Three essays on strategic disclosure detection and government financial reporting review: Harnessing machine learning, large language models, and intelligent process automation

This dissertation examines corporate disclosures and financial reports using emerging technologies to uncover insights from corporate disclosure and financial reports, overcoming challenges of conventional analytical methods facing unstructured textual data. The first essay (Chapter 2) investigates whether firms strategically misclassify negative material events into Item 8.01 ("Other Events") in Form 8-K filings to exploit investor cognitive constraints. Grounded in rational inattention theory, which suggests investors focus limited processing capacity on disclosures with lower processing costs, this study employs large language model (LLM) embeddings and machine learning techniques to detect and investigate misreporting behavior in Item 8.01. Findings reveal that firms strategically misclassify negative events, delaying investor reactions and impacting market transparency. These results underscore the need for enhanced regulatory scrutiny and offer practical insights to improve disclosure practices.The second essay (Chapter 3) explores firms’ strategic use of Environmental, Social, and Governance (ESG) disclosures in Form 8-K filings prior to negative earnings announcements. Anchored in impression management and legitimacy theories, this study investigates how firms, aware of impending poor financial results, strategically increase ESG disclosures to shape investor perceptions and mitigate adverse market reactions. Leveraging FinBERT, a specialized LLM in the financial context, to detect ESG-related disclosure, the empirical analysis shows a pronounced increase in ESG disclosures when firms anticipate negative earnings, particularly after quarter-end when outcomes are known internally. This strategic timing effectively softens negative market responses, highlighting ESG disclosures as legitimacy tools during periods of heightened information asymmetry. Thus, this essay contributes to understanding corporate communication strategies under adverse conditions.The third essay (Chapter 4) addresses the limitations of conventional robotic process automation (RPA) in processing unstructured financial text data. It introduces an intelligent automation framework integrating LLMs with RPA, demonstrated through a case study involving the Government Finance Officers Association (GFOA)’s annual review of government financial reports. The integrated framework significantly enhances accuracy, efficiency, and cost-effectiveness compared to traditional methods, demonstrating the potential of generalized intelligent automation applications. Overall, this dissertation demonstrates how advanced analytical techniques can leverage unstructured textual disclosures, offering practical insights and contributing significantly to financial reporting transparency and efficiency.