ENVIRONMENTAL KUZNETS CURVE IN GREECE IN THE PERIOD 1960-2014

CO2 emissions in Greece have increased significantly in recent decades. Many researches point out that Greek’s economy structure is characterized by a strong interdependence of energy use and economic growth, so that any increase in national production leads to an increase in CO2 emissions. In this sense, Greek’s economy structure does not allow policies to de-escalate the level of CO2 emissions without cost in terms of economic growth. This paper focuses on investigating the short and long term relationship between Gross National Product (GDP), CO2 emissions and gross energy use over the period 1960-2014, using appropriate econometric tools. Additionally, the impact of the urbanization rate on CO2 emissions is examined. The exploration of these relationships is part of the Kuznets Environmental Curve (EKC) theory that economic growth ultimately leads to a degradation of environmental degradation.


INTRODUCTION
The Environmental Kuznets Curve (EKC) theory first appears in the early 1990s and examines the relationship between economic growth and environmental degradation. The relationship developed between environmental degradation and economic growth takes the form of an inverse U, confirming the theory developed by Simon Kuznets about the income inequality -economic growth nexus (Kuznets, 1955).
The EKC case has been empirically studied for many types of pollutants (Shafic and Bandyopadhyay, 1992). Most studies indicate that the EKC hypothesis is confirmed for pollutants that have local and perceived impact and low cost of treatment (Stern et al., 1996). However, the results are rather mixed when CO 2 emissions are considered as a form of environmental degradation and the EKC hypothesis is neither rejected nor confirmed adequately on an empirical basis (Dinda, 2004;Kaika and Zervas, 2013a). Many studies indicate a rather positive relationship with economic growth (Shafic and Bandyopadhyay, 1992;Grossman and Krueger, 1995), implying that the level of CO 2 emissions rises with income. An interpretation for this is that CO 2 emissions are not immediately felt locally and the cost of dealing with them is high, as it is directly linked to the productive structure of an economy (Shafic and Bandyopadhyay, 1992;Grossman and Krueger, 1995). In addition, CO 2 emissions have a global impact and even if their growth rate is reduced as a result of the decline in economic growth (mainly from the developed countries), global CO 2 emissions will continue to increase due to the rapid economic growth of developing economies (Kaika and Zervas, 2013a). Moreover, Generalised Method of Moments and ARDL has been also used to test if there is any relationship between GDP and CO 2 emissions, with the result to be EKC evidence for some countries including Greece (Acaravci and Ozturk, 2010;Jaunky, 2011).
Recent studies set in the center of consideration energy use as a crucial factor affecting the income -CO 2 emissions nexus. For this purpose, several studies attempt to examine the relationships between CO 2 emissions, income and energy use, either by using panel data techniques (Apergis and Payne, 2009;Marrero, 2010;Pao and Tsai, 2011) or by adopting time series techniques (Acaravci and Ozturk, 2010;Ang, 2007;Halicioglu, 2009;Hatzigeorgiou et al., 2011;Jalil and Mahmud, 2009;Saboori and Sulaiman, 2013;Shahbaz et al., 2013). What comes up from the literature review is that the causal relationships between the variables of economic growth, energy use and CO 2 emissions are not clear-cut. The dynamic relationships among the variables in question may move into different directions depending on the time period used, the econometric methodology adopted, the assumptions implied and the country/ies examined (Kaika and Zervas, 2013b).
The case of Greece is quite peculiar, since there are several economic cycles (growth, stagnation, recession) between 1960 and 2014. After the 2 nd World War and the Greek Civil War (1946)(1947)(1948)(1949), the rural population moved to urban centers for the needs of industrial production. This urbanization resulted in important changes in the productive structure of the Greek economy. However, a long-run relationship between real GDP and CO 2 emissions caused by the urbanization trend has not been sufficiently studied using econometric applications. This paper investigates the causality among Gross Domestic Product (GDP), CO 2 emissions and energy use on a per capita basis over the period 1960-2014 in Greece. This study is crucial for policy makers in Greece in order to achieve a sustainable growth pattern with respect to energy limitations and the necessity to comply with global strategies to curb CO 2 emissions. In order to achieve this, econometric methods adopting time series analysis is used. For this purpose, the paper is organized as follows: The following section analyses the economic and environmental parameters of Greece. Next, the methodology adopted and the data used are presented. The paper continues with the presentation of results. Finally, the results are discussed in section 5.

ANALYSIS OF ECONOMIC, CO 2 EMISSIONS AND ENERGY PARAMETERS IN GREECE
As can be seen in Figure 1, per capita GDP, per capita CO 2 emissions and per capita energy use follow an almost common trend through the years. By taking 1960 as base year, all three variables increase until the beginning of the last economic crisis started in 2008. During the 1960-2008 period, real GDP per capita has grown 4 times compared to 1960s levels, while energy use per capita and CO 2 emissions have grown respectively 9 and 7,5 times compared to 1960s levels. Therefore, the growth of energy use and CO 2 emissions per capita overcomes the growth of GDP per capita ( Figure 1). However, from 2008, all three variables report declining trends indicating that the negative GDP growth leads to decreased energy use and CO 2 emissions.
The growth of urbanization rate is almost stable over the 1960-2014 period and does not seem to be affected by the 2008s economic crisis or other economic cycles in Greece ( Figure 1).
CO 2 emissions and energy use in Greece seem to be coupled over the years as shown in Figure 1 and both variables seem to be determined by GDP changes over time. Hondroyiannis et al. (2002) argue that total energy consumption is an important inherent determinant of economic growth during 1960-1996. According to the authors, from the early 1960s until the Second Oil Crisis in 1979, the Greek economy is experiencing high rates of economic growth fuelled by the significant growth of energy consumption in the industrial sector. This outcome is verified concerning CO 2 emissions in Greece in Figure 2, where the growth of CO 2 emissions of the manufacturing industries and construction sector is greater than the grown of total CO 2 emissions. After the two oil crises, the slowdown in economic growth is followed by a slowdown of energy use the industrial sector (Hatzigeorgiou et al., 2011) while, according to Polemis (2007), since 1980, energy consumption of the industrial sector is growing at a lower rate compared to total energy consumption (Polemis, 2007).
In the sub-period 1980-1996, the Greek economy records low growth rates of real GDP, while growth rates of total CO 2 emissions and total energy use remain higher ( Figure 1). Over this period, the growth of CO 2 emissions of the manufacturing industries and construction sector fall behind the growth of total CO 2 emissions ( Figure 2).
The Greek economy seems to recover during the period of 1997-2007: the growth of real GDP is higher than the growth rates of total energy consumption and total CO 2 emissions (Figure 1), while energy consumption of the industrial sector remains low (Hondroyiannis et al., 2002). Contrary to the evolution of energy consumption in the industrial sector, energy consumption of the services sector and the residential sector has increased significantly since the mid-1970s (Hondroyiannis et al., 2002). An interpretation of this statement is the improvement of living conditions in Greece (Hondroyiannis et al., 2002), resulting in increases in the use of private cars (Zervas et al., 2006), electrical appliances (Hondroyiannis et al., 2002) and/or cooling systems in the summer months (Rapanos and Polemis, 2006). In accordance, CO 2 emissions of the services sector -as allocated in commercial and public services and transport in Figure 2 -and residential sector boost the growth of total CO 2 emissions in Greece ( Figure 2). Finally, from 2009 onwards, all sizes report negative growth rates (Figures 1 and 2) due to the last economic crisis in Greece.
In overall, the transition of the Greek productive structure from an economy based on industry to a service-based economy may explain the evolution of the level of CO 2 emissions in Greece for the period 1960-2014. Some researchers attempt to study the underlying relationships among GDP, energy use and CO 2 emissions in Greece. For instance, Jaunky (2011) The present study deviates from these studies in the following ways. First, the period under consideration is larger including the recent economic crisis in Greece. This allows testing whether the drop in the growth of CO 2 emissions after 2008 is due to a reduced growth of GDP or to improvements in energy use by testing short and long term causalities among the variables in question. Second, the study adopts an estimation model based on the EKC theory allowing testing whether an EKC pattern is confirmed in the period under consideration. Finally, the study considers the possible effect of urbanization rate on CO 2 emissions, following Kuznets proposal in his initial work considering the income inequality -economic growth nexus that urbanization is an affecting factor (Kuznets, 1955). Also, the effect of urbanization has not been tested in the case of Greece while is of high interest as an exogenous variable in other studies. For example, Shahbaz et al. (2014) include urbanization as an exogenous variable in their test relationship and conclude that urbanization increases CO 2 emissions in United Arab Emirates, while Liobikienė and Butkus (2019) conclude that urbanization has a "positive but not significant effect on GHG emissions" in a panel consisting of 147 countries. However, other studies that include urbanization as an exogenous variable do not support evidence in favor of an EKC-pattern (Dogan and Turkekul, 2016;Farhani and Ozturk, 2015).

Econometric Methodology
The relationship among per capita CO 2 emissions (ECO2), per capita energy use (CENER) and per capita gross domestic product (CGDP) is represented by the long-term equilibrium relationship: Where, Υ= CO 2 emissions per capita (ECO2), X= GDP per capita income (CGDP), B= GDP 2 per capita income (CGDP2), Ζ= per capita use of energy (CENER), urb = the percentage of the urban population in year t, dum1974 = the dummy variable for the extreme value of CGDP in 1974.
In order to investigate how Y, X, B, Z variables are related, as well as to determine the dynamic direction of causality among these variables, we assume that the relationship can be approximated by a vector autoregressive models VAR at levels (Hondroyiannis et al., 2002). The VAR approach allows us to consider all series Y, X, B, Z as endogenous system variables. CGDP2 is inserted as an endogenous variable, following the EKC approach that considers the inclusion of a squared term of GDP in order to test the EKC hypothesis. In particular, the EKC hypothesis is confirmed if a 2 is negative and statistically significant. Moreover, we do not suppose that any of these variables is determined exogenously and unilaterally affect the other variables of the model. The vector auto regression model includes also annual urbanization rate (urb) as well as a dummy variable for the extreme value of per capita GDP (cgdp) in 1974 as exogenous interpretative variables. We choose to include the urbanization rate as an exogenous variable in order to test whether urbanization exerts a significant effect on CO 2 emissions between 1960 and 2014 following the proposal of other studies (Shahbaz et al., 2014;Liobikienė and Butkus, 2019). Additionally, the extreme value of per capita GDP in 1974 is likely to be due to the political change from dictatorship to democracy in Greece this year, or due to the oil crisis in 1973 or a combination of both above factors.
The representation of the relationship through a VAR gives us the advantage to investigate the underlying causality relationships between variables.
Therefore, the vector regression model (VAR) is given by the following relationship: Where: Y t = [ECO2 t CENER t CGDPt CGDP2 t ] vector of endogenous variables, X t = [urb t dum1974] vector of exogenous variables, a 0 = [a ECO2 a CENER a CGDP a CGDP2 ] vector of fixed terms, Α J the matrix of VAR parameters for the delay j and, ε t vector of the error conditions.
The following steps are applied in this work. First, we test whether endogenous series are stationary. All endogenous series are transformed into logarithmic form allowing normalization of the series, that is, series have a linear trend (Asteriou and Hall, 2011). Then, we apply the augmented Dickey-Fuller unit root test (Dickey and Fuller, 1979;Dickey and Fuller, 1981), the Phillips-Perron test (PP statistic) (Phillips and Perron, 1988) and the KPSS test (Kwiatkowski et al., 1992).
The second step is to check for Johansen co-integration for each of the VARs. The Johansen co-integration process tests for longrun equilibrium among the variables (Johansen, 1991;Al-mulali et al., 2012). If the series are co-integrated then the series have a long-run equilibrium, even if there is imbalance in the short-run period. More specifically, the co-integration resolves the problem of non-stationary series. The Johansen co-integration test requires that all endogenous series are stationary or they are integrated at the same order (Asteriou and Hall, 2011) Otherwise, results may be spurious (Asteriou and Hall, 2011).
At this step, it is crucial to specify the optimal lag -length of the VAR that minimises the Schwarz (SC) criterion and satisfies tests for normality, autocorrelation and heteroscedasticity of the residuals (Asteriou and Hall, 2011). Next, in order to achieve the optimal specification, we adopt the Pantula principle. Johansen (1991) suggests that the joint hypothesis of both the rank order and the deterministic components need to be tested, applying the Pantula principle. The Pantula principle involves the estimation of all three plausible models in terms of economic theory and the presentation of the results from the most restrictive hypothesis to the least restrictive hypothesis. The method and the corresponding test statistic for determining the number of cointegrating relations is based on the trace statistic (Asteriou and Hall, 2011).
In case of co-integration, Granger causality analysis is applied among the variables based on the multi-variable Vector Error Correction Model (VECM) in order to check long-term direction of causality. If co-integration is rejected, a simple Granger method is applied, to check for short-term direction of causality.

Data
Data on CO 2 emissions per capita (named ECO2), real gross domestic product per capita (named CGDP), gross use of energy per capita (named CENER) and annual urbanization rate (named URB) are derived from the World Development Indicators (WDI, 2019). Per capita CO 2 emissions are per capita emissions from fossil fuel combustion and cement production and are expressed in metric tons per capita. Real GDP per capita is expressed in constant 2010 U.S. $. Per capita use of energy includes per capita gross per capita energy (before its transformation into any other form of energy) and is expressed in kg of oil equivalent per capita. The urbanization rate refers to the percentage of total population living in urban areas. All data cover the period from 1960 to 2014. Table 1 shows the results concerning the stationarity of series. The series contain a unit root at their logarithmic levels, but are stationary in their first logarithmic differences. Therefore, the series are integrated at the same order (I(1)).

Johansen Co-integration Test
Since series are integrated at the same order (I(1)), we use the vector autoregressive relation (Shafic and Bandyopadhyay, 1992) in order to specify the optimal number of time lags in accordance to Schwarz criterion (SC). The optimum number is set to k=1. The results of Johansen co-integration test are shown in Table 2 according to the Pantula principle, in order to specify the type of co-integration. According to the trace statistic criterion, in three forms of specification (Asteriou and Hall, 2011), the null hypothesis for the existence of a co-integration relationship between the variables is accepted. So, the long-term co-integration function is shown in Table 3. According to the obtained results, all coefficients are significant at 5% significance level in the cointegrating equation. Per capita use of energy is positive, indicating that a rise in per capita energy use increases per capita CO 2 emissions. More specifically, a 1% increase in per capita energy use -holding all other variables stable-leads to approximately a 0.74% increase in per capita CO 2 emissions.
However, the EKC -hypothesis between per capita CO 2 emissions and per capita GDP is not confirmed over the 1960-2014 period, since the coefficient of the squared-term of per capita GDP is positive, indicating a U relationship between the variables in question rather than an inverse U pattern (Table 3). Moreover, the variable of urbanization is positive and statistically significant, indicating that a 1% rise in the growth of urbanization increases marginally per capita CO 2 emissions 0.0073% over the 1960-2014 period. Also, the dummy variable for year 1974 is negative and significant.
The error correction term (ECT) a ii is significant and implies that the system needs approximately 5 years and 4 months to return to long-run equilibrium, which is a rather slow rate of convergence.

Granger's Causality Test
The results of Granger's causality test are shown in Tables 4 and 5. In particular, Table 4 depicts results concerning Granger causality in the short run, while Table 5 summarizes results in the long-run period based on the joint test including the error correction term (ECT).
According to Table 4, there is a bi-directional Granger causality between per capita GDP (LCGDP) and per capita CO 2 emissions (LECO2) in the short run. The same outcome is confirmed in the long-run period according to Table 5.
No Granger causality is confirmed between per capita GDP (LCGDP) and per capita energy use (LCENER) in the short-run (Table 4); however, per capita energy use (LCENER) Granger causes per capita GDP (LCGDP) in the long run (Table 5).
Concerning causality between per capita CO 2 emissions (LECO2) and per capita energy use (LCENER), there is a uni-directional   causality running from per capita CO 2 emissions (LECO2) to per capita energy use (LCENER) in the short-run (Table 4), while there is a bi-directional causality between the variables in question in the long run (Table 5).

DISCUSSION AND CONCLUSION
The long-run relationship between real GDP and CO 2 emissions in Greece under the EKC-hypothesis has not been studied adequately using econometric applications under the EKC perspective. This paper investigates the short-term and long-term relationship between per capita income, per capita use of energy and per capita CO 2 emissions under the Environmental Kuznets Curve (EKC) concept in Greece over the period 1960-2014 by using appropriate econometric methods. Furthermore, the impact of the urbanization rate on CO 2 emissions per head is considered.
Econometric analysis suggests that time series of per capita CO 2 emissions, per capita GDP and per capita use of energy are stationary in their first differences and there is a co-integrated relationship. According to the results, CO 2 emissions are not declining in Greece following an EKC-pattern in the period under consideration. Rather, the relationship between per capita GDP and CO 2 emissions follow a U pattern contrary to findings of Acaravci and Ozturk (2010), who confirm an EKC-pattern in Greece in the 1960-2005 period. In addition, according to our results, urbanization rate has a significant but minor impact on per capita CO 2 emissions. Possibly, urbanization in Greece leads to increased energy use and increased energy consumption of transport, as seen in many high and middle income countries (Al-mulali et al., 2012).
Based on the results concerning Granger causality, per capita CO 2 emissions and per capita GDP are coupled over time since there is a bi-directional causality between the variables in the short and long term. This indicates that policies to reduce CO 2 emissions have cost in economic growth terms or, put differently, the growth of the Greek economy is of high carbon intensity. However, per capita energy use Granger causes per capita GDP in the long run while there is a bi-directional causality between per capita energy use and per capita CO 2 emissions in the long run. This implies that policies to reduce the carbon intensity of energy use in Greece could delink the long-run relationship between GDP growth and CO 2 emissions in Greece. In any case, there is no confirmation of an Environmental Kuznets Curve (EKC)-pattern in Greece during 1960-2014.
From these results it can be said that the adoption of policies targeting at the optimal energy use in Greece could lead to economic growth with lower CO 2 emissions. For example, a shift to the use of renewable energy sources to complement energy needs in Greece would lower carbon intensity of energy use in Greece contributing to the de-escalation of CO 2 emissions without inhibiting the economic growth process.
Furthermore, improving transport infrastructure in large urban centers, where the majority of the population is concentrated, could help rationalize the energy use and de-escalate CO 2 emissions. In addition, policies able to enhance energy efficiency may be useful, for example by increasing competition, security of supply, sustainability and transition to a low-carbon economy. All these policies have to be measured for industrial appliances, vehicles, households and the building stock.
Finally, policymakers in Greece could adopt strategies proposed in Paris Treaty for climate change for the reduction of CO 2 emissions such as funding measures for mitigating climate change.