Uncovering Turkish Audit Firms' Transparency Report Textual Attributes Through Computer Based Approaches and Linking Them to Audit Quality

Abstract

This paper assesses the textual attributes of transparency reports (TRs) prepared by audit firms (AFs) in Türkiye between 2013-2020. To achieve this goal, we utilize a range of computer-based techniques and expertise, including natural language processing, named entity recognition, business text processing, web scraping, data mining, and bot building. The findings reveal that TRs exhibit boilerplate tendencies, lacking full transparency. They are characterized as medium-to-hard documents with a generally positive tone and diverse EoBP at the firm level. Additionally, a customized Turkish readability scale derived from Fog index is introduced. This paper also investigates AF-based determinants of TR textual attributes. Regression model results indicate the impact of some AF characteristics (e.g., license duration, number of partners, the Big 4, female ratio, non-audit services) on most textual attributes. Finally, we explore the role of textual features, alongside AF and audit partner (AP) attributes, in influencing audit quality. To address potential endogeneity concerns, the study adopts an instrumental variables probit model. The analysis reveals that the readability of TRs by Fog index significantly influences audit quality. Factors such as AF tenure, AP busyness, CF's sales ratio, and loss contribute to audit quality, aligning with prior research.

Keywords: Audit Quality, Transparency Reports, Readability, Natural Language Processing, Business Text Processing, Named Entity Recognition, Document Similarity C26, C88, G18, M42