Is FDI inflow bad for environmental sustainability?

Abstract

In the last few years, discussions and debates around environmental sustainability have become a key goal of the global agenda. This is because the latest studies and our direct experiences of the consequences of environmental degradation and climate change are showing that economic models need some urgent changes. The goal of this paper was to investigate the effects of foreign direct investment (FDI) flow on environmental sustainability globally. The balanced panel data of 100 countries from 2000 to 2020 were used for empirical analysis to understand the relationship between FDI inflow and environmental performance index (EPI) for countries with different income levels. The estimation is made using the two-step system GMM model as it gives more robust and efficient results in this study. The results of the dynamic panel model do not support the pollution haven hypothesis, i.e., FDI inflow was not found to have any statistically significant effect on EPI. The findings of the paper suggest that to improve environmental quality globally, it is important for governments to invest more in renewable energy projects as clean energy consumption can be one of the most efficient solutions for reducing the impact of climate change. Moreover, countries need to reduce the unemployment rate as it significantly improves the environmental quality according to our estimation results.

1 Introduction

Environmental degradation is an issue for all communities, and it is becoming increasingly important due to climate change, the growing number of climate refugees and environmental migrants, as well as effects on health. In the last few years, discussions and debates around environmental sustainability have become a key goal of the global agenda. This is because the latest studies and our direct experiences of the consequences of environmental degradation and climate change are showing that economic models need some urgent modifications.

According to the pollution haven hypothesis, FDI is one of many possible factors that worsen environmental quality. A negative nexus between FDI and environmental quality is considered because industrialized countries relocate their polluting factories and set up offices in developing countries due to their weak environmental regulations, which result in environmental degradation in the host developing countries.

As statistics show, FDI flows have increased much in the last decades in almost all countries in the world, especially in developing countries. Increased FDI has raised debates about advantages and disadvantages of FDI inflows, with emphasis on the environmental effects and employment generation.

There are two categories of theoretical and quantitative research investigating the costs and benefits of FDI inflows. The first category of studies is about the effects of FDI on economic growth at the aggregate level (e.g., Lucas 1988; De Mello 1999; Carkovic and Levine 2002; Choe 2003; Addison and Heshmati 2004; Moura and Forte 2010). From theoretical aspects of the modern endogenous growth theory, countries that have more open and liberal policies toward FDI inflows influence and generate long-term economic growth of the countries. On the other side, quantitative research that focuses on the impact of FDI on economic growth of the receiving countries with different levels of development shows divergent and inconclusive results.

To mention a few studies aligned with the endogenous growth theory considering the FDI–growth nexus, De Mello (1999) argues that the growth effect of FDI is found positive within OECD countries, but it has negative growth effect in other countries. Carkovic and Levine (2002) showed that FDI does not impact the economic growth of the host country. Choe (2003) investigated the causal relationships between economic growth and FDI in a sample of 80 countries with different levels of development. By using a panel vector autoregressive (VAR) model, the author showed that as expected FDI inflow Granger causes economic growth, but the reversed effects are stronger from economic growth to FDI inflows than from FDI to growth. Addison and Heshmati (2004) investigated the new global determinants of FDI flows, with a particular focus on the impact of information and communication technology (ICT) and democratization process, on FDI inflows to developing countries. The main findings suggest that democratization and ICT increase FDI inflows to developing countries. Moura and Forte (2010) tried to explain the divergence of the results from existing theoretical and empirical studies about the economic impacts of FDI. They argue that the effects of FDI on economic growth depend on internal conditions and characteristic factors of the receiving country, such as economic, political, social, and cultural factors.

The second category includes studies related to the effects of FDI on the environmental conditions in the developing host countries (e.g., Taylor and Copeland 2003; Aliyu 2005; Baek and Koo 2009; Chakraborty and Mukherjee 2013; Yoon and Heshmati 2021). Recent studies on the pollution haven hypothesis show that under rapid globalization, developed countries tend to relocate “dirty” industries to developing countries due to their comparatively weaker environmental policies and regulations. The relocation of dirty production allows them to avoid costly pollution taxes in their home countries (Taylor and Copeland 2003).

Aliyu (2005) investigated two possible empirical consequences of the pollution haven hypothesis, namely: the correlation between FDI outflow in developed countries and environmental policy stringency, and the effect of FDI inflow on pollution in developing host countries with lax environmental policies and regulations. Using disaggregated FDI data, they showed that environmental control is one of the main factors that affect the outflow of FDI originating from OECD countries to developing destination countries. However, FDI inflow was not found to have a significant impact among others on the total concentration of known pollutants, level of temperature, and energy use, but it was correlated with CO₂ emissions. Baek and Koo (2009) using the commonly used autoregressive distributed lag (ARDL) model showed that FDI inflows in China negatively impacted the environmental quality in both the short- and long-run perspectives. But in the case of India, FDI was found to have a significant detrimental effect on the environment in the short run, but FDI had little negative effect on the environment the long run. Chakraborty and Mukherjee (2013) using panel data conducted empirical analysis over period of 2000–2010 for 114 countries. They showed that consistent with the pollution heaven hypothesis, the environmental sustainability of studied countries is negatively related to FDI inward movements, but as expected the environmental sustainability is positively influenced by FDI outward movements. Yoon and Heshmati (2021) investigate the environmental regulation’s effects on FDI decisions of the Korean manufacturing sector. The results showed that the stricter the regulations in host countries in Asia, the lower the FDI flows in these countries. The findings are in line with the prevalence of a pollution haven hypothesis.

The goal of the current research was to quantitatively analyze the impact of the inflow of FDI on the environmental performance index (EPI) globally, which will answer the main question of this paper: Does FDI inflow negatively impact environmental performance? Besides the main question described above, our study also examines the indirect effects of FDI on the environment and elaborates on other factors that cause environmental sustainability worldwide. For the estimation, the current study uses the two-step system GMM model, which alleviates endogeneity, heteroscedasticity, and autocorrelation problems.

The remainder of this study is organized as follows. In Sect. 2, we review the existing studies related to FDI–environmental degradation nexus. Section 3 describes the data and variables used in the estimation. Section 4 outlines the empirical models. Section 5 summarizes the results of the empirical analysis, Sect. 6 presents the conclusions and policy recommendations, and finally, Sect. 7 provides some future research directions and mentions of limitations of the paper.

2 Review of the Literature

FDI is viewed one of the important drivers of economic growth of host countries. Evans and Lucy (2020) found that over the period 2013–2018, Ghanaians benefitted from FDI-registered projects that created around 85 percent of the total jobs in Ghana. Hung (2005) showed that FDI inflows in Vietnam accelerate the economic growth and significantly reduce poverty rate in Vietnamese provinces.

On the other hand, rapid globalization and problems related to environmental sustainability made policymakers rethink the pros and cons of FDI. In environmental economics, the phenomenon of a negative relationship between FDI and environment is called the pollution haven hypothesis labeled as PHH. Many empirical papers tested the pollution heaven hypothesis but have found diverse and inconclusive results. This section will discuss some empirical studies that identified the FDI–environmental pollution relationship with different outcomes, covering different countries, years of observation, and estimation models.

Zhu et al. (2017), using spatial econometric models, examined the nexus relationship between FDI and sulfur dioxide emissions (SO₂) in one of the main industrial areas in China, namely the Beijing–Tianjin–Hebei mega region. Their study confirmed the presence of the pollution haven hypothesis, supported by evidence of a significantly positive relationship between FDI and SO₂ emissions found from 2000 to 2013. The authors also proved the existence of a spatial spillover effects of SO₂ emissions in the Beijing–Tianjin–Hebei mega region. The results showed that a 1% increase in the SO₂ emissions of surrounding regions would result in a 0.118% increase in spatial SO₂ emissions in local cities.

Some studies showed that FDI is linked with the income level of the host country. For example, Abid and Sekrafi (2021) examined both the direct and indirect impacts of trade openness on CO₂ emissions in 31 developed and 100 developing countries. They used the dynamic panel estimation technique. The estimation results of their study showed that the effect of FDI on CO₂ emissions is different for the two country groups distinguished by their levels of income. According to Abid and Sekrafi (2021), the issues of pollution haven hypothesis were confirmed in developing countries, whereas the opposite or pollution halo hypothesis¹ was found in developed countries. The authors explain that the positive impact of FDI on the environment in developed countries is attributed to environmentally friendly transfers of production technologies and management practices. Shahbaz et al. (2015a, 2015b) provided analogous results. According to their study, the effect of FDI on environmental and quality is different for countries with different levels of income. The findings show that FDI increases the environmental quality in high-income countries, supporting a “pollution halo.” An inverted U-shaped relationship was found between FDI and CO₂ emissions in middle-income countries. Considering low-income countries, FDI increases environmental pollution, which supports pollution haven hypothesis.

There exist some studies that empirically proved the positive nexus between FDI and environmental quality. For example, Rafindadi et al. (2018) explored whether FDI and energy consumption increase the environmental pollution in six member states of the Gulf Cooperation Council (GCC) over the period 1990–2014. Using Pooled Mean Group estimation methodology, the authors found that FDI inflows significantly decrease environmental degradation in the GCC. Such a result supports both the pollution halo hypothesis and the related technique’s effects of FDI. The results of their study also showed that energy use negatively impacts environmental quality. The authors argue that this is attributed to the scale effect of intensive energy consumption on the environment in the GCC region.

Another study that evidenced the pollution halo effect is a paper by Polloni-Silva et al. (2021). With a large sample of 592 municipalities located in São Paulo state in Brazil observed over a period 2010–2016, the authors investigated the relationship between FDI and CO₂ emissions. The estimation result is conditional on using other control variables, e.g., GDP per capita, distinguishing industrial, and service sectors, accounting for the share of residential electricity consumption of total consumption, and so on. The results of their empirical analysis show the presence of a pollution halo effect in São Paulo state. There is a linear and negative nexus found between FDI and environmental pollution. Polloni-Silva et al. (2021) state that such a result is possible because most of the foreign firms that were used in the sample originated from developed countries and were operating in medium and high-tech sectors.

Other studies by Lee (2013), Jugurnath and Emrith (2018), Albulescu et al. (2019), and Bernard and Mandal (2016) have examined whether the FDI-receiving countries are classified as pollution havens or pollution halo but could not prove any significant impact of FDI on environmental sustainability.

Various papers have focused on different factors that may impact environmental sustainability conditions in the host countries. In this regard, some of the control variables used as the determinants of environmental sustainability in our study with their linear and nonlinear forms, as well as interactions with each other, will be tested, discussed, and compared to some previous studies in Sect. 5 (Table 1).

Table 1 Summary of the findings from the literature reviewed

The divergent results of existing studies can be due to different model estimations as well as different periods of observation and samples of countries. Most of the reviewed studies used CO₂ emissions as a measure of environmental pollution, and only a few existing studies used EPI. The current study uses the two-step system GMM model, which alleviates endogeneity, heteroscedasticity, and autocorrelation problems. Our study will reduce the literature gap on the FDI–environment nexus, as it includes both squares and interactions of some independent variables.

3 Data and variables

The observation period of this research includes years between 2000 and 2020.² A total of 100 countries, for which the data for both endogenous and exogenous and control variables were provided, were used for the empirical analysis. Our sample of countries was somewhat differently divided into low-income, lower-middle-income, upper-middle-income, and high-income countries. The resulting constructed panel data consist of 1,600 observations.

For a response variable and a measure of environmental sustainability, the current study uses the environmental performance index (EPI). The index’s goal is to focus on two environmental goals, namely “reducing environmental stresses to human health” and “promoting ecosystem vitality and natural resource management” (e.g., Esty et al., 2008). The index ranges in intervals between 0 and 100 with 100 indicating a high rate of sustainability. Although ideally, it is better to estimate EPI parametrically, in this paper, we employ the given data on EPI, which was obtained from the database of Yale Center for Environmental Law and Policy reporting. The index is computed based on the non-parametric calculation and aggregation of more than 30 environmental sustainability indicators³ that reflect country-level environmental conditions.

Foreign direct investment inward (FDIin), measured as share of GDP, was taken as exogenous variable. The data for FDI inward were accessed from the UNCTAD Statistics. The inwards FDI variable is the key explanatory variable used to estimate the possible effects of pollution haven and pollution halo behavior of investors.

Along with the key exogenous variables, some control variables were also included, and estimated their effects. Summary statistics of the data and means of variables by country group and over time are presented in Tables 2, 3, and 4.

Table 2 Summary statistics of the data, NT = 100 × 16 = 1600 obs

Table 3 Mean of variables by country income group

Table 4 Mean of variables by year

The data on the following control variables and definitions of the variables used in the estimation were obtained from the World Bank Statistics. They include Trade Openness Index (OPEN), measured as the ratio of the country's total trade to the GDP, and GDP per capita (GDP) based on Purchasing Power Parity, measured in US dollars; the share of clean energy defined as alternative and nuclear energy (SCLE) is expressed as a share of total energy use; the Unemployment rate (UNEM) is measured as the share of the labor force that is without employment but is available for and seeking employment opportunities; Share of industry measured as value added by industry as a percentage of aggregate GDP (SIND); and Education Index expressed as a score between 0 and 100 (EDUC). Education is calculated using mean years of schooling and expected years of schooling, (100 = perfect education attainment).⁴

As pollution haven hypothesis states, the nexus between FDI inflows and pollution levels in the developing countries should be positive, indicating that industrialized countries relocate their polluting factories and set up offices in developing countries due to the host countries’ weak environmental regulations, which results in environmental degradation in the host developing countries. In several studies, such an expected relationship is observed (Shahbaz et al., 2015a, 2015b; Baek 2016; Solarin et al., 2017; Wang et al., 2020; Singhania and Saini 2021; Bulut et al., 2021; Musah et al., 2022).

According to environmental Kuznets curve (EKC) hypothesis, a positive correlation is expected to be found between pollution levels and GDP measuring the development level. The positive relation indicates that the level of environmental degradation increases as the country develops it level of production and energy use.

4 Model

This section presents the employed dynamic panel data estimation method that was used to analyze the possible direct and indirect effects of FDI on EPI, our proxy for environmental sustainability.

Following the previous studies on the empirical relationship between FDI and environment (e.g., Baek and Koo 2009), given the data structure and availability, let variations in the environmental performance index (EPI) for country i, at time t be explained by several possible determinants modeled as:

$$EPI_{it} = f(X_{it} ,X_{i} ,X_{t} )$$

(1)

where f(.) is the functional relationship between EPI and its determinants. In practice, the chosen functional form is assumed to be linear or nonlinear. The vectors of X variables include those which are country- and time-variant variables, X_it, X_i, and X_t, which are country-specific and time-specific indicators, respectively. The latter is represented by country groups classified by income level and continental location, and time dummy (or time trend) variables.

The equation for the environmental performance index can be written by expressing EPI as a function of its determinants linearly (Model 1) and adding an error term ε_it:

$$EPI_{it} = \beta_{0} + \sum\limits_{j = 1}^{J} {\,\beta_{j} X_{jit} + \varepsilon_{it} }$$

(2)

where

$$\varepsilon_{it} = \mu_{i} + \lambda_{t} + \nu_{it}$$

(3)

where the vector X includes all three types of determinants classified by variability in both country and time, country, and time dimensions. The error term is a two-way error component containing country-specific and time-specific effects, as well as appended with a random error term. The country-specific and time-specific effects are effects considered fixed or random and to be estimated capturing heterogeneity in environmental effects between countries and over time, while the random error term following the tradition is assumed to have zero mean and constant variance unless accounting for possible heteroscedasticity and autocorrelation. The vector X includes determinants of variations in EPI, e.g., foreign direct investment (FDI), openness (OPEN), gross domestic product per capita (GDP per capita), clean energy share (CENE), unemployment rate (UNEM), the share of industry (SIND), and education index (EDUC).

Model 1 does not account for nonlinearity and interaction effects between the determinants. This can be accounted for by including squares and interrelationships of the X variables. The square captures nonlinearity in their effects on EPI, while interactions account for indirect substitution and complementarity effects depending on the sign of the coefficient. The generalized Model 2 assuming translog functional form is written as:

$$EPI_{it} = \beta_{0} + \sum\limits_{j = 1}^{J} {\,\beta_{j} X_{jit} + \sum\limits_{j = 1}^{J} {\,\beta_{jj} X_{jit}^{2} } + \sum\limits_{k = i}^{K} {\sum\limits_{m = 1}^{M} {\,\beta_{km} X_{kit} X_{mit} + } } \varepsilon_{it} }$$

(4)

where the squares and interactions of vectors of X do not necessarily include all variables listed above. The expected effects and substitution/complementarity between the variables will determine the form of the nonlinearity such that the functional form reflects the true nature of EPI and its determinants relationships. To avoid problems related to the endogeneity, heteroscedasticity, and possible autocorrelation and obtain more robust and efficient estimation results, the current study employs a dynamic panel estimation model. Model 2, described above, can be estimated by the two-step system GMM model with one lag and written as follows:

$$EPI_{it} = \beta_{0} + \beta_{1} EPI_{i,t - 1} + \sum\limits_{j = 1}^{j} {\beta_{j} X_{jit} } + \sum\limits_{j = 1}^{j} {\beta_{jj} X_{jit}^{2} } + \sum\limits_{k = i}^{k} {\sum\limits_{m = 1}^{M} {\beta_{km} X_{kit} X_{mit} } + \varepsilon_{it} }$$

(5)

Before regressing our dynamic panel data model (Eq. 5), some diagnostic tests were applied, i.e., nonlinear correlation test, spatial or cross-sectional dependence test, and unit root test. These tests are necessary for determining and proving the validity of the GMM model.

5 Empirical results

5.1 Specification and estimation test results

Before running a regression of the models in Eqs. (2) and (4), accounting for nonlinearity and interaction effects between the determinants, we first checked whether there is a linear correlation between the endogenous variable (EPI) and vector of key exogenous variables (determinants of EPI). The scatterplots showed the curvilinear relationship between endogenous variable and exogenous variables, meaning that the nonlinear correlation between EPI and its determinants should be investigated using nonlinear regression to obtain best-fit values of the parameters.

Next, to check the multicollinearity among the regressors, the study employed the correlation coefficient test. Tables 3 and 4 present the correlation coefficients between the dependent variable environmental performance index (EPI) and key independent variables FDI inflow, trade openness, GDP per capita, the share of clean energy, unemployment rate, the share of industry, education level, and time trend. It is desirable that the correlation between EPI and the explanatory variable is high, suggesting a higher explanation power, while the correlation between the explanatory variables should be very low, suggesting no serious collinearity problem resulting in confounded effects.

Trade openness, gross domestic product per capita, education levels, and trends are statistically positively correlated with EPI. However, the clean share of energy and the industry share of GDP is negatively correlated with EPI. Correlation coefficients between EPI and independent variables are computed pairwise and as such are unconditional on other changes. Unlike correlation coefficients, the regression coefficients reflect pairwise correlation relationships but are conditional on all other explanatory variables. As such, they are more important, but high correlation coefficients indicate the prevalence of collinearity problems and confounded effects.

To confirm the results of the Pearson correlation coefficient test, we also employed the variance inflation factor (VIF) test. Table 5 presents a low VIF value (less than 3) for each independent variable, meaning no multicollinearity between exogenous variables (Table 6).

Table 5 Pearson’s correlation coefficients, 1600 observations

Table 6 Variance inflation factor

The yearly average EPI level has grown during the period studied, but at a very low rate: It grew from 51.77 in 2000 to 52.48 in 2020. The sample descriptive statistics also show that EPI has not remained high. The average EPI for the period from 2000 to 2020 was only 55.52. Table 3 shows a summary of the means of the variables by target country groups. The average EPI varies by country’s income level. Mean EPI was higher in countries with higher level of income. For the period from 2000 to 2020, the average EPI score for the low-income country group is 47.440 and it follows by lower-middle-income country, upper-middle-income country, and high-income country group, i.e., 48.166, 52.038, and 63.584, respectively.

Before the construction of the panel data model, several diagnostic tests were performed to determine the validity of the estimated regression model.

First, the cross-sectional dependence tests were carried out to decide which generation of panel unit root test should be applied further for identifying whether the variables in the current study are stationary or not. In this paper, we performed three tests, including Pesaran’s cross-sectional dependence test, and Friedman’s and Frees’ tests of cross-sectional independence. The results of all three tests strongly rejected the null hypothesis of no cross-sectional dependence among the selected regressors at a 1% level of significance. Hence, we deployed a second-generation panel unit root test that accounts for possible cross-sectional dependence. The results of the latter panel unit root test are shown in Table 7.

Table 7 Pesaran’s CADF panel unit root test

The results of the cross-sectionally augmented Dickey–Fuller test show that some of the regressors have a unit root at the levels, whereas variables are found to be stationary at their first difference. Thus, in the two-step system GMM model, we employed variables that are first-order differenced.

5.2 Estimation results

The results based on estimation of the two-step system GMM model are presented in Table 8.

Table 8 Estimation results, 100 countries observed during 2000–2020

Similar to the results of the paper by Bernard and Mandal (2016), the coefficient of FDI inflow is insignificant in our model. The negative sign of the coefficient of FDI inflow was expected because of correspondence with the pollution haven hypothesis. Although some cross-country studies (Chakraborty & Mukherjee, 2013; Shahbaz et al., 2015a, 2015b; Singhania & Saini, 2021) and single-country case studies (Solarin et al., 2017, Bulut et al. 2021) have confirmed that such a relationship, our study using two-step system GMM model did not find a negative relationship between FDI inflow and environmental sustainability.

The positive signs of the coefficients of share of clean energy were found to be statistically significant at a 5% significance level. The positive effect of renewable energy was expected as it produces zero waste and reduces CO₂ emissions (Qi et al., 2014; Waheed et al., 2018; Saidi & Omri 2020; Szetela et al., 2022). Hence, it seems that by reducing CO₂ emissions in the sample countries, the share of clean energy increases the overall environmental performance index worldwide.

According to some studies, education is one of the important factors that affect environmental sustainability as it may provide and raise peoples’ awareness of a variety of environmental issues that people face day by day. Zafar et al. (2020) suggests that to ensure environmental sustainability, countries should invest more in education and renewable energy sectors. However, in our study, the coefficient of education has a positive sign but with no statistically significant impact on EPI.

The log of per capita GDP and openness was also statistically insignificant according to our dynamic panel data model. The coefficient of unemployment is negative and significant in the current study. Such a relationship between the two variables can be due to the decrease in income and pro-environmental behavior.⁵

According to the pollution haven hypothesis, a negative nexus between the share of industry and EPI was predicted. The results presented in Table 8 show a positive sign of the share of industry but is statistically insignificant.

Although FDI inflow was not found to have a significant relationship with EPI, the coefficient of squared FDI inflow was found to have a statistically positive sign at a 5 percent level of significance. The U-shaped correlation between FDI inflow and EPI indicates that a high level of FDI inflow worsens environmental quality first and improves it later as the FDI inflow passes beyond the turning point. We suggest that such nexus can be due to the promotion of “dirty” FDI from developed countries that decrease environmental quality until the critical point, and the promotion of “green” FDI after the countries reach a certain inflow of FDI.

The indirect effect of FDI inflow on environmental sustainability is estimated through the nexus between the interaction of FDI inflow and trade openness and EPI. The results of the dynamic panel data model show FDI inflow decreases environmental quality when the countries increase the level of trade openness.

The results of the diagnostic tests reported in Table 8 suggest that all instruments used in estimating the impact of FDI inflow on EPI are valid since we accept the null hypothesis of both Sargan and Hansen tests. Moreover, the Arellano–Bond test for AR(2) did accept the null hypothesis of no autocorrelation. Diagnostic tests show that our model is appropriate and confirms the robustness of our results.

6 Conclusion and policy recommendations

In the current paper, we examined the factors that influence EPI. The period of observation is from 2000 to 2020, and a sample of 100 countries, for which the data for both endogenous, exogenous, and control variables were provided, was used for the empirical analysis. Our sample of countries was divided into four groups, including low-income, lower-middle-income, upper-middle-income, and high-income countries. Preliminary tests were applied, e.g., test for a nonlinear correlation between variables, spatial cross-sectional dependence, and panel unit root. The results of the diagnostic tests proved the validity of the two-step system GMM model.

The findings of our analysis do not support the hypothesis of pollution haven. The findings indicate that there is no significant nexus between FDI and EPI; thus, we cannot tell exactly whether FDI inflow is bad or good for environmental sustainability.

The positive sign of the coefficients of clean energy and squared FDI inflow was found in the GMM model results. According to the results of our model, a 1.0 percent growth in renewable energy consumption leads to about a 0.4 percent increase in the environmental performance index globally. The positive effect of renewable energy was expected as it produces zero waste and reduces CO2 emissions according to some previous studies. The statistically positive sign of the coefficient of squared FDI inflow shows evidence of a U-shaped correlation between FDI inflow and EPI. We assume that such nexus between FDI inflow and EPI can be due to the promotion of “dirty” FDI from developed countries that decreases environmental quality until the turning point, and the promotion of “green” FDI that raises the environmental quality after the countries reach a certain inflow of FDI.

The findings of the paper suggest that to increase environmental sustainability globally, it is important that governments invest more in renewable energy projects as clean energy can be one of the most efficient solutions in reducing the impact of climate change. Given significant investments and a future-ready infrastructure, renewable energy projects are one of the main solutions in reducing emissions worldwide and addressing climate change. Moreover, even though FDI brings up economic growth and wealth of the nations, it is important to attract “greener” FDI rather than “dirty” FDI that creates jobs in developing countries.

7 Future research directions and limitations

Neither the correlation matrix nor the estimation results of the paper showed a statistically significant correlation between FDI inflow on EPI. Such findings do not support the pollution haven hypothesis. Such insignificant coefficient can be due to aggregate country level. By examining FDI by economic sectors, e.g., FDI by services, by industry, and by agriculture, considering Green FDI and other relevant determinants or controls might provide better results. Due to the inaccessibility of such data at the sector level, we cannot conduct it in the present study but must rely on aggregate country-level data.