Impact of SO2 Emission on the Gross Domestic Product Growth of China

Received: January 5, 2020
Revised: March 8, 2020
Accepted: March 9, 2020
Download Citation: RIS|BibTeX|https://doi.org/10.4209/aaqr.2020.01.0018  

Cite this article:

Xu, J. and Yang, Y. (2020). Impact of SO2 Emission on the Gross Domestic Product Growth of China. Aerosol Air Qual. Res. 20: 787-799. https://doi.org/10.4209/aaqr.2020.01.0018

HIGHLIGHTS

• A panel VAR model is proposed to gauge the pollution and health costs in China.
• The pollution and health costs are calculated as a percentage of GDP.
• The results show large heterogeneities among the east, central and west China.

Keywords: Structural panel VAR; Pollution costs; Health costs.

INTRODUCTION

China’s industrialization initiated in the early 1980s has brought economic prosperity over the past decades (Zhang et al., 2016). Meanwhile, a side effect has loomed out of the rapid growth: The air pollution has become ever more severe, which has attracted considerable attention from academics and government agencies across the world (Banister, 1998; Cai et al., 2018; Wang et al., 2018; Alonso-Carrera et al., 2019; Lin et al., 2019; Liu et al., 2019). There is little disagreement that air pollution, as a factor to both respiratory and cardiovascular diseases, poses a major environmental risk to human health (Lee et al., 2014; Abe et al., 2018; Guttikunda and Jawahar, 2018; Hung et al., 2018; Widiana et al., 2019; Chen et al., 2020). For example, Chay and Greenstone (2003) find a strong correlation between air pollution and mortality, where respiratory-related incidents usually increase as air quality deteriorates. Despite its direct effect on health, air pollution also exerts an indirect impact on economic development. As Bloom et al. (2004) point out, pollution leads to inflating medical costs, therefore reduces the capital available to production.

These findings prompt us to investigate the possible interactions among economic growth, industrialization, air pollution, and health in China. In particular, this paper tries to answer the question: Given these interactions, how high price has China paid for its economic prosperity?

To achieve that, this paper applies a structural panel vector autoregression (PVAR) model to measure the magnitudes of the costs. In contrast to cross-section, time series and single-equation panel data models, PVAR approach not only allows us to solve the endogenous problem, but also controls for the invisible time-invariant individual features. In doing so, this paper relates to a rapidly growing literature on the interaction between economic development, air pollution and health. The increase in energy consumption has boosted economic growth, but also increased pollution. This fact can be seen in Fig. 1 which shows four different energy consumption in China. Coal still contributes to over 60% of total energy consumption from 1978 to 2018, and this trend is probably to be persistent in the future (Liu et al., 2016). 

Fig. 1. Energy consumption. Source: National Bureau of Statistics of China.

Accompanying the heavy consumption of coal, SO₂ emissions are inevitably correlated with coal consumption. Fig. 2 depict the unconditional correlation between coal consumption and SO₂ emissions at the national level as well as the separated area levels from 2000 to 2015. Along with the emissions of SO₂, mortality may also rise. From Fig. 2, the three areas display similar pattern in coal emissions nexus, but this result masks considerable heterogeneity. For instance, the degree of industrialization and intensity of environmental regulation in east and west are bound to be heterogeneous. Therefore, knowledge of the relationship between growth, industrialization, SO₂ emissions and mortality is of considerable interest. Figs. 3(a) and 3(b) report the relationship between mortality and SO₂ emissions for the whole country and central China, respectively. Clearly, these two variables are positively correlated in central China. 

Fig. 2. Coal consumption and SO₂ emissions for the whole country (2000–2015). Source: National Bureau of Statistics of China.

Fig. 3(a). Mortality and SO₂ emissions for the whole country (2000–2015). Source: National Bureau of Statistics of China.

Fig. 3(b). Mortality and SO₂ emissions for central China (2000–2015). Source: National Bureau of Statistics of China.

Figs. 3(a) and 3(b) depicts the relationship between mortality and SO₂ emissions. A large number of epidemiological studies show that air pollution is harmful to human health (Lee et al., 2011; Abe et al., 2018; Guttikunda and Jawahar, 2018; Hung et al., 2018), such as the occurrence of respiratory diseases directly caused by air pollution (Chen et al., 2017; Abdolahnejad et al., 2018). SO₂ is a common irritant that may cause diseases like bronchitis and bronchial asthma, leading to health damage (Hansell et al., 2011; Cerón-Bretón et al., 2018). SO₂ emissions, accompanying the increasing coal consumption, adversely affect health and enhance mortality. This can be reflected in Figs. 3(a) and 3(b), especially in central China, where SO₂ emissions increase and mortality increases.

Historically, many studies analyze the impact of air pollution on health. For example, Pope et al. (2002) find that SO₂ is significantly associated with mortality. Luechinger (2014) finds an elasticity of 0.07–0.13. Maji et al. (2017) show that environmental pollution in Agra is becoming increasingly serious and the burden of health economic costs is increasing. Abe et al. (2018) find environmental pollution (PM₁₀) has a significant impact on mortality burden of cardiovascular and respiratory diseases. Zhou et al. (2020) find that, compared with PM_1-2.1 and PM_2.1-9.0 particles, PM_1.1 particles cause more serious harm to human health in China. Kao et al. (2020) show that PM_2.5 affects the neurodevelopment of infants in southern Taiwan. Vormittag et al. (2018) show that improving air quality can reduce government public health fiscal expenditure and save economic costs.

There are very informative literatures have analyzed the relationship between SO₂ pollution and mortality, but no conclusion has been reached. Some studies find positive effects (Joyce et al., 1989; Bobak and Leon, 1992; Shinkura et al., 1999; Bobak and Leon, 1999). The chronic effects from prolonged exposure are generally not significant (Chinn et al., 1981; Woodruff et al., 2008).

There are also many reviews analyzing the correlation between development and pollution. The inverted U-shaped relationship between environmental pollution and income per capita - now what we call the environmental Kuznets curve (EKC) is hotly discussed and empirically tested. The early studies are found to support the EKC hypothesis. While later, among various studies, such as Wagner (2015), Apergis (2016), and many others, reach mixed conclusions. The work of Shafik (1994) even indicate that rising income levels will increase pollutant emissions. Lee and Rosalez (2017) indicate that a high-energy price can improve the energy structure by inciting energy efficiency use and result in decoupling CO₂ emissions from economic growth. Cheng et al. (2019) show that the potential sources of CO₂ and SO₂ are similar, mainly distributed in poor countries or regions, such as India and Pakistan, and Inner Mongolia and Gansu and Guizhou Provinces in China.

An assessment of the existing literature suggests that most studies focus either on the nexus of development-pollution or pollution-health where little effort has been made to test these links under the same framework. This study is an attempt to fill the gap. China is the largest developing country in the world, and it has experienced both high growth, severe pollutant emissions and health costs in the past decades. Hence, this paper proposes a structural PVAR model to gauge the magnitudes of the pollution and health costs by examining the dynamic relationships between economic development, pollution and health.

This paper formulates and estimates a structural PVAR model using the generalized method of moments (GMM) by treating economic development, industrial structure, SO₂ emissions, and mortality as endogenous variables. This paper proposes a novel method to approximate the pollution and health costs, and the paper is arranged as follows. First of all, the basic structural PVAR framework including model specification and identification is introduced. And then, the data and empirical results are reported. Finally, the conclusion is summarized.

ECONOMETRIC METHODOLOGY

Model Specification

To analyze how economic growth and pollution affect health, Ruhm (2000), and Chay and Greenstone (2003) consider a fixed-effect panel data model:

where H_it is a proxy for health of individual i in period t, such as life expectancy or mortality; E_it is economic development, such as GDP per capita; P_it is a proxy for pollution, captured by the emissions of a specific pollutant; α_i is the individual fixed effect, δ_t is the time fixed effect, and ε_it denotes the idiosyncratic disturbances.

Eq. (1) is a static model, which fails to capture the simultaneous and dynamic effects of these variables. Coondoo and Dinda (2002) find that in Asian countries, pollutant emissions and income affect each other. Likewise, Xu et al. (1994) find that lagged deaths can explain current mortality in Beijing, China. As a result, when accounting for the pollution or health costs incurred by economic development, a static model like Eq. (1) will give rise to biased estimates.

To remedy this, this paper extends Eq. (1) to a structural PVAR model: 

where the matrix A captures the contemporaneous relationship between the variables;

Multiply both sides of Eq. (2) by A^–1, then:  

In matrix form, Eq. (3) can be rewritten as:

For the definitions of the variables, please refer to Appendix A.

To analyze the impulse-response functions (IRFs), transform the VAR(k) process into a structural vector moving average (VMA) model with infinite lags: 

In compact form:

Identification

Rewrite the error term in Eq. (3) as:

In matrix form:

Here, B = (B_ab), a, b = 1, 2, 3, 4 is a 4 × 4 matrix with 16 unknowns, and B = A^–1. Since the covariance matrix Ω is symmetric, BBʹ = Ω induces only 10 equations about B_ab; in other words, 6 additional restrictions are needed to identify B.

To identify the shocks, this paper employs the Cholesky decomposition to ensure identification, namely, imposing a recursive structure on the matrix B. The basic rationales of the recursive identification scheme are that, first of all, the speed of the dynamic adjustment of industrial structure is slow, so that it reacts to economic development, SO₂ emissions and mortality shocks with a lag, hence setting B₁₂ = 0, B₁₃ = 0, and B₁₄ = 0. Secondly, Bovenberg and Smulders (1996) argue that pollution does not increase in proportion to economic growth. Meanwhile, the direction of causality goes from pollution to mortality, rather than the opposite. These facts imply that economic development and mortality have subtle contemporaneous effects on pollution, so sets B₂₃ = 0 and B₂₄ = 0. Lastly, Bloom et al. (2004) show that improvements in health may increase labor productivity and the accumulation of capital, and then increase the output, implying that the effect of mortality on development may not be instantaneous. Therefore, B₃₄ = 0.

In sum, this paper postulates that (i) development, SO₂ emissions and mortality shocks have no instantaneous effect on industrial structure, that is, B₁₂ = 0, B₁₃ = 0, and B₁₄ = 0; (ii) development and mortality shocks have no instantaneous effects on SO₂ emissions, namely, B₂₃ = 0 and B₂₄ = 0; (ii) mortality shocks show no instantaneous impact on development, B₃₄ = 0. This leads to a lower triangular matrix B which justifies Cholesky decomposition:

Bootstrapping Standard Errors

Due to the correlation between the lagged dependent variables in the panel VAR system, the standard errors of the estimated coefficients are not efficient. Therefore, this paper uses the bootstrap method to obtain the correct standard errors.

Denote p the lag order of the panel VAR model, T the total time periods, and

EMPIRICAL RESULTS

Data

Our original data covers 120 cites in China across the period 2000–2015. The variables used are collected from the National Bureau of Statistics, which is accessed via Qianzhan Database. 11 cities whose economic scale is too small are dropped; that is, each has a yearly average GDP less than 100 billion yuan. These cities are Yangquan, Zhangjiajie, Sanya, Haikou, Beihai, Panzhihua, Lhasa, Tongchuan, Shizuishan, Jinchang, Karamay. Furthermore, deleting Nanchong, as it is rich in oil and natural gas, violating our assumption that a city relies on coal as its energy source. These deletions result in 108 cities in our final sample.

To control the city heterogeneity, following the 7^th Five-Year Plan initiated in 1986, dividing the sample into three parts, east, central and west, and label them as Group 1, 2, and 3, respectively. This classification was based on geography, but later approved by the government (Eastern China: Beijing, Tianjin, Hebei, Liaoning, Shanghai, Jiangsu, Zhejiang, Fujian, Shandong, Shandong, Guangdong, and Hainan. Central China: Shanxi, Inner Mongolia, Jilin, Heilongjiang, Anhui, Jiangxi, Henan, Hunan, and Guangxi. Western China: Sichuan (and the later Chongqing), Guizhou, Yunnan, Tibet, Shaanxi, Gansu, Qinghai, Ningxia, and Xinjiang).

Note that in terms of the volume of SO₂ emissions, cities in Hebei and Shandong are akin to the central provinces. Accordingly, the cities in these two provinces are grouped into the central group (Group 2). The final lists of cities and their geographical distribution are shown in Appendix B. In total, there are 30 eastern cities, 54 central cities, and 24 western cities.

Following previous empirical related studies, the following data are collected:

1. Industrial structure, S_it: In most of the literature, industrial structure is often measured by the proportion of the secondary industry in the GDP. China is experiencing rapid industrialization and urbanization to date, and the secondary industry is still the foundation of its economy. Meanwhile, the secondary industry consumes more than 70% of total energy, and it is the largest contributor to GDP. Hence, define S_it as the proportion of the secondary industry output in GDP.
2. Pollution, P_it: Chen and Xu (2010) and Tanaka (2015) show that three quarters of China’s aggregate power are coal; hence, SO₂ is directly linked to coal combustion. According to National Bureau of Statics, SO₂ emissions are categorized by its sources, which include those from industry, agriculture, urban living, automobile, and pollution abatement. However, only SO₂ from industry can be traced back to 1994. Consequently, P_it is defined as the log of industrial SO₂ emissions (originally in 10,000 tons) per capita.
3. Economic development (GDP), E_it: Following most of the literature, the real GDP in 2000 is used as an endogenous variable, and it is adjusted by the population of city i and taken logs.
4. Health, H_it: The log of mortality per 100,000 persons is used to proxy health. In the literature, many authors use infant mortality to proxy for health, but due to the availability of data, here the overall mortality is used.

Empirical Evidence

The first step describes the dataset employed, the presence of panel unit root using LLC test is conducted. The results are shown in Table 1. The nulls of panel unit root test are all highly rejected, indicating the variables are stable.

Next, employing the model selection criteria proposed by Andrews and Lu (2001). The results are shown in Table 2. Based on results of MMSC-AIC, MMSC-BIC, MMSC-HQIC, an order of one are appropriate for all the models.

Long-run Elasticity

As indicated above, the lag order of our PVAR model is one. The standard errors of the coefficients are obtained by bootstrap with 500 iterations, and the algorithm is detailed in Section Bootstrapping Standard Errors. 

Because development, SO₂ emissions and mortality are measured in logs, the coefficients in the VMA can be interpreted as short-run elasticity. Likewise, the sum of the VMA coefficients yields the long-run elasticity. For example, the development equation E_it can be approximated as:

Note that 

Referring to the third row and the second column (GDP to SO₂) of Table 3, the income elasticity of pollution is negative across all regions: –1.15, –0.28 and –0.27 for eastern, central and western cities, respectively. This suggests that China may already be on the second stage of the inverted U-shape EKC, where SO₂ emissions reduce as the income goes across a high threshold value.

From Column 3 in Table 3, 1% increases in SO₂ emissions raises GDP per capita by –0.06 in the east, 0.92 in the central cities, and 0.20 in the west. This result indicates that central and western China is still at a stage that pollution can promote economic growth, but in the east, it seems that pollution already begins to hinder economic growth.

This result needs some scrutiny. In 1999, the State Council of China launched the Western Development Strategy, by which 11 policy-targeted provinces experienced quick growth (Lu and Deng, 2011). However, central and western regions are historically endowed with low physical capital stock. In fact, in the 1990s, more than 50% of the total investment took place in eastern China, 28% in central and only 21% in western China. Meanwhile, the infrastructure in central and west is a heritage of the planned economic system, where efficiency and profits are not of top priority. Thirdly, China’s development is compatible with the flying-geese model; that is, low-profit, high-energy-consuming, and labor-intensive industries tend to migrate from east to west. These factors force the central and western regions to adopt capital-saving technology that relied heavily on natural resources. Thus, pollution still plays a role in development.

Column 3 of Table 3 shows that 1% increases in mortality corresponds to a 0.4, 0.53 and 0.18 increase in income of eastern, central, and western China, separately. This implies that the health loss may play a positive role in economic growth.

Pollution and Health Costs

To gauge the pollution costs, first define a concept called green GDP. In economic sense, it is the GDP that eliminates the marginal contributions of the environmental (pollution) inputs. So, to quantify how “green” the economy is, the costs of being “non-green,” the pollution costs, which are measured as a percentage of GDP need to be examined. Since 

Similarly, calculate a healthy GDP by eliminating the health loss from GDP. Since 

The definitions above avoid assigning subjective prices to the calculations of pollution and health costs, implying that the results here may be more insightful. The results for pollution and health costs are shown in Tables 4 and 5, respectively.

From Table 4, the pollution costs of the eastern and central cities account for –0.3% and 26% of GDP, respectively. The small and negative costs in the east may reflect that the east is slowly transiting to a modern service economy. The high costs for central China may be that cities there still tend to adopt technologies that are more detrimental to environment. Meanwhile, pollution cost of west is about 5%; it is comparable to the annual economic growth rate, showing that the west is still undergoing an underdevelopment process. These results are in line with the long-term elasticity presented in the second row of Table 3 (SO₂ emissions to GDP), where the elasticity is positive for central and western China, but negative for eastern China. Table 4 reflects the pollution costs in eastern, central and western China. For clarification, we average the pollution costs of cities in the same province as the representative of the province, and then use a map to show it. Since the differences among 2007, 2011, and 2015 are not significant, we take 2015 as an example. Fig. 4 shows the pollution costs of 2015 in a much more visual way. It indicates that central and western China are still at a stage that pollution can promote growth and pollution costs are higher in central China, but in the east, pollution is small and negative, which means that it starts to hinder growth. 

Fig. 4. Pollution costs (2015).

The main conclusions shown from Table 3 and 4 are as follows:

First, SO₂ emissions have promoted GDP growth. These promotions are more pronounced in central and western China, probably because central China has a higher degree of industrialization, mainly the secondary industry, and more resources and environment investment, so it has a greater effect on promoting GDP.

Second, the pollution costs of eastern, central and western China have large heterogeneities. The cost of pollution is highest in central China, reaching more than 20%, which implies that more than one-fifth of the GDP is contributed by environmental pollution. From the perspective of green GDP, its actual amount is only about four-fifths of the original GDP. In western China, pollution cost is about 5%, and the green GDP is about 95% of the original GDP. In eastern China, pollution cost is negative, which means that environmental pollution has a negative effect on economic growth, confirming that the cleaner the production methods, the higher the GDP.

Turning to the health costs in Table 5, the specific numbers of the costs are about 2% for east, 3% for central region, and 1% for west, respectively. Obviously, east and central cities bear higher costs than the west, of which central region undertakes the highest. The reason may be that central region is undergoing industrialization, while east begins to go into the stage of post-industrialization. Both these processes need large physical capital stock and human capital, so that people there bear higher health costs. As regards to west, this region may still stay at the dawn of take-off stage in Rostow’s sense. Due to the lake of capital stock and advanced human resources, the health costs are relatively small. This finding is also consistent with the results in Table 3. 

Impulse Response Functions

Fig. 5 shows the IRFs of SO₂ emissions to shocks in industrial structure, pollution, development and mortality. When examining the impulse of development on SO₂ emissions, a negative reaction of SO₂ emissions to shocks in income is observed in all the three regions. This shows that the Chinese economy seems to be on the second stage of the EKC. This result reconciles with our long-run elasticity analysis in Section Empirical Evidence. Furthermore, we found that the largest effect of development on pollution occurs in eastern China. A possible reason is that people there are more conscious about pollution, and prone to put pressures on the local government when environment deteriorates. Western region, which is lagged in development, is less sensitive to the environmental issues. 

Fig. 5. IRFs of SO₂ emissions in eastern, (b) central, and (c) western cities.


Fig. 6. IRFs of economic development in (a) eastern, (b) central, and (c) western cities.

The impulse response function of mortality on SO₂ emissions shows that when there is a positive shock to mortality, SO₂ emissions decreases in east and west. In central China, the effect is a delayed one, with the SO₂ emissions initially increasing, before being affected negatively in a prolonged fashion. When comparing these results, SO₂ emissions decreases more to changes in mortality in the east than in central and west. This seems to be surprising at first glance, but it implies that the east is more concerned about health, the central cities are still at a stage to sacrifice health to exchange for development. For the west, it does not concern much about the negative relationship between mortality and pollution.

Fig. 6 plots the IRFs of development to industrial structure, pollution, development and mortality shocks at a five-year horizon. The response of income to a shock to pollution is negative in the east, though the effect is a delayed one. For the central and western region, following the shock in pollution, income decreases on impact, damping the positive effect and turning to negative. The negative effect reaches the smallest at the first year, and about six months later, it becomes positive in the subsequent years. Three years later, the effect reverts back to zero. Note that for the whole time horizons, the pattern is quite similar across both the central and the western cities, although the impact of the lagged determinants on the level of development is smaller in the central cities than it is in the western cities. Thus, the findings support the notion that environmental factors are closely related to development. For the east, environmental inputs no longer drive development, but for the central and western region, environment still plays an positive role.

Next, the IRFs illustrate that a shock to development increases income for all the three regions, but the magnitudes of the effects are very different. The effect is largest in the east, and smallest in the central region. This result partly demonstrates that China is caught in the “rich get richer” trap (i.e., the Matthew effect). Furthermore, the unconditional convergence hypothesis seems to be rejected, as no sign of the poorest west converging to the wealthiest east is found. Despite these, there may exist a glimpse of hope. The poorest west is likely to converge to the middle-income central region, a sign of “convergence club.” However, the central region shows no sign of converging to the east, a strong signal of middle-income trap.

In addition, the results suggest that the impact of mortality shock on development is positive in all the three regions, though the effect is very small. This result shows that the fast growth is at the expense of health in a way.

CONCLUSION

Our results indicate that SO₂ emissions promote the GDP in China, although the effect displays high heterogeneity between the western, central, and eastern regions. Whereas the costs of this pollutant total approximately 5% in the west, they surpass 20% (the national maximum) in the center, indicating that the latter still tends to adopt more environmentally detrimental technologies. In the east, however, the costs of SO₂ pollution are small or negative, demonstrating that this region is slowly transitioning to a modern service economy. We also find that the health costs total approximately 2%, 3%, and 1% of the GDP per capita for the eastern, central, and western regions, respectively, revealing that rapid economic growth has been achieved at the expense of health.

ACKNOWLEDGEMENT

We thank the editor for their valuable comments and suggestions. All errors remain our own. The authors thank Xiangjun Wu at Hangzhou Dianzi University and Zhe Peng at Wilfrid Laurier University for excellent research assistance. This work is supported by the Ministry of Education Project of Humanities and Social Sciences (Grant No. 18YJC790190), the Fundamental Research Funds for the Central Universities Zhongnan University of Economics and Law (Grant No. 2722020JCT029) and Educational Commission of Anhui Province of China (Grant No. SK2019A0469).