The driving force of carbon emissions in china: 1995-2015 hierarchically provincial evidence

This paper computes the provincial carbon emissions, use a “region-province-time” three-dimensional panel data set from 1995 to 2015, and conduct an empirical study to explore the driving force of carbon emissions in China. A hierarchically spatial autoregressive error (HSEAR) model is established, which takes into account the hierarchical structure and spatial error effect. The empirical results suggest the carbon emission and per capita GDP forming an “N” shape Environment Kuznets Curve (EKC). Meanwhile, by the growth of coal population, consumption proportion and the number of private cars will significantly increase the carbon emissions. The significantly positive spatial correlation of the error terms implies that the error impact of carbon emissions in the area has a significant positive spatial correlation with the adjacent area, which corrects the error of the general spatial error model to the socio-economic reality.


Introduction
There is a lot of evidence in the statistical data and literature that carbon dioxide (CO2)-based greenhouse gas emissions are the main cause of global warming. With the rapid economic development in China, the amount of carbon emissions becomes a serious Chinese issue. It is important to explore the factors which drive the carbon emissions. The existing research on the driving forces of carbon emissions in China including the environmental Kuznets curve (EKC model) and environmental impact factor decomposition method. In the empirical research literature of EKC model, Du Zhao et al. (2018) introduced the spatial panel data model to amend the influencing factors of China's regional carbon emissions and their impact. However, the spatial panel data does not take into account the nested or hierarchical structure of data, such as different provinces belong to different regions, regions may have different characteristics due to different government policies, economic conditions, and geographical location. This study will consider the nested relationship between different levels of research objects and hierarchical spatial data, spatial data econometric models, and introduce the EKC analysis framework of the factors affecting carbon emissions, consider the hierarchical effects of regional carbon emissions. Cons represents the consumption of carbon emission source k within province j in year t , k  represents the coefficient of carbon emission source k (i.e.  (coal, coke, gasoline, kerosene, diesel, fuel oil, natural gas, cement)={ 1.647, 2.848, 3.045, 3.174, 3.150, 3.064, 21.670, 0.527}).

Figure 1. Trend graph of Regional Carbon Dioxide Emissions 1995-2015
As is shown in Figure 1, the four curves show the fluctuation of regional carbon dioxide emissions, the increase of

The theoretical model of the Driving Force of Chinese Carbon Emissions
Ye and Long (2016) suggested the GMM-FGLS estimate for a hierarchically spatial error autoregressive (HSEAR) model, which consider the hierarchical structure of the data and spatial effects simultaneously. Based on the above data description, the carbon emissions' hierarchically spatial error autoregressive (HSEAR) model established as follows: where ijt CDE represents the total carbon dioxide emissions of province j nested in region i in year t , the explanatory variables vector including the per capita GDP, its quadratic form, and its cubic form, provincial population, the ratio of coal consumption and the number of private cars.  Represents the vector of parameters to be estimated. ijt u Denotes the disturbance term of the j province nested in the region i in year t time period. The nested random effects are introduced via the disturbance of ijt u which follows an error component structure.  Denotes the scalar spatial autoregressive coefficient to be estimated. i  Denotes the itch unobservable region-specific effect which is assumed to be i.e.   is constructed by the Latitude and longitude distance between capital cities.

Estimation results of the HSEAR model
After estimated by using the suggested approach of GMM-FGLS via GAUSS 15.0, the results of HSEAR model and general spatial autoregressive error model (SEAR) as comparison are reported in Table 1. Note: *** indicates significant at 1% confidence level, ** indicates significant at 5% confidence level, * indicates significant at 10% confidence level.
It can be seen from the Table 1 that all coefficients of HSEAR and SEAR are significant at the 1% level, implying that the estimators are theoretically acceptable. The adjusted R2 of model HSEAR is higher than model SEAR, AIC and BIC of model HSEAR is lower than model SEAR, revealing that the model HSEAR has better performance than SEAR. Under the circumstance of dividing China into 4 regions by HSEAR model, the PGDP, its quadratic form, and its cubic form are 0.2601, -0.4479 and 0.2243, respectively. This suggests that after carbon emission growth enters the inflection point, carbon emissions gradually become smooth. With the rapid development of economy, the carbon emissions enter the rising stage once again, forming the environment's Kuznets "N" shape EKC curve, which became a national symbol that the quick advancement of China's industrialization and urbanization.
The carbon emissions elasticity on population and Civil Car Parc are 0.0485 and 0.1598, which point out that more carbon dioxide will be emitted with the number of population and private cars increase. Furthermore, on average, a 1% proportion of coal consumption in total energy consumption increase will lead a 0.8277% increase in the carbon dioxide emissions, enlightening that it is the main reason of the carbon emission. This suggests that we can reduce carbon dioxide emissions by reducing the proportion of total coal consumption in total energy consumption, adopt energy conservation and emission reduction measures, and population control. For various provinces in the same region of China, there is a relatively small regional difference   = 0.0014 by HSEAR model, and compared with the previous inter-provincial effect 0.0805 by SEAR model, it is significantly lowered; the standard deviation of effect   = 0.0833 that the province is nested in regions represents the group difference between provinces among provinces and regions. Compared with various provinces in the same region, various provinces of different regions in China have a greater difference, which highlights the feature that "there is a larger difference between groups and a small difference within groups". An interesting finding of the estimates of spatial autoregressive error correlation is that compared with the non-significant ˆS EAR  = -0.1118 by SEAR model, HSEAR model is significantly positive, ˆH SEAR  = 0.1398. We infer that this reason is due to the sensitivity of regional grouping. It is further demonstrated that we use HSEAR model can identify the positive spatial correlation of the error terms in the model. It shows that the error impact of carbon emissions in the area has a significant positive spatial correlation with the adjacent area, which corrects the error of the general spatial error model to the socio-economic reality.

Conclusion
In this paper, we consider a "region-province-time" three-dimensional panel data set for 30 provinces from 1995 to 2015. Firstly, we calculate the carbon emission for every province and group them as four regions (the Eastern, Central, Western, and Northeast China) by the unbalanced development of economy and analyze the distribution of interprovincial carbon emission. By introducing the hierarchical structure and spatial autoregressive error term into the error term, we construct an HSEAR model to investigate the impact factors of carbon emissions. The empirical results imply that the carbon emission and PGDP performing an "N" shape EKC Curve. It is important for China to reduce the proportion of total coal consumption in total energy consumption, restrict the issuance of new car license, and control the population growth rate. The standard deviations of the hierarchical effects feature that "there is a larger difference between groups and a small difference within groups". HSEAR model can identify the positive spatial correlation of the error terms in the model. In other words, the status of carbon emissions in China is unbalanced and regional agglomerate. Therefore, China has the challenge to adjust the energy composition, develop clean energy, and create an energy sustainable development system.