Bailouts in a fiscal federal system: Evidence from Spain

European Journal of Political Economy 27 (2011) 154–170

Contents lists available at ScienceDirect

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

Bailouts in a fiscal federal system: Evidence from Spain Pilar Sorribas-Navarro ⁎ Facultat d’Economia i Empresa, Universitat de Barcelona, Dept. Economia Política i Hisenda Pública Avda. Diagonal 690, torre 4, planta 2ona, 08034 Barcelona, Spain Institut d'Economia de Barcelona, Spain

a r t i c l e

i n f o

Article history: Received 22 October 2009 Received in revised form 7 June 2010 Accepted 11 June 2010 Available online 19 June 2010 JEL classification: H1 H77 O4

a b s t r a c t This paper investigates bailouts in the fiscal federal system of Spain. An equation is estimated in which central government grants to regions depend on the regions' issue of debt. The data is for the period 1986–2006. The results reveal partial bailouts. This holds for discretionary and inprinciple non-discretionary grants. Bailouts are greater or more likely when there are limits on the borrowing autonomy of a region, when the region is responsible for providing health care; and when the region has a high proportion of swing voters). © 2010 Elsevier B.V. All rights reserved.

Keywords: Intergovernmental grants Bailouts Fiscal federalism

1. Introduction In many decentralized countries, sub-central governments are heavily indebted. These include regional governments in Brazil, Italy and Spain. Argentina has been a salient case (Rezk, 1998). The lack of fiscal discipline of sub-central governments is a major source of macroeconomic instability (Ahmad et al., 2006). The expectations of sub-central governments that they will receive additional resources if necessary can weaken fiscal discipline and can result in strategic use of debt. There is therefore moral hazard. A sub-central government has a soft budget constraint when it has a budgetary deficit and issue debts in the expectation of receiving additional resources from the central government (Kornai et al., 2003). The additional resources are termed a bailout (Wildasin, 1997). The bailout can be explicit or implicit. In an explicit bailout, the central government provides a supplementary grant to the sub-central government.11 Such bailouts may come with or without conditionality. In an implicit bailout, the central government provides an additional amount from existing grants to a sub-central government due to debt. In this paper, I empirically investigate implicit bailouts of Spanish regional governments by the central government. Theoretical analyses (e.g., Goodspeed, 2002; Inman, 2003; Schaltegger and Feld, 2009) and the evidence from a number of case studies, including the USA (Inman, 2003), Canada (Bird and Tassonyi, 2003) and Hungary (Wetzel and Papp, 2003), indicate that fiscal decentralization per se does not necessarily weaken fiscal discipline of sub-central governments. The effect on fiscal discipline depends on whether the institutional setting provides expectations for sub-central governments that they will be bailed out (Rodden et al., 2003).

⁎ Facultat d’Economia i Empresa, Universitat de Barcelona, Dept. Economia Política i Hisenda Pública Avda. Diagonal 690, torre 4, planta 2ona, 08034 Barcelona, Spain. Tel.: + 34 93 403 90 12; fax: + 34 93 402 18 13. E-mail address: [email protected]. 1 The central government might also assume some of the sub-central governments' liabilities or a share of their debt. 0176-2680/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.ejpoleco.2010.06.001

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

155

A bailout requires some discretion by the central government in the allocation of resources among sub-central governments (Rodden, 2001; Oates, 2005). The literature emphasizes the fact that, when regional governments are responsible for providing a key public service, such as health care or education, it is difficult for the central government not to bail them out when they face financial difficulties (Rodden et al., 2003; Bordignon and Turati, 2009). Likewise, an absence of sub-central fiscal autonomy (over taxes, expenditure or borrowing) can also induce the central government to bail out regions, because the regions lack the instruments to make fiscal adjustments (von Hagen and Eichengreen, 1996). The number of people living in the region can affect the bailout decision. According to the “too big to fail” hypothesis (Wildasin, 1997), the larger is the population in a region, the greater is the probability of a bailout, because the negative externalities associated with financial difficulties affect a higher share of the population. The political characteristics of the regions can also affect the central government's bailout decision, as I shall show. A number of empirical studies have examined the effects of bailouts on sub-central fiscal discipline.2 Pettersson-Lidbom (2010) estimates the effect of the expectations of being bailed out on the fiscal discipline of Swedish municipalities for the period 1979– 1992. He proxies these expectations by the explicit bailout-grants received by neighbouring municipalities and reports that such an expectation increases local public debt by 20%. Bordignon and Turati (2009) estimate the effect of bailout expectations, proxied by political characteristics, on regional public health care expenditure for the Italian regions during the period 1990–1999. They also find that bailout expectations increase these expenditures. Garcia-Milà et al. (2001) show that, for the period 1984–1995, the amount of additional grants expected by Spanish regions, proxied by the share of regions in terms of population, led to higher levels of regional public debt.3 By contrast, Lago (2005) empirically analyses the determinants of Spanish regional deficits for the period 1984–1996, and finds a role for, not bailout expectations, but lack of regional financial autonomy. The above studies are concerned with the fiscal-federal moral hazard problem or the effect of bailouts or expectations of being bailed out on the fiscal discipline of sub-central governments. The studies proxy bailout expectations using several characteristics identified as determinants of bailout expectations but do not empirically study the existence and determinants of implicit bailouts.4 Empirical evidence is lacking on the existence and determinants of bailouts. My aim is to provide such empirical evidence. To the best of my knowledge, my research question has not been previously addressed empirically. Spain is a particularly well suited institutional environment for my study, because of an ongoing process of decentralization, gradually introduced over time, and characterised by significant differences across regions. Contemporary regional public expenditure accounts for approximately 40% of overall public expenditure. During the period analysed, there was a substantial increase in the issue of public debt by regional governments. Although the greater part of their resources was from formula-based grants, there was substantial scope for discretion in distribution of grants. Spanish regions have experienced considerable variation in their fiscal autonomy; moreover, they are responsible for providing basic public services, such as health care and education, from different points in time. These differences across regions and in time allow us to identify the existence and determinants of implicit bailouts.5 I find that the Spanish central government uses discretionary and (a priori) non-discretionary grants to partially bail out regional governments. Bailouts are more prevalent when there are limits on the borrowing autonomy of the regions; the region is responsible for providing health care; and the region has a high proportion of swing voters. The paper is organized as follows. In Section 2 I describe when a bailout takes place and the main institutional characteristics that affect this decision. I also identify testable predictions. In Section 3 I set out the empirical strategy and briefly describe the characteristics of the intergovernmental relationships in Spain. The data base and econometric procedures are also discussed in this section. The results are presented in Section 4. The last section summarises the conclusions. 2. Existence and determinants of bailouts The natural way to model the bailout decision is in terms of a sequential game (Carlsen, 1998; Goodspeed, 2002; Inman, 2003). The central government makes a no-bailout claim. The sub-central governments assess the credibility of this claim and make their budgetary decisions. If they fail to find the central claim credible, they issue debt. In the final stage, the central government decides whether to bail out the sub-central governments. The decision depends on the payoffs of each action. The benefits of refusing to provide fiscal rescue operations accrue in the future when the central government's no-bailout claim becomes credible and subcentral governments behave responsibly. The costs of no bailout are immediate, through welfare of the population of a sub-central government and electoral consequences for the incumbent at the centre. This temporal gap between costs and benefits, together with the fact that politicians mainly focus in the short run, suggests that politicians only consider the costs of not bailing out a region.6 This reduces the credibility of the no-bailout claim. 2

See Besfamille and Lockwood (2008) for a theoretical approach to this issue. They develop a model in which they consider aggregate bailouts, i.e., the central government cannot increase the grants allocated to one region, but rather has to increase the grants assigned to all regions. These additional grants are distributed in proportion to each region's population. Thus, expectations of bailout are proxied by the regions' share of population. 4 Pettersson-Lidbom (2010) uses data on explicit and effective bailouts, since there is a grant devoted to bail out (under certain conditions) Swedish municipalities. In Italy and Spain there is no such a grant. 5 From here on I use the term bailout to refer solely to implicit bailouts since, in Spain, there is no grant whose specific purpose is to explicitly bailout regions in financial difficulties. 6 This assumption appears in all theoretical papers that study this issue. See e.g. Desai and Olofsgård (2006) for a theoretical and detailed approach that justifies that governments only focus on the political benefits. This paper analyses the political advantage of soft budget constraint (i.e. it is a focus on the view of the regional government, rather than on the central government). 3

156

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

2.1. Existence of a bailout Grants are the most common fiscal instrument used by central governments for implicit bailouts. A central government uses grants to bail out a regional government, if it assigns additional grants to the region in a given period due to the region's issue of debt in a previous period, i.e. if ∂ Gi/ ∂ Dit − 1 N 0, where Git are the grants assigned to region i at period t and Dit − 1 is the stock of public debt of region i at t − 1. Thus, I define an implicit bailout as the increase in the grants that a region receives because of issue of debt. The bailout can be partial or total, that is ∂ Gi/ ∂ Dit − 1 ≤ 1 + r, where r is the interest rate. 2.2. Potential determinants of a bailout 2.2.1. Type of intergovernmental grants The existence of bailouts is conditioned by the types of grant in use. If the distribution of grants is discretionary, a central government can use them to bail out regional governments. By contrast, a bailout cannot be undertaken if grants are completely non-discretionary, debt is not one of the distributional criteria, and none of the distributional criteria correlate with debt (Rodden, 2001).7 However, the existence of distributional rules can have the opposite effect and increase the likelihood of bailouts. Equalization grants, for example, depending on their design, can result in the central government's bailing out regional governments (Seitz, 2000). Moreover, politicians can manipulate the distribution of grants that are, in principle, formula-based (Khemani, 2003). Such manipulation generally takes place during the bargaining stage among politicians, when the variables to be considered in the formula and their weight are determined. Hence, formula-based grants can, in practice, allow for some discretion in grant allocation. Thus, the existence of a bailout is conditioned on the design of grants. The decision whether to bail out a regional government also depends on the costs of not providing a bailout. If the central government were benevolent, these costs would only depend on economic characteristics of the region. However, objectives of governments include their staying in office (e.g. Cox and McCubbins, 1986; Lindbeck and Weibull, 1987; Case, 2001; Dahlberg and Johansson, 2002). A central government thus takes into consideration political characteristic of the regions when deciding on a bailout. Goodspeed (2002) proposes as the objective function of the central government when deciding whether to bail out a region maximization of the expected number of votes. The literature has identified economic and political characteristics of regions that determine whether a bailout takes place. 2.2.2. Assignment of expenditure responsibilities If the central government were fully benevolent, the direct cost of not bailing out a region would be the reduction in individuals' welfare due to either a reduction in the level and/or quality of the public goods or services or an increase in the regional tax burden. If a regional government is responsible for providing key public goods or services, it is especially difficult for the central government to refuse to bail it out when that region can no longer provide such services. This is particularly true in those cases in which the regional government has little fiscal autonomy, i.e., a low adjustment capacity. In that case grants act as insurance. It is also true when the central government has established a minimum or standard level of provision of public services that are the responsibility of regional governments (Rodden et al., 2003; Bordignon and Turati, 2009). 2.2.3. Externalities Externalities affect the bailout decision. If a regional government cannot provide the optimal level of these services that generates positive externalities, the central government will tend to bail it out since any reduction in the provision of these services affects not only the utility of the consumers in this region, but also that of the consumers of other regions (Wildasin, 1997). An equal increase in per capita debt affects the interest rate more if the region has a large population. The incentives for the central government to bail out a region increase with the magnitude of these externalities, as stated in the “too big to fail” hypothesis. However there is also evidence that the probability of being bailed out decreases with the size of the region, since it is cheaper to bail out a small than a large region.8 Thus, the effect of the size of the region on the bailout decision is ambiguous. 2.2.4. Degree of borrowing autonomy Many countries impose borrowing restriction on regional governments9 in order to avoid regional financial problems that would require bailouts. Success of the restrictions depends on the regional governments' ability to circumvent them (MilesiFerretti, 2000) and on their credibility (Mikesell, 2002). If they are not effective, the limits can have the opposite effect. If complying with the restrictions is important for the central government, their establishment can result in the central government bailing out regions.

7 For instance, in an equalization grant where Income is one of the distribution criteria (the higher the income of the region, the lower the grant it is assigned) and, if Income positively correlates with Debt (cov(Income, Debt) N 0), a bailout can be implemented reducing the weight of Income in the formula that determines the distribution of this grant. 8 An example is the bailout provided to two German Länder (von Hagen et al., 2000). 9 These restrictions can take many forms, such as balanced-budget constraints or restrictions on the ability to issue debt. See Ter-Minassian and Craig (1997) for a summary.

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

157

2.2.5. Political characteristics The political characteristics of the regions also affect the bailout decision. Empirical studies that have analysed the influence of political factors on the distribution of grants identify political alignment, electoral productivity and partisanship as the main determinants. Recent papers claim that the party in central government assigns more grants to politically-aligned regions (i.e., regions that are ruled by the same party as the one in central government), since in this case, all the political credit derived from the grants is fully captured by the party (Solé-Ollé and Sorribas-Navarro, 2008; Arulampalam et al., 2009). Some papers suggest that parties will allocate more resources to regions in which the marginal gains are higher, that is, in regions with a high proportion of “swing voters” (e.g. Case, 2001; Johansson, 2003). Some of these papers have sought to contrast this hypothesis with an alternative theory that claims that, if politicians are risk averse, more grants will be allocated to those jurisdictions in which voters are clearly attached to the incumbent party (i.e. regions with a high proportion of “core supporters”) (Dahlberg and Johansson, 2002; Castells and Solé-Ollé, 2005). These arguments identify potential determinants of the bailout decision.10 If the distribution of bailouts aims at ensuring the re-election of the incumbent party, in order to avoid loss of political support, the central government will bail out (or devote more resources to bail out) regions that are politically aligned with it. A risk-averse central government will bail out core regions and otherwise swing regions. 2.2.6. Testable predictions The testable predictions derived from the literature are accordingly the following: Prediction 1: The central government will use grants to implicitly bail out a regional government. Prediction 2: The decision to bail out a region is conditioned by the type of grants; in principle, it is more difficult to use nondiscretionary grants to bail out regions. Prediction 3: The decision to bail out a region does not only depend on its issue of debt, but also on economic and political characteristics of the regions.11 3. Empirical implementation 3.1. Econometric specification Given the definition of a bailout, I use the following basic equation to estimate whether the central government uses grants to implicitly bail out regions: k

Git = βDit−1 + ∑ αk Xit + Fi + Ft + uit k

ð1Þ

where Git are the intergovernmental grants that the central government assigns to region i in period t; Dit − 1 is the public debt of region i in period t − 1; Xkit are the control variables picking up the main structural aspects of region i that affect the long-run desired level of grants and its political characteristics, as well as any other characteristics, different from debt, that capture a downturn in the economy that could also affect the distribution of grants; Fi and Ft are fixed region and time effects, respectively, which pick up structural and temporal characteristics that affect the distribution of grants that are not included in the control variables; and, uit is the error term, with zero mean and constant variance. If the estimated value of β (i.e. ∂ Git/ ∂ Dit − 1) is positive and statistically significant, it can be claimed that the central government uses supplementary grants to bail out regions. If β differs from one, the bailout is partial and otherwise total. The central government can also assign additional grants to a region to compensate for the negative economic impact of an economic downturn or an external macroeconomic shock. In that case, grants act as an insurance or, in other words, as a risksharing mechanism.12 There are several papers that, using a different econometric approach, provide empirical evidence on the role of intergovernmental grants as a risk-sharing mechanism.13 To be certain that the estimated parameter β identifies what I have defined as an implicit bailout, I include among the control variables a set that aims at capturing the negative effect of an external shock or an economic downturn that can potentially affect the distribution of grants. I estimate Eq. (1) with and without these control variables. If the estimated coefficient is quite stable, it can be claimed that it is capturing the existence of bailouts. It could be the case that debt is endogenous, in which case OLS estimates would be biased. Suppose there is an external negative shock that affects a region in period t − 1. This shock generates an increase in the stock of public debt. If grants act as insurance, in that period this region could receive additional grants. That could induce the region to spend more and thus increase its debt. If this is the case, OLS estimates will be upwards biased. To take into account this possible endogeneity, I use IV estimates. The exogenous

10 Although Goodspeed (2002) takes the central government to have a political objective, he does not provide a prediction about which political characteristics matter. 11 These characteristics are country specific and are dependent on the institutional environment. See Neyapti (2010) on the effect of the institutional environment on economic outcomes and decisions. 12 During the period analysed the main tax bases were in hands of the central government. Moreover, regional governments have limited responsibilities (shared with the central government) regarding the main expenditures that they provide. Thus, even if they made use of their fiscal autonomy it could be insufficient to smooth the effects of an external negative shock. 13 See, e.g. Asdrubali et al. (1996) for the states in the USA, estimating a VAR and Sorensen et al. (2000) on states and local government in the USA.

158

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

Table 1 Financing system of common regions. Average values, €2001 per capita.

Other revenues Grants Total non-financial revenues % Discretionary grants % Non-discretionary grants Public debt

1987–1991

1992–1996

1997–2001

2002–2006

189.19 715.17 904.36 15.45% 84.55% 2569.76

230.26 1025.49 1255.75 15.42% 84.58% 8185.67

415.17 1311.92 1727.09 26.63% 73.37% 11,228.10

1342.18 1238.94 2581.12 18.58% 81.42% 12,848.54

Source: BADESPE and Banco de España. BADESPE: Base de Datos Económicos del Sector Público Español, data base of the Spanish public sector, elaborated by the Spanish Institute of Fiscal Studies (www.ief.es). Banco de España: Spanish central Bank (www.bde.es).

variation of debt used is based on the evolution of the interest rate and the structure of regional public debt.14 The instrument is the share of the stock of debt that each region has at a variable interest rate interacted with the increase in the interest rate. Variation over time in the interest rate is considerable, and also in the structure of the public debt over regions. The estimation of Eq. (1) will answer the question whether the central government bails out regional government. If the answer is yes, the estimation of the following equation will identify the determinants of bailouts: ! Git =

β1 +

j ∑ βj Yit J

c

Dit−1 + ∑ αc Xit + Fi + Ft + uit c

ð2Þ

ˆ k estimates the effect where Yjit are the J variables that summarize the potential economic and political motives of bailouts. Thus, β ˆ k is of the interaction of debt with a variable that captures a characteristic that can have an effect on the bailout decision. If β positive and statistically significant from zero, it can be claimed that the characteristic captured by Yk induces the central government to use grants to bail out regions. 3.2. Fiscal federalism in Spain 3.2.1. Fiscal decentralization Spain's decentralization process started in 1978 with the passing of the Constitution whereby 17 regional governments, the Autonomous Communities (AC), were created. Each AC is composed of one or more provinces. The AC have quite important spending responsibilities in the provision of several public goods and services, including health care, education and welfare. However, not all regions have been responsible for providing the same public goods and services. Indeed, one of the main differences among them, until 2002, was that not all regions had the responsibility for the provision of health care and education. These responsibilities were assigned to regions at different times. Provision of health care and education are considered key services by the Spanish central government, which fixes a minimum standard and passes the basic law that regulates the provision of these services. In 1983 the revenue from some taxes was first ceded to regional governments, but the sum was insufficient to cover all their expenditure needs, which continued to be financed primarily by grants from the central government. From 1986 to 2001, the regional financing system was negotiated and modified every 5 years,15 with the aim of increasing the fiscal autonomy of the regions.16 In 1992 the Spanish central government passed a law (Escenarios de Consolidación Presupuestaria, ECP) limiting central and regional borrowing in order to fulfil the requirements of the Maastricht Treaty. This law established a maximum value of the ratio of the regional public debt to the GDP, which was different across regions. In 2001 the Spanish parliament approved the General Law of Budgetary Stability, which requires, from 2002 on, that all public administrations keep their budgets balanced or in surplus. The aim of both limits was to ensure the fiscal discipline of regional governments. However, the data shows that the two limits were not effective. 3.2.2. Grants As can be seen in Table 1, during the period analysed (1986–2006), intergovernmental grants were the main source of revenue for the Spanish regions, providing on average around 70% of total regional resources. Of these, some were completely discretionary (18%), in the main capital grants, while the greatest share of grants was, a priori, non-discretionary (82%) since their distribution is formula-based. The PIE (Participacion en los Ingresos del Estado) share of state revenues, called Fondo de Suficiencia, (sufficiency fund, from 2002 on) was the main grant (44%) and its distribution is mainly formula-based. The funds that an AC receives from PIE are computed as the difference between its estimated expenditures needs on some common expenditure responsibilities17 and 14 Change in debt has two components: primary deficit and debt servicing. Primary deficit can be affected by the behavior of regional governments. Debt servicing mainly depends on the interest rate, which is exogenous to regional governments' decisions. 15 The financing system approved in 2001 did not have any deadline. However, a new financing system was passed in July 2009. Thus, during the period analysed there have been four different financing systems, between 1987–1991, 1992–1996, 1997–2001, and 2002–2006 respectively. Nevertheless, it should be noticed that each new financing system merely includes small modifications to the previous one. 16 See Lago (2005) and Bosch and Duran (2007) for a broad description of the financing system of Spanish regions. 17 These expenditure needs are determined on the basis of the population (main variable), surface, and density of the regions.

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

159

revenue from the ceded taxes.18 The goal in principle of this grant is to guarantee horizontal fiscal equalization. This formula is only computed for the first year of the financing system. For the remainder of the years, each AC receives the funds corresponding to the first year assessment, but the revenue level is up-dated on the basis of a common index. The PIE formula can be modified every 5 years when the regions renegotiate their financing system. Although the distribution of this grant is formula-based, there is scope for discretion in its assignment. In fact, this formula is itself the result of political negotiation.19 In addition to this formula, there are a number of rules and financial guarantees that determine distribution.20 These additional rules have no economic justification and result in a final distribution of the grant that is significantly different from the formula allocation. Thus, these two traits provide discretion in distribution. The second most important grant in quantitative terms, until 2002, was that devoted to financing the regional provision of health care. It is also, a priori, non-discretionary since it is determined on a formula basis21 but not one applied yearly. Here there is also room for discretion during bargaining. Moreover, until 1994 the implementation of the health financing system was based on bilateral agreements between the regions and the central government, which introduced an additional element of discretion in its distribution.22 In 2002, when all the regions were responsible for providing health care, this grant disappeared, being integrated into the sufficiency fund and included in the computation of expenditure needs for health-care provision. As Khemani (2003) claims, formula-based grants are, in principle, less sensitive to political interests but there is discretion exercised in the political negotiation of the allocation criteria. This holds for Spanish intergovernmental grants.

3.2.3. Regional public debt As shown in the last row of Table 1, the issue of debt by regional governments has increased substantially, especially since 1992, when the ECP established the limits on debt. After the Stability Law was passed in 2001, the stock of regional public debt also increased but at a much lower rate.

3.2.4. Political environment In central and regional elections, the electoral districts are the provinces. The number of representatives elected in each province depends on its population size, and the d'Hondt formula with a threshold23 is used to translate votes into representatives. As such, the system is not entirely proportional and, in practice, it is much easier to win a seat in certain provinces (rural areas) than in others. In Spain, central and regional elections are usually held at regular four-year intervals, although they can be called before the end of the term-of-office. During the period analysed (1986–2006), six central and six regional24 elections were held. The incumbent at the centre was the left-wing (PSOE) during the first 10 years (1986–1995), right-wing (PP) during the following 8 years25 (1996–2003), and then the left-wing (PSOE) again during the last years of the period (2004–2006). Both parties had both majority and minority governments.26 Among the regions, the incumbent has always been left-wing in just 3 and always rightwing in just 2, and there have been majority governments and coalitions. Hence, during the period analysed there were significant variations in the political characteristics of both central and regional governments.

3.3. Sample, variables and data source The empirical analysis uses data for the 15 common regime Spanish regions27 for the period 1986–2006. In this section the variables used to measure grants and debt are defined, as are also those used to proxy the potential bailout determinants enumerated in the second section. A summary of the definitions of these variables, their summary statistics and sources, can be found in the Appendix. 18 Note that the importance of grants over the total non-financial resources diminishes substantially after the negotiation of the last general financing system in 2001. The reason is that when the sufficiency fund was negotiated, simultaneously more taxes were ceded to the regional governments (in some cases just the revenue generated by them was ceded and in some others some normative capacity over them too). 19 There is no economic analysis that has proposed the variables that are used to determine the expenditure needs and the fiscal capacity of the regions and their weights. They are the result of a multilateral agreement by all the common regime regions. There is only one formula that is applied to all the regions. 20 A small share of the total resources of this grant is not allocated according to its formula. 21 The greatest share of these resources is allocated according to the population and population over 65 years old at each region. This formula is also the result of a multilateral agreement. 22 During the period analysed there have been two health financing systems, covering the periods 1994–1997 and 1998–2001, respectively. 23 Any party that does not receive at least 3% of the votes at a given electoral district will not have any seat allocated to it. 24 Central elections took place in June 1986, October 1989, June 1993, March 1996, March 2000 and March 2004. Regional elections are not held simultaneously in all regions. See Solé-Ollé and Sorribas-Navarro (2008) for a broader description of Spain's electoral system and political parties. 25 PSOE: Partido Socialista Obrero Español; PP Partido Popular. PSOE is recognised as being a left-wing party and PP as right-wing. 26 PSOE ruled from June 1993 to March 1996 in a parliamentary coalition with two regional governments, CiU (Convergència I Unió, from Catalonia) and PNV (Partido Nacionalista Vasco, from the Basc Country). PP ruled from May 1996 to February 2000 in a parliamentary coalition with tree regional governments: CiU, PNV and CC (Coalición Canaria, from the Canary Islands). From April 2004 to March 2008 the PSOE had a minority government. Otherwise there have been majority central governments. 27 For historical reasons, two types of regions exist: the foral communities of Navarra and the Basque Country, and the rest (common regime AC). Given the great differences in their financing system, only the 15 common regime regions are considered in the analysis. The description in the Fiscal Federalism subsection refers to the common regime regions.

160

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

3.3.1. Intergovernmental grants Discretionary grantsit are defined as those resources that region i receives at year t from grants that do not have a formula or clear rule to determine their distribution. In the Spanish case, these are mainly capital grants. In principle, these are the only grants that could be used by the central government to assign additional resources to regions due to use of debt. However, in the Spanish case, the distribution of formula-based grants has scope for discretion when distribution criteria are negotiated. Therefore, I also test if non-discretionary grants are used to bail out regional governments. Non-discretionary grantsit are defined as those grants based on a formula for assignment.28 3.3.2. Assignment of expenditure responsibilities The main difference across regions in expenditure responsibilities is in responsibility for the provision of health care and education at different points in time. These differences are captured by an Expenditure responsibilityit index that proxies the increase in regional expenditure needs due to the assignment of the provision of health care and/or education.29 This index is computed on the basis of two dummy variables, health careit and educationit, which are equal to one if a region i is responsible for providing health care and/or education in year t and zero otherwise. 3.3.3. Externalities Following Wildasin (1997), the externalities generated by the public goods and services provided by regional governments are proxied by the size of the regions in terms of population, Population Shareit. 3.3.4. Degree of borrowing autonomy To test the effect of the limits on the regional borrowing autonomy established in 1992 and 2002 on the central government's decisions, the temporal dummy Limits on Debtt is defined, which is equal to one during the period 1992–2006, and is equal to zero otherwise. This variable is decomposed into two to capture whether the effect on the bailout decision differed for the two types of limits over this period. Share GDPt captures the limits on the borrowing autonomy that existed during the period 1992–2001 and Deficitt = 0 those of the period 2002–2006. 3.3.5. Political benefits To test the effect of political alignment, the dummy variable Alignedit is defined. This is equal to one when either the singleparty or the leader of the coalition in the regional government is of the same party as that in the central government (also a single party or coalition leader) (see Solé-Ollé and Sorribas-Navarro, 2008 for a discussion of the definition of the alignment status). Since Spain is a multi-party system with a large number of parties competing at each general election, votes are quite dispersed. Looking at the election results, it can be observed that, first, obtaining more than around 40% of the votes implies, nearly in all the cases, having the most votes and winning the elections. Second, the vote share obtained by the party with the second highest number of votes when elections are close is around 30% to 35%. Taking these observations into account, a core and a swing region are defined based on the vote share and the electoral margin of the incumbent. To test the effect of partisanship, the dummy variable Coreit is defined. This is equal to one when the party in the central government obtained more than 40% of the votes in region i at the general elections held prior to t with an electoral margin greater than 10%. This Core variable aims to identify those regions in which the incumbent had sufficient support to assume that he or she will win the election. To test for the effect of electoral productivity, the dummy variable Swingit is defined as equal to one when the party in the central government obtained more than 35% of the votes at the previous general elections and its electoral margin is lower than 10%.30,31 Since Spain does not have a twoparty system, the electoral margin is measured as the difference (in absolute value) between the vote share of the party in government and the vote share of the next party, with either more or fewer votes (Johnston et al., 1999). This Swing variable aims to capture those regions in which the greatest share of voters is not specially attached to any party, and thus votes depend on both on political ideology as well as on political decisions.32 In the swing regions, not always the same party wins the elections. The swing regions can be aligned or unaligned, but the core regions are nearly always aligned. Thus, the Aligned and Core variables are highly correlated and the excluded category is the Non-aligned regions. 3.3.6. Regional public debt The public debt of regional governments, Debtit, is the debt defined in the Protocol on the Excessive Deficit of the European Union.33 As discussed above, the greatest share of grants is formula-based and their criteria distribution can only be modified 28

In the Appendix the grants included in each category are enumerated. See the Appendix for a description of how this index is calculated. 30 As the definitions of Swing and Core regions are somewhat arbitrary, as a robustness check in the empirical analysis these variables are defined employing different limits for the share of votes and the electoral margin. 31 The threshold fixed on the vote share to define both variables is not the same since Swing regions should not only capture those regions where the incumbent wins by a small margin, but also those where the incumbent looses by a small margin. 32 Note that the regions that are swing can at the same time be pivotal regions. This is the case of the regions where the regional parties have been members of a parliamentary coalition. To test whether being pivotal has any effect, the regressions have been estimated including swing and a dummy variable for the pivotal regions and the estimated coefficient for being pivotal is not significant. Thus, for those swing regions the fact of being pivotal does not have any additional effect on the distribution of grants and nor on the bailout decision. 33 This includes the long- and short-run debt of the regional public authorities, but it does not consider the debt of all public firms and organisms. The debt of the latter bodies should also be considered, since many regions have used them to externalise public debt. Nonetheless, these data are not available. 29

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

161

when the financing system of the Spanish regions is renegotiated. It is quite reasonable to expect that these grants can only react to the regional public debt existing when their formula is determined. That is the point in time at which the central government can fix the distribution criteria such that more grants are assigned to regions due to their use of debt. If this assumption holds, formulabased grants should only react to the existing debt when the general financing system is negotiated.34 To test this hypothesis, the stock of public debt at period t is decomposed as follows. The Initial Debtit is defined as the public debt that a region i has when the financing system of year t is negotiated; and the Accumulated Debtit as the increase in the stock of debt since that time. If the estimated coefficients of both variables are positive and statistically significant, it can be concluded that the central government's decision to bail out regions depends on the stock of debt observed at every period. By contrast, if only the estimated coefficient of Initial Debtit is statistically significant, then the decision of the central government is determined by the stock of debt existing when the financing system is negotiated. The central government can also assign additional grants to regions when the health-care financing system is negotiated. In order to consider this effect for those regions with responsibility for providing health care, Accumulated Debt is decomposed into accumulated debt until the health care financing system comes under negotiation and the remainder.35 If the estimated coefficients of the Initial Debt and the Accumulated Debt when negotiating the health care financing system are positive and statistically significant, it can be concluded that the central government's decision to bail out regions depends on the stock of debt when the general financing system is negotiated and, for those regions that are responsible for providing health care, on the accumulated debt. 3.3.7. Control variables Apart from the issue of debt, several structural aspects of the regions might affect the distribution of grants. These structural aspects are mainly related to the expenditure needs and the fiscal capacity of the regions. The expenditure needs are proxied by the Populationit of the region (Solé-Ollé, 1999) and by the index, Expenditure Responsibilitiesit, which accounts for the differences in the expenditure responsibility assigned to regions. During the period analysed, Spanish regions mainly had taxes levied on wealth. There were no significant differences in their tax autonomy. Thus, their fiscal capacity is proxied by their gross domestic product, Incomeit (Esteller et al., 2005). In order to control for the negative economic impact of an economic downturn or an external shock that can also affect the distribution of grants, the region's rate of unemployment observed in the previous period, Unemploymentit − 1, and the outflow rate in the previous period, Outflow rateit − 1 are included as control variables. The lag of the dependent variable, Grants it − 1, is also included in the regressions to control for the dependence and persistence in the distribution of grants. The distribution of discretionary grants could be independent from one year to another, but this is not the case due to the design of the non-discretionary grants. Moreover, in the case of discretionary grants, the fact that there are investment projects that last more than 1 year and the political cost that a substantial change in their criteria distribution could have, can motivate persistence in their distribution. In addition to these structural aspects, the political variables (Alignedit, Swingit, and Coreit) are also considered as control variables. All monetary variables are expressed at constant 2001 prices and in per capita terms in order to standardize by region size. 3.4. Econometric issues With the lag of the dependent variable included to control for the persistence in the distribution of grants, we have a dynamic panel that covers data over 21 years. As the literature on dynamic panel data has emphasized, given the rather long time period available, the Nickell (1981) bias should not be a problem. Thus, although the lagged dependent variable is included as a regressor, there is no need to use instrumental techniques to obtain unbiased estimators.36 The empirical analysis proceeds as follows. I start with a relatively parsimonious specification in which grants are regressed on public debt and time dummies. Then, first, the economic control variables that capture the structural and cyclical characteristics of the regions are included; second, the lag of grants; and third, the political characteristics of the regions. When the model is estimated using instrumental variables to control for potential endogeneity of debt, the Limited Information Maximum Likelihood estimator is used, since in finite samples it provides a finite sample bias reduction in relation to TSLS (Wooldridge, 2002). 4. Results The empirical analysis starts by focusing on discretionary grants (Tables 2 and 3) and then moves to the analysis of (a priori) non-discretionary grants (Tables 4 and 5). Table 6 shows the central government's reaction to the stock of regional public debt when the financing system is being negotiated. The tables also report specification statistics. Both a fixed and a random effects version of the model were estimated. However, the hypothesis of no correlation between the fixed effects and the variables included in the regression is rejected, i.e. the Hausman Test is overcome, and the random coefficients model is rejected. For this reason, the within-estimator results are reported. In all cases, all the variables are jointly significant and the time effects are also significant.37 34 Although this assumption only refers to non-discretionary grants, it could also hold for discretionary grants. It could be the case that discretionary grants react more when a negotiation takes place, since the distribution of these grants can be used to facilitate the multilateral agreement needed by all regions. 35 See the Appendix for a detailed definition of these variables. 36 Empirical papers with T around 20 use OLS estimators. See e.g. Buettner and Wildasin (2006). 37 Due to space restriction, all these tests are only reported for the basic specification. However, these results hold for all the specifications estimated.

162

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

Table 2 Discretionary grants: Existence of bailouts. [1] Existence of bailouts Debt (t − 1)

0.017⁎⁎ (0.008)

[2] 0.021⁎⁎ (0.009)

Economic controls Discretionary grants (t − 1) Population (10−3)

−0.026 (0.022) 9.757⁎⁎

Income (10−3)

(4.005) 9.865 (8.160) −0.083 (0.891) 0.894⁎⁎⁎

Expenditure responsibilities (Health and/or Education) Unemployment (t − 1) Outflow rate (t − 1)

(0.250)

[3]

[4]

[5]

0.016⁎⁎ (0.008)

0.016⁎⁎ (0.008)

0.016⁎⁎ (0.008)

0.407⁎⁎⁎ (0.104) −0.018 (0.011) 5.685 (3.462) 6.236 (5.184) −0.099 (0.824) 0.812⁎⁎⁎

0.407⁎⁎⁎ (0.101) −0.018 (0.011) 3.638 (3.773) 5.918 (5.201) −0.284 (0.827) 0.730⁎⁎⁎

0.401⁎⁎⁎ (0.103) −0.018 (0.012) 4.902 (3.884) 5.774 (5.319) −0.093 (0.860) 0.774⁎⁎⁎

(0.252)

Political controls Swing Aligned

Year dummies R-squared N. observations Hausman Test (FE vs. RE) F-test (zero slopes) F-test (zero year) F-1st stage excluded instrument

5.055 (63.547) Yes 0.564 300 38.256 [0.045] 57.421 [0.000] 45.789 [0.000]

8.661 (56.817) Yes 0.637 300 39.354 [0.036] 68.003 [0.000] 45.671 [0.000]

0.428⁎⁎⁎ (0.088) −0.019 (0.010) 3.734 (3.521) 5.874 (5.033) −0.312 (0.774) 0.786⁎⁎⁎

(0.264)

(0.239)

5.926⁎ (3.267) 7.161⁎⁎

5.210⁎ (3.184)

5.742⁎ (3.059) 7.892⁎⁎

(3.029)

57.050⁎⁎⁎ (8.721) Yes 0.534 300 33.120 [0.033] 57.011 [0.000] 44.457 [0.000]

0.013⁎⁎⁎⁎ (0.003)

(0.261)

Core Constant

[6]

23.453 (61.081) Yes 0.646 300 39.267 [0.035] 68.229 [0.000] 45.012 [0.000]

(3.406) 3.884 (2.998) 10.065 (62.937) Yes 0.639 300 38.971 [0.041] 68.437 [0.000] 44.655 [0.000]

29.788 (58.370) Yes 0.648 300

70.458 [0.000] 42.008 [0.000] 69.444 [0.000]

Notes: 1. Figures in parentheses are robust standard errors. 2. ⁎,⁎⁎,⁎⁎⁎ denotes significance at 1, 5 and 10% level. 3. Figures within brackets are p-values. 4. See the Appendix A for a detailed definition of these variables.

4.1. Discretionary grants Table 2 reports the baseline results for discretionary grants (Eq. (1)). The results reported in column [1] of Table 2 show that the central government's distribution of discretionary grants depends positively on the issue of debt by regional governments (i.e. ∂ Git/ ∂ Dit − 1 N 0 and statistically significant). Thus, the initial conclusion is that Spain's central government uses discretionary grants to implicitly partially bail out regional governments. If a region increases its debt by 100 € per capita, the central government will, on average, assign to this region an additional discretionary grant of 1.7 € per capita. When the economic control variables are included (column [2]), the estimate of the debt variable is basically unchanged. More discretionary grants are assigned to richer regions and to regions with a positive outflow rate in the previous period. The expenditure needs and the unemployment rate38 do not play a role in the allocation of discretionary grants. The lagged value of grants (column [3]) is a significant determinant of the allocation of discretionary grants, with income no longer significant. If a region increases its debt by 100 €, the central government will assign to this region an additional discretionary grant of 1.6 € in the following period, and 2.7 € in the long run.39 Regarding the political characteristics, the results suggest evidence of tactical distribution in the assignment of the grants (column [4]). More grants are assigned to aligned and swing regions. There is no evidence that the central government assigns more resources to the core regions (column [5]). Thus, the distribution of the resources is intended to maximize electoral success. As a robustness check, these regressions were estimated with alternative

38

As a robustness check, the regressions were estimated using the unemployment rate for young people (less than 25 years old) and the results did not change. β is the reaction of the central government in the short run, while β/(1 − αG) is its reaction in the long run, where αG is the coefficient associated to the lagged value of grants. 39

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

163

Table 3 Discretionary grants: Determinants of bailouts. [1] Determinants of bailouts Debt (t − 1) × Population share × Limits on debt

[2]

0.015⁎⁎ (0.007) −0.063 (0.084) 0.042 (0.026)

share GDP Deficit = 0 × Expenditure resp.

0.006 (0.013)

[3]

[4]

[5]

0.015⁎⁎ (0.007) −0.061 (0.082)

0.014⁎⁎ (0.006) −0.028 (0.086)

0.014⁎⁎ (0.006) −0.063 (0.085)

0.014⁎⁎ (0.007) −0.088 (0.085)

0.042 (0.026) 0.044⁎⁎

0.044 (0.027) 0.051⁎⁎

0.043 (0.027) 0.041⁎⁎

0.043 (0.028) 0.041⁎⁎

(0.021) 0.006 (0.012)

(0.025)

(0.020) 0.012 (0.013)

(0.021) 0.013 (0.012)

0.019⁎ (0.010) 0.004 (0.006)

0.018⁎ (0.010)

Health

0.006 (0.014) 0.005 (0.016)

Education × Swing × Aligned

−0.011 (0.007)

× Core

Economic controls Discretionary grants (t − 1) Population share Income Expenditure responsibilities (Health and/or Education)

0.388⁎⁎⁎ (0.103) −536.716 (566.870) 4.948 (4.349) 7.688 (7.798)

0.388⁎⁎⁎ (0.103) −553.761 (625.981) 4.876 (4.550) 7.359 (7.275)

Health

Outflow rate (t − 1)

0.369⁎⁎⁎ (0.106) −406.851 (641.788) 5.804 (4.594) 11.570 (7.392)

0.361⁎⁎⁎ (0.107) −301.569 (633.593) 7.346 (4.771) 13.259 (9.478)

0.067 (0.776) 0.730⁎⁎⁎ (0.245)

0.029 (0.811) 0.764⁎⁎⁎ (0.251)

6.123⁎ (3.658)

−0.109 (0.781) 0.800⁎⁎⁎ (0.261)

−0.116 (0.779) 0.745⁎⁎⁎ (0.261)

6.201 (4.699) 2.029 (1.952) −0.016 (0.767) 0.750⁎⁎⁎ (0.262)

5.520⁎ (3.237) 6.057⁎⁎

5.433⁎ (3.288) 6.033⁎⁎

5.413⁎ (3.022) 6.433⁎⁎

6.561⁎ (3.899) 6.150⁎⁎

(2.984)

(3.034)

(3.091)

(3.155)

Education Unemployment (t − 1)

0.381⁎⁎⁎ (0.103) −567.561 (625.788) 4.877 (4.483)

Political controls Swing Aligned Core Year dummies R-squared N. observations

Yes

Yes 0.651 300

Yes 0.651 300

Yes 0.655 300

0.654 300

3.045 (2.874) Yes 0.651 300

Notes: 1. Figures in parentheses are robust standard errors. 2.⁎,⁎⁎,⁎⁎⁎ denotes significance at 1, 5 and 10% level. 3. Figures within brackets are p-values. 4. See the Appendix A for a detailed definition of these variables.

definitions of the Core and Swing regions (i.e., establishing different limits on the share of votes and the electoral margin40) but the results obtained were not significantly different.41 Note that the estimate of the debt variable remains quite stable when all the control variables are included.

40 First, I have reduced the electoral margin from 10 to 5 percentage points, aiming at capturing as swing regions those in which electoral competition is greater. In this case, being a swing region positively affects assignment of grants and the bailout decision. The coefficient estimated is slightly higher, but it is less precisely estimated (standard errors increase). Core regions do not have any special treatment. Second, I have increased the threshold on the vote share up to 45% to define the core regions, and to identify those regions in which support is stronger, and the results obtained hold, independently of the threshold fixed on the electoral margin (5 or 10 percentage points). Third, I have increased the threshold on the vote share to 40% to define swing regions. The results obtained hold, but the coefficient estimated is slightly smaller and are less precise. 41 As a robustness check, the regressions were also estimated introducing one political variable at a time and the results hold. Only being swing affects the distribution of grants and the bailout decision.

164

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

Table 4 (A priori) Non-discretionary grants: Existence of bailouts. [1] Existence of bailouts Debt (t − 1)

0.689⁎⁎⁎ (0.055)

[2]

Income (10−3) Expenditure responsibilities (Health and/or Education) Unemployment (t − 1) Outflow rate (t − 1)

[4] 0.090⁎⁎⁎ (0.026)

0.093⁎⁎⁎ (0.027)

0.085⁎⁎⁎ (0.021)

−0.349 (0.273) −5.039 (16.277) 735.346⁎⁎⁎ (26.370) 10.852 (9.617) 0.869 (1.009)

0.536⁎⁎⁎ (0.044) −0.252 (0.252) −13.906 (11.719) 399.508⁎⁎⁎ (36.746) 8.524 (6.652) 0.879 (2.162)

0.522⁎⁎⁎ (0.043) −0.258 (0.250) −7.167 (11.533) 415.889⁎⁎⁎ (36.787) 7.627 (6.596) 0.684 (2.005)

0.519⁎⁎⁎ (0.043) −0.264 (0.251) −6.826 (11.326) 417.026⁎⁎⁎ (36.769) 7.698 (6.567) 0.610 (1.995)

0.531⁎⁎⁎ (0.039) −0.284 (0.201) −7.221 (11.022) 416.297⁎⁎⁎ (35.881) 7.825 (6.033) 0.702 (1.985)

62.349⁎⁎⁎ (18.925) 4.652 (12.634)

56.215⁎⁎⁎ (19.559)

61.942⁎⁎⁎ (18.556) 4.812 (11.241)

Aligned Core

F-test (zero slopes) F-test (zero year) F-1st stage excluded instrument

Yes 0.257 300 34.156 [0.025] 61.003 [0.000] 49.882 [0.000]

[6]

0.073⁎⁎⁎ (0.025)

Political controls Swing

Year dummies R-squared N. observations Hausman Test (FE vs. RE)

[5]

0.202⁎⁎⁎ (0.032)

Economic controls Non-discret. grants (t − 1) Population (10−3)

[3]

Yes 0.908 300 39.850 [0.030] 62.398 [0.000] 51.289 [0.000]

Yes 0.961 300 40.377 [0.027] 69.854 [0.000] 52.342 [0.000]

Yes 0.962 300 39.971 [0.029] 71.490 [0.000] 50.785 [0.000]

−14.100 (15.981) Yes 0.963 300 38.966 [0.037] 69.882 [0.000] 51.603 [0.000]

Yes 0.960 300

73.521 [0.000] 49.225 [0.000] 54.480 [0.000]

Notes: 1. Figures in parentheses are robust standard errors. 2. ⁎,⁎⁎,⁎⁎⁎ denotes significance at 1, 5 and 10% level. 3. Figures within brackets are p-values. 4. See the Appendix A for a detailed definition of these variables.

The results hold when IV estimates are used and the coefficient estimated is only slightly smaller but not statistically different. These IV estimates should be interpreted as the effect of the exogenous variation of debt on the distribution of grants. First stage regression results show that the instrument is a statistically significant determinant of regional public debt and with the expected positive sign. The F-test of excluded instruments takes a value above 10. Moreover, it is reasonable to assume that the structure of debt interacted with the interest rate is uncorrelated with grants. Since the model is just identified, an over-identification test cannot be computed. However, there is no reason to expect the instrument to be correlated with a shock to grants. Hence, the instrument proposed seems to be a good one. These results suggest that debt is not endogenous and thus OLS estimates are not biased. The results in first column of Table 3 show no evidence that economic characteristics of the institutional setting affect the central government's bailout decision.42 Thus, expenditure responsibilities, population, and limits on debt do not affect the decision of the central government to use grants to bail out regions. However, when I allow the effect of limits on borrowing autonomy to differ according to design (column [2]), the results show that the limits that from 2002 were intended to force the regional governments to keep their budgets balanced or in surplus, induce the central government to use discretionary grants to bail out regions. On average, after the establishment of this limit, if a region increases its debt by 100 €, the central government will assign to this region an additional discretionary grant of 5.9 € in the following year (as opposed to 1.5 € before its establishment). From these results, it is not possible to conclude that the limits introduced were effective. In fact, as mentioned earlier, regional governments did not abide by the limits. This can be explained by the design of the limits, which did not include a penalty for noncompliance and was too restrictive.43 These two facts undermine the credibility of the limits and induce the central government to use grants to bail out regions. Regarding the expenditure responsibilities, even when I decompose the effects of health care provision and education, there is no evidence that the difference in expenditure responsibilities has any effect on the bailout decision (column [3]). 42

In all these regressions the variable population is not included since its correlation with the share of population is higher than 90%. In fact, in December 2005 the Stability Law was modified in order to allow public administrations to have their budget in deficit in very specific situations (defined in terms of economic downturns) and not exceeding a fixed threshold. 43

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

165

Table 5 (A priori) Non-discretionary grants: Determinants of bailouts. [1] Determinants of bailouts Debt (t − 1) × Population share × Limits on debt

[2] 0.067⁎⁎ (0.032) 0.143 (0.454) 0.284⁎⁎⁎ (0.058)

× share GDP × Deficit = 0 × Expenditure resp.

0.124 (0.142)

[3]

[4]

[5]

0.065⁎⁎ (0.031) 0.354 (0.435)

0.061⁎⁎ (0.031) 0.446 (0.443)

0.061⁎⁎ (0.031) 0.469 (0.426)

0.060⁎⁎ (0.030) 0.463 (0.416)

0.299⁎⁎⁎ (0.057) 0.226⁎⁎⁎

0.289⁎⁎⁎ (0.062) 0.210⁎⁎⁎

0.300⁎⁎⁎ (0.064) 0.215⁎⁎⁎

0.359⁎⁎⁎ (0.071) 0.273⁎⁎⁎

(0.057) 0.115 (0.139)

(0.059)

(0.061)

(0.068)

0.143⁎⁎ (0.072) 0.114 (0.082)

0.160⁎⁎ (0.073) 0.117 (0.082) 0.059⁎

0.131⁎ (0.071) 0.087 (0.081) 0.054⁎

(0.033) 0.007 (0.042)

(0.031)

× Health × Education × Swing × Aligned × Core

Economic controls Non-Discretionary grants (t − 1) Population share Income Expenditure responsibilities (Health and/or Education)

0.107 (0.087)

0.483⁎⁎⁎ (0.044) −1786.18 (1141.73) −21.643 (18.560) 443.409⁎⁎⁎

0.477⁎⁎⁎ (0.043) −1787.26 (1282.01) −18.006 (12.468) 435.962⁎⁎⁎

0.482⁎⁎⁎ (0.045) −1752.27 (1164.78) −19.567 (12.709)

0.479⁎⁎⁎ (0.045) −1815.71 (1316.36) −18.651 (12.677)

0.463⁎⁎⁎ (0.04) −1959.74 (1333.6) −18.757 (12.182)

(34.152)

(32.785)

324.888⁎⁎⁎ (46.670) 281.870⁎⁎⁎

326.986⁎⁎⁎ (46.233) 276.148⁎⁎⁎

299.567⁎⁎⁎ (43.893) 313.962⁎⁎⁎

7.168 (6.266) 0.352 (1.759)

9.360 (7.232) 0.623 (1.672)

(44.479) 9.656 (7.226) 0.287 (1.687)

(44.821) 9.695 (7.186) 0.433 (1.647)

(44.522) 10.007 (7.073) 0.799 (1.579)

45.049⁎⁎⁎ (16.922) 3.927 (12.815)

42.491⁎⁎ (17.011) 4.350 (13.201)

43.991⁎⁎ (17.740) 4.497 (13.847)

43.765⁎⁎ (17.797) 4.771 (13.251)

43.844⁎⁎ (17.573)

Health Education Unemployment (t − 1) Outflow rate (t − 1)

Political controls Swing Aligned Core Year dummies R-squared N. observations

Yes

Yes 0.968 300

Yes 0.969 300

Yes 0.970 300

0.970 300

−10.360 (16.232) Yes 0.971 300

Notes: 1. Figures in parentheses are robust standard errors. 2. ⁎,⁎⁎,⁎⁎⁎ denotes significance at 1, 5 and 10% level. 3. Figures within brackets are p-values. 4. See the Appendix A for a detailed definition of these variables.

For political characteristics, the results in column [4] suggest that swing regions receive a higher additional discretionary grant than non-swing regions, independently of their alignment status. On average, if a swing region increases its debt by 100 €, the central government will assign to this region an additional discretionary grant of 3.3 € (compared to 1.4 € if it is a non-swing region44). Thus, being a swing region is a characteristic that not only affects the decision of the central government to use discretionary grants to bail out regions, but also its general decision regarding how to distribute these grants.45 These results are in

44

After the establishment of the limits on debt in 2002, this difference is 7.7 € vs. 5.9 €. As in all the papers cited in Section 3.3, when I enumerate the political characteristics that can affect the bailout decision (as well as the distribution of grants), I assume that voters vote retrospectively i.e. the distribution of grants in period t will affect the voting decision in the future elections. Thus, these political characteristics are not endogenous. 45

166

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

Table 6 Bailout-grants: importance of the regional public debt. Discretionary grants [1] Initial debt when negotiating General Financing System (FS) Accumulated debt

0.011 (0.008) 0.022⁎⁎ (0.011)

Accumulated debt when negotiating Health Financing System Accumulated debt if responsible of Health Accumulated debt if not responsible of Health

Non-discretionary grants [2] 0.010 (0.008)

0.016⁎⁎ (0.008) −0.003 (0.012) 0.031⁎⁎

[3] 0.129⁎⁎⁎ (0.026) 0.018 (0.032)

(0.015) Economic controls Grants (t − 1) Population (10−3) Income Expenditure responsibilities Unemployment (t − 1) Outflow rate (t − 1)

Political controls Swing Aligned Year dummies R-squared N. observations

0.410⁎⁎⁎ (0.099) −0.017 (0.012) 3.943 (3.745) 8.040 (5.258) −0.283 (0.825) 0.735⁎⁎⁎

0.404⁎⁎⁎ (0.098) −0.017 (0.012) 4.187 (3.665) 8.526 (5.295) −0.305 (0.816) 0.715⁎⁎⁎

(0.255)

(0.243)

5.300⁎ (3.158) 6.668⁎⁎

5.243⁎ (3.133) 6.891⁎⁎

(2.938) Yes 0.647 300

(2.969) Yes 0.653 300

[4] 0.115⁎⁎⁎ (0.026)

0.057⁎⁎ (0.023) 0.046 (0.046) −0.059 (0.038)

0.507⁎⁎⁎ (0.042) −0.306*** (0.051) −15.355 (11.034) 431.091⁎⁎⁎ (35.510) 6.850 (5.520) 0.773 (2.002)

0.523⁎⁎⁎ (0.043) −0.285*** (0.047) −9.914 (10.428) 417.742⁎⁎⁎ (35.488) 6.149 (5.345) 0.976 (1.848)

67.054⁎⁎⁎ (18.994) 4.955 (5.986) Yes 0.965 300

66.766⁎⁎⁎ (18.736) 5.562 (5.991) Yes 0.966 300

Notes: 1. Figures in parentheses are robust standard errors. 2.⁎,⁎⁎,⁎⁎⁎ denotes significance at 1, 5 and 10% level. 3. Figures within brackets are p-values. 4. Columns [1] and [3] report the results of estimating equation [8] decomposing Debt in Initial Debt when negotiating General Financing System (FS) + Accumulated Debt. 5 Columns [5] and [4] report the results of estimating equation [8] decomposing Debt, for the regions responsible for providing health care, in Initial Debt when negotiating General Financing System (FS) + Accumulated Debt when negotiating Health FS + Accumulated Debt if Responsible of Health; for the regions not responsible for providing health care Debt is decomposed in Initial Debt when negotiating General Financing System (FS) + Accumulated Debt if not Responsible of Health.

line with Goodspeed's (2002) assumption of politically motivated bailouts. The results suggest that the central government is risk averse and, in order to increase re-election prospects, assigns additional resources to marginal regions. Thus, the results provide evidence of the importance of institutions (e.g. De Mello, 2000; Neyapti, 201046). 4.2. (A priori) non-discretionary grants The baseline results for non-discretionary grants are reported in Table 4. The first column shows that the distribution of nondiscretionary grants also depends positively on the issue of debt. Thus, although in principle non-discretionary grants cannot be used to bail out regional governments, there is evidence that in practice Spain's central government does use non-discretionary grants for this purpose. This result can be due to the fact that the most quantitatively important a priori formula-based grants (PIE and Health Care Grants) allow for discretion in their distribution when they are negotiated. Contrary to the case of discretionary grants, when the economic control variables are included (column [2]), the estimate of the debt variable remains positive and statistically significant but is reduced substantially in magnitude. The differences in responsibilities assigned to regions regarding expenditure determine the distribution of grants.47 Income, the unemployment rate and the outflow rate do not seem to affect the allocation of non-discretionary grants. As expected because of design, the lagged value of grants (column [3]) is a significant determinant of the allocation of non-discretionary grants.

46 These two papers use a cross-country analysis to study the effect of fiscal decentralization on sub-central government budget balances. The first paper emphasizes the role of grants and how this role can vary depending on the institutional environment. 47 This is an expected result, since the resources that regional governments receive to finance health care and education come from these non-discretionary grants. This is also the reason why this coefficient is significant and much larger in the case of non-discretionary grants.

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

167

Among the political control variables, only being a swing region positively affects non-discretionary grants assigned by the central government. Controlling for all these control variables, on average, if a region increases its debt in 100 € per capita, the central government will assign to this region an additional non-discretionary grant of 9 € in the short run and 18.8 € per capita in the long run. All these results hold when IV estimates are used (column [6]). As in the case of discretionary grants, the estimated coefficient is only slightly smaller and the instrument is a good one. Regarding the economic determinants of the central government's decision to use non-discretionary grants to bail out regions, the following conclusions should be highlighted (Table 5). First, the establishment of limits on the borrowing autonomy of regional governments affects the decision whether to bail out a regional government. On average, after the establishment of those limits, if a region increases its debt by 100 €, the central government will assign to this region an additional non-discretionary grant of 35.1 € in the following year (as opposed to 6.7 € before their establishment). When I allow the two limits to differ, the results show that the magnitude of the bailout is higher when a threshold on the share of GDP was established in 1992 (29.9 € vs. 22.6 €). However, given that discretionary grants are only used to bail out regions when deficit were disallowed for regional governments (5.9 €), overall it cannot be claimed that the effect of these two limits on the borrowing restriction have a different effect on the bailout decision. Thus, it cannot be claimed that the limits on debt were effective. Also, in the case of the limits established in 1992, the regional governments knew that it was important for the central government to meet these criteria in order to fulfil the requirements of the Maastricht Treaty. That undermined their overall credibility. Second, responsibility for health care provision leads the central government to bail out regional governments. If a region increases its debt by 100 €, the central government will, on average, assign to this region an additional non-discretionary grant of 20.4 € in the following year if this region is responsible for providing health care (as opposed to just 6.1 € if it is not).48 The assignment of the provision of education to regional governments does not affect the central government's decision. Perhaps expenditure on health care increases because of technological change and the aging of the population and thus is difficult to control. Third, as in the case of discretionary grants, swing regions receive a higher bailout than that received by non-swing regions. On average, if a swing region increases its debt by 100 €, the central government will assign to this region an additional nondiscretionary grant of 12 € (as opposed to just 6.1 € for a non-swing region). These results are robust to broader and narrower definitions of both swing and core regions. Thus economic and political considerations influence the decision of the central government to use non-discretionary grants to bail out regions. Moreover, the magnitude of the bailout can be significant. For instance, following the establishment of limits on the use of debt in 1992, a swing region with responsibility for health care provision that increased its debt by 100 € in a one-year period received 58 € of additional non-discretionary grants in the following year. 4.3. How discretion can be introduced in (a priori) non-discretionary grants? Table 6 shows significant differences depending on the type of grants. When the stock of debt is decomposed into that existing when the general financing system is negotiated and the increase in debt since that time, the results show that discretionary grants respond to the issue of debt all over the period. In contrast, (a priori) non-discretionary grants only react to the debt in existence when the general financing system is being negotiated. These results indicate that there is room for introducing discretion in the allocation of these grants, but only when their allocation criteria are negotiated. When the increase in debt is decomposed for the regions responsible for providing health care, the distribution of both types of grants reacts to the stock of regional public debt observed when the health financing system is being negotiated. For the regions that do not have this responsibility, discretionary grants react to the variation in debt during all the years of the period. Thus, non-discretionary grants are used to bail out regional governments when the general and health financing systems are negotiated. Discretionary grants are sensitive to each year's variation in the stock of debt for those regions that are not responsible for providing health care and to the stock of debt existing when the health care financing system is negotiated for those regions that have this responsibility, which can be used to convince these regions to approve the new financing system. 5. Conclusions This paper has empirically investigated bailouts in Spain's fiscal federal system. Although theoretical models suggest a focus on discretionary grants, the main in-principle non-discretionary grants in Spain also allow discretion. Spain's central government uses grants to partially bail out regions, using both discretionary and non-discretionary grants. The size of bailouts can be considerable and differs significantly among regions. The distribution of grants is sensitive to each year's variation in the stock of public debt for those regions that are not responsible for providing health care and to the stock of debt when health care financing is negotiated for the others. Non-discretionary grants are used to bail out regions when either the general or the health care financing system is negotiated. Thus, although in principle distribution of grants is formula-based, in practice there is scope for political discretion when negotiation of the distribution criteria takes place. The expectation of a bailout can result in moral hazard. The distributional criteria applied in the allocation of the main non-discretionary grant, PIE, are in need of redefining so as not to be used to facilitate bailouts. A reform option is also increasing the fiscal autonomy of the regions. However, it is important to stress 48 For a region responsible for providing health care, these numbers are 49 € vs. 6 € when the limits on debt where established in 1992, and 41 € vs. 6 € after introducing the balanced-budget restriction in 2002.

168

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

that our results cannot be generalized to all grants and neither to all countries. The feasibility to distribute grants to bail out regions is conditioned to the institutional framework and to their design.

Acknowledgments I gratefully acknowledge comments received from Albert Solé, Alejandro Esteller, Antoni Castells, Teresa Garcia-Milà, Julio López, Javier Suárez, Ben Lockwood, Jordi Jofre and participants at workshops, conferences and seminars where I have presented the paper and from two anonymous referees and the editor. This paper has benefited from the financial support of ECO200912680/ECON (Spanish Ministry of Education and Science) and the project 2009 SGR 102 (Generalitat de Catalunya).

Appendix A Definition of the variables, descriptive statistics and data source. Variable

Definition

Mean (s.d.)

Source

Discretionary Grantsit

Discretionary grants per capita (pc) received by region i in period t (2001 €, pc) Non-discretionary grants pc received by region i in period t (2001 €, pc) Regional public debt pc of region i in period t (2001 €, pc) Population of region i in period t

38.814 (33.705)

BADESPE

678.09 (533.10)

BADESPE

543.307 (358.829)

Bank of Spain

2,516,820 (2,135,283)

National Institute of Statistics (INE)

Non-Discretionary Grantsit Debtit Populationit Expenditure Responsibilitiesit Incomeit Unemploymentit Outflow rate Healthit Educationit Population Shareit Limits on Debtt Share GDPt Deficitt = 0 Alignedit

Coreit

Swingit

Instrument: Δit ⋅ ShareVDit

Increase in regional expenditures needs due to the assignment of the provision of health and/or education. Gross domestic product of region i in period t (2001 €, pc) Unemployment rate in region i in period t (Immigrantsit − emigrantsit)/Populationit − 1 1 if region i is responsible for providing health in period t 0 otherwise 1 if region i is responsible for providing education in period t 0 otherwise Populationit = ∑i Populationit 1 if t = 1992–2006 0 if t = 1986–1991 1 if t = 1992–2001 0 if t = 1986–1991, 2002–2006 1 if t = 2002–2006 0 if t = 1986–2001 1 if a single party or a party leading a coalition at the regional government in i is the same party in the central government (also a single party o a coalition leader) at year t 0 otherwise 1 if the party at central government obtained more than 40% of the votes in the general elections hold previous to t and the electoral margin is bigger than 10 percentage points 0 otherwise 1 if the party at the central government obtained more than 35% of the votes in the general elections hold previous to t and its electoral margin is lower than 10 percentage points 0 otherwise Increase in the interbank interest rate ((it − it − 1)/it − 1) share of debt at a variable interest rate (Debt at a variable iit/Debtit)

0.700 (0.654) 8753.266 (1810.06)

INE

15.713 (6.42) 0.8281 (4.641) 0.444 (0.497)

INE INE

0.577 (0.494) 0.067 0.714 0.476 0.238 0.524

(0.056) (0.452) (0.500) (0.426) (0.500)

INE

Ministry of Interior (MI) and Eleweb

0.390 (0.488)

MI

0.263 (0.441)

MI

(0.029) (0.124)

Spanish Central Bank

1. See footnote Table 1. 2. INE: Instituto Nacional de Estadística, National Institute of Statistics (www.ine.es). 3. Eleweb: Xarxa temàtica en “eleccions, comunicació política i opinió pública”, net on elections, political communication and public opinion (www.eleweb.net).

Grants Non-discretionary transfers include PIE (called Fondo de Suficiencia, since 2002), Health Care Grant, Guarantee Fund (Fondo de Garantía), Fund of Inter-territorial Compensation (Fondo de Compensación Interterritorial) and expenditure responsibilities assigned to regional governments that are not included in PIE.

Discretionary Transfers include transfer to enterprises and families, investment projects (convenios y contratos), other conditional transfers, extraordinary compensations and self-government grants (órganos de autogobierno).

Regional Public Debt Debtit = Initial Debtit + Accumulated Debtit Initial Debtit: Public debt that a region i has when the financing system of year t is negotiated

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

169

Initial Debtit is Debti1986 for t = 1987–1991, Debti1991 for t = 1992–1996 and Debti1996; for t = 1997–2001 Accumulated Debtit = Debtit − Initial Debtit For the regions that are responsible for providing health care: Debtit = Initial Debtit + + Accumulated Debt when negotiating the health care financing systemit + + Accumulated Debt if responsible for healthit Accumulated Debt when negotiating the health care financing systemit = Debt when negotiating the health care financing systemit − Initial Debtit. It is only different from zero for those regions that are responsible for providing health care and during the years, within a period of the general financing system, after the health care financing system has been negotiated. Thus, for the regions that are responsible for providing health care, Debt when negotiating the health care financing system is Debti1993 for t = 1994–1996, and Debti1997 for t = 1998–2001. It is 0 otherwise. Accumulated Debt if responsible for healthit = t1 × ðDebtit – Initial Debtit Þ + t2 × ðDebt when negotiating health care financing systemit – Initial Debtit Þ

where t1 = 1 the years before the health care financing system is negotiated, and 0 otherwise; t2 = 1 the years after the health care financing system is negotiated, and 0 otherwise. Expenditure responsibilities This index is computed as follows: IERit = ðHit EH + Edit EEdÞ = CE where Hit and Edit are dummy variables that are equal to one if the region i has been assigned the provision of health or education, respectively, in period t; EH = 564.67 €; EEd = 428.05 €; CE = 714.48 € is the average per capita expenditure, at 2001 constant prices, during the period 1986–2001 on health, education, and on the provision of the public goods and services that are assigned to all regions, respectively. Hence, IERit is equal to 0 if the region has not been assigned health neither education; is equal to 0.8 (0.6) if it is responsible for providing health (education); and, is equal to 1.4 if it is responsible for providing health and education. References Ahmad, E., Albino-War, M., Singh, R., 2006. Subnational public financial management: institutions and macroeconomic considerations. In: Ahmad, E., Brosio, G. (Eds.), Handbook on Fiscal Federalism. Edward Elgar, Cheltenham, UK, pp. 405–430. Arulampalam, W., Dasgupta, S., Dhillon, A., Dutta, B., 2009. Electoral goals and centre-state transfers in India. Journal of Development Economics 88, 103–119. Asdrubali, P., Sørensen, B.E., Yosha, O., 1996. Channels of interstate risk sharing: United States 1963–1990. Quarterly Journal of Economics 111, 1081–1110. Besfamille, M., Lockwood, B., 2008. Bailouts in federations: is a hard budget constraint always best? International Economic Review 49, 577–593. Bird, R., Tassonyi, A., 2003. Constraining fiscal behaviour in Canada: different approaches, similar Results? In: Rodden, J., Eskenland, G.S., Litvack, J. (Eds.), Fiscal Decentralization and the Challenge of Hard Budget Constraints. MIT Press, Cambridge, USA, pp. 85–132. Bordignon, M., Turati, G., 2009. Bailing out expectations and public health expenditure. Journal of Health Economics 28, 305–321. Bosch, N., Duran, J.M. (Eds.), 2007. Fiscal Federalism and Political Decentralization: Lessons from Spain, Germany and Canada. Edward Elgar, Cheltenham UK. Buettner, T., Wildasin, D.E., 2006. The dynamics of municipal fiscal adjustment. Journal of Public Economics 90, 1115–1132. Carlsen, F., 1998. Central regulation of local authorities. Public Finance Review 26, 304–326. Case, A., 2001. Election goals and income redistribution: recent evidence from Albania. European Economic Review 45, 405–423. Castells, A., Solé-Ollé, A., 2005. The regional allocation of infrastructure investment: the role of equity, efficiency and political factors. European Economic Review 49, 1165–1205. Cox, G.W., McCubbins, M.D., 1986. Electoral politics as a redistributive game. Journal of Politics 48, 370–389. Dahlberg, M., Johansson, E., 2002. On the vote-purchasing behavior of incumbent governments. American Political Science Review 96, 27–40. De Mello, L., 2000. Fiscal decentralization and intergovernmental fiscal relations: a cross-country analysis. World Development 28, 365–380. Desai, R.M., Olofsgård, A., 2006. The political advantage of soft budget constraints. European Journal of Political Economy 22, 370–387. Esteller, A., Navas, J., Sorribas-Navarro, P., 2005. El diseño de las cestas impositivas desde una perspectiva financiera. Investigaciones Regionales 7, 69–89. Garcia-Milà, T., Goodspeed, T.J., McGuire, T.J., 2001. Fiscal decentralization policies and sub-national government debt in evolving federations. Working Paper no. 549, Universitat Pompeu Fabra, Barcelona. Goodspeed, T.J., 2002. Bailouts in a federation. International Tax and Public Finance 9, 409–421. Inman, R., 2003. Transfers and bailouts: enforcing local fiscal discipline with lessons from US federalism. In: Rodden, J., Eskenland, G.S., Litvack, J. (Eds.), Fiscal Decentralization and the Challenge of Hard Budget Constraints. MIT Press, Cambridge, MA, pp. 35–84. Johansson, E., 2003. Intergovernmental grants as a tactical instrument: empirical evidence from Swedish municipalities. Journal of Public Economics 87, 883–915. Johnston, R.J., MacAllister, I., Pattie, C.J., 1999. The funding of constituency party general election campaigns in Great Britain. Environment and Planning C: Government and Policy 17, 391–409. Khemani, S., 2003. Partisan politics and intergovernmental transfers in India. World Bank Policy Research Working Paper no. 3016. World Bank, Washington D.C. Kornai, J., Maskin, E., Roland, G., 2003. Understanding the soft budget constraint. Journal of Economic Literature 41, 1095–1136. Lago, S., 2005. Evolving federations and regional public deficits: bailout hypothesis in the Spanish case. Environment and Planning C: Government and Policy 23, 437–453. Lindbeck, A., Weibull, J.W., 1987. Balanced-budget redistribution as the outcome of political competition. Public Choice 52, 273–297.

170

P. Sorribas-Navarro / European Journal of Political Economy 27 (2011) 154–170

Mikesell, J.L., 2002. Subnational government bankruptcy, default and fiscal crisis in the United States. International Studies Program Working Paper Series no. 0221, International Studies Program, Andrew Young School of Policy Studies, Georgia State University, Atlanta. Milesi-Ferretti, G.M., 2000. Good, bad or ugly? On the effects of fiscal rules with creative accounting. IMF Working Papers no. 00/172. International Monetary Fund, Washington D.C. Neyapti, B., 2010. Fiscal decentralization and deficits: international evidence. European Journal of Political Economy 26, 155–166. Nickell, S., 1981. Biases in dynamic models with fixed effects. Econometrica 49, 1417–1426. Oates, W.E., 2005. Toward a second-generation theory of fiscal federalism. International Tax and Public Finance 12, 349–373. Pettersson-Lidbom, P., 2010. Dynamic commitment and the soft budget constraint: an empirical test. American Economic Journal: Economic Policy 2 (3). Rezk, E., 1998. Argentina: fiscal federalism and decentralization. In: Bird, R., Vaillancourt, F. (Eds.), Fiscal Decentralization in Developing Countries. Trade and Development. Cambridge University Press, Cambridge UK, pp. 206–237. Rodden, J., 2001. And the last shall be first: The political economy of federalism and deficits in Germany. Unpublished paper, MIT. Rodden, J., Eskenland, G.S., Litvack, J., 2003. Decentralization and the Challenge of Hard Budget Constraints. MIT Press, Cambridge MA. Schaltegger, C.A., Feld, L.P., 2009. Are fiscal adjustments less successful in decentralized governments? European Journal of Political Economy 25, 115–123. Seitz, H., 2000. Subnational governments bailouts in Germany. Working Paper R-396. Inter-American Development Bank, Research Network, Washington D.C. Solé-Ollé, A., 1999. Estimació de les necessitats de despesa dels gooverns sub-centrals: aspectes metodològics i aplicació al càlcul d'una subvenció anivelladora. Unpublished PhD thesis, Universitat de Barcelona. Solé-Ollé, A., Sorribas-Navarro, P., 2008. The effects of partisan alignment on the allocation of intergovernmental transfers. Differences-in-differences estimates for Spain. Journal of Public Economics 92, 2302–2319. Sorensen, B.E., Wu, L., Yosha, O., 2000. Output fluctuations and fiscal policy: US state and local governments 1978–1994. European Economic Review 45, 1271–1310. Ter-Minassian, T., Craig, J., 1997. Control of subnational borrowing. In: Ter-Minassian, T. (Ed.), Fiscal Federalism in Theory and Practice. International Monetary Fund, Washington D.C., pp. 156–172. von Hagen, J., Eichengreen, B., 1996. Federalism, fiscal restraints and European Monetary Union. American Economic Review 86, 134–138. von Hagen, J., Bordignon, M., Dahlberg, M., Grewal, B.S., Petterson, P., Seitz, H., 2000. Subnational government bailouts in OECD countries: four case studies. Working Paper R-399. InterAmerican Development Bank Research Network, Washington D.C. Wetzel, D., Papp, A., 2003. Strengthening hard budget constraints in Hungary. In: Rodden, J., Eskenland, G.S., Litvack, J. (Eds.), Fiscal Decentralization and the Challenge of Hard Budget Constraints. MIT Press, Cambridge MA, pp. 393–428. Wildasin, D.E., 1997. Externalities and bailouts. Hard and soft budget constraints in intergovernmental fiscal relations. Working Papers — Macroeconomics and Growth Stabilization, Monetary/Fiscal Policy no. 1843. World Bank, Washington D.C. Wooldridge, J.M., 2002. Econometric Analysis of Cross Section and Panel Data. MIT Press, Cambridge MA.