Firm-Level Tax Audits: A Generative AI-Based Measurement

Abstract

By using generative AI to analyze tax-audit-related narrative disclosures, we develop a novel measure of tax audit periods at the firm level. Empirically analyzing the firm-level economic effects of corporate tax audits has been challenging, primarily due to the confidential nature of audits. Our measure shows considerable within-firm variation and has high conformity with data released by the IRS. We extensively validate the measure and use it to answer several important questions. First, we document that firms reduce tax avoidance during tax audits and that this effect persists even after conclusion of the audits. Notably, firms with more pre-audit tax avoidance exhibit stronger effects. However, tax audits also lead to negative economic outcomes, such as reduced capital investments and increased stock volatility. We further find that the stock market responds negatively to tax-audit-related narrative disclosures, especially when firms discuss forecasts of unfavorable tax settlement. Overall these findings suggest that our measure can offer new insights on the economic impact of tax audits on firms.

Keywords: Tax audit, tax enforcement, IRS, tax avoidance, textual disclosure, large language models, GPT, generative AI, effective tax rate, investment, volatility

JEL Classification: C45, D80, G32, H25, H26, H71, M41, O16