Global CO₂ Emission Reduction Disparities After and Before COVID-19

Abstract

The relationship between economic progress and environmental quality remains a central focus in global sustainability discourse. This study examines the link between per capita economic growth and CO₂ emissions across 128 countries from 1996 to 2022, controlling for energy consumption, trade volume, and foreign direct investment (FDI) inflows. It also evaluates the role of governance quality—measured by regulatory quality and its volatility—while considering the globalization index as a confounding factor influencing CO₂ emissions. We test the Environmental Kuznets Curve (EKC) hypothesis, which suggests that emissions initially rise with income but decline after reaching a certain economic threshold. Our findings confirm the global presence of the EKC. The analysis further shows that trade openness, governance, and globalization significantly influence FDI inflows, with FDI, in turn, reinforcing institutional quality through improved governance and globalization indicators. However, in countries with weaker governance and regulatory frameworks, FDI tends to promote pollution-intensive industrial growth, lending support to aspects of the Pollution Haven Hypothesis (PHH). We find a significant departure in EKC explained by post-COVID governance and globalization compromises, which induced the environment towards the PHH phenomenon. These results highlight the need for context-specific policy measures that align economic development with environmental constraints.

1. Introduction

1.1. Settings

There is an optimistic view on global emission, popularly known as the Environmental Kuznets Curve (EKC) hypothesis, that environmental quality degrades in the early stages of economic expansion as income rises, and the situation gradually improves. The degree of environmental degradation starts improving after a certain income threshold, where economic progress enables environmental improvement. There is a self-correcting mechanism between environmental degradation and economic growth, where economic expansion is the solution to a country’s environmental deterioration [1,2,3].

The EKC hypothesis believes that as the country gets richer with changing economic activities, the environmental quality also improves with shifts in consumption patterns, technological advancements, and more stringent environmental policies. McConnell [4] suggests that under the weaker assumption of environmental amenities as normal goods, the demand for environmental amenities by individuals increases with a potential increase in income. Supporting this argument, Mor and Jindal [5] explain this relationship as a change in economic structure and technology along with regulation and enhanced environmental awareness that offset the impact of economic growth on the environment. However, achieving such improved or rejuvenated environmental quality requires the design, implementation and enforcement of cooperative multi-national approaches, as Nordhaus [6] rightly points out, since individual nations act independently in their own interests, and the benefits of cooperation extend largely to other nations.

Contrary to the optimistic outlook of the EKC, the Pollution Haven Hypothesis (PHH) presents a more cautious perspective on the relationship between economic growth and environmental quality. According to the PHH, stringent environment regulations in developed countries may lead polluting industries to relocate their operations to developing countries with laxer environmental standards. This dynamics creates “pollution havens”, where developing countries may experience increased environmental degradation as they attract foreign investment in pollution-intensive sectors [7]. The PHH suggests that rather than economic growth leading to automatic environmental improvements, it can exacerbate environmental problems in lower-income countries that lack robust regulatory framework. Thus, this hypothesis challenges the universality of the EKC by emphasizing that economic progress alone does not guarantee environmental betterment, especially for the lower-income countries.

Another pessimistic view that is equally strong is that the global mean temperature could rise by 4 degrees Celsius or more above pre-industrial levels by the end of the twenty-first century if carbon emissions continue to rise at the current pace [8]. This upward trajectory of emissions contributes to environmental deterioration, posing a substantial threat to long-term economic stability. Recognizing this, economists [2,6,9,10,11,12] advocate for integrating environmental considerations into the framework of global economic growth. Addressing climate change is an environmental necessity and a sound economic strategy. The costs of mitigating climate change are far less than the costs of inaction, and addressing environmental issues can spur innovation, create jobs, and promote economic growth [1,2].

The Historic Paris Agreement of 2015 introduced a more flexible and inclusive approach, with each country setting its emissions reduction targets. The agreement’s goal of keeping global temperature increases well below 2 degrees Celsius above pre-industrial levels, with an aspirational aim of limiting it to 1.5 degrees Celsius, signaled an unprecedented global consensus on the urgency of climate action. It acknowledges the disproportionate impact of climate change on vulnerable regions, providing provisions for financing developing countries to combat climate change and bolster their resilience [13].

The Environmental Kuznets Curve (EKC) emerged from debates on population growth, consumption patterns, and resource scarcity, while the Pollution Haven Hypothesis (PHH) arose from the concerns about the environmental consequences of globalization, particularly the relocation of polluting industries to countries with weaker environmental regulations. While some economists argue that economic growth can ultimately resolve environmental issues, others maintain it exacerbates them. Challenges to the validity of EKC and PHH arise, with differing views on their applicability and the role of government policies in shaping environmental outcomes. Low-income countries face the dilemma of balancing economic growth with environmental sustainability amid international commitments to reduce emissions.

Kuznets’ inverted ‘U’ relationship between economic growth and income inequality shows an initial rise in inequality followed by a decline during industrialization in advanced economies. Grossman and Krueger [14] extend this concept to environmental degradation, forming the Environmental Kuznets Curve (EKC). Their study finds that pollutants increase and decrease at the initial phase of economic growth, suggesting that economic growth can eventually reduce environmental harm. To Grossman and Krueger, the scale, composition, and technical effect are key determinants that shape EKC. The scale effect occurs when capital accumulation consumes more pollution-intensive input and increases throughput. The environment suffers from rising pollution levels, depleting natural resources and causing biodiversity loss [1,15]. The second mechanism, the composition effect, occurs when the economy transforms from an abundance of polluting industries to a less polluting service sector, accompanied by an increase in environmental consciousness and the implementation of environmental policies. Finally, the technical effect implies reducing environmental pollution by adopting cleaner technologies. Subsequent research broadens this concept. The Pollution Haven Hypothesis (PHH) stems from trade and environmental economics, emphasizing the uneven global distribution of environmental regulations. It suggests that multinational corporations tend to shift pollution-intensive production to countries with weaker environmental laws, thereby creating so-called “pollution havens” [16]. Copeland and Taylor [17] formalize the hypothesis within a theoretical trade framework, identifying three key effects: the scale, technique, and composition effects—similar to the EKC framework but with distinct implications. While the scale effect leads to increased emissions through higher output, the composition effect under PHH refers to a country’s specialization in pollution-intensive industries due to lax regulation. The technique effect, by contrast, is often limited in developing countries, where technological transfer is slow and enforcement is weak. As a result, global trade may exacerbate environmental degradation in low-income countries unless complemented by strong international regulatory mechanisms.

Meanwhile, the validity of the EKC and its applicability in the current world scenario is challenged, as the very source of the problem is suggested as a solution. Stern [1] believes a single model cannot generalize the dynamic interaction between economic growth and environmental degradation. According to Roberts and Grimes [18], the EKC growth patterns were only relevant for industrialized nations prior to the 1970s oil shocks. These advanced economies had a colonial past and geopolitical strength to exploit their colonies’ resources and markets, and similar growth circumstances are not accessible to today’s developing world. Raymond [19] also refuses the idea of economic growth coming to the aid of the environment. Similarly, Gill and Viswanathan [20] raise the concern that by adhering to the EKC development trajectory of “grow now, clean later”, global economies have intensified the threat of global warming.

Despite criticisms, EKC remains the widely accepted model for investigating the environment and economic growth nexus. Against this backdrop, this paper will analyze the status of global CO₂ emissions and their relation to per capita GDP, energy consumption, FDI, and trade openness. Moreover, we will examine whether different countries’ groups exhibit the same trend.

1.2. Some Stylized Facts

There has been a consistent upward trend in global annual emissions. Since the mid-20th century, emissions from the energy sector have steadily increased, rising from approximately 11 billion tons of carbon dioxide per year in the 1960s to an estimated 36.6 billion tons in 2022 [21,22]. While fossil CO₂ emissions are declining in certain regions, such as Europe and the USA, the global trend continues to rise. Efforts to reduce fossil fuel usage are not advancing quickly enough to mitigate the severe impacts of climate change. Scatter plots of emissions and their contributing factors, derived from the dataset constructed for this paper and presented in Figure 1, show a strong positive correlation with GDP and energy consumption, and light correlation with FDI and global trade.

These relationships are complex and multifaceted, and it is important to understand how they manifest globally in aggregate terms. While much of the existing literature focuses on specific regions or country clusters, there has been limited empirical work that examines these dynamics from a more global and unified perspective. Our study addresses this gap by providing a comprehensive analysis of the overarching patterns and relationships at the global level. We further contextualize our findings within the broader literature in the discussion section of our results.

The key research questions guiding this study are as follows: First, do greenhouse gas emissions follow the Environmental Kuznets Curve (EKC) hypothesis—rising during the early stages of economic development and declining at more advanced stages? Second, does a greater inflow of foreign direct investment (FDI) contribute to elevated emission levels? Third, what is the impact of international trade on a country’s emissions profile? Finally, how has COVID-19 altered the dynamics of trade and environment?

1.3. Some Empirical Works on EKC and PHH

Environmental economics explores the intricate relationships among carbon emissions, FDI, energy consumption, and economic growth. The literature abounds in probing such relationships regarding scale, efficiency, and technological aspects of the Environmental Kuznets Curve hypothesis, and Pollution Haven Hypothesis with conflicting results. Zhang and Cheng [23] find a unidirectional Granger causality from GDP to energy consumption and from energy consumption to carbon emission in the long run. The study firmly suggests that neither energy consumption nor carbon emission leads to economic growth.

One of the articles by Ghosh [24], using the Johansen–Juselius maximum likelihood procedure in a multivariate framework, probes the cointegration and causality between carbon emission and economic growth. Incorporating energy supply, investment, and employment data from 1971 to 2006 of India, this paper shows that while carbon emission is positively associated with economic growth in the short run, there is no long-run causality between the two variables. In this context, Shreezal and Adhikari [25], with a multivariate analysis framework, find no long-run relationship between carbon emission and the economic growth of Nepal. Their ARDL approach, followed by the TY non-Granger causality test, shows that in the realm of carbon emission and economic growth, the EKC hypothesis does not hold. Regmi and Rehman [26] also uncover similar results. Using data on related variables from 1971 to 2019, they establish that Nepal’s carbon emission and economic growth have an inimical relationship in both the short and the long run. In the same vein, no statistically significant U-shaped EKC link exists among the prosperity and quality of environment indices in Nepal [27].

Contrary to the literature mentioned above that does not show a significant existence of EKC, a study by Mor and Jindal [5] reveals that the environmental quality improves with per capita growth. The study employs the least squares method on 39 countries covering Appendix A and Non-Appendix A Parties under the Kyoto Protocol. Karedla, Mishra [28] show that carbon emission positively affects economic growth. Using 1971 to 2016 data from India, they uncover the long-run positive relationship between carbon emission and economic growth, thus validating the EKC hypothesis. The paper also uncovers the long-run negative association between trade openness and carbon emission. Chng [29] tests the validity of the EKC hypothesis in six ASEAN countries from 1971 to 2013. Their study shows the long-run positive association between carbon emission and economic growth in Singapore, Thailand, and Vietnam, validating the existence of the EKC hypothesis. Rabbi, Akbar [30] test the validity of EKC in Bangladesh using data from 1972 to 2012. They posit a significant positive relation between carbon emission with energy consumption and economic growth.

The EKC hypothesis and the significant relation between carbon emission and trade openness are validated in their paper. Khan [31] proposes a similar conclusion using data on the relevant variables in Pakistan from 1965 to 2015. Applying ARDL bound test estimation, their study suggests a positive relationship exists between carbon emission, increased energy consumption, and economic growth. Moza-hid and Akter [32] analyze the causal relationship among carbon emission, energy consumption, FDI, and GDP of SAARC nations from 1980–2016. Their study validates the EKC, as the national income shows a long-run positive association with carbon emission. Bouznit and Pablo-Romero [33] similarly validate the EKC hypothesis in Algeria, using data from 1970 to 2010. Their analysis concludes that carbon emission is positively associated with the economic growth of Algeria in the long run.

Similarly, Aung et al. [34] examine the validity of EKC in Myanmar using GHG as an indicator of environmental pollution and GDP for economic development. This study indicates a positive relation between GDP and CO₂ in both the short run and long run. Similarly, trade and financial openness have a significant negative relationship with all forms of GHG emissions. Another study by Zhang [35] explores EKC in South Korea from 1971 to 2013. The empirical results supports the EKC hypothesis because both short- and long-run estimations show that increasing non-fossil power electricity consumption can help slow environmental deterioration, whereas expanding trade increases carbon dioxide emissions.

This study contributes to the empirical literature on the Environmental Kuznets Curve (EKC) by analyzing the pollution–income and pollution–FDI relationships over the period 1996–2022, employing panel regression techniques on a balanced panel of 128 countries. It further tests for the validity of both the EKC and the Pollution Haven Hypothesis (PHH), providing evidence on the dynamic linkages between economic activity, foreign investment, and environmental degradation. This paper further examines how the EKC phenomenon has evolved following the 2019 pandemic.

2. Materials and Methods

Data and Tools

We compile the World Bank’s World Development Indicators and Good Governance data and combined them with Globalization-related data from the KoF index. We end up with a balanced panel data set containing information for 128 countries over the years 1996–2022. Details on the definition of variables, their sources, and descriptive statistics are in

Table 1. Nature and sources of variables used.

Variables (Code)	Definition (Source)
CO2 Emissions (CO2)	The “CO2 emissions in megatons (Mt)” refers to the amount of carbon dioxide (CO2) emissions of each country, measured in megatons (Mt). (Source: The World Bank)
Energy Consumption (energy_pc)	The “Per Capita Energy Consumed in Terajoules (Tj)”, which is per person total energy used within a country. (Source: ibid)
GDP Per Capita (ln_gdp_pc)	The Natural log of GDP per capita (GDP), refers to the logarithmic transformation of the gross domestic product (GDP) per person in a country. Its squared term is also used in analysis as “GDP Per Capita Square (ln_gdp_pc_sq)”. (Source: The World Bank)
Annual GDP Growth (gdp_g_annual)	The annual rate of growth of GDP of each country. (Source: The World Bank)
Foreign Direct Investment (fdi_inflow)	Foreign Direct Investment (FDI) is an investment made by an entity or individual in one country to establish a lasting interest and significant control (at least 10% ownership) in a business operating in another country. Here, we consider the net FDI inflows, the total value of inward investments minus outward investments, capturing the net movement of foreign capital into a country. (Source: ibid)
Predicted FDI(fdi_hat)	This variable is the estimated FDI inflow based on our first model. It is not raw data but the model output of the regression of our first model. (Source: Author’s Calculation)
Trade Openness (open)	It is the sum of a country’s exports and imports of goods and services, expressed as a percentage of its GDP. It reflects how an economy integrates into the global trading system and relies on international trade. (Source: ibid)
Governance Indicators(pca_gov)	This variable measures the principle component analysis score of five governance indicators, as provided by the world Bank. The sixth indicator, Regulatory Quality, is not included here. (Source: The World Bank)
Regulatory quality (rq_dev)	The sixth component of the governance indicator, which would be used separately from the pca_gov score of the governance indicator. (Source: The World Bank)
Globalization Indicators (pca_glob)	This variable measures the principle component analysis score of the three globalization indicators, Economic (Trade and Finance), Social, and Political indicators. (Source: KOF)

.

The study employs CO₂ emissions as the dependent variable, and GDP per capita and its quadratic form as the primary independent variables. Similarly, foreign direct investment inflow, per capita energy consumption, and trade openness, which is the total volume of trade, are control variables. The secondary data are collected from the World Bank and KOF Globalization data, both openly available sources. We follow Grossman and Krueger [14], which establishes the relationship between environmental degradation and economic growth, mirroring the Environmental Kuznets Curve (EKC). The framework suggests that as GDP per capita increases, CO₂ emissions may initially increase, reflecting the positive relationship between economic growth and emissions. However, beyond a certain threshold, an inverse relationship indicates a decrease in emissions as GDP per capita increases.

The baseline model to evaluate the EKC hypothesis is $\ln E_{it}={\beta }_0+{\beta }_1\ln Y_{it}+{\beta }_2{\left(\ln Y_{it}\right)}^2+{ϵ}_{it}$, where $E_{it}$ represents environmental degradation (CO₂ emissions) in year t for country i, and $Y_{it}$ is the output per capita of country i in year t. The coefficients ${\beta }_1$ and ${\beta }_2$ represent positive and negative slopes, respectively, and ${ϵ}_{it}$ is the error term. This model is expanded to include control variables $z_{it}$ as shown in Equation (1):

$$\ln E_{it}={\beta }_0+{\beta }_1\ln {\mathrm{GDP}}_{it}+{\beta }_2{\left(\ln {\mathrm{GDP}}_{it}\right)}^2+\sum _j{n}_j{z}_{it}^j+{ϵ}_{it}$$

(1)

Replacing $E_{it}$ by CO₂ emissions, and confining $z_{it}^j$ to control variables FDI Inflow, per capita energy consumption, the regulatory quality and globalization index gives Equation (2) as

$$\begin{array}{cc}\hfill C{O}_{2_{it}}=& {\beta }_0+{\beta }_1\ln \left({\mathrm{GDP}}_{it}\right)+{\beta }_2{\left(\ln \left({\mathrm{GDP}}_{it}\right)\right)}^2+{\beta }_3\ln \left({\mathrm{EC}}_{it}\right)\hfill \\ \hfill & +{\beta }_4{\mathrm{FDI}}_{it}+{\beta }_5{\mathrm{RegQuality}}_{it}+{\beta }_6{\mathrm{Globalization}}_{it}+{\epsilon }_{it}\hfill \end{array}$$

(2)

We estimate this regression and report the summary in

Table 2. Panel regression results after and before COVID-19.

	(1)	(2)	(3)	(4)	(5)	(6)
Variables	ln_co2pc	ln_co2pc	ln_co2pc	ln_co2pc	ln_co2pc	ln_co2pc
ln_gdp_pc	0.597 ***	4.043 ***	2.756 ***	2.747 ***	2.777 ***	2.841 ***
	(0.0148)	(0.0940)	(0.0802)	(0.0801)	(0.0799)	(0.0818)
ln_gdp_pc_sq		−0.206 ***	−0.149 ***	−0.148 ***	−0.149 ***	−0.151 ***
		(0.00556)	(0.00459)	(0.00459)	(0.00457)	(0.00458)
ln_energy_pc			0.593 ***	0.596 ***	0.614 ***	0.611 ***
			(0.0132)	(0.0132)	(0.0135)	(0.0135)
fdi_inflow				−0.000480 ***	−0.000462 ***	−0.000435 ***
				(0.000124)	(0.000124)	(0.000124)
rq_stddev					0.208 ***	0.140 ***
					(0.0362)	(0.0411)
kofgi						−0.00293 ***
						(0.000852)
Constant	−4.399 ***	−18.35 ***	−17.19 ***	−17.18 ***	−17.56 ***	−17.79 ***
	(0.143)	(0.397)	(0.319)	(0.318)	(0.324)	(0.330)
Observations	3456	3456	3456	3456	3456	3456
ln_gdp_pc	0.593 ***	3.637 ***	2.597 ***	2.594 ***	2.635 ***	2.674 ***
	(0.0152)	(0.0994)	(0.0845)	(0.0843)	(0.0840)	(0.0859)
ln_gdp_pc_sq		−0.182 ***	−0.138 ***	−0.138 ***	−0.139 ***	−0.140 ***
		(0.00590)	(0.00488)	(0.00487)	(0.00485)	(0.00485)
ln_energy_pc			0.546 ***	0.547 ***	0.570 ***	0.568 ***
			(0.0135)	(0.0135)	(0.0138)	(0.0138)
fdi_inflow				−0.000472 ***	−0.000441 ***	−0.000415 ***
				(0.000137)	(0.000136)	(0.000137)
rq_stddev					0.218 ***	0.176 ***
					(0.0341)	(0.0392)
kofgi						−0.00188 **
						(0.000876)
Constant	−4.358 ***	−16.65 ***	−16.20 ***	−16.20 ***	−16.67 ***	−16.82 ***
	(0.146)	(0.418)	(0.340)	(0.339)	(0.345)	(0.351)
Observations	2944	2944	2944	2944	2944	2944
ln_gdp_pc	0.685 ***	2.880 ***	1.230 ***	1.230 ***	1.415 ***	1.403 ***
	(0.0336)	(0.278)	(0.258)	(0.256)	(0.257)	(0.262)
ln_gdp_pc_sq		−0.120 ***	−0.0533 ***	−0.0536 ***	−0.0633 ***	−0.0628 ***
		(0.0152)	(0.0135)	(0.0134)	(0.0134)	(0.0135)
ln_energy_pc			0.569 ***	0.573 ***	0.563 ***	0.560 ***
			(0.0400)	(0.0400)	(0.0396)	(0.0402)
fdi_inflow				0.000132	0.000139	0.000142
				(0.000108)	(0.000107)	(0.000108)
rq_stddev					−1.112 ***	−1.103 ***
					(0.300)	(0.301)
kofgi						0.000931
						(0.00314)
Constant	−5.223 ***	−15.00 ***	−11.13 ***	−11.15 ***	−11.69 ***	−11.65 ***
	(0.305)	(1.255)	(1.043)	(1.032)	(1.026)	(1.035)
Observations	512	512	512	512	512	512
Number of id	128	128	128	128	128	128

Standard errors in parentheses; *** p < 0.01, ** p < 0.05.

in the Results and Discussion section. We also estimate a two-stage least squares (2SLS) instrumental variable model using panel data, assuming a balanced panel. The first stage estimates the fitted values of FDI_inflow using lagged FDI and other instruments as

$$\begin{array}{cc}\hfill {\text{FDI\_inflow}}_{it}=& {\alpha }_0+{\alpha }_1{\text{L.FDI\_inflow}}_{it}+{\alpha }_2({\mathrm{gov}}_{it}\times {\mathrm{globalization}}_{it})\hfill \\ \hfill & +{\alpha }_3{\mathrm{open}}_{it}+{\alpha }_4{\text{gdp\_g\_annual}}_{it}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}\hfill \end{array}$$

(3)

And, we predict ${\widehat{\text{FDI\_inflow}}}_{it}$ and use again as an independent variable as

$$\begin{array}{cc}\hfill {\mathrm{CO}}_{2_{it}}=& {\beta }_0+{\beta }_1\ln \left({\text{gdp\_pc}}_{it}\right)+{\beta }_2\ln {\left({\text{gdp\_pc}}_{it}\right)}^2+{\beta }_3{\text{energy\_pc}}_{it}\hfill \\ \hfill & +{\beta }_4{\widehat{\text{FDI\_inflow}}}_{it}+{\beta }_5\left({\widehat{\text{FDI\_inflow}}}_{it}\times {\text{rq\_sd}}_{it}\right)\hfill \\ \hfill & +{\beta }_6{\text{rq\_sd}}_{it}+{\mu }_i+{\lambda }_t+u_{it}\hfill \end{array}$$

(4)

where ${\text{rq\_sd}}_{it}$ is the standard deviation of regulatory quality (a proxy for governance uncertainty), $u_{it}$ is the second-stage error term. The summary of this regression results are in

Table 3. EKC and PHH phenomenon: a panel estimation.

	(1)	(2)	(3)	(4)	(5)
Variables	FDI Inflow	CO2	CO2	CO2	CO2
L.fdi_inflow	0.671 ***
	(0.0136)
gov*globalization	0.142 *
	(0.0793)
open	0.0264 ***
	(0.00593)
gdp_g_annual	0.265 ***
	(0.0672)
ln_gdp_pc		675.5 ***	638.8 ***	664.3 ***	1427 **
		(136.7)	(136.7)	(139.8)	(694.2)
ln_gdp_pc_sq		−20.24 **	−17.18 **	−18.78 **	−85.29 **
		(8.187)	(8.203)	(8.410)	(41.17)
energy_pc		0.00384 ***	0.00420 ***	0.00423 ***	0.00354
		(0.000767)	(0.000770)	(0.000771)	(0.00246)
fdi_hat		−17.37 ***	−15.13 ***	−13.02 ***
		(3.686)	(3.720)	(4.450)
fdi_hat*rq_sd			72.94 ***	71.27 ***	88.16 *
			(18.17)	(18.28)	(49.75)
rq_sd				65.89
				(76.27)
Constant	−1.800 ***	−4074 ***	−4011 ***	−4122 ***	−5703 **
	(0.635)	(575.5)	(574.5)	(588.6)	(2818)
Observations	3328	3456	3456	3456	2187
Number of id	128	128	128	128	81

Standard errors in parentheses. *** p < 0.01, ** p< 0.05, * p < 0.1.

.

We estimate Equation (2) using variables as defined in Table 1, which is on the nature of the variables, their definitions, and sources of the data. According to the EKC framework, we expect a positive coefficient for GDP per capita and a negative coefficient for its square. Initially, economic growth increases pollution due to capital accumulation, leading to resource depletion and biodiversity loss. As the economy grows, the shift to less polluting sectors and greater environmental awareness help improve environmental quality. Furthermore, FDI and energy consumption are expected to positively correlate with CO₂ emissions. The impact of trade openness varies depending on the stage of economic development.

We also estimate Equation (2) using an alternative set of independent variables to examine whether there were any changes in the size and shape of the Environmental Kuznets Curve (EKC) before and after the COVID-19 pandemic. To do this, we include regulatory quality and the globalization index as additional explanatory variables. We estimate the complete model over the entire period, as well as two separate subsamples: one covering the period up to 2018 and another for the period from 2019 to 2022.

It should be noted that escalating emissions trajectories contribute to environmental deterioration, posing a substantial threat to long-term economic stability. Recognizing this, economists [2,6,9,10,11,12] advocate for integrating environmental considerations into the framework of global economic cooperation, FDI, and trade. They discuss how addressing issues on climate change is not only an environmental necessity but also a sound economic strategy. The costs of mitigating climate change are far less than the costs of inaction, spurring innovation, creating jobs, and promoting economic growth [1,2,3]. Thus, diversifying the electricity mix by introducing more renewable energies should be a priority. Global coordination, financial aid packages, and technical cooperation are important, as uncontrolled free trade damages the environment [36]. These provide justifications for the selection of variables in the model specified above.

The panel dataset constructed for this study comprises 128 countries observed over a 27-year period, from 1996 to 2022. This broad coverage allows for a comprehensive analysis of global emission patterns across varying levels of development and economic structures. Countries are selected based on the availability and consistency of data for key variables related to emissions, economic performance, energy use, foreign direct investment, and trade. The final sample represents a diverse set of economies, ensuring both geographic and developmental heterogeneity. A detailed list of countries included in the study is provided in

Table A1. Alphabetical list of countries.

Afghanistan	Albania	Algeria
Argentina	Armenia	Australia
Austria	Azerbaijan	Bahamas The
Bangladesh	Belarus	Belgium
Belize	Benin	Bhutan
Bolivia	Botswana	Brazil
Burkina Faso	Burundi	Cabo Verde
Cameroon	Canada	Chile
China	Colombia	Comoros
Congo Rep.	Costa Rica	Czechia
Denmark	Dominican Rep.	Ecuador
Egypt	Estonia	Fiji
Finland	France	Gambia (The)
Georgia	Germany	Ghana
Greece	Guinea	Honduras
Hungary	Iceland	India
Indonesia	Iran	Ireland
Israel	Italy	Japan
Jordan	Kazakhstan	Kenya
Kiribati	Korea Rep.	Kuwait
Kyrgyz Rep.	Latvia	Lebanon
Lithuania	Luxembourg	Madagascar
Malaysia	Maldives	Mali
Mauritius	Mexico	Moldova
Mongolia	Morocco	Namibia
Nepal	Netherlands	New Zealand
Nicaragua	Niger	Norway
Oman	Pakistan	Panama
Paraguay	Peru	Philippines
Poland	Portugal	Qatar
Russia	Saudi Arabia	Seychelles
Sierra Leone	Slovak Rep.	Slovenia
Solomon Islands	South Africa	Spain
Sri Lanka	Sudan	Sweden
Switzerland	Tajikistan	Tanzania
Thailand	Togo	Turkiye
Uganda	Ukraine	United Kingdom
United States	Uruguay	Vanuatu
Vietnam	Zambia	Zimbabwe

in the Appendix A.

We conduct a unit root test to determine whether the data series is stationary or exhibits unit roots. We choose to use the Levin, Lin, and Chu (LLC) unit root test as specified in Equation (3). The presence of unit roots signals a potential misspecification of our model. If the data are non-stationary or contain unit roots, indicated by non-constant mean and standard deviation, it suggests that the regression model might be misspecified or the variables may not be appropriately transformed. The LLC test we select is based on the Augmented Dickey–Fuller (ADF) test and extends it to allow for structural breaks. Its functional form is

$$\Delta y_{i,t}=\alpha y_{i,t-1}+\sum _{j=1}^{p_i}{\beta }_{i,j}\Delta y_{i,t-j}+X_{i,t}^{\prime }\delta +{ϵ}_{i,t}$$

(5)

where $\alpha =\rho -1$ is assumed. The lag order of difference terms ($p_i$) is changing over cross-sections. The residuals are obtained from the auxiliary regressions of $\Delta y_{i,t}$ and $y_{i,t}$ on lagged terms $y_{i,t-j}$ and on exogenous variables $X_{i,t}^{\prime }$. $\alpha$ is calculated using a data regression. For the Levin, Lin, and Chu unit root test, $H_0:\alpha =0$ implies there is a unit root, and $H_1:\alpha <0$ implies stationarity.

Table 4. Levin, Lin, and Chu unit root test statistics.

Variable (Code)	Unadjusted t	Adjusted t	p-Value
Carbon emission (CO2)	−22.1039	−4.3822	0.0000
Lag carbon emission (L.CO2)	−20.8348	−3.1067	0.0009
Log of per capita GDP (ln_gdp_pc)	−21.3360	−4.0111	0.0000
Log of per capita GDP squared (ln_gdp_pc_sq)	−21.6806	−4.1575	0.0000
Per capita energy consumption (energy_pc)	−24.0560	−4.3462	0.0000
Trade openness (open)	−27.0282	−4.4788	0.0000
Foreign direct investment inflow (fdi_inflow)	−29.7964	−8.9768	0.0000
Lag foreign direct investment inflow (L.fdi_inflow)	−28.7533	−8.2081	0.0000

Source: Author’s calculation.

summarizes the results of the Levin, Lin, and Chu (LLC) unit root test for stationarity in the panel data. The null hypothesis of the LLC test assumes that each time series contains a unit root (i.e., is non-stationary), while the alternative hypothesis asserts that all series are stationary. As shown in the table, all variables exhibit high negative adjusted t-statistics and p-values well below the 1% significance level. This reveals robust statistical evidence to reject the null hypothesis of non-stationarity for all variables. Therefore, we conclude that the variables under consideration are stationary in the panel data structure.

3. Results and Discussion

3.1. Results

The descriptive summary of the variables, as defined and elaborated in Table 1, is provided in

Table 5. Descriptive statistics.

Variables	Min	Max	Mean	S.D.
CO2 (in Mt) Emission	0.90	10,081.30	482.24	1227.85
GDP per Capita (gdp_pc) in USD	164.29	103,553.84	18,475.47	19,788.81
Electricity Consumption (ec_pc) in ‘000Tj	35.22	91,352.26	5787.35	12,670.26
Foreign Direct Investment (FDI) in Billion	−257.33	511.43	20.61	50.87
Trade Openness (open) in Billion	0.20	5669.25	467.86	797.43

. The panel data consists of 27 years and 128 countries, forming a balanced panel with a total of 3328 observations for each variable. The CO₂ emission variable spans a wide range, from a minimum of 0.90 Mt to a maximum of 10,081.30 Mt, showing a significant deviation. This highlights the varying levels of emissions across countries over time.

The average CO₂ emissions stand at 482.24 Mt, with a standard deviation of 1227.85 Mt. GDP per capita ranges from 164.29 to 103,553.84, with a mean of 18,475.47 and a standard deviation of 19,788.81. Energy consumption (EC) ranges from 35.22 Tj to 91,352.26 Tj, with a mean value of 5787.35 Tj and a standard deviation of 12,670 Tj. Net Foreign Direct Investment (FDI) ranges from −257.33 billion to 511.40 billion. The United States received the highest FDI in 2015 and had the highest trade openness, amounting to 5669.25 billion in 2018. FDI averaged 20.61 billion, with a standard deviation of 50.87 billion. Similarly, trade openness had a mean of 467.86 billion and a standard deviation of 797.43 billion.

The correlation coefficients among the variables considered in this study are summarized in

Table 6. Correlation coefficients.

Variables	lnCO2	lnGDPPC	lnEC	FDI	OPEN
Log of CO2 Emission (lnCO2)	1
Log of per Capita GDP (lnGDPPC)	0.397 **	1
Log of Electricity Consumption (lnEC)	0.948 **	0.286 **	1
Foreign Direct investment (FDI)	−0.094 **	0.148 **	−0.156 **	1
Trade openness (OPEN)	0.188 **	0.267 **	−0.296 **	0.512 *	1

Note: * and ** indicate for 0.05 and 0.01 level significance (2-tailed).

. The coefficients reveal that CO₂ emissions and GDP exhibit a positive correlation of 0.397 **, suggesting that as GDP increases, CO₂ emissions tend to rise. Similarly, CO₂ emissions and energy consumption show a strong positive correlation of 0.948 **, indicating that higher energy consumption is associated with increased carbon emissions. On the other hand, CO₂ emissions have a negative correlation with FDI of −0.094 **, implying that countries attracting more foreign investment tend to have lower levels of carbon emissions. Foreign trade has a positive correlation of 0.188 **, suggesting that foreign trade is associated with higher emissions. These findings indicate robust relationships between CO₂ emissions, GDP, energy consumption, FDI, and foreign trade, with strong associations among them.

In our panel regression estimation, we estimate the Hausman test to select among the random effect or fixed effect models. The RE model assumes that entity-specific effects are uncorrelated with the included independent variables, while the FE model assumes that unobserved heterogeneity is correlated with the included independent variables. The Hausman test is an asymptotic chi-square test based on the quadratic form obtained from the difference between a consistent estimator under the alternative hypothesis and an efficient estimator under the null hypothesis [37]. The null hypothesis is given by $H_0:\mathrm{Cov}({\alpha }_i,X_{it})=0$, i.e., the RE model is preferred over the FE model or $\mathrm{SE}\left({\hat{\beta }}^{RE}\right)<\mathrm{SE}\left({\hat{\beta }}^{FE}\right)$. Our test result, which is summarized in

Table A2. Hausman test.

Test Hypothesis	All	Before 2019	After 2019	Emerging
Cross-Section Random	0.0000	0.1828	0.0001	1.0000
Period Random	NA	NA	NA *	1.0000
Cross-Section and Period Random	NA	NA	0.0000	1.0000

Source: Author’s calculation; Note: NA * period random effect is zero.

in Appendix A, rejects the null hypothesis and concludes that the RE model is not appropriate. Similarly, the turning point is estimated from the income and income square terms. We estimate the turning possible turning points; the method is similar to that of De Bruyn et al. [38].

We suspect that FDI inflow is endogenous in the estimation system; therefore, we prefer to estimate a two-stage regression, where we first estimate the determinants of FDI inflow. The regression coefficients are listed in the first column of the regression table in Table 3. The lagged value of FDI, trade openness, and GDP growth significantly determine the level of FDI inflow globally, which is in line with the existing literature. Moreover, we use the interaction term between governance and globalization, which is closely related to the institutional quality of each country. This interaction term also significantly and positively determines the inflow of FDI. We predict the estimated FDI inflow, called the FDI hat, and use it to assess its impact on CO₂ emissions. By taking this step, we bypass the endogeneity problem, even if we were to estimate a one-step regression. We estimate the second stage of the regression, where CO₂ emission is dependent on variables and various combinations of CO₂ emission determinants. The results are in columns 2–5 of the regression coefficients in Table 3.

We use globalization, governance (both measured as principal component analysis values), and annual GDP growth as instrumental variables. These instruments are correlated with FDI inflows but not with CO₂ emissions (see

Table A3. Correlation coefficients.

Variables	lnFDI Inflow	GDP Growth	Open	Gov*Globalization
Log of FDI Inflow	1
Annual GDP growth	0.1334 *	1
Trade Openness	0.4447 *	0.0518 *	1
Gov times Globalization	0.0996 *	−0.1031 *	0.0559 *	1

Note: * indicate for 0.05 level significance.

and

Table A4. Correlation coefficients.

Variables	CO2	GDP Growth	Open	Gov*Globalization
CO2	1
Annual GDP growth	0.0361	1
Trade Openness	−0.1611 *	0.0518	1
Gov times Globalization	0.0299	−0.1031 *	−0.1031 *	1

Note: * indicate for 0.05 level significance.

in the Appendix A). Although trade openness is weakly correlated with both FDI and CO₂ emissions, we exclude it from the explanatory variables in subsequent estimations. Thus, we assert that our chosen instruments are conceptually valid.

In Table 3, Model (1) estimates the determinants of FDI inflow using a dynamic specification. The coefficient of lagged FDI inflow is highly significant and positive, indicating strong persistence—past inflows are highly predictive of the current FDI. Both trade openness and GDP growth are also statistically significant and positively associated with FDI, suggesting that countries with more liberal trade regimes and higher economic growth tend to attract more foreign investment. Furthermore, the significant and positive interaction between political governance and globalization implies that stronger institutions amplify the effect of globalization on FDI inflow.

Using the estimated values of FDI inflow from Model (1), Models (2) through (5) examine the Environmental Kuznets Curve (EKC) and Pollution Haven Hypothesis (PHH) using overall CO₂ emissions as the dependent variable. All models confirm the inverted-U shape of the EKC: both income per capita and its squared term are statistically significant, with the expected positive and negative signs, respectively. This finding aligns with the Environmental Kuznets Curve (EKC) criterion, which posits that environmental degradation initially increases with economic progress but eventually declines after surpassing a certain income threshold. Additionally, the positive and significant coefficient of energy consumption per capita indicates that higher energy usage is associated with greater CO₂ emissions.

Notably, the coefficient of instrumented FDI inflow is negative and statistically significant in Models (2) through (4), supporting specific aspects of the efficiency gains through FDI channels into the recipient economy. This suggests that FDI inflows may be directed toward countries with laxer environmental regulations—resulting in reduced domestic emissions—or they may introduce cleaner technologies that lower pollution levels. Models (3), (4), and (5) include an interaction between the FDI instrument and volatility in regulatory quality, which has positive and highly significant coefficients. This suggests that the pollution haven effect is contingent upon uncertainty in regulatory quality: in countries with unstable regulatory environments, FDI may exacerbate environmental degradation, whereas, in countries with more consistent regulatory frameworks, the adverse ecological effects of FDI are likely to be mitigated. The pollution-reducing coefficient due to FDI inflow is significantly less than the positive coefficients of the interaction term. The volatility in regulatory quality alone is not significant, though positive; when interacting with FDI inflow, CO₂ emissions significantly increase. Therefore, the FDI channel is one of the plausible channels for PHH.

In Table 2, we present the regression results on the determinants of per capita carbon emissions, structured into three panels. The first panel reports the estimates for the full sample, capturing the total effects. The second panel isolates the pre-COVID-19 period, while the third panel focuses on the post-COVID-19 years. Moreover, we use the log of per capita CO₂ emissions as the dependent variable. There has been a sustained debate in the literature that using CO₂ emissions in aggregate and following estimations are biased towards extensive and highly populated countries, mostly emerging nations.

In the first panel, the results support the Environmental Kuznets Curve (EKC) hypothesis. Specifically, the coefficient on per capita GDP is positive and highly significant, while its squared term is negative and significant, indicating an inverted U-shaped relationship between income and carbon emissions. Additionally, FDI inflows are found to have a negative and statistically significant effect on emissions, suggesting that foreign investments may contribute to cleaner technologies or more efficient production. The standard deviation of regulatory quality (a proxy for policy instability) shows a positive and significant effect in models (5) and (6), implying that higher regulatory volatility is associated with higher emissions.

The second panel presents results from the pre-COVID-19 period, and the estimates largely mirror those from the full sample. This is expected, as the majority of observations in the dataset are from before the pandemic. The EKC pattern holds, and FDI continues to exhibit a negative and significant association with emissions. Regulatory quality volatility also remains positively significant, reinforcing the findings from the first panel.

In contrast, the third panel, which focuses on the post-COVID-19 period, reveals some important shifts. The EKC relationship remains intact, though the magnitude of the coefficients is generally smaller, indicating a potentially weaker economic-emissions link during this time. Notably, the effect of FDI inflows becomes positive, although statistically insignificant, suggesting that the nature or sectoral composition of FDI might have changed during the pandemic. Moreover, the regulatory quality volatility now has a negative and significant effect, a reversal from earlier periods. This suggests that in the post-COVID-19 context, greater institutional stability is associated with higher carbon emissions, a counterintuitive result that may reflect short-term policy responses or shifts in regulatory enforcement. Lastly, the coefficient on the globalization index (KOF) becomes positive but remains statistically insignificant after COVID-19.

The patterns observed in Table 2 are further illustrated in Figure 2. Figure 2a of the figure visually confirms the change in the relationship between per capita GDP and CO₂ emissions: although the Environmental Kuznets Curve (EKC) pattern remains intact, the curvature appears to have slightly shifted in the post-COVID-19 period. Figure 2b indicates a modest deceleration in the globalization index, suggesting that global interconnectedness may have plateaued following the pandemic. More notably, Figure 2c,d show a clear decline in regulatory quality and a simultaneous increase in its standard deviation, implying that regulatory governance has weakened and become more volatile in the post-COVID-19 years.

3.2. Discussion

Our findings on the Environmental Kuznets Curve (EKC), both before and after the COVID-19 pandemic, are consistent with previous studies [5,6]. They reaffirm the notion that early stages of economic growth tend to be associated with environmental degradation, but as income rises, environmental concerns begin to be addressed. This inverted U-shaped relationship provides some justification for the environmental toll of development in lower-income countries. However, this raises a critical concern: excessive environmental degradation in the name of economic progress can cause irreversible damage, compromising the well-being of future generations.

Another important issue stems from the shifting behavior of firms. Multinational corporations, facing stricter environmental regulations in developed countries, often seek opportunities in developing countries where environmental laws and regulatory enforcement are weaker or more volatile [17]. This “pollution haven” dynamic poses a serious threat to environmental sustainability. Our findings suggest that the COVID-19 crisis may have intensified this threat by weakening regulatory quality in many countries. The post-pandemic increase in regulatory uncertainty may have altered the direction of foreign direct investment (FDI), channeling it toward regions with laxer environmental oversight, thereby contributing to rising CO₂ emissions.

4. Conclusions and Recommendations

The empirical findings reaffirm the existence of a differentiated Environmental Kuznets Curve (EKC), illustrating a consistent inverted U-shaped relationship between economic growth and environmental degradation as measured by CO₂ emissions. In line with the EKC hypothesis, emissions tend to rise during the early stages of economic development but decline after a certain income threshold is reached. This turning point is robust across all model specifications, with stronger explanatory power when interactions involving regulatory quality and foreign direct investment (FDI) are included.

Importantly, the results highlight that countries with higher levels of global integration and stronger governance frameworks are more likely to attract FDI that supports cleaner growth. In our emissions models, instrumented FDI is associated with reduced CO₂ emissions in certain contexts—supporting the Pollution Halo Hypothesis. However, when the host-country regulatory quality is weak or deviates from international best practices, the relationship reverses: FDI correlates with higher emissions, consistent with the Pollution Haven Hypothesis (PHH). This duality underscores the critical role of domestic institutions in shaping the environmental impact of international investment.

The COVID-19 pandemic further complicates this landscape. Our findings suggest that the post-pandemic period has seen a decline in regulatory quality and a rise in policy uncertainty, which may have exacerbated environmental risks. The weakened institutional environment in many developing countries post-COVID likely made them more vulnerable to environmentally harmful investments, intensifying CO₂ emissions and threatening progress toward sustainability.

Policymakers in developing and emerging economies should strengthen regulatory frameworks to meet global environmental standards, promote green FDI through incentives for clean technologies, and use the post-COVID recovery to build more resilient and sustainable institutions. Prioritizing energy efficiency, renewable energy, and transparent governance can help decouple growth from environmental harm. Developed countries, meanwhile, should support this transition by embedding environmental conditions in trade agreements and providing technical and financial assistance for cleaner industrial development. Further studies should investigate sector-specific patterns of emissions, the evolving role of digitalization and circular economy practices, and the effectiveness of green financing tools in curbing pollution. Understanding how these mechanisms interact with governance and global shocks like COVID-19 will be essential for designing resilient and inclusive pathways to sustainable development.