In a broad sample of publicly traded firms, we observe that the share of firms annually reporting pre‐tax book losses increased from about 20% to 40% during 1988–2023. We also observe that 68% of those loss firms have positive cash tax payments (taxpaying loss firms). The amount of taxes paid by these loss firms is substantial and increasing over time. Surprisingly, we observe that taxes paid increase with the magnitude of pre‐tax losses. This study seeks to understand the prevalence of taxpaying loss firms. We examine whether both the extensive margin—the likelihood that a loss firm pays taxes—and the intensive margin—the magnitude of taxes paid—are explained by firm characteristics. We find that multinational status, state taxes, consolidation differences, goodwill impairments, asset write‐downs, extraordinary items, discontinued operations, depreciation differences, the frequency and magnitude of losses, and firm size are key determinants of both the likelihood and the amount of taxes paid by loss firms. We find that the decrease in the statutory tax rate included in the Tax Cuts and Jobs Act of 2017 did not decrease the tax burden on loss firms.

We develop a linear corporate tax function where taxes paid are regressed on pre-tax income and an intercept. We show that if the intercept is positive, cash ETRs are a convex function of pre-tax income. We present large-sample evidence consistent with this ETR convexity. Thus, although firms may have stable linear tax functions (i.e., constant parameters in the linear tax model) representing stable tax avoidance behavior, ETRs can change over time because of growth in pre-tax income. Consequently, simply examining changes (or differences) in cash ETRs is nondiagnostic about whether tax avoidance has changed over time (or differs across firms). We illustrate our argument by showing that all of the observed downward trend in cash ETRs documented by Dyreng et al. (2017) can be explained by growth in pre-tax income. The wholesale concern about increased tax avoidance over time might be overstated. Data Availability: Data are available from the public sources cited in the text.