Chinese growth and dilemmas: modelling energy consumption, CO₂ emissions and growth in China

Abstract

Global warming appears world challenging problem of current age. International communities have agreed in recent Paris agreement to reduce global pollution to certain level and have shown great concern with Chinese pollution. China is not only the world second largest and fast growing economy but the highest CO₂ emitters as 29.4% world emissions source is China. This paper adds in this debate by exploring cointegration and causal relation between real GDP per capita, CO₂ emissions per capita, energy consumption per capita and urban population for China utilizing annual data for the period of 1971–2014. Autoregressive distributed lag model has been used to confirm the cointegration among variables. Results reveal short run income coefficient is positive while it turns to negative and insignificant in long run marking that growth and development is supportive in CO₂ emissions reduction in China. Its insignificancy gives impression that only growth dependency to overcome CO₂ emissions will not be appropriate choice for China. By introducing income cubic function, N-shape relation between income and CO₂ emissions was found in long run that re-emphasis though income is helping in emissions reduction, however, there is need to dependent on multiple environmental policies. Further, results show that energy consumption has positive and significant impact on CO₂ emissions and energy coefficient was becoming fatter with the time. Urban population was insignificant while it turns to negative and significant with income cubic model suggesting that urbanization process is not adding to pollution emissions in China. Overall, results give feelings that model with income cubic function is suited well for Chinese economy. FMOLS and GMM estimators confirm ARDL results are robust. In addition, this paper also explores causal relation between the variables by utilizing error correction based on granger causality models.

1 Introduction

Global warming appears world challenging problem and its consequences seems severe. Intergovernmental panel on climate change (IPCC) has reported that climate change is raising sea level and average temperature around the globe that causes extreme weather events (Tsai et al. 2016). Researchers like Stern (2007) have warned from their projection about the economic loss from global warming that will be higher than that of World War I and World War II. Met-office and NASA data have reported the year 2015 as the warmest year in history dated back to 1850 (Met-Office 2015). International energy agency has documented that energy consumption accounts 80% of CO₂ emissions worldwide that is the major source of global warming. Thus, any effort to reduce CO₂ emissions will involve energy sector. Talking about the future, it has been predicted that in 2030 greenhouse gas emissions will increase 25–90% as compare to 2000 and energy related CO₂ emissions will increase around 40–110% till 2030 (IPCC 2007). These facts offer footprints for future’s danger if proper measures are not taken in time.

China is the source of motivation for many developing and developed countries. Despite the second largest economy and having social and political shocks, China has maintained high economic growth around 10% since last three decades. China’s economic growth is highly energy dependent. Yearly trend in Fig. 1 shows that real GDP and energy consumption have co-movement in long run, however, this heavily energy dependency leads to pollution emissions (Ahmed et al. 2016). The relation between energy consumption and CO₂ emissions in Fig. 2 confirm that these variables are interlinked in long run. Infact, China is not only the second largest and fastest growing economy but the highest CO₂ emitters in the world as CO₂ emissions in China has been reached more than 10 billion tons, amounting around 29.4% of total global emissions. Carbon dioxide emissions in China is higher than the combination of European Union and United States (Friedlingstein et al. 2014). Such environmental pressure and environmental consequences in the form of different unwanted events put pressure to reduce emissions. No doubt, controlling pollution and low pollution emissions is the dream of each and every economy and particularly, for China. It’s the reason, Chinese government has set up greenhouse gas reduction goals like by 2020, China will lower 40–45% CO₂ emissions per unit of GDP from its level in 2005. Additionally, Chinese government ambitious plan to work on CO₂ emissions to curb it up to 15% by increasing share of non-fossil energy in primary energy around 20% till 2030. These facts show that Chinese government is very willing to reduce pollution emissions to support global warming reduction plan (Li and Hu 2012).

Fig. 1

Source: Authors’ construction using world development indicators data

Real GDP and energy consumption co-movement in China.

Fig. 2

Source: Authors’ construction using world development indicators data

Energy consumption and pollution in China. Energy use (kg of oil equivalent per capita), CO₂ emissions (metric tons per capita).

China’s tremendous growth and development since its reform and open up policy (1978), subsequent of 30 years of growth is known as China miracle. However, this high growth is due to high inputs, high energy consumption and high pollution emissions caused by high energy consumption. Yearly trend between economic growth and CO₂ emissions (Fig. 3) shows that with the passage of time these variables are moving together with increasing trend. Rapid increase in energy consumption and pollution emissions forcibly suggest to work as environmental protection become central theme of political leaders and people debate where they seem more concerned about environmental pollution and its consequences in long run. Chinese public are sensitive and aware about the environmental issues and have great concern for pollution reduction (He and Lin 2017). Recently, in Paris agreement 2015, international communities have shown great concern about the Chinese pollution emissions.

Fig. 3

GDP and CO₂ emissions co-movement in China

This paper tends to extend previous thoughts and adds in literature by investigating the relation between energy consumption, CO₂ emissions and economic growth for China under environmental Kuznets curve (EKC) framework utilizing two novel approaches. Environmental Kuznets curve states that initially, rise in income raises environmental pressure but after the specific period, in long run, this relation turns to inverse and thus, further increase in income decreases environmental pressure. Grossmann and Krueger (1991, 1993, 1995) work has opened a new window of this direction and onward, there is small growing literature to test environmental Kuznets curve as EKC offers ideal solution for environmental pollutions i.e. economic growth itself is the solution to reduce environmental pollution. However, previous studies results are mixed for China and EKC has been investigated using income and income square as independent variables while CO₂ emissions as dependent variable in the model. Generally, income positive and income square negative signs confirm the validity of EKC. Econometrically, income and income square in one model can cause collinearity or multicollinearity problem (Narayan and Narayan 2010). In 2010, another criticism was from Lind and Mehlum (2010) who proved that taking income and it’s square in one model to prove U-shape or inverted U-shape relation is only necessary rather than a sufficient condition. In this situation, ignoring multicollinearity problem, one may wish to continue with the model by taking income and income square to prove U-shape or inverted U-shape relation that will only offer necessary condition. Additionally, she has to use U-test proposed by Lind and Mehlum (2010) to confirm necessary as well as sufficient condition. To overcome the issues and inspired by Narayan and Narayan (2010), aim of this paper is to test environmental Kuznets curve for China based on short run and long run income coefficients sign that will not only overcome multicollinearity problem caused by income and income square in one model but will make model irrespective from necessary and sufficient condition as suggested by Lind and Mehlum (2010).

This paper contributes in the existing literature by testing environmental Kuznets curve for China based on short-run and long-run income coefficients’ sign. It is built up on the idea that initially, in short-run, rise in income raises environmental pressure but after the specific period, in long run, this relation turns to inverse i.e. short run positive income coefficient and long run negative income coefficient will validate environmental Kuznets curve. Actually, this reversion of sign is important to confirm and validate environmental Kuznets curve to represent that with the growth and development, environmental pressure is going to reduce. The problem in this approach can be the shape of environmental Kuznets curve that will be unknown if she will make her judgment based on short and long run income coefficient signs. Thus, to overcome the shape problem of EKC for China that can be raised by making the judgment based on short run and long run income coefficients sign has been covered with the flexible function by introducing income cubic function in the model. Pablo-romero and De Jesús (2016) has pointed out that estimations with cubic form offers greater range and flexibility to the model in system. Omitted variables problem that can cause misleading results have been covered with additional variables such as energy consumption and urban population. Autoregressive distributed lag (ARDL) model has been to test the relation between variables because of its several advantages on other techniques, particular, it outperforms in small sample size and generally, it does not require pre-testing of unit root property of data. Cointegration among variables is the signal of causality, however, ARDL model offers only cointegration and does not explain about causality. To overcome this, vector error correction model has been used to detect direction of causality in short and long run.

The results of the paper are unique and show that in short run income coefficient is positive marking that initially, in short run, rise of economic growth raises pollution emissions but in long run, this relation turns to inverse and insignificant marking that economic growth will be supportive to reduce pollution emissions. Its insignificancy gives impression that only growth dependency to overcome pollution emissions will not be appropriate choice for China. There is need to depend on multiple policies to overcome pollution emissions as China is the highest CO₂ emitters i.e. 29.4% world CO₂ emissions is from China. When we try to robustify results with fully modified OLS (FMOLS) and generalized method of moment (GMM), results remain consistent with ARDL, however, long run income coefficient turns to significant re-confirming the validity of environmental Kuznets curve for China based on short run and long run income coefficients' sign. In order to test the shape of environmental Kuznets curve for China, flexible function has been introduced with cubic form. Results show N-shape relation between pollution emissions and economic growth in China. Further, results reveal that energy consumption coefficient was becoming fatter from short to long run showing that rise in energy consumption will raise more pollution emissions from short to long run. It gives signal to introduce alternative energy sources and increase energy efficiency. Urban population coefficient was insignificant but it turns to negative and significant by introducing income cubic function in the model marking that urbanization is helpful in pollution reduction. Error correction term (ECT) coefficient and its significance level has improved with the income cubic function. Results have been robustify with FMOLS and GMM estimators as FMOLS estimators are free from endogenity issue, serial correlation problem and provide unbiased results in small sample (Phillips and Hansen 1990; Stock and Watson 1993). Similarly, GMM overcomes endogenity issue by making model dynamic with the introduction of lags in the model. Causality results reveal bidirectional causal relation between energy consumption and economic growth, economic growth and pollution emissions, energy and pollution emissions that marked these variables are interlinked with each other in China. In this situation, ideal path seems to adopt alternative energy sources and increase energy efficiency that will not hurt economic growth and pollution problem can be eradicated.

Rest of the paper is structured as: Sect. 2 is for literature review, Sect. 3 is for data, model and estimation procedure via ARDL model to cointegration, Sect. 4 is for results and discussion while Sect. 5 concludes the paper.

2 Literature review

There is growing literature to investigate cointegeration and causal relation between energy consumption, CO₂ emissions and economic growth as these variables have policy concerns for the growth and development as well as for the environment. Wang (2013) has documented that the investigation of relation between CO₂ emissions and economic growth is necessary as these variables play focal role in the sustainable development and global pollution reduction debate. Furthermore, CO₂ emissions is directly linked with energy consumption that is an important factor for world economy and economic growth has an important implication for growth and development as well as for environmental policies. Recent studies like Sohag et al. (2017) have documented that energy use and industrial sector growth positively contribute in the explanation of CO₂ emissions. Researchers like Hwang and Yoo (2014) found bi-directional causal relation between energy consumption and economic growth while there was uni-directional causality running from economic growth to energy and to CO₂ emissions without any feedback. In the similar pattern, Ben et al. (2017) found bi-directional causality between CO₂ emissions and energy consumption. Indeed, studies use diverse approach, methods, variables and country/group of countries and results remain different and contradictory. Further, one country’s results cannot be generalized on other economy¹.

The field of study reveals the relation between CO₂ emissions and economic growth in the form of environmental Kuznets curve that postulates the relation between two variables is an inverted U-shape. It states that as income increases, energy consumption increases and as a result CO₂ emission increases in the environment. As income reaches to specific level, governments and people become aware and concerned about the regulations and environmental protection and as a result, they increase resources in pollution reduction plans. In the consequences, relation between CO₂ emissions and economic growth turns to inverted U-shape. In other words, as income increases, relation between CO₂ emissions and economic growth turns from positive to zero and then to negative marking inverted U-shape relation between two variables. The term environmental Kuznets curve emerged in the literature after Grossman and Krueger (1991, 1993, 1995) work where they have shown inverted U-shape relation between income and environmental pollution. Panayotou (1993) work was the first study that have introduced the term “environmental Kuznets curve” similar to original Kuznets curve (1955) where Kuznets have stated that there is an inverted U-shape relation between income and income inequality. The implication of these studies depict that additional economic growth will improve environmental quality.

Apergis and Payne (2010) investigate the cointegration and causal relation between GDP, GDP square, CO₂ emissions and energy for 11 commonwealth countries. They found inverted U-shape relation between GDP and CO₂ emissions that validate environmental Kuznets curve. Energy consumption impact was positive on CO₂ emissions. Saboori and Sulaiman (2013) explore cointegration and causal relation among GDP, GDP square, carbon dioxide emissions and energy consumption for five ASEAN countries (Singapore, Malaysia, Thailand, Philippines and Indonesia). They confirm the existence of environmental Kuznets curve in long run for Thailand and Singapore only. Narayan and Narayan (2010) tested environmental Kuznets curve for 43 developing nations. GDP and CO₂ emissions was used to validate environmental Kuznets curve and judgement was made based on short run higher income coefficient than that of long run. They found environmental Kuznets curve was valid for 35% of sample countries. Kivyiro and Arminen (2014) tested nexuses between energy consumption, CO₂ emissions, economic growth and FDI for six Sub-Saharan African countries. GDP and GDP square were used to confirm the validity of environmental Kuznets curve. They found that EKC is valid for three countries namely democratic Republic of Congo, Kenya and Zimbabwe in long run. Pablo-romero and De Jesús (2016) tested the relation between energy consumption and economic growth for 22 Latin American and Carbibean countries and concluded that energy Kuznets curve is not valid for these economies.

Talking about individual country’s studies, Pao et al. (2011) explore the relation among energy use, carbon dioxide emissions and GDP and GDP square for Russian. They fail to find the validity of environmental Kuznets curve and suggested to conserve energy and growth to reduce carbon dioxide emissions. On the other hand, Ahmad et al. (2017) investigate environmental Kuznets curve with the help of GDP, GDP square and CO₂ emissions for Croatia and has confirmed environmental Kuznets curve validity in long run. Study from Lau et al. (2014) has tested environmental Kuznets curve for Malaysia. GDP, GDP square and CO₂ emissions relation confirm the existence of EKC in short as well as in long run. On the other hand, Begum et al. (2015) have investigated the dynamic impacts of economic growth, energy consumption and population growth on CO₂ emissions for Malaysia. They have confirmed using various method that economic growth coefficient was with negative sign and insignificant while its square was with positive and significant marking environmental Kuznets curve is not valid for Malaysia. The energy consumption coefficient was having positive significant impact on CO₂ emissions. They suggested to introduce renewable energy and increase energy efficiency to reduce CO₂ emissions. At another point, Sohag et al. (2015) found inverted U-shape relation between household’s energy consumption and CO₂ emissions in short and long run. They concluded that CO₂ emissions decrease as household energy consumption increases. For Turkey, Ozturk and Acaravci (2013) explore cointegration and causal relation between GDP, GDP square, carbon dioxide emissions, energy consumption and employment ratio. They fail to find the validity of environmental Kuznets curve. On the other hand, Yavuz (2014) took CO₂ emissions, energy consumption, GDP and GDP square in the model and found the validity of EKC in long run for Turkey while fail to find EKC in short run. Ang (2007) concludes that CO₂ emissions and economic growth has quadratic relation in France.

Talking about Chinese economy, there is small but growing literature to investigate the linkages between CO₂ emissions, energy consumption and real GDP in the framework of environmental Kuznets curve. China is not only fastest growing economy but energy dependent economy and highest CO₂ emitters in the world i.e. 29.4% world pollution sources is China. Previous literature offer signal for mix results. For example, Jalil and Feridun (2011) explore cointegration among GDP, GDP square, CO₂ emissions, energy consumption and foreign trade. By utilizing ARDL confirm the validity of environmental Kuznets curve in short as well as in long run. On the other hand, Wang et al. (2011) results show U-shape relation between CO₂ emissions and economic growth for China using 28 China provinces data. They found that GDP is having negative sign while GDP square was with positive sign. They have concluded that China has to sacrifice economic growth to reduce carbon dioxide emissions from the environment. Recently, Dong et al. (2017) have utilized panel data for 30 Chinese provinces with the time spanned 1995–2014 to investigate the relation between CO₂ emissions, GDP per capita, GDP per capita square and natural Gas. Results reveal the existence of an inverted U-shape relation between CO₂ emissions and GPD per capita for Chinse economy.

Above literature review offer footprints that results from different studies are mix and particularly, for Chinese economy there are limited studies offering mix results where researches are using income and income square as independent variables to validate environmental Kuznets curve. This idea seems econometrically weak as income and income square in one model may cause collinearity or multicollinearity problem (Narayan and Narayan 2010) and ignoring this econometrics problem, one may wish to continue, however, it can only offer necessary rather than sufficient condition to confirm U-shape or inverted U-shape relation (Lind and Mehlum 2010). To overcome these issues, this paper aims to investigate environmental Kuznets curve based on short run and long run income’s coefficients sign i.e. short run positive income coefficient and long run negative income coefficient will grant the guarantee of validity of environmental Kuznets curve. In other words, positive short run income coefficient and long run negative income coefficient will validate environmental Kuznets curve as EKC states that initially, rise in income will raise environmental pollution, but after the specific period, further rise in income will reduce environmental pollution. Though the idea seems interesting based on short run positive and long run negative income coefficient signs, however, the problem in such approach will be the unknown shape of Kuznets curve. To counter this issue, this paper introduces cubic form of the income with the flexible function of the model as estimations with cubic form of model offer greater range and flexibility to the model in system.

3 Data, model and estimation procedure

3.1 Data and model

Inspired by recent trend and literature, GDP per capita (constant 2010 US$), CO₂ emissions (metric tons per capita), energy use (kg of oil equivalent per capita) and urban population (total urban population living in urban areas) data has been collected for the period of 1971–2014 from world development indicators (WDI) 2017, World Bank. Variables were converted in logarithmic form to interpret coefficients in elasticities. Data descriptive statistic i.e. Mean, Median, Maximum, Minimum, skewness and standard deviation are reported in Table 1 where Jarque-Bera stats show that all variables are normally distributed as P value is greater than 5%.

Table 1 Descriptive statistics.

In order to test environmental Kuznets curve based on short-run positive and long-run negative income coefficients, CO₂ emissions is treated as dependent variable while real GDP is taken as main independent variable. Energy consumption and urban population are added in the regression to overcome omitted variable bias. General form of the model will be as:

$${\text{CO}}_{{ 2_{t} }} = \beta_{0} + \beta_{1} Y_{t} + \beta_{2} E_{t} + \beta_{3} URB_{t} + \varepsilon_{t}$$

(1)

t is time trend and $\varepsilon$ is white noise error term, CO₂ is carbon dioxide emissions per capita, Y is real GDP per capita, E is energy consumption per capita and URB is urban population, B₀ is constant, B₁ is coefficient of GDP, B₂ is energy consumption coefficient and B₃ is urban population coefficient.

To test the shape of environmental Kuznets curve for China, CO₂ emissions per capita is treated as dependent variable while income per capita, income square per capita and income cubic function as main independent variables with energy consumption and urban population as additional variables to overcome omitted variables bias. Objective was to test environmental Kuznets curve with flexibility function to figure out the shape of environmental Kuznets curve as Pablo-romero and De Jesús (2016) has pointed out that estimations with cubic form of the model offer greater range and flexibility to model in the system. Following Dinda (2004), Akbostanci et al. (2009) and Pablo-romero and De Jesús (2016), cubic form of the model has been introduced as follows:

$${\text{CO}}_{2} = F(Y,Y^{2} ,Y^{3} ,Z)$$

(2)

CO₂ is natural logarithm of carbon dioxide emissions per capita, Y is natural logarithm of real GDP per capita, Y² is natural logarithm of real GDP per capita square, Y³ is natural logarithm of real GDP per capita cubic functions. Z are those factors that can influence pollution emissions in China. We have added energy consumption and urban population to test their role in the explanation of pollution emissions and to overcome omitted variables bias. Additional variables addition is important step to not only overcome omitted variables bias in analysis but also for policy suggestions to test their relation with pollution emissions for China.

Econometric model can be written as:

$$CO_{{2_{t} }} = \chi_{t} + \alpha_{1} Y_{t} + \alpha_{2} Y^{{^{2} }}_{t} + \alpha_{3} Y^{3}_{t} + \alpha_{4} E_{t} + \alpha_{5} URB_{t} + \varepsilon_{t}$$

(3)

E is natural logarithm of energy consumption, Y is natural logarithm of real GDP, Y² is natural logarithm of real GDP square, Y³ is natural logarithm of real GDP in cubic function and URB is natural logarithm of urban population.$\chi_{t}$ is constant, $\alpha_{1} ,\alpha_{2} ,\alpha_{3} ,\alpha_{4} ,\alpha_{5}$ are coefficients of respective variables that will be interpreted into elasticities since data is in log form. $\alpha_{1} ,\alpha_{2} ,\alpha_{3}$ are parameters of real income to test environmental Kuznets curve for China with flexible function. Different income coefficients’ sign will offer different functional forms. Some possible alternative combinations can be as: $\alpha_{1} > 0{\text{ and }}\alpha_{2} = \alpha_{3} = 0$ will yield monotonic increasing relation with economic growth while $\alpha_{1} < 0{\text{ and }}\alpha_{2} = \alpha_{3} = 0$ will show monotonically decreasing relation. $\alpha_{1} < 0,\alpha_{2} > 0{\text{ and }}\alpha_{3} = 0$ will indicate U-shape relation between income and pollution emissions while $\alpha_{1} > 0,\alpha_{2} < 0{\text{ and }}\alpha_{3} = 0$ will show inverted U-shape relation between income and pollution that is typical environmental Kuznets curve validity that marks economic inverted U-shape relation between pollution emissions and real income. If $\alpha_{1} > 0,\alpha_{2} < 0{\text{ and }}\alpha_{3} > 0$, it will indicate N-shape relation between income and pollution emissions. To sum, validity of environmental Kuznets curve can be if $\alpha_{1} { > 0, }\alpha_{2} < 0,\alpha_{3} \le 0$.

3.2 Estimation procedure: ARDL model to cointegration

Autoregressive distributed lag (ARDL) model by Pesaran et al. (2001) has been used to test the relation among variables as ARDL has several advantages on other techniques. Main contribution of ARDL is to solve stationary problem in time series data, otherwise, she has to decide method according to the stationary of data. For example, ordinary least square is usable if all variables in the given model are level stationary i.e. I(0). Usually, time series data shows trend with time and move to first difference stationary i.e. I(1). Engle and Granger (1987) technique can be used for two variables and represents causality between two variables i.e. X is causing Y or Y is causing X etc. Johansen co-integration (1988) method has advantages on Engle and Granger as it can be used for more than two variables, however, it is suitable for large sample and can provide misleading results for small sample and furthermore, pre-condition is the intergration of variables in same order i.e. I(1). In most cases, some variables get stationary at level and some move to first difference stationary. In this situation, OLS, Engle and Granger and Johansen co-integration methods are not applicable. To overcome pre-mentioned issues, Pesaran et al. (2001) proposed method i.e. Autoregressive distributed lag (ARDL) model can be ideal choice as it can be used for purely I(0) variables or purely I(1) or mixture of both I(0) and I(1) variables (Pesaran et al. 2001). This approach provides unbiased results in small sample (Pesaran and Shin 1999). In ARDL model, error correction model (ECM) can be obtained through simple OLS and this ECM shows short run to long run equilibrium in case of disturbance in equilibrium (Laurenceson and Chai 2003). Instrumental variables are employed to avoid endogenity issue, however, there is no ideal variable to be used as an ideal instrument. The best way to tackle this issue is to introduce lags of variable (s) by making regression dynamic (Borensztein et al. 1998) and ARDL makes model dynamic by introducing lags in regression (Pesaran and Shin 1999). Thus, ARDL has several advantages as it can be used irrespective to the variables integration order i.e. I(0) or I(1) or mixture of both and usually variable (s) does not move to second difference stationary, particularly, it is rare in small sample. It seems obvious that stationary property of variables is less important while using ARDL approach and if even any variable (s) is turning to second difference stationary, still ARDL is applicable by taking difference of second difference variable and then using it in the model (Ahmad and Du 2017). Thus, stationary property is less important while dealing with ARDL.

Having advantages, ARDL has been applied with the given unrestricted error correction regressions:

$$\Delta CO_{{2_{t} }} =\, \beta_{0} + \sum\limits_{i = 1}^{n1} {B_{1i} \Delta } CO_{{2_{t - i} }} + \sum\limits_{i = 0}^{n2} {\beta_{2i} } \Delta Y_{t - i} + \sum\limits_{i = 0}^{n3} {\beta_{3i} } \Delta E_{t - i} + \sum\limits_{i = 0}^{n4} {\beta_{4i} } \Delta URB_{t - i} + \delta_{0} CO_{{2_{t - 1} }} + \delta_{1} Y_{t - 1} + \delta_{2} E_{t - 1} + \delta_{3} URB_{t - 1} + \varepsilon_{3t}$$

(1.3)

$$\Delta Y_{t} =\, \beta_{0} + \sum\limits_{i = 1}^{n1} {B_{1i} \Delta } Y_{t - i} + \sum\limits_{i = 0}^{n2} {\beta_{2i} } \Delta CO_{{2_{t - i} }} + \sum\limits_{i = 0}^{n3} {\beta_{3i} } \Delta E_{t - i} + \sum\limits_{i = 0}^{n4} {\beta_{4i} } \Delta URB_{t - i} + \delta_{0} Y_{t - 1} + \delta_{1} CO_{{2_{t - 1} }} + \delta_{2} E_{t - 1} + \delta_{3} URB_{t - 1} + \varepsilon_{4t}$$

(1.4)

$$\Delta E_{t} =\, \beta_{0} + \sum\limits_{i = 1}^{n1} {B_{1i} \Delta } E_{t - i} + \sum\limits_{i = 0}^{n2} {\beta_{2i} } \Delta CO_{{2_{t - i} }} + \sum\limits_{i = 0}^{n3} {\beta_{3i} } \Delta Y_{t - i} + \sum\limits_{i = 0}^{n4} {\beta_{4i} } \Delta URB_{t - i} + \delta_{0} E_{t - 1} + \delta_{1} CO_{{2_{t - 1} }} + \delta_{2} Y_{t - 1} + \delta_{3} URB_{t - 1} + \varepsilon_{5t}$$

(1.5)

$$\Delta URB_{t} =\, \beta_{0} + \sum\limits_{i = 1}^{n1} {B_{1i} \Delta } URB_{t - i} + \sum\limits_{i = 0}^{n2} {\beta_{2i} } \Delta CO_{{2_{t - i} }} + \sum\limits_{i = 0}^{n3} {\beta_{3i} } \Delta Y_{t - i} + \sum\limits_{i = 0}^{n4} {\beta_{4i} } \Delta E_{t - i} + \delta_{0} URB_{t - 1} + \delta_{1} CO_{{2_{t - 1} }} + \delta_{2} Y_{t - 1} + \delta_{3} E_{t - 1} + \varepsilon_{6t}$$

(1.6)

$$\Delta CO_{{2_{t} }} =\, \beta_{0} + \sum\limits_{i = 1}^{n1} {B_{1i} \Delta } CO_{{2_{t - i} }} + \sum\limits_{i = 0}^{n2} {\beta_{2i} } \Delta Y_{t - i} + \sum\limits_{i = 0}^{n3} {\beta_{3i} } \Delta Y_{t - i} + \sum\limits_{i = 0}^{n4} {\beta_{4i} } \Delta Y_{t - i} + \sum\limits_{i = 0}^{n5} {\beta_{5i} } \Delta E_{t - i} + \sum\limits_{i = 0}^{n6} {\beta_{6i} } \Delta URB_{t - i} + \delta_{0} CO_{{2_{t - 1} }} + \delta_{1} Y_{t - 1} + \delta_{2} Y_{t - 1} + \delta_{3} Y_{t - 1} + \delta_{4} E_{t - 1} + \delta_{5} URB_{t - 1} + \varepsilon_{5t}$$

(2.1)

In Eq. 1.3, ∆ is difference operator, t is time trend, $\varepsilon$ error term, β₁–β₄ are dynamic of error corrections. In second part of Eq. 1.3, δ₀–δ₃ are long run coefficients of carbon dioxide emissions, real GDP, energy consumption and urban population respectively. Other Eqs. (1.4, 1.5 and 1.6) can be explained in similar way. Moving to Eq. 2.1, ∆ is difference operator, t is time trend, $\varepsilon$ error term, β₁–β₆ are dynamic of error corrections. Second part of Eq. 2.1, δ₀–δ₅ are long run coefficients of carbon dioxide emissions, real GDP, real GDP square, real GDP cube, energy consumption and urban population respectively. This equation will be utilized to check the shape of environmental Kuznets curve for China.

First task is to estimate each equation by ordinary least squares to obtain coefficients. Next step is to compute f-statistics through wald test to test the validity of co-integration among variables. Null of no cointegration or no long run relation will be as H₀: δ₀ =  δ₁ =  δ₂ =  δ₃ = 0 against alternative H₁:δ₀ ≠ δ₁ ≠ δ₂≠ δ₃≠ 0 for each equation (equation from 1.3 to 1.6) while for Eq. 2.1, null of no cointegration will be as H₀:δ₀= δ₁=  δ₂=  δ₃= δ₄= δ₅= 0 against alternative H₁:δ₀≠  δ₁≠ δ₂≠ δ₃≠ δ₄≠  δ₅≠  0. Third step is to compare calculated f-statistics with the critical values given by Pesaran et al. (2001). Two sets of critical values are available for I(0) and I(1) variables. Lower critical bounds (LCB) are for I(0) variables and upper critical bounds (UCB) are for I(1) variables. If calculated f-statistics is higher than the upper bounds, null of no cointegration will be rejected against alternative. If calculated f-statistics is below the lower bounds, in this case, null of no cointegration cannot be rejected. However, calculated f-statistic between UCB and LCB will bring inconclusive results. In this case, negative and significant error correction term (ECT) will confirm long run relationship among series.

Once long run relation has been confirmed, next step is to estimate error correction model (ECM). General form of ECM model for Eqs. 1.3, 1.4, 1.5, 1.6 and 2.1 will be in the form of Eqs. 1.7, 1.8, 1.9, 1.10 and 2.2. respectively.

$$\Delta CO_{{2_{t} }} = \beta_{0} + \sum\limits_{i = 1}^{n1} {B_{1i} \Delta } CO_{{2_{t - i} }} + \sum\limits_{i = 0}^{n2} {\beta_{2i} } \Delta Y_{t - i} + \sum\limits_{i = 0}^{n3} {\beta_{3i} } \Delta E_{t - i} + \sum\limits_{i = 0}^{n4} {\beta_{4i} } \Delta URB_{t - i} + \eta_{1} ECT_{t - 1} + \varepsilon_{t}$$

(1.7)

$$\Delta Y_{t} = \beta_{0} + \sum\limits_{i = 1}^{n1} {B_{1i} \Delta } Y_{t - i} + \sum\limits_{i = 0}^{n2} {\beta_{2i} } \Delta CO_{{2_{t - i} }} + \sum\limits_{i = 0}^{n3} {\beta_{3i} } \Delta E_{t - i} + \sum\limits_{i = 0}^{n4} {\beta_{4i} } \Delta URB_{t - i} + \eta_{2} ECT_{t - 1} + \varepsilon_{t}$$

(1.8)

$$\Delta E_{t} = \beta_{0} + \sum\limits_{i = 1}^{n1} {B_{1i} \Delta } E_{t - i} + \sum\limits_{i = 0}^{n2} {\beta_{2i} } \Delta CO_{{2_{t - i} }} + \sum\limits_{i = 0}^{n3} {\beta_{3i} } \Delta Y_{t - i} + \sum\limits_{i = 0}^{n4} {\beta_{4i} } \Delta URB_{t - i} + \eta_{3} ECT_{t - 1} + \varepsilon_{t}$$

(1.9)

$$\Delta URB_{t} = \beta_{0} + \sum\limits_{i = 1}^{n1} {B_{1i} \Delta } URB_{t - i} + \sum\limits_{i = 0}^{n2} {\beta_{2i} } \Delta CO_{{2_{t - i} }} + \sum\limits_{i = 0}^{n3} {\beta_{3i} } \Delta Y_{t - i} + \sum\limits_{i = 0}^{n4} {\beta_{4i} } \Delta E_{t - i} + \eta_{4} ECT_{t - 1} + \varepsilon_{t}$$

(1.10)

$$\Delta CO2_{t} = \beta_{0} + \sum\limits_{i = 1}^{n1} {B_{1i} \Delta } CO_{{2_{t - i} }} + \sum\limits_{i = 0}^{n2} {\beta_{2i} } \Delta Y_{t - i} + \sum\limits_{i = 0}^{n3} {\beta_{3i} } \Delta Y_{t - i} + \sum\limits_{i = 0}^{n4} {\beta_{4i} } \Delta Y_{t - i} + \sum\limits_{i = 0}^{n5} {\beta_{5i} } \Delta E_{t - i} + \sum\limits_{i = 0}^{n6} {\beta_{6i} } \Delta URB_{t - i} + \eta_{1} ECT_{t - 1} + \varepsilon_{t}$$

(2.2)

In Eq. 1.7, ECT is error correction term that shows speed of adjustment. It shows disturbance in equilibrium will take how much time to reach to its equilibrium path. $\eta$₁ is coefficient of error correction term that direct the time duration to converge system in long run equilibrium path. Other Eqs. (1.8, 1.9, 1.10 and 2.2) can be interpreted in the similar way.

4 Empirical results and discussion

Although ARDL is flexible method and can be used irrespective to I(0) and I(1) variables, and generally, variable (s) does not move to second difference stationary and particularly, for small sample size. However, to be safe side to avoid for I(2) variable (s) and to overcome Ouattara (2004) comment that ARDL cannot be used for I(2) variable since critical bounds produced by Pesaran et al. (2001) are based on I(0) and I(1) variables. Three different unit root tests (Dickey and Fuller 1979, 1981; Phillips and Perron 1988 and Kwiatkowski et al. 1992) has been applied and found none of the variable was second difference stationary. To conserve space, results are not reported, however, are available upon request. In ARDL estimation, first step was to estimate each equation by ordinary least squares to obtain coefficients. Next step was to compute f-statistics through wald test to test the validity of co-integration among variables. F-statistic results are reported in Table 2. First, CO₂ emissions is treated as dependent variable and income, energy consumption and urban population as independent variables. For this equation, f-statistic was found 4.307 that is higher than the upper bounds at 5 percent level of significance. It confirms long run relationship among variables. Results also reveal the existence of cointegration when real GDP and energy consumption was treated as dependent variable for their corresponding equation since f-statistic was higher than the upper bounds in each equation. However, we fail to find cointegation when urban population is treated as dependent variable as f-statistic was lower than the lower bounds. The f-statistic was found statistically highly significant at 1% level of significance with coefficient 4.112 when CO₂ emissions was treated as dependent variable and income cubic function has been introduced in the model to confirm the shape of relation between pollution emissions and income in China.

Table 2 Bound tests for the existence of cointegration

After confirming co-integration among variables, next step was to check short-run and long-run income coefficients when CO₂ emissions was treated as dependent variable to confirm the validity of environmental Kuznets curve for China. Table 3 shows that a 1% increase in economic growth raises CO₂ emissions by 0.47% in short run while long run income coefficient reveals interesting fact by turning negative (− 0.13) and statistically insignificant. At one side, long run coefficient is negative that confirm the validity of environmental Kuznets curve for China. At second side, it turns to insignificant that means only growth dependency to reduce pollution emissions will not be appropriate choice for China. As, China is not only the largest developing economy but high performing economy and highest CO₂ emitters in the world. So, there is need to depend on multiple polices to overcome pollution emissions from China. IEA (2009) has reported that China is facing high pressure of CO₂ emissions worldwide. Xu et al. (2014) has viewed that low CO₂ emissions is not only better for China but also for the rest of world. Moving to other results, it is shown that energy consumption raises CO₂ emissions at high rate from short to long run as energy consumption coefficient is becoming fatter with the passage of time and in long run it is unitary. In this situation, one may think direct option to reduce energy consumption to overcome CO₂ emissions but this energy reduction will hurt economic growth of China. Solution lies to introduce cleaner energy like renewable energy, bioenergy and nuclear energy. Results show that urban population coefficient is insignificant in short and long run. Its insignificancy gives the impression that urban population is less important determinants of CO₂ emissions as compare to energy consumption and economic growth. The error correction term is statistically significant at 10% level of significance having coefficient 0.19 that shows that the disturbance in system will be adjusted 19% in first year. This low adjustment speed (19% yearly) again gives signal to activate environmental policies to control emissions rather than just depending on economic growth to fight with environmental pollution. Of-course, growth and development is one input factors and thus, growth reduction policy will not be suggested in the given situation. The goodness of fit model has been judged with different diagnostic tests reported in Table 4 that confirm model is free from serial correlation and heteroskedasticity issue. Jarque-Bera test for normality fails to reject the null hypothesis that means the estimated residuals are normally distributed. Ramsey reset test confirms model is correctly specified. Stability of the coefficients is checked through CUSUM and CUSUM square given by Brown et al. (1975). The CUSUM (Fig. 4) and CUSMSQ (Fig. 5) show that variables in the model are stable in long periods of time as critical bounds remain within the bounds. Overall, model is reliable for policy purpose.

Table 3 Short-run and long run results when EKC confirmation is based on short run and long run income coefficients’ sign.

Table 4 Diagnostic tests for the equation when EKC confirmation is based on short run and long run income coefficients’ sign.

Fig. 4

CUSUM test for stability

Fig. 5

CUSUM square for stability

Results reported in Table 5 for Eq. 3 to test the shape of relation between environmental pollution and income reveal several facts as: in short run income coefficient was negative, income square coefficient was positive and income cubic coefficient again turns to negative marking that in short run, initially, rise in income reduces pollution emissions, later rise in income raises pollution emissions and finally with the passage of time rise in income helps to reduce pollution emissions that seem to suggest inverted N-shape relation between pollution emissions and income for China. However, this short term trend is not an adequate to guide about policy for long run. Moving to long run results, surprisingly, income’s coefficients turn inverse. Results show that initially, rise in income raises pollution emissions, later, it reduces with the rise of income and again finally, this relation turns to positive again. Overall, marking N-shape relation between income and pollution emissions in long run. Figure 6 clarify that initially rise in income raises pollution emissions and after reaching to the peak, this relation turns to inverse where further rise in income is helping in pollution reduction and finally, this relation again turns to positive marking N-shape relation between the two variables. It seems to suggest that Chinese CO₂ emissions will not be reduced only by the growth and development and therefore, alternative measures are needed. On the other hand, it also shows that Chinese government have stricter policies to overcome pollution emissions. More efficient technology should be applied to overcome CO₂ emissions. Increase in natural gas consumption can help to reduce pollution emissions from China. China is always committed to reduce its CO₂ emissions level. For example, in 2009, in Copenhagen talk, China has made commitment to reduce carbon intensity i.e. CO₂ emissions/GDP around 40–45% till 2020 that shows China’s commitment to reduce its pollution emissions. Additionally, Chinese government ambitious plan to work on CO₂ emissions to curb it up to 15% by increasing share of non-fossil energy in primary energy around 20% till 2030. China has set targets to reduce its pollution emissions intensity per capita GDP around 40–60% till 2030 as compared to 2005. In this aspect, though relation between pollution emissions and growth is encouraging, however, it appeals to focus on environmental policies rather than only growth and development dependency. Results show that in short and long run, energy consumption coefficient is near to unitary and highly significant. It reveals the fact that energy consumption is an important determinant of pollution emissions that raises pollution emissions in China with high rate. Being industrialize economy, it will not be an appropriate choice for China to reduce energy consumption to overcome pollution emissions. Solution can be the introduction of cleaner energy like renewable energy, bioenergy and nuclear energy. The urban population variable is found statistically highly significant with negative coefficient in short and long run, revealing that urbanization process is helping to support pollution reduction. Recalling results from Eq. 1, urban population was found insignificant with positive sign in short and long run while in Eq. 3, urban population does not only turn to significant but also negative marking urbanization process is helpful to reduce pollution emissions. Error correction term is found highly significant at 1% level of significant with the coefficient size − 0.86 showing system will show quick converge to long run equilibrium path if disturbance occurs. The model with income cubic function seems more fit and closer to reality for China with high speed of adjustment rather than first model where ECT was found with low adjustment speed. In the second model, ECT was highly significant and with high adjustment rather than low adjustment with 10% level of significance. Error correction term seems quite sensitive to model selection. The goodness of fit for model 3 has been confirmed with different diagnostic reported in Table 6 that confirm model is free from serial correlation and heteroskedasticity issue. Jarque-Bera test for normality confirm the estimated residuals are normally distributed. Ramsey reset test confirms model is correctly specified. The CUSUM (Fig. 7) and CUSMSQ (Fig. 8) show variables in the model are stable in long periods of time as critical bounds remain within the bounds.

Table 5 Short-run and long run results with income cubic function to test EKC shape.

Fig. 6

N-shape relation between pollution emissions and income level

Table 6 Diagnostic tests for the equation to test N-shape Kuznets curve.

Fig. 7

CUSUM test for stability

Fig. 8

CUSUM square for stability

4.1 Robustness check via FMOLS and GMM

Fully modified ordinary least squares (FMOLS) and generalized method of moment (GMM) has been employed for the robustness of long run coefficients. FMOLS estimators are free from endogenity issue, serial correlation problem and provide unbiased results in small sample (Phillips and Hansen 1990; Stock and Watson 1993). Similarly, GMM overcomes endogenity issue by making model dynamic with the introduction of lags in the model. Results in Table 7 for Eq. 1 show that income variable turns to significant in long run and a 1% increase in income reduces pollution emissions by 0.35% in long run while GMM results reveal that a 1% increase in income will help to reduce pollution emissions by 0.38% in long run. It reconfirms the validity of environmental Kuznets curve in China though to depend on multiple policies. There was a positive and significant relation between energy consumption and CO₂ emissions in long run with the elasticity range 1.11–1.12 in FMOLS and GMM respectively that was in line with ARDL results. Urban population coefficient remains positive and turns to significant with coefficient range 0.81–0.91 in long run. It shows a 1% increase in urban population raises pollution emissions by 0.81% in FMOLS while GMM shows a similar increase in urban population raises pollution emissions by 0.91% in long run that seems to suggest that rise in urban population increases pollution emissions in China. What about Eq. 3? FMOLS and GMM estimators show that income coefficient is with positive sign, income square coefficient is with negative coefficient and income cubic coefficient again turn to positive confirming the existence of N-shape relation between income and pollution emissions in China. This N-shape relation mark that initial rise in income raises pollution emissions, then this relation turns to inverse and finally, it moves to positive again stressing to introduce environmental policy rather than only income dependency to fight with environmental pollution. Further, energy coefficient was found with 1.17 in FMOLS and 1.23 in GMM confirming that in long run rise energy consumption raises pollution emissions with the given high elasticities. There is need to introduce alternative energy sources like renewable energy consumption. Urban population was found negative and highly significant showing that rise in urban population reduces pollution emissions in long run. Overall, FMOLS and GMM confirm the ARDL results for both equations.

Table 7 Robustness check

4.2 Causality analysis

ARDL model offers the presence or absence of co-integration between variables but cannot explain the direction of causality. To overcome this, two step procedure has been adopted from Granger (1988) to test the causal relation between variables i.e. X is causing Y or Y is causing X. In the absence of co-integration, vector autoregressive (VAR) model is used to check causal relation among variables, however, in the presence of co-integration, conventional VAR can provide misleading results (Loizides and Vamvoukas, 2005). By following Granger (1988), we use two step procedure to check short run and long run causal relation: In the presence of co-integration, we obtain error correction term (ECT) through long run equation (Eq. 1) and use it with its lag as additional independent variable in the model. The short run causality/weak causality is detected through F-statistic or Wald test with first difference variables. The negative and significance error correction term (ECT) will show long run equilibrium. Thus, long run causality will be detected through t-test of lagged error correction term in each equation.

The general form of VECM will be as follows:

$$(1 - L)\left[ \begin{array}{llll} CO_{{2_{t} }} \hfill \\ \, Y_{t} \hfill \\ \, E_{t} \hfill \\ URB_{t} \hfill \\ \end{array} \right] = \left[ \begin{array}{llll} \beta_{1} \hfill \\ \beta_{2} \hfill \\ \beta_{3} \, \hfill \\ \beta_{4} \hfill \\ \end{array} \right] + \sum\limits_{i = 1}^{q} {(1 - L)} \left[ \begin{array}{llll} C_{11,i} {\text{ C}}_{{ 1 2 , {\text{i}}}} {\text{ C}}_{{ 1 3 , {\text{i}}}} {\text{ C}}_{{ 1 4 , {\text{i}}}} \hfill \\ C_{21,i} {\text{ C}}_{{ 2 2 , {\text{i}}}} {\text{ C}}_{{ 2 3 , {\text{i}}}} {\text{ C}}_{{ 2 4 , {\text{i}}}} \hfill \\ C_{31,i} {\text{ C}}_{{ 3 2 , {\text{i}}}} {\text{ C}}_{{ 3 3 , {\text{i}}}} {\text{ C}}_{{ 3 4 , {\text{i}}}} \hfill \\ C_{41,i} {\text{ C}}_{{ 4 2 , {\text{i}}}} {\text{ C}}_{{ 4 3 , {\text{i}}}} {\text{ C}}_{{ 4 4 , {\text{i}}}} \hfill \\ \end{array} \right]\left[ \begin{array}{llll} CO_{{2_{t - i} }} \hfill \\ Y_{t - i} \hfill \\ E_{t - i} \hfill \\ URB_{t - i} \hfill \\ \end{array} \right] + \left[ \begin{array}{llll} \delta_{1} \hfill \\ \delta_{2} \hfill \\ \delta_{3} \hfill \\ \delta_{4} \hfill \\ \end{array} \right]ECT_{t - 1} + \left[ \begin{array}{llll} \varepsilon_{1t} \hfill \\ \varepsilon_{2t} \hfill \\ \varepsilon_{3t} \hfill \\ \varepsilon_{4t} \hfill \\ \end{array} \right]$$

(1.11)

$(1 - L)$ is lag operator, $ECT_{t - 1}$ is error corrector term with one period lag, $\varepsilon_{t} 's$ are normally distributed residuals term with zero mean and constant variance. Generally, VECM allow to create difference between short run and long run causality. Wald test with differenced variables give short run causal information while t-test of lagged error correction term contains long run causal relation information. Results in Table 8 show bi-directional causality between energy consumption and economic growth in short and long run that shows energy and economic growth are complement of one another in short as well as in long run meaning that energy consumption will boost economic growth and high economic growth will cause high energy consumption. The existence of bi-directional causality between energy and CO₂ emissions indicate that China should reduce energy consumption to reduce CO₂ emissions from environment. At one side, this energy consumption is raising CO₂ emissions and at second side, it is the source of economic growth of China. As more energy is needed to sustain China’s growth and development, so, it is necessary to increase the share of clean energy. Improving energy efficiency can be the best solution to reduce CO₂ emissions and to maintain high economic growth. Uni-directional causality from CO₂ emissions to income in short run implies that decrease in pollution emissions will reduce income level as being industrialize economy, China depends on energy consumption and this energy injects pollution in the environment. It gives impression that reduction in pollution will negatively influence growth. Alternative energy sources will be helpful tools to overcome pollution issue and will not hurt growth and development. Uni-directional causality running from income to urban population reveals that rise in income level will encourage urbanization process in China. Uni-directional causality from urban population to energy consumption reveals that urbanization process will raise energy consumption in China. The negative and significant ECT for first three equations show bi-directional causality among variables in long run. Range of adjustment for first three equations remain between 0.42 to 0.55 that reveal the convergence will take place roughly within 2 years in the system in case of disturbance.

Table 8 Granger causality results

5 Conclusion and policy implications

This paper has extended previous discussion on the relationship between energy consumption, CO₂ emissions and economic growth for China under the environmental Kuznets curve framework. In this paper, environmental Kuznets curve for China has been investigated in two novel ways: one was to make judgment based on short-run and long-run income coefficients’ sign. It is to say if short run income coefficient sign is positive and long run income coefficient sign is negative, we can confirm the validity of environmental Kuznets curve as it states that “At initial level, rise in income will raise environmental pressure but after the specific period, further increase in income will decrease environmental pressure”. The reason behind the reversion of income sign is that as initially in short run, with the economic growth, environmental pollution will increase but after the specific period (when move to long run), this relation turns to inverse. Making judgement of environmental Kuznets curve based on short run and long run income coefficients sign will not only overcome the multicollinearity problem in the model that can be because of income and income square or income-cubed in one model but will make the model irrespective from the necessary and sufficient condition. However, problem in the idea may raise the issue for the shape of environmental Kuznets curve. So, in next step, income cubic function has been introduced in the model that offer flexible function in the model to test the shape of environmental Kuznets curve for China.

Annual data has been used for the period of 1971–2014 for CO₂ emissions per capita that is treated as dependent variable while real GDP per capita, energy consumption per capita and urban population are used as independent variables. ARDL findings were unique and offer several facts as: economic growth coefficient was positive and significant in short run that shows that initial increase in economic growth raises pollution emissions in China while in long run coefficient turns to negative and insignificant. At one side, it is the confirmation for the validity of environmental Kuznets curve in China and at second side, insignificant coefficient in long run gives signal that there is need to activate environmental policies instead of only depending on economic growth to fight with pollution. These results make sense that only depending on economic growth to fight with pollution will not be an appropriate choice for China but other environmental policies are needed to control environmental pollution as China is not only largest economy but also highest CO₂ emitters. The reversion of sign from short run (positive sign) to long run (negative sign) confirm the validity of EKC in China. FMOLS and GMM results show that economic growth coefficient turns to significant in long run that reconfirm the validity of EKC for China. By introducing income cubic function in model, there was the existence of N-shape relation between income and environmental pollution that remark to activate environmental policies. Results show that increase in energy consumption rises pollution emissions and its coefficient was becoming fatter over the time. In this situation, direct option can be thought to reduce energy consumption to overcome environmental pollution. However, this energy reduction will hurt economic growth of China. Solution lies in the introduction of cleaner energy like renewable energy, bioenergy and nuclear energy. Results reveal that urban population coefficient is positive and statistically insignificant that seems to suggest that energy consumption and economic growth are more important determinants of pollution emissions than that of urban population. However, by introducing income cubic function in the model, urban population coefficient does not turn significant but also negative, marking that urbanization process is helping to reduce CO₂ emissions from China. Error correction terms was negative and statistically significant that show the disturbance in the system will be corrected 19% annually when income was introduced in the model to test EKC based on short run and long run income coefficient sign. This low adjustment speed (19% yearly) again gives signal to activate environmental policies to control emissions instead of just depending on economic growth to fight with environmental pollution. By introducing income cubic function, error correction improves and becomes − 0.86 showing fast speed of adjustment and system convergence less than a year to long run in case of disturbance in the system. The highly significant and fast adjustment seems to suggest model with cubic function is more reliable for China. FMOLS and GMM confirm the results from ARDL are robust for both models. Different diagnostics tests confirm the perfectness of models.

Granger causality results show bi-directional causality between energy consumption and economic growth in short and long run that shows energy and economic growth are complement of one another in short and long run meaning that energy consumption will boost economic growth and economic growth will also depend on energy consumption. The existence of bi-directional causality between energy and CO₂ emissions suggest that China should reduce its energy consumption to reduce CO₂ emissions from environment. At one side, this energy consumption is raising CO₂ emissions and at second side, it is the source of economic growth for China. As more energy is needed to sustain China’s growth and development, so, it is necessary to increase the share of clean energy. Improving energy efficiency can be the solution to reduce CO₂ emissions and maintain high economic growth. Uni-directional causality from CO₂ emissions to income in short run implies that decrease in pollution emissions will reduce income level that shows being industrialize economy, China depends on energy consumption and this energy injects pollution in the environment. It gives impression that reduction in pollution will negatively influence growth. Alternative energy sources will be helpful to overcome the issue and economic growth and development will not be hurt. Uni-directional causality from income to urban population reveals that rise in income level will encourage urbanization process in China. The existence of uni-directional causality between urban population to energy consumption reveals that urbanization process will raise energy consumption in China. Policy suggestions from the results are: (1) Chinese economic growth is supportive to fight with pollution emissions since there are the signals for the validity of environmental Kuznets curve in China, however, results point out that there is need to dependent on multiple policies to overcome pollution problem. Further, at one side, energy consumption is the source of growth and development and at second side, it is causing pollutions emissions with high rate as coefficient was becoming fatter from short to long run. In this situation, ideal solution will be the introduction of cleaner energy like renewable energy, bioenergy and nuclear energy that will not hurt growth and development and will tackle environmental pollution problems. Urbanization is important not only for the growth and development but also for the pollution reduction from China. In future research, different pollution indicators can be used, for example SO2, NH4 or carbon footprints to test the impact of income level on them. Further, more additional variables can be incorporated in the regression like foreign direct investment, tourism, financial development and son on. It will also be equally interesting to explore this relation based on firm level data or provincial level considering provincial heterogeneity across the Chinese provinces.