Macroeconomic implications of modeling the Internal Revenue Code in a heterogeneous-agent framework⋆

Tax policy analysis in macroeconomic models requires the incorporation of a tax system that can approximate complex present tax law and proposed deviations. One challenge is that policy changes affect each taxpayer differently. Despite the household heterogeneity featured in early heterogeneous-agent models, the entire tax system is often approximated using a single parameterized tax function that smoothly maps household income into tax liabilities or effective tax rates, such as those developed in Easterly and Rebelo (1993), Gouveia and Struass (1994), Bénabou (2002), and Li and Sarte (2004).¹ While the recent work of DeBacker et al. (2019) builds upon this literature by using a large number of tax functions conditioning on household age, two shortcomings of this conventional approach remain: First, functional form assumptions and the imposition of smoothness on the tax system are questionable approximations of an actual tax system that features non-convexities.² Second, failing to explicitly condition on idiosyncratic household characteristics such as filing status, number of dependents, and tax-preferred consumption choices ignores the variance in tax liability for households earning similar incomes. Growing public interest in the macroeconomic effects of tax policy changes for the United States warrants a critical evaluation of these shortcomings.³

To examine the extent to which the shortcomings of the conventional approach have implications relevant for tax policy analysis, we develop an alternative approach: Within a large-scale overlapping generations (OLG) model, we embed a tax calculator that explicitly incorporates tax provisions of the United States Internal Revenue Code (IRC) applied to labor income and conditions on idiosyncratic household characteristics when computing tax liabilities. Specifically, we model the present law⁴ statutory tax rate schedule, standard deduction, earned income credit, child tax credit, home mortgage interest deduction, state and local income, sales and property tax deductions, charitable giving deduction, net investment income and Medicare surtaxes, and dependent care credit. Unlike the conventional approach, we do not impose functional form assumptions on the effective tax rate schedules. Instead, endogenous household behavior generates deviations from the statutory tax rate schedule either by employment choices, which could affect tax credits, or consumption choices, which could affect deductions.⁵ This is important because certain policy proposals may include measures with potentially offsetting effects on incentives which vary across households (Gravelle and Marples, 2015), or may have interactions with underlying provisions in the tax code. By explicitly modeling a wide range of key tax provisions in lieu of parsimony, our approach ensures that tax policy changes involving large, discrete effects on a relatively small group of households are not washed out as a smaller change for the wider population.

We choose one of the commonly-used smooth tax functions analyzed in Guner et al. (2014), the Bénabou (2002) tax function, as a benchmark. Relative to this, our approach offers two innovations: First, the internal tax calculator can target specific households affected by a policy change. Homeownership is a choice in the model, so if the home mortgage interest deduction were repealed or its value were to be reduced due to interactions with another change to the tax code, for example, only households claiming that deduction would experience an increase in their tax liability. Alternatively, the tax function would shift for every household, indiscriminately. The second advantage of the tax calculator is that households have the opportunity to react optimally to the policy change. If the deduction is repealed or its value is reduced directly or indirectly, renters who would purchase a house in the near future under present law may continue to rent, or decide to purchase a smaller property. Under the tax function, the extent to which households are affected by the change is implicitly assumed.

We simulate two tax policy changes using the internal tax calculator and the smooth tax function: (i) a ten percent reduction in statutory tax rates on ordinary income; and (ii) an expansion of the earned income tax credit for childless adults. In both policy experiments, labor and consumption behavior vary across the two systems because of the differing levels of tax detail. For the statutory rate cut, the internal tax calculator captures the reduction in the value of itemized deductions for high-income households while the tax function does not. This results in substitution away from tax-preferred consumption choices, such as charitable giving and owner-occupied housing, and towards saving under the internal tax calculator environment. For the EITC expansion, the internal tax calculator can specifically target lower-income childless households, while the tax function instead treats all lower-income single households similarly. Further, the tax function models this policy change as a simultaneous average rate decrease and effective marginal rate increase for many EITC recipients, contrasting with the explicit modeling of the expansion and change to phase-in and -out regions for households under the internal tax calculator. As a result, the net change to labor hours is a wash under the tax function environment while net labor hours increase post-expansion under the internal tax calculator environment.

Although we find that for a debt-constant steady state analysis the variation in underlying household behavior across tax systems is unimportant for the response of macroeconomic aggregates, we find that for a debt-financed transition path the variation in household behavior is associated with significant quantitative and qualitative differences in the time path of labor supply, private capital, and public debt accumulation. Taken together, our results indicate that while unconditional smooth tax functions may be an appropriate modeling simplification for performing steady state analysis with heterogeneous-agent models, they are less suitable for transition analysis of tax policy changes. Our findings provide support for the incorporation of explicit tax detail in lieu of parsimony for purposes of tax policy analysis in this class of models.