Modern growth theories have also pointed out the various channels such as ICT-trade and financial development which help to attain sustainable economic development and protect the environment. Hence, we can confer that ICT-trade and banking sector development can significantly help to decouple economic growth and CO2 emissions (Ullah et al., 2021). Therefore, to capture the impact of ICT-trade and banking sector development on green growth and CO2 emissions, we have followed Murshed et al. (2020) and Yu et al. (2021) and constructed the following models:
Where green growth (GG) and CO2 emissions (CO2) in equations (1 & 2) are dependent on ICT-trade openness (ICT-trade), bank credit (BC), educational attainment (Education), stock market development (SM), research and development (R&D), and is randomly distributed error term. Equations (1 & 2) reflect the long-run effects of green growth and CO2 emissions. To suppose the short-run effects, we re-write equations (1 & 2) in an error-correction format. The study uses the following linear CS-ARDL regression equation.
Where, and are set of the independent and control variables in equations (3 & 4). The first step in the empirical estimation of the model is to confirm the stationarity of the variables. To that end, we have applied two unit root tests known as the Levin–Lin–Chu (2002) and Im–Pesaran–Shin (2003). After the application of the unit root test and confirmation of the order of integration of the variables, we move to our main model.
For getting short and long-run estimates of the variables, we have relied on the cross-sectional augmented autoregressive distributive lag order (CS-ARDL) model proposed by Chudik and Pesaran (2013). This method has various benefits over other methods. The first-generation estimation techniques such as FMOLS, DOLS, etc can’t address the issue of cross-sectional dependence and provide biased and inefficient estimates (Chen et al., 2022). Whereas, due to the inclusion of lagged dependent variable and cross-sectional averages, the CS-ARDL can deal with the most celebrated problems of the cross-sectional dependence and also deal with an additional issue of slope heterogeneity; hence, providing efficient results (Chudik and Pesaran, 2013). Another benefit of this method is that it can provide both short and long-run estimates simultaneously. Further, pre-unit root testing is not mandatory during the application of this method because it can also account for the variables that are integrated at different orders. Lastly, this method can provide efficient results even in the case of a limited number of observations across time.
To check the robustness of our results, we have applied PMG-ARDL and the quantile regression model. As compared to ordinary least square (OLS), which defines the variation in the dependent variables with the averages of the explanatory variables. Nevertheless, in the case of panel data analysis, some of the basic assumptions of OLS (zero mean, homoscedasticity, and normal distribution) are violated resulting in biased estimates. Therefore, quantile regression analysis is appropriate because in this method dependent variable is the conditional quantile of all independent variables and thereby providing regression for all quantiles. Therefore, quantile regression is better as compared to the OLS in the case of panel data due to its ability to analyze the impact of explanatory variables on the range of variation and conditional diffusion of the dependent variable.
Data
This study is exploring the effects of ICT-trade and bank credit on green growth and carbon neutrality in BRICS economies for the period 1990 to 2020. Table 1 provides the details about the description and abbreviations of variables and descriptive statistics of data. The green growth variable is determined through pollution-adjusted GDP growth. The carbon neutrality variable is measured by CO2 emissions in kilotons. Trade of ICT goods as a percent of total trade is taken to measure the ICT-trade. Bank credit is determined by domestic credit to private sector by bank as percent of GDP. Human capital is determined through education that is measured as average years of schooling. The study has used stock market development and research and development as control variables. Stock market development is measured as stock market capitalization to percent of GDP. Research and development expenditure are taken in percent of GDP. Data for these variables have been extracted from various sources such as OECD, WDI, and IMF.
Table 1: Definitions and data description
Variables
|
Definitions
|
Mean
|
Median
|
Maximum
|
Minimum
|
Std. Dev.
|
Skewness
|
Kurtosis
|
Sources
|
GG
|
Pollution-adjusted GDP growth
|
5.178
|
4.983
|
13.13
|
-4.730
|
3.045
|
-0.457
|
3.602
|
OECD
|
CO2
|
CO2 emissions (kt)
|
13.94
|
14.13
|
16.21
|
12.48
|
1.096
|
0.487
|
2.210
|
WDI
|
BC
|
Domestic credit to private sector by banks (% of GDP)
|
3.805
|
3.966
|
5.108
|
-6.725
|
1.355
|
-5.565
|
40.32
|
IMF
|
ICT-trade
|
ICT goods trade(% total trade)
|
17.13
|
10.00
|
56.78
|
3.128
|
15.41
|
1.506
|
3.640
|
WDI
|
Education
|
Average years of schooling
|
12.38
|
12.90
|
15.50
|
8.200
|
1.964
|
-0.680
|
2.518
|
WDI
|
SM
|
Stock market capitalization to GDP (%)
|
4.049
|
4.027
|
5.794
|
1.807
|
0.794
|
0.266
|
2.804
|
IMF
|
RD
|
Research and development expenditure (% of GDP)
|
1.026
|
0.970
|
2.192
|
0.555
|
0.368
|
1.531
|
5.163
|
WDI
|
Empirical results
The cross-sectional ARDL technique is based on four steps such as cross-sectional dependence test, stationarity test, cointegration test, and long-run and short-run coefficient estimation. As a first step, the study is testing the cross-sectional dependence properties of data. For this purpose, the study adopted the cross-sectional dependence technique developed by Pesaran et al. (2004). Table 2 reports the findings for the cross-sectional dependence test. The findings clarify that all the variables in the model are cross-sectionally dependent. It infers that any change in one of the BRICS economies will affect other BRICS economies as well. The second step is to check the stationarity of variables. In the current study, LLC, IPS, and ADF tests are used to check the panel stationarity of data. Table 3 displays the results of all unit root tests. The findings of LLC unit root test clarify that GG, BC, ICT-Trade, and R&D possess level stationary properties while CO2, Education, and SM variable possess first difference stationary properties. The findings of IPS unit root approach declare that GG and BC are level stationary variables and the remaining variables are first difference stationary. The ADF approach clarifies that GG, BC, and SM variables hold level stationary properties while CO2, ICT-trade, education, and RD hold first difference stationary properties.
Table 2: Cross-sectional dependence tests
|
GG
|
BC
|
ICT-trade
|
Education
|
SM
|
RD
|
Pesaran's test
|
4.527***
|
0.689
|
-1.200
|
4.791***
|
5.326***
|
-1.655*
|
Off-diagonal elements
|
0.313
|
0.289
|
0.412
|
0.402
|
0.382
|
0.474
|
|
CO2
|
BC
|
ICT-trade
|
Education
|
SM
|
RD
|
Pesaran's test
|
6.186***
|
0.898
|
-1.631*
|
2.422**
|
5.099***
|
-1.180
|
Off-diagonal elements
|
0.472
|
0.165
|
0.375
|
0.374
|
0.354
|
0.437
|
Note: ***p<0.01; **p<0.05; *p<0.1
Table 3: Panel unit root tests
|
LLC
|
|
|
IPS
|
|
|
ADF
|
|
|
|
I(0)
|
I(1)
|
|
I(0)
|
I(1)
|
|
I(0)
|
I(1)
|
|
GG
|
-3.750***
|
|
I(0)
|
-3.626***
|
|
I(0)
|
-5.659***
|
|
I(0)
|
CO2
|
-0.790
|
-2.001*
|
I(1)
|
-1.051
|
-4.142***
|
I(1)
|
1.321
|
-6.926***
|
I(1)
|
BC
|
-5.472***
|
|
I(0)
|
-2.898*
|
|
I(0)
|
-3.256***
|
|
I(0)
|
ICT-trade
|
-2.296**
|
|
I(0)
|
-1.971
|
-4.905***
|
I(1)
|
-1.157
|
-8.708***
|
I(1)
|
Education
|
-0.422
|
-1.289*
|
I(1)
|
-0.019
|
-3.421***
|
I(1)
|
1.702
|
-5.126***
|
I(1)
|
SM
|
-1.005
|
-5.919***
|
I(1)
|
-2.023
|
-4.505***
|
I(1)
|
-1.313*
|
|
I(0)
|
RD
|
-1.528*
|
|
I(0)
|
-1.812
|
-3.870***
|
I(1)
|
-0.726
|
-6.285***
|
I(1)
|
Note: ***p<0.01; **p<0.05; *p<0.1
Table 4: Panel cointegration tests
|
Green growth
|
CO2
|
|
|
Statistic
|
Value
|
Z-value
|
P-value
|
Value
|
Z-value
|
P-value
|
Gt
|
-3.690***
|
3.090
|
0.001
|
-8.170***
|
-13.08
|
0.000
|
Ga
|
-6.349
|
2.222
|
0.987
|
-0.197
|
4.440
|
1.000
|
Pt
|
-6.470*
|
1.502
|
0.067
|
-3.241***
|
-2.408
|
0.008
|
Pa
|
-4.351
|
2.030
|
0.979
|
-0.277
|
3.535
|
1.000
|
Note: ***p<0.01; **p<0.05; *p<0.1
In the third step, the long-run relationship between the variables has been examined by employing a cointegration test. The outcomes of panel cointegration tests are given in Table 4. The long-run cointegration association is found among the variables of the models. After confirmation of long-run association among variables, the study employed a cross-section ARDL approach for deducing short-run and long-run coefficient estimates of models. The long-run and short-run coefficient estimates of the green growth model and carbon neutrality model are given in Table 5. In the green growth model, the long-run findings demonstrate that bank credit reports a significant and positive effects on green growth confirming that increase in bank credit results in enhancing green growth in BRICS economies. The finding exhibits that a 1 percent rise in bank credit brings 1.097 percent upsurge in green growth in the long-run. The impact of ICT-trade on green growth is significant and positive in the long-run confirming the increasing pattern of green growth due to an increase in ICT-trade. It infers that a 1 percent increment in ICT-trade intensifies green growth by 0.302 percent in the long-run. The impact of education, stock market development, and research and development is statistically insignificant in the long-run. In the short-run, ICT-trade reports a significant and positive effects on green growth. In contrast, bank credit, education, stock market development, and research and development report statistically insignificant effects on green growth in the short-run. The coefficient estimate of the error correction term is statistically significant and negative confirming the tangency of achieving equilibrium in the long-run. In the quantile regression model in Table 6, the estimates of only ICT-trade are significantly positive in the green growth model at higher quantiles; while, the estimates of bank credit and education are insignificant in most quantiles.
In the carbon emissions model, the long-run findings display that bank credit reports significant and negative effects on CO2 emissions in the long-run confirming that an increase in bank credit results in improving environmental quality. It implies that a 1 percent increase in bank credit reduces CO2 emissions by 1.198 percent in the long-run. ICT-trade and education both variables have a significant and negative effects on CO2 emissions in the long-run. The coefficient estimates display that a 1 percent increase in ICT-trade and education tends to mitigate CO2 emissions by 0.102 percent and 0.289 percent, respectively. Stock market development and research and development have a statistically insignificant influence on CO2 emissions in the long-run. In the long-run, the findings of all independent variables are consistent in terms of direction and magnitude in the robust model. However, control variables report a significant positive effect on CO2 emissions in the robust model.
In the CS-ARDL model, the impact of all the variables on CO2 emissions is statistically insignificant in the short run. However, bank credit and ICT-trade produce a significant effects on CO2 emissions in the robust model in short-run. The findings of the robust model display that Bank credit reports a significant positive effect on CO2 emissions revealing that increase in bank credit results in intensification of CO2 emissions in the short-run. In contrast, ICT-trade reports significant and negative effects on CO2 emissions in a robust model revealing the increase in ICT-trade leads to a significant reduction in CO2 emissions in the short-run. Lastly, the error correction term reports a significant and negative coefficient estimate that confirms the possibility of convergence towards stability in the long-run. However, in Table 6, the estimates of bank credit, ICT-trade, and education are significant and negative in the CO2 emissions model only at higher quantiles.
Table 5: Short and long-run estimates of green growth and CO2 emissions
|
GG
|
|
CO2
|
|
GG
|
|
CO2
|
|
|
CS-ARDL
|
|
CS-ARDL
|
|
PMG-ARDL
|
PMG-ARDL
|
|
Coefficient
|
Z-Stat
|
Coefficient
|
z-Stat
|
Coefficient
|
t-Stat
|
Coefficient
|
t-Stat
|
Long-run
|
|
|
|
|
|
|
|
|
BC
|
1.097***
|
2.890
|
-1.198***
|
3.030
|
1.054***
|
5.744
|
-1.190***
|
2.850
|
ICT-trade
|
0.302*
|
1.840
|
-0.102*
|
1.760
|
0.125***
|
7.348
|
-0.150***
|
3.476
|
Education
|
0.108
|
0.950
|
-0.289*
|
1.750
|
0.145
|
0.774
|
-0.281***
|
9.193
|
SM
|
0.254
|
1.000
|
-0.145
|
1.250
|
0.109
|
0.430
|
-0.159**
|
1.963
|
RD
|
0.842
|
0.840
|
0.689
|
0.170
|
0.621**
|
2.459
|
-0.486***
|
2.849
|
|
|
|
|
|
|
|
|
|
Short-run
|
|
|
|
|
|
|
|
|
D(BC)
|
1.655
|
0.840
|
0.195
|
1.490
|
1.739
|
0.430
|
0.097**
|
2.556
|
D(BC(-1))
|
|
|
|
|
0.769
|
0.260
|
|
|
D(ICT-trade)
|
0.785**
|
2.540
|
0.105
|
1.320
|
0.038
|
0.209
|
-0.003*
|
1.857
|
D(ICT-trade(-1))
|
|
|
|
|
0.507
|
1.358
|
|
|
D(Education)
|
0.154
|
0.050
|
0.265
|
1.360
|
1.288***
|
2.739
|
-0.061
|
1.493
|
D(Education(-1))
|
|
|
|
|
-1.169
|
0.844
|
|
|
D(SM)
|
1.534
|
0.930
|
0.162
|
1.310
|
1.811***
|
4.662
|
-0.014
|
0.736
|
D(SM(-1))
|
|
|
|
|
0.596**
|
2.463
|
|
|
D(RD)
|
1.026
|
0.050
|
0.178
|
0.390
|
1.411
|
1.492
|
-0.034
|
0.309
|
D(RD(-1))
|
|
|
|
|
0.698
|
1.132
|
|
|
C
|
|
|
|
|
1.206***
|
3.741
|
0.394
|
0.531
|
ECM(-1)
|
-0.565
|
2.012
|
-0.501
|
2.987
|
-0.455***
|
6.076
|
-0.442*
|
1.732
|
Note: ***p<0.01; **p<0.05; *p<0.1