Economic Growth , CO 2 Emissions , and Financial Development in Jordan : Equilibrium and Dynamic Causality Analysis

This article contributes to the existing literature by investigating equilibrium and dynamic causality relationships among economic development, CO2 emissions, energy consumption, financial development, foreign direct investments inflows, and gross fixed capital formation in the case of kingdom of Jordan over the 1976-2010 period. The ARDL approach has been employed to detect co-integration between the series. The VECM Granger causality is also applied to evaluate causal relationships among the variables. The findings suggest the existence of co-integration between economic growth and its determinants. In addition, CO2 emissions, foreign direct investments inflows, and gross fixed capital formation have positive and significant impact on economic growth in long-run. Interestingly, the results of this paper indicate that environmental Kuznets curve hypothesis does exist between economic development and CO2 emission in long-run and short-run.


Introduction
Jordan is one of the smallest economies in Middle East, with a total (GDP) gross domestic product of JD10.5 billion in 2012 and a population of 6.32 million, of which 13.3% live below the poverty line (CD-ROM, 2014).Unlike other neighboring Arab countries, it is a non-oil-producing country with limited natural resources and minerals.Jordan is rapidly growing, both as a result of its population demographic and due to an influx of refugees over the past decades, with an estimated 1.5 million non-Jordanian residents.It is heavily dependent on imports of energy to meet the growing of (EC) energy consumption due to the lack of energy resources.The EC is expected to double from 7.58 (mtoe) million tons of oil equivalents in 2007 to 15.08 mtoe by 2020 (The Jordanian Energy Sector Report, 2012).
The prices of energy have soared recently, which encouraged the Jordanian government to update its energy master plan.The energy sector master plan transforms the existing energy mix from one heavily reliant on oil, and natural resources to a more balanced mix with a higher proportion of energy supplied by nuclear power, oil shale, and renewable sources.In addition, the Jordanian government is currently supporting various policies, initiatives, and programmes aimed at achieving a green economy.These are: the complete removal of subsidies for oil in 2008; the adoption of the renewable energy law and fiscal incentive package on renewable energy and energy efficiency equipment in 2010; and the establishment of the Eco-cities forum, the Eco-financing seminar and the Zarqa river rehabilitation project (Green Economy Report, 2011).

Model Structure and Data Used
The cure objective of this paper is to investigate equilibrium and causality relationships between GDP and its determinants including CO 2 emissions, EC, FD, FDI, and GCF using annual time-series data for the 1976-2010 period.The empirical model is structured as in Eq. (1). LGDP where, L denotes the natural logarithmic form to remove non linearity in parameters; t represents the discrete time period; β 0 is the intercept term;

Econometrics Methodology
The present paper employs four steps of econometrics methodology to achieve research objectives.Initially, the descriptive statistics tests have been utilized to check if the (ε t )s error terms are ((ε t )s ~ N(0, σ 2 )) normally distributed with zero mean and constant variance.Ng-Perron (2001) unit root test has been used in the second step to determine the integration levels of variables.Also, the current paper employs the ARDL approach to test the (H 0 ) null hypothesis of no co-integration and estimate equilibrium relationships.This approach is suitable with small sample size and can be employed if the variables are having mixed order of integration (Bekhet & Al-Smadi, 2015;Bekhet & Mugableh, 2013;Mugableh, 2015;Narayan, 2005;Pesaran, Shin, & Smith, 2001).If the calculated F-statistics value is greater than the (I (1)) upper critical F-statistics value, then the H 0 of no co-integration would be rejected.In contrast, if the calculated F-statistics value is lesser than the (I (0)) lower critical F-statistics value, then the H 0 of no co-integration would be accepted.The ARDL approach is modelled as in Eq. (2).
where, ∆ is the first difference operator; α 0 denotes the intercept term; α it (i= 1,….6) represent long-run coefficient for testing the H 0 of no co-integration.If α it ≠ 0, then the H 0 of no co-integration would be rejected, implying that the variables are shared long-run relationships among each other.α is (i= 7,….12) signify short-run coefficients; h denotes the lag length that is obtained using the Akaike information criterion; and ε t is the error term.If the co-integration exists between variables, then the causality in long-run and short-run should be evaluated.The VECM Granger causality has been developed as in Eq. ( 3) to determine the directions of causality.
here, 1-L is the backshift operator; α it (i = 1, …, 6) denote the intercept terms; δ ijs (i, j = 1, …, 6) signify the coefficients to test the H 0 of no Granger causality directions in short-run; λ it (i= 1,….6) represent the coefficients of (Ec t-1 )s error correction terms.These coefficients are employed to test the H 0 of no Granger bidirectional causality in long-run.Engle and Granger (1987) argued that the differenced vector autoregressive model is not sufficient to examine the causality directions in short-run, especially if co-integration exists.Therefore, the inclusion of (Ec t-1 )s is necessary to evaluate the bidirectional causality in long-run.

Descriptive Statistics Tests
Table 1 shows the results of descriptive statistics tests.The H 0 of non-normality has been rejected implying that the (ε t )s ~ N(0, σ 2 ).That is, the p-values of Jarque-Berra are greater than 10%.In addition, the correlation matrix results show that the variables are linearly correlated between each other.

Unit Root and Co-integration Tests
The Augmented Dickey-Fuller (A.D-F) and Phillips-Perron (P-P) unit root tests have not been employed in this paper due to their low prediction power.However, the Ng-Perron unit root test is suitable for small sample size and provides efficient results regarding integration levels of variables.The results of Ng-Perron unit root test are reported in Appendix A. The outcomes show that all variables are stationary at the first difference, I (1), with intercept and time trend (i.e., ∆LGDP t , ∆LCO 2t , ∆LEC t , ∆LFD t , ∆LFDI t , ∆LGCF t ).This lead us to apply the ARDL approach to examine co-integration and equilibrium relationships.In addition, the VECM Granger causality would be employed to evaluate causality directions in long-run and short-run.(2) * denotes the significance at 1% level.
Source: The computed F-statistic value was obtained from Micro-Fit software package (version 5.1).
The results reported in Table 2 reveal that the computed F-statistic value (6.42) is more than the tabulated F-statistic value (5.33) at upper integration level, I(1), and 1% significance level.Thus, the long-run relationship is found between LGDP t and its determinants.These results are in line with the findings obtained for Bahrain (Hamdi et al., 2014) and Saudi Arabia (Alshehry & Belloumi, 2015).

Equilibrium and Dynamic Causality Analysis
After confirming the co-integration among variables, the equilibrium relationships have been tested by applying the ARDL approach.In addition, Bekhet and Al-Smadi (2014), Bekhet and Matar (2013), Bekhet andMugableh (2012), andGranger (1969) mentioned that if the variables are co-integrated then the causality should be found at least from one direction in both long-run and short-run.Table 3 shows long-run relationships analysis results when the ∆LGDP t is dependent variable.The LCO 2t-1 adds in ∆LGDP t and it is statistically significant at 10% level.A 1% increase in the LCO 2t-1 is linked with 2.95% ∆LGDP t in long-run.Keeping other things are constant, a 1% increase in LFDI t-1 adds in ∆LGDP t by 0.31%.LGCF t-1 is significant at the 5% level and positively associated with the ∆LGDP t .LGCF t-1 0.99 0.44 2.28[0.033]5% Source: Micro-Fit software package (version 5.1).
The short-run relationships results analysis is illustrated in Table 4.The ∆LGDP t is negatively influenced by ∆LCO 2t-1 , ∆LFD t-1 , and ∆LFDI t-1 .The impact of ∆LFDI t and ∆LGCF t on ∆LGDP t are positive and significant at the 1% and 5% levels, respectively.The diagnostic tests are detailed in Table 4 in lower segment.The results show that the short-run model seems to pass serial correlation, non-normality, and Heteroscedasticity.
Table 5 illustrates the results of the VECM Granger causality test.There is a unidirectional Granger causality in short-run from ∆LEC t to ∆LGDP t .The unidirectional Granger causality from energy consumption to economic growth is similar to the finding obtained for South Africa (Menyah & Wolde-Rufael, 2010).Therefore, the non-neutrality hypothesis is existed in Jordan because economic development is highly dependent on energy consumption.Turning to the long-run causality results, Table 5 indicates long-run bidirectional Granger causality between ∆LGDP t and its determinants.These results are in line with the findings obtained for India (Boutabba, 2014).The long-run bidirectional Granger causality between financial development and economic development confirms the existence of supply and leading hypotheses in Jordan.Financial development is an important driver of economic growth through the allocation of resources, capital accumulation, and technological innovation.Source: E-views software package (version.8.1).

Final Remarks
The current article analyses dynamic relationships between economic growth and it determinants in the kingdom of Jordan over the 1976-2010 period.The Ng-Perron unit root test is applied to determine the integration levels of variables.The results show that the variables are stationary at I(1), which confirm the use of ARDL and VEC models.The findings suggest the existence of co-integration between economic growth and its determinants.The carbon dioxide emissions, foreign direct investments inflows, and gross fixed capital formation add in economic growth in long-run.The findings show that carbon dioxide emissions are positively associated with economic growth in long-run, whereas they negatively linked to economic growth in short-run.Hence, the EKC (environmental Kuznets curve) hypothesis does exist between economic development and CO 2 emission in both long-run and short-run.This hypothesis argues the relationship between economic development and CO 2 emissions.Specifically, the emissions of greenhouse gases levels increase as a country develops but decrease when a certain level of economic development is achieved (i.e., an inverted-U shape curve).These results are in line with the findings obtained for Malaysia (Azlina et al., 2014;Lau et al., 2014).The causality analysis reveals bidirectional between economic development and its determinants in long-run as the coefficients of (Ec t-1 )s are significant with negative signs.
The bidirectional Granger causality between energy consumption and economic growth suggests the implementation of energy exploration policies to sustain economic development in long-run.In addition, the bidirectional Granger causality between carbon dioxide emissions and economic growth recommends the Jordanian policy makers to update energy master plans to achieve the green economy.The increases of foreign direct investment flows into Jordan would firstly create employment and secondly increase energy consumption.Thus, economic growth would be increased.In this manner, the Jordanian policy makers ought to focus on establishing industrial projects and improving the total factor productivity strategy.However, financial development does lead economic growth in both long-run and short-run.This implies that loans used by both consumers and investors add to economic growth.Consequently, the Jordanian policy makers are proposed to promote financial development by increasing the levels of performing loans.An interesting future research could investigate the determinants of economic growth by using a heterogeneous panel cross sectional data for a pooling of Arab countries in the Middle East.

Table 2 .
Co-integration test results

Table 3 .
Long-run relationships analysis results

Table 4 .
Short-run relationships analysis results