Using Narrative Disclosures to Detect Financial Fraud

Abstract

This dissertation measures the information content in narrative financial disclosures to identify linguistic differences in manager and analyst language when fraud versus when it is not. The first chapter describes the motivation for this research and an overview of the research domain. Next, I review the literature covering textual analysis of narrative disclosures and present a heuristic and classification scheme for studies in this context. In Chapter 3, I compare the language across two common narrative disclosure types: quarterly earnings calls and the Management’s Discussion & Analysis (MD&A) section of quarterly and annual financial statements and find evidence of restricted incremental information from the CFOs of fraudulent companies. Chapter 4 uses a quasi-experiment to compare analyst the frequency and topics of analysts’ question during earnings calls. I find that relative to nonfraudulent firms, analysts ask the managers of fraudulent firms more questions overall, and are more persistent in asking questions as a call progresses. Chapter 5 is an exploratory study of dominance and linguistic style matching from managers and analysts when interacting in the question-and-answer portion of an earnings call. The dissertation concludes with a discussion of the work, limitations, and avenues for future research.