B. Results of the Pedroni panel cointegration test
Before the long run relationships of pollution emissions growth, economic growth, squared economic growth, fossil fuel energy consumption growth, and renewable energy consumption growth were empirically estimated, it had to be first established that the variables adhere to long run equilibrium (Zoundi, 2017; Le & Quah, 2018; Hasnisah et al., 2019). To achieve this objective, the study employed the Pedroni panel cointegration (PPC) test to establish whether the null hypothesis of no cointegration is present across the panel samples or not (Pedroni, 1999, 2004). Specifically, the PPC test was implemented to establish if pollution emissions growth has long run relationships with the economic growth variables and energy consumptions. The PPC test reported three test statistics, namely the ADF t-stat, the PP t-stat and the modified PP t-stat that correspond to the three Z statistics mentioned in the methodology section, namely: \(Z, \tilde{Z}\) and\(\tilde{Z}\)*, respectively.
Table 3 showed that the cointegration tests of economic growth and energy consumption variables with pollution emissions for the total panel of Asian countries (with or without China) and by income groupings were generally found to be statistically significant at α = 1% in all the three versions of the tests. Therefore, it can be concluded that there exists strong evidence for the rejection of the null hypothesis of no cointegration. This means that economic growth, squared economic growth, fossil fuel energy consumption growth, and renewable energy consumption growth exhibited long run equilibrium with respect to pollution emissions growth. This result was found to be robust in the total and disaggregated income panels and consistent even after controlling for the presence of China. Hence, the estimated coefficients of the long run relationships from the dynamic panel data regressions done in this study would prove to be reliable.
Table 3
Summary of the results of the Pedroni panel cointegration test for the total Asian panel and by income groups (with and without China).
REGRESSORS
|
TEST STATISTIC
|
Modified PP t-stat
|
PP t-stat
|
ADF t-stat
|
Full panel with China
|
|
|
|
lnn
|
0.8635
|
-5.1062**
|
-3.7630**
|
lnr
|
2.1751*
|
-1.9664*
|
0.0524
|
lny, lny2
|
2.6165**
|
-4.8740**
|
-4.2723**
|
lny, lny2, lnn
|
4.1851**
|
-6.9757**
|
-6.0003**
|
lny, lny2, lnr
|
4.2593**
|
-5.6934**
|
-5.9633**
|
lny, lny2, lnn, lnr
|
5.5358**
|
-9.9776**
|
-9.3861**
|
Full panel without China
|
|
|
|
lnn
|
2.0387*
|
-2.2081*
|
-0.2487
|
lnr
|
0.9741
|
-4.9615**
|
-3.6511**
|
lny, lny2
|
2.5450**
|
-4.9264**
|
-4.3004**
|
lny, lny2, lnn
|
4.2909**
|
-5.0192**
|
-4.6984**
|
lny, lny2, lnr
|
4.2675**
|
-5.4951**
|
-5.8379**
|
lny, lny2, lnn, lnr
|
5.5899**
|
-8.1653**
|
-8.1420**
|
Group 01
|
|
|
|
lnn
|
1.1725
|
-0.6401
|
-0.5038
|
lnr
|
0.5583
|
-3.6395**
|
-2.6586**
|
lny, lny2
|
1.1432
|
-3.7121**
|
-4.0595**
|
lny, lny2, lnn
|
2.5690**
|
-5.7198**
|
-4.4298**
|
lny, lny2, lnr
|
2.4253**
|
-5.3702**
|
-5.3094**
|
lny, lny2, lnn, lnr
|
3.4754**
|
-9.6864**
|
-5.9631**
|
Group 02 with China
|
|
|
|
lnn
|
1.6319
|
-0.5598
|
1.9793*
|
lnr
|
0.9078
|
-3.6273**
|
-1.6077
|
lny, lny2
|
1.7489*
|
-2.3586**
|
-2.1834*
|
lny, lny2, lnn
|
2.4274**
|
-6.3164**
|
-4.9661**
|
lny, lny2, lnr
|
2.7434**
|
-2.7527**
|
-3.2039**
|
lny, lny2, lnr, lnn
|
3.4263**
|
-8.0545**
|
-6.9260**
|
Group 02 without China
|
|
|
|
lnn
|
1.4181
|
-0.9345
|
1.5596
|
lnr
|
1.1068
|
-3.4081**
|
-1.3939
|
lny, lny2
|
1.6364
|
-2.4203**
|
-2.2124**
|
lny, lny2, lnn
|
2.5976**
|
-3.1558**
|
-2.8601**
|
lny, lny2, lnr
|
2.7627**
|
-2.3942**
|
-2.9781**
|
lny, lny2, lnn, lnr
|
3.5179**
|
-5.1291**
|
-4.9115**
|
Group 03
|
|
|
|
lnn
|
0.7367
|
-2.3938**
|
-1.7730*
|
lnr
|
0.0211
|
-2.5894**
|
-2.3191*
|
lny, lny2
|
1.4467
|
-1.7532*
|
-0.8495
|
lny, lny2, lnn
|
2.3748**
|
-0.8879
|
-0.3926
|
lny, lny2, lnr
|
2.1998*
|
-1.6399
|
-1.4412
|
lny, lny2, lnn, lnr
|
2.8182**
|
-2.9526**
|
-2.9195**
|
Regressand: Carbon dioxide emissions growth
|
Ho: All panels are cointegrated.
|
AR parameter: panel-specific
|
** and * denotes significance at α = 0.01 and at α = 0.05, respectively.
|
C. Results of the panel ARDL regression model estimation
Table 4 presents the estimates of the full panel Autoregressive Distributed Lag (ARDL) model with and without China, while Table 5 presents the results of the disaggregated analyses by income groups. The Pooled Mean Group (PMG) and the Mean Group (MG) regression models were both estimated and the Hausman specification test (1978) was performed to choose the better model. The test revealed that the null hypothesis cannot be rejected in the both the full and disaggregated panels. This means that the long run coefficients estimated by the Pooled Mean Group (PMG) model are not statistically different from estimates of the Mean Group (MG) model. Thus, the result of the Hausman test implies that the long run coefficients from the two models are statistically homogeneous. Therefore, the foregoing analyses are focused on the results of the PMG model regressions as basis in analyzing the long run relationships of pollution emissions growth to economic growth, squared economic growth, fossil fuel energy consumption growth, and renewable energy consumption growth, since the parameter estimates from the PMG model are deemed more efficient (Pesaran et al., 1999).
According to the results of the full panel regression, economic growth and its squared form are significantly significant at α = 0.01 with their expected signs, i.e., positive and negative, respectively. This supports the hypothesis that the growth of carbon dioxide emissions exhibits an inverted-U relationship with economic growth, which provides proof for the validity of the Environmental Kuznets Curve (EKC) hypothesis under the Asian setting (Le & Quah, 2018; Shahbaz & Sinha, 2019). That is, a one percent increase in the growth of real per capita GDP of the Asian economies leads to an approximate increase in its carbon dioxide emissions by 1.5046 percent, holding all other factors constant. Moreover, since the established income-pollution relationship behaved in a quadratic fashion, it must be that the response of per capita emissions diminishes for every percent increase in the growth of real income. Therefore, the rate at which the growth of the emissions of per capita carbon dioxide decreases by 0.1424 percent for every one percent increase in the growth of real per capita GDP.
With respect to the relationships of the growth in carbon dioxide emissions with energy consumption, it was found that a one percent increase in the growth fossil fuel energy consumption leads to an approximately 1.0850 percent increase in the growth of per capita carbon dioxide emissions, holding all other factors constant. On the other hand, a one percent increase in the growth of renewable energy consumption leads to an approximately 0.3548 percent decrease in the growth of per capita dioxide emissions, holding all other factors constant. The signs exhibited by the energy consumption variables are consistent with what were hypothesized, i.e., positive and negative, respectively. Note that the expected relationships of growth in emissions with the key variables in the model are robust with the omission of China in the sample panel.
Additionally, the effects of fossil fuel and renewable energy consumption growths were estimated separately due to their likely problem of multicollinearity. The results showed that the EKC hypothesis was found to be valid only on the following conditions: a) fossil fuel energy consumption growth and renewable energy consumption growth were both included in the model; and b) only renewable energy consumption growth was included in the model. The model which omitted renewable energy consumption growth failed to provide evidence for the prevalence of the EKC phenomenon. These findings are robust and consistent even when controlling for the inclusion of China in the sample panel.
In the case of the fossil fuel consumption, the generation of energy from fossil fuels is expected to emit more carbon dioxide and is thus considered destructive for the atmosphere (Rasiah et al., 2018, Sarkodie & Strezov, 2018). Energy generation from renewable sources, on the other hand, is more sustainable for the environment in the sense that such processes emit lesser to no carbon dioxide emission (Zoundi, 2017; Shahbaz & Sinha, 2019). The signs of the coefficients of the energy consumption variables are robust even when one energy consumption variable is omitted and also when controlling for the inclusion of China. The turning point level of real per capita income—the level at which real income growth and pollutions emissions decouple (Grossman & Krueger, 1991, 1995), was found to correspond to 11 percent economic growth, which in per capita terms translates to US$ 38,793.
Table 4
Summary of the results of the full panel ARDL regression model estimations with and without China
VARIABLES
|
FULL PANEL WITH CHINA
|
FULL PANEL WITHOUT CHINA
|
MG
|
PMG
|
MG
|
PMG
|
MG
|
PMG
|
MG
|
PMG
|
MG
|
PMG
|
MG
|
PMG
|
Long run elasticity coefficients
|
|
|
|
|
|
|
|
|
|
|
|
|
Log per capita GDP
|
191.0866
|
1.5046**
|
-1.5147
|
-0.5497
|
-33.8731
|
2.5907**
|
197.1218
|
1.4981**
|
-1.3274
|
-2.9131**
|
-34.8381
|
2.6432**
|
Log squared per capita GDP
|
-16.6148
|
-0.0712**
|
0.1104
|
0.0332
|
1.6642
|
-0.1234**
|
-17.1337
|
-0.0709**
|
0.0989
|
0.1464**
|
1.7097
|
-0.1274**
|
Log fossil fuel energy consumption
|
-1.7549
|
1.0850**
|
-2.1033
|
1.6623**
|
--
|
--
|
-2.0413
|
1.0862**
|
-2.4123
|
1.5631**
|
--
|
--
|
Log renewable energy consumption
|
-63.8356
|
-0.3548**
|
--
|
--
|
-0.8425**
|
-0.4520**
|
-65.7677
|
-0.3541**
|
--
|
--
|
-0.8380**
|
-0.4473**
|
Short run elasticity coefficients
|
|
|
|
|
|
|
|
|
|
|
|
|
Error correction term
|
-1.0417**
|
-0.6665**
|
-1.0696**
|
− .5583**
|
-1.1241**
|
-0.6193**
|
-1.0173**
|
-0.6668**
|
-1.0501**
|
-0.4523**
|
-1.0942**
|
-0.6123**
|
D.(Log per capita GDP)
|
13.1922
|
-12.9743
|
-24.4255
|
-32.7713**
|
-30.1568
|
-7.6245
|
13.2255
|
-13.1787
|
-25.5907
|
-48.7197**
|
-34.3653
|
-9.1510
|
D.(Squared log per capita GDP)
|
-1.2203
|
0.6924
|
1.3346
|
1.9554**
|
1.7399
|
0.4163
|
-1.2325
|
0.7032
|
1.4037
|
3.0072**
|
1.9959
|
0.5054
|
D.(Log fossil fuel energy consumption)
|
-1.0134
|
1.5248
|
-0.1537
|
0.7783
|
--
|
--
|
-0.9307
|
1.5229
|
0.0183
|
1.1611
|
--
|
--
|
D.(Log renewable energy consumption)
|
0.6298
|
0.3842
|
--
|
--
|
0.1494
|
-0.1954
|
0.6319
|
0.3997
|
--
|
--
|
0.1337
|
-0.1954
|
Constant
|
102.8675
|
-6.9918**
|
3.7255
|
-2.1178**
|
54.6420
|
-6.7811
|
105.9873
|
-6.9950**
|
4.0650
|
3.8424**
|
55.6509
|
-6.8299**
|
Regressand: Carbon dioxide emissions growth
|
No. of observations
|
408
|
408
|
408
|
408
|
408
|
408
|
396
|
396
|
396
|
396
|
396
|
396
|
Log Likelihood
|
--
|
825.7931
|
--
|
743.1152
|
--
|
743.8522
|
--
|
800.9872
|
--
|
711.889
|
--
|
718.9039
|
Hausman Test Chi2 statistic
(Ho: PMG is a more efficient estimator)
|
0.1100
|
0.5500
|
1.8100
|
0.1100
|
0.1500
|
0.6243
|
Resulting model
|
PMG
|
PMG
|
PMG
|
PMG
|
PMG
|
PMG
|
Does the EKC hypothesis hold?
|
YES
|
NO
|
YES
|
YES
|
NO
|
YES
|
Turning point GDP per capita
(constant 2010 US$)
|
38,793
|
--
|
36,213
|
38,751
|
--
|
32,003
|
** denotes significance at α = 0.01
|
Table 5
Summary of the results of PMG models for the disaggregated panel ARDL regressions by income groups.
VARIABLES
|
(1)
|
(2)
|
(3)
|
(4)
|
Long run elasticity coefficients
|
|
|
|
|
|
|
|
|
|
|
|
|
Log per capita GDP
|
-1.6771
|
6.8989**
|
6.8506**
|
8.4603**
|
-0.9731
|
3.7942**
|
0.4910
|
-1.4267
|
-1.3129
|
-4.6231
|
-8.3923**
|
16.4359
|
Log squared per capita GDP
|
0.1769
|
-0.4391**
|
-0.4407**
|
-0.5167**
|
0.0594
|
-0.2064
|
-0.0068
|
0.0872
|
0.1214
|
0.2391
|
0.4092**
|
-0.8024
|
Log fossil fuel energy consumption
|
0.9414**
|
1.1158**
|
--
|
-4.1269**
|
3.9827**
|
--
|
1.0264**
|
3.3996**
|
--
|
-2.4912**
|
0.0852
|
--
|
Log renewable energy consumption
|
-0.6147**
|
--
|
-0.5880**
|
-0.2543**
|
--
|
-0.3137**
|
-0.3426**
|
--
|
-0.3546**
|
-0.2636**
|
--
|
0.0379
|
Short run elasticity coefficients
|
|
|
|
|
|
|
|
|
|
|
|
|
Error correction term
|
-0.5044**
|
-0.5848**
|
-0.6736**
|
-0.4518**
|
-0.6453**
|
-0.7409**
|
-0.8192**
|
-0.6858**
|
-0.7561**
|
-0.5619
|
-0.5728
|
-0.5480**
|
D.(Log per capita GDP)
|
25.3147
|
9.8822
|
-2.9388
|
-56.7679**
|
-42.5187**
|
-47.8476**
|
-36.0575**
|
-46.8093
|
-11.9691
|
-9.7964
|
-40.9111
|
-49.9823
|
D.(Squared log per capita GDP)
|
-1.8760
|
-0.8986
|
0.1270
|
3.3863**
|
2.4504**
|
2.8326**
|
2.1557**
|
2.7090**
|
0.8008
|
0.5125
|
2.0269
|
2.4115
|
D.(Log fossil fuel energy consumption)
|
1.6001
|
0.0850
|
--
|
-0.2566
|
-0.4758
|
--
|
-1.7385
|
-0.5808
|
--
|
5.6368
|
3.1208
|
--
|
D.(Log renewable energy consumption)
|
0.9163
|
--
|
-0.4831
|
-0.1099
|
--
|
-0.0811
|
0.0947
|
--
|
-0.0585
|
0.1595
|
--
|
-0.0376
|
Constant
|
0.6709**
|
-18.2238**
|
-16.4462
|
-6.4310**
|
-8.0677**
|
-11.3955**
|
-5.1929**
|
-5.5292**
|
3.3000**
|
20.0207
|
25.8250
|
-44.8627**
|
Note: (1) – low to lower-middle income economies (n = 13); (2) – upper-middle income economies (n = 13); (3) – upper-middle income economies without China (n = 12); (4) – high-income economies (n = 8)
|
Regressand: Carbon dioxide emissions growth
|
No. of observations
|
156
|
156
|
156
|
156
|
156
|
156
|
144
|
144
|
144
|
96
|
96
|
96
|
Log Likelihood
|
310.0069
|
287.6600
|
293.9480
|
330.9667
|
296.0795
|
301.8599
|
317.0680
|
272.0597
|
275.9868
|
185.1126
|
176.9273
|
169.6980
|
Does the EKC hypothesis hold?
|
NO
|
YES
|
YES
|
YES
|
NO
|
NO
|
NO
|
NO
|
NO
|
NO
|
NO
|
NO
|
Turning point GDP per capita
(constant 2010 US$)
|
--
|
2,580
|
2,374
|
3,594
|
--
|
--
|
--
|
--
|
--
|
--
|
--
|
--
|
** denotes significance at α = 0.01
|
As shown in Table 5, the estimated parameters of the full model for the low to middle income economies (Group 1) showed that the income growth regressors were found to be both statistically not different from zero, and thus the EKC hypothesis was found to be non-valid in this income group. In fact, there had been no evidence supporting a functional income-pollution relationship for Group 1 when both energy consumption regressors were included in the model, particularly the growth in fossil fuel consumption. This implies that economies within this group are possibly not yet prioritizing initiatives that aim to minimize environmental pollution emissions, such that their main concern is still biased towards rapid economic growth.
Surprisingly, the EKC hypothesis was found to have been validated in Group 1 regressions whenever the energy consumption regressors lnn (growth of fossil fuel consumptions) and lnr (growth in renewable energy consumption) are included separately in the model. The turning point level of per capita income in the EKC of the low-to-lower-middle income group was found to be approximately US$ 2,580 whenever only fossil fuel energy consumption was included, and that it was about US$ 2,374 whenever only renewable energy consumption was included. For this income group, arguably, the scale effect of the EKC was deemed more dominant than the composition effect and the technique effect. This can be due to the fact that, for the low-to-lower-middle income group in Asia, there will always be a proportional increase in pollution emissions whenever economic growth is achieved (Grossman & Krueger, 1991; Stern, 2004, 2015).
It is rather evident that the turning point was less in the model that includes only renewable energy consumption growth compared to that where only fossil fuel consumption growth was included. This supports the argument that economies which pursue the utilization of renewable energy like wind energy, solar energy, geothermal energy, and the like, will have beneficial effects that would prove to be favorable for the environment. Thus, the results of the low-income group regressions reveal that when economic growth in developing countries can be environmentally sustainable due to the use of greener technologies, the decoupling of economic growth and pollution growth can be achieved much faster.
On the other hand, in the case of the upper-middle income Asian economies (Group 2), only the full model in this respective income group provided evidence that support the validity of the EKC hypothesis with a turning point of approximately US$ 3,594. Interestingly, it was found that this result was not robust when China was excluded in the panel sample (Group 3). This observation, therefore, sheds light on the argument that China could be a significant determinant in the formation of the inverted-U-shaped income-pollution model that is consistent with the EKC hypothesis. This comes to no surprise given that China is arguably economically more developed compared to the other Asian economies falling in the same income category. Since China had already surpassed the estimated turning point in the EKC for this particular income group, its absence in Group 3 sample rendered the remaining economies to be relatively homogeneous in terms of income. Therefore, China was found to be some sort of a leader in the upper-middle income economy group, thus it eventually ushers in the potential decoupling between pollution emissions and economic growth in the Group 2 sample holding all other factors constant. This implies that China had been a significant contributor for the composition effect that signals the potential transition for the upper-middle income Asian economies to progress towards the second half of their group’s EKC (Sarkodie & Strezov, 2018; To et al., 2019).
Lastly, in the case of high-income economies in Asia (Group 4), the results of the analysis provide no evidence for the validity of the EKC hypothesis across all model variants. In fact, according to the results of the ARDL regressions, only the model where the fossil fuel energy consumption growth regressor was included did the study find evidence for a functional nonlinear income-pollution model. However, it was found to be contradictory to the EKC hypothesis as the results argue a quadratically decreasing income-pollution model. That is, pollution emissions growth has been decreasing at an increasing rate with respect to economic growth as being experienced by the relatively affluent economies in Asia. Therefore, whenever only the high-income economies in Asia are considered, it can be said that the technique effect had dominated, such that it begun to offset the scale effect in the income-pollution model. The finding that pollution emissions will eventually decrease at an increasing rate is consistent with the argument summarized by the economic phenomenon of the technique effect (Copeland & Taylor, 2004; Grossman & Krueger, 1991). The technique effect is notably evident in the case of the high-income economies Asia like South Korea, Hong Kong, Japan, Singapore, and the like, which are known to have integrated environmental initiatives onto their development policy frameworks (Hanif, 2018; Hasnisah et al., 2019). Examples of these programs include the Japan-Korea Environmental Policy Dialogue, Singapore’s Voluntary National Review (VNR) on the Sustainable Development Goals (SDGs), etc. This supports the argument that these economies are largely prioritizing programs and initiatives which not only curb the emissions of environmental pollution but also the commendable reduction of pollution emissions itself.
Table 6 presents the summary of the turning points that were estimated from each resulting regression run that yielded evidence supporting the validity of the EKC hypothesis. It can be observed that the turning point levels of real per capita GDP in the full Asian panel (with and without China) were estimated to be within the range of US$ 32,003 to US$ 38,793. Specifically, the estimated turning point level of per capita income in the estimated Asian EKC is approximately US$ 38,793 for the full panel with China, US$ 36,213 for the full panel with China when fossil fuel energy consumption growth was omitted, US$ 38,751 for the Asian panel without China, and US$ 32,003 for the Asian panel without China when fossil fuel energy consumption growth. On the other hand, the turning point in the low income economy group’s EKC was estimated to be within the range of US$ 2,374 to US$ 2,580, while the turning point in the upper-middle income economy group’s EKC was estimated to be about US$ 3,594.
Table 6
Turning point levels of per capita GDP per panel sample
PANEL SAMPLE
|
TURNING POINT(constant 2010 US$)a
|
Full panel
|
38,793
|
Full panel (lnn was omitted)
|
36,213
|
Full panel without China
|
38,751
|
Full panel without China (lnn was omitted)
|
32,003
|
Low income economies (lnr was omitted)
|
2,580
|
Low income economies (lnn was omitted)
|
2,374
|
Upper-middle income economies
|
3,594
|
a Based on authors’ calculations
|
Given the full panel estimated turning point level of per capita GDP ($38,793), it was revealed that only 3 out of 34 Asian economies, specifically, Singapore, Japan, and the United Arab Emirates have reached the benchmark turning point level of per capita income, suggesting that these economies have more environmentally-favorable pollution management strategies compared to their other Asian neighbors. On the other hand, some economies like Hong Kong, Israel, Cyprus, and South Korea are almost at the turning point level of per capita income. Unfortunately, those not mentioned are considerably far away from this turning point level of the estimated EKC. This implies that in the context of Asia, pollution emissions will keep on rising alongside economic growth since the turning point level of per capita income is still far from being reached by most Asian economies.