Wisdom or Whims? Decoding Investor Trading Strategies with Large Language Models

This paper analyzes the trading strategies used by retail investors. By applying large-language models (LLMs) to over 100 million messages posted by nearly 800,000 users on a popular social media platform, we are able to obtain sharp inferences about investor strategies. Messages describing technical analysis strategies tend to increase in periods without fundamental news and technical message sentiment negatively predicts future returns, is associated with less informative retail order flows, and a higher likelihood of Robinhood herding episodes. In contrast, messages describing fundamental analysis exhibit sentiment that is informative of future returns. Furthermore, the highly profitable technical signal extracted using machine learning image analysis techniques (Jiang et al. 2023) generates even stronger profits when trading against retail technical sentiment and the profit disappears when the signal coincides with retail technical sentiment. The findings suggest that retail investors tend to misinterpret technical analysis, but tend to use fundamental analysis correctly. The evidence provides insights into the diverse investment approaches traders use and the interaction of different players in the age of AI-powered trading.

Social media Retail investors ChatGPT BERT Technical analysis Fundamental analysis