Sentiment analysis of corporate disclosures with generative AI
Virginia Sotiropoulou1 and Dimitris Tzelepis2
1Postdoctoral researcher, Accounting Lab, School of Social Sciences, Hellenic Open University, Greece
2Professor, Department of Economics, University of Patras and Adjunct lecturer, Hellenic Open University, Greece
Correspondence: Virginia Sotiropoulou, Accounting Lab, School of Social Sciences, Hellenic Open University, Greece. E-mail: virginia.sotiropoulou@ac.eap.gr
Abstract
Purpose – The purpose of this study is to examine the association of textual attributes of U.S. annual reports, identified using a generative AI tool, with future firm performance. This research work attempts to investigate the adoption of a generative AI approach in the process of textual analysis of annual reports of U.S. firms to capture the information content of their sentiment regarding future firm performance. A statistically significant association reveals that textual characteristics captured with generative AI are informative regarding firm performance. Thus, the main research question of this study is formulated as: “Do textual attributes of annual reports, extracted with a generative AI tool, have a significant relation with future firm performance?”
Design/methodology/approach – First, an algorithmic approach in Python programming language is used for retrieving the U.S. annual reports (10-Ks) from SEC’s Electronic Data Gathering, Analysis, and Retrieval system (EDGAR). Then, the cleaning process is implemented using again an algorithm in Python programming language. When this procedure is completed, the generative artificial intelligence approach is employed to analyze textual information. The database created after the implementation of textual analysis of annual reports is merged with the database with financial data. Stata software is used for the regression analyses of the final dataset.
Findings – The empirical findings of this study indicate that the generative AI approach can efficiently capture the information content of corporate disclosures regarding future firm performance. Therefore, the results indicate that the generative AI tool can be successfully implemented to measure the attributes of narratives of corporate documents, such as annual reports.
Research limitations/implications – This study examines exclusively the association of textual attributes of annual reports with firm profitability without shedding light to other aspects of firm performance.
Practical implications – The investigation of the association of textual attributes of U.S. annual reports, identified using a generative AI tool, with future firm performance aims to equip with deeper comprehension the users of accounting information - academics, managers, investors, analysts, and auditors.
Originality/value – To our knowledge this is the first research effort that analyzes this research question. In specific, after extensive literature review, we find that this is the first study that investigates the association of textual attributes of U.S. annual reports, identified using a generative AI tool, with future firm performance. It adopts a comprehensive textual analysis framework, using an innovative method to capture the sentiment of narratives of 10-Ks. Future research may expand the sources of financial information beyond annual reports.