The Impact of Intermunicipal Cooperation on Local Public Spending

Abstract

The purpose of this paper is to assess the effects of intermunicipal fiscal cooperation on municipal public spending, based on the French experience. A model of municipal spending choice is estimated using panel data and spatial econometrics for municipalities over the period 1994–2003. Two main results are provided. First, intermunicipal cooperation has no significant impact on the level of municipal public spending, which suggests that cooperation does not achieve its goal of reducing municipal spending by the sharing of local responsibilities. Second, there are no spending interactions between municipalities belonging to the same intermunicipal community. This is in line with the goal assigned to cooperation in terms of internalisation of spatial externalities. However, the results show that benefit spillovers remain highly significant outside intermunicipal communities.

1. Introduction

Since the 1950s, local governments in many European countries (Austria, Sweden, Finland, Germany, Switzerland, France, etc.) have been cooperating and, nowadays, encouraging cooperation among local authorities in the provision of local public goods is on the political agendas of many central and local governments (Hulst and van Montfort, 2007). There are several reasons for this widespread and persistent phenomenon of cooperation (see details for example in Blume and Blume, 2007). First, in a globalised world, larger spatial units are expected to be more competitive. Second, as governments try to reduce the cost of providing public goods, the achievement of economies of scale in the provision of local public services is a strong incentive to cooperate. Third, fiscal cooperation allows jurisdictions to internalise spending spillovers: the benefits of public expenditure (infrastructure, road building, cultural facilities, etc.) often spread across the boundaries of the supplying jurisdiction and affect the welfare of the citizens in neighbouring localities.

Despite frequent claims that cooperation among local governments is a potential solution to inefficiencies, there are few studies on that topic. This paper tries to bridge this gap using the French experience and providing estimates of the different effects of fiscal cooperation on municipal spending decisions between 1994 and 2003. The French case offers a favourable setting for research on intermunicipal cooperation. In 1971, there was an unsuccessful attempt by the central government to force the country’s municipalities to merge. Since then, and unlike the situations in Belgium, England and Germany for example, the French central government has suggested that municipalities should cooperate voluntarily within larger jurisdictions known as intermunicipal communities or Etablissements Publics de Coopération Intercommunale (EPCI). Thus, municipalities that want jointly to finance and provide some public services can create or join a community. These supramunicipal structures benefit from high levels of fiscal autonomy and co-exist with the municipal structures but have different responsibilities depending on the municipalities’ choices regarding which competences to transfer to the community.

Our central concern is to disentangle the different possible effects of cooperation on the spending behaviour of municipalities. Our first aim is to investigate how fiscal co-operation influences the level of municipal public spending. To our knowledge, very few papers analyse the impact of intermunicipal cooperation on municipal decisions although its effect is a priori not known. Some first attempts by Leprince and Guengant (2002) and Guengant and Leprince (2006), using cross-section municipal and intermunicipal spending data, show that intermunicipal spending has a significant and negative impact on municipal spending. Thus, community and municipality public goods will tend to be substitutes. However, these studies use only cross-sectional data and ignore spatial interactions between neighbouring municipalities, although this should be a central concern in the municipal spending model studied. Indeed, among other goals, fiscal cooperation is intended to internalise spending spillovers between municipalities in the same EPCI. Therefore, as well as studying the possible effect of cooperation on the level of municipal spending, we want also to develop and estimate models of municipal spending that are fully specified: they should include spatial interactions between neighbouring municipalities but also should allow adequately for the impact of fiscal cooperation on the nature and extent of these possible spatial interactions.

While many empirical papers investigate the extent of tax interactions between local governments, studies on the existence and the nature of local governments’ interactions over spending are scarce. Sollé-Ollé (2006) uses a cross-section of 2610 Spanish municipalities in 28 metropolitan areas and finds a negative spatial dependency in neighbouring municipalities’ overall spending decisions. He shows also that this broad result is driven by urban municipalities in the suburbs and that the significance of these spending interactions disappears if the focus is on non-urban or city centres’ spending decisions. Schaltegger et al. (2009) study a panel dataset of 107 Swiss municipalities in the canton of Lucerne and find that in this small metropolitan area, overall spending spatial interactions are slightly significant and positive. However, these interactions tend to be highly significant and negative for major categories of spending such as education, health, and environment. Revelli (2003) uses cross-sectional data on the spending decisions of the two overlapping levels of local government (238 districts and 34 counties) in the non-metropolitan parts of England. Empirical evidence shows that public goods provided by districts and counties tend to be complements but that when this effect is controlled for, the extent of spatial spending interactions at the lower level of local governments (districts) is low but still positive. Using panel data on the spending decisions of more than 50,000 inhabitants French municipalities, Foucault et al. (2008) provided strong empirical evidence of positive strategic interactions between the biggest French municipalities in relation to primary (i.e. both current and investment) and investment spending. However, their empirical model ignores the possible effects of fiscal cooperation on both the level of municipal spending and the extent of spatial interactions between municipalities’ spending.

The study that is the closest to ours is by Ermini and Santolini (2010). They investigate the impact of interjurisdictional agreements in Italy on the extent of spending interactions, focusing on specific categories of expenditures. They find that for the two spending categories where the partnerships are very active—police and road maintenance—strategic interactions among jurisdictions included in voluntary partnerships are lower than among isolated municipalities. This outcome suggests that spillovers may be internalised in specific cases.

This literature survey shows that several institutional contexts and empirical models with spatial interactions have been studied. However, it is difficult to make clear inferences about the effects of fiscal cooperation on the level of municipal spending and on the existence and nature (positive or negative) of spatial spending interactions between neighbouring municipalities. We try to fill these gaps using a general model of municipal spending choice, which combines spatial interactions terms and fiscal cooperation terms. Our panel dataset of French urban municipalities for the 1994–2003 period and our use of spatial econometric techniques allow us to provide two main results. First, intermunicipal cooperation has no significant impact on the level of municipal public spending, which suggests that cooperation does not achieve the goal of reducing municipal spending through the sharing of local responsibilities. Second, there are no spending interactions between municipalities belonging to the same intermunicipal community. This is in line with the objective of cooperation in relation to internalisation of spatial externalities.

The paper is organised as follows. Section 2 presents the local governments in France. Section 3 discusses the empirical design of the estimations and the data. Section 4 presents the estimation results. Section 5 concludes.

2. The French Institutional Context

French municipalities underwent huge changes at the beginning of the 1980s. The decentralisation process introduced in March 1982 greatly modified the budgetary choices of local authorities which became responsible for implementing public policies on urban infrastructures, economic and social aspects, health, supply of transport for school children, first degree education, supply of school equipment and culture. Prior to the publication of the decentralisation laws, municipalities were responsible for items such as elections, administrative and civil registration, first degree education (since the 1881 Ferry Law) and local road safety and road maintenance. The transfer of additional responsibilities to municipalities resulted in increased tax receipts and benefits deriving from higher grants from central government.

The current French local institutional context is characterised by three overlapping tiers of local government. The lowest tier consists of some 36,600 municipalities; the middle tier consists of 96 counties (or départements); and at the highest level of local government are 22 regions. Counties administer social assistance and maintain the counties’ roads and middle schools. Regions are responsible for the provision of vocational training, economic development and building, and high school provision.

Most local revenues come from taxation (54 per cent) and grants (23 per cent), the remaining 23 per cent coming from user charges and loans. The local business tax (or Taxe Professionnelle) is the major source of local government tax revenue, accounting for approximately 45 per cent of the revenues derived from direct local taxes.¹ The tax-base consists mainly of capital goods and is based on the rental values of buildings and equipment (assumed to be 16 per cent of the cost of the equipment). The remaining three taxes are collected from households in the form of residential tax (taxe d’habitation), property tax (taxe foncière sur le bâti) and land tax (taxe foncière sur le non bâti).

In 1992, 1999 and 2004, three laws were passed that increased municipalities’ scope for intermunicipal cooperation.² In practice, municipalities decide which local public services (with responsibilities such as space planning, economic development, transport and environment) will be delegated to the community. The EPCI is governed by a board of delegates elected by municipal councils among their members.³ Therefore, unlike council members in municipalities, départements or regions, EPCI officials are not directly elected by the population. However, the EPCI’s revenue comes from tax receipts, grants allocated by central government and loans, so that communities benefit from a similar level of fiscal autonomy to that of the three other levels of local governments (municipalities, départements, and regions).

Besides, in 2003, the intermunicipal map already covered a large part of the country, with 2360 EPCIs which grouped 81 per cent of the French municipalities and more than 79 per cent of the French population. On average, a community comprised 13 member municipalities and 21,000 inhabitants. Then, the 2004 law for decentralisation was passed; it recommended simplifying the organisation of intermunicipal communities: sharing of responsibilities between municipalities and the community had to be defined more precisely and municipalities had the opportunity to leave the EPCI in case of conflicts and under some conditions. As a consequence, this reform has introduced dramatic changes in some EPCIs. For this reason, we focus on the 1994–2003 period in order to avoid bias due to breaks in the data.

3. Empirical Design

In this section, we first discuss the empirical spending model used to estimate the impact of fiscal cooperation on municipal spending choices and then describe the econometric method used. We also present the data.

3.1 Empirical Models of Municipal Spending

The basic spatial model of municipal spending

As a first step, we use a simple municipal model of public spending with spatial interactions among local jurisdictions (see Brueckner, 2003; Revelli, 2005). Each municipality i chooses its spending level $Z_{i}$ , which is also affected by the level of spending chosen in the other jurisdictions $Z_{j}$ . Thus, the municipality’s objective function can be written

U (Z_{i, t}, Z_{j, t}, X_{i, t})

(1)

where $X_{i}$ is a vector of the characteristics of the municipality i.

In order to maximise its objective function, the municipality i chooses the spending level that satisfies $\partial U / \partial Z_{i . t} = 0$ . Basically, the solution of this maximisation problem is a reaction function such as

Z_{i, t} = R (Z_{j, t}, X_{i, t})

(2)

Thus, the spending decision of the municipality i depends on the other municipalities’ spending choices and on the municipality i’s characteristics. The sign of the reaction function’s slope can be positive or negative, depending on the properties of municipal preferences (Brueckner 2003, p. 177). Therefore, the specification of such a spending model can be written as

Z_{i, t} = α + β W Z_{j, t} + X_{i, t} η + ε_{i, t}

(3)

where $α$ is a constant term; $ε_{i, t}$ is a random term; $β$ and $η$ are the unknown parameters to be estimated.

The significance of parameter $β$ is expected to reveal whether there are spatial interactions between municipalities when they choose their level of public spending. A negative sign of $β$ would indicate the presence of spending spillovers: the inhabitants of neighbouring municipalities benefit from the local services provided by the municipality i. In comparison, a positive sign of $β$ would reveal local competition, due either to tax-base mobility (see Wilson, 1999, for a survey of tax competition) or to a yardstick competition mechanism with a re-election goal (Salmon, 1987; Besley and Case, 1995). Whatever the sources of spatial interactions, this model of municipal spending is not satisfactory in the French local government context because it ignores fiscal cooperation between municipalities.

The impact of intermunicipal cooperation on the level of municipal spending

In a second step, we extend this basic spatial framework in order to investigate the impact of local cooperation on municipalities’ spending choices. First, we focus on the direct effect of cooperation on the level of municipal spending by including a dummy variable Coop which is equal to 1 if the municipality i is part of an intermunicipal community, and 0 otherwise. Thus, a municipality’s policy reaction function can be written as

Z_{i, t} = R (Z_{j, t}, Coo p_{i, t}, X_{i, t})

(4)

with the same notations as those used in model (3). The equation to be estimated then becomes

Z_{i, t} = α + β W Z_{j, t} + δ Coo p_{i, t} + X_{i, t} η + ε_{i, t}

(5)

where $δ$ is expected to be negative: a municipality belonging to a community transfers some spending responsibilities that are endorsed by its community and thus the municipality’s spending level is expected to be lower.

However, this impact of communities on their municipalities may differ according to the magnitude of their expenditures. Therefore, in a second estimation, we replace the Coop dummy with $Z_{I, t}$ , the spending level of the community I at time t. We estimate the following model

Z_{i, t} = α + β W Z_{I, t} + δ Z_{I, t} + X_{i, t} η + ε_{i, t}

(6)

This allows us to test whether intermunicipal and municipal public goods and services are either independent (in the case of a non-significant estimated value of $δ$ ), substitutes (which would imply a negative impact of $Z_{I, t}$ ) or complements (which would imply a positive impact of $Z_{I, t}$ ).

The impact of cooperation on the extent of spatial interactions between municipalities

Besides the direct effect of cooperation on the level of municipal spending, we focus on the indirect effect of cooperation on the extent of spatial interactions in municipal spending. To do so, we decompose the spatial interaction term $W Z_{j, t}$ used in equations (3), (5) and (6) because it rests on the simplifying hypothesis that interactions do not differ whether or not neighbouring municipalities are members of the same intermunicipal community. However, this hypothesis should be studied more deeply: by sharing some responsibilities, municipalities that are members of the same community intend precisely to internalise spillovers in order to improve the efficiency of local public spending within the community. To study this effect of intermunicipal cooperation on the nature and the extent of spatial interactions, we decompose $W Z_{j, t}$ into two terms: $W^{SAME} Z_{j, t}$ and $W^{OTHER} Z_{j, t}$ . The first term ( $W^{SAME} Z_{j, t}$ ) allows us to estimate the impact of the spending of neighbouring municipalities belonging to the same intermunicipal community as the municipality i. The second term ( $W^{OTHER} Z_{j, t}$ ) allows us to estimate the impact of the spending of neighbouring municipalities that do not belong to the same community as municipality i. In other words, we distinguish intracommunity ( $W^{SAME} Z_{j, t}$ ) and extracommunity ( $W^{OTHER} Z_{j, t}$ ) spatial spending interactions among municipalities.

Using this detailed identification of spatial interactions, we can specify two spatial interactive terms, so that the model to be estimated is as follows

Z_{i, t} = α + μ W^{SAME} Z_{j, t} + σ W^{OTHER} Z_{j, t} + δ Z_{I, t} + X_{i, t} η + ε_{i, t}

(7)

Since cooperation is expected to internalise externalities among local governments, we expect a lower absolute value for the parameter $μ$ (which measures the extent of intracommunity spatial spending interactions) than for the parameter $σ$ (which measures the extent of extracommunity spatial spending interactions). As shown by the discussion on $β$ in equation (3), a negative sign of $μ$ (resp. $σ$ ) will indicate that there are significant spending spillovers among cooperation (resp. non-cooperating) municipalities. In contrast, a positive sign would indicate that there are spending interactions between neighbouring municipalities due either to tax-base or yardstick competition.

Now, intermunicipal cooperation is expected to reduce the extent of spatial interactions in municipal spending, whatever its source. On the one hand, since intermunicipal cooperation allows the financing and production of local public goods in larger geographical areas, it partially internalises spending spillovers. On the other hand, it also reduces the number of competing local government units and therefore weakens the intensity of fiscal competition (Hoyt, 1991). Therefore, we expect $μ$ to be significantly smaller than $σ$ .

3.2 Econometric Issues

The empirical literature on the spatial spending model highlights two main econometric issues, which we need to deal with in our estimation strategy.

First, we need to identify the precise nature of the spatial interaction phenomenon we analyse in our model. This is especially important since one of the possible effects of intermunicipal cooperation we focus on is an effect on the extent of spatial interactions. Following the empirical literature, we choose a geographical definition of neighbourhood based on the Euclidean distance between municipalities. This scheme imposes a smooth distance decay and the weights w_ij are given by 1/d_ij where d_ij is the Euclidian distance between the municipalities i and j for j≠i. However, in order to ensure that our estimation results are not specific to one (and only one) definition of neighbourhood, we checked their robustness by testing various weight matrices. We study three matrices

—W^{DIST<20 km}, where w_ij = 1/d_ij if d_ij < 20 km, and w_ij = 0 otherwise;

—W^DIST<15km, where w_ij = 1/d_ij if d_ij < 15km, and w_ij = 0 otherwise;

—W^CTG, where w_ij = 1 if j is contiguous to i, and w_ij = 0 otherwise.

All these weight matrices are standardised so that $\sum_{j} w_{ij} = 1, \forall i$ .

Note that, although our study is based on urban municipalities, we do not limit our empirical approach to spatial interactions between these urban units, in order to avoid possible border effects. Indeed, such a restriction would bias the estimation results: it would implicitly assume that urban areas are like ‘islands’ in the country and that urban municipalities only interact with other neighbouring urban municipalities but not with neighbouring rural municipalities. To avoid this, we allow spatial interactions to take place between urban and rural municipalities, and we use an extended definition of neighbouring municipalities. More specifically, the neighbouring municipalities’ decisions on spending—either with the W matrix in (3), (5) or (6) or with the W^SAME and W^OTHER matrices in (7)—are computed for urban municipalities and also for all municipalities (even rural) considered to be neighbours—that is, all municipalities located within a distance of 20 km, or a distance of 15 km or which are contiguous.

However, our model (7), which distinguishes between intracommunity and extracommunity spatial spending interactions with the W^SAME and W^OTHER matrices, raises a second border effect problem. In order to illustrate this, we depict the case of two municipalities (Talant and Perrigny) belonging to the same community: the Communauté d’agglomération du Grand Dijon (see Figure 1). Those two municipalities have four contiguous neighbours, but Talant is located in the middle of the community whereas Perrigny is located on its border. Thus, Talant’s four neighbours belong to the same community (municipalities in dark grey in Figure 1.1) but none of them is outside the community. In comparison, Perrigny has one neighbour that belongs to the same community (municipality in dark grey in Figure 1.2) and three neighbours outside the community (municipalities in light grey in Figure 1.2). As a consequence, Talant’s W^OTHER matrix is empty when based on the contiguity neighbourhood definition. If we compare this with the case where we use the 20 km neighbourhood definition (see Figures 1.3 and 1.4), W^SAME and W^OTHER are not empty for either municipality. This led us to prefer the distance-based neighbourhood definition over the contiguity based neighbourhood definition in our model (7) with two spatial interactions matrices.

Figure 1.

Contiguity matrix vs distance-based matrix in the refined model: an illustrative example with the Communauté d’agglomération du Grand Dijon.

The second econometric issue rests on the endogeneity of our spatially lagged variable. Indeed, if localities do react to each others’ spending choices, then neighbours’ spending decisions are endogenous and correlated with the error term ( $ε$ ). As a consequence, ordinary least squares (OLS) yields a biased parameter estimate (Anselin, 1988). Basically, there are two possible approaches that provide consistent estimates of the spatial parameters in our equations of municipal spending. The first is based on an instrumental variables (IV), two-stage least squares (2SLS) method. It consists of finding variables that are correlated with neighbours’ spending choices but uncorrelated with the error term. The IV approach suggests the use of the weighted average of neighbours’ exogenous or control variables (WX) as instruments (Kelejian and Robinson, 1993; Kelejian and Prucha, 1998). The second method is based on maximum likelihood (ML). Under this method, a non-linear reduced form of the estimated equations is computed by inverting the system. A non-linear optimisation routine is used to estimate the spatial coefficient.

In this paper, we compute heteroscedasticity-robust IV/GMM estimators for three main reasons. First, we suspect endogeneity in our main variable of interest (the Coop_i,t dummy or the intermunicipal spending level Z_I,t). However, Fingleton and Le Gallo (2007, 2008) show that IV/GMM estimators are useful in those cases where spatial dependence models contain one or more endogenous explanatory variables other than the spatially lagged dependent variable (see also Elhorst, 2010, p. 15). Secondly, we have chosen a broad definition of neighbourhood in order to avoid border effects. As a consequence, neighbours can belong to rural areas that are not part of our initial sample of urban municipalities and, in that case, the usual ML routines cannot be used. Thirdly, since local public spending may be persistent over time, serial correlation may emerge. Basically, in order to estimate a spatial dynamic model, we have two options.

The first is to estimate a spatial dynamic panel model à la Arellano and Bond (1991), including time-lagged municipal spending as an explanatory variable. However, this method introduces some correlation with the municipal fixed effect and requires the time-lagged dependent variable to be instrumented.

The alternative option is to estimate a non-dynamic spatial panel model by a method providing results that are robust to serial correlation. Here, we can use the Arellano (1987) clustering method in the following way: since it provides robust standard errors for intracluster estimators, we define a cluster for each municipality. Therefore, we obtain estimation results that are robust to any temporal correlation for each municipality—i.e. results robust to serial correlation and time persistency of local public spending. Since we already have a high number of instruments, we chose this alternative way to tackle this issue.

Finally, if neighbours’ localities are subject to correlated shocks, we may find correlation between jurisdictions’ spending choices. The omission of explanatory variables that are spatially dependent could generate spatial dependence in the error term, which is given by the following equation

ε_{i, t} = W ε_{i, t} + u_{i, t}

(8)

If we ignore spatial error dependence, estimation of our equations could provide false evidence of strategic interaction (for example Case et al., 1993). In order to identify the source of spatial autocorrelation, we usually compute the LM and RLM tests. Then, a SAR or a SEM specification of the model is estimated. However, in order to avoid important border effects that could bias our estimation results, our data sample includes some buffer zones (municipalities in light grey on Figure 2)—i.e. some non-urban municipalities within a 20 km distance from an urban municipality. In that case, the LM tests routines cannot be run because they require the Z_i,t vector to be the same as the Z_j,t vector.⁴ In other words, the neighbouring municipalities should be exactly the same as those municipalities where we explain spending levels. To tackle this issue, we use the IV approach developed by Kelejian and Prucha (1998, 1999) and based on the generalised moments method (GMM) to provide estimation results robust to spatial error dependency.

Figure 2.

Spatial distribution of the French urban municipalities studied and their neighbours.

Also, we control both for individual fixed effects using within differences (each variable is expressed in difference to the individual mean), and for year fixed effects using a set of nine year dummies.

3.3 Data

Our study focuses on urban municipalities⁵ located in the French metropolitan areas (pôle urbain according to the INSEE).⁶ We selected only those urban municipalities that existed throughout our entire period of study (1994–2003), which yielded a sample of 2895 municipalities over 10 years and a total of 28,950 observations. Expenditures data come from the Direction Générale des Collectivités Locales (DGCL, Ministère de l’Intérieur) for municipal governments⁷ and the remaining control variables from French census data. Descriptive statistics are given in Table 1.

Table 1.

Descriptive statistics, 1994–2003

Description	Symbol	Mean	S.D.	Minimum	Maximum
log(municipality’s expenditure in hundred € p.c.)	Z_i,t	2.34	0.48	−1.57	5.45
W^{DIST < 20 km}* log(municipality’s expenditure in hundred € p.c.)	W Z_i,t	2.19	0.27	1.45	4.14
W^{SAME, DIST< 20 km}* log(municipality’s expenditure in hundred € p.c.)	W^SAME Z_i,t	1.39	1.11	0.00	3.79
W^{OTHER, DIST< 20 km}* log(municipality’s expenditure in hundred € p.c.)	W^OTHER Z_i,t	2.16	0.29	0.00	4.14
Cooperation dummy	Coop_i,t	0.62	0.48	0.00	1.00
log(community’s expenditure in hundred € p.c. + 1)	Z_I,t	0.51	0.55	0.00	3.10
log(population density)	Density_i,t	6.15	1.23	2.84	10.11
log(proportion of the population below 14 y.o. + 1)	Pct_Young_i,t	0.18	0.03	0.09	0.33
log(proportion of the population above 60 y.o. + 1)	Pct_Old_i,t	0.17	0.04	0.02	0.38
log(after-tax yearly mean income)	Mean_income_i,t	5.16	0.27	4.17	6.90
log(tax capacity in hundred € p.c. + 1)	Tax_capacity_i,t	1.80	0.49	0.00	4.40
log(grants received by the municipality in hundred € p.c. + 1)	Grant_i,t	0.95	0.28	0.00	2.26
Number of municipalities					2895
Number of observations					28,950

Note: Monetary terms are expressed in constant euros, base 2005.

Figure 2 shows the spatial distribution of the French urban municipalities that we study in this paper. Recall that neighbours’ spending decisions are computed over our 2895 urban municipalities and also over all municipalities considered as neighbours (i.e. located within 20 km) to avoid possible border effects between urban and rural municipalities.

In line with the literature, we include four socio-demographic controls and two budgetary variables that might influence local demand for municipal public goods and services

—Population density (Density_i,t) is expected to have a positive sign since a big city is likely to supply a high level of local public goods to citizens living within its borders as well as to the citizens of neighbouring localities.

—Proportions of the population aged below 14 years (Pct_Young_i,t) and above 60 years (Pct_Old_i,t) should take account of municipal demographic heterogeneity influencing demand for local spending. We would expect these control variables to be associated with a positive sign because of the particularly high demand from these two populations for local public services.

—Since we expect demand to be higher when income increases, after-tax yearly mean income (Mean_Income_i,t) should have a positive impact on the level of municipal expenditures, meaning that municipal public goods are normal goods.

—A municipal tax capacity per capita variable (Tax_Capacity_i,t) is included as an explanatory variable to control for the fact that wealthier municipalities can afford higher levels of public spending. Tax capacity is defined and computed by central government as the tax revenues a local government would receive were its local tax-bases taxed at the mean tax rates calculated across the country for the same level of local government.

—A municipal grant variable (Grant_i,t) refers to the amount per capita of Dotation Globale de Fonctionnement (DGF), the main grant that each municipality receives from central government according to a national formula that rests mainly on municipal population.

4. Results

We present our estimation results in the following steps. In a first step (4.1), we provide the results for the estimations of models (3), (5) and (6). Model (3) is the basic spatial model of municipal spending and ignores fiscal cooperation. Models (5) and (6) introduce the potential effect of cooperation on the level of municipal spending. In a second step (4.2), we refine our estimation strategy and investigate whether spatial interactions among municipalities are different if we introduce two types of matrix in the model, to control for the fact that some municipal neighbours belong to the same community but others do not. This gives us the results for the estimation of model (7).

4.1 Baseline Results

Table 2 in column (2.1) presents the estimation results of model (3) and the results for models (5) and (6) are in the remaining columns. Columns (2.2) and (2.3) include the Coop dummy as an explanatory variable to capture the effect of fiscal cooperation on the level of municipal spending. Columns (2.4) and (2.5) provide the estimation results including the level of intermunicipal spending in the model. In all the columns in Table 2, spatial lag variables with a unique distance matrix W are instrumented (see the instruments in the notes under Table 2) while the variables of interest (i.e. Coop and Z_I,t) are instrumented only in columns (2.3) and (2.5).

Table 2.

Estimation results of the simple spatial model of municipal spending using a distance-based weight matrix with a threshold at 20 km, 1994–2003

	(2.1)	(2.2)	(2.3)	(2.4)	(2.5)
	GMM-cluster	GMM-cluster	GMM-cluster	GMM-cluster	GMM-cluster
	W^DIST<20km	W^DIST<20km	W^DIST<20km	W^DIST<20km	W^DIST<20km
Parameter estimates (p-values)
W Z_j,t	0.744**	0.740**	0.775*	0.744**	0.703*
	(0.002)	(0.002)	(0.023)	(0.002)	(0.024)
Coop_i,t	—	−0.002	0.019	—	—
	—	(0.715)	(0.899)	—	—
Z_I,t	—	—	—	0.001	−0.029
	—	—	—	(0.927)	(0.839)
Density_i,t	−0.037	−0.036	−0.038	−0.037	−0.037
	(0.614)	(0.616)	(0.604)	(0.615)	(0.609)
Pct_Young_i,t	0.315	0.318	0.291	0.315	0.304
	(0.337)	(0.333)	(0.438)	(0.337)	(0.360)
Pct_Old_i,t	0.127	0.127	0.126	0.127	0.123
	(0.652)	(0.652)	(0.652)	(0.651)	(0.664)
Mean_Income_i,t	−0.051	−0.051	−0.052	−0.051	−0.050
	(0.061)	(0.062)	(0.064)	(0.061)	(0.070)
Tax_Capacity_i,t	0.122***	0.122***	0.122***	0.122***	0.123***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Grant_i,t	0.020	0.020	0.021	0.020	0.020
	(0.633)	(0.634)	(0.629)	(0.633)	(0.637)
Year_Fixed_Effects_t	yes***	yes***	yes***	yes***	yes***
General statistics
Number of clusters	2895	2895	2895	2895	2895
Number of observations	28,950	28,950	28,950	28,950	28,950
Validity of instruments
Hansen J statistic	2.304	2.310	2.260	2.306	2.274
	(0.512)	(0.511)	(0.323)	(0.511)	(0.321)
1st step R²
W Z_j,t	0.4316	0.4322	0.4316	0.4317	0.4316
Coop_i,t	—	—	0.2217	—	—
Z_I,t	—	—	—	—	0.2831

Notes: p-values in parentheses * p < 0.05; ** p < 0.01; *** p < 0.001. (2.1)–(2.5) instruments: spatial lag with the W^DIST<20km weight matrix of pct_young, pct_old, mean_income, grant. The Hansen J statistic follows a chi-squared distribution under the joint null hypothesis that the instruments are valid instruments—i.e. that they are uncorrelated with the error term and that the excluded instruments are correctly excluded from the estimated equation. Besides, this statistic is robust to within-cluster correlation. In our results, we are always in the acceptance region, which allows us to conclude that our sets of instruments are always valid.

Estimation results of the basic spatial model (3) indicate that there are very significant spatial interactions between municipalities located within a distance smaller than 20 km. More specifically, the estimated spatial lag parameter is 0.744 (column 2.1 in Table 2). This result holds with a small variation in the estimated parameter when a direct effect of cooperation on the level of municipal spending is introduced in the model (see the remaining columns in Table 2). This is evidence of positive spatial interactions in public spending between municipalities and is consistent with the evidence in Jayet et al. (2002) and Charlot et al. (2010) for different cross-sections of municipal tax rates. However, this is the first evidence based on panel data and inclusion of spatial interactions in municipal spending for municipalities of different sizes included in the same ‘bassin de vie’. Foucault et al. (2008) found positive spatial interactions but focused on a panel of municipalities with more than 50,000 inhabitants and ignored the possible effect of cooperation on municipal spending.

Finally, in our spatial model, apart from significant time fixed effects, only one explanatory variable appears to be significant at 5 per cent: a higher tax capacity per capita leads to higher municipal spending, which is in line with results obtained in previous studies on the French case (see for example Leprince and Guengant, 2002). The other explanatory variables in the model are never significant in Tables 2 and 3.

Table 3.

Estimation results of the refined spatial model of municipal spending using a distance based weight matrix with a threshold at 20 km, 1994–2003.

	(3.1)	(3.2)	(3.3)	(3.4)	(3.5)
	GMM-cluster	GMM-cluster	GMM-cluster	GMM-cluster	GMM-cluster
	W^DIST<20km	W^DIST<20km	W^DIST<20km	W^DIST<20km	W^DIST<20km
Parameter estimates (p-values)
W^SAME Z_j,t	−0.006	0.049*	0.029	−0.005	0.022
	(0.058)	(0.021)	(0.183)	(0.219)	(0.266)
W^OTHER Z_j,t	−0.224*	−0.147	−0.164	−0.232*	−0.360**
	(0.013)	(0.125)	(0.086)	(0.012)	(0.006)
Coop_i,t	—	−0.122**	−0.078	—	—
	—	(0.010)	(0.107)	—	—
Z_I,t	—	—	—	−0.006	−0.148
	—	—	—	(0.648)	(0.153)
Density_i,t	0.081	0.057	0.064	0.081	0.070
	(0.263)	(0.429)	(0.374)	(0.260)	(0.343)
Pct_Young_i,t	0.426	0.432	0.435	0.421	0.306
	(0.205)	(0.194)	(0.193)	(0.210)	(0.378)
Pct_Old_i,t	0.380	0.364	0.353	0.378	0.338
	(0.184)	(0.200)	(0.213)	(0.186)	(0.242)
Mean_Income_i,t	−0.041	−0.040	−0.040	−0.041	−0.036
	(0.140)	(0.154)	(0.154)	(0.141)	(0.209)
Tax_Capacity_i,t	0.126***	0.125***	0.124***	0.127***	0.131***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Grant_i,t	0.015	0.014	0.016	0.015	0.0122
	(0.737)	(0.739)	(0.717)	(0.736)	(0.783)
Year_Fixed_Effects_t	yes***	yes***	yes***	yes***	yes***
General statistics
Number of clusters	2895	2895	2895	2895	2895
Number of observations	28,950	28,950	28,950	28,950	28,950
Validity of instruments
Hansen J statistic	5.464	1.773	2.993	5.326	3.251
	(0.362)	(0.621)	(0.559)	(0.377)	(0.517)
1st step R²
W^{SAME Z_j,t}	0.9752	0.9752	0.9752	0.9752	0.9752
W^{OTHER Z_j,t}	0.3854	0.3849	0.3854	0.3859	0.3854
Coop_i,t	—	—	0.9988	—	—
Z_I,t	—	—	—	—	0.6496

Notes: p-values in parentheses * p < 0.05; ** p < 0.01; *** p < 0.001. (3.1), (3.3), (3.4), (3.5) instruments: spatial lag with the W^SAME weight matrix of pct_young, mean_income, tax_capacity, grant; spatial lag with the W^OTHER weight matrix of density, pct_young, tax_capacity. (3.2) instruments: spatial lag with the W^SAME weight matrix of tax_capacity, grant; spatial lag with the W^OTHER weight matrix of density, pct_young, tax_capacity. The Hansen J statistic follows a chi-squared distribution under the joint null hypothesis that the instruments are valid instruments—i.e. that they are uncorrelated with the error term and that the excluded instruments are correctly excluded from the estimated equation. Besides, this statistic is robust to within-cluster correlation. In our results, we are always in the acceptance region, which allows us to conclude that our sets of instruments are always valid.

Result 1. When controlling for the possible effect of cooperation on the level of municipal spending, our baseline estimation results show that spatial interactions in municipal spending are very significant and positive between French municipalities.

In the next step, we focus on the effect of cooperation on the level of municipal spending; we provide our results in Table 2 column (2.3) (resp. 2.5) controlling for possible endogeneity of Coop (resp. the level Z of intermunicipal spending). We show that the estimated parameters are never significantly different from zero: local cooperation does not modify the level of municipal public spending. This outcome suggests that the single fact that an urban municipality is a member of an intermunicipal community does not lead to significantly different spending behaviour, compared with isolated municipalities.

Result 2. Intermunicipal cooperation per se does not have any impact on the level of municipal spending.

There are several possible reasons for this result of independence between municipal spending and intermunicipal or community spending. First, the community might provide public goods not previously supplied by rather small municipalities, such as a public swimming pool: this is the ‘zoo effect’ identified by Oates (1988) (see some evidence for France in Frère et al., 2011). In such cases, the scope of municipal public goods is not reduced by intermunicipal co-operation and both levels of public spending are independent. Second, in some cases, there may be two phenomena, which compensate for each other (see Leprince and Guengant, 2002). On the one hand, local cooperation among municipalities is a more efficient way to provide public goods that municipalities were already providing before the cooperation. In such cases, the Coop dummy would be significant and negative because the scope of municipal provision is reduced by cooperation. On the other hand, municipalities may react to the extended scope of intermunicipal provision of public goods by improving the quality of the supplied goods and services or by extending the scope of their public goods provision to satisfy previously unfulfilled demand. In some communities, this municipal behaviour might induce a positive impact of the Coop dummy. These effects with opposite signs might be compensating, producing a non-significant result overall.

Finally, we observe that our Result 2 is obtained in a spatial model of municipal spending. When the spatial lag variable is omitted,⁸ the effect of fiscal cooperation on municipal spending becomes significant and negative in our panel approach, in line with cross-sectional evidence provided by Leprince and Guengant (2002). However, these significant effects of cooperation are obtained from a municipal model that is incorrectly specified because it ignores significant spatial interactions in spending, leading to biased estimation results (see next sub-section). Thus, our central concern in the rest of our paper is to correctly introduce spatial effects and to test alternative weight matrices in spending models designed to study the effects of intermunicipal cooperation fully.

4.2 Municipal Spending Models where Cooperation Impacts on the Extent of Spatial Interactions

Our baseline Results 1 and 2 suggest significant spatial interactions in municipal spending, and a non-significant effect of fiscal cooperation on the level of municipal spending. However, the spatial nature of our spending model should be studied in more depth in order to test the robustness of our first results.

First, we want to make sure that different weight matrices would lead to the same results. Indeed, our spatial model rests on one weight matrix, which defines neighbours as municipalities located within a distance of 20 km. To provide robustness checks, we estimate our models with two other matrices: a 15 km distance, and a contiguity matrix, where neighbours are the only municipalities that share a common border with the urban municipalities scrutinised in our study.

Results are available upon request and provide two new insights. First, whatever the matrices used, Result 2 remains valid: the impact of cooperation on the level of municipal expenditures is never significant. Estimation of the direct impact of cooperation on municipal spending is thus not dependent on the way spatial interactions are modelled. However, concerning Result 1, we show that spatial weights matter. First, the 15 km distance matrix leads to spatial interactions that remain significant but with a somewhat lower absolute value of the spatial lag parameter. Second, the contiguity matrix leads to non-significant spatial interactions, the rest of the model being stable with the tax capacity variable remaining the main significant variable. These results obtained when the distance between urban municipalities and their neighbours declines contradict our a priori, according to which the smaller the distance between urban units, the higher spatial interactions.

We think that these results suggest the need to examine more carefully the nature of spatial interactions in our model. Our hypothesis is that the spatial models used at the beginning of this paper neglect a second effect of cooperation on municipal behaviour. More specifically, models (3)–(6) rest on the hypothesis that interactions do not differ if neighbouring municipalities are members of the same intermunicipal community. We intend to analyse this hypothesis more deeply since members of the same community share spending responsibilities in order, among other objectives, to internalise spatial externalities. Thus, cooperation should play a role in the nature and in the extent of spatial interactions among neighbouring municipalities.

To study the possible effect of cooperation on spatial interactions, we decompose $W Z_{j, t}$ into two terms, $W^{SAME} Z_{j, t}$ and $W^{OTHER} Z_{j, t}$ . As explained in section 3, the estimation of the $W^{SAME} Z_{j, t}$ parameter allows us to estimate the impact on the urban municipalities i of neighbouring municipal spending only if the neighbours j are included in the same intermunicipal community as the municipality i: thus, these spatial interactions come from intracommunity interactions among municipalities. The second term $W^{OTHER} Z_{j, t}$ allows us to estimate the impact of neighbouring municipalities when neighbours j do not belong to the same community as municipality i. This is model (7) presented in section 3.

The results are displayed in Table 3. While the rest of the model remains stable, with a significant impact of the tax capacity variable and a non-significant impact of cooperation on the level of municipal spending, the distinction between the two types of spatial interactions strongly modifies the estimation results in a way that is more in line with expectations.⁹ When the possible endogeneity of the variables Coop_i,t, (or Z_I,t) and $W^{SAME} Z_{j, t}$ and $W^{OTHER} Z_{j, t}$ is accounted for in columns (3.3) and (3.5), the results strongly support our hypothesis that cooperation modifies the nature of spatial interactions between municipalities in their spending behaviours. Indeed, the estimated coefficient of $W^{SAME} Z_{j, t}$ is not significant in these columns.

Result 3. There are no spending interactions among municipalities belonging to the same intermunicipal community. Intermunicipal cooperation seems to internalise spending spillovers among cooperating local governments.

The absence of spending interactions between neighbours in the same community may be one of the main consequences of the decrease in responsibilities when municipalities belong to a community. This is the expected result of internalisation of spatial externalities, either in terms of benefit spillovers or in terms of incentives to mimic neighbours’ decisions (in tax competition or yardstick competition models). Therefore, our results suggest that unlike expectations, intermunicipal cooperation does not significantly cut municipalities’ spending levels (Result 2) but it efficiently reduces spatial spending interactions between the municipalities that are members of a same community.¹⁰

On the other hand, the estimated parameter of $W^{OTHER} Z_{j, t}$ is significant¹¹ and negative. This suggests the existence of significant spending spillovers among neighbouring municipalities that do not belong to the same community. Since the coefficient is negative, spatial interactions among local governments cannot be induced by yardstick competition or spending competition over a mobile tax-base. Similar to the results obtained by other studies (see, for example, Solé-Ollé, 2006; and Schaltegger et al., 2009) when studying local government behaviours in spatial models, this result confirms that there are strong spending interactions among neighbouring municipalities outside intermunicipal communities. This result is in line with a well-known special report by the Cour des Comptes (2005), according to which communities are too small, especially in urban areas where benefit spillovers are especially high and may be more efficiently internalised by enlarging the existing communities. For this reason, a law passed in 2010 recommends merging communities and even forcing smaller ones with less than 5000 inhabitants to merge. Thus a larger community is expected to favour greater internalisation of overall spatial spending spillovers among municipalities, so that the local public spending level becomes closer to the socially optimal level.

Result 4. Benefit spillovers among municipalities that do not belong to the same community remain significant, suggesting that the communities’ size is too small to internalise spatial spending externalities fully.

5. Conclusion

Intermunicipal cooperation is a widespread phenomenon occurring in most European countries, but its different effects on municipal choices are not a priori known since it might combine reduced costs due to economies of scale, internalisation of spending spillovers and reduced tax competition, among other expected effects.

The main aim of this paper was to test the impact of intermunicipal cooperation on municipal spending behaviour. We estimated a model of municipal spending choice using panel data for 1994–2003. We found that intermunicipal cooperation does not have per se any impact on the level of municipal spending: contrary to our expectations, municipalities that belong to a community—and where therefore some competences previously supplied by the municipality have been transferred to the community—do not spend less per capita than municipalities that do not belong to a community. We found also that intermunicipal cooperation internalises spending spillovers among municipalities in the same community. However, we found that benefit spillovers among municipalities not members of the same community remain significant, suggesting that communities are too small as regards this particular goal of intermunicipal cooperation.

We believe that this paper will contribute to the on-going debate on the reorganisation of sub-national jurisdictions since our work promotes the idea that cooperation through the creation of a new level of local government (i.e. the intermunicipal community) may reduce spending spillovers among cooperating local governments. Future work could extend the research described here, in various ways. For example, it could investigate whether the results change if different categories of public expenditures are considered.

Footnotes

Acknowledgements

The authors would like to thank S. Charlot, J. P. Elhorst, H. Hammadou, V. Piguet, S. Riou, F. Revelli, Y. Rocaboy, participants at the 2011 NARSC conference, at the 2011 EPCS meeting, at the 5th J2E workshop and at the 28th JMA conference, as well as two anonymous referees for helpful comments and suggestions.

Funding

Part of this research has been funded by the French Agence Nationale de la Recherche (ANR), grant number ANR-08-GOUV-054.

Notes

References

Anselin

(1988) Spatial Econometrics: Methods and Models. Dordrecht: Kluwer Academic Publishers.

Arellano

(1987) Computing robust standard errors for within-group estimators, Oxford Bulletin of Economics and Statistics, 49, pp. 431–434.

Arellano

Bond

(1991) Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations, The Review of Economic Studies, 58, pp. 277–297.

Besley

T. J.

Case

(1995) Incumbent behavior: vote-seeking, tax-setting, and yardstick competition, The American Economic Review, 85, pp. 25–45.

Blume

(2007) The economic effects of local authority mergers: empirical evidence for German city regions, Annals of Regional Science, 41, pp. 689–713.

Brueckner

J. K.

(2003) Strategic interaction among governments: an overview of empirical studies, International Regional Science Review, 26, pp. 175–188.

Case

A. C.

Rosen

H. S.

Hines

J. R.

Jr (1993) Budget spillovers and fiscal policy interdependence: evidence from the states, Journal of Public Economics, 52, pp. 285–307.

Charlot

Paty

Piguet

(2010) The effects of fiscal cooperation on local taxation: the French case. Working Paper No. 2010–1, CESAER-INRA, Dijon.

Cour des Comptes (2005) L’intercommunalité en France. La documentation française (http://www.ladocumentationfrancaise.fr/rapports-publics/054004449/index.shtml).

10.

Elhorst

J. P.

(2010) Applied spatial econometrics: raising the bar, Spatial Economic Analysis, 5, pp. 9–28.

11.

Ermini

Santolini

(2010) Local expenditure in Italian municipalities: do local council partnerships make a difference?, Local Government Studies, 36, pp. 655–677.

12.

Fingleton

Le Gallo

(2007) Finite sample properties of estimators of spatial models with autoregressive, or moving average disturbances and system feedback, Annales d’économie et de statistiques, 87/88, pp. 39–62.

13.

Fingleton

Le Gallo

(2008) Estimating spatial models with endogenous variables, a spatial lag and spatially dependent disturbances: finite sample properties, Papers in Regional Science, 87, pp. 319–339.

14.

Foucault

Madies

Paty

(2008) Public spending interactions and local politics, Public Choice, 135, pp. 57–80.

15.

Frère

(2012) Coopération intercommunale et offre de biens publics locaux: une approche microéconomique des enjeux et des choix politiques des collectivités territoriales. PhD thesis, University of Burgundy.

16.

Frère

Hammadou

Paty

(2011) Range of local public services and population size: is there a ‘zoo effect’ in French jurisdictions?, Louvain Economic Review, 77, pp. 87–104.

17.

Guengant

Leprince

(2006) Évaluation des effets des régimes de coopération intercommunale sur les dépenses publiques locales, Économie et Prévision, 175, pp. 79–99.

18.

Hoyt

W. H.

(1991) Property taxation, Nash equilibrium and market power, Journal of Urban Economics, 30, pp. 123–131.

19.

Hulst

Montfort

A. van

(Eds) (2007) Inter-municipal Cooperation in Europe. Amsterdam: Springer.

20.

Jayet

Paty

Pentel

(2002) Existe-t-il des interactions fiscales stratégiques entre les collectivités locales?, Economie et Prévision, 154, pp. 95–105.

21.

Kelejian

H. H.

Prucha

I. R.

(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, pp. 99–121.

22.

Kelejian

H. H.

Prucha

I. R.

(1999) A generalized moments estimator for the autoregressive parameter in a spatial model, International Economic Review, 40, pp. 509–534.

23.

Kelejian

H. H.

Robinson

D. P.

(1993) A suggested method of estimation for spatial interdependent models with autocorrelated errors, and an application to a county expenditure country police expenditure, Papers in Regional Science, 72, pp. 297–312.

24.

Leprince

Guengant

(2002) Interactions fiscales verticales et réaction des communes à la coopération intercommunale, Revue Economique, 53, pp. 525–535.

25.

Oates

W. E.

(1988) On the measurement of congestion in the provision of local public goods, Journal of Urban Economics, 24, pp. 85–94.

26.

Revelli

(2003) Reaction or interaction? Spatial process identification in multi-tiered government structures, Journal of Urban Economics, 53, pp. 29–53.

27.

Revelli

(2005) On spatial public finance empirics, International Tax and Public Finance, 12, pp. 475–492.

28.

Salmon

(1987)Decentralization as an incentive scheme, Oxford Review of Economic Policy, 3, pp. 24–43.

29.

Schaltegger

C. A.

Torgler

Zemp

(2009) Central city exploitation by urban sprawl? Evidence from Swiss local communities. Working Paper No. 246, QUT School of Economics and Finance.

30.

Solé-Ollé

(2006) Expenditure spillovers and fiscal interactions: empirical evidence from local governments in Spain, Journal of Urban Economics, 59, pp. 32–53.

31.

Wilson

J. D.

(1999) Theories of tax competition, National Tax Journal, 52, pp. 269–304.