Table 1 Unit root test results without structural breaks.
Variable
|
Augmented Dickey and Fuller test
|
Phillips and Perron test
|
At level
|
At first difference
|
At level
|
At first difference
|
t-Statistic
|
Lag
|
t-Statistic
|
Lag
|
Adj. t-Stat.
|
Bandwidth
|
Adj. t-Stat.
|
Bandwidth
|
COAL
|
-1.892
|
0
|
-4.815a
|
1
|
-1.455
|
6
|
-5.265a
|
1
|
GDP
|
-4.289a
|
1
|
-3.603a
|
1
|
-3.871b
|
4
|
-2.565d
|
4
|
GDP2
|
-4.347a
|
1
|
-3.313a
|
0
|
-4.884a
|
2
|
-2.344
|
5
|
REN
|
-1.691
|
0
|
-3.963a
|
0
|
-1.707
|
4
|
-3.780a
|
4
|
OP
|
-2.419
|
0
|
-5.929a
|
0
|
-2.439
|
1
|
-5.966a
|
3
|
Note: a, b, and d indicate significance at the 1%, 5%, and 11% level, correspondingly. Intercept and trend are included for testing in the level equation while only intercept is included when testing in the first difference.
Table 2 Unit root test results with structural breaks.
Variable
|
At level
|
At first difference
|
Intercept break
|
Intercept break
|
|
t-Statistic
|
Time break
|
t-Statistic
|
Time break
|
COAL
|
-1.694
|
2007
|
-6.358a
|
2010
|
GDP
|
-6.564a
|
2002
|
-6.420a
|
2007
|
GDP2
|
-5.138a
|
1994
|
-4.734b
|
2000
|
REN
|
-5.626a
|
1988
|
-5.656a
|
1989
|
OP
|
-3.903
|
2003
|
-6.333a
|
2000
|
Note: a and b represent significance at the 1% and 5% level, respectively. The test assumes an innovation outlier break. We choose both intercept and trend for testing in the level. Only intercept is included when testing in the first difference.
Table 3 Results from bounds tests.
Model specification
|
Lag
|
Foverall
|
tDV
|
FIDV
|
Decision
|
COAL=f(GDP, GDP2, REN_POS, REN_NEG, OP_POS, OP_NEG)
|
2, 1, 2, 2, 2, 2, 2
|
9.704a
|
-11.087a
|
7.238a
|
Cointegration
|
Diagnostic test
|
Normality
|
Serial corr.
|
Hetero.
|
CUSUM
|
CUSUMSQ
|
p-value
|
0.294
|
0.133
|
0.862
|
Stable
|
Stable
|
Note: a denotes significance at the 1% level. _POS and _NEG stand for the partial sum of positive and negative changes. The critical values are extracted from Narayan (2005) for overall F-test (Foverall), Pesaran et al. (2001) for t-test (tDV), and Sam et al. (2019) for F-test (FIDV). The critical value bounds are based on case II with restricted intercept and no trend (Pesaran et al., 2001). See Table A.4.
Table 4 Long-run coefficients using the NARDL approach.
Variable
|
Coefficient
|
t-Statistic
|
p-value
|
GDP
|
-7.5904
|
-3.2782
|
0.0066
|
GDP2
|
0.8092
|
4.5055
|
0.0007
|
REN_POS
|
-0.6471
|
-4.5685
|
0.0006
|
REN_NEG
|
2.3005
|
4.3070
|
0.0010
|
OP_POS
|
0.7809
|
3.4859
|
0.0045
|
OP_NEG
|
-0.1055
|
-0.5041
|
0.6233
|
Constant
|
17.4969
|
2.2722
|
0.0423
|
Table 5 Results of the nonlinear ECM regression.
Variable
|
Coefficient
|
t-Statistic
|
p-value
|
∆COAL(-1)
|
0.2000
|
2.2196
|
0.0465
|
∆GDP
|
93.2622
|
9.3918
|
0.0000
|
∆GDP2
|
-7.2168
|
-9.0618
|
0.0000
|
∆GDP2(-1)
|
-0.3667
|
-3.3661
|
0.0056
|
∆REN_POS
|
-0.1280
|
-1.3409
|
0.2048
|
∆REN_POS(-1)
|
0.4738
|
3.9972
|
0.0018
|
∆REN_NEG
|
-0.7327
|
-4.3947
|
0.0009
|
∆REN_NEG(-1)
|
-2.3677
|
-6.6893
|
0.0000
|
∆OP_POS
|
0.4083
|
3.7158
|
0.0029
|
∆OP_POS(-1)
|
-0.3490
|
-2.8969
|
0.0134
|
∆OP_NEG
|
0.4199
|
4.0527
|
0.0016
|
∆OP_NEG(-1)
|
0.4379
|
4.9423
|
0.0003
|
DUMMY2015
|
0.3917
|
6.4130
|
0.0000
|
ECT(-1)
|
-1.2290
|
-11.0869
|
0.0000
|
Note: A dummy (the year 2015) accounts for years with transitioning to a net coal importer.
Table 6 Testing the presence of asymmetries.
Variable
|
Short-run
|
|
Long-run
|
F-statistic
|
p-value
|
Asymmetry
|
|
F-statistic
|
p-value
|
Asymmetry
|
REN
|
16.981
|
0.001
|
Yes
|
|
30.596
|
0.000
|
Yes
|
OP
|
2.235
|
0.161
|
No
|
|
7.678
|
0.017
|
Yes
|