Using Narrative Disclosures to Predict Tax Outcomes

Abstract

We examine whether the narrative discussion in financial disclosures can help corporate stakeholders better predict tax outcomes. To measure qualitative discussion, we map the thematic content of 10-K disclosures estimated with topic modeling analysis to the tax planning determinants framework of Wilde and Wilson (2018). We find that qualitative discussion in financial disclosures can substantially improve the prediction of tax outcomes in in- and out-of-sample tests. We further find that prediction-relevant discussion is distributed throughout the 10-K, supporting that disclosures should be analyzed holistically rather than examining only limited pieces of larger disclosures. We also find that analysts struggle to use this information effectively, resulting in substantial and predictable forecast errors. These findings empirically illustrate the wealth of qualitative information in 10-K disclosures for stakeholders concerned about tax outcomes and offer a practical approach to examining qualitative disclosures and using them to predict tax outcomes.

Keywords: tax outcomes, tax planning, textual analysis, topic modeling, forecasting

JEL Classification: G39, H25, H26, M21, M41