AI-Driven Textual Analysis of Chairperson Traits for Predicting Corporate Debt Policies: Evidence from Taiwanese Listed Firms

Abstract

This study compares traditional financial indicators with generative AI text analysis in an internet-embedded pipeline to examine how chairperson traits, including extraversion and overconfidence, shape corporate debt policy. Extraversion is quantified by classifying linguistic cues in the annual report’s “Letter to Shareholders.” A web crawler harvests filings and related public disclosures and normalizes HTML/PDF to text with versioned logs for auditability. Financial data for Taiwanese listed companies from 2004 to 2023 is collected from TEJ. Traditional determinants are controlled in dynamic target-adjustment regressions estimated with GLS cross-sectional weighting to mitigate heteroskedasticity. We find that extraverted or overconfident chairpersons tend to adopt optimistic debt policies by lowering leverage. However, as firms move toward target capital structure, overconfident or extraverted leaders pursue faster adjustments, suggesting that both measurement systems deliver consistent predictions. AI-driven textual analysis, continuously refreshed via the internet pipeline, partially substitutes traditional metrics by capturing subtle psychological and behavioral signals in near real time. Integrating AI-driven personality assessments and internet-based monitoring into governance can help investors, boards, and policymakers anticipate executive decisions, strengthen strategic forecasting, and improve financial stability and risk management.