Abstract: The issue of information content of critical audit matters (CAM) is a cutting-edge issue that has received considerable attention from the auditing community. Using Shanghai and Shenzhen A-share listed companies from 2017 to 2021 as the research sample, this paper empirically investigates the relationship between signed auditor narcissism and information content of key audit matters by applying natural language processing techniques in the field of machine learning. It is found that narcissistic auditors can effectively improve the information content of key audit matter disclosures, and it is mainly project auditor narcissism not review auditor narcissism that helps improve the information content. The phenomenon is more pronounced for impairment-type matters and in the presence of audit fee premiums. In addition, project auditor narcissism facilitates CAM to provide more ongoing incremental information and the increased information content makes the text more difficult to understand. These findings hold after robustness tests, one of which eliminates the effect of a third auditor. The findings not only enrich the research related to the information content of key audit matters from the perspective of auditor psychological characteristics, but also provide reference for firms to formulate business development strategies and regulators to further regulate the disclosure of qualitative content of audit reports.

Key words: critical audit matters, information content, auditor narcissism, text similarity