Fiscal Federalism and the impact of Intergovernmental Grants: the European Regional policy in Spain

1 Title: Fiscal Federalism and the impact of Intergovernmental Grants. The European Regional policy in Spain Author: Juan González-Alegre Affiliation: Universitat Autònoma de Barcelona Contact: [email protected] Phone: (+34) 935811740 Fax: (+34) 935812292 Dpto. Economía Aplicada Universidad Autónoma de Barcelona Edifici B Campus de Bellaterra 08193 Cerdanyola (Barcelona) Spain 2 Abstract: We suspect that the efficiency of intergovernmental grants is related to the level of fiscal autonomy of the subsidized government. In this paper we construct and estimate a panel data model capturing the role of fiscal federalism on the effectiveness of EU Structural Actions in enhancing public expenditure in selected policy areas. We use data from the seventeen Spanish regions for the period 1993-2007. Results unambiguously support the hypothesis that the effectiveness of the ERDF decreases with larger fiscal autonomy. The role of the European Social Fund is still under analysis. These results could reflect the fact that fiscal decentralization in Spain has been focused to larger taxation autonomy without affecting regional income redistribution. JEL classification: H72, H77, C33, C23. Keywords: Fiscal Federalism; Intergovernmental Grants; Regional Policy; Panel Data; Acknowledgements I am grateful for comments and helpful suggestions at the Irish Society of New Economists meeting 2010, the XIV Encuentro de Economía Aplicada, the XXXVI Simposio de la Asociación Española de Economía and the ESF Exploratory Workshop on Fiscal Policy at the Regional Level and Intergovernmental Relations. Funding The author acknowledges financial support from the “Institut d´Economia de Barcelona” under its “Research grants on fiscal federalism” program. 3 1. INTRODUCTION The Cohesion Policy designed by the European Union has been contributing actively to the achievement of sustainable economic growth in European regions over the last decades. The recent political and economic developments in the EU may justify the revision of some of the principles driving the Cohesion Policy so as it can perform its duty with equal success in the coming years. One of the main challenges to tackle, which is already taking place, is the transition of the Cohesion Policy to the new European Union after the more recent enlargements of the Union, which have leaded to a larger and, in particular, more heterogeneous, field of application of the policy. The recent economic crisis, will, in addition, put more pressure on the consolidation of the public budget in all levels of the public administration. One of the aspects that must be put into consideration, and the issue covered in this paper, is the role that the different levels of fiscal decentralization achieved in every Member State have on the mechanisms ruling the Cohesion Policy. In particular, we will study the programs design under the Structural Actions1 that pursue the increase of public investment on key areas for growth. We will, therefore, focus our attention in these policies whose purpose is enhancing Public Investment, and will try to evaluate whether the level of fiscal decentralization of the member states play a role in their effectiveness. Both issues, Fiscal decentralization and EU intergovernmental grants, have been addressed separately in numerous empirical studies. In most of the cases the focus of the studies has been centered in estimating the effect of these policies on economic growth. Only very recently, some researchers have 1 In the nomenclature of the European Union, the term “Structural Funds” usually refer to the four Funds conforming the so-called Regional Policy (European Regional Development Fund, or ERDF; European Social Fund, or ESF; Financial Instruments for Fisheries Guidance, or FIFG; and the European Agricultural Guidance and Guarantee Fund, or EAGGF which has been replaced by the European Agricultural Guarantee Fund in 2007) while the “Structural Actions” include, in addition, the Cohesion Fund. In this paper, we will use both terms indistinctively. 4 put their attention on the impact on the distribution of public expenditures. But, to our knowledge, there is no previous work trying to address the importance of the simultaneous effect of both policies. Economic theory has also traditionally modeled the issues of fiscal decentralization and effectiveness of intergovernmental grants separately. Nevertheless, very recent developments of economic theory in the field of intergovernmental grants have identified the role of fiscal autonomy of granted government in the efficiency of the grants. Results, if not totally contradictory, are not coincident among the few studies. Volden (2007), for example, finds that the effect of grants depends on the capacity of the recipient government to efficiently raise taxes. Governments with greater tax-efficiency2 would experience higher crowding-out induced by the grant, meaning that the grant becomes less effective in enhancing public expenditure in a particular policy area3. Kappeller (2007) finds, instead, that the granted governments would under-invest when tax-autonomy is restricted, particularly in rich regions. In this case, the level of matching-grants is also suboptimal. Economic theory probably needs of further empirical studies identifying stylized facts over which build assumptions and develop richer models. But also the public administrations and the society in general, need of better instruments to judge the results of the several policies taken over. Based on the declared target that the Structural Actions –exclusive of the ESFare intended to promote Public Investment in key areas for growth, this paper tries to show that the effectiveness of these policies will depend on the level of fiscal decentralization of the country or region of application. Being this the case, the policy implication yield by this result would include taking into account the different levels of fiscal federalism achieved in the Member States in the rules governing the Structural Actions. The onesize-fits-all strategy, that has given reasonably good results in the past, may be improved in order to 2 Defining tax-efficiency as the capacity that the subsidized government has to efficiently raise taxes. One could think that this variable may be closely linked to the level of fiscal autonomy. 3 Gil-Serrate and López-Laborda (2005) link the causality in the other direction, stating that economies with a higher “flypaper effect” (expenditure response to an intergovernmental grant) would have a lower optimal level of tax-decentralization. 5 serve a larger and more heterogeneous European Union in a new scenario in which, most likely, taught constrains in the public budget are going to remain for years after the crisis is overcome. Spanish regions are, probably, the better example of the development on both policies over the past few years. Spain have, simultaneously, experienced an important decentralization process as well as benefited greatly of the Cohesion Policies run through the Structural Actions. Both processes have been asymmetric and independent: asymmetric because while fiscal federalism has affected differently in time and degree the several Spanish regions, the allocation of Structural Action shows also important differences across regions; and independent, because both policies are completely unrelated, since there is no economical, social or geographical aspects running the processes of decentralization. Therefore, the stronger effect of the Structural Actions devoted to poorer regions affect, equally, to regions with high or low level of fiscal autonomy. The paper proceeds as follows: Section 2 gives an overview of the main facts and figures describing fiscal decentralization and Structural Funds in Spain; Section 3 presents the theoretical framework that will help to interpret the results; Section 4 presents the data and variables; Section 5 describes the methodology and the results; and Section 6 concludes. 2. FISCAL DECENTRALIZATION AND REGIONAL POLICY IN SPAIN In this section, we introduce some figures that show how Spanish regions are a suitable illustration of the two policies under consideration (fiscal decentralization and EU cohesion policy), as they affect these regions with a relatively large degree of cross-sectional variability. The recent process of decentralization of public financing in Spain starts with the Spanish Constitution of 1978. The Constitution set the bases for the ulterior establishment of the seventeen regional bodies, defined as “Autonomous Communities”, which are the main beneficiaries of the decentralization process. The Constitution states that the level of competencies assumed by each regional government 6 and the pace at which these competencies are assumed is not homogeneous among all regions. The constitution of the regional governments finished in 19834. Figure [1] about here Simultaneously to the process of political adaptation to the new Constitution, occurred the most important increase of public spending. Total public spending moved from representing less than thirty percent of GDP in the late seventies to lay around fifty percent in the last years. Figure [1] shows how the main beneficiary of the decentralization in the last years has been the regional sector. Local public expenditure has only increased its share over total expenditure two percentage points in thirteen years, while the regional level has increased to over 30% of total public expenditure in 2008, compared to 1995 when it represented around 17 %. Figure [2] shows that the process of decentralization that Spain has experienced is not a general pattern of behavior of the countries on its economic environment. Figure [2] about here The Spanish Constitution discriminates between two types of regions: the so-called "historic nationalities" or regions with a high level of competencies5 and the ten remaining regions6 (and the two autonomous cities) that in principle assume a lower level of competencies. In practice, the regions with high levels of competencies experienced a higher level of fiscal autonomy in the beginning, but the gap between both types of regions have been reduced as long as the decentralization process has been taking place. We can observe this phenomenon if we build a ratio of fiscal decentralization as the coefficient between per capita expenditure at the regional level to the per capita expenditure at the central level, which is shown in Figure [3]: 4 Although later, in 1995, were constituted the Statutes of the two Autonomous cities, Ceuta and Melilla. These have been excluded from our analysis due to data availability. 5 Andalusia, Basque Country, Canary Islands, Catalonia, Galicia, Navarre and Comunidad Valenciana. 6Aragon, Asturias, Balearic Islands, Cantabria, Castile La-Mancha, Castile and Leon, Comunidad de Madrid, Extremadura, Murcia and La Rioja. 7 Figure [3] about here A deeper analysis of the functional categories7 reveals that the category "Social Public Goods" -using the nomenclature of the functional classification used by the Spanish "Ministerio de Economía y Hacienda"- is the main area of decentralization for the regions with a low level of competencies as well as the main component of the public budget. Other functional categories that have experienced a significant level of decentralization have been "Social Security and Promotion", "Economic regulation of Productive Sectors", "General Public Services" and "Economic Public Goods" While the level of fiscal autonomy is almost identical among the regions with low level of competencies, fiscal competencies among the regions with high level of competencies is more heterogeneous8. This is also reflected in the distribution of the revenue-side of the budget shown in Figure 4. In particular, two of the regions (Basque Country and Navarre) have particular privileges about the collection of taxes in their territory. Spanish regional financing scheme has gone through several revisions over time (See Lopez-Laborda, 2006; De la Fuente, 2010). For the time window analyzed in this paper, there have been three different systems into force (1991-1996; 1997-2001; and 2002-2008). These revisions have established, subsequently, a larger share of tax revenues in the budget of regional governments as well as larger leeway to decide on the level of taxation, and normative capacity to establish of abolish certain taxes. The redistributive mechanism, however, has not been modified until the revision implemented on 2009 (See Bassols et al., 2010) which is outside the period under consideration in this paper. [Figure 4 about here] Both groups of regions are not representing either geographical concentration or economic characteristics, meaning that there is no other common denominator between regions with high level of 7 See González-Alegre (2010) 8 See Molero (2001) 8 autonomy or between regions with low level of autonomy other than their political status. In both groups of regions there are objective 1 regions (which are eligible for most of the Structural Funds) and regions with per capita income larger than the national and European averages. Figure [5] has the purpose of showing that there is no systematic difference in the amount of Structural Funds from the EU that both groups of regions receive. Figure [5] about here The increase in the size regional governments has also affected the distribution of public regional spending among the different economic categories. The regions have augmented the share of current spending, devoting a minor part of their funds to increasing their stock of capital (Figure [6]). One might think that this situation could be induced by a certain reallocation of competencies between the central and regional governments. However, the Central Government has not increased its share of capital expenditure, but has, on the contrary, slightly decreased it. The fall in the capital share of public expenditure is clearly more relevant in the regions with low levels of competencies, which are also those that have undergone a more profound process of decentralization. Figure [6] about here 3. MODELLING FISCAL DECENTRALIZATION This section shows how theoretical predictions about the impact of intergovernmental grants on public administrations gaining fiscal autonomy within a federation, depend largely on the assumptions used to model fiscal autonomy. As mentioned before, the existence of theoretical models that combine fiscal decentralization and intergovernmental grants is relatively limited. 9 Nevertheless, the literature about the conditions that make fiscal decentralization desirable is extremely prolific9. Arguments in favour of fiscal decentralization rely largely on preference heterogeneity for public goods provision among regions (Besley and Coate, 2003; Brueckner, 2005; RubinchickPessach, 2005) and its impact on multifactor productivity (Martínez-Vazquez and McNab, 2006) while arguments against are usually based on internalization of spillover effects among regions (Chu and Yang, 2012) , strategic behaviour towards redistribution (Oates, 2005) or fiscal competition (LeiteMonteiro and Sato, 2003; Hatfiel and Padro, 2008). For the sake of this paper, however, it is important to distinguish between fiscal autonomy based on the capacity to decide on the distribution of the public budget –both in the expenditures and the revenues side of the budgetand fiscal autonomy affecting the mechanisms of regional redistribution of income. Based on one of the sub-games10 included in Volden (2007), let us assume a sub-national government which acts as a representative agent of its constituents and that benefits from providing an investment good and keeping taxes low, so that: Us = (yi)fc,s – (ts + tn) f b,s - d - xs The utility of the sub-national government depends positively on income, which is related to the public investment good according to the production function yi=iqi. The function fc,s represents the fraction of credit that the sub-national government obtains from the provision of the public investment good; ts and tn represent taxes issued by the sub-national and national governments respectively, and fb,s represents the fraction of blame that the sub-national government obtains from taxation. Finally, the last term captures the disutility associated from taking a policy direction which deviates, in a one dimensional line, from the optimal direction preferred by the representative agents. 9 Being Oates (1972) considered as the blast-off that stimulated most subsequent research. An intuitive review of the evolution of the literature may be found in Weingast (2009) 10 Introducing two main innovations to the model: the consideration of a public investment good instead of a consumption good an the introduction of a redistributive mechanism among sub-national governments. 16 In order to test the hypothesis that public investment may be affected by European Structural Funds ' grants, we have constructed a model in which the dependent variable is Public Investment at the regional level for the seventeen Spanish regional bodies. The set of explanatory variables includes our main variable of interest, EUSF, represent the capital transfers from the EU to the regional government allocated to the region "i" in the current year "t". We have also introduced in the model other control variables: private investment, public consumption, GDP growth, population growth, and central government investment, included in the vector x: i,t i,t i,t ,PubInv = eusf + x + i i t      (1) Where  is the coefficient that describes the impact of Structural Funds on Public Investment and the main target of our estimation; x is a vector, (1x5), of explanatory variables and  is the set of parameters, (5x1) associated to these control variables that must be estimated; i  is the unobservable unit-specific effect and ,it  is the unobservable error term. In order to estimate equation (1), we have split the sample attending to the level of fiscal autonomy of the regions. We have taken two alternative criteria into consideration in order to consider sub-samples: firstly, we have classified Spanish regions into two subgroups according to the level of fiscal autonomy that the Spanish Constitution recognizes them. Therefore, we create a group of what the Spanish Constitution considers16 “Historic Nationalities”, and a second group of the remaining ten regions17, for which the Constitutions recognized a lower level of Autonomy. And secondly, as a robustness check, we have also considered the time-dimension of the series in order to identify two alternative subgroups with remarkable differences in their level of fiscal autonomy. We have selected the year 2000 as the break point, which will leave us two subsamples of similar length. 16 Andalusia, Basque Country, Canary Islands, Catalonia, Galicia, Navarre and Valencian Community. 17 Aragon, Asturias, Balearic Islands, Cantabria, Castile and Leon, Castile-La Mancha, Extremadura Comunidad de Madrid, Murcia, La Rioja. 17 The evolution of fiscal autonomy across time in both subsamples is quite stable in both groups of regions18 although with a remarkable gap between regions with low and high level of competencies. The use of two alternative criteria to divide the sample will let us overcome some of the shortcomings which are attached to each criterion. On the one hand, splitting the sample according to the role recognized in the Constitution may arise the doubt that we may be accounting for a systematic difference between both groups of regions that may not come from the level of fiscal autonomy but from an ignored source19. On the other hand, breaking the sample into two time-periods may be interpreted as the identification of some structural change across time. Primary estimations of equation (1) suggest the presence of autocorrelated errors. Therefore, the original model in equation has been estimated in the presence of serially correlated errors20. Initially, we also assume strict exogeneity of the explanatory variables ( ,,[ , ] 0i s i tEx   ; t,s=1,2,...T). This assumption may be considered too strong for our model. Many results21 show that the allocation of public expenditure may be endogenous to the allocation of grants. The distribution of the Structural Funds may be thought to respond to some unobserved necessities and conjuncture that simultaneously drives decisions on public investment. We must admit the possibility that some of the explanatory variables, in particular eusf, must be correlated to the error term since the propensity to increase public investment may incentive larger allocation of Structural Funds (thus, making causality run in the opposite direction to the one assumed in the paper). The immediate solution to the problem could be to find some instrumental variables correlated to structural funds but orthogonal to public investment. Alternatively, we can use lags of the dependent and explanatory variables as instruments. The GMM estimation method developed by Arellano and 18 See González-Alegre (2008) 19 Despite the fact that there are no remarkable differences in the level of economic development among both groups of regions. Neither there are geographical, commercial or cultural differences among them. 20 Preliminary estimations suggest also the use of fixed-effects models. The results of the random effects estimations, as well as the Hausman test are omitted for the sake of brevity. 21 Knight, 2002; Becker, 1996; Besley and Case, 2000. 18 Bond (1991) relies on the orthogonality of the dependent and explanatory variables with the first differences of the error component in lagged periods. This method allows us to include endogenous and predetermined dependent variables. These GMM methods construct moment conditions that reflect this orthogonality, under assumption of serially uncorrelated shocks, error components and predetermined initial conditions22. The problem would be, therefore, that we have previously admitted the possibility of the existence or AR(1) errors in the original model, which implies that lagged values of the dependent and explanatory variables are correlated with past shocks and the moment conditions that should be used23, are no longer valid in the original model For that reason, we transform the static model into a dynamic one with serially uncorrelated shocks by subtracting the autocorrelation term attached to the original errors: i,t i,t i,t ,PubInv = eusf + x + i i t      where , , ,1i t i t i te e u     i,t i,t i,t i,t i,t i,t ,PubInv = *PubInv + eusf - eusf + x - x +(1- ) i i tu         (2) Equation (2) represents a model with serially uncorrelated shocks that we can estimate using Arellano and Bond (1991) GMM estimator for dynamic panels. The explanatory variables are correlated with the individual effects and are assumed to be endogenous with respect to the serially uncorrelated shocks. Estimation Results. Table 3 shows the results of estimating equations (1) and (2) when we divide our sample according to the level of autonomy recognized for the regions in the Spanish Constitution. Columns [1] to [4] include the estimation for the regions with a lower level of autonomy while 22 ,,[ ] [ ] [ ] 0i i t i i tE E E        ; ,,[ ] 0i s i tE   for ts and ,1 ,[ ] 0i i tE PubInv   t=2,...T respectively. 23 ,,[ ] 0i t s i tE PubInv   for t=3,…T and 2s ; ,,[ ] 0i t s i tEx   , for t=3,…T and 2s if variables in x are endogenous 19 columns [5] to [8] include the estimations for the remaining seven regions with a larger level of fiscal autonomy. Columns [1]-[2] and [5]-[8] assume a fixed-effects24 model with autocorrelated errors, while [3]-[4] and [7]-[8] are estimates for equation (2) obtaining assuming engogeneity of explanatory variables using one-step version of the GMM estimator developed by Arellano and Bond (1991). In addition, we assume two sets of control variables, one more general and one more restrained. Results are quite homogeneous among models for every set of regions. The Structural Actions (eusf) seem to be a significant determinant of Public Investment in the regions with low level of competencies, being the coefficient estimated significantly positive 0 and smaller than one. However, for the regions with a high level of competencies, the coefficients estimated are smaller and generally insignificantly different from zero. As for the remaining control variables, the main source of variability between both data-sets in the coefficient attached to Public Consumption that show a behaviour quite similar to the one described for eusf. Public consumption will capture the effects of the size of the regional administration. It is expected to increase with larger fiscal autonomy and, therefore, induce further increases also in Public Investment. Private investment is a positive determinant of Public Investment in all cases while the remaining control variables do not seem to play a key role. Having estimated a different effect of the Structural Actions on Public Investment for the two groups of regions, we also make a second estimation by splitting the sample through the time dimension. If we examine Figure [3], we can see how the level of fiscal autonomy of both groups of regions has increased over time. By splitting the sample around year 2000, the level of fiscal autonomy remains relatively stable for both groups of regions across time25, keeping a significant difference among them. The results of the equivalent estimations are shown in Table [4]. We have estimated an impact of the 24 The selection of the fixed-effects model has been made upon estimation of the equivalent random-effects model and the corresponding Hausman (1979) test. Acordingly, the autocorrelated errors have been included upon estimation of preliminary models. 25 See González-Alegre (2010) 20 EUSF on Public Investment larger and significantly positive for the period 1993-1999, while the estimates for the period 2000-2007 show poor levels of significance. Regarding Public Consumptions, the differences observed in the previous estimation remain but are less strong. The behavior of the other control variables remains stable. Interaction Term In order to take into account for the effect of the evolution Fiscal Decentralization on the relationship between the Structural Actions and Public investment, we will make use of an Interaction term. Interaction terms may be added to a model in order to incorporate the joint effect of two variables on a dependent variable, over and above their separate effects. These are usually added as the cross-product of two independent variables, typically placing them after the simple "main effects". In this subchapter, we will analyze the interaction of fiscal decentralization (represented by the variable “dec”) and the capital transfers received by regional governments (represented by “eusf”). The separate effect of both variables are expected to be positive, since an increase in the level of fiscal decentralization (measured as the ratio of per capita regional over national public expenditures) is assumed to increase the size of regional governments and, therefore, increase on public expenditures – compressive of public investment-. The effect of the capital transfers through the Structural Actions (eusf) would follow the arguments examined in the previous subchapter. i,t (1) i,t (2) (3) i,t ,PubInv = eusf + dec+ eusf*dec+ x + i i t        (3) The interaction term would capture, therefore, the joint effect of these two variables. We can see in table 5 the results of estimating the model represented by equation (3) for the whole sample. We have expanded our set of alternative control variables, since we expected that the correlation between “Public Consumption” and “dec” might be problematic. The results, however, look quite robust with 21 respect to this issue. As for the estimation assumptions and methodology, we have followed similar guidelines as tables 3-4. We have estimated a negative coefficient attached to the interaction term in all cases. The level of significance, however, is variable and seems to depend on the set of controls. The negative coefficient means that the join effect of additional decentralization and public investment becomes weaker. If we assume a fixed level of decentralization, for example, additional EUSF will induce an effect on Public investment equal to the coefficient estimated for EUSF plus the coefficient estimated for the interaction term multiplied by the value of decentralization. Given that the coefficient estimated for the interaction term is negative, the effect of EUSF on public investment is positive, but decreasing for larger decentralization. Cross-product interaction terms may be highly correlated with the corresponding simple independent variables in the regression equation, creating problems with assessing the relative importance of main effects and interaction effects. Because of this, sometimes it may well be desirable to use centered variables (where one has subtracted the mean from each datum). This transformation often reduces multicollinearity. For the sake of robustness, we have also run equivalent estimation using a centered interaction term and the results do not change significantly (see table 6) Simultaneous Equation model We want to check the robustness of our result to the introduction of a simultaneous equation model (SEM), in which we capture causality in both directions. One could think that the two variables in which we focus our interest: eusf and public investment, are jointly determined by a system of equations. In fact, the political decision of investing is closely related to the political decision of allocating –or making use ofthe Structural Funds. Also the economic realization of the payments is closely related, given that both variables are often related to common investment projects. In addition, each one of the variables may be a determinant of the other one. So far we have considered that the 22 allocation of Structural Funds may encourage Public Investment, but me must be aware that the propensity to invest in the public sector may also incentive the allocation of Structural Funds in a particular region. The system consists of two structural equations, one in which the dependent variable is Public Investment, while in the other the capital transfers allocated through the Structural Funds. Each of the equations includes one of the variables as dependent variable but also the other one as an explanatory – endogenousvariable. In addition to these, we also include a set of exogenous variables26: i,t (1) i,t (1) (1)i,t (1) (1) ,PubInv = eusf + x + i i t      (4) i,t (2) i,t (2) (2)i,t (2) (2) ,eusf = PubInv + x + i i t      (5) Where (1) x and (2) x are two vectors, (1xm) and (1xn) respectively, of exogenous explanatory variables. Both vectors are not identical, but they can share some variables. (1)  and (2)  are the set of parameters, (mx1) and (nx1) respectively, associated to the exogenous variables that must be estimated. (1)i  and (2)i  are the unobservable unit-specific effects and (1) ,it  and (2) ,it  are the unobservable error terms. We estimate the model above following different alternative estimation methods in order to check for the robustness of the results27. First of all, we assume that the source of endogeneity is present through a positive correlation between the endogenous variables and the error term (1) ,it  . In this setup, the model may be estimates assuming Fixed-Effects through two-stage least-squares (FE-2SLS) and 26 One might be tempted to think that Public Current Expenditures should take part of the simultaneous equations model as an endogenous variable. As Wooldridge (2002) describes for an example relating hours devoted to crime with hours devoted to work, the choice of the share of the public budget devoted to current expenditures and to investment is the solution of the maximization problem of the utility function of the government and depends on exogenous factors –like the population, level of education, private investment, etc-. Of course, some endogeneity may arise when estimating the relations among both variables, but we consider that this possibility is more related to an omitted variables problem –or even to measurement errorrather than to simultaneity. The case for Public Investment and Capital Transfers (EUSF) is different since in this case both expenses are accrued simultaneously when referred to the same investment project. 27 We use limited information estimators, which means that every equation of the system is estimated at a time, in contrast to full-information systems, in which the estimators are based on the entire systems of equations. 23 assuming Random-Effects through the Error-Component two-stage least-squares estimator developed by Baltagi (1981)28. Results obtained using this estimation strategy are presented in table 8. Alternatively, we may assume that the source of endogeneity comes from the positive correlation between the idiosyncratic term and the endogenous variables. The explanatory variables are, then, orthogonal to the structural errors and the exogenous variables are, in addition, orthogonal to the idiosyncratic term,  . If we assume Fixed-Effects, the model can be estimated by OLS after the within transformation, as shown by Cornwell et al. (1992). For the cases in which the unit-specific effects are random, we make use of the Two-stage least-square Hausman and Taylor (1981) procedure (HT-2SLS) estimator29. The method of 2SLS is the most common method used for estimating simultaneous-equations models, because of their simplicity and asymptotic efficiency. In this case, we include also additional variables on equations (6) and (7), (1) z and (2) z respectively, which are two vectors of time-invariant explanatories, including both endogenous and exogenous variables: i,t (1) i,t (1) (1)i,t (1) (1)i (1) (1) ,PubInv = eusf + x + z + i i t       (6) i,t (2) i,t (2) (2)i,t (2) (2)i (2) (2) ,eusf = PubInv + x + z + i i t       (7) Usually, as we are not interested in their effect, time-invariant variables are omitted since their effect may be captured by the idiosyncratic-term. However, for the HT 2SLS estimator, they are used as instruments to estimate the system, so it may be useful to include them. We describe in table 7 the time invariant variables included in the HT 2SLS regression. These are, basically, determinants of the Investment needs and economic performance at the beginning of the sample and its selection has been made upon consultation of several studies addressing public investment30. Results upon the assumption that the endogenous variables are correlated with the unit-specific term are shown in table 9. 28 See Baltagi (2005) for details on this estimator. 29 There are alternative procedures to the HT, for example the Amemiya and Mc Curdi (1986) or the Breusch et al. (1989), which make use of additional instruments but at the cost of additional assumptions about the exogeneity of the explanatory variables and all their future and past values. See Cornwell et al. (1992) for a detailed description of the different estimators and their properties. 30 With a particular attention to Mitze (2007), since he uses also this simultaneous equation estimator 24 Both, tables 8 and 9, show similar results with respect to most of the variables under consideration. The results previously observed with respect to the impact of the Structural Funds on Public investment are reinforced in after this estimation, although it must be stressed that the option in which we assume fixed-effect and orthogonality of the endogenous variables with the error term (table 9) yields poor significant coefficients. Public Investment, simultaneously, seems to be a key determinant of the volume of Structural Funds allocated to each region in each period, although the coefficients attached to this direction of the causality are significantly smaller than those from equation (4) and (6). Decentralization is a positive determinant of public investment. This result was know from the previous estimations of the paper and it was also expected, since the variable fiscal decentralization is an indicator of the size of the regional government31 and, therefore, of its expenditure power. Nevertheless, we find a significant negative coefficient when estimating the impact of fiscal decentralization on the Structural funds (equations (5) and (7)). At a first glance, one might be tempted to think that after increasing the level of fiscal autonomy of a region, Public investment may be spread over more heterogeneous policy areas. This expansion may be attached to competencies that are not eligible for the Structural Funds, reducing, therefore, the possibility of the government to maintain the relationship between Structural Funds and public investment. The coefficient estimated for the exogenous variables are, in general, expected and consistent across models. Among them, the level of significance of “population growth” as a negative determinant of public investment becomes relevant with respect to previous subchapters of this paper. 3131 As well as the variable “Public Consumption. In fact, one might expect a significant level of colinearity between both variables, which would justify the use of the alternative models estimated, as introduced before. 25 Tables 10 and 11 replicate the SEM estimation after splitting the sample into the group of regions with low level of competencies and the regions with high level of competencies. Table 10 presents the results assuming that the source of the endogeneity is the correlation of the variables with the unit-specific effect, while table 11 assumes correlation with the error term. We have estimated only the reduced versions of the models assuming both, random and fixed-effects. The estimations have to be taken cautiously since the number of observations is a bit limited. In general, the coefficients estimated for the regions with low level of autonomy are larger in absolute value and level of significance for the variables of our interest which is in line with our previous results. Public investment, as a determinant of EUSF is stronger also in the regions with low level of autonomy, while in the regions with high level of autonomy, EUSF seem to depend very few of the propensity of the government to invest. Finally, also the level of fiscal decentralization as a determinant of EUSF seems to be more – negativelyimportant in regions with low level of autonomy. That is somehow an expected result since these regions have experienced the larger decentralization process and, in any case, confirms our previous suspicious that by gaining fiscal autonomy regions find it more difficult to be eligible for additional grants. 6. CONCLUSIONS The impact and efficiency of the Structural Actions carried over by the European Union in order to enhance sustainable development in European Regions may depend of the level of fiscal federalism of the Member States. In this paper, we address the particular case of the Structural Actions designed to enhance Public Investment in key areas for growth. Spain has experienced a process of fiscal decentralization in the recent years and, simultaneously, has been recipient of an important share of the Structural Actions. Due to the heterogeneous level of economic development and also to the diverse political status of Spanish regions, both policies have affected these regions in an asymmetric way. These conditions make Spanish regions the perfect 32 Wooldridge, JM (2002) Econometric analysis of cross section and panel data. MIT press, Cambridge, MA Zhang, T, Zou, HF (1998) Fiscal decentralization, public spending and economic growth in China. Journal of Public Economics 67: 221-240 Appendix. Table A1. Fiscal Decentralization and European Cohesion Policy: Previous Empirical Studies with Spanish regional data. 33 Autor/s (year) Main Issue Data coverage Methodology Main Results Alvárez et al. (2000) The impact of fiscal decentralization on the size of the public sector Spanish public sector at the regional level, 1993 Estimation of a model for crosssectional data Fiscal decentralization has a negative impact on the size of the public sector Molero (2002) Public Spending and Fiscal Federalism in Spain Spanish Public Administrat ions, 1988 1998 Descriptive Statistics Fiscal decentralization is more related to Public Expenditure related to economic intervention in regions with low level of competencies, while in regions with a high level of competencies is more related to Redistribution. De la Fuente (2002) Impact of EU Cohesion Policy on Spanish Objective 1 Regions Spanish Regionlevel data, 1994-2006 Panel-data estimation of growth model, and calibration of the impact. The EU Funds add one percentage point to annual output growth in and 0.4 percentage points to employment growth. Pardo Garcia (2003) European Cohesion Policy in Spanish Regions Spanish Regions, 1988-1999 Descriptive analysis of the Community Support Framework The weakest regions have improved their infrastructures, but there are many differences about their innovation capacity, knowledge access, information, and the training of human resources. Farrell (2004) Effect of European Cohesion Policy on Spanish and Irish Economies. Nationallevel data, 1990-2000 Descriptive Statistics EUSF promoted economic growth, but more efficiently in Ireland. In Spain, regional disparities actually increased. Part of the explanation lies with the institutional differences and policy decisions taken in each Public Administration. Sosvilla Rivero (2005) Impact of EUSF on growth and employment Spanish Objective 1 Regions (1989 2006) Adaptation of the HERMIN model to the Spanish regions (demand and supply effects of the EUSF). Average increase of 0.56 percentage points in the growth rates . Average increase in per capita income of 425 euros at 1999 prices. Increase of 1.46 per cent in employment. Perez González and Cantarero Prieto (2006) Fiscal decentralization and Economic Growth Spanish Regional Data, (19862001) Panel data model. Fixed-Effects and Instrumental Variables. The impact of fiscal decentralization on economic growth is insignificant. Gil-Serrate and LópezLaborda (2005) Taxdecentralization and economic growth Spanish national and regional data, 19801997 Calibration of growth model that accounts for tax decentralization An increase in the level of tax decentralisation in the Spanish economy compared to the level existing in the taken period would result in economic growth. Carrion-iContribution of Spanish Panel Data model For the regions with higher level of 34 Silvestre, Espasa and Mora (2008) fiscal decentralization to economic growth Regional Data, 19642000 estimated by GMM competencies, fiscal decentralization has positive and significant effects on economic growth, but fiscal decentralization has negative effects on the regions with lower level of competencies. GonzálezAlegre (2012) Effectiveness of EU Structural Actions; 1993-2005 EU15 and Spanish regional data Panel Data model estimated by GMM Public investment in the member countries makes up around 60% of the increase in EU funds. Figures 35 FIGURE 1: Shares of Public Expenditure by level of administration Source: Eurostat, Government Finance Statistics FIGURE 2: Ratio of state and local public expenditure to general government expenditure. 1995 2007. 0 0.1 0.2 0.3 0.4 0.5 0.6 Austria France Germany Greece Italy Portugal Spain UK 1995 1999 2003 2007 Source: Eurostat, Government Finance Statistics. 36 FIGURE 3: Decentralization Ratio. Ratio of per capita public expenditure of the regional government to the per capita public expenditure of the central government (excluding social security). Source: Badespe and “Liquidación del presupuesto de las CC AA” FIGURE 4: Sources of Revenues as percentage of GDP 37 FIGURE 5: Capital transfers from the EU to the Spanish regional governments. (% GDP and Euro per capita) Source: “Liquidación del presupuesto de las CCAA” FIGURE 6: Ratio Capital to Total Expenditure Source: “Liquidación del presupuesto de las CCAA” 38 Tables TABLE 1: Variables and sources of data Variable Label Definition Units Source Public Investment PubInv Gross fixed capital formation in the Regional Government %GDP Badespe database (Instituto de Estudios Fiscales, Ministry of Economy) EU Structural Funds EUSF EU expenditure executed corresponding to Structural funds, by Member State. %GDP Liquidación de Presupuestos de las CC AA, Ministry of Economy Public Consumption PCons Public Current expenditure in the Regional Government %GDP Badespe database Private Investment PrivInv Investment of tangible and intangible assets in the private sector %GDP IVIE (Valencian Institute of Economic Research) Central Government Investmeng CGInv Public investment from the central government disaggregated at the regional level %GDP 1984-1999 IVIE 2000-2007 PGE (General Public Budget) GDP growth GDPgr Real GDP growth Growt h rate INE (National Statistical Office) Population growth Popgr Population in miles persons Growt h rate Eurostat Fiscal Decentralization DEC Ratio of per capital public expenditure of the regional government to per capital public expenditure of the central government Ratio Badespe database (Eurostat for population) TABLE 2: Summary Statistics Variable Obs Mean Std. Dev. Min Max PubInv 255 0.0285 0.012 0.0054 0.0586 EUSF 255 0.0054 0.005 0.0000 0.0307 PCons 255 0.1011 0.048 0.0144 0.2213 PrivInv 255 0.2335 0.043 0.1404 0.3702 CGInvest 255 0.0180 0.010 0.0017 0.0681 GDPgr 255 0.0735 0.027 0.0152 0.2088 POPgr 255 0.0083 0.010 -0.0046 0.0383 DEC 255 0.5700 0.295 0.1055 1.6255 DEC*EUSF 255 0.0030 0.003 0.0000 0.0164 39 TABLE 3: The impact of Structural Actions on Regional Public Investment. Regions with different levels of Autonomy [1] [2] [3] [4] [5] [6] [7] [8] Regions with low level of competencies (art. 151) Regions with high level of competencies (art.143) F-E F-E GMM-AB GMM-AB F-E F-E GMM-AB GMM-AB PubInv (t-1) 0.5081*** 0.4679*** 0.6888*** 0.7257*** 0.080 0.077 0.080 0.071 eusf 0.5913*** 0.5981*** 0.6366*** 0.5858*** 0.0176 0.0092 0.3573* 0.2690 0.143 0.142 0.156 0.152 0.184 0.178 0.217 0.219 PubCons 0.0647*** 0.0526** 0.0732** 0.0693*** 0.0451 0.0533 0.1075 0.1102* 0.023 0.021 0.028 0.027 0.054 0.048 0.071 0.061 PrivInv 0.0820*** 0.0782*** 0.0847*** 0.0815*** 0.0656*** 0.0624*** 0.0611*** 0.0623*** 0.020 0.020 0.025 0.025 0.019 0.018 0.022 0.022 CGInvest -0.0322 0.0280 -0.1112 -0.0215 0.066 0.074 0.124 0.131 GDPgr -0.0131 -0.0270* -0.0031 -0.0121 0.012 0.016 0.009 0.013 POPgr -0.1418 -0.1138 0.0235 0.0285 0.120 0.140 0.113 0.112 F test group 4.33 (.001) 7.26 (.000) 6.15 (.000) 6.78 (.000) R2 within 0.321 0.3092 0.152 0.1502 Autocorr. Test D-W = .7933 D-W = .7951 AB(1) -2.67 (.00) AB(1) -2.65 (.00) D-W = .6732 D-W = .5007 AB(1) -1.97 (.04) AB(1) -1.96 (.04) B-W = .9651 B-W = .9639 AB(2) 1.18 (.23) AB(2) 1.38 (.16) B-W= 1.0279 B-W= .8483 AB(2) 0.76 (.44) AB(2) 0.17 (.86) Sargan test 94.743 97.339 75.742 78.676 stat 0.77 0.75 0.35 0.39 Obs (groups) 140 (10) 140 (10) 130 (10) 130 (10) 98 (7) 98 (7) 91 (7) 91 (7) *,**,*** denote significance levels at the 10%, 5% and 1% respectively Ab(order) denotes Arellano and Bond (1991) test for autocorrelation in the error term D-W: modified Durbin-Watson test for autocorrelated errors; B-W: Baltagi Wu LBI 40 TABLE 4: The impact of Structural Actions on Regional Public Investment. Time-Evolution [1] [2] [3] [4] [5] [6] [7] [8] 1993-1999 2000-2007 F-E F-E GMM-AB GMM-AB F-E F-E GMM-AB GMM-AB PubInv (t-1) 0.4551*** 0.3934*** 0.1643 0.0831955 0.108 0.096 0.115 0.107 eusf 0.5886*** 0.5884*** 0.7426*** 0.4373** 0.0851 0.1854 0.0316 0.0562522 0.165 0.161 0.237 0.222 0.220 0.215 0.236 0.244 PubCons 0.0929* 0.1021** 0.0907 0.1182* 0.0232 0.0338 0.1027** 0.0947564 0.047 0.044 0.068 0.069 0.035 0.034 0.047 0.037 PrivInv 0.0669** 0.0705*** 0.0467 0.0628* 0.0623** 0.0619** 0.1199*** 0.1281358 0.028 0.025 0.037 0.032 0.024 0.024 0.041 0.037 CGInvest -0.0148 0.1040 -0.1555* -0.00038 0.130 0.180 0.081 0.123 GDPgr -0.0072 -0.0349*** -0.0051 0.0166 0.008 0.013 0.040 0.048 POPgr 0.0167 -0.4982 -0.1063 -0.1841 0.252 0.463 0.120 0.176 F test group 3.52 (.000) 3.73 (.000) 9.79 (.000) 14.21 (.000) R2 within 0.3059 0.2973 0.142 0.0997 Autocorr. Test D-W = .9509 D-W = .8448 AB(1)-2.26 (.02) AB(1)-2.04 (.04) D-W = 1.227 D-W = 1.175 AB(1)-2.20 (.02) AB(1)-2.38 (.01) B-W = 1.350 B-W = 1.264 AB(2) -.065 (.94) AB(2) -.491 (.62) B-W = 1.553 B-W = 1.491 AB(2) .566 (.57) AB(2) .851 (.39) Sargan test 27.128 31.83 61.767 64.890 stat 0.98 0.74 -0.48 0.16 Obs (groups) 102 (17) 102 (17) 85 (17) 85 (17) 119 (17) 119 (17) 102 (17) 102 (17) *,**,*** denote significance levels at the 10%, 5% and 1% respectively Ab(order) denotes Arellano and Bond (1991) test for autocorrelation in the error term D-W: modified Durbin-Watson test for autocorrelated errors; B-W: Baltagi Wu LBI 41 TABLE 5: The impact of Structural Actions on Regional Public Investment. Interaction Term [1] [2] [3] [4] [5] [6] [7] [8] F-E F-E F-E F-E GMM-AB GMM-AB GMM-AB GMM-AB PubInv (t-1) 0.5881*** 0.5807*** 0.5366*** 0.5400*** 0.05 0.05 0.05 0.05 eusf 0.6482*** 0.6895*** 0.6477*** 0.6757*** 0.7860*** 0.7898*** 0.5452** 0.5262** 0.19 0.18 0.19 0.18 0.22 0.20 0.22 0.21 dec 0.0189** 0.0173*** 0.0164*** 0.0154*** 0.0222*** 0.0208*** 0.0169*** 0.0167*** 0.004 0.00 0.00 0.00 0.00 0.00 0.00 0.00 dec*eusf -0.5612* -0.6267* -0.5269 -0.5732* -0.7154* -0.7325** -0.5390 -0.4977 0.338 0.33 0.33 0.33 0.37 0.34 0.36 0.35 PrivInv 0.0462*** 0.0475*** 0.0444*** 0.0456*** 0.0492*** 0.0508*** 0.0594*** 0.02 0.02 0.01 0.02 0.02 0.02 0.02 PubCons -0.0198 -0.0134 -0.0169 0.00073 0.0608*** 0.03 0.03 0.03 0.03 0.02 CGInvest -0.0177 -0.0203 0.0432 0.0427 0.05 0.05 0.06 0.06 GDPgr -0.0129* -0.0119* -0.0263*** -0.0247*** 0.01 0.01 0.01 0.01 POPgr -0.1606* -0.1629** -0.1253 -0.1273 0.08 0.08 0.09 0.09 F test group 5.93 (.000) 5.76 (.000) 7.80 (.000) 7.54 (.000) R2 within 0.3014 0.297 0.2743 0.2699 AR Test D-W = .654 D-W = .618 D-W = .652 D-W = .612 AB(1) -3.13 (.00) AB(1)-3.10 (.00) AB(1)-3.05 (.00) AB(1)-3.02 (.00) B-W = .896 B-W = .860 B-W = .902 B-W = .866 AB(2) .858 (.39) AB(2) .891 (.37) AB(2) .526 (.59) AB(2) .519 (.60) Sargan test 173.888 174.296 168.892 169.070 stat 0.63 0.65 0.71 0.67 Obs (groups) 238 (17) 238 (17) 238 (17) 238 (17) 221 (17) 221 (17) 221 (17) 221 (17) *,**,*** denote significance levels at the 10%, 5% and 1% respectively Ab(order) denotes Arellano and Bond (1991) test for autocorrelation in the error term D-W: modified Durbin-Watson test for autocorrelated errors; B-W: Baltagi Wu LBI