Foreign direct investments and environmental quality in sub-Saharan Africa: the merits of policy and institutions for environmental sustainability

This study investigates the association between foreign direct investment (FDI) and environmental quality, taking into account policies and institutions for environmental sustainability across 23 sub-Saharan African (SSA) countries. We applied the Generalised Method of Moment (system-GMM) to analyse the data for the period 2005 to 2019. The results revealed that FDI improves environmental quality in the long run, whereas in the short run, FDI diminishes environmental quality when interacted with policies and institutions for environmental sustainability. Furthermore, policies and institutions for environmental sustainability and domestic investment improve environmental quality in both the long and short run. The study concludes that policies and institutions for environmental sustainability in SSA are important as they improve environmental quality. The study also finds policies and institutions for environmental sustainability complements with FDI to improve environmental quality in the long run. Finally, the study further establishes that domestic investment is important to improve environmental quality in SSA. The key findings call for strengthening policies for improving environmental quality in SSA.


Introduction
Environmental conditions, such as pollution (water and air), poor sanitation, and depletion of forest reserves and natural resources, in recent years, have been a major concern for both developed and developing countries. Poor environmental conditions are detrimental to the health and economic welfare of the citizenry. Baloch and Wang (2019) asserted that the essentials of human life, such as health, natural and physical capital, and access to water, food, and land, are prone to climate change. These environmental concerns have brought about a global movement to combat climate change, with the launch of the Paris Agreement and the Kyoto Protocol. The primary goals of these global movements are to mitigate the effects of carbon dioxide (Co2) emissions on the environment. Despite the efforts to reduce Co2 emission, Global Energy and Co2 Status (2019) reported that, in 2018, carbon dioxide emissions increased by 1.7% globally. The report again revealed that the 1.7% increase in global carbon emissions is the highest rate of growth since 2013 and that it is 70% higher, compared to the average increase in carbon emissions since 2010. In sub-Saharan Africa, carbon emissions (in kilotons of Co2) increased from 806,537.52 in 2013to 853,107.13 in 2016(World Bank 2020. World Bank (2020) data further shows that carbon emissions in sub-Saharan Africa (SSA) increased by 1.88% in 2013 and further increased by 7.51% (more than double) in 2014. However, SSA experienced a decline in carbon emissions by 4.11% in 2015 but increased again by 2.6% in 2016. Given the increment in Co2 emissions in SSA, although a marginal fall in 2015, there is no doubt that Co2 emissions adversely affect environmental quality in SSA, which has a negative impact on citizenry welfare and, therefore, calls for urgent attention. Less attention will worsen the negative repercussions of climate change on human life, economic growth, and climatic and ecological systems (Stern et al. 2006;Baloch and Wang 2019).
Foreign direct investment (FDI) is generally argued (both theoretically and empirically) to complement domestic capital to foster economic growth and development, given its benefits (see Bosworth et al. 1999;Alfaro 2003;Bustos 2011;Sakyi et al. 2015). FDI flows to SSA rose by US$32 billion in 2018, representing a 13% increase compared to the preceding 2 years, which were US$28.5 billion in 2017 and US$31.8 billion in 2016 (UNCTAD 2019). For the past 4 decades, the average FDI flows to SSA were 12.5,40.2,20.0,and 37.2 (billion US dollars) for the periods 1980-1989and 2010-2019, respectively (World Bank 2020. Based on the neoclassical theory regarding capital flows (like FDI), which asserts that foreign capital complements domestic resources to augment economic growth (Duodu and Baidoo 2020), one can construe that SSA countries are on the verge of developing, given the flow of FDI and its associated benefits. Lee (2013) and Abdouli and Hammami (2017) documented that FDI flows lead to economic stimulation through access to skills and management expertise, exchange of technology, and employment creation. Despite the positive impact of FDI on economies (especially sub-Saharan Africa), FDI, with its interaction with the environment, somehow affects environmental quality. As a result, researchers as well as policymakers are interested in unravelling whether FDI causes Co2 emissions and depletion of natural resources and forest reserves. A significant amount of literature (see for instance, Ma et al. 2019;Ganda 2020;Farooq et al. 2020) points to the fact that FDI deteriorates environmental quality (i.e. increasing Co2 emission). However, other scholars (e.g. see Hao et al. 2020;Khan et al. 2020;Ahmad et al. 2020) also argue that FDI can improve environmental quality (i.e. reduce Co2 emissions), making the debate on the subject matter still imperative to help policymakers and other stakeholders settle on policies to improve environmental quality.
In their efforts to unveil the environmental effect of FDI, past studies, especially those on sub-Saharan Africa (see for instance, Kivyiro and Arminen 2014;Adams et al. 2020;Asongu and Odhiambo 2020), failed to consider the merit of policy and institutions for environmental sustainability (which reflects the extent to which environmental policies foster the protection and sustainable use of natural resources and the management of pollution), even though policy and institutions for environmental sustainability is an important variable that needs attention when addressing FDI and environmental quality nexus, in the sense that economies with effective policies and institutions for environmental sustainability are likely to implement policies to improve environmental quality. For instance, a country with effective policies and institutions for environmental sustainability is most likely to implement policies that scrutinise the flow of FDI, which enhances environmental quality. This indicates that policy and institutions for environmental sustainability has vital role to play in the relationship existing between FDI and environmental quality.
In the present study, we contribute to the literature on SSA by incorporating environmental policies and institutions for environmental sustainability as well as their interaction with FDI. The existing literature largely uses the quality of political institutions variables (e.g. see Abid 2016;Baloch and Wang 2019;Farooq et al. 2020). This study uses a measure of policy and institutions for environmental sustainability index, which focuses directly on the quality of the environment. This is a departure and, therefore, an important contribution. This inclusion allows us to assess the extent to which the implementation of policy and institutions for environmental sustainability affects environmental quality. The interaction variable helps to further assess the complementary effect of policy and institutions for environmental sustainability with FDI on environmental quality. Furthermore, foreign direct investments are channelled to different sectors of the economy, such as manufacturing, mining, and other extraction activities, and, therefore, could have various impacts on the quality of the environment, especially on pollution and depletion of forest reserves and natural resources. However, most past studies considered the impact of FDI on Co2 emissions but neglected the depletion of forest reserves and natural resources. Most of the studies, especially those on SSA, focused on carbon emissions as a measure of environmental quality (e.g. see Ojewumi and Akinlo 2017;Ssali et al. 2019;Ganda 2020). This does not completely depict the depth to which FDI affects environmental quality. We fill this gap by considering the effect of FDI on Co2 emissions, natural resource depletion, and forest reserve depletion individually. Also, we create an environmental quality index, using the three measures of carbon dioxide emission, natural resource depletion, and forest reserve depletion to assess the overall impact of FDI on environmental quality.
The general objective of the study is to explore the impact of FDI on environmental quality in SSA, taking into account policy and institutions for environmental sustainability as well as its interaction with FDI. The remaining of the paper is structured as follows: the section that follows is devoted to theoretical and empirical review, and the third and fourth sections focus on the methodological framework employed and the analysis of empirical results, respectively. The final section of the study dwells on conclusion and policy suggestions.

Theoretical and empirical review
There are divergent theoretical and empirical views about the impact of FDI on environmental quality. In this section, we briefly review some of the theoretical strands and empirical studies relevant to this study. Concerning the theoretical strands, we review three hypotheses: the pollution haven hypothesis, the scale effect hypothesis, and the pollution halo hypothesis.
The pollution haven hypothesis was, initially proposed by Walter and Ugelow (1979) and later, confirmed by Baumol et al. (1988) and then accepted as an impeccable theory. The pollution haven hypothesis postulates that advanced economies tend to have stringent environmental regulations, making industries, including foreign companies, pay a substantial cost for emitting pollution or degrading the environment (Javorcik and Wei (2003) as cited in Opoku et al. (2021)). As a result, these multinational companies move to economies with more relaxed environmental regulations because of their goals of maximising profits, resulting in more pollution in these underdeveloped countries. No doubt that the World Investment Report (2019) revealed that FDI flows in Africa, including SSA, rose by 11% in 2018. The pollution haven hypothesis, therefore, argues that foreign companies, through FDI, deteriorate environmental quality. With regard to the pollution haven hypothesis, Esty and Dua (1997) documented that economies with less stringent environmental regulations attract foreign companies, which, in effect, adversely affects environmental quality. Considering that stringent environmental regulation is associated with good institutions, some scholars (see Abid 2017;Bokpin 2017;Adams et al. 2020) demonstrated that good institutions improve or complement foreign companies to enhance environmental quality. However, Farooq et al. (2020) in OIC countries observed that institutional quality degrades environmental quality. Their findings could mean that weak institutions fail to provide good environmental regulation and as documented by Hao et al. (2020) in 30 China provinces that environmental regulation deteriorates environmental quality. This notwithstanding ample literature corroborates the pollution haven hypothesis based on how foreign investment (FDI) increases Co2 emissions. For instance, Kivyiro and Arminen (2014), Ojewumi and Akinlo (2017), and Asongu and Odhiambo (2020) provided clear evidence of the pollution haven hypothesis existing in SSA as they reported that FDI causes Co2 emissions to be increased. However, Ssali et al. (2019) explored the nexus between environmental quality (measured by Co2 emission), economic growth, and FDI in 6 SSA countries and found no evidence of the pollution haven hypothesis. Not far from these studies in SSA, past studies elsewhere in the world also revealed that the pollution haven hypothesis is key in explaining the link between FDI and environmental quality. For example, Shahbaz et al. (2015) in both middle-and low-income countries, Seker et al. (2015), and Jiang (2015) pointed out that FDI deteriorates environmental quality (i.e. it has a significant positive impact).
The scale effect hypothesis also explains that foreign or multinational companies, although they increase the output of the industrial sector, heavily consume more energy, which eventually leads to pollution in the recipient or host countries (Pao and Tsai 2011). This hypothesis further augments the argument that FDI results in higher Co2 emissions and has empirically proved valid in the above empirical findings. Even so, in countries with higher Co2 emissions (second pollution emitters) like Middle East and North Africa, Shahbaz et al. (2020). Abdouli and Hammami (2017), Abid (2017), and Shahbaz et al. (2019) indicated that FDI influences Co2 emissions positively. Not far from such conclusion, Mert et al. (2019), Naz et al. (2019), and Solarin et al. (2021) in recent years has pointed out that FDI degrades environmental quality which accords with the pollution haven and scale effect hypotheses.
From a different standpoint, the pollution halo hypothesis postulates that investment in underdeveloped countries, like SSA, by foreign companies from well-developed countries improves environmental quality. The theorists are of the view that foreign companies with good management ideas and forward-thinking technology exert a positive impact on the environment. Eskeland and Harrison ( 2003) posited that foreign (multinational) companies employ clean energy, which leads to more efficient use of energy and which does not expose the environment to any damage. Furthermore, several studies (see, for instance, Saud et al. 2019;Adams et al. 2020;Ahmad et al. 2020) have revealed that FDI through efficient technology improves environment quality. Recently, Khan et al. (2020), Hao et al. (2020), and Bulus and Koc (2021) have demonstrated that the pollution halo hypothesis respectively holds in the context of 58 Belt and Road Initiative countries, 30 provinces in China, and Korea as they reported that FDI reduces Co2 emissions, although the theoretical strands between FDI and environmental quality hinge on the pollution haven hypothesis, the scale effect hypothesis, and pollution halo hypothesis. Ganda (2020) and Aluko and Obalade (2020) indicated that financial development is one way to curb Co2 emissions in SSA. They reported that financial development affects Co2 emissions negatively. This suggests that through the business effect channel as documented by Acheampong et al. (2019), investors can purchase advanced technology that reduces Co2 emissions. Provided in Table 1 are the summary of empirical studies relating to FDI and environmental quality, which support the above theoretical strands or not.
While a plethora of cross-country studies are found in other parts of the world, there exist fewer studies (see for instance, Ojewumi and Akinlo 2017;Ssali et al. 2019;Asongu and Odhiambo 2020) on sub-Saharan Africa, specifically. This, therefore, reveals the essence of further research to complement previous studies to aid policymakers in SSA for effective policy implementation. It was also observed that past studies, especially those on sub-Saharan Africa, ignored the relevance of a country's policies and institutions for environmental sustainability to environmental quality and, thus, failed to depict the complementary role of same with FDI and the impact on environmental quality. Although some studies (see Abid 2017;Bokpin 2017;Hao et al. 2020 and others) attempted to use quality of political institutions (measured by government effectiveness, rule of law, control of corruption, political stability, and others), which may not have direct effect on environmental quality given its wide scope, the results of those studies may render policy implementation ineffective, compared to policy and institutions for environmental sustainability, which is solely geared towards environmental protection. Furthermore, most studies have virtually focused on Co2 emissions as a measure of environmental quality in analysing FDI effects on environmental quality, considering the fact that FDI is mostly channelled to different sectors of the economy (mining and other extraction activities), which could have different effects on the environment. In that respect, this study will measure environmental quality by the depletion of natural resources and forest reserves, aside Co2 emissions to ascertain the diverse impact of FDI on environmental quality. Again, from the empirical literature reviewed and to the best of the author's knowledge, no studies (see empirical review section) have ever created a composite index to assess the full effect of FDI on environmental quality. Thus, there is limited information for policy implications. In the light of the foregoing, this study further contributes to the extant literature by creating an environmental quality index to deliver adequate information to aid policymakers in sub-Saharan Africa in the area of environmental quality. Upon the theoretical and empirical review, we proposed two theoretical hypotheses: (i) stringent policies and institutions for environmental sustainability enhance environmental quality in SSA and (ii) the presence of strong policies and institutions for environmental sustainability complement FDI to improve environmental quality in SSA.

Methodology and model specification
This section focuses on the model specification, data and variable descriptions, and estimation techniques employed.

Theoretical and empirical model specification
The study relies on the Stochastic Impacts Regression on Population, Affluence, and Technology (STIRPAT) model to examine how FDI, policy and institutions for environmental sustainability (PIES), and its interactions affect environmental quality. The STIRPAT model builds on the Ehrlich and Holdren (1971) IPAT (Impact Population Affluence Technology) model, considering the major shortfall of the IPAT model, which is a mathematical identity equation that tends to make hypotheses intolerable (Li and Lin 2015). As a result, Dietz and Rosa (1994) transformed the IPAT model to the STIRPAT (stochastic) model, which accounts for such limitation. The STIRPAT model explains the environmental effect of population, affluence, and technology. Equation (1) is the STIRPAT model specified: where I, P, A, and T denote environmental quality, population (proxied by urbanisation), affluence (proxied by economic growth), and technological progress, respectively. a denotes the constant term, and β, γ, and τ represent the elasticity of population, affluence, and technological progress on environmental quality. Also, i, t, and ε denote crosssectional units, time trend, and the stochastic error term, respectively. Technological progress in this study is assumed to be influenced by FDI, domestic investment, and trade openness, that is, T = f(FDI, DI, TO). Equation (1) can, then, be transformed into Equation (2), which is presented as follows: where FDI, DI, and TO represent foreign direct investment, domestic investment, and trade openness, respectively, and, τ, φ, and σ are their respective elasticities. Equation (2) is then linearised into Equation (3), which is presented below: I, P, A, FDI, DI, and TO are explained in Equations (1) and (2). From Equation (3), it is observed that environmental quality is influenced by urbanisation, economic growth, foreign direct investment, domestic investment, and trade openness. The general specification of Equation (3) is expressed in Equation (4), which is presented as follows: where EQ, URB, and EG represent environmental quality, urbanisation, and economic growth.
Equation (4) is then modified to capture other variables, such as policy and institutions for environmental sustainability (PIES) and international tourism (IT). Every economy implements policies and establishes institutions that protect the environment. Therefore, to examine the extent to which such policies and institutions influence environmental quality, it is imperative to include the policy and institutions for environmental sustainability variable in the environmental quality model. Udemba (2019) asserted that international tourism influences environmental quality positively, thus deteriorating environment quality. In that regard, we also introduce international tourism into the model, considering the significant number of tourist sites that exist in SSA countries, to examine how that influences the environment in SSA. Also, following Abid (2017), Bokpin (2017), Aluko and Obalade (2020), Hao et al. (2020), andFarooq et al. (2020), the study introduces an interaction term between FDI and policy and institutions for environmental sustainability (FDI*PIES) to ascertain how FDI affects environmental quality when policy and institutions for environmental sustainability is improved or is at its mean. From this, the final specification of the generalised model is specified in Equation (5): PIES, IT, and FDI*PIES are defined above. Equation (5) is transformed to its panel dynamic estimable form as expressed in Equations (6) and (7): where a 1 and a 2 are the constant terms. β's (1, 2, 3, . . ., 7) and δ's (1, 2, 3, . . ., 7) denote the variable coefficients to be estimated in Equations (6) and (7), respectively. ε, ln, ϱ i , and η t represent the stochastic error term (which is normally distributed with a mean of zero and constant variance), logarithm, country-specific effect, and time-specific effect, respectively. The λ and γ's (1 and 2) are the coefficients of the lagged dependent variable and the interaction term which captures the impact of FDI on environmental quality, if policy and institutions for environmental sustainability improves. It must be emphasised that Equations (6) and (7) are estimated twice and are referred to as models 1 and 2 in Equation (6) and models 3 and 4 in Equation (7). Models 1, 2, and 3 are the estimations with carbon dioxide emission, natural resource depletion, and forest reserve depletion, respectively. Model 4 is the estimation with environmental quality index (created using principal component analysis). The difference between models 1 and 2 and models 3 and 4 is that carbon dioxide emissions and natural resource depletion are in their log forms, and this explains why we have Equations (6) and (7).

Data and variable description
This study used a balanced panel data of 23 sub-Saharan African countries, spanning 2005 to 2018, and the choice of the study period was as a result of the limited data available for an important variable in the model, that is, policy and institutions for environmental sustainability. Data for the variables (CO 2 emission, natural resources depletion, forest reserve depletion, policy and institutions for environmental sustainability, FDI, domestic investment, international tourism, trade openness, urbanisation, and economic growth) used in this study are extracted from the World Bank's World Development Indicator (World Bank 2020). From the data, we observed that there exist some missing values, and by following Zaman et al. (2016), Bhuiyan et al. (2018), Naz et al. (2019), and Khan et al. (2020), we employed a suitable technique to fill the missing gaps. Table 2 shows a brief description of the variables used in this study.

Estimation technique
To empirically delve into how FDI and policy and institutions for environmental sustainability as well as their interaction affect environmental quality, this study employs the dynamic System Generalised Method of Moment (system-GMM) estimation technique. The system-GMM, proposed by Blundell and Bond (1998), is designed to handle and produce reliable estimates when there exist small T and large N, as in our case of 14 years and 23 cross-sectional countries, which makes the system-GMM an appropriate estimator in our study. Furthermore, due to the dynamic nature of Equations (6) and (7), which results from the presence of the lagged dependent variable (lnEQ it − 1 and EQ it − 1 ), the use of static panel estimators, like the Pooled Ordinary Least Squares, the Fixed Effect, the Random Effect, and the rest, makes the estimates biased and inconsistent. The failure of these panel static estimators to produce reliable estimates in a dynamic model is a result of an endogeneity issue arising from the introduction of the lagged dependent variable, which is correlated with the stochastic error term (ε it ) when Equations (6) and (7) are transformed to their difference form. The system-GMM, however, overcomes this issue, hence the choice of the dynamic panel system-GMM estimation method, which is an extension of the differenced-GMM, by Arellano and Bond (1991) and Arellano and Bover (1995). The general GMM specification of our models to be estimated is specified in Equation (8) as follows: where EQ is the measure of environmental quality, X is a vector of the variables employed, and ε is the error term. The accuracy of the system-GMM estimations is diagnosed, using the Hansen (1982) and Sargan (1958) tests for instrument validity (overidentification restriction) and the Arellano-Bond test for second-order [AR(2)] serial correlations. The nonrejection of the null hypothesis of the Hansen and Sargan tests and the AR(2) test indicates that the instruments are valid and lack second-order serial correlation, respectively, hence the consistency of the estimates. Given that the long-run effect is more vital for policy implications, we adopt the delta method [i.e. β k /(1 − λ)] by Papke and Wooldridge (2005) to estimate the long-run coefficient from the short parameters. This is done by dividing the short-run coefficients by one, minus the lagged dependent variable coefficient. Before using the system-GMM, it is important to consider some preliminary tests, such as cross-sectional dependence test, unit root test, and cointegration test, to prevent spurious, bias, and inconsistent results (Pesaran 2007). The study employs the cross-sectional dependence (CD) test proposed by Pesaran (2004) to test whether there exist dependencies in the data from the 23 countries studied. In this test, the null hypothesis of cross-sectional independence is tested against the alternative hypothesis of cross-sectional dependence. The test follows the standard normal distribution with zero mean and a unit variance [i.e. CD~N(0, 1)] for large cross-section (N) and small time dimension (T). The CD test is specified in Equation (9): After the cross-sectional dependence test, the study proceeds to establish the stationary properties of the variables. To determine the unit root of the variables, we employ both parametric and nonparametric unit root tests as proposed by Wu et al. (2016). Because of the statistical weaknesses, we employ Im et al. (2003), Harris and Tzavalis (1999), and Maddala and Wu (1999)'s Fisher unit root tests and the cross-sectionally augmented IPS (CIPS) unit root test proposed by Pesaran (2004) to account for the cross-sectional dependencies. With respect to these tests, the rejection of the null hypothesis of the panel containing unit roots (nonstationary) implies that the panels are homogenously stationary. However, in the CIPS unit root test, a rejection of the null hypothesis of no homogenous stationarity indicates that there exists heterogeneous stationarity. With respect to the panel cointegration test (long-run relationship), this study applies Pedroni's (2004) cointegration test, and as a robustness check, we further use Kao's (1999)  Note: WDI denotes World Development Indicators Source long-run relationship among the variables) as against the longrun relationship (cointegration) of the alternate hypothesis. The rejection (nonrejection) of the null hypothesis implies the presence of a long-run relationship among the variables (no cointegration).
In the attempt to include an interaction between FDI and PIES, Brambor et al. (2006) argued that it is imperative to generate the marginal effect and its significance. In spite of that, we generate the marginal effect of FDI in the estimable models by calculating the partial derivatives of environmental quality with respect to FDI, and we obtain the standard error of the marginal effects to establish its significance. The general partial derivatives specification for our estimable models is shown in Equation (10): Furthermore, to assess the overall impact of FDI and policy and institutions for environmental sustainability as well as their interaction on the environment but not just some aspects of the environment, the study employs the panel principal component (PCA) analysis to create environmental quality index, using the three aforementioned measures of environmental quality (i.e. carbon dioxide emission, natural resource depletion, and forest reserve depletion). The environmental quality index is constructed using the normalised formula below as adopted by Owusu-Ankamah and Sakyi (2020). It is worth noting that the relative standard deviation is used as weight for the sub-indices.
where EQI, CO 2 , NRD, FRD, and represent environmental quality index, carbon dioxide emission, natural resource depletion, forest reserve depletion, and the relative standard deviation, respectively. The EQI is normalised to zero and one, with zero implying poor (or low) environmental quality and better (or high) environmental quality, as the value is closer to one.

Analysis and discussion of empirical results
Under this section, we discuss and analyse the empirical estimation results. We first discuss the descriptive statistics and the correlation among the sample variables. After that, we discuss the preliminary (cross-sectional dependence, unit root, and the cointegration) test results. Following that, we analyse the long-and short-run results as well as the marginal effect results of the interaction between FDI and policy and institutions for environmental sustainability. Finally, we present and discuss the principal component analysis report.

Descriptive statistics and correlation analysis
Reported in Tables 3 and 4 are the variables descriptive statistics and the correlation matrix results, respectively.
It is observed from Table 3 that the average values of carbon dioxide emission, natural resource depletion, and forest reserve depletion are about −1.66, 2.01, and 6.17, respectively. It is observed that, on average, there is a relatively smaller dispersion around the mean values of the respective variables, as the standard deviation of the variables does not deviate significantly from the average values. It is further noticed that the maximum and minimum values of the sample data are 32.26 and −5.10, respectively.
With respect to the correlation among the variables in Table 4, we noticed that FDI and policy and institutions for environmental sustainability (PIES) have a positive association with the environmental quality measures, with the exception of FDI tending to have a negative relationship with forest reserve depletion measure. While domestic investment has a positive correlation with carbon emissions and environmental quality index, the association tends to be negative with natural resource depletion and forest reserve depletion. On average, we observe from the correlation matrix that the associations among the variables are less than 0.50, with few exceptions, which is an indication that there is less likelihood for multicollinearity to exist in the dataset employed.  Among the variables used in the study, it is revealed from Table 5 that only 3 variables, namely, FDI, domestic investment, and trade openness, exhibit cross-sectional independencies across the 23 SSA countries. This is because the probability values (0.118, 0.632, and 0.347) of these variables exceed the conventional 5% significance level, implying a nonrejection of the null hypothesis of cross-sectional independencies. Aside from these variables, all other variables indicate cross-sectional dependence, given that the probability values lead to rejection of the null hypothesis at a high significance level (1%). With that, we can say that there exists crosssectional dependence across the 23 SSA countries, as the majority of the variables indicate cross-sectional dependencies, and with that, policymakers should consider cross-sectional dependencies among countries when implementing policies. Table 6 shows the results of the panel unit root test. Given the cross-sectional dependencies among the variables, we employ the CIPS unit root test to account for the cross-sectional dependencies existing among the variables (Pesaran 2007). A different unit root test is employed for consistency and robustness.

Panel unit root test results
From Table 6, it is observed that all sample variables are stationary. Specifically, all the tests (IPS, H-T, Fisher, and CIPS) show that carbon dioxide emissions, natural resource depletion, forest reserve depletion, policy and institutions for environmental sustainability, and trade openness are stationary at the first difference [I(1)], whereas FDI and domestic investment are stationary at the levels [I(0)]. While the H-T, Fisher, and CIPS (IPS, H-T, and Fisher) tests show that the environmental quality index (international tourism) is stationary at the levels (first difference), the IPS (CIPS) shows that the environmental quality index is stationary at the first difference (levels). Moreover, the Fisher and CIPS tests indicate that urbanisation is stationary at the levels, but the IPS shows that it is stationary at the first difference. With regard to economic growth, all the tests reveal that it is stationary at the first difference, except the Fisher test, which shows that it is stationary at the levels.  After a valid confirmation of panel stationarity among the variables, the study then proceeds to establish whether there exists a long-run relationship between the variables, using both Pedroni's (2004) and Kao's (1999) cointegration tests. Table 7 is the cointegration test results.

Reported in
From Table 7, it can be seen that there is the validation of a cointegration relationship among the dependent variable (environmental quality) and the explanatory variables in all the estimable models (1, 2, 3, and 4), as shown by the Pedroni's and Kao's cointegration tests. This is because the test statistics in both tests indicate significance at 1% and 5% error levels. This implies a rejection of no cointegration null hypothesis. Given the presence of a long-run relationship among the variables, the study continues to estimate the long-and short-run results as well as the marginal effect of FDI on environmental quality, using the Generalised Method of Moment (system-GMM) estimation technique.

Estimation results and their marginal effect
The long-run results and their marginal effects are reported in Tables 8 and 9, respectively, while the short-run results and their marginal effects are shown in Tables 10 and 11, respectively. We first analyse the long-run results and their marginal effects and then followed by the short-run results and their marginal effects. In this analysis, it is important to note that a positive and negative effect on carbon emissions, natural resource depletion, forest reserve depletion, and environmental quality index means deterioration of environmental quality and improvement in environmental quality, respectively.
Starting with model 1, where environmental quality is measured by carbon emission, it is observed that policy and institutions for environmental sustainability is positively and significantly associated with Co2 emissions, thereby deteriorating environmental quality. However, we find the effect of policy and institutions for environmental sustainability on natural resource depletion (model 2) and forest reserve depletion (model 3) to be negative and significant (improving   Note: *** and ** denote significance levels at 1% and 5%, respectively. Models 1, 2, and 3 represent the estimated model with Co2 emission, natural resources depletion, and forest reserve depletion as a measure of environmental quality, respectively, whereas Model 4 denotes the estimated model with environmental quality index as dependent variable environmental quality), but its overall impact is insignificant on the environment (model 4) in the long run. Specifically, the coefficient in model 1 indicates that an additional improvement in policy and institutions for environmental sustainability worsens environmental quality (increasing Co2 emissions) by about 0.62% at a 1% error level. This outcome can be attributed to the fact that the attitude of policy and institutions towards carbon dioxide emissions in SSA is weak, and this could explain why Co2 emissions continue to increase in SSA countries. Also, the coefficient in model 2 (model 3) reveals that an improvement in policy and institutions for environmental sustainability improves environmental quality in SSA (reducing natural resource and forest reserve depletion) by about 1.01% (0.04 points) at a 1% (10%) significance level.
These results indicate the significance of policy and institutions for environmental quality in SSA countries, and for that matter, policymakers should pay utmost attention to policies that enhance environmental quality. The negative outcome of policy and institutions for environmental sustainability is in line with Bokpin (2017), Abid (2017), and Adams et al. (2020), whereas the positive effect is consistent with Hao et al. (2020), Farooq et al. (2020, and Asongu and Odhiambo (2020). The outcome in model 1 refutes the hypothesis stringent policies and institutions for environmental sustainability enhance environmental quality in SSA, whereas the outcomes in models 2 and 3 accord with the hypothesis. With regard to FDI, it is noticed that the unconditional effect (when there exists no policy and institutions for environmental sustainability) of FDI (Table 8) improves environmental quality in models 1 and 3; that is, it decreases carbon emissions and forest reserve depletion. It, however, tends to worsen environmental quality (increasing natural resource depletion) in model 2 but has an insignificant effect on the overall environment (model 4). To understand the real effect of FDI on environmental quality, it is prudent to account for the conditional impact of FDI on environmental quality, that is, the marginal effect from the interaction term. This is because the partial derivatives in the estimable models show how FDI affects environmental quality when policy and institutions for environmental sustainability is at its mean (i.e. δEQ it δFDI it ¼ β þ γPIES ). We estimate the long-run marginal effect in Table 9, and we observe that an additional increase in FDI improves environmental quality in models 1, 3, and 4, but the effect is insignificant in model 4. However, FDI worsens environmental quality (depleting natural resources) in model 2. Specifically, the coefficients in model 1 show that if policy and institutions for environmental sustainability in SSA is at its mean (or improves), FDI will improve environmental quality (and reduce carbon emissions) by about 0.65 and 1.02% at the 50th-95th percentiles, respectively, if FDI flows in SSA increase by one point and are significant at 5% error level, holding all other variables constant. In model 3 (i.e. forest reserve depletion), the coefficients indicate that a single increase in FDI will improve environmental quality by approximately 0.20, 0.13, and 0.08 points at all percentile levels, respectively, if policy and institutions for environmental sustainability in SSA is at its mean (or improves) and is significant at 1% error level when all other covariates are constant. These results are consistent with the pollution halo hypothesis, which argues that foreign companies use modern technology and, hence, improve environmental quality. The coefficient in model 2 (i.e. natural resource depletion), however, reveals that FDI will deteriorate environmental quality by about 0.70% at the 50th percentile for any additional increase in FDI when policy and institutions for environmental sustainability in SSA is at its mean (or improves) and is significant at 10% error level, holding all other variables constant. This outcome supports the pollution haven hypothesis and the scale effect hypothesis. Overall, we notice that FDI  Note: ***, **, and * denote 1, 5, and 10% significance levels, respectively. Standard errors are in the parenthesis. Models 1, 2, and 3 represent the estimated model with CO 2 emission, natural resources depletion, and forest reserve depletion as a measure of environmental quality, respectively, whereas Model 4 denotes the estimated model with environmental quality index (EQI) as dependent variable improves environmental quality by 0.07 points at the 5th and 25th percentiles, respectively, even though it tends to degrade the environment from the 50th to 95th percentile by about 0.19 and 0.46 points, respectively, for an additional increase in FDI, if policy and institutions for environmental sustainability in SSA is at its mean (or improves), holding all covariates constant but is insignificant, and this could be because the measures for environmental quality do not move in tandem, as shown by the different signs in Table 8. The findings in models 1, 3, and 4 support the hypothesis the presence of strong policies and institutions for environmental sustainability complement FDI to improve environmental quality in SSA. However, model 2 outcome disputes such hypothesis.
Moving to the control variables in the long-run results, it is found that domestic investment in the long run improves environmental quality, specifically, reducing carbon emissions and natural resources depletion, in models 1 and 2 as well as in model 4 (overall effect), but the effect is insignificant in model 4 (environmental quality index). Domestic investment's impact on environment quality tends to be positive (deteriorating the environment) in model 3 (forest reserve depletion). The coefficient in models 1 and 2 shows that, when all other things held constant, a 1% increase in domestic investment in SSA countries will enhance environmental quality by about 0.23% at 1% and 5% significance levels, respectively. This outcome could be attributed to the fact that domestic firms, knowing very well how poor environmental quality will harm their welfare, are less likely to engage in activities that will deteriorate the environment. Bokpin (2017) reported similar results. However, the coefficient in model 3 (forest reserve depletion) indicates that environmental quality will be worsened by about 1.33 points and is significant at the 1% significance level, if domestic investment in SSA countries rises by 1%, when all other variables are held constant. Citizens, in their quest to undertake businesses, may damage or destroy the forest reserve, leading to poor environmental quality. This result is in line with Abid (2017).
Furthermore, we observe that, while international tourism improves environmental quality, that is, it has a negative effect on natural resource depletion and the environmental quality Note: ***, **, and * denote 1, 5, and 10% significance levels, respectively. Model 1, 2, and 3 represent the estimated model with CO 2 emission, natural resources depletion, and forest reserve depletion as a measure of environmental quality, respectively, whereas Model 4 denotes the estimated model with environmental quality index as dependent variable index in models 2 and 4, it tends to worsen environmental quality in models 1 and 3, though we find the effect on the environmental quality index to be insignificant. The coefficient shows that a 1% rise in international tourism will enhance environmental quality by about 0.12%, significant at 1% error level, and 0.74 points in models 2 and 4, respectively, but will deteriorate environmental quality by about 0.04% and 0.76 points in models 1 and 3, respectively, at a 5% and 1% significance levels, respectively. The positive effect of international tourism on environmental quality is consistent with findings by Udemba (2019). Trade openness in all the estimation models, although insignificant in model 4, is observed to deteriorate environmental quality. The coefficients, specifically, reveal that environmental quality will deteriorate by about 0.61%, 0.53%, and 1.68 points in models 1, 2, and 3, respectively, at a 1% and 5% significance levels, if trade openness increases by 1%, holding all other variables constant. This result agrees with the scale effect of the trade openness concept, which argues that trade openness increases carbon emissions and, thus, deteriorates environmental quality, due to the heavy consumption of energy and natural resources. Abid (2017) and Asongu and Odhiambo (2020) reported similar findings, but this result contradicts those of Ansari et al. (2019) and Ahmad et al. (2020). With respect to urbanisation, we observe that urbanisation deteriorates or has positive and significant effects on environmental quality and, thus, increases natural resource depletion in model 2, and its effect on other measures on environmental quality is insignificant. Specifically, the coefficient depicts that a 1% rise in urbanisation induces poor environmental quality by about 0.44% at the 1% significance level, holding all other variables constant. Pressure on the environment as population tends to grow can lead to poor waste disposal as well as depletion of natural resources, resulting in bad environmental quality. While economic growth deteriorates environmental quality by increasing carbon emissions in model 1, the effect is revealed to enhance environmental quality as it decreases natural resource and depletes forest reserve in model 2 and model 3, respectively. The coefficient indicates that a 1% increase in economic growth will deteriorate environmental quality by about 1.26% in model 1, 0.61%, and 0.07 points in model 2 and model 3, respectively, which are all significant at 1% significance level. The negative effect is in line with the findings by Bokpin (2017), whereas the positive impact on carbon emissions conforms with studies by Gunarto (2020), Aluko and Obalade (2020), Khan et al. (2020), Ganda (2020), andHao et al. (2020).
Turning to the short-run results reported in Table 10, although there exist some differences in the signs, magnitudes, and significance, the short-run results do not differ much from the long-run results.   Note: ***, **, and * denote 1, 5, and 10% significance levels, respectively. Standard errors are in the parenthesis. Models 1, 2, and 3 represent the estimated model with CO 2 emission, natural resources depletion (NRD), and forest reserve depletion (FRD) as a measure of environmental quality, respectively, whereas Model 4 denotes the estimated model with environmental quality index (EQI) as a dependent variable The short-run results in Table 10 reveal that the lagged environmental quality measures, that is, carbon emission, natural resource depletion, forest reserve depletion, and the index, all have a significant positive effect on environmental quality, except natural resource depletion, which is insignificant. The coefficients depict that a previous 1% increase in carbon emissions, forest reserve depletion, and the environmental quality index induces carbon emissions (forest reserve depletion and environmental quality index) to rise by about 1.58% (2.20 and 0.87 points, respectively) at a 1% error level when all other factors are constant. This outcome indicates that there exists no convergence as the positive coefficient indicates a drift. This estimate confirms findings by Abid (2017), Farooq et al. (2020, Khan et al. (2020), Asongu and Odhiambo (2020), and Adams et al. (2020). In the short run, we observed that policy and institutions for environmental sustainability enhances environmental quality, specifically, decreases carbon emissions and natural resource depletion, in models 1 and 2. The coefficient shows that, if all other things are controlled, a 1% improvement in policy and institutions for environmental sustainability decreases carbon emissions and natural resource depletion by about 0.36% and 0.85%, respectively, at 1% significance level. However, its effect on forest reserve depletion (which differs from the long-run results) and environmental quality index is insignificant. The short-run effect of policy and institutions for environmental sustainability also emphasises the vital role of policy and institutions geared towards environmental protection and the need for it to be given keen attention in SSA countries to improve environmental quality. This effect is consistent with Abid (2017), Bokpin (2017), and Adams et al. (2020). The short-run results in models 1 and 2 support the hypothesis stringent policies and institutions for environmental sustainability enhance environmental quality.
Contrary to the long-run results, we notice that the unconditional and conditional impacts of FDI deteriorate environmental quality in the short run in all models (1, 2, 3, and 4). Specifically, the marginal effect, that is, the conditional effect of FDI, in Table 11 reveals that, when all other variables are held constant, a single increase in FDI will deteriorate Note: ***, **, and * denote 1, 5, and 10% significance levels, respectively. Models 1, 2, and 3 represent the estimated model with CO 2 emission, natural resources depletion, and forest reserve depletion as a measure of environmental quality, respectively, whereas Model 4 denotes the estimated model with environmental quality index as a dependent variable environmental quality (increasing carbon emissions, natural resource depletion, and forest reserve depletion and decreasing the index value closer to zero) by about 0.38 and 0.59% at the 50th-95th percentile, respectively, in model 1; 0.59% at the 50th percentile in model 2; 0.23, 0.17, and 0.10 points at the 5th-95th percentile in model 3; and 0.02 and 0.06 points at the 50th-95th percentile, respectively, in model 4, if policy and institutions for environmental sustainability is at its mean (or improves) and is all significant at the 5%, 10%, 1%, and 10% significance levels in models 1, 2, 3, and 4, respectively. This outcome could be that policies and institutions that protect the environment in SSA countries are less stringent in the short run to monitor or scrutinise the FDI flows properly, assessing foreign companies with modern technology, and, in turn, cause FDI to deteriorate the environmental quality in SSA. These short-run results, therefore, validate the pollution haven hypothesis, which holds that there exists environmental pollution in developing countries because of less stringent policies. Jiang (2015) and Naz et al. (2019) reported similar outcomes. The results, however, contradict the findings of Abid (2017) and Bokpin (2017). The hypothesis the presence of strong policies and institutions for environmental sustainability (PIES) complement FDI to improve environmental quality in SSA in the short run is rejected.
With respect to domestic investment in the short run, we realise that domestic investment enhances environmental quality in models 2, 3, and 4. However, its impact in model 1 (carbon emission) deteriorates environmental quality, which differs from that of the long-run results. The coefficient indicates that a percent (or single) increase in domestic investment improves environmental quality by approximately 0.19% (1.59 and 0.89 points) in model 2 (model 3 and model 4, respectively) and is significant at 1% and 5% significance levels in models 2, 3, and 4, whereas the coefficient in model 1 shows that environmental quality will deteriorate by about 0.13% at 5% significance level, if domestic investment increases by 1%, holding all covariates constant. As argued earlier, in the case of the long-run results, domestic investors concerned about their welfare are less likely to engage in activities that harm the quality of the environment, and that could explain why domestic investment improves environmental quality in models 2, 3, and 4. This effect contradicts Bokpin (2017) and Hao et al. (2020), but the positive impact on carbon emissions in model 1 supports the findings of Bokpin (2017).
Unlike the results in the long run, we observe from the short-run results that international tourism improves (i.e. has a negative impact) environmental quality in all the models, though its effect in model 4 is observed to be insignificant. The coefficient indicates that, with all things being equal, a percent rise in international tourism is associated with 0.03 and 0.10% (0.91 points) in models 1 and 2 (model 3) at a 5% and 1% (10%) significance levels, respectively. This result contradicts the study by Udemba (2019). Compared to the long run, we notice that trade openness in the short run improves environmental quality in model 1 (model 2). On the contrary, its effect is insignificant in models 3 and 4. Precisely, a percent increase in trade openness induces environmental quality to improve by about 0.36 and 0.45% in model 1 and model 2, respectively, and is significant at 1% significance level, when controlled for all other variables. While the long run supports the scale effect concept of trade openness, the short run only supports it in model 2. Further, economic growth is observed to improve environmental quality in SSA in models 1 and 2 (models 3 and 4), but the impact is insignificant in model 4. The coefficient reveals that quality of the environment in SSA countries will improve by about 0.73 and 0.52% (0.08 points) in models 1 and 2 (model 3), respectively, if economic growth increases by a percent and is significant at 1% (5%) significance level in both models 1 and 2 (model 3) and when all other factors are held constant. The negative effect of economic growth is in line with Bokpin (2017), but Ojewumi and Akinlo (2017) and Ahmad et al. (2020) reported positive effects. Just like what we observe in the long run, urbanisation tends to have a positive significant effect (deteriorate environmental quality) only in model 2, and the coefficient depicts that, with all things being equal, a percent increase in urbanisation is associated with 0.37% deterioration of environmental quality in SSA countries. Adams et al. (2020) and Ahmad et al. (2020) reported similar results.
The second-order [AR(2)] serial correlation and overidentification tests in Table 10 indicate that there is the lack of a second-order serial correlation and there exist valid instruments, respectively, in all estimable models (1, 2, 3, and 4). This is because the probability values of the AR(2) [0.098, 0.231, 0.974, and 0.145) and the Hansen test (0.753,0.073,0.269,and 0.194) fail to reject the null hypothesis of no second-order serial correlation and instruments validity at a 5% significance level. This reveals the accuracy and consistency of the estimable models' parameters. Table 12 shows the report of the principal component analysis.

Principal component analysis report
It is observed from Table 12 that all the components (carbon emission, natural resource depletion, and forest reserve depletion) have a strong positive association with the environmental quality index. Furthermore, we notice that carbon dioxide emissions and natural resource depletion are the predicted variables for the index, since its eigen values (1.4809 and 1.08470) are greater than one and show a higher proportion (about 0.50 and 0.36) of the environmental quality index created. The report from Table 12 may be an indication that carbon emissions and natural resource depletion are key environmental issues in SSA countries and should be a concern, when implementing policies for environmental protection. Given that the Bartlett's probability value is significant at 1%, we conclude that the three variables used for the index are intercorrelated.

Conclusion and policy suggestion
With environmental issues being a major concern globally, of which SSA countries are not an exception to this phenomenon, this study examines the association links between FDI, policy and institutions for environmental sustainability, and environmental quality, using a balanced panel data across 23 SSA countries from 2005 to 2018 (Table 13). The study employs the Generalised Method of Moment (system-GMM precisely) as an estimation technique for the analysis. The results reveal that foreign direct investment (FDI) improves environmental quality in the long run (models 1, 3, and 4), whereas in the short run, FDI deteriorates environmental quality in all models, when it interacted with policy and institutions for environmental sustainability. Policy and institutions for environmental sustainability, in this study, is observed to improve environmental quality in SSA in both the long run (models 1 and 3) and the short run (models 1 and 2). With regard to the control variables, the GMM results indicate that domestic investments in SSA improve environmental quality in both the long run (models 1, 2, and 4) and the short run (models 2, 3, and 4). Furthermore, international tourism is revealed to improve environmental quality in the short run but deteriorate environmental quality in the long run. Moreover, trade openness deteriorates environmental quality in all models in the long run, but the effect improves environmental quality only in model 1 in the short run. While urbanisation in both the long run (model 2) and the short run (model 2) deteriorates environmental quality, economic growth, on the other hand, improves environmental quality in both the long run (models 2 and 3) and the short run (models 1 and 2). The study, therefore, concludes, based on the outcome, that policy and institutions for environmental sustainability in SSA is worthy, as it improves environmental quality and complements FDI to improve environmental quality in the long run. The study further concludes that domestic investment is also important to improve environmental quality in SSA.
The implication of our results suggests that policies to attract FDI in SSA should focused on the long run but not the short run as the immediate effect (short run) of FDI depletes environmental quality. In addition, the finding further suggests that in an attempt to devise policies to enhance environmental quality, governments, policymakers, and other stakeholders in SSA should not only focus on Co2 emissions but should also consider natural resource and forest reserve depletion. Regarding the theoretical implication, our findings support the theoretical argument that FDI improves environmental quality on condition that policy and institutions for environmental sustainability is improved or strengthened.
On the policy front, the study, based on the impact of policy and institutions for environmental sustainability, suggests that governments, policymakers, and other stakeholders in SSA should implement more stringent policies that focus on environmental protection. This can be accomplished if policies and institutions that are geared towards environmental quality are free from political interference and corruption. Furthermore, policies that scrutinise the flows of FDI must Note: The eigen value greater than one, based on the Kaiser criterion, is selected for the principal components. The null hypothesis of no intercorrelation between the variables for the Bartlett's sphericity test is tested against the alternate of variables intercorrelation. CO 2 , NRD, and FRD denote carbon dioxide emission, natural resource depletion, and forest reserve depletion, respectively be keen to policymakers and governments. This will ensure that companies, especially foreign ones, adopt technologies that do not harm environmental quality in SSA, and this in turn reduces the level of pollution and resource and forest depletion. Moreover, on the outcome of domestic investment on environmental quality, the study suggests that governments and other stakeholders in SSA implement policies that enhance domestic investment to improve environmental quality. This can be attained if governments and stakeholders ensure a more friendly business environment in SSA. Doing so will enhance domestic investments and, hence, improve environmental quality, according to the results and, in turn, improve citizenry welfare and, to larger extent, foster economic growth and development in SSA.
The study suggests that future studies should extend the panel to Africa (not just SSA) and beyond to corroborate the present findings. In doing so, differences in policy and institutions for environmental sustainability could be incorporated to access how that influenced the environmental effect of FDI among countries.