Ecological Footprint, CO2 Emissions and Economic Growth in Qatar : Evidence from a Markov Switching Equilibrium Correction Model

Ecological footprint, CO2 emissions and economic growth in Qatar : Evidence from a Markov Switching Equilibrium Correction Model

Charfeddine Lanouar1, Department of Finance and Economics, College of Business and Economics, Qatar University, Qatar.

Abstract Reducing the impact of air pollution and global environment degradation on Human health and the quality of Qatari living is one of the most important pillars of Qatar 2030 vision. With respect to this vision, we examine the effects of economic growth, energy consumption, urbanization, openness trade and financial development on environment quality during the period 1975-2011 in Qatar. Unlike the existing studies, we use the ecological footprint and the 퐶푂2 emissions as indicators of environment degradation. Moreover, we use Markov Switching Equilibrium Correction Model with shifts in both the intercept and the income per capita coefficient. Our finding show strong evidence for cointegration with Markov shifts. We found, for both pollutants, that the EKC hypothesis holds for the Qatar economy when accounting for possible shifts. Moreover, we found that financial development, urbanization and openness trade worse environment. In contrast, we found that the effect of electricity consumption on 퐶푂2 emissions is positive and negative for ecological foot print.

Keywords : Environment degradation, EKC, CO2, foot print, economic growth, Cointegration with Markov shifts.

JEL classification :

1 P.O.Box: 2713-Doha-Qatar. Email : [email protected]. Office : (+974) 4403-7764, Fax : (+974) 4403-5081.

1. Introduction

“Long live the planet. Live Humanity. Long live life itself.” With this slogan of the COP21 Paris climate change conference of November-December 2015, there is actually no doubt in the priority of preserving the planet from the growing warmer of the earth’s atmosphere in the coming years. The main outcome from this conference, is the large commitment of more than 187 countries around the world to reducing their greenhouse gas in order to keep the rise in temperature below the level of 2°C. Human activity and rapid population increases are the most two important factors largely advanced to explain increases of Greenhouse Gases (GHS). The U.S and China are the largest emitter in the world with 21 percent and 15 percent respectively in absolute term (World Resources Institute, 2005). However, focusing on absolute pollutant level will gives a partial understanding and does not lead to completely understanding the overall reality about environment degradation. Thus, it is very important to have a deeper investigation in term of relative pollutants and in its key determinants. This is of great importance and for a particular interest especially when dealing with a countries characterized by high level of pollutant per capita such as CO2 emissions per capita or ecological footprint per capita.

This paper investigates the relationship between environment degradation and economic growth as well as with some others key determinants of environment degradation in Qatar. This paper is mainly motivated by the controversy effects of the rapid economic development that have experienced Qatar in the last four decades due to its abounded hydrocarbon resources. This rapid economic growth has severely affected the environment quality of the country and has increased the need to preserve and protect its ecosystem including water, air and lands. The current situation of Qatar in term of environment degradation is very critical. For instance, according to the World Bank Indicators (WBI, 2011), Qatar is ranked the first in term of 퐶푂2 emissions per capita and the second in term of ecological footprint per capita according to the Living Planet Report of 2014. Following the WWF’s Living Planet Report of 2014, “if all people on the planet had the Footprint of the average resident of Qatar, we would need 4.8 planets”.

Moreover, Qatar 2030 vision has given a high importance to questions related to air pollution, climate change and their impacts on economic sustainability. For example, three out of the four pillars of Qatar 2030 vision are directly or indirectly related to preserving the environment2. Air pollution, global greenhouse gases (GHG), water pollution and water resources degradation are among the most serious environmental concerns that actually encounter the country. In addition, the needs for a high level of economic growth accompanied with a rapid increase of urbanization and international trade have made a strong pressure in the country energy use considered as the main source of CO2 emissions of the country.

The Qatar situation in term of air pollution is also very worrying, the local air pollution levels in Qatar has frequently exceeded recommended levels and are more time higher than the international standards. In fact, compared to the WHO’s standards of the 24-hour and annual averages concentration of 50 휇g/m3and 20 휇g/m3 for PM10 the Qatar’s national air quality standards are far from these values. For instance, the values for PM10 is around 150 휇g/m3 for 24 hours average concentration and around 50 휇g/m3 for the annual average concentration (see the world health

2 The four pillars of Qatar 2030 visions are : (1) economic development, (2) social development, (3) human development, and (4) environmental development.

organization, WHO). These high level have increased the likelihood of diseases related to the respiratory system such as asthma, chronic obstructive pulmonary disease among many others.

Given that situation, several important questions arise for this country.

1. Is a continuous rise of income positively monotonically associated with environmental degradation proxy? Or is the relationship nonlinear (U-or Inverted U-shaped)? 2. Which are the most important determinants (macroeconomic and financial variables) of environmental degradation? 3. What it the type of the relationship between environmental degradation proxy and its determinants? Is the relationship linear or non-linear? 4. Is the environment degradation- economic growth nexus environmental degradation proxy dependent? 5. Finally, how is the causal relationship between environment degradation proxies and its key determinants?

The answers to these questions are critical at this stage of economic development of the Qatar economy and are of particular interest for building and designing the appropriate strategies for reducing environment degradation.

The first question which is related to the relationship between environment degradation and economic growth can be answered by examining the type of the relationship between the environment degradation proxy and income proxy. In the empirical literature, recent studies have focused on whether the Environmental Kuznets Curve (EKC) hypothesis holds or not. Following the EKC hypothesis, the relationship between economic growth and environment degradation is inverted-U-shaped. From the economic perspective, this means that initially economic growth increases environment degradation and then declines it after a threshold point of income per capita. More specifically, at initial level of economic growth, an increase in income is linked with an increase in energy consumption that raises environment degradation. However, after reaching a critical level of income, the spending on environment protection is increased, and hence environment degradation tend to decrease. In the economic theory, three effects-types has been advanced as a channels through which economic growth impacts environment quality. The first channel is the scale effect which postulates that as the economy develop more inputs are needed in term of energy consumption, water, etc… inducing more degradation of the environment quality. The second channel is the composition effect following which as the economy develop the economic structure changes and switch from an agriculture based economy to an industrial based economy, and then to a service based economy. All along these switching processes, the needs of the economy changes and their impact on the environment changes also. Finally, the last channel corresponds to the technique effect which has a positive effect on the environment quality and which drives the curve downward. Following this relationship, to improve environmental quality the best way is to become rich (see for instance Beckerman, 1992; and Cole, 1999). From an econometrical or statistical perspectives, the EKC hypothesis can be tested by estimating the EKC equation which relies the environment degradation proxy to the real GDP and to a nonlinear term of the real GDP (the squared real GDP).

If the EKC hypothesis holds then the real GDP and the squared real GDP have respectively a positive and negative signs. This EKC hypothesis has been firstly introduced by Kuznets (1955) when examining the relationship between economic growth and income inequality which shows that this relationship is inverted U-shaped. Grossman and Krueger (1995) are the first to examine this relationship between environment degradation and economic growth in their seminal paper published on the Quarterly Journal of Economics. They found that this relationship is inverted U- shaped which validates the EKC hypothesis. However, until now there is no consensus about the true nature of the relation between real GDP and environment degradation. Evidence for the EKC hypothesis is very mixed. Overall, the results seem to depend in many factors including the specification, the pollutants and the econometrics technique used (Shafik and Bandyopadhyay, 1992; and Stern, 1996, 2001). First, empirical studies show that the results in term of positive and negative relationships as well as in term of magnitude differ significantly for the same country depend on the specification studied, linear, quadratic or cubic. Moreover, the inclusion of other factors in the right hand of the regression such as urbanization, trade openness, financial development and political stability have a significant impact on the magnitude of the income per capita variables coefficients. Second, the results differ significantly following the environment degradation proxy used. For instance, Horvath (1997) and Holtz-Eakin and Selden (1995) suggest that the use of global pollutants leads to continuously rise the levels of environment degradation or to a high levels of income per capita turning point, see also Esteve and Tamarit (2011). Third, the results also seem to depend in the econometric approach employed. To answer the second question, we augment the basic EKC equation by adding several macroeconomic and financial variables such as energy use (electricity consumption), urbanization, openness trade and financial development. So, by determining which factors explain environment degradation, policymakers, researchers and international institutions can help on recommending the adequate economic policies that can improve the environment quality and the live standing of inhabitants.

In the literature, energy use is considered as the second key determinant of environment degradation. In the empirical literature several proxies of energy use have been employed such as energy consumption, energy electricity among many others. The relationship between energy consumption and environment quality is expected to be positive as energy consumption is supposed to increase the scale of the economy and worse environment. Several empirical studies have documented that high level of energy electricity leads to a high level environment of degradation. Others studies suggest that electricity consumption is an efficient energy and it decreases environment degradation.

The third potential determinant of environment degradation is urbanization. Previous literature is inconclusive on the relation between urbanization and environment degradation. Urban studies advance three theories to explain the channels through which urbanization affects environment degradation : ecological modernization; urban environmental transition; and compact city theories, see for instance Poumanyvong and Kaneko (2010). Empirically, a positive relationship between environment degradation and urbanization is found in several studies (see for instance, Cole and Neumayer, 2004; York, 2007; Liddle and Lung, 2010; Poumanyvong and Kaneko, 2010; Kasman and Duman, 2015; among others). In the other hand, a negative relationship is found by Fan et al.

(2006), Dodman (2009), Sharma (2011), Hossain (2011) among others. For these authors, the increase of urban population contributes positively to the development of the structures of modernity via orienting urbanization towards smart cities, new ICT to shorten the time of transportation and to spread the usage of e-works, the use of clean and renewable energy sources which improve environment quality, see for instance Ehrhardt-Martinez (1998). In addition to these two previous groups of studies, many others studies show that this relationship is ambiguous and depend in many others factors and especially in the economics structures of each country. For example, Poumanyvong and Kaneko (2010) show based on a study for 99 countries, using the stochastic impacts regression on population affluence and Technology model over the period 1975-2005, that urbanization decreases the level of energy use in low-income countries, while it increases energy use in middle and high-income countries. Recently, Sadorsky (2014) found mixed results about this relationship and shows that the result depends on the estimation techniques. More recently, Chikaraishi et al. (2014) show that improving urbanization make countries more environmentally friendly when the country's GDP per capita and the percentage share of service industries in GDP are sufficiently high.

The fourth potential determinant of environment degradation is openness trade. Many studies have showed that the impact of openness trade on environmental quality can be decomposed into scale, technique, and composition effects, see for instance Antweiler et al. (2001). Empirically, mixed results have been reported in the previous works. Some studies find a negative relationship between trade openness and environmental degradation (Shahbaz et al.,2013; Hossain, 2011; Shaabaz et al., 2012; Jayanthakumaran, 2012). On contrary, other studies find evidence of positive effect on environmental degradation (Suri and Chapman, 1998; Alber et al., 1999; Kasman and Duman, 2015 and Ozturk and Acaravci, 2013). Recent studies in the field supports evidence that financial development is also a key determinant of environment degradation. Most of studies argue that financial development can affect environment degradation through efficient/inefficient allocation of financial resources. Moreover, the development of financial markets can affect positively the environment quality by providing more funds to investments in research and development on modern and efficient technologies relating to clean energy. Empirically, there is some evidence of a negative effects of financial development on environment degradation, see for instance Shahbaz et al. (2013b) for the Indonesia case, Shahbaz et al. (2013a) for the Malaysia case, Charfeddine and Khediri (2016) for the case of UAE and Al-mulali and Sab (2012) for thirty Sub Saharan African countries. Ozturk and Acaravci (2013) find no significant effect of financial development on per capita carbon emissions in the long- run. In contrast to all these previous results, Boutabba (2014) finds that financial development increases 퐶푂2 emissions for the India case. Question three concerns the type of the relationship linking environment degradation and economic growth and the other explanatory variables. Recent studies in the field provide strong evidence that this relationship is nonlinear (see for instance Charfeddine and Khediri, 2016; and Shahbaz et al. 2014). Charfeddine and Khediri (2016) show that the 퐶푂2 emissions, real GDP per capita, electricity consumption, financial development and urbanization are cointegrated and have a long run relationship. However, the authors show using Gregory and Hansen (1996) and Hatemi-J (2008) that this long run relationship is characterized by the presence of structural breaks. Esteve and Tamarit (2011a) examine the long run relationship between 퐶푂2 emissions and income using threshold cointegration for Spain during the period from 1857-2007. They found evidence for nonlinear

relationship using Hansen and Seo (2002) approach. In a second paper Esteve and Tamarit (2011b) show that linear cointegrated regression model with multiple structural breaks describe better the relationship between 퐶푂2 emissions and income per capita. The authors found evidence for two breaks (three regimes) and that the estimated coefficient associated to the income-per capita in the long run relationship (long run-elasticity) is decreasing overtime. The authors conclude that even if 퐶푂2 emissions per capita is monotonically increasing with income, the long run elasticity of income decreases overtime and is less than one. In this paper, we follow another way in examining the EKC hypothesis for Qatar. We consider the possibility of non-linear relationship between environment degradation proxy and real GDP per capita using Markov switching equilibrium correction model. The nonlinearity is allowed in both variables (U-shaped versus inverted U-shaped) and parameters (the intercept and the coefficient associated to the real GDP). To the best of our knowledge this specification have not been yet used before to investigate the Kuznet curve hypothesis.

To answer the fourth question which aims to examine whether the environment degradation- economic growth nexus is environment degradation proxy dependent, we use two different proxies of environment degradation namely : (1) 퐶푂2 emissions per capita which is the most widely used proxy in the empirical literature, and (2) the ecological footprint proxy which is rarely used in empirical studies. The ecological footprint proxy is a more global proxy of environment degradation compared to 퐶푂2 emissions. Ecological footprint is a measure of the demand populations and activities place on the biosphere in a given year, given the prevailing technology and resource management of that year.

Finally, the answer to the last question concerning the causality direction between all used variables is critical for recommending the more appropriate economic policy that are able to reduce environment degradation. For instance, the causality direction between electricity consumption and economic growth will provide us with which of the following : growth, conservation, neutral or feedback hypothesis holds for Qatar.

This paper contributes to the empirical literature of the EKC hypothesis in many ways. First, to our knowledge this paper is the first to consider the case of the Qatar economy as a single country to test the EKC hypothesis as well as the different directions of causality between variables. Second, in addition to the 퐶푂2 emissions largely employed in the empirical literature, in this paper we employ also the ecological footprint as a new proxy of environmental degradation. Third, we use recent development of cointegration approach with structural breaks which is also rarely used for the case of EKC hypothesis. As tests of cointegration with shifts in the cointegration vector, we use the Gregory and Hansen (1996), Hatemi-J (2008) and to investigate the causal relationship between all variables using standard Granger causality tests. Fourth, to our knowledge this paper is the first study that uses Markov Switching Equilibrium Correction Model with shifts in both the intercept and the income per capita coefficient for the long run relationship between environment degradation and its key determinants. The remainder of the paper is organized as follows. Section 2 presents the empirical methodology to follow. Precisely, it presents how the data is collected and the variables are measured, unit roots tests with structural breaks and cointegration with single and multiple structural breaks techniques. Section 4 discusses the empirical findings. Finally, section 5 concludes and propose some policies that help to reduce environment degradation.

2. 푪푶ퟐ emissions and ecological footprint in Qatar

Which environment degradation proxy is more suitable when investigating the EKC hypothesis? The answer to this question seems to depend in many factors such as the economic structure and the stage of economic development of the country under study. However, there is some evidence among researchers that the EKC hypothesis holds more for local and regional pollutants than for global pollutants. Recent empirical studies have focused more in global pollutant such as CO2 emissions rather than in local pollutants (see for instance Khedir and Charfeddine, 2016 and references therein).

In this study, we follow the second trend of literature by using two proxies of environment degradation namely the CO2 emissions and the ecological footprint indicator. While the data availability motivate the use of the CO2 indicator, the ecological footprint indicator has emerged in recent years as an important indicator of environment degradation level. In contrast to the CO2 emissions pollutant largely considered in empirical studies due to the CO2 emissions undesired effects, or to it is adversely effects on the resource usefulness, the ecological foot print indicator have the advantage of measuring the total humanity's demand on nature at individual, national, regional or world level. Moreover, compared to the CO2 emissions, the ecological footprint indicator is a more global environment degradation proxy which account for cropland, forest, forest, grazing and crop land as well as fishing land and Carbon emissions.

2.1. Qatar Carbon Dioxide Emissions by source

The actual situation of the state of Qatar in term of both environment degradation proxies (CO2 emissions and ecological footprint) are presented in Figures 1 and 2. Figure 1 report the CO2 emissions per capita for Qatar and the five other GCC countries for comparison. The Qatar, Kuwait, UAE and Bahrain are among the top 10 emitters of CO2 emissions per capita. Moreover, as previously indicated Qatar is ranked the first in the world in term of CO2 emissions. For instance, according to the World Bank data (2010), Qatar greenhouse gases emissions is evaluated at 40.3 tonnes of carbon dioxide per capita being emitted every year. Compared to the world average, this figure is around 10 times higher than the global world average evaluated at 4.6 tonnes per capita. 120

100

Qatar Bahrain Kuwait Oman Saudi Arabia UAE

Figure 1 : CO2 emissions per capita evolution

Four sectors contribute to more than 99% of total CO2 emissions in Qatar. The electricity and heat production and the manufacturing industries and construction sectors are the largest contributors to CO2 emissions with a share that present approximately 75% to 85%, see Figure 2. The electricity and heat production sector contributes highly to the CO2 emissions, this sector contributes approximately between 45% to 63% during the period of our study. The second contributor to the CO2 emissions is the manufacturing industries and construction sector with a contribution between 20% and 45%. In contrast to the electricity and heat production sector, the manufacturing industries and construction sector show in the last years a decreasing of CO2 emissions levels which can be explained by the regulations laws that limit the amount of CO2 emitted by manufacturing industries. In addition to these two important sectors, in the last ten years, the transport sector emerges as the second contributor to CO2 emissions with approximately a same level than the manufacturing industries and construction sector.

100 Electricity and heat production in (%) 90 Manufacturing industries and construction Residential buildings, commercial and public services 80 Transport 70 60 50 40 30 20

10 Years 0

Figure 2 : Qatar Co2 emissions by components 2.2. Qatar foot print by source

In term of ecological footprint, the situation does not differ significantly, Qatar in ranked the second in the word. Moreover, from Figure 3, one can conclude that Carbon dioxide is the bigger contributor to ecological footprint indicator. This can be largely explained by the rapid increase of the transportation, industrial and domestic fuel burning as well as the industrial process since discovering the oil and gas resources at the end of 1970.

12 Carbon

10 Built_up_Land

Fishing_Ground 8 Forest_Land

6 Grazing_Land Crop_Land 4

Number of Earths demanded Earths of Number 2

1975 2003 1961 1963 1965 1967 1969 1971 1973 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2005 2007 2009 2011 2013 2015

Figure 3 : Qatar ecological footprint by components

3. Empirical methodology

Since the objective of this paper is to investigate the true nature of the relationship between 퐶푂2 emissions and ecological foot print to their macroeconomic and financial variables, we begin in subsection 2.1. by presenting the data sources and how variables are constructed. Then, in subsection 2.2., we introduce the Markov Switching Equilibrium Correction Model (MS-ECM), the Silvestre et al. (2009) unit root tests with structural breaks, the Gregory and Hansen (1996) and Hatemi-J (2008) tests for cointegration with structural breaks and the Granger causality test.

3.1. Data and variables description 3.1.1. Data This paper uses macroeconomics and financial data, including CO2 emissions, ecological foot print, real GDP per capita, energy use, urbanization, financial development and openness trade, to investigate the EKC hypothesis for the Qatar economy. All the dataset except the ecological foot print variable are collected from the world Bank’s development indicators (WDI). The ecological footprint data is obtained from the National Footprint Accounts (NFAs) of the Global Footprint Network. This variable is employed as second proxy of environment degradation measures. The data set used in this paper is annual data which covers the period 1975Q1 to 2007Q4 for variables used for ecological footprint equation and covers the periods 1980Q1 to 2010Q4 for the CO2 emissions equations variables. To account for possible existence of structural break in the cointegration vector we convert the annual frequency data to quarterly frequency data using the

quadratic match sum method3. The use of quarterly frequency data instead of annual data increases the sample size and then allows us to investigate the possibility of long run relationship with multiple shifts in the cointegration vector (Gregory and Hansen, 1996; Hatemi-J (2008), Arai and Kurozumi, 2007; and Kejriwal and Perron, 2010).

3.1.2. Variables description This section is devoted to present all variables used to investigate the environmental Kuznets curve hypothesis (EKC). We classify these variables into three categories. The first category presents the two variables used as proxies of environment degradation named the 퐶푂2 emissions and ecological footprint variables. The second and third categories includes all independent variables supposed determine the environment degradation variables. The second category includes the macroeconomics and financial variables and the third category includes the socio-demographic variables. a. Environment degradation proxies i. CO2 emissions, dependent variable, is defined as the carbon dioxide emissions that are stemming from the burning of fossil fuels and the manufacture of cement. It includes carbon dioxide produced during consumption of solid, liquid, and gas fuels and gas flaring. It is measured in metric tons per capita. ii. Ecological footprint, dependent variable, is measured in global hectares. Following the NFAs, the ecological footprint is a measure of how much area of biologically productive land and water an individual, population or activity requires to produce all the resources it consumes and to absorb the waste it generates, using prevailing technology and resource management practices. b. Macroeconomics and Financial variables i. Income per capita is measured as the gross domestic product divided by midyear population (GDP per capita). Data are in constant 2005 U.S. dollars. To investigate the EKC hypothesis we use the real GDP per capita in level and its squared form. The economic theory suggests that if the EKC hypothesis holds the expected sign of the real GDP per capita in level will be positive and negative for its quadratic form. ii. Electricity consumption is equal to the production of power plants and combined heat and power plants less transmission, distribution, and transformation losses and own use by heat and

3 In the theoretical literature several methods have been proposed that allow to convert low frequency data (annual or quarterly) to more high frequency data (quarterly or monthly). In this paper, we use the local quadratic-match sum method which has many advantages compared to others methods. In particular, this method ) is not sensitive to the presence of outliers and structural changes in the original series. This is of particular interest for us as our main objective is to allow of existence of breaks in unit root and cointegration tests. The local quadratic-match sum method is the by default method used by the Database of Global Economic Indicators (DGEI) from the Federal Reserve Bank of Dallas. For more details about this method we refer to the work of Grossman [79].

power plants. It is measured in KWh per capita. We expect that electricity consumption variable will be negatively correlated to the environment degradation variable as this variable is friendly sources of energy. iii. Openness trade is the percentage of the total value of exports and imports as a share of GDP. Following Grossman and Krueger (1991) the reduction of trade barriers affects the environment quality via increasing the economic activity. Then, it is expected that the openness trade variable will have a positive sign if the production processes is heavy 퐶푂2 emitter. However, environmental protection laws have a significant impact on reducing 퐶푂2 emissions and other pollutants. In this paper, as Qatar is an oil exporting country where the share of gas and oil exports is very important in the total GDP, then we expect that this variable will have a positive sign. iv. Financial development is measured by the ratio of domestic credit to private sector scaled by GDP. Indeed, financial development may play a role in explaining environment degradation. For instance, financial development may help firms in adopting advanced cleaner and environment friendly technology in the energy sector which in turn improve environment quality. Financial development may also attract foreign direct investment, which enhance research and development (R&D) activities that improve economic activities, and hence, influence environmental quality (Frankel and Romer, 1999). On the other hand, Financial development can has a negative impact on environment quality by increasing manufacturing activities which in turn leads to an environment degradation through an increases of industrial pollution. c. Socio-demographic variables

As socio-demographic variable we use urbanization. This variable is calculated using World Bank population estimates and urban ratios from the United Nations World Urbanization Prospects. It is equal to the percentage of country’s population living in urban areas. While this variable is generally associated with greater level of environment degradation, an increase of this variable can improve environment quality through economies scale in the provision of sanitation facilities. Empirically, the existing literature is inconclusive on the impact that urbanization has on environment degradation. Then, we suggest that, a priori, no sign expectation can be made on the relationship between this variable and environment degradation.

3.2. Econometric Model 3.2.1. The Markov Switching Equilibrium Correction Model (MS-ECM) To investigate the EKC hypothesis for the Qatar economy, we propose to use a Markov-Switching equilibrium correction model with shifts in the intercept and the slope coefficients of independents variables. The general form of the proposed model can be written as follow,

푟 푠 훥푦푡 = 훼 퐸퐶푀푡−1,푠푡 + ∑푖=1 훤푗훥푋푡−푖 + ∑푗=1 휋푗훥푦푡−푗 + 푢푡, Where,

′ 퐸퐶푀푡−1,푠푡 = 훽푠푡푋푡−1 − 휇푠푡,

Where 훼 measures the rapid adjustment towards the long run equilibrium. 푋푡 is a vector of independent variables. 푦푡 is the dependent variable and 훥 is the first difference operator. 푟 and 푠 are the order lags of the independent and dependent variables in the short run relationship.

푠푡 is a dummy variable which takes values of 0 and 1. It is governed by a first order Markov chain which is given by, 0 with probability 푝 푠 = { 푡 1 with probability 푞

2 Where 푝 = 푃[푠푡 = 푖|푠푡−1 = 푖], 푞 = 푃[푠푡 = 푗|푠푡−1 = 푗] for 푖, 푗 = 1, 2 and ∑푖=1 푝푖푗 = 1. Under these notation the intercept and slopes coefficients can be written as follow,

휇푠푡 = 휇1( 1 − 푠푡) + 휇2푠푡 ,

푠푡 1 2 1 2 1 2 훽 = ((훽1 , 훽1 ), (훽2 , 훽2 ), … , (훽푛, 훽푛)), where 푛 is the number of independent variables. For 1 2 example (훽1 , 훽1 ) are the slope coefficients of the real GDP under the first and second regimes respectively. 3.2.2. Model specification The general form of the empirical model investigated in this study is given by the following equation,

퐸퐷 = 푓 ( 푌, 푀푎푐푟표, 퐹푖푛, Socio)

Where 퐸퐷 refers to the environment degradation variable measured in this paper by 퐶푂2 emissions or ecological footprint; 푌 refers to the income per capita variable, 푀푎푐푟표 refers to macroeconomic variables that explain environment degradation such as openness trade and electricity consumption. 퐹푖푛 is the financial development variable and finally 푆표푐푖표 represents the urbanization variable. In this paper, we propose After taking the logarithm of all variables, the specification is given by,

2 퐿퐸퐷푡 = 휇푠푡 + 훽1푠푡 퐿푅퐺퐷푃푡 + 훽2퐿푅퐺퐷푃푡 + 훽4퐿퐸푡 + 훽5퐿퐹퐷푡 + 훽6퐿푈푅퐵푡 + 훽6퐿푂푃퐸푁푡 + 휖푡

Where 퐸퐷 is the 퐶푂2 emissions and ecological footprint (퐹푃) proxies. The empirical methodology used to investigate the EKC hypothesis is summarized in the following four steps. First, we start by testing whether the 퐶푂2 emissions, environment degradation, the real GDP, electricity consumption, urbanization, trade openness and financial development are characterized by the presence of unit root or not. This step is an important precondition before testing for the existence of long run relationship between environment degradation proxy and explicative variables. For this end, we use both standard unit root tests (Dickey and Fuller, 1979; Phillips and Perron, 1988; and Kwiatkowski et al.; 1992) and unit root tests allowing for multiple structural breaks of Silvestre et al. (2003)). Second, for cointegration with structural breaks by using the Gregory and Hansen (1996), Hatemi-J (2008) and Arai and Kurozumi (2007) tests. Third, we estimate the short and long run relationship using the two steps of Engle and Granger (1987) cointegration approach. In addition, we check for residuals stability using the CUSUM and CUSUM squared tests. In the last

step, we investigate the linear and nonlinear Granger causality direction sense between all variables. In the following subsections, we presents the unit roots tests with structural breaks and the testing procedure for cointegration with structural breaks.

3.3. Unit roots test with structural break This subsection is largely inspired from Silvestre et al. (2009) paper. It is devoted to describes the unit root tests with structural breaks recently developed by Silvestre et al. (2009). These unit roots tests have many advantage compared to the existence literature on testing for unit roots when series are subject to change in regime. First, Silvestre et al. (2009) propose a large variety of tests including the M-tests introduced by Stock (1999). Second, the proposed tests allow for multiple changes under both the null hypothesis of unit root and the alternative hypothesis of stationarity. Third, the Silvestre et al. (2009) tests use the quasi-GLS detrending method advocated by Elliot et al. (1996) which permits tests that have local asymptotic power functions close to the local asymptotic Gaussian power envelope. Silvestre et al. (2009) treat the breaks dates as known and then they show that the limiting distribution is the same when the breaks dates are unknown. Silvestre et al. (2009) consider the following model,

푦푡 = 푑푡 + 푢푡

푢푡 = 훼푢푡 + 휈푡 푡 = 0, … , 푇

Where 푦푡 is a time series, {푢푡} is an unobserved mean-zero process. 푢0 is supposed to be equal to ∞ ∞ 0. The disturbance 휈푡 is defined by 휈푡 = ∑푖=0 훾푖휂푡−푖 with ∑푖=0 푖|훾푖| < ∞ and {휂푡} a martingale difference sequence adapted to the filtration 퐹푡 = 휎 − field{휂푡−푖; 푖 ≥ 0}. The long and short run 2 2 2 2 −1 푇 2 variance are defined as 휎 = 휎휂 훾(1) and 휎휂 = lim 푇 ∑푡=1 퐸(휂푡 ), respectively. 푇→∞

The deterministic component in (1) is given by, ′ ′ ′ ′ 푑푡 = 푧푡(푇0)휓0 + 푧푡(푇1)휓1 + ⋯ + 푧푡(푇푚)휓푚 ≡ 푧푡(휆)휓 Where ′ ′ ′ ′ ′ ′ ′ ′ 푧푡(휆) = [푧푡(푇푗), 푧푡(푇1), … , 푧푡(푇푚)] and 휓 = (휓0, 휓1, … , 휓푚) .

′ The different forms taken by the 푧푡(푇0) deterministic component and the 휓 coefficients define the three models considered by Silvestre et al. (2009) in this paper. These three models are defined by,

퐷푈푡(푇푗) Model 0 ′ 0 ∗ 푧푡(푇푚) = { 퐷푇푡 (푇푗) Model I ∗ ′ (퐷푈푡(푇푗), 퐷푇푡 (푇푗)) Model II

Where Model 0 refers to the “crash” or level shift model, Model I refers to the “changing growth” slope change model and Model II refers to the mixed change model as in Perron (1989)4. ′ ′ ′ Silvestre et al. (2009) define 푧푡(푇0) ≡ 푧푡(0) = (1, 푡) and 휓0 = (휇0, 훽0) and for 1 ≤ 푗 ≤ 푚 ′ 휓푗 = 휇푗 in Model 0, 휓푗 = 훽푗 in Model I and 휓푗 = (휇푗, 훽푗) fin Model II. 퐷푈푡(푇푗) = 1 and ∗ 퐷푇푡 (푇푗) = (푡 − 푇푗) for 푡 > 푇푗 and 0 otherwise. 푇푗 = [푇휆푗] which denotes the j-th break date, with [. ] defines the integer part, and 휆푗 ≡ 푇푗⁄푇 ∈ (0,1) is the break fraction parameter. To estimate the breaks dates, they use the global minimization of the sum of squared residuals (SSR) of the GLS-detrended model,

휆̂ = arg min 푆(훼̅, 휆) 휆∈⋀(휖)

Where 푆(훼̅, 휆) is the minimum of an objective function, see Silvestre et al. (2009) for more details. 훼̅ = 1 + 푐̅⁄푇 is a non-centrality parameter. ⋀휖 = {휆 ∶ |휆푖+1 − 휆푖| ≥ 휖, 휆1 > 휖, 휆푘 > 1 − 휖} and 휖 is a small arbitrary number, in practice the common value of 휖 = 0.15. Proposition 2 in Silvestre et al. (2009) show the rate of convergence is fast enough to guarantee a same limiting distribution as when the breaks dates are known.

The proposed tests are defined by,

퐺퐿푆 −1 2 2 −2 푇 2 −1 푀푍훼 (휆) = (푇 푦̃푇 − 푠(휆) )(2푇 ∑푡=1 푦̃푡−1) ,

퐺퐿푆 2 −2 푇 2 1/2 푀푆퐵 (휆) = (푠(휆) 푇 ∑푡=1 푦̃푡−1) ,

퐺퐿푆 −1 2 2 2 −2 푇 2 −1/2 푀푍푡 (휆) = (푇 푦̃푇 − 푠(휆) )(4푠(휆) 푇 ∑푡=1 푦̃푡−1) ,

퐺퐿푆 2 −2 푇 2 −1 2 2 푀푃푇 (휆) = ( 푐̅ 푇 ∑푡=1 푦̃푡−1 + (1 − 푐̅) 푇 푦̃푇 )⁄푠(휆) ,

̂′ ′ ̂′ 2 Where 푦̃푡 = 푦푡 − 휓 푧푡(휆) and 휓 are the estimated values of 휓. 푠(휆) is an estimate of the spectral density at frequency zero of 휈푡.

3.4. Cointegration with structural break

Testing for parameter instability in macroeconomic relationships is an important issue for both macroeconomic and econometric sides. As noted by many authors, the luck of control for the presence of structural breaks inside time series can lead to model misspecification problems and misleading interpretation (Gregory and Hansen, 1996; and Arai and Kurozumi, 2007). In Addition, the fact that we investigate the EKC hypotheses during a period of 36 years -from 1975 to 2011, then it is more likely that the cointegration relationships may be subject to structural breaks. Testing for cointegration with structural breaks is conducted using the Gregory and Hansen (1996)

4 See also Zivot and Andrews (1992), Ng and Perron (1995) and Lee and Strazicich (2003).

and Hatemi-J (2008) tests. The Gregory and Hansen (1996) approach is used when only break is detected in the long run. The Hatemi-J approach is employed to investigate possible cointegration with two structural breaks. The approach proposed by Hatemi (2008) is an extension of the Gregory and Hansen (1996) approach to the case of cointegration vector with two breaks. In what follow, we present these three approaches in more details.

3.4.1. Gregory and Hansen (1996) cointegration with one break

Gregory and Hansen (1996) propose ADF, 푍훼 and 푍푡-type tests designed to test the null of no cointegration against the alternative of cointegration in the presence of level or regime shifts. Under the general hypothesis of cointegration with structural breaks, the model is given by:

′ ′ 푦푡 = 훼0 + 훼1퐷1푡 + 휃 푡 + 훽0푥푡 + 훽1퐷1푡푥푡 + 푢푡 , (1)

Where (훼0, 훼0 + 훼1) are the intercept coefficients both for the first and second regimes, ′ ′ respectively. In the same manner, 훽0 and (훽0 + 훽1) are the slope coefficients vectors under the first and second regimes, respectively. 휃 is the linear trend coefficient and 푢푡 is the error term. 퐷1푡 is a dummy variable defined by,

0 if 푡 ≤ [푛휏1] 퐷1푡 = { 1 if 푡 > [푛휏1] In this paper, we limit our investigation to the level shift (C) and regime shift models of Gregory and Hansen (1996). The level shift with trend (C/T) model is not considered in this paper for the simple reason that most economic time series can be adequately described by level shift and regime shift models, see for instance (Perron, 1989; Lee and Strazicich, 2003). The level shift (C) model as described by Gregory and Hansen (1996) is given by equation (1) ′ when we restrict 휃 = 0 and 훽1 to the zero vector. Under this assumption the level shit model is given by,

′ 푦푡 = 훼0 + 훼1퐷1푡 + 훽0푥푡 + 푢푡 , , Concerning the regime shift (level shift and slope coefficients change) (C/S) model, it is given by equation (1) when 휃 = 0.

′ ′ 푦푡 = 훼0 + 훼1퐷1푡 + 훽0푥푡 + 훽1퐷1푡푥푡 + 푢푡 ,

Gregory and Hansen (1996) proposed the three tests of cointegration with shifts that are an extensions of the standard ADF test of Dickey and Fuller and the Phillips two tests statistics Z  and Z t of cointegration. These three test are given by,

퐴퐷퐹∗ = inf 퐴퐷퐹(휏), 휏∈푇 ∗ 푍푡 = inf 푍푡(휏), 휏∈푇 ∗ 푍훼 = inf 푍훼(휏), 휏∈푇 15

These three tests of cointegration tests are calculated for each possible regime shift τ ∈ T, and then we take the smallest value for each test among all possible breaks points. We refer to Gregory and Hansen (1996) for more details. 3.4.2. Hatemi-J (2008) cointegration with two breaks Hatemi-J (2008) extends the Gregory and Hansen (1996) cointegration approach with one structural break to the case of cointegration with two structural breaks. As in Gregory and Hansen (1996), Hatemi-J (2008) consider the case where both intercept and the slopes are subject of structural breaks (two breaks here). The model considered by Hatem-J (2008) is given by,

′ ′ ′ 푦푡 = 훼0 + 훼1퐷1푡 + 훼2퐷2푡 + 휃 푡 + 훽0푥푡 + 훽1퐷1푡푥푡 + 훽2퐷2푡푥푡 + 푢푡 ,

퐷1푡 and 퐷2푡 are dummy variable defined by,

0 if 푡 ≤ [푛휏1] 0 if 푡 ≤ [푛휏2] 퐷1푡 = { and 퐷2푡 = { 1 if 푡 > [푛휏1] 1 if 푡 > [푛휏2]

Where 휏1 and 휏2 represent the location of breaks (unknowns). They are supposed to lies in the interval (0, 1) and that 휏2 > 휏1. In that case, (훼0, 훼0 + 훼1, 훼0 + 훼1+훼2 ) are the intercept ′ coefficients both for the first, second and third regimes, respectively. In a same manner, 훽0, ′ ′ (훽0 + 훽1) and (훽0 + 훽1 + 훽2) are the slope coefficients vectors under the first and second regimes, respectively. 휃 is the linear trend coefficient and 푢푡 is the error term. For this case of cointegration with two structural breaks, the level shift and regime shift models are given by,

′ 푦푡 = 훼0 + 훼1퐷1푡 + 훼2퐷2푡 + 훽0푥푡 + 푢푡 , and

′ ′ ′ 푦푡 = 훼0 + 훼1퐷1푡 + 훼2퐷2푡 + 훽0푥푡 + 훽1퐷1푡푥푡 + 훽2퐷2푡푥푡 + 푢푡 , As in the case of Gregory and Hansen (1996), Hatemi-J (2008) test statistics are the smallest values of these three tests across all values of 휏1 and 휏2. Hetemi-J (2008) suppose that 휏1 ∈ 푇1 = (0.15, 0.70) and 휏2 ∈ 푇2 = (0.15 + 휏1, 0.85)

∗ 퐴퐷퐹 = inf 퐴퐷퐹((휏1, 휏2)), (휏1,휏2)∈푇

∗ 푍푡 = inf 푍푡(휏1, 휏2), (휏1,휏2)∈푇

∗ 푍훼 = inf 푍훼(휏1, 휏2), (휏1,휏2)∈푇 3.5. VECM Granger causality Investigating causality direction within the framework of VECM is an important step after estimating the long run and short run relationship. The non-rejection of cointegration with

structural break suggests that at least one causal link exist among the series. To investigate this issue of causality, we use both standard linear Granger causality as introduced by Engle and Granger (1987). After refining the estimation of the long-run relationship, the Granger causality testing procedure can be done in the following four steps. a. The first step consists to recuperate the residuals of the long-run dynamics between all variables. These residuals are called 퐸퐶푇1푡 and 퐸퐶푇2푡 for the CO2 emissions and the ecological footprint environment degradation proxies.

These 퐸퐶푇1푡 and 퐸퐶푇2푡 residuals are obtained as follows,

̂ ̂ 2 ̂ ̂ ̂ 퐸퐶푇1푡 = (퐿퐸퐹푃푡 − 훽1푠푡퐿푅퐺퐷푃푡 − 훽2(퐿푅퐺퐷푃)푡 −훽3퐿퐸퐿퐸퐶푡 − 훽4퐿퐹퐷푡 − 훽5퐿푈푅퐵푡 − 휇푠푡 ) And,

̂ ̂ 2 ̂ ̂ ̂ 퐸퐶푇1푡 = (퐿퐶푂2푡 − 훽1푠푡퐿푅퐺퐷푃푡 − 훽2(퐿푅퐺퐷푃)푡 −훽3퐿퐸퐿퐸퐶푡 − 훽4퐿퐹퐷푡 − 훽5퐿푈푅퐵푡 − ̂ 훽6퐿푂푃퐸푁푡 − 휇푠푡)

̂ where, 훽푙, 푙 = 1, … ,6 are the parameters estimates of the coefficients of long-run relationship. b. In the second step, we estimate the short-run relationships for the CO2 emissions and the ecological footprint given by the two equations below,

푑1,1,푗 푑1,2,푗 푑1,3,푗 푑1,4,푗 푑1,5,푗 훥퐿퐸퐹푃푡−푗 훥퐿퐸퐹푃푡 휔1 훥퐿푅퐺퐷푃 푘 푑2,1,푗 푑2,2,푗 푑2,3,푗 푑2,4,푗 푑2,5,푗 훥퐿푅퐺퐷푃푡−푗 푡 휔2 휔 훥퐿퐸퐿퐸퐶푡 = 3 + ∑ 푑3,1,푗 푑3,2,푗 푑3,3,푗 푑3,4,푗 푑3,5,푗 훥퐿퐸퐿퐸퐶푡−푗 휔 훥퐿퐹퐷푡 4 푗=1 푑4,1,푗 푑4,2,푗 푑4,3,푗 푑4,4,푗 푑4,5,푗 훥퐿퐹퐷푡−푗 훥퐿푈푅퐵 휔5 ( 푡 ) ( ) (푑5,1,푗 푑5,2,푗 푑5,3,푗 푑5,4,푗 푑5,5,푗 ) ( 훥퐿푈푅퐵푡−푗 ) 휗 휋1 1,푡 휗 휋2 2,푡 휋 + 3 퐸퐶푇2푡−1 + 휗3,푡 휋 4 휗4,푡 휋5 ( ) (휗5,푡) 17

And, 푐1,5,푗 푐1,6,푗 훥퐿퐶푂2푡−푗 훥퐿퐶푂2푡 휃1 푐1,1,푗 푐1,2,푗 푐1,3,푗 푐1,4,푗 훥퐿푅퐺퐷푃 휃 푐 푐 푐2,5,푗 푐2,6,푗 훥퐿푅퐺퐷푃푡−푗 푡 2 푘 푐2,1,푗 푐2,2,푗 2,3,푗 2,4,푗 훥퐿퐸퐿퐸퐶 휃 훥퐿퐸퐿퐸퐶 푡 3 푐3,1,푗 푐3,2,푗 푐3,3,푗 푐3,4,푗 푐3,5,푗 푐3,6,푗 푡−푗 = + ∑ 훥퐿퐹퐷푡 휃4 푐4,1,푗 푐4,2,푗 푐4,3,푗 푐4,4,푗 푐4,5,푗 푐4,6,푗 훥퐿퐹퐷푡−푗 푗=1 훥퐿푈푅퐵푡 휃5 푐5,1,푗 푐5,2,푗 푐5,3,푗 푐5,4,푗 푐5,5,푗 푐5,6,푗 훥퐿푈푅퐵푡−푗 (훥퐿푂푃퐸푁 ) (휃 ) 푐 푐 푐 푐 푐 푐 푡 6 ( 6,1,푗 6,2,푗 6,3,푗 6,4,푗 6,5,푗 6,6,푗 ) (훥퐿푂푃퐸푁푡−푗) 휆1 푢1,푡 휆2 푢2,푡 푢 휆3 3,푡 + 퐸퐶푇1푡−1 + 푢 휆4 4,푡 푢 휆5 5,푡 (푢6,푡) (휆6)

c. Using these two specifications, the long run causality can be examined by testing the significance of the lagged error correction term using t-test statistic, 휋푗 = 0 and 휆푗 = 0 for 푗 = 1, . . ,6 . d. In the other hand, the short run causality is examined by testing the coefficient of the first differences of the variables. For example, 푐1,2,푗 ≠ 0 and 푑1,2,푗 ≠ 0 ∀ 푗 shows that the economic growth (훥퐿푅퐺퐷푃푡) Granger cause ecological foot print (훥퐿푃퐹푡) and CO2 emissions (훥퐿퐶푂2푡), respectively. The feedback that ecological foot print and CO2 emissions Granger cause economic growth is obtained when 푐2,1,푗 ≠ 0 and 푑2,1,푗 ≠ 0 ∀ 푗.

4. Empirical results 4.1. Preliminary analysis

Descriptive statistics of the ecological footprint, 퐶푂2 emissions, economic growth, electricity consumption, financial development and urbanization variables in term of mean, median, standard deviation, Skewness, Kurtosis and Jarque-Bera statistics are reported in Table 2 below. The results show that there is significant difference between the mean and the median for the both ecological foot print and real GDP variables. The normality hypothesis is rejected for all time series. Table 1 : Descriptive statistics and traditional unit root test

푳푬푭푷 푳푪푶ퟐ 푳푹푮푫푷 푳푬푳푬푪 푳푭푫 푳푼푹푩 푳푶푷푬푵 Mean 0.479 2.458 9.317 8.365 2.128 3.159 1.695

Median 0.581 2.563 8.856 8.327 2.173 3.167 1.703 Std.dev 0.455 0.293 0.952 0.141 0.347 0.032 0.106 Skew. -1.305 -0.695 1.093 0.398 -0.907 -0.311 -0.642 Kurt. 4.256 2.238 2.823 2.034 4.115 1.675 3.208 J-B 46.18 13.00 24.87 8.088 23.42 11.07 9.583

In the other hand, Table 3 reports the empirical results of pair-wise correlation (under the first matrix diagonal) and their corresponding t-statistics (above the matrix diagonal). Table 3 shows

that all correlations coefficients are highly significant except for the pair (퐿퐶푂2, 퐿퐸퐿퐸퐶). This non significance of the correlation between 퐶푂2 emissions and electricity consumption is not surprising as the electricity, as source of energy and input of the production process, is known to be a clean energy source compared to other energy sources such as oil, coal, gas etc… The pair- wise correlations between real GDP and both environment degradation proxies are positive suggesting that economic growth degrades environment. Moreover, financial development and urbanization are positively correlated to ecological foot print. In the other hand, the CO2 emissions is negatively and positively correlated with financial development and urbanization, respectively. Table 2 : Pair-wise correlation 푳푬푭푷 푳푪푶ퟐ 푳푹푮푫푷 푳푬푳푬푪 푳푭푫 푳푼푹푩 푳푶푷푬푵 푳푬푭푷 1 - 7.809 11.88 12.89 12.30 0.064 푳푪푶ퟐ - 1 3.302 1.564 -4.246 3.648 8.195 푳푹푮푫푷 0.565 0.278 1 12.93 3.775 15.93 4.803 푳푬푳푬푪 0.721 0.136 0.750 1 5.378 13.44 7.607 푳푭푫 0.749 -0.349 0.314 0.427 1 8.208 -6.652 푳푼푹푩 0.733 0.305 0.813 0.763 0.584 1 2.666 푳푶푷푬푵 0.006 0.596 0.399 0.567 -0.516 0.235 1 Pair-wise correlation under the first diagonal and their corresponding t-statistics are above the diagonal In addition to this preliminary analysis based on descriptive statistics and pair-wise correlation, we report in Figure 1 the trajectories evolution of the ecological footprint, 퐶푂2emissions, real GDP, electricity consumption, financial development and urbanization variables. This Figure shows that all trajectories has an upward trend during all the period of study, except for the 퐶푂2 emissions series which shows a downward trend until 1990, and then it remains stable with a new downward at the end of the period.

LFP LCO2

1.2 3.0

0.8 2.8

2.6 0.4 2.4 0.0 2.2 -0.4 2.0

-0.8 1.8

-1.2 1.6 1975 1980 1985 1990 1995 2000 2005 2010 1975 1980 1985 1990 1995 2000 2005 2010

LRGDP LELEC

12 8.8

8.6 11 8.4

10 8.2

9 8.0 7.8 8 7.6

7 7.4 1975 1980 1985 1990 1995 2000 2005 2010 1975 1980 1985 1990 1995 2000 2005 2010

LFD LURB

3.0 3.22

2.5 3.20

3.18 2.0

3.16 1.5 3.14

1.0 3.12

0.5 3.10 1975 1980 1985 1990 1995 2000 2005 2010 1975 1980 1985 1990 1995 2000 2005 2010

LOPEN

1.9

1.8

1.7

1.6

1.5

1.4

1.3 1975 1980 1985 1990 1995 2000 2005 2010

Figure 1 : Evolution of the 퐶푂2 emissions, Ecological foot print, real GDP, electricity consumption, financial development and urbanization variables

4.2. Unit roots tests with and without structural breaks

Non-stationarity of time series, e.g. I(1), is an important pre-condition before investigating possible long run relationship between environment degradation proxy and economic growth, electricity consumption, financial development and urbanization using cointegration analysis. In this study, to investigate non-stationarity, we use both traditional unit root tests (Ng-Perron, 1996) and unit root tests with structural break (Silvestre et al., 2009).

Empirical results are reported in Table 3 below for standard unit root test and Table 5 for unit root tests with structural breaks. Standard unit root tests results show that the ecological footprint, 퐶푂2 emissions, economic growth and financial development are non-stationary in level. For the electricity consumption and urbanization time series, we find that these series are stationary in level. Using first difference transformation, all variables are non-stationary except the 퐶푂2 emissions series. As noted by many researchers, macroeconomics time series are generally subject to structural breaks and the use of standard unit root test leads to highly reject the true behavior.

Table 3 : Results of Ng and Perron (1996) unit root tests without breaks. Level First difference 푮푳푺 푮푳푺 푮푳푺 푮푳푺 푮푳푺 푮푳푺 푮푳푺 푮푳푺 푴풁휶 푴푺푩 푴풁풕 푴푷푻 푴풁휶 푴푺푩 푴풁풕 푴푷푻 푳푬푭푷풕 0.550 1.301 0.716 102.85 -0.223 0.722 -0.162 30.917 a a a 푳푪푶ퟐ풕 -7.443 0.249 -1.854 3.568 -9.125 0.231 -2.109 2.793 푳푹푮푫푷풕 2.175 1.134 2.466 108.80 -1.368 0.510 -0.698 14.762 a a a a 푳푬푳푬푪풕 -10.71 0.213 -2.289 2.386 0.033 0.994 0.033 55.806 푳푭푫풕 -0.217 0.714 -0.155 30.42 -1.956 0.449 -0.880 11.36 a a a a 푳푼푹푩풕 -91.63 0.073 -6.714 0.374 -3.808 0.362 -1.378 6.433 푳푶푷푬푵풕 -6.613 0.275 -1.817 3.711 -7.640 0.254 -1.943 3.248 퐺퐿푆 퐺퐿푆 퐺퐿푆 The critical values are -8.10, 0.233, -1.98 and 3.17 for the 푀푍훼 , 푀푆퐵 , 푀푍푡 , and 퐺퐿푆 푀푃푇 tests respectively. The null hypothesis of Ng-Perron tests is the non-stationarity. The null hypothesis is rejected if the statistic is lower than critical values.

To account for structural breaks under both the null and alternative hypothesis when testing for unit root, we propose to use Silvestre et al. (2009) unit root tests. Empirical results reported in Table 5 shows strong evidence for the existence of unit root and for structural breaks in the level series at conventional level. The bottom panel of Table 5 shows that all series are stationary in first difference which means that all series are I(1) in level. Moreover, our results show that the structural breaks are in the origin for the rejection of the stationarity hypothesis when using standard unit root tests.

The first break occurs for all time series at the end of the seventy decade. The second break correspond to the middle of the eighty decade except for the electricity consumption series (1990Q4) and financial development (1981Q4). year 1990Q2 for the 퐶푂2 emissions and urbanization time series, and to 1981Q1 for electricity consumption, 1983Q3 for financial development, 1986Q4 for real GDP and 1999Q4 for ecological foot print. Finally, the last break correspond to the years 2004Q2, 1993Q3, 1997Q3, 1985Q4, 1992Q4 and 1995Q4 for the

ecological foot print, the CO2 emissions, the real GDP, electricity consumption, financial development and urbanization respectively.

Overall, Silvestre et al. (1994) unit root tests suggest that the ecological foot print, 퐶푂2 emissions, economic growth, electricity consumption, financial development and urbanization are I(1) in level and I(0) in first difference. This allow us to examine the possibility of long run relationship among investigated variables. This result suggest also that the presence of break in the level of these time series may be viewed as an indication of possible existence of shift in the cointegration vector.

Table 4 : Results of Silvestre et al. (2009) unit root tests with structural breaks. 푮푳푺 푮푳푺 푮푳푺 푮푳푺 푴풁휶 (흀) 푴푺푩 (흀) 푴풁풕 (흀) 푴푷푻 (흀) Breaks dates In level 퐿퐸퐹푃푡 -12.06 (-46.71) 0.200 (0.103) -2.421 (-4.828) 35.403 (8.99) 1978Q1, 1984Q3, 1990Q4, 1999Q4, 2004Q2 퐿퐶푂2푡 -11.20 (-45.74) 0.208 (0.105) -2.333 (-4.739) 37.339 (9.27) 1982Q3, 1985Q3, 1988Q3, 1992Q4, 1997Q4 퐿푅퐺퐷푃푡 -15.23 (-46.51) 0.181 (0.103) -2.759 (-4.827) 27.648 (8.88) 1981Q1, 1986Q4, 1996Q1, 2000Q4, 2005Q3 퐿퐸퐿퐸퐶푡 -8.87 (-46.05) 0.235 (0.103) -2.089 (-4.787) 47.55 (9.08) 1978Q1, 1990Q4, 1996Q3, 2000Q4, 2004Q3 퐿퐹퐷푡 -15.69 (-46.98) 0.177 (0.103) -2.785 (-4.826) 27.64 (9.12) 1978Q1, 1981Q4, 1986Q4, 1992Q4, 2000Q3 퐿푈푅퐵푡 -7.713 (-47.79) 0.234 (0.102) -1.802 (-4.874) 54.88 (9.38) 1979Q2, 1985Q4, 1989Q4,1995Q3,2004Q3 퐿푂푃퐸푁푡 -16.62 (-46.31) 0.168 (0.103) -2.793 (-4.80) 25.15 (8.88) 1980Q1, 1983Q4, 1987Q3, 1996Q4, 2001Q4 In first difference 퐿퐸퐹푃푡 -25.93 (-22.25) 0.138 (0.149) -3.593 (-3.301) -4.059 (-3.301) 1979Q4 퐿퐶푂2푡 -43.55 (-32.49) 0.107 (0.123) -4.651 (-3.991) 5.365 (7.295) 1978Q4, 1983Q4, 1986Q4 퐿푅퐺퐷푃푡 -44.97 (-35.22) 0.105 (0.118) -4.74 (-4.171) 6.282 (8.125) 1985Q4, 1989Q4, 1996Q4 퐿퐸퐿퐸퐶푡 -46.26 (-32.38) 0.104 (0.124) -4.81 (-4.016) 5.034 (7.175) 1978Q4, 1982Q4, 2003Q4 퐿퐹퐷푡 -42.864 (-36.52) 0.108 (0.116) -4.627 (-4.255) 7.028 (8.227) 1982Q4, 1990Q4, 1999Q4 퐿푈푅퐵푡 -41.25 (-40.92) 0.0110 (0.110) -4.54 (-4.51) 8.923 (8.939) 1982Q4, 1986Q2, 1997Q3, 2004Q1 퐿푂푃퐸푁푡 -45.31 (-32.95) 0.104 (0.121) -4.731 (-4.018) 5.665 (7.740) 1981Q4, 1984Q4, 1987Q4 Critical values for the 5% level significance are in parenthesis.

4.3. Cointegration with Markov shifts results

4.3.1. Testing for cointegration with structural break

The results of testing for cointegration with one and two structural breaks using the 퐴퐷퐹, 푍훼 and 푍푡-type tests proposed by Gregory and Hansen (1996) and Hatemi-J (2008) are reported in Table 5. Panel A corresponds to the results of the 퐴퐷퐹, 푍훼 and 푍푡-type tests of Gregory and Hansen (1996) when the cointegration vector is characterized by the presence of only one structural break and to the 퐴퐷퐹, 푍훼 and 푍푡-type tests of Hatemi-J (2008) when the cointegration vector is subject to two structural breaks. Empirical results of applying these tests on the level and slope coefficients shifts model are reported in column 3 of table 5. Critical values for the 퐴퐷퐹, 푍훼 and 푍푡 tests are obtained by simulation and reported in columns 4 and 5 for the 5% and 10% level of significance. These critical values depend on the number of explicative variables used, see Gregory and Hansen (1996) and Hatemi-J (2008) for more details. Table 5 : Results of Gregory and Hansen (1996), and Hatemi-J (2008) tests of cointegration with structural breaks 5% 10% Test Estimated Break Break Equation Critical Critical statistics value date 1 date 2 Value Value Panel A : Gregory and Hansen results 퐴퐷퐹∗ -15.383 -6.800 -6.561 1982Q2 - ∗ FP 푍푡 -15.442 -6.800 -6.561 1982Q2 - ∗ 푍훼 -175.00 -88.83 -81.94 1982Q2 - 퐴퐷퐹∗ -15.258 -7.191 -6.764 1986Q1 - ∗ CO2 푍푡 -15.305 -7.191 -6.764 1986Q1 - ∗ 푍훼 -164.30 -97.78 -91.58 1986Q1 - Panel B : Hatemi-J results 퐴퐷퐹∗ -4.755 -8.345 -8.017 1978Q3 2000Q1 ∗ FP 푍푡 -18.095 -8.345 -8.017 1980Q1 2000Q2 ∗ 푍훼 -185.99 -148.63 -142.33 1980Q1 2000Q2 퐴퐷퐹∗ -11.684 -8.987 -8.702 1991Q4 2000Q3 ∗ CO2 푍푡 -14.297 -8.987 -8.702 1993Q3 2001Q1 ∗ 푍훼 -176.89 -167.53 -159.28 1992Q2 2001Q4 The critical values are obtained by simulations, see Gregory and Hansen (1996) and Hatemi-J (2008).

Our empirical findings show strong evidence for cointegration with structural breaks using both Gregory and Hansen (1996) and Hatemi-J (2008) tests. For the Gregory and Hansen test, empirical results show that the null hypothesis of no cointegration is rejected at conventional level significance and using the three tests statistics which means that the alternative hypothesis of cointegration with one structural break cannot be rejected. For instance, the estimated values of ∗ ∗ ∗ the 퐴퐷퐹 , 푍푡 and 푍훼 for the 퐸퐹푃 and 퐶푂2 equations are -15.383, -15.442 and -175.00, and - 15.258, -15.305 and -164.30, respectively. As Table 4 shows, these estimated values are lower than the critical values of -6.800 and -88.83 for the 퐸퐹푃 equation and -7.191 and -97.78 for the

∗ ∗ ∗ 퐶푂2 equation respectively for the 퐴퐷퐹 and 푍푡 , and 푍훼 tests. To summary, this first result indicates that the hypothesis of cointegration with one structural break is highly accepted. Concerning the break date, the empirical findings show that the break date corresponds to the second quarter 1982 for the footprint equation and to the first quarter 1986 for the CO2 emissions equation. To overcome the limits of considering only one structural break for the Gregory and Hansen H (1996) approach, we discuss now the results of the Hatemi-J (2008) test. Panel B of Table 5 shows that for both equations (FP and 퐶푂2 emissions) the alternative hypothesis of cointegration with ∗ ∗ two breaks cannot be rejected by the 푍푡 and 푍훼 tests. Only for the ADF test that the null hypothesis of no cointegration cannot be rejected as the estimated statistic value of the ADF test (-4.755) is higher than the corresponding critical value (-8.345). Concerning the dates of breaks, the Hatemi- J (2008) procedure finds two breaks, the first one is approximately a common break and corresponds to the first quarter 2000 for both FP and CO2 equations. The second break corresponds to the third quarter 1978 for the ADF test and second quarter 1980 ∗ ∗ for the 푍푡 and 푍훼 tests for the FP equation. For the 퐶푂2 equation, we found a break that corresponds ∗ to the fourth quarter 1991, third quarter 1993 and second quarter 1992 when using the ADF, 푍푡 ∗ and 푍훼 tests. 4.3.2. Estimating the short and long run relationship After confirming the existence a long run relationship between the environment degradation proxy and its key determinants, we proceed in the second step to estimate the short and long run equation by using the two steps Engle-Granger technique. First, we start by estimating the long run relationship by applying a Markov switching model on time series in level I(1). In this paper, we consider that only the intercept and the slope coefficient of the real GDP over time5. Second, we recuperate the residuals of the first step and we use it as an error correction when estimating the short run equation. Empirical results of the two long run relationships for the FP and CO2 emissions equations are reported in Table 6 and in Table 7 for the short run relationships. 4.3.3. Long run analysis The estimated long run relationships allows to determine the long run marginal impact of the different independent variables such as economic growth, electricity consumption, financial development, urbanization and trade openness on the FootPrint and 퐶푂2 emissions environment degradation proxies. Empirical results for the two long run relationships with Markov shifts for the for the ecological Foot Print and 퐶푂2 emissions dependent variables are reported in Table 6 below. Table 6 shows that the majority of coefficients are statistically significant at conventional level. The Likelihood Ratio (LR) bound test proposed by Davies (1977, 1978) show that a two states Markov switching specification with shifts in both the intercept and slope coefficient of the real GDP per capita variable is more appropriate than the linear specification. Moreover, we found that the estimated values of the 푝11 and 푝22 probabilities are statistically significant at the 1% level significance. Moreover, these estimated values are close to the unit which means that each regime

5 The use of this specification of long run relationship with only Markov shifts in the intercept and real GDP slope coefficient is retained as the more appropriate specification after estimation of several long run relationship models with Markov shifts in the intercept and slopes coefficients of all independent variables.

6 last enough of time. In average, the first regime last for 31.765 Quarter for the 퐶푂2emissions pollutant equation and 68.21 Quarter for FP pollutant. For the second regime, it last in average approximately about 108.542 Quarter and 63.36 Quarter for the 퐶푂2 emissions and FP pollutants respectively. Table 6 : Estimation of the long run relationship with Markov shifts

푳푪푶ퟐ풕 푳푬푭푷풕 No shifts Markov shifts No shifts Markov shifts Coef. t-stat Coef. t-stat Coef. t-stat Coef. t-stat 흁ퟏ -14.992 -4.357 1.341 0.938 -21.709 -5.893 -41.66 -33.62 흁ퟐ - - -4.401 -3.309 - - -34.54 -27.85 ퟏ 휷ퟏ -3.864 -3.369 1.766 8.592 3.185 4.793 3.549 21.01 푳푹푮푫푷풕 ퟐ 휷ퟏ - - 2.354 11.50 - - 2.803 16.30 ퟐ 푳푹푮푫푷풕 0.172 3.198 -0.108 -10.59 -0.163 -4.821 -0.170 -19.32 푳푬푳푬푪풕 0.457 1.865 0.517 4.963 0.354 2.326 -0.183 -1.841 푳푭푫풕 -0.398 -4.646 -0.208 -4.470 0.517 9.530 -0.229 -4.371 푳푼푹푩풕 1.206 8.032 0.315 -7.287 0.919 0.768 0.818 21.96 푳푶푷푬푵풕 1.502 5.975 0.189 2.156 - - - - 풑ퟏퟏ - - 0.968 3.707 - - 0.985 3.427 풑ퟐퟐ - - 0.990 4.259 - - 0.984 5.022 Log-Likelihood 46.184 99.97 23.667 91.461 LR-test 107.57 135.59 Expected duration Regime 1 - 31.765 Q - 68.21Q Regime 2 - 108.542 Q - 63.36 Q

Figures 2 and 3 below show the evolution of the estimated probabilities smoothing. These figures can be used to date exactly the breaks dates. For the 퐶푂2 emissions case, the first regime covers the period from the first Quarter 1980 to the second Quarter 1991 and from the second Quarter 2001 to the fourth Quarter 2010. For the second regime it covers the period from the third Quarter 1991 to the first Quarter 2001. For the Foot Print case, the probabilities smoothing figure show that the first regime covers the periods from the first Quarter 1975 to the second Quarter 1981 and from the second Quarter 2005 to the fourth Quarter 2007. For the second regime, it last from the third Quarter 1981 to the first Quarter of 2005.

6 Under the Markov switching model each regime last in average for 1⁄(1 − 푝푖푖) periods for each regime 푖 = 1,2, see Hamilton (1989) for more details.

Regime 1 Regime 2

1.0 1.0

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0.0 0.0 1980 1985 1990 1995 2000 2005 2010 1980 1985 1990 1995 2000 2005 2010 Figure 2 : Smoothed probabilities for the CO2 emissions equation.

Regime 1 Regime 2

1.0 1.0

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0.0 0.0 1975 1980 1985 1990 1995 2000 2005 1975 1980 1985 1990 1995 2000 2005 Figure 3 : Smoothed probabilities for the Ecological FootPrint (FP) equation

It is important to remember that after we have estimated different specifications models before selecting the final retained specification which is characterized by Markov switching in both the intercept and the slope coefficient of only the real GDP dependent variables. The real GDP 1 2 coefficients (훽1 , 훽1 ) are positive and highly significant under both regimes and for both pollutants. The slope coefficient of on the nonlinear term of real GDP per capita ( squared real GDP per capita) is negative and significant at the 1% level significance. In more details, the results indicate that 1 per cent increase in real GDP per capita increases the 퐶푂2 emissions per capita between 1.766 per cent in regime 1 and 2.35 per cent for regime 2. For the FP dependent variable, an increase by 1 per cent of the real GDP per capita increases FP by 3.549 per cent under regime 1 and 2.803 per cent under regime 2. On the other hand, the negative coefficient of the nonlinear term of real GDP per capita is equal to -0.107 for the 퐶푂2 emissions pollutant and -0.170 for the FP pollutant. As a result, the positive and negative signs coefficients of the real GDP and squared real GDP respectively under each regime indicate that the EKC hypothesis holds for both pollutants (퐶푂2emissions and EFP).

The impact of electricity consumption on 퐶푂2 emissions and FP is different. For the electricity consumption, the effect on 퐶푂2 emissions is positive and highly significant, significant at the 1% level. In contrast, the effect on the ecological footprint (FP) is not significant at conventional level. For the 퐶푂2emissions, an increase of 1% in electricity consumption increases 퐶푂2 emissions by 0.515 per cent. This results confirm some empirical evidence in the energy use-environment nexus literature which suggest that electricity consumption will negatively affects CO2 emissions if the country uses a friendly environment energy source. However, as in Qatar the whole electricity production is provide by burning gas then it positively CO2 emissions.

The third determining variable of the 퐶푂2 emissions and FP pollutant it concerns the role played 7 by trade openness. In our empirical findings, we found that it determine only the 퐶푂2emissions . The results show that openness trade affects positively and significantly the 퐶푂2 emissions. We found that 1 per cent increase in the rate of openness trade is associated with an increase of 0.189 per cent of 퐶푂2 emissions. This empirical finding suggest that trade openness destroy environmental quality.

The fourth determining variable used to explain the 퐶푂2 emissions and EFP evolution is urbanization. The estimated slope coefficient of urbanization is positive and significant at the 1% level significance for both pollutants. The empirical findings show that an increase of 1 per cent of the urbanization level increases the 퐶푂2 emissions and EFP pollutants by 3.154 per cent and 0.818 per cent respectively. This result confirm some empirical studies in the empirical literature which found a positive relationship between environment degradation and urbanization level, see for instance Shahbaz et al. (2014) for the case of UAE. This empirical findings can be explained by the fact that an increase of urbanization lies in increases in energy demand and 퐶푂2emissions. Turning now to the Financial development variable, the empirical findings show that the coefficient on domestic credit to private sector scaled by GDP is negative and statistically significant at 1 per cent level of significance. This results support the idea following that financial development decreases 퐶푂2 emissions and EFP. Our result indicate that policymakers and government should enhanced more financial sector in order to better benefit from its positive impact on the environmental quality. The result indicates that an increase by 1 per cent of domestic credit to private sector scaled by GDP will decreases 퐶푂2 and FP by 0.208 per cent and 0.229 per cent respectively. 4.3.4. Short run analysis The estimation results of the short run relationship is reported in Table 7 below. We discuss the results on the basis of each pollutants, the 퐶푂2 emissions and the ecological Foot Print. This Table shows that the first lag of 퐶푂2 emissions, the real income per capita, the electricity consumption, the financial development, the trade openness and the first lag of the urbanization variables significantly affect, in the short run, the 퐶푂2 emissions. We found that at the short run and as expected, the lagged 퐶푂2 emissions have a positive and significant impact on the actual 퐶푂2 emissions. In particular, the results show that the real income per capita is positively and significantly linked to the 퐶푂2 emissions in the short run. Moreover, we found that electricity

7 Trade openness variable is not a key determinant of the Foot Print environment degradation proxy.

consumption, financial development and urbanization affect positively 퐶푂2 emissions in the short run. In contrast, we found that openness trade affects negatively 퐶푂2 emissions. For the case of ecological Foot Print, we found that the first lag of the ecological Foot Print is statistically significant and positively affects the level of current FP. Our empirical findings show also that the real GDP per capita, financial development, and the first lag of squared real GDP per capita impact positively the current level of FP. In contrast, we found that the current level of squared real GDP per capita and the first lag of real GDP per capita affect negatively the current level of FP.

The coefficient on the ECMt-1 for the two short run relationships is negative and significantly different from zero. This coefficient which measures the rapid adjustment towards the long run equilibrium varies between -0.106 and -0.113. The negative and significant sign of this coefficient indicates the existence of adjustment towards the long run equilibrium. Moreover, the results indicate that variation in energy emissions are corrected between 10.6% and 11.30% each year. Table 7 : Estimation of the short run relationship 푳푭푷 푳푪푶ퟐ Variables 풕 풕 Coef. t-stat Coef. t-stat 푰풏풕풆풓풄풆풑풕 0.003 0.738 -0.030 -4.118 푬푪푴풕−ퟏ -0.106 -3.592 -0.113 -3.698 휟푳푭푷풕−ퟏ 0.644 9.797 - - 휟푳푪푶ퟐ풕−ퟏ 0.628 9.796 휟푳푹푮푫푷풕 5.562 3.158 0.313 4.299 ퟐ 휟푳푹푮푫푷풕 -0.303 -3.095 -0.030 -1.191 휟푳푭푫풕 0.106 1.795 0.205 3.921 휟푳푬푳푬푪풕 - - 0.448 3.339 휟푳푶푷푬푵풕 - - -0.383 -2.928 휟푳푹푮푫푷풕−ퟏ -5.327 -2.993 - - ퟐ 휟푳푹푮푫푷풕−ퟏ 0.297 2.975 - - 휟푳푼푹푩풕−ퟏ - - 0.249 3.406 B-G LM(2) test 0.2208 0.273 RESET test (2) 0.6658 0.189 ARCH(3) 0.1866 0.112

Empirical results concerning residual diagnostic analysis reported in the bottom of Table 7 above show that the two specifications of 퐶푂2 emissions and Foot Print are appropriate and adequate to describe the evolution of the two pollutants. For instance, the Breusch-Godfrey Serial Correlation LM test (B-G LM(2) test), the RESET test for higher order omitted variables and the ARCH test of heteroscedasticity show a p-values that are higher than the 5% level significance. This means that for all these three tests the null hypothesis cannot be rejected which means, absence of serial autocorrelation, absence of higher order omitted variables and absence of heteroscedasticity.

4.3.5. CUSUM and RS-CUSUM tests After selecting the more appropriate specifications for both pollutants, the following step consists in testing for residuals stability under each specification using the CUSUM and squared residuals

CUSUM tests. The results show that the solid blue lines of the cumulative sum of residuals and squared residuals do not deviate from the critical bounds at the 5% level of significance (see Figures 4 and 5). This results indicate the mean and variance stability of the residuals of each specification which confirm that the selected specifications are more appropriate to describe the evolution of the 퐶푂2 emissions and EFP pollutants. This results allow us also to investigate the

40 1.2

30 1.0

20 0.8 10 0.6 0 0.4 -10 0.2 -20

-30 0.0

-40 -0.2 76 78 80 82 84 86 88 90 92 94 96 98 00 02 04 06 76 78 80 82 84 86 88 90 92 94 96 98 00 02 04 06

CUSUM 5% Significance CUSUM of Squares 5% Significance Figure 4 : CUSUM and squared CUSUM test for the Footprint equation

40 1.2

30 1.0

20 0.8 10 0.6 0 0.4 -10

-20 0.2

-30 0.0

-40 -0.2 82 84 86 88 90 92 94 96 98 00 02 04 06 08 10 82 84 86 88 90 92 94 96 98 00 02 04 06 08 10 CUSUM 5% Significance CUSUM of Squares 5% Significance Figure 5 : CUSUM and squared CUSUM test for the 퐶푂2 emissions equation

4.4. Granger causality analysis

The most interesting and important step after selecting the relationship between environment degradation and its key determinants is the determination of the causality direction between each couple of variables. The econometric theory suggests that at least one causal relationship should exist if there exist a long run relationship between variables. In this paper, we use the standard VECM Granger causality approach to investigate both the short and long run causal relationships. The empirical results of the short and long run VECM Granger causality are reported in Table 8 and 9 for the 퐶푂2 emissions and FP pollutants, respectively..

Our empirical findings concerning the causal relationship differ significantly between the two pollutants employed in this paper. using the 퐶푂2 emissions pollutant, the empirical results for the long and short run causality is reported in Table 8. Empirical results show that bidirectional causality is running from 퐶푂2 emissions to the real GDP per capita, financial development and openness trade. We found also bidirectional causality running from openness trade to real GDP per capita and financial development. In the short run, our empirical findings show that urbanization causes the 퐶푂2 emissions, we found also that financial development and urbanization causes real GDP.

Table 8 : The short and long run VECM Granger causality analysis Type of Granger causality Dependen Short-run Long-run t 훥퐿퐶푂2 훥퐿푅퐺퐷푃 훥퐿퐸퐿퐸퐶 훥퐿퐹퐷 훥퐿푈푅퐵 훥퐿푂푃퐸푁 E C M 푡 푡 푡 푡 푡 푡 t 1 [t- Variable F-statitics [p-values] statistics] 6.100* 2.626** 7.564* 0.654 5.929* -0.100* 훥퐿퐶푂2 - [0.003 * 푡 [0.000] [0.522] [0.000] [-2.881] ] [0.077] 2.682** 16.63* 3.049** 0.679 8.007* -0.064** 훥퐿푅퐺퐷푃 * - [0.000 * 푡 [0.509] [0.000] [-2.276] [0.073] ] [0.052] 0.178 0.695 1.699 0.028 0.525 0.006 훥퐿퐸퐿퐸퐶 - [0.837 푡 [0.502] [0.129] [0.973] [0.593] [0.280] ] 4.524** 12.39 0.333 1.181 4.680** -0.052 훥퐿퐹퐷 - 푡 [0.013] [0.000] [0.717] [0.311] [0.011] [-1.106] 0.043 1.731 1.206 0.061 0.008 0.001 훥퐿푈푅퐵 [0.958 - 푡 [0.182] [0.312] [0.941] [0.992] [0.332] ] 5.916* 3.945** 4.249* 0.041 0.397 0.025 훥퐿푂푃퐸푁 [0.004 - 푡 [0.023] [0.003] [0.959] [0.673] [0.213] ] *, **, *** indicate significance at the 1%, 5% and 10% level respectively.

Using the ecological Foot Print variable, empirical results reported in Table 9 show that, in the long-run, only the ECT coefficient associated with the ecological foot print variable is significantly different from zero and have a negative sign. This means that, for only this variable, there is an evidence of an error correction mechanism which derives this variable back to their long-run relationship. In the short-run, we found strong evidence for bidirectional causality between the ecological foot print and real GDP per capita, electricity consumption and financial development variables. Bidirectional causality is also running between the real GDP per capita, and electricity consumption and urbanization. On the other hand, unidirectional causality relationship is found between the urbanization and the foot print variable and between the financial development to the real GDP per capita variable.

Table 9 : The short and long run VECM Granger causality analysis Type of Granger causality Dependent Short-run Long-run

훥퐿퐹푃푡 훥퐿푅퐺퐷푃푡 훥퐿퐸퐿퐸퐶푡 훥퐿퐹퐷푡 훥퐿푈푅퐵푡 E C M Variable t 1 F-statitics [p-values] [t-statistics] 훥퐿퐹푃 3.162** 6.255** 5.419* 2.515 -0.0563** 푡 - [0.047] [0.015] [0.006] [0.116] [-2.165] 훥퐿푅퐺퐷푃 2.378*** 5.830* 1.666 4.942* 0.005 푡 - [0.098] [0.004] [0.194] [0.009] [0.707] 훥퐿퐸퐿퐸퐶 5.104* 3.852** 2.641*** 3.633** -0.019 푡 - [0.262] [0.025] [0.100] [0.030] [-1.322] 훥퐿퐹퐷 5.684* 28.93* 0.333 1.092 -0.026 푡 - [0.005] [0.000] [0.718] [0.339] [-0.638] 훥퐿푈푅퐵 2.733*** 3.278** 1.568 1.253 0.0003 푡 - [0.070] [0.015] [0.214] [0.213] [0.231] *, **, *** indicate significance at the 1%, 5% and 10% level respectively.

5. Concluding remarks and Policy implications 5.1. Summary

The papers examines possible long run relationship between 퐶푂2 emissions and ecological foot print pollutants considered as proxies of environment degradation and economic growth, electricity consumption, financial development, urbanization and openness trade. In contrast to the previous studies in this area, this paper is novelty at least in three elements. We investigate the case of the Qatar country that is not yet examined as a single country. We use recent tests of unit root and tests for cointegration with structural breaks rarely examined in the empirical studies such as the Silvestre et al. (2009), and Gregory and Hansen (1996) and Hatemi-J (2008) respectively. Moreover, when estimating the long run relationship we use the Markov switching equilibrium correction model with shift in the intercept and slope coefficient of the real GDP per capita variable. We investigate the Granger causality direction among variables within the context of VECM framework. The results of the unit root tests with structural breaks show that all series in level are non- stationary and integrated with first order, e.g I(1), with the presence of structural breaks. This result can be viewed as an indicator of possible existence of breaks in the cointegration vector if there exist. After verifying that all variables are I(1) processes, the second step is to test for cointegration with structural breaks using the Gregory and Hansen (1996) for one unknown break in the cointegration vector and Hatemi-J (2008) test for cointegration with two unknown breaks. Empirical findings show that both tests of cointegration, with one unknown or two unknown breaks, cannot reject the alternative hypothesis of cointegration with shifts in both the intercept and slope coefficients indicating the existence of long run relationship among investigated variables. The results of estimation of the long run relationship with Markov shifts supports the

existence of a cointegration vector with two structural breaks as indicated in the probabilities smoothing evolution. Moreover, we found that only the intercept and the slope coefficient of the real GDP per capita variable shifts between regimes. More precisely, our empirical results show that the inverted U-shaped hypothesis between the environment degradation proxy and economic growth cannot be rejected for the two pollutants examined, the CO2 emissions and ecological footprint. Moreover, we found that the impact of electricity consumption is positive for the CO2 emissions environment degradation proxy and negative for the ecological foot print proxy. We found also that urbanization and financial development have positive and statistically significant impact on both pollutants. Concerning the openness trade, we found that this variable determines positively the 퐶푂2 emissions.

Concerning the Granger causality results, we found bidirectional causality running from 퐶푂2 emissions to the real GDP per capita, financial development and openness trade. We found also bidirectional causality running from openness trade to real GDP per capita and financial development. In the short run, our empirical findings show that urbanization causes the 퐶푂2 emissions, we found also that financial development and urbanization causes real GDP. For the ecological footprint proxy, we found strong evidence for bidirectional causality between the ecological foot print and real GDP per capita, electricity consumption and financial development variables. Bidirectional causality is also running between the real GDP per capita, and electricity consumption and urbanization. On the other hand, unidirectional causality relationship is found between the urbanization and the foot print variable and between the financial development to the real GDP per capita variable. To summarize, this paper validate the EKC hypothesis for the Qatar state behavior differ across regimes. Following this result, the location of the turning point depends on whether the Qatar economy is in the first or second regime. This indicates also that the country economic structure may play an important role in determining the speed and the time necessary to reach the turning point.

5.2. Policy implications The design of the appropriate policies required to improve the environmental quality is of great importance for the Qatari government and Qatari policy makers. The effectiveness of the proposed environmental strategy depends mainly on whether policy makers have completely understand the variables that determine the environmental degradation and the causal relationships among these determinants. The results of validating the EKC hypothesis and the Granger causality testing have many policy implications. First, the results that validate the EKC hypothesis suggests that the rise in economic growth in recent years has been associated with lower environmental degradation. However, the challenge persists as Qatar remains one of the countries with highest CO2 emissions in the world. Therefore, more efforts must be undertaken by policymakers to minimize the level of emissions in the country. Second, it is important to recognize that policymakers are restricted from imposing conservation policies due to the confirmed results of the bidirectional relationship between economic growth and electricity consumption. This is because policies are likely to lower not only CO2 emissions but also the level of economic growth. Accordingly, one measure that can be undertaken to balance environmental quality and economic growth is to generate electricity

from renewable energy sources. According to the World Bank indicators, 100% of the electricity production in Qatar in 2012 came from natural gas8. If the country were to diversify the production of electricity by using renewable energy sources such as: solar power, wind power or wave power, the emissions from electricity production will be reduced without negatively affecting economic growth. Additionally, policies should be implemented to ensure saving and efficient use of energy. This can be done by adopting energy efficiency technologies, and influencing the behavior of consumers mainly by raising the prices of energy sources which is likely to lead to better exploitation of available energy. Moreover, reducing the energy subsidy will lead to better utilization of energy sources. 5.3. Limits and extensions

One of the major limits of our paper consists on allowing for a maximum of two breaks in the cointegration vector of the long run relationships. As noted before, this choice is mainly explained by the period of study that does not allows to consider more than 2 breaks. As possible extension of this paper and in the case of collecting a data that can cover a more long period is to consider the recent tests for cointegration with multiple breaks proposed by Arai and Kurozumi, 2007; and Kejriwal and Perron, 2010. Another possible extension of our study is to include some others socio-demographics and institutional variables as determinants of environment degradation proxies such as fertility rate, life expectancy, political index as in Charfeddine and Khediri (2015).

References

Abler, D. G.; Rodriguez, A. G., and Shortle, J. (1999). Trade liberalization and the environment in Costa Rica. Environment and Development Economics; vol. 4, pp. 357–73.

Al-mulali, U.; Che, Sab, C.N., (2012). The impact of energy consumption and CO2 emission on the economic growth and financial development in the Sub Saharan African countries. Energy; vol. 39, pp. 180–186.

Antweiler W.; Copeland, B.R., Taylor MS. (2001). Is free trade good for the environment? American Economic Review, vol. 91, pp. 877–908.

Arai, Y., Kurozumi, E., (2007). Testing for the null hypothesis of cointegration with a structural break. Econometric Reviews; vol. 26, pp. 705–739.

Beckerman, W. (1992). Economic Growth and the Environment: Whose Growth? Whose Environment? . World Development, vol. 20(4): pp. 481-496.

8 http://wdi.worldbank.org/table/3.7

Carrion-i-Silvestre, J.L., Kim, D., and Perron, P., (2009). GLS-Based Unit Root Tests with Multiple Structural Breaks Both under the Null and the Alternative Hypotheses. Econometric Theory, vol. 25, pp. 1754-1792.

Charfeddine, L.; and Khediri, K. B. (2016). Financial development and environmental quality in UAE: Cointegration with structural breaks, vol. 55, pp. 1322-1335.

Chikaraishi, M., Fujiwara, A., Shinji, Kaneko S., Poumanyvong, P., Komatsu, S.; Kalugin, A., (2015). The moderating effects of urbanization on carbon dioxide emissions: A latent class modeling approach. Technological Forecasting and Social Change; 90 Part A:302-17

Cole, M.A., Neumayer, E. (2004). Examining the impact of demographic factors on air pollution. Population and Environment; vol.26, pp. 5–21.

Matthew A. Cole; and Eric Neumayer (2004). Examining the impact of demographic factors on air pollution. Population and Environment, vol. 26 (1), pp 5-21.

Davies, R.B. (1977): "Hypothesis testing when a nuisance parameter in present only under the alternative," Biometrika, vol. 64, pp. 247-254.

Davies, R.B. (1987): "Hypothesis testing when a nuisance parameter in present only under the alternative," Biometrika, vol. 74, pp. 33-43.

Dickey, D.A., Fuller, W.A., (1979). Distribution of the estimators for autoregressive time series with a unit root. Journal of American Statistics Association, vol. 74, pp. 427–431.

Dodman, D. (2009). Blaming cities for climate change? An analysis of urban greenhouse gas emissions inventories. Environment and Urbanization, vol. 21 no. 1 185-201.

Ehrhardt-Martinez (1998) Social Determinants of Deforestation in Developing Countries: A Cross-National Study. Social Forces (1998) 77 (2): 567-586.

Elliot, G., Rothenberg, T.J., Stock, J.H., 1996. Efficient tests for an autoregressive unitroot. Econometrica 64, 813–836.

Engle RF, Granger CWJ. (1987). Cointegration and error correction: representation, estimation, and testing. Econometrica; vol. 55; pp. 251–76.

Esteve V, Tamarit C (2012). Is there an environmental Kuznets curve for Spain? Fresh evidence from old data. Economic Modelling; vol. 29, pp. 2696-703.

Esteve, V., Tamarit, C.(2012). Threshold cointegration and nonlinear adjustment between CO2 and income: the environmental Kuznets curve in Spain, 1857–2007. Energy Economics. Vol. 34, (6), pp. 2148–2156.

Fan Y, Lui L.-C, Wu G, Wie Y-M. (2006). Analyzing impact factors of CO2 emissions using the STIRPAT model. Environmental Impact Assessment Review; vol. 26, pp. 377–95.

Frankel JA, Romer D. (1999). Does trade cause growth? American Economics Review, vol. 89 (3), pp. 379–99.

Gregory, A. W. and Hansen, B. E., (1996). Residual-based tests for cointegration in models with regime shifts, Journal of Econometrics, vol. 70, pp. 99-126.

Grossman G.M., Krueger A.B., 1995. Economic growth and the environment. Quarterly Journal of Economics 110. 353-377.

Hamilton, James D. (1989). “A New Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle,” Econometrica, 57, 357-84.

Hamilton, James D. (1990). “Analysis of Time Series Subject to Regime Changes,” Journal of Econometrics, 45, 39-70.

Hansen, B.E.; Seo, B. 2002. Testing for two-regime threshold cointegration in vector error- correction models". Journal of Econometrics 110.2, pp. 293-318.

Hatemi-J A. (2008). Tests for cointegration with two unknown regime shifts with an application to financial market integration. Empirical Economics; vol. 35, pp. 497-505.

Holtz-Eakin, D., Selden, T.M., (1995). Stoking the fires? CO2 emissions and economic growth. Journal of Public Economics 57, 85–101.

Horvath, R.J. (1997). “Energy consumption and the environmental Kuznets curve debate”. Mimeo, Department of Geography, University of Sydney, Australia.

Hossain, Md. S., (2011). Panel estimation for CO2 emissions, energy consumption, economic growth, trade openness and urbanization of newly industrialized countries, Energy Policy, vol. 39, 6991–6999.

Jayanthakumaran K, Liu Y. (2012). Openness and the environmental Kuznets curve: evidence from China. Economic Modelling; vol. 29, pp. 566–76.

Kasman A, Duman YS. 퐶푂2 emissions, economic growth, energy consumption, trade and urbanization in new EU member and candidate countries: A panel data analysis. Economic Modelling 2015;44:97–103.

Kejriwal, M., 2008. Cointegration with structural breaks: an application to the Feldstein–Horioka puzzle. Studies in Nonlinear Dynamics & Econometrics 12 (1), 1–37.

Kejriwal, M., Perron, P., 2010. Testing for multiple structural changes in cointegrated regression models. Journal of Business and Economic Statistics, vol. 28, pp. 503–522.

Kwiatkowski, D., Phillips, P., Schmidt, P., Shin, Y., (1992). Testing the null hypothesis of stationary against the alternative of a unit root: how sure are we that economic time series have a unit root? Journal of. Econometrics. Vol. 54, pp. 159–178.

Lee J, Strazicich MC. (2003). Minimum Lagrange multiplier unit root test with two structural breaks. The Review of Economics and Statistics; vol. 85, pp.1082-9.

Liddle B, Lung S. (2010). Age-structure, urbanization, and climate change in developing countries : revisiting STIRPAT for disaggregated population and consumption related environmental impacts. Population Environment, Vol. 31, pp. 317–43.

World Wide Fund For Nature. Living Planet Report (2014). Species and spaces, people and places.

Ng S, Perron P. Unit root tests in ARMA models with data dependent methods for the selection of the truncation lag. Journal of the American Statistical Association 1995;90:268–81.

Ozturk I, Acaravci A. (2013). The long-run and causal analysis of energy, growth, openness and financial development on carbon emissions in Turkey. Energy Economics, Vol. 36, pp. 262-267.

Perron P. (1989). The great crash, the oil price shock and the unit root hypothesis. Econometrica; vol. 57, pp. 1361-401.

Phillips, P.C.B., Perron, P., (1988). Testing for a unit root in time series regression.Biometrika 75, 335–346.

Poumanyvong, P., Kaneko, S., (2010). Does urbanization lead to less energy use and lower CO2 emissions? A cross-country analysis. Ecological Economics 70 (2), 434–444.

Sadorsky P. (2014). The effect of urbanization on CO2 emissions in emerging economies. Energy Economics; vol. 41, pp. 147–53.

Shahbaz M, Lean HH, Shabbir MS. (2012). Environmental Kuznets curve hypothesis in Pakistan: cointegration and Granger causality. Renewable and Sustainable Energy Reviews; vol. 16, pp. 2947–53.

Shahbaz M, Hye QMA, Tiwari AK, Leitão NC. (2013). Economic growth, energy consumption, financial development, international trade and CO2 emissions in Indonesia. Renewable and Sustainable Energy Reviews; vol. 25, pp. 109–21.

Shahbaz M., Lean H.H., Shabbir M.S., (2012), Environmental Kuznets curve hypothesis in Pakistan: cointegration and Granger causality. Renewable and Sustainable Energy Reviews, vol. 16, pp. 2947–53.

Shahbaz M., Solarin S.A., Mahmood H., Arouri M., (2013a). Does financial development reduce CO2 emissions in Malaysian economy? A time series analysis. Economic Modelling, vol. 35. pp. 145–152.

Shahbaz M., Tiwari A.K., Nasir M,. 2013c. The effects of financial development. economic growth. coal consumption and trade openness on CO2 emissions in South Africa. Energy Policy 61. 1452–1459.

Shahbaz M., Lean H, (2012). Does financial development increase energy consumption? The role of industrialization and urbanization in Tunisia, Energy Policy 40, 473–479

Sharma SS. (2011). Determinants of carbon dioxide emissions: empirical evidence from 69 countries. Applied Energy; vol. 88, pp. 376–382.

Shafik, N., Bandyopadhyay, S., (1992). Economic growth and environmental quality: time series and cross-country evidence. Background Paper for the World Development Report 1992. World Bank, Washington D.C.

Stern, D.I., Common, M.S., (2001). Is there an environmental Kuznets curve for sulfur? Journal of Environmental Economics and Management 41, 162–178.

Stern, D.I., Common, M.S., Barbier, E.B., (1996). Economic growth and environmental degradation: a critique of the environmental Kuznets curve. World Development 24, 1151–1160.

Stock, J.H., (1999). A class of tests for integration and cointegration. In: Engle, R.F., White, H. (Eds.), Cointegration, Causality and Forecasting. A Festschrift in Honour of Clive W.J. Granger. Oxford University Press, pp. 37–167.

Suri V, Chapman D. (1998). Economic growth, trade and the environment: implications for the environmental Kuznets curve. Ecological Economics; vol. 25:195–208.

World Bank, (2011). World Development Indicators (WDI). Accessed from: http://www. worldbank.org/data/onlinedatabases/onlinedatabases.html.

World Resource Institute, (2007). October 2006 Monthly Update: Fossil Fuel Consumption and its Implications: http://earthtrends.wri.org/updates/node/100.

York R, Rosa EA, Dietz T. (2003). STIRPAT, IPAT and ImPACT: analytic tools for unpacking the driving forces of environmental impacts. Ecological Economics; vol. 46, pp. 351–65.

Zivot E, Andrews D. (1992). Further evidence of great crash, the oil price shock and unit root hypothesis. Journal of Business and Economic Statistics, vol. 10, pp. 251–70.