From public sentiment to stock returns: ESG discourse and financial materiality in social media

Abstract

This thesis pioneers the application of a sarcasm-adjusted sentiment pipeline for Environmental, Social, and Governance (ESG) classification in Reddit-based event studies. This exploratory study examined the impact of ESG-related public perceptions on stock market price reactions. Reddit discussions in 2024 about 35 listed U.S. companies were extracted and preprocessed using four transformer-based Natural Language Processing (NLP) models. The methodology combines event-based regression, designed to account for the short-term market response to spikes in ESG sentiment, with fixed-effects panel regressions that assess the ongoing impact of daily sentiment on excess returns. Together, these approaches enable the analysis of both individual spikes and the underlying dynamics of a change in public sentiment. Despite the application of extensive data processing techniques and advanced machine learning tools, the sentiment indicators did not achieve statistical significance, suggesting that Reddit sentiment may not contain predictive ESG signals. Although in most regression models null results were obtained, the social pillar proved to be the only dimension consistently approaching significance, partially aligning with Hypothesis 3. This study develops a framework addressing sarcasm, bot activity, and financial materiality in sentiment analysis. It also highlights current limitations and identifies new directions for future research in ESG financial management. The study highlights the need for platform-specific NLP models and more nuanced contextual labeling to bridge the gap between online discourse and financial relevance. The study suggests that investor decisions are still more strongly guided by structured ESG disclosures than by informal crowd sentiment, even when that sentiment is emotionally charged or widespread.