Gök & Sodhi (2021) used system-generalized method (SYS-GMM) of moments for 115 countries clustered as high, middle, and low income over the period of 2000 to 2015, to study the environmental impact of governance. The results show that improvement in governance increases environmental quality in high-income countries.

Dincă et al. (2022) examined the impact of governance quality and educational level on environmental performance for a set of 43 countries European Union (EU) members and The Group of Twenty (G20) states. The data were collected for the 1995-2020 period. The research uses three estimations methods, respectively Pooled ordinary least squares (POLS), Fixed effects model (FE) and Random effects model (RE), together with a two-step dynamic GMM model, to address the endogeneity issue as well. The main results show that all the independent variables reflecting institutional quality from a technical point of view, included in the model when considering the PCSE estimation, have a direct and positive link to CO₂ emissions’ level, with control of corruption variable being the only one to influence in a positive manner CO₂ emissions at a significant level.

Azimi (2022) studied and examined the effects of governance on environmental quality, for 180 countries, over the period 1999 to 2021. They employed Cross-sectional augmented autoregressive distributed lags model, to analyze the stationarity properties of the variables. The findings show that CGI is crucial to improving environmental quality by decreasing CO₂ emissions across all panels. They found that while CGI sustains the same magnitude, the size of its effects substantially varies according to the income level across panels. Moreover, the findings approve that financial development and the human development index reduce CO₂ emissions, while energy consumption, population growth rate, trade openness, and urbanization are significantly decrease environmental quality.

The literature also reveals that Azimi et al. (2023) evaluated the impact of good governance on environmental quality and energy consumption for a panel of 66 developing countries over the period 1991 to 2017 using the GMM method. The authors constructed a governance index using three indicators such as political stability, administrative capacity, and democratic accountability. They found that, though good governance has been significantly positive in affecting energy consumption, globalization has been found to be insignificant in increasing environmental quality.

Jamel et al. (2017) employed causality test and OLS model for a yearly panel data of 40 European economies, during the period of study from 1985 to 2014, to analyze the link between financial development and CO₂ emissions. Their study revealed that financial development measured by domestic credit provided by banks to private sectors has an insignificant effect on CO₂ emissions.

Aluko & Obalade (2020) examined the nexus between financial development and environmental quality, for a panel of 35 in sub-Saharan Africa (SSA) countries for the period 1985-2014, using the augmented mean group estimator. The findings. First, financial development is a positive (negative) driver of environmental quality (CO₂ emissions). Second, financial development has an unfavourable technology effect on environmental quality. Lastly, lower environmental quality is associated with increase in population, affluence and technology. They also find, via the Dumitrescu-Hurlin panel causality test, that there is a bidirectional causal relationship between financial development and CO₂ emissions.

Emenekwe et al. (2022) analyzed the nexus between financial development (FD, and its key dimensions) carbon dioxide (CO₂) emissions for 37 Sub-Saharan African (SSA) countries from 2000 to 2016, using cross-sectional dependence, causality, and regressions analysis using second-generation analytical techniques. The regression analysis technique was the pooled mean group ARDL estimator. They used the dynamic generalized method of moment estimator to provide robustness checks. The findings indicate that overall FD reduces CO₂ emissions in the region and supports the environmental Kuznets curve (EKC) hypothesis. Specifically, the results reveal that a 1-unit increase in the overall FD index results in a 2.867% reduction in CO₂ emissions over the long run.

Habiba and Xinbang (2022) used two-stage system GMM and panel data over the period 2000 - 2018, to examine the impact of financial development on CO₂ emissions for developed and emerging countries. The empirical results show that the overall financial market development and its sub-indices (FM-access, FM-depth, and FM-efficiency) reduce CO₂ emissions in developed and emerging countries. The results further show that the overall financial institution development and its sub-indices, such as FI-access, FI-depth, and FI-efficiency foster the environment quality in developed economies, while these indices impede the environmental quality in emerging economies.

Wang et al. (2022b) used the dynamic panel threshold regression model is employed to explore the effects of ODA and carbon emissions in 59 low-income and lower-middle-income countries. Results indicate that in ODA leads to an increase in carbon emissions. Lee et al. (2020) investigated the relationship between official development assistance (ODA) on CO₂ emissions based on both direct and indirect frameworks, using the annual panel data of 30 recipient countries of Korea from 1993 to 2017. They employed a modified impact, population, affluence, and technology (IPAT) model and a simultaneous equation framework for the direct model and indirect model, respectively. The empirical results suggest that ODA reduces CO₂ emissions for these countries.

Li et al. (2020) examined the impact of Official Development Aid (ODA) on carbon emissions in recipient countries and if institutional quality matters for the effectiveness of green ODA. For 86 green ODA recipient countries over the period 2003-2014, using a two-step system generalized method of moment (GMM). They found a significant effect of institutional quality on the effectiveness of green ODA.

In this study, we evaluate the impact of foreign aid assistance and governance quality on environmental sustainability for a panel of 45 countries¹ according to the data availability over the period 2006-2022 ( Table 1 ).

Based on the study of Kassi et al. (2023) and Azimi et al. (2023) , we construct a composite governance index (CGI) and a composite index. Here, we use the principal component analysis (PCA)’s technique following recent studies by Asongu and Nnanna (2019) and Asongu and Odhiambo (2019) and Habiba & Xinbang (2022) . The composite index of financial development we refer to ( Kassi et al., 2023 ) of financial development to capture the extensive effects of governance and financial development on environment quality.

In this study, we use intermittent variables of governance quality. Data obtained from worldwide government indicators ( WGI, 2022 ), specially the political stability and absence of violence/terrorism (PS), Voice and accountability (VOICE), control of corruption (COR), the rule of law (LAW), government effectiveness (GE), and the regulatory quality (RQ).

In this study, the variables of World Development Indicators (WDI, 2022) for CO₂ (CO₂ emissions, metric tons per capita), (trade openness (TRADE), foreign direct investment (FDI, net inflows, % GDP), URBN (Urban population % of total population)

Unlike previous studies, we use the ODA variable for Sub-Saharan countries of Africa, the both of composite indicators to governance quality and financial development, to evaluate their impact of environment quality for a sample of Sub-Saharan countries of Africa.

To explore the impact of financial development-governance on CO₂ emissions,

we refer to Wen et al. (2022) and Hunjra et al. (2023) and innovatory, we use dynamic panel data for composite indexes of financial development and governance with ODA variable. Otherwise, to test cointegration of variables, we use a dynamic model for panel data with cointegration. The ARDL model used in this study is a type of autoregressive model of order p in the dependent variable $\text{CO}{2}_{it}$, with the form ARDL. In addition, we estimate a dynamic heterogeneous panel data model using a heterogeneous panel of observations using pooled mean group (PMG) and mean group (MG) estimator approaches. The MG estimator is based on the estimate of N time series regressions and the average of their coefficients. Giving to the PMG estimator, we must primarily identify the short-run dynamics for each country. We use this estimator to estimate the long-run effects and the adjustment to the long run rate. The utility of the PMG estimators that they are effective and consistent ( Blackburne & Frank, 2007 ), even in the presence of endogenous and non-stationary regressors. On the other hand, in order to determine whether the two estimators’ differences are systematic, we use the Hausman test.

Hence, we consider the panel equation to explain the relationship under analysis using a panel ARDL model as follows:

$C_{it}={\alpha }_{it}{P}_{it}^b{A}_{it}^c{T}_{it}^d{\epsilon }_{it}$

where i represents the country, t denotes time, $\alpha$ is a constant, and $\epsilon$ is the error term. The natural logarithm transformation give the following panel data model:

$C_{it}=ln{\alpha }_0+bln{P}_{it}+cln{A}_{it}+dln{T}_{it}+{ϵ}_{it}$

The subscript i measures the country i = 1, …, 45; t is the time period from 2000 to 2022; Furthermore, b, c, and d define the estimated percentage.

$\begin{array}{l}\text{CO}{2}_{it}={\alpha }_0+{\alpha }_{1}GO{V}_{it}+{\alpha }_{2}GD{P}_{it}+{\alpha }_{3}OD{A}_{it}+{\alpha }_{4}URB{N}_{it}\\ +{\alpha }_{5}FD{I}_{it}+{\alpha }_{6}PO{P}_{it}+{\alpha }_{7}IN{F}_{it}+{\alpha }_{8}TRAD{E}_{it}+{ϵ}_{it}\end{array}$

where i denotes countries and t signifies study period ${\alpha }_1,{\alpha }_2\text{and}{\alpha }_3$ are the slope coefficients; CO₂ represents carbon emissions.

Where ${ϵ}_{it}$ represents error terms, i is cross section and t is the period. GOV is the composite index of governance, trade indicates trade openness, POP is the population growth, ODA is official aids received, and GDP denotes Gross domestic product per capita growth.

Due to the CSD considerations, this study will not utilize only first-generation unit root tests (First-GURTs), as they could generate biased outcomes. As a result, the current study assessed the stationarity characteristics of panel data from SSA countries using second-generation unit root tests (Second-GURTs). The Second-GURTs, such as the cross-sectional augmented Dickey-Fuller (CADF) test, described in depth in Pesaran (2007) , are used in this study. In addition to the CADF, we do a Fisher-type unit root test in the way described by Oikonomou et al. (2021) .

The Hausman test makes it possible to make a choice between the PMG and MG estimators by testing the null hypothesis of a non-systematic difference between the model coefficients. Furthermore, it also serves as a criterion for choosing between MG and DFE, notably by testing the null hypothesis of a recurring minimal endogeneity bias in dynamic fixed effects models. However, it should be noted that the Hausman test does not allow the PMG and DFE estimators to be compared since the nature of the latter is in agreement with the hypothesis of a non-systematic difference between the model coefficients.

When estimating one or more covariate varias, it can ascent spurious regression. Hence, the first difference strategy, often used, is to make the variables stationary. To this end, two tests explore the effect of country-specifc slope heterogeneity: Pedroni (1999) and Westerlund’s (2005) test. Westerlund’s (2005) test is specified as follows:

$\Delta y_{it}={\omega }_i+{\alpha }_i\left(y_{it-1}-{\beta }_i{x}_{it-1}\right)+{\displaystyle {\sum }_{k=1}^p{\alpha }_{1i}\Delta y_{it-k}}+{\displaystyle {\sum }_{m=1}^p{\beta }_{1i}\Delta x_{it-m}{\epsilon }_i}$

where ${\alpha }_i$ presents the coefficient of the correction term. The null hypothesis is ${\alpha }_i=0$ for all panels, while an alternative hypothesis of ${\alpha }_i<1$ for at least one panel. Accepting the alternative hypothesis recommends the presence of a long-run equilibrium relationship.

Due to the CSD considerations, this study will not utilize only first-generation unit root tests (First-GURTs), as they could generate biased outcomes. As a result, the current study assessed the stationarity characteristics of panel data from SSA countries using second-generation unit root tests (Second-GURTs). The Second-GURTs, such as the cross-sectional augmented Dickey-Fuller (CADF) test, described in depth in Pesaran (2007) , are used in this study. In addition to the CADF, we do a Fisher-type unit root test in the way described by Oikonomou et al. (2021) .

Pesaran et al. (1999) proposed an estimator called PMGE (pooled mean group estimator), which combines both pooling and averaging. In fact, this method imposes a constraint of equality on the long-term coefficients and allows the short-term coefficients to be different from one country to another. The MG estimator is the unweighted average of the coefficients from the different individual regressions. However, the fundamental difference between the two estimators is the fact that the MG estimator does not take into account the possibility that some parameters in the groups may be the same. The PMG estimator constitutes a good estimator in the case where the imposed restriction is verified; but if there is heterogeneity this estimator is biased. The choice between the PMG and MG is made using a Hausman test whose null hypothesis is the non-systematic difference between the coefficients from the MG and PMG.

Hausman (1978) test the hypothesis of homogeneity of the long-term policy parameters cannot be accepted a prior. The impact of heterogeneity on the means of the coefficients can be dictated by the Hausman-type test. If the parameters are in certainty homogenous, the PMG regressors are more efficient than MG. In other words, the efficient regressors under the null hypothesis, which is PMG is favoured. Nevertheless, on the off chance that the null hypothesis is rejected, at that point, the efficient estimator MG is accepted.

Descriptive statistics are reported in Table 2 . This table reveals that, on average, the highest level of CO₂ is equal to 2.133, CC is equal to 1.698, FDI is equal to 4.491, CO₂ is equal to 2.133, INF is equal to 10.076, GOV is equal to 1.150, GDPgr is equal to 4.945, TRADE is equal to 5.463, and ODA is equal to 4.533. Then, we can remark that the highest level of risk (standard deviation) is for CO₂ (1.369), followed by FDI (1.305) and ODA (1.253), respectively.

Principle component analysis of governance and financial development variables

GOVE = (PGOV, EGOV, IGOV, GOV)

Governance variables were grouped into three dimensions, namely, political governance (P.GOV), institutional governance (I.GOV) and economic governance (E.GOV).

We built three components of governance indicators I.GOV I.GOV and E.GOV, using PCA method. Results demonstrate that the first component P.GOV 82.96% of the information contained in its constituents with an eigenvalue of 1.65.

Simultaneously, the first components of I.GOV and E.GOV and the overall index represent 92.96%, 96.07% and 81.65 of the variance in their constituents, with eigenvalues of 1.85, 1.92 and 4.89, respectively. PGOV =f (FRE, PSTA), EGOV=f (REG, GEF), IGOV (COR, LAW). Hence, the highest eigenvalue is for Economic index ( Table 3 ).

When the statistics associated with the tests are significant at the 5% threshold, the null hypothesis is accepted and the first-generation stationarity tests are valid. Then, second-generation tests apply ( Table 4 ).

The results of the adjusted LM test of Pesaran and Yamagata (2008) , contained in Table 5 , confirm the cross-sectional dependence of the series at the 5% threshold. In this regard, second generation unit root tests are the appropriate ones where appropriate.

Notes: The null hypothesis of cross-section dependence CD ~ N (0, 1).

Null hypothesis H0: there is no long-run relationship between the variables included in the panel. ***, ** and * denote statistical significance at 1%, 5%, and 10% level respectively.

We analyze the properties of all variables by running the unit root test in Im et al. (2003) , Levin et al. (2002) , Phillips and Perron (1988) . These tests assume the null hypothesis that all variables are non-stationary at the level against the alternative hypothesis of stationary variables at the first difference. The test in Levin et al. (2002) supposes a standard unit root process, contrary to the three others. Result of panel unit root test reveal that all variables are not stationary in level, but they are stationary at first difference i.e., all variables are I (1). Therefore, we could apply the cointegration test to show that there is a long-term relationship or lack of relationship between variables ( Table 6 ).

Pesaran et al. (1999) proposed an estimator called PMGE (pooled mean group estimator), which combines both pooling and averaging. In fact, this method imposes a constraint of equality on the long-term coefficients and allows the short-term coefficients to be different from one country to another. The MG estimator is the unweighted average of the coefficients from the different individual regressions. However, the fundamental difference between the two estimators is the fact that the MG estimator does not take into account the possibility that

Note: * and ** indicate significance at 1% and 5% levels, respectively.

some parameters in the groups may be the same. The PMG estimator constitutes a good estimator in the case where the imposed restriction is verified; but if there is heterogeneity this estimator is biased. The choice between the PMG and MG is made using a Hausman test whose null hypothesis is the non-systematic difference between the coefficients from the MG and PMG.

The PMG, or “Pooled Mean Group”, method is often used to deal with panel data containing both stationary and non-stationary variables. The main idea behind the PMG method is to combine the advantages of fixed effects models and random effects models. It makes it possible to estimate long-term relationships between cointegrated variables while taking into account individual and temporal effects.

By using the PMG method, we estimate the long-run relationship between cointegrated variables, and obtain the error correction coefficient. This coefficient measures the adjustment of non-stationary variables towards their long-term equilibrium. Its presence confirms the existence of a long-term relationship between the variables.

The PMG method offers a powerful approach for modeling and analyzing economic relationships from panel data, taking into account both stationary and non-stationary variables, as well as individual and temporal effects. The results of PMG and MG are summarized in Table 7 below.

The impact of the interaction GOV*ODA on environment quality.