Download Fiva-ngs05 762327 en

4th Norwegian-German Seminar
on Public Economics
Grand Hotel Sonnenbichl, Garmisch-Partenkirchen
16 – 17 September 2005
Fiscal Decentralization, Interjurisdictional
Competition and Welfare Spending
Jon H. Fiva
CESifo
Poschingerstr. 5, 81679 Munich, Germany
Phone: +49 (89) 9224-1410 - Fax: +49 (89) 9224-1409
[email protected]
http://www.cesifo.de
Fiscal Decentralization, Interjurisdictional Competition and Welfare Spending
Jon H. Fiva*
Center for Economic Research
Department of Economics, Norwegian University of Science and Technology,
N-7491 Trondheim, Norway
(E-mail: [email protected])
Abstract
This paper studies the relationship between fiscal decentralization and welfare spending.
Utilizing new panel data on fiscal decentralization for 19 OECD countries, a robust negative
relationship is established. The results are interpreted to be driven by fiscal competition.
Fiscal decentralization yields increased fiscal competition which makes it harder for welfare
states to redistribute between income groups.
JEL Classification: H11, H53, H73, H77
Keywords: Fiscal federalism, sub-national fiscal autonomy, fiscal competition, redistribution.
First draft: May, 2005.
This version: August 3, 2005
Comments welcome!
* I thank Dan Stegarescu for making his data available to me. Funding from the Norwegian
Research Council under the FIFOS program is highly appreciated.
1
1. Introduction
A large literature in public finance going back to Stigler (1957) and Musgrave (1959) has
warned against the consequences of decentralized responsibility for redistribution. Policies
that are redistributive in nature give rise to a phenomenon that resembles adverse selection:
net beneficiaries of redistributive policies are attracted to generous jurisdictions, while net
contributors are repelled (Wildasin 1991, 1994). This gives sub-central levels of government
incentives to behave strategically in the setting of tax rates and welfare benefits. Empirical
analyses have provided strong support for this notion of strategic interaction (for a summary
see Wilson (1999) on tax competition and Brueckner (2000) on welfare competition). The
main concern in the theoretical literature is that such fiscal competition can lead to levels of
taxation and welfare benefits that are suboptimal seen from the society’s point of view.
Nonetheless, have many countries assigned considerable responsibility for redistribution to
sub-central levels of government. Consequently, one would expect to find that more fiscally
decentralized countries will facilitate stronger intergovernmental competition, yielding a
smaller public sector.1 Following Oates (1985) a large empirical literature has looked for
evidence of downward pressure on tax revenues from decentralization of taxing powers.
There exists some empirical evidence suggesting a negative relationship between fiscal
decentralization and the size of government (see the discussion in Stein, 1999:365 and
Rodden, 2003). However, an even harder task than to establish the relationship between
decentralization and the size of government is to assert the welfare consequences of such a
relationship. The welfare consequences are dependent on how one envisions public-sector
decision-making. The literature on collective choice suggests a range of possibilities including
everything from benevolent planners who seek to maximize the ‘well-being of society’, to
utility maximizing bureaucrats with preferences for budgetary slack. In the former case fiscal
competition is clearly welfare-reducing.2 But when governments not always act in the best
interest of the citizens, fiscal competition may help to constrain a public sector that would
otherwise have been inefficiently large (the argument of Brennan and Buchanan (1980)). In
this paper I do not explicitly aim to evaluate whether fiscal competition is “good or bad”, I
simply acknowledge that a more decentralized governmental structure will make it harder to
carry out redistribution. Hence fiscal decentralization is expected to be negatively associated
1
Note that if the strategic interaction is driven by yardstick competition it is not clear that overall redistribution
goes down with increased decentralization.
2
The magnitude of the welfare loss is however an open question.
2
with welfare spending. This is also what the empirical cross country literature on
decentralization and redistribution has found. However these studies often rely on a simple
federalism dummy to capture fiscal decentralization (as for example Swank, 2002: 89, insert
more references). But as Rodden (2004:496) points out – if intergovernmental competition is
driving the negative effect of decentralization on redistribution, a simple dummy for
federalism seems to be a very poor proxy.
This paper offers a systematic analysis of fiscal decentralization on welfare state effort. To
motivate the analysis I present a model based on Wildasin (1991). Wildasin’s benchmark
model of interjurisdictional migration shows how decentralized responsibility for the
‘redistribution branch’ yields lower redistribution in equilibrium. The empirical analysis is
based on panel data from 19 OECD countries3 over the period 1965 to 1999. The analysis is
distinctive in at least two ways. First, I base inference on both cross country and within
country variation. This empirical methodology is more rigorous than what is common in the
welfare state literature, which typically base it’s inference on pure cross country variation or
pooled ordinary least squares (OLS). Here I also pursue these strategies, but in addition I
evaluate how a full battery of dummy variables for both country and year fixed effects affects
the results. Second, I apply a new data set developed by Stegarescu (2004) that focuses on
sub-central government revenue decentralization. Stegarescu’s contribution is to differentiate
between revenue of sub-central government levels according to their ability to determine
revenue sources autonomously. Strict use of account data, which is common in the fiscal
decentralization literature, may give rise to confounded results because correspondence
between budgetary items and actual decision making might be imperfect.
Measuring welfare spending as social security transfers of GDP, I find evidence consistent
with the interjurisdictional competition hypothesis. Estimation methods that rely on cross
country and within country inference both suggest that a one standard deviation increase in
fiscal decentralization reduces the share of spending on social security out of gdp with around
2 percentage points.
3
The 19 countries included in the analysis are: Australia, Austria, Belgium, Canada, Denmark, Finland, France,
Germany, Ireland, Japan, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, the United
Kingdom and the United States.
3
The rest of the paper is organized as follows. Section 2 presents the theoretical framework and
the main hypothesis. Section 3 describes the data on fiscal decentralization, while section 4
introduces data on welfare spending and presents the econometric design. In section 5 the
results are presented. Section 6 discusses possible endogeneity problems and section 7
concludes.
2. Theoretical framework
A large literature initiated by Stigler (1957) and Musgrave (1959) warns against decentralized
responsibility for redistribution. Sub-central governments face incentives to set taxes and
transfers to influence the location of households. When mobility is high this may give a level
of redistribution which is ‘too low’ seen from the country’s point of view. To clarify this
migration externality, I adapt David Wildasin’s (1991) model of income distribution in a
common labor market, also discussed by Brueckner (2000) and Saavedra (2000). Wildasin’s
benchmark model of interjurisdictional migration shows how decentralized responsibility for
the ‘redistribution branch’ yields lower redistribution in equilibrium. The central mechanism
is the mobility of welfare clients. In Wildasin’s model it is the endogenous determination of
wages that equilibrate migration flows and prevents a complete ‘race to the bottom’. Wheaton
(2000) presents an alternative framework where the equilibrating mechanism is idiosyncratic
preferences for location among the poor.
Consider a country composed of I local governments, indexed from i= 1, …I, with two kinds
of households, ‘rich’ and ‘poor’. Each poor household is endowed with 1 unit of labor and are
assumed to be perfectly mobile across local governments at no migration costs. The rich
households are endowed with other factors of production and are assumed to be immobile.
The overall sizes of both groups are fixed. The wages of the poor are determined in a common
competitive labor market and thus reflect the marginal productive of the poor. Each
jurisdiction produces a numeraire good with similar production technology given by f i li ,
which is a strictly increasing and concave function of the number of poor households, li ,
employed in jurisdiction i. The wage in a jurisdiction is hence equal to wi
households earn the remaining income yi
f i ' li and rich
f i li f i ' li li . The latter represents rents to rich
households earned by other factors of production, possibly income from rich households
labor.
4
Redistribution from rich to poor is in this model driven by altruistic preferences. The rich
households care for the poor living in their region and are willing to incur tax liabilities to
support redistributive transfers. Each welfare client (poor households) in jurisdiction i thus
receives a transfer of income denoted, bi , and each taxpayer (rich households) pays an equal
bi li
, where ni is the number of rich households
ni
share of the total transfers in the jurisdiction,
in the jurisdiction. Let ui yi , zi denote the utility of the rich, where zi
wi bi is the income
of the poor households. The utility function is increasing and concave in both arguments.
Because of costless mobility of welfare clients zi is equal for all jurisdictions. Thus across all
jurisdictions z
wi bi for some net income, z, and consequently we have the following
relation securing migration equilibrium:
f j' l j b j , i z j .
fi ' li bi
(1)
The equilibrium mechanism is the assumption of a common labor market ensuring that wages
equilibriate migration flows. Let L denote the total number of poor households in the
economy, then in equilibrium the following condition must hold:
I
¦l
L.
i
(2)
i 1
Equation (1) and (2) determine the distribution of welfare clients across jurisdictions and their
common net income, z , conditional on bi , i=1, …, I. Differentiating (2) with respect to b j
yields
I
wli
¦ wb
i 1
0,
(3)
j
and differentiating (1) with respect to b j yields
5
wz
wb j
fi '' li wli
1, for i
wb j
wz
wb j
fi '' li wli
, for i z j
wb j
j
,
(4)
and rearranging
wli
wb j
wz
1
1
˜ ''
''
, for i
wb j f i li f i li wli
wb j
wz
1
˜ ''
, for i z j
wb j f i li j
,
(5)
Substituting this into (3) to solve for z as a function of the parameters (b1 ,..., bI ) yields
wz
wb j
where V j
1
/
f l j ''
j
1
V j ! 0,
(6)
.4 And (5) can be written:
I
¦ f l ''
i
i
i 1
wli
wb j
V j 1
wli
wb j
Vj
f i '' li f i '' li ! 0, for i
j
.
(7)
0, for i z j
When b j increases, jurisdiction j is more attractive and poor households migrate from other
jurisdictions into jurisdiction j. Without the common labor market which introduces offsetting
wage movements, then all the poor would move to the jurisdiction with the highest benefits.
4
V j  > 0,1@ . When welfare clients are evenly distributed across all local governments then V j
1
.
I
6
Moving on to the choice of benefit levels, I assume that the decision is taken by a
representative rich household in each jurisdiction. Assuming that each rich household receives
an equal fraction (1/n) of total non-poor income, then his utility is given by:
§ f l fi ' li li bi li '
·
, f i li bi ¸¸ .
u yi , zi u ¨¨ i i
ni
ni
©
¹
(8)
Each jurisdiction maximizes (8), taking into account the migration effect in (7) and viewing
uz
other jurisdictions benefit levels as fixed. Letting MRS yi , zi uy
denote the marginal rate
of substitution between poor people’s consumption and own consumption, then:
MRS yi , zi wyi / wbi
.
wzi / wbi
(9)
Considering a symmetric Nash equilibrium where number of welfare clients are evenly
distributed across all jurisdictions, then
wzi
wbi
1
I
wyi
wbi
1
ni
§ 1
1
1·
˜ bi ( 1) li ˜ ¸
¨¨ ''
I
I ¸¹
© f i li ,
(10)
and consequently the first order condition can be written:
ni ˜ MRS yi , zi 1
f i li ''
˜ bi (1 I ) li .
(11)
The right hand side of (11) gives the private marginal social cost to taxpayers in jurisdiction i.
This yields ‘too low’ provision of welfare benefits seen from the society’s point of view. To
understand this, consider the first order condition in the no-mobility case:
ni ˜ MRS yi , zi li .
(12)
7
Because the RHS of (11) is larger than the RHS of (12) the welfare benefits are larger in the
mobility than in the no-mobility case. This reflects that the marginal cost of increasing the
welfare of the poor is larger in the mobility case. To see the intuition behind this result
consider the representative tax payer’s decision problem. He compares altruistic gains from
helping the poor to an increase in the tax burden. If the poor do not move, then the tax burden
rises only because each of a fixed number of poor recipients receives a larger benefit.
However, when welfare migration occurs, the size of the jurisdiction’s poor population grows
as its welfare benefit becomes more generous. Generosity is thus more costly with welfare
migration. Note that because the concern about welfare migration depress welfare benefits in
all jurisdictions, no jurisdiction succeeds in repelling welfare clients and the equilibrium is
characterized by all jurisdictions setting lower benefits than they would in the no-mobility
case. The welfare benefits are therefore “too low” seen from the nation’s point of view
(Brueckner 2000).
The key point in this model is that redistributive activity in any one jurisdiction creates
external benefits for other jurisdictions because their tax bases increase and their
redistributive burdens diminish. With decentralized responsibility for redistribution, this
externality is not internalized by taxpayers in jurisdiction i. The socially optimal benefit level
in each state corresponds to the one chosen in the immobility case. This gives rise to the
following hypothesis: Fiscal decentralization yields increased fiscal competition which makes
it harder for welfare states to redistribute between income groups.
3. Measuring fiscal decentralization – new data on decentralization of tax revenue
Fiscal decentralization reflects how responsibilities for tax revenues and public expenditures
are distributed among different tiers of government. The complexity of the governmental
structure makes it challenging to capture this in one single dimension. The standard approach
is to make use of accounting measures such as revenue or expenditure shares for sub-central
relative to central government to proxy for fiscal decentralization. As thoroughly discussed by
Ebel and Yilmaz (2003), Rodden (2004) and Stegarescu (2004) this approach might be
problematic. Whether sub-central governments’ expenditure is funded by intergovernmental
grants, some revenue sharing program or own-source revenue through independent taxes and
user charges clearly makes a difference. What one typically observes is that decentralization
appears as higher in the expenditure than the revenue dimension. Such vertical fiscal
8
imbalance implies that one should rather look for measures of ‘revenue decentralization’ than
‘expenditure decentralization’, in particular when one wants to use fiscal decentralization as a
proxy for fiscal competition (which is the case here). However utilizing account data for
revenue decentralization can also be problematic since central governments can, and often do,
significantly limit the fiscal autonomy of lower levels of government. Thus, a larger share of
sub-central revenues out of total tax revenues does not necessarily imply greater fiscal
dependence from central government. Until recently, virtually all cross-country studies have
ignored the distinction between locally determined taxes, ‘piggybacked taxes’ or shared taxes
(Stegarescu, 2004:3).5 OECD (1999) tries to overcome this problem and present cross country
data which explicitly focused on the role of taxation in determining the fiscal autonomy of
sub-national governments. The study aimed to classify taxes in terms of the kind of autonomy
they provided to state and local governments, hence focusing on ‘revenue decentralization’.
Stegarescu (2004) draws on the analytical framework provided by OECD (1999) and expands
their dataset to cover 23 OECD countries for a long time span (1965 to 2001). Stegarescu’s
data distinguishes between different kinds of sub-central government revenue according to the
degree of discretion sub-central governments have on determining them autonomously. In this
respect the data represent a major improvement compared to existing measures of fiscal
decentralization.
In this analysis I utilize Stegarescu’s data for 19 OECD countries. 4 countries are excluded
because of size (Luxembourg and Iceland) and uncertainty with respect to data availability
(Greece and Italy). The key explanatory variable in the empirical analysis conduct below is
decentralization. decentralization measure the revenue share of sub-central government
relative to general government, but contrary to what is common in the literature this variable
only includes revenues where the sub-central government has discretion over tax rate, tax
base or both.6 Stegarescu (2004) finds that using the conventional measure of revenue
decentralization typically overestimates the degree of fiscal decentralization. This is
particularly the case for Austria, Belgium, Germany and Portugal (Stegarescu, 2004: 11-12).
Descriptive statistics for decentralization is reported in appendix table A.1. The data show a
5
There are numerous cross-national studies that utilize ‘naïve’ accounting measures as proxy for fiscal
decentralization, see for example the studies of Panizza (1999) and Arzaghi and Henderson (2005) that use fiscal
decentralization as the dependent variable. Rodden (2004:481) provides an overview of studies that use fiscal
decentralization as an explanatory variable to explain the size of governments, economic growth, corruption and
macro economic stability. Most of these studies are cross-country analyses that use the Government Finance
Statistics (GFS) of the International Monetary Fund.
6
The variable is denoted ’tdec1’ in Stegarescu (2004).
9
trend towards an increasing role for sub-central governments in most of the countries. There is
a significant increase in local tax autonomy over time in for instance in Belgium, France and
Spain according to this measure. Some countries have also moved in the direction of less subcentral tax autonomy, such as Ireland, Norway and the United Kingdom.
4. Empirical specification
The full data matrix used in this paper comprises 19 OECD countries over 35 years (1965
through 1999). I follow among others Rodrik (1997), Huber and Stephens (2001), Garrett and
Mitchell (2001) and Swank (2002) and proxy for welfare spending by utilizing OECD’s
measure of transfer payments as a percentage of GDP (sstran) as dependent variable. Social
security transfers are defined as “benefits for sickness, old-age, family allowances, etc., social
assistance grants and welfare benefits paid by general government” (Armingeon et al., 2004).
Descriptive statistics for this variable for each country is reported in appendix table A.2.7
Comparing 1965 to 1995 we find that all countries have expanded their share of spending on
social security out of gdp. The data reveals large cross country differences, but also
considerable within-country differences. The average spending on social security out of GDP
increased from 9% in 1965 to 16% in 1995. This measure is certainly not a perfect measure of
welfare spending, but I believe it is reasonable to argue that it captures important aspects of
the welfare state, in particular the effort to carry out redistribution. Another approach to try to
tap ‘welfare state effort’ would be to directly try to measure reduction in inequality before and
after taxes and transfers. Bradley et al. (2003) and Iversen and Soskice (2004) follow this
approach and use data from the Luxembourg Income Study to measure “the percentage
reduction in the gini coefficient from before to after taxes and transfers” (Iversen and Soskice,
2004). Although some people are very optimistic about the “pre/post approach” (see for
example the introduction in Mahler and Jesuit, 2004) it also has it flaws (see the discussion in
Bergh, 2005). Hence, I have decided to focus on the OECD measure of social transfers.8 The
empirical specification is given by:
7
sstran is not available for all years for all countries, the total number of missing observations for sstran is 35, in
particular due to data on New Zealand (ends in 1983). Only France has interrupted time series. Decentralization
is missing for 63 observations (primarily because 6 countries only have data from 1973 onwards). Leaving out
Spain before 1977 and Portugal before 1975 leaves us with an unbalanced panel of 552 observations on 19
countries.
8
Mahler and Jesuit (2004:24) finds a raw correlation between the two measures of ‘welfare state effort’ of 0.72.
10
sstranit
D E decentralizationit J controlsit H it
(13)
where sstranit is welfare spending in country i at time t. E measures the effect of fiscal
decentralization on welfare spending and is the coefficient of interest. If fiscal
decentralization increases fiscal competition, which reduces the ability to redistribute, then
E will come out with a negative sign. To evaluate the importance of fiscal decentralization on
welfare spending it is essential to take into account all other potentially important
determinants of welfare spending, thus a matrix of controls is included.9
Cameron (1978) was one of the first to put the attention to the linkage between international
and domestic economy in explaining the growth of the public sector. He finds the openness of
the economy to be the best single predictor of the growth of public revenues relative to GDP.
He interprets this to be driven by heavier unionization in more open countries which increases
demand for public sector spending. If this assertion is correct then openness should be an
important predictor of social security spending. In my regressions I control for both openness
(measured as the sum of exports and imports divided by GDP) and share of union members in
the labor force (union). The latter variable is only included in some specification because I
haven’t been able to find data that covers all countries. In addition I control for per capita
income (l_rgdp) to capture ‘Wagner’s Law’ which implies income elastic demand for public
sector spending. Furthermore I control for country size (l_population) and the share of people
that are: living in rural areas (ruralpop), are under 15 years (under15) and are above 65 years
(above65), respectively. In addition to these proxies for political demand I also control
explicitly for partisanship by including the share of the cabinet from left (govleft) and center
(govcent) parties. Business cycle dynamics are often highlighted as central sources of
variations in social welfare provision, and unemployment is included to capture both
automatic entitlement pressures and political demands. Meltzer and Richard (1981) argue that
a more skewed income distribution will increase political support for redistribution. A similar
logic may apply to voter turnout (vturn), since it is reasonable to argue that non-voter turnout
is concentrated among the poor. Increased turnout is expected to increase support for
redistribution.10
9
Descriptives are included in appendix table A. 3.
Unemployment and voter turnout may be problematic control variables if they are dependent of current
welfare spending and consequently endogenous. The results reported below are robust to the exclusion of these
variables.
10
11
Utilizing panel data, inference can be based on both cross country and within country
variation. Pooling all the data and running an ordinary least squares (OLS) regression on (13)
provide consistent and unbiased results only if the error term can be considered as random
across countries and over time. This is a strong assumption to make. A potential remedy is to
estimate a restricted version of (13) which includes a full battery of time and country fixed
effects, given by:
sstranit
D i G t E decentralizationit J controlsit H it
(14)
A problem with this approach is that sstran and decentralization do vary considerably less
within than across countries. Including country fixed effects, thus, removes a lot of variation
in the data. Hence, it can be fruitful to base the analysis on both estimations of (13) and (14).
The standard approach in welfare state research is to rely on a specification like (13) (see for
example the book length studies of Huber and Stephens (2001) and Swank (2002)). This
approach is also common in the ‘Searching for Leviathan’ studies (initiated by Oates (1985)).
In section 4 I present estimation results both based on one year ‘snapshots’ as well as OLS on
means, pooled OLS, random effects (RE) and fixed effects (FE) estimation.11
5. Results
A first investigation of the relationship between fiscal decentralization and welfare spending
is provided in table 1.
Table 1 about here.
Specification (1), (2) and (3) are ‘snapshot’ cross country regressions from 1975, 1985 and
1995. The effect of decentralization on sstran is as expected negative for all years, but it is
only statistically significant for the 1975 regression. Pooling all the years together yields a
coefficient of -0.07 which is highly statistically significant.12 Including a full set of time
dummies to soak up common time specific shocks does not alter this result (specification (5)).
11
The estimation procedure is based on annual data, but one could also utilize data averaged over multi-year
periods. Although the latter approach reduces the degrees of freedom it might make it more sensible to include
time lags in the specification and will reduce measurement error if measurement error is not correlated over time.
In future work I will test the model on data with 5-years averages.
12
Pooling the data can give a downward bias in the standard errors. I intend to present panel corrected standard
errors in the next revision of the paper.
12
The coefficient is estimated to be -0.09. Finally I apply the pure between estimator (OLS on
means) and find similar results. A coefficient of approximately -0.1 indicates a moderate, but
far from irrelevant economical impact. A one standard deviation increase in decentralization
reduces the share of social security out of gdp with around 1.7 percentage points, or around
1/3 of a standard deviation. The result is consistent with the hypothesis presented in section 2.
When local governments have responsibility for redistribution, household mobility introduces
a migration externality which yields lower redistribution in equilibrium. Generous
redistributive programs will serve to attract low-income households and chase away those
with higher incomes whose taxes must finance the transfers. In the political decision making
process this aspect is taken into account when determining taxes and transfers. The negative
association reported in table 1 is thus interpreted to be driven by interjurisdictional
competition.
Previous research has found a negative relationship between a simple dummy for federalism
and welfare spending. Cameron (1978) for example, whose main contribution was to discuss
the role of an open economy in promoting public spending, found that federalism ‘dampens
the degree of expansion in the public economy’ (Cameron 1978:1253). Some have interpreted
the negative effect of federalism to be driven by fiscal competition. Consequently it is of
interest to investigate whether the effect of decentralization is robust to the inclusion of a
dummy for federalism. In table 2 I report alternative specifications where federation is a
dummy equal to 1 if the country is a federation and zero otherwise.13 A simple dummy for
federation does not have the same explanatory power as decentralization. However, when
both decentralization and federation are included (specification (9) and (10)), they both have a
statistically significant impact on welfare spending. Surprisingly the two effects seem to be
quite independent of each other.14 The effect of fiscal decentralization is unaltered when
federation is included in the pooled OLS specification (comparing (5) to (9)). And the effect
of federation is estimated to be in the same range as a one standard deviation increase in fiscal
decentralization. The pooled OLS specifications suggest that federal institutions yield a
reduction in social security out of gdp of around 2 percentage points. I’ve also tried an
interaction term with federation and decentralization, but it came out statistically insignificant
and is not reported (t-value of -0.7).
13
The following countries were coded as federations: Australia, Austria, Belgium (since 1993), Canada,
Germany, Spain, Switzerland and the United States.
14
Note that the raw correlation between decentralization and federation is a mere 0.33.
13
Table 2 about here.
Cross-country evidence has a number of shortcomings. As discussed above, it may be
problematic to base inference on variation between countries if cross section heterogeneity is
large. If there are some inherent features of different countries that affect welfare spending
which are not accurately captured by any of the included regressors, than the correct approach
is to include a full set of country dummies. Garrett and Mitchell (2001) criticize the standard
approach in welfare state research and argue that leaving out country fixed effects is likely to
give substantial bias in the results. In the following I report different specifications that take
into account country fixed effects (FE). These are reported in table 3. The pooled OLS
estimator is also included for comparison. Specification (11) is based on the random effects
(RE) estimator which can be considered a ‘compromise’ between OLS and FE. Baltagi
(2001:15-18) shows that RE is a matrix weighted average of the within estimator (FE) and the
between estimator (OLS on means) where each estimate is weighted with the inverse of the
corresponding variance. Pooled OLS on the other hand gives equal weight to both between
and within variation. RE is going to be closer to FE the larger the part of the variance of the
error term that is due to the country specific term.
Table 3 about here.
Controlling for country specific fixed effects, I again find the estimated effect of
decentralization on sstran to be statistically significant.15 Interestingly the coefficient of
interest is of the same magnitude in all the specifications, from -0.09 to -0.15. The effect of
decentralization is not sensitive to leaving out vturn or unemployment or including union. I
conclude that the OLS effect is robust to the inclusion of a full battery of time and country
fixed effects.
Other determinants of welfare state effort
On theoretical grounds it is not clear how increased integration into the world economy
affects welfare spending. On the one hand, is economic integration likely to create
15
The Hausman test rejects the null hypothesis that the random effects specification is appropriate (pvalue=0.003), comparing (11) and (12).
14
competition for cross country mobile factors in a similar fashion to the theoretical framework
put forward in section 2. Hans-Werner Sinn, among others, has been concerned about this
development for the European welfare states: “… opening the border and allowing factors of
production to move freely across them makes it more and more difficult to maintain the
welfare state. … The people who make a net contribution to financing the state transfer their
economic activities to the low-tax countries to rid themselves of their responsibilities, and the
net recipients of government resources congregate in the well-established welfare states where
they the make the existing financial problems even worse” (Sinn, 2003:65). However, it can
also be argued that welfare spending is expected to increase if governments expand the
welfare state to provide a cushion against external risks. The existing empirical evidence on
the effect of economic integration on welfare spending is ambiguous. Relying on cross
country data Rodrik (1997) finds a strong positive association between openness and welfare
spending which he understands to be driven by increased demand for a more active
government role in providing social insurance. However when he controls for year and
country fixed effects he finds that increased openness have resulted in reductions in social
spending. My analysis reproduces these findings and is also consistent with Garrett and
Mitchell (2001) who also rely on FE estimation. I.e. it seems like countries that traditionally
have been very open to trade (for example Belgium, Denmark, Netherlands and Norway) have
more generous welfare states and that this is reflected in the OLS regressions. However, the
FE estimations suggest a negative effect of openness on welfare spending. Relying on
specification (12), an increase in the shares of imports plus exports in GDP of 5 percentage
points (a total of 10 percentage points) results in a reduction in welfare spending of about 0.3
percentage points.16 Controlling for the share of the work force that is members in unions the
effect is almost three fold. Closer integration into the world economy seems to make it harder
to undertake redistributive taxation and sustain generous social programs.
The remaining controls show the more or less expected pattern. Based on fixed effects
estimation, I find that country size seems to be negatively associated with welfare spending.
However since openness and l_population are highly collinear one might worry that
multicollinearity makes it hard to separate out the true effects.17 I have tried specifications
where openness and l_population respectively are left out, but the coefficients are basically
16
This is close to identical to Rodrik’s (1997) estimate.
Alesina and Wacziarg (1998) argues, for example, that it is not openness, but country size that actually
explains government expenditure. This does not seem to be the case here.
17
15
unaltered. I find strong positive effects of unemployment and voter preferences (proxied for
by population shares). Based on specification (13) I find that a 1 percentage point increase in
the unemployment rate increases the share of social security out of gdp with 0.49 percentage
points. Lindert (1996:10) finds that higher voter turnout seems to raise all kinds of
government expenditures. This result is only confirmed in the cross country regressions.
Wagner’s law is also supported by the cross country estimates, but it is not robust to the
inclusion of fixed effects. The impact of unionization on welfare spending is strong and
positive. Union is nonetheless only included in specification (13) because including it reduces
the sample substantially, in particular because two countries are excluded (Portugal and
Spain). I do not find any robust effects of political color. Christian democratic government
seems to be associated with more welfare spending but the effect vanishes if unions are
controlled for. Note that my model only captures the short-term impact of Left and Center
governments.
6. Endogeneity
The results above suggest a robust negative relationship between fiscal decentralization and
welfare spending. However, one can question whether this reflects a causal relationship from
fiscal decentralization to welfare spending. The estimation method utilized above hinges upon
the assumption of strict exogeneity of fiscal decentralization. Acemoglu (2005) argues, in a
more general setting, that political institutions hardly can be taken as exogenous: “This is
because different political institutions will generate different policies, and thus lead to
different economic outcomes, and rational economic (and political) agents should understand
not only the implications of different policies but also the implications of different political
institutions. The logic of political economy therefore forces us to also think of political
institutions as endogenous” (Acemoglu, 2005: 12). Previous research on decentralization has
also found decentralization to be a complex process which is a product of many factors
including cultural heritage and geography (Arzaghi and Henderson (2005) and Panizza
(1999)). In this setting it will be particular problematic to take decentralization as exogenous
if preferences of political actors for decentralization are affected by the current level of
redistributional spending. Note that if the political pressure to centralize is stronger in
countries with less welfare spending, then this will give an attenuation bias in the estimates
above. However, whether there is a bias and in what direction it goes is an open empirical
question. To address this problem an instrumental variable which is correlated with fiscal
16
decentralization, but not with welfare spending is warranted. A variable that incorporates the
geographic variance in demands for public goods should according to Oates’ (1972)
decentralization theorem be a candidate for a valid instrument. I hope to be able to address the
potential endogeneity problem in future work.
7. Conclusions
This paper studies the relationship between fiscal decentralization and welfare state effort.
Utilizing new panel data on fiscal decentralization for 19 OECD countries, a robust negative
relationship is established. The estimated coefficient is remarkably stable through a variety of
specifications including the pure between estimator, pooled ordinary least squares, random
effects and fixed effects estimation. A one standard deviation increase in fiscal
decentralization is estimated to reduce welfare spending out of gdp with approximately 2
percentage points. The results are interpreted to be driven by fiscal competition: due to
household mobility, countries with decentralized responsibility for redistribution find it harder
to redistribute between income groups. Generous redistributive programs will serve to attract
low-income households and chase away those with higher incomes whose taxes must finance
the transfers.
In future work I will have a closer look at the within country variation. Several countries have
gone through considerable reforms in the time period under study and it would be interesting
to study these reforms in particular. Moreover I also intend to have a closer look into the
possible endogeneity problem related to fiscal decentralization.
17
References
Acemoglu, Daron (2005): “Constitutions, Politics and Economics: A Review Essay on
Persson and Tabellini’s The Economic Effects of Constitutions”, NBER working paper
no. 11235.
Aleisna, Alberto and Romain Wacziarg (1998): “Openness, Country Size and Government”,
Journal of Public Economics 69, 305-321.
Armingeon, Klaus, Phillip Leimgruber, Michelle Beyler and Sarah Menegale (2004):
Comparative Political Data Set 1960-2002, Institute of Political Science, University of
Bern.
Arzaghi, Mohammad and J. Vernon Henderson (2005): “Why Countries Are Fiscally
Decentralizing”, Journal of Public Economics 89, 1157-1189.
Baltagi, B.H. (2001): Econometric Analysis of Panel Data, John Wiley & Sons Ltd,
Chichester, England.
Bergh, Andreas (2005): “On the Counterfactual Problem of Welfare State Research: How Can
we Measure Redistribution?”, European Sociological Review 21, xxx – xxx.
Bradley, David, Evelyn Huber, Stephanie Moller, Francois Nielsen and John Stephens (2003):
“Distribution and Redistribution in Postindustrial Democracies”, World Politics, 193228.
Brennan, Geoffrey and James Buchanan (1980): The Power to Tax, Cambridge: Cambridge
University Press.
Brueckner, Jan K. (2000): “Welfare Reform and the Race to the Bottom: Theory and
Evidence”, Southern Economic Journal 66, 505-525.
Cameron, David R. (1978): “The Expansion of the Public Economy: A Comparative
Analysis”, American Political Science Review 72, 1243-1261.
Ebel, Robert and Serdar Yilmaz (2003): “On the Measurement and Impact of Fiscal
Decentralization,” in Jorge Martinez-Vazquez and James Alm (eds.) Public Finance in
Developing and Transitional Countries, Cheltenham, UK: Edward Elgar.
Garrett, Geoffrey and Deborah Mitchell (2001): “Globalization, Government Spending and
Taxation in the OECD”, European Journal of Political Research 39, 145-177.
Heston, Alan, Robert Summers and Bettina Aten (2002): “Penn World Table Version 6.1”,
Center for International Comparisons at the University of Pennsylvania (CICUP).
Huber, Evelyn and John D. Stephens (2001): Development and Crisis of the Welfare State,
University of Chicago Press, Chicago.
18
Huber, Evelyne, Charles Ragin, John D. Stephens, David Brady, and Jason Beckfield (2004):
Comparative Welfare States Data Set, Northwestern University, University of North
Carolina, Duke University and Indiana University, 2004.
Iversen, Torben and David Soskice (2004): “Electoral Institutions, Parties, and the Politics of
Class: Why Some Democracies Redistribute more than Others”, mimeo, Harvard
University.
Lindert, Peter H. (1996): “What Limits Social Spending?”, Explorations in Economic History
33, 1-34.
Mahler, Vincent A. and David K. Jesuit (2004): “State Redistribution in Comparative
Perspective: A Cross-National Analysis of the Developed Countries”, LIS working
paper no. 392.
Meltzer, Alan H. and Scott F. Richard (1981): “A Rational Theory of the Size of
Government”, Journal of Political Economy 89, 5, 914-927.
Oates, Wallace (1972): Fiscal Federalism, Harcourt, New York.
Oates, Wallace (1985): “Searching for Leviathan: An Empirical Study”, American Economic
Review 75, 748-757.
OECD (1999): “Taxing Powers of State and Local Government”, OECD Tax Policy Studies
1.
Panizza, Ugo (1999): “On the Determinants of Fiscal Centralization: Theory and Evidence”,
Journal of Public Economics 74, 97-139.
Rodden, Jonathan (2003): “Reviving Leviathan: Fiscal Federalism and the Growth of
Government”, International Organization 57, 695-729.
Rodden, Jonathan (2004): “Comparative Federalism and Decentralization, on Meaning and
Measurement”, Comparative Politics 36, 481-500.
Rodrik, Dani (1997): “Has Globalization Gone too Far?”, Washington DC: Institute for
International Economics.
Saavedra, Luz A. (2000): “A Model of Welfare Competition With Evidence from AFDC”,
Journal of Urban Economics 47, 248-279.
Sinn, Hans-Werner (2003): The New Systems Competition, Blackwell Publishing, Oxford.
Stegarescu, Dan (2004): “Public Sector Decentralization: Measurement Concepts and Recent
International Trends”, ZEW Discussion Paper No. 04-74, Mannheim.
Stein, Ernesto (1999): “Fiscal Decentralization and Government Size in Latin America”,
Journal of Applied Economics 2, 357-391.
19
Stigler, George J. (1957): “The Tenable Range of Functions of Local Government”, Joint
Economic Committee, Federal Expenditure Policy for Economic Growth and
Stability, reprinted in Wallace E. Oates (ed.), 1998 The Economics of Fiscal
Federalism and Local Finance, Cheltenham, Edward Elgar, pp 3-9.
Swank, Duane (2002): Global Capital, Political Institutions, and Policy Change in Developed
Welfare States, Cambridge University Press, Cambridge, United Kingdom.
Wheaton, William, (2000): “Decentralized Welfare: Will There be Underprovision?”, Journal
of Urban Economics 48, 536-555.
Wildasin, David E., (1991): “Income Redistribution in a Common Labor Market”, American
Economic Review 81, 757-774.
Wildasin, David E. (1994): “Income Redistribution and Migration”, Canadian Journal of
Economics 27, 637-656.
Wilson, John D. (1999): “Theories of Tax Competition”, National Tax Journal 52, 269-304.
20
23.24***
1.52**
0.21***
98.41**
139.78***
-4.17*
-1.83
0.11
L_RGDP
L_POPULATION
RURALPOP
UNDER15
OVER65
GOVLEFT
GOVCENT
UNEMPLOYMENT
No
No
Country fixed effects
18
18
OLS (1985 only)
17
17
OLS (1975 only)
Number of countries
Number of observations
OLS (1995 only)
18
18
0.07
0.16
2.76
3.10
57.71
39.07
0.09
0.96
6.81
3.02
0.05
St.error
0.667
No
No
0.22***
0.61***
3.85
-6.19**
265.61***
45.52
0.25***
-1.31
20.73***
-3.67
-0.01
Coefficient
(3)
Pooled OLS
552
19
0.02
0.03
0.48
0.41
10.31
7.25
0.02
0.17
0.82
0.74
0.01
St.error
0.583
No
No
0.12***
0.23***
3.04***
-0.66
106.87***
12.07*
0.10***
0.06
6.47***
0.07
-0.07***
Coefficient
(4)
Pooled OLS
552
19
0.02
0.05
0.46
0.40
9.71
7.03
0.02
0.16
1.09
0.71
0.01
St.error
0.651
No
Yes
0.12***
0.39***
2.25***
-0.87**
118.93***
6.07
0.15***
0.19
12.14***
2.21***
-0.09***
Coefficient
(5)
No
No
OLS on means
19
19
0.16
0.38
6.35
8.63
112.04
64.64
0.12
1.40
8.70
7.54
0.08
St.error
0.291
0.11
0.17
5.71
-2.10
113.67
19.77
0.11
-0.24
12.28
-0.14
-0.12
Coefficient
(6)
21
Note: Standard errors in parentheses. ***, ** and * denotes significance at 1%, 5% and 10% level respectively. A constant term is included in all regressions (not reported).
Estimation method
0.646
0.11
0.29
3.39
2.09
61.74
42.86
0.12
1.17
8.08
5.43
0.08
St.error
0.837
R2adj
No
0.11
0.55*
-0.49
4.68**
89.26
-19.20
0.28**
2.50**
18.16**
19.49***
-0.08
Coefficient
No
0.06
0.31
3.37
2.26
38.55
39.32
0.06
0.71
4.39
4.98
0.06
St.error
(2)
Year fixed effects
0.17***
13.31**
OPENNESS
VTURN
-0.13**
DECENTRALIZATION
Coefficient
(1)
Table 1 Determinants of welfare spending (measured as SSTRAN) based on cross and within country variation
19
552
OLS on means
19
552
Pooled OLS
Number of countries
Number of observations
Pooled OLS
552
19
0.01
St.error
0.651
No
Yes
-0.09***
Coefficient
(5)
No
No
OLS on means
552
19
0.08
St.error
0.290
-0.12*
Coefficient
(6)
Pooled OLS
552
19
0.30
0.01
St.error
0.676
No
Yes
-1.87***
-0.09***
Coefficient
(9)
OLS on means
552
19
2.11
0.07
St.error
0.486
No
No
-4.04***
-0.16***
Coefficient
(10)
table 1 is included in all regressions (not reported).
22
Note: Standard errors in parentheses. ***, ** and * denotes significance at 1%, 5% and 10% level respectively. A constant term and control variables as in specification (5) in
Estimation method
R
0.223
2.53
St.error
0.636
adj
No
No
2
Country fixed effects
-3.03
No
0.31
Coefficient
Yes
-2.01***
St.error
(8)
Year fixed effects
FEDERATION
DECENTRALIZATION
Coefficient
(7)
Table 2 Comparing federalism to fiscal decentralization as determinants of welfare spending (SSTRAN)
12.14***
L_RGDP
6.07
118.93***
-0.87**
2.25***
0.39***
0.12***
UNDER15
OVER65
GOVLEFT
GOVCENT
UNEMPLOYMENT
VTURN
No
No
Country fixed effects
19
552
RE
19
552
Pooled OLS
Number of countries
Number of observations
0.02
0.04
0.36
0.26
12.11
7.32
0.03
0.44
1.10
1.08
0.02
St.error
0.00
Yes
Yes
FE
552
19
0.873
0.41***
0.80**
-0.29
120.76***
36.43***
0.17***
-6.44***
1.11
-3.14***
-0.15***
Coefficient
(12)
0.02
0.04
0.37
0.27
14.05
7.82
0.04
2.46
1.19
1.22
0.03
St.error
Yes
Yes
FE
464
19
0.910
9.20***
0.01
0.49***
0.17
-0.44
80.82***
52.96***
0.31***
-3.57
-1.55
-8.79***
-0.15***
Coefficient
(13)
1.56
0.02
0.05
0.37
0.28
17.15
8.36
0.05
2.98
1.20
1.50
0.03
St.error
0.30
-0.25
FE
558
19
0.850
Yes
Yes
103.55***
33.20***
0.19***
-8.83***
-1.20
-3.08**
-0.14***
Coefficient
(14)
0.39
0.29
14.91
8.29
0.05
2.66
1.20
1.32
0.03
St.error
23
Note: Standard errors in parentheses. ***, ** and * denotes significance at 1%, 5% and 10% level respectively. A constant term is included in all regressions (not reported).
Estimation method
0.376
0.651
R2adj
Yes
0.03
0.41***
0.98***
-0.39
113.38***
34.56***
0.07**
-0.59
2.27**
-2.39**
-0.10***
Coefficient
Yes
0.02
0.05
0.46
0.40
9.71
7.03
0.02
0.16
1.09
0.71
0.01
St.error
(11)
Year fixed effects
UNION
0.15***
RURALPOP
0.19
2.21***
OPENNESS
L_POPULATION
-0.09***
DECENTRALIZATION
Coefficient
(5)
Table 3 Determinants of welfare spending (measured as SSTRAN) based on within country variation
24
Country
Mean
St. dev.
Minimum
Maximum
1965
1975
1985
1995
Australia
20.33
1.75
18.19
24.23
N/A
19.94
18.55
22.47
Austria
3.48
0.21
2.90
4.03
N/A
3.62
3.31
3.76
Belgium
11.32
7.38
5.31
24.55
6.59
6.30
6.88
23.61
Canada
51.56
2.19
46.31
55.36
52.14
47.08
52.44
54.66
Denmark
29.50
1.65
25.96
31.97
N/A
29.02
27.63
31.48
Finland
25.91
2.06
22.67
32.18
23.14
26.90
26.15
29.50
France
12.48
7.18
1.13
20.05
N/A
2.18
15.59
19.96
Germany
7.46
0.40
6.49
8.01
N/A
7.76
7.89
6.49
Ireland
5.85
4.00
2.14
13.78
13.78
8.66
2.72
2.81
Japan
32.02
2.41
27.80
37.54
28.33
32.17
34.07
34.13
Netherlands
3.83
0.92
1.92
5.20
N/A
1.92
4.41
4.73
New Zealand
7.47
0.90
6.29
9.07
9.07
7.18
6.37
5.17
Norway
26.96
2.91
23.11
31.51
N/A
31.51
23.11
25.53
Portugal
1.57
1.28
0.00
3.36
N/A
0.00
0.41
2.85
Spain
10.17
3.64
5.68
21.98
N/A
N/A
8.16
12.86
Sweden
39.88
4.94
30.77
47.56
30.77
36.31
40.46
47.02
Switzerland
56.71
2.10
52.41
61.50
54.33
61.50
57.06
56.99
United Kingdom
10.94
3.77
4.74
14.25
13.77
13.56
12.89
4.88
United States
36.93
1.68
33.42
39.08
34.38
38.30
37.52
38.58
Note: decentralization is defined as sub-central government own tax revenue divided by general government total tax revenue. Only tax revenue
from taxes where the sub-central government autonomously can change the tax rate, tax base or both are included in the nominator.
Appendix Table A.1 Descriptive statistics for decentralization for each country
25
Country
Mean
St. dev.
Minimum
Maximum
1965
1975
1985
1995
Australia
8.39
1.00
6.30
9.90
5.80
8.60
9.60
8.60
Austria
18.78
1.39
15.30
21.10
11.23
16.90
20.40
19.50
Belgium
17.49
3.11
12.40
22.70
12.40
18.80
21.70
16.60
Canada
10.65
2.32
6.00
14.60
6.10
10.20
12.20
13.20
Denmark
16.90
2.49
11.10
21.20
8.30
13.80
16.30
20.40
Finland
13.02
5.53
7.00
24.00
7.00
8.50
14.80
22.20
France
19.10
1.89
15.50
22.10
16.40
20.40
22.10
18.50
Germany
16.87
1.38
13.50
19.30
12.30
17.60
16.20
18.10
Ireland
11.64
3.16
5.90
17.20
5.90
12.70
16.60
11.30
Japan
9.32
3.21
4.50
14.60
4.80
7.70
11.00
13.40
Netherlands
23.27
4.92
12.60
28.80
11.50
24.10
26.10
15.30
New Zealand
12.33
2.45
9.71
16.84
9.89
11.58
N/A
N/A
Norway
15.19
0.99
13.30
17.10
9.00
13.60
14.80
15.80
Portugal
11.38
1.15
8.60
13.90
N/A
8.60
10.80
11.70
Spain
15.57
1.95
10.40
18.40
N/A
N/A
16.00
17.20
Sweden
16.78
4.08
9.20
23.40
9.20
14.20
18.20
21.30
Switzerland
11.14
2.37
6.50
14.10
7.40
12.50
13.70
11.20
United Kingdom
11.85
2.59
7.60
16.00
7.60
9.90
13.80
15.40
United States
10.22
2.24
5.40
13.00
5.40
11.40
11.00
13.00
Note: sstran is defined as social security transfers as a percentage of GDP. Consists of benefits for sickness, old-age, family allowances etc.,
social assistance grants and welfare benefits paid by general government.
Appendix Table A.2 Descriptive statistics for sstran for each country
Total trade (export + imports) as a share of RGDPL.
Gross union members out of total labor force
OPENNESS
UNION
16.71
0.04
0.03
1.27
13.51
4.20
0.39
0.32
13.40
0.28
0.28
0.17
21.93
0.22
0.13
16.62
27.72
6.10
0.35
0.24
76.86
9.72
0.52
0.40
B
E
D
D
A**
D
A**
A**
A**
C
C
C
C
C
4.78
St.dev
14.14
Mean
A*
Source
0.07
0.09
8.70
35.00
0.00
0.00
0.00
14.78
2.95
0.15
0.06
0.00
4.50
Min
0.84
1.77
10.32
95.80
1.00
1.00
24.10
19.42
71.98
0.33
0.18
61.50
28.80
Max
original source see Armingeon et al. (2004). Descriptives based on 552 observations (union excluded).
Comparative Welfare States Data Set (Huber et al. (2004)). *Original source is OECD historical statistics, various years, table 6.3. ** For
Stegarescu’s data set (2004), (C) The World Development Indicators, (D) The Penn World Table Version 6.1 (Heston et al. (2001)) and (E)
Note: The data are collected from five different data sources: (A) The Comparative Political Data Set (Armingeon et al. (2004), (B) Dan
Log of RGDPL
Social security transfers as a percentage of GDP. Consists of benefits
for sickness, old-age, family allowances etc., social assistance grants
and welfare benefits paid by general government.
Sub-central government own tax revenue divided by general
government total tax revenue. Only tax revenue from taxes where the
sub-central government autonomously can change the tax rate, tax
base or both are included in the nominator.
Share of population 0-14 years of age
Share of population 65 years of age and above
Total population
Log of total population
Share of population living in rural areas
Unemployment rates, standardized as far as possible according to
OECD criteria
Cabinet composition: share of government from left parties, weighted
by days
Cabinet composition: share of government from center parties,
weighted by days
Voting turnout in percent
Real GDP per capita in 1996 international dollars (Laspeyeres)
Definition
L_RGDP
VTURN
RGDP
GOVCENT
GOVLEFT
UNEMPLOYMENT
YOUNG
OLD
POPULATION
L_POPULATION
RURALPOP
DECENTRALIZATION
SSTRAN
Variable
Appendix Table A.3
26