Romero-Avila and Strauch (2008)

8/10/2019 Romero-Avila and Strauch (2008)

1/20

Public finances and long-term growth in Europe: Evidence from apanel data analysis

Diego Romero-vila a, Rolf Strauch b ,

aDepartment of Economics, Pablo de Olavide University, Carretera Utrera, Km. 1, Sevilla, 41013, Spainb European Central Bank, Kaiserstr. 29, 60311 Frankfurt am Main, Germany

Received 29 November 2005; received in revised form 23 January 2007; accepted 11 June 2007Available online 13 August 2007

Abstract

This paper addresses the question whether public finance reform can affect trend growth in the EU-15. Focusing on time seriespatterns, we investigate whether there have been persistent trends in economic growth and fiscal variables over the last 40 years. Inaddition, we estimate a distributed lag model, which 1) indicates that government size measured either with total expenditure orrevenue shares, government consumption and direct taxation negatively affect growth rates of GDP per capita, while publicinvestment has a positive impact, and 2) provides robust evidence that distortionary taxation affects growth in the medium-termthrough its impact on the accumulation of private capital. 2007 Published by Elsevier B.V.

JEL classification: C22; C23; H11; O11Keywords:Panel cointegration; Public finances; Economic growth

1. Introduction

The European Council set forth the Lisbon Process in order to raise growth rates of output in EU countries. TheCouncil considersan average economic growth rate of around 3% [as] a realistic prospect for the coming years.1 Forthis purpose economic policy should be geared to foster a knowledge-driven economic expansion through the spread ofnew technologies and higher human capital, more perfect goods and financial markets in Europe, a more employment-

friendly active labour market policy and a modernisation of the welfare state as well as an investment-friendly climatebrought about by regulatory changes. Many of the measures envisaged by the heads of states affect not only theregulatory setting but also public finances. In a follow-up to the process initiated in Lisbon, the Commission andECOFIN Council underscored that the qualityof public finances plays a crucial role for growth and employment.More specifically, they outlined the necessity to lower the tax burden and particularly the tax wedge for low-skilled

Available online at www.sciencedirect.com

European Journal of Political Economy 24 (2008) 172191www.elsevier.com/locate/ejpe

Corresponding author. Tel.: +49 69 1344 7376; fax: +49 69 1344 6320.E-mail address:[email protected](R. Strauch).

1 See Conclusions of the Presidency, page 2.

0176-2680/$ - see front matter 2007 Published by Elsevier B.V.doi:10.1016/j.ejpoleco.2007.06.008
mailto:[email protected]://dx.doi.org/10.1016/j.ejpoleco.2007.06.008http://dx.doi.org/10.1016/j.ejpoleco.2007.06.008mailto:[email protected]


2/20

workers, make benefit systems more supportive to employment, shift resources towards productive expenditures inhealth, education and physical infrastructure, and to ensure the sustainability of public finances.

The conclusions of the European Council on the European economic performance in the future leave open whichgrowth model actually best reflects the intentions of the heads of state. If they implicitly assume an exogenous growthframework, then we could expect output to speed up in the short and medium run but then level off again. Conversely,

the structural changes which they envisage to make Europe a more integrated, competitive and productive economymay also imply that trend growth rises permanently from currently 22.5% to 3% per year.

In this paper we will investigate which growth pattern would be the outcome of policy reforms in the future courseof the Lisbon process. Analysing past experiences in European countries should indicate whether public finances havethe potential to raise growth rates permanently, i.e. affecting trend growth, or whether one could at best expect atransitory improvement.

Numerous Barro-typeregression studies have claimed to find evidence for endogenous growth if diverse policyand institutional variables included in the regressions affect long-term performance. However, these studies mainlyexploit the cross sectional variation of very large samples and are not very informative if we want to focus on theEuropean context. Using the standard set-up of these studies, relying on long-term averages of output growth, wewould quickly exhaust the degrees of freedom for European countries. Moreover, this approach would not be

appropriate since the European sample is rather homogeneous in several of these explanatory characteristics.

2

As an alternative, a small literature has emerged with a main focus on the time series implications of the two strandsof theory. If the policy variable follows a specific time series pattern, economic growth should exhibit the samebehaviour under endogenous growth theory. Conversely, the time series properties of the policy-variable do notnecessarily have to coincide with output growth according to exogenous growth models. Fiscal variables are a goodtesting ground for these hypotheses, since distortionary taxation and productive expenditures are assumed to have apermanent effect on economic growth according to endogenous growth theory, whereas they should have only leveleffects from a neoclassical perspective. We will use these predictions as a basis to interpret the observable pattern ofeconomic growth and public finance developments, which has not systematically been done for Europe. This is the gapin the empirical literature that our study wants to close. In first place, we will analyse the time series properties of outputgrowth and fiscal policy variables in order to determine their degree of persistence. In second place, we will estimatethe impact of fiscal policies on trend growth by employing the distributed lag approach.

The paper is organised as follows. Section 2 briefly describes the theoretical background and the shortcomings ofthe empirical evidence existing in this area. Section 3 describes the data used in the empirical exercise. In Section 4 weanalyse the time series properties of real per-capita output growth and public finance variables. This analysis will showthat there are persistent developments in per-capita GDP growth and fiscal variables, which are broadly in line withsome theoretical predictions on long-term growth. In Section 5 we conduct distributed lag estimations as a moresystematic check of the long-term impact of public finances and Section 6 concludes.

2. Theory and existing empirical evidence

Since the mid-1980s the theoretical growth literature has above all tried to endogenise the growth rate of output inthe long-run. Earlier growth models, formulated by Solow (1956)andCass (1965)among others, conceived trend

growth largely as a function of factors exogenous to public policy such as technological progress and populationgrowth. Endogenous growth theory pioneered by the work ofRomer (1986, 1990),Lucas (1988),Barro (1990)andRebelo (1991)among others, points out mechanisms by which policy variables cannot only affect the level of output,but also steady-state growth rates. Barro (1990)constitutes one of the first attempts at endogenising the relationshipbetween growth and fiscal policies. He distinguishes four categories of public finances: productive vs. non-productiveexpenditures and distortionary vs. non-distortionary taxation. We present a simple sketch of the Barro model in order toshow that both productive public expenditure and distortionary taxation can affect long-run output growth. We assumethat the population of consumers is normalised to one. Consumers both consume and produce final output according tothe following production function:

yAk1ggg 1

2 For a meta-analysis of studies on the effect of fiscal policies on long-run growth seeNijkamp and Poot (2004).

173D. Romero-vila, R. Strauch / European Journal of Political Economy 24 (2008) 172191


3/20

wherekstands for privately accumulated physical capital andg is productive government expenditure that directlyenters the production process. It is assumed that the government budget constraint is balanced in every period and isgiven by:

gGsdyT 2whereGrepresents other government expenditure that does not directly enter the production function as an input, Trepresents lump-sum taxation and is the proportional tax rate on output which distorts the investment decision.Consumers maximise their intertemporal utility function that is given by

Rl

0 eqtc

1r 11r dtsubject to the standard budget

constraint. represents the time preference rate and is the elasticity of intertemporal substitution of consumption.The growth rate of consumption and output in steady state takes the form:

c:

cy

:

y1

r 1s1gA 11g g

y

g1gq

" # 3

Eq. (3) shows that productive government expenditure as a share of output positively affects long-run growth while

distortionary taxation has a negative impact on growth. Neither unproductive expenditure nor lump-sum taxation affectoutput growth in steady state. From this model, we see that fiscal variables from both sides of the budget constraintmatter for growth, and the failure to include both productive government expenditures and distortionary taxation ingrowth regressions would lead to mis-specified models.

Jones (1995)constitutes the first attempt at exploiting time series properties to test exogenous vs. endogenousgrowth theory. He starts with the simple argument that according to endogenous growth theory, permanent shifts incertain policy variables should have a permanent effect on the growth rate of output. Hence, if growth rates in the USand other OECD countries exhibit no large persistent changes, the underlying policy variables should also either notshow large persistent changes or the persistent movements in these variables must be off-setting. Using theconventional augmentedDickey and Fuller (1979, ADF) test, he finds considerable evidence for a stochastic trend inthe data-generating process for total investment, producer durables investment and R&D expenditure in a large share ofcountries. The estimation of distributed lag models showed that shifts in investment and R&D expenditures did not

exert any permanent effect on growth, thereby rejecting the predictions of endogenous growth theory.Karras (1999)largely follows the approach ofJones (1995), focusing on the effect of taxation on per capita GDP

growth. He analyses a panel of 11 OECD countries, finding that the real GDP growth rate is generally stationary, whilethe null of a unit root cannot be rejected for the total and direct tax rates in most of the countries. He concludes thatadjustments of tax rates cannot be associated with permanent changes of real GDP growth, unless permanent changesin taxes are cancelled out by permanent changes in other policy variables. But he does not investigate this finalpossibility by including the expenditure side of the budget in his analysis.Evans (1997)attains similar results byanalysing the impact of government consumption on growth for a sample of 92 countries.

Kocherlakota and Yi (1997)test whether taxes or public investment have any permanent effect on output growth,based on time series for up to 100 years for the US and 160 years for the UK. Thus, they incorporate both sides of thebudget into their analysis and find that predictions of exogenous growth theory are usually rejected when taxes and

public investment are included in the econometric model. However, they do not formally test for the co-movement ofpolicy variables. This also holds true forKneller et al. (1999)andBleaney et al. (2001)who estimate the long-run effectof public finances on growth for the OECD countries, finding a significant growth impact of productive expendituresand distortionary taxation.

The main contributions of the present paper are as follows. Our analysis constitutes the first attempt at investigatingthe possibility that two countervailing forces may impart an opposite (and possibly offsetting) effect on growthfollowing the logic of Jones' analysis. This is done by putting together two main theories. On the one hand, we take theendogenous growth model ofBarro (1990)which predicts that productive government expenditure and distortionarytaxation exert a growth effect of opposite sign. On the other, we consider the theory of intertemporal sustainability offiscal policies, which predicts that government expenditure and revenues are cointegrated in order to ensure that theintertemporal government budget constraint in present value terms holds. Therefore, even if we find that individualfiscal variables exhibit non-stationarity while output growth appears mean-stationary, this diverging pattern could stillbe reconciled with endogenous growth predictions if another policy variable with an offsetting persistent effect on

174 D. Romero-vila, R. Strauch / European Journal of Political Economy 24 (2008) 172191


4/20

growth existed. Thus, unlike the studies ofEvans (1997) and Karras (1999) and based on the theory of fiscalsustainability, we look at the opposite side of the government budget constraint since any additional expenditure needsto be financed, thus leading to a higher excess burden of taxation.

3. Data

As the previous short review of the literature has shown, the existing evidence clearly supports endogenous growthpredictions of a long-term growth impact when both sides of the budget are taken into account, but evidence is still

incomplete. Except for the study ofBleaney et al. (2001)the sample of countries and the budgetary categories used havebeen very selective.Bleaney et al. (2001)incorporate most European economies in their sample and include all budgetitems, but they focus on central government data only. The drawback of this approach is, of course, that ideally one wouldlook at general government figures. First, overall government activity and not only central government activity shouldcount from an economic point of view. Second, general government provides a more homogeneous data set than centralgovernment, which may vary strongly according to the organisation of national and subnational authorities.

Therefore, we use data for general government outlays and revenues in all EU member states from 1960 to 2001(Commission AMECO data set, Autumn 2002). All time series are computed in logs and fiscal variables are measuredas shares of GDP. Budgetary aggregates are classified according to an economic criterion rather than functionally. Thisis fairly unproblematic with respect to taxation, because the classification of direct taxation on property and income, onthe one hand, and indirect taxation on imports and production on the other, largely reflects the theoretical distortionary/non-distortionary classification. For public expenditures the link is less immediate. Evidently, public capital formationcould be counted as productive expenditures. Government consumption comprises wage payments going to teachers

Fig. 1. Growth rates in European countries, 19612001.



5/20

and professors, i.e. they are investments in human capital, as well as salaries and purchases for the social securitysystem, whichBleaney et al. (2001)assume to be unproductive. In addition, it also comprises expenditures on healthand education, which are clearly of productive use. In other words, empirical evidence on the impact of these spending

categories jointly evaluates the validity of the theoretical predictions and assesses the productive vs. non-productivecontent of the expenditure flow under consideration. Data to compute the level and growth rate of real GDP per capitaas well as the private investment share of GDP are obtained from the OECD Economic Outlook. 3

4. The time series properties of growth and public finances

To assess the potential impact of fiscal policies on growth we take a look at the time series properties of the data. Theconjecture that permanent shifts in policy variables should be associated with a permanent shift in the growth pattern, ifendogenous growth theory holds, is compatible with different time series patterns. Therefore, we first search fordeterministic long-run movements and then for possible persistent stochastic processes.

4.1. Deterministic trends

Fig. 1 presents the growth rates of real GDP per capita in EU member states from 1961 to 2001. From the charts there isnot a clear time series pattern which holds for all countries. Growth rates in several countries, such as France, Spain,Germany, Greece, Italy and Sweden, have slowed down over the last few decades. In Ireland and Luxembourg medium-term growth has however picked up over this time period, while it has remained fairly stable in Austria and Belgium.

More formal evidence for these patterns is presented inTable 1where we show different estimates for deterministictrends existing in EU countries. The first row shows a trend estimate imposing a common mean and trend coefficientfor all countries. The second row shows the within estimate of the trend coefficient (i.e. allowing the intercept to vary)and the third row a mean-group estimate of the trend coefficient which is computed as the average of individualestimates.4 The estimated coefficient for a deterministic trend in per-capita GDP growth carries a negative sign and ishighly statistically significant. Turning to public finances,Fig. 2shows that there has been a clear increase in public

spending up to the early 1980s in all countries. This trend flattens thereafter or is even reversed. In Belgium, Ireland,Luxembourg and the Netherlands, the reversal sets in the 1980s, while it is of a more recent nature in most othercountries. Overall public revenues show a similar but often less pronounced increase in the 1960s and 1970s. This trendthen also flattens in most cases, but is not reversed.

Since aggregate spending and revenues are inaccurate measures of productive expenditures and distortionarytaxation,Table 1presents the trend estimates for different spending and revenues categories which might have a growthimpact according to Barro's model. The second column confirms the long run upward trend in government spending,which appears mainly driven by transfers and to a somewhat lesser extent by government consumption. Interestingly,public investment shows the opposite development, as indicated by the negative and statistically significant coefficient

3 See Table A1 for the notation of the variables employed in the analysis and data sources. Descriptive statistics and the correlation matrix are

provided in Tables A2 and A3, respectively. These three tables are provided in Appendix A.4 Pesaran et al. (1996)showed the asymptotic consistency of the mean-group estimator.

Table 1Deterministic time series patterns of growth and selected fiscal variables

DLYPC LTEXP LTR LG LPI LTREV LTDIST

Deterministic trend (common coefficient and intercept) 0.0005

(0.00013)0.015

(0.0015)0.018

(0.003)0.010

(0.002)0.014

(0.003)0.014

(0.001)0.017

(0.003)

Deterministic trend (fixed effects, common coefficient) 0.0005

(0.0001) 0.015

(0.001) 0.015

(0.002) 0.010

(0.0009) 0.013

(0.002) 0.013

(0.0008) 0.015

(0.001)Deterministic trend (mean group estimate) 0.0005

(0.0002)0.015

(0.0019)0.015

(0.003)0.011

(0.001)0.014

(0.005)0.014

(0.002)0.015

(0.002)

Note: The deterministic trend pattern is estimated usingyit= +Ti+ eitfor different assumptions about the heterogeneity ofand, whereyis thevariable under consideration for countryiat timetandTrepresents the time trend. The table reports the estimates ofand the NeweyWest standarderrors allowing for serial correlation in parenthesis. Three asterisks indicate statistical significance at the 1% level.



6/20

for the whole panel. In parallel, distortionary taxationwhich is computed as the sum of total direct taxation and socialsecurity contributions increased steadily over the last decades. The long-term decline of public investment and risingdistortionary taxation are both compatible with the lower trend growth apparent in our estimates.

4.2. Stochastic trends

Given the observable patterns inFigs. 2 and 3, our policy variables may not only exhibit deterministic but alsopersistent stochastic trends. To test this hypothesis we carry out panel unit root tests which overcome the low powerassociated with individual unit root tests. We employ the Im et al. (2003)(henceforth, IPS) and theBreitung (2000)

tests. Our panel specification will be of the form:

Dyitaidithtgiyit1Xpij1

bijDyitjeit 4

wherepiis the required country-specific degree of lag augmentation to make the residuals white noise,5 andNandT

are the number of cross-section units and time observations, respectively.ianditrepresent the country-specific fixedeffects and deterministic trends respectively, and tstands for the time dummies used to account for cross-sectionalcorrelation which could result from common shocks affecting all panel members in a given period. The null hypothesis

5 The degree of augmentation of individual ADFt-statistics was computed following the general-to-specific procedure of removing insignificant

lag-differenced terms until the last term is significant at conventional significance levels. For the panel as a whole, we use a degree of lag-truncationof 2 and 4, since 4 was in general the longest degree of augmentation required to whiten the residuals of individual ADF specifications.

Fig. 2. Public expenditures in European countries, 19612001.



7/20

H0of the IPS test implies thati= 0, for alli, i.e. all series have a unit root, which is tested against the alternative H1thatib0 fori =1,2,,N1andi= 0, fori =N1+ 1,N1+ 2,,N. Assuming that theNcross-section units are independentlydistributed, thet-statistic can be computed as an average of the individual ADFt-statistics such that:

tNTp;q

XNi1

tiTpi;qi

N 5

wheretiT(pi,i) is thet-statistic for testingi=0 in each individual ADF specification. Assuming that the second-ordermoments oftiT(pi,i) exist, the t

NTp

;qstatistic is corrected for small sample size as follows:

P

Zt

ffiffiffiffiN

p tNTp;q 1N

PNi1

EtiTpi; 0=gi0

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1

N

PNi1

VartiTpi; 0=gi0s Yd N0; 1 6

whereE[tiT(pi,0)/i=0] and Var[tiT(pi,0)/i=0] are the adjustment factors obtained via stochastic simulation. Thestandardised statistic weakly converges to a one-sided standard normal distribution under the null and diverges underthe alternative. Therefore, we need to compare the value of

P

Ztto the critical values from a lower-tailed standard normaldistribution.

The main strength of the IPS test compared to theLevin and Lin (1992)test is that the autoregressive coefficient isallowed to differ across countriesand only a fraction of units is required to be stationary under the alternative. However, the

Fig. 3. Public revenues in European countries, 19612001.



8/20

IPS test suffers from a substantial decrease in power when heterogeneous trends are included in the specification as a resultof the small-sample bias correction applied to thet-statistics (Baltagi and Kao, 2000a,b; Breitung, 2000).6 To address thisissue,Breitung (2000)proposes a panel unit root test which employs unbiasedt-statistics. By allowing for heterogeneous

deterministic trends and short-run dynamics across countries without the need of bias adjustment, the Breitung test hasmore power to reject a false null and is not sensitive to the degree of augmentation of the ADF specifications.7

Table 2presents the results for real per-capita growth, current revenues and aggregate expenditures along with theireconomic subcategories (i.e. direct taxation, social security contributions, indirect taxation, government consumption,transfers and public investment).8 The results of the tests show that per-capita GDP follows an I(1) pattern, while realGDP per-capita growth is stationary for the sample of the EU-15. The results for fiscal variables are also clear. Asregards the revenue side of the budget, the IPS and Breitung tests indicate a unit root in all specifications for totalrevenues and distortionary taxation (LTDIST). This result seems to be mainly driven by the direct tax component whichclearly contains a unit root, since the results for social security contributions are rather mixed. While the IPS test rejectsthe null of a unit root in social security contributions at the 1% level when trends are included and at 10% in thespecifications without trends, the Breitung test does not reject it. With regard to the expenditure side of the budget, total

expenditure and its sub-categories appear to be driven by a stochastic trend. The unit root statistics for the variables infirst-differences indicate that no fiscal variable is integrated of second order.Taken as a whole, output growth is found to be stationary while fiscal policy variables are in general non-stationary.

Given that public investment is generally considered as productive expenditure and direct taxation as distortionarytaxes, this result could be considered as a challenge to endogenous growth theory. Since persistent changes in fiscalcategories do not appear to be accompanied by persistent changes in GDP per capita growth rates, there is no supportfor endogenous growth predictions according to the logic expressed byJones (1995)andKarras (1999).

6 We present the IPS results of the specifications with and without heterogeneous deterministic trends. We include time dummies so as to controlfor common shocks affecting all EU Member States in a given period.7 Aswiththe IPStest, the Breitung testconverges to a one-sided standard normaldistribution under the null, soweneed to compare thevalue of the statistic to

the critical values from a lower-tailed standard normal distribution. Large negative values would reject the null of joint non-stationarity.8 We include all economic categories for the sake of completeness, although only some results will be discussed in this section.

Table 2Panel unit root tests for growth and fiscal variables

Variables IPS test 2 lags IPS test 4 lags Breitung test

No trend Trend No trend Trend Trend

LYPC 4.30 3.14 3.87 2.32 2.33

LTREV 1.97 0.73 0.06 1.01 0.40LTDIR 1.75 2.04 0.43 2.47 0.62LSSC 1.49 2.60 1.60 2.77 0.69LTDIST 0.48 0.31 1.94 0.16 1.17LTIND 2.23 0.43 2.67 1.44 1.57

LTEXP 1.26 1.38 1.51 1.00 3.03LPI 1.45 2.61 0.80 2.85 0.54LTR 0.68 0.06 1.16 0.12 4.38LG 1.02 0.38 1.60 2.24 1.37DLYPC 5.28 5.54 2.27 2.19 11.27

DLTREV 9.50 8.26 5.64 3.96 12.79

DLTDIR 8.93 8.41 3.63 3.50 13.71

DLSSC 8.77 6.94 5.90 3.06 13.13

DLTDIST 9.70 8.69 5.96 5.02 12.93

DLTIND 8.24 6.91 5.52 5.15 12.69

DLTEXP 6.96 5.40 3.18 1.96 10.63

DLPI 6.80 5.28 2.12 0.15 16.60

DLTR 6.96 5.46 4.57 2.84 10.97

DLG 8.33 6.72 4.67 3.64 13.11

Note: Time dummies were includedin allADFspecifications. Thevalue of theIPS andBreitungtestsmust becomparedto thecriticalvaluesfrom a lower-tailedstandard normal distribution. , and imply rejection of the null of a unit root at the 10%, 5% and 1% level of significance respectively.



9/20

4.3. Cointegration of expenditures and revenues

The diverging time series pattern of per-capita GDP growth and the fiscal policy variables could still be reconciled ifanother policy variable with an offsetting persistent effect on growth existed. In other words, two conditions have to bemet: first, the variable has to co-move with the policy instrument under consideration and, second, it should exhibit a

persistent growth effect according to endogenous growth theory. The obvious candidate here is to look at the oppositeside of the budget, since any expenditure increase has to be financed, and this may lead to a higher excess burden oftaxation.

It can be shown that the intertemporal budget constraint does indeed imply a cointegrating relationship betweenrevenues and expenditures (seeAfonso, 2005; Santos Bravo and Silvestre, 2002; Trehan and Walsh, 1988; Bohn,1991). However, whether this relationship exists for our sample and whether it could explain the observable pattern ofeconomic growth is still an empirical question. First, growth theory focuses on productive expenditures anddistortionary taxation. Thus, it is empirically unclear whether the cointegrating relationship holds for relevant spendingand revenue items. For example, higher public investment could be financed through non-distortionary consumptiontaxes. Second, the intertemporal budget constraint has to hold over an infinite horizon and binds under the assumptionthat the economy operates efficiently. The implications over a finite horizon are not clear ex ante, particularly if the

economy operates inefficiently and growth rates exceed interest rates. Under these circumstances, countries can engagein a deficit gambleeven for an extended period of time (Ball et al., 1998; O'Connell and Zeldes, 1988).Therefore, we conduct panel cointegration tests for different combinations of spending and revenue aggregates or

sub-categories. The panel cointegration tests ofPedroni (1999)make use of estimated residuals from the hypothesisedlong run regression of the form:

yit aidit gtb1ix1itb2ix2it Nbmixmiteit 7

whereMis the number of regressors. This can be seen as a fixed effects model, where iand itare country-specificintercepts and deterministic trends respectively, andtrepresents a set of time dummies common to all members of thepanel. The coefficients,mi(m =1,2,...M), are allowed to differ across individuals. Evidence in favour of cointegrationis provided wheneitin Eq. (7) is found stationary. This is done by rejecting the null thati=1 in eit

qi ei;t1

uit.

In order to identify long-run relationships, we present the results of the pooled and mean-group ADF t-statisticsdeveloped byPedroni (1999)for the specifications with and without heterogeneous deterministic trends.9 Pedroni(1999)rescales the mean-group ADFt-statistic withN1/2 so that it is distributed as a standard normal distribution. Thestandardisation of the cointegration statistics can be expressed as:

jNTlffiffiffiffiN

pffiffiffim

p ZN0; 1 8

whereNTis the standardised form of the test statistic with respect to Nand T. The values of the mean () and thevariance () are tabulated inPedroni (1999). The values of the normalised statistics are to be compared to the criticalvalues implied by a lower-tailed standard normal distribution. We opt for normalising on the variable standing for the

revenue side of the government budget constraint, without implying that the direction of causality is running from thespending to the revenue side.10

Results inTable 3support the existence of a long-run relationship between revenues and expenditures. First, totalspending as a share of GDP appears to clearly cointegrate with total current revenues when deterministic trends areincluded in the cointegrating vector. Similar cointegrating patterns are found for the long-run relationship between totalexpenditures and revenue subcategories in the specification with deterministic trends. Given this result we would expectbig ticket items to co-move with the other side of the budget. Indeed, we find clear indications that total transfers

9 The pooling of information is done along the within dimension of the data for the pooled version of the test and along the between dimension forthe mean-group ADFt-statistic. The latter, thus, allows for an additional source of cross-country heterogeneity under the alternative. The panel unitroot and cointegration analyses were carried out with routines kindly provided by Peter Pedroni.10

Empirical evidence on Granger-causality for expenditures and revenues indicates different patterns running unidirectional from expenditures torevenues or vice versa, or being bi-directional in some countries (Belessiotis, 1995).



10/20

cointegrate withall revenuesubcategories, in particular with distortionary taxation. Governmentconsumption cointegrateswith direct and distortionary taxation. Finally, there is firm evidence that public investment cointegrates with total revenueas well as with distortionary taxation and its components, especially in the specification without trends.

Table 4presents the estimates of the long-run coefficient for the variables entering the cointegrating vector. The

long-run estimates should provide evidence for the co-movement between fiscal categories from both sides of thebudget. We base our inferences on the mean-group Fully Modified OLS estimator (FMOLS) proposed byPedroni(2000), which corrects the standard OLS for the bias induced by the endogeneity and serial correlation of theregressors. It also allows for heterogeneous cointegrating elasticities and more flexible hypotheses testing than thepooled FMOLS estimator. The cointegrating vector is again normalised on the revenue category.

Looking at these relationships from the expenditure side, we find that the long-run coefficient of the relationship ofaggregate revenues with total expenditures is around 0.7 and is robust to the inclusion of time effects. This implies thatan increase in total government spending is under-compensated by an increase in total revenues.11 We tested for a one-to-one relationship between total revenues and total expenditures, rejecting the existence of a proportional relation atthe 1% significance level.12 This finding can be partly explained by the deficit bias leading to a continuous debt build-up during the 1970s and 1980s in many European countries.13 Furthermore, the coefficient of the relationship of

government transfers and consumption with total revenues and their sub-categories falls significantly when timedummies are included, though remaining positive and significant. This indicates that part of the co-movement betweenthese budget categories is accounted for by factors which are invariant across EU countries, such as common supply ordemand shocks. The long-run elasticity of public investment to total revenues and distortionary taxation equals 0.15,and 0.32 for direct taxation in the specification with time dummies.

Table 3Panel cointegration results for public finances

Countrydummies

Countrydummiesand trend

Countrydummies


Countrydummies


Countrydummies


Dep. variable Indep. variable Indep. variable Indep. variable Indep. variableLTREV LTEXP LPI LTR LGPanel adf-stat 0.398 5.750 1.493 0.049 0.328 3.359 1.898 0.315Group adf-stat 1.269 7.121 2.479 1.903 0.363 6.489 1.661 1.509

LTDIR LTEXP LPI LTR LGPanel adf-stat 2.239 6.573 2.261 1.409 0.289 5.104 2.871 2.119

Group adf-stat 3.954 11.657 3.633 3.372 0.688 10.979 3.966 4.519

LSSC LTEXP LPI LTR LGPanel adf-stat 0.309 3.235 2.745 0.530 0.947 3.859 0.398 0.967Group adf-stat 4.673 7.101 2.885 0.157 3.340 5.713 1.049 4.903

LTDIST LTEXP LPI LTR LGPanel adf-stat 1.985 8.215 2.603 1.300 0.417 4.752 1.769 1.298

Group adf-stat 4.994 11.341 2.638 1.750 1.181 7.637 2.304 5.280

LTIND LTEXP LPI LTR LGPanel adf-stat 1.605 5.815 2.381 5.786 0.177 6.381 0.388 2.659

Group adf-stat 1.351 10.883 7.803 12.426 0.273 18.580 0.231 6.006

Note: The cointegrating vector is normalised on the revenue category. The normalised panel cointegration statistics must be compared to the criticalvalues of a lower-tailed standard normal distribution. , and imply rejection of the null of non-cointegration at the 10%, 5% and 1% level ofsignificance, respectively.

11 Again, when we talk about undercompensating movements in revenues as a result of expenditure changes, we do not mean causality going fromexpenditures to revenues.12 Quintos (1995)showed that a cointegrating slope lower than one may still be compatible with public deficit sustainability, provided the debt

ratio grows at a lower rate than that of mean interest rates. Santos Bravo and Silvestre (2002)note that such condition holds for most EU countriesover the period 19602001.13

This appears in line with the findings in Afonso (2005)andSantos Bravo and Silvestre (2002)who conduct cointegration tests for individual EUmember states.



11/20

In short, this exercise has shown a strong cointegrating relationship between public revenues and expenditures. Thisholds particularly true for budget aggregates, but also applies to sub-categories. Moreover, the cointegrating vectorsgenerally have the expected sign, showing that expenditures and revenues co-move in the same direction. To the extentthat their persistent component may have opposite effects on growth, they would therefore cancel each other out. Thatwould be particularly true for the cointegrating relationship between direct taxation and public investment. Aninteresting case is the cointegrating relationship found between government consumption and indirect taxation. Thus, ifgovernment consumption exerts a persistent growth-enhancing effect, this would not be fully offset by distortionary tax

developments.

5. The impact of public finances on long-term growth A distributed lag test

The previous exercise searching for deterministic processes has shown that per-capita GDP growth rates and fiscalpolicy variables are marked by persistent, long-run developments. Real per-capita GDP shows a declining trend in thegrowth rate and public finance variables follow some persistent upward or downward trends which could be compatiblewith this growth pattern. When we look at stochastic processes, fiscal variables also reveal a persistent component,while GDP growth is mean-reverting. This pattern is still compatible with the assumption that the persistent impacts ofproductive expenditures and distortionary taxation cancel each other out. But even under the condition that both effectsoffset each other, fiscal policies may have a short or medium term impact on the growth rates of the economies. In this

section we will therefore analyse more systematically whether fiscal variables affect economic performance over thebusiness cycle.

5.1. Estimation procedure

The long-term effect of fiscal policy on growth can be estimated using a distributed lag approach, controlling forboth sides of the government budget. Thus a simple model of the following form will be estimated:

DyitXIj0

bgjgitjXIj0

bsjsitjaieit 9

whereyindicates the log of per capita output,gand represent the log of government expenditures and revenuescategories expressed as shares of GDP, iis a set of country dummies and Irepresents a finite number of lags. This

Table 4Panel cointegrating vectors for public finance (aggregates and economic sub-categories)

Dep. variable Indep. variable Indep. variable Indep. variable Indep. variable

Countrydummies

Countrydummies and

time dummies

Countrydummies

Countrydummies and

time dummies

Countrydummies

Countrydummies and

time dummies

Countrydummies

Countrydummies and

time dummiesLTEXP LTR LG LPI

LTREV 0.66 0.71 0.46 0.37 0.75 0.26 0.01 0.15

(31.25) (26.6) (15.88) (11.26) (20.6) (10.48) (8.45) (57.42)1-to-1 rel test (13.52) (13.84)

LTDIR 0.74 1.27 0.61 0.46 0.92 0.35 0.10 0.32

(17.82) (15.89) (10.99) (6.3) (14.2) (7.42) (3.5) (7.96)LSSC 0.87 0.52 0.65 0.39 0.81 0.01 0.13 0.02

(26.23) (13.06) (24.35) (11.73) (14.95) (1.22) (7.62) (2.6)LTDIST 0.76 0.74 0.60 0.38 0.80 0.20 0.03 0.15

(32.64) (18.06) (20.54) (12.27) (17.1) (4.58) (6.61) (7.08)LTIND 0.16 0.15 0.12 0.70 0.41 0.10 0.12 0.02

(2.43) (4.12) (1.85) (3.6) (4.41) (1.17) (1.1) (5.18)

Note: The cointegrating vector is normalised on the revenue category. The row labelled by 1-to-1 rel test relates to testing the existence of long-runproportionality between total government revenues and aggregate expenditure. T-statistics are given in parenthesis below the estimates., and

imply rejection of a zero long-run elasticity at the 10%, 5% and 1% level of significance, respectively.



12/20

equation can theoretically be used to test exogenous vs. endogenous growth theories. As Evans (1997) andKocherlakota and Yi (1997)show, exogenous growth theory implies

XI

j

0

bgj XI

j

0

bsj0

as the lag order goes to infinity. Conversely, endogenous growth theory implies for productive expenditures

XIj0

bgjN0

and for distortionary taxation

XIj0

bsjb0

In other words, the sum of coefficients has to be different from zero for a sufficiently large lag order if endogenousgrowth predictions are to be valid. It is not clearex ante what constitutes the right lag order in this context. Even transitorychanges to the new equilibrium state after a fiscal reform can expand over several years. Thus, we may not be able tosensibly discriminate between the two growth theories, but only confirm whether public finances have a consistent impacton growth over the cycle and possibly affect trend growth. Therefore, we use a lag-length equal to eight since spectralanalysis tends to indicate a business cycle of six to eight years for European countries (see Bouthevillain et al., 2001).Moreover, this specification is in line with the literature in the public finance field (e.g.Bleaney et al., 2001).

Eq. (9) can be rewritten as a function of the lag operator as follows:

DyitALgitBLsitaieit 10whereA(L) andB(L) represent two lag polynomials with a unit root outside the unit circle. We re-parameterise Eq. (10)in line withJones (1995)in order to separate long-run from short-run effects as follows:

DyitA1gitCLDgitB1sitDLDsitaieit 11whereC(L) andD(L) are (p1)th-order lag-polynomials such that:

cis Xp

js1aij

dis Xp

js1bij

wheres = 1,,p1.

In sum, the coefficients forA(1) andB(1) capture the long-run effect of government spending and revenue categorieson growth, while the first-difference terms will capture short-run interactions between fiscal policies and growth.

The baseline estimation approach is amplified in two respects. First, a set of other standard variables is added tocontrol for further growth factors. Second, we use variables for short-term domestic and international shocksaffecting growth performance below business cycle frequency to ensure that our results are not distorted byshorter-term correlations. Therefore the estimation specification takes the following form:

DyitA1gitCLDgitB1sitDLDsitKwitaieit 12

where K is a vector of parameters and it represents a set of control variables which are commonplace in theempirical growth literature.14 This includes the private investment rate, the average years of schooling in the

14 The data set used for the empirical analysis is available from the authors upon request.



13/20

working age population,15 changes in the trade openness ratio computed as exports plus imports over GDP,16 theinflation rate and US GDP growth. Our coefficient estimates for the fiscal variables will thus measure the influencethat individual fiscal policies have on growth beyond their effect on the decision to invest in physical and humancapital. Inflation is used as domestic conjunctural indicator assuming that short to medium-term fluctuations ofoutput are mainly demand-driven, i.e. inflation and output move in the same direction. Private investment andinflation are instrumented using two lags to cope with the endogeneity problem affecting these two variables. USGDP growth is employed as a variable to control for the international environment.17 In addition, we allow forcountry effects, which should capture all time invariant sources of growth, such as persistent features of thepolitical system and initial conditions. Finally, time dummies for the years 1990 to 1992 are incorporated to correctfor the impact of German reunification.18

15 Though not entering directly in the model sketched in Section 2, direct taxation on personal income, due to its progressivity, can reduce theformation of human capital. Higher tax rates levied on higher incomes reduce the return to investment in human capital, thus leading to lesseducation and lower growth.16 The IPS test consistently showed for the specifications analysed in Section 4.2 that the trade openness ratio follows an I(1) process. Therefore,

the variable enters the model in first differences.17 SeeGali and Perotti (2003), who also employ this variable. In addition, we used real oil price growth as a measure for the international

environment. It did however not yield any significant estimates and was therefore dropped.18 These are the only years for which we could find robust and significant estimates when all time effects are included in the model.

Table 5Public finances and long-term growth

(1) (2) (3) (4) (5)

Private investment 0.015

(0.006)0.01

(0.006)0.01

(0.006)0.006(0.007)

0.004(0.007)

Human capital 0.03(0.02) 0.04

(0.02) 0.04

(0.02) 0.03(0.02) 0.02(0.02)Openness 0.10

(0.03)0.01(0.02)

0.01(0.02)

0.02(0.03)

0.003(0.03)

Inflation 0.29

(0.03)0.20

(0.04)0.25

(0.03)0.21

(0.04)0.25

(0.04)US GDP growth 0.002

(0.001)0.001

(0.000)0.001

(0.001)0.001

(0.001)0.001

(0.001)Total revenues 0.06

(0.01)0.07

(0.03)Direct taxes 0.006

(0.01)0.02

(0.01)Social security contributions 0.01

(0.01)0.03(0.02)

Indirect taxes 0.02(0.01)

0.02(0.02)

Total expenditures 0.05

(0.009)0.05

(0.02)Government consumption 0.06

(0.02)0.08

(0.02)Government transfers 0.006

(0.01)0.04

(0.02)Government investment 0.03

(0.005)0.03

(0.007)Rsquare (within) 0.44 0.56 0.67 0.59 0.67Wald test 1071.8 1478.22 1621.83 1288.5 1571.2

Usable observations 468 468 407 416 407

Note: The dependent variable is the growth rate of per-capita GDP. The estimates shown represent long-term coefficients. Standard errors are given inparenthesis., and imply the rejection of the null of insignificant parameter estimates at the 10%, 5% and 1% levels, respectively.



14/20

5.2. Estimation results

5.2.1. Public finances and growthTable 5reports the estimates of the growth effects from aggregate revenues and expenditures as well as from different

subcategories. Model 1 controls for the conditioning information set in addition to total revenues which measure the overall

effect that government size has on growth.19

The coefficient on total revenues carries a negative sign and is statisticallysignificant at the 1% level. Likewise, model 2 renders a comparable coefficient on total expenditures (as an alternativemeasure of government size) equal to 0.05, which should capture the benefits of spending minus the cost of taxation inaddition to the reduction of growth due to deficit financing. When we try to include both total revenues and aggregateexpenditures in the same (unreported) specification, only the latter remains significant. The fact that one variable losesstatistical significance should not be surprising given the high correlation between both aggregates with a correlationcoefficient larger than 0.9. In order to disentangle whether the balance of benefits and costs looks different for individualspending categories, model 3 analyses the sign and size of the growth effects due to government spending categories whilecontrolling for total revenues. Government consumption and public investment are significant at the 1% levels with anegativeand positiveparameterestimate, respectively. Government transfers enter insignificantly witha negativecoefficient.When we control for different revenue categories in model 5, the results for public investment and government consumption

remain robust, while the negative coefficient estimate for government transfers becomes statistically significant.According to the simple theoretical model presented earlier, productive spending items should have a positive growtheffect, while non-productive spending would be at worst neutral. The empirical results support the notion that publicinvestment can be considered productive spending contributing to long-term growth. The negative estimates for the twoother major spending items cannot reflect the tax financing costs since this is properly accounted for in the model.Therefore it should either relate to the deficit financing costs in terms of growth performance or a growth-reducing effectwhich is not apparent in our model, where among others labour supply is inelastic. In a more complete theoretical model,social benefits or government wages may reduce labour supply in the private sector and thereby undermine growth.20

As stressed by Widmalm (2001), not all taxes are expected to have the same impact on growth.21 It is, thus, necessary toinvestigate the growth effects of the tax mix. As regards capital taxation, which includes personal and corporate incometaxes, it is clear that it reduces growth by decreasing the accumulation of capital. In a growth model like the one sketchedabove, the assumption of an inelastic labour supply makes a consumption tax or a flat tax on labour income useless to

influence growth. It would not affect intertemporal decisions on consumption and capital accumulation. On these grounds,one may expect direct taxation and, to a lower extent, social security contributions to be growth inhibiting, while indirecttaxation would be neutral. Models 4 and 5 can help us shed some light on this issue. The former controls for aggregategovernment expenditures as wellas for the tax categories.Totalexpenditures enter significantly witha negative coefficient,while all tax coefficients are negative though none is statistically significant at conventional levels. However, when thegrowth effect associated with different spending categories is captured in model 5, the coefficient for direct taxationremains negative and becomes statistically significant at the 5% level. Thus in the most complete model, we find on theexpenditures and on the revenue side growth effects which stand much in line with theoretical predictions.

For theregressionresults reportedabove to be valid, one has to assume that the fiscal variables on the right hand side areexogenous. However, this can be questioned due to the existence of automatic stabilizers, i.e. mechanisms built into fiscalregulations that lead to counter-cyclical fluctuations of transfers and taxation over the business cycle. Spending items such

as public investment or government consumption are generally not affected by automatic stabilisers.Under the assumptionthat this is thesource of endogeneity, parameter estimatesfor social transfers would be downward biased because thesocialtransfers to GDP ratio should decline in high growth periods and vice versa. The parameter estimated for social transfersmay therefore be the result of this source of endogeneity. The estimates for taxes would be upward biased since the tax to

19 We also report thet-statistics and the significance level of the estimates, although these do not allow proper inference since the GDP growth rateis stationary and the policy variables areI(1). However, it should be noted that the results are largely in line with those reported later on, where wecontrol properly for both sides of the budget. To the extent that spending and revenue items are in general cointegrated, the t-statistics can be usedfor statistical inference.20 These negative coefficients are much in line with the results reported in De la Fuente (1997)for industrialised countries. De la Fuente also

discusses the productivity and investment channels through which government expenditures could have a negative externality for growth. These arehowever partly captured by our tax controls.21

Widmalm (2001) find that the proportion of tax revenue raised through taxes on personal income negatively affects growth in 23 OECDcountries over the period 19651990.



15/20

GDP ratio should increase in high growth periods. On this account, the negative parameter estimate for direct taxationwould be the upper bound of plausible estimates, while the lack of statistical significant results mainly for social security

contributions could be driven by this source of endogeneity. Indirect taxation generally tends to show an output gapelasticity close to one, which holds the tax to GDP ratio constant over the cycle and should therefore not produce asignificant bias. In order to control for this problem, we estimate a set of specifications using leads, cyclically adjustedfiscal data (either produced by theEuropean Commission (various years)or trends derived by HP-filtering) as well asinstrumental variables GMM estimates. Details are reported in Appendix B. None of these methods suggests that theresults reported above are driven by an endogeneity problem caused by the functioning of automatic stabilisers.22

5.2.2. Taxation and private investmentEndogenous growth theory suggests that distortionary taxation affects the investment decision, and provided private

investment exerts a positive impact on growth, distortionary taxation in turn affects growth. In order to check for thistransmission channel of physical capital accumulation, we run distributed lag regressions with private investment as the

dependent variable on different revenue categories, also controlling for aggregate government outlays. The controlvariables often employed in the growth literature are dropped from Eq. (12) since one cannot presume ex ante that theyhave a direct impact on private investment.23 The investment model includes fixed time and country effects in additionto the set of fiscal variables.

Table 6reports the results when using revenues to GDP ratios as proxies for tax effects. Models (6) to (8) combinetotal expenditures with different revenue categories. Model (9) then combines the three main expenditure and taxcategories. In all specifications direct taxation carries a negative coefficient which is mostly statistically significant atstandard levels. Indirect taxation yields a positive and highly statistically significant parameter estimate. A possible

Table 6Public finances and private investment

(6) (7) (8) (9) (10) (11)

Total expenditures 0.12(0.15)

0.13(0.16)

0.53

(0.19)0.77

(0.19)

Government consumption 0.53

(0.17)Social transfers 1.26

(0.13)Public investment 0.18

(0.07)Direct taxes 0.24

(0.11)0.28

(0.11)0.15(0.11)

0.23

(0.09)Social security contributions 0.45

(0.13)0.35

(0.13)0.83

(0.19)Indirect taxes 0.85

(0.14)1.15

(0.12)Average effective tax on labour 1.22

(0.16)0.68

(0.20)

Average effective tax on capital 0.09(0.11)

0.18(0.11)

Average effective tax on consumption 1.20

(0.15)1.16

(0.15)R2 (within) 0.33 0.36 0.44 0.57 0.47 0.53F-test 3.35 3.11 3.54 4.35 4.81 4.96

Usable observations 416 416 416 407 345 345

Note: The dependent variable is the private investment share of GDP. The estimates shown represent long-term coefficients. Standard errors are givenin parenthesis. The sample period for models including average effective tax rates from 19702001. , and imply the rejection of the null ofinsignificant parameter estimates at the 10%, 5% and 1% levels, respectively.

22 The fact that the cyclically adjusted and the GMM estimates do not corroborate the result reported in model 5 for direct taxation could be seen aslag of robustness. However, as also noticed in the annex, it should be taken into account that the cyclical adjustment is subject to substantialmethodological problems and GMM estimates may suffer from efficiency losses and finite sample bias given the relatively restricted time and cross-

sectional dimension of our data set.23 Moreover, these variables became statistically insignificant when we experimented with variations of Eq. (12) as a robustness check.



16/20

explanation for this result would be the systematic substitution from direct to indirect taxes pursued by severalEuropean countries in the 1980s and 1990s as part of their reform strategy. Finally, the coefficient estimate for socialsecurity contributions as a share of GDP is negative when total expenditures is included but becomes positive when allspending categories are incorporated in the model. The estimate is statistically significant in all specifications, whichcomplicates the interpretation since multicollinearity would not only be associated with parameter instability, but also

with lack of statistical significance. Arguably, the change in the sign of the coefficient is thus related to the fact thatsocial security contributions capture an effect attached to a specific spending category in model (9). The most robustresult from this exercise is therefore that direct taxes have a negative impact on private investment.

Since direct taxation is a broad revenue category, we further investigate the link between investment and taxation byusingan updatedversion of a more differentiated data set of the effective taxrateson labour, capital and consumption goodsfromMartnez-Mongay (2000).24 The effective tax rates should capture the average negative effect that taxation exerts ongrowth, thereby constituting a better proxy than the revenue shares of GDP. Models (10) and (11) give a clear message.Labour taxation consistently affects investment. The coefficient takes on a negative value of around0.7, implying that a1% increase in capital taxation would cause a fall in the private investment rate by 0.7%. We could, however, not find anysignificant effect from the effective tax rate on capital on private investment.25 While this result may look surprising at firstsight, one has to notice that the bulk of direct taxation comes from labour income. Moreover, as Alesina et al. (2002)and

Ardagna (2006)show, increases in direct taxation put pressure on union

s wage claims, leading to higher private sectorwages and lower profit margins and investment in unionized labour markets in industrialized countries. The coefficient forconsumption taxes is consistently positive in line with the earlier results reported for indirect taxes.

Overall, our results point to the existence of a significant impact from aggregate government expenditure and itsmain subcategories on growth. The size of the public sector and the government consumption negatively affect long-run growth. This may reflect the fact that the average size of the public sector in the EU is above its optimal level.Conversely, public investment has a positive impact on growth which indicates the likely gains in economicperformance from shifting welfare expenditure to productive investment. Furthermore, we find that above all directtaxation negatively affects growth through its impact on private capital accumulation.

6. Conclusions

The Lisbon Process assigns a prominent role to public finance reform in order to foster economic growth. The mainpurpose of our analysis is therefore to shed some light on the relation between public finances and growth in the EU-15.Most importantly, this requires determining whether public finances provide policy instruments contributing to highertrend growth, or whether they can at best be expected to have a short-run impact on economic performance. Following theapproach of some studiesin this fieldthat exploit the time series properties of the data, we find some persistent deterministicchanges in per-capita GDP growth rates and public finances. Nevertheless, when we look at stochastic trends, it appearsthat public finance variables have generally shown persistence over time while growth rates of output have remained fairlystable. This pattern does not excludea long-termeffect of fiscalvariables per se, if expenditures and revenues have oppositelong-term effects and co-move. Using recently developed panel cointegration techniques, we have indeed foundoverwhelming evidence of cointegration between both sides of the budget, as would be expected on theoretical grounds.

We then estimate the long-run effect of fiscal policies on growth using a distributed lag approach. We improve on

previous studies on the nexus of fiscal policies and growth by better controlling for real business cycle effects andreverse causality as well as by using better proxies for taxation. The main findings are that the expenditure side of thebudget appears to consistently affect long-run growth over the business cycle. Specifically, government size andgovernment consumption are found to have a clear negative effect on growth, while public investment positively

24 For a discussion on different measures of the tax burden on labour seede Haan et al. (2004). Although there are substantial differences, the authorsfind that this does not affect strongly conclusions regarding the economic impact found in several well-known empirical studies using these measures.25 This finding is not in line with Mendoza et al. (1997). The reason for the discrepancy of the results cannot easily be explained since Mendoza

et al. (1997)use a different methodology and data sample. As a robustness check, Portugal has been dropped from the sample, since it shows fairlylarge swings in capital tax revenues in the 1970s and therefore could present an outlier. For the reduced sample, the capital taxation coefficient

becomes indeed negative (0.17), but does not attain standard significance levels, while the coefficients for labour and consumption taxes remainstable and statistically significant. Amplifying the model by including our human capital variable and changes in trade openness, to use a model

more closely reflecting the specification chosen by Mendoza et al., also yields a negative coefficient when Portugal is omitted, which is however notstatistical significant at the 10% level.



17/20

affects growth. On the revenue side, evidence exists for a negative effect of direct taxation on growth in our preferredspecification. Moreover, a robust negative impact of direct taxation on physical capital accumulation is confirmed byour data. This impact seems to work primarily through the taxation of labour income which may lead to wage pressure,thus lowering profits and investment in unionized European labour markets. Taken as a whole, our results stand incontrast to previous studies on this issue (e.g.Evans, 1997; Karras, 1999) which did not appropriately take into account

the financing relations implied by the intertemporal budget constraint.Before concluding the paper, it is important to point out some limitations of the present analysis. First, the growth effects

estimated represent reduced-form partial correlation coefficients, which do not necessarily imply causation. Second, thefact that the panel unit root tests employed in the analysis fail to allow for structural breaks may bias the results towards thenon-rejection of the joint non-stationarity null, which was required for the conduct of the cointegration analysis. Third,given the limited lengthof thedata seriesavailable, it may be impossible to know with certainty whether the growth impactestimated is the result of a continuum of level shifts along the transitional path as held by neoclassical growth theory orrepresents a genuine growth effect as advocated by endogenous growth defenders.26 Finally, according to the endogenousgrowth models being tested, it would be desirable to use marginal tax rates rather than average or effective tax rates.However, we have refrained from employing marginal tax rates due to the difficulties in consistently computing them forthe EU-15 countries. This is because the system of exemptions and scalesused to tax different types of income vary widely

across countries. Future research should be conducted in the direction of providing reliable estimates of marginal tax ratesso that new results can be confronted with the ones obtained in the present paper.

Acknowledgements

We would like to thank Jrg Breitung, Antonio Fats, Manfred Kremer, Chiara Osbat, Jonathan Temple, Jrgen vonHagen, Simon Wren-Lewis, and two anonymous refereesfor very helpful commentsand discussion. Pedro Pedroni kindlymade his RATS codes for the panel cointegration analysis available to us. All remaining errors are ours. The opinionsexpressed herein are those of the authors and do not necessarily represent those of the European Central Bank.

Appendix A. Data sources and descriptive statistics

Table A1Notation of variables

Variable Definition Source

LYPC Gross domestic product in per capita terms nominatedin constant dollars

Economic outlook N.74, OECD.

LPRINV Private physical investment share of GDP Economic outlook N.74, OECD.LHK Average years of education of working age population Bassanini and Scarpetta, 2002; De la Fuente and Domnech, 2006.LOPEN Ratio of exports plus imports to GDP Penn World Table 6.1.INF Inflation rates computed with consumer price index data International Financial Statistics (IMF).DLYPC Growth rate of real GDP per-capita Economic Outlook N.74, OECD.LTREV Total current revenues as a share of GDP Ameco StatisticsAutumn 2003, European Commission.LTEXP Total expenditures as a share of GDP Ameco StatisticsAutumn 2003, European Commission.

LPI Total public investment as a share of GDP Ameco StatisticsAutumn 2003, European Commission.LTR Total transfers as a share of GDP Ameco StatisticsAutumn 2003, European Commission.LG Government consumption spending as a share of GDP Ameco StatisticsAutumn 2003, European Commission.LTDIR Total direct taxation as a share of GDP Ameco StatisticsAutumn 2003, European Commission.LSSC Social security contributions as a share of GDP Ameco StatisticsAutumn 2003, European Commission.LTDIST Total distortionary taxation as a share of GDP computed as

the sum of direct taxation and social security contributionsAmeco StatisticsAutumn 2003, European Commission.

LTIND Total indirect taxation as a share of GDP Ameco StatisticsAutumn 2003, European Commission.LTL Effective labour tax rate Martnez-Mongay (2000).LTK Effective capital tax rate Martnez-Mongay (2000).LTC Effective consumption tax rate Martnez-Mongay (2000).

Note: All variables except for DLYPC and the inflation rate are expressed in log-levels.

26 SeeTemple (2003)for a more extended discussion of this issue.



18/20

Table A2Descriptive statistics, European countries 19612001

Obs. Mean S.D. Min Max

LYPC 630 16,145.601 7507.463 3619.817 66,398.890LHK 615 9.240 1.836 4.370 13.600

LOPEN 615 59.808 44.077 8.985 276.545INF 615 0.060 0.049 0.007 0.255LINV 600 20.415 4.644 5.013 35.411LTREV 597 40.628 9.711 17.438 62.859LTDIR 544 12.925 5.763 2.369 30.644LSSC 544 11.398 4.924 1.463 21.056LTDIST 544 24.323 7.227 7.692 39.626LTIND 544 13.601 3.274 5.875 30.644LTEXP 597 42.818 10.277 17.385 70.075LG 536 17.323 4.313 7.845 28.858LTR 544 15.284 5.057 3.029 28.473LPI 544 3.287 1.053 1.034 6.468DLYPC 615 2.835 2.683 11.049 11.552LTL 465 32.503 9.381 11.100 54.100

LTK 465 20.009 6.451 6.800 38.000LTC 465 20.263 4.463 9.100 31.400

Notes: All variables except growth rates of per capita GDP and inflation rate are expressed in levels.

Table A3Correlation matrix of main variables, European countries 19612001

LYPC LINV LHK LOPEN INF LTREV LSSC LTIND LTDIR LTDIST LTEXP LG LTR LPI DLYPC

LYPC 1.00 0.47 0.80 0.70 0.39 0.78 0.33 0.42 0.68 0.80 0.70 0.53 0.67 0.14 0.17LINV 0.47 1.00 0.41 0.43 0.39 0.45 0.12 0.29 0.41 0.39 0.47 0.37 0.39 0.25 0.12LHK2 0.80 0.41 1.00 0.53 0.41 0.84 0.23 0.62 0.73 0.78 0.75 0.70 0.68 0.17 0.26LOPEN 0.70 0.43 0.53 1.00 0.42 0.57 0.17 0.35 0.60 0.56 0.51 0.25 0.48 0.09 0.02INF 0.39 0.39 0.41 0.42 1.00 0.41 0.17 0.27 0.35 0.38 0.24 0.23 0.29 0.22 0.25LTREV 0.78 0.45 0.84 0.57 0.41 1.00 0.30 0.57 0.87 0.95 0.92 0.79 0.82 0.06 0.31LSSC 0.33 0.12 0.23 0.17 0.17 0.30 1.00 0.25 0.07 0.46 0.34 0.02 0.54 0.05 0.17LTIND 0.42 0.29 0.62 0.35 0.27 0.57 0.25 1.00 0.59 0.39 0.49 0.59 0.33 0.04 0.07LTDIR 0.68 0.41 0.73 0.60 0.35 0.87 0.07 0.59 1.00 0.82 0.76 0.76 0.58 0.05 0.22LTDIST 0.80 0.39 0.78 0.56 0.38 0.95 0.46 0.39 0.82 1.00 0.89 0.68 0.86 0.08 0.33LTEXP 0.70 0.47 0.75 0.51 0.24 0.92 0.34 0.49 0.76 0.89 1.00 0.75 0.89 0.06 0.42LG 0.53 0.37 0.70 0.25 0.23 0.79 0.02 0.59 0.76 0.68 0.75 1.00 0.47 0.11 0.35LTR 0.67 0.39 0.68 0.48 0.29 0.82 0.54 0.33 0.58 0.86 0.89 0.47 1.00 0.11 0.36LPI 0.14 0.25 0.17 0.09 0.22 0.06 0.05 0.04 0.05 0.08 0.06 0.11 0.11 1.00 0.08DLYPC 0.17 0.12 0.26 0.02 0.25 0.31 0.17 0.07 0.22 0.33 0.42 0.35 0.36 0.08 1.00

Note: The matrix shows the correlations of relevant variables.

Appendix B. Public finances and growth: Robustness results

This appendix reports the results of various specifications controlling for the possible endogeneity of fiscal variablesin Eq. (12) due to the existence of automatic stabilisers. Three different approaches are used for this purpose. First, Eq.(12) is augmented with leads of fiscal variables. Eight lags and five leads are included in our specification.27 Second,regressions are run with cyclically adjusted budgetary aggregates as provided by the European Commission. Sincethese are not available for different spending and revenue categories, we also compute cyclically adjusted spending andrevenues as shares of trend GDP using a HP-filter.28 Third, we use a GMM instrumental variables estimator to check

27 The results appear fairly robust to different lag and lead-lengths. In order to keep a reasonable number of usable observations we set to eight andfive the number of lags and leads included in the regressions respectively. Panel studies often average the data over five-year periods to cancel out

cyclical fluctuations, so we use a number of leads equal to five.28 The smoothing coefficient is set to =100.



19/20

for endogeneity.29 Considering that our regressions already include lagged values of the fiscal regressors up to 8 years,we employ as instruments for the fiscal policy variables their values lagged 9 and 10 periods. 30

In interpreting the results it should be kept in mind that the operation of automatic stabilizers would induce adownward bias for social transfers and an upward bias for taxes, in particular if taxes have an output elasticity largerthan one. For the sake of brevity, Table A1 reports only the results for the coefficient estimates for fiscal variables.

Model (13) is based on the cyclically adjusted budgetary aggregates as derived by the European Commission. Theestimated coefficients do not provide a clear picture. The coefficient for aggregate spending increases slightly as onemight expect, but the coefficient for total revenues, contrary to what one would expect, increases even more. For socialtransfers, the coefficient estimate is even lower than the one found in the comparable specification (model 5) inTable 5.For direct taxation, the result varies considerably across specifications. The specification including leads (model 12)yields a somewhat lower coefficient than the one reported inTable 5. The estimate is significant at the 10% level. HP-filtered direct taxation as a share of trend GDP yields a statistically significant positive coefficient. Parameter estimatesvary again considerably and no statistically significant coefficient is found for social security contributions and indirecttaxation. Overall, based on these results there is no evidence for a systematic bias in parameter estimates induced by theoperation of automatic stabilizers. The results reported in Table A4 may rather point to some problematic aspects ofthese specifications, in particular the reduction of the sample size and degrees of freedom through the inclusion of

leads, measurement issues for the cyclical adjustment of fiscal data and the possible inefficiency of IV-estimates.

Table A4Public finances and growth: Robustness checks

(12) (13) (14) (15)

Leads Cyclically adj. (EC) Cyclically adj. (HP-filter) GMMTotal primary expenditures 0.04

(0.03)Government consumption 0.06 0.02 0.06

(0.04) (0.06) (0.02)Social transfers 0.05 0.05 0.05

(0.04) (0.02) (0.02)Public investment 0.03 0.00 0.02

(0.01) (0.02) (0.007)Total revenues 0.01

(0.05)Direct taxes 0.04 0.05 0.01

(0.02) (0.02) (0.01)Social security contributions 0.04 0.03 0.02

(0.04) (0.03) (0.02)Indirect taxes 0.05 0.05 0.02

(0.04) (0.03) (0.02)

Note: The dependent variable is per-capita GDP growth. The estimates shown represent long-term coefficients. Standard errors are given inparenthesis. The sample period for models including average effective tax rates is 19702001. , and imply the rejection of the null ofinsignificant parameter estimates at the 10%, 5% and 1% levels, respectively.

References

Afonso, A., 2005. Fiscal sustainability: the unpleasant European case. FinanzArchiv 61, 1944.Alesina, A., Ardagna, S., Perotti, R., Schiantarelli, F., 2002. Fiscal policy, profits, and investment. American Economic Review 92, 571 589.Ardagna, S., 2006. Fiscal policy in unionized labour markets. Working paper. Harvard University, Cambridge MA.Arellano, M., Bond, S., 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Review

of Economic Studies 58, 277297.Ball, L., Elmendorf, D.W., Mankiw, N.G., 1998. The deficit gamble. Journal of Money, Credit and Banking 30, 699 720.Baltagi, B.H., Kao, C., 2000a. Nonstationary panels, cointegration in panels and dynamic panels: a survey. Working Paper. Syracuse University,

Syracuse.

29 More specifically, we employ the Generalised Method of Moments (GMM) approach ofArellano and Bond (1991)which first-differences the

data and employs lagged values of the regressors as instrumental variables for the first-differenced data.30 This choice of instruments helps us correct for the possibility of measurement errors in the variables. See more details inBond et al. (2001).



20/20

Baltagi, B.H., Kao, C., 2000b. Nonstationary panels, cointegration in panels and dynamic panels: a survey. In: Baltagi, B.H. (Ed.), Non-StationaryPanels. Panel Cointegration and Dynamic Panels Advances in Econometrics, vol. 15. Elsevier Science Inc., Amsterdam, pp. 751.

Barro, R., 1990. Government spending in a simple model of endogenous growth. Journal of Political Economy 98, 103 125.Bassanini, A., Scarpetta, S., 2002. Does human capital matter for growth in OECD countries? A pooled mean-group approach. Economics Letters 74,

399405.Belessiotis, T., 1995. Fiscal revenues and expenditure in the community: Granger-causality among fiscal variables in thirteen member states and

implications for fiscal adjustments. Economic Papers, vol. 114. European Commission, Brussels.Bleaney, M., Gemmell, N., Kneller, R., 2001. Testing the endogenous growth model: public expenditure, taxation, and growth over the long-run.Canadian Journal of Economics 34 (1), 3657.

Bond, S., Hoeffler, A., Temple, J., 2001. GMM estimation of empirical growth models. CEPR Working Paper, vol. 3048. Centre for Economic PolicyResearch, London.

Bouthevillain, C., Cour-Thimann, P., van den Dool, G., Hernndez de Cos, P., Langenus, G., Mohr, M., Momigliano, S., Tujula, M., 2001. Cyclicallyadjusted budget balances: an alternative approach. ECB Working Paper, vol. 77. European Central Bank, Frankfurt.

Bohn, H., 1991. Budget balance through revenue or spending adjustments? Some historical evidence for the United States. Journal of MonetaryEconomics 27, 333359.

Breitung, J., 2000. The local power of some unit root tests for panel data. In: Baltagi, B.H. (Ed.), Non-stationary panels. . Panel Cointegration andDynamic Panels Advances in Econometrics, vol. 15. Elsevier Science Inc., Amsterdam, pp. 161177.

Cass, D., 1965. Optimum growth in an aggregative model of capital accumulation. Review of Economic Studies 32, 233 240.De Haan, J., Sturm, J.-E., Volkerink, B., 2004. How to measure the tax burden on labour at the macro-level? In: Srensen, P.B. (Ed.), Measuring the

Tax Burden on Capital and Labour. MIT Press, Cambridge MA, pp. 289318.

De la Fuente, A., 1997. Fiscal policy and growth in the OECD. CEPR Discussion Paper, vol. 1755. Centre for Economic Policy Research, London.De la Fuente, A., Domnech, R., 2006. Human capital in growth regressions: how much difference does data quality make? Journal of the European

Economic Association 4, 136.Dickey, D., Fuller, W.A., 1979. Distribution of the estimates for autoregressive time series with a unit root. Journal of the American Statistical

Association 74, 427431.European Commission (various years). European Economy Public Finance Report, European Commission, Brussels.Evans, P., 1997. Government consumption and growth. Economic Inquiry 35, 209217.Gali, J., Perotti, R., 2003. Fiscal policy and monetary integration in Europe. Economic Policy 37, 243272.Im, K.S., Pesaran, M.H., Shin, Y., 2003. Testing for unit roots in heterogeneous panels. Journal of Econometrics 115, 53 74.Jones, C.I., 1995. Time series properties of endogenous growth models. Quarterly Journal of Economics 110, 495525.Karras, G., 1999. Taxes and growth: testing the neoclassical and endogenous growth models. Contemporary Economic Policy 17, 177188.Kneller, R., Bleaney, M.F., Gemmell, N., 1999. Fiscal policy and growth: evidence from OECD countries. Journal of Public Economics 74, 171190.Kocherlakota, N.R., Yi, K., 1997. Is there long-run growth? Evidence from the United States and the United Kingdom. Journal of Money, Credit and

Banking 29, 233

262.Levin, A., Lin, C.F., 1992. Unit root tests in panel data: asymptotic and finite sample properties. Discussion Paper. University of California, Berkeley.Lucas, R.E., 1988. On the mechanics of economic development. Journal of Monetary Economics 22, 342.Martnez-Mongay, C., 2000. ECFINs effective tax rates. Properties and comparisons with other tax indicators. Economic Papers, vol. 146. European

Commission, Brussels.Mendoza, E.G., Milesi-Ferretti, G., Asea, P., 1997. On the ineffectiveness of tax policy in altering long-run growth: Harberger's superneutrality

conjecture. Journal of Public Economics 66, 99126.Nijkamp, P., Poot, J., 2004. Meta-analysis of the effect of fiscal policies on long-run growth. European Journal of Political Economy 20, 91124.OConnell, S., Zeldes, S.P., 1988. Rational Ponzi games. International Economic Review 29, 431450.Pedroni, P., 1999. Critical values for cointegration tests in heterogeneous panels with multiple regressors. Oxford Bulletin of Economics and Statistics

61, 653678.Pedroni, P., 2000. Fully modified OLS for heterogeneous cointegrated panels. Non-Stationary Panels, Panel Cointegration and Dynamic Panels:

Advances in Econometrics, vol. 15. Elsevier, Amsterdam, pp. 93130.Pesaran, H.P., Smith, R.P., Im, K-S., 1996. Dynamic linear models for heterogeneous panels. In: Matyas, L., Sevestre, P. (Eds.), The Econometrics of

Panel Data: A Handbook of the Theory and Applications. Kluwer Academic Publishers, Dordrecht.Quintos, C.E., 1995. Sustainability if the deficit process with structural shifts. Journal of Business and Economic Statistics 13, 409417.Rebelo, S., 1991. Long-run policy analysis and long-run growth. Journal of Political Economy 99, 500 521.Romer, P., 1986. Increasing returns and long-run growth. Journal of Political Economy 94, 10021037.Romer, P., 1990. Endogenous technological change. Journal of Political Economy 98, S71S102.Santos Bravo, A., Silvestre, A.L., 2002. Intertemporal sustainability of fiscal policies: some tests for European countries. European Journal of

Political Economy 18, 517528.Solow, R.M., 1956. A contribution to the theory of economic growth. Quarterly Journal of Economics 70, 6594.Temple, J., 2003. The long-run implications of growth theories. Journal of Economic Surveys 17, 497510.Trehan, B., Walsh, C., 1988. Common trends, the government budget constraint, and the revenue smoothing. Journal of Economic Dynamics and

Control 12, 225444.Widmalm, F., 2001. Tax structure and growth: are some taxes better than others? Public Choice 107, 199219.


Documents

Romero-Avila and Strauch (2008)