Sample, Data and Variables
Our sample includes a total of 23 countries across different world regions. These include Brazil, India, China, Russia, South Africa, Nigeria, Ghana, Gabon, Algeria, Egypt, Tunisia, Tanzania, Togo, Mauritius, Namibia, Senegal, Kenya, Indonesia, Malaysia, Sri Lanka, UAE, Pakistan, and Germany. Our panel dataset covers from 2000 to 2022 and includes 529 countryyear observations. All data are gathered from the World Development Indicators, while statistical analysis is done with the aid of EViews and Excel.
Financial Development
This is measured by domestic credit to private sector by banks as a percentage of GDP. According to WDI, this measure comprises bank loans, nonequity securities, and trade credits that are provided to the private sector by deposittaking corporations. Higher credit to private sector ratio implies higher financial development relative to economic growth.
CO2 Emission
This is measured in kiloton and refers to carbon dioxide emitted from the production and consumption of fossil fuels (such as solid, liquid, and gas fuels and gas flaring) and cement manufacturing.
Energy Consumption
This is measured in terms of energy intensity, which is the ratio of energy supply over GDP measured at PPP (Purchasing Power Parity). This measure also indicates the extent of energy used in producing a unit of economic output. Lower ratio implies higher energy efficiency which means that less energy is expended to produce a unit of output.
Trade Openness
This is measured by the ratio of total trade (sum of imports and exports of goods and services) over GDP.
Foreign Direct Investment
This constitutes the net inflow of financial investment to acquire a longterm management interest in domestic enterprises by nonresident investors. It is expressed as a percentage of GDP.
Economic Growth
This is measured by GDP per capita expressed in constant 2015 US$.
Table 1 shows the summary statistics for pooled data. Figures 1 and 6 plot the first and second moments of CO2 emissions, domestic credit to private sector by banks ratio to GDP, energy intensity, trade openness, foreign direct investment, and real GDP per capita for the individual countries.
Table 1
Pooled Descriptive Statistics
Variable

Mean

Std. Dev.

Skewness

Kurtosis

JB(Pvalue)

Obs.

CO2E

611332.4

1689884.0

4.5

24.2

0.0000

483

CPSBY

45.7

34.8

1.3

4.5

0.0000

517

EIL

5.4

2.4

1.0

3.3

0.0000

486

TOP

67.4

36.0

1.5

5.6

0.0000

520

FDIY

2.5

2.2

1.5

6.5

0.0000

529

GDPPC

7342.1

11396.8

2.8

10.3

0.0000

529

Model Specification
We specify the relationships on interest (in functional form) as follows:
$$CO2E = f(CPSBY, EIL, TOP, FDIY, GDPPC)$$
1
$$CPSBY = f(CO2E, EIL, TOP, FDIY, GDPPC)$$
2
Where;
CO2E = CO2 emission
EIL = energy intensity level ratio to GDP (constant PPP)
CPSBY = credit to private sector by banks (% of GDP)
TOP = Trade openness
FDIY = Foreign direct investment ratio to GDP
GDPPC = Real GDP per capita
Method of Analysis
First, we employ the fixed effects regression method to explicitly estimate unobserved countryspecific effects and periodspecific effects that may characterize the relationship between financial development and CO2 emission across several countries. We specify our fixed effects model (in logarithmic form) as follows:
$${\text{L}\text{C}\text{O}2\text{E}}_{it}= {\beta }_{0}+{\gamma }_{i}+{\theta }_{t}+{{\beta }_{1}\text{L}\text{C}\text{O}2\text{E}}_{it1}+{{\beta }_{2}\text{L}\text{C}\text{P}\text{S}\text{B}\text{Y}}_{it}+{{\beta }_{3}\text{L}\text{E}\text{I}\text{L}}_{it}+{{\beta }_{4}\text{L}\text{T}\text{O}\text{P}}_{it}+{{\beta }_{5}\text{L}\text{F}\text{D}\text{I}\text{Y}}_{it}+{{\beta }_{6}\text{L}\text{G}\text{D}\text{P}\text{P}\text{C}}_{it}+{ϵ}_{it}$$
3
$${\text{L}\text{C}\text{P}\text{S}\text{B}\text{Y}}_{it}={\lambda }_{0}+{\gamma }_{i}+{\theta }_{t}+{{\lambda }_{1}\text{L}\text{C}\text{P}\text{S}\text{B}\text{Y}}_{it1}+{{\lambda }_{2}\text{L}\text{C}\text{O}2\text{E}}_{it}+{{\lambda }_{3}\text{E}\text{I}\text{L}}_{it}+{{\lambda }_{4}\text{L}\text{T}\text{O}\text{P}}_{it}+{{\lambda }_{5}\text{L}\text{F}\text{D}\text{I}\text{Y}}_{it}+{{\lambda }_{5}\text{L}\text{G}\text{D}\text{P}\text{P}\text{C}}_{it}+{\epsilon }_{it}$$
4
Where \({\gamma }_{i}\) and \({\theta }_{t}\) are countryspecific and periodspecific fixed effects. The significance of these latent parameters in our model is determined in the usual way by applying the Likelihood ratio and Hausman specification tests. While the Likelihood ratio test is used to test the direct role of these parameters in the specified models, the Hausman test is used to establish the extent of their correlation with other explanatory variables. Besides, both models (3) and (4) are dynamic as they incorporate one lagged value of the dependent variable in as additional explanatory factor. Hence, \({\beta }_{1}\) and \({\lambda }_{1}\) are the persistence terms for CO2 emission and financial development respectively.
The fixed effects estimation approach is built on the assumption that the explanatory variables are strictly exogenous, thereby ignoring the endogeneity issue typically associated with panel data. Hence, our specified models may be plagued by endogeneity bias due to the presence of lagged dependent variables (\({\text{L}\text{C}\text{O}2\text{E}}_{it1}\) and \({\text{L}\text{C}\text{P}\text{S}\text{B}\text{Y}}_{it1}\)) in the RHS of the equations (Arellano and Bond 1991). Endogeneity issue can also arise due to the welldocumented feedback nature of the relationship between CO2 emission and financial development. According to Hao et al. (2016) monetary and financial policies are typically formulated along environmental policies to cope with the simultaneous relationship between financial development and growth of CO2 emissions.
To address the endogeneity concern, we employ the DGMM (Differenced Generalized Method of Moment) approach suggested by Arellano and Bond (1991). This estimation framework differences out the heterogeneity parameter and other constants in the model but relies on instrumental variables to deal with the endogeneity bias. The DGMM models for the dynamic relationship between financial development and CO2 emission are specified as follows:
$${\varDelta \text{L}\text{C}\text{O}2\text{E}}_{it}={{\beta }_{1}\varDelta \text{L}\text{C}\text{O}2\text{E}}_{it1}+{{\beta }_{2}\varDelta \text{L}\text{C}\text{P}\text{S}\text{B}\text{Y}}_{it}+{{\beta }_{3}\varDelta \text{L}\text{E}\text{I}\text{L}}_{it}+{{\beta }_{4}\varDelta \text{L}\text{T}\text{O}\text{P}}_{it}+{{\beta }_{5}\varDelta \text{L}\text{F}\text{D}\text{I}\text{Y}}_{it}+{{\beta }_{6}\varDelta \text{L}\text{G}\text{D}\text{P}\text{P}\text{C}}_{it}+\varDelta {ϵ}_{it}$$
5
$$\varDelta {\text{L}\text{C}\text{P}\text{S}\text{B}\text{Y}}_{it}={{\lambda }_{1}\varDelta \text{L}\text{C}\text{P}\text{S}\text{B}\text{Y}}_{it1}+{{\lambda }_{2}\varDelta \text{L}\text{C}\text{O}2\text{E}}_{it}+{{\lambda }_{3}\varDelta \text{E}\text{I}\text{L}}_{it}+{{\lambda }_{4}\varDelta \text{L}\text{T}\text{O}\text{P}}_{it}+{{\lambda }_{5}\varDelta \text{L}\text{F}\text{D}\text{I}\text{Y}}_{it}+{{\lambda }_{5}\varDelta \text{L}\text{G}\text{D}\text{P}\text{P}\text{C}}_{it}+{\varDelta \epsilon }_{it}$$
6
The endogeneity bias in the models is linked to the correlation between \({\varDelta \text{L}\text{C}\text{O}2\text{E}}_{it1}\) and \(\varDelta {ϵ}_{it}\), and \({\varDelta \text{L}\text{C}\text{P}\text{S}\text{B}\text{Y}}_{it1}\) and \({\varDelta \epsilon }_{it}\). As implied by Arellano and Bond (1991), \({\text{L}\text{C}\text{O}2\text{E}}_{it1}\) is uncorrelated with \(\varDelta {ϵ}_{it},\) and hence can be used as instrument for \({\varDelta \text{L}\text{C}\text{O}2\text{E}}_{it1}\). Similarly, \({\text{L}\text{C}\text{P}\text{S}\text{B}\text{Y}}_{it1}\) can serve as an instrument for \({\varDelta \text{L}\text{C}\text{P}\text{S}\text{B}\text{Y}}_{it1}\). To avoid under identification problem, we include as instruments several periods of the dependent variable from period 2 and use the Sargan test (Jstatistic) to test the validity of all included instruments.
To determine whether financial development and CO2E converge in the long run, we employ the Fisher (Johansen) cointegration test and the pooled mean group (PMG/ARDL) estimation framework. We also employ the Granger causality test framework to determine the direction of causation between financial development and CO2 emission. The PMG/ARDL models are specified as follows:
$${\varDelta y}_{it}={\lambda }_{i}\left({y}_{it1}{X}_{it}^{{\prime }}\delta \right)+\sum _{j=0}^{p1}\varDelta {X}_{it{j}^{{\prime }}}{\beta }_{ij}+\sum _{j=1}^{p1}{\theta }_{i{j}^{*}}\varDelta {y}_{itj}+{ϵ}_{it} \left(7\right)$$
Where \(y\) = dependent variables (CO2E and CPSBY), \(X\) = explanatory variables, \(\delta\) = longrun coefficients, \({\lambda }_{i}\) = speed of adjustment coefficients, and \(\beta\) and \(\theta\) are the shortrun coefficients.
Empirical Results and Discussion
Model Estimation and Analysis
Table 2 displays the dynamic fixed effects regression results for the empirical nexus between financial development and CO2 emission. Table 3 presents the estimated unobserved countryspecific effects in the context of the relationship between financial development and CO2 emissions. To guarantee the robustness of our findings, we also report the DGMM estimation results. The DGMM estimation approach effectively deals with the endogeneity problem that characterize the relationship between most macroeconomic and financial variables through instrumental variables.
As the results show, the model diagnostic tests show that both fixed effects and the DGMM regression frameworks are adequate for estimating the specified relationships under investigation. For fixed effects method, both Likelihood ratio (for both crosssection and period effects) and Hausman test statistics are highly significant for both LCO2E and LCPSBY models, thereby rejecting the assumption that unobserved countryspecific effects are irrelevant in emerging markets in the context of the specified relationships. These results are expected since our sample comprises countries from different regions and with different financial systems and political evolutions. For DGMM, the Jstatistic lacks statistical significance for both models, thereby validating all the instruments used to control the endogeneity of the variables under study. Also, for both LCO2E and LCPSBY models, the AR(1) statistic is negatively signed while the AR(2) is not statistically significant. This is consistent with the expectation that the estimated models are free from secondorder autocorrelation, and hence, are well behaved.
The results show that in emerging markets, as in developed countries, both CO2 emission and financial development are strongly persistent as evident from the positive, sizable, and statistically significant \({y}_{it1}\) for all models. For fixed effects results, the coefficient on LCPSBY is small and statistically insignificant, which shows that financial development is not an important factor driving CO2 emission in emerging markets. Similarly, both trade openness and foreign direct investment are not significant explanatory factors for CO2 emission. However, economic growth and energy consumption both are significant drivers of CO2 emission, with positive coefficients. When compared to the DGMM estimates, we can observe that the results for economic growth, energy consumption, and foreign direct investment are robust, while the results for financial development and trade openness are sensitive to different methodological frameworks.
Turning to the CPSBY model, none of the estimated coefficients is statistically significant for the fixed effects regression, while only the coefficients on LFDIY and LTOP are statistically significant for the DGMM regression. However, the significance of these DGMM estimates occurs at the 10% level, thereby underscoring the weak linkage of financial development to both foreign direct investment and trade openness in emerging markets. Hence, the effects of trade openness and foreign direct investment are sensitive to different methodological frameworks, whereas the lack of significance of the effects of CO2 emission, energy consumption, and economic growth is robust to different methodological frameworks.
From Table 3, the differences in the estimated latent parameters or unobserved countryspecific effects are linked to the differences in the financial system, environmental policies, political environment, and governance mechanisms of the individual countries. These unobserved factors also moderate the relationship between financial development and CO2 emissions. In general, while all the BRICS countries have positive countryspecific effects that affect both CO2 emission and financial development, most African countries are associated with negative unobserved countryspecific effects.
Table 2
Panel Regression Results; pvalues are in parenthesis
Variable

LC02E

LCPSBY

Fixed Effects

DGMM

Fixed Effects

DGMM

Constant

0.1166
(0.5048)


0.2082
(0.3449)


\({y}_{it1}\)

0.8719
(0.0000)

0.6735
(0.0000)

0.8664
(0.0000)

0.6713
(0.0000)

LCPSBY

0.0109
(0.5762)

0.0996
(0.0000)



LCO2E



0.0134
(0.6149)

0.0785
(0.3553)

LEIL

0.1370
(0.0366)

1.0058
(0.0000)

0.0708
(0.1902)

0.2506
(0.5620)

LTOP

0.0177
(0.4167)

0.1181
(0.0093)

0.0070
(0.8470)

0.1577
(0.0797)

LFDIY

0.0038
(0.4187)

0.0047
(0.1486)

0.0027
(0.7018)

0.0165
(0.0985)

LGDPPC

0.1226
(0.0424)

1.0263
(0.0000)

0.0092
(0.8519)

0.4094
(0.1321)

Adj. Rsquared

0.9991


0.9855


Fstatistic (pvalue)

0.0000


0.0000


DurbinWatson

1.9072


1.9126


LR (Crosssection)

79.075
(0.0000)


64.862
(0.0000)


LR (Period)

27.982
(0.0838)


40.879
(0.0025)


Hausman

79.171
(0.0000)


19.112
(0.0040)


Instrument


24


23

Jstatistic


23.207
(0.1827)


21.585
(0.2011)

AR(1)


0.0008
(0.9993)


3.7894
(0.0000)

AR(2)


0.0001
(0.9999)


0.8628
(0.3882)

Table 3
CountrySpecific Fixed Effects
S/n

Country

LCO2E

LCPSBY

1

Brazil

0.0497

0.0365

2

China

0.2064

0.0864

3

India

0.2270

0.0403

4

SA

0.0347

0.0207

5

Russia

0.0813

0.0217

6

Nigeria

0.0143

0.0747

7

Pakistan

0.1024

0.0363

8

Malaysia

0.0098

0.0666

9

Mauritius

0.1767

0.0522

10

Indonesia

0.1250

0.0078

11

Gabon

0.2027

0.0976

12

Ghana

0.0159

0.0572

13

Algeria

0.0355

0.0185

14

Egypt

0.0759

0.0042

15

UAE

0.0887

0.0380

16

Sri Lanka

0.0288

0.0171

17

Tanzania

0.0601

0.0806

18

Tunisia

0.0435

0.0274

19

Togo

0.1622

0.0654

20

Senegal

0.0488

0.0275

21

Namibia

0.1573

0.0115

22

Kenya

0.0456

0.0243

23

Germany

0.0063

0.0567

Cointegration and Causality Analysis
To test whether the variables in our empirical models are cointegrated, we apply the Johansen Fisher Panel Cointegration test, with the results presented in Table 4. As the results clearly show, all the tested cointegrating hypotheses are strongly rejected by both the Trace and Max Eigen statistics. Hence, there is strong evidence that the relationships between financial development, economic growth, trade openness, foreign direct investment, energy consumption, and carbon emission have a longrun dimension.
We estimate the longrun, shortrun, and the error correction coefficients using the Pooled Mean Group/ARDL framework and the results are presented in Table 5. As expected, the error correction term is statistically significant, moderately sized, and negatively signed for both LCO2E and LCPSBY models, thereby confirming that both CO2 emission and financial development have a stable longrun equilibrium relationship with their determinants. However, the estimated ECT coefficients show that financial development has a higher speed of adjustment compared to CO2 emission. The adjustment of CO2 emission after suffering a shock occurs at a speed of 30% per annum, while the adjustment of financial development occurs at a speed of 39% per annum. In the short run, energy consumption and economic growth are the main drivers of CO2 emission while in long run, all the variables are significant explanatory factors for CO2 emission, with trade openness being the only negative factor. For financial development, energy consumption and foreign direct investment are the main explanatory factors in the short run, whereas CO2 emission, trade openness, and foreign direct investment are the significant long run drivers.
Also, we formally test the direction of causality between the study variables. Table 6 presents the Granger causality test results. From the results, we can verify that the relationship between financial development and CO2 emission is not a causal one. Also, there is no evidence of causality between CO2 emission and trade openness. However, we can observe a bidirectional causality between energy consumption and CO2 emission and between economic growth and CO2 emission. Further, there is evidence of a unidirectional causality from foreign direct investment to CO2 emission. Hence, the main drivers of increasing CO2 emissions in most African and emerging markets are foreign direct investment, energy consumption, and economic growth.
Our causality test results also provide evidence of a feedback causation between financial development and economic growth, which validates both supplyleading and demandfollowing theories of financegrowth nexus. There is evidence of a oneway causality from energy consumption to financial development, financial development to trade openness, economic growth to trade openness, economic growth to energy consumption, and trade openness to foreign direct investment. However, there is no causal evidence for foreign direct investment and financial development, trade openness and energy consumption, foreign direct investment and energy consumption, as well as foreign direct investment and economic growth.
Table 4
Hypothesized Cointegration Rank

Trace

MaxEigen

Statistic

Pvalue

Statistic

Pvalue

None

686.7

0.0000

347.2

0.0000

At most 1

494.5

0.0000

357.0

0.0000

At most 2

424.9

0.0000

281.3

0.0000

At most 3

281.7

0.0000

203.0

0.0000

At most 4

146.4

0.0000

104.0

0.0000

At most 5

124.7

0.0000

124.7

0.0000

Table 5
PMG/ARDL Estimates for LCO2E and LCPSBY
Variable

LC02E

LCPSBY

Short Run

Long Run

Short Run

Long Run

ECT

0.3009
(0.0011)


0.3905
(0.0000)


LCPSBY

0.0214
(0.5292)

0.2379
(0.0000)



LCO2E



0.3712
(0.1042)

0.1968
(0.0186)

LEIL

0.8616
(0.0000)

1.1863
(0.0000)

0.6615
(0.0541)

0.0500
(0.5415)

LTOP

0.0148
(0.6850)

0.1055
(0.0176)

0.1194
(0.3017)

0.2221
(0.0000)

LFDIY

0.0081
(0.2986)

0.0237
(0.0030)

0.0226
(0.0393)

0.0432
(0.0000)

LGDPPC

1.0708
(0.0000)

1.4765
(0.0000)

0.3660
(0.3220)

1.5553
(0.0000)

Constant

0.3983
(0.0010)


1.0639
(0.0000)


Linear Trend

0.0011
(0.2337)


0.0003
(0.8907)


Table 6
Causal Direction

Fstatistic

Pvalue

LCPSBY \(\to\) LCO2E
LCPSBY \(\leftarrow\) LCO2E

1.0664
0.4907

0.3452
0.6125

LEIL \(\to\) LCO2E
LEIL \(\leftarrow\) LCO2E

2.5728
7.5709

0.0775
0.0006

LGDPPC \(\to\) LCO2E
LGDPPC \(\leftarrow\) LCO2E

8.7860
3.9439

0.0002
0.0201

LTOP \(\to\) LCO2E
LTOP \(\leftarrow\) LCO2E

0.5423
0.5815

0.1149
0.8914

LFDIY \(\to\) LCO2E
LFDIY \(\leftarrow\) LCO2E

3.1833
1.9665

0.0425
0.1413

LEIL \(\to\) LCPSBY
LEIL \(\leftarrow\) LCPSBY

3.2241
0.4917

0.0408
0.6119

LGDPPC \(\to\) LCPSBY
LGDPPC \(\leftarrow\) LCPSBY

7.3019
10.313

0.0008
0.0000

LTOP \(\to\) LCPSBY
LTOP \(\leftarrow\) LCPSBY

0.6108
2.3575

0.5433
0.0958

LFDI \(\to\) LCPSBY
LFDI \(\leftarrow\) LCPSBY

1.6990
0.7137

0.1841
0.4904

LGDPPC \(\to\) LEIL
LGDPPC \(\leftarrow\) LEIL

11.020
1.4788

0.0000
0.2290

LTOP \(\to\) LEIL
LTOP \(\leftarrow\) LEIL

0.1296
0.5608

0.8784
0.5711

LFDIY \(\to\) LEIL
LFDIY \(\leftarrow\) LEIL

0.1344
0.4867

0.8742
0.6150

LTOP \(\to\) LGDPPC
LTOP \(\leftarrow\) LGDPPC

1.6648
3.2710

0.1903
0.0388

LFDIY \(\to\) LGDPPC
LFDIY \(\leftarrow\) LGDPPC

0.1678
1.6175

0.8455
0.1995

LFDIY \(\to\) LTOP
LFDIY \(\leftarrow\) LTOP

1.0994
5.1234

0.3340
0.0063
