The Informational Content of Key Audit Matters: Evidence from Using Artificial Intelligence in Textual Analysis

Abstract

Key Audit Matters (KAMs) are informative for future accounting outcomes. Using FINBERT, a deep learning model for natural language processing that allows human-like text comprehension, we show that goodwill-related KAMs are predictive for firms' future impairments. In fact, we find that using KAMs as a stand-alone predictor for future impairments significantly outperforms a random classifier. Delving deeper into the semantic meaning of reported KAMs, we find that their predictive power is driven by text passages that elaborate how the firm and the auditor exercised their judgement in respect to the accounting and auditing of goodwill. Further analyses indicate that the informational content of KAMs is also incrementally predictive beyond key firm-level determinants of impairments identified in prior studies. Taken together, our findings contribute to the overall understanding of the informational content of KAMs, a key rationale for their introduction.

Keywords: audit reporting; key audit matters; prediction; natural language processing; FINBERT; goodwill impairment

JEL Classification: M41, M42, C45, G32, M48