1. Introduction
Global warming has caused huge damage to the natural ecological environment on which human beings depend [
1]; the culprit behind global warming is the accumulation of carbon dioxide. The negative impact of global warming has made all countries realize the necessity of jointly dealing with CO
2 emissions. The international community has developed a series of agreements to urge countries worldwide to cut CO
2 emissions in steps, such as the Copenhagen Accord (2009) and the Paris Agreement (2015) [
2].
Currently, China’s CO
2 emissions rank first in the world [
3], and statistics show that China’s total of CO
2 emissions in 2021 was 10.6 billion tons. The heavy industry contributed 55% of China’s CO
2 emissions during the period 2005–2019. Carbon intensity is a key factor affecting CO
2 emissions of the heavy industries [
4]. The motivation for this article is to investigate the effects of the influencing factors on the carbon intensity of the heavy industries, and to propose targeted countermeasures. The research results can provide empirical support for local governments to formulate industrial policies and can offer new paths for achieving carbon neutrality goals.
The importance of carbon intensity has attracted many scholars to conduct in-depth research into it. Analysis of the existing relevant literature found that these studies have the following two notable characteristics: (1) Existing studies have investigated the impact of environmental regulations on carbon intensity, but they do not break down environmental regulations. These regulations can be further subdivided into mandatory environmental regulations and incentive-based environmental regulations [
5]. Moreover, the ways in which these two environmental regulations affect carbon intensity are markedly different; (2) Most literature investigates carbon intensity in the industrial sector using the ordinary least squares. However, a prerequisite for the ordinary least squares to obtain robust parameters is that the sequence of economic variables is normally distributed. In fact, economic phenomena are complex and the data of economic variables have obvious characteristics of “sharp peak” and “thick tail”. This results in the series of economic variables being skewed. Using the ordinary least squares to estimate non-normally distributed variable data will lead to serious adverse consequences, such as increased variance and biased parameter estimators.
The possible contributions of this paper are mainly the following two points: (1) Considering the availability of data, this paper divides environmental regulations into mandatory environmental regulations and incentive-based environmental regulations. This paper investigates the impact of these two environmental regulations on carbon intensity separately. The findings help local governments to flexibly use corresponding environmental policies to promote carbon intensity reduction; (2) Preliminary test results show that the series of economic variables in this paper are not normally distributed. Quantile regression models do not require that the series of economic variables follow a normal distribution. Under the condition that the series of economic variables is not normally distributed, the results of quantile regression are more robust than those of the ordinary least squares [
6]. Quantile regression can estimate the effects of influencing factors on carbon intensity at different levels, including maximum, minimum and median values. The ordinary least squares can only give an average effect of influencing factors on carbon intensity. Thus, the research results can provide empirical support for local governments to formulate effective emission reduction policies.
Apart from the introduction, this paper has the following sections: the literature review is placed in
Section 2; the theoretical model and the construction of the empirical model are placed in
Section 3; the empirical results are placed in
Section 4; the discussion section is placed in
Section 5; the results of the robustness test are presented in
Section 6; policy recommendations are placed in
Section 7; and the Limitation of the study and future recommendations are placed in the last section.
2. Literature Review
This section selects several representative influencing factors of carbon intensity for comment. China is now the world’s largest CO2 emitter; in order to achieve low-carbon growth, the Chinese government has taken many measures to control fossil energy use and reduce CO2 emissions, such as expanding investment in environmental governance and promulgating environmental laws. Most of these measures fall into the category of environmental regulations. Therefore, this section first reviews the existing literature on environmental regulations.
(1) Mandatory environmental regulations. In a survey in China, Liang et al. [
7] found that mandatory environmental regulations promoted the improvement of environmental efficiency and the decrease in carbon intensity. The scale of China’s manufacturing industry was the largest in the world, and its carbon intensity was high [
8]. Jiang et al. [
9] pointed out that mandatory environmental regulations had not played a role in improving energy-saving technologies and reducing carbon intensity. However, Wang and Zhu [
10] reached the opposite conclusion. Using a spatial econometric model, they found that mandatory environmental regulations helped to reduce carbon intensity. The main reason was that mandatory environmental regulations prompted heavy industrial enterprises to move from the provinces with strict environmental regulations to the provinces with low environmental regulations. A difference-in-difference model survey indicated that mandatory environmental regulations were beneficial for reducing the carbon intensity of the heavy industry [
11]. The main reason was that mandatory environmental regulations increased the pressure on industrial enterprises, prompting industrial enterprises to change environmental strategies [
12].
Compared with mandatory environmental regulations, incentive-based environmental regulations were flexible. Therefore, incentive-based environmental regulations could play a better role in environmental governance.
(2) Incentive-based environmental regulations. Incentive-based environmental regulations include many means of environmental governance, such as resource tax, carbon tax, carbon emission permits, and wastewater discharge permits [
13]. However, the main task of developing countries was to facilitate the rapid economic development and social employment [
14]. Therefore, many developing countries implemented loose incentive-based environmental regulations [
15]. This attracted many investment projects in the heavy industries, such as the petrochemical, plastics processing, and metallurgy industries [
16]. The lax incentive-based environmental regulations had made many heavy industrial companies pay little attention to improving energy efficiency and had resulted in high carbon intensity [
17]. During the 12th and 13th Five-Year Plans, China’s incentive-based environmental regulations had continuously strengthened and played a critical role in reducing carbon intensity [
18]. Using a stochastic frontier approach, Yao et al. [
19] found that incentive-based environmental regulations had prompted industrial enterprises to increase investment in technology research and development. Technological innovation could not only decrease the use of fossil energy but also expand the application of abatement equipment, thereby contributing to the reduction of carbon intensity.
Environmental regulations were mainly formulated by the governments. Energy prices were not only affected by domestic factors, but also by international factors. The heavy industry was energy-intensive and sensitive to energy prices. Soaring oil prices would prompt industrial enterprises to expand clean energy use, helping to reduce carbon intensity. The existing literature on the relationship between energy prices and carbon intensity is reviewed below.
(3) Energy prices. The production activities of the heavy industries often required a large amount of energy input [
20]. Rising fossil energy prices turned to increase cost for heavy industry producers [
21]. To reduce production costs and enhance market competitiveness, heavy industry production companies had begun to focus on improving energy efficiency and reducing carbon intensity [
22]. Natural gas was a clean energy source, and many countries strongly encouraged natural gas consumption [
23]. Government departments had expanded financial subsidies that had lowered natural gas prices. The decline in clean energy prices incentivized heavy industry producers to gradually increase the use of clean energy, thereby promoting carbon emission reductions and mitigating carbon intensity [
24]. In recent years, major energy-consuming countries have accelerated the development of wind power and photovoltaic power [
25]. Governments have also provided financial subsidies for wind power and photovoltaic power generation, thereby reducing renewable energy prices [
26]. This induced heavy industry to expand the use of clean electricity, contributing to CO
2 emission reduction and carbon intensity mitigation.
Energy prices could affect the energy structure of the heavy industries. A sharp drop in oil prices would help increase the share of fossil fuels in energy structure, resulting in a high carbon intensity. A literature review of the relationship between energy structure and carbon intensity is given below.
(4) Energy structure. In general, the heavy industry still relied mainly on highly polluting coal to meet energy needs [
27]. Coal had a high carbon content, as its excessive use resulted in the emission of carbon dioxide and high carbon intensity [
28]. Using comparative analysis, Rojas-Cardenas et al. [
29] examined the carbon intensity in Mexico and the world’s energy-consuming countries; they showed that the carbon intensity of Mexico’s steel industry was low, since the main source of energy consumption was clean natural gas. Similarly, Switzerland’s heavy industrial production plants increased biomass energy use and significantly reduced the share of coal in energy consumption [
30]. The carbon intensity of China’s heavy industrial sector had historically been high, mainly because many heavy industries still relied on high-emission coal [
31]. A coal-dominated energy mix remained a major obstacle to cutting down carbon intensity [
32].
In the long run, technological progress was an important factor of carbon intensity. More low-carbon technologies would certainly help mitigate carbon intensity. The literature on the relationship between technological progress and carbon intensity was described as follows.
(5) Technological progress. Employing the decomposition approach, Song et al. [
33] studied the heavy industry and showed that technological advances could help reduce carbon intensity. Moreover, the level of technology had obvious regional heterogeneity. It was because there were significant regional differences in R&D funding and technological talent [
34]. This made the contribution of technological innovation to carbon intensity in different regions significantly different [
35]. Findings from the Swedish steel industry showed that updating furnace technology and equipment could significantly improve energy efficiency and reduce coal use, thereby contributing to the mitigation of carbon intensity [
36]. The European steel industry analysis also found that updating blast furnace combustion technology and carbon capture technology could significantly reduce carbon intensity [
37].
Technological progress could help industrial enterprises to update energy utilization technology and reduce energy intensity, thereby contributing to the reduction of carbon intensity. The literature on the relationship between energy intensity and carbon intensity is reviewed below.
(6) Energy intensity. The heavy industry was an energy-intensive industry, as its energy intensity was significantly higher than that in other industries such as the light industries, services, and agriculture [
38]. Under the background of China’s energy consumption mainly relying on highly polluting fossil energy, high energy intensity made the reduction in carbon intensity a difficult task [
39]. Shandong and Liaoning provinces were two major industrial provinces in China, where many heavy industrial enterprises gathered [
40]. Therefore, reducing energy intensity was key in facilitating the decline of carbon intensity. A survey of the heavy industries in India yielded similar findings that energy intensity was one of the key factors influencing the changes in carbon intensity [
41].
Reviewing the above research literature, it is found that most of the sources use mean models to analyze the impact of environmental regulations and other determinants on carbon intensity. The obtained research results have little reference value for local governments to manage the environment and reduce carbon intensity. Therefore, this paper first subdivides environmental regulations into incentive-based environmental regulations and mandatory environmental regulations. Quantile regression methods were then used to investigate the impact of these two environmental regulations and other determinants on carbon intensity. The quantile regression can estimate the heterogeneous impact of the influencing factors on carbon intensity, including the effect of the maximum, minimum, and median values.
4. Results
4.1. Unit Root Test
Since the 1960s, the development of econometrics has entered a fast lane. Econometric methods have become mainstream economics, sociology, and management research methods. However, the premise of econometric model regression is that the variable series is stationary. The unit root test is generally used to judge whether an economic series is stationary. If the socio-economic variable sequence has unit roots, it becomes a non-stationary sequence. Many existing studies have proved that a unit root process in an economic sequence implies non-stationarity and that a pseudo-regression problem in regression analysis may occur. The unit root test has now become an indispensable step in econometric analysis. If the probability value of the statistic value of the unit root test is less than 10%, it means that the tested economic variable sequence is stationary. On the contrary, it means that the series of variables is non-stationary. This paper uses the IPS, ADF, ADF–Fisher, Breitung test, Fisher-PP, and CIPS tests to examine the economic series used. Different from other testing methods, the CIPS test can determine whether the series of economic variables is stationary on the basis of considering the possible cross-sectional correlation. In order to improve the reliability of the test results, this paper uses these six methods to perform unit root tests. The formula for the main tests is listed in
Appendix A. The results in
Table 3 show that all economic series are not stationary.
4.2. Cointegration Tests
Cointegration theory was first proposed by Engle and Granger [
53]. It lays a theoretical foundation for finding equilibrium relations between two or more non-stationary variables and establishing error correction models with cointegration relations. Cointegration is the sequence when multiple economic variables show as non-stationarity, but a specific linear combination of these variables shows stability. Under these circumstances, a long-term stable relationship can exist between these variables. The methods suitable for testing the cointegration relationship of panel data mainly include the Pedroni test and the Kao test. Based on the logarithmic variable data, this paper uses these two methods to perform cointegration tests (
Table 4). The probability values of the statistical values in
Table 4 are all less than 10%, indicating that there is a causal nexus between carbon intensity and its determinants.
4.3. Multicollinearity Test
Econometric theory states that multicollinearity problems will arise if the explanatory variables are closely related. Multicollinearity includes perfect multicollinearity and approximate multicollinearity. Perfect multicollinearity will have some serious consequences, such as non-existence of parameter estimates and infinite variance of parameter estimates. Approximate multicollinearity produces undesirable results as follows: (1) the variance of the parameter estimates increases; (2) the confidence intervals of the parameter estimates become larger; and (3) the t-test of the estimated parameters may not be significant. The methods of multicollinearity test mainly include: (1) simple correlation coefficient method; (2) variance inflation factor (VIF) method; and (3) Klein’s discriminant rule. The VIF method has the advantages of easy operation and objective test process. Therefore, this paper uses the VIF method to perform a multicollinearity test. The formula of VIF is as follows:
where
represents the goodness of fit.
J indicates that the
j-th explanatory variable is used as the dependent variable for regression estimation. If
VIFj ≥ 10, it indicates there is a severe multicollinearity between the explanatory variable and other explanatory variables. Moreover, this multicollinearity has serious adverse effects on the results of the ordinary least squares estimation. Conversely, if
VIFj is less than 10, it indicates that the multicollinearity is weak and will not have a serious adverse effect on the estimation results. The results in
Table 5 show that all the values of VIF are less than 10, indicating that model 4 does not have severe multicollinearity.
4.4. Tests of Normal Distribution
For traditional econometric models, an important prerequisite for obtaining good estimates is that the variable sequence obeys a normal distribution. However, many research results have shown that the series of economic variables is often non-normally distributed. Under the condition that the economic series is not normally distributed, the result of quantile regression is more robust. Therefore, before the model regression, this paper uses the Q-Q diagram (i.e., quartile–quartile) to test whether the economic series used is normally distributed. The results in
Figure 1 show that the blue variable curve does not completely fit the X = Y line, which indicates that these economic series are not normally distributed. In this case, it is more reasonable to use quantile regression to investigate carbon intensity.
4.5. Quantile Regression Results
According to the level of carbon intensity, this paper divides the 30 administrative units into 6 groups (
Table 6). The results of quantile regression are listed in
Table 7, and the graph of quantile regression results is placed in
Figure 2. In addition, this paper also uses the mean models for regression, and compares its results with that of quantile regression (
Table 7):
(1) Incentive-based environmental regulations. The carbon intensity in the 10th–25th, 25th–50th, and 50th–75th quantile groups receives a larger impact from incentive-based environmental regulations, with coefficients of 0.418, 0.479 and 0.418, respectively. These quantile groups mainly include Jiangsu, Shandong, Fujian, Hunan, Zhejiang, Henan, Hebei, Guangxi, Liaoning, Hebei, and Shanxi provinces. These provinces are home to many industrial enterprises, such as coal processing plants, petrochemical, iron and steel, and cement enterprises. To achieve low-carbon growth, local governments strengthen environmental regulations. Strict environmental regulations have prompted industrial companies to upgrade technology and equipment, which further cut down carbon intensity;
(2) Mandatory environmental regulations. Carbon intensity receives a negative effect from mandatory environmental regulations, as its regression coefficients ranging from the lower 10th quantile group to the upper 90th quantile group are −0.010, −0.053, −0.102, −0.136, and −0.151, respectively. A negative coefficient indicates that increasing the formulation and implementation of environmental laws has the effect of reducing carbon intensity. Furthermore, mandatory environmental regulations have a greater impact on carbon intensity in the 25th–50th, 50th–75th, and 75th–90th quantile groups. This is mainly because these provinces have enacted more environmental laws to deal with industrial pollution;
(3) The regression coefficients of energy consumption structure in the 25th–50th, 50th–75th, and 75th–90th quantile groups are negative, −0.328, −0.852, and −0.948, respectively. This result implies that changes in energy structure have contributed to the decline in carbon intensity. This is because the rate of clean energy in total energy consumption continues to expand, which is conducive to reducing CO2 emissions. These groups include Yunnan, Sichuan, Guangxi, Shanxi, Shaanxi, Inner Mongolia, Liaoning and Jilin. There are many rivers in Yunnan, Sichuan, and Guangxi, and the vertical drop of the rivers is large. The central government and local governments continued to increase hydropower development, and their hydropower output increased rapidly. Local industrial enterprises expand the use of hydropower, which improves energy structure and further drives down carbon intensity. Shanxi, Shaanxi and Inner Mongolia are major coal producing provinces. The local government encourages coal-to-gas production. Local industrial companies expand gas use, which helps reduce carbon intensity;
(4) The parameter estimates of technological progress in all quantile groups are negative, namely −0.287, −0.327, −0.336, −0.275, and −0.170, respectively. This means that technological innovation is conducive to reducing carbon intensity. Technology is an important contributor to achieving carbon intensity reduction. In recent years, economic development has enabled government finance and industrial enterprises to allocate more funds to scientific and technological innovation. The number of granted patents has grown rapidly, and the application of patented technology helps industrial companies reduce carbon intensity;
(5) Foreign direct investment is negatively related to carbon intensity, which means that expanding foreign investment can contribute to the decline of carbon intensity. To improve the technical level of industrial enterprises, the government vigorously introduces foreign investment projects with advanced technology and less pollution. These technologically advanced and low-energy-consuming projects have driven domestic industrial enterprises to upgrade their technologies and equipment, reducing carbon intensity;
(6) The parametric estimates of oil prices are negative in the high quantile groups, indicating that oil prices have contributed to the decline in carbon intensity. International oil prices have generally risen, although with occasional sharp fluctuations. This brings a heavy economic burden to local industrial enterprises and is not conducive to the long-term stable development of this industry. Therefore, local governments actively develop clean energy and accelerate the replacement of oil resources. Expanded use of clean energy reduces CO2 emissions, thereby contributing to a reduction in carbon intensity;
(7) Economic growth. The regression coefficients of economic growth in all groups are negative, −0.489, −0.688, −0.652, −0.584, and −0.820, respectively. This denotes that economic growth is contributing to a reduction in carbon intensity. The coefficient values are all between −1 and 0. It indicates that for every 1% increase in economic growth, carbon intensity decreases by less than 1%. The decline in carbon intensity is less than the increase in economic growth, and the main reasons for this are as follows: the mode of economic growth is undergoing a transition from an extensive economic growth to green economic growth; governments at all levels vigorously develop the high-tech industries and tertiary industries, and strictly control the scale of heavy industry; technological innovation and equipment renewal brought about by the development of high-tech industries are applied in the heavy industries, which promotes the reduction in carbon intensity; and economic growth still emits carbon dioxide, but the carbon intensity of economic growth is gradually decreasing.