Does corruption matter for the environment? Panel evidence from China

This article examines the income-energy-SO2 emissions nexus by taking a corruption variable into account. To that end, the panel cointegration methods are applied to 29 Chinese provinces over 1999–2012. The authors ́ empirical evidence shows that an increase in the number of anti-corruption cases tends to drive down SO2 emissions in China. It is also found that income growth appears to have a beneficial effect on decreasing SO2 emissions over the past two decades. Finally, energy consumption is found to increase SO2 emissions. JEL C23 Q56


Introduction
China has achieved rapid economic growth at an average rate of almost 10 percent annually over the past three decades. This economic success, however, comes at the cost of deterioration of the environment. One of the most severe environmental problems that China is currently facing are air pollution. For example, the State Environmental Protection Administration of China (SEPA) reports that about 70% of the 300 cities in China fail to meet the air quality standards set by the World Health Organization (WHO) and seven out of the ten most polluted cities in the world are located in China. The World Bank estimates that the direct cost of air pollution -such as acid-rain damage to crops, medical bills and job-loss from illness -ranges between 8 percent and 12 percent of China's GDP annually. In addition, it is estimated that, because of heavy air pollution, more than three million people die prematurely each year and the average life expectancy is more than 5 years lower for residents in northern China than those living in the south (Wang, 2007;Pope III and Dockery, 2013).
The Chinese government has made substantial efforts to reduce air pollution by introducing various emission reduction measures such as environmental taxes/charges, pollution treatment programs and even closure of inefficient power/industrial facilities.
Moreover, under the 12th Five-Year Plan (2011)(2012)(2013)(2014)(2015), China has paid considerable attention to energy and climate change issues and has established a new set of targets and policies for the plan period. The main targets include a 16 percent reduction in energy intensity (energy consumption per unit of GDP), an increase in non-fossil energy up to 11.4 percent of total energy consumption and a 17 percent reduction in carbon intensity (carbon emissions per unit of GDP). In 2015, however, China still recorded the world's largest increment in energy consumption for the thirteenth consecutive year and became the world's largest emitter of both carbon dioxide (CO 2 ) and sulfur dioxide (SO 2 ) emissions. Therefore, a fundamental question would be certain to arise regarding China's environment: what are the main determinants affecting air pollution in China?
A number of studies have sought to isolate the independent effects of various factors on air pollution in China. Traditional specification of this subject includes a growth variable (i.e., income per capita) and investigates the environmental Kuznets curve (EKC) -an inverted Ushaped relationship between income per capita and certain types of pollutants (typically measured by CO 2 emissions). Then, as we glance through the literature more, we come across empirical studies that claim that energy consumption could be an important determinant of environmental outcomes and analyze the so-called income-energy-environment relationship.
Examples include, but are not limited to, Song et al. (2008), Jalil and Mahmud (2009), Jalil and Feridun (2011), Wang et al. (2011, Govindaraju and Tang (2013), Michieka (2014), Qu and Yan (2014), Yuan et al. (2015), Wang et al. (2016) and Li et al. (2016). The findings from these studies generally show that there is the ambiguous evidence in favor of the EKC for China, and strong evidence that China's growth in energy consumption indeed causes environmental degradation. Woods (2008). However, the existing literature does not directly address the issue in China.
The main objective of this paper is to take a measure of corruption into account in a model when examining the income-energy-environment relationship in China. Although China is currently the world's largest SO 2 emitting country along with CO 2 emissions, empirical studies have paid little attention to SO 2 emissions in their analyses. 1 Empirical focus is thus on assessing the effects of corruption, income and energy consumption on SO 2 emissions using panel data of 29 provinces in China from 1991 to 2012. To that end, the panel cointegration methods are utilized. This paper is organized as follows: in Section II, we outline the empirical model to be estimated and the data used for the estimation. In Sections III and IV our empirical procedures 1  is perhaps the only study addressing the issue; they find that growth has a beneficial effect on reducing SO 2 emissions in China. However, they only examine the income-environment nexus. and major findings are discussed, respectively. Finally, section V makes some concluding remarks.

The model to be estimated
In the empirical model adopted here, we extend the so-called standard model of the income-energy-environment nexus to include a measure of corruption. Letting i denote the crosssectional unit (Chinese provinces in this paper) and t the time period, we can write a model in a log-linear form as: where it so ) ( 2 is the sulfur dioxide (SO 2 ) emissions for province i in China; it y is the 2000 real income for province i; 2 it y is the square of the 2000 real income for province i; it ec is the energy consumption for province i; it cor is a measure of corruption for province i and is the number of anti-corruption cases; and ε it is the error term. The variables are measured on a per capita basis.
We are particularly interested in the parameter β 4 -that is, the ceteris paribus effect of corruption on SO 2 emissions.
Given that numerous studies commonly show the crucial role of income plays in influencing environmental outcomes, it would be proper to directly test the Environmental Kuznets Curve (EKC) hypothesis into our modeling. In Eq. (1), to the extent that β 1 >0 and β 2 <0, the EKC hypothesis is predicted to hold; that is, income has a diminishing effect on SO 2 emissions after the turning point (or maximum point of the income), achieving a parabolic shape.
It is expected that β 3 >0 due to the fact that an increase in energy consumption mainly driven by growth is likely to push SO 2 emissions up. Finally, it is expected that β 4 <0 because the increasing number of anti-corruption cases is likely to result in improved environmental regulations, thereby reducing SO 2 emissions. Our (balanced) panel dataset contains the 29 Chinese provinces from 1999 to 2012 2 Some scholars (e.g., Damania et al., 2003;Cole, 2007) use governmental honesty taken from the International Country Risk Guide (ICGR) as a proxy for corruption in their models. At the sub-national level, however, the data are not available.

Data
(N*T=406 observations, where N=29 provinces and T=14 years). This time period is chosen by availability of the data for all the variables. All variables are converted into natural logarithms.

Empirical Results
The first requirement for estimating our model in Eq.
(1) using the panel cointegration method is that the variables must be nonstationary I(1) series. Accordingly, the panel cointegration modeling normally starts with testing whether a panel series follows a unit root.
However, the possibility of cross-sectional dependence in panels is likely to invalidate the test statistics of conventional panel unit root tests such as the LLC (Levin et al., 2002) and IPS tests (Im et al., 2003). These tests commonly assume the cross-sectional independence in panels.
Before applying a unit root test, therefore, we must test whether a panel series is crosssectionally independent. A cross-sectional dependence (CD) test of Pesaran (2004) can be used to achieve this goal. The results show that the null hypothesis of no cross-sectional dependence can be strongly rejected (the p-values for all five variables are zero to two decimal places), providing compelling evidence of the cross-sectional dependence in the sample ( processes. When estimating a nonstationary panel model, there is serious concern about spurious regression. In one important case, a regression estimating nonstationary I(1) series is not spurious, and that is when the series are cointegrated. Hence, the presence of cointegration relationship among the variables is tested using the various tests developed by Pedroni (1999 and and Kao (1999). The results show that the null hypothesis of no cointegration can be rejected even at the 1% level of significance for all five tests, evidence that SO 2 emissions and its determinants have a long-run relationship (Table 3). In other words, whenever deviations from the long-run equilibrium take place, they would be transient: there are economic forces that drive SO 2 and its main factors back to restore the long-run equilibrium relationship.
Having learned about a potential long-run relationship among the five series, we now apply the FMOLS and DOLS panel estimators of Mark and Sul (2003) and Kao and Chiang (2000) to Eq. (1) in order to estimate the long-run parameters. We also report the estimated effects of the fixed effects estimator here for comparison. Table 4 reports the long-run effects for all independent variables and for each of the three estimated models. The estimates generated by the models seem remarkably consistent. The signs of the coefficients are the same across models, and the same variables are generally statistically significant in each model.

Discussion
The coefficients on the income and the quadratic term are similar across models. Because the coefficient on it y is always positive and the coefficient on 2 it y is always negative, this equation literally implies that, at low value of income, an additional rise in income tends to increase SO 2 emissions. At some point, the effect becomes negative, and the quadratic shape means that the elasticity of SO 2 emissions with respect to income is decreasing as income increases. In other words, the finding seems to be supportive of the EKC hypothesis for Chinese SO 2 emissions. It turns out, however, that all of the 29 provinces in the sample have more than the calculated turnaround values of income, and so the part of the curve to the left can be ignored. Thus, SO 2 emissions in fact have monotonically fallen with income growth in China over the past decade.
From policy perspectives, this result can be interpreted that the Chinese government's policies targeted to reducing air pollution are likely to work effectively without costing economic growth.
The partial effect of energy consumption on SO 2 emissions is always positive, and the magnitudes of the coefficient estimates are very similar across all three models. In the FMOLS model, for example, a one percent increase in energy consumption is estimated to increase SO 2 emissions by about 0.80% in China. Given the fact that growth largely leads to an increase in energy use, this finding suggests that any favorable growth effect on air pollution could be offset by a detrimental energy consumption impact.
The key policy variable, it cor , seems to have the desired effect. The estimated coefficient is negative for all three models. For example, the FMOLS coefficient (-0.17 show strong bidirectional causation for 4 cases and unidirectional causation for 5 cases (Table 5).
For example, the relationships between SO 2 emissions and energy consumption, and SO 2 emissions and corruption are characterized by bidirectional causality. This means that SO 2 emissions are significantly affected by changes in energy consumption (corruption) and energy consumption (corruption) is also influenced by changes in SO 2 emissions. On the other hand, there is unidirectional causality running from income to SO 2 emissions. This suggests that SO 2 emissions are significantly affected by changes in income, while income is not affected by 11 changes in SO 2 emissions. Together, these findings provide evidence that all independent variables can be used to forecast future SO 2 emissions and justify the use of our model in Eq. (1).

Concluding Remarks
Although corruption can have an effect on the environment in China, no study has directly addressed this issue empirically. In this short article, therefore, we take into a corruption variable into account in a dynamic panel model when estimating the income-energy-SO 2 emissions nexus. Our results show that anti-corruption cases seem to have a beneficial effect on reducing SO 2 emissions in China. Other findings shows that income growth tends to lower SO 2 emissions, while energy consumption increases SO 2 emissions.  -2.14 -3.17** -3.11** -3.84** Notes: CADF and CIPS represent cross-sectionally augmented Dickey-Fuller and crosssectionally Im-Pesaran-Shin tests, respectively. ** denotes rejection of null hypothesis at the 1% level.   One-way causality from lny 2 to lncor lny 2 → lncor 92.69*** 0.00 lnec → lncor 94.19*** 0.00 One-way causality from lnec to lncor lncor → lnec 66.27 0.21 Notes: ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively.