find no evidence—despite the large power provided by the large dataset—that StockTwits sentiment has predictive power on daily stock return, thus calling for caution in interpreting the findings from past studies favoring the use of social media sentiment to predict stock returns. On the other hand, we find, for the first time, strong evidence that daily stock return predicts StockTwits sentiment. The effect of daily stock return on StockTwits sentiment peaks almost immediately (i.e., on the next day) and quickly wears out. The effect of daily stock return on negative sentiment is much stronger than that on positive sentiment.