The local distribution of endowments matters: Modelling tax competition with heterogeneous local residents

Abstract

Chinese

This study expands on the tax competition literature by incorporating the heterogeneity of resource endowments into the tax competition framework. It theoretically elaborates that the local distribution of resource endowments affects both the level of tax rate and the degree of spatial dependence in tax competition, and empirically confirms the theory using the data for 60 urban municipalities in the Seoul metropolitan area (SMA), Korea in the years 2004–2006. A spatial panel model for tax cut confirms the presence of tax competition in the SMA and the effects on tax cut of the resource endowment distribution. Another regression model for local indicators of spatial dependence uncovers the fact that the spatial dependence in tax cut is also determined by the local endowment distribution.

Keywords

heterogeneous endowments spatial dependence spatial panel model tax cut tax competition

Introduction

Public finance scholars have long paid attention to the inter-local tax competition since cases of tax competition were frequently reported in Europe and the USA. Several scholars point out the strategic traits intrinsic to the tax competition between local governments in the urban or metropolitan areas of decentralised countries (Brueckner, 2000; Brueckner and Saavedra, 2001; Heyndels and Vuchelen, 1998). Applying Tiebout’s votes-by-foot to the capital movement across the local areas, their studies develop the inter-local reaction model where local governments make a strategic choice of tax cuts (or not) in the consideration of their neighbours’ behaviours as well as the internal conditions.

Since Tiebout’s (1956) seminal study, inter-local spatial correlation between local governments is not new in public policy studies. The literature in urban economics and political science backs up the spatial interdependence between subnational governments (Burbidge and Myers, 1994; Doreian, 1980; Hernandez-Murillo 2003; Heyndels and Vuchelen, 1998; Ladd, 1992).

While there are many theoretical studies devoted to tax competition, only few studies empirically verify it using relevant methodologies (Brueckner and Saavedra, 2001). The breakthrough research on budgetary spillovers is Case et al.’s (1993) study on budget spillovers and fiscal policy interdependence. They construct an empirical model based on a model of interstate benefits from spillovers that were introduced in the early theoretical literature. Using spatial econometric techniques, they estimate strategic interaction among state governments in the USA, where they regard expenditures as the strategic variable. Heyndels and Vuchelen (1998) also construct a spatial econometric model of tax mimicking to analyse the strategic choice of property tax rates in Belgian municipalities. Brueckner and Saavedra (2001) estimate a model of strategic property-tax competition using data for Boston metropolitan-area cities. These studies provide an empirical support for the theory that the spatial effect of tax competition or policy mimicking exists across local jurisdictions.

Another gap in the literature is that the literature on tax competition does not clearly consider the influence of local tax payers on tax competition although the tax, regardless of its type, is the transfer from the private sector to the government (Borck, 2003). Because the tax rate decision by the government determines directly tax payers’ net incomes or their property values, they are very sensitive to it and sometimes face up to a policy which is unfavourable to their assets.

This study aims to fill the gaps in the tax competition literature by investigating how tax payers’ utility maximising behaviours affect the property tax rate in empirical terms as well as in theoretical terms. We employ the property-tax competition framework to model how local governments determine, following Brueckner and Saavedra (2001). But the assumption of homogeneous individuals within a jurisdiction is relaxed so that the tax payers in a jurisdiction are heterogeneous in their endowment and income. We incorporate this heterogeneity’s influence into the property-tax competition framework to see how the presence of heterogeneous tax payers changes the government’s strategy in property-tax competition.

A theoretical model is tested using the data for municipalities in the Seoul Metropolitan Area (SMA),¹ Korea. In the mid-2000s, the local governments in the SMA competitively cut their property-tax rates right after a property tax hike driven by the central government’s reassessment programme. The empirical study verifies the results of the theoretical tax competition model with heterogeneous individuals.

Property tax competition with heterogeneous individuals

Heterogeneous residents in tax competition

Studies of tax competition commonly start with a few assumptions about the factors of production, such as mobile tax base, immobile population and a fixed total amount of capital (Bucovetsky, 1991, 1995; Mintz and Tulkens 1986; Wildasin, 1988a). Capital invested in a local jurisdiction is a local government’s main tax base, and capital moves from a jurisdiction to another in response to changes in economic conditions such as tax rates or social infrastructure (An et al., 2014). A typical strategic behaviour in tax competition is that one jurisdiction takes into account the others’ tax rates when it decides its own tax rate (Wilson, 1985).

This study accepts the typical assumptions of the tax competition literature: First, the individual residents seek to maximise the net return from their investment in capital. Consequently marginal returns on invested capital across local jurisdictions will be equalised. When the property tax is levied on capital, the after-tax net return will also be equal over local jurisdictions in a similar manner (Zodrow and Mieszkowski, 1986). Second, the metropolitan area contains many jurisdictions, and the population in each jurisdiction is a locally fixed labour resource. The population produces a private good and gets paid for labour. The private good is directly consumed by individuals and indirectly delivered by the local government in the form of public goods purchased with the taxes. Third, total capital in the metropolitan area is fixed. Then total fixed real estate capital in the metropolitan area is the budget constraint of the metropolitan area. Fourth, housing is produced using only land as in Brueckner and Saavedra (2001). As the total quantity of land in each jurisdiction is fixed, the supply of housing is also fixed. The property tax of the same rate as the capital tax is levied on housing with appropriate choice of units for land. Individuals are assumed to consume housing as much as they are initially endowed.² Given that housing consumption persists for a relatively long time, this assumption is not much unreasonable. These two assumptions on the housing market guarantees market clearance as well. In addition to this classical setting, we introduce the condition that individuals in one jurisdiction are heterogeneous, who are identified by their different endowments of production factors (Borck, 2003; Fuest and Huber, 2001). The typical tax competition model assumes that all individuals within a community and across communities have one unit of labour and capital endowment. We relax the assumption of homogeneity in the resource endowments among individuals. It makes the tax competition model more realistic, contributing to the tax competition literature. In fact, the condition of heterogeneous individuals is a key concept for analysing the political aspect of the tax competition because it paves a way for linking the majority voting model to the tax competition model (Buchanan, 1999). While those who do not own the residential real estate would likely vote for the redistributive policy with a high tax rate, those who own high-priced real estate would likely vote for a low property tax. The heterogeneous preference for tax rates may strongly influence the tax decision making through the pre-electoral competition.

As individuals can have more than one resource such as labour and capital, the modelling of the endowment distribution will become readily complicated. Following Borck (2003), however, one can assume that an individual’s capital endowment is negatively correlated with her labour endowment. An individual who has more labour endowment has less capital endowment, which is quite natural as poor residents should work harder to make up the lack of capital rent. Formally, an individual j in jurisdiction i provides $(1 + a_{i j})$ unit of labour, $(1 - a_{i j}) k^{*}$ unit of capital, and $(1 - a_{i j}) q_{i}^{*}$ unit of land, where the mean of $a_{i j}$ is 0 and so the average labour, capital and land endowment is (1, $k^{*}$ , $q_{i}^{*}$ ). This assumption will make it possible to handle the problem of heterogeneous endowments in the one dimension of tax preference.

Tax rate choice in the tax competition

Given the one-dimensionality of preference, the median voter theorem suggests that the local government supposedly chooses the tax rate to maximise the utility of the median voter to get the support from the majority of its electorates (Downs, 1957). Incorporating the distribution of endowments into the tax competition model, we can derive the reaction function in terms of tax rate choice in the tax competition game. With a simple set of utility and production functions in Tiebout tax competition model, Brueckner and Saavedra (2001) derive the reaction function in a linear form. Applying the median voter theorem to the tax competition model a la Brueckner and Saavedra’s, one can derive the tax rate choice that depends on the distribution of local residents’ endowments.

The tax rate choice for the tax competition game is derived into a linear form as $t_{i}$ = $A t_{l} + B$ , where $t_{i}$ and $t_{l}$ are tax rates for jurisdictions i and l, and A and B are parameters to be determined by local variables such as the local distribution of endowments and the preference for public goods. For the details of the process of derivation, see Appendix 1. Note that these results for tax competition between two jurisdictions can be extended to the competition among many jurisdictions, as the tax rate choice is a linear function with parameters constant for each jurisdiction. In the extended setting, jurisdiction i’s tax rate is expressed as a function of a linear combination, such as a weighted average, of its neighbours’ tax rates.

The linear reaction function shows that tax rates respond to each other. The slope A in the reaction function represents the responsiveness of a community’s tax rate choice to others, while the intercept B represents the level of tax rate chosen by a community regardless of the others’ choice. We can see that both the slope and the intercept depend on the local distribution of endowments and the preference for public goods. The endowment distribution can be represented by the skewness of the capital distribution: A positive value for it means that the proportion of residents with capital endowments smaller than the average exceeds a half, as illustrated in Figure 1. The higher the value, the larger is the proportion.

Figure 1.

Illustration of the distribution of capital endowments.

Consider first the intercept B in the linear reaction function. When the marginal utility is large enough, the intercept is increasing in the skewness of the local capital distribution. It implies that a community with a more positively skewed distribution of capital among individuals will more likely adopt a higher tax rate regardless of the other communities’ tax rates. This makes sense in this way: individuals with capital endowments smaller than the average are the majority of residents in that community, and under the pre-electoral competition condition, the political leaders of the local government will respond strategically to the majority’s preference (Buchanan, 1999; Downs, 1957). Individuals poorer than the average prefer higher tax rates as they get more public goods than they pay in the form of tax. Thus, in a community with the majority of residents poorer than the average, the local government will raise the tax rate.

Consider next the slope A. We can see that the slope, i.e. the responsiveness of a community’s tax rate choice to others is strictly decreasing in the skewness of the local distribution of capital endowments. It implies that in a community with positively skewed distribution of capital endowments and negatively skewed distribution of labour endowments, the local government is less sensitive to tax cut in its neighbouring communities. It makes sense because in a community where the distribution of capital endowments is more positively skewed and thus the proportion of poor individuals is larger, the local government will respond less concurrently to other communities, i.e. less sensitive to other jurisdiction’s decision making.

Hypotheses

Based on the above theoretical discussion on tax competition with heterogeneous endowments, we can set up two hypotheses on the relationship between the distribution of capital endowments and the local government’s choice of tax rates. As data for tax cut are available for an empirical study, tax rate choice is translated into tax cut choice in the hypotheses.

H 1: Tax cut is decreasing in the skewness of the local capital distribution, holding the other jurisdictions’ tax rates.

The theory in the previous section suggests that the tax rate is increasing in the skewness of the local capital distribution, holding constant the other jurisdictions’ tax rates. A positive value of the skewness means that the proportion of residents with capital endowments smaller than the average exceeds a half. As the poor prefer high tax rates, the level of tax cut should decrease with the proportion of the poor, and thus with the skewness of the local capital distribution.

H 2: The responsiveness of tax cut is decreasing in the skewness of the local capital distribution.

The theory also suggests that a community’s responsiveness to others in tax rate choice depends on the skewness of the local capital distribution. The responsiveness in tax rate choice, specially, in tax cut decreases with the skewness, and hence with the proportion of residents poorer than the average. Since poor residents prefer higher tax rates, a community where the majority of residents are poor will be less responsive to its neighbours’ tax cut.

The case of the SMA tax cut race

The Korean local government system has two tiers under the central government: provincial governments at the higher level and local (municipal) governments at the lower level. A local government is governed both by the elected official, mayor, and by the city council. When it comes to property tax, the central government and municipal governments have the right to decide the rates. The central government determines the property tax rate,³ and then local governments can flexibly adjust the rate within a given range, for instance, 50% of the rate, which is called Flexible Tax Rate.

In 2004 the Korean government introduced a package of tax hikes on real estate in order to make the tax levy more equitable. Citizens owning high priced real estate had long enjoyed relatively low tax rates under the old taxation system where the assessed value of real estate was appraised much lower than its fair market value – about 50% of the fair market value. The revised tax act, however, raised the assessed value of real estate almost up to the market value, which in turn increases the property tax burden as much.

In response to this central government policy, some local governments started to cut their property tax rates. Followed by higher income earners, homeowners in the median income group were major supporters of the property tax cuts. These residents opposed to the radical reassessment as they feared that the tax hike would reduce the value of their investments in residential real estate. Responding to such a strong pressure for property tax cut from their residents, several local jurisdictions of the Seoul metropolitan area (SMA) cut their property tax rates, which prompted the tax competition across the metropolitan area. Of 60 urban municipalities, 25 cut their tax rates by 10–30%. Figure 2 shows geographical distribution of the percentage of property tax cuts for Year 2004 in the SMA.

Figure 2.

Geographical distribution of tax cut for 2004.

The case of the SMA tax competition is appropriate for an empirical test of the tax competition model, elaborated in the previous section. Individuals’ preferences for property tax rates vary, depending on their heterogeneous endowments – the values of housing. Some municipalities have positively skewed distribution of their residents’ endowments, while others have negatively skewed distribution. As individuals’ preferences are translated into the local government’s decision making on tax rates, how the local endowments are distributed will determine the tax competition game. It is expected, therefore, that the analysis of the tax competition in the SMA will contribute to empirically examining the theoretical model of tax competition with heterogeneous local residents.

Empirical methods and data

We construct two different empirical models using the data for 60 SMA municipalities in the years 2004–2006 to test the impacts of the local endowment distribution as indicated by the theory in the previous section. The first model aims to test the impacts on tax rate and the presence of tax competition as indicated in the first hypothesis H 1. It is a spatial panel model that has the percentage of property tax cut⁴ (TCUT) as the dependent variable and its spatial lag term as one of the explanatory variables to capture tax competition. The second one aims to test the impact on the spatial dependence degree as indicated in the second hypothesis H 2. It is a classical panel model where the spatial dependence is measured by the local Moran’s I for tax cut.

The spatial panel models for tax cut

Consider first the spatial panel model for tax cut. As Figure 2 shows, there are certain spatial patterns in tax competition in the SMA. The empirical model tests the possibility that local governments’ property tax rate decisions are spatially dependent – in turn, whether there is strategic tax competition between local governments. To test the existence of strategic tax competition correctly, this study employs a spatial panel model in two different specifications.

The general model of tax competition implies that the property tax rate is determined strategically by a local government, considering changes in tax rates by other local governments. At the same time, the tax rate is determined by socioeconomic conditions specific to the local government. Then the equation for tax rate for a jurisdiction i at time t can be written in the first specification, a spatial autoregressive model (SAR) as follows:

\begin{matrix} y_{i t} = α + ρ \sum_{j = 1}^{n} w_{i j} y_{j t} + \sum_{k = 1}^{K} x_{i t k} β_{k} + μ_{i} + v_{i t} \\ μ_{i} ~ i . i . d . N (0, σ_{μ}^{2}) and v_{i t} ~ i . i . d . N (0, σ_{ν}^{2}) \end{matrix}

(1)

where ρ is the spatial lag parameter; w_ij is the weight for jurisdiction j; y is the dependent variable; x is the explanatory variable; α is the constant; β is the coefficient for x; μ is the individual-specific effect; v is the error term; and $σ_{μ}^{2}$ and $σ_{ν}^{2}$ are variance for μ and v. Note that we are using a random effects model, as our model introduces some time-invariant explanatory variables. It is generally known that a fixed effects model cannot be used when the model includes time-invariant explanatory variables as in this study.⁵

The second specification, a spatial error model (SEM) captures the possibility that the spatial dependence exists only in the error term, not in the spatial lag term. It is written as follows:

\begin{matrix} y_{i t} = α + \sum_{k = 1}^{K} x_{i t k} β_{k} + μ_{i} + ε_{i t} \\ ε_{i t} = λ \sum_{j = 1}^{n} w_{i j} ε_{j t} + v_{i t} \\ μ_{i} ~ i . i . d . N (0, σ_{μ}^{2}) and v_{i t} ~ i . i . d . N (0, σ_{ν}^{2}) \end{matrix}

(2)

where λ is the spatial error parameter; ε is the spatially dependent error and v is the normal error. Note that equation (2) includes the spatial lag term not for the dependent variable, but for the error term unlike in equation (1).

If the property tax cuts in the SMA have spatial correlation, then the weight will have a significant coefficient (ρ). This means the slope of reaction function is non-zero and tax rates in one local jurisdiction are strategically chosen in response to tax rate changes in other local jurisdictions.

The matrix of weights reflects adjacency patterns among municipalities. Two kinds of weighting schemes were used to capture different forms of capital mobility. The first matrix, denoted by W^c, reflects a simple contiguity and so it assigns a weight of ‘1’ to local jurisdictions that share borders with a given jurisdiction, and a weight of ‘0’ to others. As a variant of W^c, the second weight matrix W^c2 has a value 2 for the first order neighbours and 1 for the second order neighbours based on the contiguity definition of neighbours. Both weight matrices are then row-standardised for ease of interpretation as usual. We estimated the spatial panel models using a Stata module called XSMLE developed by Belotti et al. (2013).

Spatial dependence model

Consider next the regression model for the relationship between spatial dependence and the local endowment distribution. The second hypothesis H 2 indicates that the slope of the reaction function depends on the local endowment distribution as captured by the parameter $a_{i m}$ . In other words, the degree of dependence of a municipality on its neighbours in making a decision on tax rate is determined at least in part by the local distribution of its residents’ endowments.

As the local distribution varies over jurisdictions, the degree of spatial dependence also should vary. But the spatial models introduced in the previous section measure the spatial dependence averaged over all municipalities, as indicated by a constant coefficient ρ. In order to measure the varying spatial dependence, we need a more complicated model where the spatial lag term has a varying coefficient depending on the local endowment distribution. As it will be very challenging to estimate such a complicated model, we adopt an indirect method to approach the question.⁶

Instead of measuring the slope of the reaction function, we investigate how the local endowment distribution affects spatial dependence in tax cut. As the spatial dependence should vary over space, we need a local measure of spatial dependence rather than the global measure. Therefore, we introduce the local Moran’s I as the local measure of spatial dependence. For a variable x in location i, the local Moran’s I is defined by Anselin (1995) as follows:

I_{i} = \frac{x_{i} - \bar{x}}{S_{i}^{2}} \sum_{j = 1, i \neq j}^{N} w_{i j} (x_{j} - \bar{x})

where $S_{i}^{2} = \sum_{j = 1, j \neq i}^{N} (x_{j} - {\bar{x}}^{2}) / (N - 1)$ , $\bar{x}$ is the mean of x, N is the total number of regions, and w_ij is the (i, j) element of the weight matrix. With two different ways of weighting, i.e. W^c and W^c2, the indicator can have two different values for each region.

The local Moran’s I represents the degree and direction of spatial dependence for a specific locality, while the global Moran’s I does for the whole area. Its sign represents the direction of spatial dependence and its magnitude does the degree of spatial dependence. Spatial dependence in tax competition means that the choice of tax rate in a community is dependent on the decision in its neighbours. A positive value means that the decision goes in the same direction, while a negative value means the opposite. The higher the magnitude is, the higher the degree of the dependence is.

With this local measure of spatial dependence, we introduce a classical panel model that measures the impact on spatial dependence of the local distribution of residents’ endowments.⁷ The dependent variable (STDEV_I) that represents the degree of spatial dependence in tax cut is measured by the standard deviate of local Moran’s I, i.e. $S T D E V_I_{i} = [I_{i} - E (I_{i})] / \sqrt{V a r (I_{i})}$ ; see Anselin (1995) for details. Thus the model is specified as follows:

\begin{matrix} S T D E V_I_{i t} = δ + \sum_{k = 1}^{K} x_{i t k} γ_{k} + u_{i} + υ_{i t} \\ u_{i} ~ i . i . d . N (0, σ_{u}^{2}) and υ_{i t} ~ i . i . d . N (0, σ_{υ}^{2}) \end{matrix}

where x are the explanatory variables; δ and γ the regression parameters; u is the individual-specific effect; υ is the error term; and $σ_{u}^{2}$ and $σ_{υ}^{2}$ are variance for u and υ. As the model is a classical linear panel model, it is estimated by the generalised least squares (GLS) method.

Data and variables

We estimate the empirical models using the data for 60 urban municipalities in the SMA in the mid-2000s,⁸ excluding six rural municipalities where residents hardly respond to the residential real estate value because most of area is the farmland and pasture. While the first model uses the percentage of tax cut as the dependent variable, the second uses the local Moran’s I of tax cut. Even if the dependent variables are different in the two models, we introduce the same set of explanatory variables except the spatial lag terms in the spatial econometric models.⁹ It is because that equation (A-9) in Appendix 1 indicates that the intercept and slope are determined by a similar set of variables. The set of explanatory variables is listed in Table 1.

Table 1.

Variables and data sources.

Concept	Variable	Description	Data source
Dependent variable
Strategic Tax Policy	TCUT	Percentage of property tax cut (for example, if a municipality cut the tax 30%, the variable takes the value 30)	Public announcement by local governments
Explanatory variables
Distribution of Capital Endowment	SKEW	Skewness of estimated apartment values (Apartment value is measured by mean apartment price per m² of March multiplied by apartment size)	Statistics Korea (2005b)
Demand for Public Services	PROBUD	Proportion of policy project budget to gross budget in the general account	MOGAHA (2004a, 2005a, 2006a)
Demand for Social Welfare	NUM_NURS	Number of nursery facilities per 1000 children aged four years or less	MOHW (2004, 2005, 2006)
Fiscal Independence	DEGIND	Percentage of own source revenue (local tax receipt plus non-tax receipt) in the gross budget by general account	MOGAHA (2004a, 2005a, 2006a)
Tax Capacity	LN LAG_LT	Natural logarithm per capita local tax in the previous year	MOGAHA (2003, 2004a, 2005a, 2006a)
Political Support for Conservative	CONS_PARTY	1 if the mayor’s political party is conservative in the 2002 Local Election; 0 otherwise	NECK (2002)
Tax Base	LN APT_PRICE	Natural logarithm of the average apartment price per m²	KREIA (2004, 2005, 2006)
Labour Resource	NUM_WORKERS	Proportion of workers in major industrial sectors to the registered population	Statistics Korea (2004, 2005a, 2006)
Population	LN POP	Natural logarithm of registered population in a jurisdiction	MOGAHA (2004b, 2005b, 2006b)

Data for the dependent variables were collected directly from the public announcement of the property tax cut that each local government made. The explanatory variables, all local level data, include the home (apartment) sales price skewness (SKEW), the ratio of project budget to total budget (PROBUD), per capita local tax revenue lagged one year (LN LAG_LT), the ratio of own source revenue to total budget, the index of fiscal independence (DEGIND), the number of nursery facilities per 1000 children (NUM_NURS), the number of workers in major industrial sectors per 1000 people (NUM_WORKER), the apartment price per square metre (LN APT_PRICE), dummy variable for the mayor’s political party as of the 2002 local election (CONS_PARTY). Tables 2 and 3 show the descriptive statistics and bivariate correlations for these variables, respectively.

Table 2.

Descriptive statistics.

Variable		Mean	Std. dev.	Min	Max	Observations
TCUT	overall	14.56	17.04	0	50	N = 180 n = 60 T = 3
	between		13.48	0	43.33
	within		10.51	−18.77	47.89
SKEW	overall	1.023	0.406	−0.281	2.121
	between		0.409	−0.281	2.121
	within		5.18*10^-17	1.023	1.023
PROBUD	overall	55.80	10.78	35.2	79.8
	between		4.910	45.57	67.63
	within		9.613	33.90	78.44
NUM_NURS	overall	11.55	2.529	6.96	18.9
	between		2.290	7.923	17.79
	within		1.099	8.536	14.32
DEGIND	overall	48.37	15.82	20.1	92.7
	between		15.56	22.07	91.47
	within		3.305	36.24	56.94
LN LAG_LT	overall	11.90	0.851	10.42	13.66
	between		0.850	10.56	13.50
	within		0.097	11.67	12.10
CONS_PARTY	overall	0.867	0.341	0	1
	between		0.343	0	1
	within		0	0.867	0.867
LN APT_PRICE	overall	5.394	0.505	4.340	6.781
	between		0.506	4.381	6.645
	within		0.059	5.293	5.577
NUM_WORKERS	overall	376.9	366.3	145.1	2762.03
	between		368.1	152.8	2727.04
	within		14.82	313.4	425.88
LN POP	overall	12.68	0.633	11.01	13.88
	between		0.636	11.053	13.86
	within		0.024	12.60	12.78

Note: The overall standard deviation is decomposed into the between and within components. The overall refers to the variable $x_{it}$ , the between refers to ${\bar{x}}_{i}$ , and the within refes to $x_{it} - {\bar{x}}_{i} + \bar{\bar{x}}$ , where ${\bar{x}}_{i}$ and $\bar{\bar{x}}$ are the group mean and the global mean, respectively.

Table 3.

Correlation of variables.

	TCUT	SKEW	PROBUD	NUM_ NURS	DEGIND	LN LAG_ LT	CONS_ PARTY	LN APT_ PRICE	NUM_ WORKERS	LN POP
TCUT	1
SKEW	−0.255	1
PROBUD	−0.125	0.076	1
NUM_NURS	−0.002	0.185	0.079	1
DEGIND	0.449	−0.124	−0.036	−0.326	1
LN LAG_LT	0.280	0.073	0.154	0.058	0.541	1
CONS_PARTY	−0.078	−0.060	0.076	0.248	−0.098	−0.156	1
LN APT_PRICE	0.461	−0.357	−0.182	−0.299	0.541	−0.042	0.004	1
NUM_WORKERS	0.136	−0.046	−0.081	−0.187	0.543	0.426	−0.215	0.296	1
LN POP	0.243	−0.197	−0.020	−0.087	0.270	−0.293	0.075	0.200	−0.284	1

The key explanatory variable is the home value skewness. It is used to capture the local distribution of capital endowments, which determines the collective preference for the tax rate, say, the median voter’s preference. The home value skewness is calculated using the data for the size distribution of apartments, which are of the most dominant housing type in the SMA.¹⁰ As the tax base for apartments used to be calculated by multiplying the standard apartment price by the size, the home value skewness is equivalent to the size skewness. But the size distribution data is provided only every five years and thus we inevitably use the data for 2005 since it is closest to the period of study. Hence this variable is time-invariant over the period of years 2004–2006 in our modelling.

Another important explanatory variable is DEGIND, the indicator of a local government’s fiscal independence, which is defined by the ratio of own-source revenues to total budget.¹¹ Since a local government’s high dependence on external resources sometimes undermines its fiscal stability, the own-resource revenues such as local taxes are one of the most important factors that determine its local autonomy. It is true particularly in the SMA tax competition, as the central government tried to cut the grant to prevent property tax cut dominoes by local governments. Threfore, local governments with weak fiscal health are less fiscally independent of the central government and will rarely, if ever, cut the tax down (Choe, 2005; Man, 1999).

In addition, local tax revenue per capita (LN LAG_LT) is complementally used in the empirical model to capture the effect of the tax capacity of a local government. The variable is measured for the previous year to avoid the endogeneity problem. Local governments with higher tax capacity can afford to cut the property tax to a greater degree and better make up the fiscal deficit by the tax cut and meet their fiscal needs.

The ratio of project budget to the total budget (PROBUD) is included to take into account the influence of local demand for public services on local government’s tax choice. The project budget represents the comprehensive plan for fundamental public services such as local infrastructure – services such as the road pavement and public transportation management, water supply and drainage system construction. It also includes the budget for basic local social services such as construction of nursing homes for the elderly, public education, health, and so on. As this variable represents needs for public services local government should meet, a local government with a higher value would less likely cut the property tax or cut it to a small extent.

Also, the number of nursery facilities per 1000 children (NUM_NURS) is included as a proxy for the demand for social welfare services because the project budget does not include the expense for public child care services. There is a high demand for nursery services in metropolitan areas because a lot of families with children need the organised child care services and they have no other informal options such as care by relatives (Hofferth et al., 1991; Seo et al., 2005; Swensen, 2008). As a small value for the variable implies a high demand for public child care, the local government faces high pressure of building public nursery facilities and thus will hold their property tax rate high.

In addition, the model includes a political variable of whether the mayor’s political party was conservative¹² in the the 2002 Local Election (CONS_PARTY) to control for the party competition in the tax hike policy. The need for compromise between conflicting parties creates a potential roadblock in policy adoption (Boyne, 1994; Dubois et al., 2007; Hansen, 1983). It is, therefore, harder for a split government to introduce a new policy than a unified government. A tax with the goal of redistribution is frequently controversial in the political process because it often gives a cause for civil strife between social classes (Lowi, 1970). The property tax hike in 2004 had the policy goal of wealth redistribution, and was initiated by the then progressive ruling party in Korea. It is expected, therefore, that mayors of the conservative parties may not agree on the progressive tax hike.

The model also includes variables on labour resources and tax base. The level of labour resources is measured by the number of workers in major industrial sectors (NUM_WORKER).¹³ Local governments with large labour resources are less motivated to join the property tax competition in our theoretical model; refer to the Appendix 1. Also, since there were no local income and local sales taxes in Korea until 2010, only the property tax base should be taken into account rather than taxes on income or earnings. The property tax base of local government is measured by the average apartment price per square metre of a local jurisdiction (LN APT_PRICE). The apartment value may well represent the main tax base of a local government because apartments are a typical housing type in the SMA (Gelézeau, 2006). Local governments with high apartment prices will cut the property taxes to a large degree because they can meet their fiscal demands for public services with relatively low tax rates.

Major empirical findings

Local endowment distribution and tax cut

First, we estimated the spatial models for tax cut to analyse the relationship between tax cut and the local endowment distribution. Table 4 provides the results of different spatial panel models, spatial autoregressive model (SAR), and spatial error model (SEM), using two contiguity matrices, respectively. The results demonstrate that there exists tax competition as manifested by the highly significant values for spatial lag parameter ρ and that several variables account for a considerable amount of the variation in local governments’ tax cuts.¹⁴ The AIC and BIC values indicate that the two models are similarly appropriate, though SAR is slightly better than SEM. Also, ρ and λ are both highly significant, which justifies the use of spatial models.

Table 4.

Tax cut model estimation results.

Model variables	Non-spatial	1st order contiguity (W^c)		2nd order contiguity (W^c2)
Model variables	Non-spatial	SAR	SEM	SAR	SEM
SKEW	−6.434^† (3.32)	−7.260* (3.193)	−6.408* (3.155)	−6.576* (3.219)	−6.181^† (3.221)
PROBUD	−0.197* (0.088)	−0.165* (0.084)	−0.202* (0.101)	−0.177* (0.0852)	−0.199* (0.0963)
NUM_NURS	1.689*** (0.511)	1.171* (0.518)	1.197* (0.61)	1.157* (0.556)	1.110^† (0.645)
DEGIND	0.287* (0.13)	0.239^† (0.125)	0.230^† (0.129)	0.243† (0.127)	0.230^† (0.13)
LN LAG_LT	5.056* (2.085)	4.450* (2.005)	5.554* (2.194)	4.850* (2.022)	5.643** (2.168)
CONS_PARTY	−6.239 (3.899)	−5.144 (3.753)	−5.521 (3.816)	−4.56 (3.863)	−5.017 (3.878)
LN APT_PRICE	13.43*** (3.44)	9.497** (3.537)	12.70*** (3.673)	11.03** (3.521)	13.00*** (3.604)
NUM_WORKERS	−0.0107* (0.004)	−0.00796^† (0.00413)	−0.00922* (0.0043)	−0.00898* (0.0042)	−0.00997* (0.0042)
CONSTANT	−124.4*** (33.06)	−95.41** (33.06)	−119.2*** (34.9)	−108.2** (32.94)	−121.4*** (34.44)
ρ		0.299** (0.099)		0.274* (0.13)
λ			0.351** (0.118)		0.363* (0.166)
$σ_{μ}$	6.715	6.488	6.755	6.662	6.766
$σ_{ν}$	11.72	11.07	11.01	11.31	11.24
N	180	180	180	180	180
R2	0.3923	0.399	0.391	0.4180	0.392
AIC		1447.6	1448.5	1451.9	1452.2
BIC		1486	1486.8	1490.2	1490.5
Log lik.		−711.8	−712.3	−714	−714.1

Note: SAR stands for spatial autoregressive model, and SEM for spatial error model; standard errors in parentheses; ^†p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001.

The focal variable, SKEW, the skewness of home value distribution is significant consistently in all the models. This result underpins the argument that the tax base distribution skewed to the right may bring more public goods or higher tax rate than the distribution skewed to the left. Thus we accept the first research hypothesis, H1. Individuals’ heterogeneous resource endowments are a critical condition for the tax competition, when it is a tax base.

Previous studies argue that a high grant-in-aid from the superior government leads local governments to choose a low tax rate with less fiscal loss under tax competition (Brueckner and Saavedra, 2001; Wildasin, 1988b). In contrast, we find that local governments with high fiscal independence (DEGIND) choose a higher percentage of tax cut. Even if it looks in contradiction to rational choice theories, the behaviour of local governments is rational in the case of the SMA (Choe, 2005). Tax competition in the SMA is partly a tax resistance against the central government property tax hike that aims at a more progressive tax scheme. The central government utilises a grant-in-aid and designs the fiscal sanction against the property tax cuts by local government (Kim, 2008a, 2008b). That is, because the central government will not complement any local fiscal loss from the property tax cut and may give hidden disadvantages in the distribution of national tax resources to local governments, the local governments with small own source revenue cannot but yield to the central government’s property tax hike.

Per capita local tax revenue (LN LAG_LT), the proxy variable for fiscal health is significant and its coefficient has a positive sign as expected. This supports the expectation that a local jurisdiction’s tax capacity has an influence on the decision making of property tax cut. The results also support Brueckner and Saavedra’s (2001) argument that a community’s tax rate depends on how strong its demand for public goods is. The budget plan for local infrastructure and basic social services supply (PROBUD) has a significant and negative coefficient, implying that the high demand for public goods decreases the degree of tax cut. The number of the nursery facilities per 1000 children (NUM_NURS) is also significant and its coefficient is positive as expected.

The control variable to capture the vertical party competition, CONS_PARTY, has the negative sign and has insignificant coefficients in all the models, which does not support the existence of political competition between the central and local governments. This implies that the political leaders of local governments put focus more on the economic motive of residents than their political parties. Lastly the economic control variables to capture the tax base and the labour resource are also significant and have positive coefficients as expected.

Local endowment distribution and spatial dependence

Next, we estimated the panel model for the local Moran’s I of tax cut using the same data set. With two different weighting schemes, we have two different results of estimation, as shown in Table 5.

Table 5.

Local Moran’s I model estimation results.

	1st order contiguity (W^c)			2nd order contiguity (W^c2)
	I	II	III	IV	V	VI
SKEW	−0.595*	−0.558*	−0.559*	−0.670*	−0.618*	−0.618^†
SKEW	(0.263)	(0.263)	(0.268)	(0.312)	(0.31)	(0.32)
PROBUD	−0.0214*	−0.0218**	−0.0218*	−0.0263*	−0.0270*	−0.0276*
PROBUD	(0.008)	(0.008)	(0.00858)	(0.011)	(0.011)	(0.011)
NUM_NURS	−0.0761^†	−0.089n0*	−0.0852^†	−0.152**	−0.171**	−0.171**
NUM_NURS	(0.042)	(0.043)	(0.0465)	(0.0509)	(0.0521)	(0.057)
DEGIND	0.021*	0.0202*	0.0215^†	0.0259*	0.0247*	0.0244
DEGIND	(0.009)	(0.009)	(0.0126)	(0.011)	(0.011)	(0.015)
LN LAG_LT			−0.0296			0.00337
LN LAG_LT			(0.191)			(0.232)
CONS_PARTY		0.4	0.389		0.566	0.567
CONS_PARTY		(0.317)	(0.33)		(0.375)	(0.395)
NUM_WORKERS	−0.0008*	−0.000713^†	−0.000720^†	−0.0006	−0.0005	−0.0005
NUM_WORKERS	(0.00039)	(0.0004)	(0.00041)	(0.00047)	(0.00047)	(0.00049)
LN POP	−0.534**	−0.532**	−0.552*	−0.648**	−0.643**	−0.641*
LN POP	(0.202)	(0.201)	(0.244)	(0.24)	(0.238)	(0.294)
CONSTANT	9.376***	9.134***	9.655*	11.84***	11.50***	11.47*
CONSTANT	(2.612)	(2.604)	(4.274)	(3.114)	(3.091)	(5.146)
$σ_{u}$	0.399	0.390	0.416	0.298	0.270	0.363
$σ_{υ}$	1.140	1.140	1.113	1.495	1.495	1.436
N	180	180	180	180	180	180
R²	0.145	0.155	0.155	0.183	0.194	0.194

Note: Standard errors in parentheses; ^†p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001.

The focal variable, SKEW, has significantly negative coefficients in both specifications. These results are consistent with theoretical expectation that a municipality with the capital endowment distribution skewed to the right will respond less sensitively to its neighbouring communities. Thus we accept the second research hypothesis, H 2.

The other explanatory variables have the same signs as in the tax cut model except one variable NUM_NURS. The financial independence variable, DEGIND has significantly positive coefficients, which implies that a financially independent municipality is willing to participate in the tax cut race. The two variables PROBUD and NUM_NURS representing the demand for public goods have the opposite signs. While PROBUD has the same sign as in the tax cut model as expected, NUM_NURS has the opposite sign. It may be due to the fact that municipalities with low values of NUM_NURS are located within the City of Seoul, where tax cut competition was high in the mid-2000s. This contrast between low values of the variable and high dependence in tax rate can lead to the negative sign of the coefficient of NUM_NURS.

Concluding remarks

This study investigates how the local distribution of resource endowments affects the property tax competition among local governments. Under the heterogeneity of endowments within a municipality, the marginal utility of public goods cannot determine the pattern of tax competition by itself, but the combined effect of the demand for public goods and the resource endowment distribution determines the property tax cut. The heterogeneity of labour and capital endowments implies the pre-electoral competition about the property tax rate or political actions by home owners, which in turn affects the tax competition. The theory implies that the local distribution of resource endowments affects not only a municipality’s decision on the tax rate, but also the degree of spatial dependence in tax cut. The results of empirical analysis also support these expectations.

First, the spatial lag parameter is statistically significant and has the positive sign, which indicates a positive spatial interdependence of the property tax rates among municipalities. This implies that positive-sum tax competition in the SMA occurs: holding all other things constant, if one local government cuts the tax rate, others also cut the tax rate strategically. Also, this reaction pattern is stronger as local jurisdictions are closer to each other. These results confirm that the Tiebout tax competition model among local governments is valid in Korea as well, not specific to the USA, for analysing the strategic tax interaction.¹⁵

Second, the local distribution of resource endowments, as measured by skewness of home values, affects tax rate. A municipality with higher skewness in home values is adopting a higher tax rate, in other words, a lower tax cut. In that community, individuals who own low-priced homes are the majority, and they have a strong political will that pushes the local government to raise or maintain the property tax. This is consistent with the previous studies on the tax rate choice in the pre-electoral competition (Borck, 2003; Downs, 1957; Persson and Tabellini, 2000).

Third, the local distribution of resource endowments also affects how much a municipality is dependent on its neighbours in making a decision on tax rate. A municipality with higher skewness in home values is less sensitive to its neighbouring communities, and thus it is less likely to participate in the tax cut race. In that community, individuals who own low-priced homes are the majority, and they do not push the local government to cut its tax rate in spite of tax cut dominoes. This finding is new to the literature, and can be an empirical evidence for the theoretical prediction of the previous studies.

Lastly, a municipality with high demand for public goods will maintain a high tax rate to meet the demand. The results demonstrate that the ratio of project budget and the number of nursery facilities per 1000 children as proxies for the demand for public goods restricts the tax cut decision. The results of the local Moran’s I analysis on the ratio of project budget show that a high demand for public goods reduces the degree of spatial dependence in tax competition, implying that the local government tends to be less sensitive to its neighbours.

In conclusion, this study theoretically elaborates that the local distribution of resource endowments affects both the level of tax rate and the degree of spatial dependence in tax competition, and empirically confirms the theory using a panel dataset for the SMA municipalities in 2004–2006. It extends the tax competition literature by incorporating the heterogeneity of resource endowments into the tax competition framework, and reaching a unique conclusion about the relationship between the spatial dependence in tax cut and the local endowment distribution.

Footnotes

Appendix 1. Tax rate choice in the tax competition

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Notes

References

Kang

Lee

(2014) A study on the impact of soft location factors in the relocation of service and manufacturing firms. International Journal of Urban Sciences 18(3): 327–339.

Anselin

(1995) Local indicators of spatial association: LISA. Geographical Analysis 27(2): 93–115.

Belotti

Hughes

Mortari

(2013) XSMLE: Stata module for spatial panel data models estimation. Statistical software components S457610, Boston College Department of Economics (revised 15 Mar 2014).

Borck

(2003) Tax competition and the choice of tax structure in a majority voting model. Journal of Urban Economics 54: 173–180.

Boyne

(1994) Party competition and local spending decisions. British Journal of Political Science 35: 210–222.

Brueckner

(2000) A Tiebout/tax-competition model. Journal of Public Economics 77(2): 203–210.

Brueckner

Saavedra

(2001) Do local governments engage in strategic property-tax competition? National Tax Journal 54(3): 231–253.

Buchanan

(1999) Public Finance in Democratic Process: Fiscal Institution and Individual Choice. Indianapolis, IN: Liberty Fund.

Bucovetsky

(1991) Asymmetric tax competition. Journal of Urban Economics 30(2): 167–181.

10.

Bucovetsky

(1995) Rent seeking and tax competition. Journal of Public Economics 58(3): 337–363.

11.

Buettner

(2001) Local business taxation and competition for capital: The choice of the tax rate. Regional Science and Urban Economics 31: 215–245.

12.

Burbidge

Myers

(1994) Population mobility and capital tax competition. Regional Science and Urban Economics 24(4): 441–459.

13.

Case

Rosen

Hines

Jr (1993) Budget spillovers and fiscal policy interdependence Evidence from the states. Journal of Public Economics 52(3): 285–307.

14.

Charlot

Paty

(2010) Do agglomeration forces strengthen tax interactions? Urban Studies 47(5): 1099–1116.

15.

Choe

(2005) The political economy of property tax competition in Korea: Considering the difference in fiscal conditions. The Korean Journal of Local Government Studies 9(4): 47–69.

16.

Doreian

(1980) Linear models with spatially distributed data: Spatial disturbance or spatial effects. Sociological Methods and Research 9(1): 29–60.

17.

Downs

(1957) An economic theory of political action in a democracy. The Journal of Political Economy 65(2): 135–150.

18.

Dubois

Leprince

Paty

(2007) The effects of politics on local tax setting: Evidence from France. Urban Studies 44(8): 1603–1618.

19.

Elhorst

Fréret

(2009) Evidence of political yardstick competition in France using a two-regime spatial Durbin model with fixed effects. Journal of Regional Science 49: 931–951.

20.

Fell

Haynie

(2013) Spatial competition with changing market institutions. Journal of Applied Econometrics 28(4): 702–719.

21.

Fuest

Huber

(2001) Labor and capital income taxation, fiscal competition, and the distribution of wealth. Journal of Public Economics 79(1): 71–91.

22.

Gelézeau

(2006) Globalization, housing culture, and new urban space of Seoul. Asia Studies 125: 7–36.

23.

S-K

(2013) Housing markets and government intervention in East Asian countries. International Journal of Urban Sciences 17(1): 32–45.

24.

Hansen

(1983) The Politics of Taxation: Revenue Without Representation. Westport, CT: Praeger.

25.

Hayashi

Boadway

(2001) An empirical analysis of intergovernmental tax interaction: The case of business income taxes in Canada. Canadian Journal of Economics 34: 481–503.

26.

Hernandez-Murillo

(2003) Strategic interaction in tax policies among states. Federal Reserve Bank of St. Louis Review 85: 47–56.

27.

Heyndels

Vuchelen

(1998) Tax mimicking among Belgian municipalities. National Tax Journal 51(1): 89–101.

28.

Hofferth

Brayfield

Deich

et al . (1991) National Child Care Survey, 1990. Washington, DC: Urban Institute Press.

29.

Kim

(2008a) The choice of local governments under two principals: The case of tax hike policy of South Korea – Noncompliance of local governments to tax hike policy. Korean Journal of Policy Studies 23(1): 181–207.

30.

Kim

(2008b) Three actors and three perspectives in property tax competition of Seoul Metropolitan Area: Economic motives and political actions. PhD Thesis, Florida State University, USA.

31.

Korean Real Estate Information Association (KREIA) (2004, 2005, 2006) Quarterly apartment price. Available at: http://price.joinsland.joins.com/area/?areaCode=1168000000&mcateGroup=A1 (accessed 31 March 2006).

32.

Ladd

(1992) Mimicking of local tax burdens among neighboring counties. Public Finance Review 20(4): 450–467.

33.

Lowi

(1970) Decision making vs. policy making: Toward an antidotes for technocracy. Public Administration Review 30: 314–325.

34.

McMillen

Singell

Waddell

(2007) Spatial competition and the price of college. Economic Inquiry 45: 817–833.

35.

Man

(1999) Fiscal pressure, tax competition and the adoption of tax increment financing. Urban Studies 36(7): 1151–1167.

36.

Ministry of Government Administration and Home Affairs (MOGAHA) (2003, 2004a, 2005a, 2006a) Annual local fiscal almanacs of Korea. Report, Republic of Korea.

37.

Ministry of Government Administration and Home Affairs (MOGAHA) (2004b, 2005b, 2006b) Registered population. Available at: http://kosis.kr (accessed 5 August 2013).

38.

Ministry of Health and Welfare (MOHW) (2004, 2005, 2006) Statistics of child care by local jurisdiction. Available at: http://kosis.kr (accessed 17 January 2016).

39.

Mintz

Tulkens

(1986) Commodity tax competition between member states of a federation: Equilibrium and efficiency. Journal of Public Economics 29: 133–172.

40.

National Election Commission of Korea (NECK) (2002) Local election data 2002 from election information archive of national election commission of Korea. Available at: http://www.nec.go.kr:7070/dextern/ (accessed 5 August 2013).

41.

Persson

Tabellini

(2000) Political Economics: Explaining Economic Policy. Cambridge, MA: The MIT Press.

42.

Revelli

(2001) Spatial patterns in local taxation: Tax mimicking or error mimicking? Applied Economics 33: 1101–1107.

43.

Seo

Stafford

Seiling

(2005) Mothers’ share of child care in rural low-income families. Consumer Interests Annual 51: 107–120.

44.

Statistics Korea (2004, 2005a, 2006) Nation-wide business and employment investigation. Available at: http://kosis.kr (accessed 17 January 2016).

45.

Statistics Korea (2005b) 2005Population and housing census. Available at: http://kosis.kr (accessed 11 August 2014).

46.

Statistics Korea (2011) Aggregate statistics of the sample survey for the 2010 Population and Housing Census. Press release for 27 October 2011. Available at: http://kostat.go.kr/smart/news/file_dn.jsp?aSeq=251652&ord=1 (accessed 10 January 2016).

47.

Swensen

(2008) Child care arrangements in urban and rural areas. Report for Office of the Assistance Secretary for Planning and Evaluation. Washington, DC: US Department of Health and Human Services.

48.

Tiebout

(1956) A pure theory of local expenditures. The Journal of Political Economy 64(5): 416–424.

49.

Wildasin

(1988a) Nash equilibria in a model of fiscal competition. Journal of Public Economics 35: 241–249.

50.

Wildasin

(1988b) Interjurisdictional capital mobility: Fiscal externality and a corrective subsidy. Journal of Urban Economics 25: 193–212.

51.

Wilson

(1985) Optimal property taxation in the presence of interregional capital mobility. Journal of Urban Economics 18: 73–89.

52.

Zodrow

Mieszkowski

(1986) Pigou, Tiebout, property taxation, and the under-provision of local public goods. Journal of Urban Economics 19: 356–370.