Master’s Thesis Economics Version March 13th, 2011
Bieuwe Geertsema (1434276) University of Groningen
Supervisor: Dr. M.A. Allers
Political Budget Cycles in
Dutch local governments
Abstract
This thesis studies the occurence of Political Budget Cycles (PBC’s) at the municipality level in the Netherlands. A data panel with data of 431 municipalities in the period 2002-2010 is constructed. Two different regression models are used: the first one specifically chosen to deal with problems resulting from the short time dimension of the data panel, the second one to control for spatial effects. The results show some evidence of increased expenditures in election years and decreased expenditures in the pre-election year. However, these results become insignificant in the spatial lag model.
I
I
NNTTRROODDUUCCTTIIOONNWhen looking at the influence of politics on government finances, there are two well known phenomena that are believed to prevent politicians from setting steady and fully technocratic policies. The first well known effect is being caused by the fact that politicians have the desire to maintain or enlarge (individual or party) power when possible. This has led Nordhaus (1975) to publish the first theory on Political Budget Cycles (PBC’s). He has stated that politicians are opportunistic in the sense that they aim specifically at maintaining power. According to Nordhaus, incumbents are tempted to stimulate the economy in the period before elections, resulting in a temporal improvement of economic circumstances and a corresponding decline shortly after. This is what we call an opportunistic PBC.
Secondly, we have the so called partisan effect. Left-wing politicians have a different view on what would be the ideal policies than right-wing politicians. As a result, once elections have shifted the balance of power within a government, we can expect to see the effects of that shift of power resulting in changes in expenditures and taxations. As long as there is no long term stability in which political groups have a majority vote -and in most democracies this is the case- we will see what economics call a partisan PBC. These two phenomena have been studied extensively, not only at the level of national governments, but also for lower governments within various countries. In the Netherlands, there are three territorial levels of government. In addition to the central governement, the country is divided into 12 provinces and -on a local level- into 431 municipalities (as of 2010), with an average population of about 38,000. This thesis will extend the research on PBC’s by looking at the occurrence of PBCs at a local level in the Netherlands. It will make an attempt to answer the following research question:
1
1
.
.
L
L
IITTEERRAATTUURREERREEVVIIEEWWIn the literature about PBCs, two schools of thought about the cause for PBC behavior are distinguished. In this section I will give an overview of the main issues of both of the schools. Next I will give an overview of the more recent empirical findings. The section concludes with some theoretical background for the reseach question of this thesis.
Opportunistic Political Business and Budget Cycles
The first author to mention any politically related cycles was Nordhaus (1975) with his opportunistic behavior model. The key assumption of this model is that politicians have no policy preferences of their own, but are always aiming for re-election of their party and as a result choose policies that maximize their chances for that. No matter what their ideology is, they aim for low unemployment and high economic growth in the election year, and are willing to accept high inflation in and just after that period. Since voters heavily discount the past, the current year’s performance greatly influences their voting behavior. Later on, modifications of this model have been created by Rogoff and Sibert (1988) and Rogoff (1990), amongst others. In these articles, a rational component is introduced, building on the assumption that there is information asymmetry regarding the competence of incumbent politicians. As a consequence, these politicians signal their competence by choosing expansionary fiscal policy. Indeed, by then the focus had shifted from macroeconomic business cycles to fiscal budget cycles. This means that instead of studying the macroeconomic results of the government policy, economists started looking at the budget size and composition directly.
Partisan Theory
A second strand of literature focuses on partisan motivation of politicians. According to this train of thought, started by Hibbs (1977), the major causes of budget cycles are the differences in politicians’ preferences with regard to unemployment and inflation. Hibbs states ‘that the objective economic interests as well as the subjective preferences of lower income and occupational status groups are best served by a relatively low unemployment-high inflation macroeconomic configuration, whereas a comparatively high unemployment-low inflation configuration is compatible with the interests and preferences of upper income and occupational status group.’ As a result, left-wing politicians tend to invest in lower unemployment, allowing higer inflation figures, while right-wing politicians show opposite movements with regard to these two variables.
Empirical findings
When we look for more recent findings and focus on research studying aggregate spending in different countries, two articles are especially worth mentioning. One of the most influential articles in this light is the one written by Brender and Drazen (2005a). They have found that in their sample of 69 countries, any observable PBC behavior disappears when the relatively new democracies are removed from the sample. Hence, they conclude that PBCs only help politicians in their quest for re-election when voters are relatively inexperienced. In well-established democracies, an effect of punishment for fiscal manipulation kicks in and neutralizes any rewards. Alt and Lassen (2006) have taken another approach, and have found other potential sources of PBC behavior. They have studied a sample of 19 OECD countries, and found that PBCs mostly prevail when fiscal transparency is relatively low, or when political polarization is relatively high.
The shift from business cycles to budget cycles in empirical research lead economists to look for opportunistic and/or partisan behavior by studying variations in government budgets. This approach opened the door for empirical research focused on governments on a sub-national level. A number of articles in this category have been published since then.1
Most of these studies have focused on opportunistic PBC’s. Both in Canadian provinces (Blais and Nadeau, 1992) and in Israelian cities (Rosenberg, 1992) evidence has been found for increased spending in election years, mainly observable in the spending targets most visible to the voters. The same goes for cities in the United States (Bhattacharyya and Wassmer, 1995) and Western German Länder (Galli and Rossi, 2002). More recently, three articles have been published on PBCs on a local level in European countries. Considering Portuguese municipalities, classical symptoms like increased spending and decreased taxes in election years were observed by Veiga and Veiga (2007). Foucault and Francois (2006) have found opportunistic behavior when studying French municipalities, showing that capital spending increases in election years and decreases in the year thereafter. Finally, in Italy evidence has been found of the existence of opportunistic PBCs in current expenditures (Bartolini and Santolini, 2009).
This thesis will extend the previous research with a first-time study of the Dutch situation considering PBCs. I will study PBCs both in local taxation, as well in local government expenditures. Of these expenditures, certain categories in the budget as well as more detailed sub-categories will be looked at. As another extension of current literature, I will also take into account the influence of the ideology of the local government on PBC behavior. Thirdly, spatial interaction effects will be controlled for in a second model.
2
2
.
.
R
R
EESSEEAARRCCHHSSEETTUUPPThe standard method to study political budget cycles is to specify a dynamic panel data model. For the Netherlands, we have data available of more than 430 municipalities for the years 2002-2010 (a more detailed description of the data will be given in the next section). Since we need a 1-year lagged dependent variable in order to execute a dynamic test, we can only run regressions with data from 2003-2010. This shape of my data panel, where the number of studied instances is relatively large whereas the time dimension (T=8) is very small, is an important issue to take into account when setting up a research model. It is well known that using dummy variables to estimate individual effects in a dynamic model results in biased estimates in such a case.
Nickel (1981) was the first to publish about this bias. He has derived a formula for it, and has proven that the bias can safely be ignored when T approaches infinity. This has raised the question how to cope with the bias when using data sets for which this is not the case. Judson and Owen (1999) have compared various estimating methods that have been developed to do this. They mention Anderson and Hsiao (1981) who have made a first step in the research field with their first difference estimator. In the same article, Judson and Owen also study different variations of the Generalized Method of Moments (GMM). Especially the difference GMM and the system GMM (Arellano and Bond, 1991; Arellano and Bover, 1995; Blundell and Bond, 1998) are estimators that are widely used in panel data analysis. Thirdly, Judson and Owen look at an approach known as the Corrected Least Squares Dummy Variable (CLSDV) developed by Kiviet (1995, 1999). Kiviet has developed an approximation of the size of the bias and through that a method to correct the basic LSDV estimator. Judson and Owen (1999) have compared these three methods using a Monte Carlo approach, and have found that for panels with a time dimension of less than or equal to 10 observations, the CLSDV method is the preferred method.
In line with this conclusion, I will employ the CLSDV method, but in a somewhat adapted form. First, Bun and Kiviet (2003) have found that the approximation for the bias correction can be calculated with more simple formulae. Next, Bruno (2005) has adapted this method in order to be able to use it with unbalanced panel data. The method is based on a standard dynamic panel data model
y
it= γ y
i,t-1+ x
itβ + η
i+ ε
itwhere yit is the dependent variable, xit is the vector of (strictly exogenous) explanatory variables, ηi is an
unobserved individual effect and εit is an unobserved white noise disturbance. This LSDV estimation is
than corrected by a bias approximation which corrects any bias in the observed values of γ and β .
My first extension of this standard model is the addition of an election variable ELEit. It is common
municipal elections, which were held in March of 2002, 2006 and 2010. In addition, in November of every year (except 2007), a small number of intertemporal elections are held in those municipalities
amalgamating on the 1st of January of the next year. I will treat these elections in different ways to detect
PBC behaviour in spending, or in taxation.
In order to detect PBC behaviour for taxation, I do not alter the election year. Whether elections are held in March or in November, the rates published at the beginning of the year are relevant for the voter since these are in both cases the most recent ones. The election variables is 1 in the election year in both cases, and 0 in other years. For spending however, this is a different story. An election in March of year T can be influenced by past expenditures in the year T-1, and by announced expenditures in year T itself. I set the
election variable as 1 in year T (and 0 in other years), and the pre-election variable prELEit as 1 in year T-1
(and 0 in other years). However, if elections are held in November, the different timing demands a different treatment of the (pre-)election variables. In November of year T, the year has almost passed, and the past expenditures in year T are most likely to have influence on elections, instead of expenditures in year T-1. At the same time, since proposed municipality budgets are already published in October, the announced expenditures that are most relevant at the time of the elections are those of year T+1. Hence, I choose to treat the intertemporal elections in year T as regular elections in year T+1, shifting both the election and the pre-election variable for intertemporal elections back on year.
A second extension of the model is that I want to look at the effect of the ideology of the local government on the emergence of any PBC effects. In order to do this, in addition to introducing an
ideology variable IDEit that captures any partisan influence on the dependent variable, I will also introduce
an interaction variable ELEit×IDEit to measure the effect of ideology on the correlation of the election
variable ELEit with the dependent variable y. One final addition to the model is that I want to look at
specific time effects that are not yet captured by other control variables by introducing a time variable τt .
This leads to the following extended model:
y
it= γ y
i,t-1+ x
itβ + α
1ELE
it+ α
2IDE
it+ α
3ELE
it×IDE
it+ η
i+ τ
t+ ε
itSpatial econometric analysis: controlling for policy interaction
dependent variable like Allers and Elhorst do, I do want to control for these spatial effects. To do this, I
will conduct a spatial analysis of the dependable variables, in the form of a spatial lag model 2.
A problem that arises here is that the CLSDV method that I have selected before is only applicable when all the independent variables are strictly exogenous. This is not the case for a model with a spatial lag. Hence, I have to use another method, taking notion of the fact that any other method will -according to Judson and Owen (1999)- be sub-optimal given the short time dimension of my data panel. For the spatial panel analysis I use the Matlab modules written for this kind of model by Elhorst as described in his recently presented paper (2010). His modules are suitable for various forms of spatial analysis. For the spatial lag approach he defines the following model:
In the model, y stands for the dependent variable. The parameter ρ measures the spatial interaction, using
the spatial weight matrix wij. With xit a set of control variables is introduced, while ηi can be inserted to
capture any specific municipality effects, τt is also optional and does the same for any specific time effects,
and εit is an error term, independent and identically distributed with zero mean and variance σ2. To the
model above I will add a lagged dependent variable, as well as the election, ideology and interaction variables as defined in the previous section. Because I have chosen a fixed effects model for the first part of the research, this will also be done for this part. Accordingly, I will also set the module to calculate fixed time effects.
2 Spatial analysis exists in two forms. Spatial error models are appropriate when the observed spatial effect should result from
certain unobserved spatially concentrated charachteristics. A spatial lag model is chosen here because we look for direct spatial interaction between the dependent variables.
N
Σ
j=1
3
3
.
.
D
D
AATTAADDEESSCCRRIIPPTTIIOONNThis research is built on a data set provided by Statistics Netherlands, which contains detailed information about the municipality budgets for the years 2002 through 2010. In addition to total expenditures and revenues, the budget data is also split up into 10 categories and even into 120 subcategories. This creates options for a very thorough study of some very specific expenditures.
A point of caution should be made, however: the data in this set is provided by the municipalities themselves. Although efforts have been made by Statistics Netherlands to minimize freedom in categorizing certain expenditures and thus create uniform data from all municipalities, it can not be guaranteed that some expenditures are differently categorized among municipalities due to incorrect interpretation of the categorization system or even due to political motivations. This effect becomes larger the lower the level of aggregation of the budget. A second potential bias in the data lies in the possibility that municipalities are able to manipulate data by applying different accounting procedures which leads to different budget data for similar expenditures. Although these possibilities for potential deviations are known and most likely present in my data, we cannot know how they influence the data set and the outcomes of the analysis, nor do we know whether we can speak of any structural bias.
A second issues concerns the composition of the population. During the time period 2002-2010, a series of amalgamations has reduced the number of municipalities from 496 in 2002 to 430 in 2010. I choose to handle this by rebuilding the dataset in such a way that all the mergers are retroactively applied to the data; I act as if all the mergers have been implemented before 2002. Although this could result in some biased data (e.g. organization costs could go down after a merger) the only alternative - removing data from all the affected municipalities before or after a merger - would leave us with unnegligable gaps in the data. Dependent variables
To study PBCs as effectively as possible, one should consider which variables are most likely to be influenced by politicians for re-election purposes. Firstly, these variables must be influenceable by local politicians. Secondly, variations in the variables should be highly noticeable for the voters. For research aimed at national governments, it is common practice to study aggregate figures like total expenditures and the total budget surplus/deficit. This makes sense considering that national governments are sovereign and are able to influence these variables to a great extense. However, for municipalities in general and Dutch municipalities specifically, this is another story.
Next to these grants, roughly one third of revenues consists of municipal levies and income from property and market activities. It must be noted that income from property is largely market driven and income from market activities cannot be spent freely because it is largely offset by the costs of these activities. All in all, it is clear that municipalities have limited influence on their total revenues compared to national governments. Additionally, it is determined by law that local budgets should at all time be balanced, so the limited revenue autonomy has a direct effect on the total expenditures of municipalities. Hence, it would be uninformative to study total expenditures, total revenues or the budget balance.
On a local level, it makes more sense to study revenues or expenditures that are significantly steerable by the local government. On the revenue side, this leads us to the municipal levies which are divided into local taxes and user charges. User charges are not allowed to be higher than budgeted costs, but can be lower which leaves some room for PBC behavior. Local taxes consist of propery taxes for residential and
non-residential properties and can be set by local governments within certain nationally imposed rules3. I
will use a variable Housing charges as calculated by the Center for Research on Local Government Economics (COELO). They have add the average property taxes payed by a household and the average amount payed for two user charges for households (sewage charges and waste disposal charges). By taking this variable, any fluctuations that are caused by shifting income between these taxes and charges are eliminated and I get a look at the final tax changes that voters experience.
On the expenditure side, I make a selection of the 10 budget categories that are identified in the data by Statistics Netherlands. These categories are:
1. Administration
2. Public order and Safety 3. Traffic and Transport 4. Economic affairs 5. Education
Of these categories, an intuitive selection of the most noticeable and visible ones would be Traffic and Transport, Culture and Recreation and Welfare and Social services. The budgeted expenditures in these categories will be the next three dependent variables. Any revenues booked in the budget category are substracted from expenditures and as a result study net expenditures. Now, if I would only look at
3 In 2006 and 2007 upper limits for property tax rates were set by the national government. After that a macronorm was
introduced: if average property tax revenues growth was higher than this norm, the total volume of the municipality fund can be cut.
6. Culture and Recreation
7. Welfare and Social services
8. Public health and Environment
9. Spatial affairs and Housing
4. Sports 5. Arts
6. Employability measures
absolute values, variations in these dependent variables would be largely explained by control variables coïnciding with the size of the municipality, such as the number of inhabitants and the size of the municipality fund. Therefor, I will take per capita expenditures.
The next step is to delve deeper into the budgets by looking at specific subcategories within these main categories. Again I make an intuitive selection of expenditure subcategories that can easily be noticed by voters and choose the six following:
1. Roads and Streets investment and maintenance
2. Other Recreative Facilities (e.g. playgrounds, local media) 3. Social-cultural services (e.g. neighbourhood activities)
Each subcategory will be tested individually, where it should be noted that I study net expenditures, again per capita.
Data used in CLSDV model
Variable Obs. Mean Std. Dev. Min. Max.
Housing Charges 3448 612 105 268 1168
Net expenditures per capita on:
Traffic and Transport 3200 143 53.1 6.52 623
Culture and Recreation 3200 196 75.7 9.77 1076
Welfare and Social services 3200 244 93.5 10.4 824
Sports 3193 41.8 20.8 -47.0 256
Arts 3094 12.5 17.5 -103 186
Other recreative facilities 3029 10.8 20.4 -4.38 830
Roads and Streets 3200 120 46.4 -16.2 532
Social Cultural Work 3187 32.0 20.2 -18.6 238
Employability measures 3171 11.8 17.7 -132 353
Data used in Spatial lag model
Variable Obs. Mean Std. Dev. Min. Max.
Housing Charges 1768 609 104 268 1118
Net expenditures per capita on:
Traffic and Transport 1768 142 49.7 6.52 487
Culture and Recreation 1768 192 65.7 9.77 961
Welfare and Social services 1768 242 90.8 10.3 788
Sports 1768 40.9 20.1 -2.70 256
Arts 1693 11.7 17.2 -1.23 186
Other recreative facilities 1768 10.3 22.7 -4.38 830
Roads and Streets 1768 119 42.7 -16.2 468
Social Cultural Work 1765 30.8 17.7 -7.12 168
Employability measures 1760 11.9 15.4 -132 147
The most striking observation in the table above is that in the balanced dataset, used to estimate the spatial lag model, in some cases almost half of the observations are omitted due to gaps in the data. This does not lead, however, to very different mean values. The largest deviation is found in Arts, where the mean is 6.3% lower in the second dataset. On average the means in the second dataset are 2.2% lower.
Control variables
In my model description, I have introduced an as yet unspecified matrix of control variables xit which is to
control for any observable influences on the dependent variables. Firstly, I control for the demographic characteristics of the municipality by introducing both the population size and the population density. Expenditures can also be influenced by the composition of the population. To control for this, I will introduce the share of the population that is respectively under 20 and over 65 years old. Secondly we have to look at general wealth developments. For year T average income per capita in the year T-2 is
taken4. We will also look at unemployment levels. As an indication I will use a summation of the number
of unemployment benefits and welfare benefits per capita, measured halfway through the year.
Some of the studied expenditures could be classified as ‘luxury services’. This would imply that they greatly depend on the financial position of a municipality. As a measure of this, I will introduce the general (non-earmarked) central government grant per capita. This variable could also influence local taxes and charges, since a decrease of the general grant could push a municipality to raise these rates in order to maintain a certain level of services.
Finally, one additional control variable is used specifically when we look at local taxes and charges, in line with Allers and Elhorst (2005). They define the tax price as the proportion of the raised property taxes by owners of residential property (holiday homes excluded) and state that property tax on holiday homes and non-residential property is to a large extent paid by non-residents. They expect that when a smaller part of property tax revenues is paid by non-residents, there is an upward pressure on taxes and charges for households. All the data mentioned above is obtained from Statistics Netherlands, with the addition of local tax rates provided by COELO which were used to calculate the tax price variable.
Other variables
In addition to the control variables I need to explain my ideology variable and the spatial weight matrices. I will run different regressions using two ideology variables extending the PCCOUNCIL variable of Allers et al. (2001). The first one defines ideology of the council on a scale from 0 to 1, with a value of 0 meaning a council with only right wing parties and 1 meaning that the whole council is left wing. This is done by counting the number of seats for left wing parties, adding one half of the seats of parties of ‘neutral’ ideology (e.g. local parties) and dividing the sum by the total number of council seats. I count the national parties PvdA (Social Democrats), Groen Links (the Green Left), SP (Socialist Party), D66 (left wing liberals) and CU (Social Christians) as left wing parties, whereas VVD (Conservative Liberals), CDA (Christian Democrats) and SGP (Conservative Christians) are counted as right wing parties. Local parties
4Statistics Netherlands provides this information with a two year lag, which means that at the time of setting the budget for year
that have a clear right or left wing signature are treated accordingly. In Dutch municipalities, a coalition in the council appoints an executive board which in principle sets policy guided by a coalition program. For majority coalitions (which is the most common form) this means that all decisions in line with the coalition program are approved by the coalition council members, and for these decisions the executive board does not have to take into account preferences of any non-coalition parties. The composition of the executive board and its ideology can therefore also be determinative of the set policies. That is why I will also run the regressions using the ideology of the executive board. This is calculated by the method explained above, with the restriction that I disregard the seats of the non-coalition parties. Hence, we get the weighted average of the ideology of the coalition parties, and by that take into account the relative bargaining power of all coalition parties during coalition program negotiations. This variable does not take into account the number of seats of each party on the executive board, since I assume the coalition program to be leading for policies in all areas. All election data is obtained from COELO.
Finally, in my extended model I will use a set of spatial weight matrices based on the idea that municipalities sharing a common border or borderpoint (so-called queen contiguity) are used as areas of reference, and that municipalities watch their neigbors most closely. Since the dataset is rebuilt for every dependent variable, the corresponding spatial weight matrix is also rebuilt everytime to include the right municipalities. Information on municipality borders is obtained from Statistics Netherlands.
Data used in CLSDV model
Variable Obs. Mean Std. Dev. Min. Max.
Inhabitants (number) 3448 37968 59257 942 767457
Municipality Fund (euros per capita) 3200 756 208 38.0 2765
% up to 20 years (% of total population) 3448 0.25 0.026 0.18 0.43
% from 65 years (% of total population) 3448 0.15 0.028 0.065 0.27
Average Income (x1000 euros) 3442 13.4 1.84 8.3 24.9
Unemployment (% of total population) 3448 0.025 0.012 0.005 0.094
Tax Price (% of property tax revenues) 3448 0.65 0.12 0.22 0.95
Ideology (council) 3448 0.46 0.11 0.045 0.80
Ideology (exec. board) 3200 0.44 0.22 0.00 1.00
Data used in Spatial lag model
Variable Obs. Mean Std. Dev. Min. Max.
Inhabitants (number) 1768 33733 65592 3456 767457
Municipality Fund (euros per capita) 1768 750 184 38.0 1976
% up to 20 years (% of total population) 1768 0.25 0.026 0.18 0.43
% from 65 years (% of total population) 1768 0.15 0.025 0.065 0.23
Average Income (x1000 euros) 1768 13.2 1.66 8.3 23.3
Unemployment (% of total population) 1768 0.025 0.012 0.006 0.094
Tax Price (% of property tax revenues) 1768 0.66 0.11 0.30 0.93
Ideology (council) 1768 0.46 0.11 0.053 0.80
4
4
.
.
T
T
EESSTTRREESSUULLTTSS4.1 Estimates of the CLSDV model
I present the results of the regressions using the CLSDV method for unbalanced panels by Bruno (2005) in the five tables below. For all regressions I have included year dummies that control for nationwide temporal effects, like law changes or national budget cuts that affect the local playing field. Bun and Kiviet (2003) introduce three methods for approximation of the bias with LSDV estimations on data panels with a short time dimension, of which I have chosen to use the most accurate one. Standard errors are approximated by a bootstrap algorithm with five repetitions. The initial, biased estimations of γ, β and α are obtained employing the Arellano and Bond-estimator (1991). Since the CLSDV module provides no information on the goodness of fit of the model, I have rerun all the regressions as a regular LSDV test with fixed effects (including a lagged dependent variable), and provide the R2 of these estimations.
Although these values give no accurate measure of the goodness of fit of the CLSDV model, they do give a good indication of the relative goodness of fit of the various regressions.
Table 3 shows the results of the regressions taking Housing charges as the dependent variable in columns (1) to (6). The first thing to notice is that the results for the election effect are insignificant. Although the
Regression (1) (2) (3) (4) (5) (6) Lagged dependent (0.020) 0.81*** (0.019) 0.81*** (0.020) 0.81*** (0.027) 0.80*** (0.019) 0.81*** (0.027) 0.80*** Municipality Fund -0.000 (0.006) -0.000 (0.006) -0.001 (0.006) -0.005 (0.010) -0.000 (0.006) -0.005 (0.010) Inhabitants 0.000 (0.000) 0.000 (0.000) 0.000 (0.000) 0.000 (0.001) 0.000 (0.000) 0.000 (0.001) % up to 20 yrs (117) -137 (116) -138 (119) -148 -98.8 (162) (118) -146 (165) -101 % from 65 yrs (72.7) -136* (74.3) -136* (72.0) -120* (209) -104 (72.3) -129* (215) -113 Average Income 2.53*** (0.78) 2.56*** (0.87) 2.66*** (0.76) 2.45 (2.21) 2.58*** (0.75) 2.35 (2.28) Unemployment -406*** (153) -406*** (153) -393** (158) -365 (279) -380** (157) -350 (278) Tax Price 62.0(30.4) ** 61.7(30.7) ** (29.9) 63.3** 59.1(12.5) *** 62.2(30.0) ** 58.8(12.7) *** Election (8.91) 2.35 (10.0) 6.45 (4.48) 5.76 Ideology (Council) 17.6** (7.02) 20.0*** (6.93)
Ideology (Exec. Board) 7.14
(4.97)
8.77 (5.19)
Election × Ideology (5.67) -8.43 (5.55) -6.90
R2 indication 0.9022 0.9023 0.9022 0.8999 0.9024 0.9000
Table 3: Results for Housing. Bootstrapped standard errors are given between parentheses. Coeffecients marked with *
, **
or ***
coefficients have positive value, standard errors are high and the interaction effect is negative in all cases. This means that for local governments with an average ideology, the combined coefficient ends up near zero. Ideology variables consistently show positive correlations as expected, with council ideology having a significantly positive effect on the Housing charges level.
In table 4 I present the results of the regressions with expenditures on Traffic and Transport. In addition to the election dummy, I also run regressions with the pre-election dummy. Both dummies show a 1% significant effect when I run regressions without ideology and interaction variables. Expenditures on Traffic and Transport appear to be decreased in pre-election years, and increased in election years. These coefficients keep their expected sign when ideology effects are added but become insignificant in these regressions. The coefficients of ideology variables are ambiguous. The effect of council ideology is insignificant, but expenditures appear to be significantly lower when the executive board is relatively left-wing. This is somewhat peculiar since ideology of the council and the executive board are closely related. The coefficients of the interaction variable shows that both the positive election effect and the negative pre-election effect appear to be magnified when left-wing politicians are in charge.
Regression (1) (2) (3) (4) (5) (6) (7) (8) (9) Lagged dependent (0.027) 0.36*** (0.027) 0.37*** (0.027) 0.36*** (0.028) 0.36*** (0.017) 0.37*** (0.027) 0.37*** (0.028) 0.37*** (0.016) 0.37*** (0.017) 0.37*** Municipality Fund 0.10*** (0.018) 0.10*** (0.018) 0.10*** (0.018) 0.10*** (0.018) 0.10*** (0.015) 0.10*** (0.019) 0.10*** (0.018) 0.10*** (0.015) 0.10*** (0.014) Inhabitants 0.0015*** (0.0006) 0.0015*** (0.0006) 0.0015*** (0.0006) 0.0016*** (0.0005) 0.0014* (0.0008) 0.0015*** (0.0005) 0.0016*** (0.0005) 0.0014* (0.0008) 0.0014* (0.0008) % up to 20 yrs -216 (177) -221 (180) -216 (177) -219 (174) -329** (138) -221 (176) -236 (173) -321** (139) -338** (137) % from 65 yrs 358(106) *** 359(106) *** 358(106) *** 364(107) *** (247) 287 387(111) *** 346(107) *** (245) 309 (255) 277 Average Income 2.97(0.86) *** 3.08(0.81) *** 2.97(0.86) *** 3.01(0.79) *** (2.62) 3.92 3.40(0.73) *** 3.18(0.81) *** (2.65) 4.19 (2.58) 4.00 Unemployment (357) -321 (357) -323 (356) -341 (342) -318 (253) -349 (365) -354 (338) -348 (257) -374 (251) -360 Election 13.4*** (5.04) 3.44 (8.50) 9.65 (7.09) Pre-Election -12.3*** (2.95) -1.00 (6.68) -3.17 (3.17) Ideology (Council) 7.68 (18.8) 3.41 (16.4) 15.4 (20.3) Ideology (Exec. Board) -10.1** (4.17) -11.6*** (3.83) -6.96 (4.25) (Pre-)Election × Ideology 22.7 (16.4) -26.1*** (9.55) 9.66 (5.90) -10.2 (7.56) R2 indication 0.2612 0.2626 0.2638 0.2612 0.2723 0.2636 0.2650 0.2743 0.2738
Table 4: Results for Traffic and Transport. Bootstrapped standard errors are given between parentheses. Coeffecients marked with *
, **
or ***
Table 5 presents the results for Culture and Recreation. Again we see significant election and pre-election effects in columns (2) and (3), which are significant on a 5% and 10% level. These become partly insignificant and partly contradicting when ideology effects are introduced in columns (4) to (6), but this can be explained by looking at the interaction effects. The 1% significant negative coefficient for council ideology on the pre-election effect (column 7) explains the intitial positive value for the pre-election dummy. Other interaction variable coefficients again show some signs of increased PBC behavior by left-wing politicians in both election and pre-election years. The coefficients for ideology are consistently negative but insignificant.
In table 6 we see the results for Welfare and Social services. These results give some weaks signs of an election effect when we take into account the interaction variable and look at the combined effect. The pre-election effects are consistently negative and for the regression without ideology variables it is significant (1%). The coefficients of council ideology are significantly positive in all relevant regressions. Again, the interaction effects are positive in election years and negative in pre-election years, two of them significantly, showing signs of PBC behavior being more likely in municipalities governed by left-wing politicians. Regression (1) (2) (3) (4) (5) (6) (7) (8) (9) Lagged dependent 0.31*** (0.031) 0.31*** (0.031) 0.31*** (0.031) 0.31*** (0.031) 0.24*** (0.013) 0.31*** (0.030) 0.31*** (0.031) 0.24*** (0.013) 0.23*** (0.014) Municipality Fund 0.12*** (0.016) 0.12*** (0.016) 0.12*** (0.015) 0.13*** (0.016) 0.13*** (0.013) 0.12*** (0.016) 0.13*** (0.016) 0.13*** (0.012) 0.13*** (0.012) Inhabitants 0.0007 (0.0005) 0.0007 (0.0005) 0.0007 (0.0005) 0.0007 (0.0005) 0.0007 (0.0007) 0.0006 (0.0004) 0.0007 (0.0004) 0.0006 (0.0006) 0.0007 (0.0006) % up to 20 yrs -15.9 (146) -18.8 (148) -19.2 (146) -9.61 (143) -198* (118) -11.1 (144) -21.8 (143) -195 (119) -204* (118) % from 65 yrs 249*** (92.6) 250*** (94.0) 247*** (93.7) 235*** (93.7) 134 (198) 249*** (96.7) 221** (93.2) 147 (195) 129 (204) Average Income -2.98*** (0.81) -2.92*** (0.76) -2.90*** (0.83) -3.10*** (0.75) -2.48 (2.06) -2.85*** (0.68) -3.01*** (0.77) -2.30 (2.08) -2.38 (2.02) Unemployment -46.9 (303) -48.3 (302) -56.1 (302) -56.7 (290) -261 (222) -80.3 (309) -71.0 (286) -277 (227) -269 (220) Election 7.37* (4.19) 0.61 (7.03) 6.62 (5.96) Pre-Election -5.57(2.49) ** (5.68) 4.06 -5.00(2.58) * Ideology (Council) -15.6 (15.9) -18.4 (13.8) -9.03 (17.1) Ideology (Exec. Board) -2.51 (3.62) -3.37 (3.46) -1.01 (3.70) (Pre-)Election × Ideology 15.0 (13.9) -22.1*** (8.16) 5.71 (4.97) -4.53 (6.37) R2 indication 0.3939 0.3944 0.3945 0.3941 0.3958 0.3951 0.3955 0.3968 0.3968
Table 5: Results for Culture and Recreation. Bootstrapped standard errors are given between parentheses. Coeffecients marked with *
, **
or ***
In table 7, I present the results of the regressions for six budget subcategories. In order to keep the number of regressions limited, I have chosen only to run regressions including ideology and interaction effects here. Also, the fact that the results above have not given us a specific reason to keep testing with both ideology variables, has lead me to decide to only run regressions including the ideology of the council from this point onwards. We see that the results concerning election effects are ambiguous. For the subcategory Sports we see (pre-)election effects of the expected sign, but only the pre-election effect is 5% significant. Ideology and interaction variable effects are all insignificant. For Arts, none of the studied effects is significant and only ideology coefficients are of the expected sign. The regressions with Other Recreative Facilities give various significant results, which we should analyze simultaneously. Although the election effect is significant and negative, it is likely to be neutralized or turned around by the interaction effect. The same goes for the pre-election effect, but then in the opposite directions. We also see a strongly negative influence on expenditures of ideology (1% significant).
All results regarding expenditures on Roads and Streets are insignificant, as are those on Social Cultural Work. For Employability measures, the regression with election effects gives some 1% significant results for the election dummy and the interaction variable: in election years, right-wing councils tend to spend
Regression (1) (2) (3) (4) (5) (6) (7) (8) (9) Lagged dependent (0.030) 0.42*** (0.030) 0.42*** (0.030) 0.42*** (0.031) 0.41*** (0.013) 0.39*** (0.031) 0.41*** (0.031) 0.42*** (0.014) 0.39*** (0.015) 0.40*** Municipality Fund 0.28*** (0.012) 0.28*** (0.012) 0.27*** (0.012) 0.27*** (0.012) 0.29*** (0.011) 0.27*** (0.013) 0.27*** (0.013) 0.29*** (0.011) 0.29*** (0.010) Inhabitants (0.0004) 0.0002 (0.0004) 0.0002 (0.0004) 0.0002 (0.0004) 0.0003 (0.0005) 0.0002 (0.0004) 0.0002 (0.0004) 0.0003 (0.0005) 0.0002 (0.0005) 0.0002 % up to 20 yrs -499(115) *** -501(117) *** -502(115) *** -519(113) *** -544(103) *** -519(114) *** -526(112) *** -539(105) *** -549(104) *** % from 65 yrs 93.0 (65.7) 93.3 (66.7) 92.4 (66.4) 121* (66.4) 97.4 (171) 138** (68.5) 115** (66.4) 112 (168) 92.0 (175) Average Income -9.01(0.87) *** -8.98(0.82) *** -8.92(0.88) *** -8.75(0.82) *** -8.75(1.75) *** -8.51(0.72) *** -8.64(0.83) *** -8.55(1.76) *** -8.66(1.71) *** Unemployment (253) 85.3 (253) 84.8 (253) 75.3 (245) 113 (192) 231 (262) 82.7 (241) 101 (195) 204 (189) 225 Election (3.37) 3.82 (5.91) -3.94 (5.13) 0.19 Pre-Election -6.29*** (2.06) -0.75 (4.83) -2.29 (2.27) Ideology (Council) 32.9** (13.4) 29.1** (11.5) 36.8** (14.6) Ideology (Exec. Board) 1.79 (3.11) -0.18 (3.02) 4.23 (3.21) (Pre-)Election × Ideology 18.2 (11.9) -12.9* (6.85) 9.57** (4.33) -7.89 (5.57) R2 indication 0.8800 0.8800 0.8802 0.8805 0.8791 0.8807 0.8807 0.8794 0.8793
Table 6: Results for Welfare and Social services. Bootstrapped standard errors are given between parentheses. Coeffecients marked with *
, **
or ***
almost 10 euros per capita extra on Employability, but left-wing councils would do the opposite and decrease spending by almost 10 euros.
4.2 Estimates of the spatial lag method
In tables 8 to 12 I present the results of the regressions employing the spatial lag model. I have run all the regressions with fixed time and unit effects, and present the regular R2 values generated by the Matlab
routine of Elhorst (2010). These values are remarkably higer than the ones obtained for the regressions without the spatial lag effect. One might thus argue that the better model fit is a direct result from adding the spatial lag, but this would be incorrect considering that the coefficient for the spatial lag variable is not even significant in most of the regressions. Some other technical explanation has to be found for this in
future comparison of the models. For now, the R2 values are only comparable when looking at different
regressions using the same model.
Table 8 shows the results of the regressions with Housing charges. Although the absolute values and in some cases even the signs of control variable coefficients do differ from their counterparts in the CLSDV results, there are no contradicting findings in any of the significant variables. The spatial lag effect is
Subcategory Sports Arts Other Recreative
Facilities Roads and Streets
Social Cultural Work Employability Lagged dependent 0.58*** (0.010) 0.58*** (0.009) 0.55*** (0.017) 0.55*** (0.019) 0.15*** (0.022) 0.15*** (0.020) 0.36*** (0.026) 0.36*** (0.027) 0.54*** (0.019) 0.54*** (0.019) 0.49*** (0.015) 0.49*** (0.015) Municipality Fund 0.020*** (0.002) 0.020*** (0.003) 0.012*** (0.004) 0.013*** (0.004) 0.001 (0.011) 0.001 (0.011) 0.091*** (0.016) 0.092*** (0.015) 0.023*** (0.003) 0.023*** (0.003) 0.041*** (0.004) 0.040*** (0.005) Inhabitants 0.0002 (0.0001) 0.0002 (0.0002) -0.0001 (0.0001) -0.0001 (0.0001) -0.0000 (0.0003) -0.0000 (0.0003) 0.0008* (0.0004) 0.0008* (0.0004) -0.0001 0.0001 -0.0001 0.0002 -0.0006*** (0.0001) -0.0006*** (0.0001) % up to 20 yrs 32.7 (39.6) 33.9 (39.7) -21.3 (41.2) -20.7 (40.5) 197 (179) 190 (177) -372** (146) -376*** (144) -104** (45.7) -102** (45.0) -2.57 (69.7) 1.91 (70.1) % from 65 yrs (69.5) 49.9 (72.8) 49.9 (51.6) -82.9 -84.7(48.7) * (156) 109 (155) 91.9 (92.0) 210** (88.8) 182** (25.5) 31.0 (30.5) 34.9 (60.9) 90.4 (62.8) 107* Average Income (0.75) 0.39 (0.72) 0.38 (0.43) -0.68 -0.72(0.43) * -4.02(1.30) *** -4.12(1.29) *** 4.20(0.60) *** 3.93(0.68) *** (0.38) -0.08 (0.37) -0.05 (1.44) -0.59 (1.45) -0.39 Unemployment (90.8) -187** -187(93.9) ** (59.3) -95.4 (59.3) -89.3 501(95.6) *** (97.1) 501*** (303) -480 (280) -455 (108) -140 (101) -145 (156) 271* (154) 247 Election (2.09) 2.38 (1.86) -1.47 -2.86(1.71) * (7.01) -0.95 (3.21) 3.68 9.65(3.27) *** Pre-Election -3.68(1.80) ** (1.58) -1.16 9.28(2.51) *** (5.54) -1.33 (1.97) -2.67 (2.85) -0.60 Ideology (Council) 1.64 (7.41) 0.73 (7.37) 1.57 (3.85) 1.96 (3.69) -38.6*** (6.15) -31.7*** (6.71) 2.54 (13.5) 9.47 (16.8) -1.29 (8.56) -3.02 (8.11) 11.6 (13.8) 6.81 (14.8) (Pre-)Election × Ideology 1.30 (2.94) 3.72 (3.54) 3.88 (5.39) 1.61 (3.23) 7.07 (11.6) -21.2*** (5.53) 21.6 (13.6) -8.55 (8.01) -2.74 (6.40) 3.52 (2.59) -18.7*** (6.52) 3.23 (4.68) R2 indication 0.2948 0.2950 0.2711 0.2704 0.0319 0.0346 0.2424 0.2410 0.4068 0.4066 0.1671 0.1630
Table 7: Results for various subcategories. Bootstrapped standard errors are given between parentheses. Coeffecients marked with *
, **
or ***
significantly positive for the level of Housing charges. Election and ideology effects are consistent and of the expected sign but unlike the coefficients in the CLSDV model, these effects are never significant.
Regression (1) (2) (3) (4) Lagged dependent 0.65*** (0.018) 0.65*** (0.018) 0.65*** (0.018) 0.65*** (0.018) Municipality Fund (0.013) 0.002 (0.013) 0.003 (0.013) 0.002 (0.013) 0.002 Inhabitants -0.00 (0.00) -0.00 (0.00) -0.00 (0.00) -0.00 (0.00) % up to 20 yrs (192) 130 (192) 130 (193) 123 (193) 126 % from 65 yrs -85.5 (168) -88.0 (168) -84.9 (168) (168) -93.0 Average Income 4.10* (2.36) 4.09* (2.36) 4.14* (2.36) 4.00* (2.37) Unemployment -526(272) * -530(272) * -520(272) * -513(272) * Tax Price 94.5(28.6) *** 94.5(28.6) *** 95.2(28.6) *** 94.9(28.6) *** Spatial lag 0.039* (0.022) 0.040* (0.022) 0.043** (0.022) 0.043** (0.022) Election (8.17) -6.16 (9.62) -2.97 Ideology (Council) 7.39 (15.5) 8.34 (15.7) (Pre-)Election × Ideology (11.6) -6.94 R2 0.9538 0.9538 0.9538 0.9538
In table 9 the results of regressions for Traffic and Transport are presented. The spatial lag does not show any significant effect on the expenditures in this category. The election effects are positive for average ideology values, whereas the pre-election effects are consistent and even 1% significant when studied before introducing ideology. Results indicate that ideology possibly has a positive influence on expenditures, but these results are not significant. The interaction variables indicate magnified (pre-) election effects for left-wing councils compared to right-wing councils.
Next, table 10 shows us what the results regarding Culture and Recreation are. When looking at the outcomes regarding the control variables, we can see that both the income from the municipality fund and the percentage of people above 65 years old lose their significance, even though the spatial lag variable again shows no significant effect. We also see that both the election and the pre-election effects are now insignificant but still of the expected sign when combined with the interaction effect. Ideology also fails to show any significant effect. The interaction variable for election and ideology indicates that in pre-election years left-wing councils seem to decrease expenditures in this category significantly.
Table 8: Results for Housing charges with spatial lag effects. Standard errors are given between parentheses. Coeffecients marked with *
,
**
or ***
Table 11 presents the results of the regressions run with expenditures on Welfare and Social Services. The spatial lag variable is introduced with a highly significant positive value. When expenditures in neighbouring municipalities are on average one euro per capita higher, this has a positive effect of about seven cents on local expenditures. Election and pre-election effects are insignificant, but the effect of ideology on expenditures is significantly positive, as was also the case in the CLSDV model. The effect is even stronger in this model, with a difference between left- and right-wing councils of about 70 euros, compared to around 30 euros in the earlier regressions. According to the interaction variable coefficient, this difference could be increased by 23 euros in election years.
Finally, table 12 gives us the results of a series of regressions using the six expenditure subcategories. These results largely correspond with their CLSDV counterparts. Except for some changes in the nominal values of the coefficients, almost all significant election, ideology and interaction effects from the CLSDV regressions are found again in the spatial lag model as a significant effect, except for the election effect for Other Recreative Facilities. On the other hand, the interaction effect for Roads and Streets is now significantly positive. Regression (1) (2) (3) (4) (5) (6) Lagged dependent (0.026) 0.22*** (0.026) 0.22*** (0.026) 0.22*** (0.026) 0.22*** (0.026) 0.22*** (0.026) 0.22*** Municipality Fund 0.052(0.017) *** 0.052(0.017) *** 0.051(0.016) *** 0.051(0.017) *** 0.048(0.017) *** 0.050(0.017) *** Inhabitants 0.002(0.001) ** 0.002(0.001) ** (0.001) 0.002** 0.002(0.001) ** 0.002(0.001) ** 0.002(0.001) ** % up to 20 yrs -323 (242) -320 (242) -312 (241) -329 (242) -328 (242) -332 (242) % from 65 yrs 152 (208) 154 (208) 158 (207) 153 (208) 177 (208) 146 (207) Average Income 2.10 (2.90) 2.32 (2.90) 2.16 (2.89) 2.19 (2.90) 2.93 (2.92) 2.34 (2.90) Unemployment -55.8 (341) -52.6 (341) -65.4 (341) -52.2 (341) -96.8 (342) -72.2 (341) Spatial lag -0.00 (0.03) -0.00 (0.03) -0.00 (0.03) -0.00 (0.03) -0.00 (0.03) -0.00 (0.03) Election 8.84 (7.04) -3.44 (9.56) Pre-Election -14.6*** (5.52) -2.89 (8.50) Ideology (Council) 7.72 (19.5) 2.39 (19.7) 14.5 (19.8) (Pre-)Election × Ideology (14.7) 28.0* (14.3) -26.0 R2 0.6915 0.6917 0.6927 0.6915 0.6924 0.6933
Table 9: Results for Traffic and Transports with spatial lag effects. Standard errors are given between parentheses. Coeffecients marked with *
, **
or ***
Regression (1) (2) (3) (4) (5) (6) Lagged dependent 0.15*** (0.025) 0.15*** (0.025) 0.15*** (0.025) 0.15*** (0.025) 0.15*** (0.025) 0.15*** (0.025) Municipality Fund 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) Inhabitants (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 % up to 20 yrs -95.0 (229) (229) -94.1 -93.1 (229) -89.1 (230) -91.4 (230) (229) -107 % from 65 yrs (199) 210 (199) 211 (199) 211 (199) 209 (199) 224 (199) 189 Average Income -7.36*** (2.74) -7.27*** (2.75) -7.36*** (2.74) -7.45*** (2.75) -6.91*** (2.77) -7.24*** (2.75) Unemployment -50.6 (333) -49.7 (333) -53.3 (333) -55.0 (333) -95.6 (334) -76.7 (332) Spatial lag 0.01 (0.03) 0.01 (0.03) 0.01 (0.03) 0.01 (0.03) 0.01 (0.03) 0.01 (0.03) Election (6.66) 4.18 (9.06) -5.32 Pre-Election -3.40 (5.23) 12.7 (8.07) Ideology (Council) -6.97 (18.4) -11.5 (18.6) 2.02 (18.6) (Pre-)Election × Ideology 21.4 (13.9) -35.7*** (13.6) R2 0.8367 0.8367 0.8367 0.8367 0.8370 0.8368 Regression (1) (2) (3) (4) (5) (6) Lagged dependent 0.37*** (0.022) 0.37*** (0.022) 0.37*** (0.022) 0.37*** (0.022) 0.37*** (0.022) 0.37*** (0.022) Municipality Fund 0.001(0.000) *** 0.001(0.000) *** 0.001(0.000) *** 0.001(0.000) *** 0.001(0.000) *** 0.001(0.000) *** Inhabitants -0.002*** (0.000) -0.002*** (0.001) -0.002*** (0.001) -0.002*** (0.000) -0.002*** (0.000) -0.002*** (0.000) % up to 20 yrs -276 (200) -275 (200) -269 (199) -339* (199) -344* (199) -341 (199) % from 65 yrs 591(173) *** 591(173) *** 591(173) *** 609(172) *** 626(172) *** 602(172) *** Average Income -11.1*** (2.40) -11.0*** (2.41) -11.0*** (2.40) -10.3*** (2.40) -9.85*** (2.41) -10.2*** (2.39) Unemployment 603** (289) 604** (289) 599** (289) 646** (288) 600** (288) 634** (287) Spatial lag 0.067*** (0.023) 0.069*** (0.023) 0.074*** (0.023) 0.068*** (0.023) 0.064*** (0.023) 0.070*** (0.023) Election (5.79) 0.64 (7.82) -9.15 Pre-Election (4.54) -5.71 (6.96) 0.10 Ideology (Council) 72.5*** (15.9) 67.6*** (16.1) 75.7*** (16.1) (Pre-)Election × Ideology 23.3* (12.0) -12.9 (11.8) R2 0.9419 0.9419 0.9420 0.9426 0.9427 0.9421
Table 10: Results for Culture and Recreation with spatial lag effects. Standard errors are given between parentheses. Coeffecients marked with *
, **
or ***
are significant on a 10%, 5% or 1% level respectively.
Variable Sports Arts Other Recreative
Facilities Roads and Streets
Social Cultural Work Employability Lagged dependent 0.43*** (0.023) 0.43*** (0.022) 0.35*** (0.024) 0.35*** (0.024) -0.11*** (0.026) -0.11*** (0.026) 0.19*** (0.026) 0.19*** (0.026) 0.40*** (0.024) 0.40*** (0.024) 0.37*** (0.023) 0.37*** (0.023) Municipality Fund -0.00 (0.00) -0.00 (0.00) 0.00*** (0.00) 0.00*** (0.00) 0.00 (0.00) 0.00 (0.00) 0.00*** (0.00) 0.00*** (0.00) 0.00* (0.00) 0.00* (0.00) 0.00*** (0.00) 0.00*** (0.00) Inhabitants 0.00 (0.00) 0.00 (0.00) -0.00*** (0.00) -0.00*** (0.00) -0.00 (0.00) -0.00 (0.00) -0.00 (0.00) -0.00 (0.00) 0.00 (0.00) 0.00 (0.00) -0.00*** (0.00) -0.00*** (0.00) % up to 20 yrs (78.5) -26.5 (78.4) -22.1 (48.4) -21.5 (48.0) -40.9 (179) 403** 382(178) ** -496(202) ** -498(202) ** (62.2) -56.8 (62.3) -55.6 (107) 189* (107) 188* % from 65 yrs (67.9) 86.5 (67.8) 89.3 -158(40.9) *** -180(40.4) *** (150) 282* (150) 261* (174) 59.4 (174) 34.0 (53.8) 68.5 (53.8) 67.3 353(92.3) *** 371(92.5) *** Average Income (0.94) 0.95 (0.94) 0.82 -1.71(0.58) *** -1.79(0.57) *** -8.02(2.13) *** -7.96(2.11) *** (2.42) 3.79 (2.41) 3.23 (0.75) -0.88 (0.74) -1.01 (1.28) -1.46 (1.28) -1.15 Unemployment -251** (114) -252** (114) -134** (70.2) -157** (69.5) 713*** (258) 698*** (256) -367 (292) -318 (292) -83.2 (90.3) -78.4 (90.2) 654*** (155) 621*** (155) Spatial lag 0.007 (0.025) 0.007 (0.025) -0.030 (0.023) -0.031 (0.022) -0.016 (0.027) -0.016 (0.027) -0.008 (0.026) -0.007 (0.026) 0.008 (0.026) 0.009 (0.026) -0.023 (0.025) -0.025 (0.025) Election (3.10) 4.27 (1.94) -1.16 (6.64) -1.53 (7.93) -10.9 (2.45) 1.64 10.7(4.18) ** Pre-Election -6.72** (2.75) -2.26 (1.72) 14.6** (5.91) 3.78 (7.07) -1.55 (2.18) -3.21 (3.75) Ideology (Council) 5.35 (6.35) 4.43 (6.37) 5.16 (3.76) 5.69 (3.77) -64.6*** (14.4) -55.6*** (14.4) 2.72 (16.3) 11.6 (16.4) -4.75 (5.07) -4.63 (5.08) 11.7 (8.66) 4.54 (8.71) (Pre-)Election × Ideology 0.92 (4.75) 3.67 (4.64) 2.93 (2.92) 3.86 (2.86) 3.81 (10.5) -32.0*** (10.2) 28.6** (12.2) -10.7 (11.9) 2.78 (3.77) 1.46 (3.69) -20.8*** (6.50) 10.3 (6.36) R2 0.7955 0.7960 0.9191 0.9187 0.3230 0.3273 0.7191 0.7183 0.8501 0.8499 0.6340 0.6324
The general conclusion to be drawn from the extra regressions with the spatial model is that the spatial lag effect does not change the original outcomes very much. It does enter the regression results with a significant coefficient in some cases, but almost all election and ideology effects persist in the second set of regressions. However, the significance of most election and ideology effects are decreased in almost all cases. Whether this is a result of a change of the model in general, or of the introduction of the spatial effect specifically remains for further research.
Table 12: Results for the expenditure subcategories with spatial lag effects. Standard errors are given between parentheses. Coeffecients marked with *
, **
or
***
5
5
.
.
C
C
OONNCCLLUUSSIIOONNThe main question that this thesis was meant to answer, is whether there is evidence of the occurrence of PBCs on a municipal level in the Netherlands. The answer to this question however is not an easy one to give, even after analyzing the results of 89 regressions. In table 13 I have summarized the results of all regressions with Housing charges and expenditures on main budget categories per model in single symbol. Although I admit that this is a somewhat blunt method, it does give an overview of the main outcomes of these regressions.
Election Pre-election Ideology
(Council) Ideology (Exec. Board) Housing charges 0 0 x x + 0 0 (+) x Traffic and Transport + 0 (+) - - 0 (+) 0 - x Culture and Recreation + 0 (+) - 0 (-) 0 (-) 0 0 (-) x Welfare and Social Services 0 (+) 0 - 0 (-) + + 0 x
With the exception of expenditures on Welfare and Social Services, results regarding ideology ar rather weak and conflicting. For election effects, the results are stronger. Although there are quite some insignificant results, we see that the regressions on expenditure categories do show a pattern. As it appears, expenditures are increased in the election year, whereas in pre-election years we see a decrease in expenditures. This pattern could be a sign of the phenomenom mentioned in the introduction: politicians might want to hold back on expenditures in the pre-election year, to create fiscal room for expensionary fiscal policies in the election year itself. The voter is than expected to value the announced increased expenditures in the election year as being more important than the results of the decrease in expenditures in the pre-election year.
However, it must also be noted that with the introduction of the spatial lag the effects persist, but most of the coefficients become insignificant. And even the seemingly significant results should be interpreted with caution. As an illustration: by mistake I have executed half of the regressions using a data set of which some columns had been erroneously sorted, and even with this bogus dataset I managed to get some highly significant evidence of pre-election and ideology effects. It goes to show that even the significant outcomes of this research should not make anyone jump to conclusions.
Table 13: Schematic overview of main outcomes of the regressions with Housing charges and expenditures on budget categories. Results from CLSDV model are given in the first line, results of the spatial lag model in the second. Symbols: +: significantly positive;
Possibly the ambiguity of the results can be partly explained by the limited time dimension of the data panel. Although the CLSDV method is chosen specifically to compensate for this, the fact that my data only includes two election years and two pre-election years for the vast majority of the municipalities, calls for great caution while interpreting the results. A logical but also long-term improvement of this research is thus to extend the data set with future observations. Especially when we wait for another year of nationwide local elections, more thorough research can be done.
A second improvement for future research should be looked for in the choice of control variables. For this thesis I have used the same control variables for all expenditure (sub-)categories. This limits the amount of time needed for the research, but also the explanatory power of the control variable set. A third improvement for research in the field of PBC’s in general, is the addition of spatial effects as a control variable. With the exception of the working paper of Bartolini and Santolini (2009), none of the articles discussed in the literature review take spatial effects into account. It seems quite logical that spatial effects do play a role in budgetting policies, specifically for local governments since these are much more
prone to the so-called ‘foot voting’ mechanism5 than central governments are. The fact that the
significance of my results regarding election and ideology effects was decreased in almost all cases by introducing a spatial lag in the model, calls for spatial effects to no longer be ignored.
5A concept introduced in 1956 by Charles Tiebout, implying that people physically move to another jurisdiction where policies
L
L
IITTEERRAATTUURREEAllers, M.A., De Haan, J. and Sterks, C. (2001) Partisan influence on the local tax burden in the Netherlands, Public Choice, Vol. 106: 351-363.
Allers, M.A. and J.P. Elhorst (2005) Tax mimicking and yardstick competition among local governments in the Netherlands, International Tax and Public Finance, Vol. 12: 493-513.
Anderson, T.W. and Hsiao, C. (1981) Estimation of dynamic models with error components, Journal of the American Statistical Association Vol. 76: 589–606.
Alt, J.E. and Lassen, D.D. (2006) Transparency, Political Polarization, and Political Budget Cycles in OECD Countries, American Journal of Political Science, Vol. 50, No. 3: 530-550.
Arellano, M. and Bond, S. (1991) Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations, Review of Economic Studies, Vol. 58: 277–297.
Arellano, M. and O. Bover (1995) Another look at the instrumental variable estimation of error-components models, Journal of Econometrics, Vol. 68 (1): 29-51.
Bartolini, D. and Santolini, R. (2009) Fiscal Rules and the Opportunistic Behavior of the Incumbent Politician: Evidence from Italian Municipalities, Cesifo Working Paper, No. 2605
Bhattacharyya, D.K. and Wassmer, R.W. (1995) Fiscal dynamics of local elected officials, Public Choice, Vol. 83: 221-249
Blais, A. and Nadeau, R., (1992) The electoral budget cycle, Public Choice, Vol. 74: 389-403.
Blundell, R. and Bond, S. (1998). Initial conditions and moment restrictions in dynamic panel data models, Journal of Econometrics, Vol. 87: 115–143.
Brender, A. and Drazen, A. (2005a) Political budget cycles in new versus established democracies, Journal of Monetary Economics, Vol. 52: 1271-1295.
Bruno, G.S.F. (2005) Approximating the bias of the LSDV estimator for dynamic unbalanced panel data models, Economics Letters, Vol. 87: 361-366.
Bun, M.J.G. and Kiviet, J.F. (2003) On the diminishing returns of higher order terms in asymptotic expansions of bias, Economics Letters, Vol. 79: 145–152.
Franzese, R.J. (2000) Electoral and Partisan Manipulations of Public Debt in Developed Democracies, 1956-1990, in Strauch, R. and Hagen, J. von (eds) Institutions, Politics and Fiscal Policy, 61-83, Kluwer Academic Press, Dordrecht.
Foucault, M., and Francois, A. (2005) Local Political Business Cycles: Empirical analysis of French municipal budget, Preliminary version for 2006 European Public Choice Society Meeting.
Galli, E. and Rossi, S.P.S. (2002) Political budget cycles: The case of the Western German Länder, Public Choice, Vol. 110: 283-303.
Hibbs, D.A., Jr., (1977) Political Parties and Macreconomic Policy, American Political Science Review, Vol. 71, 146-87.
Judson, R.A. and Owen, A.L. (1999) Estimating dynamic panel data models: a guide for macroeconomists, Economics Letters, Vol. 65: 9-15
Kiviet, J.F. (1995) On bias, inconsistency and efficiency of various estimators in dynamic panel data models, Journal of Econometrics, Vol. 68: 53–78.
Kiviet, J.F. (1999) Expectation of expansions for estimators in a dynamic panel data model; some results for weakly exogenous regressors. In: Hsiao, C., Lahiri, K., Lee, L.-F., Pesaran, M.H. (Eds.), Analysis of Panel Data and Limited Dependent Variables, Cambridge University Press, Cambridge.
Kmenta, J. (1971) Elements of econometrics, Macmillan, New York.
Nickel, S. (1981) Biases in dynamic models with fixed effects, Econometrica, Vol. 49: 1417-1426. Nordhaus, W. (1975) The Political Business Cycle, Review of Economic Studies, Vol. 42: 169-90.
Rogoff, K. and Sibert, A. (1988) Elections and Macroeconomic Policy Cycles, Review of Economic Studies, Vol. 55: 1-16.
Rogoff, K. (1990) Equilibrium Political Budget Cycles, American Economic Review, Vol. 80: 21-36.
Rosenberg, J. (1992) Rationality and the political business cycle: The case of local government, Public Choice, Vol. 73: 71-81.
Literature reference Country, level Instances Period
(years) Dependent variable Model
Conclusions
Blais and Nadeau (1992)
Canada,
provinces 10
1951-1984 (34)
Budget Balance, Total
Expenditures, Expenditures on Social Services/ Roads/ Agriculture
Modified "cross-sectionally heteroskedastic and time-wise autoregressive model", Kmenta (1971)
Evidence of PBC's, focussed on Social Services and Roads
Rosenberg
(1992) Israel, cities 10
1964-1982
(19) Development Expenditures
Linear and logarithmic
regressions Evidence of PBC's Bhattacharyya and Wassmer (1995) United States, large cities 20 1959-1987 (28)
Total Expenditures, Total Revenues Seemingly Unrelated Regressions Evidence of influence of elections
Galli and Rossi (2002)
Germany,
Länder 11
1974-1994 (21)
Budget Balance, Total
Expenditures, Expenditures on Administration/ Healthcare/ Education/ Roads/ Social Services Cross-section weighted Generalized Least Squares Some evidence of opportunistic cycle
Veiga and Veiga (2006)
Portugal,
municipalities 278
1979-2001 (23)
Budget Balance, Taxes, Total Expenditures, Capital
Expenditures, Exp. on Roads
System-GMM Evidence of PBC's in all dependent variables Foucault and Francois (2008) France, large municipalities 91 1977-2002 (26)
