European Journal of Political Economy 24 (2008) 562570

Contents lists available at ScienceDirect

European Journal of Political Economy

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e j p e

Fiscal federalism and the composition of public investment in Europe

Andreas Kappeler a, Timo Vlil b,
a
b

Munich Graduate School of Economics, Ludwig-Maximillians-Universitt, D-80539 Munich, Germany

Economic and Financial Studies Division, European Investment Bank, L-2950 Luxembourg

a r t i c l e

i n f o

Article history:
Received 10 October 2007
Received in revised form 6 June 2008
Accepted 9 June 2008
Available online 16 June 2008
JEL classiaction:
H54
H77
H72
C23
C24

a b s t r a c t
We analyse the determinants of the composition of public investment in Europe, with a special
focus on the role of scal decentralisation. The results suggest that scal decentralisation boosts
economically productive public investment, notably infrastructure, and curbs the relative share
of economically less productive investment, such as recreational facilities. While not readily
reconcilable with the traditional theory of scal federalism, these ndings can be interpreted in
terms of the literature on scal competition, with not only tax rates but also the quality of public
expenditure weighing in rms' location decisions.
2008 Elsevier B.V. All rights reserved.

Keywords:
Public investment
Fiscal federalism
Dynamic panel data

1. Introduction
Public investment has received only limited academic attention as an aggregate variable, and its composition has to our
knowledge received none at all, at least in the European context. This paper seeks to ll that gap at least in part by presenting an
empirical analysis of what drives different types of public investment, with a special focus on the impact of scal federalism.
Perhaps because of lack of academic attention, misconceptions abound concerning the nature, drivers, and impact of public
investment. Most notably, there is often confusion about what it is in the rst place. Perhaps the most prominent example of
this type of confusion is the customary synonymous use of public investment and infrastructure investment in much of
economic literature. There is, however, a great deal of infrastructure investment that is not public, and there is a great deal of
public investment that is not infrastructure investment. While it is well-known that many roads, water and sanitation
networks, and municipal swimming pools are publicly funded and provided, neither economic theory nor empirical analyses
have really distinguished between them when studying what determines public investment or how productive public
investment is.
As a starting point for a more nuanced analysis and understanding of public investment, we rst break it down into different
types with distinctly different economic characteristics in Section 2. We then propose to use the traditional theory of scal
federalism and some of its more recent extensions, reviewed in Section 3, to derive hypotheses about the link between scal

The authors would like to thank, without implicating, two anonymous referees as well as seminar participants at the EIB and the University of Munich (LMU)
for comments on earlier drafts of the paper.
Corresponding author. Tel.: +352 4379 88679; fax: +352 4379 68895.
E-mail address: t.valila@eib.org (T. Vlil).
0176-2680/$ see front matter 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.ejpoleco.2008.06.002

A. Kappeler, T. Vlil / European Journal of Political Economy 24 (2008) 562570

563

Table 1
Functional breakdown of public investment
Aggregation

ESA 95 COFOG

1. Infrastructure (INF)
2. Hospitals and Schools (HS)

Economic Affairs
Health
Education
Defence
General Public Services
Environment
Order and Safety
Housing
Recreation
Social Protection

3. Public Goods (PG)

4. Redistribution (RED)

Source: Eurostat; own aggregation.

decentralisation and the composition of public investment. Section 4 seeks to articulate empirical tests of the hypotheses, and their
results are interpreted from an economic perspective in Section 5, before concluding in Section 6.
2. Data on the composition of public investment in Europe
To the best of our knowledge, no empirical analyses have been conducted with a focus on the composition of public investment,
at least in the European context. Therefore, we start off by describing the available data in this section.
Based on the functional classication of government expenditure in the 1993 UN System of National Accounts and in the 1995
European System of Accounts (ESA 95), Eurostat provides a breakdown of public investment for EU countries starting in the early
1990s. Complete data are available for EU15 countries from 1995 (i.e., the introduction of ESA 95) through 2005.1 However, many
countries have back-dated their time series to 1990.
The public investment variable is gross capital formation of the general government. This includes changes in inventories,
which may create some undesired noise for our analysis; however, the breakdown between gross xed capital formation and
changes in inventories is not available.
The functional breakdown of public investment is presented in Table 1. The right-hand side column shows the functional
classication (Classication of Functions of Government, COFOG for short) in ESA 95. The left-hand side shows our aggregation
of the 10 available functions into four types of public investment with economically distinct roles. This aggregation will be
used in the remainder of this paper; however, we also consider alternative groupings as a robustness check in the empirical
analysis below.
The four different types of public investment affect the economy through different channels, with varying degrees of directness,
and over different time horizons. Public investment in Infrastructure, consisting of just Economic Affairs in the ESA 95 COFOG2,
seeks to measure public investment in traditional infrastructure, mainly transport. This type of public investment has the most
direct economic impact by reducing rms' production and transaction costs. The economic impact of public investment in
Hospitals and Schools is more long-term and less direct in character, as it facilitates the building up and maintenance of the
economy's stock of human capital. Investment in Public Goods affects the economy's allocative efciency indirectly through
framework conditions for productive activity. Finally, Redistribution affects the economy's income distribution rather than
allocative or productive efciency per se.
In addition to the composition of Infrastructure investment, some other aggregates shown in Table 1 contain undesirable
noise as no further breakdowns of the right-hand side functions are available. For example, public investment in water supply
and wastewater management are not part of Infrastructure as one would wish; instead, they are part of Redistribution (Housing)
and Public Goods (Environment), respectively. Similarly, one would wish to include street lightning in Public Goods; now it is in
Housing and thereby Redistribution. However, as with Infrastructure, we expect such noise to be of sufciently small magnitude
so as not to invalidate the empirical analysis below.
3. Public investment and the theory of scal federalism
The theory of scal federalism or any other theory for that matter does not deal explicitly with the composition of public
investment. At best, it distinguishes between consumption-oriented public expenditure and public expenditure to produce public

1
EU15 comprises Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal, Spain, Sweden, and the
United Kingdom.
2
Economic Affairs comprise a number of different sectors, including agriculture; fuel and energy; mining, manufacturing, and construction; transport;
communication; R&D; and others. Among these sectors, transport is likely to be by far the dominant recipient of public investment. Note that investment by
energy companies owned by the public sector, for example, is classied as corporate investment in national accounts statistics as long as such companies are
commercially run.

564

A. Kappeler, T. Vlil / European Journal of Political Economy 24 (2008) 562570

inputs for the production processes of private rms. In what is to come we do not consider differences between current public
spending and public investment per se; rather, we consider the various types of public investment as enhancements of production
potential for different public services. Thus, infrastructure investment is considered to produce more future transportation
services, and redistribution investment is considered to produce, e.g., more future recreation services. This perspective allows us to
link the theory of scal federalism with the kind of data on the composition of public investment that we have.
The traditional theory of scal federalism is based on the seminal contributions by Tiebout (1956), Oates (1972), and Musgrave
(1959). The underlying assumptions include, most importantly, the benevolence of the policy-maker in the centre (that is, his
objective is the maximisation of social welfare); the existence of pure local public goods and global public goods (whose benets
accrue locally and nation-wide, respectively); benet taxation (same incidence for the cost and benet of public spending); factor
mobility; and absence of spillover effects of scal decisions horizontally (between regions) and vertically (between regions and the
centre).
Considering the responsiveness of public spending to local preferences and the creation of incentives for economic efciency as
policy goals, the theory derives normative conclusions about the optimal task assignment between the central and sub-national
levels of government. Responsiveness to local preferences implies that decentralisation and scal competition are preferable in the
provision of local public goods whenever local preferences are heterogeneous. On the other hand, centralisation is warranted in the
provision of public goods whose optimal supply cannot be achieved by scal competition. Such goods include most notably global
public goods, but also the macroeconomic stabilisation and income redistribution functions of the government (which may be
interpreted as providing global public goods as well). Public goods may also have spillover effects, with one region beneting from
a highway built by its neighbouring region, for example. Fiscal competition among sub-national levels of government will result in
a sub-optimally low level of provision of such goods, as regions do not consider the spillover benets in their individual decisionmaking. Oates (1972) suggests that the optimal provision can be achieved by means of matching grants from the centre, which act
to internalise the externality.
We have thus far identied three types of public goods (local, global, and spillover public goods) and the optimal level of
government to provide each of them. We can now consider the different types of public investment in Table 1 against this
background. Infrastructure, such as roads and other transportation infrastructure, provide both local benets and positive spillover
effects, in so far as it connects localities and regions. Hospitals and Schools provide also local benets and positive spillover effects;
the latter is especially the case when the labour force and population at large are mobile and move across regions. Public Goods, as
dened in Table 1, is a mixture of local and global public goods, while Redistribution comprises chiey local public goods.
So how would one expect scal decentralisation to affect the different types of public investment? Investment in local public goods,
most notably Redistribution, would unambiguously increase with decentralisation. Investment in Infrastructure as well as Hospitals
and Schools would also increase with decentralisation, especially if supplemented with grants from the centre. Investment in Public
Goods could go either way, depending on whether the aggregate Public Goods is more local or global in character.
More recent literature on scal federalism has relaxed the assumption of no spillover effects in policy-making. Focussing on
horizontal policy spillovers, consider regional tax competition.3 With capital mobile across regions that seek to attract it, tax
competition can lead to sub-optimally low tax rates (race to the bottom) and, as a consequence, insufcient provision of public
services (both public consumption goods and infrastructure). The standard reference is Zodrow and Mieszkowski (1986);
however, Sinn (2003) has come out strongly against their analysis.4 Hulten and Schwab (1997) discuss the circumstances where tax
competition can lead to a sub-optimally low level of public capital. Competition between regions for an industry with external
scale economies is a case in point: in competing for the location of such an industry, regions may reduce their tax rates so low as to
unduly suppress public investment.
Considering the impact of scal competition on the composition of public expenditure, Keen and Marchand (1997) argue that
uncoordinated scal competition induces regions to over-invest in local public inputs at the cost of (consumption-oriented) local
public goods. Investment in public inputs increases the potential of regions to attract mobile private capital, since public inputs
reduce production costs for private rms. This generates distortions in the composition of public expenditure. Decentralisation
leads to a relative over-supply of public inputs and an under-supply of local public goods.
To sum up, scal competition has been argued to reduce public investment across the board (tax competition), but it has also
been argued to boost productive public investment, at least relative to local public goods (broader scal competition). In terms of
the public investment types in Table 1, these results would imply that decentralisation increases investment in Infrastructure as
well as Hospitals and Schools, while reducing investment in Redistribution, at least in relative terms. This contrasts, notably, with
the hypotheses above based on the older scal federalism literature.
4. Empirical analysis
4.1. Model specication
Although it is possible to formulate hypotheses of the relationship between decentralisation and the composition of public
expenditure (investment) as in Section 3, there is no explicit theoretical framework that could be used to derive a model of the
3
We ignore here the literature of vertical scal externalities (see, e.g., Dahlby, 1996; Dahlby and Wilson, 2003; Martinez-Lpez, 2005). The predictions of that
literature are ambiguous, hinging on assumptions whose relevance for our data sample we cannot assess.
4
See also Matsumoto (1998).

A. Kappeler, T. Vlil / European Journal of Political Economy 24 (2008) 562570

565

determination of different types of public investment. We will therefore proceed directly to the specication of a reduced-form
model to be estimated. In so doing we seek to identify exogenous variables measuring the impact of decentralisation on public
investment, as well as a set of control variables that render the model empirically well-specied.
The reduced-form specication to be used is as follows:
Ic;it 1 taxit1 2 capit 3 gdpit1 4 debtit1 5 lendit1 6 popit1 i uit

where uit i.i.d. (0, 2), with subscript i referring to observations in the cross-section dimension (individual countries) and t to
observations in the time dimension. The dependent variable Ic represents public investment of type c, c = 1,, 4, as shown in Table
1. Ic is expressed relative to trend GDP5, thus in theory assuming values in R+.
The rationale for using trend GDP to scale both the dependent variable and some regressors, as indicated below, is to minimise
spurious correlation between variables. Using, e.g., nominal GDP instead would add to the scaling variable cyclical, seasonal, and
random GDP uctuations. This in turn would imply a risk of introducing spurious correlation between the dependent variable and
regressors as well as among the regressors. Besides, using trend GDP improves the time series properties of the variables and
facilitates the interpretation of the estimation results, as the coefcient estimates are not directly affected by short-term GDP
variability.6
Fiscal decentralisation is measured by two explanatory variables. First, our primary interest is in the share of tax revenue
attributed to sub-national levels of government (regional and local governments), which is denoted tax. Second, we control for
investment grants from the central government to sub-national levels of government (cap); in the empirical analyses it is
measured in relation to trend GDP.7 The tax share is lagged by one period to reect the fact that investment decisions are most
often taken a year before, based on knowledge about the revenue situation at that time. In contrast, capital transfers are
contemporaneous with investment, as they nance investment the same year it is undertaken (Rodden, 2003).
We also consider alternative measures of scal decentralisation for comparison. We focus on alternatives proposed by
Stegarescu (2005), which distinguish between different types of control over tax revenues. Specically, such measures focus on to
what extent sub-national levels of government control tax rates, tax bases, both, or none. Note that we do not consider alternative
measures based on expenditures (e.g., expenditure share of sub-national levels of government or the ratio of sub-national tax
revenue to expenditure), as they include the dependent variable by construction, and we also do not consider alternative measures
based on total revenue shares, as they include capital transfers.
Turning then to the control variables, they seek to capture the general economic, scal, and demographic developments of
signicance for the determination of public investment. As shown theoretically by Turrini (2004) and empirically by Mehrotra
and Vlil (2006), aggregate public investment varies alongside GDP as well as the short- and long-term scal position. There
are, of course, no a priori grounds to expect that these variables affect different types of public investment similarly, or that
they are signicant determinants of every type of public investment. To take an example, an episode of scal consolidation may
be associated with a reduction in less productive public investment (such as recreational facilities) while keeping the more
productive infrastructure investment intact, or it may be associated with cuts across the board. Another control variable, not
considered in the studies above focussing on aggregate public investment but conceivably signicant for some types of public
investment is population density, as the needs for and cost of infrastructure networks differ between densely and sparsely
populated areas.
Real GDP, denoted gdp in (1), is measured in per capita terms and lagged by one period. The short- and longer-term scal
environment is captured by the budget surplus of the general government (lend) and public debt (debt). Both are measured in
relation to trend GDP and lagged by one period. A reason for using the one-period lags of the real GDP and scal variables is to
address any bias arising from the possible joint determination of these variables and the dependent.
Population density is denoted pop. i denotes unobserved time-invariant country-specic effects that are included in the
estimations (unless otherwise indicated).
Finally, we also considered a number of other regressors to test the robustness of our results. They included unemployment,
birth rates, migration rates, and mortality rates, all of which could plausibly affect the different types of public investment (see, e.g.,
De Haan and Sturm, 1997, and Falch and Ratts, 1997). These additional regressors were, however, insignicant most of the time
and their inclusion did not change the estimation results materially. We therefore focus on model specication (1) in what is to
come.
4.2. Sample properties
The main sample used in the estimations consists of a panel of EU10 countries (EU15 less the Cohesion countries less
Luxembourg) during the period 19902005. We exclude the Cohesion countries from our sample, because public investment in
those countries has been signicantly inuenced by the receipt of EU support. As explained in Section 2, not all countries have
back-dated all relevant series to 1990, so the panel is unbalanced.

5
6
7

Trend GDP is calculated using the HodrickPrescott Filter with a smoothing parameter = 100.
Using nominal GDP instead of trend GDP as a scaling variable does not change any of the signs or the statistical signicance of the estimation results.
The interaction term of tax and cap turned out to be insignicant in most of the estimations below and is therefore omitted.

566

A. Kappeler, T. Vlil / European Journal of Political Economy 24 (2008) 562570

Table 2
Im, Pesaran and Chin (IPS) panel unit root tests

1. INF
2. HS
3. PG
4. RED
4. RED (trend)
4. RED (1st diff)
Lend
Lend (trend)
Lend (1st diff)
Debt
Debt (trend)
Debt (1st diff)
Cap
Gdp
Gdp (trend)
Gdp (1st diff)
Otaxshl
Otaxshl (trend)
Otaxshl (1st diff)
Pop
Pop (trend)
Pop (1st diff)

Statistic

P-value

2.57763
2.27868
1.83078
1.19257
1.31247
4.33572
1.04089
1.27982
3.80819
0.58686
2.64085
4.51514
1.94034
5.18343
4.58604
3.53691
0.03522
1.84063
9.02904
4.39171
3.25404
2.05281

0.0050
0.0113
0.0336
0.1165
0.0947
0.0000
0.1490
0.1003
0.0001
0.7214
0.0041
0.0000
0.0262
1.0000
0.0000
0.0002
0.4860
0.0328
0.0000
1.0000
0.0010
0.0200

Note: individual effects included throughout in the test equation, trend when indicated. 1st diff denotes testing for unit root in rst differences. Lag length
selection based on the AIC.

The main data source is Eurostat's New Cronos database. Only the data on the scal variables (budgetary surplus and public
debt) as well as population come from the OECD.
Table 2 reports the results of the Im, Pesaran and Chin panel unit root tests (Im et al., 1997).8 Of the dependent variables,
Infrastructure, Hospitals and Schools, and Public Goods are all level stationary (at 5% signicance level), while Redistribution is not.
It is trend stationary (albeit only at 10% signicance level) and difference stationary. The variable lend (net lending, or budgetary
surplus, of the general government) is neither level nor trend stationary at 10% signicance level, but it is difference stationary. Cap
(capital transfers) is level stationary at 5% signicance level. All other variables are trend and difference stationary, but not level
stationary.
In sum, the variables that are not level stationary are either trend or difference stationary. Trend stationarity is sometimes a
borderline case, but difference stationarity is always clear.
The dependent variables are highly autocorrelated and persistent, with rst-order autocorrelation coefcients between 0.8 and
0.9 for all types of public investment. Correlation among our explanatory variables is mostly negligible. Only correlation
coefcients between the tax share variable and GDP per capita and population density are rather high at 0.65 and 0.44,
respectively.
4.3. Estimation methodology and results
4.3.1. Estimation methodology9
As our dependent variables are highly autocorrelated, we choose a dynamic specication of the model (1), including the lagged
dependent variable as another explanatory variable. The dynamic model specication thus becomes:
Ic;it 0 Ic;it1 1 taxit1 2 capit 3 gdpit1 4 debtit1 5 lendit1 6 popit1 i uit

The estimation of specication (2) will have to account for the correlation between the regressors (lagged dependent) and the
composite term (i + uit), which renders least squares estimators inconsistent even asymptotically. To circumvent this problem we
employ General Method of Moments (GMM) estimation, which has become the workhorse in estimating dynamic panel data
models.10,11

8
We focus on the Im, Pesaran and Chin panel unit root test as it allows for cross-section heterogeneity in the autoregressive coefcients. We have, however,
also performed the Levin, Lin and Chu (2002) panel unit root test, which does not allow for cross-section heterogeneity. Both tests yield exactly the same
conclusions about the stationarity properties of our series.
9
All estimations are conducted using eViews 5.1 or Stata 8.e or 9.
10
See Arrelano and Bond (1991) and Bond (2002).
11
We performed the DurbinWuHausman endogeneity test, which conrms that GDP, capital transfers, and the tax share variable are weakly exogenous. The
only possible exception may include the tax share variable with Hospitals and Schools as the dependent variable; however, at 5% level of signicance even this
variable is weakly exogenous throughout.

A. Kappeler, T. Vlil / European Journal of Political Economy 24 (2008) 562570

567

Table 3
1-step GMM estimation results (dependent variable type of investment relative to trend GDP)
1-step GMM

1.INF

2.HS

3.PG

4.RED

Ic(lag)

0.47711
(4.22)
0.01199
(3.01)
0.11478
(4.11)
0.32491
(2.53)
0.00022
(0.15)
0.00014
(0.19)
0.00006
(1.53)
0.4182
0.0856
0.2171
98
118

0.50599
(5.14)
0.01008
(2.11)
0.08925
(4.53)
0.32920
(1.63)
0.00009
(0.19)
0.00155
(1.84)
0.00010
(2.78)
0.8188
0.0776
0.5063
98
118

0.59714
(5.58)
0.01392
(4.30)
0.09606
(2.23)
0.45610
(4.61)
0.00129
(1.60)
0.00241
(2.62)
0.00011
(2.48)
0.6152
0.0226
0.1775
96
118

0.39505
(4.01)
0.00387
(0.77)
0.03845
(0.56)
0.36452
(2.85)
0.00217
(5.56)
0.00025
(0.26)
0.00006
(1.76)
0.7143
0.0640
0.9958
95
118

Tax
Cap
Gdp
Lend
Debt
Pop
Sargan (p-value)
m1 (p-value)
m2 (p-value)
Nobs.
No. moment conditions

Note: Heteroskedasticity-robust standard errors. Signicance at 10% level indicated in bold. t-values in parentheses.

4.3.2. Results
Table 3 presents the results of one-step difference-GMM estimation of Eq. (2). As the estimation is done in rst differences, the
stationarity properties of our data, as reported in Table 2, are appropriate. The Sargan test for over-identifying restrictions is used to
determine the instrument set, which includes further lags of the dependent variable and also three lags of the real GDP and the
capital transfers variables. The residual autocorrelation tests are labelled m1 for rst-order and m2 for second-order
autocorrelation.
The large number of moment conditions (see Table 3) warrants a special mention. There is a possible trade-off between bias and
efciency when the number of instruments (moment conditions) is increased with small samples (see, e.g. Roodman, 2007). In our
case the choice of a large set of instruments is based on the residual autocorrelation and Sargan test results (these tests often
suggest misspecication of the model and/or the instrument set with smaller instrument sets) and, more importantly, on the fact
that the parameter estimates are quite robust to using smaller instrument sets. For instance, using just 4 lags of the dependent as
instruments will not change the sign or signicance of the estimates for 1, which is of key interest to us, and it results in only
relatively minor changes to their numerical values. The only exception is Hospital and Schools which turns insignicant; however,
the residual autocorrelation tests suggest misspecication when the smaller instrument set is used. As bias is therefore not of
major concern, we use a large instrument set to gain efciency.
Considering the results in Table 3, we conclude that a higher sub-national tax share increases the aggregate level of investment
in Infrastructure, Hospitals and Schools, and Public Goods, but it has no statistically signicant impact on the aggregate public
investment in Redistribution. The parameter estimates imply that an increase in the sub-national tax share by one percentage
point leads to an increase in investment in Infrastructure and Public Goods of about 0.01 percentage points of GDP, or 1.6 and 2.3%,
respectively, evaluated at sample mean. The parameter estimate for Hospitals and Schools implies that a one percentage point
increase in the sub-national tax share leads to an average increase of 2.0% in investment.
These results are robust to alternative measures of scal decentralisation, as detailed in Appendix A.
Considering the coefcient estimates for capital transfers, we observe a signicant positive impact on investment in
Infrastructure, Hospitals and Schools, as well as Public Goods. An increase of capital transfers by 1% of GDP boosts these types of
investment by 0.11, 0.09, and 0.10 percentage points of GDP (9%, 15%, and 13%, respectively, at sample mean).
As regards other control variables, real per capita GDP is positive in all four models, with a coefcient estimate of 0.30.5 and
generally signicant: only for Hospitals and Schools its p-value is above 10% (0.103). The scal variables are mostly insignicant,
except that higher budgetary surpluses reduce investment in Redistribution and that higher public debt goes hand in hand with
higher investment in Public and Schools as well as Public Goods.12 Finally, population density turns out to be signicant for
investment types two to four.
As a robustness check, we also consider alternative plausible groupings of the different types of public investment (cf.
Table 1).13 Combining Hospitals and Schools with Public Goods into one single group yields a positive and signicant coefcient
estimate for the sub-national tax share variable (0.028), with all other parameter estimates retaining their sign and signicance.
Similarly when traditional Infrastructure and Hospitals and Schools (human capital infrastructure) are combined into one
group (coefcient estimate for the sub-national tax share variable 0.024) or when Public Goods and Redistribution are merged

12
13

The results do note change materially if the scal variables are excluded from the estimation.
The complete estimation results with the alternative groupings are available from the authors upon request.

568

A. Kappeler, T. Vlil / European Journal of Political Economy 24 (2008) 562570

(0.021). In sum, the signs and signicance of the coefcient estimates are robust with respect to alternative groupings of public
investment.
5. Economic interpretation of results
We saw in Section 3 how the traditional theory of scal federalism could be used to derive some hypotheses about the
composition of public investment. Most notably, it suggests that a higher degree of scal decentralisation should result in more
public investment in spillover goods, such as Infrastructure as well as Hospital and Schools, especially if accompanied by capital
transfers from the centre to internalise the spillover effects of such investments. Furthermore, more decentralisation should result
in more investment in Redistribution (local public goods) a result challenged by the more recent theory of scal competition.
Finally, the impact of decentralisation on our variable Public Goods was considered ambiguous, depending on whether it is
dominated by local or global public goods.
As shown in the empirical analysis, decentralisation in terms of tax shares increases public investment in Infrastructure,
Hospitals and Schools, and Public Goods. Public investment in Redistribution is not signicantly affected by decentralisation. An
interesting corollary of these results is that decentralisation reduces the relative share (but not the level) of Redistribution
investment in total public investment. As regards the supplementary role of capital transfers, it was found that they have a positive
impact on the same types of public investment as tax share (Infrastructure, Hospitals and Schools as well as Public Goods, but no
signicant impact on public investment in Redistribution).
Especially the result that decentralisation reduces the share of Redistribution investment is difcult to reconcile with the
traditional theory of scal federalism. Our Redistribution variable is meant to capture consumption-oriented local public goods,
such as recreational facilities, and decentralisation should lead to an increase, not relative decline, in their provision. On the other
hand, the positive impact of capital transfers on public investment in especially spillover goods (Infrastructure, Hospitals and
Schools) is in line with the traditional theory of scal federalism.
Our results can be interpreted in terms of scal competition (Keen and Marchand, 1997). Decentralisation increases the
level of investment in especially Infrastructure as well as Hospitals and Schools, both providing public inputs. What is more,
this increase suppresses the relative share of investment in Redistribution (local public consumption-oriented goods). It is
noteworthy that decentralisation does not lower the level of any type of public investment. This being the case, we do not see
any evidence of decentralization being associated with tax competition that would have a detrimental impact on public
investment.
As suggested above, scal competition may play a role by biasing sub-national governments' spending in favour of public
inputs, at the expense of consumption-oriented local public goods. However, the relative decline in Redistribution investment can
be given another interpretation as well. That its share actually declines with decentralisation can signal over-investment in
Redistribution in centralised systems (Ratts, 2003). With lower level governments competing for a common pool of resources,
there may be strategic reasons for them to misrepresent the local (or regional) demand for public services captured in
Redistribution. This being the case, decentralisation would reduce such strategic behaviour and bring Redistribution in line with
local demand.
6. Conclusion
The empirical analysis of the relationship between scal decentralisation and the composition of public investment is rst-of-akind, at least in the European context. Our results suggest that decentralisation increases economically productive public
investment, notably investment in public spillover goods (Infrastructure; Hospitals and Schools) and in local and global public
goods. There is no statistically signicant impact of decentralization on public investment in consumption-oriented local
public goods (Redistribution). Decentralisation changes the composition of public investment by reducing the relative share of
public investment in Redistribution.
While not readily reconcilable with the traditional theory of scal federalism, especially as regards the provision of local public
goods, these ndings can be interpreted in terms of the literature on scal competition, with not only tax rates but also the quality
of public expenditure weighing in rms' location decisions. The nding that decentralisation reduces the relative share of
Redistribution investment can also signal over-investment in more centralised system with competition for a common pool of
resources.
Clearly, this is but a rst step in the analysis of the composition of public investment. There is plenty of scope for future
research to tackle issues that our analysis leaves open. The theoretical foundations for studying the composition of public
investment remain thin, especially as regards the articulation of an explicit link between scal federalism and different types of
investment. Empirical examination of different types of public investment could usefully focus on differences in their
productivity, as well as on a more nuanced examination of what drives the different types of investment, including but not
limited to scal federalism.
Appendix A. Estimation results using alternative measures of scal decentralisation
To measure decentralisation, Stegarescu (2005) also accounts for the autonomy of sub-national governments to determine their
tax base and/or tax rate. He argues that federal systems are more decentralized if sub-national governments rather than central

A. Kappeler, T. Vlil / European Journal of Political Economy 24 (2008) 562570

569

governments can control regional tax revenues even if the regional tax share remains unchanged. This perception of
decentralisation does not only capture the regional tax share, but it also includes regional autonomy to raise these funds. The
concept can be translated into the classication of taxes by decreasing order of regional autonomy over revenue sources proposed
by the OECD (1999):
Table A1
Classification of taxes (in decreasing order of control over revenue sources)
(a)
(b)
(c)
(d)
(d.1)
(d.2)
(d.3)
(d.4)
(e)

Sub-central government (SCG) determines tax rate and tax base

SCG determines tax rate only
SCG determines tax base only
Tax sharing:
SCG determines revenue-split
Revenue-split only changed with consent of SCG
Revenue-split unilaterally changed by central government (CG) (legislation)
Revenue-split unilaterally changed by CG (annual budget)
CG determines tax rate and tax base

Source: OECD (1999).

Indeed, based on this classication of tax autonomy, Stegarescu (2005) proposed three decentralisation measures relative to the
tax revenue of sub-national governments. We are particularly interested in two of them:
TDec1

SCG own tax revenue a to c

GG total tax revenue

Decentralisation measure TDec1 only accounts for taxes whose base and/or rate is unilaterally xed by sub-national
governments. Thus, this denition of decentralisation is stricter than the overall tax share, used in Table 3. On the other hand
decentralisation measure TDec3 accounts for all seven categories of tax revenue:
TDec3

SCG own tax revenue a to e

GG total tax revenue

This denition is closer to the tax share variable used in Table 3 since it also contains taxes whose tax base and/or share is at
least partly determined by the central government.14 Re-estimating our model by using TDec1 instead of the tax share yields the
following results:
Table A2
Estimation results using decentralisation measure TDec1
1-step GMM

1. INF

2. HS

3. PG

4. RED

Ic(lag)

0.47100
(4.28)
0.01700
(2.46)
0.11021
(4.54)
0.39141
(2.23)
0.00051
(0.38)
0.00006
(0.06)
0.00005
(1.15)
0.4784
0.0748
0.1975
99

0.51573
(5.23)
0.01096
(2.23)
0.08736
(3.74)
0.38170
(1.67)
0.00042
(0.83)
0.00192
(1.56)
0.00010
(2.40)
0.7866
0.0729
0.6452
99

0.60644
(6.17)
0.01590
(4.78)
0.08683
(2.39)
0.55935
(4.31)
0.00073
(1.01)
0.00299
(2.36)
0.00013
(2.79)
0.7139
0.0270
0.1680
97

0.44764
(3.90)
0.00414
(0.91)
0.02990
(0.44)
0.36190
(2.44)
0.00185
(3.40)
0.00042
(0.38)
0.00005
(1.23)
0.8171
0.0410
0.8730
96

TDec1
Cap
Gdp
Lend
Debt
Pop
Sargan (p-value)
m1 (p-value)
m2 (p-value)
Nobs.

Note: Heteroskedasticity-robust standard errors. Signicance at 10% level indicated in bold. t-values in parentheses.

14
We do not consider decentralisation measure TDec2 in Stegarescu (2005), which includes tax revenues of type (a) to (c) as well as (d2). Statistical properties
of decentralisation measures TDec1 and TDec3 are similar for our sample. Therefore, a further distinction among categories of tax autonomy as suggested by
TDec2 would not allow for enough variation to gain additional insights.

570

A. Kappeler, T. Vlil / European Journal of Political Economy 24 (2008) 562570

Substituting TDec1 by TDec3 yields the results in Table A3:

Table A3
Estimation results using decentralisation measure TDec3
1-step GMM

1. INF

2. HS

3.PG

4.RED

Ic(lag)

0.47828
(4.53)
0.01210
(2.96)
0.11690
(4.22)
0.26827
(2.08)
0.00033
(0.25)
0.00063
(0.92)
0.00007
(1.81)
0.4763
0.0846
0.1993
99

0.49404
(6.39)
0.01169
(3.26)
0.08637
(3.87)
0.36648
(2.20)
0.00033
(0.78)
0.00165
(2.03)
0.00010
(3.09)
0.8228
0.0787
0.5695
99

0.60530
(6.05)
0.01408
(4.82)
0.09127
(2.51)
0.47233
(4.57)
0.00092
(1.16)
0.00243
(2.26)
0.00011
(2.29)
0.6772
0.0286
0.1542
97

0.43787
(3.87)
0.00526
(1.15)
0.02723
(0.42)
0.36714
(2.77)
0.00194
(3.77)
0.00037
(0.35)
0.00006
(1.40)
0.8159
0.0424
0.9413
96

TDec3
Cap
Gdp
Lend
Debt
Pop
Sargan (p-value)
m1 (p-value)
m2 (p-value)
Nobs.

Note: Heteroskedasticity-robust standard errors. Signicance at 10% level indicated in bold. t-values in parentheses.

Signicance of parameter estimates in general (and for decentralisation in particular) is similar to the results in Table 3: TDec1
as well as TDec3 are signicant for investment in Infrastructure, Hospitals and Schools, as well as Public Goods, whereas
coefcients for Redistribution are not. Thus, measures of decentralisation as proposed by Stegarescu (2005) conrm that higher
regional autonomy leads to more investment in public infrastructure, whereas Redistribution remains unaffected. This nding is in
line with our hypothesis that decentralisation generates strategic competition among regions resulting in increased public input
provision.
References
Arrelano, M., Bond, S., 1991. Some tests of specication for panel data: Monte Carlo evidence and an application to employment equations. Review of Economics
Studies 58, 277297.
Bond, S., 2002. Dynamic panel data models: a guide to micro data methods and practice. CeMMAP Working Paper CWP09/02. Institute for Fiscal Studies, London.
Dahlby, B., 1996. Fiscal externalities and the design of intergovernmental grants. International Tax and Public Finance 3, 397412.
Dahlby, B., Wilson, L.S., 2003. Vertical scal externalities in a federation. Journal of Public Economics 87, 917930.
De Haan, J., Sturm, J.-E., 1997. Political and economic determinants of OECD budget decits and government expenditures: a reinvestigation. European Journal of
Political Economy 13, 739750.
Falch, T., Ratts, J., 1997. Political economic determinants of school spending in federal states: theory and time-series evidence. European Journal of Political
Economy 13, 299314.
Hulten, C.R., Schwab, R.M., 1997. A scal federalism approach to infrastructure policy. Regional Science and Urban Economics 27, 139159.
Im, K.S., Pesaran, M.H., Shin, Y., 1997. Testing for unit roots in heterogeneous panels. Unpublished paper, Department of Applied Economics, University of
Cambridge.
Keen, M., Marchand, M., 1997. Fiscal competition and the pattern of public spending. Journal of Public Economics 66 (1), 3353.
Levin, A., Lin, C.-F., Chu, C.-S.J., 2002. Unit root tests in panel data: asymptotic and nite-sample properties. Journal of Econometrics 108, 124.
Martnez-Lpez, D., 2005. Fiscal federalism and public inputs provision vertical externalities matter. European Central Bank Working Paper Number 484,
Frankfurt.
Matsumoto, M., 1998. A note on tax competition and public input provision. Regional Science and Urban Economics 28, 456473.
Mehrotra, A., Vlil, T., 2006. Public investment in Europe: evolution and determinants in perspective. Fiscal Studies 27, 443471.
Musgrave, R.A., 1959. The Theory of Public Finance. McGraw-Hill, New York.
Oates, W., 1972. Fiscal Federalism. Harcourut Brace Jovanovich, New York.
OECD, 1999. Taxing powers of state and local government. OECD Tax Policy Studies No. 1. OECD, Paris.
Ratts, J., 2003. Fiscal federation or confederation in the European Union: the challenge of the common pool problem. Unpublished paper, Department of
Economics, Norwegian University of Science and Technology.
Rodden, J., 2003. Reviving Leviathan: Fiscal federalism and the growth of government. International Organization 57, 695729.
Roodman, D., 2007. A short note on the theme of too many instruments. Center for Global Development Working Paper Number 125, Center for Global
Development, Washington.
Sinn, H.W., 2003. The New Systems Competition. Basil Blackwell, Oxford.
Stegarescu, D., 2005. Public sector decentralization: measurement concepts and recent international trends. Fiscal Studies 26 (3), 301333.
Tiebout, C.M., 1956. A pure theory of local public expenditures. Journal of Political Economy 64, 416424.
Turrini, A., 2004. Public investment and the EU scal framework. Economic Paper No. 202, European Commission, Directorate-General for Economic and Financial
Affairs, Brussels.
Zodrow, G., Mieszkowski, P., 1986. Pigou, Tiebout, property taxation, and the underprovision of local public goods. Journal of Urban Economics 19, 356370.