A joint initiative of Ludwig-Maximilians University’s Center for Economic Studies and the Ifo Institute

CESifo GmbH · Poschingerstr. 5 · 81679 Munich, GermanyTel.: +49 (0) 89 92 24 - 14 10 · Fax: +49 (0) 89 92 24 - 14 09E-mail: office@CESifo.de · www.CESifo.org

CE

Sifo

Co

nfer

ence

Cen

tre,

Mun

ich

Area Conferences 2013

CESifo Area Conference on

Public Sector Economics11–13 April

Beyond the Labor Income Tax Wedge: The Unemployment-Reducing Effect of Tax Progressivity

Etienne Lehmann, Claudio Lucifora, Simone Moriconi and Bruno van der Linden

Beyond the Labor Income Tax Wedge: The

Unemployment-Reducing Effect of Tax Progressivity∗

Preliminary version - Please, do not quote without authors’

permission

Etienne LEHMANN †

ERMES (TEPP) University Pantheon-Assas Paris 2 and CREST

Claudio LUCIFORA‡

Universita Cattolica del Sacro Cuore and IZA

Simone MORICONI§

Universita Cattolica del Sacro Cuore, Milano

Bruno VAN DER LINDEN¶

FNRS and IRES Universite Catholique de Louvain

March 26, 2013

Abstract

This paper argues that, for a given overall level of labor income taxation, a more progres-sive tax schedule reduces the unemployment rate and increases the employment rate. Wefirst review some theoretical arguments: progressivity increases employment because em-ployment is much more responsive for low-paid workers and because progressivity induces awage-moderation effect. We then test our arguments on a panel of 21 OECD countries over1998-2008. We empirically capture progressivity by the change in the burden of taxationbetween 67% and 167% of the average tax rate. Our empirical findings address potential en-dogeneity issue and confirm that controlling for the burden of taxation at the average wage,a more progressive taxation reduces the unemployment rate and increases the employmentrates.

Keywords:

JEL code:

∗We wish to thank participants at a seminar at CREST, with a particular mention to Pierre Cahuc, FrancoisFontaine and Jean-Baptiste Michau for their valuable comments and suggestions. Usual caveats apply.†Email: etienne.lehmann@ensae.fr. Address: CREST, Timbre J360, 15 boulevard Gabriel Peri, 92245,

Malakoff Cedex, France. Etienne Lehmann is also research fellow at IRES-Universite Catholique de Louvain,IDEP, IZA and CESifo. Etienne Lehmann is also a junior member of IUF which is gratefully acknowledged forits financial support.‡Email: claudio.lucifora@unicatt.it. Address: Department of Economics - Universita Cattolica 1, Largo

Gemelli, 1 - 20123 Milano, Italy.§Email: simone.moriconi@unicatt.it. Adress: Department of Economics - Universita Cattolica 1, Largo Gemelli

- 20 123 Milano, Italy. Simone Moriconi is also associated to CREA, University of Luxembourg.¶Email: bruno.vanderlinden@uclouvain.be. Address IRES, Universite Catholique de Louvain, Place Mon-

tesquieu 3, B1348, Louvain-la-Neuve, Belgium. Bruno Van der Linden is also research fellow at IZA, CESifo andexternal member of ERMES - Universite Paris 2.

1

I Introduction

There is a general consensus among economists that higher labor income taxation has a detri-

mental effect on unemployment. Empirical studies have shown the existence of a positive rela-

tionship between tax wedge on labor income and equilibrium unemployment1. Income taxation,

however, is a rather complex policy with multiple dimensions that are only partially accounted

by the average tax wedge on labor income. Many OECD countries, after the Great Crisis, have

experienced higher unemployment and larger public debt, being confronted with the need to

combine fiscal consolidation measures with policies to curb unemployment. For this reason, a

better understanding of the effects of labor income taxation on the functioning of the labor mar-

ket seems important. In particular, we need a better description on how the different dimensions

of labor income taxation affect labor market performance and how to limit their adverse effects

on unemployment. This paper proposes a first step in this direction by investigating the effect

of both average labor income taxation, and tax progressivity on unemployment. It argues that,

for a given aggregate level of labor income taxation, a more progressive tax schedule can reduce

unemployment.

We first review theoretical arguments in support of the unemployment-reducing effect of tax

progressivity. The first mechanism is a composition effect that occurs because the labor demand

for low-skilled workers is more sensitive to taxation as compared to high-skilled workers, also

gross wage for low-skilled workers is deemed to be more rigid than that of high-skilled workers.

The second mechanism is a more traditional wage moderation effect. Since a more progressive

tax schedule makes any increase in the after-tax wage more costly for employers, the latter will

resist more any wage increase claimed by employees, thus moderating the adverse wage effect

on labor demand.2 We also argue that these unemployment-reducing effects of tax progressivity

are very likely to remain in the presence of labor supply responses. Although tax progressivity

may be detrimental to labor supply along the intensive margin, as typically argued in public

finance, tax progressivity remains beneficial for employment (and unemployment), with only an

ambiguous total effect on aggregate production. Next, we empirically investigate the effects of

taxation on labour market performance using a panel data of 21 OECD countries over 1998-

2008. We derive our tax indicators for the total tax wedge - for workers at the average wage -,

and tax progressivity - measured comparing the fiscal wedge of workers at 67% and at 167% of

the average wage -, from the OECD tax database. In particular, we exploit differences in output

shocks and labour market institutions, across countries and over time, to assess the overall effects

of fiscal policies on unemployment.

Our paper contributes to the well-established literature that uses panel data to explain un-

employment patters across OECD countries.3 Empirical studies typically investigate the impact

1e.g. Layard and Nickell (1999), Daveri and Tabellini (2000), Nickell et al. (2005) among others.2See e.g. Hersoug (1984), Pissarides (1985, 1998), Lockwood and Manning (1993).3See the time series analysis on different OECD countries by Bean Layard and Nickell (1986), Layard Nickell

and Jackman (1991) and Nickell and Layard (1999) and the panel data analyses by Blanchard and Wolfers (2000),Daveri and Tabelinni (2000), Belot and Van Ours (2004), Nickell Nunziata and Ochel (2005), Griffith et alii (2007),

2

of macroeconomics shocks and (time-varying) labor market institutions on aggregate unemploy-

ment. In this literature, labor income taxation is generally captured by a single indicator, the

tax wedge for the average worker, which is found to be a key determinant of the unemployment

rate (Daveri and Tabellini (2000), Nickell et alii (2005)).

Another empirical literature to which our paper is related explores the effect of tax progres-

sivity on wages using time series data, either at the aggregate level or at specific industry or

sub-groups of workers.4 The typical result is that tax progressivity reduces the gross wage, at

least for blue-collar workers.5 Note, however, that the change in aggregate wage may confound

true wage responses with heterogeneous employment responses that alter the wage distribu-

tion. Such composition effects therefore make the results hard to interpret in term of a wage

moderation effect. 6

Finally, there is an extensive literature that uses micro-data to evaluate the effect of tax

reforms and making-work-pay programs in selected countries. For example, policies such as the

EITC (Earned Income Tax Credit) in the US, or the WFTC (Working Families Tax Credit) in

UK7 have been shown to improve transitions from nonemployment to employment. As these

policies shifts the tax burden away from some disadvantaged groups of employed (e.g. employed

lone parents), we interpret them in a way that is consistent with an increase in progressivity.

One advantage of the literature that uses micro-data to evaluate the effects of programs tar-

geted to specific subgroups, is the possibility to exploit quasi-experimental methods to identify

their causal effect on labor market outcomes - e.g. on the extensive margin of labor supply.

However this also comes at a cost, since the focus on selected programs in a specific country has

implications for the external validity of the results and their relevance for other countries.

Our main results support the usual findings that higher labor taxation increases unemploy-

ment. Moreover, we show that controlling for tax progressivity the estimated impact of the tax

wedge on unemployment becomes larger. The novelty of our findings, however, is the sizable

unemployment-reducing effect of tax progressivity: for any given level of taxation, we find that

a more progressive tax schedule reduces unemployment. Although not typically done in the

panel-data literature, we also address the potential endogeneity of fiscal policy. Since the fac-

tors that drive changes in taxation are often correlated with other developments in the economy,

as well as with changes in unemployment, both reverse causality and omitted variable bias are

likely to affect the estimates. Our empirical strategy exploits exogenous variations in institu-

tional, social and political factors that influence a country’s fiscal policy, to identify the effect

Bassanini and Duval (2009) among others. See also Prescott (2004), Rogerson (2005, 2007) and Rogerson andWillenius (2009) who account for the different trends in total hours of works across OECD by difference in taxpolicy in calibrated macroeconomic models.

4Malcomson and Sartor (1987), Lockwood and Manning (1993), Holmlund and Kolm (1995), see also thesurveys by Sørensen (1997) and by Røed and Strom (2002).

5Hansen Pedersen and Sløk (2000).6Brunello and Sonedda (2009) use panel data to estimate the effect of tax progressivity on wages. Manning

(1993), uses quarterly time series for the UK, and estimates also an auxiliary unemployment equation augmentedwith a tax progressivity indicator.

7e.g. Eissa and Liebman (1996) and Blundell and Shepard (2012) among many others

3

of taxation on unemployment (and employment). In practice, we instrument the tax wedge and

tax progressivity indicators using the following variables: a narrative record for the tax com-

ponents of fiscal consolidation policies, an index of trust in civil services, and a measure of the

political orientation of the parliament. Results show that OLS tend to underestimate the effects

of taxation on unemployment. According to our preferred specification, a one percentage point

decrease in the average tax rate reduces unemployment by 0.95 percentage points. This has

to be complemented by the unemployment effect of a unit increase in tax progressivity8 which

decreases unemployment by 1.14 percentage points. We also find that employment rates9 tend

to decrease with labor income taxation and to increase with tax progressivity. In line with the

common wisdom in public finance which emphasizes the disincentive effects of tax progressivity

along the intensive margin of labor supply, we find that a more progressive tax schedule has

a negative impact on average production per worker. Hence, we argue that the design of an

optimal tax schedule should not simply focus on the tradeoff between the equity gain of a more

progressive tax schedule against the adverse effect on the incentives to work harder, but should

also include the employment-enhancing (unemployment-reducing) effect of tax progressivity (See

Hungerbuhler et alii (2006) and Lehmann et alii (2011)).

The paper is organized as follows. We review the theoretical arguments on the unemployment

effects of tax progressivity in Section II. The empirical strategy is outlined in Section III, where

we also discuss the validity of our instruments. The dataset is described in Section IV and the

empirical results are discussed in Section V. Finally, Section VI concludes.

II Theoretical framework

In this section, we use a matching model a la Diamond-Mortensen-Pissarides to derive theoretical

predictions about the effects of taxation on the steady-state unemployment rate. We start with

a model without labor supply responses and then introduce participation and in-work decisions

into the model.

II.1 The model without labor supply responses

Time is continuous and discounted at the common rate r > 0. All agents are risk-neutral. An

homogeneous consumption good is produced by different types of perfectly substitutable labor

indexed by the corresponding productivity level i. The different labor markets are perfectly

separated: a type i individual can only occupy a type i job.10 Let yi, Li, Ni, ui, δi and wi

denote respectively the productivity, the level of employment, the number of participants, the

unemployment rate, the exogenous destruction rate and the pre-taxed (gross) wage (or labor

8We define a unit increase in tax progressivity as the combination of a half percentage point decrease in thetax wedge for workers at 67% of the average wage with a half percentage point increase in the tax wedge forworkers at 167% of the average wage

9The employment rate is the share of the working age population that is employed. It is thus equal to theproduct of the participation rate to one minus the unemployment rate.

10These assumptions of perfect substitution and of perfect separation are only made to simplify the exposition.

4

cost) for individuals of type i, with obviously 1− ui = Li/Ni.

We assume congestion externalities in the matching process. Following Diamond (1982),

Mortensen and Pissarides (1999) and Pissarides (1985, 2000), we assume that the flow of hirings

is given by a matching function Mi(υi, Ni − Li) of the stock υi of vacant jobs of type i and

the stock Ni − Li of unemployed of type i. Function Mi(., .) is assumed to be increasing and

concave in each of its argument and to exhibit constant returns to scale. The rate at which jobs

are filled is a decreasing function qi(.) of the tightness ratio θi = υi/ (Ni − Li) in labor market

i, with q(θi) ≡ Mi(1, 1/θi). Symmetrically, the rate at which an unemployed finds a job is an

increasing function pi(.) of tightness θi, with pi(θi) ≡Mi(θi, 1) = θiqi(θi). The equality between

exitsMi(υi, Ni−Li) out of and entries δiLi into unemployment gives the employment level and

the unemployment rate at the steady state:

Li =pi (θi)

δi + pi (θi)Ni and ui =

δiδi + pi (θi)

(1)

On each labor market i, a filled job generates a profit flow yi−wi and it provides an expected

intertemporal profit denoted Ji:

Ji =yi − wir + δi

(2)

Firms create vacancies as long as the flow cost of search ci > 0 is smaller than the expected

returns to search. The latter is equal to the rate qi (θi) at which jobs are filled times the

intertemporal value Ji of a filled job. As firms create more and more vacancies, tightness

increases and the job filling rate decreases. This occurs until the following free-entry condition

is met:ci

q (θi)= Ji =

yi − wir + δi

(3)

Combining (1) and free-entry condition (3) leads to labor demand in market i. The fraction

1 − ui of participants of type i who are employed is a decreasing function of the gross wage

wi.This relationship and its equivalent when wi is replaced by ni are denoted LD in Figure 1.

A rise in the gross wage reduces the intertemporal profit (2) made on a filled job. So, firms find

it less profitable to create jobs, tightness decreases and the unemployment rate increases.

We consider a nonlinear income tax function T (.) that only depends on wages w. The average

tax rate at wage wi is denoted τi ≡ T (wi)/wi and the net wage is ni ≡ (1− τi)wi. We call 1− τithe retention rate. A change in the average tax rate τi does not affect job creation if it holds

for a fixed gross wage wi. In the left part of Figure 1 where the gross wage wi is on the vertical

axis, a rise in the average tax rate τi does not shift the LD curve. Conversely, keeping the net

wage ni fixed, a rise in the average tax rate increases labor cost wi, thereby inducing firms to

create fewer jobs. In the right part of Figure 1 where the net wage ni is on the vertical axis, a

rise in the average tax rate τi shifts the LD curve inwards.

We denote bi the instantaneous value of being unemployed among type i individuals. This

value sums untaxed unemployment benefits and the value of time out of work. The expected

lifetime utility of a type i employed individual Ei, respectively an unemployed one Ui, verifies

5

wi WS

A

B

Decline in ψi

Rise in τi

θi

1-ui

LD

C

ni WS

Rise in τi

Decline in ψiA

θi

1-ui

LDRise in τi

B

C

Figure 1: The effects of taxation

the following asset equation:

r Ei = wi − T (wi) + δi (Ui − Ei) (4a)

r Ui = bi + pi (θi) (Ei − Ui) (4b)

The permanent income rEi of an employed worker of type i is equal to her instantaneous

utility wi − T (wi) minus the average loss δi (Ei − Ui) in case of job destruction. A symmetric

interpretation applies for unemployed workers. We assume Nash bargaining over the wage and

denote γi ∈ [0, 1) workers’ bargaining power. We show in Appendix A that:

Ei − Ui1− τi

=γi

1− γiΨi Ji (5)

where Ψi is the Coefficient of Residual Income Progression (hereafter the local CRIP) at wage

wi defined by Musgrave and Thin (1948) as:

Ψ(w) ≡ 1− T ′ (w)

1− T (w)w

= 1 +d ln

(1− T (w)

w

)d lnw

(6)

A one percent increase in the gross wage increases the net wage by a relative amount equals

to the CRIP. The latter is also equal to one plus the wage elasticity of the retention rate

1− (T (w)/w). A higher CRIP comes from a lower marginal tax rate T ′(wi) or a higher average

tax rate τi and is associated with less tax progressivity. With a lower CRIP, a marginal increase

of the gross wage remains as costly for the employer, but provides less additional income to the

worker. A more progressive tax schedule therefore affects the sharing rule (5) as does a decrease

in the workers’ bargaining power (Pissarides (1985, 1998), Lockwood and Manning (1993)). This

is the wage moderating effect of a more progressive tax schedule. We then obtain the following

gross and net wage setting equations, denoted WS in Figure 1 (See Appendix A):

wi =γi Ψi

1− γi + γi Ψi(yi + ci θi) +

1− γi1− γi + γi Ψi

bi1− τi

(7a)

ni =γi Ψi

1− γi + γi Ψi(1− τi) (yi + ci θi) +

1− γi1− γi + γi Ψi

bi (7b)

6

In the absence of taxation, Nash bargaining implies that workers extract a share γi of the total

surplus Ji+Ei−Ui while the firm extracts a share 1−γi. Therefore, the worker’s pay is a weighted

average of the instantaneous value in unemployment on the one hand and of the productivity

level yi augmented with the expected hiring cost per unemployed ci θi on the other.11 As the

latter increases when the labor marker becomes more tight, i.e. when the unemployment rate

decreases, the wage Equations (7a) and (7b) are represented by upward-sloping curves labeled

WS in Figure 1.

To analyze the impact of labor taxation on wage formation, we first consider a rise in the

average tax rate τi, holding tax progressivity Ψi constant. This reduces the total surplus of the

match that the worker and the employer share when they bargain over the wage. The surplus

Ji accruing to the firm decreases via a rise in the gross wage wi, which corresponds in the left

panel of Figure 1 to an upward shift of the WS curve (see (7a)). The surplus Ei − Ui accruing

to the worker shrinks via a reduction in the net wage ni, which corresponds in the right panel

of Figure 1 to a downward shift of the WS curve (see (7b)). Taking labor demand into account,

the equilibrium shifts from A to B, where employment and the net wage are lower while the

unemployment rate and the gross wage are higher. It is worth noting that if bi consisted only

of unemployment benefits indexed on the net wage, the tax rate would have no effect on the

unemployment rate nor on the gross wage (see e.g. Pissarides (1998, 2000)). This is because a

rise in the tax rate, by decreasing the net wage, decreases the level of unemployment benefits.

This effect turns out to offset the direct increasing impact of the tax rate on the gross wage.12

Put another way, the effect of the average tax rate τi relies on an imperfect indexation of

unemployment benefits plus the value of time out of work on the net wage.

Second, we consider a more progressive tax schedule (a lower Ψi) holding the average tax

rate τi unchanged. From (5), this affects the sharing rule as does a decrease in the relative

bargaining power γi/(1 − γi). A locally more progressive tax schedule therefore reduces the

workers’ effective bargaining power γi Ψi/(1− γi + γi Ψi), that is the weight of productivity in

the wage Equations (7a) and (7b). The WS wage curve shifts downwards in the two panels of

Figure 1, so the equilibrium moves from A to C. Employment increases while the unemployment

rate, the net and the gross wages decrease. This wage moderating effect of tax progressivity thus

provides a first argument justifying why we expect a more progressive tax schedule to decrease

the unemployment rate.

The local CRIP at at given wage w only captures progressivity in the neighborhood of w.

The following example illustrates the limitation of the local CRIP at w. Consider a piecewise-

linear tax system. Assume the only change from one year to the next is a small shift of a

threshold between two tax brackets from above to below a given wage level w (e.g. a “bracket

creep” phenomenon). Then, the marginal tax rate at w experiences a dramatic change, inducing

11For ci θi = ciυi/ (Ni − Li). When a job-seeker is hired, the employer saves the cost of searching for an otherapplicant.

12Plugging bi = ρ (1 − τi)wi into (7a), Equations (3) and (7a) would become a two-equation system in wi andθi that does not depend on τi.

7

a large shift of the CRIP at wage w, while the overall tax system has been almost unchanged.

We therefore choose to capture the progressivity of the tax system by a “global CRIP” that is

equal to the (log of) the ratio of the retention rates at wages levels w1 and w0 with w1 > w0.

The global CRIP between w0 and w1 is defined as:

Ψw1w0≡ ln

(1− T (w1)

w1

1− T (w0)w0

)=

∫ w1

w0

(Ψ(w)− 1)dw

w(8)

where the last equality is obtained by the integration of (6) between w0 and w1. The global

CRIP aggregates the local CRIPs between w0 and w1 and a lower global CRIP is associated

with a globally more progressive tax schedule between wages w0 and w1. The equality

n1

n0=w1

w0exp(Ψw1

w0)

stresses that the global CRIP measures how the tax system reduces after-tax wage inequality

between gross wage w0 and w1.

In addition to the wage moderating effect of a lower local CRIP, a lower global CRIP may

also reduce the aggregate unemployment rate if employment of low-paid workers at wage w0 is

more elastic to a change in the retention rate than employment for high-paid workers at wage

w1 > w0. The idea that low-paid employment is more responsive than high-paid employment

is quite common in the literature (e.g. Juhn et alii (1991), Kramarz and Philippon (2001),

Immervoll et alii (2007)). An increase in the log of the retention rate 1 − τ0 at w0 would

lead to an increase in low-paid employment that outweighs the decrease in employment induced

by an equivalent decrease in the log of the retention rate 1 − τ1 at wage w1. This leads to a

composition effect according to which a a lower global CRIP reduces unemployment. In addition

to this mechanism, there is no way to decrease the global CRIP between w0 and w1 without

reducing the local CRIPs in between, which also contributes to reduce unemployment trough

the wage moderating effect.

Prediction 1. In the model with exogenous labor supply, a rise in the retention rate (a decline

in the average tax rate) and a more progressive tax schedule (a decline in the global CRIP) reduce

the unemployment rate and increase the employment rate.

II.2 Extensive and intensive margins of labor supply

Extensive margin. The next step introduces participation decisions in each labor market.

For this purpose, we assume that individuals have different values I of being out of the labor

force. When an individual enters into the labor force, she is unemployed and searches for a job.

Among individuals of type i, only those for which I ≤ Ui choose to participate to the labor

market, where:

Ui =(r + δi)bi + pi(θi) nir (r + δi + pi(θi))

from (4a) and (4b). The value I of being out of the labor force is assumed continuously dis-

tributed among individuals of type i according to the CDF Gi(.). Let N i denote the exogenous

8

size of the working age population of type i. The participation rate in the population equals

Gi (Ui), the size of the labor force is Ni Gi (Ui).

From Equations (1) and (3), the fraction of employed participants 1 − ui is independent of

the size of the labor force. Because of congestion externalities, firms recruit workers more easily

when the size of the labor force increases. Whenever the participation rate increases holding

the gross wage wi constant, firms create more jobs until tightness, thereby the unemployment

rate, return to their initial values. The level Li of employment thus increases in the same

proportion as the size of the labor force. Consequently, total employment Ni Gi (Ui) (1− ui)and the employment rate Gi (Ui) (1− ui) are decreasing in the gross wage through labor demand

and are increasing in the net wage through the participation margin.

As displayed by Figure 1, a decline in the local CRIP Ψi, holding the average tax rate τi

constant, decreases the net wage ni but increases the exit rate out of unemployment pi (θi). The

total impact on the incentives to participate is therefore theoretically ambiguous. We show in

Appendix A that a decline in Ψi increases (decreases) the participation rate if the unemployment

rate is inefficiently high (low). This is because a decline in the local CRIP affects the labor market

outcome only by reducing the workers’ effective bargaining power. For a given retention rate τi,

participation is maximized whenever the effective bargaining power satisfies the Hosios (1990)

condition.13 However, the effect of tax progressivity on the unemployment rate is unchanged by

the introduction of the participation decision.

Figure 1 also shows that a rise in the tax rate τi, holding progressivity Ψi constant, decreases

the net wage ni, increases the gross wage wi and reduces the exit rate out of unemployment

pi (θi). Searching for a job thus becomes less interesting (i.e. Ui decreases), inducing pivotal

individuals to exit the labor force. The participation rate decreases. The employment rate and

the level of employment decrease because the size of the labor force is lower (a labor supply

effect) and a smaller fraction of participating agents is employed (a labor demand effect).

The empirical literature concludes that the extensive responses (i.e. participation decisions)

are concentrated on low-wage subgroups such as low-skilled workers or secondary earners (e.g.

Juhn et alii (1991), Immervoll et alii (2007) or Meghir and Phillips (2010)). This suggests

that increasing by 1% the retention rate 1− τi on the bottom half of the wage distribution and

decreasing the retention rate by 1% on the top half would increase overall participation through

a composition effect on participation. From (8), such a change in the profile of the retention

rate is associated with a decline in the global CRIP, that is, with a globally more progressive

tax schedule.

Prediction 2. In the model with bargained wages and endogenous participation, a rise in the

average tax rate increases the unemployment rate and decreases the employment rate. A rise

in tax progressivity (a decline in the CRIP) decreases the unemployment rate. It increases

(decreases) the participation rate if the unemployment rate is inefficiently high (low).

13This point was suggested by Pierre Cahuc.

9

Extensive and intensive margins in a competitive setting. The preceding conclusions

that tax progressivity reduces unemployment are in deep contrast with a long tradition in public

finance where tax progressivity reduces incentives to work. We argue that these two views are

not contradictory as they focus on different objectives, employment on the one hand, in-work

effort on the other hand.

The public finance literature typically assumes away frictions on the labor market, so full

employment prevails. However, even in such a frictionless environment, one should not only

consider the intensive margin of the labor supply. The empirical literature suggests that the

extensive margin is key (e.g. Heckman (1993), Meghir and Phillips (2010), Roed and Strom

(2002)). In a competitive setting, an individual chooses to work if and only his net income

wi − T (wi) is larger than his instantaneous value of staying out of the labor force i, which

is assumed continuously distributed. The extensive margin elasticity is empirically higher for

low-skilled workers. An increase in the log of the retention rate 1 − τ0 at w0 would lead to an

increase in low-paid participation that outweighs the decrease in participation induced by an

equivalent decrease in the log of the retention rate 1−τ1 at wage w1. This leads to a composition

effect in participation according to which a a lower global CRIP increases participation, thereby

employment. In sum, we expect that a more progressive tax schedule increases employment and

decreases output per worker, leading to an ambiguous impact on total production.14

Introducing the intensive margin in the matching framework. Turning back to en-

vironments with unemployment and wage negotiation, Sorensen (1997, 1999), Hansen (1999),

Fuest and Huber (2000) Roed and Strom (2002) and Parmentier (2006) have considered the

impact of tax progressivity when in-work effort is endogenous. Increasing income tax progres-

sivity reduces in-work effort through the traditional labor supply effect. Therefore jobs tend

to become less productive, which may also be detrimental for employment. Moreover, a more

progressive tax schedule still reduces the share of the surplus that accrue to the workers, which

is beneficial for employment through the above-mentioned wage moderation effect. In general,

the total effect on employment is ambiguous. This ambiguity can be resolved under some further

specific assumptions. For instance, Cahuc and Zylberberg (2004) obtain that the wage moder-

ation effect dominates the labor supply effect under multiplicatively separable preferences. A

more progressive tax schedule then always increases employment and always decreases the un-

employment rate and output per employed. The effect on total production remains ambiguous.

Hansen (1999) obtains similar analytical results under different specifications. Numerical simu-

lations under different assumptions for individual preferences find that employment is increasing

in tax progressivity while output per worker is reduced (Sorensen (1999), Parmentier (2006)).

The following predictions are derived in Appendix A under additively separable preferences.

14Immervoll et al. (2007) uses a microsimulation model with intensive and extensive labor supply responsesto compute the efficiency costs of two prototypical tax reforms that increase tax progressivity. They show thattheir “working poor policy” which pays more attention to the disincentive effects along the participation marginentails less efficiency costs than their “demogrant policy”. This suggests distortions along the extensive marginare quantitatively much more important to consider than the intensive margin.

10

Prediction 3. In the model with bargaining over wages and in-work effort, a rise in tax pro-

gressivity (a decline in the CRIP) reduces in-work effort. It also reduces the unemployment rate

if taxation is initially not too progressive. The effect on total production is ambiguous.

Predictions 2 and 3 confirm that while the negative effect of tax progressivity on unem-

ployment is robust to considering participation effects on the extensive margin, results are less

straightforward when we consider the intensive margin. Thus the sign of the effect of tax pro-

gressivity on the total hours of work is at best ambiguous.

III Empirical strategy

We estimate the effects of average tax rates and tax progressivity on unemployment rates and on

other labor market indicators in 21 OECD countries15 using information drawn from different

data sources, over the period 1997-2008.

Our measures of labor taxation are based on average tax rates (ATR) of single individuals

at different points of the earnings distribution, namely: 67% of the average wage, the aver-

age wage (i.e. 100%) and 167% of the average wage, provided by the OECD tax database.16

These indicators encompass income taxation by central and local governments and employers

and employees social security contributions, and are harmonized over time and across OECD

countries.17 While the focus on singles allows us to avoid many confounding factors originating

from household composition and intra-household participation decisions, it also has a drawback

since we are missing the contribution of specific policies which are restricted to households with

kids. Another feature worth noticing is that, since the wage distribution is likely to differ across

countries and over time, the average tax rates computed at 67%, 100% and 167% of the average

wage, may reflect actual taxation at a percentile of the wage distribution that are different.

From the above information, we then computed the tax retention rates

retji,t = 1− T (j ×AWi,t)

j ×AWi,t= 1−ATRji,t for j ∈ {67%, 100%, 167%}

with respect to the average wage (AWi,t) in country i and year t. The two measures of taxation

that we use in the empirical analysis are: the average tax burden, measured by the retention

rate at 100% of the average wage, ln(ret100); and the tax progressivity indicator captured by

the global CRIP, measured as the ratio of retention rates at 167% and 67% of the average wage,

ln(

ret167%i,t /ret67%

i,t

). Note that the latter follows closely the global CRIP we derived in (8),

where w0 and w1 are measured, respectively, at 67% and 167% of the average wage. 18

15The countries we consider are: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany,Greece, Ireland, Italy, Japan, Netherlands, Norway, New Zealand, Portugal, Spain, Sweden, Switzerland, UnitedKingdom and the United States.

16The OECD Tax database are drawn from http://www.oecd.org/tax/tax-policy/oecdtaxdatabase.htm, SectionB) 3.b, Table I.5.

17Since the OECD tax database only starts in 2000, we use the information provided by the OECD taxing wagedatabase to extend the relevant time series back to 1997. Details on the two database and their harmonizationare given in the Data Appendix.

18Given w0 and w1, the ratio w1/w0 is constant over time and across countries.

11

Our baseline specification is the following equation:

Yi,t = β1 ln(

ret100%i,t−1

)+ β2 ln

(ret167%

i,t−1

ret67%i,t−1

)+ ν ·Xi,t + αi + ϕt + εi,t (9)

where Yi,t is an indicator of labor market performance in country i and year t, Xi,t is a vector of

control variables, αi and ϕt are, respectively, country- and time-fixed effects, while εi,t is the error

term. The parameters of interest for the empirical analysis are β1 and β2. The vector of control

variables Xi,t includes a set of macroeconomic shocks, such as the output gap (outputgap), the

change in inflation (inflchange), and the long term interest rate on government’s bonds (irate).

It also includes a baseline set of labor market institutions, namely, the average unemployment

benefits replacement ratio (UBRR), union density (UnionDensity), an index of the degree of

coordination in wage bargaining (wcoord) and an synthetic index of employment protection

(EPL) (Nickell, Nunziata and Ochel, 2005).19.

The labor market performance variable, Yi,t, measures either the aggregate unemployment

rate, or group-specific unemployment rates. According to our theoretical predictions we expect

a rise in the retention rate (a decrease in the average tax rate) to reduce the unemployment rate

(i.e. β1 < 0), while a rise in the global CRIP (a less progressive tax schedule) is expected to

increase unemployment (i.e. β2 > 0). Alternatively, when Yi,t measures the employment rate,

we expect β1 > 0 > β2.

One reason for concern in estimating equation (9) by OLS is due to the potential endogeneity

of the tax indicators. Changes in taxation can be driven by different economic considerations,

such as (exogenous) long-term fiscal consolidation plans, (endogenous) fiscal policies induced by

cyclical variation in output (and unemployment), as well as other developments in the economy

affecting both the choice of taxation as well as unemployment. For example, a negative shock

to unemployment that reduces the tax base and increases social expenditures, when the public

budget has to be balanced, also generates a negative externality on retention rates. Alternatively,

a government may react to an increase in unemployment by cutting taxes and reducing the CRIP

to stimulate employment. In both cases above, reverse causality is likely to affect the estimates.

A further source of bias may also derive from different types of (unmeasured) labor market

policies - such as ALMP, direct job creation or early retirement - that, at the same time, lower

the unemployment rate and increase tax rates to finance the measures taken. In this case, the

omitted variable that affects unemployment is also correlated with the tax indicators. Note that,

while the macro-empirical literature on the determinants of unemployment does not generally

address such endogeneity issues 20, we take a number of steps in this direction. First, since tax

reforms take time to be decided, we enter our tax indicators in equation (9) with a one-year

lag, which also contribute to mitigating reverse causality biases. Second, we use an instrumental

variable estimator, based on exogenous variations in institutional, social and political factors that

19These data are drawn from: OECD Labor Force Statistics, OECD Economic Outlook, OECD Main EconomicIndicators, ICTWSS database and World Bank Development Indicators.

20See Nickell et alii (2005), Blanchard and Wolfers (2000), Griffith et alii (2007), Bassanini and Duval (2009)and Arpaia and Mourre (2010) for a survey. A notable exception is Nunziata (2005).

12

influence a country’s fiscal policy, to identify the (causal) effect of taxation on unemployment

(and employment). In practice, we instrument the tax wedge and tax progressivity indicators

using the following variables: a narrative record for the tax components of fiscal consolidation

policies, an index of trust in civil services and a measure of the political orientation of the

parliament.

Our first instrument is based on information drawn from country’s fiscal consolidation plans

and using the narrative approach pioneered by Romer and Romer (2010). The data we use

record multi-year tax increases and spending cuts for 17 OECD countries from 1978 to 2009

(Devries et alii 2011). Information on fiscal reforms is reconstructed using historical records,

available in official documents (i.e. budget reports, central bank reports, IMF reports, OECD

Economic Surveys and other country-specific sources), with the aim of identifying size, timing,

and main motivation for the fiscal actions undertaken by each country. In particular, in order to

guarantee the ’exogeneity’ of fiscal measures with respect to cyclical fluctuation, only long-term

structural fiscal plans primarily designed to put public debt on a sustainable path are considered.

To illustrate what in our data is a typical change in taxation motivated by long-term structural

consideration, consider the fiscal consolidation efforts (i.e. mainly in terms of higher taxation)

undertaken by several European countries to conform to the ’Stability Pact’ for accession to

the Monetary Union. While this was agreed under the terms of the 1992 Maastricht Treaty, it

was not up until the very last moment that these measures were taken and implemented. In

other words, it is the precise timing of the tax hikes, for long-waited reforms, what the narrative

indicator measures; while tax changes motivated by business cycle fluctuations are not recorded.

To construct our instrument we focus only on the changes in taxation induced by a country’s

long-term structural fiscal consolidation plan. In particular, we exploit that part of the tax wedge

on labour which originates from all the fiscal consolidation policies implemented over the years.

In practice, we construct the (Taxhike) variable as an index which, in country i at time t,

cumulates all the tax hike episodes occurred up to t-1.

Our second instrument relies on the idea that redistributive policies require a high level of

trust and confidence in public services. The more people believe that public institutions are

trustworthy, the more they are willing to let the State implement redistributive policies. So, we

expect that where confidence in public services is high, a more progressive taxation is observed.

This instrument has to property to be highly correlated with tax progressivity (i.e. the global

CRIP), but excludable from the unemployment equation. Available evidence supports these

properties: Sapienza et alii (2007) argue that respondents’ beliefs regarding the functioning of

the civil service are not directly correlated with economic activity, thus with unemployment

performance. Studies that obtain a direct effect of trust on economic activity, in general use

a measure of ‘trust in others’, rather than trust in specific institutions such as public services

(see e.g. Algan and Cahuc (2006, 2010) and Guiso et alii (2008)).21 Since low confidence in

21Algan et alii (2011) construct a composite index of distrust for public institutions, which combines distrustfor the civil service with distrust for the parliament and the justice system. We neglect the latter two dimensionsof distrust, as we are mostly interested in describing the social propensity to finance public good provision and

13

public services may also affect unemployment insurance generosity, it is important to include

the unemployment benefit replacement ratio among the controls. In practice, our measure of

(dis)trust in civil services (Trust) is given by the percentage of individuals, in each country, that

report low or no confidence in civil service; that is, the share of people that responded “none

at all” to the question: “how much confidence do you have in the civil service?” in the World

Value Survey (WVS).22

The third instrument we use is grounded on the notion that the political orientation of

parties in Parliament, quite independently of the state of the economy, is correlated with the

propensity to tax. In general, left-wing politicians are typically expected to support higher tax

levels and more progressive taxation as compared to right-wing politicians (Summers et alii

(1993), Persson and Tabellini (2000), Ardagna (2004) or Nunziata (2005) among others). In

this context, it is worth noticing that a long standing literature in political economy argues that

the political orientation of parties may affect, besides taxation, also monetary and fiscal policies,

as left-wing politicians, compared to right-wing ones, are more likely to implement keynesian

policies. We take this into account in different ways. First, we include cyclical indicators, such as

the output gap, as well as indicators of monetary policy, such as the real interest rate and change

in inflation, among the controls. 23 Second, there is evidence suggesting that adverse economic

performances may reduce the probability of reelection of incumbent politicians (Drazen, 2000).

In other words, this would imply after a negative shock to the economy an increase in leftism,

if the incumbent is right-wing, and the opposite if the incumbent is left-wing. This ’probability

of reelection channel’ induces de facto a nonlinearity in the effects of the business cycle (e.g.

unemployment) on leftism, which we exploit for identification. 24 In practice, our Leftism

instrument is measured as the difference between the shares of seats of left-wing parties with

respect to the share of seats of right-wing parties in the parliament.25 More details on all the

variables used in the empirical analysis are provided in the data Appendix B.

IV Data and descriptive statistics

We assembled a unique data set which combines information, drawn from different sources on

taxation and other labor market institutions, on labor market performance and other socio-

economic characteristics for 21 OECD countries over the period 1997-2008.26

redistribution.22Information on the (dis)trust indicator in WVS is only available every five-years. We introduce it in first

stage regression with a five-years lag.23These variables are expected to control for the keynesian channel in (9) such that there is no residuals impact

of leftism on the business cycle.24We find support to this hypothesis in the data: out of 35 country-specific economic crisis episodes - defined

as GDP being one percentage point below ist potential level in a given year, or by half a point for two consecutiveyears -, leftism obtained a relative majority in the first election after the crisis in 21 cases, while right-wing partiesprevailed in the other 14.

25The leftism variable in our first stage regression is entered with a lag - i.e. with t−2 given that tax indicatorsare already lagged t− 1 in the unemployment equation.

26The countries we consider are: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany,Greece, Ireland, Italy, Japan, Netherlands, Norway, New Zealand, Portugal, Spain, Sweden, Switzerland, United

14

To get a broad overview of our tax data, we plot in Figure 2, the average value by country,

over the sample period, of both the average tax retention rates ret100%i,t and (the exponential of)

our global CRIP, namely ret167%i,t /ret67%

i,t . Figure 3 shows, for the same indicators, the country’s

change between the 1997-1999 and the 2006-2008 sub-periods. For convenience, we also report

the overall sample means (dashed horizontal and vertical lines) that partition the graph into

four quadrants, according to the level and change of OECD countries’ tax burden (i.e. in terms

of average taxation and tax progressivity). In practice, each quadrant identifies for each country

a different long-term tax structure (Figure 2), and the major tax reforms that occurred over the

sample period (in Figure 3).

In terms of the levels of average tax burden and progressive taxation (Figure 2), the overall

patterns shows, with notable exceptions, a positive correlation between the two tax indicators,

suggesting that countries that tax more (i.e. with lower average retention rates, ret100% are

also characterized by a higher progressivity (i.e. a lower retention rates ratio, ret167%/ret67%).

The bottom-left quadrant shows countries with a high tax burden, these are mainly Nordic

and continental European countries. In the upper-right quadrant we find mostly Anglo-Saxon

and Mediterranean European countries (as well as Switzerland and Japan) characterized by a

relatively lower tax burden.

Figure 2: Average of the tax retention rates (100%APW) and the progressivity (167% vs67%APW) by country over the 1997-2008 period. Sources: OECD Tax Database and OECDTaxing wages and authors’ calculation.

When we focus on changes in the tax structure (Figure 3), we notice that the majority of

countries, over the 1997-2008 period, exhibit small variations and appear to be located in the

upper-right quadrant suggesting an overall trend, for most OECD countries, in the reduction

of tax burdens. Some countries (e.g. France, Greece, New Zealand, Austria and Great Britain)

show an increase in global progressivity. The overall pattern of changes in countries’ tax structure

(i.e. both changes in the level and progressivity of taxation) show a (weakly) positive correlation:

a number of countries, however, stand out from the fitted line showing substantial changes in the

Kingdom and the United States.

15

structure of taxation which suggest that a significant reform of the tax system occurred. The

latter shows three different patterns: first, Japan and The Netherland, where the average level

of taxation increased while progressivity remained relatively stable, and second, Ireland which

decreased substantially the tax burden both in terms of average taxation and progressivity.

Figure 3: Changes of the tax retention rates (100%APW) and the progressivity (167% vs67%APW) by country over the 1997-2008 period. Sources: OECD Tax Database and OECDTaxing wages and authors’ calculation.

Figures 4 and 5 highlight our main argument showing a snapshot of the correlation between

taxation (i.e. both average tax and tax progressivity) and aggregate unemployment across

countries. To account for (time invariant) unobserved factors, that can be correlated with both

taxation and unemployment, thus playing a counfounding role, we specify change in each variable

between the 1997-1999 and the 2006-2008 sub-periods.

In line with our theoretical priors, changes in average tax retention rates Figure 4 are shown

to be negatively related to changes in unemployment (i.e an increase in the average tax burden

is associated to a higher unemployment rate); while changes in our global measure of progressive

tax retention ratio 5 is shown to be negatively correlated with changes in unemployment (i.e an

increase in the ratio of our global CRIP is associated to a lower unemployment rate). Coun-

tries that, over the period 1997-2008, experienced an increase in average taxation (in particular,

Japan and The Netherland), generally show an increase in unemployment. The opposite occurs

in countries where a decrease in average taxation (in particular, Ireland, Sweden and Finland)

is associated with a decrease in unemployment. Interestingly, the changes in tax progressivity

and unemployment show a (weakly) positive relationship. This suggests that countries which

increased tax progressivity (in particular, France, New Zealand and Italy) experienced a (mod-

erate) decrease in unemployment; while the opposite is true for countries where progressivity

decreased.

16

Figure 4: Change in average tax retention rates and change in unemployment between 2006-2008 and 1997-1999 by country. Sources: OECD Tax Database and OECD Taxing wages andauthors’ calculation.

Figure 5: Change in tax progressivity and change in unemployment between 2006-2008 and1997-1999 by country. Sources: OECD Tax Database and OECD Taxing wages and authors’calculation.

V Empirical results

V.1 Baseline specification

Table 1 presents results for our baseline specification with the aggregate unemployment rate

as dependent variable (i.e. OLS estimates are reported in columns (1) and (2), IV estimates

are shown in columns (3) to (5)). In Column (1) we report results that are consistent with

the specification typically used in the literature, whereby taxes affect the unemployment rate

only through the average tax wedge (e.g. Nickell and Layard (1999); see Bassanini and Duval

(2009) for a review). A one percentage point increase in the average retention rate, ln(ret100)

(i.e. a one percentage point decrease in the labor tax wedge) has a positive impact on the

unemployment rate, for the average OECD country, of about 0.09 percentage points. This

order of magnitude is consistent with previous findings, which suggest that one percentage point

change in average taxation may entail a change in the unemployment rate in the range of 0.02 and

17

0.2 percentage points, the former as reported in Blanchard and Wolfers (2000), the latter as in

Nickell and Layard (1999). Daveri and Tabellini (2000) report a much bigger effect in Continental

European Countries, they suggest that the unemployment reducing impact of one percentage

point tax cut may be close to 0.5 percentage points, due to the interactions between labor

taxation and different welfare models. In Column (2) we show results with a specification that

also considers the impact of tax progressivity on unemployment. Consistent with Prediction 1, a

lower average tax rate (as described by a marginal increase of ln(ret100)) and a more progressive

tax schedule (as described by a marginal decrease of ln( ret167ret67 )) are both associated with a lower

unemployment rate. The effect of the average tax rate is statistically significant at the 5% level,

while the effect of progressivity is only weakly significant. From the coefficient on ln( ret167ret67 ) in

Column (2), we can compute the effect of a half-percentage point decrease in the average tax

rate at 67% of the average wage, together with a half-percentage point increase in the average

tax rate at 167% of the average wage. Such a tax reform induces a rise in tax progressivity

that decreases the global CRIP ln( ret167ret67 ) by 0.016 points, thereby reducing unemployment by

about 0.06 percentage points27. After accounting for tax progressivity, the estimated effect of

one percentage point increase in the average retention rate on the unemployment rate rises to

0.11 percentage points.

However, as previously discussed, OLS estimates may be biased and inconsistent due to

the endogeneity of the tax variables. In columns (3), (4) and (5), we deal with those issues

implementing an instrumental variable estimator (IV). In particular, in Column (3), we re-

estimate the specification of column (1) using the variables Taxhike and of Leftism (lagged

twice) to instrument ln(ret100) 28. In Column (4), we introduce tax progressivity but consider

this variable as exogenous and instrument ln(ret100) only, as in Column (3). In Column (5),

we instrument both ln(ret100) and ln( ret167ret67 ), using the same instruments as before, as well as

the Trust variable 29 First-stage estimates show that the Taxhike and Trust indicators have

a strong statistically significant impact on the ln(ret100) variable, while the Trust indicator

is mostly correlated with the global CRIP ln(ret167ret67

)variable (i.e. first-stage estimates are

shown in Table 7 Appendix C). First-stage estimates reported in Table 7 in the Appendix C

are reassuring about instruments’ correlation with the endogenous regressors, and the Hansen

J test confirms the validity of the over-identifying restrictions.30.

27The mean of ret67, ret100 and ret167 over the sample are respectively 66.35%, 62.19% and 57.42%. Thus,the decrease in the global CRIP induced by a typical reform computed at the mean is (0.5/66.35) + (0.5/57.42) =0.0162.

28Note that the Taxhike variable in the Devries et alii dataset is only available for 17 out of the 21 countriesin our sample. For this reason, the observations for the missing countries (i.e. Greece, New Zealand, Norway andSwitzerland) are set to zero letting the country dummies capture this effect in the estimation. In the robustnesschecks section we provide additional evidence imputing the information on (exogenous) long-term structuralchanges in taxation from an external source.

29Since the number of instruments exceeds the number of endogenous regressors, we can test the validity ofoveridentifying restrictions.

30The p values of the Kleibergen-Paap under-identification test show that we can reject the null hypothesisthat our model is under-identified, while the F statistics with one instrument (in Columns (3) and (4)) is slightlybelow the threshold level of 10, conventionally used as a rule of thumbs by Stock and Yogo under the assumptionof i.i.d residuals. Moreover the Anderson and Rubin test, which is robust to weak instruments, rejects at the 1%

18

OLS estimates IV estimates(1) (2) (3) (4) (5)

ln(ret100) –5.893** –6.680** –25.382*** –30.554*** –59.181***(2.664) (2.750) (9.466) (10.956) (22.709)

ln( ret167ret67 ) 3.859 10.156** 70.128***

(2.522) (4.404) (26.841)

UBRR 0.031 0.037 0.117** 0.146** 0.309**(0.025) (0.025) (0.052) (0.061) (0.123)

UnionDensity 0.066 0.078 –0.042 –0.029 0.059(0.055) (0.054) (0.077) (0.081) (0.184)

wcoord –0.541*** –0.545*** –0.588*** –0.605*** –0.701*(0.191) (0.188) (0.205) (0.221) (0.397)

EPL 0.753* 0.735* 1.096** 1.102** 1.103(0.405) (0.393) (0.457) (0.453) (0.749)

outputgap –0.378*** –0.375*** –0.460*** –0.465*** –0.487***(0.064) (0.064) (0.082) (0.085) (0.138)

inflchange –0.068 –0.064 –0.167* –0.172* –0.190(0.076) (0.077) (0.101) (0.104) (0.177)

irate –0.273* –0.299** –0.546** –0.658** –1.292**(0.141) (0.142) (0.252) (0.282) (0.524)

R2 0.89 0.89 0.86 0.85 0.53N 231 231 231 231 231Hansen J test 0.1380 0.1629 0.6423Anderson Rubin F test 0.0093 0.0057 0.0000

Table 1: Tax wedge, tax progressivity and the unemployment rate. Robust standard errorsin parentheses. Significance levels: ∗: 10%, ∗∗: 5%, ∗ ∗ ∗: 1%. p-values of Hansen J over-identification tests and of Anderson and Rubin F test of significance of endogenous regressorsare provided.

A comparison of estimated coefficients reported in Columns (3)-(5) with those in Column (2)

indicate that OLS estimates of ln(ret100) may suffer from attenuation bias. IV estimates indeed

suggest that one percentage point increase in the average retention rate implies a reduction

of the unemployment rate, for the average OECD country, between 0.41 and 0.95 percentage

points - i.e. an order of magnitude that is closer to the one suggested by Daveri and Tabellini

(1995). Symmetrically, an increase in tax progressivity ln( ret167ret67 ) - computed as described above

- determines a reduction in aggregate unemployment between 0.16 and 1.13 percentage points.31.

In other words, a reduction in the average tax rate combined with an increase in tax progressivity

can have a sizeable effect on the unemployment rate. Estimates of the other control variables

are in line with usual findings in the literature, namely: higher unemployment benefits tend to

increase unemployment, while more coordination in wage bargaining reduces it. An increase in

the output gap is associated with lower unemployment, while the negative coefficients of the

the null hypothesis that the endogenous regressors are not significant in the second stage.31It should be noted, however, that the lower bound estimate of 0.16 is obtained from a specification in which

tax progressivity is assumed to be exogenous (Column 4), while the estimated impact of 1.13 is obtained from aspecification in which also tax progressivity is instrumented.

19

real interest rate on government’s bonds and the change in inflation suggest that policies that

guarantee prices stabilization (in either the goods and capital markets) may be associated with

higher unemployment.

Overall, estimates in Table 1 yield three interesting considerations regarding the impact

of taxation on the unemployment rate. First, omitting the role of tax progressivity leads to

underestimate the impact of average taxation on unemployment. Second IV estimates, when

compared to OLS, show a considerable increase in the magnitude of coefficients, both in terms

of ln(ret100), as well as ln( ret167ret67 ). We interpret this finding as evidence that reverse causality

introduces attenuation bias on OLS estimates, - e.g. a government may react to an increase in

unemployment by reducing average labor taxes, in particular on low-paid jobs. Also omitted

factors affecting the unemployment rate, such as public employment and relief jobs, would be an

additional source of attenuation when negatively correlated with the ln(ret100). Last, but not

least, the impact of tax progressivity is shown to be as important as the impact of the average

taxation.

OLS estimates IV estimates(1) (2) (3) (4) (5)

ln(ret100) 3.125 4.814 20.994* 28.271** 60.140**(3.750) (3.938) (11.323) (12.973) (26.889)

ln( ret167ret67 ) –8.278** –14.465*** –81.155**

(3.527) (5.428) (36.104)

R2 0.97 0.97 0.97 0.97 0.92N 231 231 231 231 231Hansen J test 0.1713 0.2032 0.6706Anderson-Rubin F test 0.0330 0.0126 0.0007

Table 2: Tax wedge, tax progressivity and the employment rate. Robust standard errors inparentheses. Significance levels: ∗: 10%, ∗∗: 5%, ∗ ∗ ∗: 1%. p-values of Hansen J over-identification tests and of Anderson and Rubin F test of significance of endogenous regressorsare provided.

In Table 2 we replicate the same regressions of Table 1 replacing the unemployment rate with

the employment rate as the dependent variable. Most previous findings are confirmed also when

the effects of taxation are estimated on the employment rate, though, as expected, the signs are

now reversed as compared to Table 1. Focussing on our preferred specification in Column (5),

we find that a one percentage point increase in the average retention rate determines an increase

of the employment rate, for the average OECD country, up to 0.96 percentage points. Similarly,

a one percentage point increase in ln( ret167ret67 ) - defined as a half-percentage point decrease in the

average tax rate at 67% of the average wage and a half-percentage point increase in the average

tax rate at 167% - causes an increase in the employment rate up to 1.32 percentage points. Here

as well, a comparison of the coefficients on ln(ret100), in Column (3) and Column (5), indicate

that neglecting progressivity tends to underestimate the impact of the average tax wedge on

employment.

20

log(1− urate) log(prate) log(erate) log(GDPEmp ) log(GDPPop )

(1) (2) (3)=(1)+(2) (4) (5)=(3)+(4)

OLS estimates

ln(ret100) 0.100*** –0.003 0.096 –0.026 0.070(0.036) (0.040) (0.063) (0.072) (0.090)

ln( ret167ret67 ) –0.104*** –0.042 –0.146*** 0.135** –0.012

(0.035) (0.037) (0.055) (0.062) (0.065)

R2 0.86 0.98 0.97 1.00 1.00N 231 231 231 231 231

IV estimates

ln(ret100) 0.856*** 0.173 1.029** –0.533 0.496(0.317) (0.177) (0.455) (0.585) (0.391)

ln( ret167ret67 ) –0.981** –0.382 –1.362** 1.933*** 0.570

(0.383) (0.292) (0.595) (0.594) (0.494)

R2 0.38 0.97 0.90 0.98 0.99N 231 231 231 231 231Hansen J test 0.5803 0.8811 0.7333 0.6019 0.1605Anderson Rubin F test 0.0000 0.5671 0.0006 0.0000 0.0026

Table 3: Decomposing the effects of taxation on different indicators. Robust standard errorsin parentheses. Significance levels: ∗: 10%, ∗∗: 5%, ∗ ∗ ∗: 1%. p-values of Hansen J over-identification tests and of Anderson and Rubin F test of significance of endogenous regressorsare provided.

As a final exercise, in Table 3, we decompose the effects of taxation on labour market

performance using the same identification strategy as in Columns (5) of Tables 1 and 2. The

top panel reports OLS estimates, while the bottom panel shows IV estimates. In particular, we

consider the effect of taxation on different dependent variables, namely: in Column (1) we use

the logarithm of one minus the unemployment rate, in column (2) we take the logarithm of the

participation rate, while in Column (3) the logarithm of the employment rate is used instead.

In this way, adding the estimates of Column (1) and Column (2), by construction, we obtain

estimates reported in Column (3)32. Similarly, in Column (4) we use the logarithm of GDP per

worker as dependent variable, while in Column (5) the logarithm of the GDP per working-age-

population is used. Since, GDP per working-age-population is the product of the employment

rate times GDP per employed worker, the sum of estimates in Column (3) and Column (4),

by construction, give estimates reported in Column (5). When the logarithm of one minus the

unemployment rate is used as the the dependent variable (Column (1)), the estimated coefficients

provide the elasticity of tax indicators with respect to the fraction of the active population that is

employed. Results show that a higher taxation increases unemployment while more progressivity

decreases it (e.g. note that there is also an improvement in the statistical significance of tax

progressivity with respect to Table 1). Focusing on Column (3), we find that lower taxation and

32Note that, for consistency with the decomposition exercise, in Column (1) we use the (non-harmonised)unemployment rate instead of the standardised unemployment rate provided by OECD, as in Table 1. We havechecked that this change has little effects on the estimates.

21

higher progressivity decrease the employment rate considerably, however these effects seems to

be mainly driven by unemployment rather than by the the effects played by the participation

rate (e.g. which is non statistically significant both in OLS and IV estimates of Column (2)).

This results suggest that the emphasis that is usually placed, in the public finance literature,

on labor supply decisions along the extensive marginto to explain the employment effects of

changes in taxation, instead of wage moderating effects, may be misleading.

In order to quantify the impact of labor taxation along the intensive (in-work effort) mar-

gin, in Column (4) we use as dependent variable the (log of) GDP per worker. In line with

Prediction 3, we find that a rise in progressivity (a decline in CRIP ln( ret167ret67 )) reduces GDP per

employed worker. Therefore our finding that progressivity reduces unemployment and increases

employment is not inconsistent with the common wisdom that tax progressivity may also have

a detrimental effect on incentives to work harder. These estimates are highly significant. Con-

versely, the effect of ln(ret100) is never statistically significant. Despite the negative impact

of progressivity on output per worker shown in Column (4), we find that the overall effect on

output is not statistically significant. This may suggest that the effects of progressivity on em-

ployment are indeed large enough to offset the negative impact on the incentive margin. In sum,

Table 2 reconciles our view that tax progressivity reduces unemployment (Column (1)) and in-

creases employment (Column (3)) with the traditional view that it generates negative incentives

in terms of in-work effort (Column (4)). It also shows that the total effect on production is

ambiguous (Column (5)).

Skill AgeLow-skilled High-skilled Young Adults

OLS estimates

ln(ret100) –9.311** –5.428 –17.203** –8.776***(3.946) (3.355) (6.807) (3.011)

ln( ret167ret67 ) 8.504** 9.963*** 14.659* 9.828***

(3.588) (2.916) (7.826) (2.859)

R2 0.85 0.78 0.91 0.87N 231 231 231 231

IV estimates

ln(ret100) –82.798** –51.221** –139.108*** –69.584***(36.825) (25.118) (49.359) (24.623)

ln( ret167ret67 ) 123.079*** 64.475** 139.764** 78.557**

(40.349) (31.508) (61.439) (30.915)

R2 0.23 0.44 0.69 0.44N 231 231 231 231Hansen J test 0.8169 0.1195 0.3678 0.5088Anderson Rubin F test 0.0000 0.0110 0.0000 0.0000

Table 4: Effects on the unemployment rates of different subgroups. Significance levels: ∗: 10%,∗∗: 5%, ∗ ∗ ∗: 1%. p-values of Hansen J over-identification tests and of Anderson and Rubin Ftest of significance of endogenous regressors are provided.

22

Estimates in Tables 1, 2 and 3 implicitly assume that workers are homogeneous both in

terms of labour demand characteristics of the available jobs, as well as in terms of labor supply

responses to fiscal and labor market incentives. In Table 4, we report estimates of our preferred

specification (i.e. Columns (5) in Table 1) for selected sub-population of workers, namely using

the unemployment rate by skill and age (i.e. Columns (1) and (2) for skill, and Columns (3) and

(4) for age, respectively). The top panel report OLS estimates while the bottom panel shows

IV estimates. Some caution is needed when interpreting these estimates, since the distribution

of wages can vary a lot between these groups, while retention rates are not specific to each sub-

population considered. Still, a rough comparison of the magnitude and statistical significance of

the coefficients of ln(ret100) and ln( ret167ret67 ) can improve our understanding of the diverse effects

of taxation on unemployment across different groups of workers.

Estimates in columns (1) and (2) suggest that the impact of taxation and progressivity is

relatively larger (and statistically more robust) for low skilled workers as compared to high

skilled workers.33 Results reported in Columns (3) and (4) show a similar patters, whereby the

impact of taxation and progressivity on unemployment is larger for younger workers (i.e. age

group 15-24), as compared to older workers (i.e. age group 25-54).

V.2 Robustness checks

Baseline Identification Using Guichard et alii Using Public Debt(1a) (1b) (2a) (2b) (3a) (3b)

Second-stage results

UNR erate UNR erate UNR erate

ln(ret100) –59.181** 60.140* –56.339* 58.699 –61.979*** 65.666**(22.709) (26.889) (25.404) (30.567) (18.001) (21.855)

ln( ret167ret67 ) 70.128** –81.155* 72.217** –85.020* 70.128* –80.502*

(26.841) (36.104) (27.142) (36.796) (27.567) (36.990)

R2 0.53 0.92 0.53 0.92 0.51 0.92N 231 231 231 231 231 231Hansen J 0.6423 0.6706 0.8642 0.5608 0.6903 0.6153Anderson Rubin F 0.0000 0.0007 0.0000 0.0083 0.0000 0.0000

First-stage statistics

ln(ret100) ln( ret167ret67 ) ln(ret100) ln( ret167

ret67 ) ln(ret100) ln( ret167ret67 )

F of Excluded 5.976 6.391 6.260 5.680 16.571 7.518Shea Partial R2 0.0778 0.0430 0.0725 0.0433 0.1078 0.0431KP under 0.0009 0.0008 0.0007KP weak 4.12 4.23 4.37

Table 5: Robustness checks about instruments

In this section, we consider a number of potential robustness concerns for our empirical

33These two skill categories are based on educational attainment. Low skilled workers completed up to secondaryeducation, while high skilled workers completed tertiary education or more. The inclusion of individuals with ahigh school degree into the low skilled category can be disputed. However, this aggregation choice is unavoidabledue to the switch from the ISCED76 to the ISCED97 classification, which recoded lower-secondary educationfrom level 1 (primary) to level 2 (secondary).

23

analysis and run several sensitivity checks. In Table 5 we test the robustness of our results

using an alternative identification strategy based on different sets of instruments. To facilitate

comparisons, in Columns (1a) and (1b), we reproduce our results from the baseline specification

(Column (5) of Table 1, for the unemployment rate; Column (5) of Table 2 for the employment

rate). The bottom part of Table 5 shows first-stage results (i.e. note these are identical whether

the dependent variable in the second stage is the unemployment or the employment rate) 34

In Columns (2a) and (2b), first we replace the missing information (for Greece, New Zealand,

Norway and Switzerland) in the Taxhike instrument (from the Devries et alii dataset) with

information drawn from an alternative OECD dataset on fiscal consolidation (Guichard et alii

2007). Although the information on fiscal consolidation episodes provided in the Guichard et

alii dataset relies on an statistical algorithm (i.e. based on countries’ structural changes in the

cyclically adjusted primary balance), rather than on the ’narrative’ approach, the sequence of

tax hikes show a rather similar pattern in both dataset 35. Finally, in Columns (3a) and (3b), we

use the ratio of public debt over GDP (t−2) as instrument to replace the tax hike indicator. The

latter captures the idea that highly indebted governments need to use fiscal policy to consolidate

their budget (Gali and Perotti, 2005). To mitigate the risk of reverse causality we lagged the

variable twice. Moreover, the inclusion of cyclical controls is expected to account for any effect

of public debt on unemployment going through the channels of demand-management policies

(i.e. note that the typical Keynesian fiscal policies are better described by the ratio of public

deficit over GDP, rather than public debt over GDP ratio). In general results reported in Table

5, where we use alternative instruments, support earlier findings providing additional evidence

for the robustness of our estimates.

Additionally, we run a set of more traditional robustness checks. In particular, we experiment

a different set of control variables, alternative specifications to the baseline equation, various

estimation method to control for unobserved country specific shocks, as well as a different

clustering of the error term to account for common unobserved effects. In Table 6, for each

sensitivity check, we report the estimated coefficients of the tax variables and the Hansen test

for the overidentifying restrictions. The various robustness checks experimented are discussed

hereafter. We present all estimates using both the unemployment rates (Columns (1) to (3))

and the employment rates (Columns (4) to (6)).

In our baseline equation we use the output gap to control for overall business cycle conditions,

assuming that output fluctuations are exogenous with respect to the unemployment rate. To

check for the potential (endogeneity) bias that the inclusion of the output gap may determine, in

row (1), we re-estimated the model omitting it. Next, while in our baseline specification all the

tax variables are assumed to be pre-determined and entered with a year lag, in row (2) we check

the sensitivity of tax estimates when we replace the lagged terms with the contemporaneous

34Column (1a) displays the Shea Partial R2 and the F statistic of the significance test of excluded instrumentsfor the first-stage estimation of ln(ret100); Column (1b) provides the same statistics for ln( ret167

ret67).

35We checked the consistency of the two indicators - i.e. ’narrative’ and ’statistical’ - for the countries in whichinformation on both approach was available.

24

tax indicators. We further experiment how results are altered by considering only country

fixed effects (i.e. excluding the time dummies; row (3)), replacing country fixed effects and

time dummies with country specific trends (row (4)) and including in the specification both

time dummies and country specific trends (row (5)). To account for the presence of cross-

section correlated error terms, in row (6), we replace the time-fixed effects with a Correlated

Common Effect Pooled (CCEP) estimator (Pesaran (2006)), while in row (7) we implement

the Newey-West HAC estimator that delivers heteroskedasticity and autocorrelation consistent

estimates. In all the above cases the results are not significantly altered with respect to our main

specification. In row (8) to (11), we progressively exclude from our sample the countries that,

in the period considered in the empirical analysis, show significant changes in the structure of

taxation suggesting major tax reforms (i.e. France, Ireland, Japan and The Netherlands). In all

above cases, the overall results are unaffected by the change in the set of countries considered.

Finally, to capture the medium-run effects and to partially smooth the year-to-year variations,

we replicated the estimates using three years averages. Also in this case the overall results are

virtually unchanged.

VI Discussion and Conclusions

In this paper, we review a number of theoretical arguments that support the idea that tax pro-

gressivity may have an unemployment-reducing effect. We develop a simple theoretical model

in which while tax progressivity may be detrimental to labor supply along the intensive mar-

gin, its overall effects on employment and unemployment may be beneficial for labor market

performance (Boone and Bovenberg (2004)). We empirically test the effects of tax progressivity

on employment and unemployment using a panel data of 21 OECD countries for the 1998-2008

period. We propose a new measure global tax progressivity which is defined as the difference

in the fiscal wedge between 67% and 167% of the average wage for a single worker. We then

include our tax progressivity index to the list of time-varying indicators of labor market institu-

tions traditionally used to explain differences in equilibrium unemployment across countries. We

find that tax progressivity has a significant unemployment-reducing impact of a similar order of

magnitude as the unemployment-increasing effect of average labor market taxation. This effect

is more concentrated among low-skilled workers and among young workers. These results are

line with theories that claim tax progressivity reduces unemployment because it shifts tax bur-

den away from group of workers that are more responsive to taxation, and because it generates a

wage moderation effect in collective wage-setting that is favorable to labor demand. We also find

that a more progressive tax schedule increases the employment rate, but decrease production

per worker, leading to a negligible total impact on total production. Our central result that tax

progressivity increases employment comes thus hand by hand with the usual disincentive effect

of tax progressivity on the intensive margin of the labor supply that is central in the public

finance literature. In sum we add a new effect to take into account in the optimal design of

25

labor income taxation. Optimal progressivity not only trades off the equity gain of a higher

progressivity against the efficiency loss due to the disincentive effect along the incentive margin.

One has also to take into account the efficiency gains of a more progressive tax schedule through

a reduction in unemployment and a rise in employment. Our results need to be extended in

different directions, the most obvious one being to build indicators of tax progressivity over a

longer time period.

A Theoretical model

We here consider the general model with an intensive labor supply margin. Let `i be the effortprovided by employed individuals and the money-equivalent of this disutility.36 One can thinkof `i as hours of work or as the intensity of in-work effort. We assume that providing effort `ileads to a flow of output fi(`i), with f ′i(.) > 0 > f ′′i (.) and a flow of disutility `i. Denoting wias the total gross wage (and not the wage rate), Equation (4a) becomes:

r Ei = wi − T (wi)− `i + δi (Ui − Ei) (10)

The case without the intensive margin is retrieved by setting `i = 0 and f(`i) = yi below.Equations (2) and (10) imply respectively:

Ji =fi(`i)− wir + δi

and Ei − Ui =wi − T (wi)− `i − r Ui

r + δi(11)

Maximizing the (log of) the Nash product with respect to the wage and in-work effort leads to:

γi (1− T ′i )Ei − Ui

=1− γiJi

andγi

Ei − Ui=

(1− γi) f ′i(`i)Ji

where T ′i ≡ T ′(wi). Using (6) and Ψi = Ψ(wi), the first of these two equations gives the sharingrule (5). The second of these equations imply the labor supply condition:(

1− T ′i)f ′i(`i) = 1 (12)

which determines effort `i as a decreasing function of the marginal tax rate, independently ofany other variable. Combining the sharing rule (5) with (11) lead to:

γi Ψi (fi(`i)− wi) = (1− γi)(wi −

`i + r Ui1− τi

)Moreover, we get from (4b) and the sharing rule (5):

r Ui1− τi

=bi

1− τi+ pi(θi)

γi Ψi

1− γiJi =

bi1− τi

+ ciγi Ψi

1− γiθi (13)

where the second equality uses the free-entry condition (3). The (gross) wage equation is

wi =γi Ψi

1− γi + γi Ψi(fi(`i) + ci θi) +

1− γi1− γi + γi Ψi

bi + `i1− τi

which gives (7a) in the case of exogenous labor supply by taking `i = 0 and f(`i) = yi. Equation(7b) follows directly. Combining this wage equation with the free-entry condition (3) leads to:(

1 +γi Ψi

1− γi

)r + δiqi (θi)

+γi Ψi

1− γiθi =

1

ci

(fi (`i)−

bi + `i1− τi

)(14)

36Introducing a more complex increasing and convex expression for this disutility leads to more cumbersomeexpressions without adding any new insight.

26

Equation (14) determines implicitly equilibrium tightness θi as a decreasing function of effort`i, of the average tax rate τi and of the CRIP Ψi. Using (1), it also determines the fraction1− ui of participants of type i that are employed.

With an endogenous labor supply margin, productivity is a function of effort and the disutil-ity of effort plays a role similar to the value in nonemployment bi. For a given level of effort `i, arise in progressivity (a decrease in the CRIP Ψi) increases employment through the same wagemoderating. Moreover, a rise in tax progressivity decreases effort `i. This reduction has twoopposite effect on the total surplus. On the one hand, production decreases, on the other handthe disutility of work also decreases. The net effect of hours of work on employment depends onhow the term fi (`i)− (bi+`i)/(1−τi) varies with effort `i. Using (12) a reduction in effort `i in-creases (decreases) this term whenever taxation is regressive (progressive) i.e. when Ψi > 1 (i.e.Ψi < 1). Therefore, by continuity, and provided that the taxation is not too progressive so thatthe labor supply effect of tax progressivity remains dominated by the wage moderating effect, arise in tax progressivity (a reduction in the CRIP Ψi) increases employment and decreases theunemployment rate.

To get the Let us denote Xi ≡ γi Ψi

1−γi . From (14), the effect of the CRIP Ψi on tightnesshappens only through a change in the “effective” bargaining power that is captured by a changein Xi. A rise in the CRIP Ψi holding the average tax rate τi constant increases participationonly if it increases the value of unemployed. Rewriting (13) as:

r Ui = bi + ci Xi θi(1− τi)

implies that a rise in the CRIP for a fixed τi increases participation if and only if Xi θi isincreasing in Xi. Denoting the elasticity of the job filling rate by η(θ) ≡ −θq′(θ)/q′(θ) anddifferentiating (14) in θi and in Xi = γi Ψi

1−γi gives

dθiθi

= −Xi

r+δq(θi)

+Xi θi

η(θi)(1 +Xi)r+δq(θi)

+Xi θi

dXi

Xi

Hence, Xi θi is increasing in Xi, thereby participation is decreasing in progressivity, if and onlyif Xi > η(θi)(1 +Xi). Given the definition of Xi, the latter condition is equivalent to

γi Ψi

1− γi + γi Ψi> η(θi)

that is to effective bargaining power being higher than the efficient one prescribed by the Hosios(1990) condition (see Pissarides (2000)), i.e. by unemployment rate being inefficiently high.

B Data Appendix

Instruments

DisTrustCivil: percentage of respondents which gives answer 4 (i.e., ‘none at all’) to ques-tions E069 8 in WVS1-5, V212 in EVS4, V207 in EVS3, q553i in EVS2, v546 in EVS1 (howmuch confidence in civil service). The period is as follows:

1980-89 : coverage by EVS1/WVS1 but for CHE, covered by EVS2. Surveys carried in 1981for AUS, BEL, DEU, DNK, ESP, FIN, FRA, GBR, IRE, JPN, NLD; 1982 for CAN, NOR, NOR,SWE, USA;

1990-94 : coverage by EVS2/WVS2. Surveys carried in 1990 for AUT, BEL, CAN, DEU,DNK, ESP, FIN, FRA, GBR, ITA, JPN, NLD, NOR, PRT, SWE, USA. Notice that we havetwo observations for ESP (in year 1990) corresponding to both WVS2 and EVS2 being carriedthat year.

27

1995-99 : coverage by EVS3/WVS3. Surveys carried in 1995 for AUS, ESP, JPN, USA; 1996for CHE, FIN, NOR, SWE; 1997 for DEU; 1998 for GBR, BEL, GBR, NZL; 1999 for AUT,BEL, DEU, DNK, ESP, FRA, GBR, GRC, IRE, ITA, NLD, PRT, SWE, USA. Notice that wehave two observations for ESP (1995 and 1999), DEU (1997 and 1999), GBR (1998, 1999), andUSA (1999), corresponding to both WVS3 and EVS3 being carried in those countries.

2000-04 : coverage by WVS4 but for FIN and NZL, covered by EVS3 and WVS5, respectively.This period is generally not covered by any EVS wave, thus the majority of European countriesis not surveyed. Surveys carried in 2000 for CAN, ESP, FIN, JPN; 2004 for NZL.

2005-08 : coverage by EVS4/WVS5. Surveys carried in 2005 for AUS, FIN, ITA, JPN; 2006for CAN, DEU, FRA, GBR, NLD, SWE, USA; 2007 for CHE, ESP; 2008 for AUS, CHE, DEU,DNK, ESP, FRA, GRC, IRE, NLD, NOR, PRT. Notice that we have two observations for AUS(2005 and 2008), CHE (2007 and 2008), DEU (2006, 2008), ESP (2007, 2008), FRA (2006,2008), NLD (2006, 2008), POL (2005, 2008), corresponding to both WVS5 and EVS4 beingcarried.

Observations were averaged out by country and period thus obtaining an unbalanced panelof 21 countries covering the period 1990-2008 in five years averages. Missing observations wereobtained by linear interpolation. The initial observation covering the period 1980-89, has notbeen used in the empirical analysis, but provided the basis to obtain the observation for theperiod 1990-94 by linear interpolation rather than extrapolation for countries where observationswere missing for this period.

Leftism: Unweighted distance between left party and right party in legislative seats, as a per-cent of all legislative seats (authors’ calculation on data from ‘Electoral, Legislative, and Govern-ment Strength of Political Parties by Ideological Group in Capitalist Democracies, 1950-2006: ADatabase’, (by Duane Swank. See http : //www.marquette.edu/polisci/facultyswank.shtml).

Pcdebt: General government gross financial liabilities, as a percentage of GDP (OECDEconomic Outlook).

Taxhike: dcoumented tax increases drawn by historical sources and records based onthe methodology developed by Romer and Romer (2010). (see Devries et alii, 2011: http ://www.imf.org/external/pubs/cat/longres.aspx?sk = 24892.0.

Other variables used in the analysis

EPL: Unweighted sum of the OECD synthetic index of employment protection legislation(OECD Employment Outlook).

UnionDensity: union density (% of unionized workers; OECD Employment Outlook).UBRR: average unemployment benefit replacement rates (average of replacement rates across

various earnings levels, family situations and durations of unemployment; OECD Benefits andWages Database).

Irate: Long-term interest rate on government bonds (OECD Economic Outlook).Outgap: Percentage deviation of output from trend (OECD Economic Outlook).Rexch: Ratio of home country’s prices to a weighted average of competitor country’s prices,

relative to a base year (2000) and measured in US dollars. Therefore an increase is an appreci-ation of the home country’s real exchange rate (OECD Main Economic Indicators).

Inflchange: Change in the inflation between two consecutive years (authors’ calculation ondata from the OECD Economic Outlook).

C First-stage results

28

Un

emp

loym

ent

rate

Em

plo

ym

ent

rate

ln(ret

100)

ln(ret

167

ret

67

)H

anse

nJ

test

ln(ret

100

)ln

(ret

167

ret

67

)H

an

sen

Jte

st(p

valu

e)(p

valu

e)

1.n

oou

tpu

tgap

-55.

31**

77.5

4***

0.60

-55.3

1**

77.

54*

**0.

496

(25.

32)

(28.

55)

(0.4

37(2

5.32)

(28.5

5)

(0.4

81)

2.

conte

mp

ora

neo

us

tax

ind

icato

rs-6

6.85

**76

.10*

*1.

09-6

6.8

5**

76.

10*

*0.4

17(2

7.47

)(3

3.45

)(0

.295

)(2

7.47)

(33.4

5)

(0.5

18)

3.

cou

ntr

yfi

xed

effec

ts,

on

ly-4

8.30

***

59.1

9***

1.44

3-4

8.30*

**59.

19*

**0.

003

(18.

67)

(17.

81)

(0.2

29)

(18.

67)

(17.8

1)

(0.9

57)

4.

cou

ntr

ysp

ecifi

ctr

end

s,on

ly-5

5.09

**77

.26*

**0.

339

-55.0

9**

77.

26*

**0.

316

(22.

89)

(26.

25)

(0.5

61)

(22.

89)

(26.2

5)

(0.5

74)

5.

cou

ntr

ysp

ecifi

ctr

end

san

dti

me

du

mm

ies

-58.

99**

*70

.19*

**0.

225

-58.

99*

**70.

19*

**0.

226

(22.

67)

(26.

73)

(0.6

35)

(22.

67)

(26.7

3)

(0.6

35)

6.

cou

ntr

yfi

xed

effec

tsan

dC

CE

Pes

tim

ator

-51.

80**

*62

.51*

**0.

485

-51.

80*

**62.

51*

**0.

068

(19.

28)

(22.

31)

(0.4

86)

(19.

28)

(22.3

1)

(0.7

94)

7.

HA

Cst

an

dar

der

rors

-59.

18**

70.1

3**

0.17

4-5

9.1

8**

70.

13*

*0.1

48(2

8.35

)(3

0.56

)(0

.676

)(2

8.35)

(30.5

6)

(0.7

00)

8.

Fra

nce

excl

ud

ed-6

5.36

***

59.8

9**

1.37

1-6

5.36*

**59.

89*

*1.

390

(20.

43)

(23.

66)

(0.2

41)

(20.

43)

(23.6

6)

(0.2

38)

9.

Irel

an

dex

clu

ded

-58.

18**

*62

.98*

*0.

321

-58.

18*

**62.

98*

*0.

104

(21.

59)

(25.

00)

(0.5

71)

(21.

59)

(25.0

0)

(0.7

47)

10.

Jap

an

excl

ud

ed-8

2.95

**76

.99*

*0.

283

-82.9

5**

76.

99*

*0.5

34(4

1.40

)(3

3.72

)(0

.594

)(4

1.40)

(33.7

2)

(0.4

65)

11.

Net

her

lan

ds

excl

ud

ed-6

7.97

**80

.52*

*0.

166

-67.9

7**

80.

52*

*0.0

61(3

1.88

)(3

4.74

)(0

.683

)(3

1.88)

(34.7

4)

(0.8

06)

12.

thre

eye

ars

aver

ages

-51.

69*

51.6

5**

0.13

2-5

1.69*

51.6

5**

0.498

(26.

51)

(25.

39)

(0.7

16)

(26.

51)

(25.3

9)

(0.4

80)

Tab

le6:

Rob

ust

nes

sch

ecks.

IVes

tim

ates

wit

hro

bu

stst

and

ard

erro

rs.

Han

senJ

pro

vid

esth

ep

-valu

eof

the

Han

sen

test

for

over

-id

enti

fica

tion

.S

ign

ifica

nce

leve

ls:∗

:10

%,∗∗

:5%

,∗∗∗

:1%

29

[3] [4] [5]ln(ret100) ln(ret100) ln(ret100) ln( ret167

ret67 )

Taxhike –0.0135** –0.0126* –0.0127* –0.0062(0.0068) (0.0068) (0.0068) (0.0060)

Trust –0.0830 0.1936***(0.0786) (0.0674)

Leftism –0.0475*** –0.0421*** –0.0461*** –0.0301(0.0143) (0.0154) (0.0148) (0.0192)

ln( ret167ret67 ) 0.2025**

(0.0855)UBRR 0.0045*** 0.0046*** 0.0044*** –0.0003

(0.0009) (0.0009) (0.0009) (0.0006)UnionDensity –0.0064*** –0.0054*** –0.0065*** –0.0043**

(0.0014) (0.0016) (0.0014) (0.0017)wcoord –0.0014 –0.0017 –0.0018 0.0021

(0.0064) (0.0064) (0.0063) (0.0029)EPL 0.0152 0.0135 0.0177* 0.0026

(0.0110) (0.0110) (0.0104) (0.0079)outputgap –0.0034* –0.0032* –0.0035* –0.0011

(0.0018) (0.0018) (0.0018) (0.0011)inflchange –0.0047** –0.0043** –0.0047** –0.0017

(0.0022) (0.0020) (0.0022) (0.0019)irate –0.0102 –0.0114* –0.0105 0.0062*

(0.0071) (0.0068) (0.0072) (0.0032)

R2 0.98 0.98 0.98 0.93F Test of Excluded instruments 8.444 6.828 5.976 6.391Shea Partial R2 0.109 0.090 0.0778 0.0430Kleibergen-Paap under-identification test 0.0000 0.0002 0.0009Kleibergen-Paap weak-identification test 8.44 6.83 4.12N 231 231 231 231

Table 7: First stage results. Robust standard errors in parentheses. Significance levels: ∗: 10%,∗∗: 5%, ∗ ∗ ∗: 1%. p values of Kleibergen-Paap rank LM test for under-identification test

30

References

[1] Algan, Y., P. Cahuc and A. Zylberbrg, 2002, Public employment and labour market per-formance, Economic Policy, 17(34), 7-66.

[2] Algan, Y., P. Cahuc and M. Sangnier, 2011, Efficient and Inefficient Welfare States’ IZADiscussion Paper No. 5445.

[3] Algan, Y. and P. Cahuc, 2006, Civic Attitudes and the Design of Labour Market Institu-tions: Which Countries Can Implement the Danish Flexicurity Model?’ CEPR DiscussionPaper No. 5489, February, Centre for Economic Policy Research.

[4] Algan, Y. and P. Cahuc (2010) Inherited Trust and Growth, American Economic Review,100(5): 2060-92.

[5] Arpaia, A. and G. Mourre (2010) ‘Institutions and performance in European labor markets:taking a fresh look at evidence’ Journal of Economic Surveys.

[6] Bassanini A., R. Duval (2009) Unemployment, institutions and reform complementarities:re-assessing the aggregate evidence for OECD countries Oxford Review of Economic Policy,25, 1.

[7] Bean, C., R. Layard, and S.J. Nickell (1986) The Rise in Unemployment: A multicountrystudy Economica, 53: S1-S22.

[8] Belot, M. and J. Van Ours, 2004, Does the recent success of some OECD countries inlowering their unemployment rates lie in the clever design of their labor market reforms?,Oxford Economic Papers, 56(4), 621-642.

[9] Bertola G., 2009, Fiscal Policy and Labor Markets at Times of Public Debt, CEPR Dis-cussion Paper 8 037.

[10] Blanchard, O. and J. Wolfers, 2000, The role of shocks and institutions in the rise ofEuropean unemployment: the aggregate evidence, Economic Journal 110/ 462: C1-C33.

[11] Blundell, R. and A. Shepard, 2012, Employment, hours of work and the optimal taxationof low-income families, Review of Economic Studies, 79(2), 481-510.

[12] Boone, J. and Bovenberg, L., 2004, The optimal taxation of unskilled labor with job searchand social assistance, Journal of Public Economics, 88(11), 2227-2258.

[13] Botero, J.C., Djankov, S., La Porta, R. and F. Lopez-de-Silanes, 2004, The regulation ofLabor, Quarterly Journal of Economics, 119, 1339-1382.

[14] Brunello G. and D. Sonedda, 2009, Progressive taxation and wage settino when unionsstrategically interact, Oxford Economic Papers, 59, 127-140.

[15] Butler J., P. Giuliano, and L. Guiso, 2009, The Right Amount of Trust, NBER WorkingPaper 15 344.

[16] Cahuc, P. and A. Zylberberg, 2004, Labor Economics, MIT Press, Cambridge, MA, US.

[17] Daveri, F., and G. Tabellini, 2000, Unemployment, growth and taxation in industrialcountries, Economic Policy, 15, 47-88.

[18] Devries P., J. Guajardo, D. Leigh and A. Pescatori, 2011, A New Action-based Dataset ofFiscal Consolidation, IMF Working Paper 11/128, International Monetary Fund.

31

[19] Diamond, P., 1982, ‘Wage determination and efficiency in search equilibrium’, Review ofEconomic Studies, 49, 217-227.

[20] Doerremberg, P. and A. Peichl (2010), ‘Progressive Taxation and Tax Morale’, IZA Dis-cussion Paper, No. 5378.

[21] Drazen, A., 2000, Political Economy in Macroeconomics, Princeton University Press,Princeton.

[22] Eichhorst, W., Feil, M. and P. Marx, 2010, Crisis, What Crisis? Patterns of Adaptationin European Labor Markets, IZA Discussion Paper, 5045.

[23] Eisa, N. and J. Liebman, 1996, Labor supply responses to the earned Income tax credit,Quarterly Journal of Economics, 111(2), 605-637.

[24] Falk, M. and Koebel, B., 2001, A Dynamic Heterogeneous Labour Demand Model forGerman Manufacturing, Applied Economics, 33, 339-348.

[25] Fitzenberger, B. (1999), Wages and Employment Across Skill Groups. An Analysis forWest Germany, ZEW Economic Studies, 6, Physica, Heidelberg.

[26] Fuest, C. and B. Hubert, 2000, Is tax progression really good for employment? A modelwith endogenous hours of work, Labour Economics, 7, 79-93.

[27] Freier, R. and V. Steiner, 2007, Marginal Employment and the Demand for HeterogenousLabour: Empirical Evidence from a Multi-Factor Labour Demand Model for Germany,IZA Discussion Paper 2577.

[28] Gali, J. and R. Perotti (2003) ”Fiscal Policy and monetary integration in Europe”Economic Policy, October 2003, pp.533-572.

[29] Giavazzi, F., Schiantarelli, F. and M. Serafinelli, 2009, Culture, Policies and Labor MarketOutcomes, NBER Working Paper 15 417.

[30] Griffith, R., Harrison R. and G. Macartney, 2007, Product market reforms, labour marketinstitutions and unemployment, The Economic Journal, 117, C142-C166.

[31] Guichard, S., M. Kennedy, E. Wurzel and C. Andre, 2007, What Promotes Fiscal Con-solidation: OECD Country Experiences, OECD Economics Department Working Paper,No. 553.

[32] Guiso, L., Sapienza, P., and L. Zingales, 2008, Long term persistence, NBER Workingpaper 14 278.

[33] Hamermersh, D., 1993, Labor Demand, Princeton University Press.

[34] Hansen, C. T., 1999, Lower tax progression, longer hours and higher wages, ScandinavianJournal of Economics, 101, 49-65.

[35] Hansen, C., Pedersen, L and. Sløk, 1995, Progressive taxation, wage setting and unem-ployment: theory and swedish evidence, Swedish Economic Policy Review, 2, 423-460.

[36] Heckman, J., 1993, What Has Been Learned about Labor Supply in the Past TwentyYears?, American Economic Review, 83, 116-121.

[37] Hersoug, T., 1984,Union Wage Responses to Tax Changes, Oxford Economic Papers, 36,37-51.

32

[38] Holmulund, B. and A. Kolm, 1995, Progressive taxation, wage setting and unemployment:theory and swedis evidence, Swedish Economic Policy Review, 2, 423-460.

[39] Holm-Hadulla, F. Leiner-Killinger, N., and M. Slavik, 2011, The response of labour taxa-tion to changes in government debt, ECB Working Paper, No. 1307, March.

[40] osios, A.,1990, On the Efficiency of Matching and Related Models of Search and Unem-ployment, Review of Economic Studies, 57, 279-298.

[41] Hungerbuhler, M., E. Lehmann, A. Parmentier and B. Van der Linden, 2006, Optimalredistributive taxation in a search equilibrium model, Review of Economic Studies, 73,743-767.

[42] Immervoll, H., Kleven, H., Kreiner, C. T. and Saez, E., 2007, Welfare reforms in Europeancountries: a Microsimulation analysis, Economic Journal, 117, 1-44.

[43] Juhn, C., K.M. Murphy, R.H. Topel, J.L. Yellen and M.N. Baily, 1991, Why has theNatural Rate of Unemployment Increased over Time?, Brookings Papers on EconomicActivity, 75-142.

[44] Koskela, E. and R. Schob, 2007, Tax progression under collective wage bargaining andindividual effort determination, CESifo Working Paper 2024.

[45] Koskela, E. and R. Schob, 2009, Is tax progression good for employment? Efficiency wagesand the role of the pre-reform tax structure, FinanzArchiv: Public Finance analysis, 65,51-72.

[46] Lehmann, E., A. Parmentier and B. Van der Linden, 2011, Optimal income taxation withendogenous participation and search unemployment, Journal of Public Economics, 95,1523-1537.

[47] Lockwood B. and Manning A., 1993, Wage Setting and the Tax System: Theory andevidences for the United Kingdom, Journal of Public Economics, 52, 1-29.

[48] Lockwood, B., Sløk, T. and T. Tranaes, 2000, Progressive taxation and wage setting. someevidence for Denmark, Scandinavian Journal of Economics, 102, 707-723.

[49] Malcomson, J. and N. Sartor, 1987, Tax push inflation in a unionized labor market, Eu-ropean Economic Review, 31, 1581-1596.

[50] Manning, A., 1993, Wage bargaining and the Phillips curve: The identification and spec-ification of Aggregate wage equations, Economic Journal, 103(416), 98-118.

[51] Meghir, C. and Phillips 2010, Dimensions of Tax Design, in The Mirrlees Review, Institutefor Fiscal Studies (eds), Oxford University Press, Oxford, UK.

[52] Mortensen, D. and Pissarides, C. 1999, New developments in models of search in theLabor Market, in O. Ashenfelter and D. Card (eds.), Handbook of Labor Economics, vol 3B, North-Holland, Amsterdam.

[53] Musgrave, R. A. and Thin, T. 1948, Income tax progression, 1928-48, Journal of PoliticalEconomy, 56, 498-514.

[54] Nickell, S.J. and R. Layard, 1999, Labour market institutions and economic performance,in: Ashenfelter, O. and Card, D. Handbook of Labor Economics, Amsterdam, North Hol-land, 3029-3084.

33

[55] Nickell, S.J., L. Nunziata, and W. Ochel, 2005, Unemploment in the OECD since the1960s. What do we know?, The Economic Journal, 115, 1-27.

[56] Nunziata, L., 2005, Institutions and Wage Determination: a Multi-country Approach,Oxford Bulletin Of Economics And Statistics, 67, 435-466.

[57] OECD Taxing Wages, 2009, Special feature: consumption taxation as an additional burdenon labour income Organization of Economic Co-operation and Development.

[58] OECD (2012), OECD Tax database, Organization of Economic Co-operation and Devel-opment.

[59] Pagano, M., and P. Volpin, 2006, The Political Economy of Corporate Governance, Amer-ican Economic Review, 95, 1005-1030.

[60] Parmentier, A., 2006, The effects of the marginal tax rate in a matching model withendogenous labor supply, IRES Discussion Paper 2006− 11.

[61] Pesaran, M., 2006, Estimation and inference in large heterogeneous panels with a multi-factor error structure, Econometrica, 74(4), 967-1012.

[62] Persson, T. and G. Tabellini, 2000, Political Economics: Explaining Economic Policy,MIT Press, Cambridge, MA, US.

[63] Pissarides, C. A., 1985, Taxes, Subsidies and Equilibrium Unemployment,Review of Eco-nomic Studies, 52, 121-34.

[64] Pissarides, C. A., 1998, The impact of employment tax cuts on unemployment and wages:the role of unemployment benefits and tax structure, European Economic Review, 42,155-183.

[65] Pissarides, C. A., 2000, Equilibrium Unemployment Theory, Second Edition, MIT Press,Cambridge, MA, US.

[66] Røed, K. and S. Strøm, 2002, Progressive Taxes and the Labour Market: Is the TradeoffBetween Equality and Efficiency Inevitable?, Journal of Economic Surveys, 16, 77-110.

[67] Romer, Christina D., and David H. Romer, 2010, The Macroeconomic Effects of TaxChanges: Estimates Based on a New Measure of Fiscal Shocks, American Economic Re-view, 100 (3), 763-801.

[68] Sapienza, P., A. Toldra and L. Zingales, 2007, Understanding Trust, NBER Working paper,13 387.

[69] Sørensen, P. B., 1997, Public finance solutions to the European unemployment problem?,Economic Policy, 25, 223-64.

[70] Sørensen, P. B., 1999, Optimal Tax Progressivity in Imperfect Labour markets, LabourEconomics, 6, 435-52.

[71] Summers L., Gruber J. and R. Vergara, 1993, Taxation and the structure of labor markets:the case of corporatism’ The Quarterly Journal of Economics, 108, 385-411.

34