Information Fusion in Financial Markets: A Hierarchical Framework for Combining Social Sentiment, Fundamental Valuation, and Technical Analysis

This thesis tests whether Reddit attention signals can predict stock returns around earnings announcements using 2.7 million posts matched to 115,000 earnings events for S&P 500 companies from 2010 to 2023. The two headline findings are: first, sudden spikes in Reddit discussion for heavily-followed stocks are followed by a significant price reversal of around 0.6%, consistent with retail-driven overvaluation correcting itself; and second, stocks with low Reddit activity before earnings show larger and more persistent post-announcement price drift than heavily-watched stocks, consistent with the investor inattention hypothesis — inattentive investors react slowly to news, while attentive ones price it in immediately. A machine learning model combining all signals generates modest positive returns in the test period, though the social media features alone have no predictive power without market structure variables, meaning the main contributions rest on the event study results rather than the ML model.