Master thesis, 15 hp Master’s Programme in Economics
Spring term 2020
Local government
fragmentation and impact on local government debt
A panel data analysis of Swedish municipalities
Anton Ögren
Abstract
This paper sets out to investigate the impact of government fragmentation on local government debt. The weak government hypothesis states that government fragmentation leads to higher budget deficits and higher public debt. The hypothesis is tested on a panel dataset comprised of 285 out of 290 Swedish municipalities over the period 2000-2017. The impact of government fragmentation is explained using government inaction theory and the common-pool problem. I find no evidence supporting that coalition governments generally accumulate higher public debt compared to single-party governments. However, there is evidence for the that ideological differences within coalition governments have a positive effect on the municipal debt level.
Keywords: Government fragmentation, public debt, local government, Weak Government Hypothesis, panel data
Table of content
1. Introduction ... 1
2. Institutional background ... 3
2.1 Political parties ... 6
3. Review of literature ... 7
3.1. Theory ... 7
3.1.1. Government inaction ... 8
3.1.2. Common pool problem ... 9
3.2. Empirical literature... 11
4. Empirical analysis ... 14
4.1. Data ... 14
4.1.1. Fragmented variables ... 15
4.1.2. Control variables ... 16
4.2. Econometric specification ... 19
5. Empirical results ... 21
6. Discussion ... 25
7. Conclusion ... 26
References ... 28
Appendix ... 30
1
1. Introduction
The degree of robustness in the public finances, institutional system and the political establishment varies across the developed economies. Where weakness in any of them may create an imbalance that results in sustained public deficits which in turn pushes up total debt per capita and thus the debt-to-GDP ratios over time. Following the 2008 economic crisis, many southern European countries saw their debt-to-GDP level soar. The soaring public expenditures arise from stimulants packages and at the same time, a fall in the tax-base. Similar to the Swedish situation following the 1990's economic crisis. Where the debt-to-GDP ratio reached the highest point in 1996 of 73% debt-to-GDP. Through political cooperation, the debt ratio has fallen down to 29%in debt-to-GDP by 2017, with fiscal policy performance playing an important role.
Previous empirical studies have investigated how differences in political, structural and institutional conditions affect national and local budgetary policy decisions and how it may affect public debt accumulation. Two early studies by Roubini and Sachs(1989a,b) sets a precedent for research regarding the subject of government fragmentation, they establish the hypothesis that more government fragmentation lead to higher budget deficits and debt accumulation. Following Roubini and Sachs studies, the Weak Government Hypothesis (WGH), becomes a subject of interest for other researchers, wanting to apply the hypothesis on numerous other local and national datasets. Some of those early studies set out to test Roubini and Sachs results, conducting studies based on the same data set but with a modified approach.
Subsequent studies find evidence that can be seen to either corroborate or refute the hypothesis that Roubini and Sachs puts forth. There is consensus in the economic literature that larger ideological variance within coalition governments may result in higher public budget deficits and larger accumulation of public debt. In the literature there is also prominent evidence for that left-wing governments run higher budget deficits than right-wing governments, mainly due to larger spending on social welfare programs (Pettersson-Lidbom, 2003). That ideological differences within government coalitions can affect the fiscal outcome, is a quite established concept in literature.
The objective of this study is to investigate how the WGH applies to the Swedish municipal circumstance, implementing the hypothesis on a dataset of Swedish municipalities and analysing the impact of government fragmentation on the public debt. The study contributes
2
possible insight on how well political parties keep budget deficits in check when in government coalitions, and what effect the coalition constellation has. In order to capture the effect of government fragmentation, three different fragmentation variable configurations are used. The first configuration is used to measure the effect of a coalition government against single-party majorities on the per capita public debt. The second configuration splits different government coalitions in groups, setting dummy variables for different coalition sizes in order to determine if the size of the coalition governments significantly affect municipal indebtedness. The third configuration estimates the fragmentation effect of mixed government coalitions.
An advantage of analysing local municipalities is that the data is more homogenous, eliminating possible problems that might occur when analysing national data, like differences in institutional, structural and political systems. Another advantage of analysing regional data is that the larger amount of available datapoint allows for a larger dataset, in this paper a dataset covering 290 Swedish municipalities is used. Compared to many studies based on national- level data, studies only include a range of between ten to twenty entities resulting in a much smaller dataset. Over the time period studied the average number of municipalities running budget deficits each year is around fifty. Public budget deficits might not be a problem, as long as the absolute debt level does not increase faster than the economic growth, since the size of debt is relative to the size of the economy.
The local government sector consists of both municipalities and regions, being an essential part of the Swedish welfare system, responsible for such welfare systems such as, healthcare, elderly care, low and middle schooling. The local government sector expenditures account for around 24% of the total Swedish-GDP. Given the size, the local government sector is an important employer, accounting for around a fourth of the entire Swedish employed workforce. As the local government sector plays an essential part in the economy, it is crucial to understand what drives local government spending and the accumulation of public debt. This study sets the focus on the political-economic drivers. The economic situation of the Swedish local government sector has in recent years been covered more frequently in both the news media and political debates, as many municipalities are experiencing increased economic difficulties. Difficulties that force municipalities to implement saving plans to cut back on public expenditure, to avoid future budget deficits and rising the public debt further. This done on the expanse of lowering the quality of the welfare services the municipalities provide.
The paper sets out to investigate the impact of fragmented local governments on the accumulation of debt in Swedish municipalities. Testing the WGH on a dataset comprised of
3
285 of the 290 Swedish municipalities under the period 2000-2017, with the general hypothesis that coalition governments accumulate more debt than single-party governments. Furthermore, to what degree the size of fragmentation affects the indebtedness. The remainder of the paper is organised as follows. The next section, section 2 presents the institutional background.
Section 3 sets up the theoretical framework, providing an overview of the theoretical and empirical literature. Section 4 introduces the empirical analysis, introducing the dataset and the econometric specification. The results are presented in section 5, the discussion in section 6 and section 7 concludes.
2. Institutional background
The Swedish municipal political design.
The Swedish local government system is characterised by self-government, a long tradition dating back more than 150 years, based on the idea that the local inhabitants knows what is best for them. Today the municipal self-government plays an important role in the Swedish society (Lars Nilsson & Hålan Forsell, 2013). The main motives for local self-government and the decentralization of power are democratic-values and efficiency. The self-government enables inhabitants to gain increased insight into how political decision-making works and thus becoming more involved in the democratic process. "The democratic rules of the game" of local self-government for municipalities are laid down in the Local Government Act, determining what degree of autonomy and obligations the 290 Swedish Municipalities have when governing local affairs. A democratically elected council leads the municipalities, which the electorate votes on a fixed four-year cycle. A parliament majority then forms a local “new” government following each election cycle. This local government are then responsible for a wide range of facilities and vital welfare services such as primary schooling, childcare, care for the elderly and other social services. Additional responsibilities apart from the welfare is to provide technical services like water and sewage, cleaning and waste management. The municipalities are also responsible to provide the rescue services, the local libraries and provide housing. What the municipalities are obligated to provide to its citizen are specified in The Swedish Local Government Act (SFS 2019:835).
The Municipalities are not mere local regions, the municipalities have a considerable degree of autonomy and independent powers of taxation. The Swedish Constitution protects the
4
municipalities power of self-governing and finance autonomy, however controlled by ample restrictions. The municipalities responsibilities also include the administration of the local budget, being endowed with tax setting autonomy to be able to cover the expenditures required to provide the municipal services. However, the taxation autonomy is restricted as the taxation system is characterised by a high degree of centralisation to the national level, the tax authority of municipalities is limited to income taxation.
The income tax-rates vary across and within different municipalities over time. In 2018 the lowest municipal income tax rate was 17,12% in Solna, while the highest municipal income tax-rate was recorded in Dorotea, 23,85%. Dorotea is a small rural municipality located in the north, which is a less populated area, and Solna a sizeable urban municipality located in Stockholm county. There is also a sizeable fiscal equalisation system harmonising revenue across municipalities. The system's primary purpose is to balance out structural differences so that municipalities has the more equal capabilities in providing their services. The equalisation system is an essential source of revenue, especially for small and rural municipalities. If revenues from the first two revenues sources fail to cover the municipal fiscal expenditures, the policymakers have one source left to them, which is to borrow.
There is an important aspect to consider in the regard to public budget deficits. In the Local Government Act it is specified that:
"Municipalities shall exercise good economic management in their activities and activities conducted by the agency of other legal entities" (SFS 2019:835)
What "good economic management" implies is not clearly defined in the act. The proposition (Prop.2003/04:105) to the act, have some clarifying requirements of what proper economic management implies. The first requirement states that there should exist an economic control system which evaluates the finances in the municipality, that also includes a financial goal that is to be evaluated in an annual report. Which is defined in chapter 11 of the local government Act (SFS 2018:597). The second requirement is that the economy is in balance, which implies that an economic deficit is to be balanced over the business cycle, implying that each generation carries its costs. Thirdly, municipalities are to set up plans for dealing with unexpected occurrences, that there is a scheme where necessary actions are outlined to correct for imbalances between revenues and expenditures. However, oversite responsibility lies wholly on the inhabitants as there is no threat of sanctions from the state for a municipality that fails to
5
keep the budget in check. The idea is that the local government is to be evaluated by the electorate in the election cycle. If the incumbent government fails in its obligation to uphold
"good economic management" among other equally important obligations, it is up to the electorate to vote the incumbent government out of office. This comes down to how well informed the electorate is and in what degree the voters' favour fiscal discipline. In a paper by Jochimsen and Nuscheler (2011), they found that the German voters favour governments being fiscally responsible, not running unwarranted public deficits. This might not be the case in other nations or in regional governments.
Table 1: Number of parties in Swedish municipal governments
1998-2001 2002-2005 2006-2009 2010-2013 2014-2017
1 party 49
17.9%
47 16.2%
34 11.7%
37 12.7%
23 7.9%
2 parties 74
25.6%
72 24.8%
46 15.8%
38 13.1%
46 15.9%
3 parties 54
18.6%
60 20.6%
47 16.2%
71 24.5%
91 31.4%
More than three 112
38.7%
111 38.2%
163 56.2%
144 49.7%
130 44.8%
Average number of parties
3.10 3.08 3.55 3.38 3.41
Total 289 290 290 290 290
Source: SKR
In table 1, the political fragmentation in Swedish municipalities is displayed, from one party government to coalition government with more than four parties. The average number of parties forming local governments in the period is just over 3. In the period studied the number of municipalities having single-party governments has been decreasing, from 17.9% in 2000 to only 7.9% of municipal governments being run by a single party majority after the 2014 election. The majority of Swedish municipalities have coalition governments with more than three parties, ranging from around 40% - 50% in the period studied. In the period 1994-1997, nearly 40% of the municipalities had single-party governments, and the average number of
6
parties forming local governments was 2.27. Over the years following, the average number of parties in local government has steadily been increasing, to 3.4 in 2014.
2.1 Political parties
In Sweden under the period studied 2000-2017 there are eight parties large enough to get into the Swedish parliament, The Social Democratic Party (S), The Moderate Party (M), Swedish Democrats (SD), Centre Party (C), Left Party (V), Christian Democrats (KD), Liberals (L) and The Green Party (MP). At the Municipal parliament level, other local political parties qualify for the local parliament, which are not to be expanded upon further in the scope of the study.
Every major party, except SD, has formed a coalition in the period studied. Only taking part in a coalition until after the period. However, throughout the period, the political power structure in many municipalities has changed, affecting the political structure of many local governments, from being controlled by either political wing, to being composed by a larger proportions of mixed coalition governments.
Table 2: Political rule in Swedish municipal governments
1998-2001 2002-2005 2006-2009 2010-2013 2014-2017 Left-wing 127
44%
126
43%
91
31%
110
38%
99
34%
Right-wing 105
36%
104
36%
157
54%
141
49%
89
31%
Mixed 57
20%
60
21%
42
14%
39
13%
102
35%
Source: SKR
The political control under the period studied is relatively stable with few large fluctuations.
However, some interesting changes can be seen throughout the election cycles. The number of local left-wing governments seem to decrease, being replaced by a larger proportion of right- wing government. Left-wing control at its highest at the beginning of the period, at 44% and at its lowest in 2006-2009 with 31%. Right-wing government control is at its highest at 54% of the Swedish municipalities, and lowest at the last period with 31%. In the last period 2014-2017 we see a large increase in the proportion of mixed local governments.
7
3. Review of literature
In this section, an overview of the theoretical and empirical literature is presented, starting with theories expanding on the theory of government fragmentation and the formations of public deficits and public debt accumulation, followed by a review of the previous empirical studies of governmental fragmentation and weak government hypothesis.
3.1 Theory
In the literature regarding to the subject of government fragmentation, some theories are more recurrent, which focus on the effect on fiscal policy. In a paper, Alesina and Perotti (1995) conducts an critically survey of the literature on political institutional determinants of the government budget. They focus their inquiry into why certain OECD countries accumulate higher public debts and not others. Exploring the effect on the formation, and drivers of public budget deficits and accumulation of debt. They find that common-pool problems are a driving factor for public deficits and public debt.
A related theory to that of the common-pool problem, is the 'war of attrition' theory which is also found to play a significant role, A theory presented by Alesina and Drazen (1991), where each involved party attempts to wait out the others expecting more substantial benefit than their counterparts. The theory especially plays a role in coalition governments. Where the risk of delays in fiscal adjustments increases, resulting in government inaction and promotion of the status quo, (A. Alesina & Perotti, 1995)
The focus of this thesis is on the government fragmentation and the WGH, the hypothesis states that: government fragmentation leads to higher public deficits and public debt. In the following subsections, theories affirming why the WGH may be correct are presented. Expanding on what characteristics of government fragmentation have on fiscal policy performance. The two main theories presented in the paper are the government inaction theory and the common-pool problem, with a shorter mention of the partisan theory.
The partisan theory is relevant to take into consideration when discussing political economics, as it suggests that ideological motives act as a primarily drive for fiscal policies. And that the
8
deficit will be higher for left-wing governments than their counterpart, as left-wing parties are more inclined towards expansionary policies (Jochimsen & Nuscheler, 2011). Therefore, it is expected that left-wing governments accumulate higher public debt compared to right-wing governments.
3.1.1 Government inaction
In a model presented by Alesian and Drazen (1991) it is hypothesized that the number of parties involved in a decision-making process will negatively affect the timing efficiency when reaching policy decisions. In the study Alesina and Drazen (1991) investigates why stabilizing actions are delayed within coalition governments, when a proposed legislation would mitigate a large budget deficit. In their model they present arguments for that the delay is due to a ‘war of attrition’, in which decision is delay due to a political gridlock between the government parties, which is not resolved until the cost of this delay become to big for one of the decision makers. The number of involved parties in the decision process is also assumed to have an exacerbating affect on the degree of government inaction. Implying that fragmentation causes a degree of inefficiency related to the decision process, this inaction is translated into the reluctancy to deviate from the status quo. In the event of a shock that disrupts the government budget, a status quo bias delays mitigating actions resulting in higher budget deficits and public debt compared to a single-party government (Roubini & Sachs, 1989a). Bulow and Klemperer (1999) sets up a model for war of attrition with K+N competing firms, they use the results to explain how long it takes to form winning coalitions in politics. However, they conclude that there is no relationship between the number of involved parties and government inaction, finding supporting evidence that player’s time of passivity is unaffected by the number of competitors and their action. The situations where this may occur are in contests in legislation processes or when trying to build a voting majority. Bulow and Klemperer illustrate an example of this with budgetary disputes in the US Senate, where the budget bill will not pass until 51 senators have more to gain from voting for the bill. The hypothesis is that fragmented governments have a higher tendency to be less efficient in their decision making than a single- party government. This would imply that policies for fiscal stabilization are delayed longer than otherwise necessary, and the public debt is allowed to accumulate further.
9
The veto-player theory is similar to the war of attrition, the characteristics of the theories reaches corresponding result, the affirmation of the status quo. In a study Tsebelis (2000) proposes a model in which each coalition partner is a veto-player. That is a player whose agreement is required for a policy decision. Tsebelis suggests that the veto-player would rather block political decisions that goes against their political interests, thus preferring the status quo.
He also hypothesises that a coalition government with a larger number of parties is likely to experience a higher degree of inaction, due to the increased number of interest groups present in the government. Similar studies find that coalition partners with substantial ideological differences are slower to take fiscal policy actions than coalitions comprised by parties with similar political views (Tsebelis & Chang, 2004). Finally, a study by Blais et al. (2010) affirms that the coalition governments that are more ideologically diverse are more likely to sustain the status quo bias. This by analysing public spending in 33 parliamentary democracies between 1970 and 2000. The veto player theory gives support to the government inaction theory, that coalition governments are more ineffective compared to single-party governments. Both theories support the hypothesis that fragmented governments are more ineffective in making policy decisions compared to non-fragmented governments. As governments become increasingly fragmented the decision-making process also becomes increasingly ineffective, especially so in case of mixed coalitions governments.
3.1.2 Common-pool problem
The common-pool problem highlights possible problems within the budgetary process within a coalition government with parties that have different areas of interest. As parties involved in the negotiation will condition their agreement on a policy or project that favours their constituents. The budgetary process can be understood as an allocation of a shared resource pool of municipal spending. It is typically assumed that coalition governments are less able to resist pressure for higher spending and are less able to internalise the expenditures fully. This results in overspending and accumulation of higher public debts for municipalities controlled by governments that have a higher degree of fragmentation (R. Weingast et al.,1981). The common-pool problem is also an increasing function of the number of parties in the government. Velasco (2000) demonstrates this by formulating a model in which groups can
10
extract resources from the common pool, finding that spending increases with the number of interest groups represented in the government.
There is a consensus in the literature that coalition governments have a propensity to spend more than single-party governments. As the number of parties involved in the fiscal process increase, the parties will take less responsibility for the future tax burdens (R.Weingast et al., 1981). Wehner (2010) finds in a study where he investigates the cabinets of 58 countries between 1975 and 1998, that there is a strong positive association between the numbers of spending ministers and budget deficits. Giving validation to the theory behind the common- pool problem, that the number of parties involved in the spending process will positively affect the public deficit and thereby the public debt. Jochimsen and Nuscheler (2011) connects the common pool problem to the strength of the finance minister within the local government. They argue that the reason that coalition governments accumulates significantly more debt than single-party governments are closely related to the strength of the finance minister. They arrive at the finding that coalition governments with a strong finance minister is not significantly different from single party governments. And thus, they conclude that the strength of the finance minister has a significant negative effect on borrowing. Hence, a strong finance minister is expected to mitigate the common-pool problem within a coalition government resulting in lower borrowing compared to a government with a weak finance minister.
The theoretical literature establishes a strong support for the WGH. As the theories highlight the problems caused by government fragmentation leads to increased budget deficit and accumulation of debt. The inefficiency and overspending of coalition governments occur from the increased number of interests present within the government. Resulting in numerous problems such as those discussed in the theories above, that may result in increased public debt.
For example, when fiscal policy changes are needed in case of an economic shock and the decision is delayed due to clashing political interests the public deficit problem might increase as a result. The increased areas of interest may also increase the public debt accumulation as each involved party want a “win” in the budgetary negotiations which adds a significant risk of overspending.
11
3.2 Empirical literature
This section reviews the previous empirical studies providing a framework and record of previous findings on the subject of government fragmentation. In the field of political economy, the subject of public government debt has been thoroughly studied, where various studies investigate the determinants of public deficit and public debt from different perspectives and models. One early study represented in the literature examining deficit and debt problems is by Barro (1979), who defines "The Equilibrium approach" to fiscal policy, also the so-called Tax Smoothing approach. Defining that the cause for public deficits are motivated by the concept of tax-smoothing, a countercyclical model akin to that of the Keynesian model method of fiscal deficit. This approach argues that the government minimises the distortion of taxation, setting a tax rate which "smooths" the deficits over time, keeping tax-rates fixed. In periods of economic downturn, the public deficit would increase to be paid off in later periods of economic expanse.
Studies examining the effect of government fragmentation on public finances is commonly based on National government data, with a large percentage of the studies focused on the industrialised nations, particularly OECD countries. Roubini and Sachs (1989a, b) studies are early contributions into the subject, analysing the fiscal behaviour of 13 OECD countries, introducing the WGH. Since their contribution numerous studies have studied government fragmentation and the WGH. Some subsequent empirical studies question the robustness of their finding (Edin & Ohlsson, 1991). Other author finds results that support their conclusions that fragmented governments have higher budget deficits and accumulate more public debt compared to single-party governments. After first presenting an overview of the national based studies, a presentation of locally based studies follows.
Roubini and Sachs (1989a, b) develops a model, where they find evidence supporting the WGH theory on data of 13 OCED countries under the period 1960 – 1985. They conclude that much of the budget deficits in the period could be explained by regular cyclical factors, though not entirely. They are observing that in the period 1975-1985 coalition governments showed a higher budget deficit compared to single-party majority governments. Edin and Ohlsson (1991) re-examine Roubini and Sachs findings, based on the same dataset and with a revised version of Roubini and Sachs model Edin and Ohlsson argue that the political cohesion variable used
12
by Roubini and Sachs only captures the effect of minority coalition governments. Thus, Edin and Ohlsson find no evidence supporting the hypothesis that the number of parties in the majority government has any effect on the government's ability to reduce the public deficit.
Haan and Sturm (1994) use a different dataset on 21 OECD countries spanning the period 1982 – 1989 where he finds evidence which are in contrast with that of Roubini and Sachs that there is no coalition effect, in line with results of Edin and Ohlsson. However, Haan and Sturm could not find significant evidence supporting the claim of Edin and Ohlsson that minority governments cause higher budget deficits than a single-party majority government.
In a paper De Hann et al., (1999) investigates whether coalition governments are more challenged to keep their budget inline after adverse shocks than single-party majority governments. The study is conducted on a broad sample of 21 OECD countries, for the period 1979-1995. Depending on what dependent variable are used, the results are varied. If they employ general government debt as the dependent variable, they find that the number of political parties participating in the central government is not related to the fiscal performance of countries in their sample. When they apply the growth rate of central government debt-to- GDP as the dependent variable, they find that the numbers of parties within government have a significantly positive effect. Also pointing out that the type of government (majority or minority) has no effect on the variation in fiscal policy. Hence, their finding gives ambiguous evidence supporting the WGH. Their result does not validate the finding of either Edin and Ohlsson and Roubini and Sachs regarding majority or minority-controlled governments. Perotti and Kontopoulos (2002) explores the role of government fragmentation on the fiscal performance of 19 OECD countries. Reporting findings that support the conclusions drawn by De Hann et al,. They further conclude that the size of the government cabinet is at least as significant a determinant of fiscal performance as the size of the government coalition. Lastly, they affirm the partisan theory, that the political ideology of the coalition government affects the accumulation of public debt.
Many of the early papers investigating government fragmentation and the WGH where based on OCED datasets. However, the studies based on local government data has become more ample over time. Example of such are; The study by Ashworth et al. (2005) who investigate the WGH using data on 296 Flemish municipalities. And in two more recent studies, Artés and Jurado (2018) and Goodman (2019) studies the effect of local government fragmentation on data of Spanish municipalities and US counties respectively.
13
Ashworth et al. (2005) utilize an error correction model in order to test the hypothesis that government fragmentation leads to higher public deficit and debt compared to single-party governments, on a large panel dataset of 296 Flemish municipalities. The error correction model enables them to separate the short- and long-term effect of government fragmentation, giving them the ability to analyse the effects separately, and identify general support for that the numbers of parties in a coalition have a positive effect on the short-term municipal indebtedness. However, they find no significant evidence for that government fragmentation affects the long-term public indebtedness. They conclude that fragmented governments are less effective in event of negative shocks to the municipal budget, resulting in higher public debt levels under fragmented governments
Artés and Jurado (2018) employ both a non-parametric and a parametric estimation models to analyse whether single majority governments run more balanced budgets than coalition governments. Using a panel of Spanish municipalities, they confirm their hypothesis that single party majority governments run more balanced budgets. However, they can neither confirm or refute Edin and Ohlsson (1991) findings that it is minority coalition governments and not majority coalition governments that run higher budget deficits compared to single party governments. In an analysis on US county data Goodman (2019) finds supporting results that increased government fragmentation leads to higher public expenditures per capita. Goodman also finds that the level of centralisation and size of the public sector are important determinants for the local government spending. Lastly, in an empirical analysis of Norwegian local governments, Borge (2005) uses a Herfindahl index as a measurement for political power in the local government council in a time series model. Capturing the party fragmentation of the local council and finding that the higher party fragmentation has a highly significant effect on the budget deficit.
The previous empirical studies into the effect of government fragmentation provide ambiguous evidence for what relevance the WGH has in regard to the performance of local coalition governments. On whether government fragmentation lead to higher public deficits and debt compared to single-party governments. The findings might depend on how the dependent variables are defined, and statistical models employed maybe foremost on the institutional, political and societal system of the study subjects.
14
4. Empirical analysis
In this section a closer examination of the dataset and the variables implemented in the analysis are presented. Inspiration for which variables to include in the model are taken from previous empirical studies that have been discussed above. The econometric model is estimated using panel data estimation models, given the dataset and type of analysis being conducted, a panel data model is the most suitable for the task. A few other models have been implemented in previous empirical studies, which are often based on smaller sample sizes and over a limited period. The econometric model will be presented in more detail further down. The dataset used does not suffer from any considerable data losses, after considering some missing data points and possible errors and excluding some municipalities from the dataset the panel data is balanced. More details on this in section 4.1.
The variables used in the analysis are categorized into two groups; fragmentation variables and control variables, where the control variables in turn are categorised into two sub-groups;
economic variables and political variables. The control variables included in the analysis are in line with what is seen in the literature. Which are implemented to capture factors that may affect the municipalities economic performance. The econometric model is first estimated with a classic OLS regression, then the model is developed to account for fixed municipal characteristics. Resulting in that the econometric model are estimated using a fixed effect estimation model.
4.1 Data
In the empirical analysis on how local government fragmentation impacts the total local public debt per capita uses a panel dataset that consists of data on 285 of the 290 Swedish municipalities over the period 2000-2017. Due to complicating circumstances in the data sources and to avoid possible data problems the period studied are restricted to no further back than to the year 2000. The complicating factors arise from the creation of new municipalities as well as the restructuring of some existing municipalities. The municipality of Gotland is also excluded from the dataset, due to the municipality of Gotland having extended obligation traditionally reserved to the regions, which goes beyond the normal commitments for a
15
municipality. These circumstances prompt the exclusion of five municipalities from the dataset.
The primary data sources used are Statistics Sweden and Kolada.
The dependent variable used in the empirical analysis is the total debt per capita of Swedish municipalities. The total debt per capita is defined as the long-term liabilities and the short-term liabilities added together, and then divided by the municipal population. Short-term liability are the financial obligations that are expected to be paid off within a year, and long-term liabilities are financial obligations that are due in more than a year. There is some variation on the choice of the dependent variable in previous studies. Example of such variable variations are;
government deficit (Edin & Ohlsson, 1991), government revenue (Alt & Lowry, 1994) and government spending/expenditures. To analyse what effect local government fragmentation has on the formation of local government public debt per capita, the analysis is conducted with different model configurations. Consisting of different fragmentation variable variations which are necessary to determine what impact government fragmentation has on the degree of municipal borrowing.
4.1.1 Fragmentation variables
The central interest of the analysis is to determine what impact political fragmentation has on the local government fiscal policies. The study attempts to accomplish this by an analyse of the effects of local government fragmentation with three different fragmentation configurations.
The first fragmentation dummy variable is if coalition government, COAL = 1, or if one party government = 0.
The second fragmentation variable is a set of dummy variables, to analyse what impact the size of the government fragmentation has on the per capita local public debt. Following Edin and Ohlsson’s (1991) dummy variable specification, the municipalities are split into groups, in which each group is assigned a dummy variable. The Edin and Ohlsson specification is an modification of Roubini and Sach’s (1989) political fragmentation indexation of the different type of government fragmentation sizes. The first dummy variable, SOC1, is equal to 1 where there are two coalition partners, 0 otherwise. The second, SOC2, is equal to 1 where there are three coalition partners, otherwise 0. The third, SOC3, is equal to 1 where there are four or more coalition partners, otherwise 0. Lastly, SOC4 is one if the coalition government are in minority.
The hypothesis is that, as the number of coalition partners increases, the expected coefficient
16
sign for all variables is to be significantly positive. Below a shorter overview of the dummy variables are presented:
SOC0: One party majority parliamentary government
SOC1: Majority coalition government with two coalition partners SOC2: Majority coalition government with three coalition partners SOC3: Majority coalition government with four or more coalition partners SOC4: Minority coalition government
Finally, a variable for ideological fragmentation variable, IDEO, is 1 when the parties in the government coalitions do not belong to the same political wing. Following political consensus and political history, S and V are members of the left-wing, while M, C, L and KD comprise the right-wing. On the national stage, MP is generally seen as a member of the left-wing.
However, on the local stage, MP is considered more neutral and may belong to either block.
SD is considered situated on the right in many political issues but belongs to neither the left nor right-wing. As there is no third political block comprising SD in the period studied, it is of no concern. The hypothesis is that fiscal policy decisions will take longer to be reached in case of a mixed coalition, the ideological differences of coalition partners associated with different political wings can be assumed to be wider. It is expected that mixed or rainbow government coalitions have a significantly positive impact on debt accumulation.
4.1.2 Control variables
In line with the literature, several economic control variables are included in the model, to control for factors that may affect the municipal economic conditions. The variability of these factors might result in different economic opportunities and strength of municipalities. The first variable included is that of the population of the municipality. The population size can vary significantly between municipalities and less within observed municipalities. The largest recorded population is in Stockholm and the smallest in Bjurholm. The hypothesis is that increased population size is likely to result in increased pressure on the local welfare system.
Resulting in increased demands for higher government spending which may cause higher levels of public borrowing (Ashworth et al., 2005). An potential counteracting effect to the increased
17
public spending is economies of scale due to an larger population size, having a counter negative effect on the municipal economy, the welfare cost per inhabitant decreasing as population size increase (Edmark & Ågren, 2008).
The change in taxable income per capita and the change in per capita government grants are included in the analysis to control for differences in the municipalities revenues. The taxable income per capita reflects two elements. The first is the increased demand for public goods, expected to have a positive effect if the public goods are considered normal. Secondly, the taxable income per capita reflects the municipal ability to finance expenditures. A positive change in the taxable income would decrease the need for loan financing. The interplay of the two elements might make the taxable income sign ambiguous (Ashworth et al., 2005). The per capita government grant is expected to have a positive effect on the financial situation of municipalities, as the government grants are a vital additional income source for many smaller municipalities. And thus, a negative effect on the total debt per capita. The fiscal equalisation system consists of several elements that determine the level of the municipal grants, which makes the system relatively complex. One such element is income-equalisation process, which is determined based on the municipal taxation power and financed by the state. Another is cost- equalisation system that are based on structural differences between municipalities in demographics, geography and socioeconomic situation, among other elements.
A large proportion of either younger or older inhabitants are expected to increase the demand on the welfare system. Hence, an increase in the proportion of either age group is expected to have a positive effect on the total public debt per capita. To control for the effect of demographic changes the proportion of inhabitants 0-19 and the proportion of inhabitants 65+ are included in the analysis.
To determine the effects of ideological differences two political control variables are incorporated, whether the coalition government is left-wing or right-wing. It is often maintained that left-wing governments are more inclined to keep more expansionary policies than right- wing governments, and thus are more willing to accept rising the public deficit, following the hypothesis of the partisan theory (Hagen & Vabo, 2005), (Haan & Sturm, 1994). Suggesting that if left-wing governments are in office, the public debt level will be higher than otherwise.
The empirical evidence for the hypothesis that left-wing are more inclined to run deficits are ambiguous. Jochimsen and Nuscheler (2011) cannot find any significant difference in deficits of left-wing governments and deficits of right-wing governments in German regions.
18
Table 3 provides a list in which the descriptive statistics of the included variables are presented.
Displaying that the per capita public debt is 23992 SEK on average over the period, with a standard deviation of 15108, min of 1483 and max of 105530 SEK. Examining some of the descriptive statistics of explanatory variables, the mean municipal population is 31573, and a standard deviation of 31464. The recorded minimum is 2421 and a maximum of 949761. The table shows that the average proportion of young and old averages to 23.4 % and 20.9%
respectively over the period. The standard deviations are relatively small, 2.5 and 4.2. With larger discrepancies between minimum and maximum, the proportion of young inhabitants has a min of 16.5% and a maximum of 31.4%. In comparison, the proportion of old inhabitants has a min of 8.2% and a max of 34.2%.
Table 3: Descriptive statistics
Variable Obs Mean Std Min Max
Total public debt 5130 23992 15108 1483 105530
Population 5130 31573 63704 2421 949761
Tax base 5130 148933 31464 81249 358341
Grants 5130 9666 5971 -16123 34227
Proportion young (0-19) 5130 23.4 2.5 16.5 31.4
Proportion Elderly (65+) 5130 20.9 4.2 8.2 34.2
Left-wing 5130 .375 .484 0 1
Right-wing 5130 .417 .493 0 1
Mixed 5130 .208 .406 0 1
Coalition government 5130 .869 .337 0 1
Index0 5130 .12 .32 0 1
Index1 5130 .193 .395 0 1
Index2 5130 .229 .420 0 1
Index3 5130 .448 .497 0 1
Index4 5130 .010 .100 0 1
Source: Statistics Sweden and KOLADA
19
4.2 Econometric specification
In this section, the econometric model used when estimating what effect the local government fragmentation has on the municipal public debt per capita is presented. The three different fragmentation configurations used, will be estimated separately with two different panel data estimation models, the Pooled Ordinary Least Squares model and the Fixed Effect Regression Model.
Which were deduced to be the most suitable model to apply after being tested against the random effect model with the Hausman test, which will be explained in further detail below.
All regression models are estimated using the robust standard errors option to account for any heteroscedasticity problems. In Table A1, the set of explanatory variables used in the models are presented, see appendix. The base econometric equation to test the government fragmentation hypothesis with Pooled OLS is:
Yit = β0 + βnxn,it + δkdk,it +uit (1)
Where Yit is the total public debt per capita for municipal i, at time t, β0 is the constant term, βn
is the coefficient for the estimated control variables, where Xn are the explanatory control variables used in the model. Likewise, δk is the coefficient for the fragmentation dummy variables, and dk are the fragmentation dummy variables. uit represents the error term. As stated above, Model (1) is estimated with the Pooled OLS regression model. The estimate of δ is the parameter of most interest since it shows what effect, if any government fragmentation as on municipal public debt per capita. Regarding analysing Panel data, the pooled OLS estimator may not be the most suitable. Other estimation models might be more suitable. There are two primary candidates, the Fixed effects regression model and the Random-effects regression model.
The second model is an expansion of the Base Model (1) presented above. Model (2) is estimated with the Fixed Effect Regression Model:
Yit = βnxn,it + δkdk,it + αi +uit (2)
20
This model specification is very similar to the one specified above: Yit, Xn and dk have the same specifications, the same is true for the coefficients. The two differences are that the constant term β0 has disappeared and that αi has been introduced. The αi are the entity-specific intercept capturing the unobserved individual effects, αi = β0 + βn+1. The Fixed effect model is suitable when analysing the impacts of a variable that vary over time and entities, as the relationship between the explanatory variable and the dependent variable within entities are explored, in this case, municipalities. When using the Fixed Effects model entity, individual characteristics may or may not affect the explanatory variables. Examples of such characteristics could be urban against rural areas, and other characteristics that might affect the size and activity in a municipality is if a Hospital or University is located in the municipality. The Fixed effect model accounts for such as these, as they do not vary over time. In order to determine which estimation model is the most suitable to implement a Breusch-Pagan and Hausman test were conducted. The Breusch-Pagan tests the Pooled OLS model against the Random effect model.
And the Hausman tests random effect model against the fixed effects model.
Starting with a more extensive explanation of the Breusch and Pagan's Lagrange multiplier (LM) test, which examines if the individual specific or time-specific variance components are zero, H0: σ𝑖2 = 0. The null hypothesis of the LM test is that the variance across individual- specific and time-specific errors are zero (Breusch & Pagan, 1980). That is that there is no significant difference across units; implying that there is no panel effect. By rejecting the null hypothesis, it is concluded that there is a significant individual-specific and time-specific error variance. Hence, there is a significant random effect in the panel data, and that the random effect model is more suited to deal with heterogeneity compared to the pooled OLS model (Greene 2008). The result of the Breusch – Pagan LM test applied on the dataset, is that the null hypothesis is rejected. And thus, conclude that there is significant evidence for a random effect in the panel data, and that the random effects model is more suited for dealing with heterogeneity than the Pooled OLS model.
A more thorough description is given into the Hausman test and the finding that the fixed effect model is more relevant when applied to the panel data. The Hausman specification test compares the fixed and random effects under the null hypothesis that individual effects are uncorrelated with any regressors in the model (Hausman, 1978). The issue is whether the conditional mean of αi can be regarded as independent or correlated to the xit. If the uncorrelation null hypothesis, E(µi|Xit,) = 0, fails, the fixed-effect model is consistent, and the random effect model is biased and inconsistent. The statistical difference from zero is evidence
21
against the random-effects model (Wooldridge, 2001). If the null hypothesis uncorrelated individual effects are rejected. Individual effects, ui, can be concluded to be significantly correlated with at least one regressors in the model and thus, the random effect model is problematic. Hence the fixed effect model is more relevant than the random effect counterpart.
Application of the Hausman test on the panel dataset used in this study gives the result that the null hypothesis is rejected. That is the conditional mean αi is correlated with xit; this way, its deduced that the Fixed effects model is the appropriate model to estimate the effects of government fragmentation.
5. Empirical results
The estimation results of model (1) is presented in Table A2, see appendix. Model (1) is estimated using the pooled OLS model. In the table, all three configurations are presented, and all the model configurations are estimated with robust standard errors, to correct for suspected heteroskedasticity in the error term.
The WGH states that fragmented governments accumulate more debt compared to single-party governments, and increasingly so as the size of the government fragmentation increases. Thus, the expected sign for the coefficient estimated is positive for all fragmentation variables, which the results of Model 1 does not reflect. The coefficient estimates of the first coefficient, COAL is insignificant and has the incorrect sign, the opposite of what is expected. A significant negative sign implies that coalition governments would accumulate less public debt per capita than compared to single-party governments. The coefficient estimate results for the coalition size dummy variables shows similar results, significantly negative estimates for two of the dummy variables. The exception is the coefficient estimate for coalition governments with two parties, showing a positive but insignificant effect. The only fragmentation coefficient estimation result, which gives a significant positive estimate in line with the expectations is for the ideological fragmentation coefficient. However, as concluded in the previous section 4.2, the Pooled OLS estimation method is not the most suitable; the coefficient estimates of the pooled OLS estimation are likely to be inconsistent and biased.
The fixed effects estimation model is implemented to control for the municipal specific effects.
In Table 4 the estimation results of the fixed effects estimation model are displayed, where all three fragmentation configurations are estimated with the robust standard error, controlling for
22
heteroskedasticity. In contrast to the coefficient estimation results presented in table 3, the COAL coefficient has an insignificant positive estimate showing the expected sign. Given this finding it cannot be generally stated that coalition government accumulate more public debt per capita than single-party governments. Thus, there is no evidence supporting that the WGH applies across the board on Swedish municipalities in the period 2000-2017. The result is consistent with the findings of both Edin and Ohlsson (1991) and Haan and Sturm (1994).
Neither finding conclusive evidence that coalition run regional governments accumulate higher public debt. However, the opposite of two newer studies based on Spanish and German regional data that finds that there is supporting evidence that the WGH is present in local government coalitions (Artés & Jurado, 2018; Jochimsen & Nuscheler, 2011).
Continuing by examining the dummy variables for coalition size for local governments to determine if the degree of fragmentation can be found to have an effect on regional government debt accumulation per capita. In the estimation results of coalition size, only the dummy variable for a coalition government of three parties (SOC2) has a coefficient that has a significant positive estimate. The coefficient estimate results for the other two size dummy variable representing coalition governments of two parties (SOC1) or coalition government of four or more (SOC3) are insignificant. The estimate result capturing the effect of minority government coalitions SOC4 are significant negative. Implying that local minority coalition governments has a negative effect on total per capita municipal debt. Which is an unexpected result, the hypothesis being that minority governments are expected to have a harder time reaching fiscal policy decisions than a single majority government due to a weaker position in negotiations. Previous studies have found that the effects of minority governments either are insignificant (Haan & Sturm, 1994), or that minority governments have larger budget deficits than single-party governments (Borge, 2005). The unexpected result regarding minority governments in this paper requires further investigation. Conclusive evidence supporting the theory that the size of the government fragmentation has positive effect on the per capita public debt levels in Swedish municipalities cannot be found.
The coefficient result for the ideological fragmentation variable estimate is significantly positive. Likely the only fragmentation variable able to corroborate the WGH. Specifically, that coalition governments with larger ideological differences among the policymaker enhances the risk of budget deficits and the accumulation of public debt per capita. This result fall in line with the reasoning of the government inaction theory resulting in a slower decision process and the affirmation of the status quo. Blais (2010) recognises that coalition governments with a
23
higher degree of ideological differences tend to run higher budget deficits, i.e., that they accumulate more public debt per capita than non-mixed coalition governments.
Below the estimation results for the control variables included in the model are presented.
Starting with estimation results of the two political control variables. Examining whether the political ideology of the coalition governments influences the local public debt per capita level.
In configuration one and two, the coefficient estimates of both left- and right-wing coalition governments are significantly negative. The coefficient estimate for right-wing governments shows a larger negative effect than the coefficient for left-wing governments. These findings support the hypothesis of the partisan theory that left-wing governments typically are inclined to have higher public spending than right-wing governments, resulting in higher public debt per capita for municipalities run by left-wing coalitions. These results are also in line with what is found in the political-economic literature (Jochimsen & Nuscheler, 2011).
Finishing with the economic control variables, starting off with the population size variable.
The result of the population size coefficient estimate can be seen in Table 4, showing that the estimate for population size has a significantly positive effect on the per capita municipal public debt. This result would indicate an increased demand for public expenditure within municipalities with higher population sizes. The coefficient estimate result of the second control variable, change in taxable income is shown to be significantly negative, i.e. having a negative effect on the per capita indebtedness. Implying a lower public deficit in the case of a positive change in taxable income for the local government. The coefficient estimate for per capita state grants is not significantly different from zero. This would imply that the equalisation system and the state grants are not sufficient to affect the local government indebtedness.
Both the proportion of young and old inhabitants in the population has a significantly positive effect on the municipal indebtedness per capita. The coefficient estimate for younger inhabitants is much larger than that for older inhabitants. A result that are expected given how average municipality costs are comprised, municipal expenditures for school and childcare is twice as large as the expenditures for elderly care.
24
Tabel 4: Estimation result of model (2), Fixed effects model estimation.
(1) (2) (3)
Dependent variable Total debt per capita Total debt per capita Total debt per capita Independent variable
Coalition government 244
(903.7) Coalitions size
A) Two parties (SOC1) -863.3487
(1165)
B) Three parties (SOC2) 2726.862
(1398)**
C) Four or more parties (SOC3) 162.4436
(1341)
D) Minority Government (SOC4) -4900.063
(1814)**
Ideological fragmentation (IDEO) 2385.732
(689)***
Left-wing - 2091.5
(818.0)**
-2173 (857.5)**
Omitted
Right-wing - 2559.9
(724.3)***
-2361 (754.7)***
Omitted
Population 0.511
(0.240)**
0.511 (0.235)**
.5095936 (0.24)**
Tax base -.4926447
(0.060)***
-0.471 (0.060)***
-.490633 (0.061)***
Grant 0.031
(0.077)
0.0313 (0.076)
.0301166 (0.077)
Proportion of Young 22044.17
(7308)**
22350.43 (7327)**
22052.97 (7294)**
Proportion of Old 12233.68
(5058)**
12224.35 (5091)**
12257.96 (5066)**
R2 0.101 0.114 0.106
Robust standard errors Yes Yes Yes
F-test 0.000 0.000 0.000
Fixed effects Yes Yes Yes
Observations 4845 4845 4845
Note: Standard errors in parenthesis, *** one per cent significance level, ** five per cent significance level, * ten per cent significance level.
25
6. Discussion
Some of the empirical estimation results are worth taking a closer look at. Failing to find significant evidence for the WGH, that coalition governments generally accumulate more debt than single party governments are not surprising given the finding of previous empirical studies.
Suggesting that coalition governments in general does not experience more problems in the fiscal policy decision process than single party governments. Opposing the theories presented in section 3. However, as many previous empirical studies find similar results, based on other datasets, such as in the study by Jochimsen & Nuscheler (2011) or that of De Haan et al. (1999).
The empirical results for the dummy variables for the coalition size is also inconclusive. The results do not give support to any of the theories that the budget deficit and public debt level increases with the number of parties involved in the government.
That there is no general presence of the WGH in Swedish regional governments might be due to the political system and traditions. That the Swedish society and electorate values and expects its representatives to keep more prudent fiscal policies. Such as the Jochimsen & Nuscheler (2011) states is true for the German electorate. However, the studies inability to capture any evident effect of government fragmentation might be due to the model or the chosen dependent variable. De Hann et al., (1999) demonstrates what effect different dependent variables might have on the estimation results. Illustrating that the WGH cannot be disregarded for the Swedish example due to this study. As there are many characteristics that the model does not capture or explore. The study does not account for power structure within the local coalition governments, which may be a significant determinant for the degree of fragmentation in the coalition government. Creating a Herfindahl index to capture the power fragmentation within municipalities might help the model to explain more of the impact of government fragmentation on the level of public debt.
Variable for minority governments does not yield the hypothesized coefficient estimation result. The significantly negative coefficient estimation might be explained by that the dataset of minority government is comprised of single party governments and coalition governments.
When examining the available data on how the minority governments are comprised, a large part of the minority governments is comprised of a single party, generally (S) a party which traditionally wins a large portion of the mandate in some regions. Which could be interpreted as a "stronger" minority government, having a stronger negotiation position than a minority
