This study investigates the influence of low-quality information, specifically fake news, on short-term stock returns, contributing to the discourse on signal theory and behavioral economics. Employing a machine learning approach, the authors developed a robust fake news dataset from publicly available sources. Sentiment analysis was conducted using a FinBERT model fine-tuned for stock market contexts, alongside extensive dataset cleaning through the NER-BERT model and conventional methods. Fixed-effects regression analysis was applied to assess the relationship between news volume and stock returns. The findings reveal that negative fake news significantly impacts stock returns, with its effects varying across firms and economic conditions. Notably, fake news has a stronger influence on relatively illiquid stocks and becomes more pronounced during periods of market uncertainty. These results underscore the importance of addressing misinformation in financial markets. This study offers originality in its approach by introducing a novel definition of fake news, addressing the lack of consensus in economic literature. It also identifies new categories of fake news previously unexplored in the field and provides the first large-scale econometric analysis of fake news effects on stock market performance. The findings hold critical implications for investors, regulators, and corporate management, emphasizing the necessity of countering misinformation to safeguard market stability and asset valuation integrity.
Федорова et al. (Fri,) studied this question.