Some Methodological Considerations for the relationship between Environmental Degradation, Economic Growth and Energy Consumption for South Asian Countries

The present study is designed to investigate that how Economic Growth (EG) of the South Asian region has affected the Environmental Degradation (ED) from 1980 to 2018. This study has used Newey and West (1987) robust standard errors approach to overcome the problem of autocorrelation and heteroskedasticity in panel data. The results of the statistical model confirmed the existence of the Inverted U-shaped Environmental Kuznets Curve (EKC). Furthermore, the results also confirmed that the use of energy is also deteriorating the environment significantly. One significant contribution of the study is to check the causality between EG and CO2 by applying a relatively new approach namely Granger non-causality test presented by Dumitrescu and Hurlin (2012). The results confirmed that economic growth is contributing towards more CO2 emissions. The study concluded that South Asian countries should use environment-friendly renewable energy sources to achieve a higher growth rate.


INTRODUCTION
The alarming situation of Environmental Degradation (ED) after the industrial revolution of the 19 th century drew the attention of international institutions like the World Bank to guide policymakers that all development policies should be integrated with environmental considerations. The World Bank's notion that there exists a trade-off between Economic Growth (EG) and ED (World Bank, 1992) paved the way for researchers to empirically check the relationship between EG and environmental quality. Several studies were conducted to check the relationship between EG and ED by using time series, cross-sectional, and panel data (Jalil and Mahmud, 2009;Joo et al., 2015;Lee and Yoo, 2016;Mehmood, 2021;Mugableh, 2015;Riti et al., 2017). After the pioneer studies of Grossman and Krueger (1995) and Holtz-Eakin and Selden (1995), a significant number of studies, based on the Environmental Kuznets Curve (EKC) hypothesis, show how EG first increase and then gradually decrease the ED (Apergis and Ozturk, 2015;Ardakani and Seyedaliakbar, 2019;Dinda, 2004;Özokcu and Özdemir, 2017;Rahman et al., 2020). In addition to the EG-ED relationship, a significant literature also investigates the role of energy consumption towards ED and studies the nexus among energy consumption, ED, and EG (Ahmad et al., 2016;Bekhet and Lojuntin, 2020;Ito, 2017;Khan et al., 2020;Munir et al., 2020;Omri, 2013;Salari et al., 2021).
The present study is designed to test the EKC hypothesis and role of energy consumption on ED by highlighting some of the common This Journal is licensed under a Creative Commons Attribution 4.0 International License estimation errors made by previous researchers. Furthermore, this study will also check the causality among the energy consumption, ED and EG by using a relatively new technique.
Unprecedented growth in the use of energy, in the 21 st century, to fuel the industry and household consumption has affected the environment adversely. Industrialization in South Asian countries, like the other Asian regions, started in the mid-late twentieth century. The introduction of the free trade concept and the idea of trade-led growth presented by Bhagwati (1988) was well bought by the world. The economies are opening up by following World Trade Organization's (WTO's) principle of integration and new sources of energy has catalyzed the production and industrial growth. This resulted in high growth rates in South Asian countries. However, this growth was at the cost of ED (commonaly measured in terms of Carbon Dioxide (CO 2 ) emission in Literature).
The following figures show the CO 2 emissions per capita, GDP growth and energy consumption per capita (kwh) for the South Asian countries over years.
Bangladesh, India, Pakistan and Sri Lanka become member of WTO in 1995 while Bhutan and Nepal becomes member of WTO in 1999 and 2004 respectively. Following this, the trade liberalization policies and industrialization in South Asian countries increased the CO 2 emissions ( Figure 1). As shown in Figure 2, the GDP growth rate of Bhutan, Bangladesh, India, Nepal and Sri Lanka increased after joining WTO but the average GDP growth rate of Pakistan decreased due to many economicstupor. Moving on to the trend for energy consumption, Figure 3 shows that, since yar 2000, there is a continuous increase in the energy consumption per capita for the South Asian countries.
The high EG rates, use of energy by the world, and ED were well observed by the researchers in the last three decades. The researchers tried to check the nexus among these three variables by taking CO 2 emission as a proxy of ED (Acheampong, 2018;Ardakani and Seyedaliakbar, 2019;Aye and Edoja, 2017;Chontanawat, 2020;Erdoğan et al., 2019;Salari et al., 2021). The ED issue came under the limelight in the last two decades or so. The most important air pollutants include nitrogen oxide, carbon mono oxide, Sulphur dioxide, led, and particulate matter out of which the CO 2 is the worst pollutant that deteriorated the environmental quality (Houghton, 1996). CO 2 absorbs the heat emitted by the planet's surface and retains it in the atmosphere instead of letting it go to space. Although the concentration of CO 2 is just only 0.041% of the earth's atmosphere, its effects are immense on global warming. This led the researchers to link economic activities to CO 2 emissions.
The data used in Figure 3 is taken from https://ourworldindata.org/energy   , 2021;Shaari et al., 2014). If these potential problems are neglected then the standard least square estimation may produce specious regression (Arellano, 2003;Baltagi, 2008;Drukker, 2003;Hsiao, 2007;Kao, 1999;Pesaran, 2015b;Qasim et al., 2021;Wooldridge, 2010). The present study will take into consideration the consequential issue associated with the disturbance term. This study verifies the possible problems associated with an error term and by following Newey and West (1987) estimates the robust standard errors to avoid the spurious regression.
Furthermore, many studies had tested the causal nexus between EG and CO 2 emission by using the traditional Granger causality (Ahmed et al., 2017;Al-Mulali and Che Sab, 2018;Chang, 2010;Erdoğan et al., 2020;Fodha and Zaghdoud, 2010;Gorus and Aydin, 2019;Islam et al., 2017;Lee and Yoo, 2016;Mikayilov and Hasanov, 2020;Mugableh, 2015), which may lead to methodological bias in the presence of CD (Dumitrescu and Hurlin, 2012;Wursten, 2017). In contrary to past studies, this study has applied a relatively new approach of causality presented by Dumitrescu and Hurlin (2012). This approach produces robust estimates in the presence of CD. The current paper revisits the "EKC" hypothesis by using robust standard errors and also checks the causal relationship between EG and CO 2 emission by using an updated methodology for six South Asian countries namely, Bangladesh, Bhutan, India, Nepal, Pakistan, and Sri Lanka.

Data Sources
The present research aims to empirically examine the Environmental Kuznets Curve (EKC) hypothesis and causality between Environmental Degradation (ED), measured through Carbon Dioxide (CO 2 ) emission per capita (CO2EPC), and real GDP per capita (LRGDPPC) for the six South Asian courtiers namely Bangladesh, Bhutan, India, Nepal, Pakistan, and Sri Lanka. Afghanistan and Maldives are excluded from the present research due to the non-availability of data on the variables of interest. Moreover, the analysis for the study was conducted by using data from 1980 to 2018. Data on the variables used in the study is taken from World Development Indicator (WDI) and Penn World Table (PWT).

The Model
Previous studies tested the "EKC" hypothesis by taking LRGDPPC and LRGDPPCSQ as independent variables and LCO2EPC as a dependent variable (Abbasi et al., 2020;Apergis and Ozturk, 2015;Kasman and Duman, 2015;Shabani et al., 2021;Sharma et al., 2021). The present study has also used the same variables to develop a model with the addition of the EU as an independent variable. The model is presented below: (1) Where: LCO2EPC it = is the log of CO 2 emission per capita LRGDPPC it = is the log of GDP per capita, at constant prices LRGDPPCSQ it = is the log of GDP per capita Square LEU it = is the log of energy use i = indicates the number of cross-sections t = shows the number of time periods Ψ = is the slope coefficient ε it = is the white noise error term

Econometric Modeling
Keeping in mind the nature of the data and objectives of this research, the present study has used panel data. In conventional panel data analysis, there are three classifications of models namely the fixed effects (FEM), Random Effects (REM), and common constant model (Baltagi, 2008;Studenmund, 2014). Hausmann specification test was applied to check which of these three is suitable for the present study. Hausmann (1978) proposed a model choice test between FEM and REM. The null and alternative hypotheses are given below:

D 0 FEM is consistent and efficient
And the test statistic is, The results of the Specification test suggested FEM is more efficient and appropriate for the model used in the present study.

Fixed effect model
The general econometrics form of FEM is as follow; And the matrix form Where y it = is the dependent variable X it = is the 1×K vector of independent variables β = is the k×1 matrix of slope coefficients α i = captures the unobserved time-invariant effect of cross-sections The model is referred to as the "fixed-effect model" because this model produces an independent intercept term for every individual entity. The subscript "i" of the slope coefficient considers that each entity may have a different intercept value but this value does not change over time i.e., it is time-invariant (Baltagi, 2008;Greene, 2003;Qasim et al., 2021;Wooldridge, 2010;2016).

Robustness checking
To avoid misleading results, the econometricians proposed different robustness tests to check the assumptions of Ordinary Least Square (OLS). The present study has applied an autocorrelation test for panel data proposed by Drukker (2003) and modernized by Wooldridge (2010). The null hypothesis for this test is the absence of autocorrelation. Furthermore, Modified Wald test is applied to check the problem of heteroscedasticity in the data. The null hypothesis for this test is the absence of heteroscedasticity. The results of both the tests reject their respective null hypotheses, indicating that there is a problem of autocorrelation and heteroscedasticity in the data To account for the issue of unequal variance and serially correlated disturbance, the present study has used robust standard error method proposed by Newey and West (1987).

Newey and west robust standard error
To cope with biased standard errors, as a result of unequal variance and serially correlated disturbance, Newey and West (1987) put forward the statistical inference to correct the standard errors.
The corrected standard errors provide efficient and optimal results (Arellano, 2003;Baltagi, 2008;Hsiao, 2007;Pesaran, 2015b). The Newey and West estimator is given below: To produce robust standard errors, the first step is to calculate the maximum number of lags. Andrews (1991) and Newey and West (1987) have suggested the criteria for best lag length selection. The formula for lag selection is as follows: Similarly, Hoechle (2007) recommended the general principle for approximating the lag length.

( )
Where m(T) = optimal lag length T = number of time periods

Dumitrescu and hurlin non-causality test
This study has applied the non-causality Granger test for the heterogeneous panel data model proposed by Dumitrescu and Hurlin (2012). The general model of panel non-causality test for two stationary variables, x and y observed for individual crosssection, i and time, T is given below:

= represents the regression coefficients
The null and alternate hypotheses of homogenous non-causality (HNC) test are as follows: The test statistic for NHC is given below: Where, W i,T = is the Wald statistics for cross-section for H 0 : β i = 0 and = is the standardized test statistic for fixed T.

Cross-sectional dependence (CD)
The non-causality panel causality test assumes that x and y are stationary. Before testing the panel unit root, the first step is to confirm the presence of CD in cross-sections. CD is defined as the correlation between spaces instead of time. This study has applied the Pesaran CD (Pesaran, 2004) and Pesaran Scaled LMS tests to check the existence of CD (Pesaran, 2015a).
2.3.5.1.1. Pesaran's CD tests CD is the potential problem of the panel or cross-sectional data (Baum, 2011;Froot, 1989;Hoechle, 2007). The most common test of CD for the panel data was devised by Pesaran (2004). The Stata package of the CD test was developed by Wursten (2017) in STATA. CD test statistic is given below: H 0 : No cross-sectional dependence 2.3.5.2. Choice of unit root test After checking the CD, the last step is to check the stationarity of the variables. The choice of unit root test depends on the absence or presence of CD in the variables. First-generation unit root tests are used if the problem of CD does not exist in the data Arellano (2003); (Im et al., 2003;Levin et al., 2002). Second generation unit root test is applied if there exists the problem of CD in the data (Im et al., 2003;Pesaran, 2007). In this study, a second-generation test namely the Crosssectionally-augmented IPS (CIPS) test is applied due to the existence of CD. In the presence of CD, the first-generation tests are misleading.

CIPS test
The second-generation unit root test was developed by (Im et al., 2003;Pesaran, 2007). The general model of the CIPS test is given below: H 0 : Presence of a Unit Root

RESULTS AND DISCUSSION
Before estimating the model, the matrix of correlations among three variables of the study is obtained. The results are given in Table 1. Table 1 analyzes the cross-correlation of the variables used in the study. All variables are positively correlated with each other. The correlation of LCO2EPC with LRGDPPC, LRGDPPCSQ and LEU is 0.999, 0.435, and 0.555 respectively.
Hausman specification test suggested that FEM is an appropriate estimator. Table 2 shows the regression results with (Newey and West, 1987) robust standard errors approach. These standard errors corrected the problems of serial correlation and heteroscedasticity in the model. To obtain unbiased results, this study has used the Wooldridge test for autocorrelation and Baum Modified Wald test for group-wise heteroscedasticity. The results suggested the presence of serial correlation and heteroscedasticity in the model as null hypotheses of both tests are rejected at a 1% level. Therefore, this study has used the Newey and West (1987) standard errors. These standard errors have produced efficient and unbiased estimates as propagated by Arellano (2003), Baltagi (2008), and Raj and Baltagi (2012).
The results presented in Table 2 confirm that at the beginning LRGDPPC is posing a negative effect and after reaching the point of inflection the effect becomes positive on the environment. The relationship of LRGDPPC and LRGDPPCSQ with Carbon Dioxide (CO 2 ) emission confirmed the Inverted-U shape EKC for the selected countries. It implies that initially, EG worsens the environmental quality. After reaching the turning point, it will ameliorate the environment (Grossman and Krueger, 1995). Several studies in the literature have confirmed the existence of the EKU hypothesis (Ardakani and Seyedaliakbar, 2019;Aye and Edoja, 2017;Özokcu and Özdemir, 2017;Salazar-Núñez et al., 2020;Shahbaz et al., 2020). Similarly, LEU has a direct and statistically significant effect on LCO2EPC. This suggests that increased energy use has caused Environmental Degradation (ED). These findings are similar to Bölük and Mert (2015), Chen et al., (2016), Franklin and Ruth (2012), Heidari, Katircioğlu, and Saeidpour (2015)  The present study has also checked the causation between LRGDPPC and LCO2EPC by using the non-causal Granger test proposed by Dumitrescu and Hurlin (2012). The first step to estimate the non-causality test is to check expected Cross-sectional Dependence (CD) in panel data.
The results of Pesaran's CD test and Pesaran's LMS test are presented in Table 3, indicate the presence of correlation between cross-sectional units at a 1% level of significance. Hence, as discussed earlier, the Pesaran Cross-sectionally-augmented (CIPS) second-generation unit root test is applied. These results are represented in Table 4 below: The results of CIPS rejected the null hypothesis that time series are non-stationary at a 1% level suggesting that the time series LRGDPPC and LCO2EPC are stationary. Based on the results of pre-tests, the study has used Dumitrescu and Hurlin (2012) noncausal Granger test. The results are presented in Table 5.
The null hypothesis of panel causality is that the independent variable does not homogeneously cause the variable under consideration. The results of the causality test indicate the unidirectional causality running from LRGDPPC to LCO2EPC. The p-value of W-Stat and Zbar-Stat is less than 0.000 suggesting the rejection of the null hypothesis at a 1% level of significance.
The results also indicate that causality running from LCO2EPC to LRGDPPC does not exist as some researchers pointed out in literature (Islam et al., 2017;Lee and Yoo, 2016;Wang et al., 2020).

CONCLUSION AND POLICY IMPLICATIONS
The present study is designed to empirically re-examine the Environmental Kuznets Curve (EKC) hypothesis and the causality between Economic Growth (EG) and Carbon Dioxide (CO 2 ) emissions for six South Asian courtiers, namely Bangladesh, Bhutan, India, Nepal, Pakistan, & Sri Lanka, using latest estimation methodologies. The fixed-effect model with Newey and West (1987) robust standard error indicates that LRGDPPC has positive and significant impact on CO 2 emission while LRGDPPCSQ has a negative and significant impact on CO 2 emission. These findings confirm the existence of inverted U-shaped EKC. Furthermore, the results also confirm the existence of positive and significant impact of Energy Use (EU) on CO 2 emissions.
The present study has also applied a non-granger causality test and the result suggest unidirectional causality running from EG to CO 2 emissions. In literature, most of the studies applied the traditional Granger causality test but this study has applied a relatively new non-causal Granger test presented by Dumitrescu and Hurlin (2012) to check the causality between EG and CO 2 emissions. The results confirm that EG is significantly contributing towards CO 2 emissions. The results of the study reconfirm the EKC hypothesis for South Asia using improved estimation methodologies. Based on empirical findings; this study recommends that the policymakers of the six South Asian courtiers selected for this study should formulate such policies which encourage adopting environment-friendly technologies. The energy sources should be shifted from non-renewable to renewable energy sources to minimize the environmental hazard.