RETRACTED ARTICLE: International trade and environmental pollution in sub-Saharan Africa: do exports and imports matter?

Abstract

Sub-Saharan Africa (SSA) and Africa in general are known as the lowest emitters of carbon dioxide (CO₂) emissions. However, CO₂ emissions in SSA are increasing, making it a problem of concern and calls for attention given its adverse consequences on human health and climate change. International trade is argued to have a vital role in global and SSA emissions in diverse ways, leading to doubts of whether trade is good or bad to the environment. As a result, we explore the environmental effect of international trade in 33 SSA countries from 1990 to 2020. The study further evaluates the differential effect of exports and imports on environmental pollution. The generalized method of moment estimator and Dumitrescu and Hurlin (D-H) causality test were utilized. The results revealed that the overall effect of trade reduces environmental pollution by about 0.10% and 0.79% in both the short and long run, respectively. Again, we observe that exports and imports minimize environmental pollution of about 0.07% and 0.45% (0.08% and 0.58%) in the short run (long run), respectively. Regarding D-H results, we noticed the existence of bidirectional causality between total trade and environmental pollution, whereas exports and imports have a unidirectional causality from CO₂ emissions to exports and imports. We conclude based on the findings that international trade causes pollution reduction in SSA. Furthermore, we establish that exports and imports have a homogeneous impact on environmental pollution in SSA. Given the results, we call for trade initiatives that ensure improvement in environmental and energy efficiency technologies related to production and transportation of exported and imported goods and services.

Introduction

Fighting climate change remains the essential target of the global economy due to the surge of carbon dioxide (CO₂) emissions and its deleterious impact on the global environment. For example, global CO₂ emissions recorded¹ an unsurpassed level of 36.3 billion tons in 2021 after a 5.2% reduction in global CO₂ emissions in 2020 due to the COVID-19 crunch, which slows the global economy. The substantial increase in CO₂ emissions is alarming, as CO₂ emissions adversely influenced economic wealth, human health, and, to a greater extent, the green environment. For instance, Liu et al. (2022a) argued that global warming arising from CO₂ emissions leads to a reduction in food production and biodiversity and increased in ocean levels and mortality rate. Although sub-Saharan Africa (SSA) is not among the leading emitters of CO₂ emissions, SSA is prone to major harmful effects of CO₂ emissions, which inhibit its economic growth and development. Despite the low contribution of SSA, CO₂ emissions in SSA are increasing, making it a problem of concern and calls for attention given the dangerous effect of CO₂ on the region. For instance, CO₂ emission in SSA increased from 784,540.02 kilotons in 2016 to 823,770.02 kilotons in 2019, indicating a growth rate of 2.75% in 2019 from 1.57% in 2016 (World Bank 2022). Therefore, the greatest need to reduce the soaring trend of SSA CO₂ emissions makes this study worthwhile.

International trade plays a crucial role in the emission of carbon dioxide and other gases via emissions resulting from production and transportation of goods and services, which lead to climate change. Evidence shows that about a quarter of CO₂ emissions is associated with trade flows (Brenton and Chemutai 2021). The indication is that economies largely engaged in international trade are likely to experience higher CO₂ emissions. SSA trade as a share of GDP increased from 45.98% in 2016 to 50.03% in 2020 (World Bank 2022). This trade flow undoubtedly has a share in the increase in CO₂ emissions in SSA, as argued above. Although international trade plays a role in global emissions, its associated benefits, such as reallocation of resources and diffusion of technology, have been argued to enhance economic growth and the environment (Duodu and Baidoo 2020; Wan et al. 2015). Consequently, the environmental effect of international trade is questionable in SSA and beyond, as to whether trade is good or bad to the environment.

On the one hand, scholars (see Shahbaz et al. 2014; Wan et al. 2015) alleged that trade through its benefit of exchange of technology can lead to the adoption of advanced green technologies (such as pollution abatement technologies), which reduces pollution emission and its adverse impact on the climate. As a result, some scholars opine that trade openness reduces environmental pollution by reducing CO₂ emissions (see Muhammad et al. 2020; Iheonu et al. 2021). On the other hand, others debate that trade has a deleterious effect on the environment (Boamah et al. 2017; Du et al. 2020; Duodu et al. 2021). Among the contentions includes the fact that trade increases the intensity of fossil energy consumption via exports of goods and services that require the industrial sector to rely heavily on fossil energy. The consequent effect is higher CO₂ emissions as economies tend to amass trade surplus via exports. Undoubtedly, the conflicting results of trade on environmental pollution can, to some extent, be attributed to methodological weaknesses employed.

However, to some degree, it is also possible that the environmental effect of trade might depend on the trade targets² of economies, though the trade effect on the environment has been established empirically as shown above. However, the most intriguing question that previous studies³ on trade and environment have failed to address (especially within SSA) is whether exports and imports of trade have homogeneous or heterogeneous effects on environmental pollution. This question of concern is vital to address as it helps policymakers to identify which form of trade (exports or imports) should be focused on or targeted to ensure economic growth without deteriorating the environment. Most studies in SSA have focused on trade openness (the combined effect of exports and imports), which did not specifically reveal the effect of exports and imports on the environment (see Tenaw and Beyene 2021; Iheonu et al. 2021; Okelele et al. 2022). Therefore, neglecting the potential effects of exports and imports of trade could lead to inappropriate policies of trade targets that induced environmental quality. For example, importation makes it easy to access technologies that could upgrade the industrial sector from the use of fossil fuel combustion to renewable energy that limits CO₂ emissions. On the other hand, exportation, especially in the case of SSA where most countries export precious natural resources like gold, iron, copper, limestone, diamonds, bauxite, petroleum, and uranium could lead to higher CO₂ emissions through the extraction of such resources for exports (see Adedoyin et al. 2020; Erdogan et al. 2021; Oteng-Abayie et al. 2022a; Oteng-Abayie et al. 2022b). These suggest that trade exports and imports could have a diverse impact on the environment. Therefore, it is worth investigating whether exports and imports of SSA economies have a homogeneous or differential effect on the environment for policy purposes.

In this regard, this study complements previous studies in SSA by assessing the effect of trade on environmental pollution. In doing so, we deviate from previous studies⁴ in SSA as a contribution to knowledge by investigating the total effect of trade, as well as the heterogeneous impact of trade exports and imports on environmental pollution. While trade openness helps to evaluate the total effect of trade on the environment, exports and imports help to assess the disaggregated effect of trade on environmental pollution, which previous studies in SSA have ignored. This helps to assess whether the effect of imports and exports of trade aligns with the total effect of trade, and more specifically, which form of trade (exports or imports) improves environmental sustainability in the SSA region in Africa. Therefore, this study explores the heterogeneous impact of trade on environmental pollution in SSA. Thus, the study minimizes the research gap and makes a substantial contribution to the trade and environmental pollution nexus and the implementation of vital policies. To the authors’ knowledge, we only know of Nwani et al. (2022), who have considered exports and imports in assessing trade effects on the environment in SSA. However, this study suffers from methodological flaws, as the study does not account for a possible endogeneity problem, which may result from a reverse causality between trade and CO₂ emissions. We provide more robust evidence by using the generalized method of moment estimator which controls for such potential endogeneity. Again, this study focused on SSA nations instead of the net-importing countries in SSA considered by Nwani et al. (2022). We considered SSA nations because all nations in SSA engage in trade. Therefore, the environmental effect of international trade is likely to affect all SSA economies but not only the net-importing countries in SSA. Furthermore, unlike Nwani et al. (2022), we expand the data span to 2020 to reflect contemporary changes in trade policies, which may likely affect trade volumes and their effect on the environment. This helps policymakers with the current implications of trade on the environment.

This paper is structured as follows. The next section reviews relevant past studies related to trade and environmental pollution, followed by the third section, which shows the empirical methods adopted for the study. The empirical results and their discussion are presented in the fourth section, whereas the final section concludes the paper with policy implications.

Literature review

This section provides a review of theoretical and empirical studies regarding international trade and environmental pollution.

Theoretical and empirical review

Theoretically, the pollution haven hypothesis (PHH) (Walter and Ugelow, 1979; Baumol et al., 1988) has been used as the basis for international trade and pollution emissions. The PHH asserts that countries that adopt trade liberalization policies with less stringent environmental regulations attract pollution-intensive industries. The hypothesis postulates that advanced economies with stringent environmental regulations require a higher cost of pollution (Ren et al. 2014). As a result, pollution-emitting companies tend to move to countries with trade liberalization and lenient environmental standards or policies because of the lower cost of pollution, particularly in underdeveloped economies. Consequently, environmental quality in many developing countries is compromised due to ineffective environmental regulations. Given the assertion of the PHH, many scholars have focused on the role of international trade on the environment due to the growing globalization and integration of economies in the world. For example, Copeland and Taylor (1994) examined the relationships between environmental degradation and international trade. They observed that developing countries with free trade policies worsened environmental quality while developed countries with free trade and stringent environmental regulations improve their environment. Another theory in explaining the theoretical link between trade and environmental pollution is the scale effect of the trade openness hypothesis. The theory explains that foreign or multinational companies in developing countries through trade intensify energy consumption (particularly fossil energy). Therefore, international trade increases CO₂ emissions by being heavily dependent on energy consumption and natural resources (Duodu et al. 2021).

Given these theoretical concerns, many studies in SSA have validated the PHH and the scale effect hypothesis on trade and pollution emissions. They empirically argued that international trade in SSA increases environmental pollution. For example, Kwakwa and Adu (2015) explored the link between income, energy consumption, and trade openness on pollution emissions in SSA from 1977 to 2012 and observed that trade openness increases CO₂ emissions in SSA. Similarly, Acheampong et al. (2019) employed the generalized method of moment (GMM) to examine the globalization and renewable energy effect on CO₂ emissions in SSA. Their results confirm that of Kwakwa and Adu (2015), that trade openness results in higher CO₂ emissions in SSA. Using the GMM method, Asongu and Odhiambo (2021) investigated the trade and FDI thresholds of CO₂ emissions in SSA and found that trade induces CO₂ emissions. The above adverse impact of trade openness on the SSA environment has also been validated by other recent studies (see Tenaw and Beyene 2021; Nwani et al. 2022) reporting that the CO₂ emissions in SSA are attributed to the trade flows. Given that the above findings are subject to the methodology and data span employed, other studies with different approaches debate that trade openness in SSA enhances environmental quality by reducing CO₂ emissions. For instance, Ali et al. (2016) employed the autoregressive distributed lag (ARDL) method to examine the dynamic effect of urbanization, economic growth, energy consumption, and trade openness on CO₂ emissions from 1971 to 2011. Their study revealed that trade openness reduces CO₂ emissions in Nigeria. Iheonu et al. (2021) also used panel quantile regression in 34 SSA countries to analyze whether economic growth, international trade, and urbanization uphold environmental sustainability. They found that international trade improves environmental sustainability in the SSA region. On a similar argument, Okelele et al. (2022) examine the trade effect on the ecological footprint in SSA, using the feasible generalized least square (FGLS), and observed that trade openness enhances the environment by decreasing the ecological footprint. These studies did not validate the PHH and scale effect hypothesis in SSA.

It is obvious from the above that studies on international trade and the environment in SSA are limited with mixed results and therefore call for further examination. Furthermore, these studies fail to assess the differential effect of trade exports and imports on environmental pollution. Mention can be made to Nwani et al. (2022), which attempt to access the exports and imports effect on the environment in SSA. However, the caveat of the work arises from methodological weaknesses. Nwani et al. (2022) used the method of moment quantile regression (MM-QR), which failed to account for potential endogeneity that cannot be overlooked when working with panel data. Furthermore, the authors used data from 1995 to 2017, which does not reflect contemporary changes in trade policies that can affect trade. As a result, estimates may render policies ineffective. Given these knowledge gaps, we complement the past literature in SSA (including Nwani et al. 2022) by investigating the disaggregated effect of trade on the environment using the two-step generalized method of moment (GMM), which is robust to possible endogeneity. Additionally, we expand the data period to 2020 to determine the current effect of trade, exports, and imports on the environment in SSA.

Regarding studies beyond SSA, a plethora of literature has supported the PHH and the scale effect hypothesis, while others refute such a hypothesis. For example, in China where most economies trade, Ren et al. (2014) used the system-GMM to analyze the association between international trade, FDI, and CO₂ emissions from 2000 to 2010 and found that trade surplus causes higher emissions. Likewise, Boamah et al. (2017) examine the role of international trade on Chinese CO₂ emissions from 1970 to 2014 and reported that China’s trade induces higher CO₂ emissions. Furthermore, Du et al. (2020) in their study of 116 countries employed the fixed-effect method to investigate whether trade promotes CO₂ emission performance from 1986 to 2014. Their study shows that trade increases emissions in the 116 countries. Similarly, Gulistan et al. (2020) examined the relationship among economic growth, energy, trade openness, tourism, and environmental degradation in 112 countries from 1195 to 2017 and found that trade openness induces higher emissions but has mixed results across the subsamples. In the Association of Southeast Asian Nations (ASEAN), Nathaniel and Khan (2020) used the augmented mean group (AMG) to explore the nexus between urbanization, renewable energy, trade, and ecological footprint. Their findings confirm that of Ren et al. (2014) and Boamah et al. (2017), that international trade promotes environmental pollution as trade increases CO₂ emissions. The above evidence that trade induces pollution has also been confirmed in a recent study by Anwar et al. (2022a), stating that trade openness increases CO₂ emissions in seven emerging countries. On the contrary, Dogan and Turkekul (2016), Dogan et al. (2017), and Muhammad et al. (2020) observed that international trade reduces pollution emissions in the USA, OECD, and Belt and Road Initiative (BRI) economies, respectively. With regard to Muhammad et al. (2020), while exports decreased CO₂ emissions in low- and high-income countries, it depletes the environment in middle-income countries. Furthermore, imports decreased CO₂ emissions in middle- and high-income countries but increased CO₂ in low-income countries. This likely points to the fact that SSA exports and imports could have a diverse impact on the environment. Studies on both developed and developing countries have also established the negative impact of trade on CO₂ emissions. In the study by Ibrahim and Ajide (2022) in African countries, they used the system-GMM to evaluate trade facilitation and environmental pollution from 2005 to 2014. Their study shows that trade facilitation reduces CO₂ emissions in Africa. In a similar study, Yazdi and Beygi (2018) further confirm that trade in African countries reduces pollution emissions. Again, Khan et al. (2021) and Ma and Wang (2021) used data from both developed and developing countries to examine the trade effect on carbon emissions from 1980 to 2017 and 1995 to 2014, respectively. They found that international trade reduces environmental pollution. In the next 11 (N11) economies, Nathaniel et al. (2021) examined the nexus between economic growth, energy use, international trade, and ecological footprints from 1990 to 2016 and found the long-run impact of trade to increase ecological footprints. We present in Table 9 (see the Appendix) the abridged literature review of the trade and environmental pollution nexus.

Aside from trade openness, recent studies (see Liu et al. 2022b; Sun et al. 2022; Anwar et al. 2022b; Wen et al. 2022) have shown other determinants (such as renewable energy consumption and economic growth) of CO₂ emissions. For example, the above studies revealed that renewable energy consumption mitigates CO₂ emissions in the seven emerging economies and the top ten polluted countries. However, Liu et al. (2022b) and Sun et al. (2022) further indicated that economic growth increases CO₂ emissions in the same countries. Therefore, it is essential to account for these variables in the study of the trade-environmental pollution nexus in SSA.

Empirical methods

In this section, we describe the data and variables used in this study. We also present the empirical model and the estimation techniques used for the analysis of the study.

Data and variable description

The study relies on balanced panel data spanning 1990–2020 in 33 SSA countries. The study period and the selection of 33 SSA countries are influenced by the availability of data. The variables used for empirical evaluation include environmental pollution, trade, foreign direct investment, renewable energy consumption, economic growth, and industrialization. Following previous literature (Muhammad et al. 2020; Duodu et al. 2021; Zheng et al. 2021), we measured environmental pollution by carbon dioxide emissions (metric tons per capita) and trade by trade openness. However, to account for the heterogeneous effect of trade exports and imports, we further measured trade by exports as a share of GDP and imports as a share of GDP. Foreign direct investment (FDI) and renewable energy consumption were measured by the net inflows of foreign direct investment (share of GDP) and renewable energy consumption (share of total final energy consumption), respectively. Finally, economic growth and industrialization were measured as GDP per capita and industry value added, respectively. It must be emphasized that the variables and their measurement were motivated by previous studies (see Acheampong et al. 2019; Muhammad et al. 2020; Duodu et al. 2021; Zheng et al. 2021) that employed the above measurements as proxies for the variables used in this study. Data for the sample variables were obtained from world development indicators (World Bank 2022). In Table 10 (see the Appendix), we present a brief description of the variables used for the analysis.

In Table 1, we report the descriptive statistics of the variables. The average CO₂ emissions (environmental pollution) in SSA is about 0.97 metric tons. This indicates higher CO₂ emissions in SSA and hence poor environmental quality. Regarding trade (TRD1), we noticed the average trade is about 17.67% of GDP whereas exports (TRD2) and imports (TRD3) are 28.42% and 36.35% of GDP, respectively. The average mean of imports suggests that SSA depends more on imports compared to exports. This is an indication that exports and imports could have a differential impact on the environment. Regarding the other variables, we observed that foreign direct investment, renewable energy consumption, economic growth, and industrialization have an average of 2.82% of GDP, 65.41% of energy consumption, 2,118.74 per capita, and 23.78% of GDP, respectively. Regarding the correlation, we observed that all variables except renewable energy consumption have a positive association with environmental pollution.

Table 1 Descriptive statistics and correlation matrix

Empirical model

Following the empirical model of Iheonu et al. (2021) and Duodu et al. (2022), we augmented the STIRPAT⁵ model by Dietz and Rosa (1994) for the model specification. Thus, we specified environmental pollution (EP) as a function of trade (TRD), foreign direct investment (FDI), renewable energy consumption (REC), economic growth (EG), and industrialization (IND). Therefore, the empirical model to explore the environmental effect of trade is expressed in a dynamic panel Eq. (1).

$$\ln {\textrm{EP}}_{it}={\delta}_0+{\delta}_1\ln {\textrm{EP}}_{it-1}+{\delta}_2\ln {\textrm{TRD}}_{it}+{\delta}_3{\textrm{FDI}}_{it}+{\delta}_4\ln {\textrm{REC}}_{it}+{\delta}_5\ln {\textrm{EG}}_{it}+{\delta}_6\ln {\textrm{IND}}_{it}+{\gamma}_t+{\varphi}_i+{\varepsilon}_{it}$$

(1)

where EP, TRD, FDI, REC, EG, and IND represent environmental pollution, trade (trade openness, exports, and imports), foreign direct investment, renewable energy consumption, economic growth, and industrialization, respectively. i denotes the cross-sectional units (33), and t represents the time dimension (1990–2020). The δ₀ … … … δ₆ are the parameters to be estimated, and the ε_it is the stochastic error term. The γ_t and φ_i denote the fixed effect and individual heterogeneity effect, respectively.

Estimation techniques

The study begins its empirical estimation by performing some preliminary tests such as cross-sectional dependence (CD), unit root, and cointegration tests. These tests are performed to avoid spurious estimations (Pesaran 2007).

Cross-sectional (CD) dependence test

Panel data analysis is most likely to exhibit cross-sectional correlation, leading to biased results (Pesaran 2007). As a result, the CD test remains crucial in panel data analysis. CD is likely to occur when there exist spatial or spillover effects or unobserved factors existing among cross-sectional units. Given that economies are more integrated than ever, the economic policies of one country are likely to influence each other, resulting in dependence between countries. Therefore, it is important to check CD in this study. As a result, we used the Pesaran (2004) CD test to check for cross-sectional dependence. The CD test statistic is given in Eq. (2).

$$\textrm{CD}=\sqrt{\frac{2T}{N\left(N-1\right)}}\left[\sum\nolimits_{i=0}^{N-1}\sum\nolimits_{j=i+1}^N\rho ij\right];\textrm{CD}\sim N\left(0,1\right)$$

(2)

where ρij, N, and T denote the cross-sectional correlation between errors i and j, cross-sectional units, and time dimensions, respectively. In the CD test, the rejection of the null hypothesis implies the presence of cross-sectional dependence.

Panel unit root test

After confirmation of CD, we employed the second-generation unit root tests for stationarity properties of the series used for the study. Specifically, Pesaran (2003) cross-sectional augmented Dickey-Fuller (CADF) and Pesaran (2007) cross-sectional augmented IPS (CIPS) were used for the unit root test. The choice of these second-generation tests is due to the fact that it accommodates or overcomes CD in the presence of CD in the data. However, first-generation unit root tests become invalid because it assumes that there is no CD among cross-sectional units. As postulated by Pesaran (2007), the CADF incorporates the unobserved factors in the model to overcome CD. In the CADF and CIPS tests, rejection of the null hypothesis implies that the series are stationary. Equation (3) gives the regression to ascertain CADF statistic.

$$\Delta {y}_{it}={a}_i+{b}_i{y}_{it-1}+{c}_i{\overline{y}}_{t-1}+\sum\nolimits_{j=0}^s{d}_{ij}\Delta {\overline{y}}_{t-j}+\sum\nolimits_{j=1}^s{\gamma}_{ij}\Delta {\overline{y}}_{t-j}+{\varepsilon}_{it}$$

(3)

where $\overline{y}$ and $\overline{\Delta y}$ represent the cross-sectional outcome variable averages at lagged levels and first difference, respectively. The test statistic obtained from Eq. (3) is utilized to derive the CIPS test statistic expressed in Eq. (4).

$$\textrm{CIPS}=\frac{1}{N}\sum\nolimits_{i=1}^N{\textrm{CADF}}_i$$

(4)

where CADF_i is the t-statistic obtained from Eq. (3).

Cointegration test

To establish the long-run relationship among the sample variables, we used the Westerlund (2007) second-generation cointegration test. We choose the Westerlund test over the first-generation cointegration tests based on its superiority to overcome CD problems and nuisance resulting from endogeneity (Westerlund 2007). Therefore, the Westerlund (2007) test for cointegration is an appropriate test for this study, since the interdependence of economies is likely to result in CD issues in our sample data. This test has the null hypothesis of no long-run relationship against the alternative hypothesis of cointegration. The test proposed four test statistics under the null hypothesis. Two of them are group mean statistics, and the remaining two are panel mean statistics. The group mean test for cointegration for the entire panel, while the panel mean test for at least the existence of cointegration in one cross-sectional unit. Equations (5) and (6) present the group mean and panel mean statistics, respectively.

$${G}_t={N}^{-1}\sum\nolimits_{i=1}^N\frac{{\overline{\alpha}}_i}{\textrm{SE}\left({\overline{\alpha}}_i\right)}\ \textrm{and}\ {G}_a={N}^{-1}\sum\nolimits_{i=1}^N\frac{T{\overline{\alpha}}_i}{{\overline{\alpha}}_i(1)}$$

(5)

$${P}_t=\frac{{\overline{\alpha}}_i}{\textrm{SE}\left({\overline{\alpha}}_i\right)}\ \textrm{and}\ {P}_a=T\overline{\alpha}$$

(6)

where ${\overline{\alpha}}_i(1)$ is the semiparametric kernel estimator of ${\overline{\alpha}}_i$ and $\textrm{SE}\left({\overline{\alpha}}_i\right)$ denote the standard error.

Parameter estimation

After performing the above tests, we utilized the two-step system generalized method of moment (system-GMM) to unveil the overall and heterogeneous impact of trade on the environment. The dynamic model specified in Eq. (1) makes the use of panel estimations such as fixed effect and random effect inappropriate for this study. This is because of possible endogeneity that is likely to exist in the model due to the omission of relevant variables, measurement errors, or reverse causality. However, the system-GMM is capable of overcoming such a potential problem in this study (Blundell and Bond 1998). Furthermore, the system-GMM is applicable for a panel sample of N > T, which is consistent with our study. Therefore, the system-GMM used for this study is suitable. The consistency and efficient estimates of the system-GMM rely on the instruments’ validity and absence of second-order serial correlation [AR(2)]. As a result, we diagnosed the estimates using the Hansen (1982) test of instrument validity and the Arellano-Bond test for second-order serial correlations. Failure to reject the null hypothesis of the Hansen test and Arellano-Bond test implies valid instruments and the absence of AR(2), respectively. The system-GMM specification of Eq. (1) is expressed in Eq. (7).

$$\ln {\textrm{EP}}_{it}-\ln {\textrm{EP}}_{it-1}={\delta}_1\left(\ln {\textrm{EP}}_{it-1}-\ln {\textrm{EP}}_{it-2}\right)+{\delta}^{\prime}\left(\ln {X}_{it}-\ln {X}_{it-1}\right)+\left({\gamma}_t-{\gamma}_{t-1}\right)+\left({\varepsilon}_{it}-{\varepsilon}_{t-1}\right)$$

(7)

where all variables are defined already. X_it represents a vector of control variables as shown in Eq. (1). To access the total effect of trade, as well as the differential or heterogeneous impact of trade exports and imports, Eq. (7) is estimated three times. In the first estimate, Eq. (7) is estimated with trade openness to provide the overall effect of trade on the environment, while in the second and third estimates, Eq. (7) is estimated with exports and imports of trade, respectively. This helps to access whether there exists a homogeneous or heterogeneous impact of trade exports and imports on the environment. Furthermore, it helps to ascertain whether the effect of trade exports and imports aligns with the overall effect of trade. Moreover, the study follows Duodu et al. (2021) and applied Papke and Wooldridge (2005) delta method to ascertain the long-run parameters from the short-run coefficients. This is done to access the level at which international trade in the long run influences the environmental pollution of SSA. Again, since policy interventions are often based on long-run effects, it becomes imperative to access the long-run parameters. The Papke and Wooldridge (2005) delta method is specified in Eq. (8).

$${\delta}_k^{\ast }=\frac{\delta_k}{\left(1-\lambda \right)}$$

(8)

where ${\delta}_k^{\ast }$ are the long-run parameters estimated from the short-run parameters (δ_k) in Eq. (7). λ is the coefficient of the lagged-dependent variable in Eq. (7).

For robustness and consistency of the system-GMM estimates, we used panel-corrected standard error (PCSE) and dynamic common correlated effects (DCCE) estimators. The PCSE and DCCE estimates are robust to cross-sectional dependence, heteroskedasticity, and serial correlation (Reed and Ye 2011; Chudik and Pesaran 2015). Therefore, using PCSE and DCCE as robustness is ideal.

Causality test

Given the vital role that causal relationship plays in policy implementation, it is imperative to ascertain the causal relationship among the variables. As a result, we applied the Dumitrescu and Hurlin (DH) (2012) causality test to determine the causal relationship between the sample variables and environmental pollution. More specifically, we provide the causal association among trade openness, exports, imports, and environmental pollution. As documented, the D-H causality test controls for heterogeneity in slope parameters and overcomes cross-sectional dependence (Dumitrescu and Hurlin 2012). Therefore, the advantages of the D-H causality test make it appropriate and efficient for causal relationships compared to the usual Granger causality test, which assumes slope homogeneity across cross-sectional units. The D-H causality test is obtained from Eq. (9).

$${y}_{it}={\tau}_i+\sum\nolimits_{i=1}^p{\vartheta}_i^{(p)}{y}_{it-n}+\sum\nolimits_{i=1}^p{\varphi}_i^{(p)}{x}_{it-n}+{\varepsilon}_{it}$$

(9)

where ${\vartheta}_i^{(p)}$ and ${\varphi}_i^{(p)}$ denote the autoregressive and regression parameters, respectively. The constant (τ_i) and the coefficient ${\varphi}_i^{(p)}=\left({\varphi}_i^{(1)}\dots \dots \dots {\varphi}_i^{(p)}\right)$ are fixed.

Empirical discussion

This section presents the findings of the empirical estimations. The study starts with a discussion of the preliminary test results and proceeds to analyze the results from the system-GMM and robustness checks. Finally, the results of the D-H causality test are discussed.

Cross-sectional dependence (CD) test results

Table 2 shows the cross-sectional dependence results. The results reveal that there is a cross-sectional dependency in all variables except trade openness (TRD1) and industrialization. This is because the p value of trade openness and industrialization indicates a non-rejection of the null hypothesis of no CD. Given that almost all variables exhibit CD, we conclude that there exists CD among the sample panels. The presence of CD suggests that policies, including trade and environmental policies, in one economy, could influence another economy. Therefore, policymakers in all economies need to consider other economies when formulating policies.

Table 2 Cross-sectional dependence test

Stationarity and cointegration test results

The presence of CD nullifies the cross-sectional independence assumption of the first-generation unit root and cointegration tests. As a result, we tested for series stationarity and the long-run relationship using second-generation tests (CADF, CIPS, and Westerlund), which overcome cross-sectional dependency issues (Pesaran 2007; Westerlund 2007). The unit root and cointegration results are reported in Tables 3 and 4, respectively. From Table 3, both CIPS and CADF show that all the variables were not stationary at the levels. Specifically, while the CIPS shows that exports, imports, and FDI are stationary at the levels, the CADF reveals that only imports and FDI are stationary at the levels. However, after taking the first difference, all the series in both tests were stable or stationary at the first difference. Therefore, the series used are stationary at the first difference.

Table 3 Panel unit root test

Table 4 Westerlund cointegration test results

Regarding the cointegration results, we noticed in Table 4 that there is a long-run relationship between environmental pollution, international trade, FDI, renewable energy, economic growth, and industrialization. This is because the significance levels of the 3 test statistics (G_t, P_t, and P_a) in all models indicate the existence of cointegration between the variables. Therefore, estimating the long-run effects in addition to the short-run impact is justified. The presence of stationarity and cointegration avoids spurious estimates. Hence, the study continues with its estimations.

Trade effect on environmental pollution (system-GMM)

We report both the short- and the long-run results of Eqs. (7) and (8) in Tables 5 and 6, respectively. In each of the analyses, we estimated 3 models. Models (1, 2, and 3) are the estimations with trade openness, exports, and imports as the variable of interest, respectively. Given that an increase in CO₂ emissions has a detrimental effect on the environment, a negative sign in this study implies a reduction in environmental pollution, while a positive sign suggests an increase in environmental pollution.

Table 5 Effect of trade on environmental pollution

Table 6 Effect of trade on environmental pollution (long-run estimates)

Beginning with the short-run results, we observe in Table 5 that previous environmental pollution (CO₂ emissions) does not exhibit convergence in all models. Specifically, the results indicate that previous CO₂ emissions in SSA induce an increase in environmental pollution of approximately 0.87%, 0.84%, and 0.85% in models (1, 2, and 3) respectively. This outcome suggests the need to intensify environmental policies in SSA to improve environmental quality. Studies (Ren et al. 2014; Duodu et al. 2021) reported similar outcomes in China and SSA that previous CO₂ emissions positively influence current CO₂ emissions. Turning to the variable of interest (trade), we observed that total trade (trade openness) in model 1 lowers environmental pollution. The coefficient indicates that a 1% increase in overall trade is associated with about 0.10% reduction in CO₂ emissions, holding other covariates constant. The implication is that participation in international trade has the capacity to decrease the environmental pollution. This result could be attributed to the fact that international trade induces diffusion of green technologies, which developing countries such as SSA adopt to transform their economic structures (manufacturing, industrial, and services sectors) to control the harmful impact of CO₂ emissions on the environment. Indeed, the results suggest that engaging in international trade provides access to essential technologies, which support the adaptation and mitigation of the changing climate and its consequences. This result supports previous studies in SSA and beyond (Ali et al. 2016; Iheonu et al. 2021; Ma and Wang 2021; Okelele et al. 2022) reporting that international trade minimizes environmental pollution.

To assess whether exports and imports of trade have a homogeneous or heterogeneous effect on the environment, we estimated models 2 and 3. The results in Table 5 reveal that both exports and imports lessen environmental pollution by decreasing CO₂ emissions in SSA. The results suggest that exports and imports in SSA have a homogeneous effect on the environment, which aligns with the overall effect of international trade.

The results imply that a 1% increase in SSA exports and imports causes a decline in environmental pollution by about 0.07% and 0.08%, in models 2 and 3, respectively. The effect of exports on the environment could be that SSA does not engage in exports of high carbon-intensive products, which may lead to higher emissions. Although SSA enhances its economic growth through exports of goods and services to accumulate a trade surplus, they are cautious about the environment. As a result, they tend to implement trade policies (including exports policies) that favor environmental quality in SSA. The environmental effect of imports on another hand could be attributed to the fact that SSA countries import goods and services that are environmentally friendly and do not harm the environment. For example, SSA economies may imports goods and services that meet environmental requirements implemented by them. Consequently, exporters to SSA adopt the use of cleaner production processes and technologies, which limit the environmental effect of exporting to SSA. Although both exports and imports have the same effect on the environment, we observed that imports lower environmental pollution more than exports. The indication is that developing countries like SSA capitalize on international trade to access modern environmental technologies from developed economies, which improves the environment. The findings contradict Nwani et al. (2022), who argue that imports and exports increase CO₂ emissions. However, our results support Muhammad et al. (2020) reporting that exports reduce environmental pollution. Ren et al. (2014) and Boamah et al. (2017) in China find an insignificant impact of exports and imports on the environment, which also contradicts our result in SSA.

Regarding the control variables, the results reveal that FDI, renewable energy consumption (significant in models 2 and 3), and economic growth increase environmental pollution in SSA in all models. However, the environmental effect of industrialization was found to reduce environmental pollution in all models but only significant in model 1. In particular, the results show that an additional increase in FDI increases environmental pollution in models (1, 2 and 3) by about 1.45%, 1.59%, and 1.64%, respectively, with other variables, held constant. This outcome is consistent with the pollution haven hypothesis, which claims that environmental policies in general are less stringent in developing economies (such as SSA). As a result, foreign investors move from countries with stringent environmental policies to those with lenient policies, and this tends to induce higher pollution. Ren et al. (2014) and Muhammad et al. (2020) reported a similar outcome that FDI results in pollution emissions in China and BRI countries, respectively.

With respect to renewable energy consumption, the results indicate that a 1% increase in renewable energy consumption increases environmental pollution by approximately 0.29% and 0.20% in models (1 and 2), respectively. The implication is that the SSA economies are heavily dependent on fossil energy consumption compared to renewable energy, leading to low consumption of renewable energy in the subregion. Consequently, the positive effect of renewable energy does not offset the adverse effect arising from the consumption of fossil energy. Therefore, the expected positive impact of renewable energy on the environment does not manifest in SSA due to the region’s low consumption of renewable energy. This finding contradicts the study by Nathaniel and Khan (2020) and Iheonu et al. (2021) arguing that renewable energy consumption ensures a clean environment.

Furthermore, Table 5 reveals that a 1% increase in economic growth promotes environmental pollution in SSA by about 0.29%, 0.35%, and 0.29% in models (1, 2, and 3), respectively, maintaining other variables constant. The indication is that the economic growth initiatives in SSA are not environmentally friendly to promote environmental quality. This is due to the fact that the quest for SSA economies to achieve growth and development makes them to instigate policies that stimulate growth but compromise the quality of the environment. Therefore, it is essential that policymakers in the SSA focus on policies that stimulate green economic growth. Du et al. (2020) and Duodu et al. (2021) reported similar findings that economic growth is among the factors that contribute to CO₂ emissions. With regard to industrialization, the coefficient suggests that a 1% increase in industrialization lowers environmental pollution by about 0.12% keeping other covariates constant. The result implies that industrialization in SSA with improvement in energy efficiency will lessen the environmental pollution. The findings in SSA are contrary to Zheng et al.’s (2021) outcome, who argues that industrialization induces higher CO₂ emissions in China.

Turning to the long-run analysis, we observed from Table 6 that the estimated long-run results statistically do not differ from the short-run results in Table 5. However, it must be emphasized that the long-run impacts (in terms of the magnitudes) are larger than the short-run case. Specifically, the results suggest that a 1% increase in total trade (trade openness) is associated with an impact of 0.79% reduction in CO₂ emissions, keeping other covariates constant. This outcome suggests that the long-run impact of international trade on the SSA environment has a much greater impact on minimizing its environmental pollution compared to the short-run impact. Again, the long-run results further confirm the homogeneous effect of exports and imports, as we observed that a 1% increase in exports (imports) lowers environmental pollution in SSA by approximately 0.45% (0.58%), while other variables remain constant. The implication is that regardless of whether SSA economies target more exports or imports, the environmental quality will be improved. This is because international trade brings out green technologies that make local production processes more efficient by reducing the use of inputs that are environmentally harmful. The long-run results of international trade are again consistent with the findings of Ali et al. (2016), Iheonu et al. (2021), Ma and Wang (2021), and Okelele et al. (2022).

For the control variables, the long-run results revealed a similar conclusion as in the short run (Table 5). We found that FDI, renewable energy consumption (significant in model 2), and economic growth encourage environmental pollution in SSA in all models. Nevertheless, the environmental effect of industrialization was observed to lessen environmental pollution in all models but significant in model 1 only. The implications of these long-run results remain the same as in the short-run analysis. However, it must be stressed that these effects were slightly greater than in the short-run case, as shown in Table 6.

In Table 5, we noticed that our estimated results are robust to second-order serial correlation and instrument validity. This is because the p values of AR2 (0.445, 0.448, and 0.423) and the Hansen test (0.680, 0.675, and 0.668) indicate nonrejection of the null hypothesis of the absence of second-order serial correlation and instrument validity, respectively. Furthermore, the p values of the Jarque-Bera (0.5378, 0.1753, and 0.2262) and Shapiro-Wilk (0.4392, 0.5265, and 0.2154) suggest that the series for the estimated models are normally distributed. Given the above, we conclude that the estimated results are robust and consistent.

Robustness results (PCSE and DCCE estimates)

Given that the total effect of trade on environmental pollution aligns with that of trade exports and imports as shown in Tables 5 and 6, we, therefore, provide a robustness to the total effect of trade. We do so because we observed that imports and exports have no differential impact on the environment, and the effects are consistent with that of total trade as revealed in the system-GMM results. The results of PCSE and DCCE are presented in Table 7. We observed that the total trade effect on environmental pollution in both estimators is not different from that of the system-GMM (in both the short and long run) in terms of signs. In fact, the PCSE and DCCE results suggest that for a 1% increase in international trade in SSA, a 0.01% reduction in environmental pollution would be achieved with other variables kept constant. This result corroborates the fact that international trade has the potential to diffuse green technologies to minimize environmental pollution. Although the total trade effect in both estimators is statistically insignificant, they validate the overall effect of trade (in terms of signs) as observed from the system-GMM estimates.

Table 7 Robustness results

D-H causality results

The causality results are shown in Table 8. Focusing on the variables of interest, we observed the existence of a bidirectional relationship between total trade and CO₂ emissions (environmental pollution). However, we noticed a unidirectional causality from CO₂ emissions to trade exports and imports. The results seem plausible as higher CO₂ emissions make economies import green technologies to lower the emissions level. Also, higher CO₂ emissions cause net-exporting economies to improve in energy efficiency associated with the production and transportation of exported goods and services. This result is consistent with Hossain (2011) who reports that trade causes CO₂ emissions in newly industrialized countries. Regarding the other variables, we found that there is bidirectional causality between FDI, renewable energy consumption, economic growth, industrialization, and CO₂ emissions. The conclusion from these results is that each variable (FDI, renewable energy consumption, economic growth, and industrialization) is a possible determinant of CO₂ emissions in SSA.

Table 8 D-H causality test results

Conclusion and policy implications

Foreign trade can induce economic growth and development. However, international trade has the potential to affect global greenhouse gas emissions in several ways. Therefore, the total effect of trade on the environment is complex to determine. As a result, this study has explored the environmental effect of international trade in 33 SSA countries. The study further evaluates whether there is a homogeneous or heterogeneous effect of exports and imports on environmental pollution. We utilized balanced panel data from 1990–2020 for the investigation. The results of the system-GMM revealed that the overall effect of trade lowers environmental pollution in both the short and long run. The results again show that SSA exports and imports also minimize environmental pollution in both the short and long run, which is consistent with the effect of total trade. Furthermore, we found that FDI, renewable energy consumption, and economic growth increase environmental pollution in SSA. However, industrialization has been shown to reduce environmental pollution. Regarding the D-H causality results, the study observed that total trade and environmental pollution (CO₂ emissions) have bidirectional causality whereas exports and imports tend to have a unidirectional causality running from CO₂ emissions to exports and imports. Based on the results, the study concludes that international trade is among the factors that will induce environmental quality or sustainability in SSA. We further establish that trade exports and imports in SSA do not have a differential impact on environmental pollution.

Regarding policy implications, the finding implies that engaging in international trade provides an avenue to access green technologies and technological innovations that ensures a considerable reduction in pollution emissions associated with the production of exported and imported goods. Therefore, we suggest, based on the effect of international trade (trade openness, exports, and imports), that SSA countries should move to green consumption and production by implementing policies to accelerate trade in green environmental goods and services. For example, policies that ensure a reduction in import tariffs on green environmental goods and services will facilitate the trade of environmental goods and services that improve environmental quality. Additionally, we suggest that SSA countries implement initiatives that ensure the improvement of energy and environmental efficiency technologies associated with the production and transportation of exported and imported goods. For example, implementing policies to ensure carbon-efficient technologies for the production process in SSA have the tendency to reduce CO₂ emissions related to trade. Furthermore, the SSA countries should invest in renewable fuel technologies (such as solar fuels, e-fuels, and biofuels) associated with the production of exported goods and services. Doing this will ensure that local firms in exporting countries in SSA adopt renewable energies that promote environmental sustainability. In all, implementing the above suggestions will ensure that international trade diffuses green technologies that reduce environmental pollution by improving carbon efficiency.

Recently, institutions on the environment have been shown to enhance the sustainability of the environment. However, this study did not consider the role of institutional quality in the environmental effect of trade. As a result, we recommend future studies to complement the present study by examining how the quality of institutions (specifically institutions toward environmental sustainability) influence the effect of international trade (total trade, exports, and imports) on environmental pollution in SSA.

Appendix

Table 9 Abridged literature review

Table 10 Variable(s) description