Using Natural Language Processing Approach to Extract Risk Profile from 10-K Filings: Evidence from Audit Opinions and Internal Control

• In Event: Auditing (Auditing and Information Quality)

Abstract

The objective of this study is to introduce an innovative text mining approach to examine the relationship between corporate textual risk disclosures in 10-K filings and audit opinions and internal control over financial reporting. Using Natural Language Processing (NLP) techniques, we develop a novel firm-specific risk measure based on the COSO’s (Committee of Sponsoring Organizations) risk management framework in four categories: financial, strategic, operational, and hazard risks. We find that auditors are more likely to issue modified audit opinion and identify material weaknesses in internal control when firms’ self-disclosed risks increase. We further find such a positive relation is mainly driven by financial and strategic risks. In contrast to prior studies documenting that 10-K filings are more likely to be boilerplate, our findings suggest that corporate textual risk disclosures provide additional and valuable information to external users of financial statements, such as auditors, financial analysts, investors, and regulators.

Authors

• Manlu Liu, Rochester Institute of Technology

• Kean Wu, Rochester Institute of Technology

• Rong Yang, Rochester Institute of Technology

• Yang Yu, Rochester Institute of Technology