Abstract
This study investigates the role of urbanization in carbon dioxide (CO2) emission in the case of the Group of Seven (G7) countries, except Germany. The conventional environmental Kuznets curve (EKC) was revised by augmenting and controlling for urbanization for this purpose. The results of this study confirm the long-term effects of urbanization on CO2 emission in the selected G7 countries. Only France and Italy seem to manage environmental issues when promoting urbanization; this is because urban development exerts negative effects on the climate levels in these two countries, while it generates positive impacts in the case of the other G7 countries. The inverted U-shaped and revised EKC is also successfully confirmed for only France and Italy among the others.
Introduction
Cities increasingly expand their boundaries as a result of growing population sizes; as Poumanyvong and Kaneko 1 mention human history is defined as the history of urbanization. Increased industrialization and urbanization in recent years have dramatically affected the number of urban buildings which resulted in major effects on energy consumption. Therefore, the role of urbanization as a particular segment in interactions between environmental quality and energy consumption has recently attracted considerable attention from researchers. Urbanization exerts several independent influences on energy use2–4; it constantly displaces traditional use of energy with modern types, substantially raising the energy intensity of some activities such as air jets for pollution purification in underground highways and urban transportations. 5 On the other hand, urbanization has been a major demographic trend worldwide solely in the case of Africa and Asia with potential and major consequences for better development. This issue is solely important for energy policy and planning. Urbanization plays a key role in what has become known as the “energy transition”.6,7 The urbanization process has a direct and noteworthy impact on energy use, which is usually quantified by energy intensity. 8
The empirical link between urbanization and energy consumption has been extensively investigated in the energy economics literature.7–14 The vast majority of studies confirm that urban development leads to increased energy use at both per capita and per unit of output and that aggregate energy use rises with the level of urbanization, signaling a strong link between consumption of energy and emissions of greenhouse.15,16
On the other hand, the validity of the environmental Kuznets curve (EKC) hypothesis has been tested in many works of literature, which examined the interactions between gross domestic product (GDP) and environmental pollution.17–23 In further studies, carbon dioxide (CO2) emission is related to the level of income per capita, population size, urban population, industrial structure, and energy intensity of each country. 18 The results of empirical studies generally show urbanization drives CO2 emission, following an EKC, as predicted by the ecological modernization theory (EMT). 18
Growth in CO2 emissions is likely to occur as a result of urban development, through construction, modernization, and industrialization, as well as other aspects of urban areas. 24 For example, Al-mulali et al. 25 find a long-run, bidirectional positive relationship between urbanization, energy consumption, and CO2 emission in the Middle East and North Africa (MENA) region. But, the nature of the long-run relationship among urbanization, energy consumption, and CO2 emission varies across the MENA countries based on their respective levels of income and development. Poumanyvong and Kaneko, 1 on the other hand, investigate the relationship among urbanization, energy consumption, and CO2 emission in 99 countries as divided into three groups (low-, middle-, and high-income); their findings suggest a positive effect of urbanization on CO2 emission for all of the income levels although the results for the middle-income group was more expressed. Zhang and Lin 26 find that in China, the effects of urbanization on energy consumption vary across regions and decline continuously from the western to the central and eastern regions. However, the impact of urbanization on CO2 emission in the central region was greater than that of the eastern region. Zhang and Lin 26 find that the effect of urbanization on energy consumption is greater than its impact on CO2 emission in the eastern region.
Shi 27 finds a direct relationship between population changes and CO2 emissions in 93 developing countries varying with their respective levels of affluence. Shi 27 also shows that the impact of the population on CO2 emission is more pronounced in lower-income than in higher-income countries. Hossain 28 empirically examine the dynamic causal relationships among CO2 emission, energy consumption, economic growth, trade openness, and urbanization for the panel of newly industrialized countries (NICs). Although Hossain 28 finds no evidence of long-run causal relationships, short-run causal relationships running (1) from economic growth and trade openness to CO2 emission, (2) from economic growth to energy consumption, (3) from trade openness to economic growth, (4) from urbanization to economic growth, and finally, (5) from trade openness to urbanization are found.
Urbanization may have a statistically significant impact on environmental quality, as mentioned above. However, many panel-data-based studies focus on panel data analysis rather than country-level time series analysis which would exhibit better conclusions. Although heterogeneity and cross-section dependency in panel data are controlled by various econometric approaches, time-series analyses with a longer span of data are likely to provide better individual conclusions for countries. Against this backdrop, this article investigates the role of urbanization in the conventional EKC in the selected G7 countries using the latest available econometric procedures. The present study contributes to the relevant literature in several ways: Firstly, the G7 countries are the UK, the US, France, Germany, Italy, Canada, and Japan where they hold a significant share in the global economy, overall population, and urban population as presented in Table 8 as of 2020. Therefore, studying the link between urbanization and pollution would be interesting for the G7 group. Secondly, on the other hand, urbanization and high urban densities might influence economy-wide patterns of resource use and global environmental quality. Particularly, the indirect and direct energy requirements of urban living could significantly contribute to the contrary effects of urbanization in the environment. 29 Therefore, searching for the link between urbanization and environmental quality for developed countries as well as country level-time series analysis again deserves attention from researchers. Thirdly, the present study contributes to the existing literature by paying detailed attention to the G7 group which composes of developed countries; that is, time series analyses are done for each country using the largest span of data available. Thus, time series analyses provide better insights and criticisms than panel data analyses do. And, finally, to the best of our knowledge, there isn’t any study for the link between urbanization and pollution in the case of the G7 countries till the moment although there are some studies testing the validity of the conventional EKC without urbanization effect. Therefore, owing to these significant research gaps in the current literature, the present study is expected to provide important implications.
The rest of this article is structured as follows. The next section presents the theoretical background of the study, Then, the data and methodology is defined; in the penultimate section, the empirical results and discussion are explained, and finally, the conclusion of the study is presented in the final section.
Theoretical setting
CO2 emissions (kt) extensively is the main factor of Environmental pollution, as indicated in the relevant literature.
30
As a starting point of the theoretical background, the study can propose that urbanization might have a determinative role in CO2 emission levels through additional economic activities and energy consumption. The main determinant of CO2 emission is GDP based on the framework of conventional EKC. Squared GDP (GDP2) was extensively used in this study because some countries might have an inverted U-shaped EKC, although rare studies considered the third power of GDP in the framework of EKC. Some works of literature of energy economics area suggested energy consumption is playing an important role as a determinative of CO2 emission, which was added to the relationship between GDP and CO2 emission regards to the framework of conventional EKC, in the case that the EKC might still exist as a pattern of an inverted U-shaped.
31
Therefore, the urbanization-induced EKC model was suggested in this study is presented as follow:
Where CO2 refers to carbon dioxide emission (kt), E signifies energy consumption (kt of oil equivalent), GDP is a gross domestic product, and OU stands for the overall urbanization proxy. The parameters of
To capture the impacts of growth over the economy in the long-term period, a double logarithmic regression equation of the urbanization-induced EKC model in equation (1) could be expressed32,33
Where the term “ln” stands for the natural logarithm of regressors in equation (2) at period t, while
Regards to equation (2), the dependent variable might not immediately adjust to its long-term equilibrium path, following any changes in its determinants. Therefore, between the short-run and the long-run levels of the dependent variable the speed of adjustment could be captured by estimating the following error correction model:
Where
Data and methodology
Data
The data used in this research were annual figures covering the 1960–2016 period, 35 and the variables were CO2 emission (kt), energy use (E) (kt of oil equivalent), constant GDP (2010 = 100), squared constant GDP (GDP2) (2010 = 100), and percentage of the urban population in total population in the G7 countries. The G7 countries under consideration in this study were the UK, the US, France, Italy, Canada, and Japan. The data for Germany were not available before 1992 (when East and West Germany merged); therefore, Germany is omitted in this study.
Methodology
This study investigated the urbanization-induced EKC model in the G7 countries, except Germany. Gauss codes which are the second-generation of econometric procedures had been adopted in this study to take consideration of multiple structural breaks. In the first stage, the quasi-generalized least squares (GLS)-based unit root tests (which were developed by Carrion-i-Silvestre et al. 36 ) had been applied for the time series for this study which is considering multiple structural breaks up to five, because multiple breaks might be exhibit in series over the years, as can also be observed in Figures 1 to 6. As a second stage, cointegration tests, 37 had been accomplished to confirm that there is a cointegrating vector in equation (2) which again considered multiple structural breaks until five. As a third step, long-run and short-run models plus ECTs had been applied by using the Autoregressive Distributed Lag (ARDL) method. Finally, for providing further support to the earlier results of this study, Granger causality tests through the block exogeneity approach, impulse response functions, plus variance decompositions had also been examined.

Time series plot of variables under consideration in the natural logarithm in Canada.

Time series plot of variables under consideration in the natural logarithm in France.

Time series plot of variables under consideration in the natural logarithm in Italy.

Time series plot of variables under consideration in the natural logarithm in Japan.

Time series plot of variables under consideration in the natural logarithm in UK.

Time series plot of variables under consideration in the natural logarithm in USA.
Estimation of long-term and short-term coefficients
Once the cointegrating vector is acquired, the next level would be to compute long-term coefficients, as expressed above in equation (2), which would be estimated through the channel of the ARDL approach. Stock and Watson 38 suggested including differenced and lagged structures of independent variables in addition to their level forms, to omit any deviation and problems of internality in the estimators of OLS.
The ARDL models could be employed, regardless of the order of variables’ integration (whether regressors were purely I (0), purely I (1), or mutually cointegrated), but the dependent variable would need to be I (1). Providing the strong ARDL approach would be consistent estimations in the presence of problems of autocorrelation and internality.39,40 The ARDL model would be applied to compute equation (2), which could be presented as follows:
After computing long-term coefficients through the mechanism of ARDL, coefficients of short-term plus the ECT would also be estimated. In addition to equation (3), to meet if their coefficients are statistically significant, the breaking years, estimated in Maki’s
37
cointegration test, would also be included.
Where Di shows dummy variables of breaking years, which are again allowed up to five. 37
Granger causality tests, variance decompositions, and impulse responses
Regarding the long-term relationship in equation (2), Granger causality tests would also be estimated under the block exogeneity Wald tests and through the mechanism of the ECM (error correction model) mechanism. Therefore, this study could be determined the framework of Granger causality tests as follows:
In equation (6), Δ presents the difference operator and ECTt-1 stands the lagged ECT applied from the long-term equilibrium model. Finally, ε1,t, ε2,t, ε3,t, ε4,t, and ε5,t denote serially independent random errors with a mean of zero and a finite covariance matrix. Regards to the causality tests of ECMs, having statistically significant χ2- (chi-square) statistic(s) for ECT-1 in equation (6) would see the condition of having causation(s) in long-term and short-term.
In the final level, the variance decompositions for CO2 emission and urbanization would be carried out to determine what percentage of the forecasted error variance of the dependent variable could be described by exogenous shocks to independent variables. Once variance decompositions are estimated, impulse responses would be computed to forecast the reaction of exogenous shocks in the others for the selected variable under consideration.
Results and discussion
Unit root test results
Table 1 presents the results of the GLS-based unit root test 36 for the variables mentioned above. The three successful and significant structural breakpoints in all of the series are observed according to unit root tests, as can be seen in Table 1 (panels a through f). Regards to those break years from Table 1 into account, because the null hypothesis of a unit root cannot be rejected in the case of each variable, the unit root tests conclude that all of the series under consideration are non-stationary at their levels. Moreover, these series will be stationary at their first differences, while the null hypothesis of a unit root can be rejected again in the case of each variable. According to the GLS-based unit root tests, this study can suggest that lnCO2, lny, lnE, and lnOU are integrated with order one, I (1); thus, equation (1) of this study may be a cointegration model for the G7 countries.
The GLS-unit root tests under multiple structural breaks.
Note: Table 1 presents PT statistics of the GLS Unit Root Tests.
Cointegration tests under multiple structural breaks
In this study, all of the series are integrated in the same order; thus, it is suitable to estimate cointegration tests for equation (1), employing Maki’s 37 approach. Table 2 presents the results of the cointegration tests under multiple structural breaks (panels a through f).
Maki37 cointegration test results under multiple structural breaks.
Note: * and *** denotes the rejection of the null hypothesis of ‘no cointegration’ at the 0.01 and 0.10 levels, respectively.
As seen in Table 2, the null hypothesis which is no cointegration can be rejected by the existence of various structural break years, (panels a through f), and through two out of the four models suggested by Maki’s 37 methodology, which are presented in the previous section. The results reveal that equation (1) is a cointegration model for the G7 countries and the forecasting parameters in equation (2) are expected to be robust over the long-term period. It is important to note that those break years, which have been successfully obtained and provided in Table 2 (panels a through f), are also added to the estimation of long-term coefficients in equation (2) via dummy variables. 37
Estimation of long-term coefficients
The long-term coefficients in equation (2) are applied through the ARDL approach which is given in Table 3 (panels a through f). In Canada’s case, the coefficient of the GDP is positive and that of the GDP2 is negative, but both are statistically insignificant, which do not confirm the hypothesis of inverted U-shaped EKC. From the other point of view, energy consumption applies a positive and elastic impact on CO2 emission (β = 1.78, p < 0.01). Most importantly, the coefficient of urbanization is elastic, positive, and statistically significant (β = 7.31, p < 0.01), suggesting that a 1% change in urbanization leads to a 7.31% change in CO2 emission in the same path, which reveal that “damages” in climate quality at further levels of urbanization. These outcomes reveal that urbanization growth applies positively to significant impacts on climate change in Canada, which reflect unsuccessful policies in energy conservation on urban development. Finally, the results presented in panel (a) show that the coefficients of two break years (1969 and 1997) are statistically significant. The coefficient of the intercept is negative and statistically significant, which is quite acceptable, notice that if there is not any change in its determinants of equation (1), CO2 emission is likely to decline significantly in Canada.
Long-run estimates. With the overall CO2 emissions
Note: Numbers in square brackets are t ratios. Autocorrelation and heteroscedasticity problems in the model have been eliminated by means of the Newey-West approach. The model also passes the normality test, with a Jarque-Bera test statistic (χ2JBN) of 2.78. The five dummy variables in the model are: K1 (1963), K2 (1969), K3 (1975), K4 (1979) and K5 (1998) from Model 3 that includes a constant and a deterministic trend. * denotes statistical significance at the customary 0.05 significance level.
Table 3 (panel b) illustrates that in France’s case, the GDP (without squaring) has positive coefficient which is statistically significant (β = 123.46, p < 0.01), while that of the GDP2 is negative and again significant (β = −1.97, p < 0.01). These findings indicate strong evidence of the inverted U-shaped EKC hypothesis in France’s case. On the other hand, energy consumption exerts a negative and elastic effect on CO2 emission and damages on the EKC, as expected (β = −2.14, p < 0.01). The coefficient of urbanization is elastic, negative, and statistically significant (β = −7.51, p < 0.01), suggesting that a 1% change in urbanization volume leads to a 7.51% change in CO2 emission in the opposite direction. This concludes that urbanization growth applies negatively significant effects on climate change in France, signaling successful energy conservation policies on urban development. Finally, the results show that breaks in 1975 and 1994 (as estimated from cointegration tests) exert statistically significant and negative effects on CO2 emission (β = −0.11, p < 0.01 and β = −0.16, p < 0.01); however, breaks in 1990 and 1998 (as estimated from cointegration tests) exert statistically significant and positive effects on CO2 emission (β = 0.04, p < 0.01 and β = 0.23, p < 0.01), while the coefficients of the other break in 1983 are statistically insignificant. The coefficient of the intercept is negatively significant, notice that if there is not any change in its determinants of equation (1), CO2 emission is likely to decline considerably in France.
Table 3 (panel c) shows that in Italy’s case, GDP (without squaring) has positive and statistically significant coefficient (β = 12.70, p < 0.01), while that of the GDP2 is negative and again significant (β = −0.23, p < 0.01). This result is quite parallel with the inverted U-shaped EKC. From the other point f view, energy consumption applies a positive and elastic impact on CO2 emission (β = 1.02, p < 0.01). The coefficient of urbanization is elastic, negative, and statistically significant (β = −1.92, p < 0.01), which explains that a 1% change in the volume of urbanization leads to a 1.92% change in CO2 emission in the opposite direction. The result reveals that urbanization growth exerts a negatively significant effect on climate change in Italy, manifesting successful energy conservation policies on urban development. Finally, the break-in 1970 (as obtained from cointegration tests) exerts a statistically significant and negative effect on CO2 emission (β = −0.08, p < 0.01); the coefficients of the other break years are statistically insignificant. The coefficient of the intercept is negative, which is acceptable, indicating that if there is not any change in its determinants of equation (1), CO2 emission is likely to decline considerably in Italy.
Table 3 (panel d) shows that in Japan’s case, GDP (without squaring) has negative and statistically significant coefficient (β = −4.60, p < 0.01), while that of the GDP2 is positive and again significant (β = 0.07, p < 0.01). These findings do not confirm an inverted U-shaped EKC in Japan’s case. From the other point of view, energy consumption again applies a positive and elastic impact on CO2 emission (β = 1.15, p < 0.01). The coefficient of the urbanization variable is elastic, positive, and statistically significant (β = 0.80, p < 0.01), suggesting that a 1% change in the urbanization volume leads to a 0.80% change in CO2 emission in the same direction. This concludes that urbanization growth applies a positively significant effect on climate change in Japan, demonstrating unsuccessful energy conservation policies on urban development. Finally, the break-in 1998 (as obtained from cointegration tests) exerts a statistically significant and negative effect on CO2 emission (β = −0.04, p < 0.01); as well, the other break years’ coefficients are statistically insignificant. The coefficient of the intercept is negative, indicating that if there is not any change in its determinants of equation (1), CO2 emission is likely to decline considerably in Japan.
Table 3 (panel e) shows that in the UK’s case, GDP (without squaring) has positive and statistically significant coefficient (β = −19.96, p < 0.05), while that of the GDP2 is negative and again significant (β = 0.34, p < 0.05). These findings do not confirm an inverted U-shaped EKC in the UK’s case. Energy consumption exerts a positive and elastic effect on CO2 emission (β = 2.61, p < 0.01). The coefficient of urbanization is elastic, positive, and statistically significant (β = 16.47, p < 0.01), which suggests that a 1% change in the urbanization volume leads to a 16.47% change in CO2 emission in the same direction. This concludes the deteriorating effects of urbanization growth on the UK climate, indicating unsuccessful energy conservation policies on urban development. Finally, the break-in 1992 (as obtained from cointegration tests) exerts a statistically significant and positive effect on CO2 emission (β = 0.04, p < 0.01); however, the coefficients of the other break years are statistically insignificant.
Table 3 (panel f) illustrates that in the US case, GDP (without squaring) has negative and statistically significant coefficient (β = −2.55, p < 0.01), while that of the GDP2 is positive and again significant (β = 0.04, p < 0.01). These findings confirm an inverted U-shaped EKC. Energy consumption exerts a positive and an elastic effect on CO2 emission (β = 1.03, p < 0.01). The coefficient of urbanization is elastic, positive, and statistically significant (β = 0.82, p < 0.01), suggesting that a 1% change in the urbanization volume leads to a 0.82% change in CO2 emission in the same direction. This concludes that urbanization growth applies a positively significant effect on climate change in the US, indicating unsuccessful energy conservation policies on urban development, similar to the Canadian case.
It is effective to enable the EKC figures in the case study, with and without the urbanization volume, before proceeding with the ECM regressions. Figures 7 to 12, based on the estimations presented in Table 3 (panels a through f), plot the EKC for the G7 countries, using two EKC model options: (1) conventional EKC without urbanization but with estimated CO2 emission relating to GDP and energy consumption and (2) revised EKC, including the urbanization variable. Figures 8 and 9 show that the EKCs of France and Italy are always inverted U-shaped, regardless of whether or not energy consumption and/or urbanization volume are/is added; in other words, successful urban development seems to be adopted in France and Italy. In contrast, Figures 7, 10, and 11 indicate minor shifts in the EKCs of Canada, Japan, and the UK when urbanization and energy consumption variables are added. In the US case (Figure 12), a very minor shift in the EKC over longer periods does not signify successful urban development.

Actual and estimated EKCs for Canada.

Actual and estimated EKCs for France.

Actual and estimated EKCs for Italy.

Actual and estimated EKCs for Japan.

Actual and estimated EKCs for United Kingdom.

Actual and estimated EKCs for United States.

Impulse Responses – Canada.

Impulse responses – France.

Impulse responses – Italy.
Estimation of error correction models
The ECM regressions associated with the cointegration model from equation (2) have been done as the next level. The results are given in Table 4 (panels a through f). In Canada’s case, the ECT term in equation (3), where lnCO2 (CO2 emission) is the dependent variable, is statistically significant and negative (β = −0.117, p < 0.01). This concludes that lnCO2 converges to its long-term equilibrium path by 11.7% speed of adjustment through the channels of energy consumption, real income, and urbanization growth. In the short-term, GDP (without squaring) has the negative, elastic, and statistically significant coefficient (β = −28.515, p < 0.01). Furthermore, the short-term coefficient of the GDP2 is positive, inelastic, and statistically significant (β = 0.521, 1.65 < t), another proof of not having an inverted U-shaped EKC in the short-term Canadian economy. The short-term coefficients of the energy consumption and urbanization variables are positive and inelastic but statistically insignificant.
Conditional error correction models through the ARDL approach and short-term coefficients.
In France’s case (Table 4, panel b), the ECT term for equation (3), where lnCO2 is the dependent variable, is statistically significant and negative (β = −0.464, p < 0.01). This shows that lnCO2 converges to its long-term equilibrium path by 46.4% speed of adjustment through the channels of energy consumption, real income, and urbanization growth. In the short-term, GDP (without squaring) has the negative, elastic, and statistically significant coefficient (β = 128.49, p < 0.01). Furthermore, the short-term coefficient of the GDP2 is negative, elastic, and statistically significant (β = 2.290, p < 0.01), another evidence of not having an inverted U-shaped EKC even in the short-term French economy. The short-term coefficient of energy consumption is negative, inelastic, and statistically significant (β = −0.967, p < 0.01), as expected. It is important to note that, one more time, in the short-term, the urbanization variable has a negative, elastic, and statistically insignificant coefficient (β = −5.39, p < 0.01). This again proposes the success in the policies of energy-saving to reduce environmental pollution in France, even in the case of the short term. Additionally, the coefficient of the intercept is negative and statistically significant.
In Italy’s case (Table 4, panel c), the ECT term for equation (3), where lnCO2 stands for the dependent variable, which is statistically significant and positive (β = −0.544, p < 0.01). This proves that lnCO2 converges to its long-term equilibrium path by 54.4% speed of adjustment through the channels of energy consumption, real income, and urbanization growth. In the short-term, GDP (without squaring) has negative, elastic, and statistically insignificant coefficient (β = −7.653, p < 0.01). Also, in the short-term GDP2 has a positive, inelastic, and statistically insignificant coefficient (β = 0.142, p < 0.01), another proof of not having an inverted U-shaped EKC, even in the short-term Italian economy. In the short-term, energy consumption has negative, inelastic, and statistically significant coefficient (β = −0.514, p < 0.01), as expected. Therefore, it is important to note that, one more time, in the short-term, the urbanization variable has a negative, inelastic, and statistically significant coefficient (β = −0.056, p < 0.01). This again proposes that policies of energy-saving minimize environmental pollution seem successful in Italy, even in the short term. Moreover, the coefficient of the intercept is positive and statistically significant.
In Japan’s case (Table 4, panel d), the ECT term for equation (3), where lnCO2 shows the dependent variable, which is statistically significant and positive (β = −0.588, p < 0.01). This concludes that lnCO2 converges to its long-term equilibrium path by 58.8% speed of adjustment through the channels of energy consumption, real income, and urbanization growth. In the short-term, GDP (without squaring) has negative, elastic, and statistically significant coefficient (β = −28.527, p < 0.01). Also, in the short-term, GDP2 has a positive, inelastic, and statistically significant coefficient (β = 0.501, p < 0.01), another indicator of not having an inverted U-shaped EKC, even in the short-term Japanese economy. In the short-term, energy consumption has positive, inelastic, and statistically insignificant coefficient (β = 0.121, p < 0.01), as expected. Notice that, one more time, in the short-term, urbanization variable has negative, inelastic, and statistically insignificant coefficient (β = −0.665, p < 0.01). It reflects Japan’s success in implementing energy-saving policies to decrease environmental pollution, even for the short term period. Also, the intercept’s coefficient is positive and statistically insignificant.
In the UK’s case (Table 4, panel e), the ECT term for equation (3), where lnCO2 shows the dependent variable, which is negative and statistically significant (β = −0.550, p < 0.01). This shows that lnCO2 converges to its long-term equilibrium path by 55% speed of adjustment through the channels of energy consumption, real income, and urbanization growth. In the short-term, GDP (without squaring) has negative, elastic, and statistically significant coefficient (β = −80.015, p < 0.01). The short-term coefficient of the GDP2 is positive, elastic, and statistically significant (β = 1.445, p < 0.01), another evidence of not having an inverted U-shaped EKC, even in the short-term UK economy. The short-term coefficient of energy consumption is positive, inelastic, and statistically significant (β = 0.651, p < 0.01), as expected. The momentous is that one more time, in the short-term, the urbanization variable has a negative, elastic, and statistically insignificant coefficient (β = −3.422, p < 0.01). It demonstrates that policies of energy-saving to reduce environmental pollution seem successful in the UK, even in the short term. Moreover, the coefficient of the intercept is positive and statistically significant.
In the US case (Table 4, panel f), the ECT term for equation (3), where lnCO2 shows the dependent variable, which is statistically significant and negative (β = −0.425, p < 0.01). It demonstrates that lnCO2 converges to its long-term equilibrium path by 42.5% speed of adjustment through the channels of energy consumption, real income, and urbanization growth. In the short-term, GDP (without squaring) has negative, elastic, and statistically insignificant coefficient (β = −12.864, p < 0.01), whereas, in the short-term, GDP2 has positive, inelastic, and statistically insignificant coefficient (β = 0.215, p < 0.01). This is another proof of not having an inverted U-shaped EKC, even in the short-term US economy. In the short-term, energy consumption has positive, inelastic, and statistically insignificant coefficient (β = 0.842, p < 0.01), as expected. The momentous is that one more time, in the short-term, the urbanization variable has a positive, elastic, and statistically insignificant coefficient (β = 1.479, p < 0.01). One more time, it proposes that policies of energy-saving minimize environmental pollution seem unsuccessful in the US, even for the short term period. Also, the intercept’s coefficient is positive and statistically insignificant.
In the next stage, the causality direction can now be searched within the Granger causality tests under the block exogeneity Wald tests, through the mechanism of error correction for the short-term and long-term periods. The χ2-statistics for both long-term and short-term causations are implied in Table 5, as calculated in equation (6).
Granger causality tests under block exogeneity approach.
Note: i ** and *** denote the rejection of null hypothesis respectively at alpha 0.05% and 0.10% level.
Variance Decomposition Results.
The results in Table 5 (panels a through f) presents various causalities both in the long-term and short-term periods. Canada’s case indicates an absence of causality in the EKC model of equation (1), where it is controlled for urbanization. However, changes in urbanization are preceded by the changes in CO2 emission, energy consumption, and real income, both in the short and long terms. In France’s case, long-term causality is apparent in equation (1), which is also supported by short-term causalities running from real income, energy consumption, and urbanization to CO2 emission. Italy’s case shows no causality in equation (1), but short-term causality that runs from energy consumption to CO2 emission is inferred. The results of this study revealed similar findings for Japan, the UK, and the USA like the case of Canada.
Table 6 (panels a through f) presents the results of variance decomposition, which generally concludes that in the initial periods, low levels of the forecasted error variance of CO2 emission are explained by exogenous shocks to the energy consumption, output, and urbanization variables. The variance decomposition ratio of CO2 emission to urbanization is highest in France, at 30.382% in period 10. The second-highest ratio is in the UK, followed by the US, Italy, and Canada, respectively. The ratio is lowest in Japan, at 1.562% in period 10.
Finally, Figures 13 to 18 provide line plots of impulse responses among CO2 emission, energy consumption, output, and urbanization. The response of emission pollutants to a shock in urbanization volume in each country is provided in the last column of the first line. Figures show that the response of emission pollutants to shocks in urbanization is sharply upwards in the cases of Canada, France, and the USA denoting a positive effect and statistically significant. This finding reveals that when there is a shock in the volume of the urban population, emission pollutants will be significantly affected in the same direction; that is, i.e., an increase in urban population will increase emission pollutants as well. However, on the other side, such response is sharply downwards in the cases of Italy and the UK denoting a negative effect and again statistically significant. This finding reveals that, i.e., an increase in the urban population of Italy and the UK will not increase emission pollutants; but, there will be a decline in the volume of pollutants. Finally, emission pollutants are not responsive to a shock in urbanization in the case of Japan. Figure 16 shows that the reaction of emission pollutants in Japan is highly insignificant to shocks in urbanization revealing that any shock to the urban population in Japan will not impact its emission pollutants.

Impulse Responses – Japan.

Impulse Responses – UK.

Impulse Responses – USA.
Conclusion
This study empirically investigated the urbanization-induced EKC hypothesis and the long-term equilibrium interactions and direction of causality between urban development and CO2 emission in the G7 countries, except Germany. In this respect, the theoretical EKC framework had been taken into consideration in the empirical analysis. The results of this study are of interest to both scholars and policymakers because the G7 countries constitute the most dynamic economies in the world, which are linked with various resources of energy, high-level of urbanization, and rapid growth in industrialization. The main rationalization of this research is that urban development is expected to have a statistical interaction with CO2 emission in such economies. To the best of the author’s knowledge, this study is the first of its kind to investigate the interaction between urban development and CO2 emission, using the theoretical EKC framework.
This study’s findings show that only France and Italy have inverted U-shaped EKCs, even in the presence of urbanization. Furthermore, urbanization in these two countries exerts negative effects on their CO2 emission levels, as opposed to the positive effects in the other G7 countries.
To summarize, this study’s outcome suggests that urban development has been successfully managed in France and Italy only, as far as “green energies and green environment” are concerned. For comparison purposes, similar research can be replicated for other countries that have experienced rapid urbanization.
Footnotes
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Appendix
Abbreviations.
AIC
:
Akaike information criterion
ARDL
:
Autoregressive distributed lag
CO2
:
Carbon dioxide emission
E
:
Energy consumption
ECT
:
Error correction term
EKC
:
Environmental Kuznets curve
EMT
:
Ecological modernization theory
G7
:
Group of seven
GDP
:
Gross domestic product
GLS
:
Generalized least squares
MENA
:
The Middle East and North Africa
NICs
:
Newly industrialized countries
OLS
:
Ordinary least squares
OU
:
Overall urbanization
UK
:
United Kingdom
USA
:
United States of America
