Do equalization payments affect subnational borrowing? Evidence from regression discontinuity

European Journal of Political Economy xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

European Journal of Political Economy journal homepage: www.elsevier.com/locate/ejpoleco

Do equalization payments affect subnational borrowing? Evidence from regression discontinuity ⁎

Monika Köppl–Turynaa, , Hans Pitlikb a b

Agenda Austria, Schottengasse 1/3, 1010 Vienna, Austria Austrian Institute of Economic Research, Arsenal Objekt 20, 1030 Vienna, Austria

A R T I C L E I N F O

ABSTRACT

JEL classification: D72 D78 H74 H77

According to the fiscal federalism literature, subcentral budget constraints become softer when local governments are more dependent on revenues over which they have no discretion. As a consequence of ‘transfer dependency’, subcentral governments can expect to be bailed out by the central government and therefore tend to accumulate higher levels of debt. We test this conjecture with data from Austrian municipalities. In fiscal terms, Austria is a highly centralized federation in which tax autonomy at the municipal level is rather weak. Our identification strategy is based on a discontinuity caused by the unique regulation of population weights in the tax-sharing agreement between central government and the municipalities. Our results indicate that, in line with theoretical expectations, municipalities with higher revenue dependency are responsible for higher net borrowing per capita. The size of the additional borrowing effect equals to about 5% of average municipal debt. We also find that almost one half of the observed discontinuity works through an investment channel.

Keywords: Fiscal autonomy Subnational borrowing Vertical fiscal imbalance Regression discontinuity

1. Introduction Carefully designed fiscal decentralization can contribute to the effectiveness of the public sector and eventually to an increase in welfare. The traditional fiscal decentralization theorem (Tiebout, 1956; Oates, 1972) points to the conclusion that the local provision of public goods is better suited to satisfying the needs of local communities. Public choice identifies the role of decentralization in taming the ‘Leviathan’ government (Brennan and Buchanan, 1980), which would otherwise inefficiently overexpand at taxpayers' expense (see also Feld, 1997). However, tax competition could lead to a race to the bottom regarding tax rates (see Zodrow and Mieszkowski, 1986; Wilson and Wildasin, 2004) and an underprovision of public goods. Competition for mobile citizen-taxpayers could increase efficiency in the local provision of public goods, but could also lead to an unwanted sorting of individuals who differ in terms of income. Such a selfselection process could lead to substantial differences in income distributions among communities (Schaltegger et al., 2011). Critics also point out that ill-designed fiscal decentralization can produce ‘soft budget constraints’, which not only preclude the efficient allocation of public money, but may exaggerate the problem of overexpanding government. Soft budget constraints are facilitated if subnational governments largely depend on transfers from the central government, as it lowers the local cost of borrowing. In the present paper, we take a closer look at this result and empirically analyze whether the dependence of municipalities on transfers from central government results in higher levels of borrowing at the municipal level. Our identification strategy exploits the population threshold discontinuity present in the Austrian tax-sharing agreement

⁎

Corresponding author. E-mail addresses: [email protected] (M. Köppl–Turyna), [email protected] (H. Pitlik).

http://dx.doi.org/10.1016/j.ejpoleco.2017.07.002 Received 3 January 2017; Received in revised form 23 May 2017; Accepted 11 July 2017 0176-2680/ © 2017 Elsevier B.V. All rights reserved.

Please cite this article as: Köppl Turyna, M., European Journal of Political Economy (2017), http://dx.doi.org/10.1016/j.ejpoleco.2017.07.002

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

between the central government and the municipalities. The allocation of tax revenues is a function of the population, with a cutoff point at the level of 10,000 inhabitants. At the threshold of 10,000, per capita transfers from the pool of joint revenues to a municipality increase by roughly 15%. We explore this quasi-experimental setup by applying a regression discontinuity (RD) design. Using data for the time period 2001–2014, our results indicate that municipalities with higher revenue dependency are responsible for higher net borrowing per capita. Additionally, we show that only about half of the additional borrowing is spent on investment. Some recent papers exploring the fiscal behavior of subcentral governments have also drawn on such institutional settings to employ an RD design (e.g., Egger and Koethenbuerger, 2010; Pettersson-Lidbom, 2012; Grembi et al., 2016). The method has also been criticized, with the relevant literature showing that an RD design using population thresholds should be interpreted carefully. Yet, our results seem to be robust, as they include several political, socioeconomic and geographic factors. Moreover, our tests show that the frequently mentioned major challenges of manipulation and sorting (Eggers et al., 2015) are most likely not an issue here. This paper is structured as follows. The next section provides a brief literature review, while Section 3 presents an overview of the Austrian institutional setting and the main hypothesis. Section 4 presents the data used and the empirical strategy, after which Section 5 presents the results of the investigation. Section 6 concludes the paper. 2. Previous literature and predictions The question of whether intergovernmental transfers and vertical fiscal gaps in federations affect subnational fiscal performance has been intensively analyzed in the economic and political science literature. The link between vertical fiscal gaps and debt has been theoretically established by, among others, Goodspeed (2002) and Rodden (2006). Meanwhile, Rodden (2006) argues that grants make subcentral governments expect central assistance if they fall into fiscal difficulty. Given that the central government already funds substantial portions of subcentral budgets, local governments will find it politically difficult to resist pleas for bailouts in order to avert bankruptcy. The more that subcentral governments depend on the central government for fiscal support, the more creditors and voters will assume that the subcentral governments are simply administrative arms of the center and that the latter is responsible for the fiscal condition of its subordinates. As such, local governments will be more likely to run up large debts in the first place: the familiar phenomenon of moral hazard (Sorens, 2016). Empirical evidence points to increasing subnational deficits and debts associated with vertical fiscal gaps. Rodden (2002, 2006) finds that vertical fiscal imbalance reduces subcentral government and total net surplus, if combined with subnational borrowing autonomy. De Mello (2000) finds a nonlinear effect of grants (excluding shared revenue) on the budget surplus: the deficit increases along with grants when expenditure decentralization is high, and decreases when it is low. Similarly, Eyraud and Lusinyan (2011) find that grants, excluding shared revenue, raise the deficit when combined with borrowing autonomy. The counterevidence comes from Baskaran (2010), who finds that vertical fiscal imbalance is associated with lower debts. Foremny (2014) presents evidence that a higher share of subcentral entities' own tax resources reduces incentives to run deficits, due to lower bailout expectations. A high level of autonomy over tax instruments is key to preventing large deficits at the subnational level. As for other fiscal variables, a large body of literature analyzes whether vertical fiscal gaps are associated with higher, and potentially inefficient, subcentral expenditure. Inefficient fiscal performance, as a result of grant financing, has been recognized by both political economy and public finance scholars. Whereas political economy points to the claim by Brennan and Buchanan (1980) that the more dependent subcentral governments are on grants and shared revenue, the less they compete with each other for geographically mobile citizens, and the more they can extract from citizens for their own benefit, public finance scholars also recognize the dangers of vertical fiscal gaps, which create common fiscal problems and, in turn, negative externalities on other jurisdictions. In light of these observations, the evidence points to a dependence on the size of local governments and grant/transfer dependence. A large number of empirical studies also concludes that dependence on grants is associated with higher government spending. Cross-country evidence (see, e.g, Jin and Zou, 2002; Cassette and Paty, 2010; Ashworth et al., 2013; Prohl and Schneider, 2009) suggests that grants raise general and subnational spending. Intra-country studies, meanwhile, arrive at mixed conclusions. Dahlberg et al. (2008), using a strong identification strategy based on discontinuity in grants allocation, find that equalization grants increase general spending roughly one for one. Similarly, Volden (1999) finds that grants increase US social benefit payments, whereas cuts in grants do not decrease them. On the other hand, Gordon (2004) finds evidence that block grants boost spending, but only in the short term. The effects of vertical fiscal imbalance on government spending have been analyzed by, e.g., Fiva (2006) and Rodden (2003), both of whom conclude that vertical fiscal imbalance is associated with higher general spending. Rodden (2003) also proposes that subnational spending increases alongside vertical fiscal imbalance. An important challenge for these empirical investigations is endogeneity. Several papers in this field of research apply RD designs to mitigate the effects of reverse causality. The notion of RD is to get closer to a quasi-experimental setting. Using a population threshold RD approach, one can compare sets of municipalities that have implemented different policies, but are comparable in other important respects. For example, Egger and Koethenbuerger, 2010 exploit population size discontinuities in the law of the German state of Bavaria, which regulates the local council size of municipalities, in order to identify a causal impact on municipal spending. Pettersson-Lidbom (2012) also estimates the causal effect of council size on government spending for local jurisdictions in Sweden and Finland with a population size-based RD design. Yet, employing population thresholds in order to identify causal effects has certain pitfalls, most notably, confounded treatment, as well as manipulation and sorting issues (see the discussion in Eggers et al., 2015). Confounded treatment is related to the problem that, at a certain population threshold, more than one policy or set of institutional arrangements can change. If other factors create 2

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

discontinuities in this relationship at the same threshold level, a clear identification of the local treatment effect is not always possible. In order to identify causal effects, a deeper institutional analysis is required to rule out such effects. In our case, we report and test the potentially confounded treatment at a certain population threshold with respect to changes in council size. Moreover, identification is only possible if municipalities are unable to manipulate the assignment variable and strategically sort it into the desired category. In that respect, the use of population thresholds for identification may be problematic. For example, recent research provides evidence for the claim that a manipulation of population figures by municipalities in order to be eligible for higher per capita transfers is present in local fiscal transfer schemes in Spain (Foremny et al., 2015). Foremny et al. (2015) make the case that the responsibility of local governments for the collection of population data generates stronger incentives and opportunities to overreport population size. De Witte and Geys (2015) also find evidence that local strategic housing policy decisions surpass population thresholds in the allocation of funds in Belgian municipalities. Opportunities for strategic sorting are dependent on institutional characteristics. McCrary (2008) and Cattaneo et al. (2016b) present formal tests of strategic sorting. 3. Institutional setting, data and hypotheses 3.1. Basic institutional setting Austria is a federation of nine states (“Länder”) and 2100 municipalities1. In 2015, general government debt amounted to EUR 290.7 billion, which is 85.5% of GDP. The debt share of the states is 7.1%, while municipalities' share of total debt is 4.7%. Austria is characterized by a system of cooperative federalism, with substantial overlaps involving the competences of governmental levels, a highly complex system of revenue sharing, and a rather low level of tax autonomy in the states and municipalities. In principle, all local entities are in charge of identical public tasks, regardless of their respective population size (“Fiktion der abstrakten Einheitsgemeinde”). The responsibilities of the municipalities do not differ systematically by population size. Rather, the tasks performed by municipalities depend on historical tradition, geographical location, state law and financial situation. Moreover, many tasks provided by local governments are not carried out on a mandatory basis, but are optional. In practice, larger municipalities often provide somewhat different services, possibly generating benefit spillovers to neighboring communities. Every community with over 10,000 inhabitants is a “city” by definition, but smaller municipalities are sometimes assigned a city status as well (“Stadtrang”). However, the distinction between cities and communities is formally unrelated to a different assignment of responsibilities. Among the exceptions are so-called “statutory cities”, which typically have over 20,000 inhabitants, perform additional policing issues and have responsibility for special fiscal equalization schemes imposed on them. Statutory cities are excluded from the following analysis. The allocation of tax revenues follows the national Fiscal Equalization Law (German: “Finanzausgleichsgesetz” (FAG)). The FAG is based on federal legislation, which is negotiated between all governmental levels. The FAG is usually in effect for four to six years2. It consists of a set of regulations for tax competences, supplementary fiscal transfers, cost-sharing arrangements and fiscal equalization schemes, covering all levels of government. For the vertical distribution of revenues, tax sources are assigned to governmental tiers (“original vertical distribution”). About 14% of all tax revenues stem from so-called “own taxes”, whose receipts are allocated exclusively to one governmental level. While the states have less than 0.5% of total revenues from their own taxes, the federal level in 2015 received 8.5% of total revenues from its own sources, while municipalities received only slightly more than 5%. The most important own taxes at the local level are the real estate tax and the payroll tax (“Kommunalsteuer”). Municipalities only have the right to set tax rates for the real estate tax, yet within limits set by state regulations. The federal level is in charge of the legal definition of the tax base. Cities and towns, however, have some discretion with respect to the design of local user fees and charges. For the fiscally more important municipal payroll tax (“Kommunalsteuer”), municipalities have no discretion; the tax base and tax rate are fixed uniformly across all jurisdictions by the federal government. Around 86% of total tax revenues are so-called “shared taxes” (German: “Gemeinschaftliche Bundesabgaben”.) The tax base and tax rates of shared taxes are legislated exclusively at the federal level. Among others, revenues from personal income tax, corporate income tax, VAT and petroleum tax are shared by all governmental levels. Receipts from shared taxes are allocated between governmental levels by a fixed formula, following a distribution of pre-assigned revenues (“Vorweganteile”). Currently, with some minor exceptions, the federal level is entitled to a share of 67.4%, with the states receiving 20.7% of shared tax revenue. Austrian municipalities are entitled to the remaining 11.9% of the revenues from shared taxes. These shares are called “Ertragsanteile”. Even though shared taxes give slightly more autonomy to regional and local governments than intergovernmental grants, individual decision-making sovereignty regarding taxes at the subcentral level is very limited (cf. Blöchliger and Petzold (2009) for a theoretical discussion of the dividing lines between grants and tax-sharing arrangements). Regarding horizontal distribution, local revenues from own taxes and user fees pertain exclusively to the respective local entity. The distribution of shared taxes follows a two-step procedure. First, the total shares of the states and municipalities are divided between the nine states according to a fixed formula. In a second step, the respective shares of the local units of a state are distributed within the state. A smaller share is allocated to municipalities, for several historical reasons, mainly as compensation for previous changes in the FAG regulations, as well as to reduce disparities in fiscal capacities. About 86% is then allocated according to 1 This number was higher until the end of 2012, when a major wave of municipal amalgamations in the federal state of Styria took place. As reported in the recent literature, planned amalgamations generate additional incentives for merging municipalities to increase local debt. See Jordahl and Liang (2010) or Nakazawa (2013) for more detailed discussion. Hence, we shall exclude the municipalities located in Styria from the analysis. 2 The most recent FAG, which was in effect between 2008 and 2016, had been prolonged on two occasions. A new FAG has been in effect since 2017.

3

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

the weighted population size of the respective communities, such that larger municipalities receive a disproportionately higher amount per person. In the next subsection, we explain the relevant regulations, on which our empirical design is based, in slightly more detail. In a further step, special purpose grants and various cost-sharing agreements supplement the horizontal distribution of tax revenues. Eventually, the federal level ends up with a share of 56.3%, the states (excluding the city state of Vienna) with 13.6%, the capital city of Vienna with 7.7% and the remaining municipalities with 11.1% of total tax receipts (social contributions are not included). Meanwhile, 3.1% of tax receipts are assigned as the national contribution to the European Union budget. Incentives of states and municipalities to maintain and develop their own tax base are rather weak (for a very brief description and critique of the fiscal framework in Austria, cf. Pitlik (2014)). Substantial limitations on tax autonomy shift the focus of local economic policies almost exclusively to the expenditure side. The degree to which local governments in Austria are free to decide on their own spending figures is substantially higher than on the revenue side. In general, a principle of budgetary autonomy provides state and local governments with considerable decision-making powers. The combination of low level of tax autonomy and a relatively high level of discretion over spending certainly creates a familiar common-pool problem. Municipal governments can decide on the basis that higher expenditure does not correspond with local tax increases, as the resources stem from a common pool. Thus, the smaller the share of own tax revenues, the greater the incentives to inefficiently expand local spending, as corresponding cost increases are not fully internalized by local entities.

3.2. Weighted population size scheme Our RD design originates from Article §9(9) of the FAG, which stipulates that, depending on the population of a municipality, per capita receipts from shared taxes are multiplied. This regulation is called “abgestufter Bevölkerungsschlüssel” (aBS)). Depending on the state regulations, between 80 and 90% of the joint tax revenues are distributed according to the aBS scheme. The payments per capita are multiplied: 1. 2. 3. 4.

for for for for

municipalities municipalities municipalities municipalities

below 10,000 inhabitants by a factor of 1 41/67 between 10,001 and 20,000 inhabitants by a factor of 1 2/3 between 20,001 and 50,000 inhabitants and statutory cities by a factor of 2 above 50,000 and the city of Vienna by a factor of 2 1/3

It should be noted that the current shape of the aBS has been in force since the 2008 fiscal year. During the first seven years in our sample, the discontinuity has been larger: the population of municipalities up to 10,000 inhabitants has been multiplied by a factor 1 1/2. Before 2005, the factor equated to 1 1/3. The aBS regulation dates back to the Law on Public Finance of 1948. At that point in time, it was important to equip larger cities with additional financial means in order to support the reconstruction of city infrastructure damaged during the Second World War. Yet, the basic regulation is still in force, although the original rationale no longer exists. Currently, the main ‘economic’ argument is to compensate bigger municipalities for positive spillovers into neighboring entities, as well as compensation for seemingly higher per capita costs of service provision3. The historic origin of the aBS regulation leaves no doubt about its exogeneity. As the 10,000 threshold has remained unchanged since its introduction in 1948, the possibility that the threshold itself is endogenous to, e.g., political negotiations between the different layers of the government is excluded. At the population threshold of 10,000, there is a discontinuity in the size of the per capita payments from the pool of joint revenues to a municipality, as depicted in Fig. 14,5. Just below the cutoff point, the municipalities receive, on average, slightly less than EUR 700 per capita, whereas the payment just above the 10,000 cutoff mark increases to about EUR 800 per capita. Although shared taxes (“Ertragsanteile”) are additionally complemented by other transfers to the municipalities, as Fig. 2 shows, at a cutoff of 10,000, there is a drop in the fraction of municipalities' income resulting from own taxation from about 25% to about 20%. Unlike other transfer payments, revenue from shared taxes is not earmarked and can be free disposed of. It is important that neither legal responsibility nor overall per capita spending changes at the cutoff, as depicted in Fig. 3; if anything, expenditure is higher just below the threshold, due to several extreme values in the sample. Another important fiscal variable, lagged debt per capita, does not seem to change at the threshold, as visualized in Fig. 4. A final aspect of the discontinuity setup to be addressed is the so-called fading-in rule (“Einschleifungsregel”), which was introduced in 2001 and provides some additional payments to municipalities with between 9000 and 10,000 inhabitants. These payments increase linearly with the population, i.e., each additional inhabitant over a population level of 9000 is multiplied by a small linear multiplier. Importantly, in 2005, this regulation only applied to five municipalities, while, in 2014, this number rose to 10. Given that only a few observations are affected, we shall ignore this rule in our further calculations. As will be shown later in Section 5, the optimal bandwidths used are municipalities that have more than 1000 inhabitants, which means that the results should remain unaffected. 3 4 5

The latter statement has, however, never been broadly analyzed and quantified. Scatters represent the averages over equally spaced bins. In this study, we do not analyze the changes in borrowing at other threshold levels, as the sample is unreasonably small.

4

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Fig. 1. Local linear fit; bandwidth=1500; revenue sharing payments (“Ertragsanteile”) per capita, in euros.

Fig. 2. Local linear fit; bandwidth=1500; fraction of income from own taxation (in %).

Fig. 3. Local linear fit; bandwidth=1500; per capita expenditure in euros.

3.3. Local borrowing restrictions Limitations on local government debts and deficits stem from several sources (see Thöni et al., 2002 for more details). Municipal

5

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Fig. 4. Local linear fit; bandwidth=1500; per capita lagged debt in euros.

borrowing is, above all, regulated by specific rules, which differ across states. In principle, the most important rule in place is the obligation that ordinary and regular spending has to be covered by ordinary revenues (mainly taxes and user fees), and that only extraordinary and absolutely necessary expenditure may be (partly) financed by municipal loans or credit, in addition to surpluses from ordinary revenues. However, a clear legal definition of extraordinary expenditure does not exist. Borrowing by local governments usually requires approval by the state government; in some states, this rule is only applicable if municipal debt exceeds a certain level. The conditions under which municipal governments operate are contingent on how approval differs between states, although this does not usually involve severe borrowing limitations. According to Thöni et al. (2002), the approval process is not really an efficient instrument for limiting local debts, while Bröthaler et al. (2015) conclude that municipalities generally face a “nottoo-hard” budget constraint, which limits borrowing. Unclear and inconsistent legal definitions and rules allows for a lot of discretion, while supervision by the state government is rather weak. Moreover, Austrian municipalities face a fairly high degree of autonomy connected with ‘soft’ forms of borrowing, i.e., borrowing from publicly owned banks and state-owned enterprises, which inevitably leads to the insufficient control of borrowing through the financial market (Rodden, 2003; Oates, 2005). In addition to these regulations, Austria adopted a set of fiscal rules, which apply to all governmental levels in 1999. The Internal Stability Pact includes rules for the coordination of budgetary policies between the three governmental levels and requires both state and local governments to meet certain budgetary targets collectively as a single group, such that the republic complies with European fiscal rules. A lack of willingness to internally sanction rule-breaking behavior by local or regional units, however, has generated a prisoner's dilemma-like situation, in which incentives for municipalities not to limit deficits clearly dominate (Thöni et al., 2002). A revision of the Internal Stability Pact in 2012 somehow sharpened the rules by which to balance local budgets, while a structural budget balance rule has been in effect since 2017. However, an individual municipality's incentives to cheat and free-ride on collective budget discipline are still present. We use official data on municipal finances provided by Statistik Austria for 2001–2014. Excluding the state of Styria and the “statutory cities”, there are 1685 municipalities in the sample, 372 of which are cities. Summary statistics of all used variables are provided in Table 17 in the Appendix. The smallest Austrian municipality (Gramais) has 60 inhabitants, while the largest (Vienna) has over 1.8 million6, but as explained further on, we restrict the sample to municipalities with between 2000 and 20,000 inhabitants. 3.4. Hypothesis Our main hypothesis is derived from the “bailout” literature as well as the “soft borrowing” and “common pool” literature. As stated by Rodden (2003), “transfer-dependent governments face weak incentives to be fiscally responsible, since it is more rewarding to position themselves for a bailout”. According to Von Hagen and Eichengreen (1996), budget constraints become softer when subnational governments are more dependent on revenues over which they have no discretion, meaning that subcentral decision makers could rationally expect to be bailed out ex post by central government if they run up unsustainable levels of debt. As joint taxes (Ertragsanteile) give local jurisdictions only slightly more decision-making powers than unconditional grants, we also expect municipalities in Austria, which receive higher per capita shares of joint taxes, to have a higher propensity towards borrowing. Hypothesis 1. We expect a positive relationship between higher transfer dependency and municipal borrowing. 6 Due to the special status of the city of Vienna in the FAG, we further exclude it from any analysis. We also exclude 10 “statutory cities” to which special equalization schemes apply (Eisenstadt, Rust, Klagenfurt am Woerthersee, Villach, Krems an der Donau, St. Poelten, Waidhofen an der Ybbs, Wiener Neustadt, Linz, Steyr, Wels).

6

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

4. The empirical approach Our dependent variable is borrowing per capita, as defined in net terms, i.e., we subtract repayments of debts from ‘new’ municipal borrowing during a fiscal year. Since the assignment of tax revenues is a function of population, with a clear cutoff point at the level of 10,000 inhabitants, a natural way to explore this quasi-experimental setup is to use an RD design. The municipalities are assigned to two different groups zi ∈ {0, 1} depending on population. The outcome, in our case, municipal borrowing, can, be denoted thus:

⎧ y (0) ifpop ≤ c i yi = ⎨ i ⎩ yi (1) ifpopi > c, ⎪

(1)

where c = 10, 000 and popi denote the population size of each municipality. Assignment is a deterministic function of population size and allows us to implement a sharp RD design (Imbens and Lemieux, 2008). We analyze the sharp discontinuity in the conditional expectation of borrowing regarding the given population size in order to identify the average causal effect of the treatment:

τ = limE[yi |popi = x ] − limE[yi |popi = x ]. x↓c

(2)

x↑c

This term defines the local average treatment effect at the threshold (Imbens and Lemieux, 2008):

τ = E[yi (1) − yi (0)|popi = c].

(3)

In all specifications, we only include municipalities with between 2000 and 20,000 inhabitants. In smaller entities, forces such as suboptimal community size and/or local management incompetence may be at work and possibly drive borrowing behavior. Frequently, smaller communities are said to be inherently inefficient due to limitations in hiring well-trained administrative staff. For example, smaller communities frequently have difficulties in finding qualified candidates for local mayors or other public posts7. Hence, we prefer specifications that exclude these smaller entities, even at the cost of having fewer observations. Our results, however, do not change when we include municipalities than have less than 2000 inhabitants, as is expected in the case of local design. Additionally, we report the results for the subsample of municipalities with a city status and less than 20,000 inhabitants (but over 2000). By definition, every community with over 10,000 inhabitants is a “city”, but smaller municipalities are sometimes assigned a city status as well (see Section 3). Municipalities with a city status may serve de facto as regional centers, providing neighboring communities with regional public goods8. Hence, the smaller sample of cities could be more homogeneous with respect to spending composition. In order for the RD design to identify the local treatment effect, some assumptions must be met. Firstly, the treatment assignment must be a monotone deterministic function of the assignment variable. This holds in our case, as the exogenously established FAG fully determines assignment to treatment. Secondly, a population threshold of 10,000 may simultaneously determine other policy-determining factors, making results difficult to interpret as caused by a softened budget constraint. As mentioned above, the constitutional assignment of task responsibilities to municipalities does not depend on population size, while city status has no formal legal consequences. We document potential confounded treatment in Table 18 in the Appendix9. A potential distortion to our setup could come from a discontinuity of the council size at a threshold of 10,000 inhabitants in Carinthia, Lower Austria and Tyrol. Council size alone could cause higher expenditure and borrowing (see, e.g., Egger and Koethenbuerger, 2010), as happens with coalition governments (e.g., Persson and Tabellini, 2005, 2004; Blume et al., 2009). The common-pool problem could become more severe with larger councils, which would lead to disproportionally higher spending at a cutoff of 10,000. In all states, council size increases along with population, but most states do not employ a specific population threshold of 10,000. A possible way to deal with this issue is to include council size in regressions with control variables. Additionally, we run placebo tests as a robustness check in relation to other council-size thresholds (7000 in Lower Austria and 6000 in Carinthia and Tyrol). Moreover, we report robustness checks excluding these states. Manipulation and sorting should not be issues in our case. In Austria, population figures used for the allocation of funds across jurisdictions are collected by the Federal Statistical Office (“Statistik Austria”). When revising its fiscal equalization scheme in 2007, the Federal Ministry of Finance initiated a debate on the population accounting method. Due to the longer time period involving unadjusted census data, pre-reform statistics did not adequately reflect changes in size and structure of the local population. In 2011, the hitherto employed census method with outdated population figures was replaced by more accurate and annually available registry data, provided by the municipalities and collected under the strict supervision of Statistik Austria. The revised system increased the level of transparency and accuracy of population figures. Meanwhile, due to a clearing process by the National Statistics Office, municipal opportunities to manipulate data are rather small (see Blöchliger and Vammalle, 2012). Nevertheless, we need to establish whether manipulation of the running variable is an issue in our study. Fig. 5 presents the density estimation using the McCrary (2008) procedure and test, as implemented in R software, using a fixed bin size of 100, instead 7 8 9

See, e.g., http://derstandard.at/2000029520805/Debatte-ueber-Gemeindefusionen-Tirol-gehen-die-Buergermeister-aus All cities in the sample have had this status throughout the entire analyzed period. Other population thresholds are reported in Table 19 in the Appendix.

7

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Fig. 5. McCrary density test; bin length=100.

Fig. 6. McCrary density test without Vorarlberg; bin length=100. Table 1 Revenue mix of Austrian municipalities in 2014 (in euros per capita). Inhabitants

Revenue sharing

Own taxes

Fees

Less than 2500 Between 2501 and 5000 Between 5001 and 10,000 Between 10,001 and 20,000 Between 20,001 and 50,000 More than 50,000

786 795 782 910 1093 1224

325 418 526 559 554 746

249 290 320 313 335 216

Source: Gemeindefinanzbericht 2014.

of the automatic size, in light of the many empty bins around the threshold.10 The estimated difference in the density heights θ equals 1.01 with a standard error of 0.203. The p-value of the test equals 0.312, thus we cannot reject the null hypothesis that the running variable is not sorted. One potential critique of this approach is that, since population is a discrete variable, any observations that lie exactly at the cutoff are automatically assigned to the first bin to the right of the threshold, which potentially biases the test statistic. Eggers et al. (2015) suggest that, in such a scenario, a McCrary test at the cutoff value of -0.5 should be run. As such, the value of the test statistic θ is 0.335, with a standard error of 0.188, resulting in a p-value of 0.0751. After a closer look at the data, we establish that one particular state, Vorarlberg, distorts the results. After we run the test excluding these observations, θ equals 0.175 with a standard error of 0.195, resulting in a p-value of 0.369 (see Fig. 6). In the robustness section, we report the results excluding these observations from the estimation. Finally, as mentioned above, accounting of the population changed in 2011, as census data

10

We are grateful to an anonymous referee for pointing this out.

8

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Table 2 Local linear regression: (1) and (2) all municipalities; (3) and (4) municipalities with city status. (1)

(2)

(3)

(4)

Cutoff=1

117.83** (2.19)

110.30** (2.29) [2.32]

104.18*** (1.83)

112.36** (1.83) [2.09]

Normalized population

−49.11* (−1.96)

−63.78* (−1.90)

−30.49 (−1.29)

−12.56 (−0.39)

Cutoff=1 ×Normalized population

−39.87 (−0.84)

35.85 (0.69)

Constant

−25.08 (−1.09)

−39.12 (−1.32)

−39.51 (−1.38)

−21.94 (−0.62)

Observations 2000–20,000 Bandwidth Observations in h

6525 1580 351

6525 1580 351

1919 2019 274

1919 2019 274

Cameron et al. (2011) robust standard errors, clustered at state and year levels in (). Calonico et al. (2014) robust standard errors in []. t- and z-statistics in parentheses; significance: *** 0.01 ** 0.05 * 0.1. Note: Only municipalities with a population size over 2000 and under 20,000 are included.

Fig. 7. Net borrowing per capita; local linear fit, bandwidth=1,000; upper panel: all municipalities; lower panel: cities.

9

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Table 3 Local polynomial regression; bandwidth=CCT; municipalities with between 2000 and 20,000 inhabitants. (1) Net borrowing per capita Uniform kernel

(2) Net borrowing per capita Triangular kernel

p=1

110.34** (2.32)

71.93** (1.98)

h b p=2

1,581 3,013 95.38* (1.67)

2,280 3,869 65.61* (1.74)

h b p=3

3,148 4,132 85.85* (1.66)

3,477 4,987 50.87 (1.15)

h b p=4

4,656 6,222 66.80 (1.21)

3,489 4,292 49.11 (0.85)

h b Observations

5,278 6,729 6,572

4,982 5,780 6,572

Calonico et al. (2014) robust confidence intervals. z-statistics in parentheses; significance: *** 0.01 ** 0.05 * 0.1.

Table 4 Local polynomial regression; bandwidth=CCT; cities with between 2,000 and 20,000 inhabitants. (1) Net borrowing per capita Uniform kernel

(2) Net borrowing per capita Triangular kernel

p=1

112.36** (2.14)

96.73** (2.05)

h b p=2

2026 2761 110.52* (1.67)

2351 3473 100.66** (1.95)

h b p=3

3331 4971 102.03* (1.88)

3189 4456 52.60 (0.67)

h b p=4

3,136 4,734 116.29 (1.55)

3,317 4,348 67.41 (0.96)

h b Observations

4125 5275 1919

4476 5438 1919

Calonico et al. (2014) robust confidence intervals. z-statistics in parentheses; significance: *** 0.01 ** 0.05 * 0.1.

were not frequently updated. When we perform a McCrary test for the period before 2011, θ equals 0.630 with a standard error of 0.203, resulting in a p-value of 0.121. This means that, although previously used figures may not have reflected the population development realistically, we do not find any statistical evidence of manipulation. In our case, municipalities in Vorarlberg seem to slightly distort the test statistic, with exclusion resulting in a p-value of 0.298 during the test. As the McCrary (2008) test relies on pre-binning of the data, it is sensitive to the choice of bin size, in particular, in the case of discrete variables, such as population (Eggers et al., 2015). We therefore test potential sorting with an alternative approach, as 10

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Table 5 Linear regression and linear weights; (1) and (2) all municipalities with more than 2000 inhabitants; (3) and (4) cities with more than 2000 inhabitants; with control variables. (1)

(2)

(3)

(4)

Cutoff=1

94.08*** (3.39)

66.87* (1.72)

164.94*** (3.39)

155.54*** (3.23)

Normalized population

−12.39*** (−3.10)

−16.29*** (−4.36)

−17.02*** (−2.89)

−20.66*** (−2.77)

Cutoff=1 ×Normalized population

11.25* (1.71)

7.34* (1.80)

Urbanization

9.80 (1.23)

10.76 (1.37)

23.38 (1.41)

24.52 (1.43)

% Post-working

-400.90 (−0.89)

-407.43 (−0.90)

−290.09 (−0.45)

−292.67 (−0.46)

% Pre-working

−132.44 (−0.38)

−129.35 (−0.38)

−162.81 (−0.25)

−156.33 (−0.24)

Population density

3.41 (1.58)

3.44 (1.61)

0.24 (0.09)

0.40 (0.15)

L. debt per capita

−0.00 (−0.88)

−0.00 (−0.91)

−0.00* (−1.76)

−0.00* (−1.86)

Median income

−0.01*** (−3.56)

−0.01*** (−3.50)

−0.02* (−1.93)

−0.02* (−1.94)

Gini income

125.16 (0.39)

118.65 (0.38)

−574.85 (−1.31)

−563.90 (−1.30)

Payroll tax per capita

0.00 (0.08)

0.00 (0.06)

−0.01 (.)

−0.01 (−1.40)

Real estate tax per capita

0.27 (0.50)

0.28 (0.51)

1.31 (1.56)

1.31 (1.56)

Other taxes per capita

0.05 (0.24)

0.05 (0.26)

0.30 (1.03)

0.31 (1.08)

Council size

0.25 (0.18)

1.88 (.)

−2.85** (−2.33)

−1.43 (−1.03)

Council fragmentation

23.88 (0.24)

21.42 (0.22)

−67.35 (−0.43)

−70.35 (−0.45)

Abs. maj. SPOE

−42.73*** (−5.27)

−41.26*** (−4.93)

2.20 (0.07)

2.56 (0.08)

Abs. maj. OEVP

−39.09 (.)

−37.20 (.)

−0.31 (−0.01)

0.76 (0.02)

Abs. maj. FPOE

−82.95*** (−2.86)

−78.58** (−2.47)

0.00 (.)

0.00 (.)

Coalitions

−31.42*** (−4.19)

−29.67*** (−3.39)

−13.65 (−0.46)

−12.39 (−0.40)

Constant

230.41 (1.20)

165.14 (0.84)

440.80*** (2.88)

381.16*** (4.94)

Observations Time effects State effects

5973 YES YES

5973 YES YES

1766 YES YES

1766 YES YES

Cameron et al. (2011)robust standard errors, clustered at state and year levels in (). t-statistics in parentheses; significance: *** 0.01 ** 0.05 * 0.1. Note: Only municipalities with a population size over 2000 and under 20,000 are included.

11

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Table 6 Linear regression and linear weights: investment expenditure at the threshold; (1) and (2) all municipalities with more than 2000 inhabitants; (3) and (4) cities with more than 2,000 inhabitants. (1) Investment per capita

(2) Investment per capita

(3) Investment per capita

(4) Investment per capita

Cutoff=1

35.13 (1.04)

−17.80 (−0.50)

31.89 (0.59)

22.26 (0.35)

Normalized population

−2.69 (−0.59)

−9.15 (−1.56)

3.35 (0.43)

0.16 (0.02)

Cutoff=1 ×Normalized population

20.45 (1.50)

6.76 (0.51)

Urbanization

−8.81 (−0.41)

−7.49 (−0.36)

12.82 (0.27)

13.75 (0.30)

% Post-working

623.34 (1.43)

619.81 (1.44)

669.77*** (6.31)

675.50*** (5.99)

% Pre-working

48.06 (0.13)

55.37 (0.15)

157.61 (0.37)

164.91 (0.39)

Population density

1.23 (0.68)

1.37 (0.72)

−3.49 (.)

−3.31*** (−5.43)

L. debt per capita

0.04*** (4.68)

0.04*** (4.69)

0.05*** (3.35)

0.05*** (3.40)

Median income

−0.00 (−1.42)

−0.00 (−1.28)

0.00 (0.87)

0.00 (0.82)

Gini income

−411.08 (−0.88)

−425.26 (−0.93)

−578.35 (−0.80)

−570.11 (−0.76)

Payroll tax per capita

0.14*** (3.44)

0.14*** (3.35)

0.01 (0.17)

0.01 (0.15)

Real estate tax per capita

−0.17 (−0.39)

−0.16 (−0.38)

0.78 (.)

0.79 (.)

Other taxes per capita

0.28 (1.41)

0.28 (1.45)

0.16 (0.66)

0.17 (0.70)

Council size

−3.76** (−2.31)

−1.04 (−0.42)

−10.34*** (−3.31)

−9.07* (−1.87)

Council fragmentation

13.16 (0.35)

9.03 (0.24)

45.13** (2.23)

41.98** (2.06)

Abs. maj. SPOE

−43.97** (−2.09)

−41.60* (−1.82)

−64.60** (−1.99)

−63.89* (−1.93)

Abs. maj. OEVP

2.54 (0.13)

5.43 (0.26)

−10.82 (−0.46)

−9.60 (−0.42)

Abs. maj. FPOE

−244.89*** (-13.30)

−244.58*** (-13.73)

0.00 (.)

0.00 (.)

Coalitions

−17.42 (−0.95)

−14.90 (−0.77)

−9.86 (−0.36)

−8.80 (−0.31)

Constant

357.80 (1.33)

248.78 (0.78)

317.61 (1.28)

263.57 (0.75)

Observations Time effects State effects

7,811 YES YES

7,811 YES YES

2,144 YES YES

2,144 YES YES

Cameron et al. (2011) robust standard errors, clustered at state and year levels in (). t-statistics in parentheses; significance: *** 0.01 ** 0.05 * 0.1. Note: Only municipalities with a population size over 2,000 and under 20,000 are included.

12

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Table 7 Linear regression and linear weights: the effect of investments on borrowing; (1) and (2) all municipalities with more than 2000 inhabitants; (3) and (4) cities with more than 2000. (1)

(2)

(3)

(4)

Cutoff=1

55.03 (1.49)

69.40** (1.97)

116.71*** (2.98)

123.12*** (3.10)

Normalized population

−8.37** (−2.14)

−6.19* (−1.67)

−15.37** (−2.54)

−12.73*** (−2.90)

−6.08*** (−2.58)

Cutoff=1 ×Normalized population

−5.21 (-0.78)

Investment per capita

0.51*** (9.89)

0.51*** (9.89)

0.54*** (15.79)

0.54*** (15.84)

Urbanization

3.33 (0.21)

2.77 (0.18)

−3.37 (−0.14)

−4.23 (−0.17)

% Post-working

−514.84** (−2.25)

−511.83** (−2.23)

−92.02 (−0.33)

−90.16 (−0.32)

% Pre-working

−69.04 (−0.35)

−71.06 (−0.35)

−5.96 (−0.01)

−10.83 (−0.02)

Population density

1.80 (0.94)

1.79 (0.93)

3.45 (1.26)

3.35 (1.22)

L. debt per capita

−0.03*** (−5.27)

−0.03*** (−5.16)

−0.03*** (−4.87)

−0.03*** (−4.51)

Median income

−0.01*** (−3.55)

−0.01*** (−3.69)

−0.02*** (−3.18)

−0.02*** (−3.13)

Gini income

332.93** (2.09)

337.06** (2.11)

−55.38 (−0.13)

−63.39 (−0.15)

Payroll tax per capita

−0.05* (−1.68)

−0.05* (−1.67)

0.01 (0.35)

0.01 (0.41)

Real estate tax per capita

0.07 (0.21)

0.07 (0.20)

0.73 (1.16)

0.72 (1.16)

Other taxes per capita

−0.08 (−0.93)

−0.08 (−0.96)

0.06 (0.20)

0.05 (0.18)

Council size

2.66*** (3.64)

1.75 (.)

3.72*** (5.33)

2.70 (.)

Council fragmentation

−44.65 (−0.52)

−43.24 (−0.51)

−144.59 (−0.96)

−142.39 (−0.94)

Abs. maj. SPOE

2.56 (0.20)

1.71 (0.13)

49.12* (1.67)

48.79* (1.65)

Abs. maj. OEVP

−25.48 (.)

−26.58 (.)

23.78 (0.97)

22.92 (0.89)

Abs. maj. FPOE

−23.85 (−1.47)

−26.26 (−1.59)

0.00 (.)

0.00 (.)

Coalitions

−11.43 (.)

−12.43 (.)

1.23 (0.07)

0.26 (0.02)

Constant

−21.87 (−0.13)

14.52 (0.09)

59.69 (0.21)

102.39 (0.36)

Observations Time effects State effects

5,466 YES YES

5,466 YES YES

1,620 YES YES

1,620 YES YES

Cameron et al. (2011) robust standard errors, clustered at state and year levels in (). t-statistics in parentheses; significance: *** 0.01 ** 0.05 * 0.1 Note: Only municipalities with a population size over 2000 and under 20,000 are included.

13

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Table 8 Local polynomial regression including covariates and triangular kernel; bandwidth=CCT. (1) Net borrowing per capita All municipalities

(2) Net borrowing per capita Cities

p=1

109.89*** (2.95)

136.85** (2.50)

h b Eff. observations p=2

2107 4037 377 109.65*** (2.72)

1855 ,322 223 137.13** (2.35)

h b Eff. observations p=3

3725 5575 775 36.12 (0.51)

3177 5069 399 90.57 (1.11)

h b Eff. observations Observations

2789 4126 549 5488

3079 4200 388 1626

Calonico et al. (2016) robust confidence intervals including covariates: lagged debt per capita; investment per capita; median income; Gini of income z-statistics in parentheses; significance: *** 0.01 ** 0.05 * 0.1.

recently developed by Cattaneo et al. (2016b) and implemented by Cattaneo et al. (2016a), based on the local polynomial density estimator, which does not require pre-binning of the data and provides improvements in both size and power, compared to other approaches. The reported test statistic with a bias-corrected standard error equals -1.377, with a p-value of 0.16.11 Table 9 Local polynomial regression including covariates and triangular kernel; bandwidth=CCT; Vorarlberg excluded. (1) Net borrowing per capita All municipalities

(2) Net borrowing per capita Cities

p=1

164.41** (1.97)

121.03** (2.38)

h b Eff. observations p=2

1455 2948 236 106.12** (2.02)

1774 3208 200 134.81** (2.36)

h b Eff. observations p=3

3182 4719 593 42.83 (1.01)

3099 4990 389 108.65 (1.15)

h b Eff. observations Observations

2789 4126 549 5488

3104 4268 389 1570

Calonico et al. (2016) robust confidence intervals including covariates:. lagged debt per capita; investment per capita; median income; Gini of income 3lz-statistics in parentheses; significance: *** 0.01 ** 0.05 * 0.1.

11 The restricted model has equal bandwidths on both sides of the threshold (bandwidth=1546.320); the conventional p-value is 0.3630, which is similar to the value of the McCrary (2008) test; the restricted model assumes that F(·) and higher order derivatives are equal for both groups at the cutoff, resulting in a more conservative testing procedure (for more detail, see Cattaneo et al., 2016b).

14

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Fig. 8. Local polynomial regression, p=1; sensitivity towards the choice of bandwidth =1000, …, 9000 .

Fig. 9. Local polynomial regression, p=1; cities only; sensitivity towards the choice of bandwidth =1000, …, 9000 .

An RD design can be implemented either with parametric or nonparametric models (see Imbens and Lemieux, 2008; Lee and Lemieux, 2010; Calonico et al., 2014, for further details, see). Nevertheless, nonparametric local linear regression with a uniform (i.e., rectangular) kernel yields results equivalent to those estimated by:

yit = α + β0popit + β1zit + β2zit × popit + Γ Xit + εit ,

(4)

on the range [z − h , z + h], where h is the optimal bandwidth. The difference is in the standard errors from the correction, as proposed by Calonico et al. (2014); we will report both uncorrected (two-way clustered at state and year level) and corrected standard errors. The β1 coefficient represents the RD, while β2 represents the change in the slope of the relationship at the cutoff. Implementation in this form has the advantage that it can be further augmented with control variables X . A standard parametric approach, which assumes the same slope of the regression line on both sides of the threshold, is also reported for comparison. Alternatively, one can implement the RD design using local (linear or polynomial) nonparametric regression with a non-uniform kernel. Although we concentrate on the estimations with control variables, we also report further results and a graphical analysis around the threshold. The choice of bandwidth is conducted with the Calonico et al. (2014) (CCT) procedure, which is a more robust confidence interval estimator for average treatment effects at threshold points in RD designs. In our regressions, we control for other factors, which could possibly explain municipal expenditure and borrowing per capita. Firstly, we include the lagged level of debt per capita, as higher initial debt may restrict future borrowing behavior. We hence expect a negative coefficient. Secondly, population size itself is a running variable, which could also affect the borrowing level. The impact of population size on public borrowing and public expenditure cannot be easily predicted, as it depends on whether the demand for public goods grows faster or slower than the population (Werck et al., 2008). A number of contributions has found a negative relation with public spending in categories such as transport and communications, healthcare, defense and communal services (see, e.g., Costa-Font and Moscone, 2009). 15

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Fig. 10. Local polynomial smoothing; bandwidth=1500; all municipalities.

As for socioeconomic and geographic factors, we include population density, as well as the LAU2 urbanization typology of the European Union to control for differences in service provision between rural and urban areas (Sanz and Velazquez, 2002). We also include the fractions of those aged below 15 and above 65 years in order to control respectively for the importance of local provision of (pre-)school services and social/health services for the elderly. The purpose is to test whether these two groups of electors benefit to a greater proportional extent from the provision of certain public goods, such as healthcare or education, in comparison to the rest of the citizenry (Hayo and Neumeier, 2012), which in turn can incentivize municipalities to overborrow. Another variable often employed in studies concerning the determinants of public expenditure is the average or median level of income, which seeks to capture the per capita wealth of a community and may reflect its demand for public goods and services. Moreover, local income distribution may affect part of the expenditure, particularly the levels of expenditure on social policies. We include, as control variables, the average income in a municipality and the Gini coefficients of net disposable income. Finally, political variables potentially affect the level of municipal borrowing. Increasing council size could lead to a more severe common-pool problem in the council, while positively affecting deficits and thus the public debt (Persson and Tabellini, 2005, 2004). As mentioned before, Egger and Koethenbuerger (2010) apply a similar RD design to municipalities in the German state of Bavaria, finding a robust positive impact of local council size on local spending. We also include dummy variables for parties with an absolute majority on the council (OEVP - Christian Democrats, SPOE - Social democrats, FPOE – the populist Freedom Party), as well as instances of coalition governments 12 . Moreover, we account for council fragmentation, which may be associated with higher spending and propensity to borrow. We measure council fragmentation as the Herfindahl-Hirschman index of seat shares on the municipal council. Time and state effects are added to control for the time development of the fiscal variables over the business cycle, as well as account for the above-described institutional differences between the states.

12 The omitted reference group is formed by those cases in which a party, other than the three above-mentioned, has an absolute majority on the local council, typically, a local interest party.

16

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Fig. 11. Local polynomial smoothing; bandwidth=1500; cities.

5. Results 5.1. Baseline results In this section, we present the estimated local average treatment effects of the change in the vertical fiscal gap on the levels of new borrowing by municipalities Table 1. Linear regression results for a simple RD design are presented in Table 213. Linear regression results with additional control variables are presented in Tables 5 and 7. The main results of our estimates without additional covariates are reported in Table 2. Columns (1) and (3) correspond to linear regressions, assuming the same slope of the fit line on both sides of the threshold, whereas Columns (2) and (4) are the local linear regressions equivalent to nonparametric estimations with a uniform kernel. Across all specifications, we observe significant local treatment effects. The simple linear regressions without covariates in Table 2 show a treatment effect of about EUR 100 in increased net borrowing per capita for the sample of municipalities between 2000 and 20,000 inhabitants. Given that the change in the vertical fiscal gap is about 10 percentage points at the threshold, this result suggest an additional EUR 9-12 in new debt per capita for every one percentage point change in the fiscal gap. Our results also indicate certain ‘scale effects’, as (normalized) population size is always negatively related to net borrowing per capita. The size of the effect of about EUR 100 in new net borrowing is also rather significant in economic terms. As reported in Table 17 in the Appendix, net borrowing per capita varies between EUR -2250 and 18,949, with a mean of EUR 105. The effect at the discontinuity, also visualized in Fig. 7, is the difference between zero (or perhaps negative) new borrowing and positive new borrowing. Meanwhile, municipalities just below the threshold tend not to issue new debt or repay old debt, whereas the ones just above the threshold issue positive levels of new debt. Since the average debt level is EUR 1954 per capita, EUR 100 in new borrowing is equal to about 5% of the average debt. The results of local regressions with triangular and uniform kernels of higher-order polynomials are presented in Tables 3 and 4, which further validate the findings. For the sample of all municipalities, the results are rendered insignificant above the third-order polynomial; in the case of cities, they also remain significant with regard to the third-order polynomial and uniform kernel. Since

13

In order to present the results, the population cutoff has been normalized to 0.

17

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Table 10 Placebo test; cutoff 8500 bandwidth=CCT; uniform kernel. (1) Net borrowing per capita All municipalities

(2) Net borrowing per capita Cities

p=1

−2.53 (−0.08)

−29.18 (−0.51)

p=2

−16.87 (−0.17)

-25.18 (−0.33)

p=3

15.30 (0.37)

8.33 (0.08)

p=4

−38.13 (−0.66)

358.45 (1.37)

Observations

6,572

1,919

(1) Net borrowing per capita All municipalities

(2) Net borrowing per capita Cities

p=1

−23.30 (−0.61)

−40.44 (−0.87)

p=2

−17.08 (−0.49)

2.30 (0.05)

p=3

−15.92 (−0.40)

−4.66 (−0.07)

p=4

−24.32 (−0.55)

−41.12 (−0.56)

Observations

6,572

1,919

Calonico et al. (2014) robust confidence intervals. z-statistics in parentheses; significance: *** 0.01 ** 0.05 * 0.1.

Table 11 Placebo test; cutoff 11,500; bandwidth=CCT; uniform kernel.

Calonico et al. (2014) robust confidence intervals. z-statistics in parentheses; significance: *** 0.01 ** 0.05 * 0.1.

Table 12 Placebo test; cutoff 7000; bandwidth=CCT; uniform kernel; Lower Austria only. (1) Net borrowing per capita All municipalities

(2) Net borrowing per capita Cities

p=1

−20.48 (−0.33)

68.08 (0.90)

p=2

21.63 (0.33)

32.44 (0.29)

p=3

24.84 (0.41)

40.56 (0.28)

p=4

37.91 (0.30)

69.62 (0.30)

Observations

2,452

805

Calonico et al. (2014) robust confidence intervals. z-statistics in parentheses; significance: *** 0.01 ** 0.05 * 0.1.

18

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Table 13 Placebo test; cutoff 6000; bandwidth=CCT; uniform kernel; Carinthia and Tyrol only. (1) Net borrowing per capita All municipalities

(2) Net borrowing per capita Cities

p=1

86.48 (1.49)

−1.23 (−0.21)

p=2

8.99 (0.08)

176.01 (0.87)

p=3

−32.09 (−0.20)

−160.02 (−0.76)

p=4

131.11 (1.01)

−226.20* (−0.16)

Observations

1,171

507

Calonico et al. (2014) robust confidence intervals. z-statistics in parentheses; significance: *** 0.01 ** 0.05 * 0.1.

Table 14 Local randomization inference; all municipalities. Window

Diff. in means

P-val.

K-Sa

Rank sumb

179 257 335 413 491 569 647 725

244.908 184.96 205.450 166.273 140.408 101.112 66.56 7.308

0.00 0.00 0.00 0.00 0.00 0.00 0.02 0.81

1.00 0.80 0.78 0.64 0.63 0.47 0.35 0.21

−3.55 −3.56 −4.07 −4.25 -4.04 −3.31 −2.65 0.94

a b

Kolmogorov-Smirnov statistic. Wilcoxon-Mann-Whitney studentized statistic.

Table 15 Local randomization inference; cities. Window

Diff. in means

P-val

K-Sa

Rank sumb

179 257 335 413 491 569 647 725

221.893 183.399 213.674 162.966 119.066 52.53 15.915 −21.171

0.00 0.00 0.00 0.00 0.00 0.12 0.58 0.58

1.00 0.80 0.92 0.68 0.60 0.33 0.14 0.47

−2.71 −2.73 −3.38 −3.41 -2.93 −1.67 −0.13 2.49

a b

Kolmogorov-Smirnov statistic. Wilcoxon-Mann-Whitney studentized statistic.

recent work on RD by Gelman and Imbens (2014) suggests that estimations using polynomials of an order higher than two should be interpreted with caution, we cautiously conclude that the results remain valid. Graphically, we can present the discontinuity using nonparametric linear fits on both sides of the threshold, for which we use a kernel bandwidth of 1000. Results are presented in Fig. 7. After adding covariates to Table 5, the treatment effects are still significant in both samples. Turning very briefly to the behavior of our controls, we do not find any robust effects of urbanization or population density on net borrowing. Lagged debt per capita has an expected negative, sign but is not significant.Median income is negatively related to local net borrowing per capita in all specifications, indicating that, on average, poorer entities are indeed more reliant on higher deficits and debt. This may be due to both higher social spending or reduced revenues per capita. Results for local income distribution are inconsistent and far from significant across specifications.

19

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Fig. 12. Local randomization inference; sensitivity; all municipalities.

Fig. 13. Local randomization inference; sensitivity; cities.

We find no consistent results regarding the impact of own local taxes on debt levels. We were also unable to find any effects of municipal council size or council fragmentation on borrowing behavior. The effects of an absolute majority of one of the large national parties are close to significant and appear with a negative sign in Columns (1) and (2), suggesting that local voter associations are associated with higher borrowing, but are rendered insignificant if we only consider cities. This is likely due to independent local voter associations being more popular in smaller communities. Coalition governments of large parties are also not associated with higher net borrowing. 5.2. Role of investment spending In theory, municipal borrowing should be strongly related to local investment expenditure. In particular, spending on infrastructure may be financed through higher deficits, with larger investment projects (in relation to budget size) possibly inflating local borrowing in the respective fiscal year. Table 6 shows that investment does not jump at the 10,000 threshold, which confirms the validity of the empirical approach, while investment activities explain higher borrowing in general, as shown in Table 7. Table 7 provides evidence that additional investment spending indeed contributes to higher net per capita borrowing at the local level. In all specifications, investment spending per capita has a positive and significant effect on net borrowing. The effects are of a very similar size (+ EUR 0.5 per capita) throughout. Treatment effects become slightly weaker in the full sample of municipalities with more than 2000 inhabitants. Columns (3) and (4), which display the results for municipalities with city status only, the discontinuity remains significant at the 5% level, although coefficients are also reduced by almost 50%. This is consistent with the coefficient of the investment variable, meaning that for each EUR 1 of net borrowing, about EUR 0.5 can be attributed to investment. We cautiously conclude that almost half of the observed discontinuity works through an investment channel. The latter observation allows us to conclude that a substantial part (around 50%) of the additional borrowing above the threshold is not spent on 20

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Table 16 Spatial results. (1)

(2)

(3)

(4)

SL. net borrowing per capita

0.53* (1.95)

0.49* (1.80)

0.77** (2.10)

0.75** (2.03)

Cutoff

74.17*** (3.75)

67.58*** (3.23)

67.58*** (3.23)

70.30*** (3.13)

Normalized population

−10.19*** (−6.08)

−11.09*** (−5.64)

−11.09*** (−5.64)

−11.43*** (−3.37)

Urbanization

14.82 (1.53)

14.82 (1.53)

12.96 (1.30)

% Post-working

−350.32 (−1.51)

−350.32 (−1.51)

−264.00 (−1.12)

% Pre-working

−69.13 (−0.48)

−69.13 (−0.48)

−61.75 (−0.42)

Population density

3.23** (2.11)

3.23** (2.11)

3.01* (1.94)

L. debt per capita

−0.00 (−1.38)

−0.00 (−1.38)

−0.01* (−1.65)

Median income

−0.01*** (−4.66)

−0.01*** (−4.66)

−0.01*** (−4.14)

Gini income

190.99 (1.15)

190.99 (1.15)

44.92 (0.25)

SL. pop. density

−0.41 (−0.89)

−0.42 (−0.92)

SL. % Pre-working

43.71 (1.42)

43.40 (1.40)

SL. % Post-working

−37.75 (−1.49)

−38.24 (−1.51)

Payroll tax per capita

0.01 (0.68)

Real estate tax per capita

0.21 (1.19)

Other taxes per capita

0.03 (0.60)

Council size

0.33 (0.18)

Council fragmentation

11.10 (0.29)

Abs. maj. SPOE

−41.82*** (−2.64)

Abs. maj. OEVP

−35.35** (−2.51)

Abs. maj. FPOE

−90.46 (−0.38)

Coalitions

−31.79** (−2.32)

Constant

−4.01 (−0.37)

195.99** (2.38)

21

195.99** (2.38)

199.96* (1.91) (continued on next page)

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Table 16 (continued)

Observations Time effects State effects

(1)

(2)

(3)

(4)

6525 YES YES

5999 YES YES

5999 YES YES

5973 YES YES

Note: Only municipalities with a population size over 2000 and under 20,000 are included.

investment activities, but on current consumption. This observation is at odds with the “golden rule of public finance”, which states that debt financing should be used for investment purposes only. Initial per capita debt now shows a small, but statistically significant, negative effect on municipal borrowing. Having established that the most important fiscal variables affecting the borrowing levels are lagged debt levels and investment per capita, as well as median income and the Gini coefficient of income, we again present local polynomial regressions of different orders to test the sensitivity of the results, while including these variables. As presented by Calonico et al. (2016), this should not affect the estimated coefficients by much, but could affect the variance and thus significance. The results are reported in Table 8. As can be seen, up to a second-order polynomial, the discontinuity is clearly visible. A third-order polynomial is, unlike the baseline case, no longer significant. The same holds if we exclude municipalities in Vorarlberg Table 9. 5.3. Sensitivity Figs. 8 and 4 present the sensitivity of the treatment effect towards the choice of bandwidth in the local polynomial regressions. The results remain stable for the bandwidth choices other than the optimal bandwidth, i.e., between 2000 and 6000 inhabitants, around the threshold Fig. 10. For a larger bandwidth, the results become insignificant Fig. 11. Further falsification tests are presented in Tables 10 and 11 at population thresholds of 8500 and 11,500 for the entire sample. There is no evidence that spurious jumps in net borrowing can be observed at these thresholds. Regarding continuity of net borrowing at the population thresholds at which council size changes, regression results already suggest that no changes in borrowing are associated with changes in the size of a municipal council Table 12. The results of the falsification tests for two population thresholds, at which the council size changes at the 10,000 threshold in the three states of Carinthia, Tyrol and Lower Austria, further confirm that municipal council size does not appear to influence levels of borrowing Table 13. As a further sensitivity test, we re-estimate the results using the randomization inference framework, as developed by Cattaneo et al. (2015) and Cattaneo et al. (2016c). As the authors note, only a few observations may be available that are close enough to the threshold at which local randomization is plausible, which means that standard large-sample procedures may be suspect. The proposed methodology is intended as complementary to and a robustness check on standard RD inference approaches (Cattaneo et al., 2015). We choose the optimal window size for the inference, based on exogenous covariates, and also test the results for diverse window lengths. The results presented in Tables 14 and 15 show the local randomized inference in the very close neighborhood of the threshold, as well as further confirm that discontinuity is found at a cutoff of 10,000, while significant differences in the levels of net new borrowing per capita can be observed. Above the window of about 700 inhabitants, the results are no longer significant, but become significant again at the window of about 1000 inhabitants, as indicated in the sensitivity graph (Figs. 12 and 13), thereby confirming previous calculations that use bandwidths above 1500. The drop for the window between 700 and 1000 inhabitants is associated with a small number of observations in the analyzed part. Finally, when it comes to public expenditure at the municipal level, spillovers into neighboring communities represent an important factor. Local centers of activity typically provide infrastructure and services, from which neighboring communities also benefit, such as kindergartens, schools and old people's homes. Normalizing service expenditure in per capita terms, as well as controlling for the age structure of a community, does not resolve the issue, as users of services will typically be registered in their home community when using the resources of their neighboring community. To account for the possibility of spillovers, we need to directly include spatial components in our regression model. As a final robustness check, we include a parametric regression with the following model:

yit = α + β0popit + β1zit + ρWyjt + Γ Xit + Φ WZjt + εit ,

(5)

εit = λ Wεjt + ξij

(6)

where:

and Φ denotes the spatial dependence of independent variables Z in municipality j and outcome y in municipality i, ρ is the impact of the spatial lag of the dependent variable, and W is the weighting matrix. This general specification also allows for the spatial dependence of the error term. A GMM-IV estimator for this model has been developed by Drukker et al. (2009) and Arraiz et al. (2010) for cross-sectional cases. We implement this estimator as a pooled cross-sectional model, adding time and state effects. The results are presented in Table 16. 22

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Including the spatial lag of the dependent variable reveals that net borrowing is indeed strongly spatially correlated. Nevertheless, the local treatment effect is still significant at the 1% level, although the size of the coefficient is reduced to about EUR 50-75. 6. Conclusions The second-generation fiscal federalism literature theoretically reports that subcentral budget constraints will be softer when local governments are more dependent on revenues over which they have no discretion (Von Hagen and Eichengreen, 1996; Goodspeed, 2002; Oates, 2005). As a consequence of reduced fiscal accountability, subcentral governments can expect to be bailed out by the central government and therefore tend to accumulate higher levels of debt. We test this conjecture using data from municipalities in Austria. The highly centralized fiscal framework in Austria is characterized by an exceptionally low financial responsibility among state and municipal governments to cover spending with own taxes and user fees. The excessive dependency of states and local entities on shared tax revenues and transfers from the central government is especially likely to produce such negative incentives. Employing an RD design for identification, which, for the first time, utilizes a unique and special regulation in the Austrian taxsharing agreement, we find that, in line with theoretical expectations, municipalities with higher revenue dependency contribute to higher net borrowing per capita. Almost half the observed discontinuity works through an investment channel. The size of the effect, i.e., about EUR 100 in new net borrowing, is also quite significant in economic terms, resulting in a difference between zero (or perhaps negative) new borrowing and positive new borrowing. While municipalities just below the threshold tend not to issue new debt or repay old debt, those just above the threshold issue positive levels of new debt. The findings reported in this paper thus confirm the notion that badly designed fiscal decentralization can be a potential source of overspending and deficit bias when subnational governments fail to fully internalize the cost of financing additional spending. Recent scholarly proposals for reforms to the Austrian fiscal framework, including the notion of substantially increasing subnational tax autonomy, hence point in the right direction: a smaller fiscal gap is supposed to harden the budget constraint for Austrian municipalities, which ought to contribute to higher fiscal discipline. That said, the future prospects for a fundamental reform of intergovernmental fiscal relations, in order to reduce fiscal imbalances in Austria, are rather weak. Both regional and local governments seem to be reluctant to adopt reforms that would substantially increase fiscal transparency and accountability. Attempts to only introduce highly moderate elements of local tax autonomy and fiscal competition at the state and the municipal level have failed over recent decade, as they were obviously not in the interests of political actors at all governmental levels. A very cautious step towards greater tax autonomy at the state level, as stipulated in the 2017 reforms to the FAG, is certainly not enough to generate strong incentives for subcentral fiscal prudence.

Appendix

Table 17 Summary statistics. Variable

Obs.

Mean

SD

Min.

Max.

Population Net borrowing per capita Revenue share per capita Debt per capita Pop. density Mean income (in EUR 1000) Gini income % Pre-working % Post-working Council size Council fragmentation

19,677 13,718 19,677 19,527 19,677 15753 15,470 19,677 19,677 19,677 15,041

2400 105 684 1954 1.24 19.42 .33 .162 .224 18.9 .489

2394 416 119 1751 2.16 2.02 .02 .0274 .0388 5.78 .136

53 −2250 128 .0692 .00981 8.34 .24 .0394 .108 9 .137

19,974 18,949 2191 44,685 38.7 29.80 .52 .331 .518 37 1

23

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Table 18 Regulations affecting the 10,000 threshold. State

Regulation

Source

Burgenland Carinthia

– The size of the council increases from 27 to 31 members

– §18(1) K–AGO

The wage of the mayor changes Obligatory higher education of the head of public services

§29(1-5) K–AGO §78(2) K–AGO

The size of the council increases from 33 to 37 members

§19(1) NÖ–GO

The council can choose a third vice mayor The council executive body increases from seven to eight members

§24(1) NÖ–GO §24(1) NÖ–GO

Salzburg Tyrol

– The size of the council increases from 19 to 21 members

– §22(1) TGO

Upper Austria

Obligatory higher education of the head of public services Necessary support for calling a referendum increases from 900 to 1400 inhabitants

§37(1) OÖ. GemO §38(1) OÖ. GemO

Vorarlberg

–

–

Lower Austria

Note

There are also changes for other thresholds

There are also changes for other thresholds Second vice mayor at the 2000 threshold

There are also changes for other thresholds

Table 19 Regulations affecting other thresholds. State

Regulation

Source

Burgenland Carinthia

Council size changes at 250, 500, 750, 1,000, 1500, 2000 and 3000 Council size changes at 2000, 3000, 6000, 10,000 and 20,000 Wage changes for the council and mayor at diverse thresholds

§15 Bgld–GemO §18(1) K–AGO §29(1-5) K–AGO

Lower Austria

Council size changes at 500, 1,000, 2000, 3000, 4,000, 5,000, 7,000, 10,000, 20,000 and 30,000 The council can choose a second vice mayor at 2000 The council executive body size changes at 1,000, 5,000, 7,000, 10,000 and 20,000

§19(1) NÖ–GO §24(1) NÖ–GO §24(1) NÖ–GO

Salzburg Tyrol

Council size changes at 800, 1,500, 2,500, 3500 and 5000 Council size changes at 200, 1,000, 2,000, 4,000, 6000 and 10,000 Second and third vice mayors at 1000 and 5000

§19(1) GdO §22(1) TGO §23(3) TGO

Upper Austria

Council size changes at 400, 1,100, 1,900, 4500 and 7300 Wages of the council change at 1,000, 4500 and 1,5000 Necessary support for calling a referendum increases at 1,000 and 10,000

§18(1) OÖ. GemO §34(2) OÖ. GemO §38(1) OÖ. GemO

Vorarlberg

Council size changes at 500, 1000, 1500, 2000, 2500, 5000, 8000, 11,000 and 15,000

§34(1) V–GemG

References Arraiz, I., Drukker, D.M., Kelejian, H.H., Prucha, I.R., 2010. A spatial cliff-ord-type model with heteroskedastic innovations: Small and large sample results. J. Reg. Sci. 50 (2), 592–614. Ashworth, J., Galli, E., Padovano, F., 2013. Decentralization as a constraint to leviathan: a panel cointegration analysis. Public Choice 156 (3–4), 491–516. Baskaran, T., 2010. On the link between fiscal decentralization and public debt in oecd countries. Public Choice 145 (3–4), 351–378. Blöchliger, H., Petzold, O., 2009. Finding the dividing line between tax sharing and grants. OECD Fiscal Federalism Group Working Paper COM/CTPA/ECO/GOV/ WP(2009)10. Blöchliger, H., Vammalle, C., 2012. Reforming Fiscal Federalism and Local Government: Beyond the Zero-Sum Game. OECD Publishing, Ch. Austria: The reform of the fiscal equalisation law, pp. 39–49. Blume, L., Müller, J., Voigt, S., Wolf, C., 2009. The economic effects of constitutions: replicating - and extending - Persson and Tabellini. Public Choice 139 (1–2), 197–225. Brennan, G., Buchanan, J.M., 1980. The Power to Tax: Analytic Foundations of a Fiscal Constitution. Cambridge University Press, Cambridge, England. Bröthaler, J., Getzner, M., Haber, G., 2015. Sustainability of local government debt: a case study of austrian municipalities. Empirica 42 (3), 521–546. Calonico, S., Cattaneo, M.D., Farrell, M.H., Titiunik, R., 2016. Regression discontinuity designs using covariates. Tech. rep., working paper, University of Michigan. Calonico, S., Cattaneo, M.D., Titiunik, R., 2014. Robust nonparametric confidence intervals for regression-discontinuity designs. Econometrica 82 (6), 2295–2326. Cameron, A.C., Gelbach, J.B., Miller, D.L., 2011. Robust inference with multiway clustering. J. Bus. Econ. Stat. 29 (2), 238–249. Cassette, A., Paty, S., 2010. Fiscal decentralization and the size of government: a European country empirical analysis. Public Choice 143 (1–2), 173–189. Cattaneo, M.D., Frandsen, B.R., Titiunik, R., 2015. Randomization inference in the regression discontinuity design: An application to party advantages in the US senate. J. Causal Inference 3 (1), 1–24. Cattaneo, M.D., Jansson, M., Ma, X., 2016a. rddensity: an application to party advantages in the us senaty. Stata J., 1–18.

24

European Journal of Political Economy xxx (xxxx) xxx–xxx

M. Köppl–Turyna, H. Pitlik

Cattaneo, M.D., Jansson, M., Ma, X., 2016b. Simple local regression distribution estimators with an application to manipulation testing. Tech. rep., University of Michigan, Working Paper. Cattaneo, M.D., Titiunik, R., Vazquez-Bare, G., 2016c. Inference in regression discontinuity designs under local randomization. Stata J. 16 (2), 331–367. Costa-Font, J., Moscone, F., 2009. The impact of decentralization and inter-territorial interactions on Spanish health expenditure. Phys.-Verl. HD, Heidelb., 167–184. Dahlberg, M., Mörk, E., Rattsø, J., Ågren, H., 2008. Using a discontinuous grant rule to identify the effect of grants on local taxes and spending. J. Public Econ. 92 (12), 2320–2335. De Mello, L.R., 2000. Fiscal decentralization and intergovernmental fiscal relations: a cross-country analysis. World Dev. 28 (2), 365–380. De Witte, K., Geys, B., 2015. Strategic housing policy, migration and sorting around population thresholds. CESifo Working Paper No. 5639. Drukker, D.M., Egger, P., Prucha, I.R., 2009. On single equation GMM estimation of a spatial autoregressive model with spatially autoregressive disturbance. Tech. rep., Technical report, Department of Economics, University of Maryland. Egger, P., Koethenbuerger, M., 2010. Government spending and legislative organization: quasi-experimental evidence from Germany. Am. Econ. J.: Appl. Econ. 2 (4), 200–212. Eggers, A.C., Freier, R., Grembi, V., Nannicini, T., 2015. Regression discontinuity designs based on population thresholds. DIW Discussion Paper no. 1503. Eyraud, L., Lusinyan, L., 2011. Decentralizing spending more than revenue: Does it hurt fiscal performance? IMF Working Papers. Feld, L.P., 1997. Exit, voice and income taxes: the loyalty of voters. Eur. J. Political Econ. 13 (3), 455–478. Fiva, J.H., 2006. New evidence on the effect of fiscal decentralization on the size and composition of government spending. Finanz.: Public Financ. Anal. 62 (2), 250–280. Foremny, D., 2014. Sub-national deficits in european countries: the impact of fiscal rules and tax autonomy. Eur. J. Political Econ. 34, 86–110. Foremny, D., Monseny, J.J., Ollé, A.S., 2015. Hold that ghost: Local government cheating on transfers. In: IIPF Conference Papers. URL: 〈https://www.cesifo-group. de/dms/ifodoc/docs/Akad〉Conference/CFP CONF/CFP CONF. Gelman, A., Imbens, G., 2014. Why high-order polynomials should not be used in regression discontinuity designs. Working paper, National Bureau of Economic Research. Goodspeed, T.J., 2002. Bailouts in a federation. Int. Tax. Public Financ. 9 (4), 409–421. Gordon, N., 2004. Do federal grants boost school spending? Evidence from title I. J. Public Econ. 88 (9), 1771–1792. Grembi, V., Nannicini, T., Troiano, U., 2016. Do fiscal rules matter? Am. Econ. J.: Appl. Econ. 8 (3), 1–30. Hayo, B., Neumeier, F., 2012. Leaders' impact on public spending priorities: the case of the german laender. Kyklos 65 (4), 480–511. Imbens, G.W., Lemieux, T., 2008. Regression discontinuity designs: a guide to practice. J. Econ. 142 (2), 615–635. Jin, J., Zou, H.-f., 2002. How does fiscal decentralization affect aggregate, national, and subnational government size? J. Urban Econ. 52 (2), 270–293. Jordahl, H., Liang, C.-Y., 2010. Merged municipalities, higher debt: on free-riding and the common pool problem in politics. Public Choice 143 (1–2), 157–172. Lee, D.S., Lemieux, T., 2010. Regression discontinuity designs in economics. J. Econ. Lit. 48, 281–355. McCrary, J., 2008. Manipulation of the running variable in the regression discontinuity design: a density test. J. Econ. 142 (2), 698–714. Nakazawa, K., 2013. Amalgamation, free-rider behavior, and regulation. Int. Tax. Public Financ., 1–22. Oates, W.E., 1972. Fiscal Federalism. Harcourt, New York. Oates, W.E., 2005. Toward a second-generation theory of fiscal federalism. Int. Tax. Public Financ. 12 (4), 349–373. Persson, T., Tabellini, G., 2004. Constitutions and economic policy. J. Econ. Perspect. 18 (1), 75–98. Persson, T., Tabellini, G.E., 2005. The Economic Effects of Constitutions. MIT Press, Cambridge, Massachusetts. Pettersson-Lidbom, P., 2012. Does the size of the legislature affect the size of government? Evidence from two natural experiments. J. Public Econ. 96 (3), 269–278. Pitlik, H., 2014. Fiscal Federalism and Fiscal Decentralisation in Europe. Comparative Case Studies on Spain, Austria, the United Kingdom and Italy. Innsbruck, Wien, Bozen,, Ch. Fiscal Federalism, Austrian Style: Fear of Competition, pp. 41–59. Prohl, S., Schneider, F., 2009. Does decentralization reduce government size? A quantitative study of the decentralization hypothesis. Public Financ. Rev. 37 (6), 639–664. Rodden, J., 2002. The dilemma of fiscal federalism: grants and fiscal performance around the world. Am. J. Political Sci. 46 (3), 670–687. Rodden, J., 2003. Reviving leviathan: grants and fiscal performance around the world: grants and fiscal performance around the world. Int. Organ. 57 (04), 695–729. Rodden, J., 2006. Hamilton's Paradox: The Promise and Peril of Fiscal Federalism. Cambridge University Press, Cambridge, England. Sanz, I., Velazquez, F.J., 2002. Determinants of the composition of government expenditure by functions. Universidad Complutense de Madrid. Grupo de Economía Europea. Schaltegger, C.A., Somogyi, F., Sturm, J.-E., 2011. Tax competition and income sorting: evidence from the zurich metropolitan area. Eur. J. Political Econ. 27 (3), 455–470. Sorens, J., 2016. Vertical fiscal gaps and economic performance. Mercatus Working paper. Thöni, E., Garbislander, S., Haas, D.-J., 2002. Local budgeting and local borrowing in austria. In: Dafflon, B. (Ed.), Local Public Finance in Europe: Balancing the Budget and Controlling Debt. Edward Elgar Publishing, pp. 45–74. Tiebout, C.M., 1956. A pure theory of local expenditures. J. Political Econ. 64 (5), 416–424. Volden, C., 1999. Asymmetric effects of intergovernmental grants: analysis and implications for us welfare policy. Publius: J. Fed. 29 (3), 51–73. Von Hagen, J., Eichengreen, B., 1996. Federalism, fiscal restraints, and european monetary union. Am. Econ. Rev. 86 (2), 134–138. Werck, K., Heyndels, B., Geys, B., 2008. The impact of central places on spatial spending patterns: evidence from Flemish local government cultural expenditures. J. Cult. Econ. 32 (1), 35–58. Wilson, J.D., Wildasin, D.E., 2004. Capital tax competition: bane or boon. J. Public Econ. 88 (6), 1065–1091. Zodrow, G.R., Mieszkowski, P., 1986. Pigou, Tiebout, property taxation, and the underprovision of local public goods. J. Urban Econ. 19 (3), 356–370.

25