Does foreign capital flow into ‘greener pastures’? Exploring the potential of FDI in mitigating carbon emissions in African states

ABSTRACT

Our study provides fresh insights into FDI's impact on carbon emissions across 51 African nations. Analysing data from 1990 to 2021, we use ARDL estimates and wavelet coherence analysis. While ARDL shows insignificant cointegration for most countries, wavelet coherence reveals: (i) 17 countries with a positive FDI-emissions link, (ii) 16 countries with a negative association, and (iii) 18 countries with ‘sign switching’ nonlinear dynamics. With no clear regional patterns, our main policy implication stresses the need for aligning regional efforts, like AfCFTA, with tailored country-specific strategies to guide capital flows towards cleaner industries. We advocate for industry-level FDI data in Africa to pinpoint sectors needing monitoring for greener production. On the global policymaking level, we assert that FDI cannot significantly address climate justice for most African countries.

KEYWORDS:

• Pollution Haven hypothesis
• Pollution Halo hypothesis
• Foreign direct investment
• Carbon emissions
• ARDL
• wavelet coherence
• Africa

1. Introduction

The purpose of the series of Conference of Parties (CoP) meetings, convened biennially since 1997, has been to address the lingering issue of climate change, which threatens the habitability of the Earth for both humans and other living organisms. A re-occurring theme within the most recent CoP meetings has been the role which foreign direct investment (FDI) can play in simultaneously fostering ‘clean growth’ in all countries and achieving climate justice for less developed ‘Annex II’ countries (Doku et al., 2021). However, whilst there exists a general consensus that places FDI as a modern engine of economic growth by means of promoting job creation, improving domestic wages, productivity and knowledge spillovers to host counties; export diversification and introduction of new industries (Bello et al. 2023; Borga et al. 2022; Ofori and Asongu 2022), there is less agreement on the effects of FDI on the environment. On one hand, an optimistic view of FDI proposes that multinational corporations (MNCs) may spread cleaner environmental technologies or practices to the host countries while on the other hand, a pessimistic view suggests that industrialised economies take advantage of lax environmental laws to pass through their emissions by conducting their operations in less developed countries (Gu and Hale 2023).

In the case of Africa, the FDI-emissions situation is paradoxical. Firstly, despite exhibiting the highest global investment returns, African nations historically attract disproportionately low FDI inflows. For instance, Adegboye and Okorie (2023) illustrate a 12% return on investment for SSA, compared to the 7% global average. However, the 2022 World Investment Report reveals a mere 5% share of global FDI flows, despite the region's record-breaking $83 billion FDI influx post-pandemic in 2021. Notably, this contradicts international economic theory, which posits that capital-scarce, labor-abundant economies like Africa should attract capital from capital-intensive counterparts via arbitrage opportunities (Adegboye and Okorie 2023; Alfaro, Kalemli-Ozcan, and Volosovych 2008; Aluko and Ibrahim 2019; Azzimonti 2018; Lucas 1990). Secondly, despite historically minimal contributions to global greenhouse gas (GHG) emissions, Africa bears the brunt of climate change impacts. The 2023 IPCC report underscores Africa's heightened vulnerability, with climate change exerting a more severe negative impact on the continent's GDP compared to other regions (IPCC 2023). Africa's extensive coastlines and desert expanses amplify risks; by making the continent more susceptible to rapid sea-level rise and exacerbated desertification. Considering the continent's existing food system and security vulnerabilities, intensified by COVID-19 and the ongoing Russian-Ukraine conflict, concerns arise that unaddressed climate change could trigger famine-related conflicts and instability (Koubi 2019).

At the same time, the African Continental Free Trade Area (AfCFTA), which is a resuscitation of previous failed/abandoned attempts to form a unified economic region (Echandi, Maliszewska, and Steenbergen 2022; Fofack 2020), is hailed as a game-changer for boosting cross-border investment by eliminating tariff and non-tariff barriers and replacing existing bilateral and regional trade deals with a unified, single market (Echandi, Maliszewska, and Steenbergen 2022; Fofack 2020). A recent World Bank report estimates that Africa’s FDI flows could increase between 111% and 150% under the AfCFTA agreement (Echandi, Maliszewska, and Steenbergen 2022). These projections are most welcome by African policymakers as the increased FDI resulting from AfCFTA activities will reduce dependency on climate donor finance, which has been criticised as currently insufficient to address the mitigation and adaption needs of African countries (Doku et al., 2021; Ngwenya and Simatele 2020).

Nonetheless, there is a widely held academic consensus that Africa is marred by corruption, weak institutions, and political instability, factors that have been linked to increased emissions from FDI (Chang 2015; Khan et al., 2023; Xaisongkham and Liu 2023). This raises concerns about the compatibility of AfCFTA's goals with the climate change SDG. However, a bulk majority of the existing literature comes to this conclusion based on panel-based studies which fail to account for country-level heterogeneity. For country-specific studies, the evidence is more ambiguous and the coverage of countries is less extensive. Further compounding these empirical gaps is the issue of nonlinearities, which from a country-specific perspective has been solely examined in the papers of Odubgesan and Adebayo (2020) and Udemba and Yalcintas (2021) for Nigeria and Algeria, respectively. These authors use nonlinear ARDL models which capture cyclical asymmetries related to business cycles and find that FDI harms the environment only during expansionary periods in Algeria whereas FDI lowers emissions in all business cycles for Nigeria. Besides these cyclical asymmetries, the literature identifies other forms of location asymmetries such as ‘thresholds effects’ which imply that beyond a certain level of institutional quality (Chang 2015), FDI (Xie, Wang, and Cong 2020) and income (Wang, Yang, and Wang 2023), the sign of the FDI-emissions relationship switches. There are also quantile asymmetries which imply that the FDI-emissions relationship varies across distributional quantiles of carbon emissions hence accounting for outliers in the data (Asiedu 2021; Gyamfi et al. 2021; Halliru, Loganathan, and Hassan 2021). Notably, standard econometric techniques can at best accommodate for one type of nonlinearity and hence are not equipped to capture all dimensions of the FDI–emissions relationship under a singular comprehensive framework.

We re-examine the FDI-emissions link across 51 African nations using complex wavelet transforms. Wavelets, tools for signal-to-image processing, convert two-dimensional time series into a three-dimensional time–frequency spectrum that depicts intensity, frequency, and time-position dynamics. Wavelet coherence analysis further examines time series synchronisation in a scale-based manner, revealing co-movement in the time–frequency domain. A unique facet of multiresolution analysis is its capacity to capture the ‘sign and strength’ of co-movement across varying time periods and frequencies, thereby encompassing diverse forms of ‘location’, ‘cyclical’, and ‘time-varying’ asymmetries within a unified framework. Additionally, these methods adeptly handle non-stationary data, remain insensitive to chosen time windows, and circumvent potential ‘regression errors’, distinguishing them from conventional econometric tools, which depend on selected time spans and contain regression errors.

To demonstrate the usefulness of these tools against traditional econometric techniques, we compare the findings with those obtained from conventional ARDL models. While ARDL models suggest an insignificantly correlated FDI-emissions association across most countries, our wavelet coherence analysis reveals a distinct pattern: 17 nations exhibit a consistently positive FDI-emissions link, 16 nations display a consistent negative relationship, and 18 nations demonstrate a non-linear connection characterised by sign fluctuations across diverse frequencies. Additionally, our findings challenge prevalent explanations for heterogeneity in FDI-emissions connections, such as income disparities (Acheampong 2019; Adeel-Farooq, Riaz, and Ali 2021; Doytch and Uctum 2016; Hoffman et al. 2005; Marques and Caetano 2020; Shahbaz et al. 2015; Shao 2017), institutional quality (Bouzahzah, 2022), FDI levels (Abdul-Mumuni, Amoh, and Mensah 2022), resource abundance (Gyamfi et al. 2021), and emissions levels (Abdul-Mumuni, Amoh, and Mensah 2022), which do not align with our African dataset. Instead, we conclude that country-level heterogeneities in FDI-emissions interactions may stem from distinct distributions of FDI inflows into ‘dirty’ and ‘green’ industries across individual countries.

Our study makes significant contributions to the existing literature in two main ways. Firstly, it pioneers the application of wavelet coherence tools to examine the relationship between Foreign Direct Investment (FDI) and emissions for African countries. While previous research, such as the works of Arain et al. (2020) and Chishti (2023), has utilised wavelet coherence in exploring the FDI-emissions correlation in specific contexts like China and Pakistan, our study broadens this scope by conducting a multi-country investigation across Africa. Unlike previous African studies that took a more abstract approach, we address various time–frequency dynamics in the FDI-emissions relationship. Secondly, we extend the depth of analysis beyond single-country studies for African nations. This is crucial given that most existing studies on the FDI–emissions relationship adopt a panel study approach, grouping African countries with others internationally and providing a singular regression estimate. Moreover, previous country-specific studies have been limited in their coverage, collectively examining only 22 African countries. In contrast, our research focuses on 51 African countries, utilising more recent data. The selection of these countries is motivated by their association with the newly established African Continental Free Trade Area (AfCFTA).

Our study holds relevance for policymaking on both domestic and global scales. By discerning which specific African countries can leverage FDI to enhance their environmental outcomes and which cannot, we provide valuable insights. This distinction is particularly significant in light of the anticipated surge in cross-border capital flows following the implementation of AfCFTA. Such insights can inform global policymakers about the suitability of FDI as a tool for promoting climate justice, especially for African countries that contribute minimally to global emissions yet bear the brunt of their consequences (Doku, Ncwadi, and Phiri 2021; Doku and Phiri 2022). Additionally, our findings are pertinent for domestic policymakers, offering them the opportunity to assess the effectiveness of existing laws and regulations in steering FDI towards environmentally sustainable practices or identifying the need for more stringent measures.

Overall, our study contributes novel insights into the FDI-emissions relationship in Africa, utilising advanced analytical tools and extending the analysis to a wider range of countries. These findings offer valuable guidance for policymakers at both domestic and international levels, aiding in the pursuit of environmentally sustainable economic development strategies.

The remainder of our study is organised as follows. The following section presents the literature review. The methods are outlined in the third section. The fourth section presents the data description and empirical results whilst the fifth section concludes the study.

2. Literature review

There are at least three theoretical conjectures which encapsulate the FDI-emissions relationship. Firstly, is the pollution halo hypothesis which suggests that FDI can positively impact the host economy’s environment; as they come with the transfer of business knowledge, management practices, cleaner technology and standards that use efficient energy sources and lower carbon emissions (Copeland and Taylor 1994; Khan, Rana, and Ghardallou 2023). Secondly, the pollution haven hypothesis asserts that ‘dirty’ firms, which are mobile enough to gain a comparative advantage in pollution-intensive production, migrate from economies with stricter environmental rules to countries with lax environmental laws, regulations and rules to the detriment of environmental quality of the host economy (Cole 2004; Dou and Han 2019). Lastly, there is a hybrid hypothesis which follows the dictates of the Environmental Kuznets Curve (EKC) of Grossman and Krueger (1995) and suggests that during the early stages of development when economies are heavily dependent on dirty energy sources for economic growth, increased FDI will contribute to harming the environment. However, after reaching a certain level of development, economies begin to increasingly become more environment aware/conscious and introduce stringent laws which force multinational firms to adopt eco-friendly technologies and re-direct FDI flows towards green projects, hence reducing emissions (Khan, Rana, and Ghardallou 2023). Reverse nonlinear dynamics can also occur in which poor countries who are initially reliant on low-emitting labour-intensive production sectors eventually reach a level of development in which they begin to shift towards and specialise in immobile energy-intensive sectors resulting in ‘dirty’ growth (Pazienza 2019).

Empirically, studies have sought to examine these theoretical propositions by regressing a measure of FDI on carbon emissions alongside a set of conditioning variables and estimating these regressions using conventional econometric methods. By examining the sign on the coefficient CO₂ emissions variable, researchers are able to test the pollution haven versus halo pollution hypotheses with the former (latter) being verified by a positive (negative) coefficient estimate (see Demena and Afesorgbor 2020 for a meta-study and Al-Nimer et al. 2022). Testing the hybrid hypothesis is commonly achieved by adding square FDI term in the linear regressions or using more formal nonlinear econometric models such as TAR and STR models for threshold effects (Chang 2015; Wang, Yang, and Wang 2023; Xie, Wang, and Cong 2020), the NARDL model for cyclical asymmetries (Abdul-Mumuni, Amoh, and Mensah 2022; Odubgesan and Adebayo 2020; Udemba and Yalcintas 2021) or the quantile regression model for location asymmetries (Asiedu 2021; Gyamfi et al. 2021; Halliru, Loganathan, and Hassan 2021; Sarokdie and Strezov 2019).

In reviewing the African-related literature, we conduct an extensive search on Scopus and Web of Science (WoS) and find a total of 42 related articles. Table 1 summarises the scope (sample size and period cover), methods and findings (sign on the relationship). For convenience sake, we group the studies into three subgroups mainly, international panel studies inclusive of African countries, panel studies exclusively using African countries and country-specific evidence on individual African countries.

Table 1. Summary of Africa-related studies.

Author	Scope	Methods	Results
Panel A: international panel studies (24)
Talukar and Meisner (1991)	44 developing countries: 1987–1995	GMM	−ve
Grimes and Kento (2003)	66 countries: 1980–1996	POLS	+ve
Hoffman et al. (2005)	112 countries: unbalanced panel 1990–2000	Panel causality tests	FDI→CO2 in middle-income countries CO2→FDI in low-income countries No causality in high-income countries
Chakraborty and Mukherjee (2013)	181 countries: 1980–2009	FGLS	+ve
Omri, Nguyen, and Rault (2014)	54 counties: 1990–2011	GMM	+ve for full sample Insig for Europe and Asia +ve for Latin America and Africa
Chang (2015)	65 countries	3 regime PTAR with corruption, rule of law and political stability thresholds	+ve at low regime Insig at middle regime −ve at high regime
Neequaye and Oladi (2015)	27 developing countries	POLS	+ve
Shahbaz et al. (2015)	99 countries	FMOLS	−ve for full and high-income countries +ve for middle and low-income countries
Doytch and Uctum (2016)	132 countries: 1984–2011	GMM	+ve effect for full sample and low-income countries Insig for middle and high-income countries
Shao (2017)	188 countries	GMM	−ve for full sample, HIC, MIC and LIC
Kastratovic (2019)	63 developing countries: 2000–2014	GMM	+ve
Sarokdie and Strezov (2019)	BRICS	Driscoll-Kraay regressions and panel quantile regressions	N-shaped relationship at all quantiles
Essandoh, Islam, and Kakanika (2020)	52 countries: 1991–2014	PMG-ARDL	Confirmed PHH
Marques and Caetano (2020)	21 high- and middle-income countries: 2001–2017	ARDL	−ve effect in HIC and + ve effect in MIC
Xie, Wang, and Cong (2020)	E11 countries; 2005–2014	PSTR	V- and W-shaped relationship
Adeel-Farooq, Riaz, and Ali (2021)	76 countries: 2002–2012	PMG	−ve (insig) for LI and MI (HI) for FDI from developing countries +ve (insig) for LI (MI and HI) for FDI from developed countries
Benzerrouk, Abid, and Sekrafi (2021)	31 developed and 100 developing countries: 1980–2016	GMM	−ve developed countries Insig developing countries
Huang et al. (2021)	G20: 1996–2018	FGLS	+ve
Singhania and Saini (2021)	21 Developed and developing countries: 1990–2016	GMM	−ve for full sample insig for developed countries +ve for developing countries
Xu, Huang, and An (2021)	42 countries	RE and FE	Insign
Dagar et al. (2022)	80 countries	PMG	+ve
Khan, Rana, and Ghardallou (2023)	108 developing countries; 2000–2016	P-veCM	Above a certain threshold, FDI reduces emissions
Wang, Yang, and Wang (2023)	67 countries: 1990–2019	PTAR	+ve (−ve) below (above) threshold of $542
Xaisongkham and Liu (2023)	115 developing countries: 2002–2006	GMM	+ve
Panel B: African panel studies (15)
Bokpin (2017)	52 African countries: 1990–2013	Panel FE	+ve
Ojewumi and Akinlo (2017)	33 SSA countries: 1980–2013	PVAR and PVECM	+ve
Acheampong (2019)	46 SSA countries: 1980–2015	FE and RE	−ve for high-income countries Insign for full sample and lower-income countries
Opoku, Adams, and Aluko (2021)	22 SSA countries: 1995–2014	GMM	−ve
Opoku and Boachie (2020)	36 African counties: 1980–2014	PMG	+ve
Tenaw (2020)	20 African countries; 1990–2017	CS-ARDL	Insig
Asiedu (2021)	20 African countries: 1995–2016	FMOLS, AMG, Quantile Regression causality tests	+ve from Q50 to Q95 insign otherwise Bidirectional causality
Gyamfi et al. (2021)	15 oil-producing and 28 non-oil-producing SSA countries: 1990–2016	OLS and quantile regressions and causality tests	−ve (+ve) at 5th (50^th to 95th) quantile and FDI->CO2
Halliru, Loganathan, and Hassan (2021)	6 ECOWAS countries: 1970–2017	FMOLS, quantile regressions	+ve at all quantiles expect 0.95
Tawiah, Zakari, and Khan (2021)	50 Africa: 1992–2014	PMG	−ve
Bouzahzah (2022)	40 African countries: 1988–2016	PARDL	+ve for full sample, +ve (−ve) for lower (high) institutional quality
Adjei-Mantey and Adams (2023)	29 SSA countries: 2006–2015	GMM	Insig.
Abdul-Mumuni, Amoh, and Mensah (2022)	41 SSA countries; 1996–2018	P-NARDL	Symmetry: +ve Asymmetry: +ve for increasing FDI, −ve for decreasing FDI
Boamah et al. (2023)	41 SSA countries: 2005–2019	FE and RE	Both −ve and + ve effect
Aminu, Clifton, and Mahe (2023)	11 SSA countries: 1995–2019	FMOLS	Insig
Panel C: African country-specific studies (10)
Kivyiro and Arminen (2014)	6 SSA countries: 1971–2009	ARDL	Insig for Congo, Zambia +ve for Kenya, Zimbabwe −ve for DRC, SA
Keho (2015)	12 ECOWAS countries: 1970–2010	ARDL	−ve for Cote d’Ivoire, Ghana, Mali, Nigeria +ve for Gambia, Liberia, Niger insig. For Benin, Burkina Faso, Senegal, Sierre Leone, Togo
Hakimi and Hamdi (2016)	Tunisia and Morocco: 1971–2013	VECM and PVECM	+ve
Solarin et al. (2017)	Ghana	ARDL	+ve
Solarin and Al-Mulali (2018)	20 Countries: 1990–2015	CCEMG	Insig for Egypt −ve for Nigeria +ve for South Africa
Mahmood et al. (2019)	Egypt: 1990–2014	ARDL	−ve long-run and short-run
Odubgesan and Adebayo (2020)	Nigeria:1986-2016	ARDL and NARDL	−ve for ARDL +ve in both partitions for NARDL
Zubair, Samad, and Dankumo (2020)	Nigeria: 1980–2018	ARDL and VAR causality	−ve and FDI→CO2
Awodumi (2021)	8 ECOWAS countries: 1980–2016	ARDL	Insig for Benin, Mali, Senegal, Sierra Leone, Togo −ve for Burkina Faso, Gambia, Nigeria
Udemba and Yalcintas (2021)	Algeria: 1970–2018	NARDL	+ve (insig) for + ve (−ve) partition

Starting with the international studies (Panel A of Table 1), we firstly note that the bulk majority of studies conducted exclusively on developing countries produce a positive coefficient (Kastratovic 2019; Neequaye and Oladi 2015; Xaisongkham and Liu 2023) with the sole exception of an earlier study by Talukar and Meisner (1991) which finds a negative effect. Other international studies tend to split the data into sub-groups based on income levels or regional groups and most studies indicate that developed countries produce a negative effect whilst developing and lower income groups tend to produce positive or insignificant estimates (Adeel-Farooq, Riaz, and Ali 2021; Benzerrouk, Abid, and Sekrafi 2021; Doytch and Uctum 2016; Hoffman et al. 2005; Marques and Caetano 2020; Omri, Nguyen, and Rault 2014; Shahbaz et al. 2015; Shao 2017; Singhania and Saini 2021). Moreover, threshold effects have been observed in which countries with high incomes (Wang, Yang, and Wang 2023; Xie, Wang, and Cong 2020), high FDI (Khan, Rana, and Ghardallou 2023) tend to have negative effects whilst those low do not. In summary, these international studies imply that African countries, which are categorised as developing and low-income countries.

In further scrutinising panel studies which have been exclusively conducted in African countries (Panel B of Table 1), we similarly observe that most of these studies find a positive effect of FDI on CO₂ emissions (Bokpin 2017; Ojewumi and Akinlo 2017; Opoku and Boachie 2020; Boamah et al. 2023) while a few others either find a negative (Opoku, Adams, and Aluko 2021; Tawiah, Zakari, and Khan 2021) or insignificant effect (Adjei-Mantey and Adams 2023; Aminu, Clifton, and Mahe 2023; Tenaw 2020). Some other studies which segregate their data into low versus high-income groups (Acheampong 2019) or low versus high levels of institutional quality (Bouzahzah, 2022) and generally observe negative (positive) effects for countries with low (high) levels of income and institutional quality. Concerning nonlinearities, the most popular method amongst these studies are the quantile regressions (Asiedu 2021; Gyamfi et al. 2021; Halliru, Loganathan, and Hassan 2021) which tend to find positive effects at the upper quantiles or high levels of carbon emissions whilst the sole study of Abdul-Mumuni, Amoh, and Mensah (2023) uses a panel NARDL model to reveal positive (negative) FDI-emissions relationship during expansionary (contractionary) cycles.

Lastly, for the group of country-specific studies (Panel C of Table 1), it is interesting to observe that only 22 African countries have been covered by previous country-specific studies for individual African countries. Further scrutiny reveals that West African countries, and particularly ECOWAS countries, attract the most empirical country-specific attention. For instance, Nigeria has attracted the most empirical attention with four studies conducted at the country-specific level which mutually reveal an inverse FDI-emissions relationship for the country (Awodumi 2021; Keho 2015; Odubgesan and Adebayo 2020; Zubair, Samad, and Dankumo 2020). Other West African countries such as Benin, Burkina Faso, Ghana, Senegal, Sierra Leone and Togo each have two studies (Awodumi 2021; Keho 2015). Notably, the only non-Western African countries which have more than one paper are South Africa (Kivyiro and Arminen 2014; Solarin and Al-Mulali 2018) and Egypt (Mahmood et al. 2019; Solarin and Al-Mulali 2018) which produce conflict empirical evidence. The remaining African countries have one paper each i.e. Algeria, DRC, Kenya, Liberia, Niger, Tunisia, Morocco, Zambia, and Zimbabwe (Hakimi and Hamdi 2016; Kivyiro and Arminen 2014; Udemba and Yalcintas 2021).

Having summarised the scope, methods and findings of previous studies, we note that whilst most panel studies have extensive coverage of African countries which generally reveal positive or insignificant FDI-emissions relationships, the scope of the country-specific evidence is less extensive and more inconclusive. Moreover, from a methodological perspective, the current methods used in previous studies either estimate linear regressions or, at best, use nonlinear estimators which can only account for one form of asymmetries. Based on these observations, we formulate the following testable hypotheses:

1. The findings from a majority of previous panel data studies do not apply to each individual African country.

2. Estimates based on linear econometric methods may be hiding important nonlinear relationships which exist in several forms which can make the ‘sign, and strength’ of the relationship changes at different times and over different frequencies.

To test the above hypotheses, we conduct a comparative analysis in which we estimate the FDI-emission relationship for 51 countries using two empirical models. Firstly, we use conventional ARDL regressions which is a popular estimation method in the literature for country-specific evidence (Awodumi 2021; Keho 2015; Kivyiro and Arminen 2014; Mahmood et al. 2019; Solarin et al. 2017; Zubair, Samad, and Dankumo 2020). Secondly, we use more powerful wavelet coherence tools which allow us to examine the ‘sign and strength’ switching dynamics in the FDI-emissions correlation in time–frequency space. Computational details of these methods are provided in the following section.

3. Empirical framework

3.1 Baseline model

We specify our baseline empirical model as: (1) $\mathrm{C}\mathrm{O}{2}_t={\beta }_0+{\beta }_1\mathrm{FD}{\mathrm{I}}_{\mathrm{it}}+{\beta }_X\mathrm{CONTROL}{\mathrm{S}}_{\mathrm{it}}+\mathrm{erro}{\mathrm{r}}_{\mathrm{it}}$(1) where βs are regression coefficients, error is a well-behave residual term; CO₂ is carbon emissions, FDI is a measure of foreign direct investments, and CONTROLS are a set of theoretically sound regression conditioning variables commonly used in the literature, namely,

• GDP (Acheampong 2019; Chakraborty and Mukherjee 2013; Grimes and Kento 2003; Hakimi and Hamdi 2016; Huang et al. 2021; Kastratovic et al., 2019; Keho 2015; Khan, Rana, and Ghardallou 2023; Neequaye and Oladi 2015; Odubgesan and Adebayo 2020; Ojewumi and Akinlo 2017; Omri, Nguyen, and Rault 2014; Shahbaz et al. 2015; Tawiah, Zakari, and Khan 2021; Udemba and Yalcintas 2021; Wang, Yang, and Wang 2023; Xaisongkham and Liu., 2023; Xie, Wang, and Cong 2020; Xu, Huang, and An 2021).

• Trade (Essandoh, Islam, and Kakanika 2020; Mahmood et al. 2019; Marques and Caetano 2020; Opoku, Adams, and Aluko 2021; Opoku and Boachie 2020; Singhania and Saini 2021; Talukar and Meisner 1991; Zubair, Samad, and Dankumo 2020).

• Urbanisation (Adeel-Farooq, Riaz, and Ali 2021; Asiedu 2021; Benzerrouk, Abid, and Sekrafi 2021; Boamah et al. 2023; Bokpin 2017; Bouzahzah, 2022; Gyamfi et al. 2021; Sarokdie and Strezov 2019; Solarin et al. 2017; Solarin and Al-Mulali 2018).

With the exception of FDI and GDP, the literature predicts a positive effect of these controls on carbon emissions. For both FDI and GDP, theory depicts that the relationship between the variables can be either negative or positive. In the following sections, we outline the methods used to estimate the baseline FDI–emissions relationship.

3.2 ARDL model

We firstly use an ARDL model to estimate the long-run and short-run cointegration relationships between emissions (CO₂), foreign direct investment (FDI) and, whilst controlling for economic growth (GDP), Trade (TRADE) and urbanizations (URBAN). In its baseline form, the ARDL models can be specified as: (2) $\begin{array}{rl}\mathrm{\Delta }\mathrm{C}\mathrm{O}{2}_t& =a_0+a_1\mathrm{C}\mathrm{O}{2}_{t-1}+a_2\mathrm{FD}{\mathrm{I}}_{t-1}+a_3\mathrm{GD}{\mathrm{P}}_{t-1}+a_4\mathrm{TRAD}{\mathrm{E}}_{t-1}+a_5\mathrm{URBA}{\mathrm{N}}_{t-1}\\ & +{\sum }_{i=1}^{p-1}{b}_{1i}\mathrm{\Delta }\mathrm{C}\mathrm{O}{2}_{t-i}+{\sum }_{i=1}^{p-1}{b}_{2i}\mathrm{\Delta }\mathrm{FD}{\mathrm{I}}_{t-i}+{\sum }_{i=1}^{p-1}{b}_{3i}\mathrm{\Delta }\mathrm{GD}{\mathrm{P}}_{t-i}\\ & +{\sum }_{i=1}^{p-1}{b}_{4i}\mathrm{\Delta }\mathrm{TRAD}{\mathrm{E}}_{t-i}+{\sum }_{i=1}^{p-1}{b}_{5i}\mathrm{\Delta }\mathrm{URBA}{\mathrm{N}}_{t-i}+{\epsilon }_t\end{array}$(2) Following Pesaran, Shin, and Smith (2001), we conduct the three-step modelling process:

Step 1: Test for bounds cointegration test by testing the null hypothesis of H₀: c₁ = c₂  = c₃=  c₄ = c₅ = 0 in Equation (2) by computing the standard F-test and use the non-standard critical values developed by Narayan (2004) for smaller samples of 30 observations.

Step 2: We estimate the long-run regression of form CO₂ = β₀ + β₁FDI_t + β₂GDP_t + β₃TRADE_t + β₄URBAN_t + e_t where the long-run coefficients are computed as β₁ = a₂/a₁, β₂ = a₃/a₁, β₃ = a₄/a₁, β₄ = a₅/a₁,. Whilst the FDI coefficient can be either negative (β₁< 0) or positive (β₁< 0), all control variables are expected to produce positive coefficients in line with general economic theory, i.e. β₂> 0, β₃> 0, β₄> 0.

Step 3: We estimate the associated error correction model of the form: (3) $\begin{array}{rl}\lg (\mathrm{C}\mathrm{O}2{)}_t& =c_0+\mathrm{EC}{\mathrm{T}}_{t-1}+{\sum }_{i=1}^{p-1}{c}_{1i}\mathrm{\Delta }\mathrm{C}\mathrm{O}{2}_{t-i}+{\sum }_{i=1}^{p-1}{c}_{2i}\mathrm{\Delta }\mathrm{FD}{\mathrm{I}}_{t-i}+{\sum }_{i=1}^{p-1}{c}_{3i}\mathrm{\Delta }\mathrm{GD}{\mathrm{P}}_{t-i}\\ & +{\sum }_{i=1}^{p-1}{c}_{4i}\mathrm{\Delta }\mathrm{TRAD}{\mathrm{E}}_{t-i}+{\sum }_{i=1}^{p-1}{c}_{5i}\mathrm{\Delta }\mathrm{URBA}{\mathrm{N}}_{t-i}+{\epsilon }_t\end{array}$(3) where the coefficient γ measures the speed of adjustment to equilibrium after a ‘shock’ to the system, and the coefficient is expected to be negative. Following Banerjee, Dolado, and Mestre (1998), Pesaran, Shin, and Smith (2001) propose the use of the t-statistic of the error correction term, γ, as an additional test for cointegration, i.e. BDS cointegration test

3.3 Wavelet coherence

Next, we employ continuous wavelet analysis can be considered a major step-up from time series-based econometric techniques such as the VAR, and present a multiresolution analysis of signal or time-series in time–frequency space. Lau and Weng (1995) present an excellent analogy to explain the concept of wavelet transforms applied to time series data by comparing this transformation process to converting a two-dimensional written musical score into three-dimensional audible music tones, characterised by frequency, time position and duration, and intensity. In practice, these transforms convolute a time series with a set of complex-valued ‘daughter wavelets’ defined as: (4) $W_x(s,)={\int }_{-\mathrm{\infty }}^{\mathrm{\infty }}x(t){\displaystyle \frac{1}{\sqrt{s}}\ast \left({\displaystyle \frac{t-}{s}}\right)\mathrm{d}t}$(4) where $\ast$ is the conjugate of the complex number, τ and s are the translation and dilation parameters responsible for amplitude and phase dynamics in time–frequency space. The mother wavelet which is responsible for the shape of the daughter wavelets is defined as the following Morlet et al. (1982) function: (5) $(t){=}^{-{\displaystyle \frac{1}{4}}}\exp (\mathrm{it})\exp \left(-{\displaystyle \frac{1}{2}{t}^2}\right),$(5)

where ω₀ is set at 2π to ensure optimal joint time–frequency resolution. We can then compute the wavelet power spectrum (WPS) of the time series as: (6) $W_m^s(s)={\displaystyle \frac{t}{\sqrt{s}}{\sum }_{n=0}^{N-1}{x}_n\ast ((n-m){\displaystyle \frac{t}{s},n=0,\dots .,N-1,m=0,\dots .,N-1}}$(6) where δt is a uniformed time step. The WPS is a representation of the energy distribution of the series in a time–frequency plance and is analogous to the variance.

The wavelet coherence analysis between a pair of time series across a time–frequency plane and is closely related to the concept of Fourier coherency (Aguiar-Conraria and Soares 2014; Torrence and Compo 1998). Given the WPS for a pair of time series x(t) and y(t) (i.e. W_xx = |W_x|² and W_yy = |W_y|²,) their cross-wavelet transform (CWT) can be defined as (WPS)_xy = W_xy = |W_xy|, from which the wavelet coherence, is computed as: (7) $R_{y.x}(s)={\displaystyle \frac{S(W_{x,y})}{{[(S{W}_x^2)(S{W}_y^2)]}^{{\displaystyle \frac{1}{2}}}}}$(7) where S is a smoothing operator in both time and scale. The phase-difference dynamics are determined as: (8) ${\varphi }_{x,y}=\mathrm{Arcta}{\mathrm{n}}^{-1}\left({\displaystyle \frac{\{W_x\}}{\{W_x\}}}\right)$(8) where π < ϕ_x,y < −π and provides information on (i) whether the pair of series are in-phase (positive) or antiphase (negative) synchronised and (ii) whether x leads y or vice-versa.

4. Data and results

4.1 Data description

We use five time series variables in our study, namely, (i) Carbon emissions in metric tons per capita (CO₂) (ii) Foreign direct investment, net inflows in current US$ (FDI) (iii) Gross Domestic Product in constant 2015 US$ (GDP) (iv) Trade as % of GDP (TRADE) (v) Urban population as % of total population (URBAN). All data is sourced from the World Bank development indicators and collected on an annual frequency over the year 1990–2020 for 51 African countries. Note that the collection of our sampled data is purely dictated by the availability of time series data of which we find that all FDI data begins in 1990 hence we standardise the length of all variables to this starting date. Sine it is well known that the ARDL cointegration model are compatible with time series variables integrated of an order higher than I(1), we therefore conduct unit root tests on the first differences variables we test for unit roots to ensure that none of the variables is integrated or order I(2). The DF-GLS test results reported in Table 2 show that all series reject the unit root null hypothesis in their first differences hence confirming the compatibility of the data with the ARDL regressions.

Table 2. Unit root tests results.

Country	ΔGHG	ΔFDI	ΔGDP	ΔAid	ΔTrade	ΔUrban
Algeria	−4.28***	−8.38***	−4.45***	−5.48***	−7.56***	−8.12***
Angola	−4.18***	−3.74***	−8.32***	−8.54***	−7.10***	−5.82***
Benin	−7.03***	−4.11***	−7.25***	−6.01***	−3.02**	−6.62***
Botswana	−6.14***	−5.59***	−4.96***	−4.02***	−7.03***	−6.29***
Burkina Faso	−7.31***	−4.84***	−9.97***	−4.43***	−9.12***	−3.55***
Burundi	−4.02***	−5.73***	−3.05**	−6.03***	−5.42***	−3.69***
Cabo Verde	−5.84***	−6.90***	−5.62***	−4.54***	−7.67***	−5.36***
Cameroon	−8.62***	−9.49***	−7.78***	−5.32***	−6.75***	−6.79***
CAR	−5.88***	−8.31***	−5.64***	−5.91***	−7.10***	−4.95***
Cote D’Ivoire	−3.27**	−7.99***	−4.07***	−6.09***	−5.95***	−3.22***
Chad	−5.10***	−2.93*	−3.76***	−8.59***	−10.05***	−3.53***
Comoros	−6.27***	−9.01***	−6.42***	−6.51***	−7.80***	−4.77***
Congo	−3.64**	3.31**	−3.36**	−3.71***	−5.28***	−9.71***
DRC	−4.87***	−10.29***	−3.90***	−6.34***	−7.18***	−3.13***
Djibouti	−7.59***	−4.96***	−7.45***	−5.54***	−6.89***	−4.08***
Egypt	−2.93*	−4.14***	−4.47***	−4.41***	−9.01***	−5.52***
Equatorial	−4.84***	−7.20***	−8.64***	−4.33***	−8.04***	−9.08***
Eritrea	−6.46***	−4.18***	−5.23***	−3.66***	−8.50***	−6.06***
Eswatini	−6.44***	−11.23***	−5.34***	−5.82***	−7.82***	−4.76***
Ethiopia	−411***	−4.74***	−4.28***	−5.31***	−3.34***	−7.77***
Gabon	−3.03*	−6.62***	−4.68***	−3.92***	−8.34***	−8.53***
Gambia	−2.93*	−5.99***	−5.43***	−7.13***	−7.02***	−4.57***
Ghana	−4.59***	−4.94***	−3.02*	−3.72***	−5.21***	−4.81***
Guinea	−6.71***	−5.37***	−5.54***	−5.01***	−5.42***	−5.10***
Guinea_b	−5.05***	−3.81***	−4.15***	−6.29***	−8.06**	−3.27***
Kenya	−5.45***	−5.32***	−9.44***	−4.18***	−5.51**	−7.75***
Lesotho	−6.34***	−8.29***	−4.08***	−4.56***	−5.82***	−5.42***
Liberia	−5.04***	−4.96***	−6.94***	−5.50***	−5.68***	−5.89***
Libya	−7.76***	−5.54***	−4.91***	−5.91***	−7.71***	−8.62***
Madagascar	−9.18***	−3.30**	−4.97***	−3.79***	−6.63***	−5.07***
Malawi	−7.91***	−7.79***	−5.23***	−6.23***	−4.46***	−3.03**
Mali	−6.47***	−7.30***	−6.36***	−6.26***	−3.78***	−8.74***
Mauritius	−6.70***	−3.38**	−8.02***	−4.03***	−3.66***	−4.32***
Morocco	−7.53***	−9.70***	−9.18***	−5.36***	−4.02***	−4.14***
Mozambique	−5.05***	−3.95***	−6.95***	−5.42***	−5.12***	−3.42***
Namibia	−4.86***	−6.61***	−9.85***	−7.56***	−8.52***	−4.54***
Niger	−8.92***	−7.76***	−4.08***	−8.04***	−6.81***	−3.74***
Nigeria	−5.48***	−6.33***	−8.97***	−4.23***	−5.82***	−4.80***
Rwanda	−6.69***	−5.09***	−8.19***	−4.10***	−5.33***	−5.18***
South Africa	−4.71***	−6.13***	−4.96***	−4.17***	−7.88***	−4.45***
Senegal	−6.46***	−5.18***	−7.63***	−5.58**	−7.48***	−4.80***
Seychelles	−5.11***	−5.83***	−5.23***	−6.61***	−7.70***	−4.89***
Sierra Leone	−5.52***	−5.59***	−4.49***	−6.38***	−8.94***	−3.91***
Somali	−7.61***	−4.96***	−6.86***	−5.90***	−3.24***	−3.34***
Sudan	−7.51***	−6.63***	−3.75***	−3.56***	−8.62***	−3.86***
Tanzania	−5.30***	−3.85***	−7.56***	−3.58***	−6.01***	−8.75***
Togo	−3.28***	−3.21**	−3.41**	−5.35***	−6.18***	−4.29***
Tunisia	−6.87***	−10.02**	−5.81***	−6.23***	−4.45***	−4.47***
Uganda	−5.41***	−5.82***	−4.76***	−3.19**	−3.16**	−4.51***
Zambia	−5.53***	−8.64***	−3.12**	−8.12***	−4.40***	−6.39***
Zimbabwe	−5.31***	−7.42***	−3.42***	−4.38***	−7.35***	−4.71***

Notes: ‘***’, ‘**’, ‘*’ denote 1%, 5% and 10% critical levels, respectively.

4.2 ARDL estimates

We begin our analysis by estimating the ARDL regressions of CO₂ emissions regressed on FDI and other control variables for 51 countries. To recall, the modelling process requires that we firstly estimate the baseline regression and then testing for bounds cointegration tests by computing the standard F-statistic and comparing it to non-standard critical values. However, since we have 30 observations per country, we rely on the critical values derived by Narayan (2004) which were devised for smaller samples of 30 observations. Once, bounds cointegration are verified, we proceed to estimating the long-run, short-run coefficients of the regression and use the t-statistic of the error correction term as an additional test for cointegration (Banerjee, Dolado, and Mestre 1998; Pesaran, Shin, and Smith 2001).

Table 3 summarises the empirical findings from the ARDL modelling process for the sampled data. For the sake of space, Table 3 only reports whether the coefficient estimates are positive (+), negative (−) or insignificant (insign) and the bounds and BDS cointegration tests pass the 5% significance level (✓) or not (✗). Note that we estimate ARDL (1. 0, 0, 0, 0, 0) regressions as they are found to be optimal lag lengths in accordance with the AIC information criterion. Overall, we observe that 12 out of the 51 observed countries mutually produce F-statistics and ECTs which fail to reject the null hypothesis of no ARDL relationships between the variables hence indicating no significant FDI-emissions cointegration relations for these countries (i.e. Botswana, Equatorial Guinea, Gambia, Kenya, Lesotho, Liberia, Madagascar, Seychelles, Sierra Leone, Somali, Zambia, Zimbabwe). For the remaining 39 countries, only 10 countries produce statistically significant coefficient estimates on the FDI variable with 4 countries producing positive and significant estimates (i.e. CAR, Djibouti, Eritrea, Namibia) whilst 6 countries produce negative and significant coefficient estimates (i.e. Algeria, Benin, Ghana, Libya, Malawi, Nigeria).

Table 3. ARDL regressions.

Country	Bounds test	Long-run coefficients					Short-run and error correction coefficients
		FDI	GDP	Aid	Trade	Urban	ΔFDI	ΔGDP	ΔAid	ΔTrade	ΔUrban	ECT/BDS tests
Algeria	✓	−	+	insig	insig	insig	−	+	insig	insig	insig	✓
Angola	✓	insig	+	insig	insig	+	insig	+	insig	insig	+	✓
Benin	✓	−	+	insig	insig	insig	−	+	insig	insig	insig	✓
Botswana	✗	insig	insig	insig	insig	insig	insig	insig	insig	insig	insig	✓
Burkina Faso	✓	insig	+	insig	insig	+	insig	+	insig	insig	+	✓
Burundi	✓	insig	+	insig	insig	+	insig	+	insig	insig	+	✓
Cabo Verde	✓	insig	insig	insig	+	+	insig	insig	insig	+	+	✓
Cameroon	✓	insig	insig	insig	insig	+	insig	insig	insig	insig	insig	✓
CAR	✓	+	insig	−	+	+	+	insig	−	+	+	✓
Cote D’Ivoire	✓	insig	+	insig	insig	−	insig	+	insig	insig	−	✓
Chad	✓	insig	+	insig	insig	+	insig	+	+	Insig	+	✓
Comoros	✓	insig	+	insig	insig	+	−	+	insig	+	insig	✓
Congo	✓	insig	insig	insig	insig	+	insig	insig	+	insig	+	✓
DRC	✓	insig	insig	insig	insig	+	insig	insig	insig	insig	insig	✓
Djibouti	✓	+	−	insig	insig	+	+	−	insig	insig	+	✓
Egypt	✓	insig	+	insig	insig	insig	insig	−	−	insig	insig	✓
Equatorial	✗	insig	insig	insig	insig	insig	insig	insig	insig	insig	insig	✓
Eritrea	✓	+	+	insig	−	−	+	+	insig	−	−	✓
Eswatini	✓	insig	insig	insig	insig	+	insig	insig	insig	insig	+	✓
Ethiopia	✓	insig	insig	insig	insig	insig	insig	insig	insig	insig	insig	✓
Gabon	✓	insig	+	insig	insig	insig	insig	+	insig	insig	insig	✓
Gambia	✗	insig	insig	insig	insig	insig	insig	insig	insig	insig	Insig	✓
Ghana	✓	−	+	insig	insig	insig	−	+	insig	insig	insig	✓
Guinea	✓	insig	+	insig	+	+	insig	insig	insig	+	+	✓
Guinea_b	✓	insig	insig	insig	insig	+	insig	insig	insig	insig	+	✓
Kenya	✗	insig	insig	insig	insig	insig	insig	insig	insig	insig	insig	✓
Lesotho	✗	insig	insig	+	insig	insig	insig	insig	insig	insig	insig	✓
Liberia	✗	insig	insig	insig	insig	+	insig	insig	insig	insig	insig	✓
Libya	✓	−	+	insig	insig	insig	−	+	insig	insig	insig	✓
Madagascar	✗	insig	insig	insig	insig	insig	insig	Insig	insig	insig	insig	✓
Malawi	✓	−	+	insig	insig	insig	−	+	insig	insig	insig	✓
Mali	✓	insig	insig	insig	insig	insig	−	insig	insig	insig	insig	✓
Mauritius	✓	insig	insig	−	insig	−	+	insig	−	+	−	✓
Morocco	✓	insig	+	−	+	insig	insig	+	−	+	insig	✓
Mozambique	✓	insig	+	insig	+	insig	insig	+	insig	+	insig	✓
Namibia	✓	+	−	+	−	+	+	−	+	−	+	✓
Niger	✓	insig	insig	insig	insig	insig	insig	insig	insig	insig	insig	✓
Nigeria	✓	−	+	insig	insig	insig	−	+	insig	+	−	✓
Rwanda	✓	insig	+	insig	insig	+	insig	+	insig	insig	+	✓
South Africa	✓	insig	+	insig	insig	insig	insig	+	+	insig	insig	✓
Senegal	✓	insig	+	insig	insig	insig	insig	+	insig	insig	insig	✓
Seychelles	✗	insig	insig	insig	insig	insig	insig	insig	insig	insig	insig	✓
Sierra Leone	✗	insig	insig	insig	insig	insig	insig	insig	insig	insig	insig	✓
Somali	✗	insig	insig	insig	insig	insig	insig	insig	insig	insig	insig	✓
Sudan	✓	insig	insig	Insig	insig	+	insig	insig	insig	insig	insig	✓
Tanzania	✓	insig	insig	insig	insig	+	insig	insig	insig	insig	insig	✓
Togo	✓	insig	insig	+	insig	+	insig	insig	+	insig	+	✓
Tunisia	✓	insig	+	insig	insig	insig	insig	+	insig	insig	insig	✓
Uganda	✓	insig	+	insig	+	insig	−	+	−	+	insig	✓
Zambia	✗	insig	+	insig	insig	insig	Insig	+	insig	insig	insig	✓
Zimbabwe	✗	insig	+	insig	insig	insig	Insig	+	insig	insig	insig	✓

Notes: All coefficient estimates are evaluated at 10% critical level.

Notably, some of our findings are in coherence with previous country-specific literature. For instance, Awodumi (2021) and Keho (2015) similarly find an insignificant effect for Mali, Senegal, Sierra Leone, Togo; Solarin and Al-Mulali (2018) find an insignificant effect for Egypt whilst Kivyiro and Arminen (2014) for Congo and Zambia. Moreover, the negative effect found for Nigeria was also established in the previous works of Solarin and Al-Mulali (2018), Odubgesan and Adebayo (2020), Zubair, Samad, and Dankumo (2020) and Awodumi (2021). All-in-all, the findings from our ARDL estimates concur with those of previous studies for 8 of the 22 countries conducted at the country-specific level. Also, there appears to be more support for GDP and urbanisation as determinants of carbon emissions compared to FDI flows as the coefficient estimates for these control variables generally produce positive and significant estimates in the majority of African cases. These latter findings are line with those of Bokpin (2017); Solarin et al. (2017); Solarin and Al-Mulali (2018); Sarokdie and Strezov (2019); Adeel-Farooq, Riaz, and Ali (2021); Asiedu (2021); Benzerrouk, Abid, and Sekrafi (2021); Gyamfi et al. (2021); Bouzahzah (2022); Boamah et al. (2023).

Table 4 presents the diagnostic tests of the ARDL regressions, i.e. normality (${\chi }_{\mathrm{NORM}}^2$), Breusch–Godfrey serial correlation test (${\chi }_{\mathrm{SC}}^2$), ARCH heteroscedasticity tests (${\chi }_{\mathrm{HET}}^2$) and RESET test for functional form (${\chi }_{\mathrm{FF}}^2$). As an be observed most regressions suffer from non-normal residuals and incorrect functional form. On one hand, the evidence of non-normality indicates that the error terms are not normally distributed and possibly asymmetrically distributed. On the other hand, incorrect functional form is indicative of higher order regressions or nonlinear regressions are more appropriate. Against this evidence, we proceed to present the findings from the wavelet coherence analysis which captures different forms of time- and frequency-based asymmetries.

Table 4. Diagnostic tests of ARDL regressions

Country	${}_{\mathrm{NORM}}^2$	${}_{\mathrm{SC}}^2$	${}_{\mathrm{HET}}^2$	${}_{\mathrm{FF}}^2$
Algeria	0.02	0.89	0.21	0.03
Angola	0.83	0.37	0.15	0.35
Benin	0.99	0.88	0.27	0.58
Botswana	0.45	0.51	0.96	0.97
Burkina Faso	0.01	0.24	0.12	0.00
Burundi	0.00	0.28	0.92	0.00
Cabo Verde	0.17	0.58	0.04	0.02
Cameroon	0.01	0.67	0.49	0.86
CAR	0.02	0.51	0.04	0.74
Cote D’Ivoire	0.01	0.47	0.40	0.09
Chad	0.00	0.47	0.72	0.72
Comoros	0.00	0.63	0.00	0.11
Congo	0.00	0.42	0.83	0.45
DRC	0.11	0.67	0.13	0.26
Djibouti	0.69	0.65	0.86	0.33
Egypt	0.99	0.83	0.57	0.92
Equatorial	0.00	0.92	0.12	0.03
Eritrea	0.05	0.87	0.93	0.48
Eswatini	0.45	0.59	0.92	0.21
Ethiopia	0.23	0.00	0.00	0.02
Gabon	0.31	0.96	0.63	0.39
Gambia	0.00	0.95	0.14	0.65
Ghana	0.00	0.49	0.47	0.00
Guinea	0.00	0.94	0.32	0.03
Guinea_b	0.00	0.06	0.94	0.49
Kenya	0.00	0.59	0.49	0.26
Lesotho	0.00	0.48	0.56	0.30
Liberia	0.00	0.00	0.10	0.00
Libya	0.00	0.35	0.49	0.55
Madagascar	0.74	0.97	0.12	0.07
Malawi	0.00	0.26	0.70	0.00
Mali	0.56	0.60	0.10	0.85
Mauritius	0.31	0.93	0.57	0.11
Morocco	0.93	0.80	0.10	0.03
Mozambique	0.72	0.06	0.00	0.25
Namibia	0.71	0.66	0.46	0.70
Niger	0.50	0.42	0.54	0.00
Nigeria	0.97	0.36	0.26	0.98
Rwanda	0.10	0.07	0.61	0.06
South Africa	0.87	0.52	0.15	0.20
Senegal	0.37	0.61	0.10	0.00
Seychelles	0.00	0.86	0.10	0.02
Sierra Leone	0.00	0.49	0.89	0.15
Somali	0.00	0.63	0.99	0.65
Sudan	0.79	0.90	0.16	0.17
Tanzania	0.22	0.97	0.42	0.95
Togo	0.00	0.61	0,16	0.81
Tunisia	0.00	0.84	0.98	0.58
Uganda	0.97	0.45	0.03	0.06
Zambia	0.07	0.87	0.54	0.85
Zimbabwe	0.15	0.38	0.11	0.72

Notes: p-values of diagnostic tests reported.

4.3 Wavelet coherence analysis

Next, we present the findings from our wavelet coherence analysis and since these methods are unconventional, we provide a brief description of the interpretation of the ‘heatmaps’ used to represent the findings from the analysis. Generally, the wavelet coherency is captured on a 2-dimension contour plot which describes the ‘sign and strength’ of the correlation between FDI and emissions at different times (measured along the horizontal axis) and different frequencies (measured along vertical axis). The sign of the relationship is determined by the phase dynamics as represented by the ‘arrow orientation’ in the spectral plot within phase (anti-phase) or positive (negative) correlation shown by arrows pointing to the right (left), i.e. $\uparrow$, $\nearrow$, $\to$ and $\searrow$ ($\downarrow$, $\swarrow$, $\leftarrow$ and $\nwarrow$). Therefore, nonlinear dynamics can be captured by observing changes in the arrow notation across different times and frequencies. Furthermore, the strength of the relationships are measured by the colour contours with warmers (cooler) colours denoting stronger (weaker) coherence, which adds another dimension to the measure of nonlinearity, in the sense of capturing strength-varying dynamics in the FDI-emissions correlation. the significance of the wavelet coherence is determined by the faint white lines enveloping/encompassing the arrow orientation and colour contours.

A general overview of the wavelets (see Appendices 1–3) shows significant cyclical synchronisation between FDI-emissions co-movements for the entire African sample, which is different from the findings of the ARDL regressions which showed an insignificant relationship for most African countries. To facilitate the discission of our results, we group the findings from the wavelet coherence analysis for the individual countries into three categories:

1. Firstly, there are 17 countries which strictly showed in-phase dynamics between FDI and emissions at all time periods and at different frequencies hence lending supporting the pollution haven hypothesis i.e. Angola, CAR, Ivory Coast, Egypt, Eretria, Libya, Morocco, Madagascar, Mali, Malawi, Namibia, Senegal, Togo, Tanzania, South Africa, Zambia and Zimbabwe.

2. Secondly, there are 16 countries which consistently showed anti-phase dynamics between FDI and emissions at all time periods and across all frequencies hence lending support to the pollution halo hypothesis i.e. Algeria, Benin, Cameroon, Congo, DRC, Ethiopia, Gabon, Equatorial Guinea, Gambia, Lesotho, Niger, Nigeria, Rwanda, Somalia, Chad, and Comoros.

3. Lastly, the remaining 18 countries which have sign-switching FDI-emissions dynamics across different frequencies. These group of countries can be further segregated into two sub-groups, with 10 countries which exhibit in-phase (anti-phase) dynamics at lower (higher) frequencies, i.e. Burundi, Cabo Verde, Kenya, Liberia, Mauritania, Sudan, Sierra Leone, Uganda; and 8 countries which exhibit anti-phase (in-phase) dynamics at lower (higher) frequencies Burkina Faso, Botswana, Djibouti, Ghana, Guinea Bissau, Mozambique, Mauritius, Eswatini, Seychelles, Tunisia.

It now becomes curious to see if they are any common characteristics amongst the three groups of countries as advocated by previous literature. For instance, several studies, such as Hoffman et al. (2005), Shahbaz et al. (2015), Doytch and Uctum (2016), Shao (2017), Acheampong (2019), Marques and Caetano (2020) Adeel-Farooq, Riaz, and Ali (2021), all find that differences in income and levels of development can account for heterogeneities observed in FDI-emissions relationships across different groups of countries. Other studies such as Chang (2015) and Bouzahzah (2022) find that these heterogeneities can be explained by differences in institutional quality whilst Gyamfi et al. (2021), Khan, Rana, and Ghardallou (2023) and Abdul-Mumuni, Amoh, and Mensah (2022) argue for differences in levels of resources, FDI and carbon emissions, respectively, account for the heterogeneities. Overall, our findings do not offer support to these previous generalisations in the literature. We conclude that heterogeneities in the FDI emissions amongst African countries are most likely to be attributed to industry-specific dynamics.

5. Further discussion of the results

We now provide further discussions of our empirical results by comparing our findings to those of previous studies and provide explanations for any differences in the observed results of the various African countries.

Firstly, in comparison to previous studies, our current analysis has very little similarity to those of previous studies. For instance, most previous panel studies advocate for positive FDI-emissions, where both our ARDL analysis finds an insignificant relationship for most African countries and the wavelet analysis shows s positive relationship for only a third of our analysis. Even at a country-specific level, our findings concur with those of previous studies for one country, Nigeria. We find it interesting to note that even when applying similar ARDL methodology and control variables to previous country-specific studies, we obtain contradictory results. Since the main difference between our study and these previous studies in estimating the ARDL regressions is our sample period, we, therefore, conclude that traditional econometric techniques are sensitive to the choosen sample period and are thus not robust to structural breaks. Therefore, we consider the findings from our wavelet coherence analysis more robust and reliable as they are not sensitive to selected time period and neither are there prone to regression errors.

Secondly, the findings from both the ARDL and wavelet coherence analysis do not collaborate with those from previous studies in explaining the differences in the FDI-emissions dynamics in African countries such as differences in levels of income, FDI, natural resources and institutional quality. We therefore conclude that heterogeneities in the FDI-emissions exist at country level and can be best explained by differences in distributions FDI flows to ‘dirty’ and ‘green’ industries in each individual country. Further given the aggregated nature of our data, our findings thus imply that countries which showed a positive (negative) relationship, have a majority of their FDIs channelled towards ‘dirty’ (‘cleaner’) sectors or industries characterised by lower (higher) green technologies adoption. Overall, the findings from our study pinpoint the individual African countries whose governments should be concerned with the channelling of FDI inflows into cleaner industrial structures.

6. Conclusions

The paper examined the FDI-emissions relationship for 51 countries between 1990 and 2020 using more rigorous empirical methods. To this end, we made use of wavelet coherence analysis to capture both time-varying and frequency-varying asymmetries. This allows us to overcome the shortcomings of traditional econometric techniques which at best allows researchers to account for one type of asymmetries. Moreover, our study differs from previous African studies which either use a panel approach to cover a wide range of countries or single country approach which covers a limited number of countries, and we adopt a country-specific approach for a wide range of African countries to effectively account for country-specific heterogeneities which could influence the FDI-emissions relationship. For comparative purposes, we also estimate conventional ARDL models using control variables dictated by conventional literature.

Whilst the conventional ARDL models indicate an insignificant FDI-emissions relationship for most African countries, the wavelet coherence analysis indicates otherwise, with 17 countries showing a positive relationship, 16 countries showing a negative relationship and the remaining 18 countries showing a nonlinear relationship. In comparing our results to those of previously related research, we find that they do not concur with those of most previous panel-based and country-specific studies. In further trying to explain the differences in findings, we conclude that the heterogeneities in the FDI-relationship are not explained by regional similarities or differences but are most likely explained by industry-specific effects existing at country level. We therefore conclude that regional policies such as the AfCFTA which are aimed at increasing intra- and inter-continental FDI flows would be harmful to about a third of African countries who have signed the agreement. On a global policymaking scale, these results imply that FDI cannot be used as a conduit for climate justice for most African countries. Overall, our study highlights the necessity of augmenting regional policies with country-specific tailored strategies aimed at ushering capital flows from ‘dirty’ to ‘green’ industries and sectors.

However, identifying which specific industries policymakers should monitor is beyond the scope of our study as we employ aggregated FDI and emissions data in our analysis. Naturally, we propose that future studies be dedicated to conducting an industry-specific analysis for African countries. Nonetheless, one notable challenge is the data availability for African countries of which FDI time series is available at an aggregated level for individual countries. This is unlike the case for European countries who have publicly available FDI flow data at the sector level from the OCED statistics online database i.e. https://stats.oecd.org/index.aspx?DatasetCode=FDI_FLOW_INDUSTRY. We therefore implore researchers and national statistics departments and agencies to prioritise the collection of such sectoral FDI time series for use in future research. In the meantime, immediate future research can focus on investigating the impact of aggregated FDI flows on sectoral emissions for African countries using the industry emissions data available from ‘Our World in Data’ https://ourworldindata.org/co2-and-greenhouse-gas-emissions.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data is available from the authors upon reasonable request.

Appendices

Appendix 1. Wavelet plots dominated by in-phase dynamics

Appendix 2. Wavelet plots dominated by anti-phase dynamics

Appendix 3. Wavelet plots dominated by both in-phase and anti-phase dynamics