The informational content of key audit matters: Evidence from using artificial intelligence in textual analysis

Abstract

This study provides empirical evidence that key audit matters (KAMs) are informative for future negative accounting outcomes. We employ FinBERT—a deep learning model designed for natural language processing that allows human-like text comprehension—to demonstrate that goodwill-related KAMs are predictive of firms' future impairments. Our findings reveal that utilizing KAMs as a stand-alone predictor for future impairments provides meaningful predictive power. By exploring the semantic content of reported KAMs, we find that their predictive power is primarily driven by text passages covering how both the firm and the auditor exercise judgment in the accounting and auditing of goodwill. Furthermore, we show that KAMs are incrementally predictive beyond several firm-level determinants and disclosures in annual reports. Finally, our additional analyses indicate that (1) KAM-predicted impairment probabilities are relevant to capital markets, (2) KAMs are useful for predicting the magnitude of goodwill impairments, and (3) the predictive power extends to other KAM topics. Collectively, our findings enhance the understanding of the informational content of KAMs, which is a key rationale for their introduction.