Content uploaded by Guido Tabellini
Author content
All content in this area was uploaded by Guido Tabellini on Jan 03, 2018
Content may be subject to copyright.
Unemployment, Growth and Taxation
in Industrial Countries
by
Francesco Daveri and Guido Tabellini
(University of Brescia and IGIER) (Bocconi University and IGIER)
November 10, 1997
This paper is part of a World Bank research project on growth and labor markets. We thank the World
Bank for financial support. We are grateful for comments to Giuseppe Bertola, Tito Boeri, Mike Burda,
Carlo Favero, Andrea Ichino, Otmar Issing, John Kennan, Francesco Lippi, Alessandro Missale, Stephen
Nickell, Alessandro Penati, Torsten Persson, Martin Rama and Xavier Sala-i-Martin, as well as to the
participants in the 1997 NBER Summer Institute, the 1997 EEA Meeting, the ESF Conference ‘Growth in
closed and open economies’, the European Summer Institute in Berlin, the conference on “Dynamic
Models of Economic Policy” at the University of Rochester, and in seminars at the IMF, Paris, ULB,
Brescia, IGIER, and Trento. We are also indebted to Anna Fruttero for excellent research assistance, and to
Gian Maria Milesi-Ferretti and Gilles Saint-Paul for giving us their data on tax rates and unemployment
benefits.
1
1. Introduction
The most pressing economic problem in Europe today is the apparently endless surge in
unemployment. Other industrial countries have also seen an increase in unemployment
over time, though to a smaller extent. To the layman, the upward trend in unemployment
is related to the slowdown in economic growth, which is also apparent in most industrial
countries and in particular in Europe. Both trends are illustrated in Table 1 and Figure
1. Given the long time periods involved, these two trends are not simply the result of
business cycle fluctuations, but reflect long run tendencies. The observed negative
relation between long run growth and unemployment is at odds with the economist’s
shared opinion that the natural rate of unemployment is invariant to productivity growth1.
Despite a very large literature on growth and unemployment, few papers have jointly
studied these two phenomena; none has made a systematic effort to come to grips with
the mentioned evidence from industrial countries2.
This paper argues that the layman’s view is correct and compatible with the
economist’s view. The observed increase in unemployment and the slowdown in
economic growth are related, because they stem from a common cause: an excessively
high cost of labor. If labor markets are non-competitive, an exogenous and permanent
increase in labor costs has two effects: on the one hand, it reduces labor demand, and
thus creates unemployment. On the other hand, as firms substitute capital for labor, the
marginal product of capital falls; over long periods of time, this in turn diminishes the
incentive to accumulate and thus to grow. Then high unemployment is associated with
slow growth. Whether the growth slowdown is permanent or transitory depends on the
productive technology. In the framework of endogenous growth theory, this can be a
permanent effect, while in the neoclassical model of growth this is a transitory
1 This result was obtained by Phelps (1968) and more recently restated in Gordon (1997) and Blanchard
(1996, ch.25).
2 The main issues on European unemployment are recently surveyed in Bean (1994), Alogoskoufis et alii
(1995), and the OECD Jobs Studies (1994), but little regard is given there to the growth-unemployment
link. Search theory, as summarized in Pissarides (1990, ch.2), has made a theoretical case that growth and
unemployment are negatively related in the long run. As exogenous technical change drives productivity
up, the rate of return on the creation of job vacancies rises, which accelerates the exit rate from
unemployment. Aghion and Howitt (1994) allow for search unemployment in their model of growth
through creative destruction. More rapid growth shortens the average length of a given job match, thus
increasing job separation and reducing job finding rates. Depending on whether the ‘capitalization effect’
or the ‘creative destruction effect’ prevail, the growth-unemployment relation may be negative or positive.
2
phenomenon3. The surge in unemployment, on the other hand, is permanent. Exogenous
or endogenous labor productivity growth feeds into higher real wages and unemployment
subsidies in equilibrium. Thus productivity growth does not affect the natural rate of
unemployment, as argued by Phelps. There is nothing very profound nor very surprising
in these arguments. Yet, often the simplest explanations are also the best explanations.
In Europe, labor costs have gone up for many reasons, but one is particularly easy
to identify: higher taxes on labor. As shown in Table 2, labor taxes have gone up in
almost every country and in almost every decade. If labor markets are competitive, the
small elasticity of individual labor supply implies that the burden of a tax on labor
income is borne almost entirely by the worker, with little effect on unemployment and
the capital-labor ratio. But if workers are organized in monopolistic unions, they can
succeed in shifting the burden of labor taxes onto firms. In this case, a rise in labor taxes
permanently increases unemployment, and it (permanently or temporarily) increases the
capital-labor ratio, reduces the rate of return on capital and slows down economic
growth.
This paper formulates a simple OLG growth model with a unionized labor
market, which makes this argument more precise. The theory is then used as a basis for
identification, and its predictions are investigated against the evidence of a panel of 14
OECD countries over the period 1965-1991, contrasting Europe and other industrial
countries.
The variety in labor market institutions across countries is useful, because the
negative effect of labor taxation on employment and growth is expected to be much more
pronounced in Europe, where labor markets are clearly dominated by large trade unions.
This is what we find in the data on labor taxes and unemployment depicted in Figures 2
and 3. The difference between Europe and the other industrial countries is striking: the
high positive correlation between tax rates on labor income and unemployment is clearly
a European phenomenon, not present elsewhere.
This difference between Europe and the rest of the industrial world is also
reflected in the behavior of the capital-labor ratio. The IMF estimates that between 1970
and 1995 the capital-labor ratio more than doubled in the European Union, whereas it
See also Mortensen and Pissarides (1997).
3 Furuya (1995) has recently studied the implications of efficiency wages in a Solow growth model,
showing that a negative relation may arise along the transition to the steady state.
3
only rose by 25% in the US. The Summers-Heston data on capital stocks deliver a
similar message. Between 1965-70 and 1986-91 the average capital stock per employee
increased by approximately 65%, 70% and 100% in the US, Australia and Canada
respectively. In the same period, the cumulative growth of capital per employee reached
130%, 150%, 175% and 280% in Italy, France, Germany and Spain, even though their
growth of per-capita GDP was very similar to that of the Anglo-Saxon countries. 4
A crucial step in our argument is that the effect of higher labor taxes on
unemployment is due to higher real wages. This prediction is also strongly supported by
the data: in the paper we show that higher tax rates on labor are indeed shifted onto
higher gross wages in Europe, but not in the other OECD countries. Finally, the paper
shows that, as expected, the capital-labor ratio is strongly and robustly correlated with the
increase in real wages.
The policy implications of this view are straightforward but extremely relevant.
In a world with monopolistic trade unions, labor taxes can be as distorting and harmful to
growth as capital taxes. According to the data, the effective tax rate on capital in Europe
was on average equal to 22.7% in 1965-75 and reached 32.9% in 1976-91. In the same
group of countries, the average effective tax rate on labor rose from 31.1% on average in
1965-75 to 40.5% in 1976-91, reaching 43.2% in 1986-91. Moderating the overall level
of taxation and mostly of taxes on labor is one of the main challenges currently faced by
the European Union.
Our empirical results provide a quantitative evaluation of how costly higher labor
taxes have been in Europe, in terms of reduced growth and higher unemployment. In
spite of the variety of methods of estimation and instruments we used, all of our
estimates produce similar quantitative answers. The rise of almost 10 percentage points
in the rate of effective labor taxes can account for a 4 percentage points increase in
European unemployment. Moreover, the rise of labor taxes is also associated to a
reduction of the EU growth rate of about 0.4 percentage points a year - a respectable
loss indeed if considered over a time horizon of about fifteen years, and about one third
of the observed reduction in growth between 1965-75 and 1976-91. In contrast, the
4 As noted by Blanchard (1997), the labor share of income fell in continental Europe but not in other
industrial countries over this same time period. This paper does not attempt to address this other stylized
fact, even though it is not inconsistent with our proposed explanation of the European unemployment
problem. An increase in labor taxes causing higher unemployment can also reduce the labor share of
4
equally sizable rise in capital tax rates (about 10 percentage points between 1965-75 and
1976-91) appears to have slowed down annual growth rates in the EU by a bare 0.06% a
year. But while the small effect of capital taxes on growth is a finding already known to
growth theorists5, the sizable growth effect of labor taxes is novel and lends itself to be
further discussed with reference to other periods and groups of countries.
If the explanation is so simple and straightforward, why have so few papers
pointed out that labor taxes are a key determinant of European unemployment ? As
reported by Blanchard and Katz (1996, p.24), ‘the role of taxes was a main focus of the
multi-country study organized by Richard Layard and Steve Nickell in the mid-1980s.
(...) Yet, the cross-sectional evidence within Europe does not reveal much correlation
between tax rates and unemployment rates, nor between changes in tax rates and changes
in unemployment’. Our study confirms these previous findings in the cross-section of 14
OECD countries. Cross-sectional variations in unemployment rates tend to be dominated
by fixed effects at the country level. This is not too surprising: as documented for
instance by Nickell (1997), labor market legislation differs markedly across countries,
and such legislation did not change much after the late 1960s or early 1970s. Labor taxes
significantly predict how unemployment changes over time within each country,
however. Moreover, this correlation is strong and evident among the highly unionized
countries of Continental Europe, and much less so in countries with competitive labor
markets, and in the Scandinavian countries with highly centralized trade unions. Thus,
the correlation between labor taxes and unemployment is only captured by
simultaneously exploiting the time series and cross-country variations of the data, and by
distinguishing among countries on the basis of their labor market institutions. This
distinction as well as the emphasis on time series (as opposed to cross country)
correlations was missing in many previous studies on European unemployment.
Some of the ideas in this paper are clearly related to Bruno and Sachs (1985) and
Phelps (1994). The theoretical analysis in both books was not cast in terms of modern
growth theory, however. They had more ambitious goals, and their analysis also focused
on business cycle phenomena and on international linkages. The empirical analysis was
also different from that of this paper, in the set of economic variables considered and
income if the elasticity of substitution between capital and labor is sufficiently high.
5 As conjectured by Harberger (1964), Lucas (1990) calculates that eliminating capital taxes in a revenue-
neutral way would produce no change in the long-run growth rate of the US economy.
5
because countries were not grouped according to their labor market institutions. Finally,
Alesina and Perotti (1994) found a relation between labor taxes and unit labor costs in a
sample of OECD countries, using data on tax rates different from ours, but also
discriminating between countries on the basis of the role of trade unions. Alesina and
Perotti (1994) mainly focus on competitiveness, but the approach of our paper is
consistent with many of their results.6
The paper outline is as follows. Section 2 presents the model. In equilibrium,
growth and employment are decreasing functions of labor taxes. Section 3 looks at the
evidence. Section 4 concludes the paper.
2. The Theory
In this section, we present an OLG growth model where equilibrium unemployment is
caused by monopolistic trade unions. Depending on the assumptions about technology,
growth could either be endogenous, or just a feature of the transition to the steady state.
In both cases, there is a negative relation between unemployment, on the one hand, and
the rate of return on capital and the growth rate, on the other. This relation could be
permanent or transitory, depending on the productive technology. The OLG structure is
not central to our results and the seemingly myopic behavior of the monopoly union,
which only cares about today members’ incomes, is a simple shortcut for a more general
setting where the union is infinitely lived but cannot commit to future courses of action.
2.1 A Simple Model
Consider a two-period overlapping generations economy with constant population and
closed to international trade. Individuals are non-altruistic and have a homothetic and
separable utility function defined over consumption when young and old, c and d
respectively.
Only young individuals can work. They can be employed or unemployed. If
employed, they earn wage income net of taxes, w(1-
τ
L), where
τ
L is the labor income tax
rate. If unemployed, they earn an unemployment subsidy, s, which is not taxable.7
6 Other more specific papers investigating the empirical evidence on wages, unemployment and taxation are
Padoa Schioppa (1990), Tyrvainen (1994), and Tullio (1987). Bean (1994) and OECD (1994) surveyed
this strand of literature.
7 Alternatively, a slight modification of the model could lead to an alternative interpretation : s could
6
Individual labor supply is set equal to one. A given fraction λ ≤ 1 of young individuals
belongs to a trade union. We treat union membership as parametric, whereas the number
of employed individuals is endogenous. If the number of union members is larger than
the number of employed individuals, then we assume that all employed individuals are
also union members8. Old individuals earn a return (net of taxes) of [1+(1-
τ
K)r] on their
investments when young, where
τ
K is the tax rate on capital, and r is the rate of return on
capital. Physical capital is the only outside asset and, for simplicity, it is subject to no
depreciation; savings by the young are denoted by k’.
Under these assumptions, the individual budget constraints of a young and old
individual respectively are :
ykc
iii
=+'(1)
[]
11+− =()'
τ
Kii
rk d (2)
where
i stands for either employed (E) or unemployed (U), and yE = w(1-
τ
L), yU = s. By
homotheticity, the amount saved by a young individual with income yi is :
k’i = yi F[r(1-
τ
K )] (3)
where F(.) is a known function, whose properties depend on underlying preferences. We
assume F(.) to be increasing (i.e. the substitution effect prevails over the income effect).
Let l ≤1 be the fraction of employed individuals in the (young) population, and k
be the average holdings of capital by those currently old. Then we can write the
government budget constraint in per capita terms as:
ττγ
KL
rk wl l s
+=+−()1(4)
where γ denotes per capita government consumption. In a growing economy, fiscal policy
variables also tend to grow over time. To preserve balanced growth, we assume that both
per capita public consumption and unemployment subsidies are given linear functions of
average per capita income y (to be defined below). Thus: γ =
δ
y ; s=
σ
y. In the OECD
data analyzed in section 3, the replacement ratio (i.e. the parameter σ) is constant or
increasing over time, but certainly not decreasing. The parameters
σ
and
δ
, together with
the tax rates, are the policy variables controlled by the government. To simplify the
denote the wage rate in the underground economy or in a non-unionized sector where taxes can be evaded
more easily. In most countries, unemployment subsidies are part of taxable income, but the effective tax
rate is much lower both because of deductions on low income households and of payroll taxes on employed
workers.
7
comparative statics, we assume that the parameter
δ
is adjusted by the government
whenever tax rates change, so as to keep σ constant.
Average capital per worker evolves over time according to the following law of
motion:
klk lk
EU
''()'
=+−
1(5)
where k’ denotes average holdings of capital in the next period.
Production takes place in a large number of identical competitive firms, endowed
with the following technology:
y=
φ
(k)l1−
α
(6)
where y is the level of output (or income) divided by the labor force N, l is the
employment rate, k is capital at the beginning of the period also divided by N, and
φ
(.) is
a concave and increasing function to be specified in subsection 2.3 below. Thus, we
restrict labor demand elasticity to be constant. This simplifies the analysis in that
equilibrium employment remains constant over time. This formulation encompasses
different growth models, such as the neoclassical growth model and a variety of
endogenous growth models, depending on the specification of the function
φ
(.). All
these models have the similar implications for the equilibrium, except on one dimension,
namely how lasting are the effect of taxes on growth and investment: whereas such
effects are temporary in the neoclassical growth model, they are permanent in a variety of
endogenous growth formulations. We discuss these issues in subsection 2.3 below.
Competitive firms hire labor and capital up to the point where the respective
perceived marginal product equals the relevant input price. In particular, inverting the
firms first order conditions with respect to labor, we obtain that labor demand (i.e. the
employment rate) is a function L(k,w) defined by:
l = [(1-
α
)
φ
(k) / w ] 1/
α
≡ L( k,w) (7)
Similarly, the perceived rate of return to the firm own capital is the partial derivative of
the right hand side of (6) with respect to k, namely:
r =
φ
k (k) l1-
α
(8)
Thus, quite intuitively, for any given value of k the rate of return on capital is an
increasing function of the employment rate: when existing capital is combined with
8 This assumption simplifies the exposition but is not central to our results.
8
more labor, it is more productive. Naturally, the stock of capital is endogenous too, and
how that evolves over time depends on the assumptions about
φ
(.).
2.2 Labor Market Equilibrium
Wages are set by monopolistic trade unions. We assume that the union is large enough to
be able to set wages, but small enough to take fiscal policy variables and the interest rate
as given. Thus, unions operate at the firm or sector level. Under this assumption, the
welfare of the current old cannot be affected by the union. Moreover, the union affects
the welfare of the current young only through their current income. Thus, the union faces
a static optimization problem. To simplify the analysis, we neglect risk aversion and
postulate that unions maximize the expected income (rather than the expected utility) of
their young members, subject to the labor demand function (7). That is, in each period
the union sets wages so as to maximize in an utilitarian fashion:
Lkw wLkw s
L
(, ) () (, )
λτλλ
1−+
−(9)
The union’s optimization problem in equation (9) generalizes to an infinite horizon
setting, where the union cannot commit to future policies. Under our assumption that all
employed individuals belong to a union, the ratio l/λ is the fraction of union members
that finds a job. Thus, the first term in (9) is the net wage times the probability of finding
a job, while the second term is the unemployment subsidy times the probability of being
unemployed.9
Given our production function, the optimal net wage for the union is a constant
mark-up over the unemployment subsidy . Specifically, maximization of (9) with respect
to the wage and taking the subsidy s as given implies :
wsL
=−−()( )11
ατ
(10)
This formulation of the trade union wage setting problem is standard. Firms have the
right to hire as many workers as dictated by the perceived labor demand curve, at the
wage level preset by the monopolistic union. In general, this implies some equilibrium
9 The assumption that trade unions maximize expected income (and not expected utility) of their members
can also be interpreted as saying that there is an insurance scheme within the union against the risk of being
unemployed.
9
unemployment. We do not try to address the known weaknesses of union models10:
anyway, the main conclusions of our model generalize to other labor market
imperfections.11 The key feature of labor markets we are committed to discuss is that
labor market institutions at the country level play a role in determining whether a relation
between labor taxes, unemployment and growth shows up in the data. This feature is
parsimoniously embodied in our model.
Given the equilibrium wage rate, it is straightforward to compute equilibrium
employment by combining (4), (7) and (10), to obtain :
[]
lL
*( )( )/
=− −
11
2
ατσ
(11)
where a ‘*’ denotes equilibrium variables and
σ
is the replacement rate in the
unemployment subsidy - recall our earlier assumption that s =
σ
y. Thus, as expected, a
higher tax rate is met by a higher gross wage, which in turn forces firms to cut
employment. Moreover, a higher replacement rate (a higher
σ
) also reduces employment,
because it leads to a proportionate increase in gross wages. It is easy to show that in
equilibrium there is a positive rate of unemployment (that l* < 1).12 Finally, for a
constant tax rate and a constant
σ
, equilibrium employment is also constant. This is a
special feature due to the production function we have adopted.
2.3 Equilibrium Growth
We can now combine all the pieces to compute the equilibrium growth rate. By (5) and
(3), the new average capital stock evolves according to:
[]
kFr lw ls
KL
'[( )]( )()
=− −+−
111
ττ
(12)
Hence, by (8), (10) and (12), the equilibrium growth rate, g*(k), is:
kk
lllkFkkkg k
k)(
*]
1
*1)][1(*)([/')(*1 11
φ
σ
α
α
τφ
αα
−− −
+−=≡+ (13)
Thus, at any moment in time and for any given initial capital stock, equilibrium growth is
10 To name a few, one may ask where the market power of unions originates from, why so little wage
underbidding on the part of the unemployed is observed, and why unemployment exists even when unions
are absent. Lindbeck (1993) summarizes many of these questions.
11 Furuya (1995) derived implications reminiscent of ours in a model with Solow technology and efficiency
wages.
12 We assume here for convenience that l* < λ. If λ is sufficiently small that l* = λ, then from then on it
remains true that l = λ for all parameter values that would otherwise increase employment, and the wage
rate is no longer determined by (10) but instead by (7) together with the condition that l = λ.
10
higher the greater is the employment rate. The effect of employment on growth occurs
through two distinct channels. First, from equation (8), higher employment increases the
marginal product of capital, as captured by the term l1-
α
φ
k(k) , and this in turn induces
more savings by the young. Second, higher employment increases the average income of
young individuals, which, in turn, in a life-cycle model, leads to more savings. This is
captured by the term l1-
α
φ
(k)/k at the very end of the right hand side of (13). The second
channel is a special feature of an overlapping generations economy in which individuals
only work when young, and it would disappear if work were uniformly spread throughout
one’s lifetime. But the first channel is very robust and intuitive. Higher employment
implies a lower capital-labor ratio, and hence a more productive capital stock ; this in
turn fosters investment and hence stimulates growth.
How exactly equilibrium growth varies over time depends on the functional form
of
φ
(k). Take a Cobb-Douglas production function, as in Diamond (1965), where:
φ
(k)=Bk
α
, so that
φ
k(k)=
α
Bk
α
-1 and
φ
(k)/k =Bk
α
-1 , where B>0 is a constant parameter.
Then, the growth rate equation (13) specializes to:
1111 *]
1
*1)][1(*[/')(*1 −−−− −
+−=≡+
αααα
σ
α
α
τα
BklllBkFkkkg k(14)
Here g* (k) converges asymptotically to 0 (the right hand side of (14) converges to 1), as
k converges to the steady state. In this neoclassical set up, an exogenous reduction in
employment has a negative but transitory effect on growth. The fall in employment
induces firms to reduce investment. If
φ
(.) is concave, the smaller capital stock increases
the rate of return on capital until the growth effect vanishes and the economy is back to a
lower steady state level of output.
At the opposite extreme, suppose that
φ
(k) =Ak , so that
φ
k(k)=A=
φ
(k)/k, where
A>0 is a constant parameter. The corresponding growth rate is:
AllAlFkkkg k
αα
σ
α
α
τ
−− −
+−=≡+ 11 *]
1
*1)][1(*[/')(*1 (15)
Here it is easy to see that if A is sufficiently large, then the equilibrium growth rate is
positive and constant over time, and it is permanently affected by changes in the
employment rate. As employment is reduced, the marginal product of capital falls. Firms
scale down investment, but here the marginal product of capital is not affected by this
lower investment rate. Hence the growth effect is permanent. Models with AK
11
technologies have been extensively studied in the endogenous growth literature, and the
technological assumptions that give rise to them are well known. A specific and simple
example is Romer (1986, 1989), where each firm faces a Cobb-Douglas production
function in capital and labor with constant returns to scale, but there is an externality in
average capital. To generate the result that growth is permanently affected by the
employment rate, so that in equilibrium
φ
(k) =Ak, the externality should be defined over
capital per worker (i.e. per member of the labor force) or per individual, rather than over
capital per employee.13 A second and widely studied example concerns economies with
endogenous technological progress and imperfect competition, such as in Romer (1987),
(1990), or Bertola (1993). Here too, the equilibrium aggregate production function for
the final good is linear in average capital per individual, and changes in the economy-
wide employment rate have a permanent growth effect.
Intermediate cases are also possible. Take for instance the production function
suggested by Jones and Manuelli (1990), where
φ
(k) = Ak + Bk
α
. Here
φ
k(k) =A + Bk
α
-1
and
φ
(k)/k =A +
α
Bk
α
-1. Thus, equilibrium growth is not constant at any instant, but only
asymptotically, as k approaches infinity. Here changes in the employment rate have both
permanent and transitory effects on equilibrium growth. Moreover, the model now
predicts conditional convergence in growth rates: poorer countries experience
temporarily higher growth rates.
Summarizing, the - temporary or permanent - link between employment and
output growth is a key feature of many growth models. Oddly enough, this feature, which
plays such a critical role in business cycle analysis, has been neglected in the theory of
economic growth, because typical growth models assume competitive labor markets and
full employment. But once employment is regarded as endogenous, the variables
determining employment also affect growth, with important policy implications.
13 Specifically, we can write the production function faced by each firm as:
ααα
−−
=11 LKAkQ
where Q denotes total output produced in a typical firm, L and K denote the inputs of labor and capital
hired in that firm and k is average capital per worker (i.e. member of the labor force). Dividing both terms
by N and simplifying, we obtain that in equilibrium average output can be written as in (6), with
φ
(k) = Ak.
Note however that here
φ
k(k) =
α
A < A because of the externality. The actual presence of external effects in
production is a debated issue in macroeconomics. Caballero and Lyons (1992) provide evidence in favor of
the presence of production externalities, of a smaller size than 0.66 (the available best guess for 1-
α
),
however. Other studies reach opposite (Burnside, 1996) or mixed (Basu and Fernald, 1995) conclusions.
12
Consider in particular the consequences of higher tax rates.14 To focus on one
policy intervention at a time, we assume that the government consumption share (the
parameter δ defined above) is appropriately adjusted so as to keep the government budget
balanced at any instant of time. Then a higher capital tax rate,
τ
K, reduces the net rate of
return on investment and equilibrium growth.15 This growth-reducing effect motivates
the common view that capital taxes are highly distorting and, from an efficiency point of
view, should be avoided as much as possible (see for instance Rebelo (1991)). But in our
model, labor taxes have a similar effect: by (11) and (13), a higher
τ
L reduces equilibrium
employment and, through this channel, it also reduces growth. Which tax rate is more
harmful for growth is ambiguous, and depends on parameter values. But if tax rates are
the same at the outset, in this model a rise in labor taxes is unambiguously costlier to
growth than a rise in capital taxes.16 Naturally, taxes on labor also increase
unemployment.
2.4 Extensions
Our barebones model can be extended in several ways. Some of them can be easily
adressed, others would require more careful work. We discuss here some of the more
important and fruitful extensions.
2.4.1 Human Capital Accumulation
One might wonder what happens to our results if human - rather than physical -capital is
the actual engine of growth. In a nutshell, the learning enabled by the process of human
capital accumulation opens up another route for labor taxes to directly affect growth.
This has already been studied by Stokey and Rebelo (1995) and Jones, Manuelli and
Rossi (1993), under the assumption of competitive labor markets. One implication not
fully developed in these studies - and instead discussed by Lucas (1993) and Mendoza,
Milesi-Ferretti and Asea (1996) - is that this effect crucially depends on whether human
14 Following the tradition of the growth literature, we discuss the effects of tax rates on growth disregarding
the issue of their timing and treating all changes in tax rates as permanent and unexpected.
15 Exceptions are those models, like Lucas (1988), where growth is due to human capital accumulation
and physical capital does not enter the production function of human capital. Models of this kind are
discussed at some length in section 2.4.1.
16 The intuition underlying this result is as follows. If tax rates are the same at the beginning, the effects of
higher taxes on the rate of return on capital are approximately equal. But the adverse growth effects of the
tax on labor are augmented by their negative impact on savings.
13
capital accumulation takes place through formal or informal channels.
If human capital accumulation occurs formally through, say, schooling or formal
training, then employment today reduces the amount of time potentially devoted to
human capital accumulation. The positive growth effect of increased employment on the
marginal productivity of both types of capital is thus counteracted by this negative effect.
The overall effect of unemployment on the growth rate becomes possibly nil.
If instead human capital is accumulated as a result of learning on the job, then
working today causes one to be more productive tomorrow through a standard learning-
by-doing effect (in our model, this would have to be an intergenerational effect). In this
case, the negative link between unemployment and growth would be reinforced rather
than weakened. If labor markets are competitive, then the effect of labor taxes on growth
is the same across countries. But with non-competitive labor markets, the effects of labor
taxes on growth differ across countries, and are more pronounced in the presence of
monopolistic unions.
A simple formalization of on-the-job learning compatible with our model draws
on Lucas (1993). Suppose that human capital is unintentionally accumulated on the job
according to the following linear law of motion:
h’ = Alh
where h is the amount of human capital (ability) a young is endowed with at birth, which
is then bequeathed to the next generation of young (h’) . In turn, human capital enters the
production function of final goods, augmenting labor effort as follows:
y = B k
α
(lh)1-
α
It can be readily shown that we are in an ‘Ak’ economy where all level variables, except
for l, are now linear in h in equilibrium. From the equation for human capital
accumulation, the equilibrium growth rate of the economy is clearly an increasing
function of the employment rate17. In turn, the employment rate is constant and depends
on the usual fiscal and technology parameters.
2.4.2 Large and Centralized Trade Unions
Union models tend to blame the ‘excessive‘ size or power of unions for the existence of
17 The equality between the growth rates of human and physical capital is guaranteed by instantaneous
variations in the ratio between the two stocks of capital, like in Barro and Sala-i-Martin (1995, ch.5).
14
unemployment. Monopoly unions do not take the wage as given, but push for higher
wages at the firm or sector level. This drives firms to ration employment along their labor
demand curve in order to maximize profits. As emphasized earlier, our basic model is no
exception to this practice. Our trade union is large enough to set wages, but not large
enough to negotiate over fiscal policy with the government.
Along the lines developed by Calmfors and Driffill (1988) and Summers, Gruber
and Vergara (1993), however, one may conjecture that if wage setting is centralized and
workers are represented by a very large trade union, then they are likely to develop a
more moderate attitude in negotiations. Centralized and large unions would internalize
the budgetary implications of unemployment subsidies, as well the adverse consequences
of an economy-wide drop in the capital-labor ratio, to a greater extent than medium-sized
unions that bargain in a decentralized fashion at the firm or sector level. Wages would
thus be set to a lower level than in the monopoly union model. Addressing this issue
satisfactorily would require a more realistic model of union behavior - something which
the existing literature has yet to settle on. But these considerations suggests that, when
looking at the empirical determinants of unemployment, countries might be partitioned in
more than just two groups (i.e competitive or unionized labor markets), so as to allow for
different patterns of wage negotiations by trade unions.
2.4.3 Open Economies and International Capital Mobility
In a small open economy with international capital mobility and only one good, the
capital-labor ratio and hence the gross real wage is pinned down in the long run by the
condition that the rates of return on domestic and foreign investment are the same. Some
economists have argued that, for this reason, labor taxes are not distorting in a small open
economy, because the tax burden is entirely borne by labor (Nickell (1997)). This view
neglects an important timing issue, however. International capital mobility equalizes the
expected rate of return. This in turn depends on future, rather than current, wages. When
unions set wages, they take existing capital as given. Therefore, wage setting behavior is
not affected by the extent of capital mobility. This is literally true in our overlapping
generations economy. It would continue to be true in a richer infinite horizon economy,
as a time-consistency argument would force the union to take future wages as given when
setting current wages. This is a version of the well known capital-levy problem,
15
extensively studied in public finance with regard to capital taxation (see Persson and
Tabellini (1997) and, for an application to union behavior, van der Ploeg (1987)). Hence,
even in a small open economy model with rational monopolistic trade unions, higher
labor taxes would be shifted onto higher wages in the short run.
Opening up the economy to capital mobility, however, would affect the long run
consequences of these higher labor taxes. Now capital would leave the country, on the
expectation of higher future wages. In the new long run equilibrium, the capital-labor
ratio would return towards a level consistent with the equalization of domestic and
foreign rates of return. But the adjustment to this new equilibrium would entail both
capital outflows as well as increased unemployment. Modelling precisely this would
require additional features to ensure the existence of equilibrium.18 Moreover, a
satisfactory model would entail more than one good. Pursuing this extension is therefore
beyond the scope of this paper. But the general point we want to stress is that opening up
the economy would not eliminate the distorting effects of labor taxes. On the contrary,
new and potentially worse distortions would arise. Alesina and Perotti (1994) address
some of these issues in a related framework.
2.4.3 Employment and Growth in Developing Countries
In our model, those that are not employed in the unionized sector remain unemployed.
The channel through which unemployment is damaging for growth hinges on the non-
competitive nature of the labor market. This may be appropriate for industrial countries
where trade unions are usually large and powerful organizations. In developing countries,
however, a large fraction of the working population is employed in the informal sector,
where labor is competitively supplied. Wage underbidding on the part of the competitive
fringe of informal workers is likely to bring wage rates down to competitive levels in
these countries. This tends to restrict the scope for applying our basic model.
The employment-growth link may apply to developing countries as well,
however, though the driving forces are not the same. Suppose, following Young (1993),
that the economy features a dual structure. Then modern sector firms face the same
production function as above and negotiate wage and employment with monopolistic
18 In particular, we would need a model in which higher unemployment moderates somewhat the union
demands, or in which capital outflows are not costless.
16
trade unions, while firms in the backward sector face competitive labor markets and only
use labor inputs (i.e. s is the wage in this other sector). In a companion paper under
preparation, joint with Martin Rama, we show that, as capital grows, so does
employment in the unionized sector. The reason is that now the wage in the unionized
factor remains fixed with a mark up over the constant subsistence wage, s. Thus, the term
w/k falls over time, and labor demand in the unionized sector grows as labor becomes
more productive, since capital continues to grow. Eventually, all employment is suck in
the unionized sector. At that point, the economy behaves as in the model of subsections
2.1-2.3 above. As documented by Young (1993), this pattern of development is very
consistent with the evidence of a number of countries in East Asia.
2.5 Empirical Implications
The theoretical results summarized in the previous sections can be contrasted with the
time series and cross section evidence of a panel of industrial countries. Our main goal is
to test the following theoretical propositions: (i) In countries with strong but
decentralized trade unions, higher labor taxes lead to higher unemployment and higher
gross real wages. (ii) A higher cost of labor per employee induces firms to substitute
capital for labor, in every country. (iii) This in turn reduces the return to capital
accumulation, slowing down investment and hence growth.
Our strategy is to take the model seriously as a basis for identification. The theory
helps because it has a simple recursive structure. In particular, we consider the
equilibrium conditions for employment and growth in (11) and (13), as well as the wage
setting and labor demand equations, (10) and (7) respectively. We are not committed to
the precise functional forms of these expressions, however. In particular, we do not
attempt to discriminate between endogenous versus exogenous growth models, and thus
we do not address the issue of whether the correlation between employment and growth
is permanent or temporary.
Consider the equilibrium employment equation first, and let uit = 1- lit be the
unemployment rate in country i and period t. Taking a linear approximation of (11),
unemployment can be written as a function:
uit =
β
0 +
β
1
τ
Lit +
β
2
σ
it +
β
3 xit + vuit (16)
where xit is a vector of other observable determinants of unemployment besides the labor
17
tax rate
τ
L and the replacement rate
σ
, not captured by our simple theoretical model, and
vuit is an unobserved error term. In the empirical analysis we formulate alternative
specifications of the vector xit, such as lagged unemployment, alternative measures of
union strength, inflation, various time trends, and the contemporaneous or lagged growth
rate of GDP per capita and productivity per worker. We also use more than one measure
of unemployment. These alternative specifications are discussed in section 3.
Our main focus is in the prediction that
β
1 > 0. In particular, we test whether
β
1
differs across groups of countries, in accordance to the size and importance of trade
unions. If the labor market is competitive and trade unions unimportant,
β
1 could be
close to 0. If trade unions play an important role in wage negotiations, but are not so
centralized as to internalize all the general equilibrium repercussions of higher wages, we
expect
β
1 to be large and positive. Finally,
β
1 could be smaller again if the trade union is
very large and centralized, because of the arguments discussed in the previous subsection
and elaborated at length by Calmfors and Driffill (1988). The theory also predicts that
β
2>0. It is not clear that the size of this coefficient should also depend on trade union
strength and centralization, since even in competitive labor markets higher replacement
rates could have a large effect on unemployment through individual search or bargaining
attitudes. For this reason,
β
2 is constrained to be equal across countries.
Next, consider the determinants of the equilibrium rate of growth in country i and
period t, git. Equation (13) yields the following linear approximation :
git = b0 + b1
τ
Kit + b2 uit + b3zit + vgit (17)
where now zit is a vector of observable variables not all captured by our model, but that
nevertheless can be expected to influence economic growth, and vgit is an unobserved
error term. In the empirical analysis, in some specifications we also replace the growth
rate of per capita GDP with the share of investment over GDP.19 The vector z includes
initial per capita income and the secondary enrollment ratio in all growth and investment
regressions. The inclusion of initial per-capita income is suggested by the prediction of
convergence implied by some formulations of the production function
φ
(k), as discussed
in the previous subsection. We also ask whether the resuls are robust to alternative
specifications, by also including an index of weekly hours actually worked or paid for,
19 This also serves the purpose of checking that the positive effect of the employment rate on growth does
not have to do with a movement along the production function.
18
and the share of population in working age over total population.
Here, the main implication we want to test is that unemployment adversely affects
growth and investment: b2<0. We restrict this coefficient to be equal across countries. A
tax on labor is expected to be costlier to growth and investment in countries with
powerful but decentralized unions, because of the induced greater labor market
distortion, not because of technological differences across countries. We are also
interested in the sign of the tax rate coefficient, b1. The basic model predicts that b1<0
for all countries equally. The version of the model augmented with human capital
accumulation, however, predicts b1=0.
Equations (16) and (17) consider the direct impact of taxes and other policy
variables on unemployment and growth. But according to the model, this impact occurs
through observable channels: on the one hand, higher labor taxes and replacement rates
lead to more unemployment because they induce trade unions to demand higher real
wages. On the other hand, higher real wages are detrimental for growth because they
induce firms to substitute capital for labor, thus raising the capital-labor ratio and
reducing the productivity of capital.
Consider the wage setting equation, (10). Taking a log-linear approximation and
then differencing, we obtain that the growth rate of gross real wages,
γ
w, is an increasing
function of the changes in unemployment subsidies and labor taxes. Since unemployment
subsidies are assumed to be s =
σ
y, we can approximate equation (10) with the following
log-linear equation:
γ
wit =
δ
0 +
δ
1
∆σ
it +
δ
2
∆τ
Lit +
δ
3 git +
δ
4 pit + vwit (18)
where
∆σ
and
∆τ
L denote the change in the replacement rate and labor tax rate
respectively, and p is a vector of other observable variables, such as the unemployment
rate, neglected by our simple model but that nevertheless could play an important role in
more general bargaining models. In some specification we replace the growth rate of per
capita GDP, g, with the growth rate of labor productivity. The theory predicts that
δ
1,
δ
2,
δ
3 > 0. Our interest is particularly on
δ
2. As with unemployment,
δ
2 (but not the other
coefficients) is expected to differ across countries, and to be larger where strong but
decentralized trade unions shift the tax burden onto firms.
Finally, consider the labor demand equation, (7). It can be expressed in terms of
19
the capital labor ratio, k/l, desired by the average firm at the going wage rate. Taking
logs and then differencing, we obtain that the growth of the capital labor ratio,
γ
k/l, can
be written as:
γ
k/l it = d0 + d1
γ
w it + d2 git + vk/lit (19)
where we expect d1 >0 > d2. That is, higher wages lead firms to substitute capital for
labor. And higher growth (of output or capital - in the model the two growth rates
coincide) leads to an even greater demand for labor, thereby reducing the equilibrium
capital-labor ratio.
We now ask whether these predictions are indeed borne out by the data.
3. The Evidence
3.1 The Data
Our sample is a panel of 14 industrial countries, during the period 1965-91. The sample
size is dictated by availability of data on tax rates. To remove the effect of cyclical
fluctuations, we average each variable over a five-year period (except for the first and last
periods where the average is over six years). Thus, our panel consists of 14 countries and
5 observations per country. We also check our results within the smaller sample (only
three observations per country) obtained through 9-year averaging. All data sources are
listed in the Data Appendix.
The data on tax rates deserve some special mention. They have been computed
from the OECD Revenue Statistics by Mendoza, Milesi-Ferretti and Asea (1996) and
filled by ourselves for a few observations, following the methodology of Mendoza, Razin
and Tesar (1994). This method calculates effective tax rates as ratios between the
revenues collected from a specific source and its taxable income base, reconstructed from
national accounting data. This method trades off the disadvantage of giving up the
attempt of constructing marginal tax rates - that are relevant for economic analysis - with
the advantage of a larger database, since these measures are easier to compute and
update. Tables 2 and 3 report these tax rates. The measures of labor tax rates are
probably more reliable than those of capital tax rates, because the capital tax base is more
likely to be measured with error. There is quite a lot of variation across countries and
time in both tax rates. Moreover, as already remarked in the Introduction, in most
20
countries both labor and capital taxes show a clear upward trend.
Another issue is how to measure unemployment, and in particular how to treat
changes in labor force participation. This issue arises in both the growth and the
unemployment equations. Consider the growth equation first, and suppose that the
underlying model is the endogenous growth formulation, where the production function
φ
(k) reflects an externality, such as in the production function in footnote 13. If the
externality in the production function (6) is defined over capital per worker (that is per
member of the labor force), then employment should be scaled to the labor force and the
conventional unemployment rate is the appropriate variable to use. If, instead, the
externality concerns capital per individual in working age, then the employment rate
should be computed by scaling employment to the working age population, and the
unemployment rate should be measured as one minus the employment rate. A similar
ambiguity arises in the unemployment equation, because some changes in labor force
participation are clearly endogenous and affected by, for instance, tax rates and
unemployment subsidies, while other changes are clearly exogenous and reflect
sociological and cultural factors. In the empirical analysis we mainly rely on
conventional measures of unemployment, but we show that all the results are robust to
measuring unemployment as one minus the male employment rate.20
Table 4 reports the pairwise correlations among the main variables in the whole
sample, across countries and over time within each country. Some striking patterns
emerge. Consider unemployment first. In all countries, unemployment and labor taxes are
strongly positively correlated over time. The correlation tends to be stronger for the
countries in continental Europe, as suggested also by Figures 2 and 3 already mentioned
in the introduction. Unemployment is also positively correlated with the replacement rate
over time, in most but not all countries. Neither correlation is present across countries,
however. One plausible explanation is that there are many institutional differences in the
labor markets of different countries, that affect the natural rate of unemployment and that
do not vary over time. These other institutional effects could swamp the effect of labor
taxes and unemployment subsidies when we look at cross-country correlations.
Next, consider growth and investment. Growth is negatively correlated with both
20 This robustness is a priori not obvious, but these alternative measures of unemployment are highly
correlated in our sample: the correlation coefficient between unemployment and one minus the male
employment rate is about 0.84 in levels and 0.81 in first differences. See the Appendix for the data sources.
21
labor and capital tax rates over time; this correlation is a bit weaker than for
unemployment. In some years there remains a strong (negative) correlation between
growth and capital taxes also across countries. Investment is negatively correlated with
the capital tax rate, both over time and across countries.
Unemployment is systematically (negatively) correlated with both growth and
investment, as predicted by the theory. The negative correlation between growth and
unemployment is only evident within countries, but not across countries. Investment
instead is more negatively correlated with unemployment, and this correlation is also
present across countries, even though it is weaker than within countries.
Finally, consider real wage growth (
γ
w). It is positively correlated with the change
in the labor tax rate (
∆τ
L) over time, but only in some countries, while in others the
correlation is negative or absent. Growth of the capital labor ratio (
γ
k/l) and of real
wages are strongly positively correlated, both over time and across countries (the
correlation over time is stronger), as expected.
Finally, in the full sample all correlation coefficients have the expected sign, but
they are much smaller in magnitude.
Overall, these pair-wise correlations indicate that the time series variation in the
data is remarkably in line with the predictions of the theory. In most countries, the rise in
labor tax rates over time has been accompanied by a rise in unemployment, and this in
turn is associated with a decline in growth and investment. The surge in unemployment
is also positively correlated with unemployment benefits. It is instead harder to find a
systematic pattern of correlations across countries, except in the case of investment,
whose correlations with the other variables are also evident across countries and conform
to the theoretical predictions. Finally, there is a strong positive correlation in the growth
rates of real wages and of the capital-labor ratio, while the evidence concerning real
wages and taxes is more mixed.
3.2 Labor Market Institutions and Country Groups
A central implications of the theory is that labor taxes have different effects on
unemployment and real wages depending on the role played by trade unions. Labor taxes
are expected to have the strongest effect on unemployment if wage negotiations are
decentralized and trade unions are powerful but not too large. To test for this implication,
22
we partition countries in three groups: the European countries and Australia (EURO),
where trade unions play an important role but are decentralized; the Anglo-Saxon
countries plus Japan (ANGLO), where labor markets are quite competitive; and the
Scandinavian countries (SCAND), where unions are large and centralized. This
classification is suggested by several previous studies (Calmfors and Driffill (1988),
Bruno and Sachs (1985), Layard, Nickell and Jackman (1991), Cameron (1984)). To
decide which country belongs to which group, we classify countries on the basis of
available data on trade union density, coverage of collective bargaining, as well as
measures of centralization and coordination among unions.21 Based on the data in Table
5 and on the mentioned previous studies, the classification of countries within these three
groups is quite straightforward, except for four cases. Belgium and the Netherlands could
be classified either in EURO or in SCAND. Australia could be classified either in
ANGLO or in EURO. And the UK, whose labor market institutions have changed over
time, should be classified among EURO for the first three time observations, and in
ANGLO for the last two observations. The switch of regime in the UK is quite apparent
for both density and coverage data. We experiment with different classifications at the
margin, but they do not make much difference for the results. In the end, we break the
UK in two different countries depending on periods. Thus, at the cost of some
geographical imprecision, EURO includes Australia, Belgium, France, Germany, Italy,
the Netherlands, Spain and the first three observations of the UK (covering the ‘Labour’
years between 1965 and 1980). ANGLO includes Canada, the US, Japan and the last two
observations of the UK (the ‘Thatcher’ years between 1981 and 1991), while SCAND
includes Sweden, Norway and Finland. Then, we allow the labor tax coefficient to vary
across these groups of countries, by multiplying
τ
L by three dummy variables taking a
value of unity if the country belongs to the group and zero otherwise. The resulting three
variables are labeled E
τ
L, A
τ
L, S
τ
L, for the EURO, ANGLO and SCAND groups
respectively.
Table 6 summarizes the average behavior of unemployment and growth in these
three groups of countries and over time. Naturally, these group averages conceal large
variations in individual countries within the same group. Nevertheless, some interesting
21 These data, as well as the country rankings resulting from different previous studies, are extensively
discussed in OECD (1997).
23
patterns emerge. Everywhere, the big surge in unemployment took place between the
mid-1970s and the mid-1980s. As is well known, the rise in unemployment is more
pronounced in Europe and very mild in Scandinavia. The growth slowdown, on the other
hand, is more gradual, and it is similar in all groups of countries. Table 7 summarizes the
average behavior of tax rates on labor and capital, and of replacement rates. Tax rates on
labor income went up all the time and everywhere. But the increase is more pronounced
in Europe and it is also largest in the period between the mid-1970s and mid-1980s.
Capital tax rates on average also increased in every period; but here there is more
variation across groups of countries and over time. Moreover, the increase in capital tax
rates is most pronounced in Scandinavia. Finally, replacement rates on average also
increased, particularly in Scandinavia and to a smaller extent in Europe. In the ANGLO
group, replacement rates increased during the 1970s but then came down again somewhat
during the 1980s. Table 8 summarizes the average behavior of gross real wages and the
capital labor ratio (see the data appendix for the sources). Here a somewhat different
pattern emerges: both real wage growth and growth in the capital-labor ratio was highest
in the first two decades of our sample, for all groups of countries. Moreover, as
expected, real wage growth was faster on average in Europe.
Now we turn to multivariate correlations and present the results from our
regressions on unemployment, real wages, growth and investment, in turn.
3.3 Unemployment
The identification assumptions discussed in subsection 2.5 imply a recursive structure,
with unemployment being the first equation of the model. Hence we start with the
unemployment equation. The main result of this subsection is that, as predicted by the
theory, higher labor taxes are significantly associated with higher unemployment in
continental Europe but much less so elsewhere. This holds for a variety of specifications
and estimation methods, and in particular even if the recursivity assumption is
abandoned.
3.3.1 Basic specification, alternative estimation methods
We start by estimating the unemployment equation in levels. Labor market institutions
24
are important determinants of unemployment. They are very difficult to measure and
differ a lot across countries, but have changed only very slowly over time or not at all
since the start of our sample. Hence, the appropriate estimation method is by fixed
effects. Thus, we estimate the unemployment equation in levels by OLS, with fifteen
country dummies as intercepts (one for each country, except for the UK that we break in
two countries so as to distinguish the Labour years from the Thatcher years). Note that
fixed effects estimation removes the country mean; hence these estimates reflect the time
series (as opposed to cross-country) correlation in the data.
The results are reported in Table 9, column 1. As expected, labor taxes have a
strong positive effect on unemployment in the EURO countries. This effect is also
positive but smaller in the ANGLO countries, while it is negative but insignificantly
different from zero in the Scandinavian countries. The estimated coefficient of the
replacement rate,
σ
, is also positive and significant, as expected; but, as seen below, this
result is not robust across specifications and estimation methods.
Given that unemployment and labor taxes have risen everywhere in the sample,
but particularly in Europe, a natural question is whether the level estimates reflect some
spurious correlation due to a common trend. Hence we also estimate the equation with all
variables measured in first differences (with and without a constant term appended).22
Here we drop the fixed effects. Columns 2 and 3 in Table 9 report the results from
OLS and GLS regressions for differenced data with a constant term included. OLS
estimation in first differences gets rid of spurious correlation problems due to a common
trend, but could introduce an MA(1) component in the error term. This is taken care of by
GLS estimation, that explicitly allows for a MA(1) structure in the error term.
There is a second, more subtle, possible source of spurious correlation, due to the
possible endogeneity of tax rates and unemployment benefits. For instance, a common
EU wide shock that increased unemployment could have forced all European countries to
also increase labor tax rates to pay for increased unemployment benefits. This potential
endogeneity could also reflect our method of measurement, which relies on effective
22 Throughout the paper and unless otherwise noted, a differenced variable is measured as follows: first we
compute the five (or six) year average, and then we take the first difference of the average. In the case of
the tax rates, data for the period 1960-64 are not available. To avoid losing one observation per country,
therefore, we computed the differenced variable for the period 1965-70 (and only for this period) as the
difference between the average of the tax rate during 1965-70 and the tax rate in 1965. If tax rates have an
upward trend, this underestimates the differenced variable for that period. Tax rates did rise during 1965-
25
average tax rates rather than statutory rates. We also address this problem below, in other
ways. Here, to cope with this possible endogeneity of the right hand side variables, we
estimate the differenced unemployment equation by instrumental variables. Column 4
of Table 9 reports such estimates when the instruments for all right hand side variables
are the same right hand side variables lagged once in differences and twice in levels. 23 If
the residuals of the unemployment equation are uncorrelated over time, these are pretty
good instruments. As discussed below, this estimation method is also appropriate in the
presence of measurement error in tax rates, if such error is uncorrelated over time. Yet,
since differencing reduces the signal to noise ratio in the variables if they are measured
with error, the fit deteriorates. To increase the correlation of the instruments with the
regressors, we also include among the instruments a set of time dummies. 24
No matter how we estimate the differenced regressions, whether by OLS, GLS or
IV, the estimated coefficient of
τ
L for the ANGLO countries and that of the replacement
rate are no longer significantly different from zero. Yet, the estimated coefficient of
τ
L in
the EURO countries always remains positive and highly significant in all columns. Its
size too is quite stable around 0.5, and never drops below 0.37. Furthermore, OLS
estimates both in levels and first differences reject the restriction that the coefficients of
the labor tax rate are equal for EURO and ANGLO countries (the p-values from Wald
tests on that restriction are reported in the last row of each column in the table). The same
applies for the coefficients of EURO and SCAND countries. Only IV estimates cannot
reject the restriction that the tax coefficients for EURO and ANGLO are equal,
presumably because the estimated standard errors are now bigger, since the point
estimates for the country groups remain quite different.
Though not reported here, the computed values of the standardized beta
70, but less than in the next years (see also Table 7).
23 These instruments are uncorrelated with the moving average component of the error term. Thus, the
regression coefficients are consistently estimated, even though the estimated standard errors could be
biased due to the moving average component. Note that the instruments for ∆τLt include τLt-2. This would
force us to drop two observations per country, namely those corresponding to 1965-70 and 1971-75. We
save one observation by using τL in 1965, in place of the (unavailable) average of τL during 1960-65, as one
of the instruments for ∆τL in 1971-75.
24 Removing the time dummies does not change the results. Very similar estimates obtain if the set of
instruments is instead defined as: the first difference in replacement rates lagged once; the
contemporaneous and lagged change in social security payments as a fraction of GDP, interacted by the
country groups EURO, SCAND and ANGLO (source: OECD); and a dummy variable measuring political
majorities on a left to right dimension, obtained from Alesina and Roubini (1998), also interacted with the
country groups EURO, SCAND and ANGLO.
26
coefficients suggest that labor taxes are not only statistically significant, but also
economically relevant determinants of unemployment: their standardized coefficient is
always the largest of any regressor.25
Finally, the test diagnostics for the residuals from all regressions usually cannot
reject the assumptions of normality and no serial correlation, except in column 1 where
the variables are measured in levels.26
3.3.2 Alternative Specifications
The regressions reported in columns (1)-(4) of Table 9 are rather parsimonious as to the
set of conditioning variables included. Here we show that empirical correlation between
labor taxes and unemployment in the EURO countries is not weakened by the inclusion
of additional variables.
First we append country fixed effects to the OLS regression in differences. This
would be the appropriate estimation method if there are country-specific deterministic
time trends in unemployment. This is not the case, as shown in Table 9, column 5. The
statistical significance of E
τ
L stays unchanged, while the value of the adjusted R-squared
falls dramatically from 0.16 to 0.02, a clear symptom of over-parameterization.
Next, we add the one-period lagged unemployment rate (in levels) to the
specification in first differences. This is a check on possible sluggish dynamic behavior
of the unemployment rate, not picked up by the serial correlation tests reported at the
bottom of the Table. Lagged unemployment is borderline significant; but its inclusion
does not affect either the size or the significance of the
τ
L coefficient in EURO (see
Table 9, column 6).
Third, we drop the recursivity assumption and include the growth rate of GDP per
capita among the regressors of the unemployment equation, both in level and difference
regressions. This could be appropriate if taking the five-year average is not sufficient to
remove the cyclical component of unemployment, or more generally if there is a causal
25 The standardized betas are the estimated coefficients in a regression where the original variables have
been divided by their sample standard deviation so as to ‘purge’ the estimated coefficients of their
dependence on measurement units.
26 Given the relatively short time dimension of our sample, the test statistics are not very reliable and an
acceptable p-value for the test may be well below the usual threshold of 0.05, as pointed out by MacKinnon
(1992). Moreover, even correcting for the small sample bias as discussed by Kiviet (1986), the LM test
can only be computed over a sample substantially smaller than the one over which the estimation is
conducted, since we lose 14 observations.
27
effect of growth on unemployment, as predicted by Aghion and Howitt (1994). Growth is
treated as an endogenous variable, and the growth equation is specified as in Table 12,
columns 1 and 2 (see also the discussion in subsection 3.5). The estimates are by 2SLS
and are shown in Table 9, cols. 7 and 8. The estimated coefficient for growth is positive
but insignificant when estimates are in levels. It takes a negative sign when estimates are
in differences. In both cases, the statistical significance of the labor tax coefficient for
EURO countries is unaltered and its size slightly increased after inclusion of the growth
variable. Very similar results obtain if the growth rate of productivity per worker is
substituted for the growth of GDP per capita.
Fourth, we included among the regressors variables capturing adverse
macroeconomic shocks. If there is hystheresis in unemployment, it is possible that
temporary adverse shocks leave unemployment permanently higher. The variables
included were: the change in inflation (see Ball (1996)), the price of oil, and a set of time
dummies for unobservable economic shocks. Some of these variables are occasionally
statistically significant, depending on the specification and the estimation method, but the
correlation between labor taxes and unemployment in the EURO countries survives and
remains quantitatively large.
Finally, the results do not change if we include among the regressors a measure of
trade-union density, such as the percentage of labor force belonging to a trade union. The
variable was entered both linearly and as a quadratic term, to allow for possible non-
linearities, but was never statistically significant. We also experimented with other
measures of trade union membership and concentration, but they were similarly
statistically insignificant and their inclusion never affected the other estimated
coefficients.
3.3.3 Measurement error
The labor tax rates and our measure of unemployment subsidies are certainly measured
with error. To check the robustness of our results, we compute the consistent set of
estimates suggested by Klepper and Leamer (1984) when all variables are measured with
error. Thus, we regress equation (1) in Table 9 in the directions of all regressors,
allowing the coefficients of
τ
L on unemployment to vary across the three groups of
28
countries27. The consistent bounds for the coefficient of E
τ
L in the unemployment
regression are large but remain bounded away from zero. That is, despite likely
measurement error in all the variables, the coefficient of E
τ
L in the unemployment
regression is consistently estimated to be positive. This is not the case instead for the
other country groups. As noted above, the IV estimates also provide a check of
robustness against measurement error.
Unemployment is also likely to be measured with error: different countries have
different methods of recording the unemployed. Therefore we replace the unemployment
rate with one minus the male employment rate (i.e., the ratio between male civilian
employment and male population in working age). The results are even stronger. Table
10, column 1, reports IV estimates in first differences. The significance of the estimated
coefficient of
τ
L for EURO countries is enhanced. The regression diagnostics improve
substantially and the test that the coefficients for EURO and ANGLO are equal is
rejected at the 5% level of confidence, though not at the 1%.
3.3.4 Outliers and country groups
The results are also robust to the presence of outlier observations and to changes in the
definition of the country groups.
Given the exceptional value of Spanish unemployment since the mid-seventies,
what happens if Spain is dropped from the sample? The answer is: not much. The
estimated coefficient of labor taxes on unemployment drops somewhat in size in some
specifications, but it remains clearly significant both in levels and first differences.
Columns 2 and 3 in Table 10 present results for differenced data when Spain is left out
and respectively unemployment and one minus the male employment rate are used as
dependent variables. The estimated coefficient of E
τ
L retains the same size and statistical
significance as in the full sample, while the coefficients of the other variables are
insignificant.
A second question concerns the definition of country groups. There is a margin of
ambiguity in our classification, and a few countries are borderline cases between groups.
Thus, we experiment with alternative classifications: we assign Belgium and Netherlands
27 Thus, when we regress
τ
L on the other variables, we let the coefficients of these other variables to vary
across groups of countries, and so on.
29
to the SCAND group, we put the UK in either the EURO or the ANGLO group for the
whole period (i.e., with no break in the middle). Table 10 (cols. 4, 5 and 6) reports the
estimates. The previous findings are not affected at all, except that the point estimates of
τ
L for ANGLO and SCAND go up (but remain not significant). Moreover, a Wald test
cannot reject the joint hypothesis that the UK in the Labour years belongs to ANGLO and
that the UK in the Thatcher years belongs to EURO. 28
3.3.5 Nine-year averaging
Lengthening the period over which the average is computed washes out business-cycle
fluctuations, which may still be left in 5-year averaged data. The results of these
experiments are shown in Table 10, columns 7 and 8.29 Naturally, the sample is now
smaller. The results are very similar to those reported in Table 9, columns 1 and 2. In
particular, again
τ
L is statistically significant from zero only in EURO, even though it
drops in value somewhat. The restriction that the coefficients of E
τ
L and A
τ
L are equal is
rejected, exactly like with 5-year averaged data. Finally, the coefficient of the
replacement rate remains significant (albeit at the 10% threshold of significance) even
when estimating in differences.
3.3.6 Summary of the unemployment equations
In the countries of continental Europe, unemployment is strongly positively correlated
with labor tax rates. This positive correlation sometimes also shows up, but it is much
weaker, in the ANGLO group of countries, and it is never there in the Scandinavian
countries. This correlation is mainly due to the time series variation in the data. It had
not been found in previous studies for two reasons: first, such earlier studies had imposed
the untested and false restriction that labor taxes affect unemployment equally in
different countries. Second, most previous studies used shorter time series and often
concentrated on the cross-country variation of unemployment.
Scandinavian countries stand out as an unresolved puzzle. We were unable to
detect an effect of labor tax rates on unemployment in these countries. One reason could
28The result that the coefficient of
τ
L is statistically different for EURO and SCAND countries would be
indeed affected if, for example, Spain were assigned to the SCAND group. Yet a Wald test rejects the
restriction that the unconstrained labor tax coefficient for Spain be the same as the coefficient of SCAND
countries, while the restriction that Spain belongs to EURO is not rejected.
30
indeed be the centralized structure of wage negotiations in these countries. But we are
reluctant to draw this conclusion, also because indexes of coordination and centralization
of union behavior are admittedly not very reliable. A second possible explanation is
measurement error in unemployment in these countries. Most people fired in the Swedish
private sector were recorded as still employed during the retraining programs lasting
several months. Moreover, until recently the public sector absorbed workers that
otherwise would have become unemployed, probably to a greater extent than in other
industrial countries - see for instance Rosen (1996).
3.4 Real wages
Now we investigate an additional implication of our model, that higher labor taxes feed
into higher gross wages mostly in EURO countries. Testing this implication is important,
because it can help us discriminate between our hypothesis and other candidate
explanations of European unemployment. In the previous subsection we discussed one
such explanation: some unobserved macroeconomic shock caused high unemployment in
countries with labor market rigidities, and the correlation between labor tax rates and
unemployment detected in Europe is spurious and just due to a common trend. In this
case, however, there should not be any correlation between gross real wages and labor
taxes, and more generally this correlation should be the same in Europe and elsewhere.
This is not what we find in the data. Labor tax rates are correlated with gross real wages,
but only in the EURO countries.
We use two alternative measures of real wages and of labor tax rates. The first
one, denoted
γγ
w in Table 11 and available for almost the full data set, is the growth of
gross real wage earnings in manufacturing (see the data appendix for the sources). The
specification is as described by (16) in subsection 2.5. The variables
∆τ
Land
∆σ
are
measured as changes in the average tax and replacement rates across the five-year periods
(except for the first observation, where the variation is computed between the end and the
beginning of the period for lack of data). A positive estimated coefficient on the labor tax
rates is evidence that the burden of taxation is shifted onto firms, a zero coefficient
instead implies a reduction of net wages.
The second measure of real wages, denoted as
γγ
WN in Table 11, is the growth of
29 First differences are computed as indicated in footnote 22, but for the averaging period.
31
net real wages of the average manufacturing worker (see the data appendix). Hence, here
a zero estimated coefficient on the labor tax rate provides evidence that labor taxes
increase the cost of labor, while the estimated coefficient should be negative if the
burden of taxation is borne by the worker. This variable is available only from 1979
onwards. Thus, our panel consists of 14 countries and only 2 observations per country.
In the net wage regressions, the measure of
τ
L is different from that used in the rest of the
paper : it is obtained from the same source that provides the net wage data and with a
different method of computation (see the data appendix). The correlation coefficient
between our two measures of
τ
L is 0.60 in levels, 0.37 in first differences. These are
low numbers, that underscore how important is the measurement error problem in
τ
L.
For both measures of wages, we first report the more parsimonious specification
implied by the theoretical model (columns (1) and (3) of Table 11). In columns (2) and
(4) we then add unemployment as a regressor, to allow for the moderating effect that
unemployment could have on wage claims. We also replaced the growth rate of per
capita income (implied by the model) with the growth rate of labor productivity, and the
results (not reported) were roughly equivalent. The model is estimated by 2SLS.30 The
results are similar if the model is estimated by OLS, 3SLS or, in the case of gross wages,
also by random effects.
Consider gross real wages first, in columns (1) and (2). As expected, higher labor
tax rates lead to higher real wages in Europe but not elsewhere. Note however that,
contrary to what happened in the unemployment regressions, now the Scandinavian
countries have a higher estimated coefficient than the ANGLO countries, though it
remains statistically insignificant. The growth and unemployment estimated coefficients
are significant and with the expected sign. The only variable which, contrary to the
model predictions, does not have a statistically significant estimated coefficient is the
30 The specification of the growth equation is as reported in Table 12, column 2, with unemployment
entering the growth equation in first differences, also treated as endogenous and specified as in column 2 of
Table 9. When unemployment is added to the wage regression, the 2SLS model also consists of an
unemployment equation in levels, specified as in column 1 of Table 9, and of the growth equation specified
as in column 2 of Table 12. In other words, all RHS variables are in first differences except for
unemployment that enters the growth equation in levels but its coefficient is constrained to be equal to the
negative of that on lagged unemployment. The estimated coefficients of the wage regression are virtually
identical if the growth and unemployment equations are always specified in levels, except that the LM test
reveals a bigger serial correlation of the reduced form residual. Finally, in the case of
γγ
WN, the labor tax
rates entering the unemployment equation are the same as those used in the
γγ
WN regression.
32
change in the replacement rate,
∆σ
. As usual, Spain is a bit of an outlier, as it had very
high real wage growth. Dropping it from the sample reduces somewhat the size of the
estimated coefficients on
∆τ
L for the EURO countries, which however remains
significant and higher than that of the other groups of countries. The results are also
unaffected by the inclusion of time dummies in the real wage regression, while the
inclusion of country dummies reduces the estimated coefficient of
∆τ
L in the European
countries which remains higher than in the other groups, but with a t-statistic of about
1.5. This suggests that the estimated tax rate coefficient is also picking up some cross
country correlation in real wages. Finally, in one specification there is some evidence of
serial correlation in the reduced form residuals, which however is due to the residuals in
the unemployment equation and disappears when the wage equation is estimated by OLS.
Here too, we checked the robustness of our results against likely measurement
error in all variables and in particular in
τ
L. Following Klepper and Leamer (1984), we
estimated the wage regression by least squares in all directions, exploiting the recursivity
assumption of the model. 31 The consistent bounds for the coefficient of E
τ
L in the wage
equation are once again large but remain strictly positive and bounded away from zero.
Thus, here too, measurement error does not seem to be responsible for our results.
Next, consider net real wages. Higher labor tax rates reduce the net real wage in
the ANGLO countries, but not in EURO, as predicted. Moreover, labor tax rates do not
affect net wages in the SCAND countries either, adding further support to the idea that
the wage behavior in Scandinavia is more like in the rest of Europe than like in the
ANGLO group. Finally, net real wages rise faster the higher is the growth rate of per
capita GDP, as expected. Replacing GDP per capita with a measure of labor productivity
leads to very similar results. Unemployment has the expected sign when included, but it
is not statistically significant. Finally, the replacement rate has a positive but insignificant
estimated coefficient. The results are robust to measurement error in all variables: the
consistent bounds for A
τ
L computed as in Klepper and Leamer (1984) are strictly
negative. They are not very robust to other changes, however. They do not hold if the
model is estimated by random effects, or if the sample is redefined to also cover a third
subperiod, 1991-1994. Since not all variables are observed in this last subperiod,
31 Even though we did not perform a Hausman specification test on the recursivity assumption, the OLS
estimates are very close to the 2SLS estimates reported in Table 12.
33
however, here we change both the specification and the sample. Hence, we cannot tell
what is responsible for the different results, whether a possibly mis-specified model or
the different sample.
Overall, the data on real wages confirm the predictions of the model. Higher tax
rates lead to higher gross wages in EURO but not elsewhere. While the data on net
wages provide some, though more fragile, evidence that in a shorter and more recent
sample the tax burden is borne by the workers in the ANGLO countries but it is shifted
onto firms elsewhere.
A difference between the wage and the unemployment data concerns the behavior
of the SCAND countries, for which there is some evidence of a partial shifting of labor
taxes onto higher wages, but there is no evidence that higher taxes increase
unemployment.
3.5 Growth and investment
We now turn to the final block of our recursive model: the link between unemployment
and growth. To check that our regressions of growth on unemployment are not just
capturing a movement along the production function, we also change the dependent
variable: in some regressions we replace the growth rate with the investment share of
GDP, or with productivity growth.
3.5.1 Estimation
We start by estimating the growth and investment equations in levels. Besides
unemployment and the capital tax rate, we also control for the log of initial GDP per
capita, Y, and the initial secondary enrollment ratio, SCHOOL, as is now common in the
growth literature. The estimation method is 2SLS, and the instruments are the
contemporaneous values of all exogenous variables in the entire model, namely E
τ
L , A
τ
L
S
τ
L ,
τ
K ,
σ
, Y, SCHOOL, as well as the full set of country dummies. The results are
reported in Table 12, columns 1 and 4.
To allow for a country specific component in the error term, we also estimate the
model in first differences.32 Thus, in the growth equation, we regress growth on the first
32 Estimation by fixed effects in levels would yield biased estimates, because intial income is inclueded
among the regressors (see for instance Hsiao (1986)). See Islam (1994) and Caselli, Esquivel and Leffort
34
difference of all RHS variables (initial per capita income is then replaced by lagged
growth). In the investment equation we do the same, except that investment is also
measured in first differences. The estimation method is again 2SLS. To avoid the
correlation between the RHS variables and the MA(1) component of the error term, we
use as instruments the exogenous variables lagged once in first differences and lagged
twice in levels, plus unemployment lagged once in first differences and twice in levels, as
well as time dummies. The results are reported in Table 12, columns 2 and 5.
The estimated coefficient of unemployment is always negative and significant, as
expected. It ranges between -0.09 in levels (column 1) and -0.15 in differences (column
2) in the growth regressions. And it ranges between -0.74 in differences (column 5) and
-0.91 in levels (column 4) in the investment regressions. The results for growth are
sensitive to the instrument choice, however. If time dummies are dropped from the
instrument list, the coefficient of unemployment on growth is no longer statistically
significant.
The capital tax rate is statistically significant in the investment regression in
levels, but not in differences, and never in the growth regressions. The greater impact of
the capital tax rate on investment than on growth is a feature of other empirical studies
on industrial countries as well - see Mendoza, Milesi-Ferretti and Asea (1996).
The estimated coefficient of Y (the log of initial GDP per capita) always has a
negative sign, as expected, and it is statistically significant in the growth regressions. The
value of -3.5 of the estimated coefficient of Y in column 1 is approximately half-way
between the ‘iron law of convergence’ coefficient of about 2 found for many cross-
sections of regions and countries by Barro and Sala-i-Martin (1995, ch.12), and the much
higher values found in panel studies.33 The estimated coefficient becomes smaller ( -0.7,
that is one minus the coefficient reported in column 2), but it stays significant at the 10%
level, even when estimating in first differences. The estimated coefficient of SCHOOL
has the expected positive sign but it is statistically significant only in the investment
equation and its estimate varies somewhat across specifications and estimation methods.
To check whether these results are due to the time series or cross country
(1996) for a discussion of the estimation problems of growth regressions in panel data. See also Barro’s
(1996) defense of cross-sectional studies.
33 The main point made in panel studies is that conventional cross-sectional studies produce biased
estimates of the coefficients due to the omission of country fixed effects. See e.g. Islam (1994) and Caselli,
35
variation in the data, we added time and country dummy variables to the regressions in
levels. The results on investment are unaffected. The coefficients of unemployment and
capital taxation in the growth regression, on the other hand, lose significance if country
dummies are added to that equation, while they are only marginally affected by the
inclusion of time dummies.34
The last two lines of Table 12 report the p-values for the Jarque-Bera test (JB) for
the normality of residuals and the degrees-of-freedom-corrected Lagrange Multiplier test
(DFC-LM) for serial correlation. Both JB and DFC-LM tests detect departures from
normality and serial correlation of order one in all level equations, particularly in the
investment equation. Yet, the p-values for the LM test (though not for the JB test)
improve considerably when estimating in first differences, which suggests that we are not
over-differencing already detrended variables.
3.5.2 Sensitivity Analysis
These results are robust to a number of possible changes in the variable definitions or in
the sample.
First, we change our measure of unemployment. In column 3, Table 12,
unemployment is measured as one minus the male employment rate. The estimated
unemployment coefficient on growth retains about the same size, but the estimated
standard error is now smaller.
Second, we replace the growth rate of GDP per capita by the growth rate of
productivity per worker The statistical significance of unemployment (measured in both
ways) improves substantially (Table 12, columns 6 and 7).
Third, we average the data over 9-year intervals rather than over five years. The
results remain very similar (Table 13, columns 1 and 2). Unemployment is more
significant when estimating in levels than in first differences, and its significance in the
difference regression crucially hinges on the inclusion of time dummies in the instrument
list.
Fourth, we drop Spain from the sample. Some results worsen, others survive.
Table 13, column 3 reports the results of the growth regression in first differences. The
Esquivel and Lefort (1996).
34 As already noted, however, fixed effects estimators could be biased because initial GDP per capita acts
as a lagged endogenous variable.
36
unemployment coefficient is no longer statistically significant: its pointwise estimate
drops from -0.15 to -0.09 and the standard error of the estimate goes up to 0.10.
However, the results improve substantially when investment is the dependent variable or
if unemployment is measured as one minus the male employment rate (Table 13,
column 4).
As a final check, we added hours worked in the growth and investment
regressions as an additional right-hand side variable. It was generally statistically
insignificant and all other estimated coefficients were unaffected.
3.5.3 Capital-labor ratios
Finally, we turn to an analysis of the growth in the capital-labor ratio, which we expect to
be positively related to the growth of real wages. The last column of Table 13 reports
the estimates of the specification corresponding to equation (19) in subsection 2.5. The
estimation method is 2SLS, where the specification for real wage growth is as in column
(1) of Table 11, while growth and unemployment are respectively specified as in Table
12, column 1 and Table 9, column 1. The estimated coefficient on real wage growth is
positive and highly significant, as expected. The estimated coefficient on growth is also
positive, contrary to the model predictions, though it is only marginally significant. The
results are very similar if GDP growth is replaced by the investment share of output. The
results concerning the effect of real wages on the capital labor ratio are extremely robust,
to the estimation method, to the inclusion of time or country dummy variables, to
whether or not the estimates are weighted by initial per capita GDP to allow for possible
heteroscedasticity. Finally, there is some evidence of serial correlation. If lagged per
capita income is added to the regression, however, serial correlation disappears. The
estimated coefficient on lagged per capita income is negative and statistically significant,
while the growth coefficient becomes insignificantly different from zero. The real wage
coefficient is unaffected.
Summarizing then, the empirical evidence strongly confirms the plausible
prediction of the model, that higher real wages induce firms to substitute labor for
capital.
3.5.4 Summary and Quantitative Implications.
37
In our sample, growth and investment are negatively correlated with unemployment, as
predicted by a large class of growth models. This correlation is admittedly less robust
than that between labor taxes and unemployment in EURO countries. And our empirical
tests cannot shed light on whether this correlation is temporary or permanent. We can
nevertheless ask, based on our estimates, how costly was the rise of labor taxes in Europe
in the last twenty years, in terms of higher unemployment and slower growth.
Multiplying the estimated coefficient of
τ
L by the unemployment coefficient in the
growth equation, we can measure the induced effect of labor taxes on growth, and how it
differs according to labor market institutions. For the EURO countries, the coefficient of
τ
L in the unemployment equation is consistently significant and stable across a variety of
specifications and estimation methods. Its estimated central value ranges between 0.31
and 0.65, much larger than for the ANGLO and SCAND countries. Thus, for the EURO
group, the compounded effect of
τ
L on growth (through unemployment) lies between -
0.040 and -0.075, depending on the specifications and estimation methods. For the
ANGLO and SCAND groups, on the other hand, the estimates are smaller and less
stable. Though imprecisely estimated, the coefficient of
τ
K in the growth equation is
equal to -0.006 in Table 12, column 1. That is, for the EURO countries the estimated
impact of
τ
K on both growth and investment is always smaller than that indirectly
induced by
τ
L through unemployment.
These large numbers suggest that the surge in European unemployment and the
European growth slowdown can be largely accounted for by the rise in labor taxes over
time. Consider for instance Germany and France in 1965-1975 and 1976-1991, where
labor tax rates rose by about 8 and 10 percentage points respectively between these two
periods. According to our estimates, this could account for a reduction of growth of GDP
per capita of about 0.4% per year, and an overall increase in unemployment of 3-5%.
Taking the EURO group as a whole, the rise in labor tax rates of 9.4% can be associated
to an increase of 4% in the unemployment rate and a slowdown of per-capita GDP
growth by about 0.4% a year. These are very large numbers. But so is a 10% point
increase in labor tax rates. These figures compare with much smaller growth effects of
the capital tax rate, of -0.03% and -0.10% in Germany and France respectively, which
result from both smaller increases of the tax rate and lower values of the estimated
coefficient. The sizable effect of labor taxes on growth is consistent with the findings of
38
Mendoza, Milesi-Ferretti and Asea (1996), whose simulations broadly confirm
Harberger’s (1964) conjecture on the irrelevance of the tax structure for growth. A
reduction of 10% in tax rates on capital income and consumption is shown to have
negligible growth effects (of an order of magnitude of 0.1-0.2%). Yet, taxes on human
capital (what we call ‘labor taxes’) are instead found to be growth-enhancing.35
These results are particularly striking if compared with the reduced form
regressions reported in Table 13, columns 6 and 7. In line with most of the studies in
the applied growth literature, no effects of either capital or labor taxes show up in the
reduced form growth regressions. How can we reconcile these reduced form results with
the negative correlation of unemployment and growth, and the strong effect of labor tax
rates on unemployment in Continental Europe ? One possible explanation is that the
reduced form coefficient of E
τ
L on growth is rather small (its order of magnitude is -0.05
or less even according to our indirect estimates). Thus, in the presence of measurement
error, the reduced form estimates cannot reject that it is zero. To detect an effect of E
τ
L
on growth, we need to impose sharper restrictions on the precise mechanism through
which it takes place. Our theory says that labor tax rates affect growth insofar as they
affect unemployment, and they affect unemployment insofar as labor markets are
dominated by strong and decentralized trade unions. These over-identifying restrictions
can be tested using the LM specification test proposed by Hausman (1983, p.433). They
cannot be rejected: the p-value of the test contrasting the reduced form estimates (Table
13, column 6) against the structural estimates is 0.78 (see the last line in Table 12,
column 2). Hence, despite the statistical insignificance of the reduced form estimates,
we do not reject the over-identifying assumptions of our structural model.
4. Conclusions
The surge in European unemployment undoubtedly has more than one cause. But,
according to our empirical results, one is easy to identify: higher labor taxes have been
shifted onto higher real wages. This has led firms to substitute labor with capital and it
35 Mendoza, Milesi-Ferretti and Asea find that the effects of taxes on human capital may be even larger
than ours (a 10% decrease of human capital taxation may result in a 1.5% rise of the growth rate). This is
because they also allow for human capital to be reallocated across sectors, while our basic model does not
explicitly take into account human capital accumulation.
39
has slowed down economic growth. Thus, higher unemployment and slower growth in
Europe are related, but in the sense that they stem from the same cause.
The policy implications of this view, if correct, would be extremely relevant.
They concern the cost of the generous European welfare states, the optimal structure of
taxation across alternative tax bases such as capital, labor or consumption, and the nature
of the efficient remedies for the high unemployment.
Naturally, there are many chains to the argument put forward in this paper, and
they ought to be investigated more in detail and with better data. Here are some
questions worth exploring more in detail in future research. First, why has unemployment
been so low in the Scandinavian countries, if indeed high taxes have been causing high
unemployment in continental Europe? One possible answer suggested in the paper has to
do with the high level of centralization and coordination of trade unions in Scandinavia,
which has moderated their wage claims. But the real wage data explored in this model
are not entirely consistent with this view. An alternative explanation is the expansion of
public sector employment and the role that this might have had in preventing private
sector unemployment in Scandinavia. The validity of this alternative explanation could
be explored with disaggregated data on employment.
A second natural question concerns the role of international capital flows in the
presence of different labor costs or tax treatments across countries. The theoretical model
studied a closed economy, and the empirical analysis did not focus on the response of
foreign direct investment to differences in the cost of labor. Adding these complications
to our analysis is likely to reinforce the qualitative predictions of our model concerning
the effect of labor taxes, and would increase the sensitivity of investment to capital taxes.
Moreover, such a richer model could yield new testable implications. Hence,
investigating both theoretically and empirically these questions in an open economy
setting seems a natural and interesting extension of this work.
Finally, throughout the paper policy variables have been treated as exogenous.
This is clearly not satisfactory. The same groups that are organized to bargain over
wages are also able to influence economic policy, by lobbying or through other forms of
political participation. Making economic policy endogenous and chosen by politically
responsive governments in a dynamic framework, and possibly investigating the
empirical implications of such a richer model, would be a difficult but exciting line of
40
research. Caballero and Hammour (1996) have made some progress in this direction in a
recent interesting paper.
41
References
Aghion, P. and P.Howitt (1994), Growth and unemployment, Review of Economic
Studies, 61, 477-495
Alesina, A. and R.Perotti, (1994), The Welfare State and competitiveness, NBER
Working Paper
Alesina, A. and N.Roubini, with G.Cohen (1998), Political cycles and the
macroeconomy, Boston, MIT Press, forthcoming
Alogoskoufis, G. et al. (1995), European unemployment: is there a solution?, London,
CEPR
Arellano, M. and S.Bond, (1991), Some tests of specifications for panel data: Monte
Carlo evidence and an application to employment equations, Review of Economic
Studies, 58, 277-297
Ball, L. (1996), Disinflation and the NAIRU, NBER Working Paper #5520, July
Barro, R.J. (1996), Determinants of economic growth: a cross-country empirical study,
NBER Working Paper #5698, August
Barro, R.J. and X. Sala-i-Martin (1995), Economic growth, McGrawHill
Basu, S. and J.G.Fernald (1995), Are apparent productive spillovers a figment of
specification error?, Journal of Monetary Economics , 36, 165-188
Bean, C. (1994), European unemployment: a survey, Journal of Economic Literature,
vol.32, 573-619
Bertola, G. (1993), Factor shares and savings in endogenous growth models, American
Economic Review, 83, 1184-1198
Blanchard, O. (1996), Macroeconomics, MIT Press, ch.25
Blanchard, O. (1997), The Medium Term, mimeo, MIT
Blanchard, O. and L.F.Katz (1996), What we know and do not know about the natural
rate of unemployment, NBER Working Paper #5822, November
Bruno M., and J.D.Sachs (1985), Economics of worldwide stagflation, Cambridge,
Mass., Harvard University Press
Burnside, C. (1996), Production function regressions, returns to scale, and externalities,
Journal of Monetary Economics, 37, 177-201
42
Caballero, R. and M.L.Hammour (1996), The Macroeconomics of Specificity; NBER
Working Paper #5757, September
Caballero, R. and R.Lyons (1992), External effects in U.S. manufacturing, Journal of
Monetary Economics, 29, 209-226
Calmfors L. and J. Driffill (1988) “Bargaining Structure, Corporativism and
Macroeconomic Performance”, Economic Policy, vol. 6
Cameron D.R. (1984), Social democracy, corporatism, labor quiescence and the
representation of economic interest in advanced capitalist society, in J.H.Goldthorpe
(ed.), Order and conflict in contemporary capitalism, Oxford, Clarendon Press
Caselli, F., G.Esquivel and F.Lefort (1996), Reopening the convergence debate: a new
look at cross-country growth empirics, Journal of Economic Growth
Furuya, K. (1995), Capital, savings and unemployment, mimeo, Berkeley University
Golden, M. (1996), Union centralization among advanced industrial societies: an
empirical study, UCLA, mimeo
Gordon, R. (1995), Is there a trade-off between unemployment and productivity growth?,
CEPR Discussion Paper #.1159, April
Hsiao, C. (1986), The analysis of panel data, Cambridge, Cambridge University Press
Islam, N. (1994), Growth empirics: a panel data approach, Quarterly Journal of
Economics, 110, 1127-1170
Jones, L., R.Manuelli (1990), A convex model of equilibrium growth, Journal of
Political Economy, 98, 1008-1038
Jones, L., R.Manuelli and P.Rossi (1993), Optimal taxation in models of endogenous
growth, Journal of Political Economy, 101, 485-517
Kiviet, J.F. (1986), On the rigour of some misspecification tests for modelling dynamic
relationships, Review of Economic Studies, 53, 241-261
Klepper S. and E.E.Leamer (1984), Consistent sets of estimates for regressions with
errors in all variables, Econometrica, 52, 163-183
Layard, R., S.Nickell and R.Jackman (1991), Unemployment: Macroeconomic
performance and the labour market, Oxford, Oxford University Press
Lindbeck, A. (1993), Unemployment and macroeconomics, MIT Press
Lucas, R.J.E..Jr. (1988), On the mechanics of economic development, Journal of
Monetary Economics, 2-40
43
Lucas, R.J.E..Jr. (1990), Supply-side economics: an analytical survey, Oxford Economic
Papers, 42, 293-316
Lucas, R.J.E..Jr. (1993), Making a miracle, Econometrica, 61, 251-272
MacKinnon, J.g. (1992), Model specification tests and artificial regressions, Journal of
Economic Literature, 30, 102-146
Mendoza, E., G.Milesi-Ferretti and P.Asea (1996), On the ineffectiveness of tax policy in
altering long-run growth: Harberger’s superneutrality conjecture, CEPR Discussion
Paper #1378, April
Mendoza, E., A.Razin and L.Tesar (1994), Effective tax rates in macreconomics: cross-
country estimates of tax rates on factor incomes and consumption, Journal of Monetary
Economics, 34, 297-323
Mortensen, D.T. and C.Pissarides (1997), Job reallocation, employment fluctuations and
unemployment differences, in J.Taylor and M.Woodford, Handbook of Macroeconomics,
vol.III, North Holland, forthcoming
Nickell, S. (1997), Unemployment and Labor Market Rigidities, The Journal of
Economic Perspectives, August
OECD (1994), The OECD Jobs Study, Paris, OECD
OECD (1997), Employment Outlook, Paris, OECD
Padoa Schioppa, F. (1990), Union wage setting and taxation, Oxford Bulletin of
Economics and Statistics, 52, 143-166
Persson, T. and G.Tabellini (1997), Political economics and macroeconomics, in
J.E.Taylor and M.Woodford, Handbook of Macroeconomics, Amsterdam, North Holland
Phelps, E.S. (1968), Money-wage dynamics and labor market equilibrium, Journal of
Political Economy, 76, 678-711
Phelps, E.S. (1994), Structural slumps: the modern equilibrium theory of unemployment,
interest and assets, Cambridge, Mass., Harvard University Press
Rebelo, S. (1991), Long-run policy analysis and long-run growth, Journal of Political
Economy, 99, 500-521
Romer, P.R. (1986), Increasing returns and long-run growth, Journal of Political
Economy, 94, 1002-1037
Romer P.R. (1989), Capital accumulation and long-run growth, in R.J.Barro, Modern
business-cycle theory, Oxford, Basil Blackwell
44
Romer, P.R. (1990), Endogenous technical change, Journal of Political Economy, 98,
S71-S102
Rosen, S. (1996), Public employment and the welfare state in Sweden, Journal of
Economic Literature, 34, 729-740
Stokey N.L. and S.Rebelo (1995), Growth effects of flat-tax rates, Journal of Political
Economy, 103, 519-550
Summers, L., J.Gruber and R.Vergara (1993), Taxation and the structure of labor
markets: the case of corporatism, Quarterly Journal of Economics, 385-411
Tullio, G. (1987), Long-run implications of the increase in taxation and public debt for
employment and economic growth in Europe, European Economic Review, 20, 741-774
Tyrvainen, T. (1994), Real wage resistance and unemployment: multivaraite analysis of
cointegrating relations in 10 OECD countries, mimeo, Paris, OECD
Van der Ploeg, F. (1987), Trade unions, employment and investment: a non-cooperative
approach, European Economic Review, 31, 1465-1492
Young, A. (1993), Lessons from the East Asian NICs: a contrarian view, European
Economic Review, 38, 964-973
45
Data Appendix
List of variables: definitions and sources.
u = standardized unemployment rate. Source: OECD National Accounts. In some
regressions, u is the complement to one of the employment rate (the number of the
civilian employed divided by population in working age), expressed in percentage points.
Source: OECD Labor Force Statistics.
g = Growth rate of per-capita GDP. Source: Summers-Heston data set (PWT 5.6).
Y = level of initial per capita GDP. Source : Summers-Heston data set (PWT 5.6).
τ
L = Effective tax rate on labor income: Computed as the ratio between total taxes on
labor income (= an imputation of taxes on wages and salaries from the individual income
tax + social security contributions + payroll taxes) and the labor tax base (=wages and
salaries + employers’ contributions to social security). Missing a full series, the tax rates
for few countries are computed as the average between two data points (e.g. 1965 and
1970 for the 1965-70 quinquennium for Netherlands, Italy, Spain and Norway). Primary
source: OECD National Accounts, OECD Revenue Statistics. Again, due to data
coverage problems, the data on labor tax rates employed in the net wage equations are
drawn from another OECD publication, The tax/benefit position of production workers.
τ
K = Effective tax rate on capital income: Computed as the ratio between total taxes on
capital income (= an imputation of taxes on the operating surplus of unincorporated
enterprises and profits and entrepreneurial incomes + corporate taxes + recurrent taxes on
immovable property + taxes on financial and capital transactions) and its tax base (= total
operating surplus of the economy). Missing a full series, the tax rates for few countries
are computed as the average between two data points (e.g. 1965 and 1970 for the 1965-
70 quinquennium for Netherlands, Italy, Spain and Norway). Primary source: OECD
National Accounts, OECD Revenue Statistics.
i = Investment: Domestic investment as a share of GDP at constant prices. Source:
Summers-Heston data set (PWT 5.6).
σ
= Replacement rate in unemployment benefits: Computed as a summary measure of
replacement rates, i.e. ratio between benefit entitlements and previous earnings, both
before tax, for a variety of circumstances (period of unemployment spell, family
situation, previous level of earnings). Primary source: OECD Jobs Study, Database on
Unemployment Benefit Entitlements and Replacement Rates.
γ
k/l = growth in the stock of physical capital per employee: The variable in levels has
been computed as the ratio between the capital stock per worker (source: PWT 5.6) and
the employment rate (= 1-standardized unemployment rate; source: OECD National
Accounts).
46
γ
w = Growth of real wages: growth in hourly pre-tax wage earnings in manufacturing,
deflated with the CPI . Source: OECD National Accounts and Main Economic Indicators.
γ
wn = Growth of net real wages: growth in take-home pay and cash transfers after taxes
and from annual gross earnings of a single person. Source: OECD, Tax and benefit
position of production workers, various issues.
SCHOOL = Gross secondary enrollment ratio in the initial year of the quinquennium:
number of persons enrolled in the secondary school as a share of population aged 14-18.
47
Table 1
UNEMPLOYMENT & GROWTH IN INDUSTRIAL
COUNTRIES
Unemployment Per-capita GDP Growth
1965 - 75 1976 - 95 1965 - 75 1976 - 95
• EUROPE
2.6
8.5
3.5
1.7
• EC
2.8
9.4
3.4
1.7
• SCANDINAVIA
2.0
4.8
3.4
1.5
• NON-EUROPE
3.4
6.5
3.5
1.8
• US & CANADA
5.0
8.0
2.8
1.5
• JAPAN
1.3
2.4
6.0
3.0
• AUSTRALIA 2.3 7.7 2.3 1.2
Source: OECD National Accounts and Summers-Heston data set
48
Table 2
Effective Tax Rates on Labor Income
Country/Year 1965-70 1971-75 1976-80 1981-85 1986-91
Australia 11.7 14.1 16.5 17.9 18.2
Belgium 30.5 36.4 41.7 45.3 47.7
Canada 17.1 22.0 22.6 25.0 29.2
Finland 20.7 28.1 30.9 31.1 33.9
France 33.9 33.0 37.9 42.4 45.7
Germany 30.5 35.1 38.3 38.9 41.7
Italy 26.1 28.7. 32.0 37.0 41.4
Japan 16.0 18.1 20.6 24.4 27.5
Netherlands 36.1 42.7 47.1 50.1 51.8
Norway 31.0 38.9 38.7 38.4 39.8
Spain 15.4 20.2 26.4 32.8 36.4
Sweden 34.3 38.9 47.2 48.1 50.9
UK 22.6 24.7 26.7 27.1 25.7
USA 20.1 23.0 26.1 28.3 28.9
Legenda: Effective tax rates are constructed following the methodology suggested by
Mendoza, Razin and Tesar (1994). See also the Data Appendix.
Primary source : OECD National Accounts, various issues
49
Table 3
Effective Tax Rates on Capital Income
Country/Year 1965-70 1971-75 1976-80 1981-85 1986-91
Australia 31.7 38.8 41.4 44.5 47.4
Belgium 19.7 28.0 37.3 39.5 36.3
Canada 41.0 43.4 39.6 37.9 43.1
Finland 24.0 28.1 35.8 35.2 44.1
France 16.2 18.0 23.9 28.4 26.1
Germany 21.2 24.8 29.4 31.0 27.8
Italy 12.8 13.1 17.0 25.3 28.6
Japan 20.5 28.8 32.6 39.7 49.3
Netherlands 23.9 28.9 33.5 29.7 30.9
Norway 29.0 27.3 37.3 42.6 38.7
Spain 8.2 9.7 12.7 13.9 14.1
Sweden 36.2 39.9 54.1 47.4 63.9
UK 46.7 56.6 55.7 66.5 58.3
USA 42.0 44.3 44.7 40.9 41.0
Legenda: Effective tax rates are constructed following the methodology suggested by
Mendoza, Razin and Tesar (1994). See also the Data Appendix.
Primary source : OECD National Accounts, various issues
50
Table 4 - Pairwise Correlation Coefficients
u,
τ
Lu,
σ
g,
τ
Lg,
τ
Kg, u i, u i,
τ
K
γ
w,
∆τ
L
γ
w,
γ
k/l
.20 .25 -.27 -.30 -.37 -.51 -.44 .16 .68
Over Time
u,
τ
Lu,
σ
g,
τ
Lg,
τ
Kg, u i, u i,
τ
K
γ
w,
∆τ
L
γ
w,
γ
k/l
US .78 -.08 -.55 .17 -.38 -.68 .31 .73 .67
UK .79 -.69 -.04 .15 .51 -.90 -.91 -.20 .38
Belgium .91 .57 -.63 -.73 -.82 -.94 -.80 .50 .81
France .98 .86 -.60 -.82 -.73 -.88 -.94 -.81 .94
Germany .87 -.54 -.32 -.45 -.44 -.99 -.91 .48 .74
Italy .99 -.83 -.54 -.52 -.46 -.90 -.91 -.98 .94
Netherlands .92 .67 -.81 -.68 -.83 -.98 -.69 .77 .94
Norway .43 .96 -.13 -.34 -.93 -.96 -.76 .98 -.73
Sweden .36 .36 -.92 -.88 -.14 -.73 -.58 -.12 .72
Canada .80 .77 -.86 .07 -.73 .46 .29 .44 -.11
Japan .94 -.89 -.53 -.60 -.60 -.47 -.13 -.84 .99
Finland .76 .85 -.97 -.98 -.79 -.94 -.88 .74 .92
Spain .96 .93 -.43 -.11 -.37 -.67 -.62 -.19 .90
Australia .98 .90 -.79 -.89 -.67 -.90 -.99 .47 .90
Across Countries
u,
τ
Lu,
σ
g,
τ
Lg,
τ
Kg, u i, u i,
τ
K
γ
w,
∆τ
L
γ
w,
γ
k/l
1965-70 -.27 .-.37 -.24 -.64 -.15 -.39 -.59 -.27 .66
1971-75 -.29 -.11 0 -.33 -.14 -.60 -.53 .14 .35
1976-80 -.27 .23 .01 -.22 -.19 -.47 -.38 .23 .74
1981-85 0 .30 -.53 .53 -.70 -.57 -.22 .02 .37
1986-91 .01 .12 .15 -.44 .47 -.20 -.07 -.04 -.07
Legenda : see the data appendix for the definitions.
51
Table 5
Coverage, density and coordination of labor bargaining
in Industrial Countries
Country Coverage Density Coordination
1980 1990 1976-86 1986-91 Union Employer
ANGLO
Canada 37 38 26 32 1 1
Japan 28 23 31 26 2 2
USA 26 18 23 15 1 1
United Kingdom 70 47 48 39 1 1
EURO
Australia 88 80 46 44 2 1
Belgium 90 90 55 54 2 2
France 85 92 19 12 2 2
Germany 91 90 35 31 2 3
Italy 85 83 45 34 2 2
Netherlands 76 71 35 23 2 2
Spain 68 68 25 11 2 1
SCAND
Finland 95 95 69 71 2 3
Norway 75 75 52 54 3 3
Sweden 83 83 76 83 3 3
Source: OECD Jobs Study for coverage ratios; Nickell (1997) for cordination, Golden (1996) for
union density rates.
Notes:
“Coverage” measures the extent to which contracts signed by organized unions extend to the rest
of the labor force.
“Density” measures the rates of net union density, i.e. the number of union members net of
pensioners divided by the labor force.
“Coordination” measures the extent of contracting coordination within different union and
employer organizations in 1989-94. The index provides a qualitative ranking of countries:
“1”means “Low”, “2” is for “Medium”, “3” is for “High”.
52
Table 6
Unemployment and Growth
Europe Anglo Scand All Countries
'8 * '8 * '8 * '8 *
1965-70 0.4 4.1 -0.9 4.6 0.3 3.6 0 4.2
1971-75 0.8 2.4 1.3 2.3 0.1 3.2 0.8 2.6
1976-80 2.8 2.5 1.5 2.6 1.0 2.6 2.0 2.6
1981-85 4.6 0.6 2.5 2.1 0.6 2.3 3.0 1.5
1986-91 -0.4 2.9 -1.4 1.8 -0.9 1.0 -0.7 2.1
All Years 1.6 2.6 0.6 2.6 0.4 2.5
DU denotes the change in the unemployment rate, measured as the first difference of the five (or six) year averages.
G is the average growth rate of per capita income during each five (or six) years period.
Each column displays the average of the above variables across countries, within the relevant country groups, for each time period.
The raw “All years” refers to the simple average across both time and countries, within the relevant country groups.
The country groups are those described in Table 5.
53
Table 7
Tax Rates and Replacement Rates
Europe Anglo Scand All Countries
∆σ ∆τ ∆τ ∆σ ∆τ ∆τ ∆σ ∆τ ∆τ ∆σ ∆τ ∆τ
1965-70 3.6 1.7 1.7 -0.7 2.5 2.2 -0.1 2.8 3.2 1.6 2.1 2.2
1971-75 2.7 4.4 3.7 2.0 5.0 3.1 4.6 2.0 6.6 2.9 4.1 4.1
1976-80 1.0 4.2 4.2 2.8 0.8 2.2 12.5 10.6 3.7 4.0 4.6 3.5
1981-85 0.8 2.5 3.9 -1.1 3.1 2.0 4.6 -0.7 0.3 0.9 2.0 2.4
1986-91 3.2 0.3 3.0 -0.2 1.9 1.3 8.6 7.1 2.4 3.1 2.3 2.3
All Years 2.3 2.7 3.3 0.4 2.6 2.1 6.0 4.4 3.2
∆σ
denotes the change in the replacement rate,
∆τ
is the change in the capital tax rate,
∆τ
L is the change in the labor tax rate.
All variables are measured as the first difference of the relevant five (or six) year averages.
Each column displays the average of the above variables across countries, within the relevant country groups, for each time period.
The raw “All years” refers to the simple average across both time and countries, within the relevant country groups.
The country groups are those described in Table 5.
54
Table 8
Real Wages and Capital Labor Ratio
(% growth rate)
Europe Anglo Scand All Countries
γ γ γ γ γ γ γ γ
1965-70 5.0 7.4 3.4 7.0 4.5 4.0 4.4 6.6
1971-75 5.4 7.3 3.6 6.5 4.2 3.4 4.7 6.2
1976-80 2.6 3.8 0.8 2.8 0.5 2.4 1.6 3.2
1981-85 0.7 2.7 1.0 3.1 0.7 2.2 0.8 2.8
1986-91 1.4 2.1 0.6 3.6 1.8 3.3 1.2 2.9
All Years 3.0 4.8 1.7 4.5 1.9 3.0
γ
and
γ
denote the average growth rate of real wages and of the capital-labor ratio respectively, measured as five (or six) year averages of the
yearly growth rates.
Each column displays the average of the above variables across countries, within the relevant country groups, for each time period.
The raw “All years” refers to the simple average across both time and countries, within the relevant country groups.
The country groups are those described in Table 5.
55
Table 9
UNEMPLOYMENT AND LABOR TAXES
(5-YEAR AVERAGES)
>1@ >2@ >3@ >4@ >5@ >6@ >7@ >8@
Method of
estimation OLS OLS GLS IV OLS IV 2SLS 2SLS
Dep.variable ∆ ∆ ∆ ∆ ∆ ∆
A
τ
L
E
τ
L
S
τ
L
σ
ut-1
g
#obs
Adj.R2
SEE
JB test
DFC-LM test
Wald test:
Anglo=Euro
Notes:
Column 1 : All variables in levels, fixed effects estimation
Columns 2: All variables in first differences, estimated without fixed effects.
Columns 3: All variables in first differences, estimated by GLS allowing for MA(1) in the error term but no
correlation across countries.
Columns 4,6: All variables in first differences, except ut-1 which is in levels. Instruments: Right-hand side variables
in levels and first differences lagged twice and once respectively; time dummies.
Column 5 : Country fixed effects included as intercepts. All variables in first differences.
Column 7: All variables in levels. Estimated by 2SLS, with fixed effects in the unemployment equation. The growth
equation is specified as in Table 12, column (1).
Column 8: All variables except growth in first differences. Estimated by 2SLS. The growth equation is specified as
in Table 11, column (2). All exogenous and predetermined variables in the unemployment and in the growth equations
are instrumented with a set of instruments consisting of: all exogenous and predetermined variables lagged twice in
levels and once in first differences, plus time dummies.
Fixed-effects intercepts and constants not reported. Standard errors in parentheses.
* = 10% level of significance
** = 5% level of significance
*** = 1% level of significance
56
Table 10
UNEMPLOYMENT AND LABOR TAXES
(SENSITIVITY ANALYSIS)
>1@ >2@ >3@ >4@ >5@ >6@ >7@ >8@
Method of
estimation IV IV IV IV IV IV OLS OLS
Dep.variable ∆ ∆ ∆ ∆ ∆ ∆ u∆
Fixed-effects
Averaging
Sample and
country
groups
A
τ
L
E
τ
L
S
τ
L
σ
#obs
Adj.R2
SEE
JB test
DFC-LM test
Notes:
In all columns with IV estimates: Instruments are right-hand side variables in levels and first differences lagged twice
and once respectively;, as well as time dummies.
Column [1]: The dependent variable is one minus the male employment rate. Instruments: Right-hand side variables in
levels and first differences lagged twice and once respectively; time dummies.
Columns [2]: Spain left out of the sample. All variables in first differences.
Columns 3 : Spain left out. Dependent variable is one minus the male employment rate. All variables in first
differences.
Column 4: Belgium and Netherlands included in the SCAND group of countries. All variables in first differences.
Column 5: UK included in the ANGLO group of countries. All variables in first differences.
Column 6: UK included in the EURO group of countries. All variables in first differences.
Column 7: 9-year averages. All variables in levels; fixed effects estimation.
Column 8: 9-year averages. All variables in first differences.
Fixed-effects intercepts and constants not reported. Standard errors in parentheses.
* = 10% level of significance
** = 5% level of significance
*** = 1% level of significance
57
Table 11
REAL WAGES AND LABOR TAXES
(5-YEAR AVERAGES)
> 1 @ > 2 @ >3 @ >4@
Method of
estimation 2SLS 2SLS 2SLS 2SLS
Dependent
variable
γ γ γ γ
A
∆τ
L
E.
∆τ
L
S
∆τ
L
∆σ
g
u
#obs
Adj.R2
DFC-Lm test
SEE
Notes:
Intercepts not reported. Standard errors in parentheses.
* = 10% level of significance
** = 5% level of significance
*** = 1% level of significance
Estimation by 2SLS. The specification for u and g is as in Table 9, column (1), and Table (12), column
(2), respectively. The instruments are the current values of the exogenous variables appearing in those
equations, including country dummies in the specifications with u. See also footnote 30.
58
Table 12
GROWTH AND INVESTMENT
(5-YEAR AVERAGES)
1 2 3 4 5 6 7
Method of
estimation 2SLS 2SLS
RHS in
∆
2SLS
RHS in
∆
2SLS 2SLS
RHS in
∆2SLS
RHS in
∆2SLS
RHS in
∆
Dep.variable ∆
τ K
u
um
Y
SCHOOL
#obs
Adj.R2
SEE
JB test
DFC-LM test
Hausman-LM
Notes: Columns 1, 4 : All RHS variables in levels. Instruments: Contemporaneous values of all the exogenous
variables - that is excluding u and ut-1 - listed in this Table and in Table 9, including country dummies.
Columns 2, 3, 5 - 7: All RHS variables in first differences. Instruments: All the exogenous variables in
this Table and Table 9, expressed in first differences lagged once and in levels lagged twice; unemployment in first
differences lagged once and in levels lagged twice; time dummies.
Intercepts not reported. Standard errors in parentheses.
* = 10% level of significance
** = 5% level of significance
*** = 1% level of significance
59
Table 13
GROWTH AND INVESTMENT
SENSITIVITY ANALYSIS
1 2 3 4 5 6 7
Method of
estimation 2SLS 2SLS
RHS in
∆2SLS
RHS in
∆2SLS
RHS in
∆2SLS 2SLS
RHS in
∆2SLS
RHS in
∆
Dep.variable ∆
Sample
Averaging
τ K
u
um
Y
SCHOOL
g
gw
AτL
EτL
SτL
σ
#obs
Adj.R2
SEE
JB test
DFC-LM test
60
Notes to Table 13:
Column 1: All variables in levels. Instruments: Contemporaneous values of all the exogenous variables in
this Table, that is excluding u, including country dummies.
Column [2]: All RHS variables in first differences. Instruments: Contemporaneous values of all the
exogenous variables in first differences, that is excluding ∆u, and including labor taxes and replacement rates.
Differenced variables for the first nine period are computed as described in footnote 22.
Columns 3 - 4 : All RHS variables in differences. Instruments: All RHS variables in this table, expressed in
levels and first differences lagged twice and once respectively; time dummies; unemployment lagged once in
differences and twice in levels. Unemployment replaced by one minus the male employment rate (in percentage
points) in columns and 4 .
Column [5]: Growth of gross real wages instrumented as from column [1] in table 11 and growth of per-
capita GDP as from column [1] in Table 12.
Columns 6 - 7 : All RHS variables in first differences. Instruments: All the exogenous variables in this
Table, expressed in first differences lagged once and in levels lagged twice; unemployment in first differences lagged
once and in levels lagged twice; time dummies.
Intercepts not reported. Standard errors in parentheses.
* = 10% level of significance
** = 5% level of significance
*** = 1% level of significance
Figure 1: Growth and unemployment
1965-1991
Growth, in deviation from country means
Unemployment, in deviation from country means
-7.5 -2.5 2.5 7.5
-7.5
-5.0
-2.5
0.0
2.5
5.0
7.5
10.0
Figure 2: Labor taxes and unemployment in continental Europe
1965-1991, 5-year averages
Labor taxes, deviations from country means
Unemployment rates, deviations from country means
-12.0 -4.0 4.0 12.0
-7.5
-5.0
-2.5
0.0
2.5
5.0
7.5
10.0
Figure 3: Labor taxes and unemployment outside continental Europe
1965-1991, 5-year averages
Labor taxes, deviations from country means
Unemployment rates, deviations from country means
-12.0 -4.0 4.0 12.0
-7.5
-5.0
-2.5
0.0
2.5
5.0
7.5
10.0
