How Does Environmental Regulations Effect Pattern of Carbon Emission and Energy Efficiency: A Provincial Level Analysis of Chinese Energy-Intensive Industries?

Energy has a huge environmental and economic implications in the modern community. Despite the rapid economic growth of China in the past two decades, it can further improve through sustainable green energy with more energy-efficient industries, so as to maintain a good balance between economic and social development. The performances of energy and carbon dioxide emissions are the critical indicators. On this basis, this work measures the impact of environmental regulations on energy efficiency based on 2008-17 panel data from 30 provinces in China. The total factor energy efficiency index (TFEEI) is calculated by the non-radial distance function (NDDF). In order to study the nonlinear relationship between environmental regulations and TFEEI, the dynamic threshold panel model is used under different environmental regulations, which can solve effectively endogenous problems and regional heterogeneity. The results show that, for energy-intensive industries, the overall average TFEEI level is still very low, with average values of 0.55 and 0.58, which are well below the ideal value (i.e., 1). Further, the dynamic panel data model findings showed a U-shaped significant relationship between China's TFEEI and environmental regulation. The findings reveal that environmental regulation effect on TFEI rises steadily as the values of Market-Based Environmental Regulations (MERs) and Command and control Environmental Regulations (CCERs) and surpass the corresponding thresholds. This research can help policymakers understand the effectiveness of various levels of environmental legislation to make more informed decisions.

emissions are the critical indicators for economic modelling. That is why, both energy and carbon dioxide emissions are considered in the analysis framework (Wang et al., 2019). At present, improving energy and carbon efficiency is one of the vital public policy concerns. The historic "Paris Agreement" set a below 2°C global temperature limit and started working to keep the temperature 1.5°C to take it to pre-industrial levels 1.5°C (Li et al., 2018). A quick reduction of greenhouse gas emissions is critical to the success of this agreement. Under this circumstance, China is determined to reach a peak in carbon emission by 2030. The country set the plan to reduce carbon emission per unit of GDP by 40-45% by 2020 and by 60-65% by 2030 considering 2005 as the base period (Liu et al., 2019b).
More than one third of the global energy is consumed by industries across the world. The proportion of C02 emission is slightly higher in this regard. The development model of China relies heavily on the industry, so local infrastructure development, manufacturing of exportoriented consumer products and heavy industrial equipment are highly supported by energyintensive production mechanisms. Therefore, the proportion of energy consumption with respect to carbon emission is higher than the world average in China . The statistical report of 2010 from " the National Economic and Social Development states that processing of the raw materials of petroleum and chemical products, melting and crashing of metal, non-metal and ferrous metal products, electricity generation and distribution are considered high energy consuming activities. According to this report, all these sectors are segregated into sub-sector of Energy-intensive industries (Liu et al., 2019a). In 2015, Chinese energy-intensive companies emitted 79.68% of total carbon dioxide produced across the country (Fig. 1). Due to the rise of energy intensity and subsequent carbon emission, China faced massive international criticism, and this led the government to initiate a low carbon emission policy in its 13 th five-year plan from 2016 to 2020 by setting a maximum carbon emission limit for all the energy-intensive companies of the country. It is determined in the plan that the ferrous metal processing industry should decrease its energy use by a minimum of 10% while the rate for the petrochemical and the nonferrous metal industries, in this regard, is 18% .

Fig. 1 Carbon emission scenario of Chinese industries (2002-2017)
Since fossil energy is considered to be the main factor in global warming, emphasizing the establishment of energy-efficient production and distribution processes may be the key to solving this hazardous issue (Meng et al., 2020). Economic developments across the nation are more or less correlated with energy intensity, carbon emission and global warming issues.
Therefore, businesses and governments should care about the human and wildlife, climate, and environmental aspects while setting up their respective growth strategies. In this connection, initiation of the green movement by adopting green technology solutions for industrial production and distribution can be effective undertaking. A robust energy and carbon management and control, if possible, with a strict regulatory framework can also improve the environmental quality to a great extent. This study tries to examine the performance scale of the total factor energy efficiency indicator (TFEEI) from the year 2008 to 2017 for six highly energy-centric industries located in a different province of China. The non-radial directional distance function known as NDDF is applied to measure the performance of the aforementioned indices. Based on provincial panel data (2008 to 2017) of China's six major energy-intensive industries, this study investigates the current level of the Total Factor Energy Efficiency Index (TFEEI). The first step is to calculate the TFEEI of China's energy-intensive industries through the non-radial direction distance function (NDDF). Second, the impact of the environmental regulations (ER) and other vital factors on TFEEI is studied through the GMM method, by dividing the environmental regulations into market-based environmental regulations (MERs) and command and control environmental regulations (CCERs). The interaction term between environmental regulation and research and development (R&D) is introduced to test the innovation offset effects by assessing the possible innovation offset.
This work contributes to the literature in the following three ways: it uses the NDDF to evaluate and compare the TFEEIs of China's high energy-intensive industries at a provincial level (the NDDF minimizes environmental pollution as it maximizes economic benefits); the environmental regulations are divided into MERs and CCERs with the help of the systematic GMM estimation method (the effectiveness for various types of environmental regulations is validated through our research); the nonlinear and heterogeneous effects of environmental regulations on provincial TFEEI are assessed, and the TFEEI of China's high energy-intensive industry is measured. The industrial characteristics and regional heterogeneity of China's high energy-intensive industry are captured based on large-scale provincial panel data to calculate the effects of environmental regulations on TFEEI. This research can help form targeted environmental governance policies with guidance.

Literature review
In this fast-paced globalized world, energy is the core of strong economic growth, but this energy is also the root cause of global climate change. In this case, decision makers need to formulate economic development strategies, taking into account environmental sustainability issues (Sadiq et al., 2020). There is a mature argument that paying attention to the development of energy-saving production technology may be a major measure to solve the current climate problem. Some studies are supporting this claim. For example, Wang et al. (2017) studied the energy-saving potential of 17 of the 17 Asia-Pacific Economic Cooperation (APEC) countries. Nassiri and Singh (2009) evaluated the energy performance of 21 Organization for Economic Cooperation and Development (OECD) countries by adopting both parametric and nonparametric approaches. Khoshnevisan et al. (2013) also studied the energy efficiency of 23 developing economies and concluded that the role of a strict energy policy is the key to ensure energy efficient production and environment. Moreover, the study of Chien et al. (2021);Geng et al. (2019) on 15 EU countries revealed that creating alternative energy sources and popularizing their use among industries and households can lessen pressure on fossil energy but improve the productivity comparatively. Meanwhile, (Song et al., (2013) and (Li and Lin, 2017) studied the provincial energy efficiency of China and concluded that technological advancement in the industry sector can lead to attaining energy efficient economic growth.
Researchers have adopted a variety of energy efficiency measures to study the energy and environmental efficiency levels of many different countries and regions. Data Envelopment Analysis (DEA) is one of the most popular and effective tools for measuring energy and environmental performance. Despite having some limitations, the DEA application has some advantages in evaluating energy efficiency. This study reviews the DEA application used by previous studies on the issues of energy and environmental efficiency and the decomposition of the Malmquist index. For example, ( Shao et al., 2014) and   (2020) conducted another study to explore the green total factor production efficiency and its determinants for the case of the metal industry in China with the application of sub-boundary and global DEA approaches. Apart from these studies, several other studies were conducted on the energy and environmental efficiencies for the steel, construction, and chemical industries of Chinese provinces. It is observed that current literature is based mainly on the energy efficiency of various industries of Chinese provinces. To the best of your knowledge, no prior studies objectively measured the energy and environmental performance of highly energyintensive industries of China from the operational front. In addition, it is observed that China is a huge territory and the different industrial sub-sectors that emerge along this territory affect the energy and environmental performance (EEP) of these regions. Therefore, by covering the heterogeneity of industries and regions, the total factor energy efficiency assessment of the six energy-intensive industries of Chinese provinces might help the policy makers to develop a sustainable strategy. It is further observed that the NDDF approach of efficiency measure is more flexible than any other measurement technique as it can satisfy the requirements of maximized economic growth and minimized pollution emissions. However, NDDF was used to assess the TFEEI of power sector only and no study is evident for such assessment regarding the six highest energy-intensive sectors. Therefore, this study uses the NDDF method for the TFEEI assessment of China's provincial energy-intensive industries to fill this research gap. Due to the extensive acceleration of ecological degradation, governments have undertaken a bunch of corrective measures to tackle and promote sustainable development. Research by Poudineh et al. (2020) believes that the application of taxation can internalize the cost of external pollution, and according to this proposition, many policies have been adopted, such as command and control based environmental regulations (CCER) and market based environmental regulation (MER), etc.
Other studies have shown that the formulation of regulations forces companies to increase expenditures or conduct environmental management and governance, and such enforcement will increase their operating costs and reduce economic benefits (Greenstone and Hanna, 2014;Peng, 2020;. However, according to Porter's hypothesis, a well-structured environmental regulatory framework can accelerate technological innovation, thereby reducing compliance costs. Environmental regulations have significant causal relationship with both energy and environmental performance. For example, Jefferson et al. (2013) state that in China, such regulations can bring improved performance of the industry sector. Bi et al. (2014) also proposes that the thermal power generation sector of China can be made energy efficient if a well-regulated environmental atmosphere is in place. Recently, Lin and Chen (2020) showed that for both long run and short run energy efficiency, the contribution of MER is noteworthy but CER performed better in the short run. Zhang and Song (2021) also explored the nexus between environmental regulations and environmental control in the metal sector of Chinese territory and concluded that there is a nonlinear causal relationship among CCER and environmental control. However, they also concluded that CCER had no causality with environmental control. In addition, there are few studies to measure the degree of relationship between other control factors of energy efficiency. For example, Antonietti and Fontini (2019) considered economic development level, Y.  considered technology enhancement, and Xin-gang and Shu-ran (2020) considered energy price as the affecting variable for energy efficiency measure. In particular, it is currently unclear from the existing studies how different types of environmental regulations affect the TFEEI of the Chinese high energy intensive industries. The high energy intensive industries are the most polluting industries in China. Therefore, investigating the effect of environmental regulations on these industries is a timely attempt for ensuring an environmentally sustainable developed economy in China.

The Non-radial directional distance function (NDDF) Model
In the case of measuring the relative efficiency of any organization (Decision Making Unit), The application of DEA is widely accepted. The DEA method is mainly an input-output based measure of production efficiency. DEA is applied to measure the economic efficiency of many different sectors and nowadays this method has received popularity for measuring environmental efficiency. After (Zhou et al., 2007) applied the NDDF method to measure the energy and environmental performance of the Chinese provinces. In the NDDF approach, capital (K), labor (L), and energy (E) are considered as input and industry output (Y) is considered as output. Meanwhile, CO2 is regarded as undesirable or bad output. For applying DEA in any efficiency measure, the formation of a production possibility set (PPS) is mandatory. Therefore, this study constructs a PPS which is stated equation. (1) (Zhou et al., 2006).
In the above equation, P stands for the PPS L, K, E stand for labour, capital, and energy. The PPE states that it is possible to produce out Y and C from the input K, L, and E. Table 1 shows the descriptive statistics for the input and output variables of this study.  (2010) also applied the DDF method for measuring efficiency by considering CO2 as a bad output. However, the DDF method considers an equal proportion of changes in the input-output matrix which might give an overestimated efficiency outcome (Lin and Du, 2015). To remove this shortcoming, (Zhou et al., 2012) applied the non-radial direction distance function (NDDF) which considers real production characteristics and accepts the non-proportional changes in both desirable and undesirable output variables. Later, Zhang and Choi (2013) applied NDDF for exploring the efficiency of Chinese power plants and found the method a more robust one for efficiency analysis.
2: If , , , , 2 ∈ = 0, 2 = 0) Zhang et al. (2014) state that the hypothesis of weak disposability means reducing the undesired output C will reduce the cost of the expected output O, while the zero joint hypothesis indicates that the undesired output C will inevitably be produced during the production process Such T is regarded as the environmental production technology (Zhou et al., 2012) and following this, T is estimated by applying the non-parametric DEA method. It is assumed that there are n DMUs and T is represented by the constant return to scale in equation (2): stands for the intensity variable to form constructing T as a convex expression (Chung et al., 1997). The NDDF is further used to calculate the EEP of each DMU which is represented as Eq. (3): Where is regarded as a normalized weight vector; and represented as the directional vector and is represented as the vector for scaling factor.

Formation of the Energy Efficiency Index
The total-factor NDDF (TNDDF) is constructed to evaluate the TFEEI by taking in to account the existence substitution effect between the energy and other variables. The construction of the model is portrayed below: ⃗⃗⃗⃗⃗ ( , , , , 2; G) = .  2012) and Barros et al. (2012), this study treats the inputs and both the desirable and the undesirable outputs with the same weight. Therefore, each of the variables is assigned with a weight of 1/3. Moreover, the weight of the input variables is evenly distributed to K, L and E; that is, the weight for each input factor is 1/9. furthermore, the directional vector is set as G =

The Dynamic Panel Data Model Effect
The following equation is constructed to find the relationship between environmental regulation and total factor energy efficiency: TFEEI = α + βTFEEI , −1 + γER + θ + + + In this equation, represents the intercept and , and are coefficients to be estimated. ER is the independent variable, that is, the vector that represents the command-and-control and market-based environmental regulation stringency? EEI , −1 is the first lag term of EEI . This lagged dependent variable EEI , −1 is added as the independent variable in constructing the equation considering the impact of lagged energy efficiency index on the current energy efficiency index and lagged environmental performance index on the current environmental performance index. matrix indicates the control variables set. is fixed time effect, is a single fixed-effect, and is a random error term.

The Dynamic Threshold Model
The above-mentioned study carries some limitations as the model of moderating effects fails to identify the key areas and relevant breaks of environmental regulation. This study considers a single threshold model in line with the idea of (Hansen, 1999) non-dynamic panel threshold model, to explore the non-linear causality between environmental regulation and TFEEI, to confirm the rationality of sample interval segment and to reduce the errors in model estimate.
The following section of this study addresses the environmental regulation variable as the threshold dependent variable to form a threshold effect model as below: = + 1 −1 + 2 , ∘ ( ≤ C) + 1 , ∘ ( > C) + ∑ 5 =1 + + + In this model above, C is the estimated threshold value, and I(·) is the symptomatic function, which will be true if the corresponding condition is equal to 1 and false if the value is 0. The results of the test might come up with the presence of multiple thresholds which can further be stretched to double and multiple threshold models from the base single threshold model.

Dependent variable
The explanatory variable in this paper is total factor energy efficiency index (TFEEI).

Independent variable
This study takes i). Command and Control Environmental Regulation (CCER) and ii). Some studies have also considered levying fees on pollution to encourage companies to be innovative, thereby helping the government to strengthen environmental governance.   industries showed an upward trend. The overall level of average TFEEI is still very low, with an average value of 0.55 and 0.58, which is far below the optimal value of 1. China's energyintensive industries consume large amounts of energy and emit large amounts of carbon dioxide, resulting in low energy efficiency. This shows that China's high energy-intensive industries are responsible for high levels of carbon dioxide emissions, which ultimately makes these industries energy inefficient. However, the government of China has emphasized on pollution control measures and high emitting industries like petrochemical and metal will be under severe environmental scrutiny. For this environmental regulatory initiative, TFEEI for high carbon emission industries has greatly increased. This finding also goes in line with (Wu et al., 2020) who investigated the efficiency of energy and environment of Chinese mine sector. An increasing trend for the TFEEI scores of chemical, non-metal, ferrous, and non-ferrous metal industries is evident in Fig. 2  technological conditions, and low efficiency are evident in the above-mentioned provinces.

Figure 3
The spatial and temporal distribution of average TFEEI of China's six high energyintensive industries at the provincial level

The Empirical Results of the Benchmark Model
The system GMM approach is applied to fix the issues relevant to dynamic panel estimation stated in equation 5. System GMM takes into consideration that there is no autocorrelation within the disturbance terms. This approach also solves the endogeneity issue by taking lag variables. The result of the GMM test is presented in table 5. Apart from performing the GMM, the study also conducted Arellano-bond (AR) test, Sargan-Hansen and Wald chi-square tests to attain a more robust estimation result. The Arellano-Bond (AR) test comprises both first and second-order autocorrelation of residuals tests which are known as AR (1) and AR (2) respectively. The residuals of the equation are regarded as not autocorrelated if AR (2)   Note: Standard errors are in parentheses (). ***= 1% significant level; **= 5% significant level and *=10% significant level Meanwhile, the regression estimation of the control variables shows that both research and development and industry structure positively affect the TEFFI at 1% significant level in all cases. This affirms that spending more on research and development and making industry structure optimum might accelerate the growth of TEFFI. On the other hand, both GDP and export negatively impact the TEFFI. Besides, it is observed that both FDI and urbanization have no significant influence on TEFFI improvement.
Considering the probable lag effect of environmental regulation, the estimation of Both Model 2 and model 4 are performed taking the lagged value of MER and CER. The results show the same scenario as the baseline regression as it is seen that there is existence of statistically significant of MER over TFEEI and the control variables also consistency with the estimated coefficients. However, for the case of CCER, the coefficients (0.0795) of lagged variables are found to be positive and statistically significant at 5% level proving that TFEEI growth is affected by the lags. It refers to the fact that in case of a rigorous CCER practice the influence of "innovation offset" based on is more powerful that influence of "compliance cost" effect in the long run. This proposition matches with the findings of Guo and Yuan, (2020), who argues that taking the lagged variables instead of current variables might increase to chances to generate a positive and significant effect on energy efficiency.

The empirical results of the moderating effect of R&D
Though all the moderating coefficients are seen positive, only MER and R&D are statistically significant with a 5% level (table 6)  [0] Note: Standard errors are in parentheses (). ***= 1% significant level; **= 5% significant level and *=10% significant level

The results of the threshold model
The threshold effect model is tested following three different procedures. First, the value and quantity of the threshold is determined with sample endogeneity. Then, in accordance with progressive distribution theory, the confidence interval of the threshold parameter is formed.
Finally, by using self-sampling technique, the significance level of the threshold is evaluated.
To ascertain the number of thresholds, three successive investigations are conducted by taking the assumptions of single, double and multiple thresholds. After the investigation of the number of environmental regulation thresholds, the value of F-statistic and its corresponding P-value are obtained (Table 7). It is evident from Table 8 that both single and double thresholds test results are positive and significant at 5% level while triple threshold fail to reach the required level of significance. according to test results of the threshold effect, the confidence interval shown here is 95%.  Table 9 presents the results of regression for the threshold model. As, the values of command-and-control environmental regulation (CCER) and market based environmental regulations exceed the levels of corresponding thresholds, the positive impact of environmental regulation on TFEI gradually increases. It is observed that the coefficient estimates for threshold effect model are 0.0571, 0.012, respectively and there is an upbound of their corresponding level of significance from 10% to 1%. This proposes that when the pull-out position of the environmental regulations improves by 1%, the total factor energy efficiency of the high emitting industries increases by 1.2% to 5.7%. It proves that the "J-shape" has a marginal Growth trend. This investigation results depicts the way different levels of regulations affect the causality between the surrounded position of environmental regulations, the TFEEI, and the threshold or turning point in this relationship. Note: Standard errors are in parentheses (). ***= 1% significant level; **= 5% significant level and *=10% significant level

Robustness Tests
To avoid the outliers adversely influencing the forecasted results, a 5% tail ended test is applied for the entire dataset. In Column 11 of the  (Bosch et al., 1995). This model is applied to describe the changes take place in the conditional quantile of the dependent variable with the changes in the independent variable. It is observed from the columns 12-16 of Table 11 that the regression coefficients of the TFEEI position in the five typical quantiles (10th, 30th, 50th, 70th and 90th) are have positive and statistically significant trends and it proves that the estimation is robust and reliable  Note: Standard errors are in parentheses (). ***= 1% significant level; **= 5% significant level and *=10% significant level

Conclusion and Policy Implication
This work uses 2008-17 panel data from China's 30 provinces to assess the country's energy- Based on the findings of this study, policymakers should understand the various levels of efficacy of environmental legislation to make more informed decisions. CCERs should be obligatory, resulting in a shortage of choices for the organization's renewable creative technologies. Policymakers should aggressively apply market-based environmental regulations to expand the regional industry market platform, encourage new research on emissions taxes and carbon trading, and eventually gain greater pollution control. The government should take appropriate steps in environmental governance to boost the TFEEI of energy-intensive industries and pay attention to each sector at the regional level.