Fiscal decentralization and economic growth: spending versus revenue decentralization.
Gemmell, Norman ; Kneller, Richard ; Sanz, Ismael 等
I. INTRODUCTION
Fiscal decentralization (hereafter: FD) is a political economy
trend in both developing and developed countries. According to the World
Bank (1999), some 95% of democracies now have elected subnational
governments, and countries everywhere are devolving political, fiscal,
and administrative powers to subnational tiers of government below the
national level. In developed countries the United States, the United
Kingdom, and Canada have revived debates on FD or devolution (Xie, Zou,
and Davoodi 1999). In recent years, the U.S. Congress has been
contemplating how to devolve more expenditure responsibility to State
and local governments. FD has also become a key issue in Japan because
the law for the promotion of fiscal decentralization was enacted in
1995. These efforts at devolution in a number of Organization for
Economic Co-operation and Development (OECD) countries are accompanied
by the emergence of a new top layer of government in the European Union
(EU). Stegarescu (2009) finds that European integration has favored the
fiscal decentralization process by increasing market size and the
benefits of decentralized provision of public goods in accordance with
comparative advantage and the inter-regional division of labor. The rise
of the regional level of government in Spain, Belgium, Italy, France,
and the United Kingdom are examples of this decentralization process in
the EU.
The movement toward FD is often justified by the widespread belief
that it is an effective tool for increasing the efficiency of public
expenditures and competition among subnational governments in delivering
public services. This may also be a reaction to the failure of large
centralized bureaucracies in developing and transitional countries
(Martinez-Vazquez and McNab 2003). The World Bank (1999), for example,
has argued that alongside globalization, localization--the increasing
demand for local autonomy--is the main force shaping the world in the
first decade of the twenty-first century.
In this article, we focus on a specific debate in the FD
literature--namely that it improves economic growth performance. We
summarize a number of the relevant arguments in Section II, and then
review the existing empirical evidence on the FD-growth relationship in
Section III. We argue that the existing literature is deficient in a
number of respects, for example by rarely testing simultaneously for
revenue and expenditure decentralization. Section IV presents our data
and empirical methodology, and Section V tests for an effect of FD on
economic growth rates in OECD countries over the period 1972-2005.
Section VI checks the robustness of our findings to alternative
econometric techniques to deal with endogeneity, and alternative
measures of fiscal decentralization. Section VII summarizes the main
conclusions.
II. ARGUMENTS FOR AND AGAINST FISCAL DECENTRALIZATION
The basic argument in favor of fiscal decentralization is that it
improves the efficiency of the public sector and promotes long-term
economic development (Oates 1972). The mainstream theory of fiscal
federalism, referred to by Oates (2005) as "first-generation"
theory argues that decentralization enhances economic efficiency because
local governments have better knowledge of local conditions and
preferences in the provision of public goods than national governments
due to their physical and institutional proximity. These informational
advantages allow local governments to deliver public goods and services
that better match local preferences and/or deliver the same public goods
and services at lower cost. These arguments are reinforced where public
good characteristics are local in nature (e.g., sharing economies or
nonexcludability aspects are geographically restricted). The
first-generation theory contends that if some local outputs can produce
inter-jurisdictional spillover effects, then central governments should
provide matching grants to decentralized government that would
internalize the benefits.
Secondly, Oates (1999) argues that by diversifying government
output according to local preferences, decentralization may attain
higher levels of social welfare. If preferences for public goods differ
across regions, uniform levels of public goods and services across
jurisdictions will generally be inefficient. The larger the variance in
regional demands for public goods, the larger the benefits of FD. This
diversification also allows residents to move to the community that best
matches their demand for public goods and services, and local tax rate.
Thus, a "Tiebout sorting" of individuals into
demand-homogeneous jurisdictions further increases efficiency in
resource allocation.
In addition, subnational governments may be subject to closer
scrutiny by their constituencies. Recent theoretical models stress that
one of the major advantages of decentralization is that it leads to
greater local accountability, such that decentralization may be
preferable even in cases of perfect homogeneity of preferences across
local jurisdictions. This greater accountability may also lead to
greater producer efficiency by providing incentives to local governments
to innovate in the production and supply of public goods and services
(Martinez-Vazquez and McNab 2003). Over 30 years ago Oates (1972) argued
that this allocative efficiency benefit becomes greater when there is a
close match between revenue discretion and spending assignments at
subnational levels. Such matching, it is argued, gives local government
a stronger fiscal incentive to support local market development (Jin,
Qian, and Weingast 2005), improves accountability of subnational
governments and reduces the distorting effects of intergovernmental
transfers (Shah 1994). Local jurisdictions, therefore, need to weigh
these benefits of proposed public programs against their costs (Oates
2005).
Building on the Tiebout (1956) mechanism, Brueckner (2006) proposes
a model in which FD leads young and old consumers to live in separate
jurisdictions according to their different demands for public services:
low and high. This sorting increases after-tax income when young while
reducing it when old, increasing the incentive to save, which, in turn,
leads to an increase in investment in human capital and long-term
economic growth, (1) Kappeler and Valila (2008) also find that FD
increases productive public investment and reduces the relative share of
economically less productive public investment, because fiscal
competition increases the quality of public expenditure and affects
firms' location decisions. By changing the composition of public
investment toward the most productive, FD may therefore enhance economic
growth. Finally, Schaltegger and Feld (2008) show that, contrary to the
popular claim that decentralized governments undermine policy
makers' ability to fight fiscal imbalance, FD increases the
probability of a successful "fiscal consolidation" (lower
public debt). In particular, FD increases the credibility and
accountability of mutually agreed reductions in fiscal imbalances,
whereas federal transfer payments may reduce the costs of making these
fiscal adjustments. Thus, FD could increase economic growth if the
reduced fiscal imbalances that it encourages have expansionary economic
effects (Giavazzi and Pagano 1996).
However, the theoretical effects of FD on economic growth are not
unambiguously positive. Firstly, FD may impact negatively on the
distribution of public resources across jurisdictions, because mobility
of households and businesses can seriously constrain attempts to
redistribute income. Redistributional policies are likely to induce poor
individuals to move into the jurisdiction while higher income
individuals (who bear a greater tax burden) move out. Along these lines,
Fiva and Rattso (2006) find that decisions of Norwegian local
governments about welfare benefit levels depend on the benefit level in
neighboring municipalities and own socioeconomic characteristics. (2) To
the extent that income inequality retards economic growth (Persson and
Tabellini 1994), FD might negatively affect growth by making
redistribution more difficult. Furthermore, concentration of public
goods, with supra-local spillovers, in a few geographical locations can
also inhibit per capita growth because regional inequalities in
infrastructure, education, healthcare, and other public services may
prevent full use of factors of production (Thiessen 2003). In this case,
more centralized public sectors might redistribute resources across
jurisdictions leading to a more efficient distribution.
Other economic arguments against FD include possible damage to
macroeconomic stability via fiscal policy coordination problems (Tanzi
1996); inter-jurisdictional "leakages" associated with local
expenditures (Oates 1972); and failure to exploit economies of scale and
scope (Prud'homme 1995). (3) More recently, Inman (2003) contends
that local government may expect to be bailed out from their fiscal
deficits by central governments, given the likelihood that failure to
rescue local governments would lower local welfare with some of the
political costs placed shifted to central government by voters
(Goodspeed 2002). In addition, FD may lead local governments to engage
in a "race to the bottom" on the taxation of mobile factors,
hence under-providing productive public expenditure (Brueckner 2004), or
increase corruption because officials at the local level are more
susceptible to the demands of local interest groups (Prud'homme
1995; Tanzi 1996). (4) Finally, to the extent that individuals do not
move freely between municipalities, at least in the short term, this
allows local governments to be relatively unresponsive to local
citizens' preferences.
In summary, there are clearly arguments for both positive and
negative effects of FD on fiscal efficiency and economic growth rates.
It is perhaps not surprising then that the empirical literature
discussed below has tended to find a variety of effects in different
contexts.
III. EMPIRICAL EVIDENCE ON FD AND ECONOMIC GROWTH
As a number of authors have noted, there is surprisingly little
research devoted to measuring the impact of FD on economic growth rates,
given that economic efficiency is the central argument used to support
FD (Bardhan 2002; Martinez-Vazquez and McNab 2003). Among existing
studies a mixed picture emerges of the effect of decentralization on
growth rates. Initial contributions tended to find that FD has a
negative or negligible effect on economic growth (Davoodi and Zou 1998;
Jin and Zou 2005; Woller and Philips 1998; Xie, Zou, and Davoodi 1999;
Zhang and Zou 1998). These authors interpret their results as an
indication that FD is already high, such that further decentralization
may be harmful for economic growth. However, many of these studies focus
on developing or transition economies, with China a specific focus of
attention. (5)
A number of factors may explain this negative effect. Firstly, as
Davoodi and Zou (1998) and Zhang and Zou (1998) argue, FD may be
particularly harmful for economic growth in the early stages of
development, where the administrative capability of local governments is
insufficient, local officials may not be responsive to preferences of
local residents, and local governments in those countries may be
constrained by central government. Secondly, fiscal policy-growth
effects may be more related to the functional composition of government
spending or type of tax rather than to fiscal decentralization per se.
If subnational governments spend more on items with low growth effects
such as social welfare, whereas national governments spend more on
growth enhancing items such as infrastructure, then we could expect to
observe a negative, endogenous relationship between FD and economic
growth.
More recent studies, especially those examining the U.S. or OECD
countries, find some evidence of a positive relationship between FD and
growth; see Akai and Sakata (2002), Thiessen (2003), Ebel and Yilmaz
(2004), Meloche, Vaillacourt, and Yilmaz (2004), Iimi (2005), Jin, Qian,
and Weingast (2005) and Thornton (2007). One source of difference in
results between the early, and recent, studies may be the FD measure
used. Recognizing that high subnational spending and revenue shares do
not necessarily reflect high local autonomy, and if autonomy is the key
growth-enhancing characteristic of FD, then early studies probably
overstated the degree of effective decentralization because some local
revenues/expenditures are typically controlled or mandated by central
governments. (6)
By contrast, recent studies have focused on a more restricted
measure of FD: local government spending net of conditional or
discretionary transfers (Ebel and Yilmaz 2004; Meloche, Vaillacourt, and
Yilmaz 2004) and local revenues over which subnational governments have
some degree of control over the tax rate, the tax base, or both (Akai
and Sakata 2002; Ebel and Yilmaz 2004; Meloche, Vaillacourt, and Yilmaz
2004; Thornton, 2007). Lin and Liu (2000) and Jin, Qian, and Weingast
(2005) use the marginal retention rate of locally collected revenue to
reflect the degree of FD arguing that this captures the fiscal
incentives for local government to promote local business development.
Using these narrower FD measures, a positive impact of FD on economic
growth has found more support. (7)
Recent literature has started to examine samples of OECD alone, and
thus is more related to our work. Thiessen (2003) finds evidence of a
growth-maximizing degree of FD. That is, growth is enhanced by
converging toward intermediate levels of decentralization--from either
high or low initial levels. Thornton (2007) argues that much of the
literature has not distinguished appropriately between administrative
and substantive FD. Adam, Delis, and Kammas (2008) find that public
sector efficiency is increasing with FD, whereas fiscal dependency of
local government on intergovernmental transfers affects efficiency
negatively. Baskaran and Feld (2009) find that fiscal decentralization
is generally unrelated to economic growth, and that, if anything,
subfederal control over shared taxes leads to more economic growth.
Using more flexible dynamic econometric methods we show below that,
for a variety of measures of local fiscal autonomy, an important
characteristic appears to be convergence toward similar levels of
revenue and spending decentralization. That is, our evidence suggests
raising revenue decentralization and/or lowering spending
decentralization would be growth-enhancing on average for OECD
countries. As far as we are aware, our empirical evidence is the first
to support Oates' (1972) hypothesis that FD efficiency benefits
become greater when there is a close match between revenue discretion
and spending assignments at subnational levels. Jin and Zou (2005) also
tested simultaneously for growth effects of expenditure and revenue
decentralization across Chinese provinces, but they reject Oates'
hypothesis. We obtain our results after controlling for endogeneity; we
find some effects running from growth to fiscal decentralization in line
with the arguments of Bahl and Linn (1992) and Martinez-Vazquez and
McNab (2003) that efficiency gains from, and demand for, FD emerge as
economies grow. Most previous empirical FD studies have not controlled
for endogeneity, at least in a systematic way, an exception being Iimi
(2005). (8) Using flexible dynamic panel methods, and the pooled mean
group (PMG) in particular, recognizes that efficiency gains may take
some time to materialize and occur at different rates in different
countries.
IV. DECENTRALIZATION MEASURES, DATA AND ECONOMETRIC METHODS
A. Decentralization Measures
The data used in our econometric analysis is based on OECD General
Government Accounts (various editions). We have extended this
time-series using annual IMF (2001), Government Finance Statistics (GFS)
data. (9) We follow Stegarescu (2005) and construct two measures of
expenditure decentralization and three measures of revenue
decentralization. (10) In all cases, these annualized decentralization
measures are calculated as shares of consolidated general government
spending or revenue. For expenditures we calculate:
(1) Direct [spending.sub.t]
Subnational [spending.sub.t] - Transfers from subnational to
central [government.sub.t] = / Consolidated general government
[spending.sub.t]
(2) Self- financed [spending.sub.t]
Subnational [spending.sub.t] - Grants from other
[governments.sub.t] = / Consolidated general government [spending.sub.t]
Equation (1), "Direct spending" in year t, subtracts
transfers paid to central government in that year, thus reporting
amounts spent directly at each local administrative level. (11) Equation
(2) treats subnational expenditure net of grants received from central
government as "self-financed spending," reflecting spending
from "own resources" (Stegarescu 2005). As a measure of
locally financed spending it may be regarded as a more appropriate
indicator of local autonomy.
On the revenue side, a measure of "own revenue"
decentralization is:
(3) Own [revenue.sub.t]
Subnational [revenue.sub.t] - Grants from other [governments.sub.t]
= / Consolidated general government [revenue.sub.t]
Equation (3) subtracts grants received from other levels of
government from total subnational revenues, to capture "own
resources." (12)
However, there are also locally collected taxes over which local
governments have little or no control (Sorens 2011). Arguably, these
taxes should also be subtracted to measure autonomous local resources
appropriately. Unfortunately, there is no official OECD data
distinguishing between locally collected taxes controlled by local
versus central governments for a broad sample of countries. (13)
However, following the methodology of OECD (1999, 2001) for Central and
Eastern European Countries, Stegarescu (2005) provides data for 21 OECD
countries from 1975 to 2000 on the locally collected taxes, decomposed
into the following categories:
A Tax bases or/and rates determined by
subnational governments
B Tax revenues shared between subnational and
central governments
of which:
B1 shared taxes: subnational level determines
revenue split
B2 shared taxes: subnational level has to consent
to revenue split
B3 shared taxes: central government unilaterally
determines revenue split
C Tax bases or/and rates determined by central
governments
Extending the analysis to consider decentralization for all sources
of public revenue requires the addition of nontax revenues. According to
Stegarescu (2005) these nontax revenues may include user charges,
operational surplus of public enterprises, and capital revenue. This
allows two additional revenue decentralization measures to be
calculated: autonomous own revenue (Equation (4) below) and the
autonomous plus shared own revenue (Equation (5) below).
(4) Autonomous own [revenue.sub.t]
Owntax revenue[(A).sub.t] + Nontax and capital [revenue.sub.t] = /
Consolidated general government [revenue.sub.t]
(5) Autonomous and Shared own revenuer
Own tax revenue[(A).sub.t] + Shared tax revenue[(B1 and B2).sub.t]
+ Nontax and capital [revenue.sub.t] = / Consolidated general government
[revenue.sub.t]
Equation (4) is the share of taxes for which subnational
governments determine the tax base/ rates (category A), plus local
nontax and capital revenue. The autonomous-plus-shared own revenue
(Equation (5)) is the share of taxes in Equation (4), plus shared taxes
where the revenue split is determined, or consented, by subnational
governments (categories B1 and B2). These two revenue decentralization
measures provide a narrower definition of local autonomy in public
revenues but are only available for a more limited sample of countries
and years. Thus, for Equations (1)-(3) above our sample is composed of
annual data for 23 OECD countries from the early 1970s to 2005. For
Equations (4) and (5) data are restricted to 18 countries from 1975 to
the late 1990s. We therefore use Equations (4) and (5) as robustness
checks on the other indicators.
B. Data
Table 1 shows the period averages for each FD Indicator by OECD
country. These cover state and local governments combined since only
nine countries have a federal system showing state spending and revenue
separately. Each indicator shows substantial variation across countries,
with Canada, Switzerland, and the United States revealing the greatest
degrees of FD. In those countries, subnational governments account for
approximately half of the consolidated public spending and revenue. By
contrast, Greece, Portugal, New Zealand, and Luxembourg have highly
centralized governments which control more than 85% of the public sector
size.
Differences across countries tend to be higher toward the beginning
of the period. For example, the standard deviation of logs of state and
local direct spending--the usual [sigma]-convergence
indicator--decreased from 0.77 in 1974 to 0.68 in 2003 (from 0.74 to
0.63 for self-financed spending). The dispersion in own revenue also
diminished from 0.31 to 0.27. Countries with high (low) initial levels
of decentralization generally reduced (increased) these, confirming the
convergence trend in the FD process identified by Thiessen (2003).
An important feature of these data is that, with the exception of
Mexico, state and local direct spending shares are higher than state and
local revenue shares. That is, subnational governments depend on central
government transfers to finance their spending. Self-financed
subnational spending is generally close to the subnational own revenues;
that is, subnational governments do not run large deficits after taking
into account transfers from central governments. (14)
Over 1974-2003 the data reveal quite different patterns for revenue
and spending decentralization: Figure 1 shows annual mean values across
the OECD countries. Direct and self-financed spending decentralization
in the OECD decreased on average during the 1970s and early 1980s,
trending upward only from the early-to-mid 1990s. By contrast, own
revenue decentralization has remained fairly constant throughout the
period. "Autonomous and shared" revenues (Equation (5)) reveal
more variation without any clear trend over time, but this pattern may
partly reflect missing values for some of the countries in the series.
[FIGURE 1 OMITTED]
C. Econometric Methods
Our econometric analysis follows the approach of Davoodi and Zou
(1998) and Xie, Zou, and Davoodi (1999) who consider a production
function with two inputs: private capital and public spending. Public
spending is carried out by three levels of government: federal, state,
and local. Assuming a Cobb-Douglas production function with constant
returns to scale, these authors show that the long-run growth rate of
per capita output is a function of the tax rate and the federal, state,
and local shares in aggregate government spending. Optimal government
spending shares of each administrative level match the growth elasticity
of this administration relative to the sum of the elasticities for all
administrations. If the local spending share is below (above) this
optimum, further decentralization enhances (retards) economic growth.
The models of Davoodi and Zou (1998) and Xie, Zou, and Davoodi
(1999) recognize that consolidated government spending must be financed
by tax revenue, such that tests of the growth effects of FD need to
recognize the government budget constraint. (15) In addition to
production function-related variables, we therefore also include the
general government revenue/gross domestic product (GDP) ratio as a
measure of the overall fiscal burden. Surprisingly, most recent
empirical studies have failed to control for this fiscal burden, giving
rise to potential bias in their estimates of the FD effects on growth.
(16)
Our estimating equation uses the PMG model of Pesaran, Shin, and
Smith (1999), which allows for heterogeneous short-run effects across
countries but homogeneous long-run effects. The PMG regression takes the
following "error correcting" form:
(6) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where i indicates the country, t is time, g is the rate of growth
of GDP, F is a matrix of fiscal and control variables, [phi], [beta],
and y are parameters to be estimated and [[epsilon].sub.it] is a
classical error term. In particular, the [beta] parameter vector
measures the homogeneous long-run (level) effect of the fiscal and
control variables, [[gamma].sub.0] and [[gamma].sub.1] measure the
(heterogeneous) short-run growth responses (to lagged growth and
fiscal/control variables, respectively), and [phi] captures the
adjustment toward long-run equilibrium. (17) Our interest here is
primarily with the long-run parameters, in particular the long-run
effect of FD on economic growth.
Previous studies have typically sought to capture the long-run
effect of FD on growth by using multi-year averages (Davoodi and Zou
1998; Woller and Philips 1998) or lagged (including initial year) values
of FD in their estimations (Akai and Sakata 2002; Iimi 2005; Lin and Liu
2000; Stansel 2005; Thiessen 2003; Thornton 2007). Using dynamic panel
methods, and the PMG in particular, recognizes that efficiency gains
need some time to materialize in a highly flexible way. The
heterogeneous short-run transitory effects in the PMG also allow for
differences across countries in their short-run responses of growth to
changes in each independent variable. By focusing on a relatively
homogeneous set of high-income OECD countries we hope to overcome Akai
and Sakata's (2002) concern over international differences in
history, institutions, culture, etc., but allowing for short-run
heterogeneity facilitates a more accurate estimate of long-run effects.
A disadvantage of the PMG estimator over simpler methods, such as
fixed effects models which impose homogeneity of all marginal responses,
is that unless the available time series is long a degrees of freedom
problem is soon reached. For the dataset available here this requires
choices over restrictions to lag lengths and the set of included
right-hand-side (RHS) variables. For this reason, we generally restrict
the RHS variables to include three control variables (the investment
rate, employment growth, and the ratio of general government revenue to
GDP). Data on GDP growth, the private investment/GDP ratio, employment
growth, and GDP per capita were obtained from OECD sources. This allows
us to use up to two lags and up to four FD variables (subnational
spending and revenue decentralization; disaggregated by local and state
government where possible). Restricting our regressions to include a
maximum of two lags, nevertheless, allows the effect of shocks to
persist over many periods via the inclusion of the lagged dependent
variable.
As a robustness check we include openness and inflation as controls
(at the cost of reduced lag length) because these variables have often
been employed previously, (18) Openness is expected to affect growth
positively, via the resource allocation benefits of external competition
(Feder 1983). Inflation can have either positive or negative effects on
growth though the latter is more usually observed (Zhang and Zou 1998).
If, as argued by Treisman (2000), decentralization slows inflation in
developed countries and inflation reduces growth, estimated effects of
FD on economic growth that do not account for inflation are biased
upward. Along these lines, Martfnez-Vazquez and McNab (2006) find that
decentralization indirectly enhances economic growth through its
positive impact on price stability in developed countries, offsetting
the negative direct effect of FD on growth.
V. ECONOMETRIC RESULTS
A. PMG Results
Table 2 shows regression results using both the direct spending
decentralization measure (Equation (1): in columns 1-3) and
self-financed spending decentralization (Equation (2): in columns 4-5).
We report only the long-run ([beta]) parameters in order to save space
(full results are available from the authors on request). All
regressions include the overall revenue/GDP ratio and two production
function "controls": the investment ratio and employment
growth. (19) Investment and employment confirm the expected positive and
significant relationships with GDP growth. Regressions including
openness and inflation are discussed below; they have little impact on
the other parameters shown in Table 2. The table also shows the
importance of including the overall revenue burden which can be seen in
all regressions to impact negatively and significantly on growth. That
is, increases in overall fiscal size retard growth for a given
level/type of decentralization.
Regression results in columns 3 and 5 (using direct and
self-financed spending, respectively) represent our preferred
specifications--including both spending and revenue decentralization.
This allows us to test Oates' (1972) FD hypothesis that efficiency
is enhanced by closer "matching" of revenue and spending
decentralization. These reveal a negative and significant effect of
state and local direct spending shares, or self-financed spending
shares, on economic growth. Conversely, there is a positive, significant
effect of larger state and local revenue shares on economic growth.
Together with the evidence in Table 1 that state and local direct
spending shares are higher than revenue shares in our sample countries,
this implies that a reduction of this gap, achieved either by reducing
subnational spending shares or by increasing revenue shares, would
increase economic growth. (20) As these results represent marginal
effects associated with changes from current settings they cannot
confirm whether raising revenue shares to current spending share levels,
or vice versa, would necessarily increase growth. However, they do
confirm that reductions in state/local spending shares and financing a
greater fraction of this spending by state/local taxes would be
growth-enhancing, consistent with Oates' "matching"
hypothesis.
Including either revenue shares or spending shares (columns 1, 2,
and 4) reveals that false conclusions may be drawn when one FD variable
is omitted. Including only state and local spending continues to
generate a negative parameter but which is not always significantly
different from zero. Including only state and local revenues appears
essentially to generate a zero (but negatively signed) growth effect. It
could be argued that our "matching" evidence is due to
collinearity between revenue and spending decentralization--tending
toward equal and opposite signed parameters. Indeed, subnational direct
spending and own revenue reveal a 0.89 between-country correlation and a
0.63 within-country correlation. In order to analyze whether these high
correlations are driving our results we implement the regression
collinearity diagnostic procedures proposed by Belsley (1991), based on
the interrelationships among the independent variables. As a rule of
thumb, Belsley (1991) suggests that if the condition number is 30 or
higher, then there may be collinearity problems. (21) At 19.5 the higher
condition number for our set of variables is well below this value.
Using the variation inflation factor (VIF) (22) leads to the same
conclusion: the highest VIF is 4.69 (subnational government spending),
well below the suggested rule of thumb of 10, from which collinearity
problems should be further investigated (Hair et al. 1995).
Nevertheless, as a further check, we orthogonalized subnational spending
and revenue by creating a set of orthogonal variables, using a modified
Gram-Schmidt procedure (Golub and Van Loan 1996), such that the effects
of the preceding variable have been removed from each variable. Thus, in
column 6 we transform subnational government direct spending into a new
variable in which the effect of the constant is removed and transform
subnational government revenue into a new variable in which both the
effects of the constant and subnational government spending are removed.
(23) The interpretation of the orthogonalized variable is the
independent variable in question minus the linear influences of the
variables upon which it is orthogonalized. Results show that we find
again a negative growth impact of spending decentralization and a
positive impact for revenue decentralization. We reach the same
conclusion when orthogonalizing self-financed subnational spending and
subnational revenue in column 7.
Columns 8 and 9 disaggregate state and local direct spending and
revenues into their two components. This reduces the sample to the nine
federal countries having separate state and local spending. (24) With
one exception (state own revenues becomes zero) we continue to find
negative spending and positive revenue share effects associated with the
state and local components. The largest parameters are associated with
the local administration level, because the difference between spending
and revenue is higher for local government than for the state level.
This is consistent with there being greater efficiency gains from
convergence toward equality between subnational spending and revenue
when the initial mismatch is higher.
These results again indicate that a convergence toward more equal
expenditure and revenue decentralization, at both the local and state
level, would enhance economic growth, reinforcing the importance of
testing for the growth effects of spending and revenue decentralization
simultaneously. Surprisingly, few previous empirical studies have tested
directly for both shares simultaneously; Jin and Zou (2005) is an
exception. Note that our evidence does not necessarily imply that it is
optimal for the degree of decentralization to be equal for revenues and
expenditures--a "zero gap." It does imply, given observed
values in OECD countries on average, that a marginal shift in the
direction of closer alignment of these subnational expenditures and
revenues would enhance growth.
B. Instrumental Variables
Our estimates in Table 2 of the impact of FD on economic growth may
be biased if, as Bahl and Linn (1992) and Martinez-Vazquez and McNab
(2003) argue, the efficiency gains from FD emerge as economies grow and
mature or decentralization is generally demanded at relatively high
levels of per capita income. In this subsection, we account for
potential endogeneity bias affecting the FD variables and investment,
using their third and fourth lagged values as instruments.
Instruments must satisfy two requirements: they must be (a)
correlated with the included endogenous variables; and (b) orthogonal to
the error process. The first condition can be tested using the
F-statistic and the partial [R.sup.2] between the excluded instruments
and the endogenous regressors of the first stage. However, these
measures will not reveal the weakness of a particular instrument if
remaining instruments are highly correlated with the endogenous
variables (Staiger and Stock 1997). The Shea partial [R.sup.2] (Shea
1997) overcomes this by taking into account the cross-correlations among
the instruments. Baum, Schaffer, and Stillman (2003) suggest, as a rule
of thumb, that if the partial [R.sup.2] is large, whereas the Shea
partial [R.sup.2] measure is small, we may conclude that the instruments
lack sufficient relevance to explain all the endogenous regressors. (25)
Table 3 (lower section) shows both the Shea partial [R.sup.2] and
the partial [R.sup.2] (in brackets) for the first stage regression.
These confirm that the Shea partial [R.sup.2]s are relatively high and
differences between the two measures are small--with the possible
exception of the disaggregation between state and local
decentralization. Table 3 also reports the Anderson under-identification
test of the hypothesis that excluded instruments are uncorrelated with
the endogenous regressors.
This test is rejected in all estimations, indicating that the
excluded instruments are relevant in explaining our endogenous
variables.
Since high Shea partial [R.sup.2] and rejection under the Anderson
test does not guarantee that weak instrument problems are absent, we
also report the Stock and Yogo (2005) test for the presence of weak
instruments. Results reported in Table 3 reject the null hypothesis of
weak instruments. (26) In sum, our set of excluded instruments is highly
correlated with the included endogenous variables. Furthermore, Sargan
tests reported in Table 3 do not reject the hypothesis that the third
and fourth lagged values are valid instruments, that is, orthogonal to
the error process. (27)
Comparing results in Tables 2 and 3 (columns 1 and 2) confirms our
earlier FD findings. Subnational direct spending decreases growth,
whereas subnational own revenue enhances GDP growth, with parameter
estimates in Table 3 larger than their Table 2 equivalents, confirming
our expectations that taking endogeneity into account leads to higher
estimated growth impacts. Thus, FD continues to be associated with
faster economic growth when subnational government spending more closely
matches what it collects. Using direct spending, the same conclusion is
reached when disaggregating spending and revenues into local and state
government components for the subsample of the nine
"federalist" countries. Higher state direct spending
significantly decreases growth, whereas higher state revenues
significantly increase growth. Results are less clear for local spending
and revenues in Table 3, though when openness and inflation are included
(see below) a similar pattern to state spending/revenues is obtained.
Column 4 in Table 3--for disaggregated state and local government
and self-financed spending--appears to suffer from weak instrument
problems: both the identification test and the Shea partial [R.sup.2]s
for state self-financed spending and own revenues are low. When openness
and inflation is included, this weak instrument problem wanes, but the
Sargan test indicates that the instruments are not exogenous.
Unfortunately, we cannot investigate this further using the fifth lag,
because the time-series is insufficient and we have only nine countries
in the subsample.
VI. ROBUSTNESS CHECKS
A. Adding Control Variables
We noted earlier that several previous studies included inflation
and openness variables among their control variables (though most recent
papers fail to control for total government revenues). In regressions
equivalent to those in Tables 2 and 3 but including these additional
growth determinants (not reported) we find that the openness variable
regularly takes the "wrong" (negative) sign which is
frequently significant. (28) In addition, this variable appears to
interact counter-intuitively with the investment ratio in several
regressions. We do not regard these regressions as satisfactory.
Nevertheless, of particular interest here is that inclusion of these
additional regressors does not alter the parameter estimates or
conclusions regarding the growth effects of spending or revenue
decentralization. In all cases, these remain negative and positive,
respectively, typically significantly different from zero.
B. Using "Autonomous Revenue" Definitions
The availability of the Stegarescu (2005) database allows us to
examine Equations (4) and (5) discussed above--based on definitions of
"autonomous" and central/local "shared" revenues.
Five countries are dropped from our previous sample: Greece and Mexico
(no data), and Italy, New Zealand, and Portugal (time-series
insufficient to include in PMG estimations). (29) This reduces the
sample to 18 countries and 384 observations. Disaggregation into state
and local governments is also not available. Nevertheless, the
Stegarescu (2005) database is potentially helpful to check the
robustness of our earlier results to narrower definitions of subnational
revenues, capturing aspects of subnational "control" (Equation
(4)) and "shared revenues" (Equation (5)).
Table 4 reports results equivalent to those reported in Table 3 for
our larger sample. Using either Equations (4) or (5) again suggests that
both greater direct and self-financed spending retards growth, whereas
greater autonomous revenues (either alone or with shared revenues)
enhance growth. General government revenue is again robustly negatively
associated with GDP growth. It would appear then that changing the FD
measures (direct vs. self-financed spending, own revenues vs. autonomous
own revenues), changing the data source (OECD vs. IMF), and changing the
sample (23 OECD vs. 18 OECD vs. 9 "federalist" countries) does
not alter the conclusion: fiscal decentralization enhances economic
growth where this involves moving toward a closer match between
subnational spending and subnational revenues.
C. Government Spending/Revenue Composition by Levels of
Administration
Our evidence of negative expenditure decentralization effects on
growth could be due to the fact that local governments spend less on
growth-enhancing functions than central governments, rather than being
more inefficient. Analogously, evidence of positive revenue
decentralization effects on growth could also simply reflect the fact
that local governments collect less from growth-distorting taxes than
central governments. Hence our data may simply reflect the evidence of
Kneller, Bleaney, and Gemmell (1999) and Bleaney, Gemmell, and Kneller
(2001) that "distortionary" taxes retard growth while
"productive" expenditures enhance it, rather than the
administration level at which these fiscal aggregates are spent or
collected.
To investigate this, we follow the methodology of Kneller, Bleaney,
and Gemmell (1999) and Bleaney, Gemmell, and Kneller (2001) to produce
an aggregate "productive spending" category--the sum of
general public services, defense, public order and safety, environment
protection, housing and community amenities, health and education. We
also aggregate government revenue sources into "distortionary"
and "nondistortionary" taxes and "other revenues"
(see Kneller, Bleaney, and Gemmell 1999 for discussion), where the
former is composed of current taxes on income, wealth and capital and
social contributions and so-called "nondistortionary" taxes
are mainly indirect taxes such as value-added tax (VAT). These
aggregations are only possible for European countries (from 1995), based
on Eurostat data for the functional composition of government spending
and the composition of government revenues by levels of administration.
Our calculations show the share of state and local government in
total productive spending in the EU-15 countries over 1995-2004 to be
35%. There is some variation across types of spending; the local share
is particularly high for education, public order, and safety. The share
of productive expenditure is significantly above the equivalent share of
total nonproductive spending (28%). Using an "economic"
classification leads to a similar conclusion: local and state
governments in the EU-15 accounted for a significantly higher share of
government capital formation (68%) than for intermediate consumption
(3%), compensation of employees (6%), or transfers (4%). This would seem
to suggest that the negative effect of expenditure decentralization on
economic growth we find is unlikely to occur because of a higher
proportion of nonproductive spending within local and state budgets.
A similar pattern emerges for revenue decentralization. For
distortionary taxes, we find that local and state governments in the
EU-15 collected 14% of all distortionary taxes during 1995-2004, and 28%
of "other revenues" (Kneller, Bleaney, and Gemmell show the
latter also tend to be growth-retarding). By contrast, state and local
governments only collected 11% of all nondistortionary taxes
(growth-neutral according to Kneller, Bleaney, and Gemmell 1999). The
positive revenue decentralization growth effect found in Tables 2 and 4
is unlikely therefore to be explained by a higher proportion of
nondistortionary taxes among state and local revenues.
VII. CONCLUSIONS
The empirical literature on the efficiency gains associated with
fiscal decentralization has generally focused on the growth impact of
spending or revenue decentralization separately. However, following Jin
and Zou's (2005) evidence for China, we test simultaneously for the
growth effects of both spending and revenue decentralization across OECD
countries. From a theoretical perspective, Oates (1972) has argued that
FD efficiency benefits become greater when there is a close match
between revenue discretion and spending assignments at subnational
levels.
We find that economic growth in OECD countries has been adversely
affected by decentralization of expenditures but encouraged by revenue
decentralization. As OECD countries are, in general, substantially more
spending than revenue decentralized, this implies empirical support for
the prediction that there would be FD efficiency gains on average by
moving toward a closer match between spending and revenue
decentralization in OECD countries. Our econometric results relate to
marginal changes and, hence, do not indicate whether a shift to
"perfect matching" would maximize growth. Nor can we be sure
that raising the share of decentralized revenues to the level of current
expenditure shares (or reducing decentralized expenditure shares to
current revenue shares) would necessarily be growth-enhancing. However,
they do support the conclusion that reducing expenditure
decentralization, and simultaneously reducing the fraction which is
financed centrally would be growth-enhancing.
This evidence is robust to various definitions of decentralized
spending and "own revenues," and the use of PMG methods has
allowed for the possibility that dynamic responses of growth to changes
in spending and revenue shares may take several years. We have also
allowed these short-run responses to vary across countries rather than
impose short-run homogeneity as in the fixed effects models used by
previous investigators. Our results emphasize the importance of testing
simultaneously for expenditure and revenue decentralization to avoid
conflating the distinct, and oppositely signed, impacts of the two
aspects to FD.
Finally, testing for possible endogeneity bias of our fiscal
decentralization (and some control) variables, suggests that lagged
values can provide valid instruments and these confirm that our
FD-growth estimates do not appear to be due to endogenous responses. For
OECD countries, therefore, it would appear that, ceteris paribus, their
growth rates have been hindered by a common tendency to finance a large
fraction of their subnational expenditures using centrally raised tax
revenues together with inter-government transfers, in preference to
financing a higher fraction of subnational expenditures with revenues at
the subnational level. There may be a number of good or bad reasons why
this is the case, which we have not addressed in this article. However,
the growth consequences of those choices seem clear.
ABBREVIATIONS
FD: Fiscal Decentralization
GDP: Gross Domestic Product
GFS: Government Finance Statistics
OECD: Organization for Economic Co-operation and Development
PMG: Pooled Mean Group
RHS: Right-Hand Side
VAT: Value-Added Tax
VIF: Variation Inflation Factor
REFERENCES
Adam, A.. M. D. Delis, and P. Kammas. "Fiscal Decentralization
and Public Sector Efficiency: Evidence from OECD Countries." CESifo
Working Paper 2364, CESifo, 2008.
Akai, N., and M. Sakata. "Fiscal Decentralization Contributes
to Economic Growth: Evidence from StateLevel Cross-section Data for the
United States." Journal of Urban Economics, 52(1), 2002, 93-108.
Bahl, R. W., and J. F. Linn. Urban Public Finance in Developing
Countries. Oxford: Oxford University Press, 1992.
Bardhan, P. "Decentralization of Governance and
Development." Journal of Economic Perspectives, 16(4), 2002,
185-205.
Baskaran, T., and L. P. Feld. "Fiscal Decentralization and
Economic Growth in OECD Countries: Is There a Relationship," CESifo
Working Paper 2721, CESifo, 2009.
Baum, C., M. Schaffer, and S. Stillman. "Instrumental
Variables and GMM: Estimation and Testing." Statu Journal, 3, 2003,
1-31.
Belsley, D. A. Conditioning Diagnostics. Collinearity and Weak Data
in Regression. New York: John Wiley & Sons, 1991.
Bleaney, M. F., N. Gemmell, and R. Kneller. "Testing the
Endogenous Growth Model: Public Expenditure, Taxation and Growth over
the Long-run." Canadian Journal of Economics, 34(1), 2001. 36-57.
Brueckner. J. K. "Fiscal Decentralization with Distortionary
Taxation: Tiebout vs. Tax Competition." International Tax and
Public Finance, 11(2), 2004, 133-53.
--. "Fiscal Federalism and Economic Growth." Journal of
Public Economics, 90(10-11), 2006, 2107-20.
Davoodi, D., and H. Zou. "Fiscal Decentralization and Economic
Growth: A Cross-Country Study." Journal of Urban Economics, 43(2),
1998, 244-57.
Ebel, R. D., and S. Yilmaz. On the Measurement and Impact of Fiscal
Decentralization. Washington DC: Urban Institute, 2004.
Eurostat. Annual Government Finance Statistics. Brussels: European
Commission, various years.
Feder, G. "On Exports and Economic Growth." Journal of
Development Economics, 12, 1983, 59-73.
Fiva, J. H., and J. Rattso. "Welfare Competition in Norway:
Norms and Expenditures." European Journal of Political Economy,
22(1), 2006, 202-22.
Giavazzi, F., and M. Pagano. "Non-Keynesian Effects of Fiscal
Policy Changes: International Evidence and the Swedish Experience."
Swedish Economic Policy Review, 31, 1996, 67-103.
Golub, G. H., and C. F. Van Loan. Matrix Computations. Baltimore:
Johns Hopkins University Press, 1996.
Goodspeed, T. J. "Bailouts in a Federation."
International Tax and Public Finance, 9, 2002, 409-21.
Gramlich, E. "A Policymaker's Guide to Fiscal
Decentralization." National Tax Journal, 46(2), 1993, 229-35.
Hair, J.R., R.E. Anderson, R.L. Tatham, and W. C. Black.
Multivariate Dam Analysis with Readings. Englewood Cliffs, N J: Prentice
Hall, 1995.
Iimi, A. "Decentralization and Economic Growth Revisited: An
Empirical Note." Journal of Urban Economics, 57(3), 2005, 449-61.
Inman, R. P. "Transfers and Bailouts: Enforcing Local Fiscal
Discipline with Lessons from U.S. Federalism," in Fiscal
Decentralization and the Challenge of Hard Budget Constraints, edited by
J. Rodden, G. S. Eskeland, and J. I. Litvack. Cambridge, MA: MIT Press,
2003, 35-83.
International Monetary Fund. Government Finance Statistics Manual
2001, Washington DC: IMF, 2001.
Jin, H., Y. Qian, and B. R. Weingast. "Regional
Decentralization and Fiscal Incentives: Federalism, Chinese Style."
Journal of Public Economics, 89(9-10), 2005, 1719-42.
Jin, J., and H. Zou. "Fiscal Decentralization, Revenue and
Expenditure Assignments, and Growth in China." Journal of Asian
Economics, 16(6), 2005, 1047-64.
Kappeler, A., and T. Valila. "Fiscal Federalism and the
Composition of Public Investment in Europe." European Journal of
Political Economy, 24(3). 2008. 562-70.
Kneller, R., M. Bleaney, and N. Gemmell. "Fiscal Policy and
Growth: Evidence from OECD Countries." Journal of Public Economics,
74, 1999, 171-90.
Lin. J. Y., and Z. Liu. "Fiscal Decentralization and Economic
Growth in China." Economic Development and Cultural Change, 49(1),
2000, 1-21.
Martinez-Vazquez, J., and R. M. McNab. "Fiscal
Decentralization and Economic Growth." World Development, 31(9),
2003, 1597-616.
--. "Fiscal Decentralization, Macrostability, and
Growth." Revista de Economia Publica, 179(4), 2006, 25-49.
Meloche, J., F. Vaillacourt, and S., Yilmaz. "Decentralization
or Fiscal Autonomy? What Does Really Matter? Effects on Growth and
Public Sector Size in European Transition Countries". World Bank
Policy Research Working Paper 3254. World Bank, 2004.
Oates, W. E. Fiscal Federalism. New York: Harcourt Brace
Jovanovich, 1972.
--. "An Essay on Fiscal Federalism.'" Journal of
Economic Literature, 37(2), 1999, 1120-49.
--. "Toward a Second-Generation Theory of Fiscal
Federalism." International Tax and Public Finance, 12, 2005,
349-73.
OECD. National Accounts of OECD Countries--Vol. IV: General
Government Accounts, Paris: OECD (various editions).
--. Taxing Powers of State and Local Government, Paris: OECD, 1999.
--. Fiscal Design Across Levels of Government., Paris: OECD, 2001
Persson, T., and G. Tabellini. "Is Inequality Harmful for
Growth?" American Economic Review, 84(3), 1994, 600-21.
Pesaran, M. H., Y. Shin, and R. P. Smith. "Pooled Mean Group
Estimation of Dynamic Heterogeneous Panels." Journal of the
American Statistical Association, 94, 1999, 621-34.
Prud'homme, R. "On the Dangers of Decentralization."
World Bank Research Observer, 10(2), 1995, 201-20.
Schaltegger, C. A., and L. P. Feld. "Are Fiscal Adjustments
Less Successful in Decentralized Govermnents?" European Journal of
Political Economy, 25(1), 2009. 115-23.
Shah, A. "The Reform of Intergovernmental Fiscal Relations in
Developing and Emerging Market Economies." Policy Paper 23,
Washington DC: World Bank, 1994.
--. "Fiscal Decentralization and Macroeconomic
Management." International Tax and Public Finance, 13(4), 2006,
437-62.
Shea, J. "Instrumental Relevance in Multivariate Linear
Models: A Simple Measure." Review of Economics and Statistics,
79(2), 1997, 348-52.
Sorens, J. "The Institutions of Fiscal Federalism."
Publius: The Journal of Federalism. 41(2), 2011, 207-31.
Staiger, D., and J. H. Stock. "Instrumental Variables
Regression with Weak Instruments." Econometrica, 65(3), 1997,
557-86.
Stansel, D. "Local Decentralization and Economic Growth: A
Cross-sectional Examination of US Metropolitan Areas." Journal of
Urban Economics, 57(1), 2005, 55-72.
Stegarescu, D. "Public Sector Decentralization: Measurement
Concepts and Recent international Trends." Fiscal Studies, 26(3),
2005, 301-33.
--. "The Effects of Economic and Political Integration on
Fiscal Decentralization: Evidence from OECD Countries." Canadian
Journal of Economics, 42(2), 2009, 694-718.
Stock, J. H., and M. Yogo. "Testing for Weak Instruments in
Linear IV Regressions," in Identification and Inference for
Econometric Models: Essays in Honor of Thomas J. Rothenberg, edited by
J. H. Stock and D. W. K. Andrews. Cambridge: Cambridge University Press,
2005, 80-108.
Tanzi, V. "Fiscal Federalism and Decentralization: A Review of
Some Efficiency and Macroeconomic Aspects," in Annual World Bank
Conference on Development Economics, edited by M. Bruno and B.
Pleskovic. Washington DC: World Bank, 1996, 295-316.
Thiessen, U. "Fiscal Decentralization and Economic Growth in
High-Income OECD Countries." Fiscal Studies, 24(3), 2003, 237-74.
Thornton, J. "Fiscal Decentralization and Economic Growth
Reconsidered." Journal of Urban Economics. 61(1), 2007, 64-70.
Tiebout, C. "A Pure Theory of Local Expenditures."
Journal of Political Economy, 64(5), 1956, 416-24.
Treisman, D. "Decentralization and Inflation, Commitment,
Collective Action, or Continuity?" American Political Science
Review, 99(4), 2000, 837-57.
Woller, G. K., and K. Phillips. "Fiscal Decentralization and
LDC Growth: An Empirical Investigation." Journal of Development
Studies, 34(4), 1998, 138-48.
World Bank. "Decentralization: Rethinking Government" in
World Bank: World Development Report 1999/2000. Washington DC: World
Bank, 1999.
Xie, D., H. Zou, and H. Davoodi. "Fiscal Decentralization and
Economic Growth in the United States." Journal of Urban Economies,
45(2), 1999, 228-39.
Zhang, T., and H. Zou. "Fiscal Decentralization, Public
Spending and Economic Growth in China." Journal of Public
Economics, 67(2), 1998, 221-40.
(1.) It is not necessary, however, for individuals to have
different preferences for local public goods or to be relatively mobile
to obtain efficiency gains from FD. Thiessen (2003) argues that as long
as subnational governments better reflect the priorities of taxpayers,
this is sufficient for fiscal decentralization to offer efficiency
advantages.
(2.) Nevertheless, these authors do not find that this strategic
interaction among local governments results in under-provision or even a
race to the bottom in welfare spending, because of the centralized
grants financing of the local governments in the Norwegian system.
(3.) On the other hand, Gramlich (1993) claims that if economic
shocks are asymmetric, then decentralized systems make it easier to
achieve macroeconomic stability. Shah (2006) also suggests that central
bank independence is more probably attained under decentralized systems,
because the pressure of a unique central government diminishes, leading
to the presence of multiple governments with diverse and conflicting
interests. In this line, Treisman (2000) shows that, by creating
additional veto players, federal structure may lock in existing patterns
of monetary policy, leading to slower growth of inflation among (mostly
developed) countries that started with low inflation. Finally,
Martinez-Vazquez and McNab (2003) maintain that a well-designed fiscal
decentralization system (preventing local governments to borrow without
controls) avoids fiscal systems damaging macroeconomic stability.
(4.) Martinez-Vazquez and McNab (2003) dispute this argument
claiming that local officials are more visible to their constituents and
thus corrupt behavior is more visible than at the central level of
government.
(5.) On China, see also Lin and Liu (2000) and Jin, Qian, and
Weingast (2005) who find some evidence of positive growth effects of FD,
and the critique of early studies by Akai and Sakata (2002). A detailed
summary of the empirical evidence can be found in the working paper
version of this article at http://www.ief.es/documentos/recursos/
publicaciones/papeles_trabajo/2009_06.pdf.
(6.) Furthermore, Lin and Liu (2000) criticize the measure employed
in Zhang and Zou (1998)--the ratio of provincial spending to total
central spending--because a large province would appear to have a high
degree of fiscal decentralization merely by being more populous. More
generally, the legal distinction between locally and centrally
controlled public expenditures or taxes may not be a reliable guide to
the extent of decentralization of decision-making in practice. Thus,
central governments may formally devolve spending responsibility to
local levels but circumscribe the use of those expenditures with a
variety of legal or administrative conditions. In the United States, for
example, Federal "maintenance of effort" requirements now
preclude U.S. states from making major adjustments to Medicaid programs,
including those parts that are state funded.
(7.) An exception to these fiscal decentralization measures is
Stansel (2005) who focuses on the horizontal dispersion of power among
lower tiers of government using the number of county, municipal, and
township administrations per 100,000 residents in 314 U.S. metropolitan
areas. Using this measure, Stansel (2005) finds a positive and
significant effect of FD on the growth of both population and real per
capita income.
(8.) Zhang and Zou (1998), Xie et al. (1999), Lin and Liu (2000),
Thiessen (2003), and Jin et al. (2005) acknowledge potential endogeneity
bias but do not control for it--due to small sample sizes and the
difficulty of finding good instruments. Lin and Liu (2000) show that,
for their case, the Hausman test of the potential endogeneity of the FD
variable fails to reject the hypothesis that the marginal retention rate
is exogenous. Jim Qian, and Weingast (2005) regress marginal retention
rates on lagged growth rates and find a negative rather than positive
coefficient, rejecting a positive upward bias in their estimated FD
growth effect for China.
(9.) The OECD General Government Accounts uses accrual accounting,
providing a better picture of commitments undertaken by governments than
traditional cash accounting. However, the information available from
this source starts in 1990 or 1995 for most of the countries. We have
extended this time-series using annual IMF, Government Finance
Statistics (GFS), data. This source covers a longer period, from 1972 to
1998 or 1999, but is based on the cash criterion. Using the rate of
variation of the GFS to extend back the OECD data is a sensible
procedure because the coefficient of correlation between the overlapped
period of the 1990s is always 0.94 or higher, except for Australia
(0.87) and New Zealand (0.81). Indeed, cash and accrual accounting
coincide over the medium term because commitments undertaken end up
materializing in payments. For Mexico, we use IMF cash data for the
whole period, because there is no information about this country in the
OECD and Greece has no overlapping years for cash and accrual data.
These two countries, plus New Zealand, are excluded from the robustness
checks. The dataset is available at:
http://www.victoria.ac.nz/sacl/staff/norman-gemmell.aspx and
http://www.fcjs.urjc.es/departamentos/areas/profesores/
ficha.asp?id=trqssvwuur.
(10.) Sorens (2011) provides an interesting discussion of the
Stegarescu (2005) measures of decentralization and constructs
alternative measures.
(11.) These transfers refer to the category, "'Grants to
other general government units" (Government Finance Statistics
Manual 2001). They can be current or capital grants, depending on
purpose, and they include the tax levied by one level of government but
transferred to other levels of government. Transfers from subnational
governments to central governments are only significant for Spain and,
especially, for Greece. For the rest of the sample it accounts for a
small share of subnational government spending (average: 1.9%).
(12.) Ebel and Yilmaz (2004) contend that unconditional transfers,
and transfers given under objective criteria, could be included under
revenue decentralization. However, we subtract all transfers to leave
only those revenues generated by subnational governments and which are
not discretionarily fixed by central government (Stegarescu 2005). The
other indicator used in the literature, the marginal retention rate, is
not directly observable; calculation would require simulations for each
type of revenue; see Thiessen (2003).
(13.) It is available for some Central and Eastern European
Countries for 1997-2000: see OECD (1999, 2001).
(14.) Surprisingly, the Stegarescu (2005) database shows higher
subnational revenue shares than our OECD-based database despite the fact
that the Stegarescu measure defines local revenues more narrowly. This
could be due to different countries/time periods and/or differences in
the main data source (IMF Government Finance Statistics vs. OECD
National Accounts). There are also numerous missing values for some of
the 21 countries in the Stegarescu database.
(15.) See Bleaney et al. (2001) and Kneller et al. (1999) for
similar arguments relating to tests of fiscal policy on growth more
generally.
(16.) For example, if FD leads to a lower public sector size,
because of the increased competition among levels of administration, and
there is a negative relationship between the public sector size and
growth, then there will be a positive bias in the estimation of the
growth effects of FD.
(17.) Using a Mean Group (MG), rather than PMG, model allows
long-run, as well as short-run, heterogeneity with the PMG restricted
tested using a Hausman test. However, running an MG model requires many
more degrees of freedom and is therefore not feasible here. However,
Hausman tests on our PMG regressions in Table 2 support assumption of
long-run parameter homogeneity.
(18.) The school enrolment ratio has been also included as a
control variable in some studies on the effects of FD and economic
growth. However, this variable is not reliable on an annual basis for
OECD countries.
(19.) Like most growth regression studies, data for investment
ratios is more readily available and generally more reliable than
capital growth data. We also prefer employment to labor force growth
because the former can account for the cyclical dimension to output
growth better.
(20.) This result contrasts with the empirical evidence for China
by Jin and Zou (2005), who also introduce simultaneous spending and
revenue decentralization. For China, they find a positive effect for
revenue decentralization when this measure was higher than spending
decentralization and a negative effect when it was lower.
(21.) The condition number is the condition index with the largest
value; it equals the square root of the largest eigenvalue divided by
the smallest eigenvalue. A condition number of 1 means that independent
variables are orthogonal. Large values of condition number indicate rank
deficiency of the independent variables matrix and that estimates are
sensitive to small changes in the data. This number has been obtained
applying the Coldiag2 command in Stata.
(22.) VIF is an index which measures bow much the variance of a
coefficient is inflated by the existence of multicollinearity. Large VIF
values indicate that severe effects are present.
(23.) We orthogonalize these variables using the Stata command
Orthog.
(24.) As the PMG calculates means of individual country
estimations, it is not possible to introduce variables taking zero
values for a country in every year.
(25.) The distribution of Shea's partial [R.sup.2] statistic
has not been derived.
(26.) Stock and Yogo (2005) propose a test based on the F-form of
the Cragg-Donald statistic. This test has good power, especially when
the number of instruments is large as in our case. For the case of three
endogenous variables, a desired maximal bias of 10%, and up to 14
excluded instruments (as in Table 4) the critical value is 10.25 (Stock
and Yogo 2005, Table 1). Similarly, the critical value for two
endogenous variables, desired maximal bias of 10%, and 14 excluded
instruments, is 36.36 (Table 2).
(27.) The Sargan test rejects the use of second and third lags.
(28.) These estimations can be found in the working paper version
of this paper mentioned in footnote 5.
(29.) In addition Greece, Mexico, and New Zealand are the three
countries for which extending the data back has been more difficult (see
footnote 9). Table 4 shows that excluding these three countries does not
change the econometric results.
NORMAN GEMMELL, RICHARD KNELLER and ISMAEL SANZ *
* We thank three anonymous referees and the Editor of this journal
for helpful comments on an earlier draft.
Gemmell: University of Nottingham, UK; School of Accounting and
Commercial Law, Victoria University of Wellington, Wellington. New
Zealand. Phone +64 4 463 5843. Fax +64 4 463 5076, E-mail
norman.gemmell@vuw.ac.nz
Kneller: School of Economics, University of Nottingham, Nottingham,
UK. Phone +44 115 95 14734, Fax +44 115 95 11900, E-mail
richard.kneller@nottingham.ac.uk Sanz: Economia Aplicada I. Universidad
Rey Juan Carlos, Madrid, Spain. Phone +34 91 4887800, Fax +34 91
7750342, E-mail ismael.sanz@urjc.es
doi: 10.1111/j.1465-7295.2012.00508.x
TABLE 1
State and Local Shares in Aggregate Government Spending and Revenue
Across OECD Countries: 1970-2005
Own Calculations Based on OECD National
Accounts and IMF GFS (1972-2005)
State and State and Local State and
Local Direct Self-financed Local Own
Country Spending Spending Revenue
Australia 44.6 22.2 27.4
Austria 30.9 23.8 27.4
Belgium 22.5 10.2 10.4
Canada 60.5 51.0 52.2
Denmark 56.3 31.7 32.5
Finland 37.8 27.2 26.6
France 16.0 11.6 12.1
Germany (a) 41.6 35.6 35.1
Greece 4.8 4.8 3.6
Iceland 22.3 19.7 21.5
Ireland 32.5 15.9 14.4
Italy 24.9 13.5 11.3
Luxembourg 14.4 9.5 8.6
Mexico 18.5 18.2 20.6
Netherlands 34.8 11.7 11.6
New Zealand 11.6 11.4 10.8
Norway 35.0 29.0 24.9
Portugal 10.7 7.5 8.1
Spain 25.7 13.0 15.6
Sweden 44.2 35.3 33.6
Switzerland 57.6 50.7 48.0
United Kingdom 28.4 12.9 12.9
United States 46.8 46.8 41.6
Unweighted mean 31.4 22.3 22.2
Stegarescu (2005): Calculations
Based on IMF, GFS (1975-2000)
State and Local State and Local
Autonomous Autonomous and
Country Revenue Shared Revenue
Australia 27.4 27.4
Austria 14.3 35.7
Belgium 14.4 23.6
Canada 55.3 55.3
Denmark 31.1 31.1
Finland 32.0 32.0
France 18.3 18.3
Germany (a) 24.5 53.3
Greece -- --
Iceland 22.2 22.2
Ireland 10.5 10.5
Italy 7.7 7.7
Luxembourg 11.3 11.3
Mexico -- --
Netherlands 10.7 10.7
New Zealand 9.3 9.3
Norway 26.4 26.4
Portugal 5.4 5.4
Spain 14.0 17.2
Sweden 41.4 41.4
Switzerland 62.8 65.6
United Kingdom 15.7 15.7
United States 45.0 45.0
Unweighted mean 23.8 26.9
(a) Data for Germany before 1991 refer to West Germany.
Source: OECD: National Accounts of OECD Countries-Vol. IV: General
Government Accounts. IMF: Government Finance Statistics Yearbook,
and Stegarescu (2005).
TABLE 2
Decentralization and Economic Growth:
Pooled Mean Group Regressions (1972-2005)
Regression [1] [2] [3]
Decentralized
Spending Measure Direct Direct
PMG PMG PMG
Method (2 Lags) (2 Lags) (2 Lags)
General revenue -0.052 -0.036 -0.053
ratio (-4.41) (-3.15) (-4.50)
State and local -0.050 -0.074
spending (-3.80) (-4.92)
State and local -0.014 0.056
own revenue (-0.76) (2.83)
Investment ratio 0.053 0.051 0.051
(2.28) (2.25) (2.28)
Employment growth 0.637 0.637 0.585
(13.44) (15.03) (13.28)
Countries/ 23/726 23/726 23/726
observations
Regression [4] [5] [6]
Decentralized Self- Self- Direct
Spending Measure financed financed Orthog
PMG PMG PMG
Method (2 Lags) (2 Lags) (2 Lags)
General revenue -0.042 -0.042 -0.053
ratio (-3.43) (-3.56) (4.50
State and local -0.019 -0.052 -0.497
spending (-1.30) (-2.24) (2.19)
State and local 0.060 0.353
own revenue (1.98) (2.84)
Investment ratio 0.066 0.080 0.051
(2.66) (3.47) (2.28)
Employment growth 0.577 0.535 0.585
(11.83) (11.32) (13.28)
Countries/ 23/726 23/726 23/726
observations
[7]
Regression Self- [8] [9]
Decentralized financed Self
Spending Measure Orthog Direct financed
PMG PMG PMG
Method (2 Lags) (2 Lags) (2 Lags)
General revenue -0.050 State -0.067 -0.083
ratio (4.24) spending (-1.51) (-2.04)
State and local -0.550 State -0.141 -0.104
spending (2.40) own revenue (-3.92) (-1.52)
State and local 0.341 Local 0.067 -0.016
own revenue (2.79) spending (1.02) (-0.22)
Local -0.190 -0.162
own revenue (-3.07) (-1.67)
0.417 0.379
(2.96) (3.12)
Investment ratio 0.528 0.411 0.183
(11.61) (5.04) (2.60)
Employment growth 0.061 0.941 0.781
(2.65) (11.68) (9.60)
Countries/ 23/726 9/283 9/283
observations
Note: t-statistics in parentheses below parameters. The dependent
variable in all regressions is the annual rate of GDP growth.
TABLE 3
Decentralization and Economic Growth: Instrumental Variable
Regressions (1972-2005) Instruments: 3rd and 4th lagged values
Regression [1] [2]
Decentralized Self-
Spending Measure Direct financed
PMG/IIV PMG/lV
Method (2 Lags) (2 Lags)
General revenue ratio -0.004 -0.017
(-0.28) (-1.17)
State and local spending -0.083 -0.082 State
(-7.04) (-3.10) spending
State and local own revenue 0.119 0.115 State
(5.81) (3.53) own revenue
Local
spending
Local
own revenue
Investment ratio -0.091 -0.064
(-3.38) (-2.27)
Employment growth 0.525 0.654
(14.44) (15.57)
Countries/Observations 23/645 23/645
Correlated with the included endogenous variables: Shea partial
[R.sup.2] (overall [R.sup.2] in brackets)
Shea partial [R.sup.2]: 0.60 0.55
revenue ratio [0.73] [0.72]
Shea partial [R.sup.2]: state 0.46 0.27 State
and local expenditure [0.70] [0.59] Spending
Shea partial [R.sup.2]: state 0.31 0.24 State
and local revenue [0.61] [0.59] Own revenue
Shea partial [R.sup.2]: local Local
expenditure Spending
Shea partial [R.sup.2]: local Local
revenue Own revenue
Shea partial [R.sup.2]: 0.40 0.37
investment [0.50] [0.50]
Anderson test 216.01 174.29
p value p value
0 0.00
Weak identification test 30.23 23.62
Orthogonal to the error
process
Sargan test 1.251 5.864
p value p value
0.87 0.21
Regression [3] [4]
Decentralized Self
Spending Measure Direct financed
PMG/1V PMG/IV
Method (2 Lags) (2 Lags)
General revenue ratio 0.084 -0.091
(2.56) (-3.51)
State and local spending -0.112 0.068
(-7.87) (1.09)
State and local own revenue 0.137 -0.142
(5.25) (-2.21)
0.114 -0.415
(1.72) (-5.54)
0.021 0.566
(0.24) (6.29)
Investment ratio -0.215 -0.012
(-4.16) (-0.19)
Employment growth 0.528 0.594
(10.73) (10.99)
Countries/Observations 9/254 9/254
Correlated with the included endogenous variables: Shea partial
[R.sup.2] (overall [R.sup.2] in brackets)
Shea partial [R.sup.2]: 0.56 0.43
revenue ratio [0.73] [0.74]
Shea partial [R.sup.2]: state 0.32 0.02
and local expenditure [0.82] [0.551
Shea partial [R.sup.2]: state 0.24 0.03
and local revenue [0.66] [0.62]
Shea partial [R.sup.2]: local 0.39 0.28
expenditure [0.50] [0.56]
Shea partial [R.sup.2]: local 0.37 0.26
revenue [0.57] [0.57]
Shea partial [R.sup.2]: 0.40 0.32
investment [0.50] [0.511
Anderson test 169.84 16.47
p value p value
0.00 0.02
Weak identification test 15.20 1.31
Orthogonal to the error
process
Sargan test 7.829 7.216
p value p value
0.25 0.30
Notes: t-statistics in parentheses below parameters. The dependent
variable in all regressions is the annual rate of GDP growth.
TABLE 4
Decentralization and Economic Growth: IV Regressions Using Stegarescu
Variables and Sample (1975-2000)
Regression [1] [2]
PMG/IV PMG/IV
Method (I lag) (I lag)
General revenue ratio -0.099 -0.099
(-3.88) (-2.94)
State and local direct spending -0.077 -0.075
(-5.07) (-3.73)
State and local self-financed spending
Autonomous own revenues 0.101
(5.87)
Autonomous and shared own revenues 0.037
(2.92)
Investment -0.014 -0.058
(-0.58) (-2.04)
Employment growth 0.689 0.717
(28.26) (25.74)
Openness -0.021 -0.017
(-4.58) (-3.01)
Inflation -0.114 -0.108
(-5.35) (-4.61)
Sample N=18 N=18
Observation = Observation =
359 359
Correlated with the included endogenous variables: Shea partial
[R.sup.2] (overall [R.sup.2] in brackets)
Shea partial [R.sup.2]: revenue ratio 0.94 0.94
[0.951 [0.95]
Shea partial [R.sup.2]: direct exp 0.92 0.86
[0.971 [0.98]
Shea partial [R.sup.2]: self-financed
exp
Shea partial [R.sup.2]: own tax 0.93
[0.98]
Shea partial [R.sup.2]: own and 0.86
shared tax [0.98]
Shea partial [R.sup.2]: investment 0.71 0.71
[0.72] [0.72]
Shea partial [R.sup.2]: openness 0.96 0.97
[0.99] [0.99]
Anderson test 397.80 397.49
p value p value
0 0.00
Weak identification test 75.15 75.04
Orthogonal error process
Sargan test 7.715 6.803
p value p value
0.17 0.24
Regression [3] [4]
PMG/1V PMG/IV
Method (I lag) (I lag)
General revenue ratio -0.064 -0.054
(-2.32) (-1.84)
State and local direct spending
State and local self-financed spending -0.078 -0.039
(-3.34) (-1.54)
Autonomous own revenues 0.085
(6.14)
Autonomous and shared own revenues 0.036
(3.76)
Investment 0.019 0.000
(0.86) (-0.02)
Employment growth 0.695 0.740
(28.47) (28.48)
Openness -0.034 -0.040
(-7.30) (-7.34)
Inflation -0.154 -0.177
(-7.05) (-7.48)
Sample N=18 N=18
Observation = Observation =
359 359
Correlated with the included endogenous variables: Shea partial
[R.sup.2] (overall [R.sup.2] in brackets)
Shea partial [R.sup.2]: revenue ratio 0.94 0.94
[0.95] [0.95]
Shea partial [R.sup.2]: direct exp
Shea partial [R.sup.2]: self-financed 0.93 0.72
exp
Shea partial [R.sup.2]: own tax [00.88] [098]
[0.95]
Shea partial [R.sup.2]: own and 0.68
shared tax [0.95]
Shea partial [R.sup.2]: investment 0.69 0.66
[0.72] [0.73]
Shea partial [R.sup.2]: openness 0.96 0.97
[0.99] [0.99]
Anderson test 370.28 304.92
p value p value
0.00 0.00
Weak identification test 66.46 48.60
Orthogonal error process
Sargan test 6.704 5.748
p value p value
24 33
Instruments: 3rd and 4th lagged values.
Spending FD measures: Direct spending and Self-financed spending.
Revenue FD measures: Autonomous own revenues and Autonomous and shared
own revenues.
Notes: t-statistics in parentheses below parameters. The dependent
variable in all regressions is the annual rate of GDP growth.
