Political yardstick competition among Italian municipalities on spending decisions∗ David Bartolini† Polytechnic University of Marche
Raffaella Santolini‡ University of Teramo
February 13, 2011
Abstract We investigate the presence of political yardstick competition on current spending decisions in a sample of Italian municipalities. We find significant evidence of yardstick competition when we explicitly account for the domestic stability pact (DSP), a fiscal rule introduced to limit the budget deficit of local administrations. Firstly, we estimate a static specification of a spatial panel model, and then we check for the robustness of our results with a dynamic specification. The static analysis shows that municipalities engage in yardstick competition during pre-election years, whether their are subject to the DSP or not. The dynamic analysis shows that the yardstick hypothesis remains robust only for municipalities not constrained by the DSP. JEL Classification: C21; C23; D72; H72 Keywords: Yardstick competition; Fiscal rules; System GMM; Spatial econometrics
In a decentralised setting the economic policies of neighbouring jurisdictions show a certain degree of correlation, as shown in several empirical studies (Ladd, 1992; Heyndels and Vuchelen, 1998; Brueckner, 2003). The economic literature has come up with four different possible explanations for this interaction. Firstly, it could be that jurisdictions are simply hit by a common shock that would make politicians react in the same way, or policies implemented in one jurisdiction may produce spillovers on neighbouring jurisdictions, triggering a reaction in policy choices (Case et al., 1993; Revelli, 2001; Baicker, 2005; Sol´e-Oll´e, 2006). For instance, a tougher policy against crime in one jurisdiction may push criminals towards other jurisdictions, which as a consequence need to increase their spending to contrast crime. Secondly, the interaction may be the consequence of tax competition, where jurisdictions react to policies because they fear to lose tax base (Devereux et al., 2007; Ruiz and Gerard, 2008). Thirdly, fiscal interaction can be driven ∗
We are grateful for their helpful comments on a previous version of this paper to Timo Mitze, Christian Bruns, Barbara Ermini, Leonzio Rizzo. We also would like to thank participants at the CESifo 2nd Political Economy Workshop in Dresden, Germany, and the 3rd World Conference of Spatial Econometrics in Barcelona, Spain. Usual disclaimer applies. † DEA, Universit` a Politecnica delle Marche, Ancona (Italy). ‡ Corresponding author: DSGSS, Universit` a degli Studi di Teramo, Campus di Coste Sant’Agostino, 64100, Teramo (Italy) — Email: [email protected]
by a common political trend, where politicians sharing the same ideology would tend to mimic each other, without any electoral goal (Foucault et al., 2008; Santolini, 2008, 2009). Finally, the interaction can be the result of political opportunism. The intuition behind this motivation is related to the presence of private information about either the quality of the incumbent or the costs and benefits of the policies implemented. Citizens can get some information by comparing the performance of their politicians with the performance of politicians in neighbouring jurisdictions. As a consequence, the incumbent would mimic neighbouring jurisdictions’ policies in order not to lose political consensus. This process was firstly described by Salmon (1987), in terms of yardstick competition. The present work focuses on the yardstick competition source of fiscal interaction among jurisdictions. Our main aim is to empirically asses the presence of political yardstick competition on spending decisions of a sample of Italian municipalities. Empirically it is not easy to disentangle the yardstick competition behaviour from the other three motivations, for all of them lead to the same correlation among policies of neighbouring jurisdictions. In this respect, one of the first contributions is Case (1993), which investigates income tax policies of incumbent governors in U.S. states, showing a different mimicking behaviour of politicians under term limit with respect to politicians that can be re-elected. She shows that politicians under term limit do not react to the tax policy of neighbouring politicians. Besley and Case (1995) provide a theoretical model of a tax setting game according to the idea of yardstick competition and found evidence of such behaviour on the tax setting policy of U.S. States. Sol´e-Oll´e (2003) focuses on the relationship between tax mimicking and yardstick competition showing that tax rates are lower and tax interactions are more intense when governments are: i) in election years; ii) formed by a small majority; iii) ruled by right-wing coalitions. Those hypotheses are tested in the empirical literature: for instance, Bordignon et al. (2003) show that mayors with large majorities are not affected by neighbouring politicians’ policies; while Allers and Elhorst (2005) find that municipalities ruled by mayors with a large majority have a less intense tax mimicking behaviour than municipalities ruled by small majorities. Our identification strategy is based on the comparison of fiscal interaction among municipalities in the electoral and pre-electoral years, with respect to periods away from elections. We reckon that a greater reaction of incumbent politicians in the (pre-)election year is consistent with yardstick competition behaviour. The empirical investigation is conducted on a panel of 246 municipalities located in the region Marche (Italy), from 1994 to 2003. The choice of the lowest level of government as the objective of our analysis is consistent with the idea that the possibility of citizens to infer politicians’ quality from neighbouring policies is higher the closer jurisdictions are in terms of geographical distance. Our analysis focuses on the period 1994-2003 in order to capture the effect of the introduction of a fiscal rule, the domestic stability pact (DSP), which from the year 1999 onwards limits the possibility of municipalities to run deficits on current budget settings. Since the DSP refers to current public expenditures, we focus the empirical analysis on those type of expenditures. Moreover, the choice of current expenditures is also based on the fact that they are easily manipulated by mayors in order to get electoral consensus. Although most of the works on political yardstick competition focus on tax setting policies, the opportunistic behaviour relates also to spending policies. For instance, Elhorst and Fr´eret (2009) focus on social spending at the level of “department” in the French administration, they find significant evidence that social spending is influenced by the
opportunistic behaviour of the incumbent politician. Most likely citizens would look at the economic policies as a whole, we would expect that for a given level of taxes, citizens seeing better schooling services in a neighbouring jurisdiction would believe they are governed by relatively incompetent politicians. Anticipating this behaviour the incumbent politician may decide to increase public spending, in order not to lose political consensus. For these reasons, our objective is to investigate whether spending decisions are influenced by political yardstick competition. The empirical analysis conducted on our panel, shows that spending policies are consistent with the yardstick competition hypothesis. This strategic behaviour arises only when we specifically account for the DSP, by estimating an extended version of the tworegime spatial lag model proposed by Allers and Elhorst (2005). Thereby, the omission of the fiscal rule would jeopardize the detection of this phenomenon. Our analysis is firstly conducted on a static specification of the spatial econometric model, then we test the robustness of our results with a dynamic specification. The static specification shows that, in the pre-election year the yardstick behaviour is common to any municipality, regardless of the DSP. The dynamic specification, however, confirms the yardstick hypothesis only for municipalities not subject by the DSP. Therefore, our analysis suggests that the fiscal rule could play a relevant role in constraining the opportunistic behaviour of incumbent politicians. Our contribution is important for the following reasons. Firstly, we contribute to the literature on political yardstick competition by focusing on current spending policies. Secondly, our work represents the first panel data analysis on Italian municipalities, estimating both static and dynamic spatial econometric specifications for fiscal interactions. Thirdly, we show the importance of the fiscal rule for the identification of the yardstick behaviour, and its role as a possible constraint for the opportunistic behaviour of mayors. However, to clearly understand the role of the fiscal rule further investigation is needed. The rest of the paper is organised as follows: in the next section we give a brief description of the Italian institutional setting; then, in section 3 and 4, we describe the data and the methodology used to conduct our empirical investigation; in section 5, we report our estimation results; finally, in section 6, we conclude.
The administrative structure of Italian municipalities
The Italian state is organized in four levels of government. A central government, 20 regional governments,1 109 provincial governments, and more than 8000 municipalities. The region Marche is located in the centre of Italy and consists of 246 municipalities grouped in 5 provinces. In this context, municipalities represent the lowest level of government, and it is responsible for many activities affecting the “daily” life of citizens. In particular, municipalities deal with property issues, such as building permits, street lump, sewage, etc, and with social activities, such as nursing, elderly care, etc. The political and administrative structure of municipalities consists in the mayor (sindaco), the executive board (giunta comunale), and the city council (consiglio comunale).2 1
The Italian Constitutional law considers 20 regions, but one of those regions (Trentino-Alto Adige) is actually divided in two separate governing bodies: the autonomous province of Trento and the autonomous province of Bolzano. Those “provinces” have administrative powers similar to the other Italian regions. 2 The Law 267/2000 contains the whole set of norms regarding the administrative structure, functions and electoral rules of provinces and municipalities.
The mayor and the city council are elected, while the members of the executive board are directly appointed by the mayor — which has the power to remove them. The city council consists of the elected candidates plus the elected mayor. In municipalities with less than 15,000 inhabitants, the mayor is also the president of the city council, while in larger municipalities, the president is elected by the council in the first meeting. The mayor and the city council are elected at the same time and their mandate lasts for five years. Mayors are elected by majority voting: in municipalities with less than 15,000 inhabitants, simple majority is sufficient to be elected; in larger municipalities, unless one candidate gets absolute majority, a second round takes place among the two most voted candidates, where simple majority is required. Each list of candidates for the city council is matched to a candidate for the post of mayor. The total number of seats in the council is allocated according to the proportion of votes receives by the list and the single candidates in the list, adjusted for a premium of seats for the list matched with the winning mayor. Therefore, the majority of seats in the city councils are allocated to candidates that support the mayor. The mayor is endorsed with most of the executive powers, and with the help of the giunta is in charge of planning and implementing council policies. Since the members of the giunta are appointed — not elected — it is only the mayor which is politically accountable for the policies implemented. The city council is responsible for the political and administrative control of the activities of the municipality; for instance, the annual fiscal budget must be approved by the council. In this institutional setting, the set of policies implemented by a municipality are a direct evidence of mayors’ performance. The budget is the administrative instrument that bears evidence of those policies. It mainly consists of current and investment expenditures, tax revenue, and transfers from other levels of government. A lot of public services provided by municipalities are accounted as current spending; for instance, social services like elderly care or nursery mainly consist of costs for the personnel and services contracted out to other firms. Current spending can be classified according to the following functions: general administration services, local police, education, culture, sports and leisure, tourism activities, transports and roads, environment, social services, and services to promote economic activities. In order to given an idea of the importance of current spending on the budget of municipalities, we computed the share of current expenditure with respect to total expenditures, in our sample: it amounts to 46.5% in the period 2001, and 47.2% in 2003. In the year 1999, the budget autonomy of municipalities (and also of regions and provinces) has been constrained by the introduction of a domestic stability pact (DSP), as stated in the Law 448/1998. This type of fiscal rule has been introduced by most countries in the Euro zone as a consequence of the Stability and Growth Pact, with the aim of controlling (and possibly limiting) the budget deficit of the lower levels of government. Those fiscal rules, generally consist in a limitation to run deficits and/or a direct limit to spending growth; for instance, in most European countries local fiscal rules impose budget balance restrictions (e.g., Austria, Belgium, Germany, Italy, the Netherlands); in some of those countries (e.g., Germany and Belgium) fiscal rules encompass also a constraint on the expenditure growth rate (Sutherland et al., 2005). In Italy, the DSP mainly consists in a set of norms contained in the annual budget law.3 3
The budget law (legge finanziaria) is the legal procedure through which the government makes the necessary changes to the State budget; it is discussed in the parliament in the autumn term and must be
The fact that the actual features of the DSP are prescribed in the budget law, means that it may be subject to substantial changes from one year to the other, creating a problem of credibility. In fact, we can identify three periods within which the DSP consists of homogeneous norms. In the first period, from 1999 to 2004, the objective of the DSP is to limit the deficit which arises as the difference between revenues and current expenditures — with the exception of the year 2002 in which an explicit limit to the grow of current expenditures was added. In the second period, from 2005 to 2006, the DSP consists in a direct limit to the growth of expenditures. In the last period, from 2007 to 2011, the focus has been shifted back to the budget deficit. Our analysis takes into account the DSP as defined in the first period. A problem with the Italian DSP, common to the first two periods, is the difficulty to implement sanctions. Although, different types of sanctions were prescribed in different periods, their implementation has been hindered by legal problems. For instance, in 2002 was prescribed that a violation of the DSP should lead (among other things) to a reduction in central government transfers for the following year, but a contrast with the Italian constitutional law prevented the application of such sanctions.4 This problem has affected the implementation of the DSP until recently (2007-2011), when a more effective system of monitoring and sanctions has been put in place. The effects of this new regime of DSP, however, are still under investigation. In table 1, we present a detailed account of the DSP for the relevant period of our analysis. In our sample, the DSP is mainly a limit to the possibility to finance expenditures with debt, and, therefore, it could represent a constraint to the possibility of mayors to mimic each other.
The empirical analysis is conducted in the period 1994-2003 using a balanced panel data on 246 Italian municipalities located in the Marche region. Both economic and demographic indicators of this region, are in line with the Italian indicators, therefore we can take data from the region Marche as representative of the whole nation. Our dependent variable is current expenditures as accounted in local councils balance sheets. In order to account for the introduction of the DSP rule, that might affect the composition of the budget, we consider a DSP dummy that assumes value 1 when the municipality is subject to the DSP, and zero otherwise. In the year 2001, Italian municipalities with a population lower than 5,000 people (27% of our sample) have been excluded from the DSP. The fact that our analysis focuses only on current expenditures does not represent a limit, because some empirical studies point out that the opportunistic behaviour does not depend on whether expenses are current or capital, rather it is the visibility that matters.5 In this respect, mayors can increase the quality of existing services by increasing current expenditures. For instance, the mayor may increase the periodicity of ordinary street lights maintenance, or contract the services of more swimming instructors for the municipal pool, etc. Those are visible actions that can improve the reputation of the incumbent mayor for the upcoming elections. approved by the end of the year. 4 See Patrizii et al. (2006) for an investigation of the level of compliance of municipalities to the DSP. 5 For instance, Veiga and Veiga (2007) in the context of political budget cycle, show that the opportunistic behaviour is predominant in “visible” policies.
In order to account for the presence of electoral competition in spending policies, we include two dummies in the regression analysis: one for the election year and another for the pre-election year, which assume value 1 when municipalities are in (pre-)election year and zero otherwise. We use the pre-election year dummy to capture the opportunistic behaviour when elections are held in the first part of the electoral year, as, in this case, the mimicking action is likely to take place in the year before the elections. In our sample this is an important issue, as before the year 1999, the most part of elections (90% of municipalities) took place in the first semester, and from the year 1999 onwards, every election has been held in the first semester. Public expenditures at the local level are significantly affected by the socio-economic and the political characteristics of municipalities, therefore we include data about socioeconomic and political features in our data set. We consider population and population density, which denote the presence of congestion effects if the two variables are positively affected by current expenditure (our dependent variable), while a negative sign denotes economies of scale. Additionally, we include the percentage of old people (greater or equal than 65 year old) to account for the possibility that expenditures are driven by social programs for elderly people. Data on the economic characteristics of Italian local councils, such as per-capita income6 and per-capita grants from the central government are also considered in the regression analysis. Grants from the central government represent the most part of intergovernmental transfers: in our sample, for instance, 76.3% of total transfers (in the year 2001) are from the central government. We expect a positive correlation of both income and grants with current expenditures. For instance, a positive value of the coefficient associated with income would be consistent with the Wagner’s law, which implies an increase of public expenditures as a consequence of economic development. Finally, we also control for political effects, such as political ideology and the share of votes of the incumbent mayor. We introduce two dummies that assume value 1 when the municipality is ruled by the left(right)-wing coalition, and zero otherwise. This would account for the possibility that expenditures are affected by partisan affiliation. A positive (negative) impact on public expenditure is expected when the left(right)-wing coalition rules the jurisdiction. As regards the shares of votes, we expect that politicians elected with a large majority have less incentive to increase expenditures, as they face lower political competition. By contrast, a positive correlation between majority size and expenditures could signal the presence of a Leviathan government (Brennan and Buchanan, 1980). The summary statistics of the above-mentioned variables are reported in table 2. Although in the regression analysis we take the logarithm of both dependent and control variables, to partially remove problems of heteroschedasticity which affect data on public expenditures in the region Marche (Ermini and Santolini, 2010).
Spatial econometric models for the yardstick competition analysis
In this section we describe the spatial econometric models employed to detect yardstick competition. As a first step, we estimate a standard panel data model of expenditure interaction among neighbouring jurisdictions, including a vector of municipality effects, 6
Since a panel data of disposable income is not available for Italian municipalities, we use statistics on the income tax base — Imposta sul Reddito delle Persone Fisiche (IRPEF).
αi , and a vector of time effects, τt , as described in equation (1). expit = b + ρ
wij expjt + δ1 eleit + δ2 preleit + γxit + αi + τt + µit
The dependent variable, expit , corresponds to a N × 1 vector of cross-sectional time series observations on per capita current expenditures in municipality i (for i = 1, . . . , N ) at time period t (for t = 1, . . . , T ). The spatial dimension of the model is captured by the N × N spatial weight matrix, W , whose elements wij for i 6= j assume value 1 when municipality j shares a border with municipality i, and zero otherwise. When i = j, we assume that wij = 0 (Anselin, 1988). Usually, the rows of W are standardised to 1, so that PN j=1 wij = 1. When we multiply the elements wij of the spatial weight matrix by those P of the current spending in other municipalities, i.e. N j=1 wij expjt , we obtain the average public expenditures of contiguous municipalities, so that the coefficient ρ measures the intensity of the spending interaction. Since our sample is defined by the administrative rule of belonging to the region Marche, the neighbourhood of municipalities sharing a border with municipalities in other regions is excluded from the spatial weight matrix W . We believe, however, this sample selection does not introduce any bias in our empirical analysis, because municipalities mainly mimic fiscal policies of municipalities in the same region. The reason being that information about neighbouring municipality are mainly spread through local newspapers and regional television channels; and those news, generally, do not provide any information about municipalities that are contiguous but in other regions. Although we do not expect border municipalities to significantly affect our analysis, we control for border effects introducing a dummy variable (border ), in the dynamic specification of our spatial model.7 This dummy assumes value 1 if the municipality shares a border with a municipalities of another region, and 0 otherwise. The opportunistic behaviour of the incumbent politician during the electoral campaign is captured by the election year dummy ele, and the pre-election year dummy prele. The former dummy accounts for the opportunistic behaviour in the year of the elections: it is a N × 1 vector that assumes value 1 when elections occur at time t, and zero otherwise. A positive sign of the coefficient δ1 would signal the presence of electoral competition, as current expenditures increase during the election year. The electoral race, however, can start before the election year — particularly if the date of the election is at the beginning of the year. We capture this possibility with the pre-election year dummy prele, that assumes value 1 when the municipality is in the pre-election year, and zero otherwise. A positive sign of the coefficient δ2 is a signal that spending decisions are taken according to electoral motivations. 0 As regards the control variables, we include a 1×K vector xit = (x1it . . . , xK it ), consisting 2 of: i) population density (i.e., population per km ); ii) population size; iii) percentage of young people (0-14 year old); iv) percentage of old people (≥65 year old); v) per-capita income and pre-capita grants from the central government. Finally, the model includes a constant term b, and an error term, µit , which is independent and identically distributed with zero mean and constant variance σµ2 . 7 We can use a time invariant dummy in the dynamic model since we use a system GMM estimator (Arellano and Bover, 1995).
Since the meaning attached to the coefficient ρ is consistent with four explanations of the interaction,8 we need to identify the source of expenditure interaction. According to the yardstick competition approach, incumbent politicians in election year show a stronger intensity of fiscal interaction than incumbents not in election year, as they increase their chances of being re-elected by copy-catting neighbouring fiscal policies (Sol´e-Oll´e, 2003). Therefore, we modify model (1) introducing two regimes of spatial lag ρ1 and ρ2 , as in Allers and Elhorst (2005). These two coefficients would measure the intensity of the expenditure interaction among municipalities facing elections (ρ1 ) and not facing elections (ρ2 ), respectively. A positive sign of ρ1 stands for strategic complementarity, which can be a signal of yardstick competition only if ρ1 is significantly greater than ρ2 . Given that we estimate the slope of the reaction functions for municipalities in election and non election years, the constant term b needs to be replaced by two complementary intercepts: eleit and (1 − eleit ). Equation (2) represents this model. It is analogous to model (1), with the difference that the effect of the neighbouring expenditures is split in two separate terms, one accounting for the interaction in the electoral year and the other for the interaction in the rest of the periods.
wij expjt eleit + ρ2
wij expjt (1 − eleit )+
+ ϕ1 eleit + ϕ2 (1 − eleit ) + δ2 preleit + 0
+ γxit + αi + τt + µit The introduction of the DSP can restraint the opportunistic behaviour of incumbent politicians, reducing their ability to manipulate fiscal budget variables. Therefore, not taking into account this fiscal rule could produce a misspecification of our empirical analysis. Equation (3) describes the extension of the previous model with the introduction of the DSP. In order to consider the effects of the fiscal rule on the opportunistic behaviour stemming from yardstick competition, we consider the average current expenditures of contiguous municipalities that are in election year, discriminating between municipalities affected by the DSP and not affected. Let ρd1 measure the intensity of the expenditure interaction of municipalities affected by the DSP, and ρnd 1 the intensity of the interaction of municipalities not affected by the DSP. Given this specification, we can study the presence of yardstick competition separately for municipalities affected by the fiscal rule and municipalities not affected (in equation (3)). As regards municipalities affected by the DSP, the presence of yardstick competition is denoted by the following condition on the parameters: ρd1 > 0 and ρd1 > ρ2 . Similarly, the yardstick competition for municipalities nd not affected by the DSP is denoted by ρnd 1 > 0 and ρ1 > ρ2 . This is because the presence of yardstick competition depends on the difference between the interaction coefficient of municipalities in election or pre-election year and municipalities not facing elections, 8
That is, in addition to yardstick competition, ρ could be thought of as a measure of spill-over effects, fiscal competition, or political trend.
regardless of the imposition of the DSP. expit =ρd1
wij expjt eleit DSPit +
wij expjt eleit (1 − DSPit )+
wij expjt (1 − eleit ) + ϕd1 eleit DSPit +
ϕnd 1 eleit (1
− DSPit ) + ϕ2 (1 − eleit ) + δ2 preleit +
+ γxit + αi + τt + µit In order to further investigate the reaction of each subgroup of municipalities to neighbouring policies, we discriminate municipalities not in election year between municipalities constrained and not constrained by the fiscal rule. This further specification is described in equation (4), where ρd2 and ρnd 2 correspond to the intensity of fiscal interaction of municipalities not facing elections that are subject and not subject to the DSP, respectively. The hypothesis of yardstick competition requires ρd1 > ρd2 and ρd1 > ρnd 2 , for the subgroup of municipalities affected by the DSP, while for the subgroup not affected by the DSP the d nd nd condition is ρnd 1 > ρ2 and ρ1 > ρ2 .
wij expjt eleit DSPit +
wij expjt eleit (1 − DSPit )+
wij expjt (1 − eleit )DSPit +
j=1 N X
d ϕd1 eleit DSPit + ϕnd 1 eleit (1 − DSPit ) + ϕ2 (1 ϕnd 2 (1 − eleit )(1 − DSPit ) + δ2 preleit +
wij expjt (1 − eleit )(1 − DSPit )+
− eleit )DSPit +
+ γxit + αi + τt + µit As mentioned above, the opportunistic behaviour could take place in the period before the election year, i.e., the pre-election year. In order to consider this possibility we estimate a modified version of model (2), model (3), and model (4), where preleit substitutes eleit in the interaction terms. The spatial econometric models are estimated by a feasible two-step GMM estimator with robust standard error (Anselin, 1988; Baum et al., 2003). Since it is a two-stage estimator, it remains consistent when data suffer from heteroscedasticity and non-normally distributed error terms (Anselin, 1988). Additionally, it is a robust estimator even in the presence of spatial auto-correlated shocks (Kelejian and Prucha, 1998), thereby the correlation in the level of expit does not depend on common shocks spatially distributed among jurisdictions. The average current expenditure of municipalities needs to be instrumented to avoid endogeneity problems (Anselin, 1988). We follow the approach of Kelejian and Prucha 0 (1998) in choosing the set of instruments. They show that exogenous regressors xit and 9
P different combinations of N j wij xit are a valid set of instrumental variables for a generalised spatialP two-stage least square procedure. In particular, model (1) is instrumented 0 with xit and N j wij xit . For the other spatial econometric models, we use similar instruP ments with the difference that N j wij xit is interacted with the (no-)election year dummy 9 and the (no-)DSP dummy, according toP the model estimated. PN To clarify this point, 0 N let us consider equation (3): we use xit , j wij xit eleit DSP , j wij xit eleit (1 − DSP ), P and N j wij xit (1 − eleit ). Similar instrumental variables are implemented when the preelectoral period is considered. Additional exogenous instruments, such as electoral indexes and their average are also taken into account.10 We estimate a dynamic specification of all previous spatial econometric models, in order to check the robustness of our results. This requires the introduction of the first order lagged dependent variable expit−1 among the regressors. To provide a better specification of the model, we also add the regressor ∆expit−1 . The dynamic panel data model cannot be estimated by the fixed-effect (FE) estimator when T is fixed and N goes to infinity, as the estimator would be biased and inconsistent (Verbeek, 2008). A possible solution is to consider estimators based on the generalised method of moments (GMM), which are more efficient than the first difference estimator developed by Anderson and Hsiao (1981, 1982). In particular, we refer to the difference GMM (GMM-DIF) and the system GMM (GMMSYS) estimators (Arellano and Bond, 1991; Arellano and Bover, 1995; Blundell and Bond, 1998). Both estimators require the first-order autocorrelation in the first differenced error term to be different than zero. Their consistency, however, depends on the presence of the second-order autocorrelation in the differenced residuals. Both conditions are detected by using the specification tests developed by Arellano and Bond (1991). In the empirical analysis we refer to them as AB-AR1 test, in the case of the first order autocorrelation analysis, and AB-AR2 test for the second order autocorrelation analysis. Our spatial dynamic panel model is estimated using a GMM-SYS estimator, which is a combination of a set of standard equations in first difference and equations in levels, distinctly instrumented. For the first differences equation, we use the second order lag of the dependent variable, expit−2 . Similarly, the spatially lagged dependent variable is instrumented with its own related lags (Jacobs et al., 2009). In particular, we use the third order spatial lag of the dependent variable, interacted and not with the electoral PN and DSP dummies, according to the model estimated. Additionally, we use ∆ j wij xit , where x is a vector of control variables such as, population, density, young people, elderly people, per-capita income, per-capita grants, electoral and pre-electoral indexes. For the level equations, ∆expit−1 is used to instrument the first order lag of the dependent variable. On the other hand, the spatial lag dependent variable is instrumented PN by ∆ j wij expit−2 , interacted or not with electoral and DSP dummies, according to P the model. Both N j wij xit (as previously defined) and the constant term are additional instruments for this endogenous variable. The validity of the set of instruments is detected by the standard Sargan/Hansen test of over-identifying restrictions (Arellano and Bover, 1995; Blundell and Bond, 1998), even though the proliferation of instruments can seriously weaken this test and overfit the endogenous variables (Bowsher, 2002; Roodman, 2009). Furthermore, the validity of 9
The instruments used in model 4 are the same as in model 3. These indexes, developed by Franzese (2000),capture the exact date in which the election is held. The 1 d (m − 1) + D and preleindexit =(1 − eleindexit ), where election indexes are calculated as, eleindexit = 12 m and d are the month and the day in which the election takes place, while D is the number of days of m. 10
the additional set of instruments used in GMM-SYS estimations is tested by using the Difference Sargan/Hansen statistic (Blundell and Bond, 1998, 2000; Bond et al., 2001). In our empirical analysis, a one-step version of the GMM-SYS estimator, corrected for heteroscedasticity, is adopted because the efficiency gains of using the two-step version is quite small (Bond et al., 2001).11 Moreover, Kukenova and Monteiro (2009) show that, in some instances, the GMM-SYS estimator is more efficient than the maximum likelihood estimators developed for spatial panel models. Finally, the GMM-SYS estimator is also easier to implement, as it does not require the inversion of the spatial weight matrix (Kukenova and Monteiro, 2009).
The results of the estimation of the static specification are reported in table 3 and table 4. The first column of table 3, shows the estimation results of model (1), where we do not consider the electoral period. It turns out that the expenditure interaction (Neigh Exp) is significant and occurs with a magnitude of 0.58. This result, however, does not represent in itself any evidence of yardstick competition, we need to discriminate for the election periods in order to single-out political yardstick from the other sources of interaction. The estimation of model (2) shows that municipalities in election year are more sensitive to neighbouring expenditures than municipalities facing no imminent elections (second column of table 3), pointing at the presence of yardstick competition. We find similar results when we consider the pre-election year, as shown in the third column of table 3. Those results would provide evidence of political yardstick competition in both the election and the pre-election year, but unfortunately they are not robust, as the p-value of the difference between the ρ coefficients (associated with the interaction terms) does not allow us to reject the null hypothesis of identical values. The absence of any significant evidence of yardstick competition could depend on the omission of the DSP, which represents a possible constraint to mayors’ expenditure decisions. We, therefore, estimate model (3), where the introduction of the DSP is accounted for, distinguishing the effect in the electoral and pre-electoral years (table 4). The analysis does not provide any robust evidence of yardstick competition in the election year, because the ρ-test is not statistically significant. By contrast, the hypothesis of yardstick competition is satisfied in pre-electoral years, for both municipalities affected and not affected by the DSP. Column 2 of table 4, shows that municipalities in pre-election are more sensitive than municipalities not facing elections. In particular, municipalities in pre-election year affected by the DSP react more than the other municipalities (0.73 against 0.68). This result is robust, as the ρ-test is significant at 10% level. As regards the control variables used in the static analysis, we observe: the coefficient of density is statistically significant and consistent with the presence of economies of scale; the coefficient associated with population is also significant and positive, consistent with the presence of congestion effects; and that current spending per capita is positively correlated with the presence of young people. Among the economic control variables, only grants per head have a significant impact on current spending decisions. Finally, the dummies accounting for political ideology have a positive impact on expenditures. 11
Recently, some works have applied the GMM estimators to the analysis of spatial dynamic panel models of fiscal interaction (Foucault et al., 2008; Jacobs et al., 2007).
From the static analysis, we can draw two main conclusions. Firstly, the presence of the DSP is essential for the identification of the political yardstick behaviour, if we had not considered it we would have been misled to conclude against any evidence of yardstick competition, in our sample. Secondly, the evidence of yardstick competition is significant only for the pre-electoral period, showing that the yardstick behaviour is actually predominant in the period leading to the elections rather than in the year in which elections are held. This can be justified by the fact that most of the elections in our sample are in the first semester of the year. Moreover, mimicking neighboring policies may require some time, so that mayors need to change the budget before the electoral year. In general, expenditure levels are strongly correlated with previous years levels, that is, present spending decisions are partly the result of decisions taken in previous years. Therefore, in order to check the robustness of the results of the static model, we need to specify a dynamic version of the spatial panel, accounting for the lag of the dependent variable. Column 1 of table 5, shows that the expenditure interaction among contiguous municipalities (0.11) is statistically significant also in the dynamic specification. The introduction of the border dummy, however, removes the statistical significance of the expenditure interaction term; the dummy seems to capture the spatial interaction among municipalities. In line with the static model, only when we discriminate for the presence of the DSP we find significant evidence of yardstick competition, and only the coefficient associated with the pre-election year is statistically significant.12 In the dynamic specification, however, the yardstick behaviour is confirmed only for incumbent politicians not affected by the DSP: as shown in columns 2 of table 6, they are more sensitive (0.14) than neighbouring politicians not in the pre-electoral year (0.10). In fact, we observe a negative, but not significant, sign of the expenditure interaction term for municipalities subject to the DSP. Although these coefficients are not statistically significant, the negative sign signals that the empirical analysis is sensitive to the dynamic specification of the econometric model and, therefore, some misspecification problem could affect the static analysis. Similar results are observed when the border dummy is included in the dynamic panel regressions, the main difference being that the reaction of municipalities not in pre-electoral periods not significant (see columns 3, and column 6 of table 6). Therefore, the introduction of the border dummy confirms the robustness of the reaction of municipalities not subjected to the DSP. The dynamic specification is more appropriate than the static one, as the lag of the dependent variable and ∆expit−1 are statistically significant, and the AB-AR2 test shows that the dynamic estimations are consistent. Moreover, the robustness of the results is guaranteed by the Hansen test for overidentification of the restrictions, which confirms the validity of the set of instrumental variables used in the dynamic analysis. In conclusion, the empirical analysis shows that mayors in the region Marche engage in political yardstick competition in the year before the election. This result emerges only when we account for the introduction of the DSP. The static analysis shows that both municipalities affected and not affected by the DSP engage in yardstick competition. The dynamic specification, however, shows that the yardstick hypothesis is robust only for politicians (in pre-election year) not constrained by the fiscal rule. Since the results of the dynamic specification are more robust, our analysis suggests that municipalities do engage in yardstick competition when mayors are not constrained by the fiscal rule. 12
We obtain the same results when the border dummy is excluded from the analysis of election year.
Therefore, the introduction of the fiscal rule does limit the opportunistic behaviour of mayors. Indeed, mayors may find it more difficult to finance an increase in expenditure with a deficit when they have to comply with the fiscal rule. Moreover, it could also be the case that complying with the fiscal rule acts as a signal of the incumbent quality, so that citizens may disregard signals from neighbouring jurisdictions’ policies.
The empirical analysis conducted on the panel of municipalities from the region Marche, confirms that incumbent politicians are more sensitive to neighbouring changes on expenditures when facing elections. This yardstick competition behaviour appears only when we account for the DSP, implying an important role of fiscal rules in the opportunistic behaviour of politicians. This finding suggests that the presence of the DSP is essential for the identification of the political yardstick behaviour. Both the static and the dynamic specification of the spatial panel model, show the presence of yardstick competition in our sample, in pre-election year. This result is robust for municipalities not affected by the DSP. The timing of the elections, mainly held in the first semester, is a possible reason for the evidence of yardstick only in pre-election years, as mayors may need some time to adjust to neighbouring changes in spending policies. The effect of the DSP on the opportunistic behaviour, however, deserves further investigation. The static specification shows that the DSP strengthen the intensity of the reaction of municipalities in pre-election year, while the more robust dynamic specification show a negative sign (although not significant) of the interaction term.
maximum balance deficit 0.1% of GDP
deficit arises as the difference between revenues (net of government transfers) and current expenditures (net of interests)
only in case Italy is fined by the EU
Note: pop=population size of the municipalities
Municipalities subject to DSP
reward for complying municipalities consisting in lower interest rate on borrowings
balance between revenues (net of any transfer, either from the gov., the EU, or other local governments, and also net of revenues from selling financial assets and occasional revenues) and current expenditures (net of interests, mandatory expenses, and occasional expenses)
0.1% of GDP, net of previous year deficit
forbid to hire any new employee in the following year (2002)
balance between revenues and expenditures (not from or towards other public sector bodies), excluding all revenues and expenditure stemming from fiscal federalism procedures, and always excluding payment of interests
limit of 3% increase of the deficit with respect to the level of deficit in the previous year
pop > 5000
cut in the share of government transfers (to be redistributed to complying municipalities), forbid to open hiring position, 10% reduction in current expenditures with respect to the previous year, and impossibility to finance investment expenditures with debt
same budget balance as in the previous year
limit of 2.5% increase in the deficit with respect to the previous year, and a limit of 6% to the increase in current expenditures
pop > 5000
Table 1: Normative evolution of the domestic stability pact for Italian municipalities
the same as in the previous year
same balance budget requirement as in the previous year
limit of 0% increase in balance deficit with respect to the previous year
pop > 5000
Majority (% share of votes) Left-wing coalition (1= left-wing coalition ruling; 0= otherwise) Right-wing coalition (1= rigt-wing coalition ruling; 0= otherwise) Election (1= jurisdiction in election year; 0= otherwise) Prelection (1= jurisdiction in pre-election year; 0=otherwise) Eleindex Preleindex DSP (1= budget of jurisdiction is constrained by the DSP; 0=otherwise) Border dummy
Income per-head (Euro)
0.18 0.25 0.45 0.38
0.09 0.12 0.28 0.18
Ministero dell’Interno Ministero dell’Interno Italian legal system
Ministero dell’Interno Ministero dell’Interno
242.15 11768.11 1.83 4.94 148.01
173.23 5934.04 13.10 23.25 284.49
Regione Marche - Servizio Controllo di Gestione, Ministero dell’Interno ISTAT - Atlante Statistico dei Comuni ISTAT - Atlante Statistico dei Comuni ISTAT - Atlante Statistico dei Comuni ISTAT - Atlante Statistico dei Comuni Ministero dell’Interno, Regione Marche-Servizio Controllo di Gestione Ministero dell’Interno, Ministero dell’Economia e delle Finanze, Regione Marche Ministero dell’Interno Ministero dell’Interno
Current Public Expenditure per-head (Euro)
Density (Population per km2 ) Population Young (% population 0-14) Old (% population ≥ 65) Grants per-head (Euro)
Table 2: Summary statistics and data sources Min
0 0 0
4.22 127 4.85 10.10 5.15
0.91 0.71 1
1859.08 101545 21.08 43.78 1817.34
Table 3: Estimation results of the static specification of the model without the DSP Model (1) Neigh Exp Neigh Exp*Election Neigh Exp*No Election Neigh Exp*Prelection Neigh Exp*No Prelection Election No Election Prelection No Prelection Density Population Young Old Income per-head Grants per-head Majority Left-wing coal. Right-wing coal. Constant
0.58*** (5.32) 0.54*** (5.82) 0.52*** (5.51)
-0.004 (-0.64) -0.004 (-0.72)
1.27 (1.43) 1.38 (1.53) -0.004 (-0.78)
0.63*** (6.54) 0.60*** (6.12) -0.004 (-0.61)
-1.64*** (-6.62) 1.11*** (7.27) 0.09*** (3.02) 0.07 (1.30) -0.01 (-0.24) 0.05*** (3.75) 0.001 (0.04) 0.02*** (3.54) 0.017** (2.35) -0.08 (0.903)
-1.44*** (-11.45) 0.88*** (13.84) 0.09*** (2.98) 0.058 (1.07) -0.01 (-0.29) 0.05*** (3.74) 0.001 (0.06) 0.02*** (4.05) 0.01** (2.26)
0.54 (0.57) 0.77 (0.81) -1.38*** (-10.88) 0.87*** (13.72) 0.09*** (3.14) 0.06 (1.23) -0.017 (-0.48) 0.05*** (3.79) -0.0002 (-0.01) 0.02*** (3.86) 0.016** (2.30)
0.97 0.362 0.488 0.466
0.97 0.158 0.179 0.173
Centred R2 Hansen J test ρ-test = 0 ϕ-test = 0
Note: i) results of the tests are in p-value; ii) coefficient significant at level *** 1%, ** 5%, *10%; iii) z-value in parenthesis; iv) observations: 2460.
Table 4: Estimation results of the static specification of the model with the DSP (3) Neigh Exp*Election *DSP Neigh Exp*Election * No DSP Neigh Exp*No Election Neigh Exp*No Election*DSP Neigh Exp*No Election*No DSP Neigh Exp*Prelection*DSP Neigh Exp*Prelection* No DSP Neigh Exp*No Prelection Neigh Exp*No Prelection*DSP Neigh Exp*No Prelection*No DSP Election Election*DSP Election* No DSP No Election No Election*DSP No Election*NoDSP Prelection Prelection*DSP Prelection*No DSP No Prelection No Prelection*DSP No Prelection*No DSP Density Population Young Old Income per-head Grants per-head Majority Left-wing coal. Right-wing coal.
0.63*** (7.18) 0.71*** (8.53) 0.68*** (8.11)
0.57*** (6.10) 0.65*** (7.32) 0.69*** (8.18) 0.62*** (6.50) 0.73*** (7.01) 0.68*** (7.29) 0.63*** (6.8)
0.73*** (6.48) 0.68*** (6.54) 0.69*** (6.93) 0.62*** (4.92) -0.005 (-0.83)
-0.004 (-0.66) 0.55 (0.64) 0.02 (0.02) 0.17 (0.20)
1.13 (1.27) 0.55 (0.63) 0.32 (0.38) 0.79 (0.88) -0.005 (-0.95)
-0.005 (-0.78) -0.21 (-0.22) 0.19 (0.22) 0.46 (0.5)
0.09 (0.09) 0.42 (0.43)
-1.30*** (-10.61) 0.84*** (12.90) 0.088*** (2.82) 0.05 (0.88) -0.02 (-0.53) 0.05*** (3.88) 0.003 (0.20) 0.03*** (4.19) 0.02*** (2.69)
-1.36*** (-10.85) 0.86*** (13.62) 0.10*** (3.24) 0.078 (1.44) -0.01 (-0.43) 0.05*** (3.45) -0.001 (-0.12) 0.02*** (3.83) 0.02** (2.49)
-1.37*** (-10.85) 0.86*** (13.09) 0.09*** (2.80) 0.04 (0.74) -0.02 (-0.48) 0.06*** (4.08) 0.001 (0.12) 0.02*** (4.12) 0.02** (2.50)
0.34 (0.36) 0.84 (0.75) -1.40*** (-10.35) 0.87*** (13.24) 0.09*** (3.09) 0.05 (0.91) -0.01 (-0.46) 0.05*** (3.68) -0.002 (-0.14) 0.02*** (3.65) 0.01** (2.36)
0.975 0.211 0.210 0.179
0.107 0.093* 0.088*
0.975 0.364 0.106 0.09*
0.975 0.141 0.235 0.223
Centred R2 Hansen J test ρ-test = 0 ϕ-test = 0
Note: i) results of the tests are in p-value; ii) coefficient significant at level *** 1%, ** 5%, *10%; iii) z-value in parenthesis; iv) observations: 2460.
Table 5: Estimation results of the dynamic specification of the model without the DSP (1) Neigh Exp Neigh Exp*Election Neigh Exp*No Election Neigh Exp*Prelection Neigh Exp*No Prelection Election No Election Prelection No Prelection expt−1 ∆expt−1 Density Population Young Old Income per-head Grants per-head Majority Left-wing coal. Right-wing coal. Boarder Constant
0.09 (1.28) 0.09 (0.85) 0.06 (1.11)
0.87*** (11.73) -0.10** (-2.42) -0.01 (-0.34) 0.05 (0.96) 0.25 (1.22) 0.45*** (2.77) -0.09 (-0.80) -0.05 (-0.93) 0.22 (1.44) 0.11* (1.88) -0.13 (-0.93)
0.83*** (11.26) -0.08* (-1.90) 0.001 (0.02) 0.02 (0.34) 0.39* (1.75) 0.49*** (2.73) 0.10 (0.71) -0.04 (-0.66) 0.22 (1.55) 0.06 (0.82) -0.14 (-1.00) 0.09* (1.75) -3.67**(-2.43)
-4.12** (-2.18) -3.93** (-2.45) -0.01 (-1.05) 0.79*** (11.39) -0.07* (-1.76) 0.01 (0.37) 0.02 (0.63) 0.31 (1.49) 0.49*** (3.13) 0.17 (1.48) 0.01 (0.22) 0.19* (1.74) 0.04 (0.82) -0.06 (-0.62) 0.14** (2.07)
0.08 (1.27) 0.07 (1.16) -0.001 (-0.10) -3.42** (-2.25) -3.34** (-2.17) 0.81*** (11.33) -0.08** (-2.15) 0.02 (0.66) 0.01 (0.24) 0.30 (1.32) 0.48*** (3.11) 0.12 (1.07) -0.01 (-0.20) 0.17 (1.42) 0.04 (1.11) -0.04 (-0.43) 0.15** (2.34)
AB-AR1 test 0.000 0.000 0.000 0.000 AB-AR2 test 0.261 0.453 0.420 0.427 Hansen J test 0.730 0.778 0.596 0.802 Difference Hansen J test 0.192 0.458 0.770 0.894 ρ’s test = 0 0.744 0.812 φ’s test = 0 0.745 0.782 Note: i) results of the tests are in p-value; ii) coefficient significant at level *** 1%, ** 5%, *10%; iii) z-value in parenthesis; iv) observations: 1968; v) groups number: 246.
19 0.87*** (14.28) -0.08** (-2.53) -0.001 (-0.06) 0.03 (1.10) 0.22 (1.29) 0.35*** (3.35) 0.01 (0.06) -0.01 (-0.23) 0.14 (1.39) 0.06 (1.52) -0.05 (-0.49)
0.86*** (13.05) -0.09** (-2.50) 0.02 (0.75) 0.01 (0.31) 0.22 (1.19) 0.44*** (3.23) 0.003 (0.03) -0.04 (-0.71) 0.14 (1.31) 0.05 (1.33) -0.01 (-0.11) 0.12** (2.02)
-0.58 (-0.34) -2.51* (-1.90) -2.18* (-1.64)
-0.50 (-0.30) -2.58** (-2.07) -2.28* (-1.82)
-0.16 (-1.27) 0.13** (1.97) 0.08 (1.39)
0.004 (0.07) 0.10 (1.60)
0.11 (0.83) 0.13 (1.15)
0.77*** (10.96) -0.05 (-1.38) 0.02 (0.59) 0.02 (0.74) 0.31 (1.43) 0.50*** (3.25) 0.13 (1.23) 0.02 (0.48) 0.17 (1.51) 0.04 (0.92) -0.04 (-0.44) 0.14** (2.12)
-3.11* (-1.95) -3.71** (-2.40) -0.01 (-0.85)
-3.78** (-2.02) -3.96** (-2.35)
-0.18 (-1.34) 0.14** (2.57) 0.10** (2.04)
-1.12 (-0.65) -2.49* (-1.89) -2.04 (-1.50) -2.28* (-1.74) 0.83*** (11.40) -0.07** (-2.13) 0.03 (0.78) 0.01 (0.49) 0.25 (1.30) 0.48*** (3.22) 0.004 (0.03) -0.03 (-0.52) 0.12 (1.20) 0.07 (1.57) 0.004 (0.04) 0.14** (2.16)
-0.97 (-0.61) -2.58** (-2.10) -2.19* (-1.72) -2.36** (-1.96) 0.84*** (12.54) -0.07** (-2.17) -0.003 (-0.12) 0.04 (1.40) 0.24 (1.40) 0.38*** (3.39) 0.01 (0.10) 0.003 (0.07) 0.13 (1.31) 0.07* (1.76) -0.04 (-0.39)
0.05 (0.67) 0.09 (1.44)
0.07 (1.20) 0.10** (2.05) 0.003 (0.30)
-0.10 (-0.71) 0.12* (1.68)
-0.12 (-0.88) 0.14** (2.29)
0.000 0.000 0.000 0.000 0.000 0.000 0.458 0.412 0.402 0.199 0.289 0.287 0.555 0.909 0.899 0.577 0.820 0.915 0.634 0.671 0.854 0.627 0.560 0.904 0.694 0.065* 0.077* 0.182 0.162 0.149 0.674 0.068* 0.082* 0.223 0.199 0.194 ii) coefficient significant at level *** 1%, ** 5%, *10%; iii) z-value in parenthesis; iv) observations:
0.78*** (11.53) -0.06 (-1.61) 0.01 (0.32) 0.03 (0.88) 0.32 (1.56) 0.48*** (3.39) 0.17 (1.64) 0.03 (0.62) 0.18* (1.69) 0.03 (0.73) -0.05 (-0.52) 0.13* (1.93)
-4.03** (-2.15) -4.50*** (-2.64) -3.97** (-2.55)
0.07 (0.64) 0.14 (1.27) 0.06 (1.10)
AB-AR1 test AB-AR2 test Hansen J test Difference Hansen J test rho’s test = 0 phi’s test = 0 Note: i) results of the tests are in p-value; 1968; v) groups number: 246.
Neigh Exp*Election*DSP Neigh Exp*Election*No DSP Neigh Exp*No Election Neigh Exp*No Election*DSP Neigh Exp*No Election*No DSP Neigh Exp*Prelection*DSP Neigh Exp*Prelection*No DSP Neigh Exp*No Prelection Neigh Exp*No Prelection*DSP Neigh Exp*No Prelection*No DSP Election Election*DSP Election*No DSP No Election No Election*DSP No Election*No DSP Prelection Prelection*DSP Prelection*No DSP No Prelection No Prelection*DSP No Prelection*No DSP expt−1 ∆expt−1 Density Population Young Old Income per-head Grants per-head Majority Left-wing coal. Right-wing coal. Border
Table 6: Estimation results of the dynamic specification of the model with the DSP
References Allers, M. and J. P. Elhorst (2005). Tax mimicking and yardstick competition among local governments in the Netherlands. International Tax and Public Finance 12, 493–513. Anderson, T. W. and C. Hsiao (1981). Estimation of dynamic models with error components. Journal of the American Statistical Association 76 (375), 598–606. Anderson, T. W. and C. Hsiao (1982). Formulation and estimation of dynamic models using panel data. Journal of Econometrics 18 (1), 47–82. Anselin, L. (1988). Spatial Econometrics: Methods and Models. Kluwer. Arellano, M. and S. Bond (1991). Some tests of specification for panel data: Monte carlo evidence and an application to employment equations. The Review of Economic Studies 58 (2), 277–297. Arellano, M. and O. Bover (1995). Another look at the instrumental variable estimation of error-components models. Journal of Econometrics 68 (1), 29–51. Baicker, K. (2005). The spillover effects of the state spending. Journal of Public Economics 89, 529–544. Baum, C. F., M. E. Schaffer, and S. Stillman (2003). Instrumental variables and GMM: Estimation and testing. The Stata Journal 3, 1–31. Working Paper 545, Boston College. Besley, T. and A. Case (1995). Incumbent behavior: Vote-seeking, tax setting, and yardstick competition. American Economic Review 85, 25–45. Blundell, R. and S. Bond (1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics 87 (1), 115–143. Blundell, R. and S. Bond (2000). Gmm estimation with persistent panel data: an application to production functions. Econometric Reviews 19 (3), 321–340. Bond, S. R., A. Hoeffler, and J. Temple (2001). Gmm estimation of empirical growth models. Bordignon, M., F. Cerniglia, and F. Revelli (2003). In search of yardstick competition: a spatial analysis of Italian municipality property tax setting. Journal of Urban Economics 54 (2), 199–217. Bowsher, C. G. (2002). On testing overidentifying restrictions in dynamic panel data models. Economics Letters 77 (2), 211–220. Brennan, G. and J. M. Buchanan (1980). The power to tax: Analytical foundations of a fiscal constitution. Cambridge University Press. Brueckner, J. K. (2003). Strategic interaction among Governments: an overview of empirical studies. International Regional Science Review 26, 175 – 188. Case, A. (1993). Interstate tax competition after tra86. Journal of Policy Analysis and Management 12 (1), 136–148. 20
Case, A., J. R. Hines, and H. S. Rosen (1993). Budget spill-over and fiscal policy interdependence - evidence from the States. Journal of Public Economics 52, 258–307. Devereux, M. P., B. Lockwood, and M. Redoano (2007). Horizontal and vertical indirect tax competition: Theory and some evidence from the usa. Journal of Public Economics 91 (3-4), 451–479. Elhorst, J. P. and S. Fr´eret (2009). Evidence of political yardstick competition in France using a two-regime spatial Durbin model with fixed effects. Journal of Regional Science 49 (5), 931–951. Ermini, B. and R. Santolini (2010). Local expenditure interaction in Italian municipalities. Do local council partnerships make a difference? Local Government Studies 36, 655–677. Foucault, M., T. Madies, and S. Paty (2008). Public spending interactions and local politics. Empirical evidence from French municipalities. Public Choice 137, 57–80. Franzese, R. J. (2000). Electoral and partisan manipulation of public debt in developed democracies, 1956-1990. In R. Strauch and J. Von Hagen (Eds.), Institutions, Politics and Fiscal Policy, pp. 61–83. Kluwer Academic Press. Heyndels, B. and J. Vuchelen (1998). Tax mimicking among Belgian municipalities. National Tax Journal 51, 89–101. Jacobs, J. P. A. M., J. E. Lighthart, and H. Vrijburg (2007). Consumption tax competition among governments: evidence from the United States. International Tax and Public Finance 17, 271–294. Jacobs, J. P. A. M., J. E. Ligthart, and H. Vrijburg (2009). Dynamic panel data models featuring endogenous interaction and spatially correlated errors. CentER Discussion Paper No. 2009-92. Kelejian, H. H. and I. R. Prucha (1998). A generalized spatial two stage least squares procedure for estimating a spatial autoregressive model with autoregressive disturbances. Journal of Real Estate Finance and Economics 17, 99–121. Kukenova, M. and J. A. Monteiro (2009). Spatial dynamic panel model and system GMM: a Monte Carlo investigation. MPRA Working Paper No. 14319. Ladd, H. F. (1992). Mimicking of local tax burdens among neighboring counties. Public Finance Quarterly 20, 450–467. Patrizii, V., C. Rapallini, and G. Zito (2006). I “patti” di stabilit`a interni. Rivista di diritto finanziario e scienza delle finanze 65, 1–2006. Revelli, F. (2001). Spatial patterns in local taxation: Tax mimicking or error mimicking? Applied Economics 33, 1101–1107. Roodman, D. (2009). A note on the theme of too many instruments. Oxford Bulletin of Economics and Statistics (71), 135–158. Ruiz, F. M. M. and M. Gerard (2008). Is there evidence of strategic corporate tax interaction among EU countries? MPRA Working Paper 10094. 21
Salmon, P. (1987). Decentralisation as an incentive scheme. Oxford Review of Economic Policy 3 (2), 24–43. Santolini, R. (2008). A spatial cross-sectional analysis of political trend in the Italian municipalities. Papers in Regional Science 87, 431–451. Santolini, R. (2009). The political trend in local government tax setting. Choice 139 (1-2), 125–134.
Sol´e-Oll´e, A. (2003). Electoral accountability and tax mimicking: the effects of electoral margins, coalition government, and ideology. European Journal of Political Economy 19, 685–713. Sol´e-Oll´e, A. (2006). Expenditure spillovers and fiscal interactions: Empirical evidence from local governments in spain. Journal of Urban Economics 59 (1), 32–53. Sutherland, D., R. Price, and I. Joumard (2005). Fiscal rules for sub-central governments. OECD Economics Department Working Papers No. 465. Veiga, L. G. and F. J. Veiga (2007). Political business cycles at the municipal level. Public Choice 131, 45–64. Verbeek, M. (2008). A guide to modern econometrics (3rd edition ed.). John Wiley & Sons.