ESSAYS IN AI AND FINANCE

Abstract

This dissertation comprises three essays that examine the role of emerging technologies, information disclosure, and risk in financial markets. The first essay (Chapter 1) investigates how vocal cues embedded in social media disclosures influence retail investor behavior. The second essay (Chapter 2) examines whether firms’ discussions of artificial intelligence (AI) in earnings conference calls affect market valuation and subsequent firm performance. The third essay (Chapter 3) explores the effects of cyber risks on bank lending, borrower outcomes, and regulatory responses. The increasing adoption of generative artificial intelligence (Gen-AI) has made information environments increasingly multimodal, highlighting the importance of understanding how investors process information beyond textual content. In Chapter 1, I investigate the impact of vocal cues from social media on retail trading activity. Using audio recordings of conference calls disseminated through Seeking Alpha, I examine how vocal emotions and listenability, in addition to textual sentiment, influence future retail stock and option trading. I find that vocal emotions possess predictive power for retail trading beyond that captured by textual sentiment, suggesting that vocal cues convey unique information to investors. Furthermore, the influence of vocal emotions varies across emotion types, indicating that certain emotions, such as surprise, cannot be readily interpreted as purely positive or negative signals. I also document a disagreement channel through which vocal emotion and listenability affect retail investor disagreement, with the magnitude and direction of the effects differing across emotional categories. These findings contribute to the growing literature on social media and investor behavior by demonstrating that auditory information plays an important role in shaping retail trading decisions. Artificial intelligence has emerged as a transformative technology with the potential to reshape firm operations, decision-making processes, and competitive positioning. In Chapter 2, I examine whether capital markets value firms’ AI-related discussions and whether such discussions are associated with subsequent improvements in firm performance. Drawing on signaling theory, I develop a BERT-based measure of AI mentions using earnings conference call transcripts and analyze their effects on market valuation and operating performance. I find that firms that frequently discuss AI receive higher market valuations; however, these valuation premiums are not accompanied by corresponding improvements in financial performance, and the valuation-performance gap persists for at least eight quarters. The valuation effects are stronger among firms facing greater competitive pressure, suggesting that investors place greater value on AI-related signals when firms face stronger incentives to innovate. Additional analyses indicate that while investors interpret AI mentions as positive strategic signals, they also rely on realized firm performance to evaluate the credibility of such signals. Overall, this essay contributes to the literature on strategic signaling by documenting a systematic relationship between AI-related disclosures, investor expectations, and realized firm outcomes. Cybersecurity has become a critical concern for financial institutions and regulators, yet its implications for bank lending remain incompletely understood. These concerns have intensified with the rise of AI, which lowers the cost and increases the sophistication of cyberattacks. In Chapter 3, I investigate how cyber risks affect banks and their lending activities. Using a comprehensive dataset of cyberattack events obtained from RepRisk, I document significant value destruction for both attacked banks and their connected borrowers. Specifically, cyberattacks are associated with negative stock market reactions for banks as well as for borrowers that obtained syndicated loans from the affected banks during the preceding three years. I further find that cyberattacks reduce subsequent lending amounts, indicating lasting consequences for credit supply. In contrast, measures of ex ante cyber risk derived from a BERT-based textual analysis do not exhibit a significant relationship with lending outcomes at either the aggregate or loan level. Extending the analysis to New York’s cybersecurity regulatory framework, I show that affected banks become more attentive to borrowers’ cyber risks when determining loan spreads. These findings contribute to a deeper understanding of the interactions among cyber risks, banking activities, and regulatory responses in modern financial markets.