We examine how narrative text in financial statements contextualizes firms’ tax outcomes, which narrative topics are most informative, and whether the contextual information is useful to financial statement users. We build on prior research, which broadly finds that text is informative, by explicitly modeling context as an interaction between text and numbers that improves the mapping to firm outcomes. Taxes are an ideal setting for this analysis because the link between an accounting number (pre-tax income) and a tax outcome (GAAP tax expense or cash taxes paid) is both conceptually grounded and systematically distorted by reporting and regulatory rules that narrative disclosures can help explain. Using embeddings derived from management discussion and analysis (MD&A) sections of 10-K filings, we train deep neural networks that learn how textual context alters the relation between pre-tax income and tax outcomes. We show that context from the MD&A has significant explanatory power, improving the mapping between book income and tax outcomes by 17.7% to 21.3%. In contrast, income tax footnote narratives often obscure rather than clarify this mapping. Moreover, we find that disclosures referring to M&A activities, firms’ external environments, and firm structures and segments add the greatest informational value, whereas discussions relating to corporate governance and the application of accounting standards even reduce contextual value. Finally, we show that MD&A context is useful to financial statement users (e.g., by improving analysts’ effective tax rate forecasts). Collectively, the findings demonstrate the value of contextual information in understanding distortions between book and tax numbers.
