103
China & World Economy / 103–120, Vol. 20, No. 5, 2012
Strategic Interaction in Spending on Environmental
Protection: Spatial Evidence from Chinese Cities
Huihui Deng, Xinye Zheng, Nan Huang, Fanghua Li*
Abstract
In China, the responsibility of protecting the environment lies largely with local governments.
Within the framework of spatial econometrics, we investigate empirically the consequence of
such an institutional setting. Using city-level data for China, the present study finds that city
governments behave strategically in making spending decisions regarding environmental
protection. This paper finds that a city government appears to cut its own spending as a
response to the rise in environmental protection spending by its neighbors. Hence,
environmental protection tends to be underprovided. As a result, we suggest that centralizing
the environmental protection responsibility to a higher level of government would be beneficial
in terms of controlling pollution in China.
Key words: environmental protection, spatial econometrics, strategic interaction
JEL codes: C21, H70, Q50
I. Introduction
Over the past two decades, China has experienced rapid economic growth, with annual
increases in GDP of 8–9 percent (World Bank, 2007a). However, economic growth has
come at the cost of environmental deterioration and resource depletion. According to
Elizabeth Economy (2007, p. 1), “China’s environmental problems are mounting. Water
pollution and water scarcity are burdening the economy, rising levels of air pollution are
endangering the health of millions of Chinese, and much of the country’s land is rapidly
* Huihui Deng, Assistant Professor of Economics, Institute of International Economy, University of
International Business and Economics. Email: [email protected]; Xinye Zheng (Corresponding Author),
Professor of Economics, School of Economics, Renmin University of China. Email: [email protected];
Nan Huang, Undergra duate Student, School of Economics, Renmin University of China . Ema il:
[email protected]; Fanghua Li, Graduate Student, Department of Economics, University of
California, Los Angeles. Email: [email protected]. This research was supported by the Fundamental
Research Funds for the Central Universities and the Research Funds of Renmin University of China
(11XNL009).
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
104
Huihui Deng et al. / 103–120, Vol. 20, No. 5, 2012
turning into desert.”1 According to Baoxing Qiu, Deputy Minister of Construction, aquifers
in approximately 90 percent of China’s cities are polluted and three-quarters of China’s
rivers suffer from some degree of contamination.2 Water pollution is so severe that the
health-related damage amounted to RMB9.47bn in 2003 (World Bank, 2007a). Meanwhile,
air quality in several of China’s cities is among the worst in the world. According to the
World Bank (2007b), 12 of the 20 most polluted cities in the world are located in China.
China is the world’s biggest emitter of SO2 and probably also the world’s principal source
of CO2 emissions, and it is estimated that 13 percent of all urban deaths might be premature
due to ambient air pollution (Vennemo et al., 2009). The health damage resulting from air
pollution is estimated to be 3.8 percent of GDP in China (World Bank, 2007a).
The task of pollution control has been assumed by most governments (the Chinese
Government is no exception) as a consequence of fiscal decentralization (Feruglio and
Anderson, 2008). Under fiscal decentralization, important expenditure responsibilities are
assigned to local governments. Hence, it is important to understand the relationship between
local governments’ behavior and environmental pollution (Zhang and Zheng, 2009).
Governments play a major role in environmental protection, through setting environmental
regulations and environmental protection expenditure. In the present study, we focus only
on governments’ expenditure behavior in regards to environmental protection. When
expenditure responsibilities are decentralized, it is expected that each government will not
act independently: they will incorporate to some extent the behavior of their neighboring
government. There are at least three channels through which local governments are expected
to affect the spending behaviors of their neighboring governments (Manski, 1993). First,
expenditure competition might occur if local governors compete with their neighbors to
attract households or firms (Case et al., 1993; Revelli, 2003; Lundberg, 2006; Yu et al., 2012).
Second, yardstick competition might occur if local governments’ spending on environmental
protection promotes local economic development and growth, which is the major indicator
used by higher level governments to evaluate and promote (or reappoint) local governors
(Chen et al., 2005; Caldeira, 2012). Third, expenditure externalities might occur if public
spending (e.g. on infrastructure, environmental protection and health care) by one local
government can create benefits for or have detrimental effects on its neighboring jurisdictions
(Solé Ollé , 2003; Fréret , 2006; Yu et al., 2012). In addition, Manski (1993) suggests that
fiscal interactions might result from a “common intellectual trend” that drives fiscal choices
1
Vennemo et al. (2009) provide a broad overview of the current status of China’s environment, and a
comprehensive discussion of recent trends in air and water pollution.
2
See: http://www.chinadaily.com.cn/english/doc/2005-06/07/content_449451.htm (accessed 5 March
2011).
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
105
Strategic Interaction in Spending on Environmental Protection
in the same direction. In relation to public spending on environmental protection, the
central or provincial governments might offer common directives across city governments
to increase (or decrease) certain kinds of expenditure. Such a spatial effect could occur in
the context of China. In the following empirical analysis, these sources of fiscal interaction
among city governments are to be disentangled by applying a spatial econometric technique.
Theoretically, there is no consensus regarding whether decentralized environmental
policy-making induces a “race to the bottom” or a “race to the top” (Levinson, 1997). On
the one hand, local governments tend to “race to the top” by enacting higher environmental
standards to prevent undesirable facilities from locating within a jurisdiction (Markusen et al.,
1995; Wilson, 1996; Glazer, 1999). On the other hand, local governments might “race to the
bottom” by setting laxer environmental standards to attract and retain capital (Markusen et al.,
1995; Wilson, 1996; Esty and Geradin, 1997).
Empirically, Fredriksson and Millimet (2002) find evidence that state regulators in the
USA respond strategically to the regulatory standards of other jurisdictions, but they are
unable to find evidence supporting a race to the top or a race to the bottom hypothesis.
Using state-level data to examine the impact of decentralized environmental policy-making
under President Ronald Reagan in the USA, List and Gerking (2000) find that Reagan’s
decentralization did not lead to a race to the bottom. This finding is consistent with the
empirical work by Millimet (2003), but Millimet finds evidence of a race to the top in regards
to pollution control expenditure.
The disagreement on the effect of environmental decentralization in these studies may
be attributed to several factors, including the different tax instruments used and the range
of periods of time covered, or ad hoc models specified. From an econometric perspective,
the disagreement might be due to an omitted variable bias issue, which occurs when empirical
decentralization-related studies fail to account for tax and expenditure competitions
simultaneously given that taxation and expenditure decisions are usually simultaneously
determined by local governments (in the USA).
China serves as a good example for examining governments’ environmental expenditure
behavior in the context of fiscal decentralization for several reasons. First, in China, the
local tax rate cannot be altered by local governments. In other words, local governments
cannot engage in tax competition; hence, we are able to examine the expenditure competition
behavior of Chinese local governments. Second, despite tremendous efforts made by the
Chinese Government to control pollution over the past two decades, environmental pollution
in the form of water pollution and air pollution remains serious. Finally, China has undergone
fiscal decentralization reform since 1994 and important expenditure responsibilities (e.g.
environmental expenditure) have been gradually assigned to local governments.
Then, how is the Chinese Government’s environmental expenditure behavior
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
106
Huihui Deng et al. / 103–120, Vol. 20, No. 5, 2012
characterized under fiscal decentralization? Do Chinese city governments behave
strategically in making environmental protection spending decisions? Specifically, will the
decentralization reform induce a race to the bottom or to the top of public spending on
environmental protection? These questions are answered in the present study. Using a
cross-sectional dataset including 249 Chinese cities in 2005, the present study applies
spatial econometric methods to examine the factors that determine public environmental
protection spending in a decentralized economy. The empirical results reveal that city
governments do behave strategically in making spending decisions relating to environmental
protection. A city government tends to cut its own spending as a response to a rise in the
environmental protection spending of its neighbors.
The remainder of the paper is structured as follows. Section II presents the econometric
procedures and the empirical spatial model. Section III describes the data used in the
spatial regression. Section IV reports the estimation results and Section V provides an
additional analysis. The final section concludes with a summary of the main findings and
policy implications.
II. Model Specification
Using cross-sectional data for 249 Chinese cities, an empirical model for environmental
protection expenditure can be specified in a classical linear model as:
y = αιn+ Xβ + є, є ~ N(0, σє2In), n = 1, 2, …, 249,
(1)
where ιn is an n × 1 vector of ones associated with the constant term parameter α . y is an
n × 1 vector of the dependent variable denoting per capita city government spending on
environmental protection, and X is a set of explanatory variables that are identified to
influence a city government’s investment in pollution protection. ε is the error term across
observations and β is the parameter to be estimated in this model.
Following the spatial econometric approach developed by Anselin (1988), the
aforementioned model can be extended to account for spatial dependence in the
environmental protection spending of municipal cities. The spatial dependence can take
several forms. One is the spatial lag model specifying the spatial correlation in the dependent
variable, which more or less resembles the autoregressive model in time-series econometrics;
the other is the spatial error model allowing for spatial autocorrelation in the error term,
which is similar to the moving average model in time-series econometrics. These two types
of spatial correlation are commonly seen in the spatial econometric literature.
To account for spatial correlation in the dependent variable, that is, to specify a spatial
autoregressive model (SAR), the classical linear regression model in Equation (1) is adapted
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
107
Strategic Interaction in Spending on Environmental Protection
to be:
y = αιn + λWy + Xβ + є, є ~ N(0, σє2In), n = 1, 2, …, 249,
(2)
where W is a non-stochastic n × n spatial weights matrix in which the element mij is equal to
1/dij, with dij being the (greater-circle) distance between 2 cities, i and j (i ≠ j).3 In making
such a specification, we assume that as the distance between cities i and j increases
(decreases), mij decreases (increases), which poses less (more) spatial weight to the city
pair (i, j). The weight matrix is defined a priori and does not include parameters to be
estimated. In this model, the parameters to be estimated are the usual regression parameters
α , β and σ 2, and the spatial lag parameter λ , which measures the magnitude of interactive
behavior across city governments. According to Brueckner (2003), a negative value of λ
implies the presence of an expenditure externality, while a positive value of λ implies possible
coexistence of an expenditure externality, yardstick competition and expenditure competition,
which implies that additional research should be implemented to disentangle these sources.
In addition, if there is no spatial dependence in the vector of cross-sectional observations
y, λ takes a value of zero, and this model reduces to the classical least-squares regression
(see Equation 1). In the event of spatially-lagged dependence, the traditional OLS method
does not yield unbiased and consistent estimates as the autoregressive component of the
model is correlated with residuals (Anselin and Bera, 1998). Thus, to estimate the model
with a spatially-lagged dependent variable, we apply the maximum likelihood estimation
(MLE) method (Anselin, 1988).4
In contrast, a model incorporating a spatial dependence in the disturbances (i.e., a
spatial error model (SEM)) can be specified as:
y = αιn + Xβ + μ, μ = ρWμ + є, є ~ N(0, σє2In).
(3)
Despite OLS estimation in the presence of spatial correlation among model disturbances
yielding unbiased coefficient estimators, estimates of standard errors will be inconsistent,
which implies that t-statistics and F-statistics will be incorrect and misleading. Therefore,
the spatial error model is also estimated using MLE.
A general form of the spatial model, the spatial autoregressive moving average
(SARMA) model of Cliff and Ord (1981), combines both the spatially-lagged term as well as
the spatially-correlated error structure. It is specified as follows:
3
The greater-circle distance is calculated based on the longitude and latitude of each city, which can be
obtained (as of 3 February 2011) from http://www.gpsspg.com/maps.htm. Another commonly-used
weight matrix specification is the contiguity-based binary matrix in which each element m ij of the matrix
W is set as one if city i and j (i
4
≠
j) share a common border, and zero otherwise.
The log-likelihood function for each spatial model in this study can be found in LeSage and Pace (2009).
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
108
Huihui Deng et al. / 103–120, Vol. 20, No. 5, 2012
y = αιn + λWy+ Xβ + є, є = ρWє + v, v ~ N(0, σv2In), n = 1, 2, …, 249,
(4)
where all variables and parameters are defined as above.5 The SARMA model should also
be estimated using the MLE method.
In the empirical implementation, Lagrange multiplier (LM) tests and their robust versions
can be used to choose among the SAR and the SEM models, and likelihood ratio (LR) tests
can be used to choose between the SAR or the SEM model and the general SARMA model
(Anselin, 1988; LeSage and Pace, 2009).
III. Data Source
As Ermini and Santolini (2010) argue, verifying the existence and the magnitude of spatial
interaction remains mainly an empirical issue, although the theoretical literature often
assumes the presence of economic interaction among jurisdictions. Traditionally, empirical
models of local public expenditure include indicators of income (shares), and variables
measuring fiscal, socioeconomic, demographic and geographical characteristics of the
jurisdiction (Baicker, 2005; Ermini and Santolini, 2010; Yu et al., 2011, 2012). The explanatory
variables chosen in the present study are based on those used in the existing literature.
The dataset for this study consists of a cross-section of 249 Chinese cities in 2005.
Table 1 lists the variables used in the empirical model. All variables, unless otherwise noted,
are derived from China City Statistical Yearbook 2006 (NBS, 2006).6 The dependent variable
(EP) is per capita investment in environmental protection, which is divided into three
categories: investment in urban environment infrastructure facilities (IEIF); investment in
treatment of industrial pollution sources (ITIP); and investment in environment components
for new construction projects (IECC) (NBS, 2009). Based on statistical data from the National
Bureau of Statistics (NBS, 2009), we find that the first category (IEIF) takes up the majority
of total public investment in environmental protection. Specially, IEIF accounted for 50 percent
or more of total public investment, with the number reaching above 70 percent in 2005. As
5
According to Anselin and Bera (1998), the spatial weight matrix used in the lag component in the
SARMA model can be different from the matrix used in the error component. For simplicity, we assume
these two weight matrices are identical.
6
Unfortunately, we cannot use the datasets from recent publications. The dependent variable (EP) (the
city-level public spending on environmental protection) is not available in particular years after 2005.
We could not find this item for years 2006, 2008, 2009 and 2010. We did find EP data at the city level
in 2007 (i.e. from the China City Statistical Yearbook 2008), but several cities (such as all cities in Hebei
Province, Henan Province, and some cities in other provinces) do not have the EP data. As a result, we
did not use the datasets from China City Statistical Yearbook 2008.
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
109
Strategic Interaction in Spending on Environmental Protection
public investment in urban environmental infrastructure facilities includes mainly projects
such as natural gas supply, heating, drainage, landscaping and urban sanitation, the IEIF
is less closely related to environmental pollution treatment than the latter two categories of
investment (ITIP and IECC).
With respect to the explanatory variables, REVENUE is an indicator reflecting a city’s
fiscal capacity, which is measured by the city government’s own revenue per capita. On
one hand, Wagner’s Law states that government activities increase as economies grow
(Wagner, 1911). The law implies that as an economy develops, government will spend more
on treating environmental pollution. Thus, the influence of a government’s own revenue
on expenditure in environmental protection is expected to be positive. In addition, the
environmental Kuznets curve hypothesis posits an inverted U-shaped relationship between
environmental quality and economic development/growth. Similarly, in the present study,
we hypothesize that there exists an inverted parabolic relationship between governments’
own revenue and environmental protection spending. In the early stages of economic
development/growth, environmental regulation is weak and ineffective, and more effort
and greater local government revenue are required to reduce environmental pollution. As
fiscal revenue increases (due to economic growth), leading industries become cleaner, and
environmental regulation becomes more effective. Environmental improvements can
eventually lead to falls in government revenue in later stages of development. Thus, the
variable REVENUE2 (defined as revenue squared) is expected to be negative. The
urbanization rate (URBAN) is measured as the percentage of the total population of a city
living in urban areas. China’s rapid economic development has come at the cost of
environmental degradation. While the urbanization rate can be a good indicator of a city’s
stage of economic development, a high urbanization rate implies that a city has undergone
environmental degradation and implies further that more spending on environmental
protection is required. Hence, the variable URBAN is expected to be positive. EDUCATION,
measured as the percentage of a city’s total public expenditure allocated to education, is
another variable considered to affect a city government’s spending on environmental
protection. Given the constant budget constraints of local governments, environmental
protection expenditure and education expenditure are expected to be substitutes of each
other. Consequently, the education variable is expected to be negative. SO2 indicates
industrial sulfur dioxide emissions divided by the total population in the city. More industrial
emissions lead to greater environmental pollution, which calls for further environmental
protection efforts. Thus, we expect the coefficient of SO2 to be positive. The natural
conditions of a city, such as the existence of green plants, have air-purifying effects through
removing toxins (e.g. formaldehyde, benzene and carbon monoxide) from the air. It is expected
that a city with better natural conditions will spend less on environmental treatments.
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
110
Huihui Deng et al. / 103–120, Vol. 20, No. 5, 2012
Table 1. Summary Statistics of Variables in 249 Chinese Cities (2005)
Dependent va riable
EP (RMB)
Explanatory variable
REVENUE (RMB)
REVENUE2 (RMB)
RAIN (mm)
SO2 (ton/10 000 persons)
URBAN (%)
EDUCATION ( %)
Obse rvation
Mea n
Standard
deviation
249
169.59
249
249
249
249
249
249
867.67
1.81E+06
895.98
215.94
0.34
0.19
Minimum
Ma ximum
245.73
0.73
1632.53
1030.5
5.87E+06
386.81
283.95
0.17
0.05
70.33
4945.93
144
0.2
0.08
0.08
7835.27
6.14E+07
2050.4
2593.88
0.99
0.49
Notes: EP, per capita spending on environmental protection; REVENUE, city government’s own resource
revenue per capita; REVENUE2, the term REVENUE squared; RAIN, averaged amount of precipitation
recorded during a year; SO2, per capita sulfur dioxide emissions; URBAN, percentage of total population
living in the urban area; and EDUCATION, ratio of public education spending to total city public spending.
Furthermore, cities with greater rainfall are believed to have better air quality (partly due to
the purification function of rain) and more green plants. Therefore, it is justifiable to use the
average amount of precipitation recorded during the year 2005 as a proxy variable (RAIN) to
reflect a city’s natural conditions. The data for all Chinese cities are taken from the Provincial
Statistical Yearbook 2009.
In specifying the model, city-specific characteristics that are not captured by the
explanatory variables as mentioned above might affect a city’s public spending behavior in
relation to environmental protection in that city. Not accounting for these characteristics
could lead to potential omitted variable bias. To deal with this issue, provincial dummy
variables are included in the model. Doing so allows removal of the common directives set
uniformly by the upper governments across city governments for increasing (or decreasing)
certain kinds of expenditure (i.e. adding provincial dummy variables allows us to remove
the third sources of fiscal interactions mentioned above while focusing on other possible
sources).
IV. Empirical Results
As mentioned above, the results based on OLS (Equation 1) are inconsistent if spatial
effects exist in the model. To test the existence of spatial dependence, we apply Moran’s
(1950) I-test on the residuals from the OLS model. The test statistic of 2.543 (p = 0.011)
indicates that there is significant spatial dependence in the data, suggesting that a spatial
econometric model, a spatial lag model (Equation 2) or a spatial error model (Equation 3),
should be used (see Table 2).
The LM-lag test is applied to test for spatially-lagged dependence. The LM-error test
is applied to test for spatial error dependence. The LM-error test reveals evidence of
spatially correlated error dependence at a 10-percent level of significance, while the LM-lag
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
111
Strategic Interaction in Spending on Environmental Protection
Table 2. OLS Results for Environmental Protection Expenditure in 249 Chinese Cities (2005)
REVENUE
REVENUE2
SO2
URBAN
RAIN
EDUCATION
CON STANT
Provincial dummies
Adjusted R2
Obse rvations
Diagnostics test
Jarque–Bera normality test
Spatial error:
Moran’s I
Lagrange multiplier
Robust Lagra nge multiplier
Spatial lag:
Lagrange multiplier
Robust Lagra nge multiplier
OLS model
0. 120***
(3. 32)
–2.430E-06**
(2. 44)
0. 146***
(3. 10)
458.675***
(4. 58)
–0.179**
(2. 48)
288.707
(1. 02)
–78.468
(0. 75)
Controlle d
0. 57
249
3.715 [0.156]
2.543 [0.011]
2.839 [0.092]
3.148 [0.076]
3.744 [0.053]
4.709 [0.030]
Notes: Absolute t-statistics are reported in parentheses; p-values are shown in brackets. ***, ** and *
denote significance at the 1, 5 and 10-percent level, respectively.
test does not reveal spatially-lagged dependence. However, these two tests fail to separate
one form of spatial dependence from another if only one or both types of dependence exist.
Hence, we resort to two robust versions of these two LM tests. It is shown that either test
statistic is found to be greater than its corresponding critical values (p < 0.05 for robust
LM-lag; p < 0.10 for robust LM-error). This result suggests that a general model with spatial
error and spatial lag terms (i.e. the SARMA model) should be used. In addition, we find that
the log-likelihood for the SAR, the SEM and the SARMA models are, respectively, –1623.60,
–1607.28 and –1535.51. The LR test result for the SARMA versus the SAR (176.18, with one
degree of freedom [df], p < 0.01) or the SEM model (143.54, 1 df, p < 0.01) confirms that the
SARMA model is the best candidate to explain the data.
1. Main Findings
Column 1 of Table 3 reports the results of the SARMA model.7 We find that the spatial lag
parameter (λ ) and the spatial error coefficient (ρ) are, respectively, –0.253 and 0.627, and
both are statistically significant at 1 percent. This finding implies that, within the current
environmental management system, Chinese city governments are engaging in strategic
7
Results of the SAR and SEM are available upon request.
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
112
Huihui Deng et al. / 103–120, Vol. 20, No. 5, 2012
interactions in regards to spending on environmental protection. Furthermore, spatial
interactions across city governments appear in the form of both spatially-lagged dependent
variables and spatially-correlated disturbances. In other words, while we find that city
governments appear to act similarly following a common natural shock (e.g. acid rain or a
sandstorm) or a policy shock (e.g. the to-be-enacted Taihu Management Plan (Tai Lake
Guanli Tiaoli in Chinese)),8 or even due to other unobserved omitted variables, the result
reveals that a city government tends to cut its own spending as a response to a rise in
environmental protection spending by its neighbors.
With respect to the explanatory variables, a city government’s own revenue
(REVENUE) and its squared term are, respectively, positive and negative, and both
terms are statistically significant. This result indicates an inverted U-shaped relationship
hypothesis between fiscal capacity and environmental protection spending. 9 In other
words, per capita spending on environmental protection increases first as per capita
government revenue rises and then falls after reaching a certain turning point. SO2 is
found to positively affect environmental protection spending. This result implies that
increased industrial SO2 emissions cause more devastation to the environment, which
calls for more spending directed at protecting the environment. URBAN is found to be
statistically significant at the 5-percent level, with a positive coefficient sign. This result
is consistent with our expectation. From the demand-side perspective, China’s rapid
economic growth and fast urbanization progress are accompanied by degrading
environmental quality, and require the government to invest more to protect the
environment. From the supply-side perspective, fast urbanization progress could promote
the economic growth of a city, further raising the city’s fiscal revenue. Therefore, the city
government is more capable of paying for environment abatement. The share of education
expenditure (EDUCATION) is found to be statistically insignificant. The precipitation
8
The Plan requires that Jiangsu, Zhejiang and Shanghai Provinces take action to clean up massive water
pollution in the basin area. Tai Lake is located at the junction of Shanghai, Jiangsu and Zhejiang
Provinces. It is the third largest freshwater lake in China, after Poyang and Dongting Lakes, with a
total area of 36 900 km2. The Taihu Lake Basin contains myriads of lakes and rivers, with a total river
course length of more than 120 km (Chen, 2005).
9
According to LeSage and Pace (2009), in the spatial model, the interpretation of the parameter is
different from the conventional least square interpretation. In the traditional OLS model, the coefficients
represent marginal effects, yet in the spatial models (SAR or SARMA) one can distinguish between
(average) direct effects, indirect effects and total effects that also take into account the feedback effects
arising from the spatial dependence. However, although the calculated total effect of one independent
variable on the dependent variable is found to be the same as the coefficient for the SARMA model, our
qualitative analysis for the explanatory variables will not be affected (i.e. given the coefficient sign from
the regression table).
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
113
Strategic Interaction in Spending on Environmental Protection
Table 3. Regression Results of Environmental Protection Expenditure in 249 Chinese Cities (2005)
REVENUE
REVENUE2
SO2
URBAN
RAIN
EDUCATION
CONSTANT
λ (Lag)
ρ (Error)
Provincial dummies
Observations
Number of excluded instruments
First-stage F-test
Sargan’s overidentification test
Hausman’s endogeneity test
Spatial autoregressive moving average
model (SARMA)
0.028**
(2.36)
–7.770E-07**
(2.03)
0.387**
(6.19)
83.182**
(1.97)
0.182*
(1.76)
–288.375
(0.81)
–75.891
(0.67)
–0.253***
(2.67)
0.627***
(9.07)
Controlled
249
Generalized spatial two-stage
least squares (GS2SLS)
0.020**
(6.28)
–1.183E-06***
(2.88)
0.245***
(4.64)
131.623**
(2.05)
0.221
(1.33)
117.63*
(1.86)
–58.90
(1.41)
–0.364**
(2.18)
0.156***
(3.49)
Controlled
249
10
[0.020]
[0.092]
[0.065]
Notes: Absolute t-statistics are reported in parentheses. The results in column 2 are based on the GS2SLS
procedure by Kelejian and Prucha (1998). The instrumented variables are WX, W2 X, where X are
vectors of REVENUE, SO2, URBAN, RAIN and EDUCATION and W is the distance-based spatial
weighting matrix. ***, ** and * denote significance at the 1, 5 and 10-percent level, respectively.
variable, RAIN, is positive and carries the unexpected sign, which could be partially due
to the reverse causality problem between rainfall and environmental protection, which is
not taken into account in the SARMA model.
2. Robustness Check
The variable Wy is endogenous, which might affect the estimations in the SARMA model.
In addition, some explanatory fiscal variables can be endogenous. To check whether the
results obtained from the SARMA regression are robust to possible endogeneity bias,
Kelejian and Prucha (1998) suggest an instrumental variable method, a three-step procedure
known as the generalized spatial two-stage least squares (GS2SLS) procedure.10 The GS2SLS
procedure proceeds as follows.
10
Das et al. (2003) examine the finite sample properties of the GS2SLS estimator of the regression
parameters of an SARMA model using Monte Carlo simulations and find that in many situations the loss
of efficiency due to the approximation of the ideal instruments on which the formulation of the GS2SLS
estimator is based is minor (i.e. their results suggest that the GS2SLS estimator is virtually as efficient as
the ML estimator in finite samples).
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
114
Huihui Deng et al. / 103–120, Vol. 20, No. 5, 2012
In the first step, the regression model in Equation (4) is estimated by two-stage least
squares using a set of instrument variables, H (X, WX, W2X). That is, we regress Wy on X,
WX, W2X and use the fitted values of Wy as instruments for Wyy . In the second step, we
estimate the autoregressive parameter ρ by generalized method of moments using the
residuals obtained in the first step. In the final step, we use the estimates of ρ to perform
a spatial Cochrane–Orcutt transformation of the data and obtain efficient estimates of β
and λ .
The GS2SLS regression results are reported in the second column of Table 3. The
diagnostic tests suggest that the chosen instruments are relevant (at the 5-percent level,
p = 0.020) and exogenous (at the 10-percent level, p = 0.092), and the variables (REVENUE,
SO2, URBAN, RAIN and EDUCATION) are, indeed, endogenous (at the 10-percent level,
p = 0.065). We find that the spatial lag and error parameters are statistically significant,
indicating that both types of spatial dependence coexist in the model. In particular,
based on the GS2SLS approach, we find that the spatial lag remains statistically significant,
with a similar coefficient size to that in the SAR model (without accounting for
endogeneity). This empirical finding confirms our earlier finding that cities engage in
strategic interaction in deciding their spending on environmental protection. Specifically,
a positive spillover effect emerges from environmental protection across cities. Turning
to other explanatory variables, in general we find that the empirical results from the SAR
regression are consistent with the GS2SLS specification, except that we find that education
share is positively associated with per capita environmental protection expenditure. This
result fails to support the “crowding-out” effect of fiscal expenditure, and is beyond our
expectation. A tentative explanation could be that, under a constant budget constraint,
the city government is likely to increase expenditure on both while cutting back other
types of public expenditure.
The second issue is related to the specification of a spatial weighting matrix. In a
further robustness check, we investigate whether the results of the spatial interaction
between cities are not an artifact of the statistical procedure (in which the neighborhood
variable picks up the effect of any random set of cities), and we build an intentionally
absurd weighting matrix using a randomly generated spatial weighting matrix following
Ladd (1992), Case et al. (1993) and Yu et al. (2012). We order the cities alphabetically, and
then each city’s neighbors are defined to be the other 2 cities that follow the city in the
alphabetical ordering. The Moran test (based on residuals from the OLS model) reveals
that there is no evidence of spatial dependence, and the spatially-lagged parameter
(based on estimating the SAR model) using the new spatial weighting matrix is not
statistically significant at the 10-percent level. Hence, our spatial results are not an artifact
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
115
Strategic Interaction in Spending on Environmental Protection
of the statistical procedure.
V. Further Analysis
Our finding shows that fiscal capacity matters for city governments in making decisions on
environmental protection spending under a constant budget constraint. As a city’s fiscal
revenue increases, the ability of the city government to spend on environmental protection
increases, but with a decreasing rate. It is observed that poor cities are in a worse position
than rich cities in regards to fiscal abilities, and the poor cities, some of which are severely
harmed by water or air pollution,11 are also the cities that shoulder an excessively heavy
fiscal burden. From the national point of view of reducing overall pollution, it might be
better for fiscal responsibilities to be more centralized.
Furthermore, our finding on the spatial autoregressive parameter supports the view of
greater fiscal centralization. The empirical result shows that cities are influenced by their
neighbors. Cities respond to increased environmental protection spending in neighboring
cities by decreasing their own spending. As a result, it can be expected that city governments
will free-ride in regards to spending on environmental protection. Although our study fails
to find a “race to the bottom” in environmental protection spending, studies by scholars
such as Fredriksson and Millimet (2002), Woods (2006) and Konisky (2007) find both
evidence supportive of the race to the bottom and the race to the top in environmental
regulatory decisions. If the race to the bottom-type dynamics were observed to affect the
spending behavior of Chinese cities, services relating to environmental protection for the
whole society would be expected to be severely underprovided. In this regard, it would be
better to centralize fiscal responsibilities.
Furthermore, by running a simple regression we are able to analyze in more detail
how underprovision or inefficiency of environmental protection services could occur.
Using industrial SO2 emissions as a proxy variable for local natural environmental
conditions, we examine the issue of whether a city’s environmental conditions have any
effect on the public health spending of its neighboring cities. The empirical model is
specified as:
MEDICALi = β0 + β1W*SO2i + μi,
(5)
where MEDICALi denotes public medical expenditures of city i in 2005 and SO2i stands for
11
See the 2009 Annual Report on Urban Environmental Management and Comprehensive Control of
China (Ministry of Environmenta l Protection of the People’s Republic of China) (retrieved on
15 March 2011 from http://www.zhb.gov.cn/gkml/hbb/bgth/201011/W020101108501016751709.pdf).
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
116
Huihui Deng et al. / 103–120, Vol. 20, No. 5, 2012
SO2 emissions in city i in 2005. W is the distance-based spatial weight matrix constructed
based on the longitude and latitude of each city. μi is the disturbance term. β is the coefficient
to be estimated. The estimates (standard errors in parentheses) from Equation (5) are:
M E D IC A L i =
− 5 7 7 2 4 .8 8 + 1 .2 8 W * S O 2 i , R 2 = 0 .0 5, N = 2 7 3
( − 2 .4 2 )
( 3 .6 8 )
.
The positive and statistically significant coefficient for WW implies that the pollution
of neighbors generates substantial loss for a city in terms of medical bills. In other words,
if the environmental qualities of the neighboring cities are relatively higher, the health
spending of the city can be lower, and vice versa. Thus, this result further implies that the
environmental investment of a city can impose a positive externality on its neighboring
cities through affecting its neighbor’s health expenditure.
Economic theory states that, in the presence of no externalities, marginal revenue
of the investment has to equate marginal cost associated with the investment at the
optimal level of the local government’s environmental investment (i.e. MR = MC). In a
decentralized economic system, we find that a local government’s spending on
environmental protection can affect the health spending of its neighboring governments,
yet the local government does not take this effect into account. When positive
externalities appear, the socially optimal level of investment should be set at a level
where MC social = MR local + MR neighbor . It is easily seen that MCsocial = MR local + 1.28 >
MR local = MClocal, implying that local governments will spend on environmental protection
at a level that is less than socially optimal.
VI. Concluding Remarks
China’s central government has been calling for action to abate environmental pollution;
however, environmental protection expenditure as a share of GDP remains low (within the
range of 1–1.5 percent over the past decade). 12 The present study discusses fiscal
decentralization in the context in China, where more fiscal responsibilities like environmental
protection expenditure are borne by local governments.
Using a cross-sectional dataset of 249 Chinese cities in 2005 in the spatial econometric
framework, we reach the following major conclusion: a city government tends to cut its own
12
The calculation is based on environmental protection expenditure data from the China Environmental
Statistical Yearbook and GDP data from the China Statistical Yearbook. The share was 1.2 percent on
average for the OECD countries and 1.6 percent for the USA (http://www.nationmaster.com/graph/
env_exp_pol_con_as_of_gdp-environment-expenditure-pollution-control-gdp, accessed on 5 May 2011).
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
117
Strategic Interaction in Spending on Environmental Protection
environmental protection spending as a response to a rise in the environmental protection
spending of its neighbors. This finding implies that a city government’s decision-making in
regards to environmental protection spending results in positive externalities for its
neighboring cities.
Our main finding has important implications for fiscal arrangements across
jurisdictions. The positive spillovers under a decentralized decision-making process
imply that the externalities have to be internalized to reduce efficiency loss. In other
words, one possible way to provide a socially optimal spending level of environmental
protection is to have the positive externalities internalized by an upper-tier government,
or internalized by an independent agency, like the Tennessee Valley Authority in the
USA, which is operated by a federally-owned corporation.13 In China, such a role could
be carried out by the Tai Lake Management Agency or the Yangtze River Water
Resources Committee, for example.
Our study suffers from two potential weaknesses. First, we use cross-sectional
data for Chinese cities. Given that public spending on environmental protection is
dynamic by nature, it would be preferable to have longitudinal data to explore the
dynamic interaction effects of public spending of any kind (environmental protection
in the present study) over time. Second, due to a lack of data, our analysis does not
control for “vertical fiscal interaction effects” (see Revelli, 2003; Yu et al., 2011). In the
context of China, apparently we can observe the vertical fiscal interactions among the
city government and higher government levels (say, provincial government or the
central government). For example, the following phenomenon might be observed. If a
city government increases its spending on environmental protection, its neighboring
city within the same province is observed to increase its spending on environmental
protection as well. In fact, the rise in environmental protection spending in both cities
can be a result of the provincial government mandating them to do so. If this were the
case, a misleading conclusion could be reached that a spatial interaction exists between
these 2 cities while in reality it does not. Therefore, our results should be interpreted
with caution.
This study is a first attempt to introduce spatial econometric techniques to explore
empirically Chinese local governments’ expenditure behavior in regards to
environmental protection. Although the findings and implications originated from
this empirical study are shown to be important, the two aforementioned issues merit
13
By “independent” we mean that the agency will not be affiliated with any city government or provincial
government.
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
118
Huihui Deng et al. / 103–120, Vol. 20, No. 5, 2012
additional research in future, when data become available, to explore the true spatial
effects of local governments’ spending on environmental protection to inform
appropriate public policy.
References
Anselin, Luc, 1988, Spatial Econometrics: Methods and Models, Norwell, MA: Kluwer Academic
Publishers.
Anselin, Luc and Anil Bera, 1998, “Spatial dependence in linear regression models with an introduction
to spatial econometrics,” in Aman Ullha and David Giles, eds, Handbook of Applied Economics
Statistics, New York: Maccel Dekker, pp. 237–89.
Baicker, Katherine, 2005, “The spillover effects of state spending,” Journal of Public Economics,
Vol. 89, No. 2–3, pp. 529–44.
Brueckner, Jan, 2003, “Strategic interaction among governments: An overview of empirical studies,”
International Regional Science Review, Vol. 26, No. 2, pp.175–88.
Caldeira, Emilie, 2012, Yardstick Competition in a Federation: Theory and Evidence from China
[online; cited May 2012]. Available from: http://dx.doi.org/10.1016/j.chieco.2012.04.011.
Case, Anne, James Hines and Harvey Rosen, 1993, “Budget spillovers and fiscal policy
interdependence,” Journal of Public Economics, Vol. 52, No. 3, pp. 285–307.
Chen, Ye, Hongbin Li and Li-An Zhou, 2005, “Relative performance evaluation and the turnover of
provincial leaders in China,” Economics Letters, Vol. 88, No. 3, pp. 421–25.
Chen, Zunxuan, 2005, “Tackling China’s water pollution problem: A legal and institutional perspective
from Taihu Lake water pollution control,” Temple Journal of Science, Technology &
Environmental Law, Vol. 24, No. 1, pp. 325–50.
Cliff, Andrew and John Ord, 1981, Spatial Processes: Models and Applications, London: Pion Ltd.
Das, Debabrata, Harry Kelejian and Ingmar Prucha, 2003, “Finite sample properties of estimators of
spatial autoregressive models with autoregressive disturbances,” Papers in Regional Science,
Vol. 82, No. 1, pp. 1–26.
Economy, Elizabeth, 2007, The Great Leap Backwards? [online; cited March 2011]. Available from:
htp://www.foreignaffairs.org/20070901faessay86503/elizabeth-c-economy/the-greatleapbackward. html.
Ermini, Barbara and Raffaella Santolini, 2010, “Local expenditure interaction in Italian municipalities:
Do local council partnerships make a difference?” Local Government Studies, Vol. 36, No. 5,
pp. 655–77.
Esty, Daniel and Damien Geradin, 1997, “Environmental protection and international competitiveness:
A conceptual framework,” Journal of World Trade, Vol. 32, No. 3, pp. 5–46.
Feruglio, Nicoletta and Dallas Anderson, 2008, Overview of Fiscal Decentralization [online; cited
March 2011]. Available from: http://siteresources.worldbank.org/PSGLP/Resources/
FeruglioandAndersonoverviewofFiscaldecentralization.pdf.
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
119
Strategic Interaction in Spending on Environmental Protection
Fredriksson, Per and Daniel Millimet, 2002, “Strategic interaction and the determination of
environmental policy across U.S. states,” Journal of Urban Economics, Vol. 51, No. 1,
pp. 101–22.
Fréret , Sandy, 2006, “Spatial analysis of horizontal fiscal interactions on local public expenditures:
The French case,” Paper presented at the International Workshop on Spatial Econometrics and
Statistics, Rome, Italy, May 2006.
Glazer, Amihai, 1999, “Local regulation may be excessively stringent,” Regional Science and Urban
Economics, Vol. 29, No. 5, pp. 553–58.
Kelejian, Harry and Ingmar Prucha, 1998, “A generalized spatial two-stage least squares procedure
for estimating a spatial autoregressive model with autoregressive disturbances,” Journal of
Real Estate Finance and Economics, Vol. 17, No. 1, pp. 99–121.
Konisky, David, 2007, “Regulatory competition and environmental enforcement: Is there a race to
the bottom?” American Journal of Political Science, Vol. 51, No. 4, pp. 853–72.
Ladd, Helen, 1992, “Mimicking of local tax burdens among neighboring counties,” Public Finance
Quarterly, Vol. 20, No. 4, pp. 450–67.
LeSage, James and Kelley Pace, 2009, Introduction to Spatial Econometrics, Boca Raton, FL: CRC
Press.
Levinson, Arik, 1997, “A note on environmental federalism: Interpreting some contradictory results,”
Journal of Environmental Economics and Management, Vol. 33, No. 3, pp. 359–66.
List, John and Shelby Gerking, 2000, “Regulatory federalism and environmental protection in the
states,” Journal of Regional Science, Vol. 40, No. 3, pp. 453–71.
Lundberg, Johan, 2006, “A spatial interaction model of benefit spillovers from locally provided
public services,” Regional Studies, Vol. 40, No. 6, pp. 1–14.
Manski, Charles, 1993, “Identification of endogenous social effects: The reflection problem,” Review
of Economics Studies, Vol. 60, No. 1, pp. 531–42.
Markusen, James, Edward Morey and Nancy Olewiler, 1995, “Environmental policy when market
structure and plant locations are endogenous,” Journal of Environmental Economics and
Management, Vol. 24, No. 1, pp. 69–86.
Millimet, Daniel, 2003, “Assessing the empirical impact of environmental federalism,” Journal of
Regional Science, Vol. 43, No. 4, pp. 711–33.
Moran, Patrick, 1950, “Notes on continuous stochastic phenomena,” Biometrika, Vol. 37, No. 1,
pp. 17–33.
NBS (National Bureau of Statistics of China), 2006, China City Statistical Yearbook 2006, Beijing:
China Statistics Press.
NBS (National Bureau of Statistics of China), 2009, China Environmental Statistical Yearbook 2009,
Beijing: China Statistics Press.
Revelli, Federico, 2003, “Reaction or interaction? Spatial process identification in multi-tiered
government structures,” Journal of Urban Economics, Vol. 53, No. 1, pp. 29–53.
Solé Ollé , Albert, 2003, “Electoral accountability and tax mimicking: The effects of electoral margins,
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences
120
Huihui Deng et al. / 103–120, Vol. 20, No. 5, 2012
coalition government, and ideology,” European Journal of Political Economy, Vol. 19, No. 4,
pp. 685–713.
Vennemo, Haakon, Kristin Aunan, Henrik Lindhjem and Hans Martin, 2009, “Environmental pollution
in China: Status and trends,” Review of Environmental Economics and Policy, Vol. 3, No. 2,
pp. 209–30.
Wagner, Adolph, 1911, “Staat in nationalökonomischer hinsicht (The state in economic perspective),”
in Conrad J., L. Elster and W. Lexis, eds, Handwörterbuch der Staatswissenschaften (Concise
Dictionary of Political Science), Jena: Lexis, pp. 727–39 (in German).
Wilson, John, 1996, “Capital mobility and environmental standards: Is there a race to the bottom?”
In Jagdish Bhagwati and Robert Hudec, eds, Harmonization and Fair Trade, MA: MIT Press,
pp. 395–427.
Woods, Neal, 2006, “Interstate competition and environmental regulation: A test of the race to the
bottom thesis,” Social Science Quarterly, Vol. 87, No. 1, pp. 174–89.
World Bank, 2007a, Cost of Pollution in China: Economic Estimates of Physical Damages, World
Bank: Washington, DC.
World Bank, 2007b, World Development Indicators 2007, World Bank: Washington, DC.
Yu, Yihua, Li Zhang, Fanghua Li and Xinye Zheng, 2011, “On the determinants of public infrastructure
spending in Chinese cities: A spatial econometric perspective,” Social Science Journal, Vol. 48,
No. 3, pp. 458–67.
Yu, Yihua, Li Zhang, Fanghua Li and Xinye Zheng, 2012, Strategic Interaction and the Determinants
of Public Health Expenditures in China: A Spatial Panel Perspective [online; cited March
2012]. Available from: http://dx.doi.org/10.1007/s00168-011-0488-7.
Zhang, Li and Xinye Zheng, 2009, “Budget structure and pollution control: A cross-country analysis
and implications for China,” China and World Economy, Vol. 17, No. 4, pp. 88–103.
(Edited by Jing Qiu)
©2012 The Authors
China & World Economy ©2012 Institute of World Economics and Politics, Chinese Academy of Social Sciences