Environmental Kuznets Curve (EKC): Empirical Relationship Between Economic Growth, Energy Consumption, and CO2 Emissions: Evidence from 3 Developed Countries

In this study, the environmental Kuznets curve (EKC) hypothesis is examined for 3 developed countries, which are Denmark, the United Kingdom, and Spain, for the period between 1960 and 2014. The EKC hypothesis is examined under 2 nexuses which are GDP, CO2 and energy consumption, and GDP, CO2, energy consumption and the square of GDP. Causal and long-term relationships between GDP, CO2, and energy consumption are examined for these 3 developed countries using the ARDL bounds test, the Toda and Yamamoto Granger non-causality test, the VAR Granger Causality/Block Exogeneity Wald test, and the Johansen cointegration test. Long-term relationships between GDP, CO2, energy consumption, and the square of GDP are examined by the Johansen cointegration test. The EKC hypothesis is not confirmed for Denmark, the United Kingdom, and Spain, and the neutrality hypothesis is confirmed for these 3 developed countries. Unidirectional causality running from energy consumption to CO2 is found for Denmark, and unidirectional causality running from CO2 to energy consumption is found for the United Kingdom.


INTRODUCTION
Climate change is a topic discussed worldwide by scientists, politicians, and individuals. Carbon dioxide is also discussed besides climate change because it is one of the major causes of climate change and one of the main greenhouse gas emissions. To cope with climate change and reduce CO 2 , many initiatives have taken place at the individual country and global levels. The Paris Agreement and the Kyoto Protocol can be mentioned as 2 global initiatives.
The Kyoto Protocol is an international agreement which was signed and ratified with different parties on December 11, 1997, and is one of the main efforts by humanity to cope with climate change and reduce CO 2 emissions.
The Kyoto Protocol is discussed besides the environmental Kuznets curve (EKC), which states that income increases with CO 2 to a certain level, and after that level is reached, CO 2 starts to decrease while income continues to increase. The impact of the Kyoto Protocol on EKC is one of the determinants for the countries that are involved in the protocol to determine the implications of their policy toward their coping strategies with climate change.
seasonal and exists over a long period) level and trends of income inequalities. He analyzed data for the United States, England, and Germany which were industrialized countries. He found an inverted U-shaped relationship between income inequality and economic growth. First, as economic growth increases, income inequality increases. After a certain point, as economic growth increases, income inequality declines.
In the 1990s, the U-shaped environmental Kuznets curve (EKC) was used to examine the relationship between economic growth and environmental degradation which started at low-income levels and, as income increased, environmental degradation increased, but, after a certain point, as income increased, environmental degradation decreased. Gene Grossman and Alan Krueger (1991), Nemat Shafik and Sushenjit Bandyopadhyay (1992), and Theodore Panayotou (1993) were among the first studies for the EKC hypothesis and the U-shaped relationship between economic growth and environmental degradation. Grossman and Krueger (1991) examined NAFTA's impact on the environment. One of their findings was that, as GDP increased at the low-income level, levels of sulfur dioxide and smoke, which were 2 types of environmental degradation, increased as well. After a certain point, at high-income levels, as GDP increased, levels of sulfur dioxide and smoke declined. Shafik and Bandyopadhyay (1992) examined the relationship between economic growth and environmental quality for countries at different income levels. They used environmental indicators, such as the lack of clean water and carbon emissions per capita, to examine the relationship between economic growth and environmental quality. For most environmental indicators, as income increased, environmental indicators worsened at low-income levels but improved at high-income levels. Except fecal coliforms in rivers, none of the environmental indicators worsened in high-income levels and fecal coliforms in rivers had an N-shaped relationship with income. Shafik and Bandyopadhyay (1992) concluded that there was an EKC relationship between income and most of the environmental indicators in the study. Some exceptions to this relationship were dissolved oxygen in rivers, municipal waste, and carbon emissions. Panayotou (1993) tested and verified an EKC relationship between the types of environmental degradation, which are deforestation and air pollution, and level of economic development for a sample of developing and developed countries. He also provided policy implications for developing and developed countries for the areas of employment, technology transfer, and development assistance in his study.
For emissions, income, and energy variables, there are 4 research focuses in the literature (see Table 1).

INCOME-ENVIRONMENTAL DEGRADATION RELATIONSHIP
Literature for the relationship between income and environmental degradation is examined under 2 topics which are EKC relationships and income-emissions nexus. Within the topic of EKC relationships, studies that indicated the validity of EKC relationships are examined. Under income-emissions nexus, causality relationships and long-term and short-term relationships between emissions and income are examined.

Income-emissions nexus
There are studies that test EKC relationships alone, and there are other studies that investigate for causality, long-term, and short-term relationships between income and emissions including explanatory variables.

Income-energy nexus
Studies in this context investigate to verify the neutrality, conservation, growth, and feedback hypotheses. The neutrality hypothesis states that there is no causality between energy consumption and income. The conservation hypothesis states that there is unidirectional causality running from income to energy consumption. The growth hypothesis states that there is unidirectional causality running from energy consumption to income. The feedback hypothesis states that there is bidirectional causality between income and energy consumption.
Emissions-energy nexus Studies in this context investigate causal, long-term and shortterm relationships between emissions, and energy.
Emissions-energy-income nexus Studies in this context examine causal, long-term, and shortterm relationships between emissions, energy, and income.
The environmental Kuznets curve (EKC) general model is in the literature as below: 4 are estimated parameters. t is time index. e is error term. x is an explanatory variable which can be energy consumption, trade openness, population, etc. E is the environmental degradation type. Y is the economic activity type. P is population.
Different levels of relationships between environmental degradation and economic activity are as below (Tao Song, Tingguo Zheng, and Lianjun Tong 2008).

EKC RELATIONSHIP NOT VERIFIED
For multi-country studies, Hiroki Iwata, Keisuke Okada, and Sovannroeun Samreth (2012) indicated that their study did not support EKC relationships for 11 OECD countries which were Belgium, Canada, Finland, Germany, Japan, South Korea, Spain, Sweden, Switzerland, the United Kingdom and the United States.
For panel studies, Inmaculada Martinez-Zarzoso and Aurelia Bengochea-Morancho (2003) examined 19 Latin American and Caribbean countries and found that EKC was not confirmed for panel countries. They also found heterogeneity for EKC relationships among 19 countries and only a few countries showed EKC relationships. Among 19 countries, 9 countries showed an N-shaped curve, 2 countries showed a curve with a decreasing trend, 2 countries showed a U-shaped curve and 6 countries showed an upward sloping curve.

CO2-GDP-EN NEXUS
One of the research focuses in the literature for emissions, income, and energy variables is the emissions-income-energy nexus. Four hypotheses, which are the neutrality, conservation, growth, and feedback hypotheses, were tested for GDP-EN (energy consumption) relationships. EKC relationships are also tested under this research focus. CO 2 -GDP-EN nexus also takes into consideration additional explanatory variables to investigate the relationships between CO 2 , GDP, and EN. Relationships between CO 2 , GDP, and EN are investigated by panel studies, multicountry studies, and single-country studies in the literature.
For multi-country studies, Loesse Jacques Esso and Yaya Keho (2016) studied long-term, short-term, and causal relationships between CO 2 , GDP, and EN for 12 selected sub-Saharan countries for between 1971 and 2010 with the Granger causality test and the cointegration bounds testing approach methodologies. They found unidirectional causality running from GDP to CO 2 for Benin, the Democratic Republic of the Congo, Ghana, Nigeria and Senegal, unidirectional causality running from CO 2 to GDP for Gabon, Nigeria, and Togo, and bidirectional causality between GDP and CO 2 for Nigeria in the short run. They also found bidirectional causality between GDP and CO 2 for Congo and Gabon, unidirectional causality running from EN to CO 2 for Benin, Congo, Cote d'Ivoire, Gabon, Ghana, Nigeria, Senegal, South Africa, and Togo and unidirectional causality running from GDP to CO 2 for Benin, Congo, Cote d'Ivoire, Gabon, Ghana, Nigeria, Senegal, South Africa, and Togo in the long run.
For panel studies, Samia Gmidene, Saida Zaidi, and Sonia Zouari Ghorbel (2016) examined the causal relationships between renewable energy, nuclear energy consumption, economic growth, and CO 2 emission with panel cointegration techniques and the Granger causality test methodologies. They concluded that authorities should invest in renewable energy to reduce CO 2 emissions and decrease nuclear energy to reduce CO 2 emissions. Farhani and Rejeb (2012), Asongu, Montasser, and Toumi (2015) and Wang et al. (2011) verified the conservation hypothesis in the long run for 15 Mena countries, 24 African countries, and China.
For single-country studies, Bikash Chandra Ghosh, Khandakar Jahangir Alam, and Ataul Gani Osmani (2014) studied the relationships between CO 2 , GDP, and EN in Bangladesh for the period between 1972 and 2011 with the Johansen and Juselius cointegration test and vector autoregressive (VAR) error correction model methodologies. They found that EN had a significant positive impact on GDP, and CO 2 had an insignificant negative impact on GDP in the long run.
Aviral Kumar Tiwari (2011) confirmed the neutrality hypothesis in India in the long run and short run, and Masoud Mohammed Albiman, Najat Nassor Suleiman, and Hamad Omar Baka (2015) confirmed the neutrality hypothesis in Tazmania.

DATA
The data were derived from the World Bank's official web site for CO 2 emissions (metric tons per capita), energy consumption (kg of oil equivalent per capita), and GDP per capita (constant 2010 US$). The structure of data is annually which exceeds 30 to make it a parametrical test. Data periods are determined according to the availability of data sets for developed countries. The data used in the study are from 1960 to 2014 for developed countries, which are Denmark, Spain, and the United Kingdom.

METHODOLOGY
The augmented Dickey-Fuller (1981) unit root test is applied to find stationary levels of each variable (see Table 2). According to the combination of stationary levels of each variable, a different method is used to find cointegration. For variables which have stationary levels with combination of I(1) and I(0), the ARDL bounds test for cointegration by Pesaran, Shin, and Smith (2001) is used to find cointegration between the variables. Normality tests, namely, the Breusch-Godfrey Serial Correlation LM test and Breusch-Pagan-Godfrey heteroscedasticity test, are applied to determine the stability of ARDL model. The Johansen (1991) cointegration test is used for variables which are integrated at I(1) to find cointegration.
The VAR Model is applied for variables which are integrated at I(1) with no cointegration. The Breusch-Godfrey Serial Correlation LM test, the Breusch-Pagan-Godfrey heteroscedasticity test, and the VAR stability test are applied to determine the stability of the VAR model. Impulse response analysis and variance decomposition analysis are applied to find how each variable impacts and influences the other variables.
The VAR Granger causality method is used to find causal relationships between variables that are integrated at I(1) with no cointegration.
The Toda and Yamamoto (1995) Granger non-causality test is applied to find causal relationships between variables which have stationary levels with a combination of I(0) and I(1). First, the maximum number of stationary levels of variables is determined by ADF unit root tests. Then, a lag order is selected by determining the stability of the VAR model by AR Root Graph and the VAR Residual Serial Correlation LM test, and then the VAR model is developed with the selected lag order. Lag order (Determined lag order + maximum number of stationary levels of variables) of variables are added to exogenous variables, and the VAR model is developed by the determined lag order.
Two models are used in the study which consist of CO 2 , GDP, and energy consumption variables, and CO 2 , GDP, the square of GDP and energy consumption variables to determine the cointegration and causal relationships between variables.  (1) -4.886959(0)* Notes: * and ** show the statistical significance at 1% and 5% levels, respectively. The lag length is shown by the values in parentheses.

DENMARK 4.1.1 CO 2 , GDP, and EN NEXUS
According to the ARDL bounds test results, no cointegration is found between CO 2 , GDP and energy consumption because the ARDL bounds test is not significant at a level of 5% (see Table 3). With regard to the Normality test, the Breusch-Godfrey Serial Correlation LM test and the Breusch-Pagan-Godfrey test results, the model is stable (see Table 4, Table 5, and Table 6). There is no long-term relationship between CO 2 , GDP, and energy consumption.
The Toda and Yamamoto Granger non-causality test is applied for testing causal relationships between CO 2 , GDP, and energy consumption. The VAR model is stable, and there is no correlation between variables (see Figure 1 and Table 7). According to the VAR Granger Causality/Block Exogeneity Wald tests results, unidirectional causality running from LNEN to LNCO 2 is found but no causality from LNGDP to LNCO 2 is found. Also, there is no causality from LNCO 2 and LNGDP to LNEN and no causality from LNCO 2 and LNEN to LNGDP (see Table 8).

CO 2 , GDP, the square of GDP and EN NEXUS
According to the ARDL bounds test results, no cointegration is found between CO 2 , GDP, and energy consumption because the ARDL bounds test is not significant at a level of 5% (see Table 9). With regard to the Normality test, the Breusch-Godfrey Serial Correlation LM test, and the Breusch-Pagan-Godfrey test results, the model is stable (see Table 10, Table 11 and Table 12). Because no long-term relationship is found between CO 2 , GDP, and the square of GDP and EN, the EKC hypothesis is not confirmed for Denmark.

SPAIN 4.2.1 CO 2 , GDP and EN NEXUS
According to ARDL bounds test results, no cointegration is found between CO 2 , GDP and energy consumption because ARDL bounds test is not significant at a level of 5% (see Table 13). With regard to the Normality test, the Breusch-Godfrey Serial Correlation LM test, and the Breusch-Pagan-Godfrey test results, the model is stable (see Table 14, Table 15, and Table 16). There is no long-term relationship between CO 2 , GDP, and energy consumption. The Toda and Yamamoto Granger non-causality test is applied for testing causal relationships between CO 2 , GDP, and energy consumption. The VAR model is stable, and there is no correlation between the variables (see Table 17 and Figure 2). According to the VAR Granger Causality/Block Exogeneity Wald tests results, there is no causality from LNEN and LNGDP to LNCO 2 , no causality from LNCO 2 and LNGDP to LNEN, and no causality from LNCO 2 and LNEN to LNGDP (see Table  18).

CO 2 , GDP, the square of GDP and EN NEXUS
According to ARDL bounds test results, no cointegration is found between CO 2 , GDP, and energy consumption because ARDL bounds test is not significant at 5% level (see Table 19). With regard to the Normality test, the Breusch-Godfrey Serial Correlation LM test, and the Breusch-Pagan-Godfrey test results, the model is stable (see Table  20, Table 21, and Table 22). Because no long-term relationship is found between CO 2 , GDP, square of GDP and EN, the EKC hypothesis is not confirmed for Spain.

UK 4.3.1 CO 2 , GDP, and EN NEXUS
According to the Johansen cointegration test results, no cointegration is found between CO 2 , GDP, and energy consumption (see Table 23). There is no long-term relationship between CO 2 , GDP, and energy consumption. The VAR model is established, and the VAR Granger Causality/Block Exogeneity Wald tests are applied for causality between CO 2 , GDP, and energy consumption. The VAR Residual Serial Correlation LM test and the VAR Residual Heteroscedasticity test results show the model is stable (see Table 24 and Table 25). The VAR satisfies the stability condition (see Figure 3).
According to the VAR Granger Causality/Block Exogeneity Wald tests results, there is no causality from LNEN and LNGDP to LNCO 2 and no causality from LNCO 2 and LNEN to LNGDP. Unidirectional causality running from LNCO 2 to LNEN is found, and no causality is found from LNGDP to LNEN (see Table 26).        Impulse response analysis is applied to find how each variable impacts and influences the other variables. Energy consumption has a positive impact on CO 2 in the short run but has no impact on CO 2 in the long run. GDP affects CO 2 in the first 2 periods positively, then affects it negatively in the short run after 2 periods (see Figure  4). CO 2 affects energy consumption negatively and positively in the short run. GDP has a positive impact for the first 2 periods on energy consumption, then GDP has a negative impact on energy consumption after 2 periods in the short run.
CO 2 has a positive impact on GDP in the short run. Energy consumption has a positive impact on GDP in the short run but has no impact on GDP in the long run.  Variance decomposition analysis is applied to find how each variable impacts and influences the other variables. Energy consumption can cause a 5.79% fluctuation in CO 2 in the short run and a 5.89% fluctuation in CO 2 in the long run. GDP can cause a 5.58% fluctuation in CO 2 in the short run and a 5.60% fluctuation in CO 2 in the long run (see Table 27).
CO 2 can cause a 67.98% fluctuation in energy consumption in the short run and a 67.93% fluctuation in energy consumption in the long run. GDP can cause a 5.49% fluctuation in energy consumption in the short run and a 5.51% fluctuation in energy consumption in the long run.
CO 2 can cause a 34.38% fluctuation in GDP in the short run and a 34.42% fluctuation in GDP in the long run. Energy consumption can cause an 8.92% fluctuation in GDP in the short run and an 8.98% fluctuation in GDP in the long run.   According to the Johansen cointegration test results, no cointegration is found between CO 2 , GDP, the square of GDP and energy consumption (see Table 28). Because no long-term relationship is found between CO 2 , GDP, and the square of GDP and EN, the EKC hypothesis is not confirmed for the UK.

CONCLUSION
The EKC hypothesis states that economic growth will lead to reduction in emissions. Results of this study did not verify this statement. Our results are in line for EKC with Bruyn, Bergh, and Opschoor (1998) and Acaravci and Ozturk (2010) for the United Kingdom and Roca et al. (2001) and Esteve and Tamarit (2012a) for Spain.
One of the significant findings of our study is that the EKC hypothesis is rejected, and no causal relationships are found between CO 2 and GDP. Another significant finding of our study is that the neutrality hypothesis is confirmed for 3 developed countries which state there is no causal relationship between energy consumption and income. The other significant finding of our study is that unidirectional causality running from energy consumption to CO 2 is found for Denmark, and unidirectional causality running from CO 2 to energy consumption is found for the UK for emissions-energy nexus.
No causal relationship between GDP and CO 2 , which is found for all countries, means that a country's economic growth will not have an effect on emissions. Denmark, Spain, and the United Kingdom are likely to achieve further economic growth without causing environmental degradation because no causal relationship is found between CO 2 and GDP.
No causal relationship between GDP and EN, which is found for all countries, means that a country's economic growth will not have an effect on energy consumption. The economic growth of these countries is not dependent on oil consumption. Also, oil consumption is not a source for economic growth in these countries.
For the United Kingdom, CO 2 causes energy consumption, but in Denmark and Spain, CO 2 does not cause energy consumption. Increasing efficiency of energy technologies and increasing alternatives for replacing oil should be part of the policies that target oil consumption in the United Kingdom. The industry sector constitutes 23% of overall oil consumption in the United Kingdom which has decreased from 82% (1970). Energy efficiency improvements in the industry sector will help to decrease CO 2 emissions. Solar energy can replace further oil consumption in the industry sector. Toyota uses solar energy to power its plant in Deeside, and this example can be used by other firms in the industry sector to replace oil with solar energy and decrease emissions. Increasing natural gas and electricity use can be an alternative for oil too. Thirty-six percent and 33% of overall energy used in industry is from natural gas and electricity, respectively. The household and transport sectors account for 40% and 28% of total final energy consumption in the United Kingdom. Improving home insulation should be continued to maintain energy efficiency in the household sector. Increasing the share of green energy, such as wind in the electricity generation mix, will help decrease emissions for the household sector. The United Kingdom maintained a successful energy efficiency policy for the road, and air transportion which consume 96% of energy in the transport sector. Currently, there are 52,000 electric vehicles in the United Kingdom. Incentives to increase the number of electric vehicles should be continued in the transport sector. Increasing the share of green energy, such as wind in the electricity generation mix, will also help decrease further emissions in the transport sector. Energy efficiency policy should be continued for air transport as well.
For Denmark, energy consumption causes emissions but, in the United Kingdom and Spain, energy consumption does not cause emissions. Oil is the first in Denmark's total primary energy supply and final energy consumption. Oil is mostly used in the transport sector at 65% of all oil consumption. The transport sector is the second highest energy-consuming sector (30%) in Denmark after the residential sector (32%). Denmark should increase its energy efficiency in the transport sector. Electricity is the second in final energy consumption in Denmark, and electricity has almost no share in the transport sector (1.3%). Denmark should increase the percentage of electricity consumption in the transport sector by involving more electric vehicles and providing higher-energy efficiency policy incentives for all types of transport, such as road and air transport. Natural gas has almost no share in the transport sector (0.1%), and increasing the use of natural gas in the transport sector should be included in the climate change policy of Denmark. Investment in battery technology will help to increase the share of electricity usage in the transport sector and increasing the share of green energy, such as wind energy in the electricity generation mix will help to decrease CO 2 emissions in the transport sector.
For Spain, the fuel tax should be increased, especially for diesel fuel consumption. An increase in the fuel tax will encourage more efficient usage of oil in the transport sector. Spain's oil consumption in 2017 is 21% lower than its oil consumption in 2007, and Spain's oil consumption in 2014 is 18.5% lower than its oil consumption in 2000. Oil consumption had decreased by 19.4% in the transport sector, 44.4% in the industry sector, and 25.1% in the household sector between 2007 and 2013. Spain started an efficient vehicle incentive program to provide funding to car owners to replace over 700,000 old cars and vans with new ones until 2020 to reduce each replaced vehicle's emission levels by 30% on average. Although economic recession has helped reduce oil usage, for the overall case beginning from 2000, reductions in oil consumption are mainly due to increases in energy efficiency in the industry sector, home renovations in the household sector for energy efficiency and the aforementioned efficient vehicle incentive program and efficient driving programs for new drivers. These policies should be enhanced and include the increase in fuel tax policy.
Economic growth is not likely to help Spain, Denmark, and the United Kingdom to fight climate change by itself. Increases in the use of renewable energy and improving energy efficiency in the transport, industry, and household sectors will help Spain, Denmark and the United Kingdom to fight climate change and meet emission targets. Authorities in Spain, Denmark, and the United Kingdom should continue to invest in energy conservation and emission reduction policies because these policies are likely to not have a detrimental effect on economic growth. These countries are likely to achieve further economic growth without causing environmental degradation because no causal relationship is found between CO2 and GDP.
The limitations of our study are that results are obtained for 3 developed countries and the period between 1960 and 2014 are examined for these countries.