Inferring U.S. Tax Liability from Financial Statement Information

Abstract

Using a multi-year matched tax return-financial statement dataset, this study builds empirical models that infer U.S. tax liability on the corporate tax return from publicly available financial statement disclosures, including those of Statement on Financial Accounting Standards No. 109, Accounting for Income Taxes (SFAS No. 109). Results show that current U.S. tax expense, the tax benefit from stock options, current-year tax cushion accrual, consolidation book-tax differences, and R&D are informative in inferring actual tax, while intraperiod tax allocation is not. Additionally, the sign of pretax book income and the existence of net operating loss carryforwards are useful partitioning variables in estimating actual tax. In general, for every dollar of current U.S. tax expense reported on the financial statements, approximately $0.70 is reported in U.S. tax liability on the tax return. The models are validated using a holdout sample, providing support for the notion that public parties can reliably use these results to estimate a firm's tax position. Additional tests reveal a hierarchy of sub-samples that researchers may employ when maximizing the usefulness of tax-related disclosures in inferring U.S. tax liability.