Renewable and Nonrenewable Energy Consumption, Government Expenditure, Institution Quality, Financial Development, Trade Openness, and Sustainable Development in Latin America and Caribbean Emerging Market and Developing Economies

This study investigates the role of non-renewable and renewable energy consumption in the sustainable development in 16 Latin America and Caribbean Emerging Market and Developing Economies (EMDEs) incorporating capital, government expenditure, institution quality, financial development, and trade openness by a multivariate framework using annual data from 1990 to 2014. We apply second-generation techniques for heterogeneous panel data as the presence of cross-sectional dependence and slope heterogeneity is detected. Accordingly, CADF and CIPS unit root tests show that all variables are integrated at order 1. Westerlund cointegration test acknowledges the long-run relationship among the variables. The long-run estimation is conducted by the Augmented Mean Group (AMG), MG and Common Correlated Effects MG (CCEMG) estimators. The findings indicate that, in the long run, renewable and non-renewable energy use, along with other factors including government expenditure, gross fixed capital formation, trade openness and financial development, positively affects the economic growth in the selected countries. The empirical results imply that the EMDEs in Latin America and the Caribbean should appropriately implement fiscal policies for macroeconomic stabilization in combination with finance and international trade policies as well as effective energy strategies to attain their sustainable development objectives.


INTRODUCTION
Energy consumption is deemed a main factor boosting economic growth because it directly contributes to the input of the manufacturing industry for producing goods and services and comprises a large proportion of household consumption (Stern, 2000). Despite this beneficial effect, energy consumption is one of the major culprits of pollution (Phong et al., 2018;Phong, 2019). Thus, the use of renewable energy or the one that minimally influences the environment in economic development process is an important objective (Dong et al., 2017). This indicates the more important role and effectiveness of renewable energy compared to its non-renewable counterpart in the economic growth of many countries, especially when they strive for sustainable development.
non-renewable energy consumption, along with other factors, and economic growth for emerging market and developing economies (EMDEs) in Latin America and the Caribbean.
This study contributes to the literature in two facets. First, regarding econometric technique, we consider the heterogeneity and cross-sectional dependence that possibly exist within cross countries, thus employing second-generation panel data methods that are free from the inconsistent and biased problems encountered by the first-generation techniques (Pesaran, 2004;Bilgili and Ulucak, 2018;Sharif et al., 2019). Besides, we utilize three computation methods including the Augmented Mean Group (AMG) estimator (Eberhardt and Bond, 2009;Eberhardt and Teal, 2010), the MG estimator of Pesaran and Smith (1995), and Pesaran (2006) Common Correlated Effects MG (CCEMG) estimator to ensure robustness. Second, to avoid inconsistent and biased estimation stemming from omitted variables (Lütkepohl, 1982), we expand the production function by adding renewable and nonrenewable energy consumption together with institution, government expenditure, gross fixed capital formation, financial development and trade openness as explanatory variables, which has not been done in any prior studies about the EMDEs in Latin America and the Caribbean.
The remaining contents follow this structure: Section 2 provides the review of notable studies; Section 3 explains the Model, Data, and Methodology; Section 4 shows the empirical results; Section 5 gives essential summary of this paper as well as recommendations for policy-makers grounded in the empirical findings.

LITERATURE REVIEW
The existing literature concerning the classical and modern economic growth models acknowledges the vital importance of fundamental factors such as capital, labour, financial development, trade openness and government's roles in public expenditure and institutional quality. An appropriate financial system can foster technological advancement and efficiently allocate resources to the manufacturing sector, which is deemed a crucial basis for sustainable development (Schumpeter, 1911;Demirgüç-Kunt, 2006). Besides, trade openness plays a not inconsiderable role in facilitating the economy through different ways such as achieving high efficiency of resources allocation due to export-orientation policies, attracting foreign direct investment, gaining access to advanced technology for promoting domestic production, creating financial and economic integration and improving total factor productivity (Romer, 1994;Shahbaz, 2012). Economists also contemplate if the public sector with the financial-institutional role of the government has any effect on the economic growth of a nation. According to endogenous growth theory, government expenditure can stimulate economic growth via education and medical subsidies together with social transfer and legal-framework research and design, which in turns increases human capital (Romer, 1986;Barro, 1990;López et al., 2011). Concurrently, the higher institutional quality of a nation will contribute to a better economy thanks to the reduction of transaction costs and risks (Cohen et al., 1983;Fredriksson et al., 2004;Fosu, 2014). Besides, energy is a vital input of economic growth (Kraft and Kraft, 1978;Apergis and Payne, 2009;Ozturk, 2010Ozturk, , 2017Solarin and Ozturk, 2015).
Empirical evidences with regard to the impacts of the aforementioned factors on economic growth are very different depending on the researched countries, data and econometric methods. While many studies utilize individual elements, the others combine factors in a multivariate framework. Specifically, Shahbaz (2012) investigated the influences of financial development and trade openness on the economic growth of Pakistan. The findings showed that the stable long-run economic growth was encouraged by capital formation, labour, financial development and trade openness. Shahbaz et al. (2013) examined the roles of energy consumption, capital, financial development, international trade and capital in the economic growth of China from 1971 to 2011 and indicated that all the variables had positive and significant effects. This article substantially contributed to the energy economics discipline and opened new directions for policy-makers to take advantage of the renewable energy sources to meet the increasing energy demand in the sustainable development process. Kumar and Kumar (2013) scrutinized the long-run effects of energy consumption per capita embodying gross fixed capital formation per capita on the GDP per capita of Kenya in the period  and South Africa in the period 1971-2009. The empirical results estimated by ARDL approach demonstrated that in the short run and long run energy consumption and capital ameliorated the growth of the two countries. Also, using ARDL method for timeseries analyses, Kumar et al. (2014) inspected the nexus between capital formation per capita, energy consumption per capita and GDP per capita in Albania (1980Albania ( -2012, Bulgaria (1970-2012), Hungary (1980-2012 and Romania (1980Romania ( -2012. They found that capital formation per capita stimulated the economic growth of all researched countries in both the short run and long run. Meanwhile, energy consumption per capita improved GDP per capita of all countries in the short run, but the long run effects only happened in Bulgaria and Romania. In addition, Kumar et al. (2015) employed ARDL technique to analyse the determinants of South Africa's economic growth from 1971 to 2011 and reported that energy, capital and trade openness facilitated economic growth in both the short run and long run while financial development had negative impacts. (2019) studied the influences of renewable energy use, capital and labour on the economic growth of 15 West African countries by employing panel dynamic ordinary least squares method and annual data from 1995 to 2014. They showed that while capital and labour fostered economic growth, renewable energy consumption slowed down the economies of those countries. Besides, Zafar et al. (2019) evaluated the links between non-renewable and renewable energy consumption, capital formation, trade openness and research and development expenditures and the economic growth of Asia-Pacific Economic Cooperation (APEC) countries in the period 1990-2015 by CUP-FMOLS method. The findings indicated that all the factors enhanced the economic growth of APEC countries in the long run.

Empirical Model
This study examines the role of non-renewable and renewable energy consumption and other factors in the sustainable development in 16 Latin America and Caribbean EMDEs. The empirical model of this research is based on the extended Cobb-Douglas production function employed by Shahbaz et al. (2013), Kumar and Kumar (2013) and Kumar et al. (2014), which is written as follows: In equation (1), GDP is the real gross domestic product per capita, GCF represents capital, RE is renewable energy consumption, NRE stands for non-renewable energy consumption, GC indicates general government final consumption expenditure, INS denotes institution quality, FD is financial development, and TO is trade openness. The notations i and t respectively demonstrate country and year, and ε it is the error term.

Cross-sectional dependence test
Checking for cross-sectional dependence is deemed an important issue in panel data analysis because it can produce inconsistent estimates and misleading information (Grossman and Krueger, 1995;Pesaran, 2004;Bilgili and Ulucak, 2018).
As a result, Breusch and Pagan (1980) developed Lagrange Multiplier (LM) statistics to detect cross-sectional dependence in the panel data: Nevertheless, according to Pesaran (2004), the Breusch-Pagan LM test might be inconsistent. Thus, Pesaran (2004) introduced CD test to adjust the bias in LM test as follows: Where N is the sample size, T displays the time period, ˆi j  indicates the coefficient of pair-wise correlation obtained from OLS estimation for each cross-section dimension i.

Slope homogeneity test
According to Breitung (2005), assuming slope homogeneity can cause misleading and untrustworthy estimates if the panel data is heterogeneous. Pesaran and Yamagata (2008) developed the method of Swamy (1970) to test for the slope homogeneity phenomenon, as described in equations (5), (6) and (7): ∆ and adj ∆ are the standardized dispersion and the biased-adjusted statistics. ˆ

Panel unit root test
Under the influence of cross-sectional dependence, first-generation panel unit root tests, for example Levin-Lin Chu (LLC), Im-Pesaran-Shin (IPS), augmented Dickey-Fuller (ADF) and Phillips-Perron (PP), are not valid (Pesaran, 2007). Consequently, Pesaran (2007) developed the second-generation panel unit tests including the cross-sectionally augmented Dickey-Fuller (CADF) and the cross-sectionally augmented Im-Pesaran-Shin (CIPS), which is reliable in the presence of cross-sectional dependence. The CADF statistic can be calculated as follows: , , Where y t −1 and , ∆ i t y are the cross-sectional averages of lagged levels and first differences of individual series, respectively.
The CADF statistic can be computed by averaging the CADF i as follows: Where CADF i is the t-statistics in the CADF regression defined by equation (8). Westerlund (2007) proposed the test for panel cointegration in the presence of cross-sectional dependence, which is based on the following error-correction model:

Panel cointegration test
In equation (12), ρ i is the adjustment term that determines the speed by which the system adjusts back to equilibrium.
Westerlund (2007) test is built on the least squares estimates of ρ i with the null hypothesis assuming no cointegration. Accordingly, the group mean statistics can be computed as: When G τ and G α statistics reject the null hypothesis, it can be concluded that cointegration exists in at least one cross-sectional unit of the panel.
Meanwhile, the panel statistics are retrieved from these formulas: If the null hypothesis is rejected, it can be concluded that cointegration exists in the whole panel.

Panel long-run estimates
Under the influence of cross-sectional dependence, traditional panel regression methods might be biased and inconsistent (Pesaran and Smith, 1995;Phillips and Sul, 2003;Sarafidis and Robertson, 2009;Paramati et al., 2017). Pesaran and Smith (1995) proposed MG approach that allows all slope coefficients and error variances to vary across the panel or countries. The MG approach applies OLS technique to each panel or country to get panel-specific slope coefficients and then averages the panel-specific coefficients. Nonetheless, the MG estimator does not include any information about possible common factors that may occur in the panel data.
Pesaran (2006) introduced the CCEMG estimator which is robust to cross-sectional dependence and slope homogeneity. It includes the averages of independent and dependent variables along with the unobserved common effects f t : Where y it and x it are variables; β i represents the countryspecific slope; f t stands for the unobserved common factor with heterogeneous factor; α i and ε it are the intercept and error term respectively.
Besides CCEMG, AMG estimator introduced by Eberhardt and Bond (2009) and Eberhardt and Teal (2010) is highly robust regardless of cross-sectional dependence and slope heterogeneity. AMG estimator captures the unobservable common factors f t specified in equation (17) by the common dynamic effect parameter. To describe the AMG estimator, consider this firstdifference OLS equation: ∆ denotes the first-difference operator, β i indicates the countryspecific coefficients and θ t describes the coefficients of the time dummies.
The AMG estimator is then obtained from the across-panel averaged group-specific parameters: In equation (21),  β i are the estimates of β i in equation (20).
As the performance of the AMG method in Monte Carlo simulation are unbiased and efficient for different N (number of observations) and T (time) settings (Bond and Eberhardt, 2013), this study employs the AMG method to evaluate the long-run parameters. Besides, the MG and CCEMG estimators are also used for robustness check.

RESULTS
First, we use Pesaran (2004) CD test to examine the presence of cross-sectional dependence in the panel data. The outcomes shown in Table 1 reject the null hypothesis of no cross-sectional dependence at 1% significance level. In other words, we have strong evidence for the occurrence of cross-sectional dependence.
Next, we inspect the slope homogeneity phenomenon by Pesaran and Yamagata (2008) test. All the test statistics in Table 2 are significant at 1% level, thus signifying the existence of slope heterogeneity in our panel data.
As the presence of cross-sectional dependence and slope heterogeneity is confirmed by the aforementioned tests, first-generation unit root tests are not appropriate. Rather, we employ the second-generation unit root tests including CADF and CIPS to check for the stationarity of the variables. It can be observed in Table 3 that all variables are stationary at first difference. In other words, they are integrated at order 1 and can be referred to as I(1).
Next, we investigate the long-run relationship among the variables by Westerlund (2007) cointegration test whose outcomes are displayed in Table 4, which acknowledges the cointegration among GDP, gross fixed capital formation, renewable and non-renewable energy use, government expenditure, financial development and trade openness.
After having found the occurrence of cross-sectional dependence and slope heterogeneity as well as verified the stationarity and the cointegration properties of the variables, we now estimate the long-run coefficients of the heterogeneous panel data by the AMG estimator together with the CCEMG and MG ones for robustness check. Table 5 demonstrates that gross fixed capital formation (GCF), renewable (RE) and non-renewable (NRE) energy use, government expenditure (GC), financial development (FD) and trade openness (TO) have positive and significant impacts on the economic growth of the EMDEs in Latin America and the Caribbean. Meanwhile, the effect of institutional quality is trivial. Specifically, in the long run, 1% increases of renewable and non-renewable energy use boost GDP by 0.119% and 0.099% respectively. In addition, when gross fixed capital formation per capita rises by 1%, GDP is enhanced by 0.148%. Moreover, when general government final consumption expenditure per capita goes up by 1%, GDP per capita is promoted by 0.109%. Besides, the results also validate the finance-led growth and trade-led growth hypotheses when GDP per capita grows by 0.058% and 0.076% due to 1% increases of domestic credit to private sector per capita and trade openness per capita respectively. Obviously, the estimation outcomes of the 3 estimators AMG, CCEMG and MG are similar in terms of coefficient signs and magnitude, which indicates the robustness of our empirical findings.

CONCLUDING REMARKS AND RECOMMENDATIONS
This study examines the determinants of economic growth in EMDEs in Latin America and the Caribbean. Also, we consider the role of energy consumption in the sustainable growth objective of the aforementioned countries by employing second-generation panel data techniques on the annual data between 1990 and 2014. Our estimation methods are appropriate for the presence of cross-sectional dependence and slope heterogeneity in the panel data. CADF and CIPS unit root tests indicate that all variables are stationary at first difference, thus enabling the Westerlund panel cointegration test based on error correction. We detect the long-run relationship among the variables and evaluate the long-run coefficients by the AMG estimator. We also use the Common Correlated Effects MG (CCEMG) and MG estimators for robustness check.
The empirical findings show that renewable and non-renewable energy use, gross fixed capital formation, government expenditure, financial development and trade openness positively and significantly impact the economic growth in the selected EMDEs. Namely, 1% increases of renewable energy use, non-renewable energy use, gross fixed capital formation, general government final consumption expenditure, financial development and trade openness enhance GDP per capita by respectively 0.119%, 0.099%, 0.148%, 0.109%, 0.058% and 0.076%.
From the aforesaid results, we recommend that the policy-makers of the EMDEs in Latin America and the Caribbean should consider fiscal policies for macroeconomic stabilization, improve institutional quality and implement suitable finance-led and tradeled strategies. It is also important to develop energy policies in order to foster the shift from non-renewable energy consumption to the renewable one.