Wisdom or Whims? Decoding the Language of Retail Trading with Social Media and AI
Abstract
We leverage rich social media data and large language models (LLMs) to examine the relationship between investor trading strategies, sentiment, and market outcomes. Extracting trading strategies embedded in 96 million social media posts, we find that strategy adoption is heterogeneous and dynamic, with substantial differences in performance outcomes. Our results show that news arrivals decrease users' reliance on technical signals and increase their utilization of fundamental signals. Technical sentiment negatively predicts stock returns, particularly among short-term or inexperienced users, whereas fundamental sentiment positively forecasts returns. Additionally, message sentiment correlates positively with aggregate retail buying, with technical sentiment strongly associated with aggressive buying by Robinhood investors. Our study demonstrates the promise of using AI to understand investor behaviors and their implications for market dynamics.