Testing the Pollution Haven Hypothesis with the Role of Foreign Direct Investments and Total Energy Consumption

Abstract

The main objective of this study was to examine the nonlinear relationship between environmental deterioration and foreign direct investment for subpanels based on the country’s income level. In this study, the model’s determinants were total consumption of energy and electricity consumption, the share of renewable energy, and economic growth. Due to the observation of cross-sectional dependence, utilization of cointegration tests and panel data unit root were incorporated, which confirmed a mixed integration order. For the compliance of long-run and short-run relationships among the variables, a pooled mean group estimator panel auto-regressive distributed lag approach was incorporated. The results of long-run development support the pollution haven hypothesis; hence, ecological footprint is increased by the activities related to foreign direct investments. The obtained findings depend on the different subpanels based on the income level of countries. For the assurance of economic development sustainability in the energy sector, along with the electrical energy sector, customized policymaking is suggested by this study based on the particulars of each subpanel.

1. Introduction

The global economies have experienced a significant evolution in the areas of social reforms, trade openness, specialization in specific zones/products, foreign direct investments, and the subsequent development in the economy, which also leads to the higher generation of waste, increased total utilization of energy, and high level of pollution, although economic development and preservation of the environment are both essential for the concept of sustainable development. The Sustainable Development Goals (SDGs), which comprise combating the changing climatic conditions and ensuring the accessibility of economical and clean energy for all, were endorsed in 2015 to achieve these objectives (United Nations, 2020). However, the observed trends from 1996–2000 to 2011–2015 indicate that average values of ecological footprint (EFP, hereafter) per capita, Gross Domestic Product (GDP, hereafter) per capita, consumption of energy per capita and GDP net trade flows have increased by 10.5%, 96.8%, 16.7%, and 149.6%, respectively, in average values of 5 years throughout the world. The emissions of carbon dioxide have the most significant contribution to climate change. The 5-year average value of the emissions of carbon dioxide (CO₂), which has the most significant contribution to climate change [1], raised from 2.04 to 3.02 metric tons in the same period per capita. Therefore, increased environmental degradation in these decades has occurred due to the consumption of energy and significant economic growth. This is the reason that the deterioration of the environment should be considered among the most crucial problems to be resolved in this era.

For the last two decades, the production of carbon dioxide has soared, which is the main suspect in greenhouse gas emissions as claimed by the Intergovernmental Panel on Climate Change (IPCC). To the published report by IEA (2011), more than 60% of the content of greenhouse gas is due to the carbon dioxide concentration present in the environment. The CO₂ emissions are caused by various factors, though it has been confirmed by a few studies by Lin and Jia [2] that the most contributing factors in the production of CO₂ emissions are economic growth and usage of non-renewable energy. The fundamental sources of greenhouse gas emissions are fossil fuel consumption and economic growth [3].

To decrease the emission of greenhouse gases in the atmosphere, the adoption of the Kyoto Protocol, the Paris Agreement, and the United Nations’ Sustainable Development Goals (SDG-7, SDG-11, SDG-13, and SDG-15) have been incorporated. The increased efficiency in energy, afforestation, clean cities, and sustainable growth is a must to achieve by 2030, as suggested by Sustainable Development Goals (SDGs) 7–15. The struggles against environmental change have been a big success for some countries, especially in the last few years (SDG-13). Still, much work is needed to determine the effect of human participative economic activities on the quality of the environment in the existing literature. The majority of the current literature describes the link between growth, consumption of energy, trade openness, FDI, economic development, and degradation of the environment. At the same time, only a few focus on the environment, energy consumption, globalization, and degradation of the environment.

Moreover, the existing literature ignored the asymmetries present in the series while only giving importance to mentioned indicators’ symmetric impact on the environment. A few studies involving asymmetric analysis confirm that the decomposition of exogenous variables differently affects the endogenous variables [4,5]. Considering this logic, the current study determines the asymmetric impact of energy. The existing literature [3,6,7,8,9] explores the connection between the quality of the environment (CO₂ emissions) and the macroeconomic variables (i.e., consumption of energy, economic growth, trade openness, FDI, and other financial factors). At the same time, the environmental factors (environment, energy consumption, income, and globalization) have been focused on by only a few researchers. Due to the consideration of these particular environmental quality factors, this research study includes these factors in an empirical model [9].

This research study summarizes the greenhouse gas emission factors included in the previous literature while examining environmentally friendly factors such as energy consumption, environment, income, and globalization. According to the investigation of a few studies, uncertain results have been obtained regarding the connection between different elements and the degradation of the environment. Governments worldwide are encouraged to enhance the quality of the environment by the Sustainable Development Goals of the United Nations. To attain this objective, states worldwide are making efforts to impose restrictions on the usage of non-renewable energy sources.

Various other studies have investigated the environmental Kuznets curve (EKC) hypothesis validation in a similar context [10,11]. A few studies have confirmed the U-shape of EKC [12,13]. There is an increment in the EKC slope over time, as claimed by a few academic works [14], while studies such as Rehman and Rashid [15] claim statistical insignificance in the relationship between economic growth (EG) and degradation of the environment.

While considering the link between energy consumption and the quality of the environment, the majority of the existing literature claims that the quality of the environment has been improved by increasing the energy consumption through the utilization of non-renewable energy sources and technological advancement [6]. Another important factor in the environment is globalization, which includes various dimensions such as globalization in politics, economy, and social globalization. Globalization is indicated by trade liberalization according to a few studies [16,17], in which it has been discovered that emissions of carbon dioxide are increased by trade openness. At the same time, a few other studies claim the opposite of it [18]. FDI is also considered an indicator of globalization. Shahbaz and Nasreen [14] claimed that FDI reduces carbon emissions. However, a few other studies claimed the opposite [19,20,21]. Destek and Okumus [22] confirmed that the Kyoto Protocol has reduced carbon emissions. Furthermore, this study incorporates three different forms of globalization and claims that environmental quality is enhanced by political globalization. In contrast, the other two dimensions, i.e., economic globalization and social globalization, tend to decrease the quality of the environment.

Many research studies were incorporated to test an inverted U-shaped EKC existence to determine the relationship between environment and EG. Specifically, foreign investments are utilized by the developing countries to control their financial challenges in the way of infrastructure investments so that they can achieve their EG objectives [21]. With the help of FDI, the developing countries can benefit from the transfer of cleaner technology, positive externalities benefit, direct capital financing, techniques of innovative production resulting in increased productivity, and the expertise of fresh managerial skills that can lead to acceleration in EG [23]. However, the host countries benefit from increased EG and FDI inflows in various forms, but there might be a possibility that the environment is damaged by their activities [24]. The host country’s climate is affected by FDI inflows via three means: the effect of scale, the effect of composition, and the technique effect [25]. The utilization of natural resources and other inputs is higher in the initial phase of economic development, which becomes the cause of the high level of GHG emissions, higher rate of depletion of natural resources, and resulting by-products that further damage the natural environment. This is why decreasing the deterioration of the environment while backing the host countries’ firms with the help of positive scale effect reveals that the inflows of FDI contribute not only to EG but also to the further degradation of the environment. The suggestions by the impact of composition are that the host country’s industrial design is affected by the FDIs.

In contrast, its effects on the changes in the environment lie in the industrial sectors’ stake in the composition [26]. Ultimately, cutting-edge technology diffusion and relocation may be caused by FDI and strict regulations. FDI can cause cutting-edge technology diffusion along with stringent environmental regulations, which can be termed the technique effect [27].

The introduction of the Kyoto Protocol in 2005 involves the countries of developed nations because of their relevant proficiencies and because they are the main culprits in producing harmful atmospheric emissions. This protocol requires that the emissions produced by the developed nations should be reduced up to a pre-determined level, which helps in incentive creation for developed nations which they can utilize in outsourcing their high pollution-intensive foreign investments, although low EFP level countries have correspondingly low levels of FDIs [28,29]. If the effects of composition and scale dominate the technique effect, then the host country’s environment is negatively impacted by FDI. The pollution haven hypothesis (PHH, hereafter) eventually defines the FDI’s impact on the deterioration of the environment [30,31]. It asserts that the multinational companies (MNC, hereafter) having pollution-intensive operations would be interested in the under-developed nations having poor regulations of the environment as they would be willing to outsource the pollution-intensive productions so that they can save themselves from the costs incurred by the adaptation of the regulatory requirements, the abundance of natural resources, and comparatively cheap man-power [31,32,33,34,35]. Due to this reason, the developing nations do not possess the necessary development means try to attract foreign investments by providing them relief from the legislative requirements related to the environment [28]. Moreover, the environmental sensitivity among the less developed nations is not very significant because of the intense poverty and the importance given more towards EG than the protection of the environment. At the same time, the individuals of developed countries prefer to live in a cleaner environment and take environmental degradation seriously. Moreover, the less developed countries choose to adopt a transition from the former agricultural phase to industrialization because of the various economic development phases. In contrast, the developed nations prefer to shift their focus to the service side rather than giving attention to sectors of high industrialization. Due to these changes in their preferences, the less developed nations contribute more towards environmental degradation while the developed countries take steps to decrease it [36].

In contrast to this, the pollution halo hypothesis (P-HH, hereafter) claims that multinational companies can play their part in decreasing the damage to the environment in host countries by introducing processes which are environmentally friendly and adopting clean technologies according to the internationally accepted environmental regulations [17,37]. The domination of the scale effect over the other two results is required for this hypothesis. Due to the investment of developed countries in less developed nations, the manpower over there will become more qualified and have specialized training, skills, and the latest information [38]. In addition to this, as the MNCs take their reputation and associated goodwill seriously in this competitive world, they will not risk it by causing further environmental damage [26]. Conclusively, the host countries get exposure to innovative production techniques and new inputs because of the involvement of foreign investments [39].

This study aims to assess the effect of FDI inflows on environmental deterioration by utilizing a heterogeneous dataset involving 80 countries from 1990 to 2014, further separated into income-level based subpanels. An investigation has also been done on the effect of EG, energy consumption, and the ratio of renewable energy in total consumption on the environment. Even though PHH and P-HH assume a linear association between FDI and environmental deterioration, FDI inflows are incorporated into the model in their level and squared forms to allow for nonlinear relationships. As a comprehensive index and an environmental degradation proxy, the model has included EFP. Even though CO₂ emissions have been incorporated as an indicator in various studies of the same regard, all the regions of the degradation of the environment are not targeted by it [21]. It considers the cross-sectional dependence to refrain from making estimations based on biasness, and panel unit root tests were used to understand the series integration levels. The long-term relationships between the model’s variables have been examined with the help of panel co-integration approaches. Pooled mean grouped (PMG, hereafter) panel ARDL approach was incorporated to determine the short-term and long-term impact of FDI, FDI2, EG, renewable energy usage share in total consumption, and consumption of total energy on EFP. Results of the empirical study support PHH and that every subpanel with different income levels requires tailored policies.

The rest of the paper is arranged as follows. The next section is dedicated to a literature review. In Section 3 of this paper, an explanation of the model, data, and econometric methodology is provided, while in Section 4 of this paper, the demonstration of empirical findings is provided. The conclusion of this study gives a few recommendations for policy.

2. Review of Literature

The different factors of environmental degradation are potentially impacted by FDI, thorough investigations of which have been carried out in this literature. The different concluding results obtained might be due to the selection of country, period, environmental degradation indicator, and the choice of econometric methodology. The studies related to the FDI–environmental damage connection are shown in

Table 1. Review of relevant studies.

No	Author(s)	Sample Period	Economies/Country(ies)	Variable	Technique	Hypothesis
1	Abdouli et al. (2018) [24]	1990–2014	BRICS countries	CO2	Ordinary least squares (OLS), fixed effects (FE) model, random-effects model (RE) and system generalized method of moments (GMM)	P-HH
2	Acharyya (2009) [42]	1980–2003	India	CO2	OLS	PHH
3	Al-Mulali and Tang (2013) [17]	1980–2009	GCC Countries	CO2	Fully modified OLS (FMOLS) and Granger causality tests	PHH is rejected
4	Aliyu (2005) [43]	1990–2000	14 non-OECD Countries	CO2	Panel OLS	PHH is rejected
5	Aliyu and Ismail (2015) [44]	1990–2010	19 African countries	CO2	Pooled mean group (PMG)	Neutrality
6	Asghari (2013) [36]	1980–2011	MENA Region	BOD	Panel OLS (FE and RE)	P-HH
7	Baek (2016) [45]	1981–2010	5 ASEAN Countries	CO2	Pooled mean group (PMG)	PHH
8	Bakirtas and Cetin (2017) [32]	1982–2011	MIKTA Countries	CO2	Panel Vector Auto Regressive Model (PVAR)	PHH
9	Baloch et al. (2019) [46]	1990–2016	59 Belt and Road Initiative (BRI) Countries	EFP	Panel data regressions with Driscoll–Kraay standard errors	PHH
10	Behera & Dash (2017) [19]	1980–2012	17 countries in south and southeast Asia (SSEA) region	CO2	FMOLS, dynamic OLS	PHH
11	Chandran and Tang (2013) [47]	1971–2008	ASEAN—5 economies	CO2	Johansen cointegration, VECM Granger causality	Mixed
12	Destek and Okumus (2019) [22]	1982–2013	10 newly industrialized countries	EFP	Error correction based cointegration tests	Non-linear relation
13	Gokmenoglu and Taspinar (2016) [48]	1974–2010	Turkey	CO2	ARDL error correction model PDOLS	PHH
14	Gorus and Aslan (2019) [49]	1980–2013	MENA Region	CO2	Dumitrescu–Hurlin non-causality	Mixed
15	Kivyiro and Arminen (2014) [50]	1971–2009	6 sub-Saharan African countries	CO2	ARDL, Granger VECM causality, and Maki structural break cointegration test	Mixed
16	Kocak and Sarkgunesi (2018) [51]	1974–2013	Turkey	CO2	Stock and Watson dynamic OLS Hacker and Hatemij bootstrap causality	PHH
17	Liu and Kim (2018) [52]	1990–2016	44 BRI countries	EFP	PVAR	P-HH
18	Merican et al. (2007) [53]	1976–2002	5 ASEAN Countries	CO2	ARDL, unrestricted error-correction model (UECM)	Mixed
19	Mert and Boluk (2016) [33]	2002–2010	21 Kyoto countries	CO2	Panel ARDL, Panel VECM Causality	P-HH
20	Mert and Caglar (2020) [34]	1974–2018	Turkey	CO2	Crouching ECM, VECM	P-HH
21	Omri et al. (2014) [40]	1990–2011	54 countries	CO2	GMM	PHH
22	Pazienza (2015) [54]	1981–2005	30 OECD countries	CO2	Panel OLS (FE and RE)	P-HH
23	Rafindadi et al. (2018) [55]	1990–2014	GCC Countries	CO2	PMG	P-HH
24	Sarkodie and Strezov (2019) [35]	1982–2016	China, India, Iran, Indonesia, and South Africa	CO2	Panel data regressions with Driscoll–Kraay standard errors	P-HH
25	Seker et al. (2015) [56]	1974–2010	Turkey	CO2	Hatemi-J cointegration test, ARDL, Granger VECM Causality	PHH
26	Shaari et al. (2014) [57]	1992–2012	15 developing countries	CO2	FMOLS, Granger VECM causality	Neutrality
27	Shahbaz et al. (2015) [14]	19975–2012	99 countries	CO2	FMOLS	PHH
28	Shahbaz et al. (2018) [58]	1955–2016	France	CO2	The bootstrap ARDL	PHH
29	Shahbaz et al. (2019) [4]	1990–2015	17 MENA countries	CO2	GMM	PHH
30	Solarin and Al-Mulali (2018) [28]	1982–2013	20 developing and developed countries	CO2, EFP	Westerlund panel cointegration test Augmented meean group estimator (AMG), STIRPAT Model	Mixed
31	Solarin and Al-Mulali (2018) [28]	1980–2012	Ghana	CO2	ARDL	PHH
32	Sun et al. (2017) [20]	1980–2012	China	CO2	ARDL	PHH
33	Tang and Tan (2015) [41]	1976–2009	Vietnam	CO2	Johansen cointegration, VECM Granger causality	PHH
34	Udemba [37]	1974–2017	Turkey	EFP	Structural break analysis, ARDL Bound testing, Granger causality	PHH
35	Zafar et al. (2019) [39]	1970–2015	United States of America	EFP	ARDL	P-HH
36	Zhang and Zhou (2016) [59]	1995–2010	29 provinces in China	CO2	Pedroni cointegration test, STIRPAT	P-HH
37	Zhu et al. (2016) [60]	1981–2011	5 ASEAN-5 economies	CO2	Panel quantile regression	P-HH

Source: Compiled by authors.

. A few important points must be declared after the examination of this table. First, although the emissions of CO₂ as an environmental degradation indicator have been considered by the majority of the studies [17,21,24], few studies have utilized EFP to indicate the factor of environmental degradation [22,37,39,40]. To find the connection between FDI and ecological degradation, few studies focus on the country groups [17], and others examined individual countries [41].

The most important difference between the studies is the contradictory conclusions. While there exist studies defending PHH, such Acharyya [42,56], a few other studies have also supported P-HH [19,32,41,45,55]. Some works reach neutral conclusions about the FDIs and environmental degradation relationship [43], while others find mixed results [53]. The obtained mixed results might be due to the selection of country, research period, length of data, employed econometric methods, and the model’s variables. According to the paneling patterns and the sample, the study by Shahbaz and Nasreen [14] is considered a pioneer because of distinguishing the countries according to their income level. The obtained findings concluded that PHH-related patterns are represented by middle and low-income countries, whereas the P-HH represents high-income countries.

3. Materials and Methods

Understanding the environmental impact of FDI is the prime aim of this study by distinguish the countries according to their level of income as classified by the World Bank for the phase 1990 to 2014. The influences of FDI, EG, consumption of total energy, and renewable energy resources shared in total consumption on EFP have been investigated to serve this objective. The structure made by Shahbaz and Nasreen [14] is incorporated, and for the determination of the FDI and EFP non-linear relationship, the model included the square of FDI. Shahbaz and Nasreen [14] utilized the emissions of CO₂ on a per capita basis. To determine other environmental damage factors with the inclusion of all emissions of GHG, bio-capacity, and natural resources depletion, Zafar and Zaidi [39] and Abdo and Li [27] are followed. EFP has been utilized as the proxy of environmental damage in this study.

$$lef{p}_{it}=a_0+a_{1}lgd{p}_{it}+a_{2}lfd{i}_{it}+a_{3}lfd{i}_{it}^2+a_{4}le{n}_{it}+a_{5}lre{n}_{it}+u_{it}$$

(1)

In this model, the EFP natural logarithms per capita in global hectares are indicated by $lef{p}_{it}$, and $lgd{p}_{it}$ represents per capita Real GDP for country $i$ at time $t$, respectively. For the indicator of EG, per capita Real GDP was utilized, following Al-Mulali and Tang [17] and Shahbaz and Nasreen [14]. Following Baltagi [61], specifically for 20–30 years macro panels, the cross-sectional dependency problem should not be ignored. The dependence signifies a contemporaneous correlation which causes biases in obtained results, and inconsistency in the estimations may occur due to the increased integration [62]. For this analysis, the Pesaran and Smith [63] cross-sectional dependence (CD, hereafter) test is suitable, keeping in mind the cross-section dimensions, $N$, which is 2000, and period, $T$ is 24. As one of the fundamental assumptions for the panel data analysis is the freedom across cross-sections, the Pesaran CD test examines the correlation between entities, such as countries in this study. Pesaran and Smith [63] is being proposed for balance panels.

3.1. Data

$lfdiit$ and $lfdi2$ represent the FDI inflow’s natural logarithm and its square, respectively, for country $i$ at time $t$. Energy use per capita (kilotons of oil equivalent) has been utilized for the representation of the consumption of energy following Baek [45] and Destek and Ulucak [21]. The use of total energy is transformed in a natural log form and hence denoted by 𝑙𝑒𝑛𝑖𝑡, whereas l𝑟𝑒𝑛𝑖𝑡 represents the share of renewable energy in total consumption of energy. 𝑢𝑖𝑡 is the term meaning error.

Table 2. Descriptive statistics (before natural logarithm transformation).

Measure	EFP Per Capita (In Million Hectares)	Real GDP Per Capita (USD)	Foreign Direct Investment, Net Inflows (BoP, Million USD)	Energy Use (kg of Oil Equivalent Per Capita)	Renewable Energy Consumption (% of Total Final Energy Consumption)
Mean	4.77	383,239 million	1310 million	2436.76	28.13
Median	3.12	8547 million	135 million	1371.36	20.33
Maximum	432.33	562 million	734 million	15,108.65	95.04
Minimum	0.46	323.64 million	296 million	118.90	0
Std. Dev.	17.97	129 million	408 million	2539.54	25.54
Skewness	17.72	41.43	7.13	1.82	0.8
Kurtosis	339.32	1776.34	77.67	6.71	2.5
Obs.	2000	2000	2000	2000	2000
Source	WDI	WDI	WDI	WDI	WDI

Note: Source: Authors’ calculations based on World Development Indicators (WDIs).

demonstrates the descriptive statistics before the transformation of the variables’ natural logarithm (ln). By incorporating the framework of this model, the hypothesis validation explains the association between FDI and deterioration of the environment along with the impact of EG, consumption of energy, and the usage of renewable energy on the degradation of the environment can be investigated separately incorporating data of 80 countries based on global panel and subpanels of income level.

In the essence of the matter, $a_2$ and $a_3$ are the FDI coefficients and FDI squared term, while PHH or P-HH would be declared by distinct signs of these coefficients. The hypothesis of EKC claims that with the rise in EG, degradation of the environment also increases up to a turning point, after which it appears to decrease. Therefore, the positivity of $a_1$ is anticipated. $a_4$ and $a_5$ would have positive and negative signs because, with the increase in the consumption of total energy, emissions of pollutants will also be increased. In contrast, they would be decreased by the energy sources share in total consumption.

3.2. Cross-Sectional Dependence (CD) and Unit Root Tests

According to Baltagi (2008) [61], specifically for 20–30 years macro-panels, CD is considered an issue that should be taken into account. This dependency highlights the contemporaneous correlation and it may cause biases and inconsistency in estimations because of the increased integration [62]. The Pesaran and Smith [63] CD test is most suitable for this analysis while keeping in mind the cross-sectional dimension N, which is 2000, and period, T is 24. As one of the fundamental assumptions for the panel data is the independence across cross-sections, the CD test by Pesaran will examine the correlation existence across the entities; as countries in this study, Pesaran and Smith [63] proposed the following formula for the balanced panels.

$$CD=\sqrt{\left(\frac{2T}{N\left(N-1\right)}\right)}\left({\displaystyle \sum }_{I=1}^{N-1}{\displaystyle \sum }_{J=I+1}^N\widehat{p_{ij}}\right)$$

(2)

where,

$$\widehat{p_{ij}}=\widehat{p_{ij}}=\frac{{{\displaystyle \sum }}_{T=1}^T\widehat{u_{it}}\widehat{u_{jt}}}{{\left({{\displaystyle \sum }}_{t=1}^T\widehat{u_{it}^2}\right)}^{1/2}{\left({{\displaystyle \sum }}_{t=1}^T\widehat{u_{jt}^2}\right)}^{1/2}}$$

(3)

Here, the residual’s pair-wise cross-sectional coefficient of correlation is represented by $\widehat{p_{ij}}$ as obtained from an augmented Dickey–Fuller (ADF) test regression. The null hypothesis is no CD. The CD statistic mean is zero for N and fixed T values regarding various models of panel data

$$CD \to N\left(0,1\right) for N \to \infty and sufficiently large T.$$

(4)

As the observation of CD is across the countries, distorted results were obtained by first-generation unit root tests. Pesaran’s (2007) [64] second-generation unit root cross-sectionally augmented IPS (CIPS, hereafter) test was utilized to determine whether the series is stationary. As CIPS test depends on the average of the Dickey–Fuller (DF, hereafter)/ADF t-statistic of each panel unit. For the consideration of CD across entities, incorporating the cross-sectional averages of lagged terms and first differences of the series into the standard DF/ADF regressions was done. The cross-sectional ADF calculation is as follows:

$$∆y_{it}=ai+biyit-1+ci\overline{y}t-1+di∆\overline{y}t+eit$$

(5)

where,

$$\overline{y}t-1=N-1 \sum Ni=1 yit-1$$

(6)

and

$$∆\overline{y}t=N^{-1}{{\displaystyle \sum }}_{i=1}^N{t}_{i}yit$$

(7)

$$CIPS\left(N,T\right)=t-bar=N^{-1}{{\displaystyle \sum }}_{I=1}^{N}ti\left(N,T\right)$$

(8)

Hence, CIPS was determined from every variable’s mean cross-sectionally augmented Dickey–Fuller (CAFD, hereafter) statistics. 𝑎𝑖 is the acceptable term and (𝑁, 𝑇) represents the CADF statistic for the ith cross-section unit, obtained with the help of 𝑦𝑖𝑡−1 coefficient’s t-ratio in the CADF regression shown by Equation (5). The cross-sectional mean of 𝑦𝑖𝑡 is represented by $\overline{y}t$ and for the consideration of contemporaneous correlations among 𝑦𝑖𝑡, $\overline{y}t$ was incorporated into the equation. Different terms’ critical values were provided by Pesaran [64]. We can define the null hypothesis of the test as the panels being non-stationary because of the resultant unit-roots contained by all series. In contrast, according to the alternative hypothesis, the panels are stationary.

$$H_0=b_i=0 for all i.$$

(9)

$$H_1=b_i<0 for all i.$$

(10)

3.3. Cointegration Tests

After determining the order of integration of individual series, another modeling issue is checking the existence of long-run among the series by using cointegration tests. The applied tests are beneficial in determining the long-run, and short-run dynamics information for each entity is also allowed and followed by Behera and Dash [19].

$$G_{\tau }=\frac{1}{N}{\displaystyle \sum }_{i=1}^N\frac{\widehat{a_i}}{SE\left(\widehat{a_i}\right)}$$

(11)

$$G_{\alpha }=\frac{1}{N}{\displaystyle \sum }_{i=1}^N\frac{T\widehat{a_i}}{\widehat{a_i}\left(1\right)}$$

(12)

Westerlund’s [65] error-correction based panel test of cointegration is utilized for examining the cointegration relationships between the entities. Heterogeneity robust bootstrapping p-values were obtained by this cointegration test, which is helpful in the existence of CD, and it comprises four separate cointegration tests. The group statistics are $G_{\tau }$ and $G_{\alpha }$ whereas the panel statistics are $P_{\tau }$ and $P_{\alpha }$. The individual units’ error correction was checked by the group statistics while the panel’s error correction was determined by panel statistics. These four tests comprise normal distribution and are based on structural dynamics instead of residual. Due to this reason, the absence of common factor restrictions is evident in the tests.

According to the null hypothesis, cointegration exists across the units of cross-section or the total panel, which has the error correction term condition required for all statistics involved to be zero. The error correction term is statistically significant in the error correction model (ECM, hereafter) of the constrained panel investigated in this cointegration test.

$$\widehat{a_i}\left(1\right)=1-{{\displaystyle \sum }}_{j=1}^{p_i}\widehat{a_{ij}}$$

(13)

$$P_{\tau }=\widehat{\frac{a}{SE\widehat{\left(a\right)}}}$$

(14)

$$P_{\alpha }=T_{\widehat{a}}$$

(15)

According to the framework of this study, after experiencing a shock, the system returns to its equilibrium state, which is represented by an adjusted velocity. The error correction term is included in the formation of ECM to represent that adjusted velocity. Cointegration is represented by, in addition to Westerlund’s error-correction-based cointegration test [65,66], Pedroni’s [67] cointegration residuals-based test in the examination of long-run relationships between the entities. A dual-step regression framework was adopted to examine the panel cointegration in this test. Initially, Equation (21)’s estimation is done, and obtained residuals are noted. For the inclusion of independent and dependent variables in this test, both must be integrated of order one.

$$∆y_{it}={\delta }_i^{’}{d}_t+{\alpha }_i\left(y_{it-1}-{\beta }_i^{’}{x}_{it-1}\right)+{\displaystyle \sum }_{j=1}^{p_i}{\alpha }_{ij}∆y_{it-j}+{\displaystyle \sum }_{j=1}^{p_i}{\gamma }_{ij}∆x_{it-j}+e_{it}$$

(16)

$${\lambda }_i^{’}=-{\alpha }_i{\beta }_i^{’}$$

(17)

Here, ${\delta }_i^{’}=\left({\delta }_{i1},{\delta }_{i2}\right)’$ are the parameters’ vector, $d_t$ represents the deterministic term, where

$$d_t=\left\{\begin{array}{c}0, no deterministic term.\\ 1, constant.\\ {\left(1,t\right)}^{’}, both constant and trend.\end{array}\right.$$

(18)

is considered as a dependent variable while representing the independent variables are the residuals.

$$y_{it-1}-{\beta }_i^{’}{x}_{it-1}=0$$

(19)

$$a_i=\left\{\begin{array}{c}<0, cointegration.\\ =0 no cointegration.\end{array}\right.$$

(20)

The test’s null hypothesis claims no cointegration existence following this claim; the residuals need to be integrated of order 0, I(0), so that they can be stationary. The following panel regression framework was utilized to check if they are stationary or not.

$$y_{it}=a_i+{\delta }_{1i}t+{\beta }_{1i}{x}_{1i}t+{\beta }_{2i}{x}_{2i}t+\cdots ++{\beta }_{Mi}{x}_{Mi}t+{\epsilon }_{it}$$

(21)

Here, the dependent variable is 𝑦𝑖𝑡,𝑡 represents the independent variables whereas residuals are represented by 𝜀𝑖𝑡. No cointegration existence is claimed by the null hypothesis and in light of this claim, the residuals need to be integrated of order 0, I(0), so that they can be stationary. A panel regression framework was formed to check if they are stationary or not. The second group of statistics is called between-dimension tests and comprehends group-statistic, group Phillips–Perron (PP)-statistic, and group ADF-statistic.

$$∆{\epsilon }_{it}={\rho }_i{\epsilon }_{it-1}+{\mu }_{it}$$

(22)

$$∆{\epsilon }_{it}={\rho }_i{\epsilon }_{it-1}+{\displaystyle \sum }_{j=1}^{p_i}{\psi }_{ij}∆{\epsilon }_{it-j}+v_{it}$$

(23)

The residuals are $∆{\epsilon }_{it-j}$ according to Equation (21), the first difference, operator is denoted by ∆. The model’s residuals have the assumption of white noise and consist of normal distribution. Seven test statistics in two separate groups were suggested by Pedroni [66] for testing the cointegration. Within-dimension tests are the first group consisting of panel v-statistic, panel ρ-statistic, panel PP-statistic, and panel ADF-statistic.

3.4. Hausman Test and Pooled Mean Group (PMG)

The consistency of PMG estimators has been declared if the slope parameters across the panels have homogeneity in the constructed model; otherwise, the consistency of the mean grouped (MG, hereafter) estimator will be declared. Both models are dependent on the method of ARDL. To select the most suitable model, long-run homogeneity was examined by the Hausman test. The test’s null hypothesis is that differences among the estimations of PMG and MG estimators have no statistical significance, due to which the PMG estimators result in more efficient estimations. Following the obtained results, the rejection of the null hypothesis is not possible, hence, the PMG estimator was selected due to its efficient results under the circumstances. The construction of the PMG estimator was done to inquire about the series’ long-run and short-run causality relationships. Pesaran and Shin [68] were the first to introduce PMG, which is incorporated in estimating panel data’s cointegration equations. It is claimed to be the MG estimator’s modified version [63]. Between the MG and the dynamic fixed-effects model (DFE, hereafter), the framework of PMG is considered an intermediate model because the coefficients across the panels utilize pooling and averaging steps, which are incorporated in PMG. The claim made by this framework is that the determination of the dynamics of the short-run is essential for every variable, while maintaining the estimation of long-run impacts and the adjustment of long-run velocity. Thus, it can be claimed that the adjustment velocity, coefficients of short-run, and variances in error may be different, but the equivalency and homogeneity of the coefficients of long-run are evident across the model’s panels

The formulation of Equation (24) as an ARDL (p, q,…, q) model:

$$ef{p}_{it}={{\displaystyle \sum }}_{j=1}^p{\lambda }_{ij}ef{p}_{it-j}+{{\displaystyle \sum }}_{j=0}^p{\delta }_{ij}^{’}{x}_{it-j}+{\mu }_i+{\epsilon }_{it}$$

(24)

here, $x_{it}$ represents an independent variable’s vector (5 × 1).

$$x_{it}=\left[gd{p}_{it},fd{i}_{it},fd{i}_{it}^2,e{n}_{it},re{n}_{it}\right]$$

(25)

${\mu }_{it}$ represents the fixed effects of the entity while the error term is denoted by $\epsilon it$ which is considered to have an independent distribution among entities and time, i and t, while having a mean of zero and ${\sigma }_i^2>0$. It is considered that the distribution is independent of regressors, $xit$.

$$∆ef{p}_{it}={\phi }_i\left(ef{p}_{it-1}-{\alpha }_{0i}-{\alpha }_i^{’}{x}_{it}\right)+{{\displaystyle \sum }}_{j=1}^{p-1}{\lambda }_{ij}^{*}∆ef{p}_{it-j}+{{\displaystyle \sum }}_{j=0}^{q-1}{\delta }_{ij}^{*’}∆x_{it-j}+{\mu }_i+{\epsilon }_{it}$$

(26)

where,

$${\phi }_i=-\left(1-{{\displaystyle \sum }}_{j=1}^p{\lambda }_{ij}\right)$$

(27)

$$a_i=\left({{\displaystyle \sum }}_{j=0}^q\frac{{\delta }_{ij}}{{\phi }_i}\right)$$

(28)

$${\lambda }_{ij}^{*}=-{{\displaystyle \sum }}_{m=j+1}^p{\lambda }_{im},j=1,2,\mathrm{\dots },p-1$$

(29)

$${\delta }_{ij}^{*’}=-{{\displaystyle \sum }}_{m=j+1}^p{\delta }_{im},j=1,2,\mathrm{\dots },q-1$$

(30)

The short-run impact of each regressor is represented by various terms with lag in this model. Each regressor’s short-run influences are shown by different terms with lag variables in this model. The selection of p length is based on the criteria of information. 𝜑𝑖 is the adjustment velocity for 𝑒𝑓𝑝𝑖𝑡 after variation in 𝑥𝑖𝑡 suddenly. The long-run coefficient’s vector is represented by 𝛼𝑖. In Equation (26), the long-run impact is shown by the level of independent variables.

4. Results

Stepwise representation of results was followed. First, we present the cross-sectional dependence (CD) test results. The unit root test results are presented after that. Third, we offer the cointegration test results. Finally, we present optimal lag length determination, Hausman test, and Pooled Mean Group (PMG) estimation results.

4.1. Cross-Sectional Dependence (CD) Test

As biasness in results of both tests and estimations can be caused because of the existence of CD, checking of CD is done initially.

Table 3. The results of the Pesaran cross-sectional dependence (CD) test.

Variables	CD-Test	p-Value	Correlation	Abs(Corr.)
LEFP	28.37	0.00	0.10	0.39
LREALGDP	60.66	0.00	0.22	0.54
LFDI	121.06	0.00	0.43	0.46
LFDI2	149.42	0.00	0.53	0.54
LENERGY	59.28	0.00	0.21	0.51
LRENEWABLE	248.56	0.00	0.88	0.88

Source: Authors’ calculations.

demonstrates the Pesaran [69] CD test results. As cross-sectional independence is the null hypothesis of the test, the null hypothesis is declared as rejected for all the series. Therefore, it can be concluded that CD is present in all the series. Upon the observation of cross-dependence, the application of the Pesaran [69] second-generation unit root CIPS test was done to check the stationarity for global panels and subpanels based on income level. Unit root test results have been demonstrated separately in

Table 4. Second generation unit root test.

Global Panel
Variables	At Level				At 1st Difference
	Drift	p-Value	Drift and Trend	p-Value	Drift	p-Value	Drift and Trend	p-Value
LEFP	2.72	1	4.22	1	−7.68	0.00	−4.22	0.00
LREALGDP	−3.57	0.00	−3.43	0.00	−9.33	0.00	−4.30	0.00
LFDI	1.77	0.96	9.47	1	−6.80	0.00	−4.17	0.00
LFDI2	−2.57	0.01	1.11	0.87	−13.07	0.00	−8.64	0.00
LENERGY	3.78	1	7.33	1	−3.79	0.00	−1.63	0.05
LREN.	−0.30	0.38	−0.83	0.2	−12.70	0.00	−9.16	0.00
Low Income Panel
LEFP	0.45	0.67	−0.27	0.39	−3.65	0.00	−2.86	0.00
LREALGDP	0.47	0.68	−0.52	0.30	−2.64	0.00	−1.83	0.03
LFDI	−2.38	0.01	−2.78	0.00	−6.46	0.00	−5.92	0.00
LFDI2	−1.3	0.10	−0.85	0.20	−4.83	0.00	−3.99	0.00
LENERGY	−0.82	0.21	0.37	0.64	−2.41	0.01	−1.80	0.04
LREN.	−3.98	0.00	0.35	0.64	−4.03	0.00	−3.21	0.00
Lower-Middle Income Panel
LEFP	−0.81	0.21	1.44	0.93	−7.03	0.00	−6.15	0.00
LREALGDP	−1.90	0.03	−1.11	0.14	−6.72	0.00	−4.59	0.00
LFDI	−2.30	0.01	0.46	0.68	−8.81	0.00	−7.66	0.00
LFDI2	−4.78	0.00	−2.81	0.00	−10.14	0.00	−7.65	0.00
LENERGY	0.31	0.62	1.79	0.96	−6.83	0.00	−5.53	0.00
LREN.	1.96	0.98	3.59	1	−4.33	0.00	−2.44	0.01
Upper-Middle Income Panel
LEFP	−0.95	0.17	0.42	0.66	−8.31	0.00	−5.89	0.00
LREALGDP	−3.85	0.00	−5.29	0.00	−9.15	0.00	−7.10	0.00
LFDI	−4.14	0.00	−3.45	0.00	−10.87	0.00	−8.57	0.00
LFDI2	−5.92	0.00	−3.66	0.00	−12.28	0.00	−10.14	0.00
LENERGY	−0.51	0.31	2.70	1	−7.02	0.00	−4.95	0.00
LREN.	−1.55	0.06	−0.54	0.29	−7.25	0.00	−4.70	0.00
High Income Panel
LEFP	0.72	0.77	1.02	0.85	−10.53	0.00	−8.29	0.00
LREALGDP	−4.62	0.00	−2.45	0.01	−7.92	0.00	−4.64	0.00
LFDI	−0.90	0.19	0.37	0.64	−14.87	0.00	−12.51	0.00
LFDI2	−3.74	0.00	−2.70	0.00	−15.83	0.00	−12.85	0.00
LENERGY	−0.53	0.30	−1.33	0.09	−11.88	0.00	−8.24	0.00
LREN.	0.09	0.00	−1.91	0.03	−6.96	0.00	−5.86	0.00

Source: Authors’ calculations.

for drift and drift-linear trend cases of global, low-i, lower-middle-, upper-middle-, and high-income panels.

Upon considering FDI inflow’s natural logarithm and its squared form, 5 and 7 out of 10 tests correspond to the null hypothesis rejection of non-stationarity in the level form, respectively. Hence, at the level, stationarity is concluded. In the end, mixed results have also been obtained by the natural logarithm of real GDP’s unit root test. Following the 6 out of 10 test results, the non-stationarity of null hypothesis rejection is observed at a 1% level. From the examination of the tests having the involvement of drifts but no trends, 3 out of 5 tests conclude for the integration of series in order zero, I(0), which confirms stationarity.

4.2. Unit Root Test

The second-generation unit root test is used (Table 4). At the first difference, all series are stationary, and due to this, the EFP’s natural logarithm, consumption of energy, and the ratio of renewable energy are I(1). In contrast, the rest were I(0). For the cointegration tests, the above results were considered. To determine the long-run relation presence among I(1) variables, the second-generation [65] was applied.

The drift option can be considered if the series’ mean is non-zero, as indicated by the first-difference test results. Moreover, the trend option for the real GDP’s natural logarithm is not essential because it was converted into its “real” form. At the same time, for the rest, relevant results may be obtained with the utilization of the trend option. The EFP per capita’s natural logarithm, a dependent variable, is non-stationary in its level form, while at the first difference, it becomes stationary, hence the integration of the order one. Moreover, the natural logarithm of consumption of total energy per capita’s 9 out of 10 tests and the renewable energy’s natural log ratio in consumption of total energy’s 8 out of 10 tests revealed that they have an integration of order one.

4.3. Cointegration Test

Table 5. Westerlund ECM panel cointegration test results.

Models	Global Panel		Low-Income Panel		Lower-Middle Income Panel		Upper-Middle Income Panel		High Income Panel
Get	−2.80 c	0.00	−1.22	0.99	−2.79 a	0.08	−2.68	0.19	−2.97 c	0.00
Ga	−10.12	1	−6.99	0.90	−9.48	0.99	−9.06	1	−11.30	0.97
Pt	−25.41 c	0.00	−1.70	0.96	−10.49	0.28	−9.86	0.89	−22.61 c	0.00
Pa	−10.50	0.50	−7.02	0.77	−8.00	0.95	−8.47	0.92	−22.61 b	0.05

Note: ^a = Significant at 0.10 level; ^b = significant at 0.05 level; and ^c = significant at 0.01 level. Source: Authors’ calculations.

presents Westerlund ECM panel cointegration test results for global and four different income panels. Cointegration among the variables has been demonstrated by test statistics in the lower-middle-income panel. For panels of global and high-income, test statistics reflect the valuable presence of CD and the occurrence of the panel’s non-homogeneity. According to the EFP’s natural log test results, per capita, energy consumption, and renewable energy usage ratio are demonstrated in Table 5.

Pedroni panel cointegration test (2004) was also utilized to examine the variables’ long-run relations.

Table 6. Pedroni panel cointegration test results.

Models	Global Panel		Low-Income Panel		Lower-Middle Income Panel		Upper-Middle Income Panel		High Income Panel
Panel v-statistic	1.79 b	0.04	−0.76	0.78	2.00 b	0.02	4.55 c	0.00	−2.73	1
Panel σ-statistic	−8.24 c	0.00	−0.12	0.45	−3.51 c	0.00	−1	0.16	−9.26 c	0.00
Panel ρρ-statistic	−18.65 c	0.00	−0.81	0.21	−9.86 c	0.00	−1.99 b	0.02	−22.70 c	0.00
Panel adf-statistic	−22.39 c	0.00	−0.61	0.27	−6.39 c	0.00	−5.69 c	0.00	−22.10 c	0.00
Group σ-statistic	−0.65	0.26	0.34	0.63	−0.63	0.26	0.28	0.61	−0.81	0.21
Group ρρ-statistic	−14.83 c	0.00	−0.68	0.25	−8.23 c	0.00	−6.25 c	0.00	−10.98 c	0.00
Group adf-statistic	−12.08 c	0.00	−0.44	0.33	−6.50 c	0.00	−5.25 c	0.00	−9.01 c	0.00

Note: ^b = significant at 0.05 level; and ^c = significant at 0.01 level. Source: Authors’ calculations.

represents the test results. The demonstration of the effects of within-dimension and between-dimension test statistics revealed that except for the low-income panel, the no cointegration’s null hypothesis is not accepted in most cases. For the global, upper-middle-income, and high-income panels, 5 of 7 tests revealed the null hypothesis rejection at 5%. At the same time, the rejection of the null hypothesis appeared in 6 out of 7 tests in lower-middle-income panel results. Upon the comparison of these obtained results with test results of Westerlund cointegration, the existence of cointegration among variables of global and high-income panels can be claimed.

4.4. Optimal Lag Length Determination—AIC

Through Akaike Information Criterion (AIC), lag length for multiple panels of the countries was computed (see

Table 7. Optimal lag length selection—AIC.

Countries	LEFP	LREALGDP	LFDI	LFDI2	LENERGY	LRENEWABLE
Global	1	0	0	0	0	0
Low-income	1	1	1	1	1	1
Lower-middle-income	1	0	0	0	0	0
Upper-middle-income	1	0	0	0	0	0
High-income	1	0	0	0	0	0

Source: Authors’ calculations.

).

According to the Hausman test results demonstrated in

Table 8. Hausman test for long run homogeneity.

	Global Panel			Lower-Middle-Income Panel			Upper-Middle-Income Panel			High-Income Level
Variables	(b) MG	(B) PMG	(b-B) Difference	(b) MG	(B) PMG	(b-B) Difference	(b) MG	(B) PMG	(b-B) Difference	(b) MG	(B) PMG	(b-B) Difference
LREALGDP	−0.03	−0.04	0.02	−0.16	−0.08	−0.08	0.20	−0.02	0.23	−0.10	−0.09	−0.01
LFDI	2.34	0.01	2.32	2.39	0.53	1.86	5.89	0.02	5.87	−0.02	0.01	−0.03
LFDI2	−0.05	0.00	−0.05	−0.01	0.00	0.00	−0.19	0.00	−0.19	0.01	0.00	0.01
LENERGY	0.14	−0.11	0.25	−0.21	−0.10	−0.11	0.57	−0.10	0.67	0.06	−0.13	0.19
LRENEWABLE	−0.40	−0.05	−0.34	0.77	0.25	0.52	−4.62	−0.13	−4.49	1.62	−0.08	1.70
	Chi2(5) = 7.76; Prob > Chi2 = 0.17			Chi2(5) = 36.09; Prob > Chi2 = 0.30			Chi2(5) = 5.69; Prob > Chi2 = 0.34			Chi2(5) = 7.77; Prob > Chi2 = 0.17

Source: Authors’ calculations.

, no rejection occurs for the long-run homogeneity’s null hypothesis at a 1% level for any of the panels. According to the test’s null hypothesis, no statistical significance in the differences among PMG and MG estimators was observed, which further specifies that the PMG is involved in making more efficient estimations. Therefore, according to the present conditions, preference will be given to PMG estimators over the MG estimators based on their high consistency, and efficiency abnormalities that exist in them.

The effects of the long-run and short-run independent variables on the EFP per capita’s natural logarithm have been confirmed according to the PMG estimation results (see

Table 9. PMG estimation results.

	Global Panel		Low-Income Panel		Lower-Middle-Income Panel		Upper-Middle-Income Panel		High Income Panel
	Long-Run	Short-Run	Long-Run	Short-Run	Long-Run	Short-Run	Long-Run	Short-Run	Long-Run	Short-Run
LREALGDP	−0.04 c	0.01	0.14 c	0.11	−0.08 c	0.12 c	−0.02 b	−0.21	−0.09 c	0.12 c
LFDI	0.01 c	1.47	7.31 c	1.99	0.53 c	0.83	0.02 c	3.22	0	0.55
LFDI2	0	−0.01	−0.01	−0.00 a	−0.00 c	0	−0.00 b	−0.03	0.00 a	0
LENERGY	−0.11 c	0.44 c	−0.01	0.2	−0.10 a	0.33 c	−0.10 a	0.47 c	−0.13 b	0.46 c
LRENEWABLE	−0.05	−0.05	0.15	−0.42 c	−0.25	−0.17	−0.13	−0.7	−0.08	−1.36
No of countries	80	80	2	2	19	19	23	23	36	36
Obs.	1920	1920	48	48	456	456	552	552	864	864

Source: Authors’ calculations. ^a = Significant at 0.10 level; ^b = significant at 0.05 level; and ^c = significant at 0.01 level.

) for both global panels and subpanels. Upon examining Real GDP’s effect on EFP, long-run EFP is decreased by real GDP for the global panel. As shown, 0.042 is the obtained coefficient and it has significance at 1%. Hence, for the long-run global panel, with a 1 percentage point (pp., hereafter) raise in the Real GDP, there is a 0.042 pp. decrease in EFP. Moreover, there is no observed significance for the short-run effect.

EFP is negatively impacted by real GDP as the findings are significant at 1%, 5%, and 1% levels. Moreover, there is a positive sign of EFP by Real GDP for lower-, middle-, and high-income countries in the short run. Due to this, there is environmental damage for these subpanels with the increase in real GDP. According to a compelling result, ecological deterioration is evident as EFP increases by 0.14 pp. caused by the 1 pp. increase in real GDP for the low-income panel.

5. Discussion and Implications

Upon considering the effect of FDI inflow, the evaluation of it and its squared form is essential. It can be concluded that the empirical results support the PHH because of the positiveness of all long-run coefficients and their significance at the 1% level, with the exception of high-income countries. According to the coefficients of low-income samples, EFP is increased by 7.3 pp. When FDI inflow increases by 1 pp. with the increase in the income of the subpanel, this coefficient becomes smaller until it becomes insignificant for the panel of high-income. According to the claim made by the PHH, the developed countries’ MNCs have a preference for investment in developing countries because of the presence of relaxed environmental regulations in such countries, along with the availability of a low-cost labour force. These findings support the hypothesis. At a 5% significant level or higher, there is a significance of squared term coefficients for lower-middle-, and upper-middle-income countries, in the long run reflecting a decreasing impact. Due to very small coefficients, to guarantee the inverted U-shaped relationship between EFP and FDI, the sample size and time window should be increased.

For all samples except for low-income samples, the impact of energy consumption per capita upon degradation of the environment is decreasing at a 10% or higher level of significance in the long run. At the same time, in the short run, this effect becomes positive. Although for the low-income panel, the coefficients appear to have no significance, negative and positive signs can also be noticed for this variable for the long-run and short-run coefficients. This may be possible because, in the long run, lower degradation of the environment is created because of the step-wise transition into renewable energy sources. Moreover, even if the non-renewable energy sources use has been adopted, more consciousness about the waste production has been observed among the people; hence the negative impact significance in the long run is explained. According to the majority of the empirical results, the EFP is not significantly impacted by the renewable energy ratio in total energy consumption. The negative coefficient is observed at a 1% significance level only for the subpanel of low-income countries. The result is justified as poverty presence, inadequate resources for environmental development, lack of environmental consiousness among the people, and the consumption of renewable resources is common in low-income countries, which created a higher effect in low-income in contrast with the high-income countries.

For the global panel, the error correction term is shown in

Table A1. PMG estimation results—error correction terms for each country.

Sr. No	Country	Coef.	Std. Err.	Z	p-Value	Sr. No	Country	Coef.	Std. Err.	Z	p-Value
1	Albania	−0.74 c	0.03	−23.96	0.00	41	Jamaica	−0.91 c	0.14	−6.35	0.00
2	Algeria	−0.41 c	0.11	−3.92	0.00	42	Japan	−1.10 c	0.16	−6.87	0.00
3	Argentina	−0.80 c	0.16	−4.94	0.00	43	Jordan	−0.54 b	0.22	−2.41	0.02
4	Australia	−0.72 c	0.19	−3.83	0.00	44	Kenya	−0.62 c	0.12	−5.11	0.00
5	Austria	−0.70 c	0.19	−3.74	0.00	45	Lebanon	−0.39 c	0.11	−3.61	0.00
6	Bahrain	−0.92 c	0.14	−6.76	0.00	46	Malaysia	−0.61 c	0.11	−5.48	0.00
7	Bangladesh	−0.54 c	0.14	−3.75	0.00	47	Malta	−1.02 c	0.15	−6.75	0.00
8	Belgium	−0.43 c	0.13	−3.39	0.00	48	Mauritius	−0.64 c	0.20	−3.29	0.00
9	Benin	−0.31 c	0.10	−3.02	0.00	49	Mexico	−0.47 b	0.19	−2.49	0.01
10	Bolivia	−0.82 c	0.15	−5.38	0.00	50	Morocco	−0.96 c	0.20	−4.72	0.00
11	Botswana	−0.76 c	0.13	−5.81	0.00	51	Netherlands	−1.57 c	0.12	−12.71	0.00
12	Brazil	−0.80 c	0.16	−5.09	0.00	52	NewZealand	−1.21 c	0.16	−7.62	0.00
13	Brunei Darussalam	−0.48 c	0.09	−5.16	0.00	53	Nicaragua	−0.75 c	0.17	−4.29	0.00
14	Bulgaria	−0.60 c	0.20	−3.03	0.00	54	Nigeria	−0.35 b	0.18	−2.00	0.05
15	Côte d’Ivoire	−0.86 c	0.17	−5.22	0.00	55	Norway	−0.29 b	0.14	−2.10	0.04
16	Cameroon	−0.89 c	0.18	−4.95	0.00	56	Pakistan	−1.08 c	0.19	−5.38	0.00
17	Canada	−0.43 c	0.17	−2.48	0.01	57	Panama	−0.80 c	0.14	−5.57	0.00
18	Chile	−0.90 c	0.17	−5.44	0.00	58	Paraguay	−0.27	0.17	−1.59	0.11
19	China	−0.97 c	0.15	−6.35	0.00	59	Peru	−0.42 c	0.12	−3.52	0.00
20	Columbia	−0.80 c	0.12	−6.59	0.00	60	Philippines	−0.95 c	0.13	−7.37	0.00
21	Costa Rica	−0.78 c	0.12	−6.38	0.00	61	Poland	−0.54 c	0.19	−2.75	0.01
22	Cyprus	−0.63 c	0.23	−2.81	0.01	62	Portugal	−0.77 c	0.11	−6.83	0.00
23	Denmark	−0.63 c	0.13	−4.66	0.00	63	Romania	−0.78 c	0.13	−6.20	0.00
24	Dominican Republic	−0.54 c	0.16	−3.30	0.00	64	Saudi Arabia	−0.99 c	0.17	−5.98	0.00
25	Ecuador	−0.80 c	0.13	−5.97	0.00	65	Senegal	−0.66 c	0.17	−3.91	0.00
26	Egypt	−1.03 c	0.15	−6.66	0.00	66	Singapore	−0.90 c	0.21	−4.25	0.00
27	Finland	−0.56 c	0.16	−3.40	0.00	67	South Africa	−0.86 c	0.19	−4.45	0.00
28	France	−1.00 c	0.22	−4.50	0.00	68	Spain	−0.93 c	0.15	−6.28	0.00
29	Gabon	−0.88 c	0.19	−4.69	0.00	69	Sri Lanka	−0.79 c	0.12	−6.54	0.00
30	Germany	−0.88 c	0.19	−4.71	0.00	70	Sweden	−0.60 c	0.12	−4.81	0.00
31	Ghana	−0.62 c	0.10	−5.93	0.00	71	Switzerland	−1.48 c	0.16	−9.20	0.00
32	Greece	−0.90 c	0.15	−6.07	0.00	72	Thailand	−0.52 c	0.12	−4.20	0.00
33	Guatemala	−1.52 c	0.21	−7.14	0.00	73	Togo	−0.62 c	0.13	−4.82	0.00
34	Haiti	−0.38 a	0.20	−1.88	0.06	74	Trinidad and Tobago	−0.43 c	0.11	−3.82	0.00
35	Honduras	−0.79 c	0.16	−4.82	0.00	75	Tunisia	−0.27 b	0.13	−2.17	0.03
36	India	−0.49 c	0.15	−3.21	0.00	76	Turkey	−0.93 c	0.17	−5.45	0.00
37	Indonesia	−0.50 c	0.16	−3.10	0.00	77	UAE	−1.18 c	0.07	−16.96	0.00
38	Ireland	−0.38 b	0.16	−2.41	0.02	78	UK	−0.50 c	0.12	−3.98	0.00
39	Israel	−0.97 c	0.18	−5.39	0.00	79	USA	−0.50 c	0.13	−3.68	0.00
40	Italy	−0.77 c	0.22	−3.45	0.00	80	Uruguay	−0.95 c	0.15	−6.17	0.00

Source: Authors’ calculations. ^a = Significant at 0.10 level; ^b = significant at 0.05 level; and ^c = significant at 0.01 level.

in the Appendix A, which is calculated to be −0.74 at a 1% significance level. In the first year, the adjustment of potential shocks was made by 74%. Procedures of PMG were utilized for the short-run estimations of countries available in Table A1 in the Appendix A. The obtained results show that all the countries have negative error correction terms. All the countries have a significance level of 10% or higher except Panama. Therefore, long-run relationships among these countries’ variables have been observed. The fastest country, which is adjusted to its long-run equilibrium, is Morocco as identified by these empirical results. In contrast, Trinidad is the slowest among all, and in between them lies Tobago. Less developed and low-income countries target dirty industries of the investing countries for their operations as the less developed countries are only concerned with economic growth by FDI while ignoring the environmental risks [22,48]. These findings are affirmed by the literature [42,44]. However, the opposite view, P-HH, is also supported by the works of Tamazian and Chousa [70]. This hypothesis defends the presence of favourable FDI inflows’ influence on host countries’ environment quality. This requires MNCs to adopt an environmentally friendly manner and modern production techniques while performing their activities [37,38]. In this way, freshly learned less-polluting production techniques will be adopted in the host countries, lowering the environmental degradation, especially in the long run [27].

Customized policymaking for every income level subpanel is suggested to ensure sustainable economic development in the energy, including the electricity sector during the rapid growth of electricity consumption in the era of the electricity markets liberalization [71]. Therefore, it is crucial to assure transparency in FDIs, including regarding the protection of the environment. This largely depends on the quality of institutions in the countries concerned [72,73,74].

6. Conclusions

This study has the objective of investigating causal connections between the EFP, FDI, and economic growth indicators, i.e., real GDP, consumption of energy, and the share of renewable energy in the consumption of total energy by utilizing the data from 80 countries obtained from 1990–2014. The techniques of cointegration analysis and panel data unit root were utilized, while the PMG panel ARDL estimation approach was adopted for the estimations. The separation of data was done according to the income levels in four heterogeneous panels. In light of the cross-sectional dependence test’s results, there is a contemporaneous correlation between the entities. The application second-generation unit root test was done as it is suggested for scenarios where cross-sectional dependence is present. The results showed that the dependent variable and two out of five independent variables are integrated of order one, while others are stationary at their level forms. Two-panel cointegration tests were applied and support the long-run relationship presence among usage of energy, EFP, and renewable energy share in total energy consumption. AIC determined the variables’ optimal leg lengths in the model. PMG and MG panel ARDL approaches were chosen for estimations as they are applicable in the cases where the dependent variable is I(1) and others are a combination of I(0) and I(1). The Hausman long-run homogeneity test is utilized to choose the most efficient one between PMG and MG. Test results demonstrated that PMG is the most appropriate model for all panels.

In the long run, the EFP is decreased by Real GDP for the panels of global, lower-middle-income, upper-middle-income, and high-income panels, as demonstrated by PMG estimations. In addition to this, due to the observance of the positive effect significance of real GDP on EFP for lower-middle- and high-income countries; therefore, more environmental damage has been caused by an increase in Real GDP for these samples. According to a compelling result, ecological deterioration is evident as EFP increases by 0.14 pp. caused by the 1 pp. increase in Real GDP for the low-income panel. This may be justified by the sectors involved in creating real economic growth. According to the expectation of the economic development process, if the service sectors substitute industrialization or if industrialization substitutes agriculture in an environment-friendly manner, then there is a possibility of the former results. Furthermore, the growth direction is also important along with the source. Degradation of the environment will be decreased in long-run if the direction of real GDP growth is towards greener technology, environment-friendly regions, and a proficient labor force.

It is to be noted that the deterioration of the environment of low-, middle-, and high-income countries is increased by Real GDP growth in the long run, short run, and short run, correspondingly. Due to the different development stages, the short- and long-run influence of an increase in Real GDP on the environment differs. EKC hypothesis partially explains this, according to which when the people of developing countries give importance to their income because of the notable poverty, they do not take environmental deterioration seriously unless a point arises where environmental damage appears to decrease (Grossman and Krueger, 1991).

When the FDI inflow’s impact on the environment is conceived, FDI and its squared form should be examined together. Global panel results assert PHH because EFP is increased in the long run by a rise in inflows of FDI. The estimations of the lower-middle- and upper-middle-income panels, in the long run, suggested a nonlinear relationship between these variables, which gives the reflection that FDI and EFP have an inverted U-shaped relationship among them. PHH supports the estimations of the low-income subsample; nevertheless, the coefficient that demonstrates the strength of impact is much greater than it for the global panel. Therefore, degradation of the environment in low-income countries is influenced by FDI regulations which further creates more damage to the environment. The pollution-intensive MNCs are attracted by the relaxed environmental regulations in low-income countries. To avoid the strict environmental laws in their countries and reduce the cost required, the MNCs outsource their pollution-intensive production related policies in low-income countries [31,33,34,35]. Therefore, industrialization, which increases FDI inflow in low-income countries, creates serious environmental damage due to pollution.

The impact of energy use per capita is negative in the long run, while in the short run it is positive for all panels except for the panel of low-income countries. Therefore, in the short run, there is a degradation of the environment by total energy consumption per capita while it increases environmental quality in the long run. For a sustainable environment, the efficient utilization of energy is essential. To improve the quality of the environment, renewable energy sources and technologies for utilizing more efficient energy should be adopted. The attraction of people toward more clean energy sources is evident from these discoveries. Figure A1 in Appendix A demonstrates the consumption of renewable energy in percentage per year.

Although a rising trend of renewable energy share in total energy consumption is represented in Figure A1 in Appendix A, PMG estimations are not in favor of this fact. EFP has a decreasing impact on renewable consumption of energy as depicted by all the coefficients, but there is no significance for coefficients except for the low-income panel. Re-conduction of this research might be done in the future by incorporating a large sample, a big-time interval, and unique techniques of econometry. Each country’s subpanel should follow different policies according to its circumstances. In high-income countries, there is no sign of FDI impact on the quality of the environment. This is because, in such countries, there is an implementation of strict and well-established environmental laws both for FDIs and local production aspects. An inverted U-shaped pattern has been demonstrated in the relationship between FDI and the degradation of the environment for lower-middle- and upper-middle-income countries. Therefore, environmental damage has been created by FDI until reaching the turning point. Hence, it is important for the high-income nations and developed countries to introduce such production techniques in host countries based on cleaner technologies to preserve nature by decreasing damage to the environment in the long run.

Furthermore, companies should encourage the adoption of cleaner production techniques by incentives and also by ensuring strict environmental regulations. Degradation of the environment is increased and environmental quality is decreased by FDI in low-income countries with an essential and comparatively high coefficient, hence declaring PHH. Due to this, concrete measures should be taken by these countries specifically for the protection of the environment. The license of pollution-intensive industries should be cancelled, or operating permission should be given on following specific rules along with measures such as the provision of incentives for green technology users and implementation of strict environmental regulations. The activities of already licensed firms should be monitored for environmental protection. It is the responsibility of all the countries throughout the world to preserve their environment while reducing the depletion of natural resources, reducing waste production, and providing a healthy natural environment for living to the current and future generations, which is essential for sustainable development.