DRAFT: NOT FOR QUOTING WITHOUT PERMISSION

Mirrlees Report Chapter on the Base for Direct Taxation1

1
We wish to thank Henry Aaron, Tony Atkinson, Alan Auerbach, Mike Golosov, Jon
Gruber, Jim Poterba, Dick Tresch and Iván Werning for helpful comments, Catarina Reis
and Maisy Wong for research assistance, and the National Science Foundation for
support under grant SES-0239380.

1
 Introduction

Chapter 2 of the Meade Report,2 ‘The Characteristics of a Good Tax Structure’, is

divided into six sections: Incentives and economic efficiency, Distributional effects,
International aspects, Simplicity and costs of administration and compliance, Flexibility
and stability, and Transitional problems. To consider direct taxation in the UK, the
Committee considered each of these issues separately and then combined the insights into
a policy recommendation. It seems to us, as it seemed to Alfred Marshall, that this is an
appropriate way to proceed.3 While the capacity of computers to find equilibrium in
complex models has grown apace since the Meade Report, the models available for
analysis, like much of the underlying theory, are still quite limited and still too far from
reality to proceed in any other fashion than that followed by the Meade committee.
Whilst this should not exclude learning from complex simulations, this essay focuses
solely on theoretical findings with regard to the tax base. 4

Traditionally, the conceptual starting place of a study of tax reform, such as the
Meade Report, is a concept of an ideal tax base, one that reflects both horizontal equity
(treating equals equally) and vertical equity (those with larger ideal tax bases pay larger
taxes). This conceptual starting place is then adjusted in light of the issues raised by the
other five areas of concern identified in Chapter 2 of the Meade Report.

Since the mid-1960’s, there has been a great deal of analysis of models that
consider both equity and efficiency in an integrated modeling effort, based on
maximizing a social welfare function defined in terms of individual utilities with a

2
The Structure and Reform of Direct Taxation, Report of a Committee chaired by Professor J. E. Meade,
London: George Allen & Unwin, 1978.
3
it [is] necessary for man with his limited powers to go step by step; breaking up a complex question,
studying one bit at a time, and at last combining his partial solutions into a more or less complete solution
of the whole riddle. ... The more the issue is thus narrowed, the more exactly can it be handled: but also the
less closely does it correspond to real life. Each exact and firm handling of a narrow issue, however, helps
towards treating broader issues, in which that narrow issue is contained, more exactly than would otherwise
have been possible. With each step ... exact discussions can be made less abstract, realistic discussions can
be made less inexact than was possible at an earlier stage. [Alfred Marshall, Principles of Economics,
eighth edition. New York: The Macmillan Company. 1948, page 366.]
4
For a recent optimal tax calculation and discussion of accomplishments and difficulties, see Judd and Su,
2005.

2
heterogeneous population.5 The primary purpose of this essay is to review that literature
and draw inferences for policy that sets the tax base.6

We begin by considering lessons from the recent optimal tax literature with
specific regard to the taxation of income from capital and the related issue of the tax
treatment of savings.7 We show that a succession of papers, for a number of different
reasons, have argued that under certain conditions the optimal tax schedule should not
include taxes on capital. Our analysis discusses both single cohort versions of these
models (such as the Atkinson-Stiglitz model and its successors) as well as the infinite
horizon equivalents studied originally by Chamley and Judd. The required conditions for
the optimality of zero taxation, however, – typically relating to separability between
consumption and leisure, limited heterogeneity in preferences and the extent to which this
is correlated with other outcomes, an absence of age-varying patterns of preferences and
uncertainty over the life cycle, or systematic links between socioeconomic outcomes and
expected length of life - are argued to be too restrictive and the finding of no role for
capital taxation is therefore considered not robust enough for policy purposes. Hence
there should be some role for including capital as a part of the tax base.

In the second part of this chapter we discuss broader issues in the optimal tax
setup that matter for how these theoretical considerations ought to influence tax policy.
After reviewing the structure of a social welfare function, we begin by considering how
(and whether) to combine an underlying concept of fairness with the models of utility-
only social welfare optimization. A similar issue arises in assessing the equity dimension
of a transition to a new tax system, considered later in the chapter. Finally, we consider
the time frame for thinking about annual taxation and also the limits on tax structure
coming from limits on allowable complexity.

5
The standard basic model treats administrative costs of different taxes as zero or infinite and ignores tax
evasion. See, for example, the textbooks by Myles, 1995, Salanié, 2003, Tresch, 2002, Tuomala, 1990,
although there are articles that address administrative costs and evasion. There has not been integration
with macro issues incorporating, for example, built-in stabilizers (Auerbach and Feenberg, 2000, Diamond,
1994) nor has the incorporation of international issues (trade, investment, migration) included the macro
dimensions of those issues.
6
Other chapters contain discussions of issues not considered here, including tax rates, the presence of
families, and corporate taxation.
7
In terms of the Chapter 2 topics of the Meade report, we will not consider international aspects, analyzing
closed-economy models, nor the use of taxes as part of discretionary fiscal policy for macroeconomic
stabilization. Oddly, the Meade report ignores built-in stabilizers, which seem to us to matter.

3
 Our focus is on the relative taxation of labor and capital incomes, not a horse race
between total (Haig-Simons) income and consumption. 8 In the end, the Meade Report
effectively did the same – the Report closes with a section entitled “ULTIMATE
OBJECTIVES:”
We believe that the combination of a new Beveridge scheme (to set an acceptable
floor to the standard of living of all citizens), of a progressive expenditure tax
regime (to combine encouragement to enterprise with the taxation of high levels
of personal consumption), and of a system of progressive taxation on wealth with
some discrimination against inherited wealth, presents a set of final objectives for
the structure of direct taxation in the United Kingdom that might command a wide
consensus of political approval and which could be approached by a series of
piecemeal tax changes over the coming decade.” (Page 518.)

We leave to other chapters discussion of the provision for the very poor and
concern about inheritances. And we assume that annual measurement of wealth is not
available and so consider annual capital income taxation instead.9 While the Meade
Report was part of a long tradition contrasting taxation of income with taxation of
expenditures, its inclusion of annual taxation of wealth in its conclusion places that
discussion in a different context than if they were to choose only income or expenditure
taxation. We share their framing of the potential simultaneous use of several tax bases
and focus the chapter on a set of questions.

• If there is annual non-linear taxation of earnings, is it worth having a more

complex tax structure, particularly age-varying tax rates? Would greater
age-varying rules in capital income taxation be worthwhile?

• If there is annual non-linear taxation of earnings, how should annual

capital income be taxed – not at all, linearly (as in the Nordic dual income
8
Haig-Simons income is labor income plus accrued capital income – Haig (1921), Simons (1938).
9
While the values of some types of wealth are readily measurable, others are not. Of course the same is
true for accruing capital income. In practice, this is addressed by taxing realized incomes. Such taxation
could be, but is not, adjusted to offset the difference between accrual and realization taxation. We are not
aware of a literature exploring the relative advantages of wealth and capital income taxation (with the latter
supplemented by wealth taxation at death) as part of optimal taxation. Our conjecture is that capital income
taxation could do better, but that is just a conjecture awaiting analysis.

4
 tax), by relating the marginal tax rates of capital and labor incomes, or by
taxing all income the same?

• If there is annual non-linear taxation of earnings, should there be a

deduction for savings?

Part I. The optimal taxation of capital and labour income

There are many papers that analyze optimal taxes; and they differ in many ways.
This essay is not meant to be a survey of methods and model results, but a selective
drawing of policy inferences from the literature. Before turning to particular papers and
the lessons to be drawn from them, we say a little about the modeling strategies we do
and do not draw on.

The optimal tax literature analyzes real taxes dependent on real labor and capital
incomes (apart from analysis of money growth). We do not think there is any significant
disagreement among economists that to the extent feasible, the relevant basis for taxation
is real capital income, not nominal capital income. While such taxation is feasible
without excessive additional administrative costs for capital gains, it is harder to do for
flow capital income since the asset valuation needed to convert nominal into real is not
readily available for many assets. This could be addressed by incorporating the inflation
adjustment in the eventual capital gain taxation. This would add to the case for adjusting
capital gains taxation to remove much of the advantages of deferral. Other than pointing
out that taxing nominal interest and dividends results in taxes on real interest and
dividends at rates higher than the stated marginal tax rate, we do not explore the real-
nominal distinction. We also do not explore issues related to the realization of income,
but note that taxation of deferred realization of incomes, as with capital gains, calls for
heavier taxation than non-deferred capital income, not lighter taxation as is common
practice (Helliwell, 1969, Auerbach, 1991).

Intertemporal optimal tax models can have a single (present discounted value)
budget constraint for the government, reflecting the ability of government to save and to
borrow. Some models require period-by-period budget balance (together with

5
restrictions on allowable taxes that make this constraint binding).10 The budget constraint
assumption can have a strong influence on results. The political economy of how much
borrowing a government does is important and controversial, making it unlikely that
some specific model of political outcomes implicit in a budget balance constraint will
match actual behavior. Moreover, policy conclusions are likely to be sensitive to the
particular political constraints chosen, making it difficult to draw conclusions for a
different political constraint. For countries like the UK, the ability to borrow, to reduce
the public debt, and to save is real. Examining policy with a single budget constraint is in
keeping with looking for what governments ought to do. And it is a baseline from which
it may be easier to draw inferences if one wants to adapt policies from such a model to
reflect a perceived political bias. Thus we ignore papers with period-by-period budget
balance.

Overwhelmingly, optimal tax models assume competitive behavior by firms.

While this is not a genuinely satisfactory assumption, we have not explored the limited
literature that considers other market structures.

Typically, the labor market is modeled as if workers can choose the number of
hours to work at the wage available to them. Such a simple linear before-tax budget
constraint is not realistic for many people, given rules on overtime pay and possibly
different earnings per hour on primary and secondary jobs. Also many jobs come with a
standard number of hours, although it should be recognized that the standard number of
hours at an employer is a choice variable that plausibly reflects to some degree the hours
that workers would like to work. Some of the literature recognizes the discontinuity in
disutility of work at zero hours (e. g., from commuting) that makes withdrawal from the
labor force a possible next-best alternative to work with a significant number of hours.
The distinction between extensive and intensive labor supply margins is very important
for considerations of tax rates and acknowledging both can lead to a greater role for the
average tax rate in policy analysis. Moreover, since the relative importance of intensive
and extensive margins varies widely by age, this is highly relevant for the case for age-

10
In models with standard fully rational agents, the government can balance period-by-period budgets by
using taxes and transfers over time on the same individuals. This converts a constraint of period-by-period
budgets into a single intertemporal budget constraint without explicit government borrowing or saving.

6
varying taxes. Since it is most common in the literature, we focus on models with
adjustable hours, although the retirement literature often makes use of a zero-one model
of employment opportunities.

Most of the literature assumes that the relative wages of workers with different
skills is exogenous and we review primarily inferences from these papers. A few papers
allow the undoubtedly important issue of endogeneity. Some of these papers have just
two types of labor and no capital. With constant returns, strict concavity of the
production function and only two factors an increase in the supply of one raises the return
to the other. Such modeling calls for encouraging more skilled labor supply to lower the
skill premium. For our purposes, three factor models are needed, so that capital can
affect relative wages (e. g., Naito, 1999). Empirical work supports the finding that
increased capital raises unskilled relative to skilled wages (Krusell et al., 2000).

To a large extent we work with the assumption that capital income and labor
income are both perfectly observable. This omits issues of tax evasion and also omits
difficulties in distinguishing between the two types of incomes. This is obviously an
issue for the self employed – an issue that the Nordic dual-income tax structures have had
to face.11 It is also an issue in the conversion of labor income into corporate income,
which has received attention in the literature on the corporate tax (e. g., Gordon and
MacKie-Mason, 1995) and when recognizing that labor effort can affect returns on
capital investments - by seeking better investments or working harder at evaluating them.

The use of labor for home production is relevant in some contexts, particularly
when considering differential commodity taxation of goods. But this distinction does not
seem important for the intertemporal issues considered here, although it does matter for
evaluating the impact of withdrawal from the labor market on the utility achievable from
consumer spending.

More central to consideration of the taxation of capital income, however, is the

impact of such taxation on individual savings. Overwhelmingly, intertemporal optimal
tax analyses have assumed a life-cycle model (and frequently one with an implausible
assumption of intertemporal additive separability in the life-time utility function). Yet

11
Zzz DESCRIBE RULES

7
analysts examining the wide distribution of individual consumption, savings and wealth
holdings over the life-cycle have not found this model adequate. Uncertainties are
important, as reflected in the precautionary savings literature. Non-time-consistent
behavior has been deemed important for some issues in the quasi-hyperbolic discounting
literature and even within the standard discounting framework there appears to be a
considerable heterogeneity in underlying discounting of the future in the population (e.g.,
see Hausman on different discount rates for air conditioner purchasers, or Samwick on
the distribution of discount rates that can rationalize the distribution of retirement saving
wealth). And the behavior of those with very large wealths appears to require utility
directly from wealth holding, not indirectly from later consumption (Carroll). These
issues must be taken seriously before one can evaluate policy implications that may rely
too heavily on a savings model with some real connection with the data, but very
incomplete connection. Below we will briefly speculate how different models of savings
behavior might alter some of the optimal tax inferences.

For the most part we focus on models that do not have a bequest motive and we
do not consider inter vivos transfers between generations within families.12 This is wrong
for some people and again the life-cycle consumption literature argues that the role of
saving for bequests is diverse in the population and unclear (Hurd).13 But to incorporate
bequests we would need to consider taxes on bequests or inheritances and to explore the
appropriate social evaluation in the context of such externalities (Diamond, 2006). We
can only hope that our analysis would carry over to models that realistically reflected the
diversity in bequest behaviors (and savings for bequests) and also incorporated taxes on
such transfers.

12
The empirical evidence on the consumption patterns of parents and adult children alive at the same time
is strongly contradictory of the idea that overwhelmingly people behave as if there were a single dynastic
utility function being jointly maximized. Moreover, taking this literally and recognizing marriage (which
links dynasties to each other) leads to absurdities (Bagwell and Bernheim).
13
As an example of the importance of motivation, if all bequests are accidental from incomplete
annuitization and also unobservable, then there is a case for capital income taxation when assumed
preferences and technology would have a zero tax rate be optimal without the bequests (Boadway,
Marchand, and Pestieau. 2000). On the other hand, with the same assumptions, if bequests are given from
a utility motivation and if the utility motivation is fully respected in the government objective function,
then the optimal tax on capital may be positive or negative (Cremer, Pestieau, and Rochet. 2003). On the
issue of whether such motivation should be fully respected in the government objective function, see
Diamond, 2006.

8
 Tax tools

The models examined take as given different packages of government tax tools,
contrasting the use of tools in models with different or additional tools available. They
differ greatly in both the choice of the tax base and the degree of allowed complexity in
tax rates. When a model matches a tax structure that is employed, then drawing
inferences has all the difficulties implicit in the incompleteness of a standard social
welfare function to capture all the concerns about policy, as discussed in Part II.
Drawing inferences becomes even more complicated when a model allows a degree of
complexity well beyond what is done in practice. The prime example of this issue is the
use of mechanism design to derive some of the conditions satisfied by an optimum. For
example, in the two-period model with two consumption variables and two earnings
variables studied below, the usual mechanism design approach has each person choosing
all four variables out of the set of allowed four-variable options. With uncertainty about
second-period skills or preferences, there are two separate incentive compatibility
constraints. While first period choices are over available pairs of first-period
consumption and earnings, the available set of second-period consumption-earnings pairs
depends on the choice that was made in the first period. This is complex in both tax
structure and first-period individual optimization. While one can argue that all four
variables are reasonably observable, the complexity implicit in the tax structure goes well
beyond what is observed in practice, although there are special cases that simplify. This
is not to suggest that there is nothing to learn from these models – quite the contrary is
true, but that one must draw policy inferences carefully, not literally.

Complexity in tax policy is a difficult issue to consider. We are lacking in

analyses that take us very far in considering when additional complexity is a bad idea.
Additional complexity can affect individual choices (and the perceived fairness of
outcomes that are strongly influenced by the ability to handle complexity) and can affect
the quality of the legislative process as ex ante restrictions (formal or informal) in the
range of legislation to be considered can influence the outcome. In the absence of
extended analyses on which to draw, using complexity concerns to influence policy
inferences from formal models is subjective, but seems important. We make use of such
concerns, for example, in favoring more variation of taxation with age while not favoring

9
variation of taxation with height, despite the apparent improvement in the equity-
efficiency tradeoff that would come with using height as an additional dimension in tax
design. This is discussed further in Part II.

A second issue with these models is the small dimensionality of the population
usually assumed. We think that the dimensionality of actual populations (given
measurement on an annual basis) is not small enough for the typical analysis to
automatically transfer even if complexity were not an issue. The issue is then the kinds
of inferences one should be drawing. For example, in the one-period model with
earnings and a single consumption good, the mechanism design optimum with a single
type of worker can be implemented by a tax on earnings. As Saez (2001) has shown,
provided that the optimum has individuals making small responses to small tax changes,
the standard first-order conditions can be used with a higher dimension population, using
averages of characteristics at each earnings level. However, once we move into
intertemporal models with uncertainty, we are skeptical of the ability to have such a
simple carryover when the required optimal structure has sufficiently complex allowable
interactions.

2-period models with 1 labor supply

Atkinson-Stiglitz (1976) explored a model that had a single labor supply and
multiple consumer goods. These consumer goods can be interpreted as including goods
consumed in the first period and goods consumed in the second period. Thus the model
can be viewed as shedding light on the taxation of savings for retirement. For taxation of
savings for later consumption during working life, we turn below to models with two
separate labor supplies, representing labor supply in different periods. By examining the
relationship between the MRS and MRT between consumer goods in different periods,
one can examine the taxation of savings from first-period labor to finance second-period
consumption. Thus, if the MRS should equal the MRT between each pair of consumer
goods, then the optimum is not consistent with taxing consumer goods other than
proportionally (i. e., without relative consumption distortions), and thus inconsistent with
taxing savings at the margin. Indeed, Atkinson and Stiglitz showed that when the

10
available tools include optimal nonlinear earnings taxes, optimal taxation is not consistent
with differentially taxing consumer goods when two key assumptions are satisfied. The
two key assumptions are that all consumers have preferences that are separable between
goods and labor and that all consumers have the same sub-utility function of
consumption. Like the Fundamental Welfare Theorem, this theorem plays two roles –
one is to show that limited government action is optimal in an interesting setting, and the
second is to provide, through the assumptions that play a key role in the theorem, a route
toward understanding the circumstances where more government action (in this case
distorting taxation of savings) is called for.

Moreover, the underlying logic of the theorem extends to additional settings

beyond the full optimization of social welfare. Laroque (2005) and Kaplow (2006) show
that with the same preference assumptions, and any income tax function that gives rise to
an equilibrium, if there are distorting consumer taxes, then a move to non-distorting
consumer taxes can be done along with a permutation of the income tax that leaves every
consumer with the same utility and the same labor supply, while the government collects
more revenue. If labor supply is smooth in response to uniform transfers to all consumers
(no jumps in labor supply), then this revenue gain can be used to make a Pareto
improvement. Of course, an argument that a better policy is available is only an
argument against a policy proposal if the alternative is pursued. As with the Hicks-
Kaldor criterion, hypothetical alternatives not to be taken up are not legitimate arguments
against a policy that would increase social welfare. That is, arguing against a distorting
tax that would increase progressivity in taxation on the basis of the existence of a
dominating proposal is somewhat hypocritical if the dominating proposal is not supported
and will never be adopted.

The underlying logic behind the Atkinson-Stiglitz result is that with separability
and the same preferences, distorting consumption taxation can not accomplish any
distinction among those with different earnings abilities beyond what is already
accomplishable by the earnings tax, but still would have a cost from distorting spending.
That is, the optimum encourages work by providing the utility from the consumption that
can be purchased by earnings in the most efficient manner. This logic gives insight into

11
several changes in assumptions that would prevent the conclusion in the Atkinson-Stiglitz
model that capital income should not be taxed.

One obvious change would be that preferences do not exhibit separability

between consumption and labor. Then the Corlett-Hague (1953) style analysis in a 3-
good model (current work, current consumption, and future consumption) can examine
whether a move towards taxing savings or towards subsidizing savings raises welfare.
The key issue is the pattern of cross-elasticities between labor supply and consumptions
in the two periods. But we do not know much about the relevant cross-elasticities.
Although the commonly-used assumptions of atemporal and intertemporal separability
strike me as implausible, that does not lead to a straightforward conclusion about the
cross-elasticities. In particular, those in the second period (who are retired) have more
time to do home production and also more time to enjoy consumption opportunities that
are time-intensive. It is thus not clear which cross-elasticity is higher and so whether
savings should be taxed or subsidized. Recognition of home production is an argument
for differential taxation of different goods in a point of time (Kleven, Richter and
Sørenson, 2000), but does not appear to help clarify the issue of intertemporal taxation.

Two arguments limiting the policy applicability of the Atkinson-Stiglitz theorem

based on savings patterns in a model with skill differences have been made by Saez
(2002b) and Deaton (1979). Saez argues that it is plausible that there is a positive
correlation between labor skill level (wage rate) and the savings rate. In a two-period
certainty setting with additive preferences, this is consistent with those with higher
earnings abilities having less discount of future consumption. In terms of the conditions
of the Atkinson-Stiglitz theorem, Saez preserves separability in preferences but drops the
assumption that the subutility function of consumption is the same for everyone. With
the plausible assumption that those with higher earnings discount the future less (and so
save more) then taxation of savings helps with the equity-efficiency tradeoff by being a
source of indirect evidence about who has higher earnings abilities.

A different argument comes from Deaton (1979). Deaton notes that if the income
tax is constrained to be linear, then the Atkinson-Stiglitz conditions that are sufficient for
the non-taxation of capital income with optimal nonlinear taxation are no longer

12
sufficient for the result. A further condition is needed when the income tax function must
be linear even when preferences are weakly separable between goods and leisure (as in
Atkinson-Stiglitz) - that all consumers have parallel linear Engel curves for goods in
terms of income. Thus, even with weak separability and uniformity of preferences,
different savings rates for different earners because of nonlinear or nonparallel Engel
curves prevent the general holding of the result. If higher earners save higher fractions of
their incomes, then there should be taxation of capital income. Note that this argument
applies as well to each piece of a piecewise linear tax function, with application of the
condition to those on a single linear stretch of the tax function. That is, with a linear
income tax and differing savings rates, an increase in the tax rate can not reproduce the
tax pattern from taxing savings and so be generally a dominant policy change.

As noted above, Naito (1999) has shown that with endogenous relative wages of
skilled and unskilled workers, the Atkinson-Stiglitz theorem does not hold. In the
intertemporal interpretation, there is an aggregate production set involving first-period
consumption, second-period consumption, skilled labour and unskilled labour. By
shifting consumption demand between periods, if one can shift relative wages, then the
incentive compatibility constraint can be weakened, breaking the dominance of the
earnings tax over non-proportional taxation of consumption.

Implicit in the model used by Atkinson and Stiglitz, and by almost all analysts, is
the uniformity of prices charged to different consumers. In this multi-period setting,
uniformity in the rate of return implies that all investors are equally good at choosing
investments. Without getting into the complications that come from portfolio
opportunities that involve diverse choices on a risk-return frontier and without getting
into complications from non-diversifiable risks, such as direct loans to small businesses,
there is still the issue of differing returns for the same risk pattern. This is obvious with
mutual funds that charge different fees to investors with different size portfolios and with
apparently identical mutual funds that charge different fees. It also results from the use
by some people of investment advisers charging fees and placing investors in the same
mutual fund portfolios that others choose without paying for such advice. Plausibly, the
level of available returns is positively correlated with earnings potential. To explore
whether these facts imply that taxation of capital income could improve social welfare,

13
one would need a model that has reasons (or simply assumes) that investment decisions
are made separately person-by-person. Such a model is likely to show two effects – that
those who can get a higher rate of return are better off, ceteris paribus, and they are more
efficient at converting resources into utility. These two effects naturally point in opposite
directions. Such a model might also address implications of the gap between borrowing
and lending interest rates.

One way in which savers try for higher returns is by devoting labor effort to seek
higher returns. Thus, in a model without taxation of capital income, the pursuit of higher
earnings is taxed like leisure, less than earnings are taxed. But a more efficient solution
is likely to be achieved by taxing the fruits of the pursuit of higher returns. However, we
are not aware of formal modeling incorporating this with non-linear taxation of regular
labour earnings.

The standard model assumes that a worker knows the return to working before
deciding how much to work and, since work is in the first period, knows the earnings
from work before doing any consumption. Uncertainty about earnings given work does
not influence taxation of savings if the uncertainty is resolved before first-period
consumption - the Atkinson-Stiglitz result carries over. But consumption decisions
before earnings uncertainty is resolved does impact the Atkinson-Stiglitz result.

Modifying the model so that earnings occur in the second period (with no advance
knowledge of future earnings) would imply that the first-period consumption decision is
made before the uncertainty about earnings is resolved, while second-period consumption
occurs after.14 The Atkinson-Stiglitz result no longer holds and second-period
consumption should be taxed at the margin relative to first-period consumption (Cremer
and Gahvari, 1995). We get this result whether we have general taxation on earnings and
savings or have only a linear tax on savings with a nonlinear tax on earnings. We can
compare this result with that of taxing savings when higher earners have less discount of
the future. In the latter case a worker choosing to imitate someone with less skill values
savings more than that worker with less skill since the discount of the future is less for

14
With annual taxation, consumption during the year is happening before earnings levels later in the year
are known, at least for some workers. This parallels analyses of the demand for medical care with an
annual deductible or out-of-pocket cap.

14
the potential imitator. Thus taxing savings eases the incentive compatibility constraint.
In the uncertainty case, a worker planning to deviate by earning less than the optimized
amount in the event of high opportunities has a higher valuation of savings than a worker
not planning to deviate (assuming normality of consumption). Thus again taxing savings
eases the incentive compatibility constraint. Below, we consider models with labor
supply in both periods. Then, with uncertain second-period wages, first-period
consumption is occurring after first-period opportunities are realized but before second-
period opportunities are realized. The advantage of differential tax treatment of first- and
second-period consumptions naturally occurs in this setting.

We also note that the basic model has variation in earnings ability, and sometimes
in preferences, but not in wealth at the start of the first period, and so variation in capital
income if one were to consider a capital income tax. With variation in initial wealth
holdings and an ability to tax initial wealth, the optimum may call for full taxation of
initial wealth, particularly when higher wealth is associated with higher earnings abilities.
If immediate taxation of initial wealth is ruled out, the presence of capital at the start of
the first period, which can earn a return when carried to the second period, can also
prevent the optimality of the non-taxation of capital income. As a modeling issue, one
needs to ask where such wealth came from. Presumably gifts and inheritances are a
major source when considering a tax ignoring transition issues. But since these might
themselves be taxed and since gifts and bequests might be influenced by future taxation
of capital income, a better treatment of this issue would be embedded it in an OLG model
that incorporates the different ways that people think about bequests.15 A similar issue
arises in tax reform given past savings under the previous tax regime. This links to the
issues of government commitment and credibility and is discussed in Part II.

We have focused on the two-period model with earnings in the first period.
Models with education choice have two periods, with education in the first and earnings
in the second. Optimal tax treatment depends on the mix of opportunity and out-of-
pocket costs (Hamilton, 1987, Bovenberg and Jacobs, 2005). On-the-job investment in

15
See, for example, Boadway, Marchand, and Pestieau, 2000, Cremer, Pestieau, and Rochet, 2001. That
optimal taxation depends on bequest motivation is brought out in Cremer, Helmuth, and Pierre Pestieau,
2003.

15
human capital at the expense of foregone wages has also been studied. If labor income
tax rates are constant and there is positive taxation of capital income, then there are
unequal tax incentives, with human capital favored over physical capital. Combining a
linear income tax with a progressive labor income tax can help address this issue (Nielsen
and Sørensen, 1997). While there is some analysis of tax treatment of education and of
the implications of the presence of formal education for later taxation of earnings, we do
not explore this issue. A fuller treatment would incorporate randomness in the increase
in future earnings from investments meant to increase earnings potential, whether through
formal education, on-the-job training, or work experiences that vary with jobs.
Randomness in earnings potentials is a key part of the analysis in models with two
periods of earnings, which we turn to below.

Another approach to limiting the applicability of the Atkinson-Stiglitz theorem is

to recognize that relative wages depend on the demand for different skills, rather than
being parameters, as in the basic model. The commodity tax literature has noted that
taxes can modify labor demands as different goods make use of different proportions of
different kinds of labor. Since there is an advantage to altering relative wages which are
taxed subject to a single formula, there is a case for distorting taxes even with separable
utilities. In an intertemporal context, taxation that affects capital accumulation affects
relative wages. That is, central to the analysis would be the impact of aggregate savings
on the relative wages of high and low productivity workers. Such an impact matters for
optimal capital accumulation, and its influence on taxation depends on the link between
capital taxation and capital accumulation which is dependent on the total policy toolkit of
the government. As noted above, empirical work supports the finding that increased
capital raises unskilled relative to skilled wages (Krusell et al., 2000).

Uncertain preferences and lifetimes

In addition to uncertainty about future earnings, there is uncertainty about

future preferences. In the certainty setting, the Atkinson-Stiglitz theorem required
separability between labor and the vector of consumptions and the same subutility
function for all individuals - U i [ x1 , x2 , y ] = Ai ⎡⎣ B [ x1 , x2 ] , y ⎤⎦ . The subutility function

16
might be additive or reflect an impact of first-period consumption on second-period
utility. Above we explored the possibility of different subutility functions for workers
with different earnings potentials, opening up the use of taxation of savings to improve
the ability to tax higher earners. Here we consider a setting where everyone is the same
but there is uncertainty about preferences and incomplete markets, giving a role to
taxation in helping with the inefficiency from incomplete markets.

In making first-period savings decisions, an individual may be uncertain about the

future utility from consumption. This might hold in an additive structure or as a function
of current consumption. For example, individuals may be aware that getting used to a
standard of living in period one will affect marginal utility in period two, but not be sure
how much. In the additive structure, there may be uncertainty about how much
consumption will be enjoyed when older – either from an inability to fully appreciate
future preferences or from shocks which for some reason are not fully insured – such as
health shocks or spending shocks (medical or legal expenses) or an inheritance.16 With a
shift parameter α to reflect the different possibilities, we would write utility as
U i [ x1 , x2 , y , α ] = Ai ⎡⎣ B [ x1 , x2 , α ] , y ⎤⎦ . If earnings occur in the first period, then the impact

of utility-of-consumption uncertainty depends on the correlation between probabilities of

different levels of the shift parameter α and skill level. In the absence of a range of
models, it is not clear what sign to put on the optimal taxation of savings from this
consideration.

One example of significant uncertainty is in length of life. Moreover, longer

expected lives are positively correlated with earnings abilities. A full modeling of this
interaction would need to explore the properties of the annuities market (uniform pricing
or risk-class varying pricing and the degree of accuracy in risk classification) and the
social evaluation of bequests from incomplete annuitization. While the social security
literature has examined the impact of different life expectancies on labor market
efficiency in a setting of uniform pension rules, consideration of the implications for
taxing savings is limited – either inside annuitization or outside. Analysis would need to

16
Another source of uncertainty comes from uncertain future relative prices. This is present even with
savings in real assets based on a price index that is not precisely the right one for a given individual.

17
reflect the greater availability of historic information regularly used by social security
systems as opposed to the less historically structured income taxation.

Risky returns

Many models of optimal taxation assume safe returns to savings. Yet

returns to savings are random. The randomness may be modeled as perfectly correlated
across individuals – as would be the case with the risk coming from access to a capital
market with stocks and bonds and the same subjective probabilities for everyone.
Empirically, portfolios vary widely across households. Different people have different
subjective probabilities about assets and access to different information sources and
different investment opportunities. Also, not all investments are fully in market-traded
assets, which would matter even if everyone had the same opportunity set. This may be a
basis for using capital income taxes as part of optimization even with separable
preferences, but the literature incorporating nonlinear earnings taxation and stochastic
returns is limited.

The difference between taxing consumption and taxing income has been
described as exempting the safe rate of return from taxation, but still taxing the difference
between risky and safe rates of return the same. Lying behind this view is the analysis of
Gordon (1985) and Kaplow (1994) that linear taxation of the difference between risky
and safe returns has no effects with the uses of the revenue that they describe. Gordon
assumed that the tax revenue from each person was returned to that person in a
(stochastic) lump sum way. Kaplow’s assumptions are equivalent to having the
government sell the stochastic tax yields in the market and return the proceeds of the sale
to each consumer as a lump sum. In both cases, the imposition of the tax and lump sum
transfer policy has no effect on equilibrium.

However, if we are concerned with the distributional impact of taxing the

difference between risky and safe returns, we need to consider alternative uses of the
revenue, involving other tax changes. Such analysis needs to consider differences in
portfolios and differences in savings rates. And it needs to consider the supply of safe
and risky opportunities. We are not aware of analysis of this tax comparison in these
different settings. However, we have done some preliminary analysis of a model with

18
two types with the stochastic revenues returned to each in equal shares, drawing on the
related issue of investing social security funds in stocks (Abel, 2001, Diamond and
Geanakoplos, 2003). If the two types have the same preferences and one has access to
stocks while the other does not (perhaps reflecting a fixed cost of learning about stocks
that is higher than for the investor in stocks) and ignoring any change in the savings rate,
such a tax has no effect on the saver with a diversified portfolio while increasing
expected utility for the saver without access to stocks when the availability of safe and
risky assets is perfectly elastic. However, in the other extreme, of perfectly inelastic
investment opportunities, which are purchased from others (e. g., an earlier cohort), the
increased demand for risky assets and decreased demand for safe assets implies that the
return on the risky asset goes down and the return on the safe asset goes up, implying
further utility effects.

Another case to consider is where the risk in tax revenues is initially absorbed in
debt issues and responded to over time with tax changes, thereby spreading the risk
across (possibly all) future cohorts. Analysis of this case depends on how the
government spreads risk across future cohorts. The analysis of Gollier (2005) contrasting
optimized defined contribution and defined benefit pension plans with perfectly elastic
investment opportunities suggests that optimal adjustment of taxes and public debt can
significantly increase expected utility of a representative cohort.

2-period models – 2 labor supplies

While the model with a single labor supply decision can shed light on the relative
treatment of consumption when working and when retired, a model with two labor supply
decisions opens up issues about consumption and earnings during a career. Moreover, it
opens up issues of the relationship of modeling to implementation that are not present in
the single-labor supply model. Consider a setting where individuals work in each of two
periods and consume in each of two periods. For comparability to the simpler model (in
order to highlight what changes) assume that preferences are additive, both between
consumption and labor in a single period and across time - U = ∑ ( ut [ xt ] − vt [ yt ]) . The
t

19
implications of alternative preference structures are likely to be similar to those discussed
in simpler models above and we focus on earnings patterns over time.

Two key ingredients are the available tax tools and the pattern of productivities
for an individual over time - that the age-earnings profile is steeper for more highly paid
workers and that there is uncertainty about second-period earnings possibilities.

Mechanism design

The approach commonly taken in optimal tax analyses by public finance

economists has been to analyze tax structures that resemble actual ones. In contrast,
analyses by macroeconomists have explored optimization in a mechanism design
setting.17 This allows far more complex policy tools than we ever see. For example, in a
setting of two periods with two labor supplies, the link between second period
consumption (after-tax available spending) and second period earnings can depend in a
nonlinear way on both first-period earnings and first-period consumption. That is,
second-period consumption would depend on all three of the other variables, with no
restriction on the complexity in the functional form. Framing the problem in this way
represents a restriction that everyone has the same safe rate of return on savings since
otherwise we would also need to track capital income, involving a generalization over the
bulk of the existing literature. Once one envisions modeling longer lives, this degree of
interaction becomes implausible to implement. One strand of the literature has explored
assumptions under which the optimum can be implemented with tax structures that are
not so complex. Simplification is accomplished by taxing the wealth carried from period
t to period t+1 at a rate that varies not only with past earnings but also with earnings in
period t+1 (Albanesi and Sleet, 2006, Kocherlakota, 2005). Moreover, while taxes on
wealth need to vary with individual position, they collect no expected revenue, being a
combination of taxes and subsidies. The combination of taxes and subsidies is chosen to
preserve the wedge on savings that needs to be present – current marginal utility exceeds
future expected marginal utility assuming savings at the marginal product of capital. This

17
Public finance economists have largely focused on certainty models, while macroeconomists have paid
attention to individual uncertainties.

20
insight is intriguing, but it is not clear how directly useful this is given remaining
complexity and difficulties in measuring wealth.

It is not clear how the insights should carry over from a complex to a simple
setting. This is particularly an issue when we recognize the reality that there are more
types than there are tax tools, which raises further issues about how to learn from the
simpler models. This issue about inferences from simpler models arises with more
dimensions than types even when the number of types remains small (Judd and Su, 2005,
Tarkiainen and Tuomala, 1999). Thus a key question is what lessons to draw from such
modeling. One way to proceed is to note the properties of an optimal mechanism design
and then check to see which properties are also present in models with less general
government policies.

The literature has explored the presence of what have been dubbed
“wedges” – differences between the marginal rates of substitution and transformation at
the optimal allocation. In an economy with linear taxation, a wedge would correspond to
a nonzero tax, but wedges in a nonlinear tax structure are not generally implementable in
a (partially) linear one. There are four wedges that seem useful to consider – two
intratemporal wedges between earnings and consumption in the two periods and two
intertemporal wedges – one for consumption and one for earnings. While any three of
the wedges would permit derivation of the fourth, paying attention to all four is useful
when contemplating alternative restrictions on tax tools, related to implementation. Thus
if one has annual taxation of earnings (age varying or not) then the presence or absence of
an intertemporal consumption wedge is suggestive of the role of taxation of capital
income. On the other hand, if there is annual taxation of consumption, then it is the
intertemporal earnings wedge which is suggestive of whether the consumption tax should
be supplemented by taxation of capital income. Possibly, how the wedge varies with
other variables might be suggestive of whether taxation should be dual or integrated.

In the mechanism design setting, we can look at the relationship between

the intertemporal MRS in consumption and the matching MRT as well as the
intertemporal MRS in earnings and the matching MRT. That is, we can consider the
marginal utility impact of delaying consumption or delaying earning if there were no

21
implicit taxation (if the delay involved the market rate of interest). If, with zero implicit
taxation, there is an incentive to delay either consumption or earnings (or both) at the
optimal consumption and earnings plan, then there must be implicit marginal taxation (a
wedge) so that the full market rate of interest is not appropriate for an individual’s
calculation.

If preferences are additive, as above, with the same subutility functions for
everyone, and if there is certainty in earnings possibilities, then the Atkinson-Stiglitz
theorem tells us that there is no intertemporal consumption wedge. Nevertheless, there is
generally an intertemporal earnings wedge given the evolution of earnings possibilities.
In the case of certain future earnings, if higher earners have steeper age-earnings profiles,
the sign of that wedge is to discourage savings. Progressive annual taxes of earnings
generates such a wedge for those with rising earnings (and relates to the issue of human
capital accumulation mentioned above).18 Age-varying taxes can address this as well.
But having the PDV of lifetime taxes based on the PDV of lifetime earnings does not
have such a wedge. And note that annual taxes on consumption do not generate an
intertemporal earnings wedge. Moreover, if the age-consumption profile with optimal
taxes is not flat (as is empirically the case with existing taxes), then progressive annual
consumption taxes generate an intertemporal consumption wedge, which is not wanted
under these assumptions, while not generating an intertemporal earnings wedge, which is
wanted.

Apart from cases where the mechanism design optimum is simply implementable,
a key question is the relationship between findings in a mechanism design optimum and
findings in an optimization with fewer tools. One such analysis has been done for a
setting where the Atkinson-Stiglitz result holds (no intertemporal consumption wedge),
and taxation depends in a complex manner on earnings in all periods. That is, in a two-
period model, implementation can be done by taxing first period earnings, T1 [ z1 ] , in the

first period and then having a tax on second-period earnings, collected in the second
period, that now depends on earnings in both periods, T2 [ z1 , z2 ] . This intertemporal

18
Vickrey (1947) was concerned with the relative treatment by progressive annual taxes of those with
constant incomes and those with fluctuating incomes.

22
earnings interaction matters and the full optimum can not be implemented by separate tax
structures in each period, T1 [ z1 ] , T2 [ z2 ] when there are income effects. Gaube (2005)

shows that the zero marginal rate at the top of the tax structure with complex taxation,
T2 [ z1 , z2 ] , does not carry over to the optimum with separate tax structures, T1 [ z1 ] , T2 [ z2 ] ,

since higher earnings in one period lowers tax revenue in the other period given
normality. Thus the mechanism design optimum can not be implemented in this way.
Income effects are key to the connection between separate incentives and linked
incentives.

As indicated above, we consider uncertainty about future earnings to be a central

element to be addressed in the design of direct taxation. Even with all the Atkinson-
Stiglitz preference assumptions, uncertainty about second-period earnings implies an
intertemporal consumption wedge for optimal taxation with mechanism design (Golosov,
Tsyvinski and Werning, 2007). Moreover, an intertemporal earnings wedge is also
generally part of the optimum.

It is natural to expect some of these properties to carry over to settings with less
general taxation, and some of them are readily checked, but the literature has not
explored all of them. In contrast with the case of certain earnings discussed above, with
uncertain future earnings progressive taxation of annual earnings should be supplemented
by a tax on savings – which can be done with a tax on capital income. As noted above,
we are not aware of analysis of the case with uncertainty in both future earnings and
financial rates of return. As above, annual taxation of consumption does not generate the
intertemporal earnings wedge one would want when there is discretionary savings.

It would be interesting to explore the time shape of such intertemporal wedges in

models with realistic life spans and patterns of earnings growth, and alternative amounts
of information that individuals receive about future earnings. Higher current
consumption is then a signal of anticipated higher future earnings and may be relevant for
the optimal wedges.

With our topic being the tax base, it is natural to focus on wedges and implied
marginal tax rates. The theoretical results from the optimal tax literature have been

23
primarily focused on marginal tax rates. In contrast, calculated optimal tax structures
also shed light on average tax rates, something that has been harder to explore in more
general theoretical frameworks. Yet average tax rates clearly matter, particularly when
the intensive margin is being considered (see Choné and Laroque, 2001, 2006, Diamond,
1980, Saez, 2002c for the case of personal incomes or Griffith and Devereux (2002) for
the case of multinational corporations). They would be an integral part of the design of
the complex tax structures coming from mechanism design and matter in the attempt to
use less complex tax structures as well as possible.

3-period models – 2 labor supplies

If we consider a model with two earnings periods and one retirement period, we
need to face together issues that were faced separately above, leading to further
complications. In a model with only one earnings period and a retirement period, there is
a case based on separability for non-taxation of capital income. This case is undercut to
the extent that higher earners have less discounting of the future and longer life
expectancies that induce greater savings rates. In the model with two earnings periods
and two consumption periods, there is further strengthening of the need for some capital
income taxation from the uncertainty about future earnings. Turning to a model with two
earnings periods and a retirement period, we have all of the above working. That is, as
has been noted in a number of different contexts (da Costa, 2003, Mirrlees, 1995), the
intertemporal consumption wedge between retirement consumption and consumption in
the late earnings years depends on the issues in the two period model with only one
earnings period. The intertemporal consumption wedge between retirement consumption
and consumption in early earnings years is a compound of successive period rates and
thus also depends on the issues that call for an intertemporal consumption wedge between
early and late earnings years. Thus there is an interesting question how to address
taxation of capital income when using less complex structures than implicit in mechanism
design.

In practice, some tax systems provide special tax treatment for retirement savings.
In addition, countries have mandatory public pension systems that collect contributions

24
when working and provide benefits in retirement years. Consideration of the tax
treatment of retirement savings needs to be integrated with consideration of mandatory
retirement savings systems. The latter have focused a great deal on the behavioral issues
that, left to their own devices, many people apparently would not save enough for an
appropriate replacement rate and little use would be made of annuities. Thus there is an
inherent tension between the concern that many people do not save enough and the
optimal tax analysis above which calls for discouraging savings in the form of taxation of
capital income. To address this issue, one needs to recognize the wide disparity in wealth
to earnings levels (at any age) and the apparent diversity in the motivations behind
individual savings. Thus there is no necessary contradiction between mandatory
retirement savings, which will matter little (at the savings margin) for those with large
savings, and discouragement of savings for those who would save a great deal. Indeed, it
is tempting to think that there might be analysis recognizing such individual diversity and
finding simultaneous roles for mandatory savings that is capped for high earners and
retirement savings incentives that are also capped for high savers. But such analysis does
not yet exist. Nor have we seen analyses of the optimality properties of the different
ways of doing tax-favoring (EET (as in IRAs in the US or Personal Pensions in the UK),
TEE (as in Tax Exempt Special Savings Accounts or their successor, Individual Savings
Accounts in the UK, or Roth IRAs in the US), or both available (as in the US), and partial
taxation of accumulation (as was in Australia). 19

Overlapping Generations (OLG) models

The analysis discussed above focused on the intertemporal dimension of direct

taxation and noted a role for age-varying tax structures. A natural question is the
relevance of such analyses once one recognizes the reality of overlapping generations.
There are two aspects of the connection between analysis for a single cohort (as is
implicitly done above) and OLG analysis. One aspect is the determination of the

19
Exempt, Exempt, Taxable treatment allows an income tax deduction for deposits in an account, no
taxation during the accumulation phase and taxation at withdrawal. In contrast Taxable, Exempt, Exempt,
has no deduction for a deposit but then no taxation during accumulation or of eligible withdrawals.

25
government’s role in affecting aggregate capital available to different cohorts. Second is
the extent to which there are limits to how much taxes can vary with age.

If the government is free to use public debt and public assets to optimize national
capital and if taxes are age-varying, then a full optimization in the standard model can be
divided to include suboptimizations for each cohort in the usual formulation. That is,
from the intergenerational optimization there is a constraint on the net contribution to
national capital from each cohort. Using this net contribution as a constraint on
optimization of taxes for a cohort, then the type of optimizations we have analyzed above
hold in the basic case where there is no direct concern about relative prices. The analyses
with a concern about relative prices, particularly a concern about relative wages, do not
have this full separation. Presumably our analysis above remains strongly suggestive.
Other links would naturally arise in a setting where parents look after children.

The case that needs attention has the assumption that taxes are not allowed to vary
by cohort – that is they are period-specific rather than age-period specific. In the context
of age-varying taxes, we have seen a case for taxing capital income in the setting of a
single cohort. This section has argued that this case carries over, pretty much intact, to
embedding the single cohort in an OLG model with age-varying taxes and the
government free to coordinate debt policy with tax policy. An open question is whether
the lack of age-varying taxes would influence the use of capital income taxes in this OLG
setting. In the context of a single type of agent in each cohort, Erosa and Gervais (2002)
have examined implications of uniform labour taxes. If the utility discount rate differs
from the real discount rate, individuals will chose non-constant age profiles in both
consumption and earnings, even if period preferences are additive and the same over
time. This then calls for age-varying labour taxes if feasible and if not feasible for non-
zero capital income taxation. With multiple types and nonlinear earnings taxes, it
remains the case the age-varying earnings taxes are generally required for the optimum.
It is plausible that, the requirement of uniform labour income taxation across periods for
a single cohort would block the applicability of the Atkinson-Stiglitz result in general
given the reality of age-varying consumptions, efficiencies, and labor elasticities, all of
which can be present even with fully additive utility functions.

26
 Infinite horizon agents

When agents have long horizons, modeling their current decision-making using
an infinite horizon model can be mathematically more tractable than a long finite horizon,
while doing little violence to some conclusions from the analysis. However, when
considering the evolution of an economy over time, a model with a fixed number of
infinitely lived agents behaves very differently from an OLG model.

Consideration of tax policy in the distant future may be dependent on the

evolution of the economy and thus policy inferences from models with infinite horizon
agents may be different from policy inferences from OLG models, even ones with very
long-lived agents. Moreover, inferences from such models depend on the structure of the
model – both its positive and normative assumptions.

The starting place for a discussion with infinite horizon agents is the famous and
widely cited Chamley-Judd result that when such an economy is in a steady state, there
should be no taxation of capital income (with a linked convergence result). As Chamley
(1986) explained: “The main property of the model which is used in the proof is the
equality between the private and social discount rate in the long run.” (page 608) and, in
the altruistic dynasty interpretation: “When the social planner uses the same discount rate
for the future life cyclers as the discount rate applied in the altruistic families, the long-
run tax rate on capital income is zero. This property … requires that individuals not be
constrained at a corner solution for their bequest.” (page 613) or “This assumes that the
social planner and the individuals use the same relative utility weights for
intergenerational transfers.” (page 619). Once the weights differ, then the result
changes.20

As with the Atkinson-Stiglitz result, a key question is how robust the conclusion
is to realistic changes in the model. We reach the same conclusion in this case as in the

20
Farhi and Werning, 2005, consider the case of respecting individual dynastic preferences and also giving
weight to the dynastic preferences of later generations. As in Kaplow (1995) the thrust of such modeling is
to subsidize gifts and bequests since they benefit both the donor and the donee. The results would change
if the social welfare function treated dynastic concerns differently from utility of own-consumption in the
social welfare function, an issue considered in the context of charitable donations in Diamond (2006).

27
earlier analysis – the finding is not robust for policy purposes. Let us start with the basic
interpretation of the model before turning to detailed modeling assumptions. In the
standard OLG model, individuals have no concern for the future after their deaths and
leave no bequests. This is empirically inaccurate - most people leave some bequests and
we think that some people adjust earnings and/or savings in light of planned gifts and
bequests.21 Results vary in models extending the basic OLG model depending on how
bequests are modeled. Models with ‘accidental bequests’ because of incomplete
insurance/annuitization and models with planned bequests arising from motivation that
can influence earlier decisions will generate different positive and normative tax
implications. Empirically, how important bequest considerations are for behavior is
unclear and certainly widely varying in the population. Also key to further analysis is
how to form a social welfare function since counting both the utility of a donor and the
utility of a donee in a social welfare function has implications that can be questioned as
being normatively unattractive.

In contrast, the standard infinite horizon agent model is viewed as a dynasty

model with incorporation of future utilities in the decision-making of earlier cohorts and
a normative evaluation of the utilities-of-consumption of each generation in the same way
as viewed by the existing generation. This is typically done as if there were only one
generation alive at a time and lasting only a single period, rather than the multiple
overlapping generations that are present. In terms of the normative issue raised above,
this can be viewed as counting the utility of the donor and ignoring the utility of the
donee, and is one way to approach concern about overweighing the consequences of
concern for others.

It is useful to complement models unrealistically ignoring bequests with models

giving bequests a larger role in decision-making than they have in reality – at least until
we have better empirics and analytics around bequests. So an evaluation of the role of
other assumptions in reaching the no-capital-income-taxation conclusion is appropriate.
Note that in the asymptote of the usual models, each period is exactly the same for each

21
Part of the debate on the importance of intergenerational links for evolution of the capital stock relates to
the treatment of the financing of education and other gifts that occur well before the time of life
expectancy. This is ignored in this discussion which focuses on the transfer of financial wealth at death or
at a time when remaining life expectancy is not so large.

28
of the dynasties at the time. Recognizing that the dynasties are a collection of successive
individuals makes all of the issues considered above for a single cohort relevant in this
model as well. For example, those working have on average an age-earnings profile that
is not flat. Requiring uniformity in earnings taxation across workers of different ages
alive in the same period would presumably raise similar issues of capital income taxation
as it does in the single-cohort model. The analysis of Judd (1999) still gives an average
capital income tax that tends to zero even if it is not zero in any period.

In the single-cohort model, Naito (1999) has shown that endogeneity of relative
wages, together with a uniform earnings tax function, contradicts the optimality of zero
capital income taxes when relative wages can be influenced, even with the Atkinson-
Stiglitz separability assumptions. Correia, (1996) has shown something related in the
infinite horizon model. She assumed two kinds of labour and an inability to tax one kind.
The adjustment of capital to offset the absence of taxation of this labour results in a long-
run equilibrium with non-zero taxation of capital, with the sign depending on the details
of the technology. A similar result can occur if the two types of labour must be taxed the
same (and capital affects relative wages) or if one of the two types of labour must be
taxed the same as capital income is taxed (reflecting an inability to tell apart capital and
some labour incomes.

Also, as in the one-cohort model, uncertainty about the future earnings of those
alive and already working as well as about the earnings of those not yet in the labor
market or not yet born removes the optimality of zero taxation of capital income
(Golosov, Kocherlakota and Tsyvinski, 2003).22 Aiyagari (1995) and Chamley (2001)
considered borrowing constrained agents. Their precautionary savings leads them to over
save relative to the social optimum, which means that a positive capital tax is welfare
improving in the standard setup.23

When the model is interpreted as each generation living for a single period, a tax
on capital income is equivalent to a tax on bequests. Once individuals live longer than a

22
Analysis of uncertainty that affects all earnings possibilities proportionally is quite different. See
Golosov, Tsyvinski and Werning, 2007.
23
Chamley has an example where randomness is in the timing of future incomes, with the outcome learned
ahead of time, giving an advantage to subsidizing capital income rather than taxing it.

29
single period, then one can distinguish between a tax on interest income and a tax on
bequests, which has implications for optimal taxes. This point has been made by
Chamley (1986, page 613) “If a specific tax can be implemented on the interest income
of savings used for life-cycle consumption, its rate is in general different from zero.” The
same point can be seen in a model with time-varying period utility functions (that are not
separable between consumption and labor). Assume the period utility functions are the
same in all even-numbered years and all odd-numbered years, but different across
adjoining years. Then there will be alternating taxes that would show long run zero
taxation across pairs of years (consistent with taxation being zero on average in Judd,
1999). To preserve this long-run pattern while distinguishing between capital income
and bequest taxes, if one were taxing capital income, then one would be subsidizing
bequests. Such a starting place for analysis would focus attention, appropriately on
analysis of bequest motives (and their heterogeneity).

Additional considerations arise when there is human capital as well as physical

capital in an infinite horizon model. In the presence of both physical and human capital,
labor is supplied jointly with human capital, which means that a positive labor tax is also
taxing capital. In this setup, it is optimal to converge to zero capital and labor taxes
(Jones, Manuelli and Rossi, 1997) unless human capital is observable. If a direct subsidy
on human capital is available, then it is optimal to have positive labor taxes in the long
run accompanied by a subsidy on human capital and zero taxes on physical capital (Judd,
1999). The result with unobservable human capital suggests that the accumulation of
sufficient government resources, relative to expenditures, is a key part of the result on the
optimality of zero taxation. Thus, at a point of time with a non-optimal tax structure, it is
not clear whether the result that long run taxation of capital should stop is a call for
increasing or decreasing the current taxation of capital income.

Another source of concern about the results in existing models is that the tax on
capital income is linear. Saez (2002a) has examined the role of a linear tax with an
exemption, as opposed to a tax linear from the origin. Asymptotically no one is paying
the tax, as wealths above the exemption level decline to the exemption level – with
everyone having the same discount rate the before-tax interest rate is driven to the
discount rate in a steady-state, implying a lower after-tax return if there were dynasties

30
with wealth above the exemption level. But the tax has served to raise revenue from
those with the highest wealth, reducing their wealth to the exemption level and an
exemption level that is finite (as opposed to infinite which would be equivalent to no tax)
is part of an optimum.

The standard presentation of the intuition behind the Chamley-Judd result is to

examine the impact of repeated taxation on the relative price of current and future

consumption in period T - {(1 + r ) / (1 + r (1 −τ )) } . With a positive rate of tax this

T

expression goes to infinity as T goes to infinity. This explosive effect overrules other
considerations for taxing capital coming from preferences. But we have seen that this is
explosive effect does not overrule considerations coming from uncertain earnings or pure
rents, suggesting that the real intuition for the result is coming from something additional,
perhaps the convergence property, which would not be a strong foundation for basing
future policies.

Thus we conclude that the Chamley-Judd result that there should be no taxation of
capital income is not a good basis for policy. Nevertheless the issue remains of the
compounding of taxation of capital income resulting in a growing tax wedge the longer
the horizon for decision-making – a point also made in models with finite lives of many
periods. This is suggestive of a possible role for capital income taxation that varies with
the age of the saver and/or with the time lapse between savings and later consumption (as
with tax-favored retirement savings). The role of capital income taxation when earnings
are uncertain particularly suggests that rules might well be different for those at ages
when workers are mostly retired.

Behavioral models

Behavioral economics has become a major research area for many economists and
some of the findings are very exciting (for a survey see Bernheim and Rangel, 2005,
forthcoming 2007). Indeed, analyses of the difference in outcomes with opt-in and opt-
out rules are already influencing policy makers in both the US and the UK. And the
details of tax-favored savings programs matter for the degree of take-up. Behavioral

31
analysis of savings behavior is highly relevant for the choice of tax base. It is also key
for interpreting the role of mandatory programs that require contributions when working
and provide cash benefits when retired. And these two policy institutions need to be
considered together. A key tax policy design issue is how to combine concern that some
fraction of the population saves too little for an adequate replacement rate in retirement
while another fraction saves too much, resulting in their retiring too soon from the
perspective of social welfare optimization. And heterogeneity in life expectancy,
intertemporal preferences, and consumption history (in light of the realistic links between
past consumptions and later marginal utilities of consumption) all call for diversity in
individual saving rates. Indeed, consideration of the importance of both undersaving and
oversaving for the taxation of capital income and for tax preferences for retirement
savings is an important issue for future research. To date, the literature has addressed one
of the other but not both, limiting the available inferences. EXPAND ??

Concluding remarks

The information needed for a full implementation of an optimal tax calculation is

never going to be available, especially once one moves from simpler models to more
realistic ones incorporating uncertainties and decision-making complexities. Thus one
needs some default basis for tax policy – some basis for setting tax rules that is overruled
in a context of sufficient information to warrant a judgment that an alternative tax really
can do better. Is this a place where ability to pay might enter? Or where assumptions of
separability and uniformity might be appropriate?

After discussion of some foundational issues in the place given to social welfare
maximization in this essay (part II) we return to the policy discussion in Part III, which
can be read while skipping the intervening discussion of additional normative issues.

Part II. Broader tax base issues

Social Welfare

32
 The objective function typically used in optimal taxation studies is a concave
function of individual utilities, possibly additive over a social cardinalization of
individual utilities. Thus it is assumed that in doing the optimization the distribution of
consumer preferences is known to the optimizer, even if individual preferences are
(implicitly or explicitly) modeled as not known by the optimizer. Since empirical
demand studies give us some information about the pattern of demands, but the detailed
individual distribution is not known, usable policy inferences must be drawn carefully.
The choice of a cardinalization of utilities (when utilities are added), or the shape of the
social welfare function more generally, reflects an ethical choice in evaluating alternative
distributions of utilities. Since such ethical choosing is not part of scientific economics,
economic analysis explores the relationship between different welfare criteria and
optimal tax structures. Much of the literature assumes that preferences are identical
across people, with differences only in budget constraints (usually modeled as differences
in productivity per hour of work). The identical-preference structure can be interpreted
as a form of treating everyone similarly in the welfare structure – not paying attention to
correlations between skill and other dimensions. But as we discuss below, these
correlations have important implications.

This optimization approach is often accused of assuming a benevolent

government. This criticism has both right and wrong elements. Calculation of what a
benevolent government should do is not the same as assuming that there is a benevolent
government. Rather it is asking a key question - what policies would one want to see a
benevolent government follow. Used carefully, the answer to such a question can inform
a democratic debate about government policies – which is all that academic economic
research can hope to accomplish by itself. Moreover, it is hard to see how one gives
policy advice without knowing the link between good design of policies and the
accomplishment of social ends.24

24
As Musgrave wrote: “Just as homo economicus or a competitive Walrasian system are useful fictions to
model an ideal market, so it is helpful to visualize how a correctly functioning public sector would perform.
… Unless “correct” solutions are established to serve as standards, defects and failures of actual
performance cannot even be identified.” Buchanan and Musgrave, 1999, page 35.

33
 The relevant part of the accusation is that the political tendencies of actual
governments are highly relevant for good policy recommendations. Awareness of
political tendencies can readily take two separate forms. One is to extend optimal tax
theory to incorporate additional constraints reflecting what governments are likely to do,
either in response to recommendations or in future policies that follow after current
legislation. This is a richer, and possibly more relevant, environment than considering a
constitutional approach to limits on taxability. Indeed, the literature on tax policy
without government commitment is a form of such analysis, although one that typically
does not have a well-developed, empirically well-supported theory of government
behavior in a democracy. A second form that awareness of political tendencies can take
is through judicious use of the insights from optimal tax results when moving from basic
theory to policy recommendations. Recommendations can reflect beliefs about the
workings of the political process, based on the current state of politics and political
science and projections of political evolution. In the absence of much literature on
optimal taxes with political constraints, we try to be judicious in our inferences.

The optimal tax approach has generally considered taxes in a single period or for
the lifetime of a single cohort (sometimes in an overlapping generations (OLG) setting).
Some analysts think that when modeling individual behavior as a maximization of
preferences, those individual preferences should always form the basis for the social
welfare function (respecting revealed preferences). However, we question the
appropriateness of requiring that tight link in several settings, as well as recognizing that
not all behavior can be modeled as a time-consistent maximization. One such setting is
where people care about each other. This is particularly an issue in models including
bequests or charitable gifts. It is also an issue where the allocation of resources within
the family is important. Taking the same preferences for both positive and normative
uses has also been questioned in settings of negative feelings – particularly envy. We
note these issues, but do not explore them since we consider annual direct taxation in
model settings that ignore gifts and donations. This is not fully satisfactory since, for
example, inferences about the tax treatment of capital income when savings are to finance
later consumption may need to be adjusted for the partial use of savings for gifts and

34
bequests, gifts and bequests that may not be taxed in a way that is adequately integrated
with the tax treatment of capital income.

Another source of concern about the formulation of the objective function arises
in intertemporal settings to the extent that some people may not exhibit time consistency
in their behavior.25 Since this issue is indeed central to the analysis of the relative
taxation of capital and labor incomes, we return to it later after exploring implications of
models with fully-rational agents. For now, we simply proceed with preferences without
incorporating the complications we have just identified. But it is worth noting before
getting to the fairness issues we discuss next, that we recognize that there are questions
about the social welfare approach based solely on utilities as revealed by behavior. This
approach is based on the idea that a good starting place for policy is the policy for fully-
rational agents, a policy that can then be adjusted in recognition of the inadequacy of the
assumption that all individuals show fully-rational behavior. For example, in considering
the taxation of capital income, we first ask how that should be done in an economy with
only fully-rational agents and then ask about adjustment in recognition of some fraction
of agents who do not save enough. Even the first step is complex given the many aspects
of the economic environment that are present, which are modeled separately in optimal
tax analyses because of the difficulty in making inferences if the model has many
complications at the same time.

Fairness and Inequality

While the optimal tax literature works simply with a social welfare function, there
are two other literatures showing normative concerns relevant for taxation that we
consider. One is the attention paid to income distribution, particularly using a social
income evaluation function as developed by Atkinson (1970). Second is the long
tradition, linked to horizontal equity, of an ability-to-pay basis for taxation and the more
recent formal considerations of horizontal equity for tax analysis.

25
Examples exist of analyses with time-inconsistent quasi-hyperbolic preferences and with the simple
assumption that some people do no savings at all. A similar issue of the appropriate objective function
arises if the analyst is concerned that individuals discount the future excessively even if they are time-
consistent.

35
 Social Income Evaluation Functions

To make the distinction between a social income evaluation function and a social
welfare function clear, let us consider the special case where each individual has a utility
function that is additive, with a function of consumption and a function of labor worked -
U i [ xi , yi ] = u i [ xi ] − v i [ yi ] . A social welfare function is a function of individual utilities,

for example a sum, W = ∑ ( u i [ xi ] − vi [ yi ]) . In contrast a social income evaluation

i

function depends on the social evaluation of individual (net-of-tax) incomes or

consumption levels. Again assuming a sum, this might be written W% = ∑ ( u i [ xi ]) . And
i

it may be informative to use the same utility function for everyone even though it is
readily recognized that that is not the case. For example, this arises in considering
medical expenses. These objective functions could be considered on both annual and
lifetime bases.2627

Appropriately, a great deal of attention is paid to income distributions – such

attention can be informative about what is happening in an economy (see for example,
Piketty and Saez, 2003, 2006) and can inform a concern for its social implications. And
attention to incomes is also informative since there is so much variation in incomes.
While there is also variation in hours worked, this is studied separately (Prescott, 2004;
Greenwood and Guillaume, 2005) and does not have the same social implications. But
the question for this essay is whether, when thinking about the base for direct taxation, it
makes sense to start by considering taxes that optimize a social income evaluation
function rather than a social welfare function?28

26
This approach does not directly reflect concerns that large incomes and wealths can have impacts on the
political process.
27
One approach to extending the income evaluation function is to “discount” labour income for the cost of
effort, thus converting an income measure into utility measure (under strong assumptions). From the
perspective of this section, that would be a particular form of a social welfare function.
28
Of course one might optimize a weighted sum of the two. And one might be concerned merely to limit
the extent of income inequality, although defining that in an interesting way has difficulties, as brought out
by Atkinson, op cit.

36
 One approach to answering this question is to compare the implications of
optimizing different objective functions. Lack of concern with the disutility of labor in
the objective function will tend to seek allocations that involve more work since the
income earned (or consumption produced) will raise the objective function, while the
disutility does not impact the objective function. This can impact differently workers of
different skills. The degree of difference in the implications depends on the tools and
constraints for the optimization. In keeping with the basic income tax framework,
assume that earnings are observable but hours of work are not.

If the distribution of productivities (among those not labeled disabled) is bounded

away from zero (however low the bound), then maximizing a social income evaluation
function would call for incentives that have everyone (except those labeled disabled)
work, even those with very little productivity. Indeed, those with the lowest skill would
work the maximum possible. Heavy labor by those with little productivity would have a
zero direct cost to the social income evaluation function. The gain from having low
productivity people work is that it would weaken the incentive compatibility constraint,
making it less attractive for higher earners to imitate low earners, assuming that the basis
of taxation is earnings not adjusted for productivity. That is, the gain is because higher
productivity people, contemplating less work, and thereby imitating lower productivity
people, would find this less attractive than imitating someone with minimal income and
no work. This would be more important in a model without continuity in utility and
constancy of earnings per hour as hours approach zero.29 In that way higher incomes can
be generated, enhancing the level of the social income evaluation function.30

If there are people who can not earn (and can not be identified as such), then the
degree of inducing labor from productive people is limited by the incentive compatibility
constraint that others not stop working altogether. Doing such an optimization has some
interesting differences from optimizing a social welfare function, involving pushing

29
The importance of this issue depends on the time frame being examined – shorter time frames separately
evaluated make is harder to earn just a little money for a given annual total.
30
A standard result in optimal taxation with a social welfare function is that the marginal rate at the top of
the earnings distribution is zero. This follows since with any positive tax, there is a Pareto gain available
by lowering the tax rate. Optimizing a social income evaluation function, those with the highest
productivity are taxed (assuming a finite maximum), making clear that the use of this objective function
calls for passing up a possible Pareto gain.

37
people to work more than if their disutility entered the objective function. An interesting
comparison is with a Rawlsian social welfare function (maximin of individual utilities).
If there are people who can not work, the Rawlsian optimum has a similar lack of impact
on the objective function from the disutility of those who work. But there is also a
matching lack of impact for the utility of their consumption. With a social income
evaluation function, the utility of consumption counts in the objective function for
everyone (except in the limiting maximin solution). This highlights the push for more
work from the absence of the disutility of labor in the social objective. It is not clear why
one would want an objective function that ignores the utility cost of generating the
incomes that matter for the objective function.

In sum, while we share a concern about income distribution, a social income

evaluation function is no substitute for a social welfare function in thinking about tax
policy.31 This approach appears to give too much weight to encouraging work and we do
not think it is a useful variant on social welfare function maximization. Nevertheless, one
might consider limiting income variation (perhaps because of political implications),
which would also imply rejecting possible Pareto gains.

Ability to pay

Historically there have been two different approaches to an ideal tax base – one
drawn from ability to pay and one drawn from benefits received from government
spending. 32 We note that earmarked taxes for particular expenditures are a common
feature of advanced countries (particularly in the context of social insurance) and
recognize the important political role such earmarking can play. And there may be a
direct normative gain from such earmarking in some circumstances. Discussion of the
pattern of benefits received from government spending programs that affect the entire
population did not achieve any consensus on the distributional significance and has

31
Nor do we see a case for an objective function that combines both a social welfare function and a social
utility function.
32
For historical discussion, see Musgrave, 1959.

38
disappeared from discussion of an ideal tax base.33 Our purpose, as in the Meade Report,
is to examine the base for the provision of general revenues, not earmarked ones and, like
the Report, we consider only ability-to-pay concepts.

Going back at least to Adam Smith, economists have considered what would be a
fair base for taxation (along with the issue of the fair degree of progressivity, given the
chosen tax base).34 The Meade Report states: “No doubt, if Mr Smith and Mr Brown
have the same ‘taxable capacity’, they should bear the same tax burden, and if Mr
Smith’s taxable capacity is greater than Mr Brown’s, Mr Smith should bear the greater
tax burden. But on examination ‘taxable capacity’ always turns out to be very difficult to
define and to be a matter on which opinions will differ rather widely.” (Page 14.)35 The
Report goes on to ask: “Is it similarity of opportunity or similarity of outcome which is
relevant?” and “Should differences in needs or tastes be considered in comparing taxable
capacities?”

We begin by recognizing the core argument for consideration beyond a standard

social welfare maximization, as stated by Musgrave in Buchanan and Musgrave (1999).

The state and its public sector thus form an integral part of a multifaceted
socioeconomic order. …
That order, I hasten to add, includes not only the Pareto efficient use of
resources, important though that is but also other and no less vital dimensions of
social coexistence-distributive justice and the balance of individual rights and
obligations upon which a meaningful concept of liberty has to be built. A view of
fiscal economics, which holds that all is well if only Pareto optimality prevails,
bypasses these essential components of social coexistence and fails on both
normative and positive grounds. Without allowing for a sense of social justice the

33
For example, it is hard to see how to allocate the benefit of military spending by income level in a way
that is not so arbitrary that it can not add to a discussion of taxes.
34
“The subjects of every state ought to contribute towards the support of the government, as nearly as
possible, in proportion to their respective abilities; that is in proportion to the revenue which they
respectively enjoy under the protection of the state.” Adam Smith, Wealth of Nations, New York: The
Modern Library, 1937. page 777.
35
The Meade report is not the only examination of taxation that concludes that taxable capacity is hard to
define in a way to compel wide acceptance, as is needed for the role as an agreed-on normative basis. For
example, Vickrey (1947) writes: “In a strict sense, ‘ability to pay’ is not a quantity susceptible of
measurement or even of unequivocal definition. More often than not, ability to pay and the equivalent
terms “faculty” and “capacity to pay” have served as catch-phrases, identified by various writers through
verbal legerdemain with their own pet concrete measure to the exclusion of other possible measures.
Ability to pay thus often becomes a tautological smoke screen behind which the writer conceals his own
prejudices.” [footnote omitted] (page 3-4.)

39
 good society cannot be defined, and without it democratic society cannot
function. Page 31-32.

It seems useful to distinguish three elements in the “fair” taxation of individuals.

One, reflecting the role of individuals as ends in themselves, and not merely means to
increase social welfare, calls for fair treatment of individuals in terms of an ethical basis
for fairness. Second is the extent to which a concept of fair taxation used in tax analyses
can influence government behavior, encouraging both the design of tax institutions and
the implementation of policies that better satisfy social objectives. And third is the
citizens’ perceptions of fairness, which may or may not coincide with some philosophic
measure, and which matter for both the political process and individual compliance.

Let us consider these issues in the somewhat analogous, but much starker, setting
of punishment for criminal activity. First, severe punishments as deterrents, particularly
in the presence of limited apprehensions of those committing crimes, may go too far,
violating a sense of the proper treatment of individuals. Indeed, Amendment VIII of the
US Bill of Rights states: “Excessive bail shall not be required, nor excessive fines
imposed, nor cruel and unusual punishments inflicted.” Similarly, taxes should not be
defined differently for different people in ways that would violate “equal protection of
the laws.”

Second, reliance on selective enforcement and severe punishments might leave

too much power to the discretion of officials deciding which alleged criminal acts are
pursued in court. In the tax setting, Adam Smith argued: “The tax which each individual
is bound to pay ought to be certain, and not arbitrary. … Where it is otherwise, every
person subject to the tax is put more or less in the power of the tax-gatherer.” Page 778.

And third the perception of excessive punishment may not only violate the extent
to which actions of the state should reflect the views of the citizens, but also may be self-
defeating if juries are not willing to convict when they view the punishment as too severe.
Similarly, taxation perceived as unfair may encourage evasion.

Tax assessments do not affect individuals as sharply as some criminal

punishments, as long as tax collections are not too large relative to an individual’s ability
to pay. Nevertheless we have the same three elements. Consider the situation analyzed

40
by Atkinson and Stiglitz (1976) and Stiglitz (1982b), where social welfare maximization
calls for different tax treatment of two identical individuals.36 Total reliance on social
welfare function maximization would not be directly concerned by this difference in tax
treatment. However, a concern for fairness would strictly prefer a truly random, ex ante
equal probability mechanism for deciding which individual gets which tax assessment
(Diamond, 1967).

But there are several concerns about such an approach. Will the implementation
mechanism ensure that the randomization is done properly, avoiding improper
assessments? And will individual citizens accept this approach to fairness? These issues
arise even if there is sufficient information to conclude that unequal treatment is the right
approach, as may or may not be the case, and even if the legislature is sufficiently
sophisticated to be willing to accept and vote a suitable implementation. Randomization,
as was done for the US military draft during part of the Vietnam war, might be safe from
manipulation. But given the complexity and empirical uncertainty of an argument for
different treatment, we have doubts that the citizens would ever accept the underlying
argument that it is better than simply levying the same taxes on those in the same
circumstances. This is particularly an issue if the tax rate differences are to be long-
lasting. Such a concern, assuming it correct (without any underlying polls or focus
groups) lends itself to the idea that some aspects of horizontal equity may best be
addressed by viewing them as a limitation on allowable tax tools, as has been argued by
Atkinson and Stiglitz (1980). We accept the view that tax tools should be limited by such
considerations and that policies should be restricted to ones which are uniform over their
stated tax base.37 One would need a great deal of faith in the political process to not want
some protections against arbitrary tax assessments under the guise of “better taxation.” A
complication in structuring protections lies in the definition of ‘arbitrary.’ If one actually
can increase social welfare by drawing distinctions between individuals, are the
distinctions still arbitrary? A concern with actual motivations in the political process

36
As Atkinson and Stiglitz, 1976, Page 355 note: “If tastes are identical, the equal treatment of equals is
still not necessarily implied by welfare maximization … where the feasible set is non-convex, treating
otherwise identical individuals differently may increase social welfare.”
37
Randomized auditing of returns does not seem unfair to us or, apparently, the public as long as the
probabilities are suitably selected and the audits are not unduly unpleasant.

41
needs to lie behind restrictions on tax policies, and concepts and discussion of horizontal
equity is likely to be very helpful in addressing this issue.

Horizontal equity

Three arguments coming from horizontal equity have received considerable

attention in the tax literature. One is that Haig-Simons income38 is the best measure of
ability to pay and therefore horizontal equity calls for Haig-Simons income as the tax
base. Another is that annual consumption is a better measure of ability to pay and
therefore horizontal equity calls for consumption as the tax base.39 And a third, drawing
on Feldstein (1976a and b) is based on utility rankings with and without taxes.

But there can be tension between tax bases thought to be fair and tax bases
that optimize social welfare. What if one thinks that the best measure of ability to pay is
Haig-Simons income and one also accepts the empirical validity of the conditions under
which the social welfare optimum involves no taxation of capital income? What if one
thinks that the best measure of ability to pay is consumption expenditures and one also
accepts the empirical validity of the conditions under which the social welfare optimum
involves positive taxation of capital income? The weight that should be given to a
chosen measure of horizontal equity in offsetting the conclusions from social welfare
optimization depends on the strength of conviction that one really does have a good
(usable, widely accepted) measure for horizontal equity (and sufficient strength in the
belief that this consideration matters).40

To pursue possible formulations of horizontal equity, let us consider a basic one-

period, two-good model – where consumption and earnings are the same. As indicated in
the Meade Report, there is tension between the idea that ability to pay is based on actual
outcomes or on budget sets (potential outcomes). If everyone really does have the same
preferences over work and leisure, and the preferences have plausible properties, then

38
Haig-Simons income is labor income plus accrued capital income – Haig (1921), Simons (1938).
39
This view is generally supported by the further argument that it is better to tax people on what they take
from the economy (consumption) than a measure of what they provide (income).
40
Another concern is that the choice of tax base will influence the degree of progressivity because of
political behavioral effects – it is one thing to envision a consistent optimization across interacting
dimensions of tax policy and another to recognize that the political process has some sequential elements.

42
there is no tension between the actual and potential measures and thus earnings are
perfectly correlated with the potential earnings tax base. This convergence of different
competing measures of ability to pay would strengthen the case for paying attention to
the issue. However, this case has no bite since with identical preferences in this two-
good model, there is no conflict between this horizontal fairness concept and the standard
optimal tax calculation since individuals with the same productivities pay the same taxes
in equilibrium.41

In modeling preferences in an optimal tax problem it is common to

use u [ x ] − v [ z / n ] , where x is consumption, z is earnings, and n represents what varies in

the population. The variable n is normally interpreted as skill. With this interpretation,
there is no tension between optimal taxation and a horizontal equity measure based on
potential earnings. But, the optimal tax structure is exactly the same if n reflects the
extent of dislike of work rather than skill. In this case everyone has the same potential
earnings, as normally interpreted in terms of a budget constraint in hours-consumption
space, yet those with less dislike of work are taxed more heavily. If hours of work were
observable, the two cases could be distinguished. If hours are not observable, does the
distinction between interpretations matter for the appeal of the calculation? Is there
really a good ethical basis for treating ability to earn per hour differently from dislike of
working for an hour.

Similarly, with a continuum of skills and diverse preferences at each skill level,
different earnings levels are reached by different workers with the same skill but different
disutilities, thereby violating a measure of horizontal equity that is based on the workers’
budget sets rather than the workers earnings or consumption levels. In other words,
satisfying horizontal equity defined as workers with the same budget set should pay the
same taxes is impossible in a sensible setting.42

41
If all workers at each skill level have the same preferences, differences in preferences across skill levels
may or may not be a problem for horizontal equity, although the degree of progressivity of an optimal tax is
likely to be affected.
42
We focus on the earnings picture since it makes the same point as the one with different discount rates
and so different savings rates, which is the more common setting for calling for taxation that does not vary
with savings levels since the budget sets are the same. We see no good basis for distinguishing between
these cases.

43
 There are two directions to go from this point. One is to argue that horizontal
equity should be defined in terms of actual earnings or consumption, not the budget set –
which will happen automatically for a tax defined in terms of a tax base that matches the
equity measure, although this is more complicated if ability to pay is some combination
of labor and capital incomes. The other direction is to consider some measure of the loss
to the social objective function from differences in taxes paid by workers with the same
budget set. Such a calculation can be made if the distribution of types is known, without
knowing who is who. Such a concern will alter the degree of progressivity of the optimal
tax function. And in a more complicated setting it may alter the relative tax treatment of
different components of income or spending.

However, social reactions to chosen levels of earnings vary with the cause of the
difference in earnings. Viewing a worker as lazy (liking leisure) is very different from
viewing a worker as having difficulty working longer, perhaps for physical reasons. And
some people choose lower paying jobs because of the characteristics of the jobs, which
might reflect simply standard preferences or might reflect other concerns, such as a desire
to “do good works” by working in the nonprofit sector, or perhaps pursuing a religious
calling. That is, the realized relationship between earnings and earnings potential does
not seem a sufficient statistic for a normative judgment. Should those choosing poverty
for religious reasons be taxed on their abilities to earn in the commercial world?
Admittedly, the presence of characteristics of jobs that are not subject to taxation along
with taxation based on actual earnings implies a distortion in the choice of jobs. Perhaps
these considerations would become less important if the tax code were accompanied by
subsidies of certain activities – those viewed as generating externalities or particularly
socially worthy in a way not captured by a standard social welfare function. But then we
are choosing a complex solution, not only in taxation but also in spending, a complexity
that may be beyond the capability of the legislature. In sum, we do not find a convincing
basis for accepting the budget set as an adequate proxy for desired taxation. Nor do we
find realized earnings appropriate, for pretty much the same reason viewed in reverse –
sometimes the budget set is a better measure. We conclude that we can not see a good
argument for adjusting taxes away from an optimal tax calculation based on concerns
drawn from budget sets – which recognizes productivity but not preferences. Nor do we

44
see a strong case for deviating from an optimal tax calculation based on realized income
or consumption. As the Meade Report put it: “But on examination ‘taxable capacity’
always turns out to be very difficult to define and to be a matter on which opinions will
differ rather widely.”

Horizontal equity based on hypothetical alternatives43

The discussion of ability to pay was based on a cross-section pattern of budget

sets or actual outcomes in an existing equilibrium. Another approach is based on
comparing outcomes in an existing equilibrium with outcomes in a hypothetical
alternative. The latter may consider changed behavior by individuals one-at-a-time or by
everyone at once, thereby incorporating general equilibrium responses. 44 A general
equilibrium approach seems particularly relevant for transition issues, which are
discussed below. A set of altered outcomes one individual at a time considers what a
single individual would do if that individual were exempted from taxation, with prices in
equilibrium unchanged. Thus, horizontal equity is approached in terms of the vector of
utility calculations in the hypothetical alternative and the vector of utility calculations in
equilibrium.

A sizable literature has explored horizontal equity based on such a comparison

with particular attention to the rankings of utilities in equilibrium and in the hypothetical
alternative. The focus on rankings derives from Feldstein (1976a, 1976b), who used
them in considering tax reform (including a move from having no tax to having a tax),
not in judging whether an ongoing tax was horizontally equitable.45

43
This section draws particularly on Atkinson, 1980 and Kaplow, 1989.
44
This distinction is not as clear as appears. For example, when considering tax exempt bonds, one can
recognize that the bonds would pay higher interest if taxable, relying on an arbitrage interpretation of
current equilibrium prices without considering interest rate changes that would occur in an equilibrium
response to removal of tax exemption (as, for example, in Diamond, 1965).
45
“The principle of horizontal equity in tax reform thus requires that any tax change should preserve the
utility order, and should imply that if two individuals would have the same utility level if the tax remained
unchanged, they should have the same utility level if the tax is altered.” (Feldstein, 1976b P 124.) Feldstein
recognizes that satisfying this definition of horizontal equity is not possible and thus calls for a balance
between the degree of horizontal inequity and social welfare maximization.

45
 As an example of the literature, Rosen (1978) considers the pattern of utilities if
each person were allowed to maximize utility at equilibrium prices but without taxes.
This resembles the measurement of sacrifice in sacrifice-based theories of optimal
taxation (again, see Musgrave, 1959). Rosen then looks for utility reversals between this
vector of utilities and the vector in the actual equilibrium. We see no reason to give
normative consequence to this particular hypothetical alternative, nor have we seen one
offered. 46 And we see no reason to be particularly concerned with utility reversals in
this comparison or more generally. That is, the hypothetical alternatives depend on the
behavior of both the government (through expenditures) and other individuals (in
determining prices). Thus it is not clear why an individual has a particular claim to
protection measured from such a position, since the position depends on everyone’s
behavior – it is not generally something achievable on one’s own in a world without
government expenditures and trade with others. Indeed the taxes themselves play a role
in the determination of relative prices. Moreover, there are likely to be other hypothetical
alternatives that appear as normatively plausible as this one, for example the world with
no taxes and no government spending - no police, no regulation of markets, etc. This
would take us back to the benefit approach to taxation, which has suffered from an
inability to make useful distributional inferences. And why those best capable of looking
after themselves in some such hypothetical setting should be tax protected is not
apparent.

As to giving great importance to rankings – we agree with Kaplow’s (1989)

criticism of such measures: “Minute movements leading to order reversals count as full
violations of [horizontal equity] while substantial disturbances in the initial distribution
that result in no order reversals are ignored.” [Footnote omitted] (Page 141.) More
generally, there is no obvious reason why rankings matter at all normatively.

The end of this rambling discussion is that we reject the Meade Report view that
taxes “should” relate monotonically to taxable capacity. In addition to finding taxable
capacity not well-enough measurable and not sufficiently uniformly evaluated to be

46
In referring to Feldstein and the literature pursuing measures of inequity following his approach, Kaplow
writes: “HE [horizontal equity] is now frequently measured and applied even though there has been
virtually no exploration of why one should care about the principle in the contexts and in the manner in
which it is now being used.” Page 139.

46
usable for this purpose, we also do not see an underlying normative basis for reaching the
conclusion that taxes should be related to taxable capacity without full consideration of
the equilibrium consequences of following such an approach. Our conclusion is similar
to that reached by some economists earlier – that equal marginal sacrifice (minimized
sacrifice – equivalent to optimized social welfare) was the appropriate criterion, not equal
absolute or equal proportional sacrifices.47

Complexity and Observability

From the perspective of optimal contract theory, any costlessly observable
variable correlated with unobserved characteristics or behavior should influence payoffs,
even if it is poorly measured and the correlation is limited. In tax terms, labor income
taxes should depend on all variables correlated with the ability to earn, even those
measured poorly. While tax systems have stupefying complexity, it is not from
incorporating many such variables. Income taxes depend on various measures of
incomes from different sources, adjusted for some expenditure categories. To explore the
extent to which further complications should enter, we consider three examples.

Income taxes are based on earnings without an attempt to measure hours worked
and so average earnings per hour. Minimum wage rules and requirements for paying
higher wages for overtime both require some measurement of hours worked. In this case
the employer and the employee have conflicting interests in the measurement of hours.
This makes enforcement easier than would enforcement of a tax that depended on hours
worked, where neither the employer nor the employee have an interest in seeing higher
income taxes. Nevertheless, an attempt to incorporate a measure of hours worked in the
tax base would plausibly bear considerable correlation with actual hours. For many
workers in large firms or government employment, existing financial records would form
a good basis for estimating hours worked with reasonable accuracy. Moreover a
requirement for self-declaration of hours, subject to some form of random monitoring,

47
“Edgeworth, and later Pigou, held that there was no logical or intuitive choice between the equity
principles of equal absolute and equal proportional sacrifice. Arguing on welfare grounds, they considered
equal marginal sacrifice the only proper rule, not as a matter of equity, but because it met the welfare
objective of least aggregate sacrifice.” Musgrave 1959, page 98.

47
would fit the theoretical category of a correlated, poorly measured, but nevertheless
useful basis for further tax distinctions. And it is not as if earnings were measured
perfectly either.

Thus if it did not recognize factors other than observability, optimal tax theory
would call for basing taxes in part on estimated earnings per hour. We do not think that
using an hours measure in determining taxes would be a good idea, and it is useful to
consider why not. Basing taxation on inaccurately measured variables leaves more scope
for administrative discretion and encourages cynicism about the fairness of the tax
system. As discussed above, both features are likely to add to the difficulty of voluntary
accuracy in reporting and support for the politics of better taxation. The theory of how to
use poorly measured variables would not be intuitive to either legislators or the public,
again making good tax politics more difficult. In sum, basing taxes in part on hours
worked does not seem to be a good idea, although that intuition is not supported by
formal analysis as far as we know. In the exploration of lessons from the literature, we
do not explore the (small) literature on the use of hours in determining taxes.

As another example, this one where accuracy of measurement is not at issue,

consider the finding of Case and Paxson (2006) that there is a correlation between height
and earnings abilities. With different tax structures for adults of different heights
(possibly distinguished by gender) one can then have higher social welfare than without
such multiple tax structures. While it would be complicated for tax authorities to have
multiple tax structures, there is not much complication for the taxpayer who does not get
to choose among tax structures.48 Unlike the example of different tax structures for
different ages discussed below, a set of tax functions based on height is a setting of
consistently different structures for different (fully-grown) individuals rather than
individuals passing through the different tax structures over time.49

Consider a sequence that starts with extensive research documenting that such
differences are real and robust to alternative measurement approaches, explains to the

48
Allowing ex ante choice among tax structures is a source of potential welfare gains we do not explore – it
seems that this added complexity would challenge the ability of many to figure out which tax structure to
pick and could be viewed as inequitable as some workers successfully lowered taxes significantly by a
good choice while others regretted poor choices.
49
This ignores the shrinkage that occurs with aging.

48
public and convinces them that this is the case, and then explains to the public why this is
a useful basis for differences in taxation. Then picture a legislature considering a half
dozen or so different tax structures on this basis. (Think just about earnings, but it might
also be the case that different heights are also correlated with different abilities to invest
and so different possible rates of return and different intertemporal discount factors and
so different tendencies to save.) Presumably the incentive for parents to stunt the growth
of their children would be minimal if they also recognized that the factors that affect
height affect earnings abilities. Should one be troubled by this scenario? Does it violate
some sense of horizontal equity? If height were irrelevant, it would. But once height is
linked to earnings ability, then people of different heights are not identical (as far as the
government’s ability to infer ability is concerned). This is similar to the view that people
with different tastes for work are not identical, even if they have the same budget sets.
Whether the gain in social welfare were small or large would depend on the magnitude of
the correlation and the extent to which different tax structures had an impact on
optimized social welfare.

An appropriate question to ask is how complicated a tax structure a legislature can

use well. Historically legislatures have relied more on their own decision-making in the
realm of taxation (and other topics in economics) than in other areas – legislatures vote
money for bridges, they don’t vote blueprints. Perhaps some addressing of complexity
could be allocated to some expert group.50 And perhaps the public would accept both the
underlying idea and the use of experts. Otherwise, this seems a path unlikely to reach a
good legislated end. Moreover, the public sense of equity may well be violated since
there is only a stochastic relation between height and earnings abilities. We conclude that
this does not seem something worth exploring at present, but the case against it is a
weighting of subjective factors over the factors that can be readily captured in existing
models..

A common assumption in formal models is that taxation is based on costlessly,

perfectly observed variables while all other variables are not observable at any cost. But
this description of observability is not accurate on either side – earnings is costly to

50
As Breyer (1993) has proposed for dealing with health risks.

49
measure and is not perfectly observed and there are other observable variables that could
increase social welfare if used optimally. Thus assertions about observability are not an
adequate guide to the choice of a tax base for direct taxation. Complexity matters as
well, as does a plausible sense of both political economy and public reactions. We
simply refer to variables as taxable and non-taxable, reflecting an ex ante judgment call
reflecting these multiple dimensions of relevance for choosing a tax base, rather than
observable and non-observable.

In contrast to height, age is used by actual tax structures, but very little apart from
retirement-related rules. In the US there are distinctions for children (who can be
dependents and so provide additional deductions) and those over 65, who may receive an
additional deduction. This is not much variation in taxes across ages, nor does it
represent using knowledge of age-varying earnings abilities to explore alternative
structures of marginal tax rates.51 In contrast age does play a large role in the rules for
both public pension systems and tax-favored retirement savings opportunities. Indeed, in
the context of a one-period model of income taxation, and with a focus particularly on
younger workers, Kremer (2001) called for such distinctions on the grounds that the
distributions of income levels are so different across ages that the implied pattern of
optimal tax rates would vary greatly by age. Let us consider a political process if such an
approach is taken. The first step might be to allocate each age to one of a small set of
ages, in order to limit the number of tax schedules.52 Perhaps the set might be under-30,
30-50, 50-65, and over 65. For simplicity, there might be a given set of marginal tax
rates with only the break points varying as a function of age. This doesn’t sound too hard
for a legislature to do.53 And plausibly it could be worked out without undue pressure by

51
A similar argument could be perhaps be made even more strongly with regard to gender, given gender-
differences in life-expectancies and the shapes of life-cycle earnings profiles. As with age, gender is not
used extensively in tax systems although, again, it does play a large role in public pension system rules in
some countries, such as the UK (at present).
52
If there are joint returns for couples based on a couples total incomes, labor income might be taxed on the
basis of the age of the earner while capital income might be taxed as if each received half. Or all taxable
income could be treated as if half were taxed on the basis of the age schedule of each of the couple.
53
This assertion may be undercut by the common practice of adjusting public benefit formulae for the age
at which they start with a linear formula, when multiplicative or more complex formulae seem to make
more sense. Supporting the thought of delegation is the automatic adjustment in Sweden, done on a
roughly actuarial basis, although one with rules for the actuaries set by legislation.

50
the politically better-connected ages. As discussed above, formal models do show
advantages to age-varying earnings taxes.

A different approach looks at current earnings in the context of previous earnings

(and possibly age). This could be done in a variety of ways, including a moving average
over a fixed number of years or basing lifetime taxes on lifetime earnings, with annual
taxes viewed as withholding toward the eventual determination of lifetime taxes.54
Above, we noted how this might serve social welfare maximization. Now we consider
the ability to implement. This certainly is doable, with the government providing historic
information along with tax forms. Indeed, we can consider this as parallel to rules that
determine public pensions. In defined benefit systems pensions are based on the history
of earnings, possibly a full history (as in Sweden) or a long history (as in the US).
Pension systems also commonly make retirement benefits dependent on the age at which
they start (or started). Moreover, that relationship, while commonly involving a single
parameter (annual percentage increase from a delay), involves age-varying parameters in
the US (although not well-chosen). In a wage-indexed system for initial benefits (that are
then price-indexed) as in the US, the benefit formula relating benefits to earnings varies
with date of birth through automatic indexing. Indeed legislated future changes in the US
age for “full benefits” vary with date of birth. In the UK any move to change the state
pension age, by direct legislation or automatic adjustment would result in different
structures for workers in different birth cohorts, although it can also be viewed as
different age-varying rules year-by-year. And Sweden applies different determinants of
both initial benefits (given earnings history) and the growth of benefits from a delayed
start to each birth cohort, by use of automatic adjustments partially following actuarial
principles.

Thus a key question is whether variation in annual tax rates as a function of age is
a bad idea because of complexity or a case of theory being ahead of policy, with
feasibility present, but reform called for. We are inclined to take the former views for
countries thought to have a good legislative process.

54
These would be similar to the approach in Vickrey (1947), who cumulated annual income, not annual
earnings and who considered various lengths of time for the cumulation.

51
 One approach to optimal tax theory has been to take as given a set of allowable
tax tools, chosen to reflect actual (or plausibly potential) use and chosen to enable the
inferences from a model to seem more useful for policy discussions. From this
perspective one can ask about the payoff from additional or altered tax tools, with a need
to recognize in some way the cost of greater complexity. Some analysts have considered
it significant to replace this approach of designated tax tools by assuming that the choice
of tax tools is an endogenous part of the optimization, but subject to observability
constraints. If these constraints were a good approximation of actual constraints, there
might be significant advances from such an alternative assumption (as there have been
significant advances from allowing endogeneity in the structure of the tax given its base,
as opposed to a given function of a limited number of parameters). For example, the
common assumption that earnings are observable but hours worked are not is not a good
approximation of reality. Since there has been little learned directly by substituting the
label unobserved for the label unused in tax calculations, this purported distinction has
not been important and we have continued in the older tradition.

Time Frame and Commitment

In the Arrow-Debreu complete market model, complete (contingent) plans for the
rest of time are made before any economic activity takes place. Moreover, government is
commonly fit into the model by having a complete (contingent) set of government
policies to which the government is committed. There is a considerable literature on the
implications of an inability of government to commit, but primarily in the context of a
unitary (often infinite horizon) government. There is a small literature recognizing tax
changes arising from political changes. In a complete market and complete government
policies setting of time-consistent individual execution of lifetime contingent plans, it
seems appropriate to evaluate social welfare in terms of lifetime utilities since the entire
economic outcome lies in the future. Presumably the relevant measure of lifetime well-
being would reflect the uncertainties about both earnings and capital returns, which
would matter even with complete markets since there are social risks as well as fully
poolable individual risks.

52
 However, there are multiple limitations in the complete markets Arrow-Debreu
model that are important for tax theory. Among these are the presence of overlapping
generations, the incompleteness of markets, and the incompleteness and time
inconsistency of some individual plans and actions. Also present are the incompleteness
of government (contingent) policies, and the stochastic pattern of changes of government,
and, for both reasons, the lack of commitment to policies. The literature contains many
papers analyzing single period models and lifetime models. Single period models do not
do an adequate job of distinguishing the roles of capital and labor incomes. The models
we examine are primarily lifetime utility models. Lifetime models can provide useful
insights, but must not be taken too seriously, and particularly not literally. The
specification of social welfare for tax optimization is complex when individuals are
modeled behaviorally, particularly when the model does not have individuals being time-
consistent. And it is complex when the taxes are likely to be altered by future
governments. Thus it seems appropriate to supplement lifetime considerations assuming
preservation of the tax structure with annual (or several-year) considerations that thereby
recognize a greater likelihood of the tax being in force in the near term.

Returning to our discussion of the Chamley-Judd model, the early papers had two
findings. The first is to have no taxation of capital income, either after a finite date or
asymptotically (that is taxation can be positive indefinitely, but with a steadily shrinking
tax rate) as discussed above. The second is to tax initial wealth as heavily as possible, at
least in the representative agent version. The second finding has had little direct purchase
on the policy recommendations drawing on the literature, although the same perspective,
clearly stated, lies behind arguments in OLG models for switching from income taxation
to consumption taxation as a way particularly to transfer wealth from older cohorts at the
time of tax implementation.55

It is appropriate that these two findings have had different standings. Taxing
initial wealth as much as the available tax tools allow (whether as a wealth tax or a
capital income tax) strains the relevance of the assumption that the government can

55
This basis for a change in taxation is very sensitive to implementation. It works for taxing consumption
directly and for taxing consumption as income less savings provided initial wealth is measured, but may not
work for taxing consumption as income less savings if initial wealth is not measured.

53
commit to a tax policy (and that this taxation of wealth will end). Confiscatory wealth
taxation would adversely affect savings behavior and have serious efficiency costs (even
if the government saves the revenue) because of concern that such taxation will return. A
switch from income to consumption taxation (with limited grandfathering of existing
wealth) could be interpreted as a move against wealth which has limited implications for
future taxation since the set of politically plausible tax policies has not changed very
much – increases in the taxation of consumption are limited by the fact that they fall on
everyone.

These ex cathedra assertions raise the critical question of how to model the link
between tax legislation and expectations about future taxes. The Chamley-Judd finding
of asymptotically vanishing taxation of capital income with full commitment has been
extended to a setting without commitment (Dominguez, 2006, Reis, 2006).56 These
models have a single infinitely lived government maximizing social welfare and
infinitely lived private agents. Government tax policies are chosen in light of the threat
by households to switch to some other equilibrium of the economy and the equilibrium
achieved is assumed to be the best one, which is achieved by the threat point being the
worst one that is sustainable. It is not clear how much to draw policy conclusions from
this game-theoretic modeling. This modeling does not capture the threat of switching to
a different elected government. And it relies on modeling expectations-related behavior
of individual agents that is unlikely to hold generally across different policies. The
literature is interesting since it makes clear the issue of expectations in economic
incentives. It needs to be supplemented with models that recognize the diverse, and often
behavioral, expectation formation that is not part of standard game-theoretic equilibrium.
Thus, even when rejecting the Chamley-Judd zero tax of capital result, it is worth
recognizing the insights from the Chamley-Judd framework although recognizing that
drawing further lessons is difficult.

An alternative way to view ‘commitment’ is in the realm of precedents,

paralleling their role in legal decisions. Assume the government announces a one-time
capital levy. That is a precedent for doing the same again, and so lacks credibility that it

56
These papers assume a single infinite horizon budget constraint. Zero asymptotic taxation of capital is
not optimal when the government faces period-by-period budget constraints.

54
really is one time. Perhaps there are special circumstances, a war or meteorite strike not
likely to recur. Then the precedential cost may be much lower, although there remains
the effect of possibly increasing the perception of a widening of the precedent. Just as
individuals set rules for themselves, with bright-line rules easier to adhere to,57 so too the
government process recognizes that crossing a bright-line rule runs the risk of major
backlash – whether it is losing elections, with possible reversal of the policy, or street
demonstrations, or political backlashes in other realms. Thus one might prefer a small
annual wealth tax rather than a large one-time tax, on the grounds that expectations of
continuing and possibly slow growth of the annual tax has less of a deterrent effect on
savings through perceptions of future policies. Switching from an income tax to a
consumption tax has the effect of taxing existing wealth, with possible increases in the
tax rate as the risk discouraging savings. Again, we would expect less of an impact. This
way of approaching the issue of commitment, or its lack, differs from a game-theoretic
approach in not relying on backwards induction (which is missing in the analyses of some
people) and in recognizing the limited awareness of politics of some and the multiple
motivations affecting voting, which combine to affect the nature of a political-economic
equilibrium that determines taxes.

With governments that can change and individuals with diverse and limited
attention to the future, policy inferences need to be drawn carefully from models of single
governments and lifetime optimizations. The Meade report call for “a certain stability in
taxation in order that persons may be in a position to make reasonably far-sighted plans”
(Page 21.) also suggests seeking tax instruments that are relatively simple and transparent
to aid the formation of appropriate tax expectations by individuals. With sensible voters
and an informative political process, tax instruments should reduce the possibilities for
future governments to use obfuscation as a guise to introduce potentially unpopular
reforms.

Transition

57
It appears easier to comply with a no cookies or no cigarettes rule than trying to allow oneself only a few.

55
 Transition issues arise in two ways in a discussion of the right tax base.
One is that analysis of the right tax base needs to recognize that there will be future tax
changes, and those changes will involve transition issues. This is discussed here. Second
are the one-time transition issues if the contents of this essay (or some other) were to be
accepted as the basis for future taxation. This issue is not discussed here – not because it
is unimportant but from limitations of time and space. One difference between the two is
that when considering future changes, but not when considering the initial adoption of a
new policy framework, the expectations are endogenous to the policy framework being
created. Both circumstances call for giving some degree of respect to legitimate
expectations. For discussion of the issue of an initial change, see Auerbach (2006) who
presents many issues and highlights the importance of transition by contrasting
simulations that have the same long-run tax incentive properties but very different
transition impacts. Whether ending the taxation of capital income raises or lowers social
welfare varies with the transition impact in some simulations.

In the ongoing process of the adaptation of tax policies to economic and

demographic developments as well as to changing normative perceptions and political
balance, a set of rules/guidelines for transition issues is important both economically and
politically. From this perspective we can appreciate the Meade Report’s concern for
flexibility and stability: “A good tax structure must be flexible for two rather distinct
purposes. … there must be recognition of the need to be able to adjust total tax burdens
reasonably rapidly and frequently in the interests of demand management. … In a healthy
democratic society there must be broad political consensus – or at least willingness to
compromise – over certain basic matters; but there must at the same time be the
possibility of changes of emphasis in economic policy as one government succeeds
another. … But at the same time there is a clear need for a certain stability in taxation in
order that persons may be in a position to make reasonably far-sighted plans.
Fundamental uncertainty breeds lack of confidence and is a serious impediment to
production and prosperity.” (Page 21.)

Beyond any possibility of short-run demand management, there are

changes in long-run fiscal needs that are likely to occur from trend developments in
economics and demography as well as the spreading over the future of short run changes

56
in fiscal needs (e. g., a war). A research program that would address the need for both
adjustment and stability would seek a tax structure that would have enough political
acceptability to relegate tax changes primarily to parameter changes in a class of
parameters anticipated to adjust to circumstances.58 The design should recognize that tax
expectations are endogenous to the policy framework being created. Such modeling
would examine a standard optimization of social welfare in the context of incomplete
markets and policies, striking a balance between the different effects of changing
policies.59 Such modeling would also need to examine how the structure of taxation
influences the political process, noting variations in later government policies to different
structures. Indeed, links between the form of public pension design and anticipated
future legislation has been part of the US debate between defined benefit and defined
contribution systems.

In modeling changing governments, it is not clear how to set up a suitable social

welfare function, particularly once one recognizes the endogeneity of the election
process. Beyond standard social welfare analysis in terms of lifetime expected utilities,
there may be a further normative concern of limiting the deviations from appropriately
held expectations about policies.60 The latter requires an equilibrium analysis since the
expectations that are held and expectations that are appropriate to respect to some degree
both depend on the policy process. If it is anticipated that tax rates will adjust within
some guidelines, then some subjective probabilities of such adjustments would be a
necessary part of expectations to be respected. The presence or absence of an ongoing
political discussion would affect the appropriate degree of respect for actions based on
expectations. And one would need an evaluation of the political process that might have
very different normative treatment between “loopholes” that come from less satisfactory
aspects of politics and “appropriate” political outcomes. That is, the degree of respect to
58
Currently discretionary fiscal policy, while pursued by governments, is not in high favor among
academic economists (Auerbach, 2002). But built-in stabilizers, while not getting much active attention,
remain in positive evaluation (Auerbach and Feenberg, 2000). It is odd that there was not discussion of
built-in stabilizers in the Meade Report.
59
Such analysis might parallel for an economy the analysis for individuals in Amador, Werning and
Angeletos, 2005.
60
This might parallel the same issue in the legal analyses of contracts, where courts attempt to interpret
contracts in the light of the expectations of the contract parties. The endogeneity of legitimate expectations
to court processes that try to decide in terms of the expectations of the parties has not always received
adequate attention.

57
past taxes and the expectation of their continuation need to recognize a widely held view
that the tax structure is not satisfactory and ought to be reformed (a view that underlies
the commissioning of this report).

An individual who made commitments (in education or investment, for example)

based on a legitimate expectation of a continuation of existing taxes and is greatly
harmed by a change in tax structure would have a complaint that seems to us to be
legitimate, calling for some response. Hence the importance of a policy process that
incorporates awareness of some future changes. The complaint based on the commitment
is likely to have more salience and may have more legitimacy if the tax change matters
for some people and not for others. Indeed, an explicit commitment by an individual is
likely to increase the salience of the tax change and so pressure for a response. Analysis
of restrictions on the political process is likely to find a role for both delayed
implementation (although the announcement of delayed implementation can have
inefficient effects) and grandfathering as part of protecting expectations and preserving
incentives. Use of these tools were raised in Feldstein (1976b) as part of adjusting for the
implications of the commitments that some people will have made based on the earlier
tax structure (or more precisely on the expectation of continuation of the earlier tax
structure).

Part III. Concluding remarks

The long-standing debate on the best base for nonlinear (progressive)

annual individual taxation has been between total income and total consumption
expenditures (with recognition of special treatment for some incomes and/or some
expenditures). A focus on taxation in a single year highlights the advantage of total
income. A focus on longer periods, up to lifetimes, highlights the advantages of
consumption. A focus beyond lifetimes that incorporates bequests raises further issues
about individuals born at different times. Exclusive focus on any one of these three time
frames seems incomplete in a world with liquidity constraints, time-inconsistent
individuals, accidental bequests, incomplete government policies, and changes in

58
government. These realities, along with the complexity and uncertainty in the evolution
of individual earnings, returns to capital, and individual consumption preferences make it
hard to think about policy design and inappropriate to rely too heavily on any single
model’s implications.

These realities also lead us to the conclusion that the long-standing debate is
misdirected. A more informative debate may be about the relative taxation of different
sources of income and, relatedly, the implications for progressive taxation of different
uses of income, with the focus here on savings, but plausibly also on medical expenses,
education expenses, housing expenses, taxation by other governments. We have
proceeded as in the quote from Alfred Marshall at the start of this essay, “it [is] necessary
for man with his limited powers to go step by step; breaking up a complex question,
studying one bit at a time, and at last combining his partial solutions into a more or less
complete solution of the whole riddle.” (Marshall, 1948, page 366.) Thus we have seen
the implications of a wide variety of individual analyses and asked about policy
inferences that seemed appropriate to draw. We do not think we have “a more or less
complete solution of the whole riddle.” But policy formation, and so policy
recommendations, can not wait for a complete solution.

While there are policy recommendations that seem sensible to make

without knowing much about an individual country’s political and economic
circumstances, these tend to be of the negative sort – recognizing policies that are likely
to work badly in a wide swath of circumstances. For positive recommendations one
needs a more detailed context. And while this essay, like the Meade Report, is supposed
to focus on the UK, we frame our conclusions in a set of if-then propositions in terms of
policy debates that might occur at some time, while implicitly accepting the underlying
economics and administrative capabilities of an advanced country like the UK.

This is limiting since the constraints from the political debate can narrow the
choices, sometimes in ways that make recommendations easier to derive and justify. It
does seem good for democratic policy discussion to include analyses of better policies
given perceived political constraints and analyses of even better policies if less binding
political constraints are assumed. Focusing on a better way to collect revenue rather than

59
the advantages of more or less revenue may fit a particular political context well.
Economists viewing actual tax structures often favor base widening and rate reduction in
a revenue (and sometimes distributionally) neutral way if that is the prime available route
to more efficient taxation. Indeed, Peter Birch Sørensen (2001) has argued that basis as
part of the appeal of the Nordic dual income tax – wider and more uniform taxation of
different sources of capital income in return for lower tax rates. We have no quarrel with
this approach in such a circumstance, although one still needs to consider the particulars
of what should be treated uniformly and what is the best available approximation to
uniform treatment. A tradeoff of lower taxes for a wider base does not appear to require
linearity of taxation. But our focus has been on the relationship between taxation of
capital income and of labor income, not on the details of the taxation of either.

Any real policy recommendation must address issues of transition. These may
include particular concerns about individual equity and may include recognition of the
effects of taxing existing capital. The former is lost when equity analyses look only at
lifetimes of cohorts living under a new system. And the latter is lost with consideration
of steady-state properties that are derived as the best steady state rather than the steady
state that arises from a full optimization. If one shifts relative taxation from capital
income to labor income or from income taxation to consumption taxation, there will be
important differences in the impacts (on average) on people of different ages at the time
of implementation and different levels of accumulated wealth within an age cohort.
Whether and how these are addressed as part of a reform can matter greatly. Without
special rules, a shift of taxation from capital income to labor income benefits older
cohorts at the expense of younger ones. And a shift from income taxation to
consumption taxation costs older cohorts to the benefit of younger ones. Moreover, the
pattern of impacts within a cohort depends on how the transition is done. If a tax
deduction for savings is introduced, the pattern of impacts depends on whether the
deduction is really measuring savings, or is allowing the transfer of existing wealth
among assets to be treated as savings. Depending on the goals and the remaining features
that an inability to measure and use existing wealth as part of tax calculations may or may
not be important. Combining transition issues along with evaluation of ongoing
implications of tax structures would be too large a question for this essay. So we discuss

60
the lessons from the optimal tax literature for the ongoing effects of taxation, recognizing
that this is only part of the story.

Similarly, implicit in our focus on the tax base, separate from tax rates, is an
assumption that tax rates are being optimized for given tax bases, thereby ignoring the
political linkage that may well be present between tax base and tax rates. It is incomplete
to say that suitable choice of tax rates can make different bases of comparable overall
progressivity if that suitable choice will not happen. Recognition of the link between the
form of tax institutions and the perceptions and salience that then influence policy
making is important. This is part of the debate that has occurred between defined benefit
and defined contribution national pension systems and would be appropriately part of a
fuller debate on tax base than we will engage in. Similarly, political links (or
‘hypothecation’) between tax bases and government expenditure policies matter. This is
clearest with dedicated taxes for particular programs but they plausibly matter for general
taxation as well. Again, this is missing from our discussion.

As noted at the start, the Meade Commission Report recommends a three-part

structure made up “of a new Beveridge scheme, … of a progressive expenditure tax
regime, … and of a system of progressive taxation on wealth with some discrimination
against inherited wealth.” We have not considered issues particularly important in a
focus on those with very low or no earnings and limited wealth. Also, we have not
explored models that might shed light on the relative advantages of annual taxation of
wealth relative to taxation of capital income, as the models we have examined have
mostly been restricted to a single safe asset, available on the same terms to all, leaving
the two sources of taxation the same – neither uncertainty nor realization of income nor
measurement of the value of illiquid assets has been considered.

The Meade Report discussed measuring the ability to pay taxes as part of tax
design. It concluded that: “on examination ‘taxable capacity’ always turns out to be very
difficult to define and to be a matter on which opinions will differ rather widely.” (Page
14.) A moderate number of papers since the report have explored horizontal equity for
tax purposes. We see no reason to reach a different conclusion from that in the Report –

61
indeed, we have gone further in dismissing such a measure from a central place in tax
design.

In considering the Meade Report recommendations in light of thirty years of

additional research, experience, and economic development, we explore two questions
that shed some light on the Meade Report recommendations –

- If there is an annual earnings tax, how should capital income be taxed?

- If there is an annual earnings tax, should there be a deduction for net

savings, resulting in a tax on consumption?

In addition, we explore an issue not addressed in the Meade Report, the potential
advantages, despite increased complexity, in age-varying tax rates. Each of these three
issues has both a design dimension and a transition dimension, but we consider only the
former.

Taxation of capital income with an annual earnings tax

Support by economists and tax lawyers for exempting capital income from
direct taxation has been influenced by the well-known Atkinson-Stiglitz and Chamley-
Judd analyses.61 However, we conclude that the policy relevance of the sharp finding of
the optimality of no taxation of capital income is thoroughly undercut by the implications
of large uncertainty about future earnings and the growing disparity in earnings as a
cohort ages. Adding such uncertainty and disparity to the frameworks employed by
Atkinson-Stiglitz or Chamley-Judd results in a finding that taxation of capital income or
of wealth is indeed part of optimal taxation.

In addition, in light of the widely varying individual savings rates in the economy,
there is a natural presumption that during working years there is a positive correlation
between the tendency to save and earnings potential (although the empirical underpinning
is not so clear). This is another reason for taxing capital income as a means of more
efficiently taxing those with higher earnings potentials. A further case comes from the

61
Whether this is seen as supporting taxation of expenditures or alternatively of earnings depends on
intertemporal equivalence or lack thereof of revenue and the modeling of returns to capital.

62
difficulties in distinguishing between labor and capital incomes, which gives an
advantage to reducing the difference in taxes between them. The empirical evidence that
capital raises skilled labor earnings more than unskilled labor earnings argues for taxation
of capital income in the usual model with infinite horizon savers, but not in the OLG
model, where the government can increase capital to accomplish this effect. While we
have not explored the literature incorporating human capital investment into tax
considerations, a progressive earnings tax (particularly one that is not age-varying) and
the presumption that human capital investment steepens the age-earnings trajectory can
call for some taxation of capital income to get closer to even treatment of these two forms
of investment. While we do not know of any modeling.

Taxation of capital income needs to consider two issues. Should capital income
be taxed more or less heavily than labor income? With a thought process that starts with
the conditions for zero taxation and then adds some taxation for elements not in the
models that imply zero taxation, there is the danger of “anchoring” towards zero resulting
in a conclusion that capital income taxation should be lighter, without a good basis for
reaching that conclusion. In contrast, if one started with the recognition that with a tax
system based on comprehensive nominal income, marginal taxes on (nondeferred) real
capital income are larger than those on real labor income, one might be asking how much
of this heavier taxation should remain. Again, there does not seem a good basis for the
conclusion that the optimum lies with equal or heavier taxation. There is probably no
substitute for extensive calculation using calibrated models, with models that incorporate
the elements thought to be most important in determining relative taxation. Some
existing calculations show heavier taxation while others show lighter taxation. We did
not attempt to evaluate the relevance of different calculations, but point to the need for
lots more.

A second issue is the appropriate relationship between marginal taxation of

capital income and marginal taxation of labor income. The Nordic dual tax has linear
taxation of capital income and so a greater disparity between marginal rates for those
with higher labor incomes.62 Apart from the point that trying to discourage conversion of

62
On the Nordic dual tax, see Sørensen, 2005.

63
labor income into capital income seems to call for marginal tax rates on the two types of
income that relate positively to each other, it is not clear without extensive calibrated
calculations what the relationship should be. In the US, recent lower tax rates on
dividends do relate that tax rate to the rate on labor income.63 The old system that had
inclusion of one-half of capital gains in taxable income (for those with lower tax
brackets) also had a clear relationship. To explore the normative properties of such an
approach, one would again need extensive calculations. We think such calculations are
called for and do not see a way to draw a firm conclusion from the evidence we have
examined.

A deduction for savings with an annual earnings tax

One way to have a consumption tax base is to deduct from earnings the net
increase in savings – purchases of capital assets less the sum of capital income and
realized capital gains. In countries such as the United Kingdom that already have EET
tax-favored retirement savings accounts, this corresponds to removing limits on deposits
in such accounts along with removing limits on withdrawals. Thus, compared with an
accrual-based income tax (or an approximation from taxing realized capital gains to
adjust for deferral) a consumption tax gives the advantage of deferral on all savings for
future consumption. As Judd (1999) has pointed out this approach does not get
incentives right for human capital.

This framing of the issue is different from that in Gentry and Hubbard, 1997.
They consider consumption taxation implemented by a wage tax combined with a
business cash flow tax. Although they purportedly are addressing distributional
implications, their focus is on evaluating the difference in taxation from the perspective
of a firm’s investment decisions, as opposed to a household’s life-cycle labour supply
savings choices, and as a consequence they focus on the marginal value of immediate
depreciation of investment to a firm, which they value using the safe rate of interest,
supporting the view that consumption taxation exempts the safe rate of interest but not

63
The tax rate is 5% for those whose top tax bracket is 10% or 15% and 15% for those with higher top tax
brackets.

64
the return to bearing risk or pure rents. Modeling individual choice as a base for
examining the impact on the distribution of utilities of giving the deferral advantage is
more complicated. While stocks and bonds have the same marginal value with portfolio
optimization, the impact of deferral on the inframarginal gains from the availability of
stocks is relevant for distributional analysis. As a quick example of this issue for given
wealth and Cobb-Douglas the higher the distribution of risky returns the greater the gain
from deferral for a given portfolio mix. Since the optimized portfolios may well be
different, a full analysis is more complicated. But this seems the appropriate way to
approach the distributional impact.

It is worth noting that there are significant differences between exempting capital
income from taxation and a consumption tax base. In a model with a single safe rate of
interest, the two are the same apart from differences needed in transition rules to match
them. However, both different rates of return for different investors and uncertain rates
of return make the two approaches quite different.

Age-varying taxes

Public pension systems have age-varying rules for eligibility for claiming
benefits, for determination of the size of benefits, and for the taxation of earnings. They
also have rules that have a strong reliance on individual histories over a long period in the
determination of benefits.64 Income taxes make little use of such structures (apart from
what is inherent in measuring capital gains). An implicit exception, similar to pension
calculations, comes with tax-favored retirement savings. This incorporates explicit tax
rules based on age when withdrawing funds as well as different implicit degrees of tax-
favoring depending on the age at which funds are put into an account.

The mechanism design approach to optimal taxes uses both age-varying rules and
full history dependence. Traditional optimal tax analyses have contrasted optimal rules
for linear taxation of capital income with and without age-varying earnings taxes. In
considering whether it is worth the administrative complexity and the added political

64
Funded defined contribution and notional defined contribution systems do this in terms of the summary
statistic of balance in an account.

65
process to extend tax structures to include such features, their presence in existing
national pension rules suggests feasibility. And analyses of optimal pension systems
suggests value. It is important to recognize the difference between arguments coming
from differences in the distribution of circumstances across different ages and those
coming from individual forward-looking calculations when making decisions. The
former may have more robustness in policy implications than the latter across the very
real diversity in individual decision-making.

Because age-varying taxes can address both of these issues, we think it is useful
for governments to contemplate introducing them in some form and for analysts to
explore them in more detail than has happened so far in the literature that has tax
structures similar to actual ones. For discussion purposes, we consider a tax structure that
distinguishes between labor and capital incomes with related marginal tax rates and
revisit an example of such an approach. The first step might be to allocate each
individual to one of a small set of age groupings to limit the number of tax schedules.65
Perhaps the set might be under-30, 30-50, 50-65, and over 65. There might be a given set
of marginal tax rates for all ages with different break points between the marginal rates
for each group. Analysis of the break points would reflect the distribution of earnings
possibilities by age and the intertemporal incentives inherent in facing different break
points over time. The latter might reflect uncertainties about future earnings, human
capital accumulation, and liquidity constraints. This doesn’t sound too hard to model nor
too hard for a legislature to legislate. And plausibly this could be legislated without
undue pressure by the politically better-connected ages. Obviously, any optimal tax
analysis will find a higher-valued optimum from having more policy tools that are not
superfluous. The literature suggests that the gains from age-varying taxes may not be
trivial and detailed analysis could explore how substantial the gains might be.

The optimal tax literature recognizes that accumulated assets during working life
affect labor supply, particularly retirement decisions. This implies a basis for
distinguishing the taxation of capital income (or wealth) depending on the fraction of the

65
If there are joint returns for couples based on a couple’s incomes, labor income might be taxed on the
basis of the age of the earner while capital income might be taxed as if each received half. Or all taxable
income could be treated as if half were taxed on the basis of the age schedule of each of the couple.

66
age cohort in the labor force (or close to the decision margin of still being in the labor
force). This distinction loses bite for those adjusting their savings for strong bequest
motives or for utility that is based on wealth ownership per se. Perhaps tax-favored
retirement savings with suitable limits is an adequate way of addressing this issue. Or
perhaps the typical age-earnings and age-retirement savings profiles together with
liquidity constraints suggests that there is scope for further distinctions, as precautionary
balances are built up.66

In contrast with the concern of too much precautionary savings limiting the ability
to tax, mandatory pension systems are based on a presumption of too little retirement
savings by part of the working population. Of course, there is no contradiction in
assuming that some save too much for social purposes, while others save too little. What
is called for is an integrated analysis considering both taxation of savings and mandating
of participation in a national pension system.

The literature since the Meade Report has gone well beyond earlier literature in
considering more complex tax structures, particularly age-varying rates. Maybe it is time
for policy makers and tightly policy-relevant analyses to take this up.

Concluding remarks

The Meade Report wanted to tax both consumption and wealth annually. We
share the view that capital income (or wealth) should be part of the ideal tax base,
whether the remaining part of the ideal base is labor income or consumption. We do not
find any support in optimal tax considerations for the argument that annual capital
income should be taxed exactly as annual labor income is taxed – a tax base of Haig-
Simons income. We have also argued for the advantages of explicit and significant
variation of taxation with age. We have noted repeatedly issues that warrant further
research. A call for the obvious need for further research is not meant to undercut the
relevance of research developments to date for improving policy debates, and possibly
policy.

66
The connection to saving for home purchase could be accommodated inside or outside dedicated tax-
favored savings vehicles.

67
 References

Abel, Andrew B. 2001. ''The Effects of Investing Social Security Funds, in the Stock Market
when Fixed Costs Prevent Some Households from Holding Stocks.'' American Economic Review.
91 (1) March: 128-48.

Aiyagari, S. Rao. 1995. “Optimal Capital Income Taxation with Incomplete Markets,
Borrowing Constraints, and Constant Discounting.” Journal of Political Economy,
103(6): 1158-75.

Albanesi, Stefania and Christopher Sleet. 2006. “Dynamic Optimal Taxation with Private
Information.” Review of Economic Studies, 73(1): 1-30.

Amador, Manuel, Werning, Iván and, George-Marios Angeletos. 2005. “Commitment

Versus Flexibility.” Econometrica, 74(2): 365-96.

Atkeson, Andrew, V. V. Chari, and Patrick J. Kehoe. 1999. “Taxing Capital Income: A
Bad Idea.” Federal Reserve Bank of Minneapolis Quarterly Review, 23(3): 3-17.

Atkinson, Anthony B. 1970. “On the Measurement of Inequality.” The Journal of

Economic Theory, 2(3): 244-63.

Atkinson, Anthony B. 1980. “Horizontal Equity and the Distribution of the Tax Burden.”
In The Economics of Taxation, ed. Henry J. Aaron and Michael J. Boskin, 3-18.
Washington, D.C.: Brookings Institution.

Atkinson, Anthony B. and Joseph E. Stiglitz. 1976. “The Design of Tax Structure: Direct
Versus Indirect Taxation.” Journal of Public Economics, 6(1-2): 55-75.

Atkinson, Anthony B. and Joseph E. Stiglitz. 1980. Lectures on Public Economics.

London: McGraw Hill.

Auerbach, Alan. 1991. “Retrospective Capital Gains Taxation.” American Economic

Review, 81(1): 167-78.

Auerbach, Alan. 2002. “Is There a Role for Discretionary Fiscal Policy?” In Rethinking
Stabilization Policy: Proceedings of a Symposium Sponsored by the Federal Reserve
Bank of Kansas City, 109-50. Kansas City: Federal Reserve Bank of Kansas City.

Auerbach, Alan. 2006. “Choice between Income and Consumption Taxes: A Primer."
NBER Working Papers No. 12307.

Auerbach, Alan, and Daniel Feenberg. 2000. "The Significance of Federal Taxes as
Automatic Stabilizers." Journal of Economic Perspectives, 14(3): 37-56.

68
Bernheim, B. Douglas and Kyle Bagwell. 1988. "Is Everything Neutral?" Journal of
Political Economy, 96(2): 308-38.

Bernheim, B. Douglas and Antonio Rangel. 2005. “Behavioral Public Economics:

Welfare and Policy Analysis with Non-Standard Decision-Makers.” in Behavioral
Economics and Its Applications, ed. Peter Diamond and Hannu Vartiainen, 7-84.
Forthcoming in 2007. NJ: Princeton University Press.

Boadway, Robin, Maurice Marchand, and Pierre Pestieau. 2000. “Redistribution with
Unobservable Bequests: A Case for Taxing Capital Income.” Scandinavian Journal of
Economics, 102(2): 253-67.

Bovenberg, A. Lans, and Bas Jacobs. 2005. “Redistribution and Education Subsidies are
Siamese Twins.” Journal of Public Economics, 89(11-12): 2005-2036.

Breyer, Stephen. 1993. Breaking the Vicious Circle: Toward Effective Risk Regulation.
Cambridge, MA: Harvard University Press.

Buchanan, James M., and Richard A. Musgrave. 1999. Public Finance and Public
Choice. Cambridge, MA: MIT Press.

Carroll, Christopher D. 2000. “Why Do the Rich Save So Much?” In Does Atlas Shrug?
The Economic Consequences of Taxing the Rich, ed. Joel B. Slemrod, 463-485.
Cambridge, MA: Harvard University Press.

Case, Anne and Christina Paxson. 2006. "Stature and Status: Height, Ability, and Labor
Market Outcomes." NBER Working Papers No. 12466.

Chamley, Christophe. 1986. “Optimal Taxation of Capital Income in General

Equilibrium with Infinite Lives.” Econometrica, 54(3): 607-22.

Chamley, Christophe. 2001. “Capital Income Taxation, Wealth Distribution and

Borrowing Constraints.” Journal of Public Economics, 79, 55-69.

Choné, P. and G. Laroque, 2001, Optimal incentives for labor force participation, INSEE
working paper No 2001-25

Choné, P. and G. Laroque, 2006, Should low skilled work be subsidized?, INSEE
working paper No 2006-08

Corlett, W. J. and D. C. Hague. 1953-54. “Complementarity and the Excess Burden of

Taxation.” Review of Economic Studies, 21(1): 21-30.

Correia, I. 1996. Should capital income be taxes in a steady state? Journal of Public
Economics, 60, 147-51.

69
Cremer, Helmuth, and Firouz Gahvari. 1995. “Uncertainty, Optimal Taxation and the
Direct Versus Indirect Tax Controversy.” Economic Journal, 105, 1165-1179.

Cremer, Helmuth, and Pierre Pestieau. 2003. “Wealth Transfer Taxation: A Survey.”
CESifo working paper 1061.

Cremer, Helmuth, Pierre Pestieau, and Jean-Charles Rochet. 2001. "Direct Versus
Indirect Taxation: The Design of the Tax Structure Revisited.” International Economic
Review, 42(3): 781-99.

Cremer, Helmuth, Pierre Pestieau, and Jean-Charles Rochet. 2003. "Capital Income
Taxation when Inherited Wealth is Not Observable.” Journal of Public Economics,
87(11): 2475-90.

Da Costa, Carlos E. 2003. “Optimal Taxation with Gradual Learning of Types.”

Fundaçao Getulio Vargas paper 499.

Deaton, Angus. 1979. “Optimally Uniform Commodity Taxes.” Economic Letters, 2(4):
357-61.

Deaton, Angus, and Christina Paxson. 1994. “Intertemporal Choice and Inequality.”
Journal of Political Economy, 102(3): 437-67.

Devereux, M. and R. Griffith, ‘Taxes and the location of production: Evidence from a
panel of US multinationals, Journal of Public Economics, 1998

Diamond, Peter. 1965. "On the Cost of Tax Exempt Bonds." Journal of Political
Economy, 73(4): 399-403.

Diamond, Peter. 1967. "Cardinal Welfare, Individualistic Ethics, and Interpersonal

Comparison of Utility: Comment." Journal of Political Economy, 75(5): 765-66.

Diamond, Peter. 1980. "Income Taxation with Fixed Hours of Work," Journal of Public
Economics 13, February, 101-110.

Diamond, Peter. 1994. "Integrating Allocation and Stabilization Budgets." in Modern

Public Finance, ed. John Quigley and Eugene Smolensky, Cambridge: Harvard
University Press.

Diamond, Peter. 2006. “Optimal Tax Treatment of Private Contributions for Public
Goods with and without Warm Glow Preferences.” Journal of Public Economics, 90(4-
5): 897-919.

Diamond, Peter and John Geanakoplos. 2003. “Social Security Investment in Equities,”
American Economic Review, 93 (4) September: 1047-1074.

70
Dominguez, Begona. 2006. “Public Debt and Optimal Taxes Without Commitment.”
Forthcoming in Journal of Economic Theory.

Edgeworth, Francis Ysidro. 1925. Papers Relating to Political Economy. Vol. II. London:
Macmillan and Company.

Erosa, Andres, and Martin Gervais. 2002. "Optimal Taxation in Life-Cycle Economies."
Journal of Economic Theory, 105(2): 338-69.

Farhi, Emmanuel and Ivan Werning. 2006. “Inequality and Social Discounting.” Working
paper.

Feldstein, Martin. 1976a. “On the Theory of Tax Reform.” Journal of Public Economics,
6(1): 77-104.

Feldstein, Martin. 1976b. “Compensation in Tax Reform.” National Tax Journal, 29(2):
123-30.

Gaube, Thomas. 2005. “Income Taxation, Endogenous Factor Prices and Production
Efficiency.” Scandinavian Journal of Economics, 107(2): 335-52.

Gentry, William M. and R. Glenn Hubbard. 1997. “Distributional Implications of

Introducing a Broad-Based Consumption Tax.” In Tax Policy and the Economy, vol. 11,
ed. James M. Poterba, 1-48. Cambridge, MA: The MIT Press.

Gollier, Christian. 2005. “Intergenerational Risk Sharing and Risk Taking of a Pension
Fund.” No. 391, IDEI Working Papers from Institut d'Économie Industrielle, Toulouse.

Golosov, Mikhail, Narayana Kocherlakota and Aleh Tsyvinski. 2003. "Optimal Indirect
and Capital Taxation." Review of Economic Studies, 70(3): 569-87.

Golosov, Mikhail and Aleh Tsyvinski. 2006. “Designing Optimal Disability Insurance: A
Case for Asset Testing.” Journal of Political Economy, 114(2): 257-69.

Golosov, Mikhail, Aleh Tsyvinski, and Iván Werning. Forthcoming. "New Dynamic
Public Finance: A User's Guide." NBER Macroeconomic Annual, 2006.

Gordon, Roger H. 1985. Taxation of Corporate Capital Income: Tax Revenues vs. Tax
Distortions. Quarterly Journal of Economics, Feb. 1-27.

Gordon, Roger H., and Jeffrey K. MacKie-Mason. 1995. “Why is There Corporate
Taxation in a Small Open Economy?” In The Effects of Taxation on Multinational
Corporations, ed. Martin Feldstein, James R. Hines, Jr., and R. Glenn Hubbard.
Chicago: University of Chicago Press.

71
Greenwood, Jeremy and Guillaume Vandenbroucke. 2005. “Hours Worked: Long-Run
Trends.” NBER Working Papers No. 11629.

Haig, Robert M. 1921. The Federal Income Tax. New York: Columbia University Press.

Hall, Robert E. 1968. “Consumption Taxes Versus Income Taxes: Implications for
Economic Growth.” In Proceedings of the 61st Annual Conference on Taxation, 125-45.
San Francisco: National Tax Association.

Hamilton, 1987,

Hausman, Jerry A. 1979. “Individual Discount Rates and the Purchase and Utilization of
Energy-Using Durables.” The Bell Journal of Economics, 10(1): 33-54.

Helliwell, John. 1969. “The Taxation of Capital Gains." Canadian Journal of Economics,
2(2): 314-18.

Hurd, Michael D. 1987. “Savings of the Elderly and Desired Bequests." American
Economic Review, 77(3): 298-312.

Jones, Manuelli and Rossi, 1997)

Judd, Kenneth L. 1985. "Redistributive Taxation in a Simple Perfect Foresight Model."

Journal of Public Economics, 28(1): 59-83.

Judd, Kenneth L. 1999. “Optimal taxation and spending in general competitive growth
models,” Journal of Public Economics, 71: 1-26.

Judd, Kenneth L., and Che-Lin Su. 2005. “Optimal Income Taxation with
Multidimensional Taxpayer Types.” Working Paper.

Kaplow, Louis. 1989. “Horizontal Equity - Measures in Search of a Principle.” National

Tax Journal, 42(2): 139-54.

Kaplow, Louis. 1994. Taxation and Risk Taking: A General Equilibrium Perspective.
National Tax Journal. 47, 879-798.

Kaplow, Louis. 1995. “A Note on Subsidizing Gifts.” Journal of Public Economics,

58(3): 469-78.

Kaplow, Louis. 2006. "On the Undesirability of Commodity Taxation Even When
Income Taxation Is Not Optimal." Journal of Public Economics, 90(6-7): 1235-50.

Kleven, Henrik Jacobsen, Wolfram F. Richter, and Peter Birch Sørenson. 2000. “Optimal
Taxation with Household Production.” Oxford Economic Papers, 52(3): 584-94.

72
Kocherlakota, Narayana. 2005. "Zero Expected Wealth Taxes: A Mirrlees Approach to
Dynamic Optimal Taxation." Econometrica, 73(5): 1587-621.

Kremer, Michael. 2001. “Should Taxes be Independent of Age?”

[Link]

Krusell, P., LE Ohanian, JV Rios-Rull, GL Violante. 2000. Capital-Skill

Complementarity and Inequality: A Macroeconomic Analysis. Econometrica.

Laroque, Guy R. 2005. "Indirect Taxation is Superfluous under Separability and Taste
Homogeneity: A Simple Proof." Economics Letters, 87(1): 141-4.

Marshall, Alfred. 1948. Principles of Economics, 8th ed. New York: Macmillan.

Meade, James Edward. 1978. The Structure and Reform of Direct Taxation, Report of a
Committee chaired by Professor J. E. Meade. London: George Allen & Unwin.

Mirrlees, James. 1995. "Private risk and public action: The economics of the welfare
state," European Economic Review, 39(3-4), 383-397.

Musgrave, Robert A.1959. The Theory of Public Finance: A Study in Political Economy.
New York: Mcgraw-Hill.

Myles, Gareth D. 1995. Public Economics. UK: Cambridge University Press.

Nielsen, Søren Bo and Peter Birch Sørensen. 1997. “On the Optimality of the Nordic
System of Dual Income Taxation.” Journal of Public Economics, 63(3): 311-29.

Naito, H. 1999. “Re-examination of Uniform Commodity Taxes under a Non-Linear

Income Tax System and its Implications for Production Efficiency.” /Journal of Public
Economics/, 71(2): 165-88.

Pigou, Arthur C. 1928. A Study in Public Finance. London: Macmillan and Company.

Piketty, Thomas and Emmanuel Saez. 2003. “Income Inequality in the United States,
1913-1998.” Quarterly Journal of Economics, 118(1): 1-39.

Piketty, Thomas and Emmanuel Saez. 2006. “The Evolution of Top Incomes: A
Historical and International Perspective.” American Economic Review Papers and
Proceedings, 96(2): 200-205.

Prescott, Edward C. 2004. “Why do Americans Work So Much More Than Europeans?”
Federal Reserve Bank of Minneapolis Quarterly Review, 28(1): 2-13.

Reis, Catarina. 2006. “Taxation Without Commitment.” Working Paper.

73
Rosen, Harvey S. 1978. “An Approach to the Study Of Income, Utility, and Horizontal
Equity.” Quarterly Journal of Economics, 92(2): 307-22.

Saez, Emmanuel. 2001. "Using Elasticities to Derive Optimal Income Tax Rates."
Review of Economic Studies, 68(1): 205-29.

Saez, Emmanuel. 2002a. "Optimal Progressive Capital Income Taxes in the Infinite
Horizon Model." NBER Working Papers No. 9046.

Saez, Emmanuel. 2002b. "The Desirability of Commodity Taxation under Non-Linear

Income Taxation and Heterogeneous Tastes." Journal of Public Economics, 83(2): 217-
30.

Saez, Emmanuel. 2002c. "Optimal Income Transfer Programs: Intensive Versus

Extensive Labor Supply Responses." Quarterly Journal of Economics, 117(3): 1039-73.

Salanie, Bernard. 2003. The Economics of Taxation. Cambridge, MA: The MIT Press.

Simons, Henry. 1938. Personal Income Taxation. Chicago: University of Chicago Press.

Smith, Adam. 1937. Wealth of Nations. New York: The Modern Library.

Sørensen, Peter Birch. 2001. “The Nordic Dual Income Tax – In or Out.” Invited speech
delivered at the meeting of Working Party 2 on Fiscal Affairs, OECD 14 June 2001.

Sørensen, Peter Birch. 2005. “Dual Income Taxation: Why and How?” CESifo Working
Paper 1551.

Stiglitz, Joseph. 1982a. “Self-Selection and Pareto Efficient Taxation.” Journal of Public
Economics, 17(2): 213-40.

Stiglitz, Joseph. 1982b. “Utilitarianism and Horizontal Equity: The Case for Random
Taxation.” /Journal of Public Economics/, 18(1): 1-33.

Tarkiainen, Ritva and Matti Tuomala. 1999. “Optimal Non-Linear Income Taxation with
a Two-Dimensional Population: A Computational Approach.” Computational
Economics, 13(1): 1-16.

Summers, Lawrence H. 1981. "Capital Taxation and Accumulation in a Life Cycle

Growth Model." American Economic Review, 71(4): 533-44.

Tresch, Richard W. 2002. Public Finance: A Normative Theory. Second edition. Plano,
Texas: Business Publications, Inc.

Tuomala, Matti. 1990. Optimal Income Tax and Redistribution. New York: Oxford
University Press.

74
Varian, Hal R. 1980. “Redistributive Taxation as Social Insurance.” Journal of Public
Economics, 14(1): 49-68.

Vickrey, William. 1947. Agenda for Progressive Taxation. New York: The Ronald Press
Company.

75