ArticlePDF Available

Labor Supply Behavior and the Design of Tax and Transfer Policy


Abstract and Figures

This paper argues that recent empirical evidence on labor supply behavior - showing stronger participation responses than hours-of-work responses - has important implications for the design of tax and transfer policy. Based on a review of recent research in this field, we conclude the following: (i) Conventional ways of evaluating the impact of tax-transfer reforms on economic efficiency have to be revised. (ii) Optimal redistributional policies may well involve negative marginal tax rates at the bottom of the earnings distribution. (iii )T he introduction of the Earned Income Tax Credit (EITC) in the United States in the mid-1970s and the large expansions of the credit during the past two decades did not involve a trade-off between efficiency and equality as suggested by previous estimates. Instead, the EITC has improved both equality and efficiency. (iv) For most European countries, redistribution to the working poor involves a substantially lower trade-off between efficiency and equality than traditional redistributional policies targeted to those out of work.
Content may be subject to copyright.
Labor Supply Beha vior and the
D esign of Tax and Transfer P olicy
Henrik Jacobsen Klev en
University of Copenhagen, EPR U , and CEPR
Claus Th ustrup Kreiner
Universit y of Copenhagen, EPRU, and CESifo
Decem ber 2005
This paper argues that recent empirical evidence on labor supply behavior showing
stronger participation responses than hours-of-work responses has important implications
for the design of tax and transfer policy. Based on a review of recent research in this eld,
we conclude the following: (i) Conventional ways of evaluating the impact of tax-transfer
reforms on economic eciency have to be revised. (ii ) Optimal redistributional policies may
well involve negative marginal tax rates at the bottom of the earnings distribution. (iii)The
introduction of the Earned Income Tax Credit (EITC) in the United States in the mid-1970s
and the large expansions of the credit during the past two decades did not involve a trade-o
between eciency and equality as suggested by previous estimates. Instead, the EITC has
improved both equality and eciency. (iv) For most European countries, redistribution to
the working poor involves a substantially lower trade-o between eciency and equality
than traditional redistributional policies targeted to those out of work.
This pap er has been written on the occasion of Claus Thustrup Kreiner’s inaugural lecture as full Professor
at the University of Cop enhagen on the 4th of February 2005. Parts of the paper draw heavily on recent, joint
work with Nad a Eissa, Herwig Immervoll and Em m anuel Saez.
Redistribution to low-income individuals has grown signicantly in European countries and
in the United States since World War II. Today, most of these countries devote a sizeable
amount of public spending to various lo w-income support programs. These programs include
unemployment insurance, social assistance, in-work benets, health insurance, food programs,
housing programs, child care support, and disability benets for the disabled.
These transfer programs have been motivated by a desire to alleviate poverty and to make
the income distribution more equal. However, the presence of generous low-income support
provided by the government may have adv erse eects on the labor market. In particular, since
eligibility in most welfare programs is directly related to the size of income, individuals have an
incentive to reduce or under-report earnings so as to qualify for welfare payments. For example,
some people may want to reduce hours of work or eort on the job, or they ma y choose to opt out
of the labor market ent irely. There is convincing empirical evidence that welfare programs do in
fact induce labor supply responses of this sort (Mott, 2002). Moreover, taxes collected from
middle- and high-income earners to nance low-income support leads to additional negative
eects on labor supply and employment (Blundell and MaCurdy, 1999).
Thenegativeeects on labor supply implied by taxes and transfers create a loss of eciency
in the economy. In other words, policy makers face a trade-o between eciency and equality,
since more distributional equity generated by government programs implies lower eciency. The
size of this trade-o depends on the generosity and on the design of w elfare programs. If the
programs are inappropriately designed, they will be costly in terms of lost eciency implying
that the feasible amount of redistribution is lower. By contrast, if the programs are designed
to minimize the adverse eects on labor supply, then we can redistribute more.
An important choice in the design of low-income support is whether it should apply to the
poor or to the working poor. This policy choice has varied over time and across countries. In
continental Europe and in Scandina via, welfare programs tend to be aimed at guaranteeing a
minimum level of consumption for all individuals, including those without a job (redistribution
to the poor). By contra st, Anglo-Saxon countries tend to spend large amounts on in-work benet
programs targeted to the working poor. In the United States, for example, the so-called Earned
Income Tax Credit (EITC) has become the single largest cash transfer program for lower-income
families at the federal level. Eligibility for the EITC is conditional on having a low income and
on hav ing a job. The program attempts to achieve the dual goal of redistributing income to
low-income earners and avoiding some of the adverse eects on the labor market mentioned
above. In particular, by making the transfer conditional on employment, the EITC aims at
increasing labor market participation for the eligible population.
A program similar to the EITC was introduced in the United Kingdom in the 1988 and
expanded substantially in the late 1990’s. Moreover, a number of continental European countries
(including Denmark) have recently experimented with in-work benet programs, although on
a very small scale. In other words, the introduction and further expansion of in-work benets
targeted to the working poor is high on the agenda when discussing w elfare reform in Europe.
Since the expansion of suc h programs at the expense of traditional welfare programs
would represent a fundamental change in the approach to income redistribution, it is important
to understand their implications for distribution and the labor market.
In this paper, we provide an overview of some of the central issues in the evaluation of
welfare and tax reform, and we discuss the appropriate design of redistributional policies. The
paper pays special attention to the importance of modelling labor supply decisions correctly
when considering policy reforms aecting the labor market. Until recently, a common feature
of the public nance literature was an almost exclusive focus on hours of work for those who
are working (the intensive margin of labor supply). However, an emerging consensus in the
empirical labor market literature is that most of the observed variation in labor supply reects
changes in labor force participation (the extensive margin of labor supply), whereas changes
in hours worked are much smaller (Heckman, 1993; Blundell and MaCurdy, 1999). Consistent
with these ndings, we outline a simple theoretical framework accounting for both the intensive
and extensive margins of labor supply response, and we review a number of recent empirical
applications showing that the composition of the response on the two margins is crucial for the
evaluation of tax and transfer policy.
Considerable space will be devoted to discussing the desirability of policies targeted at the
working poor versus traditional welfare policies aimed at the poorest (non-employed) segments
of the population. Based in part on the experience in the US and the UK, we argue that there
is a promising scope for in-work benet reform in continental Europe. In fact, in-work benet
reform seems to be particularly desirable in a country such as Denmark where eective tax
rates are very high at the bottom of the income distribution resulting from generous support
to those out of work combined with high tax rates on the working population.
In Scandinavia, policy proposals of this sort are usually met by two critiques. First, it is
pointed out that in-work benet reform tends to be less desirable in a country characterized by a
highly compressed earnings distribution. We discuss this argument and nd that it is not strong
enough to overturn the case for in-w ork benet reform in Denmark. Second, it is pointed out that
most of the evidence on strong participation responses has been based on countries characterized
by relatively low female labor force participation rates. While this argument is not without
merit, we argue that working poor policies will be desirable in Denmark even at participation
elasticities far below those estimated for the United States and the United Kingdom. This is
because the design of the Danish welfare and tax system creates exceptionally high eectiv e
tax rates at the bottom, implying that even very small elasticities involves substan tial eciency
losses on the margin.
The paper is organized as follow s. Section 2 provides a short review of the empirical labor
supply literature and discusses the appropriate theoretical modelling of labor supply behavior.
Section 3 provides a simple introduction to the measurement of the welfare costs of taxation.
Section 4 shows theoretically and empirically why it is important to distinguish explicitly be-
tween intensive and extensive labor supply responses when evaluating the impact of tax reforms.
Section 5 discusses dierent issues involved in the optimal design of redistributional policy and
provides empirical estimates of the trade-o between equality and eciency for 15 European
countries. Section 6 concludes.
2 Labor supply behavior at the in tensive and extensive margins
2.1 A brief review of the empirical literature
A central nding in the empirical literature is that labor supply elasticities are low at the
intensive margin (Heckman, 1993; Blundell and MaCurdy, 1999). It is notable that this nding
is consistent across dierent methods and dierent samples, and that it holds for both males and
females. The old ndings of high elasticities for women (married w omen and single mothers)
werebasedoncensoredspecications including non-participating individuals, thereby conating
extensive and intensive responses in the estimated elasticity. Once labor supply is estimated
conditional on labor force participation, it turns out that the female hours-of-work elasticity is
close to that of males (Mroz, 1987; Triest, 1990; Blundell et al., 1992).
Hence, a strong degree of labor supply responsiveness would have to come from the margin
of entry and exit in the labor market. Indeed, there is an emerging consensus that extensive
labor supply responses may be much stronger than in t ensive responses (Heckman, 1993). In
particular, participation elasticities seem to be very high for certain subgroups of the population,
typically people in the lower end of the earnings distribution. Let us briey review some evidence
from both the United States and Europe.
One body of empirical work exploits evidence from recent tax reforms and expansions to
tax-based transfers in the United States and the United Kingdom. In particular, a number
of in-work benet reforms targeted at lower-income families with kids has provided an ideal
opportunity to estimate labor supply behavior. For the United States, Eissa and Liebman
(1996) use a quasi-experimental approach to show that the 1986 expansion of the Earned Income
Tax Credit (EITC) had large eects on the labor force participation of single mothers. This
was especially the case for single mothers with low education, where the Eissa-Liebman study
implies an elasticity around 0.6. Meyer and Rosenbaum (2001) use a more structural approach
and data from 1985 to 1997, a period of time where three large tax reforms were enacted in the
US (in 1986, 1990, and 1993). Their nding that the EITC accounts for about 60 percent of the
increase in the employment of single mothers over the period implies a participation elasticity of
about 0.7. A study by Blundell et al. (2000) considers the labor market impact of the recently
implemented Working Families Tax Credit (WFTC) in the UK. Like the EITC, this program
was designed to induce low-income people with kids, typically lone mothers, from welfare into
work. Their results indicate that the reform was quite eective in achieving this goal, increasing
the participation rate of single women with children by 2.2 percentage points (5 percent).
While the literature on labor supply in Anglo-Saxon countries is extensiv e, there are fewer
studies for continental European and Scandinavian countries. One might expect that labor
supply elasticities are smaller in the more rigid labor markets of continental Europe and Scan-
dinavia. However, several recent studies suggest that this is not the case. A number of structural
studies of female labor supply are surveyed in Blundell and MaCurdy (1999). These studies
found high elasticities typically between 0.5 and 1 across a number of European countries
such as Germany, Netherlands, France, Italy, Sweden, and the UK.
Since these results were
based on samples containing non-employed individuals, the participation response is included in
the estimated elasticities. Indeed, the results probably reect an underlying labor supply curve
where elasticities are large around the poin t of participation but fall o rapidly with increases
in working hours (Blundell, 1995).
The nding that tax incentives may have substantial eects on labor force participation is
consistent with another stream of empirical literature estimating the eect of out-of-work bene-
ts on unemployment. Krueger and Meyer (2002) surveys the evidence from a number of OECD
countries. They conclude that benets raise the incidence and the duration of unemployment,
and that the elasticity of lost work time with respect to benets tend to be around one. Since
the risk of unemployment is largest among lo w-skilled w orkers, this evidence also indicates that
strong participation responses tend to be concentrated at the bottom of the wage distribution.
2.2 Modelling en try-exit behavior in the labor m ark et
Having established that extensive labor supply responses are empirically important, we wish
to explore their implications for the welfare evaluation of tax and transfer policy. In order to
do so, we have to think about how to model participation behavior. As emphasized by Eissa
et al. (2004), a realistic model must be consistent with empirical distributions of hours worked
showing almost no workers at low annual or w eekly hours of work. In other w ords, if individuals
decide to work at all, they tend to work a substantial number of hours (say, 30 or 40 hours per
week). To explain this discreteness of participation behavior, we need a framework featuring
non-convexities in preferences and/or budget sets. The observed discrete responses along the
extensive margin cannot be captured within a standard convex labor supply model.
To see the point, Figure 1 illustrates the choice of labor supply in the standard convex
model. In the gure, we consider two individuals facing the same budget constraint, where y
is non-labor income, w is the wage rate, and (1 m) w is the marginal net-of-tax wage. The
indierence curves of the two individuals are drawn such that individual 1 has a relatively low
valuation of leisure, while individual 2 has a relatively high valuation of leisure. Consequen tly,
We are not aware of studies estimating the elasticity of labor force participation with resp ect to taxes for
Denmark. While the results for Sweden may give some indication of labor supply behavior in De nm ark, it would
b e interesting and important to carry out econometric studies based on Danish data and reforms in order to
discuss the design of future tax and welfare reform in this country.
See Eissa, K leven and Kreiner (2005) for a more detailed discussion of the m odelling of labor supply.
it is optimal for the rst individual to work many hours (at point A
), whereas the second
individual chooses to stay outside the labor market (at point A
) since there is no point of
tangency at positive hours. As we have drawn the gure, the poin t of tangency for individual
2 is exactly at zero hours worked.
1-m w
gure 1: Intens
ve versus Extens
ve Responses
n t
e Convex
Consider now a small reduction in the tax rate increasing the slope of the budget line a
little bit. In principle, we want to consider a marginal (innitessimal) tax change, but we have
exaggerated the size of the change in the gure for the point of illustration. The increased slope
of the budget line induces individual 1 to increase his hours of work a little bit from point A
to point B
(an intensive response). For individual 2, on the other hand, the higher net wage
gives rise to an interior solution so that he decides to enter the labor market at point B
extensive response). Notice that this extensive response involves a change in labor supply from
zero hours to some small (innitessimal) number of hours. Hence, the type of participation
response predicted by this framework is a marginal one, just like the change in hours of work
for those who are working. This conicts with the empirical evidence described above showing
that people do not enter the labor market at innitesimal hours of work but that they do so
at, say, 30 or 40 hours. Therefore, to be able to capture participation behavior in a realistic
way, we need to adopt a framework incorporating some type of non-convexity making very low
hours of work non-optimal.
1-m w
Figure 2: Intensive versus Extensive Responses with Fixed
Costs of Working
In the empirical literature, non-con vexities are typically introduced by way of xed work
costs (e.g. Cogan, 1981; Blundell et al., 1987). In Cogan’s (1981) analysis, the xed costs may
be monetary costs (say child care and transportation expenses) or they may take the form of a
loss of time (e.g., commuting time). In Figure 2, we extend the analysis of labor supply choice
along these lines. An individual who chooses to stay outside the labor market receives non-labor
income y
. If he decides to enter the labor market, he looses y
in income and l
in leisure
time upon entry, thereby creating a discontinuity in the budget set. In the initial situation, it is
still the case that individual 1 works relatively long hours (at point A
), while individual 2 does
). Now, if we reduce the tax rate a little bit, individual 1 responds
again with a marginal ch ange in working hours from A
to B
. By con trast, individual 2 now
reacts by making a discrete jump from not working at all to working nearly as many hours
as individual 1 (at point B
). This discreteness of the participation choice is consistent with
empirical distributions of hours worked. Hence, the incorporation of xed work costs allows for
a more realistic model of participation behavior.
To summarize, we note that tax reforms entail intensive as well as extensive labor supply
responses, and to account for the observed discreteness of extensive responses, they have to be
modelled by introducing non-convexities into the standard framework.
3 The excess burden of taxation
In general, the tax system aects labor supply along both the intensive and extensive margins.
These distortionary eects on labor supply aect government revenue and give rise to a loss of
economic eciency. This is what we call the excess burden or the deadweight loss of taxation.
The aim of this section is to explain why it is important to distinguish explicitly between
the intensive margin and the extensive margin when trying to measure the excess burden of
taxation. Notice that it is not obvious a priori that we need to distinguish between the two
margins in order to quantify the eciency cost of the tax system. Taxes distort labor supply
along both dimensions but maybe this could be captured simply by looking at the aggregate
labor supply curve thereby incorporating both the intensive margin eect and the extensive
margin eect in a single relationship. Although this reasoning may sound plausible, it turns
3.1 The excess burden of taxation with in tensiv e labor supply responses
We start with a review of the standard deadw eight loss of taxation which focuses only on the
intensive labor supply margin. Consider an individual with an hourly productivity/wage of
In addition to xed work costs, the presence of low-income supp ort programs featuring gradual phase-out
and possibly discrete e ar nings and wo rk tests cr e ate non-conve xities making low hours of work very unattractive.
In fact, Figure 2 can easily be reinterpreted to illustrate the case of non-convex tax-tran sfer program s by thinking
of the c ost y
as an out-of-work welfare benet which is lost upon entry.
w. For simplicity, assume that the individual faces a simple proportional tax scheme where m
denotes the marginal tax rate. The individual is willing to work more hours at a higher after-tax
wage rate thereby giving rise to the upward-sloping labor supply curve illustrated in Figure 3a.
With no taxes, the labor mark et would be at point A. The impact of the tax system is to reduce
the hourly take-home wage from w to (1 m) w thereby giving rise to an equilibrium at point
Figure 3a: Welfare cost of taxation with
intensive labor supply responses
Hourly wage
Deadweight loss
Government revenue
Consumer surplus
Figure 3b: Marginal welfare cost of taxation
with intensive labor supply responses
Hourly wage
Behavioral effect
Mechanical effect
At point B, the individual obtains a consumer surplus given by the area CS, the government
revenue is given by the area labelled GR, while the triangle DWL captures the deadweight
loss from the tax. The deadweight loss measures the amount that is lost in excess of what
the governments collects. If the same tax revenue were to be collected by a lump sum tax
a tax which is independent of earnings the individual would continue to supply labor at
point A, and obtain a consumer surplus equal to CS + DWL.
Hence, the dierence between
the income tax and the lump sum tax is exactly equal to the area DWL. This area therefore
measures the eciency cost of having a tax system that depends on income.
Formally, the deadweight loss may be derived by rst calculating the area GR + DWL
using integration and afterwards subtracting the rectangle GR.Thisgives
h (x) dx m · w · h ((1 m) w) , (1)
The t ax system inuence s h ours worked throu gh a substitution e ec t and an income eects. The income
eect is, however, irrelevant for the measurem ent of excess burden. Notice that the excess burden is derived by
comparing a given tax system with a lump sum tax system that generates the sam e tax revenue. A ccord ingly,
the net-income of the individual does not change when going from one system to the other.
where h (·) is the labor supply function of the individual, while h ((1 m) w) denotes hours
worked at point B in Figure 3a.
Equation (1) reects the total deadweight loss of the income tax system, but this is typically
not very interesting for tax policy. The reason is that practical tax reforms tend to involve
relatively small changes within the existing tax schedule rather than abolishing income taxation
altogether. For tax policy analysis, it is therefore more interesting to look at the eciency
impact of small adjustments in the existing system, i.e. the marginal deadweight loss.
Figure 3b illustrates the impact of a small increase in the marginal tax rate m.PointB is
the choice of the individual at the pre-reform tax system (equivalent to point B in Figure 3a).
The reform reduces the marginal net-of-tax wage and moves the optimal choice from B to C.It
can be seen in the gure that the increase in the deadweight loss triangle following the reform
is given (approximately) by the black rectangle.
To understand the source of this eciency
loss, it is useful to relate it to the eects on government rev enue. In general, tax reforms
aect government revenue through two channels the mechanical eect and the behavioral
eect. The mechanical revenue eect reects the impact of the increase in the tax rate at given
behavior, and it corresponds to the gray rectangle in the gure. This eect reects a transfer
of income from the individual to the government (the consumer surplus is reduced by the same
area), and it has no consequences for eciency. In other words, although the mechanical eect
represents a welfare loss to the tax pa yer (a loss of income), it entails no loss of ressources for
the economy and hence no eciency loss.
The behavioral revenue eect reects the impact of changed behavior at the given tax
rate. Because the increase in the marginal tax rate reduces hours worked from point B to C,
taxable earnings goes down leading to a negativ e feed-back eect on government revenue. This
behavioral revenue eectfromthetaxreform–givenbytheblackareainthegure is
exactly identical to the marginal deadweight loss of taxation. The insight that the marginal
deadweight burden is equivalent to the beha vioral revenue eect is not specic to the model
adopted here. It is an insigt which follows from any model where individuals optimize and
markets are ecient, a point which we come back to below.
From the graph, or by dierentiation of expression (1), w e see that the change in the
This is only approximately true, since the black rectangle does not include the small w hite triangle imme-
diately below it. However, this triangle is a se cond-order eect arising because the reform in the gure is not
exactly marginal (innitessimal). For a marginal reform, this eect is zero.
deadweight loss becomes
DWL = m · w · h
((1 m) w) · w · m, (2)
where h
(·) denotes the derivative of the individual labor supply with respect to the after-tax
wage rate. A small increase in the tax rate, m, induces the individual to reduce the number
of hours by h
(·) · w · m. Since a one-hour reduction in labor supply reduces the tax payment
by m · w, it follows that the change in the deadweight loss is exactly equal to the behav ioral
feed-back eect on government revenue. This insight, that the welfare loss is given by the
revenue loss created by changed behavior, is not related to the specic model adopted here but
holds in general for tax policy analysis.
For empirical applications, it is useful to rewrite the above result. Let ε denote the elasticity
of hours worked with respect to the net-of-tax wage rate.
Then the c hange in the deadweight
loss in proportion to the aggregate wage income may be written as
1 m
· ε · m. (3)
This expression is a classic Harberger-type formula for the marginal deadweight burden of
It shows that the marginal welfare cost depends on the initial level of the marginal
tax rate, the increase in the marginal tax rate, and the hours-of-work elasticity. Notice that
the marginal loss is equal to zero when the marginal tax rate is zero. That is, initially the
welfare cost of raising tax revenue is zero. As the tax rate is increased, the marginal welfare
cost of taxation also increases at a giv en labor supply elasticity. In other words, it becomes
more and more costly to raise additional government revenue as the tax rate goes up. In fact,
welfare eects can be substantial even for very small labor supply elasticities provided that initial
marginal tax rates are high. Finally, since the empirical evidence does not give any indication of
large, signicant dierences in hours-of-work responses across individuals (conditional on labor
Notice that the m arginal tax rate also enters in the lower limit of the integral in equation (1). Using Leibnitz’s
rule, the derivative of this term with resp ect to m becomes w · h ((1 m) w).
To be precise, ε is the compensated h ou rs- of-work elasticity. As described in footnote 3, inc ome eects are
irrelevant for the m easu rem ent of the excess burd en of taxation.
The result is m ore general than it app ears. Specically, the result is not restricted to a proportional tax
scheme although this is the case for Figure 3a,b and for expression (1). The relationship in (3) holds for any
tax system as long as the marginal tax rate is locally constant. The result also holds for other denitions of
the de a d-weight loss. Here, we h ave derived exp ression (3) from a simple M a rsh alian denition but the result
may also be derived from dead-weight loss m easures based on comp ensating variation, equivalent variation, or
compensating surplus.
force participation), we may conclude that the largest eciency gains of tax rate reductions are
to be found where marginal tax rates are relativ ely high.
3.2 The excess burden of taxation with extensive labor supply responses
The modelling of tax distortions along the extensive margin is a bit more complicated. As noted
in Section 2, an appropriate theory requires some type of non-convexity in order to accoun t for
the observed discreteness in labor supply behavior.
In addition, the theory has to incorporate
heterogeneity across individuals. With no heterogeneity in preferences or productivity, either
everybody participates or nobody participates in the labor market. In reality, some fraction
of those in their working age participate, and small changes in labor market incentives will
typically create small adjustments in the number of emplo yed workers. The combination of
non-convexities and heterogeneity enables us to explain discrete entry-exit behavior at the
individual level together with smooth changes in emplo yment at the macro level. Here we
present a very simple framework which illuminates the main insights. A more general analysis
is presented in Eissa et al. (2004).
Consider a group of individuals who all have the same hourly productivity/wage w.For
simplication, we disregard intensive labor supply responses and assume that all individuals
work some xed number of hours,
h, if they work at all. Those who are working receive the
same pre-tax earnings denoted b y Y w
h. The individuals face xedworkcostswhichmay
reect monetary costs, time costs or simply a xed disutility of working. These work costs
are assumed to vary (smoothly) across individuals, thereb y giving rise to the upward sloping
labor supply curve in Figure 4a. This curve displays the number of employed individuals as
a function of the net income gain of working (i.e., earnings net of taxes paid and transfers
lost). At a low net income gain, many individuals stay outside the labor market because their
work costs outweigh the gain from entry. As the net income gain increases, more and more
individuals nd it worthwhile to enter the labor market so as to generate a positively sloped
extensive labor supply curve.
A general framework for analyzing welfare eects with discrete choice is provided by Small and Rosen (1981).
Figure 4a: Welfare cost of taxation with
extensive labor supply responses
Deadweight loss
Government revenue
Consumer surplus
Income gain
from entry
Figure 4b: Marginal welfare cost of taxation
with extensive labor supply responses
Behavioral effect
Mechanical effect
Income gain
from entry
With no tax-transfer system, the equilibrium would be at point A. Assume now that an
employed wh o is earning Y has to pa y an amount T in taxes (net of any benets), while a
non-employed individual receives the amount B in welfare benets (net of any taxes). This
implies that the net income gain of working equals Y T B, thereby moving the equilibrium
number of employed from point A to point B. Those who work at point B obtain consumer
surplus CS and generate a government revenue (taxes paid plus benets saved) equal to GR.
The tax-transfer system driv es a w edge, T + B, in between the social return to work and the
private return, thereby creating a deadweight loss equal to the area DWL.
Formally, the deadweight loss may be written as
E (x) dx (T + B) · E (Y T B) , (4)
where E (·) denotes the number of employed people as function of the net income gain of
working. The rst term measures the size of the area GR + DWL in Figure 4a, while the
second term equals the area GR.
Let us now consider the impact of a reform making a small adjustment to the existing
tax system. Figure 4b displays the consequences of a small increase in the tax burden on the
employed, which moves the equilibrium from point B to point C. As in the previous model, it
is useful to distinguish between the mechanical eect and the behavioral eect on government
revenue. The mechanical eect is the increase in government revenue at given labor market
participation, and it is shown by the gra y area in Figure 4b. This eect corresponds to a
redistribution of income from the emplo yed to the government and involves no deadweight
loss. The distortionary eect of the tax change comes entirely from the behavioral eect on
government revenue. The tax change reduces the number of employed individuals and this
reduces the net-payment to the government by T + B per person. Accordingly, the change in
the deadweight loss becomes
DWL =(T + B) · E
(Y T B) · T, (5)
where E
(·) denotes the derivative of the employment function. This result may also be derived
from dierentiation of expression (4).
The sensitivity of employment with respect to changes in the net income gain from entry
may be measured by the extensive labor supply elasticity η
· (Y T B). In addition,
we dene an average tax rate a T/Y and a benetrateb B/Y . With these denitions,
expression (5) may be rewritten to
Y · E
a + b
1 a b
· η · a, (6)
where a = T/Y denotes the reform-induced change in the average tax rate. This deadweight-
loss formula reects the same basic form as the traditional one in eq. (3), but it is related to
dierent policy parameters and a dierent elasticity. In particular, the welfare cost is no longer
related to the marginal tax rate. It is instead related to the tax rate applying at the extensive
margin the participation tax rate calculated as the sum of the average tax rate and the
out-of-work benet rate, a + b. The participation tax rate measures the fraction of earnings
that is lost in taxes and forgone benets upon labor market entry. Finally, the w elfare eect
depends on the sensitivity of entry-exit behavior as measured by the elasticity of labor force
participation with respect to the net-of-tax income gain from entry, η.
In general, tax reform aects labor supply along both the intensive margin and the extensiv e
margin. The strength of the responses are determinedbythesensitivityofeachmarginwith
respect to economic incentives, captured by the two elasticities ε and η. The intensive and
extensive adjustments create a c hange in aggregate labor supply, dened as the total number
of hours worked of all employed people (E · h). It is therefore natural to ask if it is possible to
construct a measure of the deadweight burden which relies only on the sensitivity of aggregate
labor supply with respect to economic incentives. The above analysis demonstrates that, in
general, it is not feasible to construct such a measure, since labor force participation is related
to a dierent tax wedge than are working hours.
Having said that, it should be noted there is one special case where tax rates on the two
labor supply margins are identical. This is the case where the entire tax-transfer system is
characterized by a linear Negative Income Tax (NIT). This type of system provides a guaran-
teed minimum income B to all individuals in the economy (participants and non-participants)
nanced by a constan t marginal tax rate m on earnings. In this special case, the participation
tax rate a+b is exactly iden tical to the marginal tax rate m.
We may then sum the right-hand
sides of eqs (3) and (6) so as to obtain the total welfare eect in proportion to aggregate labor
income as
Y · E
1 m
· (ε + η) · m, (7)
where ε + η is the aggregate labor supply elasticity. Notice, that the only dierence between
this formula and the pure intensive welfare eect in formula (3) is the presence of the aggregate
labor supply elasticity instead of the hours-of-work elasticity.
An approach based on (7) would be simple and has been used in some empirical work, but
it is unlikely to yield accurate results. It requires that the entire welfare system is a linear NIT,
which is far from being true in empirical applications. Public benets tend to be non-univ ersal,
targeted to lo w and middle income classes through earnings or work tests (creating discon-
tinuities) or through gradual phase-outs. Consider as examples lo w-income support, in-work
benets, housing- and education subsidies, child benets, medical aid, food stamps, and public
pensions. While some benets may be universal in some countries, never are they collectively
so. Moreover, income tax systems are generally characterized by increasing marginal tax rates
further conicting with the NIT assumption.
In Section 4, we demonstrate the potential large
errors from not accounting properly from the welfare eects along the extensive margin.
3.3 Excess burden, governmen t revenue, and other dimensions of behavioral
In the previous sections, we set up two simple models to derive formal expressions for the
marginal excess burden of taxation with intensive and extensive responses, respectively. We
emphasized the equivalence of the marginal excess burden with the impact of behavioral re-
sponses on government revenue. The insigh t that the eciency eect of policy reform is giv en
To see this, notice that the net tax payment of each individual under a linear NIT is given by T = B +mY .
By dividing this relationship with earnings Y ,weobtaina + b = m.
A description of the actual tax-transfer systems prevailing in a number of European countries is provided by
Immervoll et al. (2005), while the US tax-transfer system is describ ed in Eissa et a l. (2004).
simply by the behavioral revenue eect is not specic to the simple models adopted above. It is
an insight which follo ws from any model where individuals optimize and mark ets are ecient,
i.e., except for the ineciencies int roduced by the existence of distortionary taxation.
To see that the result holds in general, consider a small (marginal) tax reform changing gov-
ernment revenue by GR. In general, this change in government revenue reects a mechanical
revenue eect, M, and a beha v ioral revenue eect, B, such that GR = M + B.In
any model where agents optimize and markets are ecient, the total utility loss (in monetary
units) on individuals will be exactly equal to the mechanical tax increase M.Toseethis,
start by noticing that in the absence of behavioral responses a higher tax burden involves
a transfer of income from individuals to the government given by M, creating a utility loss for
the individual equal to M. But of course behavioral responses are not absen t: the individual
has an incentive to re-optimize behavior so as to avoid some of the tax increase. However, as
long as the reform is small, this behavioral adjustment entails no rst-order eect on utility,
since individuals were initially in their optimum. This is an application of the envelope theorem.
The marginal deadweight loss from the reform, DW L,isdened as the marginal utility
loss to individuals in excess of the additional revenue collected. By exploiting the above rela-
tionships, we immediately obtain DWL = M R = B, showing that the marginal
deadweight loss is exactly identical to the behavioral revenue loss. Notice that, under the as-
sumptions of rational individuals and ecient markets, the result relies simply on an application
of the envelope theorem and the denition of the deadweight loss. In particular, it does not
rely on the form of preferences and tec hnology, nor does it depend on the framework being one
of contin uous choice. For example, the result encompasses the case of discrete participation
behavior due to non-convexities in budget sets or preferences.
There are several points to make in relation to this result. First, behavioral responses
are central to the creation of eciency losses, even though these responses entail no rst-order
utility e
ects on individuals due to the envelope theorem. The point is that behavioral responses
create eciency losses indirectly in the form of an externality operating through the government
budget. For example, although a marginal labor supply reduction involv es no rst-order eect
on the optimizing worker (envelope theorem), it implies a loss of governmen t revenue and
therefore less money for the nancing of, say, public goods, creating a negative externality on
everybody else.
Second, the derivation did not specify the type of behavioral response we are considering.
Following most of the literature, this paper has focused on labor supply behavior in its standard
dimensions: hours worked and labor force participation. But there are many other dimensions
of labor supply that may be responsive to taxation. For example, higher taxes may lead to lower
eort on the job, and they can have adverse eects on the incentiv es to improve skills by taking
education or engaging in on-the-job training. Taxation can also aect the type of jobs that
people take (in terms of profession, geographical location, etc.) and it may aect search and
labor mobility. All of these dierent types of labor supply responses can aect taxable earnings
and government revenue, thereby creating eciency losses in the economy. Although there has
been a tremendous focus in the empirical literature on hours worked, the other dimensions of
labor supply response are just as important from the point of view of economic eciency.
Third, the relationship between excess burden and behavioral revenue eects implies that
we should go beyond thinking just about labor supply in its various dimensions. Instead, we
should be thinking about changes in taxable income more generally. Taxable earnings may
respond to taxation through several other margins than labor supply. For example, individuals
may be able to change change the form of payment for labor services in to something which
is more leniently taxed. This includes fringe benets such as corporate cars, in-house sports
facilities, free lunch and dining, laptop computers, etc. It also includes turning labor income
into capital income (typically carrying a lower rate of tax) by being paid in stock options
instead of a traditional w age income. Alternatively, taxable income can respond if higher taxes
lead to a more aggressive interpretation of tax rules (e.g., claiming questionable deductions)
or tax evasion (understating income, claiming unjustied deductions). Tax-induced changes in
avoidance or evasion aect government revenue, hence creating eciency losses in the same way
that labor supply responses do.
This type of reasoning underlies the recent literature on the elasticity of taxable income,
starting with the important paper by Feldstein (1995). While taxation may aect behavior in
a myriad of ways as suggested by the above comments the responses matter for eciency
only to the extent that they change taxable income. While it may be extremely dicult to
estimate each of the responses separately, it is possible to estimate the elasticity of taxable
income based data from individual tax returns. A focus on this parameter might make intensiv e
responses more important than portrayed in our review of the empirical labor supply literature.
Indeed, for the United States, the literature has shown that the (intensive) elasticity of taxable
income can be very high at the top of the income distribution (see Saez, 2004 for a recent survey) .
However, this is a phenomenon ocurring only at the extreme top of the income distribution
(top 1%). We are not aware of studies estimating the elasticity of taxable income for European
Finally, based on the discussion in this section, one might be tempted to conclude that we
do not need to distinguish between dierent margins of behavio ral response. To estimate the
revenue (and hence eciency) eects, it seems that all we need to know is the elasticity of
taxable income and then apply a marginal tax wedge. However, this is not entirely correct.
An obvious point is that the tax system may in volve several dierent bases subject to selective
taxrates.Inthiscase,dierent margins of response can relate to dierent bases and tax rates,
and we then have to distinguish between dierent taxable income elasticities to get the revenue
and eciency eects right. A more subtle point is that, under non-linear tax systems, marginal
and discrete changes in taxable income are taxed dierently. As shown in the previous sections
on intensive and extensive labor supply eects, it is important to distinguish between marginal
and discrete eects in the measurement of excess burden. Hence, the discussion in this section
is entirely consistent with the previous conclusions.
4 The evaluation of tax and transfer reform
Following Harberger (1964), a large numerical literature has attempted to evaluate the distor-
tions to the labor-leisure choice induced by labor income taxation and the impact of dierent
tax reforms. The evaluation methods include simple deadweight loss calculations (Browning,
1987, 1995), Computable General Equilibrium models (Ballard et al., 1985; Ballard, 1988),
and microsimulations (Browning and Johnson, 1984; Triest, 1994). A common feature of this
literature is the assumption of a standard convex labor supply model, ruling out discrete partici-
pation responses. Although the models applied are typically more sophisticated than the theory
presented above, the basic forces driving the results are the same as those underlying formula
(3). Extensive labor supply responses are either ignored, or included implicitly by substituting
the elasticity of aggregate labor supply for the hours-of-work elasticity.
Ignoring extensive
Indeed, it is not uncomm on that simulation studies based on the standard convex model employ total
elasticities in their calibration. For example, this seems to be the case in Browning and Johnson (1984), Ballard
responses corresponds to completely disregarding the welfare eects in formula (6). The second
approach corresponds to applying a formula like (7), thereby approximating the participation
tax rate (a + b) and the change in this rate (a) with the level and change, respectively, in
the marginal tax rate (i.e. m and m). In this section, we demonstrate that these approaches,
based on the traditional labor supply model, may be seriously misleading. As a case in point,
we examine the welfare eects for single women with dependent children in the United States
following four tax acts passed in 1986, 1990, 1993 and 2001.
The welfare eects of these reforms on single mothers are particular interesting to study.
The group of single mothers experienced large tax cuts from the reforms due to expansions of the
Earned Income Tax Credit (EITC), and because of increases in the standard deduction, personal
exemptions, and a more favorable tax rate schedule. The combination of all these elements
implied substantial improvements in the incentive to supply labor along both the intensive
and extensive margin. The labor supply responses to these changes in incentives have been
studied extensive ly and, not surprisingly, all the evidence point to strong participation responses
(cf. the discussion in Section 2). On the other hand, hours-of-work responses (conditional on
participation) were very small.
4.1 Eects on labor supply and eciency of an Earn ed Income Tax Credit
Let us rst discuss theoretically how an EITC inuences labor supply and economic eciency.
Figure 5 illustrates the impact on labor supply of introducing an EITC. The straight line
represents the relationship between disposable income and earnings before the introduction of
the EITC (assuming a linear tax system for simplicity). We assume that the individuals face
a xed cost of working implying that each person will have to work a certain number of hours
before consumption is raised above the non-working level. The impact of the EITC depends
on the earnings level of the individual. The credit is rstphasedinatlowearnings,thenstays
constant, and is nally phased out at higher earnings.
The consequence for the relationship
et al. (1985), Ballard (1988), Browning (1995), and Bourguignon and Spadaro (2002a, 2002b). In all these
studies, high female labor supply elasticities (around 0.5 1.0) a re used in the calib ration, a lthough elasticity
estimates of this magnitude tend to be based on censored specications including observations with zero hours of
work (Mroz, 1987; Triest, 1990). By implication, the above studies are lumping together extensive and intensive
responses in the simulations.
Notice that the EITC recently introduced in D enm ark is dier ent. T he Danish EIT C is given to everyon e
participating in t he labor m arket, while the EITC in the Un ited States is phased-out at high earnings. In addition,
the size of the EITC in the US is related to the number of depende nt kids an d, if married, on the income of the
between disposable income and earnings is shown in Figure 5. The total tax burden on labor
income is reduced in all three intervals of the EITC. The impact on the marginal tax rate
depends on the initial earnings leve l of the person. In the phase-in interva l, the marginal tax
rate is reduced, in the plateau range it is unchanged, while in the phase-out range it is increased.
The labor supply responses along the intensive margin, illustrated by the small arrows in
Figure 5, depends on substitution and income eects. In the phase-out range both of these
eects reduce labor supply. In the at range only the income eect is present which leads to a
reduction in labor supply. Finally, in the phase-in range the substitution eect and income eect
have opposite signs, so that hours increase as long as the substitution eect is stronger than the
income eect. This is equivalent to the so-called uncompensated hours-of-work elasticity being
positive in this range.
Figure 5: Impact of an Earned Income Tax Credit
Disposable Income
Phase-in Flat Phase-out
The sign of the intensive margin welfare eect depends on the direction of the change in the
marginal tax rate (cf. formula (3)).
An individual in the phase-in region experiences a lower
marginal tax rate (m<0), yielding a positive welfare eect. On the other hand, the welfare
eect is zero for individuals in the plateau region (where m =0),whilethewelfareeect is
negative for individuals in the phase-out region (where m>0). The total eect on economic
eciency from intensive responses may go either w ay but, in practise, it is quite likely that the
negative eects will dominate. The reason is that the welfare eects have to be weighted by the
aggregate labor income in the respective income segments (observe that the deadweight loss in
Notice that ε in formula (3) is the compensated hours-of-work elasticity. Thus, only t h e substitution eects
matter for welfare.
eq. 3 is measured in proportion to the earnings of the individual). Since earnings are higher in
the phase-out range this gives, ceteris paribus, a reason to expect an overall welfare loss.
The welfare eects created along the extensive margin are unambiguously positive in all
income ranges of the EITC. In all ranges, the total tax burden decrease (i.e. a<0 in all
intervals), which yields a positive behavioral response on governmen t revenue (cf. formula (6)).
The total welfare eect of introducing the EITC depends in the end on the strength of the
dierent eects. However, if the tax wedge on participation is non-negligible it is reasonable to
expect a positive total eect since participation elasticities are very large compared to hours-
of-work elasticities for the group of single mothers.
The potential errors of basing the evaluation on the standard, convex framework is now
easy to see. A w elfare analysis disregarding extensive responses is biased downwards, and the
error may be substantial if participation elasticities are large. If we try to include extensive
responses by substituting the aggregate labor supply elasticity for the hours-of-work elasticity
in the traditional form ula, the error may very well become even bigger. Assume, for example,
that the intensive w elfare eect is negative when calculated using the hours-of-work elasticity,
while the total eect is positive after accounting for extensive eects in the proper way. If
we now try to modify the standard approach by substituting a large aggregate labor supply
elasticity for a small hours-of-work elasticity then the measured welfare eect becomes even
more negative, thereby getting futher away from the correct positive eect. This t ype of error
seems to underlie the pessimistic results in Brownin g (1995) concerning the EITC. In the next
subsection, we try to quantify these errors in an empirical application.
4.2 Welfare eects on single mothers of four tax reforms in the U nited States
In Eissa, Kleven and Kreiner (2004), we estimate the welfare eects on single mothers of a series
of tax acts passed in the United States in 1986, 1990, 1993 and 2001. Our estimation is based on
a microsimulation approach using a general theoretical framework which incorporates intensive
and extensive labor supply behavior as well as individual heterogeneity in productivities and
preferences. The empirical implementation requires information about labor supply elasticities
as well as various tax/transfer parameters and wage income shares. In order to get unbiased
results, it is particular important that we account for all elements in the tax and benets system
that drive a wedge in between the social return and the private return of supplying labor.
The labor supply elasticities are based on the empirical literature reviewed in Section 2.
To generate the tax/transfer parameters, we estimate individual eective tax rates (m and
a) on a representative sample of single women with dependent children in the United States.
The tax parameters are calculated using the Tax Simulation Model (TAXSIM) of the National
Bureau of Economic Research (NBER). These tax rates include the federal, state and payroll
tax components but do not include the transfer component.
In the United States, lower-income families are eligible to receive cash assistance from the
TemporaryAssistancetoNeedyFamilies(TANF) program, previously Aid to Families with
Dependent Children (AFDC). In addition, families may be eligible for in-kind benets in the
form of food vouchers (Food Stamps) and health insurance (Medicaid). Since most single
mothers are eligible for these types of programs, it is important to incorporate the benetside
in the calculation of eective tax rates. Accordingly, we develop a benet calculator accounting
for the dierences in eligibility and benet structures at the federal and state levels.
Figure 6a,b display tax rates along the intensive and extensive margins in the four pre-
reform years. The graphs illustrate the dramatic changes in the taxation of single mothers over
the period. Most notable is the decline in the eective tax rate on labor market participation
going from 57 percent in 1986 to 32 percent in 2000. The marginal tax rate shows a more
moderate and less systematic decline than the participation tax rate, which is not surprising
given the large expansions of the EITC during this period. As discussed above, an EITC
reduces unambiguously the tax burden, whereas its impact on the marginal tax rate depends
on the distribution of individuals on the dierent income intervals of the EITC (phase-in, at
and phase-out). The two graphs also highlight the importance of accounting for the benet
system when calculating the relevant tax wedges along the two margins. Excluding benets
would seriously underestimate tax wedges. For example, the mean of the participation tax
rates would be reduced with more than 35 percentage points if we did not account for the
benet system.
Figure 6a: Marginal tax rate
1986 1990 1993 2001
Reform year
Source: Eissa
et al.
Excl. benefits
Incl. benefits
Figure 6b: Participation tax rate
1986 1990 1993 2001
Reform year
Source: Eissa
et al.
Excl. benefits
Incl. benefits
In Figure 6a,b, the changes in tax rates across the dierent pre-reform years reect all
changes in the tax and welfare system taking place at the federal and state levels over the
period, and they incorporate as well any behavioral responses to these tax and benet changes
along with macro/time eects on income and demographic variables. To isolate the direct
impact of federal tax reform, we calculate the changes in the tax rates that are attributed to
the reforms alone. The results are illustrated in Figure 7 which conrms that the decrease in
eective tax rates over the 15-year period has been driven to a large extent by tax changes at
the federal level. This was particularly the case for the 1986 and 1993 reforms, which reduced
the participation tax rate by 8 and 13 percentage points, respectively.
Figure 7: Reform-induced reductions in tax rates
1986 1990 1993 2001
Reform year
Source: Eissa
et al.
Percentage points
Marginal tax
Participation tax
There is a substantial variation in wage income and tax rates across the single mothers
in the data. This heterogeneity highlights the need for using micro-simu lations to evaluate
the tax reforms. Large errors may occur in more aggregate studies because of the correlation
between earnings, tax rates and tax changes. In Eissa, Kleven and Kreiner (2004), we carry
out a number of simulations assuming dierent elasticity scenarios. The simulations point to
substantial welfare gains for all the reforms. In particular, this is the case for the tax reform act
of 1986. In our baseline scenario with a participation elasticity equal to 0.4 and a (compensated)
hours-of-work elasticity equal to 0.1, we obtain a welfare gain for this reform equal to 7.3 percent
of labor income.
The welfare eects are also high for the 1993 reform (2.3 percent of labor
income), which contained the single largest expansion of the EITC. For the 1990 and 2001
reforms, the welfare gains are not quite as large.
For all four reforms, most of the total welfare gain is generated on the extensive margin.
While 3/4 of the gain from the 1986 tax act is created by labor market entry, essentially all of the
positive eect from the 1990 reform is driven b y labor force participation. For this reform, the
intensive welfare eect is around zero because negative eects in the phase-out region cancel out
positive eects in the phase-in region of the EITC. For the 1993 reform, the large welfare gain
is a result of the extensive margin strongly dominating welfare losses created on the intensive
margin. Finally, for the recently enacted 2001 reform, the dierence between the intensive and
extensive welfare eects is less pronounced. This occurs for two reasons: First, the 2001 tax
cuts reduced participation tax rates only slightly and, second, by year 2000 the previous reforms
had already eliminated much of the ineciency along the extensive margin.
Our nding that extensive responses drive almost all of the welfare eects created by the
four reforms underpins the importance of accounting for this margin of labor supply. A simu-
lation based on the traditional, convex labor supply model with only intensive responses would
seriously underestimate the welfare eects. For the 1993 reform, even the sign would be wrong.
Here, the intensive welfare eect in our baseline scenario equals 0.38, whereas the welfare
Fo r this reform, it is inte resting to note that a total labor supply elasticity of 0.6 certainly not out of boun d
of the em pirical estimates would imply Laer curve eects. In this scenario, the large tax reductions granted
to single mothers are recouped entirely from the labor supply responses created by the reform.
One might argue that our ndings regarding the relative sizes of the extensive and intensive welfare eec ts
were to be expected under the assumed elasticity scenario. Notice, however, that the dierence b etween extensive
and intensive welfare eec ts cannot b e explained exclusively by elasticities, s ince the two kinds of welfare eects
are related in dierent ways to the tax-transfer system. For example, for the tax act of 1990, increasing the
hours-of-work elasticity would leave the intensive welfare eect m ore or less unch anged since the losses in the
phase-out region would continue to cancel out the gains created in the phase-in region. For the 1993 reform,
increasing the intensive elasticity would simp ly e xacerbate the welfare loss along that margin, th ereby reinforcing
the point regarding the dierence of welfare eects along the two margins.
eect equals 2.30 when including the extensive welfare eect. We ma y instead try to estimate
the welfare eect from the traditional, convex model by substituting the aggregate labor supply
elasticity for the hours-of-work elasticity, i.e., apply expression (7). This gives a welfare esti-
mate of 1.88 which should be contrasted with the 2.30 obtained when accounting properly for
the extensive welfa re eect. These numbers demonstrate clearly the large errors that may arise
from basing the evaluating of tax reforms on the traditional, convex labor supply model.
The errors reect that the two margins of labor supply response depend on taxes and
transfers in dierent ways. While the intensive margin depends on the eective marginal tax
rate, the extensive margin is related to the average tax rate and the benet rate. By implication,
the size of the error depends crucially on the properties of the tax-transfer programs being
5 T h e optimal des ign of redistrib utiona l policie s
The proper amount of redistribution and the design of transfer programs is an important and
controversial issue in the political sphere. As is well known from the theory of optimal income
taxation, redistribution gives rise to a trade-o betweenequityandeciency. Redistribution
from middle and high incomes to low incomes is desirable for equity reasons. On the other hand,
redistributive programs tend to reduce labor supply incentives, thereby creating eciency costs.
In this section, we rst discuss the implications of the recen t evidence on labor supply
behavior for the theory of optimal income taxation. Secondly, we analyse empirically the trade-
o between equity and eciency in redistribution policy across dieren t EU coun tries. Thirdly,
we discuss some implications for the tax treatment of married couples.
5.1 The optim al income tax
Following the seminal con tribution by James A. Mirrlees (1971) most of the literature exploring
the optimal income tax structure have applied the standard, convex labor supply model. One
of the lessons from this literature is that marginal tax rates should be positive at all earnings
By implication, a tax-transfer system involving an EITC creating negative tax rates at
the bottom of the earnings distribution is never optimal. In a recent interesting contribution,
Kleven and Kreiner (2004a) provides a short rev iew of the standard results in the theory of optimal income
Saez (2002) shows theoretically that this result breaks down if a ll (or m ost) of the variation
in labor supply occurs along the extensive margin rather than on the intensive margin. His
calibration of a general discrete model featuring both intensive and extensive responses indicates
that an EITC is in fact optimal for the United States.
It is possible to understand the gist of these results based on simple heuristic argumen ts. Let
us start by discussing the original Mirrlees result based on the convex labor supply model. The
theory takes as its point of departure that taxes and transfers must be observed income abilities
are private information and unobservable for the tax authorities. It is further assumed that
society puts a higher value on the marginal consumption of those with low abilities/incomes than
on the marginal consumption of the well-o. The society therefore wishes to have a relatively
high taxation at the top of the earnings distribution and low net taxes (maybe even negative)
at the bottom of the distribution. The dra wback of this redistribution policy is that it involves
positive marginal tax rates, leading to lower hours worked and generating an eciency loss (cf.
Section 3.1). If the hours-of-work elasticity is large, the eciency cost will be high which, ceteris
paribus, calls for less redistribution and lower marginal tax rates. But the eect can nev er be
so strong that it calls for negative tax rates in some range. To see why this is so, consider for a
moment a negative tax rate in some small earnings range [Y
+ Y ]. The negativ e marginal
tax rate implies that an individual with income Y
+ Y pays less in taxes than an individual
with income Y
. Suppose now that a reven ue-neutral tax reform increases the marginal tax rate
a little bit in the interval [Y
+ Y ] while leaving marginal tax rates unchanged everywhere
else. This small marginal tax rate increase allows the government to raise more money from all
individuals with income above Y
+ Y . If the proceeds are then distributed uniformly to all
individuals in the population, the reform clearly improves equity. But what about eciency?
Recall from Section 3.1 that the eect on economic eciency is determined by the change in the
marginal tax rate and that the eectisgivenbythebehavioraleect on government revenue.
Only individuals in the interval [Y
+ Y ] experience a change in the marginal tax rate.
These individuals face a higher marginal tax rate and respond by reducing labor supply. This
would normally reduce eciency. How ever, since the marginal tax rate is initially negative in
this interval, the beha vioral response actually generates a positive eect on government revenue
and thereb y an eciency gain. To conclude, if marginal tax rates are negativ e then it is possible
to increase the tax rate and thereby improve both equity and eciency, thus contradicting that
negative tax rates can be part of an optimal policy.
In the standard theory described above, labor supply responds only along the intensive
margin: hours of work change a little bit when the marginal tax rate is changed a little bit.
This stands in contrast to the labor supply evidence reviewed in Section 2.1. In particular,
this is the case at the bottom of the income distribution where the variation in labor supply
is dominated by entry/exit beha vior. Saez (2002) is the rst paper to incorporate entry/exit
decisions into the theory of optimal income taxation. Below we explain his main result, namely
that negative marginal tax rates may be optimal at the bottom of the income distribution.
Consider for a moment the opposite case where the marginal tax rate is positive at the
bottom of the earnings distribution. Individuals outside the labor market pays T (0) in net-
taxes (presumably a negative number due to out-of-work transfers), while working individuals
with very low earnings pay T (Y
) >T(0). Suppose now that a reform increases the net-
tax payment of all individuals in society by a small xed amount and giv es the proceeds to
individuals at Y
. This reduces the tax payment for the low-skilled, while the tax payment
for everybody else goes up. If the social welfare weight of the low-skilled individuals at Y
higher than the average welfare weight of all individuals in the population, this redistributional
scheme increases equity. In addition, the lower tax payment at the earnings level Y
to the zero-earnings situation induces some low-skilled unemployed to enter the labor force at
.SinceT (Y
) >T(0), this beha v ioral response raises government revenue and thereby
economic eciency.
The suggested reform therefore improves both equity and eciency. As
a consequence, it cannot be optimal to have a positive marginal tax rate at the bottom of the
earnings distribution. Instead, the tax-w elfare system should subsidize labor market entry for
low-skilled individuals.
5.2 The trade-o between eciency and equalit y in European coun tries
The traditional theory of optimal income taxation emphasized a simple trade-o between the
distribution of income/welfare and the size of aggregate income/welfare. The more we redis-
tribute from rich to poor, the higher are marginal tax rates, and the lower is labor supply,
Notice that the economic incentive to enter/exit is unchanged for individuals with potential earnings above
. The reform has increased their tax payment if they work, but the tax payment for those outside the labor
market is increased by the same am ount. He nce, the dierence in tax payment between working and n ot working
is not changed for individuals with potential earnings above Y
employment and the size of aggregate income. If hours-of-work elasticities are large, this trade-
o tends to be very unfavorable. Following this literature, the old debate on redistribution was
a classical left-right debate, with the left arguing that redistribution is desirable and the right
arguing that labor supply responses are large. By contrast, following Saez (2002) and others,
the new debate on redistributional policies focuses to a smaller extent on the size of welfare
programs and to a larger extent on the shape of tax-transfer programs and the incentive they
create in the decision to enter or exit the labor market. The new debate asks if it is desir-
able to increase the incentives to work at the bottom by redistributing from the middle- and
high-income earners to the working poor, rather than to non-workers as in the old debate.
In Immervoll et al. (2004) we try to cast light on the welfare reform debates, both the
old debate on traditional welfare programs and the new debate on redistribution towards the
working poor. We construct a simple and fully explicit model encompassing labor supply
responses along both the intensive and extensive margins and we then apply the model to
the analysis of welfare reform for 15 European Union countries using the EUR OMOD micro-
simulation model that has recently become available.
The EUROMOD micro-simulation model combines a tax and benets calculator with de-
tailed country-specic, but partly harmonised, micro data on income, earnings, labor force
participation, as well as many demographic variables. For any set of household characteristics
and country, EUROMOD is able to calculate the amount of benets the household is entitled
to and the taxes it should pay. EUROMOD has been constructed to incorporate all relevant
tax and transfer programs in place in all countries that were members of the European Union
prior to May 1, 2004. It is therefore a unique tool to obtain a complete picture of the incentives
to work generated by these programs as well as the analysis of welfare reform.
We consider two types of redistributional reform: a traditional welfare policy with univ ersal
transfers and a working poor policy with transfers targeted to the working poor. The traditional
welfare policy raises the tax rate on all units of income by 1 percentage point and returns the
collected revenue as a lump sum to all individuals in the economy. The concequence on labor
supply incentives is illustrated in Figure 8. The linear dashed curve represents the relationship
between disposable income and earnings prior to the reform, while the solid curve illustrates
Detailed information on EUROMOD is provided by S utherland (2001) and on the Inte rnet at
the relationship after the reform. The reform raises the tax burden on high-income individuals
and reduces the tax burden (net of transfers) on low-income workers as well as on non-workers.
The marginal tax rate is increased at all earninge levels, implying that workers reduce hours
worked (illustrated by the small arrows in Figure 8). Since transfers are universal, non-workers
gain more from the reform than w orkers do. The reform therefore increases the participation
tax rate, leading to a reduction in the population wanting to participate in the labor market
(illustrated by the large arrows in Figure 8).
Figure 8: Traditional redistribution policy
Disposable Income
The working poor policy diers from the traditional policy in one important aspect: the tax
proceeds now nance a transfer given only to those who work. The policy therefore improves
the standard-of-living for low-skilled individuals participating in the labor market compared to
those out of work. By implication, non-participating individuals are induced to enter the labor
market at low-earnings occupations.
The desirability of each type of welfare reform depends on several factors. In general, as
discussed extensively above, the eciency eects of tax-transfer reform depends on the size
of elasticities along eac h margin of labor supply response (hours-of-work and participation
elasticities), and on the size of tax rates along each margin (marginal and participation tax
rates). If eective marginal tax rates are initially high, the increase in taxes create large
distortions along the intensive margin. For both policies, this gives an unfavorable trade-o
betweenequityandeciency. By contrast, the tax on labor force participation aects the
impact of the t wo policies in opposite directions. The presence of high participation tax rates
makes the traditional welfare policy more costly, whereas the w orking poor policy tend to
become more desirable. The reason is that the traditional policy reduces participation, while
the working poor policy stimulates it. In particular, the working poor policy will be attractive
when participation tax rates are high at the bottom of the earnings distribution, because this
is where the additional entry takes place.
In addition to elasticities and tax rates, the distribution of earnings in the population will
be important. For example, if the degree of earnings equality is already very high, the gain
of additional redistribution will be relatively low. To see this, recall that tax and transfer
payments for each individual are functions of earned income. If the income distribution is
strongly compressed, the taxes paid and the benets received tend to be of a relatively similar
size for most individuals in the population. Hence, a reform that increases tax rates and benets
generates relatively little redistribution per dollar of deadweight burden, thereby creating an
unfavorable equity-eciency trade-o. Indeed, the presence of a high earnings equality in
Scandinavian countries is often mentioned as an argument against introducing working poor
policies in this region. It is important to notice, however, that high income inequality tends to
make both the working poor policy and the traditional welfare policy costly. If the presence
of a high earnings equality is used as an argument against w orking poor policies, it may also
be used as an argument for rolling back traditional welfare programs targeted at those out of
work. Still, the eect is somewhat less pronounced for the traditional welfare policy because it
also gives transfers to those with zero earned income. We will come back to the implications of
a high earnings equality for the design of welfare reform below.
The EUROMOD calculations in Immervoll et al. (2004) provide detailed information on the
distribution of earnings along with eective tax-benet rates applying on the two margins of
labor supply for the pre-2004-enlargement countries of the European Union. In Table I, we
briey summarize some of the results from these calculations.
[Table I]
The upper panel in the table ranks countries according to the equality of pre-tax earnings
as measured by income at the 80th percentile relativetoincomeatthe20thpercentileinthe
distribution. As one would expect, Nordic countries have the highest equality of earnings while
Anglo-saxon countries have the lowest. In the United Kingdom, earnings at the 80th percentile
is more than 3 times as large as earnings at the 20th percentile, while the ratio in Sweden
is only around 1.5. Moreover, the countries with a high degree of pre-tax earnings equality
tend to be those with very high tax rates.
Marginal tax rates tend to be highest in the
Nordic countries, somewhat lower in central/continent al Europe,
and lowest in Anglo-saxon
and Southern European countries. The variation in tax rates is substantial. At the low end, we
have Spain where the a verage eective marginal tax rate is only 36 percent. At the other end,
Belgium has a tax rate that is close to 70 percent.
The Nordic countries also have some of the highest tax burdens on labor force participation
for the low-skilled. In countries such as Denmark and Sweden, participation tax rates are largest
at the bottom because the implicit tax on working created by generous earnings- and work-
tested benets weigh more heavily on low-income people. Moreover, unemployment insurance
benets can be subject to a oor so that replacement rates for low-wage earners can be very
high in some cases. By contrast, countries such as Greece, Luxembourg, Spain, and the United
Kingdom have relatively lower tax rates at the bottom because minimum income programs do
and/or because they operate in-work benetswhichpartlyoset the loss of social assist ance
or unemployment benets at the point of entry. Notice, nally, that Ireland and France have
relatively high participation tax rates compared to their marginal tax rates.
Table II shows estimates of the equity-eciency trade-o for the two kinds of welfare policy.
We dene this trade-o as the Euro value of the welfare loss for those who lose from the reform
(the ric h) in proportion to the Euro value of the welfare gain for those who gain (the poor or
working poor). In other words, the trade-o giv es the welfare cost to the rich from the transfer of
one additional dollar of welfare to the poor. If a given redistributional reform reduces economic
This is contrary to the p olicy recommen dation in the optimal tax literature. In this literature, a high degree
of equality in abilities/productivities tend to make redistribution more costly, so that optimal marginal tax rates
are lower. Of c ourse, the pre-tax earnings distribution is in itself aected by the tax system , and does not
measure the ability distribution relevant for optimal taxation. However, the presence of high marginal tax rates
would normally lead to labor supply responses making the pre-tax earnings distribution (with the tax system in
place) more unequ al than the ability distribution without a tax system. Hence, this point serves to reinforce the
contradiction between optimal and practical policy. Of course, this gap between optimality a nd practice m ight
b e explained by o ther dierences across countries such as diere nt socia l preferences for distributional eq uity.
It sh ou ld also be ke pt in m ind that the optimal tax model is not really a model to explain actual govern m ent
b ehavior, but one to analyze to best possible behavior under a given set of gove rnm ent preferences.
Luxembourg is an exception. As other smaller and very wealthy European countries or principalities such
as Liechtenstein or Switzerland, tax rates are signicantly lower in Luxembourg than in other larger continental
European countries.
eciency, the welfare cost on the rich from transferring an additional Euro to the poor will
be larger than one Euro. That is, in the process of redistributing income from rich to poor
something is lost due to behavioral responses, giving rise to a trade-o larger than one.
benchmark case where the trade-o is equal to 1, so that the welfare gain of the poor is exactly
equal to the welfare loss of the rich, reects the situation where a reforms involves no eciency
cost (or gain). Finally, if the trade-o were to be less than one, there is no conict between
distributional equity and economic eciency for the reform under consideration. The trade-o
measure used here was originally proposed by Browning and Johnson (1984).
Redistributing income to the poor by increasing the level of universal transfers leads to
eciency losses in all countries, implying a trade-o larger than one. For most European
countries, expanding the generosity of traditional welfare programs creates large eciency costs:
for 11 out of 15 countries, redistributing one additional Euro to low-income individuals requires
a reduction in the welfare of high-income individuals by 2 Euros or more. This is due to the fact
that most European countries already impose quite large marginal tax rates on earnings. The
largest trade-os are found in the Nordic and Continental European countries where taxes and
benets are high. In fact, the 6 countries with the highest trade-o between equity and eciency
are identical to the countries having the highest marginal tax rates in Table I. Likewise, the
lowest trade-os are associated with the countries with the lowest marginal tax rates.
[Table II]
A completely dierent picture emerges once we turn to the working poor policy. For all
countries, the equity-eciency trade-o is substantially lower for this type of reform. In fact,
for Denmark, France, Ireland, Portugal, and Spain, the policy would create an aggregate welfare
gain such that the trade-o is lower than 1. Hence, in these countries it would cost less than 1
Euro for the rich to transfer 1 Euro to the working poor. In the case of Denmark, the trade-o
implying the the reform is (almost) Pareto improving. For a number of other countries, the
working poor policy creates very small eciency losses such that the trade-o is quite close
to one. This applies to countries such as Austria, Greece, Luxembourg, the Netherlands, and
Okun (1975) used the metaphor of a leaky b uc ke t to illustrate the ecienly loss taking place in income
the United Kingdom. In these countries there is no signicant trade-o between eciency and
equality when we consider redistribution from the rich to the working poor. Only in the case
of Finland and Sweden does the working poor policy involve an unfav orable equity-eciency
trade-o,whichreects in part the extremely equal earnings distribution in these two countries.
A comparison of Table I and Table II reveals that countries with relatively high participation
tax rates at the bottom of the earnings distribution, such as Denmark, Ireland and France, tend
to gain more by choosing a working poor policy rather than a traditional welfare policy. The
working poor policy creates higher incentiv es for participation in the labor force. Moreover,
participation rises mainly at the bottom deciles where participation elasticities are large. If
participation tax rates are very large at the bottom deciles, the increase in labor market partic-
ipation creates a large increase in government revenue (through reduced benet expenditures
and a higher tax take) and hence in economic eciency.
Ceteris paribus, the presence of a highly equal pre-tax earnings distribution tends to mak e
redistribution very costly. This applies to redistributional policies in general, and to the working
poor policy in particular because it attempts to redistribute only within the group of workers
who are more equal than the entire population of workers and non-workers. The presense of
strongly compressed wage distributions in Scandinavian countries (cf. Table I) are often men-
tioned as an argument against pursuing in-work benet reform in this region. It is therefore
interesting that the working poor policy seems to be extremely desirable in Denmark, more
so than in any other country. The explanation is that participation tax rates are very high
in Denmark, especially in the bottom deciles where participation elasticities are also high. In
the simulations, this eect strongly dominates the eect of a compressed earnings distribution.
On the other hand, the working poor policy is less desirable in Finland and Sweden, because
participation tax rates at the bottom are somewhat lower there and because the earnings dis-
tribution is even more equal than in Denmark. It is important to realize, however, that the
equit y-eciency trade-o is even worse for the traditional welfare policy in Finland and Sweden.
The bottom line is that both kinds of redistribution are very costly in these two countries, at
least on the margin, due to the combination of an already equal distribution and high tax rates.
Besides the presence of an equal earnings distribution, a potential argument against working
poor policies in Scandinavia is that participation elasticities might be signicantly lower in
this region than in, say, Anglo-Saxon coun tries. It is true that most of the estimates of high
participation elasticities were based on the United States and the United Kingdom where female
labor force participation rates have been lagging behind those of Scandinavia, creating more
room to increase participation through in-work benets targeted at low -income fem ales. This
argument is potentially important and points to the need for good empirical studies on the
responsiveness of labor market participation. Still, there are several counter-arguments that one
can make at this point. Firstly, as mentioned in Section 2.1, there are some empirical studies
suggesting that female participation elasticities may be high even in Nothern European countries
(Germany, Sweden, and the Netherlands). Secondly, while female labor force participation is
relatively high in Scandinavia, there is in fact substantial non-participation among a number
of demographic subgroups (e.g. immigrants, single mothers, and the young). In Denmark,
around 1/4 of those in the working age (15-64) is outside the labor market. Thirdly, and
perhaps most importantly, ev en if participation elasticities were to be substantially lower in
Scandinavian countries, a working poor policy may still be desirable due to the fact that the
existing welfare system is a lot more distortionary. This argument is very important for Denmark
where participation tax rates, in an international perspective, are extremely high at the bottom
of the income distribution. In Immervoll et al. (2005), we nd that the equit y-eciency trade-
o fortheworkingpoorpolicyisexactlyequalto1inDenmark—implyingnoeciency loss
for this type of redistribution at an average participation elasticity for the population equal
to 0.13, a number far below the existing in ternational estimates.
To conclude this section, expanding traditional welfare programs have very dierent impli-
cations than introducing in-work benets. Increasing redistribution through traditional welfare
can lead to signicant negative labor supply responses along both the intensive and the ex-
tensive margins. By contrast, in-work benet reform generates positive labor supply responses
along the extensive margin (along with a negative response along the intensive margin). As a
result, the eciency cost of redistribution through this t ype of reform is much smaller.
It need s to be em ph asized, however, that the groups who b ene t from redistribution in the two reforms are
dierent. In the traditional welfare case, those who b e ne t the m ost are people w ith no earnings at all, presumably
those who are the most in need of support. In the in-work benet case, people with no earnings receive no
additional support and redistribution benets only the working poor. Finally, notice that the implementation of
one p olicy does not exclude the other p o licy. Hence, the two policies s hould be seen as complements rather than
substitutes in t he design of redistribution p olicy.
6 Discussion
In this survey, we discussed some recent research on the eects and design of redistributional
policies, emphasizing the distortionary eects of taxes and transfers on labor supply behavior,
employmen t and economic eciency. We paid special attention to the modern labor market
literature showing that labor force participation is more responsive to taxes and transfers than
is hours worked. It was shown that the presence of signicant participation responses tend
to make in-work benet reform inducing individuals from welfare into work a desirable policy
option. By contrast, traditional welfare policies providing either univ ersal or earnings-tested
benets are very costly in terms of government rev enue and eciency resulting from adverse
eects on labor force participation. This point is reinforced by the fact that earnings- and
work-tested benets generate the highest eective tax rates on labor force particiaption at the
bottom of the earnings distribution where participation elasticities are particularly high.
It needs to be emphasized, however, that the individuals who gain from in-work benet
reform are dierent from those gaining from traditional welfare policies. In the latter case,
those who benet the most are people who are not working and have zero earned income,
presumably those in the strongest needofsupport. Within-workbenet policies, people with
no earnings receive no additional support and redistribution helps only the working poor. As
a result, if the government has extremely strong redistributive tastes and put a much higher
welfare weight on those with no earnings than on the working poor (such as in the case of
a Rawlsian wel fare criterion), it is possible that expanding traditional welfare would be more
desirable than introducing in-w ork benets. By contrast, if the go vernment puts lower welfare
weights on those with no earnings than on the working poor, the case for in-work benets would
be ev en stronger. It seems that conservativ e governments (especially in Anglo-Saxon societies)
tend to hold the latter view: those not working are seen as "lazy", whereas the working poor
are seen as "deserving".
While the implementation of benets targeted to the working poor was pioneered by the
United States (starting in the 1970s), in-work benet reform has recently been high on the
agenda in political discussions of welfare reform in Europe. The rst European countries to
introduce in-work benet programs were the United Kingdom (in 1988, subsequently expanded
in 1999) and Ireland (in the early 1990s). In the past 10 years, a n umber of other European
countries have introduced programs of a similar kind. France introduced an in-work benet
program as of 2001 (“Prime Pour l’Emploi” or premium for employment). The Netherlands
introduced an Employment Tax Credit in 2001, while Belgium has been phasing-in an Earned
Income Tax Credit program from 2002 to 2004. Germany introduced the so-called “Mainzer
Modell” program in 2002 which, more recently, has been replaced by SSC reductions available
to low-wage earners. Finland has recently introduced and then expanded an Earned Income
Tax Allowance, and Denmark introduced a similar type of program on a small scale in 2004.
Finally, while the Italian family benet has not normally been considered an in-work benet,
it increases with the number of days worked. However, with the exception of Ireland and
the United Kingdom, the European in-work benet programs are still modest in size with
maximum annual benets between 300 and 1000 Euros (see OECD, 2004, for the most recen t
and systematic description of these programs). Therefore, the "small reform" methodology
adopted in this paper would seem to provide a good appro x imation of the eect of introducing
these programs. Our results lends support to the conjecture that these policies constitute a
very ecient type of redistribution, and that the expansion of such policies should perhaps be
part of future welfare reform.
The simple labor supply model underlying the analysis in this paper abstracts from a number
of issues which we would like to discuss briey. First, and perhaps most importantly, we have
assumed that the labor market is perfectly competitive. This might be a poor approximation
to European labor markets, where minimum wages tend to be substantial, and where wage
rates are often the result of bargaining between unions and employers. Minimum wages prevent
employers from paying wages which are below a dened minimum, thereby eliminating jobs
with very low productivities and potent ially creating in voluntary unemployment among the
low-skilled. Likewise, union bargaining models, eciency wage models, and search models
imply that a fraction of individuals become in voluntary unemployed.
The eects of taxation in imperfect labor markets have been explored in a number of papers
(see, e.g., Hansen et al., 1996). The in troduction of imperfections will not change the most
important mechanisms at work in our analysis. Firstly, variation in aggregate employment
is still the result of behavioral responses along the intensive and the extensive margins. For
example, Sørensen (1999) considers optimal taxation in three dierent models of involuntary
employment (unions, eciency wages, and search) where both intensive and extensive responses
are present. Secondly, in all imperfect labor market models, a reduction of a verage tax rates
leads to higher employment, where the eect is channelled through lower equilibrium wages.
Accordingly, a working poor