Saez Zucman 2023
Saez Zucman 2023
Saez Zucman 2023
Abstract
This paper proposes a new framework to study the distribution of taxes and the effects of
tax reforms, connecting classical tax incidence analysis to optimal tax theory. To study the
distribution of current taxes, labor taxes are assigned to the corresponding workers, capital
taxes to the corresponding asset owners, and consumption taxes to consumers. The tax
rates are the wedges between pre-tax prices (relevant for production) and after-tax prices
(relevant for the work, saving, and consumption decisions of households). In contrast
to the conventional approach that shifts taxes across production factors, our approach
measures actual incomes, is internally consistent, and maximizes the comparability of tax
progressivity and inequality over time and across countries. Applying this methodology
to the United States, we find that the effective tax rate of the top 1% has declined from
about 50% in the early 1950s to 32% in 2021. It is through the corporate tax that a high
degree of tax progressivity was achieved in the middle of the 20th century. To analyze
the distributional effects of tax reforms, mechanical changes in tax liability by income
groups and aggregate revenue effects due to household behavioral responses are sufficient
statistics in neoclassical optimal tax models. The effects of taxes on pre-tax prices at the
heart of classical tax incidence analysis are irrelevant. This neoclassical framework can
be extended to incorporate non-standard behavioral responses uncovered by the recent
empirical literature. We apply this framework by providing a distributional analysis of
frequently discussed tax reforms, including replacing employer-provided health insurance
contributions by a payroll tax.
∗
Emmanuel Saez, University of California, Department of Economics, 530 Evans Hall #3880, Berkeley, CA
94720, saez@econ.berkeley.edu. Gabriel Zucman, Paris School of Economics, 48 boulevard Jourdan, 75014
Paris, France, and University of California, Department of Economics, 530 Evans Hall #3880, Berkeley, CA
94720, zucman@berkeley.edu. We thank Alan Auerbach, Youssef Benzarti, Antoine Bozio, Lucas Chancel,
Jarkko Harju, Greg Leierson, and Thomas Piketty for helpful discussions and comments. We thank Akcan
Balkir for outstanding research assistance. Funding from the Stone center at UC Berkeley and the European
Research Council is thankfully acknowledged. A preliminary version was circulated in October 2019 under the
title: “Clarifying Distributional Tax Incidence: Who Pays Current Taxes vs. Tax Reform Analysis”. Some of
the empirical results in this paper are presented in a more detailed manner in a book (Saez and Zucman, 2019)
but without the formal underpinning presented here.
1 Introduction
Who pays taxes, and how tax reforms would affect the different socieconomic groups, are ar-
guably some of the most important questions in modern democratic societies. Governments of
high-income countries collect 30% to 50% of national income in taxes. These cash payments
have a first-order effect on the disposable income of households. To inform lawmakers and vot-
ers, it is thus critical to have a sound and practical way to allocate taxes across income groups
and to analyze who would gain or lose from proposed changes to the tax system.
Theoretically, this is the classical question of tax incidence (see Kotlikoff and Summers,
1987, and Fullerton and Metcalf, 2002, for surveys). Empirically, distributional tax analysis
of the full tax system was first produced in the United States following the founding work of
Colm and Tarasov (1941), Musgrave et al. (1951), and Pechman and Okner (1974). Building on
this work, US government agencies and think tanks have developed sophisticated frameworks
to analyze the distribution of federal taxes.1 The results of these models are published in the
form of distributional tax tables that have a large impact in the public debate. A large and
growing body of academic work also mobilizes the tools of distributional tax analysis globally
(with variation in methods used) to estimate inequality and study tax progressivity.2
This paper highlights shortcomings in the conventional practice of distributional tax analysis
and proposes a new framework grounded in optimal tax theory that overcomes these limitations.
Our starting point is that distributional tax analysis serves two purposes. First, it provides
information on the current distribution of income and tax payments by income groups, which is
crucial to quantify income inequality, pre-tax and post-tax, and the direct effects of taxes. From
now on, we call this analysis distributional current-tax analysis. Second, it is used to simulate
how a change to the tax system would affect the different socioeconomic groups. From now on,
we call this distributional tax-reform analysis. In the conventional approach, the allocation of
existing taxes and the simulation of tax reforms are done using the same models of tax incidence.
But the two types of analyses, we argue, require distinct methodologies, each different from the
one conventionally used. This paper presents these methodologies, applies them to the United
1
See US Congressional Budget Office (2018), US Joint of Committee on Taxation (1993, 2019), US Treasury
(2019), and Tax Policy Center (2022) for detailed descriptions and Barthold (1993) for a summary of the practical
use of such statistics by the US congress. There is less work on how state and local taxes are distributed, even
though state and local taxes represent about a third of total US tax revenue. Institute on Taxation and Economic
Policy (2018, 2019) is the only current systematic study on the distribution of US state and local taxes.
2
Recent estimates of tax progressivity include Aaberge et al. (2021) in Norway, Advani et al. (2023) in the
United Kingdom, Atria and Otero (2021) and De Rosa et al. (2022) in Latin America, Bach et al. (2023) in
France, Blanchet et al. (2022) in Europe, Bruil et al. (2022) in the Netherlands,, Guzardi et al. (2022) in Italy,
and Saez and Zucman (2019) in the United States.
1
States, and provides a practical guide for their implementation globally.
Our first contribution is to propose a new method for distributional current-tax analysis. In
our framework, taxes based on labor income are assigned to the corresponding workers, taxes
based on capital or capital income to the owners of the corresponding assets, and taxes based on
consumption to the corresponding consumers. This current-tax analysis is economically sound:
it describes the price distortions created by the tax system, as one writes a model of optimal
taxation. The tax rates in this approach are the wedges between pre-tax prices (relevant for
production) and post-tax prices (relevant for the work, saving, and consumption decisions of
households). This approach differs from simply following statutory incidence. For example, both
employer and employee payroll taxes are a tax on labor, and hence are assigned to workers. In
contrast to the conventional approach that shifts some taxes—most notably the corporate tax—
across production factors, our approach measures actual incomes (not counterfactual incomes);
it is internally consistent; it maximizes the comparability of inequality and tax progressivity
over time and across countries with different legal systems; and it is much simpler to implement,
because it does not depend on assumptions about behavioral responses to taxes.
Applying this framework to the United States, we find that the effective tax rate of the
top 1% of the income distribution has declined from nearly 50% in the early 1950s to 32% in
2021. Thanks to a consistent treatment of business profit taxes, we illuminate the dramatic
changes in the taxation of top-end business income over the last century. Rich business owners
faced significant price distortions in terms of pre-tax vs. after-tax returns to capital in the
1950s: they paid half or more of their profits in corporate taxes, before facing the progressive
individual income tax on distributed income. We show that it is through the corporate tax that
the US tax system achieved its high degree of progressivity in the middle of the 20th century—
not through the individual income tax, which has absorbed a relatively constant fraction of the
pre-tax income of top earners since 1930.
In contrast to conventional incidence, our proposed current-tax analysis captures only the
equity aspect of existing taxes. Conceivably, the high tax rates on business income at mid-
century might have been detrimental to workers. Perhaps middle-class wages would have been
even higher with lower corporate taxes. Our current-tax analysis does not provide information
on counterfactual levels of income if the tax system was different, and hence is silent about the
efficiency costs of taxation. But it provides a crucial input required to quantify these efficiency
costs and to assess the desirability of tax reforms, namely the distortions created by the existing
tax system. It is also consistent with the classic dichotomy between equity vs. efficiency effects
that arise in all optimal tax models.
2
Our second contribution is to use optimal tax theory to identify the sufficient statistics
needed to conduct distributional tax-reform analysis in neoclassical models. In the optimal tax
models of Mirrlees (1971) and Diamond and Mirrlees (1971), all that is needed to assess the
desirability of a tax reform is: (i) mechanical changes in tax liability by income groups (which
follow directly from current-tax analysis) weighted by social marginal welfare weights to reflect
the distributional preferences of society, and (ii) the aggregate revenue effects of the reform due
to household behavioral responses, keeping pre-tax prices fixed. Revenue effects do not have
to be broken down by income groups: behavioral responses matter only for their aggregate
effect on the government budget. The effect of taxes on pre-tax prices—effects that are the
heart of classical tax incidence analysis since Harberger (1962, 1964)—turn out to be irrelevant
normatively because they can be offset at no fiscal cost with an additional tax adjustment. To
understand the intuition, consider a tax on capital. If the tax hurts wages, it also correspondingly
increases the rate of return for capitalists. Because this extra capital income can be taxed away
to make workers whole, the change in factor prices is irrelevant from an optimal tax perspective.
In a nutshell, in our paper, Harberger at long last meets Diamond and Mirrlees—and it is the
Diamond-Mirrlees insights that turn out to matter most for tax reform policy advice. Diamond
and Mirrlees provide the big picture on which directions of tax reform are desirable. Harberger
provides the narrower technical information on how to achieve such tax reforms.
In the real world, because of departures from the assumptions of neoclassical models, tax
reforms can have non-standard effects. Our framework can be extended to incorporate these
effects. In such cases, it becomes necessary for welfare analysis to include predicted changes in
pre-tax and post-tax income by income group in tax-reform tables. The result can then be com-
pared to the current-tax table side by side, providing all the information needed to assess the
desirability of the reform considered. To guide the analysis of tax reform in practice, we sum-
marize the main non-standard effects uncovered by a burgeoning new empirical tax literature.
We implement our framework by providing a distributional analysis of frequently discussed tax
reforms in the United States, including replacing employer-provided health insurance contribu-
tions by a payroll tax, when non-standard incidence effects are shown to be crucial.
The rest of this paper proceeds as follows. Sections 2 founds our concepts in standard models
of taxation. Sections 3 presents distributional current-tax analysis in practice. We provide an
application to the study of the evolution of tax progressivity in the United States in Section 4.
We then move to the practice of tax reform analysis. Section 5 applies our distributional tax-
reform framework to an increase in the corporate tax rate and in the individual income tax for
the top 1%. Section 6 extends our framework to incorporate non-standard incidence effects,
3
with an application to health insurance funding. Section 7 concludes.
Distributional current-tax analysis. In this model, pre-tax income is wl, after-tax income
is wl −T (wl), and taxes paid are T (wl). A consistent current-tax analysis assigns T (wl) in taxes
to an individual earning pre-tax income wl. Aggregating pre-tax income, after-tax income, and
taxes paid by income groups provides relevant information (in a sense made precise below)
about the equity of the current tax system.
A potential limit of such an analysis is that it is silent about any efficiency considera-
tion. To incorporate efficiency costs, the starting point is that income taxation affects labor
supply—hence pre-tax earnings—through income and substitution effects. Take the simple case
of quasilinear and iso-elastic utility functions of the form u(c, l) = c − l1+1/e /(1 + 1/e). Labor
supply is l = [w(1 − T 0 (wl))]e , so that e is the elasticity of labor supply with respect to the
net-of-tax wage rate w · (1 − T 0 (wl)). This model has the advantage of ruling out income effects
so that e is both the compensated and uncompensated elasticity, and u(c, l) is also a money-
metric measure of utility where $1 extra of consumption translates into $1 extra of utility. In
a counterfactual world without taxes, labor supply would be l0 = we and earnings would be
wl0 = w1+e . The income tax depresses pre-tax earnings by a factor (1 − T 0 (wl))e .
To assess the effect of the tax on utility, consider the simple case of a linear tax at rate τ , so
that T (wl) = τ wl. It follows that u(c, l) = wl(1 − τ ) − l1+1/e /(1 + 1/e) = wl(1 − τ )/(1 + e) as
l1/e = w(1 − τ ). The linear tax decreases utility from wl0 /(1 + e) down to wl(1 − τ )/(1 + e). The
4
money-metric welfare cost of the linear income tax is [wl0 −wl(1−τ )]/(1+e). The money-metric
cost of taxation is not simply counterfactual pretax income wl0 minus actual aftertax earnings
wl(1 − τ ). This difference has to be divided by 1 + e to account for the fact that labor supply—
hence its disutility—also changes. The tax burden can be expressed as wl0 ·[1−(1−τ )1+e ]/(1+e).
The linear tax rate τ is transformed into a burden tax rate of [1 − (1 − τ )1+e ]/(1 + e). Tax
burdens do not add up to taxes collected since tax burdens include the deadweight burden of
taxes.
Although such an analysis can yield insights, it also faces two main limitations. First, it
is sensitive to the assumption made about the elasticity e. Earnings absent taxes wl0 are an
unobserved counterfactual, the computation of which relies on estimates of behavioral responses
to taxes. With a positive e, counterfactual earnings and welfare costs depart from actual pre-tax
earnings wl and actual taxes paid T (wl). Counterfactual aggregate earnings can be substantially
higher than actual aggregate earnings. Second, counterfactual earnings absent taxes wl0 , money-
metric welfare costs [wl0 − wl(1 − τ )]/(1 + e), and actual earnings after-tax wl(1 − τ ) do not
add up in the usual way (pre-tax earnings − taxes = after-tax earnings), except when e = 0
(no behavioral response) or the tax rate τ is vanishingly small. These limitations are probably
the reason why, to our knowledge, official distributional tax tables have never been presented
using wl0 and welfare costs.
Optimal tax analysis. One could think that this added complexity is the price to pay to talk
about optimal policy rigorously. But this is not correct, because modern optimal tax analysis
is based on the analysis of small tax reforms, where the no-tax counterfactual is irrelevant.3
Optimal tax theory aims at determining whether a tax reform is desirable and ultimately what
is the best tax system (i.e., the tax system that no further reform can improve upon). The welfare
changes caused by a tax reform are aggregated across all individuals using social marginal welfare
weights. The social marginal welfare weight for a given individual is the social value of one extra
dollar of disposable income to this individual (relative to one extra dollar of public funds). If
the aggregated weighted welfare gains of a tax reform are positive, the reform is desirable.
Consider a small reform dT (.) of an initial tax schedule T (.). Crucially, because labor supply
maximizes individual utility, the money-metric welfare cost of the reform for an individual
making wl pre-reform is given by dT (wl), i.e., the change in tax liability that the individual
3
The literature on the marginal costs of public funds has also developed tools to evaluate the efficiency costs
of marginal tax reforms (as opposed to existing taxes). There is a close conceptual connection between this
literature and the optimal tax literature. See Auerbach (1985) for an early survey and Hendren and Sprung-
Keyser (2020) for a new exposition and application to many policies.
5
would incur absent any behavioral response.4 The welfare cost of the reform is obtained by
looking at the mechanical impact of the tax change, ignoring behavioral responses. Of course,
the reform also affects labor supply and pre-tax earnings by d(wl), which in turn change the
tax liability of each individual by T 0 (wl)d(wl). But because labor supply responses have no
first-order effect on the welfare of the corresponding individuals, such behavioral responses
only matter for their aggregate impact on tax revenue. The classical tax incidence question of
the burden of taxes relative to a no-tax counterfactual is not relevant for tax reform analysis.
Instead, for the equity part of such an analysis, the two relevant concepts are i) the distribution
of taxes actually paid and ii) the distribution of tax changes ignoring behavioral responses.
6
Extension: tax evasion and avoidance. This model easily extends to incorporate tax
evasion or avoidance. Suppose that labor supply l and hence earnings wl are inelastic but
reported income z is elastic to taxes due to avoidance or evasion responses. In this case, pretax
income is wl and aftertax income is wl − T (z), as wl is a better measure of economic income
than reported income z which is affected by tax avoidance. Our findings carry over in this
setting. The direct impact of the tax change absent behavioral responses is what matters for
distributional analysis, while the behavioral responses matter only for their aggregate impact
on tax revenue. Empirically, wl may not be observable.5 If wl is observable, the distributional
tax table should be reported along this dimension and reported income z matters only through
its impact on taxes.6 If wl is not observable, one would ideally want to impute wl based on
reported incomes and what is known about tax evasion/avoidance and its distribution. This
highlights the need to develop good empirical measures of the distribution of tax evasion (see
Guyton et al., 2021, for a recent attempt in the United States).
Setup of the model. On the production side, the model is competitive with an aggregate
production function Y = F (K, L) with constant returns to scale, where K is capital and L
is labor. We denote by w the economy-wide pre-tax wage rate and by r the pre-tax rate of
return on capital. Profits maximization leads to the standard conditions: w = FL and r = FK .
5
True earnings may be observable if reported earnings correspond to true earnings and evasion/avoidance
takes place through deductions, for example. But true earnings may not be observable, for example, in the case
of the self-employed.
6
This is why distributional tables presented in income before deductions such as Adjusted Gross Income in the
United States are more useful than distributional tables presented in taxable income that nets out deductions.
The United States has a commendable tradition of presenting distributional tables in income before deductions.
7
Feldstein (1974) is the classic reference for this model. Kotlikoff and Summers (1987) consider the same
model in their tax incidence handbook chapter. Both focus primarily on the inelastic capital case while we find it
more pedagogical to focus on the inelastic labor case. Our main addition in the context of this model is to show
how the analysis of tax incidence—the focus of Feldstein (1974) and Kotlikoff and Summers (1987)—relates to
optimal tax analysis.
7
Because of constant returns to scale, there are no pure profits and F (K, L) = rK + wL in
equilibrium, so that output Y can be divided into capital income rK and labor income wL. We
denote by σ the elasticity of substitution between capital and labor in the production function
and by α = rK/Y the share of capital income in the economy.8
On the supply side, we assume that labor (L) is fixed, labor income is taxed at rate τL , and
capital depends on the net-of-tax return r̄ = r · (1 − τK ) where τK is the tax rate on capital
income. We can express everything in terms of capital per unit of labor k = K/L. As L is fixed,
the supply of k depends solely on r̄: k(r̄). We can define f (k) = F (1, K/L) = F (K, L)/L as
output per unit of labor. We have FK = f 0 (k) and FL = f (k) − kf 0 (k).
This model can be micro-founded with a simple two-class economy with workers and cap-
italists. Capitalists have a (reduced-form) utility function of the form uK (c, k) increasing in
consumption c = r̄k and declining in k, reflecting the opportunity cost of supplying capital to
domestic production. If the net-of-tax return increases, capitalists are willing to supply more
capital, either by saving more or by bringing capital from another sector—e.g, capital owned
abroad—into the domestic production sector.9
Equilibrium of the model. The following three equations determine the equilibrium (w, r, k):
This simple model has the advantage of being representable in a standard capital demand and
supply diagram (Figure 1). Even though this is a general equilibrium model, the diagram is the
same as the standard textbook one-market model of tax incidence. The demand for capital is
r = f 0 (k) and is downward sloping (as f 00 (k) < 0). The supply for capital is k = k(r · (1 − τK ))
and is upward sloping (and flat when eK = ∞). The surplus accruing to workers is w =
Rk
f (k) − kf 0 (k) = 0 f 0 (κ)dκ − rk and can be read off as the area below the demand curve
and above the horizontal line at r. The surplus accruing to capitalists is the area above the
8
With Cobb-Douglas production functions of the form F (K, L) = A · K α L1−α then α = rK/Y is constant
and σ = 1. With a CES production function F (K, L) = [µK (σ−1)/σ + (1 − µ)L(σ−1)/σ ]σ/(σ−1) the elasticity of
substitution σ is constant.
9
This utility form can arise from two models. First, suppose capitalists have a fixed capital k0 and decide
how much to invest domestically k and how much to invest abroad k0 − k. Suppose capital abroad earns a rate
of return r0 but that capitalists value investing k at home by a(k) with a(.) ≥ 0 increasing and concave reflecting
home bias. In this case, money metric utility takes the form uK (c, k) = c + a(k) with c = r̄k + r0 (k0 − k), leading
to a first order condition a0 (k) = r0 − r̄ which defines an upward sloping supply of domestic capital k(r̄). With
no home bias, the supply is infinitely elastic as r̄ = r0 . Second, as in Saez and Stantcheva (2018), intertemporal
maximizers have instantaneous utility c + a(k) for consumption and wealth, discount rate δ, and start with
wealth k0 . In this case, intertemporal utility takes the simple form c + a(k) + δ(k0 − k) with c = r̄k which leads
to a0 (k) = δ − r̄ (the wealth of the individual jumps immediately from k0 to k at time zero). Without utility for
8
supply curve and below the horizontal line at r̄ = r · (1 − τK ). Capital taxes are the rectangle
(r − r̄)k. The triangle pointing toward the no-tax equilibrium f 0 (k ∗ ) = r∗ , k(r∗ ) = k ∗ is the
usual deadweight burden. It is equal to the loss in surplus of workers and capitalists created by
the tax τK over and above its revenue yield (r − r̄)k.
Tax incidence analysis. Let us now move to the analysis of tax reforms. We consider a
small increase in the capital tax rate dτK and trace out its effects dk, dr, dw. Differentiating the
3 equations in (1), we have two equations on the production side:
dk dw dr
=σ· − , dw + k · dr = 0.
k w r
The first equation is the definition of the elasticity of substitution between labor and capital,
σ. The second equation is obtained by differentiating f (k) = rk + w and using f 0 (k) = r. This
wealth, the supply is also infinitely elastic as r̄ = δ.
9
equation is key: it states that the effects of the reform on factor prices sum to zero. What labor
loses due to reduced wages is exactly what capital gains through a higher return.
On the supply side, we have:
dk dr̄ dr dτK
= eK · = eK · − .
k r̄ r 1 − τK
Combining and rearranging, we obtain:
dr (1 − α)eK dτK dk σ dτK
= · , = −eK · · , dw = −kdr.
r (1 − α)eK + σ 1 − τK k (1 − α)eK + σ 1 − τK
These equations display the usual lessons from tax incidence. First, if σ = ∞, then a capital
tax increase has no effect on factor prices r and w. It only affects capital through a pure supply-
side response: dk/k = −eK dτK /(1 − τK ). Second, if σ < ∞, then capital supply responses affect
factor prices, spreading partly the incidence of the tax onto wages. The shift to wages is small
whenever eK is small relative to σ.
This is illustrated in Figure 2. The increase in τK shifts the equilibrium. The reduction in r̄
along the supply curve is attenuated by an increase in dr along the demand curve. The response
dk is attenuated relative to the case where r is fixed. Capital tax revenue is τK rk = (r − r̄)k.
Its change can be decomposed into three terms depicted on the graph:
The first term −kdr̄ > 0 is the direct effect due to a lower net-of-tax rate of return r̄. The
second term kdr > 0 is due to a higher pre-tax rate of return r. Importantly, this term is
exactly equal to −dw, i.e., what is lost by workers due the reduction in the wage rate w. The
third term is the tax revenue lost due to the supply-side response of capital (itself triggered by
dr̄). This tax revenue loss is equal to the increase in the deadweight burden triangle of the tax.
Optimal tax analysis. Suppose the social marginal welfare weight on capitalists is zero.
Maybe capitalists are much more well-off than workers (and hence have much lower marginal
utility), or maybe all residents are workers and the country attracts capital from abroad only.
In this case, society sets τK to maximize workers’ income w + (r − r̄)k where w is the wage and
(r − r̄)k is the tax collected from capitalists. As w + rk = f (k), social welfare is w + (r − r̄)k =
f (k(r̄)) − r̄k(r̄). The government effectively chooses r̄ along the supply side curve k(r̄) to
maximize surplus—the area above the line r̄ and below the demand curve for capital (blue area
in Figure 1). The first-order condition for the optimum τK is such that:
0 r − r̄ r̄ dk τK
0 = (f (k) − r̄)dk − kdr̄ = −kdr̄ 1 − = −kdr̄ 1 − eK . (3)
r̄ k dr̄ 1 − τK
10
This leads to the usual inverse-elasticity rule optimal tax rate τK∗ = 1/(1 + eK ).
The key insight is that the optimal tax rate only depends on the supply elasticity eK , not
on whether the tax on capital is shifted to workers. In other words, the supply elasticity is a
sufficient statistics for the optimal tax rate (and the elasticity of substitution σ is irrelevant).
The intuition for this result can be seen on Figure 2. Workers’ welfare is the wage area w
plus the tax rectangle. When τK increases, the reduction in wages dw is fully offset by the
increase in tax revenue kdr. As a result, the tradeoff is only about the mechanical increase in
tax revenue kdr̄ vs. the revenue loss due to the supply side response (r − r̄)dk.10 Intuitively,
setting τK is equivalent to setting r̄ so that the implicit changes in r triggered by τK can be
neutralized.11 This result is a special case of a more general result first derived by Diamond and
Mirrlees (1971).12 Optimal tax formulas can be expressed solely in terms of supply elasticities
(for production factors) or demand elasticities (for consumption goods), and social marginal
welfare weights. Conditional on these, elasticities of substitution (e.g., between capital and
labor) are irrelevant.
This result has a key implication: the effects of taxes on pre-tax prices—effects that are at
the heart of classical tax incidence analysis—are not normatively relevant. What matters for
optimal tax policy is the behavioral responses of individuals as consumers, savers, and workers.
In the context of capital taxation, changes in wages do not matter. How can this result be
squared with the common intuition that if the tax on capital hurts wages, it makes the tax
less desirable to workers? The reasoning is the following. If the tax on capital hurts wages, it
also means that it increases the rate of return for capitalists, and therefore tax revenue from
capitalists that can benefit workers. In net, this is a wash.
Why has the normative irrelevance of price effects been ignored by the literature on tax inci-
dence? The aim of tax incidence was strictly positive and narrow: explain all the consequences
of a given tax reform that a government is contemplating. The aim of optimal tax is normative
and wider: figure out what tax reforms can improve social welfare. Policy makers always have
some social objective. Hence, in our view, it is useful to provide the bigger picture view of which
10
The derivation has been made (independently) by Piketty (2000) and Mankiw (2001) in the special case
where eK = ∞ (horizontal supply curve in our diagrams) as a way to demonstrate the uselessness of capital taxes
in the standard model in which the infinite capital supply elasticity arises from infinite horizon optimizing and r̄
is pinned down by the exogenous discount rate δ. This derivation based on long-run outcomes is distinct from the
classical Chamley-Judd zero capital tax result (see Saez and Stantcheva, 2018; and Straub and Werning, 2020).
11
This result carries over more generally even if government puts a weight on capitalists (say gK < 1 per $ of
capitalist surplus lost). The reform depicted on Figure 2 reduces the surplus of capitalists by kdr̄ < 0 so that
the optimum first-order condition simply becomes (r − r̄)dk = (1 − gK )kdr̄ (instead of (r − r̄)dk = kdr̄) leading
∗
to the classic optimal tax formula τK = (1 − gK )/(1 − gK + eK ).
12
Piketty and Saez (2013) and Saez and Stantcheva (2018) show how it applies to inheritance taxation and
capital income taxation respectively.
11
tax tools can achieve the social objective. A corporate tax increase may hurt workers’ wages
in a narrow incidence analysis but if the extra corporate tax increase is used to cut taxes on
wages, then it can help workers on net as in our basic model and fulfill the social objective. We
will see in Section 5 that connecting these two bodies of work is fruitful to inform the selection
of models used to conduct tax-reform analysis.
Distributional tax-reform analysis. Because the effects of taxes on prices do not matter
normatively, in the context of the model we consider it is both sounder and simpler to ignore
them when carrying out distributional tax-reform analysis. Consider a capital tax increase.
On the equity side of the trade-off, the relevant impact is the direct effect of the tax on capital
owners, ignoring both supply-side responses and any effects on pre-tax wages and rates of return.
For example in the case of a corporate tax increase, all that matters is the mechanical changes
in corporate tax payments by income group, which can be computed using the current-tax table
(reporting how much corporate tax each group of the population pays today). The welfare costs
of these direct effects can be aggregated across income groups using social marginal welfare
weights. On the efficiency side of the trade-off, the sufficient statistic is the total change in
tax revenue due to supply-side responses, ignoring again any price effects. The revenue change
does not need to be distributed by groups. We provide a concrete illustration in the case of an
increase in the US corporate income tax rate in Section 5.1 below.
A capital tax increase also affects factor prices and the distribution of pre-tax income. It
reduces workers’ wages and increases capitalists’ pre-tax income, typically leading to an increase
in overall income inequality. It also changes the amount of taxes paid by each group. But these
effects are normatively irrelevant, because all the pre-tax price effects can be neutralized by a
corresponding adjustment of all the other taxes which is budget neutral. Of course, this result
arises in the context of the specific neo-classical model of Diamond and Mirrlees (1971). It
is important to note, however, that conventional distributional tax analysis considers the very
same type of models. It emphasizes the effect of taxes on pre-tax prices, even though the logic of
the models implies that these effects can be ignored for optimal tax analysis. In the real world,
taxes can have effects that are more complex than what is captured by neoclassical models,
in which case price effects may not be irrelevant. We extend the analysis along those lines in
Section 6.
12
(1971). A consumption tax increase on a good can also generate price effects (on the taxed
good, on other goods, and on production factors). As detailed in Appendix A.1 and illustrated
in Appendix Figure A1, as long as there are no pure profits (or that the government can
fully tax away pure profits), Diamond and Mirrlees (1971) show that such price effects are
normatively irrelevant. This follows the same logic as described above for capital taxes. Any
change in producers’ profits created by the tax change can be neutralized by adjusting the tax
on profits. What matters is (i) the direct distributional impact of the tax ignoring price effects
and behavioral responses (basically whether the rich or the poor consume the good), (ii) the
aggregate fiscal impact due to supply-side responses to the tax change, keeping pre-tax prices
fixed. A distributional tax reform analysis of a consumption tax change should thus follow the
same template as above.13
What if other taxes are not adjusted? The irrelevance of pre-tax price effects hinges on
the key assumption that other taxes can be adjusted to neutralize them at zero budgetary cost.
But what if other taxes are not adjusted? If other taxes are set at optimal levels–namely they
satisfy the optimal tax formula of Diamond and Mirrlees (1971) so that a small change in any
such tax rate has zero welfare impact, then pre-tax price effects still have zero welfare impact
because they are equivalent to small tax rate changes. However, if other taxes are not set at
optimal levels, then pretax price incidence effects have first order welfare impacts and such
effects can overturn the conclusion. Let us take an illustrative example in the two factor model
we considered above. We provide a complete formal description of such a model of labor and
capital in appendix A.2.
Suppose the capital tax τK is only slightly below the optimal rate τK∗ while the labor tax
τL is substantially above τL∗ . In this case, increasing τK has only small positive welfare effects
when ignoring pre-tax price effects. The pre-tax price effects amount to shifting part of the τK
increase into a reduced wage which is equivalent to an increase in τL . Because τL is already far
too high, this has a large first order negative welfare effect which will swamp the small positive
effect of increasing τK . Put simply, with pretax price effects, increasing τK is like increasing
both τK and τL and the downside of increasing τL swamps the modest benefit of increasing
τK . This is obviously valuable information that classical tax incidence analysis provides but it
should not rule out increasing τK . Rather, it calls for lowering τL as a priority, as τL is far above
its optimal level. Therefore, in our view, the most useful information to provide policy makers
13
That is: (i) assign the tax to consumers in a distributional table, ignoring behavioral responses and price
effects, to inform the equity side of the trade-off; (ii) estimate the extra aggregate budgetary effects due to
behavioral responses keeping pre-tax prices constant, to inform the efficiency side of the trade-off.
13
is the effect of changing one specific tax rate with constant pre-tax prices. While such a tax
change can be shifted to other tax bases through pre-tax price effects, such shifts are relevant
only if the other tax rates are suboptimal, leading to the sound policy prescription of also
readjusting the other tax rates. We shall see below that conventional incidence can even fail to
detect second-best Pareto dominated policies such as the corporate tax in the classic Harberger
model. Our proposed approach consistent with optimal tax analysis will sharply detect and flag
such inefficient policies. To sum it, the optimal tax approach provides the big picture and tell
the policy maker which direction to go. Classic tax incidence is a narrower technical tool useful
to map out how to move in such direction taking into account–and neutralizing–price effects.
Distributional tax wedges. To implement this analysis, the starting point—as in the mod-
els studied above—is that taxes are wedges between pre-tax prices (relevant for production
decisions) and post-tax prices (relevant for work, saving, and consumption decisions of house-
holds). Because the government charges taxes on labor, producers pay labor costs in excess
of what workers receive as net compensation. Because of taxes on assets and capital income,
owners receive less than the full capital income generated by their assets. Due to consumption
taxes, buyers of goods and services pay more than what producers receive. The optimal tax
problem is about computing the optimal wedges.
14
Current-tax analysis allocates these wedges to individuals as follows. Labor taxes (which
include payroll taxes and and taxes on wage income—i.e., the full wedge between pre-tax labor
costs and net-of-tax compensation) are assigned to the corresponding workers. Consumption
taxes are allocated to the corresponding consumers. Capital taxes are allocated to the corre-
sponding assets owners: taxes on corporate profits to the individual owners of corporations,
taxes on the profits of unincorporated businesses (e.g., partnerships) to the owners of unincor-
porated businesses, residential property taxes to the owners of residential properties, business
property taxes to the owners of business property, individual income taxes on dividends, inter-
est, rents, capital gains, and royalties to the individuals who earn this income. Assets, their
income flows, and the taxes on those assets or their income are all allocated to the ultimate
owners of the assets. For instance, corporate taxes paid by companies owned by pension funds
are allocated–like the corresponding profits–to the underlying individual owners. A number of
remarks are in order.
Economic meaning of tax wedges. First, even though it does not involve the specification
of behavioral responses, this approach is more than accounting, because it respects the incentives
of economic actors and follows from the standard modeling of supply and demand functions.
Labor taxes are allocated to workers as opposed to employers, because what matters for workers’
labor supply decisions is after-tax compensation, while what matters to employers is pre-tax
labor costs. Capital taxes are similarly allocated to the respective capital owners as opposed
to capital users, because what matters for capital supply is the after-tax capital return, while
capital demand depends on pre-tax returns. Consumption taxes are allocated to consumers as
opposed to producers, because the demand for goods and services depends on post-tax prices,
while production decisions depend on pre-tax prices.
The taxes we measure capture what is effectively paid by people. They incorporate all the
features of the tax system and of its implementation above and beyond the statutory rate,
including any tax avoidance, evasion, and exemptions or deductions in the tax code.
Side of the market irrelevance. Second, this approach differs from statutory incidence—
who nominally pays the tax to the government. The analysis of tax incidence often starts from
the fact that which side of the market has to legally remit the tax is not relevant, so that the
question “who pays?” does not have an obvious answer. The canonical example is employer vs.
employee payroll taxes. Distributional current-tax analysis also features this side-of-the-market
irrelevance. Both employer and employee payroll taxes are assigned to the corresponding workers
15
(even though part of payroll taxes are nominally paid by employers and part by employees),
because both contribute to the wedge between pre-tax labor costs and post-tax compensation.
Retail sales taxes are similarly assigned to final consumers regardless of whether the tax is
nominally paid by consumers or retailers.
Neutrality with respect to income classification for tax purposes. Third, in our frame-
work a tax on a given income is allocated in the same way no matter how the income is reported
for tax purposes. This mimics a key principle underlying national accounts data, namely that
economic statistics shouldn’t be affected by purely legal changes in income reporting. Applying
this principle maximizes comparability of tax progressivity over time and across countries.
To illustrate this point, consider the case of a consultant. This worker can choose to earn
labor income as a salaried worker (W2 income in the United States), as an unincorporated
self-employed (reported on Schedule C income of the individual income tax return), as a self-
employed individual using a pass-through company (S-corporation or partnership income, re-
ported on Schedule E of the individual income tax return), or as a self-employed individual
incorporated in a company subject to the corporate tax (C-corporation income). In our frame-
work, taxes paid on this consulting income are, in all cases, allocated in the same way—to the
consultant.
Consistency with macroeconomic series. Our framework also ensures consistency be-
tween distributional analysis and macroeconomic analysis. Macroeconomics is concerned with
the distribution of aggregate income across labor and capital. For the computation of factor
shares, all pre-tax corporate profit—including 100% of the corporate tax—is considered capital
income. In our approach, individual and (properly weighted) group-level capital shares add
up to the macro capital share. Our approach is similarly consistent with the literature that
estimates effective tax rates on factor incomes and consumption, following the influential work
of Mendoza, Razin and Tesar (1994). In these macro series, the effective tax rate on capital,
for example, is the ratio of all capital taxes (corporate tax, property taxes, dividend taxes, etc.)
divided by all capital income (corporate profits, housing rents, etc.).14 Our framework in essence
extends this work to incorporate the distributional dimension. By design, our group-level capital
tax rates add up to the macro capital tax rate.15
14
See, e.g., Eurostat (2021) for cross-country series in high-income countries, and Bachas et al. (2023) for a
global panel.
15
Our approach also resembles the social accounting approach sometimes applied to distributional analysis
(see, e.g., Wolff and Zacharias 2007). But we come to it using economic reasoning rather than abstracting from
16
Link with distributional national accounts. Last, current-tax analysis is a necessary
input for the production of distributional national accounts—inequality statistics that allocate
all pre-tax and post-tax national income across socio-economic groups (see, e.g., Blanchet et
al., 2022).16 But it can also be applied independently of the distributional national accounts
framework. For instance, one may be interested in allocating only federal taxes (as opposed
to all taxes at all levels of government as in distributional national accounts). One may also
wish to consider specific definitions of income (that differ from the pre-tax national income
or post-tax national income concepts central in distributional national accounts). In all these
cases, the principles described here carry over.
17
that if only the allocation of the corporate tax varies, the choice of a particular methodology
may not matter much practically. But this choice has in fact large implications (quantified in
Section 4.2) for the measurement of trends in overall US tax progressivity. Conceptually, our
methodology has four main advantages.
Consistent trends in tax progressivity. Second, our methodology allows one to study
trends in tax progressivity and in inequality consistently, in contrast to official practice which
can lead to biased trends. Consider the CBO methodology that allocates 25% of the corporate
tax to workers (vs. 75% to capital owners) and 100% of the individual income tax to the
corresponding individuals. If a C-corporation (subject to the corporate tax) elects to be treated
as an S-corporation (subject to the individual income tax of its owners), then in the CBO
treatment the tax system becomes more progressive and pre-tax income inequality increases,
even though nothing real has changed in the tax system or in the economy. The tax system
becomes more progressive because taxes that used to be partly allocated to workers are now
fully allocated to firm owners, who are higher up in the income distribution. Income inequality
(2012), for a detailed description of the methodologies.
18
increases because income that was previously partly assigned to workers is now fully assigned to
firm owners. As shown in Section 4.2, this bias turns out to be significant in the United States,
given the rise of pass-through businesses over the last decades.19 .
Individual-level analysis. Third, our framework allows us to estimate meaningful tax rates
at the individual level, in particular at the very top of the distribution. In the conventional
approach, the corporate tax is spread across all workers and capital owners nationally, propor-
tionally to their wage income and reported taxable capital income. There is no link between
what a company pays in tax, and how much corporate tax is allocated to its owners. By con-
trast, our methodology assigns firm owners their share of corporate profits and corporate tax
payments (de facto treating all corporations as pass-through businesses). This delivers high tax
rates for the owners of corporations that pay high tax rates, and low rates for the owners of
tax-avoiding corporations, as illustrated below.
Simplicity. Last, our current-tax methodology is much simpler than the conventional ap-
proach, as it does not require to make assumptions about behavioral responses to taxes or to
specify counterfactuals. In the conventional approach, calibrating the shifting of the corporate
tax requires complex assumptions (e.g., on the labor vs. capital component of various income
forms, or the normal vs. supernormal rate of return on capital). The empirical basis for these
assumptions is evolving, leading to discrepancies in methods across agencies and over time.
Jeff Bezos. Start with Jeff Bezos, the richest person in the United States in 2018 according
to Forbes. To compute his tax rate, we need to estimate his pre-tax income from all sources
and the taxes he paid (directly and indirectly) worldwide.
Bezos derives most of his income from his stake in Amazon. As reported in its annual 10-
K report to the Securities and Exchange Commission (SEC), the company made $11.3 billion
in pre-tax income globally in 2018.20 Since Bezos owned 16.3% of Amazon, he earned 16.3%
19
The US Treasury (2019), the US Joint Committee on Taxation (2019), and the Tax Policy Center (2022)
also treat differently taxes on C- vs. S-corporate profits. In contrast to all these approaches, our series are not
affected by changes in businesses’ organizational form or income re-classification.
20
This number is net of interest and depreciation; it is conceptually close to corporate profits as included in
19
of Amazon’s profit, i.e., around $1.84 billion. Even though Amazon did not pay dividends in
2018, its profits did constitute income for Amazon’s shareholders like Bezos—income that was
fully saved and reinvested in the firm. Bezos also earned income from other investments, such
as his stake in the Washington Post. Public sources suggest he earned around $250 million
in taxable income from these other investments.21 We disregard other income sources such as
imputed rents on real estate properties and income earned on pension assets and trusts, which
are second-order for our purposes.
Bezos also realized capital gains by selling Amazon stocks, $33 million according to SEC
form 4 public reports. Since he founded Amazon, and prior to 2018 Amazon made little profit,
Bezos’s cost basis was small in 2018. Virtually all of his realized capital gains reflected pure
asset price appreciation, not the effect of past or current retained earnings (already included
in income). Therefore we include the $33 million in realized capital gains in Bezos’s income.
Because his realized capital gains are small, including these gains in income makes negligible
difference to Bezos’s effective tax rate.
We compute Bezos’s total income tax as his share of the income taxes paid by Amazon
plus the income taxes he paid directly. Amazon paid $1.18 billion in cash income taxes in
2018 to federal, state, and foreign governments combined, an effective tax rate of 10.5%.22
In our methodology, Bezos paid $193 million in corporate taxes, namely his share (16.3%) of
Amazon’s corporate income taxes (or, equivalently, 10.5% of his Amazon income of $1.84 billion).
Moreover, Amazon paid business property taxes. The amounts are not publicly disclosed. We
can estimate these taxes as roughly equal to 1% of Amazon’s capital stock, the US-wide average
business property tax rate. This adds around $100 million in taxes for Bezos. Last, according
to ProPublica, Bezos paid $43 million in federal individual income taxes. As a resident of
Washington State, Bezos did not pay state income taxes. Other taxes paid by him directly or
through Amazon are negligible for our purposes.23 His total tax payments thus amounted to
national income. There are differences between profit accounting in financial statements and in the national
accounts; for our purposes in this paper, however, these differences are second-order.
21
According to ProPublica (Eisinger et al., 2021), Bezos reported $284 million in total income on his individual
income tax return. Of this, $1.7 million corresponds to Amazon compensation ($81,840 in wage and $1,600,000
in other compensation—security detail—according to public SEC forms). Since Amazon did not distribute
dividends and since according to SEC form 4 public reports Bezos realized $33 million in capital gains by selling
Amazon stocks (see below), around $250 million in income derived from non-Amazon holdings.
22
Using provisions for income taxes instead of cash income taxes paid gives a similar effective tax rate, 10.6%.
Both measures have merits and demerits. One issue with provisions for income taxes paid is that these provisions
include tax contingencies—taxes that have not been paid but that companies estimate have a more than 50%
chance to be eventually paid as a result of audits and other enforcement activities. Because some of these tax
contingencies end up not being paid (e.g., due to a lapse in statute of limitation), provisions for income taxes
can over-estimate actual tax payments.
23
Residential property taxes paid by Bezos are likely to be negligible compared to his income. Sales taxes
20
$337 million, an effective tax rate of 15.2%.24
Two remarks about this result are worth mentioning. First, if we focus on federal taxes alone
(as US government agencies do), Bezos’s effective tax rate was only 1.9%. According to its 10-K,
Amazon did not pay any federal corporate income taxes in 2018. Property taxes are all paid
to state, local, and foreign governments. The only federal tax Bezos paid was the individual
income tax. For many economic questions (e.g., the study of behavioral response to taxes)
the relevant tax rates are those including all levels of governments. However in some contexts
(e.g., policy discussions of federal tax reforms), effective federal tax rates can also be relevant.
Second, our methodology to allocate the corporate tax implies a higher effective tax rate for
Bezos than the conventional approach. In the conventional approach the amount of corporate
tax allocated to Bezos has nothing to do with the amount paid by Amazon, since the corporate
tax is shifted to workers and capital owners nationally. In the CBO methodology, Bezos pays
about $41 million in corporate taxes, as opposed to $193 million with our methodology.25
Warren Buffett. Buffett’s situation is to some extent similar to Bezos. The company he
owns—Berkshire Hathaway—does not distribute dividends and he realizes little capital gains.
Thus the bulk of his taxes correspond to his share of Berkshire Hathaway’s corporate and
property taxes. According to the conventional approach Buffet pays essentially zero tax, but in
our methodology his effective tax rate (taking into account all taxes paid) was 18.4% in 2018.
Specifically, Buffett had $8.2 billion in income, corresponding to his share (30.2%) of Berk-
shire Hathaway’s $27.0 billion in pre-tax profit.26 According to ProPublica, and consistent with
public SEC reports, Buffett had negligible reported individual income ($24.8 million, out of
which he paid $5.36 million in federal taxes). According to its 10-K, Berkshire Hathaway’s ef-
fective corporate global cash income tax rate was 16.1%—and 18.4% when adding our estimate
paid by Bezos are likely to be negligible too. For example, in the case where he consumed $10 million of taxable
goods in Seattle, the associated sales tax would be $1.025 million (6.5% rate in Washington state plus 3.75%
rate in Seattle), increasing his effective tax rate by only 0.05 percentage point.
24
This effective rate is equal to $337 million divided by $2.2 billion in income: the $100 million in Amazon
property taxes have to be added to the income denominator because they are not counted as income in corporate
income statements.
25
Specifically, in 2018 total U.S. corporate tax revenues (federal plus state) added up to $283 billion. Bezos
earned 0.00002% of all reported taxable wages and 0.02% of all reported taxable capital income (dividends,
taxable interest, net rents and royalties, and realized capital gains); hence he gets assigned 25% × 0.00002 +
75% × 0.02 = 0.0145% of corporate tax payments (or $41 million) if one applies the CBO methodology.
26
Berkshire Hathaway’s pre-tax profits are computed as pre-tax income as officially reported in the 10-K ($4.0
billion), plus realized gains on investments ($22.5 billion), plus (imputed) business property taxes paid ($0.6
billion, computed like in the case of Amazon as 1% of net property and equipment). Unrealized gains on invest-
ments are removed because they are not taxable and not part of conventionally defined income. Consistently,
we measure income tax paid as cash tax paid (as in the case of Amazon).
21
of business property taxes. Since Buffett had negligible individual taxable income, the taxes he
paid at the individual level were negligible relative to his share of the taxes paid by Berkshire
Hathaway. Buffett’s effective tax rate was thus equal to Berkshire Hathaway’s, 18.4%.
Buffett’s case illustrates that the corporate tax—and to a lesser extent business property
taxes—serve as a backstop for the ultra-wealthy. Without these taxes, Buffett’s effective tax rate
would be 0% out of $8.2 billion in income. Moreover, like for Bezos, our methodology assigns
much more corporate tax to Buffett than the conventional approach. Since Buffett has negligible
individual taxable income, in the CBO methodology Buffett is assigned virtually no corporate
tax, even though Berkshire Hathaway, of which he owns 30%, paid more than $4 billion in cash
corporate income taxes in 2018. A complete quantification of the taxes paid by the top 400
wealthiest Americans, systematically linking businesses to owners using administrative data, is
left to future research (Balkir et al., 2023).
22
micro-files (for the post-1962 period) and tabulated series (for the pre-1962 period) used in our
analysis are taken from the February 2022 release of the PSZ series.
Our main statistic of interest is the effective tax rate, defined as total taxes paid at all
levels of government divided by pre-tax income. Following the distributional national accounts
literature, pre-tax income is defined as total income deriving from labor and capital, after the
operation of the pension system and unemployment insurance system.28 To construct income
groups, our unit of observation is the adult individual (age 20 or more) with income equally
split between married spouses, and we rank adults by their pre-tax income.
Table 2 reports the distribution of income and taxes by pre-tax income groups in 2021 using
this methodology. The US tax system appears mildly progressive, with effective tax rates (all
taxes included) ranging from about 26% in the bottom 50% to 34% for the top 0.1%. At the
bottom of the distribution payroll taxes and consumption taxes play a key role. At the top,
the individual income tax is the by far the largest tax.29 When using the CBO methodology
to allocate the corporate tax, effective tax rates at the top are slightly higher (by about 1
percentage point for the top 0.1%), as detailed in Section 4.2 below.
Effective tax rate of the top 1%. The top panel of Figure 3 reports the evolution of the
effective tax rate of the 1% of adults with the highest pre-tax income back to 1913. A number
of findings are worth noting. First, there has been a dramatic inverted-U-shaped evolution of
this tax rate, which increased from about 15% in 1913 to a high of nearly 50% during World
War II and in the early 1950s, before falling back to 32% in 2021. The tax rate of the top 1%
is about the same in 2021 as immediately before the New Deal (32% in 1932). It rose strongly
during World War II, remained at a high level of around 45% until the late 1960s, before falling
in the 1970s and 1980s. Since the 1990s, it has been on a mild downward trend, with some
business cycle volatility—due to relatively strong tax collection at the peak of the cycle—and
a clear effect of tax reforms. It increased from about 30% to 34.5% between 2012 and 2013
(Obama tax reform). It fell from 35% in 2016 to about 32% in 2018 (Trump tax reform).
Second, the effective tax rate of the top 1% is only a little bit higher than the average tax
28
That is, pre-tax national income is net of Social Security taxes, contributions to pension plans, and contri-
butions to unemployment insurance, and symmetrically includes Social Security benefits, pension distributions,
and unemployment insurance benefits. As in the example of Jeff Bezos above and for the same reason, we also
include pure realized capital gains (defined as realized capital gains above 3% of national income, the historical
average level of corporate retained earnings) in pre-tax income.
29
Appendix Table A1 reports the same statistics but focusing on federal taxes only. The federal tax system
is more progressive, with effective tax rates rising from 14.5% in the bottom 50% to close to 23% in the top
0.1%. This is due to the fact that more than 80% of consumption taxes—which are regressive, as low-income
individuals consumer a higher fraction of their income—are levied by state and local governments.
23
rate today. The tax system, by contrast, was highly progressive in the 1940s, 1950s, and 1960s,
when the top 1% rate exceeded the average tax rate by about 20 percentage points. While it is
well known that the United States had a nominally highly progressive federal individual income
tax (with top marginal tax rates exceeding 90% during and after World War II), it is also well
known from publicly available tabulations of income tax returns that few individuals were in
the tax brackets subject to these extremely high rates. The actual degree of progressivity of
the US tax system in the after-war years is thus an open question. Our series show that—all
taxes included—the tax system was progressive not only on paper but in actual facts too. It
is also interesting to note that the rise of the fiscal state—the tripling of the macroeconomic
tax rate from less than 10% of national income in the early 20th century to 30% in the late
1960s—happened in tandem with an even larger increase in the tax rate of the top 1%, from
less than 15% to up to 50%. The expansion of the US government might have been facilitated
by the highly progressive nature of its tax system, although a rigorous test of this hypothesis
falls beyond the scope of this research.30
The key role of the corporate tax. To better understand the change in tax progressivity,
the bottom panel of Figure 3 shows the evolution of the effective tax rate of the top 0.1% with
a decomposition by type of tax. The long-run evolution is even more striking than for the top
1%. The tax rate of the top 0.1% rose from barely 15% in the beginning of the 20th century
to nearly 60% in the middle of the 20th century, before gradually falling back, to about 34% in
2021, the level observed in the 1920s.
When looking at the composition of taxes, a key finding emerges: it is through the corporate
tax that the United States achieved a high level of tax progressivity in the middle of the 20th
century. More broadly, changes in corporate tax payments drive most of the changes in the
effective tax rate of the top 0.1%. Corporate and business property taxes paid by the top 0.1%
rose from about 10% of the pre-tax income of the top 0.1% in the early 1900s to a high of
35% in the 1950s, before falling back to about 7% after the Tax Cuts and Jobs Act of 2017.31
By contrast, the individual income tax has absorbed a broadly constant fraction of the pre-
30
The top 1% contributed about 30% of total US tax revenues in the middle of the 20th century. For instance
in 1950, the top 1% earned 16.5% of total national income, its effective tax rate was 45%, hence it paid 16.5%
× 45% = 7.5% of national income in taxes, which is 30% of the total tax take of 25% of national income.
31
In the 19th century and early 20th century, state and local governments relied on generalized property taxes—
a comprehensive tax on all types of property (real, personal, and financial) that was de facto one of the first
wealth taxes (Dray, Landais, Stantcheva, 2023). This explains why effective tax rate at the top are significantly
higher than 0 (and higher than the average rate) even before the creation of the federal individual income tax
in 1913 and the federal corporate tax in 1909. Generalized property taxes were then gradually phased out and
de facto replaced by the income tax.
24
tax income of the top 0.1%—around 20%—since 1930, with no trend and some business cycle
volatility. Estate taxes rose from 0 before the creation of the federal estate tax in 1916 to
about 6% in the middle of the 20th century, before falling back to about 1% of income in recent
years. If not for the dramatic changes in corporate income taxation (and to some extent estate
taxation), the effective tax rate of the top 0.1% would have exhibited little change since 1930.
Why does the corporate tax play such a large role? Figure 4 shows that there has been
dramatic variation in corporate income tax revenues over the last century in the United States.
In the middle of the 20th century the corporate tax—which had a statutory rate above 50%
and effective rates close to that level—generated about 5% of national income in revenue, and
up to 7% during World War II and the early 1950s. By contrast in recent years it has only
yielded about 2% of national income. Appendix Figure A2 contrasts this evolution to that of
the individual income tax. In 1950 both generated almost as much. Since then, the individual
income tax been growing (primarily due to a rise in state income taxes), while corporate income
tax revenues have collapsed. When the corporate tax was a major source of tax revenue in
mid-century, corporate ownership was highly concentrated—this was before the rise of pension
funds somewhat equalized equity ownership—leading to high tax rates at the top.32
Share of the corporate tax paid by the top 1%. The top panel of Figure 5 contrasts
the fraction of the corporate tax assigned to the top 1% in this approach and ours. A number
of results are worth noting. First, our current-tax analysis allocates a higher fraction of the
corporate tax to the top 1% in the middle of the 20th century: 50–60% in the 1950s–1960s vs.
30%–40% in the CBO methodology. This is because the CBO methodology allocates 25% of
the corporate tax to labor, in effect adding a notional wage tax to workers and reducing the
32
While our current-tax analysis in this paper focuses on the top of the distribution, it can be implemented to
study the effective tax rates for all groups of the population; see Saez and Zucman (2019) for such an analysis
and an interpretative synthesis.
25
burden for firm owners symmetrically. The gap is even larger earlier in the 20th century, at a
time when equity ownership was extremely concentrated. Tabulations of income tax returns
show that the top 1% earned up to 80%–90% of all dividend income through to the 1930s;
accordingly our method allocates a very high share of the corporate tax to the top 1% back
then. In the conventional approach that shifts part of the corporate tax to capital owners other
than shareholders based on reported interest and rents—which were not as concentrated as
dividends—less corporate tax goes to the top 1%.
Second, in our methodology, the top 1% pays a lower share of the corporate tax today than
in the post-World War II decades. This is due to the rise of relatively broadly owned pension
funds, negligible in the 1950s. The top 1% earns 30%–35% of the profits of companies subject
to the corporate tax today—and hence is assigned 30%–35% of corporate tax payments—as
opposed to more than 50% in the 1950s. The share of the corporate tax allocated to the top
1% is stable in our series since the 1980s, as the rise of pension funds since then has been offset
by the rising concentration of directly-held corporate equities.
Third and by contrast, in the CBO methodology the fraction of the corporate tax assigned
to the top 1% is on a rising trend since the 1980s. This is due to two issues. Pensions are
ignored by CBO: for the 75% of the corporate tax assigned to capital owners, the assignment
is proportional to taxable capital income, which excludes tax-exempt capital income earned on
retirement accounts. As taxable capital income is increasingly concentrated (Saez and Zucman,
2016), so too is the corporate tax. Moreover, as 25% is allocated to labor, the corporate tax
becomes more progressive with the rise of wage inequality.
Fourth, these biases are reinforced by the rise of S-corporations, depicted on Figure 4. Until
the 1980s, almost all US corporations were subject to the corporate tax. Today, close to 40% of
domestic corporate profits are made by S-corporations, free of corporate tax and subject solely
to the individual income tax of their owners. These profits generate about 1% of national income
in individual income tax revenues. In the CBO methodology, these taxes are fully assigned to
the owners of the respective corporations, while taxes paid on C-corporations profits are shifted
to workers and capital owners nationally. Taxes on S-corporation profits end up being assigned
in a much more progressive manner (70%-80% to the top 1% in the 1980–2021 period with no
trend) than taxes on C-corporation profits (25% to the top 1% in 1980, rising to 40% in 2021).
As S-corporation profits have risen from 0% to 5% of national income over this period, this
creates a large bias in the 1980–2021 evolution of tax progressivity in the CBO series.33
33
Like CBO, the US Treasury, the Joint Committee on Taxation, and the Tax Policy Center all treat C-
corporations and S-corporations inconsistently. The US Treasury and the Tax Policy Center assign all of the
26
Implication for the decline in tax progressivity. Because the conventional approach
allocates the corporate tax more equally than our methodology in the middle of the 20th century
and corporate income tax revenues were very high then, it delivers significantly lower effective
tax rates for the top 1% in those decades (bottom panel of Figure 5). But since corporate tax
revenues are small today, the different allocation of the corporate tax has little impact on top
effective tax rates today (cf. Table 2). As a result, while in our approach the effective tax rate
of the top 1% falls by nearly 13 percentage points between 1950 and 2021, the decline is only 7
points when applying the CBO methodology. The bias in the conventional approach is larger
as one moves up the income distribution, where business profits account for a greater share of
income.
Comparison with PSZ. The original Piketty, Saez and Zucman (2018) series, which fol-
lowed the conventional approach to distributional tax analysis, also suffer from this bias. In
these series, corporate taxes were allocated to all owners of non-residential capital (including
pensions and non-corporate businesses) and not only to shareholders, building in the tax inci-
dence effects of the standard Harberger (1962) model. This led to the issues detailed above:
internal inconsistency of shifting taxes while keeping aggregate income constant; non-neutrality
with respect to changes in business organizational forms and income classification across tax
forms. Appendix Figure A3 shows that the bias in the original PSZ series is similar to the one
in the CBO methodology, and even more pronounced in the middle of the 20th century.34
27
side responses ignoring price effects. Along with social marginal welfare weights for each group
of the population, these are sufficient statistics to evaluate the value or cost of the reform. As
we discussed in detail, pre-tax price effects can be ignored because they can be neutralized by
adjusting other taxes at zero budget cost. Without such neutralization, pre-tax price effects
are welfare relevant if other taxes are not set optimally. This is why having the full suite of
distributional tax reform for each type of tax is actually useful.
In this section we apply this methodology to two frequently discussed policies: a change in
the federal corporate income tax rate, and an increase in federal individual income taxes for the
top 1%. We contrast our approach with the conventional approach influenced by models where
the relevant elasticities are sometimes infinite by assumption.
28
small and can be neglected in distributional current-tax analysis.35 But because the gross flows
are large, taking into account foreign ownership of US corporations matters for distributional
tax-reform analysis.36 We assume a zero marginal social welfare weight on non US-residents,
but other choices are possible.37
The top panel of Table 3 reports the results. The left panel shows the distribution of current
(as of 2021) incomes and corporate tax payments by income groups. The right panel shows the
effect of the reform considered. Federal corporate tax revenues would mechanically increase by
10%, a gain of $27.9 billion. Corporate profits would shrink, leading to a loss of $3.7 billion in
aggregate tax revenue. The net tax revenue raised by the reform is $27.9 − $3.7 = $24.1 billion.
The reform would entail social welfare costs for all domestic income groups, adding up to $6.9
billion in total. The net value of the reform—i.e., after subtracting social welfare costs—is $17.2
billion, making the reform desirable.
Three remarks are in order. First, in contrast to the conventional approach, we do not shift
any of the corporate tax increase onto labor. If such a shift took place, our method implicitly
assumes that it is undone by readjusting labor and corporate taxes at budget neutral cost. As
we discussed in Section 2.2, this is theoretically possible in the neoclassical model underlying
such incidence effects. It is also important to note that neoclassical pre-tax price effects assumed
in the conventional model are hard to identify compellingly empirically. Therefore such price
effects are much more assumption than established fact (see Section 6 below). Second, there is
uncertainty about the corporate profits elasticity. With our social welfare weights, the reform is
desirable for a value of the elasticity up to 3 (and it raises net revenues for an elasticity up to 4).
Third, the fact that about 40% of the US corporate tax is paid by foreigners (with zero welfare
weight in our analysis) makes the corporate tax reform desirable even if the government has
no redistributive tastes within US residents. With equal social marginal social welfare weights
across all income groups and an elasticity of 0.5, the net value of the reform is $7.2 billion.
Individual tax increase. The bottom panel of Table 3 considers a 10% increase in the US
federal individual income tax for taxpayers in the top 1% of the pre-tax income distribution,
35
I.e., the total amount of corporate tax revenue collected by US governments is similar to the total amount
of corporate tax paid by US households to US and foreign governments, so that allocating one aggregate or the
other makes little difference to effective tax rates by income groups.
36
The US Joint Committee on Taxation (2013) assumes that 10.8 percent of the 75% of corporate income
taxes not shifted to labor are borne by foreigners, i.e., about 8% of total federal corporate income taxes, much
lower than the 39% in our analysis. The JCT allowance for non-resident ownership is insufficient because it
only factors in portfolio investments into US stock (ignoring direct investment) and it is based on data from the
2005–2012 period (while foreign investments in US equities have been on a rising trend since then).
37
For instance one could consider putting a weight on non-US residents equal to the average marginal social
29
another commonly discussed tax reform. We use the same social welfare weights and assume
an elasticity of reported individual income with respect to the net-of-tax rate of 0.25, consistent
with the large body of work estimating behavioral responses to individual income tax changes
(see Saez, Slemrod and Giertz, 2012, Scheuer and Slemrod, 2020, for reviews). Under these
assumptions the net revenue gain is $75 billion, or 86% of the $87 billion revenue gain absent
any behavioral response. The net value of the reform is $64 billion as the reform targets the
top 1% only and hence has a low welfare cost of $11 billion. The reform remains desirable for
top incomes elasticities of up to 1.75.38
Modeling corporate tax changes. Consider first the Harberger (1962) model, where capital
can be used either in the corporate sector or in the non-corporate sector (such as unincorporated
businesses and housing) with perfect substitution. Individuals care only about the net-of-tax
return of the capital they own, not whether it is invested in the corporate vs. non-corporate
sector. Thus, the net-of-tax rates of return must be equalized across the two sectors. The
corporate tax creates a production inefficiency, because too little capital is used in the corporate
sector relative to the non-corporate sector. Formally, in the Harberger model, production is
Y1 = F 1 (K1 , L1 ) in sector 1 and Y2 = F 2 (K2 , L2 ) in sector 2 with K1 + K2 = K total capital
and L1 + L2 = L total labor. Given fixed labor and capital endowments L and K, production
efficiency maximizes F 1 (K1 , L1 ) + λF 2 (K − K1 , L − L1 ) for some λ ≥ 0, so that FK1 = λFK2 and
FL1 = λFL2 and hence FK1 /FL1 = FK2 /FL1 . A tax on the return to capital in sector 1 only (with
equal taxes on labor in both sectors) violates this condition.
The corporate tax, as modeled in Harberger (1962), violates the production efficiency the-
orem of Diamond and Mirrlees (1971). The theorem states that taxes that create production
welfare weight of US residents, assuming that foreign shareholders have the same class structure as in the United
States, and that the US government has cosmopolitan redistributive objectives.
38
This analysis is consistent with the optimal income theory of Mirrlees (1971) and in particular the optimal
top tax rate formula τ = (1 − g)/(1 − g + a · e) developed in Diamond (1998) and Saez (2001) with the elasticity
e, a ' 1.5 the Pareto parameter on the tail of the income distribution, and g the social marginal welfare weight
assigned to top earners.
39
See Auerbach (2006) for a review on the incidence of the corporate income tax.
30
inefficiencies are second-best Pareto inefficient: they can be replaced by other feasible taxes on
final consumption goods or factors of production in a Pareto-improving way. In the Harberger
model, replacing the corporate tax with a lower tax on all uses of capital can be Pareto-
improving, because it would allow for more production in both the non-corporate and corporate
sectors. This result immediately follows from the assumption of perfect mobility across sectors.
This point does not seem to have been noted in the literature, perhaps because of the gap
between applied tax analysis and theoretical optimal tax analysis. To see this point within the
context of our distributional tax-reform approach, consider a small increase of the corporate tax
rate. The distributional part of the analysis assigns the extra tax to the owners of corporations,
ignoring behavioral responses and price effects. The efficiency part of the analysis considers the
supply-side response ignoring price effects. In the Harberger (1962) model, because the supply-
side response has by assumption an infinite elasticity, the loss in tax revenue due to it swamps
any distributional gain. As long as capital in the corporate sector is taxed more than capital
outside the corporate sector, it is always desirable to lower the corporate tax rate. Conventional
tax incidence, starting with the pioneering work of Harberger (1962) fails to note this important
point because pre-tax price-effects muddy this clean conclusion that comes out of the Diamond
and Mirrlees (1971) analysis. In our view, this is a decisive advantage of optimal tax pioneered
by Diamond and Mirrlees (1971) over conventional tax incidence following Harberger (1962).40
Because an infinite elasticity for corporate capital supply is not realistic empirically, the
basic Harberger (1962) model is not well suited to welfare analysis of the corporate tax. It is,
however, easy to amend the model to make the elasticity finite as we did in Section 2.2. For
example, if individuals value directly the type of capital they own over and above the net-of-tax
return it delivers, then net-of-tax returns will not be equalized across sectors. There will be a
finite supply elasticity of corporate capital with respect to the net-of-tax rate of return, leading
to non-degenerate optimal tax analysis.
Modeling capital tax changes. The same issue applies to the analysis of the incidence of
capital taxes more generally. Section 2.2 showed that in neoclassical models, supply elasticities
are sufficient statistics for the efficiency part of the trade-off involved with a capital tax reform.
In two standard models of capital taxation, the elasticity of capital supply eK is infinite. First,
in the infinite horizon consumption model of Barro-Becker, in steady-state the net-of-tax return
r̄ has to be equal to the discount rate δ. Any deviation of r̄ from δ leads to an explosive
40
Conventional distributional analysis carried out by US agencies and inspired by the Harberger (1962) model
further muddies the waters because it is not conceptually fully consistent with the Harberger model, in particular
because of its assumption that taxes cannot change aggregate output and its composition.
31
or implosive long-run capital stock. Second, in the open-economy model where capital moves
freely and costlessly internationally but capital is taxed where it is used (as is the case with
standard territorial corporate taxes), r̄ has to be equal everywhere to the worldwide net-of-tax
return, creating again an infinite elasticity from a single (small) country perspective. As shown
in Section 2.2, these models can also be amended to generate more realistic finite elasticities.
32
simplest is to assume that investors derive utility from owning specific assets (such as munis),
in which case different tax rates on different assets can be optimal. More radically, if top wealth
generates excess power in the form of concentrated business ownership, the munis tax exemption
could be a desirable tool to induce top wealth holders to divest from their businesses and invest
in local government projects.
Corporate taxes. A number of papers find non-standard effects of the corporate tax that
operate through bargaining over the distribution of value-added within businesses. Fuest, Peichl,
and Siegloch (2018) show that municipality-level corporate tax cuts in Germany affect wages,
with workers receiving about 40% of the tax windfall. Kennedy et al. (2022) show that the large
2018 cut in the US corporate tax rate also generated earnings gains for workers but concentrated
among top 10% and especially top 1% earners with no gain for the bottom 90%. Highly paid
workers capture 32% of the corporate tax cut (Table 11 in Kennedy et al., 2022), comparable
to the German estimate but concentrated solely among highly paid workers. This suggests that
bargaining power within the firm affects how a corporate tax windfall is distributed, with strong
unions in Germany perhaps able to spread windfalls more equitably among workers. In Table 4,
rates are the same.
43
The individual income tax does not exhibit major non-standard incidence effects, except for the fact that
individuals do not have perfect understanding of the tax system (see e.g. Rees-Jones and Taubinsky (2020)).
33
we use Kennedy et al.’s estimates and assign 2/3 of a corporate tax change to profits and 1/3 to
workers. Crucially, these within-firm effects have nothing to do with the macroeconomic effect
of taxes on factor prices in classical incidence models, and hence are relevant for assessing the
direct welfare effects of a reform.44
Consumption taxes. The standard model predicts that increases vs. decreases in taxes
should have symmetric effects. This result is strikingly proven false by Benzarti et al. (2020) in
the case of the value-added tax (VAT), the major form of consumption tax worldwide, using a
comprehensive analysis of VAT reforms in Europe from 1996 to 2015. While producers can pass
almost all of a VAT increase onto consumers consistent with conventional assumptions, VAT
cuts are only half passed onto consumers and hence benefit businesses–and their workers and
suppliers. These asymmetric price effects persist several years after VAT changes take place. The
most likely explanation is that businesses can justify a price increase if there is a tax increase,
but can silently pocket a tax decrease with inattentive consumers.45 This asymmetric evidence
is based on many VAT changes in Europe and hence solidly established. For distributional
tax reform analysis, this implies that a VAT tax increase can be assigned to the corresponding
consumers as in conventional analysis, but a VAT tax decrease should be shared half between
consumers and half for businesses and their workers.
There is more uncertainty on how the tax windfall going to businesses should be split between
profits and workers. The estimates from Benzarti and Carloni (2019) for a single specific VAT
cut for restaurants in France finds that of the incidence on businesses, 75% goes to profits
and 25% to workers but it is hard to know whether such numbers generalize to other sectors
or countries as they likely depend on workers’ bargaining institutions and power.46 Hence, in
our summary Table 4, we split a VAT cut 50% to consumers, 37.5% to profits, and 12.5% to
workers with the latter two figures being highly uncertain. Because some of the VAT reforms
considered by Benzarti et al. (2020) are very sector specific (e.g., hairdressers or restaurants),
we conjecture in Table 4 that similar effects would hold for excise taxes as such taxes are like
the VAT typically built in the price posted to consumers.
44
We did not incorporate these effects in our corporate tax reform analysis of Section 5.1, because the wage
effect in the United States is highly concentrated at the top of the distribution, so that accounting for it has only
minor effects relative to assuming that the full incidence is on profits (see Kennedy et al. 2022, for a detailed
analysis).
45
See Kosonen (2015) and Harju et al. (2018) for a more detailed discussion in the context of the hairdressing
and restaurant industries in Finland showing that non-standard incidence is concentrated among smaller busi-
nesses.
46
Benzarti and Carloni (2019) find that firm owners pocketed around 55.7 percent of the VAT cut and em-
ployees received 18.6 of the VAT cut, making for a 3/4 vs. 1/4 split between profits and workers in this case.
34
The persistent asymmetric result by Benzarti et al. (2020) also shows that there is no single
equilibrium, since a cut followed by an offsetting increase in VAT rate on a specific good seem
to lead to permanently higher prices for the good (Figure 2 in Benzarti et al. 2020 provides a
striking case study for hairdressers vs. beauty salons in Finland). This radical departure from
equilibrium analysis means that the no-tax counterfactual of classical tax incidence analysis
might not even be well defined.
US sales taxes are not visible on posted prices and charged at the checkout. Empirical work
shows that they are passed to consumers symmetrically (for cuts or increases) and generally fully
(see e.g. Poterba 1996 and Besley and Rosen 1999 for empirical studies). Chetty, Looney and
Kroft (2009) show that consumers also under-react to changes in sales taxes relative to changes
in excise taxes that are included in posted prices, a relevant finding to inform distributional
tax-reform tables that we point out in Table 4.
Payroll taxes. A celebrated result in classical incidence analysis is that employer and em-
ployee payroll taxes are equivalent. In the real world, this result fails to materialize. A number
of studies compellingly show that employee payroll taxes changes affect the wage earnings of the
corresponding workers one-to-one but that employers fail to pass changes in employer payroll
taxes to the corresponding workers, likely because of wage rigidities.47 As a result, an increase
in employer payroll taxes likely reduces wages across the board and probably profits inside the
business.48 This effect, however, is not a neoclassical price effect.49 It produces relevant welfare
effects on the corresponding parties that should be tracked in the distributional tax-reform ta-
ble. It is possible that these non-standard effects persist in the long-run and are asymmetric
for increases and decreases. The studies by Saez, Schoefer, and Seim (2019) for Sweden and
47
Saez et al. (2012) show that, in Greece, uncapping employer payroll tax increases the labor cost of the
corresponding workers but uncapping the employee payroll tax does not. Bozio et al. (2022) find the same
result in France when there is no close link between employer payroll taxes and benefits. Saez, Schoefer, and
Seim (2019) find that employer payroll tax cuts for the young in Sweden do not increase their net-wages and
businesses redistribute the tax cut windfall partly across all workers. Rubolino (2022) finds a female specific
payroll tax cut does not increase their net-wages but boosts female employment and firm performance. Guillot
(2019) shows that a special temporary employer payroll tax on very high wage earners in France was mostly
borne by employers but then asymmetrically increased net wages upon expiration of the tax.
48
Conceivably, it could also increase prices of output of the business benefitting consumers. There is no direct
empirical evidence on this to date but the literature on minimum wage increases shows compellingly that part
of this extra labor cost is passed on consumers (e.g. Harasztosi and Lindner 2019 find that 75% of the minimum
wage ).
49
Bozio et al. (2022) provide a meta-analysis of 21 estimates in the literature showing that employer payroll
tax changes are not passed to corresponding workers except when there a tight and salient tax-benefit linkage.
While most of these studies tried to interpret their finds within the neo-classical supply vs. demand elasticities
framework, Bozio et al. (2022) show that non-standard effects: saliency of the link between taxes and benefits
and inequity aversion within firms are a more parsimonious way to account for the disparate empirical findings.
35
Benzarti and Harju (2021) for Finland show that payroll tax incidence happens at the firm level
rather than the individual level as in standard theory. Saez, Schoefer, and Seim (2019) show
that firms which have many workers eligible for a specific payroll tax cut on young workers
increase the wages of all their workers, not just the eligible workers, and that profits also go
up. This suggests that, within the firm, workers and profits share the tax cut or tax increase
in proportion to their share in value-added but there remains considerable uncertainty on how
such findings generalize. It is likely that sharing depends on institution and bargaining power of
workers within the firm as suggested by Kim, Kim, and Koh (2022).50 Therefore, in Table 4, we
tentatively assume that an employer payroll tax change would be borne collectively within the
firm by workers for 2/3 and by profits for 1/3.51 There is also evidence of strong employment
effects of employer payroll tax changes particularly if tax changes are targeted to a specific
group (Saez, Schoefer, and Seim 2019 for youth in Sweden, Ku, Schoenberg, and Schreiner 2020
for local changes in Norway, Benzarti and Harju 2021 for small businesses in Finland, Cottet
2022 for low wage workers in France, Rubolino 2022 for female hires in Italy, Citino and Fenizia
2022 for apprentices in Italy). But with rigid wages, such employment effects fail to generate
wage responses as predicted by standard incidence.
36
To illustrate the importance of non-standard tax incidence, consider replacing the current
head-tax mandate by a payroll tax proportional to earnings for all workers currently covered by
their employers. In 2021, about 85 million workers have employer-sponsored health insurance,
covering about 155 million non-elderly individuals (see KFF, 2021). These workers have total
pre-tax wage income of $9.0 trillion, their health insurance costs nearly $1.1 trillion, so the new
payroll tax would need to be levied at a rate of about τ = 12%.53 We assume that health
insurance would remain the same worker by worker (to focus solely on the funding aspect—
ignoring the complex issues of heterogeneity in benefits).
Conventional incidence with flexible wages. In the conventional analysis and in our neo-
classical analysis, both the current head tax and the proportional payroll tax are taxes on labor,
borne by the corresponding workers. This is also the assumption used in the recent comprehen-
sive analysis of Finkelstein et al. (2023) who consider a shift to proportional payroll tax funding.
This reform would leave labor costs (cash wage earnings plus the cost of health insurance and
other fringe benefits) unchanged for each worker. After the reform, health insurance premiums
currently paid by employers convert into extra gross cash earnings dollar for dollar and worker
by worker. All gross earnings are then reduced proportionally by a factor 1 − τ due to the new
payroll tax. Any worker with health care insurance costs above τ times her earnings benefits
from the reform (and conversely). This is a progressive reform that would have significant pos-
itive effects on the disposable income of the working and middle class, at the expense of highly
paid workers. Pre-tax incomes would not change, but post-tax incomes would become more
equal, as the proportional payroll tax replaces the head tax for covered workers. As shown by
Table 5, column 6, this reform would increase after-tax earnings of the bottom 50% by about
2.5% (and reduce after-tax income at the top by 1% to 2%), making it seemingly simple to put
US health care funding on a fairer and more sustainable path.54 Non-standard tax incidence,
however, is crucial to understand how and whether this would work in practice. Three scenarios
illustrate this point.
companies, and thus are best described as non-tax compulsory payments. We refer to them as a head tax to
highlight the similarities with standard taxes. These payments are not included in our analysis of the progressivity
of the current US tax system presented in Section 4 and Section 5.1 above.
53
We assume that the payroll tax would be charged before any other tax and not be part of the tax base
for existing payroll and income taxes, mimicking the current tax-exempt status of employer-provided health
insurance.
54
The results in Table 5 report the effects of the reform on the full population including non-workers and on
total pre-tax income (including non-wage income). Restricting the analysis to the covered population (less than
half of the total population of the United States) and wage income only would show significantly larger effects.
37
Employee payroll tax with rigid wages. Suppose first that the new payroll tax is charged
to employees. Workers would see their net earnings reduced by the new payroll tax (as workers
bear this new tax one for one, cf. Table 4). The elimination of the head tax is akin to an
employer payroll tax cut. As explained in Table 4, because of wage rigidities, this payroll tax
cut would not be passed one-to-one to the corresponding workers but instead passed roughly
2/3 to workers across the board within each firm (proportionately to their wages) and 1/3 to
profits, according to existing studies. Under these assumptions, the reform becomes regressive
as illustrated in Table 5. Both pre-tax and post-tax incomes become more unequal.
Employer payroll tax with rigid wages. Suppose now that the new payroll tax is charged
to employers. The head tax–the current insurance premiums paid by employers–becomes an
employer payroll tax. Because the amounts are the same, there is no tax savings or costs for
employers. Wage rigidities imply that net earnings do not change but pre-tax labor income
becomes more unequal: labor costs for each worker change by the difference between the new
payroll tax and the former head tax.55 The reform again fails to make the progressive gains of
the conventional analysis. But it is more progressive than the previous scenario as none of the
savings made by employers goes to profits.56
With non-standard incidence effects of this kind, labor costs for workers change so that
there could be employment effects due to labor demand responses of employers (cf. Table 4).
In our setting, as labor costs for low-paid workers fall, they could become more attractive
to employers, boosting employment at the low end. In a competitive standard labor market
model, such demand effects lead to wage responses generating the conventional incidence results.
But with rigid wages, such responses may be sluggish and incomplete, as shown by empirical
evidence. There remains considerable uncertainty, and hence need for more research on how
quickly such wage adjustments would take place.57
55
This is the most plausible incidence in light of the studies analyzed above. Because there is a linkage between
the head tax and health care benefits, the analysis of Bozio et al. (2022) suggests that the incidence passed on to
workers individually (Gruber 1994 and Baicker and Chandra 2006 present US based analysis of health premiums
changes consistent with this). Therefore, it is possible that wages would not be completely rigid and that the
incidence of removing the head tax would eventually shift back to workers as in the standard incidence model.
56
In the employee payroll tax scenario, if workers have a lot of power and can recoup 100% of the saving
(instead of just 2/3), then it is likely that workers would insist on a proportional-to-earnings compensation to
offset the payroll tax. In this case, the incidence is the same in the employee vs. employer payroll tax funding
cases, but this equivalence depends crucially on strong worker bargaining power (instead of being the standard
consequence of competitive markets).
57
The classic study by Gruber (1994) on mandated maternity benefits in some US states found that wages of
child bearing women adjusted within a few years.
38
Directed tax incidence with rigid wages. Last, how can incidence be steered toward the
equilibrium of the conventional analysis in the real world with wage rigidities? As proposed
by Saez and Zucman (2019), existing employer-provided benefits could be converted one-for-
one into a permanent wage increase worker-by-worker by law. This would leave labor costs for
employers unchanged worker by worker. A new payroll tax on employees should then be created
at rate τ as in the first scenario. This tax would fall on the corresponding employees. It would
recreate the exact conventional incidence. The key difference with conventional incidence is that
the equilibrium would be reached by legislation rather than through competitive market forces.
7 Conclusion
Who pays taxes, and how tax reforms would affect the different socio-economic groups, are
some of the most important public-interest questions in democratic societies. The official prac-
tice currently used to inform the public about these questions, which has evolved somewhat
inorganically from foundational scholarship developed in the middle of the 20th century, has
shortcomings. We attempt to address these issues by founding distributional tax analysis on
modern optimal tax theory.
Two main lessons emerge from this work. First, it is possible to do conceptually consistent
and practically relevant current-tax analysis that does not merely follow statutory incidence but
rather follows economic reasoning and yet does not require to specify behavioral responses. This
analysis assigns taxes to individuals simply—labor taxes to the corresponding workers, capital
taxes to the corresponding owners, consumption taxes to consumers—as one writes a model of
optimal taxation. The tax rates are the wedges between pre-tax prices (relevant for production)
and after-tax prices (relevant for work, savings, and consumption decisions of households).
This method maximizes the comparability of tax progressivity over time and across countries,
regardless of differences in the legal tax structure and the form of business organization. Classical
tax incidence analysis is not required to study the distribution of current taxes.
Second and more deeply, classical incidence analysis also turns out to be largely irrelevant for
the distributional analysis of tax reforms. This is because the effect of taxes on pretax prices at
the heart of classical tax incidence are normatively irrelevant in optimal tax models. Moreover,
the recent applied literature on behavioral responses to taxes uncovers effects that are far from
those captured by classical incidence. Instead, the existence of a hierarchy of responses (with
tax avoidance coming first), asymmetries (tax cuts having different effects than tax increases),
intra-firm bargaining effects, and wage rigidities appear to be key. Additional work needs to be
39
carried out to understand the nature of these non-standard effects. If they generalize to other
contexts, additional theories need to be developed, for example to account for the asymmetry
of VAT tax changes and to clarify how to measure the resulting welfare effects.
In the meantime, government agencies should stop shifting taxes for the analysis of the cur-
rent tax system. Practically in the US, this concerns solely the corporate income tax which
should simply be assigned to shareholders. To study the economic effects of small tax reforms,
the analysis should focus on the relevant supply elasticities and the non-standard effects uncov-
ered by the modern empirical literature. This approach, dramatically simpler than the current
practice and yet more consistent, would improve the quality of the official information available
to lawmakers and the public about the progressivity of the tax system, historical changes in
that progressivity, and the potential effect of proposed tax reforms—core topics of democratic
interest.
References
Aaberge, Rolf, Magne Mogstad, Ola L. Vestad, and Arnstein Vestre. 2021. “Okonomisk
ulikhet i Norge i det 21. arhundre,” Statistics Norway working paper #2021/33.
Advani, Arun, Helen Hughson, and Andy Summers. 2023. “How much tax do the rich
really pay? Evidence from the UK,” Oxford Review of Economic Policy, 39(3): 406–437.
Atria, Jorge and Cristbal Otero (eds). 2021. Impuestos Justos para el Chile que Viene:
Diagnstico y Desafos Tributarios para un Nuevo Pacto Fiscal. Fondo de Cultura Econmica.
Auerbach, Alan. 1985. “The theory of excess burden and optimal taxation,” in Handbook of
Public Economics, Vol. 1, Elsevier, 61-127.
Auerbach, Alan. 2006. “Who Bears the Corporate Tax?” in J. Poterba, ed., Tax Policy and
the Economy 20, 1–40.
Bach, Laurent, Antoine Bozio, Arthur Guillouzouic, and Clément Malgouyres. 2023.
“Do Billionaires Pay Taxes?” Paris School of Economics/IPP Working Paper.
Bachas, Pierre, Anders Jensen, Matthew Fisher-Post, and Gabriel Zucman. 2023.
“Globalization and Factor Income Taxation,” NBER working paper #29819.
Baicker, Katherine and Amitabh Chandra. 2006. “The Labor Market Effects of Rising
Health Insurance Premiums,” Journal of Labor Economics 24 (3), 609–634.
Balkir, Akcan, Emmanuel Saez, Danny Yagan, and Gabriel Zucman. 2023. “How
Much Tax Do US Billionaires Pay? New Evidence from Administrative Data”, work in prepa-
ration, UC Berkeley.
Barthold, Thomas A. 1993. “Measuring the Burden of Tax Changes: What Do We Tell
Congress?” Proceedings of the Annual Conference on Taxation of the National Tax Association,
86, 164-168.
Benzarti, Youssef, and Dorian Carloni. 2019. “Who really benefits from consumption tax
cuts? Evidence from a large VAT reform in France.” American Economic Journal: Economic
Policy, 11(1): 38–63.
Benzarti, Youssef, Dorian Carloni, Jarkko Harju, and Tuomas Kosonen. 2020. “What
goes up may not come down: asymmetric incidence of value-added taxes.” Journal of Political
Economy 128, no. 12: 4438-4474.
Benzarti, Youssef, and Jarkko Harju. 2021. “Using payroll tax variation to unpack the
black box of firm-level production.” Journal of the European Economic Association 19, no. 5:
2737-2764.
Besley, Timothy J., and Harvey S. Rosen. 1999. “Sales taxes and prices: an empirical
analysis.” National tax journal 52, no. 2: 157-178.
Blanchet, Thomas, Lucas Chancel, Ignacio Flores, Marc Morgan. 2021. “Distri-
butional National Accounts Guidelines: Methods and Concepts used in the World Inequality
Database.” Technical paper, 2021.
41
Blanchet, Thomas, Lucas Chancel, and Amory Gethin. 2022. “Why is Europe more
Equal than the United States?” American Economic Journal: Applied Economics, 14(4): 480–
518.
Bozio, Antoine, Thomas Breda, Julien Grenet, and Arthur Guillouzouic 2022. “Does
Tax-Benefit Linkage Matter for the Incidence of Social Security Contributions?.” Paris School
of Economics Working Paper.
Brittain, John A. 1971. “The Incidence of Social Security Payroll Taxes.” American Eco-
nomic Review 61(1), 110-125.
Bruil, Arjan, Celine van Essen, Wouter Leenders, Arjan Lejour, Jan Muhlmann,
and Simon Rabate. 2022. “Inequality and Redistribution in the Netherlands,” CPB working
paper.
Carloni, Dorian, and Terry Dinan. 2021. “Distributional Effects of Reducing Carbon
Dioxide Emissions With a Carbon Tax”, Congressional Budget Office Working Paper 2021-11.
No. 57399.
Case, Ann and Angus Deaton. 2020. Deaths of Despair and the Future of Capitalism.
Princeton University Press.
Chancel, Lucas. 2022. “Global carbon inequality over 1990–2019.” Nature Sustainability 5,
no. 11: 931-938.
Chancel, Lucas and Yannic Rehm. 2023. “The Carbon Footprint of Capital: Evidence
from France, Germany and the US based on Distributional Environmental Accounts,” PSE
Working Paper.
Chetty, Raj, Adam Looney, and Kory Kroft. 2009. “Salience and taxation: Theory and
evidence.” American Economic Review, 99(4): 1145–1177.
Citino, Luca and Alessandra Fenizia. 2022. “The Effectiveness of Targeted Payroll Tax
Reductions”, Working Paper.
Clemens Fuest, Andreas Peichl, and Sebastian Siegloch. 2018. “Do higher corporate
taxes reduce wages? Micro evidence from Germany.” American Economic Review 108(2):
393–418.
Colm, Gerhard, and Helen Tarasov. 1941. “Who pays the taxes.” Temporary National
Economic Committee, Washington, DC.
Cottet, Sophie. 2022. “Payroll tax reductions for minimum wage workers: Relative labor
cost or cash windfall effects?” Paris School of Economics Working Paper.
Cronin, Julie-Anne, Emily Lin, Laura Power and Michael Cooper. 2013. “Distributing
the Corporate Income Tax: Revised U.S. Treasury Methodology,” National Tax Journal, 66(1),
239-262.
Cronin, Julie-Anne. 2022. “U.S. Treasury Distributional Analysis Methodology.” Washing-
ton, DC: US Treasury. Office of Tax Analysis Technical Paper 8. Online at https://home.
42
treasury.gov/system/files/131/TP-8.pdf
De Rosa, Mauricio, Ignacio Flores, and Marc Morgan. 2022. “More Unequal or Not as
Rich? Revisiting the Latin American Exception,” CUNY Stone Center Working Paper Series.
Diamond, Peter A. 1998. “Optimal income taxation: an example with a U-shaped pattern
of optimal marginal tax rates.” American Economic Review 88(1): 83-95.
Diamond, Peter A., and James A. Mirrlees. 1971. “Optimal Taxation and Public Pro-
duction I: Production Efficiency; II:Tax Rules.” American economic review 61, no. 1: 8-27 and
61, no. 3: 261-278.
Dray, Sacha, Camille Landais, and Stefanie Stantcheva. 2023. “Wealth and Property
Taxation in the United States,” NBER working paper #31080.
Eisinger, J., Ernsthausen, J., Kiel, P. 2021. “The secret IRS files: trove of never-before-
seen records reveal how the wealthiest avoid income tax.” ProPublica, June 8, 2021.
Eurostat. 2021. “Taxation Trends in the European Union,” 2021 edition, Directorate-General
for Taxation and Customs Union, European Commission, Publications Office of the European
Union, Luxembourg.
Feldstein, Martin S. 1974. “Tax incidence in a growing economy with variable factor supply.”
Quarterly Journal of Economics 88, no. 4: 551-573.
Feldstein, Martin. 1999. “Tax avoidance and the deadweight loss of the income tax,” Review
of Economics and Statistics, 81(4): 674–680.
Finkelstein, Amy, Casey C. McQuillan, Owen M. Zidar, and Eric Zwick. 2023. “The
Health Wedge and Labor Market Inequality.” NBER Working Paper No. 31091, forthcoming
Brookings Papers on Economic Activity.
Fullerton, Don, and Gilbert E. Metcalf. 2002. “Tax Incidence” in Handbook of Public
Economics, Volume 4, A. J. Auerbach and M. Feldstein, eds. (Amsterdam: North-Holland),
1787-1872.
Gruber, Jonathan. 1994. “The Incidence of Mandated Maternity Benefits.” American
economic review 84(3), 622-641.
Guillot, Malka. 2021. “Tax Incidence among the Working Superrich. Evidence from the
French 75% Tax on Millionaires. Evidence From the French 75% Tax on Millionaires”, ETH
Zurich Working Paper.
Guyton, John, Patrick Langetieg, Daniel Reck, Max Risch, and Gabriel Zucman.
2021. “Tax Evasion at the Top of the Income Distribution: Theory and Evidence. NBER
Working Paper No. 28542.
Guzzardi, Demetrio, Elisa Palagi, Andrea Roventini, and Alessandro Santoro. 2022.
“Reconstructing Income Inequality in Italy: New Evidence and Tax Policy Implications from
Distributional National Accounts,” World Inequality Lab Working Paper 2022/02.
Harberger, Arnold C. 1962. “The Incidence of the Corporation Income Tax.” Journal of
43
Political economy 70, no. 3: 215-240.
Harberger, Arnold C. 1964. “The Measurement of Waste.” American Economic Review
54(3): 58–76.
Harju, Jarkko, Tuomas Kosonen, and Oskar Nordström Skans. 2018. “Firm types,
price-setting strategies, and consumption-tax incidence.” Journal of Public Economics 165:
48-72.
Hendren, Nathaniel, and Ben Sprung-Keyser. 2020. “A Unified Welfare Analysis of
Government Policies.” Quarterly Journal of Economics 135 (3): 1209-1318.
Institute on Taxation and Economic Policy. 2018. Who Pays: A Distributional Analysis
of the Tax Systems in All 50 States. 6th ed., Washington DC: ITEP.
Institute on Taxation and Economic Policy. 2019. Who Pays Taxes in America in 2019?,
Washington DC: ITEP.
Kennedy, Patrick, Christine Dobridge, Paul Landefeld, and Jacob Mortenson. 2022.
“The Efficiency-Equity Tradeoff of the Corporate Income Tax: Evidence from the Tax Cuts and
Jobs Act.”
KFF. 2021. “2021 Employer Health Benefit Survey,” available online at https://www.kff.
org/report-section/ehbs-2021-summary-of-findings
Kim, Jinyoung, Seonghoon Kim, and Kanghyock Koh. 2022. “Labor market institutions
and the incidence of payroll taxation.” Journal of Public Economics 209: 104646.
Kosonen, Tuomas. 2015. “More and cheaper haircuts after VAT cut? On the efficiency and
incidence of service sector consumption taxes.” Journal of Public Economics 131: 87-100.
Kotlikoff, Laurence J., and Lawrence H. Summers. 1987. “Tax Incidence.” In Handbook
of public economics, vol. 2, 1043-1092. Elsevier: North Holland.
Krakower, Howard, Jennifer Lee, Kate Pinard, and Marlyn Rodriguez. 2021. “Pro-
totype NIPA Estimates of Profits for S Corporations.” Bureau of Economic Analysis.
Ku, Hyejin, Uta Schoenberg, Ragnhild C. Schreiner. 2020. “Do place-based tax incen-
tives create jobs?” Journal of Public Economics 191, 2020, 104105,
Mankiw, Gregory. 2001. “Commentary: Balanced-Budget Restraint in Taxing Income From
Wealth in the Ramsey Model.” In Inequality and Tax Policy edited by K. A. Hassett and R.
G. Hubbard. Washington, DC: AEI.
Mendoza, Enrique G, Assaf Razin, and Linda L. Tesar. 1994. “Effective Tax Rates in
Macroeconomics: Cross-Country Estimates of Tax Rates on Factor Incomes and Consumption,”
Journal of Monetary Economics 34: 293–323.
Mirrlees, James A. 1971. “An Exploration in the Theory of Optimum Income Taxation.”
Review of Economic Studies 38(2), 175-208.
Musgrave, Richard A., John J. Carroll, Lorne D. Cook, and Lenore Frane. 1951.
“Distribution of tax payments by income groups: a case study for 1948.” National Tax Journal
44
4(1): 1-53.
Nunns, James. 2012. “How TPC distributes the corporate income tax.” Urban-Brookings
Tax Policy Center, Washington, DC,
Pechman, Joseph A., and Benjamin A. Okner. 1974. Who Bears the Tax Burden?,
(Washington D.C: Brookings Institution).
Piketty, Thomas. 2000. “Theories of persistent inequality and integenerational mobility.”
Handbook of Income Distribution volume 1, 429-476.
Piketty, Thomas, and Emmanuel Saez. 2013. “A Theory of Optimal Inheritance Taxa-
tion,” Econometrica 81(5), 1851-1886.
Piketty, Thomas, Emmanuel Saez, and Gabriel Zucman. 2018. “Distributional national
accounts: methods and estimates for the United States.” Quarterly Journal of Economics 133,
no. 2: 553-609.
Poterba, James M. 1996. “Retail price reactions to changes in state and local sales taxes.”
National Tax Journal 49, no. 2: 165-176.
Ramsey, Frank P. 1927. “A Contribution to the Theory of Taxation.” Economic journal 37,
no. 145: 47-61.
Rees-Jones, Alex, and Dmitry Taubinsky. 2020. “Measuring “Schmeduling”.” Review of
Economic Studies 87, no. 5: 2399-2438.
Rosenthal, Steven, and Theo Burke. 2020. “Who’s Left to Tax? US Taxation of Corpo-
rations and Their Shareholders.” Urban-Brookings Tax Policy Center working paper.
Rubolino, Enrico. 2022. “Taxing the Gender Gap: Labor Market Effects of a Payroll Tax
Cut for Women in Italy”, CESifo Working Paper No. 9671.
Saez, Emmanuel. 2001. “Using Elasticities to Derive Optimal Income Tax Rates.” Review
of Economic Studies68, no. 1: 205-229.
Saez, Emmanuel, Manos Matsaganis, and Panos Tsakloglou. 2012. “Earnings Deter-
mination and Taxes: Evidence from a Cohort-Based Payroll Tax Reform in Greece.” Quarterly
Journal of Economics 127 (1): 493–533.
Saez, Emmanuel, Benjamin Schoefer, and David Seim. 2019. “Payroll Taxes, Firm
Behavior, and Rent Sharing: Evidence from a Young Workers’ Tax Cut in Sweden.” American
Economic Review
Saez, Emmanuel, Joel Slemrod, and Seth H. Giertz. 2012. “The Elasticity of Tax-
able Income with Respect to Marginal Tax Rates: A Critical Review.” Journal of Economic
Literature 50(1): 3–50.
Saez, Emmanuel, and Stefanie Stantcheva. 2018. “A simpler theory of optimal capital
taxation.” Journal of Public Economics 162: 120-142.
Saez, Emmanuel and Gabriel Zucman. 2016.“Wealth Inequality in the United States since
1913: Evidence from Capitalized Income Tax Data,” Quarterly Journal of Economics 131(2),
45
519-578.
Saez, Emmanuel and Gabriel Zucman. 2019. The Triumph of Injustice: How the Rich
Dodge Taxes and How to Stop Them. W.W. Norton.
Scheuer, Florian and Joel Slemrod. 2020. “Taxation and the Superrich.” Annual Review
of Economics, 12:189–211.
Slemrod, Joel. 1995. “Income Creation or Income Shifting? Behavioral Responses to the Tax
Reform Act of 1986.” American Economic Review 85(2), 175-80.
Straub, Ludwig, and Ivan Werning. 2020. “Positive long-run capital taxation: Chamley-
Judd revisited.” American Economic Review 110, no. 1: 86-119.
Tax Policy Center. 2022 “Brief Description of the Tax Model”, Washington DC, updated
March 9, 2022, online at https://www.taxpolicycenter.org/resources/brief-description-tax-model
US Congressional Budget Office. 2012. “The Distribution of Household Income and Federal
Taxes, 2008 and 2009.” Washington, DC: Government Printing Office. Online at https:
//www.cbo.gov/publication/43373
US Congressional Budget Office. 2018. “The Distribution of Household Income, 2016.”
Washington, DC: Government Printing Office. Online at https://www.cbo.gov/publication/
55413
US Joint Committee on Taxation. 1993. “Methodology and Issues in Measuring Changes in
the Distribution of Tax Burdens,” JCS-7-93. Online at https://www.jct.gov/publications/
1993/jcs-7-93/.
US Joint Committee on Taxation. 2013. “Modeling the Distribution of Taxes on Business
Income,” JCX-14-13, available online here.
US Joint Committee on Taxation. 2019. “Revenue Estimating Methodology.” Washington,
DC: Government Printing Office.
US Treasury. Office of Tax Analysis. 2019. “Distributional Analysis of the US Tax
System.” Washington, DC: Government Printing Office. Online at https://home.treasury.
gov/policy-issues/tax-policy/office-of-tax-analysis
Wolff, Edward N., and Ajit Zacharias. 2007. “The distributional consequences of govern-
ment spending and taxation in the US, 1989 and 2000.” Review of Income and Wealth 53(4),
692–715.
Zucman, Gabriel. 2023. “Cross-Border Fiscal Flows,” work in progress.
46
Figure 1: General Equilibrium with Capital Tax
𝑘(𝑟)̅
𝑤
𝑟
Tax =( 𝑟̅ − 𝑟) ⋅ 𝑘
𝑟*
= r ⋅ 𝜏! ⋅ k
𝑟̅ = 𝑟 ⋅ (1 − 𝜏! )
𝑘 " 𝑟 from
𝑟 = 𝑓′(𝑘)
𝑘 𝑘* 𝐾
𝑘=
𝐿
Notes: The figure depicts the effect of a tax on capital income at rate τK on the interest rate r, the capital
to labor ratio k = K/L, and the wage w in a general equilibrium neoclassical model with fixed labor L, CRS
production F (K, L) = L · F (K/L, 1) = L · f (k). The equilibrium is characterized by 3 equations: (1) r = f 0 (k)
(rate of return of capital equals its marginal return which generates the demand for capital k d (r)), (2) k = k(r̄)
Rk
(capital supply depends on its net of tax return r̄ = r(1 − τK )), (3) w = f (k) − kf 0 (k) = 0 f 0 (κ)dκ − rk
(the wage w can be read as the area below the demand curve and above the r horizontal line). Without taxes,
the equilibrium is (r∗ , k ∗ ). With a tax rate τK , the equilibrium shifts to (r, k). The tax collects the rectangle,
(r − r̄)k = τK rk, it increases r, and reduces r̄ and w. The tax reduces the wage and the surplus of capitalists
by an excess burden triangle ' (1/2) · rτK · (k ∗ − k) over and above taxes collected. In this economy, pre-tax
labor income is wL, pre-tax capital income is rK, and post-tax capital income is r(1 − τK )K.
47
Figure 2: Capital Tax Reform and Optimum
𝑘(𝑟)̅
𝑤 𝑟 − 𝑟̅ 𝑑𝑘
𝑟 + 𝑑𝑟
−𝑑𝑤 = 𝑘𝑑𝑟
𝑟 Optimum tax maximizes
w + r ⋅ 𝜏! ⋅ k = 𝑓 𝑘 − 𝑟𝑘
̅
⇒ 𝑟 − 𝑟̅ 𝑑𝑘 − 𝑘𝑑 𝑟̅ = 0
𝑟̅
−𝑘𝑑𝑟̅ ⇒ 𝜏! = 1/(1+ 𝑒! )
𝑟̅ + 𝑑 𝑟̅
𝑘" 𝑟
𝑘 + 𝑑𝑘 𝑘 𝐾
𝑘=
𝐿
Notes: The figure depicts the effect of a change dτK in the capital income tax rate τK in the simple neoclassical
model depicted on Figure 1. The tax change reduces capital k by dk < 0, increases the pre-tax rate of return
r by dr, reduces the net-of-tax rate r̄ by dr̄ < 0. If the government wants to maximize the welfare of workers,
it sets τK to maximize w + τK rk (wages plus tax revenue extracted from capitalists). As w = f (k) − kf 0 (k),
we have w + τK rk = f (k) − r̄k, the area below the demand curve r = f 0 (k) and above the horizontal line r̄
(the blue areas in Figure 1). The first order condition for the optimum is (f 0 (k) − r̄)dk − kdr̄ = 0 (the 2 blue
rectangles on the Figure cancel out at the optimum). As f 0 (k) = r, this can be rewritten as (r − r̄)dk/dr̄ = k
∗
or (r − r̄)/r̄ = 1/eK which is the classical inverse elasticity rule τK = 1/(1 + eK ) where eK = (r̄/k)dk/dr̄ is
the pure supply side elasticity. Therefore the classical pre-tax price incidence dr, dw is irrelevant for optimal tax
analysis, a result that generalizes to any social welfare function as shown in Diamond and Mirrlees (1971).
48
Figure 3: Changes in Tax Progressivity in the United States
40%
35%
30%
25%
20%
15%
All
10%
5%
1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 2020
Estate taxes
50%
40%
30%
Individual income taxes
20%
Corporate taxes
10%
Notes: The top panel reports average effective tax rates for the US population as a whole and for the top 1%
of the pre-tax income distribution. To construct income groups, the unit of observation is the adult individual
(aged 20 or above), and adults are ranked by their pre-tax national income, with income equally split between
married spouses. All taxes at all levels of government are included in the numerator, and all pre-tax national
income is included at the denominator. Pure realized capital gains (defined as realized gains in excess of 3% of
national income) are included in pre-tax income. The bottom panel shows the effective tax rate of the top 0.1%
of the pre-tax income distribution similarly defined, with a decomposition by type of tax. “Corporate taxes”
include both federal and state corporate taxes and business property taxes. “Individual income taxes” include
both federal and state individual income taxes and payroll taxes.
Figure 4: Corporate Tax Revenue (% of National Income)
8%
6%
5% S-corporations profits
4%
3%
2%
Income tax on
1% S-corporation profits
0%
1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 2020
Notes: The figure plots the evolution of (federal plus state) US corporate income tax revenue and of S-corporation profits, both as a fraction of US national
income. S-corporation profits are taken from the prototype BEA estimates of S-corporation profits in US national income (Krakower et al., 2021, updated),
which cover the years 2012–2018, and are estimated by us using similar methods before 2012 and after 2018. Taxes on S-corporation profits are estimated
by applying the effective average income tax rate on ordinary income (i.e., income excluding capital gains) to reported S-corporation profits, separately for
the top 1% and the bottom 99% of the fiscal income distribution.
Figure 5: Allocating the Corporate Tax
80%
70%
Distributional current-tax analysis
60%
50%
40%
20%
1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 2020
25%
Distributional current-tax analysis
20%
15%
10%
5%
Conventional approach
0%
1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 2020
Notes: The top panel contrasts the share of the US corporate income tax (federal and state) paid by the
top 1% adults with the highest pre-tax national income in our methodology and the CBO methodology. The
CBO methodology assigns 75% of the corporate tax to capital owners nationally (proportionally to reported
dividends, interest, rents, and a normalized measure of capital gains) and 25% to workers nationally. Our
current-tax methodology assigns 100% of the corporate tax to the corresponding shareholders individually. The
bottom panel plots the amount of corporate tax paid by the top 1% (as a fraction of top 1% pre-tax income) in
the two methodologies. The allocation of the corporate tax does not make a significant difference in the early
20th century and since the 1980s (when the corporate tax overall is small), but makes a significant difference in
the middle of the 20th century (when the corporate tax was high). Both the CBO methodology and our approach
distribute only the amount of US corporate tax collected by US governments (i.e., make the implicit assumption
that US residents pay in foreign corporate tax as much as what foreigners pay in US corporate taxes).
Table 1: Illustration of Current-Tax analysis: Case Studies (2018)
Warren
Millions of US$ Jeff Bezos
Buffett
US federal taxes 43 930
Individual income tax 43 5
Corporate tax 0 925
Payroll taxes 0 0
Consumption taxes 0 0
US state and local income taxes 140 241
Individual income tax 0 1
Corporate taxes 70 53
Business property taxes 69 187
Consumption taxes ~0 ~0
Residential preoperty taxes ~0 ~0
Foreign taxes 154 337
Corporate taxes 123 337
Business property taxes 31 0
Total taxes 337 1,508
Pre-tax income 2,221 8,176
Effective tax rate 15.2% 18.4%
Federal 1.9% 11.4%
State and local 6.3% 2.9%
Foreign 6.9% 4.1%
Notes: See text for complete sources and details. Corporate taxes paid are equal to global cash tax payments
reported by Amazon and Berkshire Hathaway in their SEC 10-K reports, apportioned by the ownership stake of
Bezos and Buffett respectively. No geographical breakdown of cash taxes paid is available. We use the published
breakdown of provisions for current taxes (Amazon) and provisions for current plus deferred taxes (Berkshire
Hathaway) to allocate these cash payments to federal vs. state and local vs. foreign governments. Property
taxes are computed as 1% of net property and equipment, and allocated to US state and local governments vs.
foreign governments based on the geographical location of assets reported in the 10-K item 2. Individual income
taxes are taken from Eisinger et al. (2021) for federal taxes and based on public information about state of
residency for state and local taxe. State and local consumption and residential property taxes are assumed to
be negligible relative to income. Income is equal to the apportioned share of Amazon and Berkshire Hathaway’s
pre-tax profits (excluding unrealized gains on investments, and adding imputed business property taxes) plus
any individual income (e.g., realized capital gains, wages, income from other investments) identified in Eisinger
et al. (2021).
Table 2: Current Tax Distribution in the United States, 2021
Pretax income After-tax income Taxes (all levels) Tax rate composition
Individual Property Memo: Corporate
Income Payroll Consump- Corporate
Average Share Average Share Share Tax rate income taxes (incl. tax, conventional
groups taxes tion taxes tax
taxes estate tax) approach
P0-50 $20,889 12.3% $15,526 13.0% 10.7% 25.7% 2.2% 10.7% 10.5% 1.7% 0.6% 1.1%
P50-90 $80,618 38.1% $57,498 38.6% 36.9% 28.7% 8.6% 10.3% 5.6% 2.7% 1.4% 1.1%
P90-99 $243,587 25.9% $170,579 25.8% 26.2% 30.0% 14.7% 6.3% 3.5% 3.5% 2.1% 1.8%
P99-99.9 $1,085,455 11.5% $741,550 11.2% 12.3% 31.7% 20.8% 2.4% 2.2% 3.8% 2.5% 2.8%
top 0.1% $10,288,542 12.2% $6,804,921 11.4% 13.9% 33.9% 22.8% 0.8% 1.8% 5.1% 3.2% 4.1%
All $84,672 100% $59,593 100% 100% 29.6% 12.5% 7.3% 4.8% 3.2% 1.8% 1.8%
Notes: Groups are based on pre-tax national income plus pure realized capital gains (defined as realized gains in excess of 3% of national income). Unit is
individual adult (aged 20+) with equal split of income among couples. Pre-tax income is income before all taxes but after the operation of pension systems
(public and private). Taxes include taxes at all levels of government (federal, state, and local). Refundable tax credits are not included as negative tax (as
they are treated as transfers, like other cash transfers, in the national accounts). Labor taxes are assigned to the corresponding workers, capital taxes to the
corresponding asset owners, consumption taxes to the corresponding final consumers. In the conventional approach currently used by CBO, the corporate
tax is assigned 75% to capital income reported on individual tax returns and 25% to labor income (with no adjustment for corporate profits earned through
pension funds). The current tax distribution for federal taxes only (excluding state, local, and foreign taxes) is presented in appendix Table A1.
Table 3: Distributional Tax-Reform Analysis: Applications
A. Reform of the US federal corporate income tax
Notes: Groups are based on pre-tax national income including pure realized capital gains. Unit is individual
adult (aged 20+) with equal split of income among couples. The top panel considers 10% increase in the federal
corporate income tax while the bottom panel considers a 10% increase in the federal individual income tax for
the top 1%. In the top panel, column (2) includes all corporate taxes (US, state, and foreign) paid by US
residents on their corporate ownership (in US and abroad). Column (3) includes only the federal corporate tax,
close to 40% of which is paid by non-resident owners of US corporations. For the reform analysis, the tax loss
due to supply side responses is computed assuming an elasticity of corporate profits of 0.5 in the top panel and
a top 1% reported income elasticity of 0.25 in the bottom panel. We use a marginal tax rate of 30% for top
1% individuals in the current federal individual income tax (as top 1% fiscal incomes include ordinary income
and tax preferred business income, dividends, and capital gains). In both cases, we assume a simple pattern of
social marginal welfare weights declining geometrically as one moves up the income distribution: the weight on
the top 0.1% is half the weight on the next 0.9%, which is half the weight on the next 9%, etc. The bottom
of each table shows the aggregate net revenue gain (mechanical tax increase minus tax loss due to behavioral
responses) and the net value of the reform (net revenue minus the social welfare cost which is the mechanical
tax increase weighted by the social welfare weights). A positive net value implies that the reform is desirable.
As discussed in the main text, we ignore pretax price effects because such effects are normatively irrelevant (i.e.,
can be neutralized at zero fiscal cost by adjusting labor and capital taxes).
Table 4: Lessons from the Modern Literature on Non-Standard Tax Incidence
Lessons from the Modern Literature on Non-Standard Tax Incidence
Corporate income tax Profits 2/3* Fuest, Peichl, and Siegloch (2018) for Germany and Avoidance/evasion Varies with context, can be large
Workers 1/3* Kennedy et al. (2022) for the US. Likely depends on Real responses Likely medium, varies with design
bargaining power. Asymmetric effects?
Consumers 0%*
Consumption taxes
Value-added-tax or excise tax increase Consumers 100% Benzarti et al. (2020) on VAT in Europe Evasion Varies with context, can be large
Consumer demand Larger response for tax on specific goods
Value-added-tax or excise tax decrease Consumers 50% Benzarti et al. (2020) on VAT in Europe Consumer demand Response muted by 50% price passthrough
Profits 37.5%* Benzarti and Carloni (2019). Likely depends on
Workers 12.5%* bargaining power
Sales taxes (not posted on prices) Consumers 100% Consistent with conventional incidence. Poterba Evasion Can be large for small retailers
(1996) and Besley and Rosen (1999) for local sales Consumer demand response Muted by inattentiveness (Chetty et al. 2009)
tax in the US
Payroll taxes
Employee side payroll tax Workers 100% Consistent with conventional incidence Labor supply response
Likely small (higher for less attached subgroups)
Employer side payroll tax Corresponding workers 0% Saez et al. (2012) for Greece, Bozio et al. (2022) for Employer labor demand Can be large for targeted tax changes
France, Saez et al. (2019) for Sweden responses
Workers collectively 2/3* Saez et al. (2019) for Sweden, Benzarti and Harju
Profits 1/3* (2021) for Finland. Likely depends on bargaining
Consumers 0%* power. Asymmetric effects?
Notes: Column 1 reports who bears the burden of a tax change with some explanatory notes and key references in column 2. A * denotes large uncertainty
in the estimate, and therefore where further research would be most valuable. The table ignores any neoclassical pre-tax price effects as they are normatively
irrelevant and hard to compelling estimate empirically. Therefore, incidence is always within a production unit (such as a firm) on how surplus is shared
among stakeholders in the unit (owners profits, workers earnings, and consumers’ prices). Column 3 lists the most important behavioral responses with
some notes on magnitudes in col. 4. “Small” means elasticity of the tax base with respect to the net-of-tax rate in range (0,.25), “medium” in range (.25,.5),
“large” is .5 or more. See text for more detailed justifications and more nuanced explanations.
Table 5: Replacing Employer-Provided Health Insurance Contributions By a Payroll Tax
Current system Reform replacing current employer health care contributions by flat 11.8% payroll tax
Conventional incidence and directed
Employee payroll tax with rigid wages Employer payroll tax with rigid wages
incidence
Change in Change in Change in
Current
Current New payroll % change in after-tax New payroll % change in after-tax New payroll % change in after-tax
Income Average pre- head tax
head tax tax (% pre-tax pre-tax income (% tax (% pre-tax pre-tax income (% tax (% pre-tax pre-tax income (%
groups tax income (% pre-tax
($ per adult) income) income pre-tax income) income pre-tax income) income pre-tax
income)
income) income) income)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
P0-50 $20,889 $1,440 6.9% 4.5% 0.0% 2.4% 4.5% -3.3% -0.9% 4.5% -2.4% 0.0%
P50-90 $80,618 $6,505 8.1% 7.0% 0.0% 1.1% 7.0% -2.1% -1.0% 7.0% -1.1% 0.0%
P90-99 $243,587 $7,826 3.2% 5.2% 0.0% -1.9% 5.2% 2.1% 0.2% 5.2% 1.9% 0.0%
P99-99.9 $1,085,455 $6,212 0.6% 2.7% 0.0% -2.1% 2.7% 3.5% 1.4% 2.7% 2.1% 0.0%
top 0.1% $10,288,542 $5,841 0.1% 1.3% 0.0% -1.3% 1.3% 3.8% 2.5% 1.3% 1.3% 0.0%
All $84,672 $4,259 5.0% 5.0% 0.0% 0.0% 5.0% 0.0% 0.0% 5.0% 0.0% 0.0%
Notes: This table simulates the distributional effects of replacing the premiums paid by employers for health insurance provided to their workers by a flat
payroll tax in 2021. The total amount of employer-provided health insurance premiums is taken from the National Health Expenditures accounts, Table
5.6, sum of contributions to employer-sponsored private health insurance paid by private business, households, federal government, and state and local
governments. The total amount is $1,068 billion in 2021, which is equal to 5.0% of total national income (including pure realized capital gains) and 11.8%
of the total pre-tax wage income of currently-covered employees. This total is allocated to income groups following the distribution of health insurance
contributions reported in W2 forms (with a correction at the bottom of the distribution to take into account that only employers with more than 250 workers
have to report). In column 3, the result is divided by pre-tax national income (as reported in Table 2 and in col. 1 here) to compute the current “head
tax” rate. Columns 4 to 12 consider the effects of replacing this head tax by a flat payroll tax of 11.8% on the gross wage earnings of currently-covered
employees. In cols. 4 to 6 we assume that health insurance premiums currently paid by employers convert into extra gross cash earnings dollar for dollar
and worker by worker, so that pre-tax income does not change, and after-tax incomes rise at the bottom of the distribution and fall at the top (as a head
tax is replaced by a flat tax). In cols. 7 to 9 we assume that the payroll tax is charged to employees, wages are rigid, 2/3 of what was previously paid by
employers to insurers goes to covered workers and 1/3 goes to profits. In this case the reform is regressive: both pre-tax and after-tax income become more
unequally distributed that in the current status-quo. In cols. 10 to 12 we assume that the tax is charged to employers and wages are rigid, so that pre-tax
income increases by the difference between the payroll tax and the head tax (col. 10 minus col. 3), and there is no change in after-tax income.
A Online Appendix
A.1 Consumption Tax Incidence: Price Effects are Irrelevant
In this appendix, we explain why price effects at the heart of standard incidence analysis (effect
of a tax on goods) are normatively irrelevant.
Let us consider the basic supply and demand tax incidence diagram for one good from
introductory economics, as illustrated on appendix Figure A1(a). Formally, the producer profit
is Π = pQ − c(Q) where p is the pre-tax price of the good, Q the quantity produced, and c(Q)
the increasing and convex cost of producing a quantity Q. Profit maximization implies that
p = c0 (Q) which defines the supply curve S(p). The consumer utility is V = v(Q) − p̄Q where
Q is the quantity of the good consumed, v(Q) the increasing and concave utility of consuming
Q, and p̄ = p + t the after-tax price of the good (with t the tax per unit of good). Utility
maximization implies v 0 (Q) = p̄ which defines the demand curve D(p̄). The key point is that, in
the Diamond and Mirrlees model, pure profits are assumed to be fully taxed away.58 Therefore,
taxes collected are T = tQ + Π = p̄Q − c(Q).
The classic Ramsey tax problem sets tax rates to collect a certain tax revenue while min-
imizing utility loss. Therefore, the key tradeoff is consumer surplus vs. taxes collected at the
margin. As illustrated on appendix Figure A1(b), increasing the tax t mechanically increases
tax revenue (and correspondingly reduces consumer surplus) but it also reduces taxes through
the behavioral response (and correspondingly increases deadweight burden). Because pure prof-
its are in the tax base, the increase in tax from the consumption good due to dp < 0 is fully
offset by the loss of profit dΠ and hence this margin is irrelevant.59
Mathematically, the Lagrangian takes the form
Hence, (and using the envelope conditions v 0 (Q) = p̄, c0 (Q) = p), the first order condition in p̄
takes the form: −D(p̄) + λD(p̄) + λD0 (p̄)[p̄ − p] = 0 which can be rewritten as the classic inverse
elasticity formula:
t 1 λ−1
= · , (4)
p+t εD λ
with εD = −p̄D0 (p̄)/D(p̄) > 0 the elasticity of demand for the good from the consumer and
λ > 1 reflecting the fact that the marginal dollar of tax creates a welfare loss in excess of one
dollar on the consumer. The elasticity of supply coming out of the production side does not
appear in equation (4).
58
Pure profits arise in this simple one good model but would not exist in a model with several production
factors and constant returns to scale (as in the labor and capital model discussed above).
59
This is of course the same logic as in the two-factor model where the lost wages dw were made up by more
capital income kdr.
57
With only one taxed good, the Ramsey problem is not meaningful but it is straightforward
P
to consider multiple goods. With separability V = i vi (Qi ) − p̄i Qi , the demand function for
each good Qi depends only on its own price p̄i , and the same analysis carries through and
equation (4) applies to each good with the same λ, which is the basic Ramsey inverse elasticity
rule.60
Optimal tax. Let us start with the optimal tax approach. The government chooses τL , τK , R
to maximize social welfare
SW = pL gL uL + pK gK uK + p0 g0 u0 ,
60
Ramsey (1927) did not assume that pure profits could be taxed so that Ramsey’s formulas do depend on
supply elasticities as well. However, Diamond and Mirrlees (1971) noted that constant returns to scale, which
rules out pure profits, is a better assumption in general equilibrium. Hence, the standard assumption in modern
optimal tax theory has been to assume that there are no pure profits or that they can be taxed away fully.
Stiglitz and Dasgupta (1971) is the classic reference exploring this point.
58
with gL , gK , g0 the exogenous social marginal welfare weights on each group which we assume
average to one (without loss of generality) so that
which can be plugged in the social welfare function. Hence, equivalently, the government choose
w̄ and r̄ to maximize:
Importantly, pretax prices w and r have disappeared from the objective function. The govern-
ment can use taxes τL and τK to determine the after-tax prices w̄ and r̄ ignoring the effects
on pre-tax prices, one of the key results from Diamond and Mirrlees (1971). Using the enve-
lope conditions that l and k choices maximize individual utilities, and using that FK = r and
FL = w, we obtain the following first order condition for government optimization:
dSW dK r − r̄
0= = (r − r̄) − K + pK gK k = eK K − K + gK K.
dr̄ dr̄ r̄
dSW dL w − w̄
0= = (w − w̄) − L + pL gL l = eL L − L + gL L.
dw̄ dw̄ w̄
These two equations lead to the standard optimal tax formulas:
τK∗ r − r̄ 1 − gK 1 − gK
∗
= = i.e. τK∗ = ,
1 − τK r̄ eK 1 − gK + e K
τL∗ w − w̄ 1 − gL 1 − gL
∗
= = i.e. τL∗ = .
1 − τL w̄ eL 1 − gL + eL
Optimal tax rates depend solely on the supply side behavioral responses of labor and capital eL
and eK along with the social welfare weights that the government assigns to each group gL and
gK . Tax incidence on pretax prices is irrelevant because it affects the splitting of production
into pretax labor and capital income: F (K, L) = wL + rK but what matters for the government
budget is total resource F (K, L) and what matters for individuals are aftertax prices.
Tax incidence. Let us now consider the tax incidence approach starting from a given tax
system (τL , τK ). We consider a small increase in the capital tax rate dτK > 0 and trace out
its effects dK, dL, dr, dw. Differentiating the 4 equations in (5), we have two equations on the
production side:
dK dL dw dr
− =σ· − , L · dw + K · dr = 0.
K L w r
59
The first equation is the definition of the elasticity of substitution between labor and capital,
σ. The second equation is obtained by differentiating F (K, L) = rK + wL and using FK = r
and FL = w.
On the supply side, we have two equations:
dK dr̄ dr dτK dL dw̄ dw
= eK · = eK · − , = eL · = eL · .
K r̄ r 1 − τK L w̄ w
Therefore, pretax price incidence shifts the initial capital tax increase partly onto labor:
the after-tax return on capital falls by less than the new tax but the after-tax wage also falls.
Hence, in the optimal tax approach discussed just above where the government optimizes r̄ and
w̄, dτK > 0 amounts to reducing dr̄ by less than −rdτK but at the same time reducing w̄ by dw̄.
Therefore, it mixes a (smaller) tax increase on capital with a tax increase on labor. The welfare
effects of the reform dτK amount to analyzing the welfare effects of dr̄ and dw̄ and ignoring the
irrelevant price effects as discussed above.
If the labor tax is optimal and equal to τL∗ , then dw̄ has zero first order welfare effects, and
hence the welfare effects of dτK > 0 are the same as the welfare effects of dr̄ < 0. If τK < τK∗ ,
increasing the tax rate is desirable whether or not price effects are taken into accounts.
If the labor tax is suboptimal τL < τL∗ then dw̄ < 0 has a positive first order welfare effect.
Therefore, if τK < τK∗ , then dτK > 0 is desirable both because it increases the tax on capital
and also because it implicitly increases the tax on labor.
However, if the labor tax is too large τL > τL∗ then dw̄ < 0 has a negative first order welfare
effect. Therefore, if τK < τK∗ , then dτK > 0 will be desirable if and only if the positive impact
of dr̄ < 0 is larger than the negative impact of dw̄ < 0. Which effect dominates depends on
which tax rate is furthest away from its optimum. If τK is only slightly below τK∗ and τL is
substantially above τL∗ , then the dw̄ welfare effect will dominate making the reform dτK > 0
undesirable.
While it is certainly important for a policy maker to learn from classic tax incidence that a
reform dτK > 0 may be undesirable even if τK < τK∗ , it is also important for economic advice
to explain that the reason dτK > 0 is not desirable is because τL is too low and that combining
an even greater capital tax increase with a reduction of τL can achieve the goal of policy maker.
This is why we view classic tax incidence as useful but overly narrow and why we think that
60
optimal tax analysis offers a vital broader picture view for the analysis of tax reform. Put
simply, the optimal tax approach tells the policy maker which direction to go; the tax incidence
analysis can provide the technical pathway on how to get there.
Taxes on intermediate goods. Some consumption taxes (such as tariffs, taxes on alcohol
and fossil fuels, and business turnover taxes) are levied on intermediate rather than final goods.
Intermediate goods taxes are small, less than 3% of total tax revenue in the United States.
Most countries have replaced turnover taxes by the value-added tax which only taxes final
consumption.61 Because taxes on intermediate goods distort production prices, there is no
direct model guidance on how to assign these taxes for distributional current-tax analysis.
The best way to proceed is to treat these taxes as consumption taxes on the final goods
eventually produced using the taxed intermediate goods. For example, a tax on wholesale beer
will be assigned to final beer consumers (as part of the post-tax beer price), a tax on jet fuel to the
consumers of airplane travel. A more complex case involves turnover taxes on natural resource
extraction that many extracting countries impose. If the marginal cost of extraction is equal to
the selling price (no pure profits), the tax is akin to an intermediate goods tax. However, if the
marginal cost of extraction is lower than the selling price (e.g., oil extraction in Saudi Arabia,
where marginal costs are much lower than the global oil price determined by the marginal
producer), royalties are akin to a tax on the pure profits of extracting companies. Practically,
one needs to assess whether the royalty is assessed on a resource for which production is closer to
the no pure profit vs. pure profit benchmark. In the case of US oil and gas extraction, marginal
costs are significant and we treat royalties levied by US governments (0.2% of government
revenue) like taxes on other intermediate goods.62
Taxes on depreciable capital assets. Assets used in production are subject to property
taxes. If the asset does not depreciate (e.g., land) the tax is fully assigned to the ultimate owner
of the asset. If the asset depreciates (e.g., a building) then the depreciating part of the asset
is like an intermediate good: it is consumed during the production process. The corresponding
tax is allocated like other taxes on intermediate goods, i.e., to consumers of the corresponding
61
Intermediate goods taxes create production inefficiencies and the Diamond and Mirrlees (1971) model shows
that they should not be used.
62
In 2021 taxes on the extraction of natural resources such as oil and natural gas, called severance taxes,
generated $13.5 billion in revenue (NIPA Table 3.5), out of $6.3 trillion in government tax revenue.
61
final goods. For example, if Amazon uses up 1/40 of its warehouses each year (straight-line
depreciation over 40 years), then 1/40 of the annual property tax paid on these warehouses are
included in the consumption taxes on Amazon products sold to final consumers. In practice,
because the bulk of taxes on capital assets are property taxes on buildings and land, which have
long or infinite lives, business property taxes can be fully assigned to business owners.63
Carbon taxes. Carbon taxes may become important during the transition to clean energy.
Because both consumption and investment decisions are responsible for carbon emissions, a
general carbon tax covering all forms of emissions should be allocated to both consumers and
owners, in proportion to carbon emitted. In the case of emissions due to investment (e.g., a
warehouse built with cement), the intermediate-goods logic described above continues to apply.
Since assets are partly consumed during the production process, part of the tax should be
allocated to the consumers of the final goods produced with the depreciating assets. Overall,
our framework assigns carbon taxes to consumers in proportion to consumption of final goods
and fixed capital, and to business owners in proportion to net investment (i.e., gross investment
minus consumption of fixed capital).
Because business ownership is more concentrated than consumption, with this methodology
carbon taxes are more progressive than when the assignment is only based on the consumption of
final goods (ignoring investment), the conventional approach (see Carloni and Dinan, 2021, for
a survey). But carbon taxes are less progressive than in the methodology of Chancel (2022) and
Chancel and Rehm (2023), where carbon emissions are allocated to consumers for consumption
goods and business owners for gross investment (instead of net investment as we propose). In
the United States, net domestic investment is only about 25% of gross domestic investment.
Our method is thus approximately 1/4 of the way between the conventional method and the
Chancel method.
Inheritance, gift, and estate taxes. Taxes assessed on the transfer of wealth can hurt the
welfare of two parties: the donor and the donee.64 They could be assigned to either. We follow
the conventional approach that assigns taxes to donors. This can be rationalized by the fact that
the potential negative impact on donors is the one that usually raises most concerns (transfer
63
In the case of residential property taxes, for owner-occupiers the owners and the consumers are the same
individuals, so there is no assignment issue. For rented housing, the part of the property tax corresponding to
the annual depreciation of the structure should conceptually be allocated to the consumers of housing services
(the renters). This part, however, is very small (1.25% of the property tax assuming (i) straight-line depreciation
of the structure over 40 years and (ii) that land, which does not depreciate, represents half of the taxable value
of the house) and can be neglected in practice.
64
Piketty and Saez (2013) propose an optimal inheritance tax model where both welfare effects play a role. In a
dynastic model of Barro-Becker, donor and donee are part of the same dynasty, but in the real world individuals
matter separately from dynasties (and indeed to the best of our knowledge, no distributional tax table has ever
been presented for dynasties).
62
taxes harm the property rights and incentives of donors to accumulate wealth), while the cost
for donees is secondary (as they benefit from a transfer through no effort of their own).65
Transaction taxes. Some countries impose taxes on specific transactions such as real estate
transactions, or financial transactions. The simplest treatment is to allocate such taxes to
the buyer side of the transaction (and make it flow to the ultimate individual owner if an
intermediary such as a business is buying the asset). This naturally extends our treatment of
consumption taxes where consumption taxes charged on second-hand goods are also assigned
to the buyer.66 If turnover is fast (as is often the case with financial transactions), allocating to
buyers vs. sellers does not make much of a difference.
Progressive consumption taxes. A progressive consumption tax that exempts net savings
from taxation and adds net dissaving to the tax base (i.e., that extends the traditional pension
treatment to all forms of savings) is allocated to individuals based on their consumption. As
savings are concentrated at the top of the income distribution (Saez and Zucman, 2016) with
negative savings at the bottom and positive and large savings rate at the top, moving to a pro-
gressive consumption tax would be regressive when distributional impacts are assessed relative
to income percentile.67
Flat taxes. Flat taxes have been proposed in the US tax debate by Bradford 1986 (the X-tax)
and Hall and Rabushka 1985 (the flat tax). This “flat tax” is a tax on wage income combined
with a cash flow tax on business profits with no deduction for interest income payments and full
expensing of investment instead of depreciation of capital assets over their lifetime as in regular
corporate taxes.68 Using our methodology, the flat tax would be assigned on the corresponding
wage earners and the corresponding business owners.
While the “flat tax” is economically equivalent to a flat consumption tax such as a VAT
from a dynamic perspective (and ignoring the exemption for low earners built in the flat tax),
the distributional impact is quite different when measured on an annual basis. A worker who
saves most of his income consumes little and hence pays no consumption tax, but would pay
the “flat tax” on wage earnings. As highly paid workers save more than low paid workers, the
65
Arguments in favor of assigning taxes on donees can also be provided. For example, if bequests are accidental,
then donors do not care about transfer taxes and only donees are affected.
66
The convention in national accounts is that if a second-hand good is resold through a business, it is seen as
a business activity with the used good being an input and the resold used good being like a new good with the
difference in prices reflecting value added: the cost of buying and reselling the used good for the business, and
the value of reallocating the good to a consumer with higher value on the consumer side.
67
Proponents of consumption taxation might argue that individuals should be ranked by consumption rather
than income when assessing progressivity. To our knowledge, such distributional tables have not been produced,
in large part because there is no good micro-data in the United States measuring both income and consumption
especially at the top of the distribution.
68
TCJA provides full expensing for five years 2018-2022 with a phased-in return to depreciation over 2023-2027.
63
flat tax will be more progressive than the VAT on an annual basis.69 The “flat tax” exempts
investment while the consumption tax exempts savings. Investment is made by business owners
who maybe different from savers but both business owners and savers are concentrated toward
the top of the distribution. Therefore, on net, the flat tax is likely to be more progressive than
the VAT, measured on an annual basis. Naturally, from a dynamic perspective, the two taxes
generate the same budget sets and hence are formally equivalent (See Auerbach 2019 for a recent
exposition). However, if households face borrowing constraints or do not plan according to the
classic intertemporal utility model, this equivalence is lost.
Taxes on mixed business income. Business income is a mix of labor income (the labor
effort of the owner) and capital income (the return on the business assets). Neither national
accounts nor income tax data can separate cleanly the two components. How then should we
assign the corporate income tax on a closely held business or the individual income tax on
pass-through businesses? With our methodology, such taxes are assigned directly to the owners
themselves who supply both the labor and the capital so we do not need to separate labor and
capital to assign taxes either. CBO assigns 25% of the corporate tax to workers but it assigns
100% of the tax paid by a passthrough business to its owners (because it allocates individual
income taxes to each taxpayer individually). Hence, a pure change in organizational form with
no change in economic activity, such as a change from a sole proprietorship to a C-corporation,
increases the tax rate on workers nationally, which is not satisfactory conceptually.
Foreign taxes. One limitation of both the conventional approach and ours is that cross-
border corporate income tax payments are ignored. Only corporate taxes collected by the US
government are allocated to individuals. This is because government agencies are interested
in distributing US federal tax revenues, while the distributional national accounts literature is
interested in distributing national income—and foreign corporate taxes are not part of national
income. In reality US individuals pay corporate taxes to foreign governments, and some of
the corporate taxes collected by the United States are paid by foreigners. In recent years the
two flows broadly offset each other and our series thus capture the effective rate paid by US
individuals globally. However this was not the case historically. It would be valuable to develop
distributional current-tax series adding back net cross-border corporate income tax payments,
a task we leave to future research (Zucman, 2023).
How to rank individuals? Traditional distributional tables ranks individuals (or families)
by annual pre-tax income. This is justified if annual pre-tax income is indeed the best measure of
economic status. Other rankings are conceivable such as changing the time frame (such a month,
69
Similarly, for private pension arrangements in the US, a Roth IRA is equivalent to a traditional IRA from a
lifetime perspective. But on an annual perspective, if savers who get the tax exemption through the traditional
64
multi-years, or even a lifetime) or changing the variable to after-tax income, consumption, or
wealth. There is no definitive or right answer to this question. Different measures might work
best for different purposes. At the high end, wealth plays a role over and above income to
measure economic status. A CEO earning $50 million/year with no accumulated wealth is not
in the same economic class than a wealthy owner making $50 million/year out of fortune of $1
billion. This would call for factoring wealth over and above the capital income it generates in
some way. Consumption becomes an almost irrelevant variable at the very top as even lavish
personal consumption is going to be small relative to wealth for billionaires or deca-billionaires.
At the low-end, transfers play a large role so that after-tax and transfer disposable income is
likely to be a more meaningful measure of economic well-being than pre-tax income.70 Even at
the low end, consumption may not be a better measure of economic well-being than disposable
income (available for consumption and savings) as the ability to save is clearly a marker of
economic security and hence well-being. If our view, economists have spent too little time
thinking through these important and non-trivial issues.
Pre-tax and post-tax incomes. National income includes indirect taxes. To estimate the
distribution of national income, the most sensible approach is to first estimate the distribution
of national income excluding consumption taxes (i.e., factor-price national income), and then
gross up income levels proportionally (i.e., with no impact on the distribution of income). What
follows details the reasoning.
At the micro-level, pre-tax income y and post-tax income c are related as follows:
c + tc = y − ty + g, (6)
where y is pre-tax income (from labor and capital), ty taxes paid on labor and capital generating
pre-tax income y, g are transfers from the government, c is consumption—exclusive of consump-
IRA have higher incomes than retirees who get the exemption through the Roth IRA, the traditional IRA is
less progressive than the Roth IRA. Viard and Carroll (2012) note that the flat tax is like a Roth IRA while the
VAT is like a traditional IRA.
70
Pre-tax and after-tax income rank might differ substantially if transfers are targeted to specific groups.
65
tion taxes paid—plus saving and tc are taxes paid on consumption. The relevant concepts for
inequality analysis are y (pre-tax income) and c (post-tax income). Total income c + tc is less
interesting because it is an intermediate concept that includes taxes on consumption.
Importantly, equation (6) can be defined using broad or narrow definitions of income, con-
sumption, and government transfers. At the broadest level: y includes all pre-tax income from
labor and capital (labor income cash or in-kind and capital income distributed or retained
within a business); g—and hence c—includes all forms of public spending (including collective
consumption expenditures such as defense, education, etc.).71
Taking the broadest definition of income, equation (6) can be aggregated across individuals.
Using capital letters for the macro level, we have:
66
Tax rates and transfers. We now turn to the issue of how consumption taxes should be
treated for the estimation of effective tax rates. In brief: consumption taxes should be allocated
to consumers (as explained in the paper), but the portion of consumption taxes paid out of
transfer income are best treated as reducing transfer income rather than as taxes.
To see this, note that in equation (6), ty and tc are the taxes paid by the individual on her
income y and when consuming (or saving) disposable income y − ty + g. Therefore, it makes
sense to assign tc separately to y − ty and g in proportion of the taxable consumption generated
by each component.75 Hence we split tc into tcy the consumption tax assigned to y − ty and tcg
the consumption tax assigned to g and re-write (6) as:
The net transfer received is gn = g − tcg and the total tax paid on pre-tax income is t = ty + tcy .
This tax concept is the most natural one to estimate effective tax rates by income groups.76 It
avoids the issue of assigning very large tax rates to individuals at the bottom of the pre-tax
income distribution with very low income y relative to transfers g and who pay consumption
taxes on their consumption out of transfer income.77 It makes sense to measure transfers as
gn = g − tcg , i.e., net of consumption taxes paid.
Taxes paid by nonprofits. Some nonprofit organizations pay capital taxes: corporate taxes
on the profits of the companies they invest in, property taxes on the assets the own. To the extent
that nonprofits provide collective wealth and services, they should be left out of distributional
analysis. To match national income, both their primary capital income and the corresponding
taxes should be allocated in a distributionally-neutral manner, i.e., proportionally to after-tax
disposable income.
75
If y − ty and g are both cash, then they contribute to tc in proportion. If g is a pure in-kind transfer such
as health insurance that faces no consumption tax, then tc would be assigned fully to y − ty .
76
For linear taxes ty = τ ·y and tc = τc ·c, we have (1+τc )c = y·(1−τ )+g so that c = y·(1−τ )/(1+τc )+g/(1+τc ).
Hence y −ty −tc = y ·(1−τ )/(1+τc ) and gn = g/(1+τc ). Hence τ and τc add up to the standard (τ +τc )/(1+τc ).
77
We can still have high tax rates for individuals with no income and no transfers who consume through
dissaving, but this issue is typically alleviated when aggregating by income groups.
67
Appendix References
Auerbach, Alan J. 2019. “Tax equivalences and their implications.” Tax Policy and the
Economy 33, no. 1: 81-107.
Bradford, David F. 1986. Untangling the Income Tax Cambridge, MA: Harvard University
Press.
Hall, Robert E. and Alvin Rabushka. 1985. The Flat Tax, Stanford, CA: Hoover Institu-
tion Press.
Stiglitz, Joseph E., and Partha Dasgupta. 1971. “Differential taxation, public goods, and
economic efficiency.” Review of Economic Studies 38, no. 2: 151-174.
Viard, Alan D., and Robert Carroll. 2012. Progressive Consumption Taxation: The X-Tax
Revisited. AEI Press, 2012.
68
Figure A1: Consumption Tax: Incidence and Ramsey Optimum
𝑝
𝑆 𝑝
𝐶𝑆
𝑝̅ = 𝑝 + 𝑡
Tax=t ⋅Q+PS
t⋅Q (pure profits fully taxed)
𝑝*
𝑝 D (𝑝)̅
𝑃𝑆
𝑄 𝑄* 𝑄
𝑝
𝑆 𝑝
𝐶𝑆 𝑝̅ − 𝑝 𝑑𝑄
𝑝̅ + 𝑑 𝑝̅
−𝑑𝐶𝑆 = 𝑄𝑑 𝑝̅
𝑝̅ Optimum maximizes CS +l ⋅ Tax
! $ l%$
⇒ Ramsey rule = ⋅
"#! eD l
𝑝
𝑑𝑃𝑆 = 𝑄𝑑𝑝
𝑝 + 𝑑𝑝
𝑃𝑆
𝑄 + 𝑑𝑄 𝑄 𝑄
Notes: The top panel depicts the classic consumption tax incidence in a one good model. If we assume as in
Diamond and Mirrlees (1971) that pure profits (=producer surplus in the diagram) can be fully taxed away, the
tax is represented by the blue areas: t · Q + P S. The bottom panel depicts the derivation of the optimum tax
that maximizes consumer surplus plus taxes (weighted by factor λ > 1): CS +λT = v(Q)− p̄Q+λ[p̄Q−c(Q)]. A
small tax increase dp̄ reduces CS by Qdp̄ and increases taxes collected by Qdp̄ + (p̄ − p)dQ. Because pure profits
are in the tax base, the increase in tax from the consumption good Qdp is fully offset by the loss of producer
surplus dP S and hence the price effect dp is irrelevant. The first order condition (λ − 1)Qdp̄ + λdQ[p̄ − p] = 0
leads to the classic inverse elasticity Ramsey rule t/(p + t) = (1/εD ) · (λ − 1)/λ. The supply side elasticity εS
and the price effect dp are irrelevant.
69
Figure A2: Individual vs. Corporate Income Tax Revenue (% of National Income)
14%
12%
10%
6%
4%
2%
Corporate income tax
0%
1913 1923 1933 1943 1953 1963 1973 1983 1993 2003 2013 2023
Notes: This graph shows the evolution of US corporate income tax revenues and individual income tax revenues, expressed as a fraction of US national
income. Federal, state and local taxes are included.
Figure A3: Allocating the Corporate Tax: Conventional Approach
vs. Piketty-Saez-Zucman (2018) vs. Our Methodology
70%
60%
40%
20%
1916 1926 1936 1946 1956 1966 1976 1986 1996 2006 2016
20%
15%
10%
Conventional approach
PSZ (2018)
5%
0%
1913 1923 1933 1943 1953 1963 1973 1983 1993 2003 2013 2023
Notes: The top panel contrasts the share of the US corporate income tax (federal and state) paid by the top 1%
units with the highest pre-tax national income in our current-tax methodology and the conventional practice of
distributional tax analysis, as implemented by the Congressional Budget Office (CBO) and the original series of
Piketty, Saez and Zucman (2018), denoted by PSZ. The bottom panel plots the amount of corporate taxes paid
by the top 1% (as a fraction of the pre-tax income of the top 1%) implied by each of these methodologies.
71
Table A1: Current Federal Tax Distribution in the United States, 2021
Pretax income After-tax income Taxes (federal only) Tax rate composition (federal taxes only)
Individual Property Corporate tax,
Income Payroll Consumpt Corporate
Average Share Average Share Share Tax Rate income taxes (incl. conventional
groups taxes ion taxes tax
taxes estate tax) approach
P0-50 $20,889 12.3% $17,862 13.1% 9.2% 14.5% 1.7% 10.5% 1.8% 0.0% 0.4% 0.8%
P50-90 $80,618 38.1% $65,303 38.3% 37.3% 19.0% 6.8% 10.2% 1.0% 0.0% 1.0% 0.8%
P90-99 $243,587 25.9% $195,098 25.7% 26.6% 19.9% 11.6% 6.2% 0.6% 0.0% 1.5% 1.3%
P99-99.9 $1,085,455 11.5% $855,334 11.3% 12.6% 21.2% 16.5% 2.3% 0.4% 0.2% 1.8% 2.0%
top 0.1% $10,288,542 12.2% $7,956,531 11.7% 14.2% 22.7% 18.1% 0.8% 0.3% 1.1% 2.3% 2.9%
All $84,672 100% $68,266 100% 100% 19.4% 9.9% 7.1% 0.8% 0.2% 1.3% 1.3%
Notes: Groups based on pre-tax income including pure realized capital gains (defined as realized gains in excess of 3% of national income). Unit is individual
adult (aged 20+) with equal split among couples. Pre-tax income is income before all taxes but after the operation of pension systems (public and private).
72
Taxes include federal taxes only. Refundable tax credits are not included as negative tax (as they are treated as transfers, like other cash transfers, in the
national accounts). Labor taxes assigned to corresponding workers, capital taxes to corresponding asset owners, consumption taxes to final consumers. In
the conventional approach (currently used by CBO), the corporate tax is assigned 75% to capital income on individual tax returns and 25% to labor income
(with no adjustment for corporate profits earned through pension funds).