The Impact of Sub-Sector of Economic Activity and Financial Development on Environmental Degradation: New Evidence Using Dynamic Heterogeneous Panel

Abstract

While many recent studies have used the ecological footprint as a comprehensive indicator of environmental degradation instead of CO₂ emission, these were mainly focused on consumer responsibility. This study, however, aims to cover both aspects of consumption and production to elicit a more comprehensive understanding. Furthermore, this study addresses another information gap by analyzing the effect of aggregated and disaggregated economic activities on the environment. Panel data were used and sourced from 92 countries classified by income group spanning 1992 to 2015. Comprehensive financial development indicators, energy structure, energy intensity, trade openness, and urbanization were considered in examining their impacts on environmental degradation. The pooled mean group estimation was adopted in examining the long-run and short-run relationship between variables. The main findings suggest that financial development promotes green investment in high-income and upper-middle-income countries but increases degradation in lower-middle and lower-income countries. Renewable energy improves the environment in general, and energy intensity is a crucial factor in environmental modeling across all groups. Most importantly, a U-shape relationship is found on both the consumption and the production side for all income groups except for lower-income countries (inverse U-shape) on the production side. Interestingly, a U-shape relationship was found in high-income and upper-middle-income countries in the industrial sector, but a monotonic relationship in the service sector. A U-shape relationship was found for the industrial and service sectors in lower-middle-income and lower-income countries, but an inverse U-shape for agriculture outputs in lower-middle-income countries. This finding suggests the need to shift from fast-growth strategies to strategic growth planning that considers the nature of the relationship between economic sectors and the environment while diversifying the economic structure to allow for the recovery of natural capital.

1. Introduction

Climate change, desertification, and biodiversity loss are problems that have emerged from the single fact that humans are using more environmental resources and produce more pollution than the earth can regenerate and absorb [1]. Countries depend on extracting their domestic resources or importing resources from other countries to fulfil their production and consumption needs [2]. At every stage of the production and consumption processes, we use natural resources and produce waste and pollution. Accordingly, our economic activities as well as consumption patterns need to have enough biocapacity to support them [3], otherwise our environment will degrade faster.

Scientists and researchers have worked to develop different methods to measure and count natural capital productivity [2]. One of these methods is the ecological footprint developed by Wackernagel and Rees [4]. The ecological footprint accounting system answers the research question of ‘how much of the biosphere’s regenerative capacity is occupied by given human activities?’ [1]. It compares the human demand on natural resources with the biocapacity. The ecological footprint can thus be calculated for consumption and production. The ecological footprint of consumption (EF) is calculated by adding the footprint embedded in imports to the footprint of locally produced goods and services. The ecological footprint of production (EFP) is calculated by adding the footprint embedded in exports to the locally produced goods and the footprint of services.

This approach can help countries to identify how their economic growth is causing them loss in biocapacity, which could have irreversible effects in the future [5]. Understanding the impact of our economic activities on the biocapacity provides a more comprehensive view of various environmental degradation issues including climate change, biodiversity loss, desertification, drought, hunger, air, land and water pollution, public health issues, social inequality, and wars caused by an unequal distribution of natural resources [1].

A large body of literature has studied the linear and nonlinear relationships between environmental impact, mainly measured by CO₂ emission, and human activities, mainly measured by per capita income [6,7]. Furthermore, researchers have identified other factors that have impacts on CO₂ emission besides economic growth. Those factors include: energy use, trade openness, urbanization, renewable energy, financial development, human development, natural resources, technology, institutional quality, and tax system [5,8,9,10,11,12]. Recently, researchers have used the ecological footprint as a more comprehensive indicator to study the environmental-economic nexus. Many studies have examined the linear and nonlinear relationships between economic growth and the previously mentioned factors of the ecological footprint as a measurement of environmental degradation [2,13,14,15,16,17,18,19].

Previous studies focused on the EF and neglected the EFP [2,13,14,18,19]. Only a few studies, such as [20], included both EF and EFP. Hence, this study fills this gap by studying both EF and EFP. Furthermore, previous studies have examined the impact of some economic sub-sectors on CO₂ emission, such as [21,22,23,24,25,26]. However, the impact of economic sub-sectors on the ecological footprint is not examined, especially the nonlinear relationship under the Environment Kuznets curve (EKC) hypothesis. Therefore, the present study is unique in addressing the impact of economic growth on EF and EFP and in studying the per capita sectoral-wise GDP impact on EFP to measure environmental degradation under the EKC framework.

The significance of this study is its attempt to provide insightful policy recommendations on what sub-sectors of economic activities need to be regulated or promoted to achieve sustainable growth based on their impact on the environment, which has been done for each income group and both in the short and long-run analysis. Such discussions have not been addressed in the literature previously. Another significance is using the ecological footprint to measure environmental degradation. Decisions informed by ecological footprint can bring more holistic solutions compared to policies that are only informed by CO₂ emission. Policies that only target CO₂ emission, mainly focusing on reducing fossil fuel consumption, could lead to other sets of problems. For example, shifting to biofuel could result in more pressure on different environmental resources such as forests and croplands. This can cause food security issues and reduce the ecosystem’s capacity to absorb CO₂ emissions, leading to more accumulation of CO₂ in the atmosphere.

This study aims to answer the following questions that have not been addressed in the literature: (i) What is the difference between the impact of per capita income on the ecological footprint of consumption vs. the ecological footprint of production? (ii) What is the impact of the agricultural, manufacturing, and service sectors on environmental degradation in the short and long run? (iii) What is the impact of financial development, renewable energy, trade openness, and urbanization on environmental degradation in the ecological footprint of consumption vs. the ecological footprint of production?

To do so, this study utilizes the Environment Kuznets curve (EKC) hypothesis to examine the non-linear relationship between per capita income and environmental degradation in terms of EF and EFP. The study constructed five models: two models to investigate the impact of the aggregated income level against EF and EFP, and three other models to analyze the impact of economic sub-sectors, namely, agriculture, industrial, and service, on EFP. To analyze the data, the study employed an advanced econometrics tool using Autoregressive Distributed Lag (ARDL), as it is suitable when we have stationary and non-stationary variables. Specifically, the study used pooled mean group (PMG) estimation for the dynamic heterogeneous panel proposed by Ref. [27]. The PMG estimation has an interesting feature as it allows for short run heterogeneity but restricts the long run to being homogeneous. This study uses panel data for 92 countries classified into four income groups and spanning from 1992 to 2015. In addition to the variables mentioned above, the paper includes financial development, energy structure, energy intensity, trade openness, and urbanization as essential factors that affect environmental degradation as reported in the literature.

The rest of this article is organized as follows: Section 2 is a literature review, Section 3 presents the data and methodology, Section 4 presents the results and a discussion, and Section 5 is the conclusion.

2. Literature Review

The literature review is organized into three themes: first, background on the ecological footprint, followed by a discussion of the theoretical aspects of the EKC hypothesis, and finally, an explanation of the relationship between environmental degradation and other explanatory variables.

2.1. Ecological Footprint

In the same way that supply and demand concepts are fundamental to understanding any economy, ecological footprint accounting (EFA) uses a similar approach. It compares human demand for ecosystem products and services with the supply of resources from the biosphere, which is also called biocapacity (BC) [28]. In other words, human demands on land (for food, service, transportation, consumer goods, and housing) is compared to the bio-productive land supply which is categorized into five land types; namely, land to absorb CO₂ emission, built-up land for urbanization and construction needs, cropland and pasture land for food and fibres, forest for timber, and fisheries for seafood [29].

Ref. [4] developed the ecological footprint accounting system. They suggested that the ecological footprint “represents the critical natural requirements of a defined economy or population in terms of the corresponding biologically productive area” [4]. The need for such a measurement emerged from the continuous increase in human pressure on the ecosystem that has reached a point beyond the regenerative and absorption capacity of the biosystem [28]. Recent global trends in BC and EF data showed that, despite the steady increase in biocapacity since 1961 as a result of improvements in technology and management practice, humanity’s ecological footprint has increased disproportionately more [30]. The consequences of such trends are the accumulation of CO₂ in the atmosphere, loss of biodiversity, degradation of environmental capital, poverty, famine, and war [1], which are indications of unsustainability in Daly’s principle [31].

In their study, Ref. [32] found that high-income countries were characterized by high per capita consumption, high per capita footprint, and relatively small populations. For example, the top 10 countries that emit CO₂ are high-income countries [33]. Conversely, lower-income countries have less consumption and a lower per capita footprint, but higher populations. Hence, sustainability is a challenge for everyone. Developing countries argue that they need to grow first by using intensive cheap energy before they can be able to use clean energy and advanced technology to reduce CO₂ emission [34]. In their study, Ref. [10] suggested that developing countries must decrease their dependence on pollutant natural resources, mainly mining activities, through the implementation of technology. Their findings supported the fact that resource depletion degrades environmental quality. Given that economic growth depends on natural resources, when industrialization expands, resource use increases. Some exploitation exceeds the rate of the regenerative capacity of nature. Such overexploitation compromises environmental health as well as the wellbeing of people who depend on it. Therefore, the management of natural resources is crucial to preserve natural resources and people’s wellbeing [10].

2.2. EKC Hypothesis

The main goal of EKC is to explain the income-environmental nexus. Initially, Ref. [35] found an inverse U-shape relationship between income equality and per capita income. Refs. [36,37] later found the same pattern between economic outputs and environmental quality. This hypothesis suggests that economic growth is associated with environmental degradation until a certain threshold, beyond which it will be reduced by an increase in income [38]. Since then, the EKC has played an essential role in shifting policy focus from associating economic development with resource scarcity to associating income growth with improving economic quality [39]. The proponents of EKC argued that countries should focus first on fast economic growth to achieve both economic and environmental goals at a later stage [40]. The opponents, however, have contended that the notion of “grow now clean later” can subsequently cause irreversible consequences for the environment. Moreover, the EKC hypothesis is still not appropriately proven empirically, and environmental impact is unmeasurable with a single indicator like CO₂ emission which has been widely used in experimental studies [39].

Many studies have considered the relationship between CO₂ and economic growth under the EKC hypothesis. For example, Ref. [5] found evidence for the EKC in the French economy. Ref. [6] did not find significant evidence for EKC in 35 African countries. Ref. [12] found evidence for the presence of EKC in 10 Asian countries. Ref. [41] validated the presence of the EKC in India. Due to the criticism of CO₂ emission as an inappropriate measure of environmental impact, the EKC hypothesis has become increasingly studied with the use of EF [19,42,43,44,45,46,47]. These studies have arrived at varying conclusions on the existence of the hypothesis. The inconsistency could have occurred through the use of data for different variables, timespans, countries, and methodologies [48]. For example, Ref. [19] used the Augmented Mean Group method to examine the relationship between GDP, energy consumption, financial development, and EF in 11 newly industrialized countries. They found support for EKC in some countries but a U-shape relationship in other countries.

Ref. [42] studied the impact of income, renewable and non-renewable energy as a percentage of total energy, urbanization, exports and imports, and FDV on EF. Using ARDL, they found support for EKC in the MINT countries (Mexico, Indonesia, Nigeria, and Turkey). Ref. [46] further verified EKC in their study on the relationship between income, energy consumption, trade openness, FDV, and EF, using the Generalized Method of Moments. They found support in upper-middle and high-income countries but not in the low and lower-middle-income countries. Ref. [49] studied the impact of economic growth on both CO₂ and EF. They found that the EKC holds for China with respect to CO₂ emission, but India and Japan were found to have a U-shape. However, with respect to EF, EKC held for Japan but a U-shape was found in China and India. Furthermore, Ref. [50] studied the EKC for both indicators. He found that the EKC holds with respect to EF but does not hold with respect to CO₂ emission for Asian countries. However, those studies only focus on the ecological footprint of consumption and they neglect the ecological footprint of production, except for some studies, such as [20], who found support for the EKC hypothesis for the supply side but that it was not validated for the demand side in Asian countries.

Despite the enormous amount of work considering the aggregated level of economic growth, the disaggregated level of economic growth has received little attention in the literature. Ref. [23] argued that different sectors might contribute differently to the total level of environmental degradation. According to [51], countries are different in their economic growth structure. For example, agriculture is the primary income source in most developing countries [52]. In contrast, in high-income countries there is a characteristic shift from agriculture to the service sector. In this perspective, it is relevant to examine sectoral-wise the GDP and quality of the environment. Some recent studies have pointedly included economic sub-sectors [9,24,25,26,51,53,54]. Some results suggested that added manufacturing value increases environmental degradation [9]. These studies, however, have only considered CO₂ emission, while neglecting the ecological footprint on the environment.

2.3. Other Factors

The impact of human activities on the environment has also been examined through several combinations of variables. For example, financial development (FDV), which has an impact on the environment, has become an active topic in the literature [55]. Furthermore, green finance has emerged to integrate economic profit and environmental protection [56]. FDV attracts more investment and foreign direct investment (FDI) than most other economic activities [57]. FDI conversely affects the environment through three channels: technical, scale, and structural effects. According to [58], the technical impact refers to the introduction of more efficient technology, environmentally friendly solutions, and research and development. The effect of scale captures the consequences of increased economic activities, leading to more energy consumption and more pollution. The structural impact refers to the shift in the pattern of economic activities, which thus in turn depends on the productivity specialization of the country [58].

Moreover, financial development can promote environmental sustainability through the human capital channel. Countries can direct financial resources to promoting human capital, which in turn can improve the environmental quality through better resource usage and sustainability awareness. Additionally, in the presence of strong institutional quality, countries can mitigate the negative consequences of FDV on the environment. Strong institutions enable countries to implement strict laws related to financial institutions and to promote green investment [13]. Ref. [10] found that the carbon tax was the most effective method for reducing emission.

Several studies have examined the FDV-environmental nexus, but the findings have been rather mixed [59]. For example, Ref. [60] found that FDV reduces EF. Ref. [12] showed that FDV reduced CO₂ emission in 10 Asian economies, while [61] discovered that FDV reduced environmental degradation in MENA countries. Contrarily, Ref. [62] found that FDV increased EF in eight developing countries. Ref. [57] showed that FDV in 49 one-belt-one-road initiative countries increased EF and CO₂ emission, and it also increased EF in Japan. FDV is positively related to the increase of the ecological footprint in the long and short run in 18 emerging countries [13].

Other essential factors impacting the environment include renewable and non-renewable energy consumption, population, urbanization, and globalization [42,46,63]. Renewable energy was found to reduce environmental degradation while non-renewable energy increased the degradation [34]. For example, Ref. [64] analyzed the effect of renewable energy consumption on CO₂ emissions in the G-7 countries. Their results demonstrate that renewable energy had a negative pressure on CO₂ emissions. They concluded that renewable energy plays a vital role in achieving environmental sustainability; however, the mitigating effect varies across countries depending on the share of renewables in the energy mix. Ref. [65] examined the impact of natural resources (coal, oil, natural gas, and mineral rents), renewable energy consumption, and economic growth on environmental degradation. They found that renewable energy consumption helps to reduce the environmental degradation in developing, developed, and global countries, and that natural resources as well as economic growth are the main causes of environmental degradation. However, Ref. [66] found that oil and gas production reduce CO₂ emission while fossil fuel consumption increases it in Africa. The reason behind this is the environmentally friendly technologies that are used in the extraction process in those countries.

Urbanization, on the one hand, brings infrastructure improvement (e.g., transportation systems and ICT), which changes the traditional ways of production and business with more efficient solutions [67]. Conversely, urbanization induces population movement from agriculture to industrial production, leading to higher pollution and material usage [68]. Additionally, urbanization results in higher population density, higher energy demand, and consumption, causing environmental degradation [9]. Empirical studies by [24,69,70] showed that urbanization improved environmental quality. In contrast, Refs. [9,53,63] found that urbanization increased environmental degradation. Ref. [65] found that urbanization is a cause of environmental degradation in developed countries, but in developing countries it improves environmental quality. Ref. [71] found that urbanization has a positive and significant relationship with carbon dioxide emissions in middle-high and middle-low-income countries.

Trade openness (TO) allows for direct and indirect use of the environment through traded goods and global specialization [72]. In a recent study, Ref. [73] examined the effects of TO on environmental degradation in 183 countries from 1987 to 2013, using the pollution-haven hypothesis (PHH). PHH suggests that international trade is a crucial channel by which high-income countries can consume beyond their ecological burden by importing natural resources from middle and low-income countries. They showed that trade increases the ecological footprint of exports from low-income countries. However, no evidence was found for a positive increase in the ecological footprint of imports in high-income countries. They thus suggested that other factors, such as technology advancement, policies, institutions, and awareness, mitigate expectations of the PHH. The impact of globalization on the environment remains unclear [18]. For example, Ref. [48] found that TO increases EF in six Asian countries. In contrast, Ref. [61] found that TO improves the economy of MENA countries. Ref. [18] discovered the positive effect of TO in high and upper-middle-income countries in Asia while increasing environmental degradation in lower-middle-income countries.

Given the above perspective, the present study fills relevant literature gaps in the following ways: First, by using EF and EFP, the study provides a comprehensive view of the impact of per capita income on the ecological footprint of consumption and production. This is in contrast to past studies which only focus on EF measurement such as [2,13,14,18,19,43], except for [20] who studied both EF and EFP. Second, by studying the impact of the economic sub-sector on the ecological footprint. This is different from some previous studies that examined the impact of some economic sub-sectors on CO₂ emission [7,21,22,26,53,54]. One reason behind the neglect of studying the impact of economic sub-sectors on EFP could be the extensive focus on the EF of consumption, which may not accurately interpret economic sub-sectors’ impact on the environment since it includes imports. In contrast, EFP usage will provide an exciting interpretation as it measures these activities’ effects on the investigated region’s resources only. This analysis expands the scope of EKC by analysis the impact of the economic structure measured by per capita industrial, agriculture, and service income on the countries’ resource usage, which is measured by EFP.

This study, therefore, examines the per capita industrial, agriculture, and service GDP separately to identify which income-generating activity is more harmful to a country’s ecosystem under the EKC hypothesis. The reviewed literature either includes economic sectors as controlled variables besides GDP per capita and GDP per capita squared, as in [26,53], or economic sub-sectors without examining the EKC hypothesis, as in [23,54]. In a few cases, EKC was examined for the economic sub-sector but solely focused on industrial value-added as a percentage of GDP, as in [25], or as agriculture and industrial value-added, as in [24], or focused on a narrow geographical area with many economic sub-sectors, as in [51,74,75]. However, these studies examined the effect of economic activities on CO₂ emission only. The present study is thus unique in using the per capita sectoral-wise GDP effect on EFP to measure environmental degradation under the EKC framework.

3. Data and Methodology

3.1. Data Collection and Variables Definition

The study utilized panel data representing 92 countries from 1992 to 2015. The selection of countries and the period of study were subject to data availability. Because of the lack of available data for some variables, countries with missing data were not included in this study. The selected countries were classified, based on World Bank income classification (a list of those countries appears in Appendix A), into four income groups: high-income countries (HI), upper-middle-income countries (UM), lower-middle-income (LM), and low-income countries (LI). Two dependent variables were used, the ecological footprint of per capita consumption (EF) and the per capita ecological footprint of production (EFP), as proxies for environmental degradation. Both indicators were measured using a standard global hectare. Standard global hectare (gha) is a methodology to standardize each type of land productivity in different areas. There are 100 hectares in one square kilometer, and it is primarily used in the measurement of land, extracted from the Global Footprint Network (GFN). EF was essentially calculated by adding the footprint embedded in imports to the footprint of locally produced goods and services. EFP was calculated by adding the footprint embedded in exports to the locally produced goods and the footprint of the services:

Ecological footprint of consumption (EF) = ecological footprint of production (EFP) − ecological footprint of exports + ecological footprint of imports

In this study, per capita GDP was used for aggregated economic activities. For the economic sub-sectors, industrial, agriculture, and service value-added per capita were derived by dividing each value-added by the total population. The study also utilized energy structure and energy intensity following [24,25]. According to [24], energy structure (ES) is the share of clean or fossil fuel energy in total energy consumption. In this study, renewable energy consumption was used as a percentage of total energy. Energy intensity (EIN) indicates the amount of energy used to produce one unit of economic output. It was used as a proxy for technology (energy intensity level is only an imperfect proxy for an energy efficiency indicator and it can be affected by a number of factors not necessarily linked to pure efficiency, such as climate). Trade openness (TO) is the exports and imports as a percentage of GDP. Urbanization (UR) is the share of urban population relative to the total population. All the above data were sourced from the World Bank.

For financial development indicators, the FDV index of the International Monetary Fund (IMF) was used, following [54,55]. The index summarizes the level of development of both financial institutions and financial markets in terms of depth (size and liquidity), access (the ability of individuals and businesses to access the financial services), and efficiency (low-cost service with sustainable revenue and the level of activity of the capital market) [76]. Unlike the widely used single measurements, such as domestic credit to the private sector, FDV reduces the bias of omitting variables. It also provides a comprehensive representation of diverse financial systems across countries [62] that are dependent on the countries’ monetary policies [76]. Such an index is essential in this study which compares a large number of countries. In addition, adding TO to the model helps avoid specification bias in the relationship between EF and financial development [57].

Table 1. Study variables and data sources.

Variable	Unit of Measurement	Source
Ecological footprint of consumption (EF)	Average global hectares (AGH)	Global Footprint Network (GFN)
Ecological footprint of production (EFP)	Average global hectares (AGH)	Global Footprint Network (GFN)
log of GDP per capita (ln GDP)	Constant 2010$	World Bank (WB)
log of per capita industrial value-added (ln IVA),	Constant 2010$	World Bank (WB)
log of per capita agriculture value-added (ln AVA)	Constant 2010$	World Bank (WB)
log of per capita service value-added (ln SVA)	Constant 2010$	World Bank (WB)
Financial development (FDV)	IMF index	International Monetary Fund (IMF)
Energy structure: renewable energy (ES)	Renewable energy as % of total energy	World Bank (WB)
log of energy intensity: proxy of technology (ln ein)	Energy per unit of GDP	World Bank (WB)
Globalization: trade openness (to)	Export+imports/gdp	World Bank (WB)
Urbanization (UR)	% total population	World Bank (WB)

summarizes the variables used in this study.

3.2. Methodology

Following the original Kuznets Curve [73], the majority of papers that have investigated EKC included log GDP in the quadratic form (see [19,23,24,25,77]). In some studies, the cubic of log GDP was used to obtain an N-shape relationship (see [55,78]). Additional variables that were studied included urbanization in [77], energy intensity in [25], as well as other variables such as trade openness, non-renewable and renewable energy, and financial development (see [46,48,54,60]). Following the traditional approach, log GDP in a quadratic form was used in the first two models that examined EF and EFP, respectively. Similarly, the following last three models, Models 3, 4, and 5, which investigated the effect of three economic sub-sectors (respectively, industrial, agriculture, and service GDP per capita) on EFP, were used in a quadratic form.

Thus, the five basic models were written as follows:

$$\mathrm{EF}=(\ln \mathrm{GDP}, \ln \mathrm{sqGDP},\mathrm{FDV}, \mathrm{ES}, \ln \mathrm{ein}, \mathrm{TO}, \mathrm{UR})$$

(1)

$$\mathrm{EFP}=(\ln \mathrm{GDP},\ln \mathrm{sqGDP}, \mathrm{FDV}, \mathrm{ES}, \ln \mathrm{ein}, \mathrm{TO}, \mathrm{UR})$$

(2)

$$\mathrm{EFP}=\left(\ln \mathrm{iva},\ln \mathrm{sqiva}, \mathrm{FDV}, \mathrm{ES}, \ln \mathrm{ein}, \mathrm{TO}, \mathrm{UR}\right)$$

(3)

$$\mathrm{EFP}=(\ln \mathrm{ava},\ln \mathrm{sqava}, \mathrm{FDV}, \mathrm{ES}, \ln \mathrm{ein}, \mathrm{TO}, \mathrm{UR})$$

(4)

$$\mathrm{EFP}=(\ln \mathrm{sva}, \ln \mathrm{sqsva}, \mathrm{FDV}, \mathrm{ES}, \ln \mathrm{ein}, \mathrm{TO}, \mathrm{UR})$$

(5)

where EF is the ecological footprint of consumption, EFP is the ecological footprint of production, ln GDP is the log of per capita GDP, ln sqGDP is the square of the log of per capita GDP, FDV is the financial development indicator, ES is energy structure measured by renewable energy consumption, ln ein is the log of energy intensity, TO is trade openness, and UR is urbanization. In addition, ln iva, ln ava, and ln sva are, respectively, the industrial, agriculture, and service values added per capita, whereas ln sqiva, ln sqava, and ln sqsva are, respectively, their squared values. Specifically, Models 1 and 2 examine the effect of per capita income and other explanatory variables on EF and EFP, respectively, whereas Models 3, 4, and 5, respectively, examine the impact of the industrial, agriculture, and service sectors and other explanatory variables on EFP.

Firstly, to evaluate cross-sectional dependency among variables, the study adopted the Pesaran (2004) CD test [79], where the null hypothesis is cross-sectional independence. The study then proceeded to check whether the variables were stationary by using the Im–Pesaran–Shin (IPS) test, as proposed by [80], and the cross-sectional augmented Dickey–Fuller (CADF) test by [81] for panel unit root in the presence of cross-sectional dependency. The null hypothesis for both tests was that all series were non-stationary.

For data analysis, according to [23,77,82] the ARDL panel dynamic approach offers several interesting features. For example, some variables may be stationary I(0) and others non-stationary I(1). The panel ARDL approach includes two different estimators: the mean group (MG) estimator and the pooled mean group (PMG) estimator. The MG estimator proposed by [83] allows the coefficients to be heterogeneous in the long and short-runs. Ref. [27] later proposed that the PMG estimator allow the intercept, short-run coefficient, and error variance to differ freely across countries, but constrain the long-run coefficient to be identical. By including the ARDL (p, q) lags for the dependent (p) and independent variables (q), the framework is developed as follows:

$$Y_{it}={\displaystyle \sum }_{j=1}^p {\lambda }_{ij} Y_{it-j}+{\displaystyle \sum }_{j=0}^q {\delta }_{ij} x_{it-j}+{\mu }_i+{\epsilon }_{it}$$

(6)

where i = 1, 2, …, N indicates groups (countries), and t =1, 2, 3, …T indicates time (annual), and p and q represent the number of optimum time lags. In Equation (6), Y represents EF per capita or EFP per capita for i countries and t periods as the dependent variable; $x_{it}$ is the vector of independent variables, namely, ln GDP, ${\left(\ln \mathrm{GDP}\right)}^2$ or ln iva$, {\left(\ln \mathrm{iva}\right)}^2$ or ln ava$, {\left(\ln \mathrm{ava}\right)}^2$ or ln sva$, {\left(\ln \mathrm{sva}\right)}^2$, FDV, ES, ln EIN, TO, and UR, while ${\mu }_i$ represents the fixed effects. After parameterization and grouping the variables in levels, the equation will be as follows:

$$\Delta Y_{it}={\varnothing }_i(y_{it-1}-{\theta }_i^{’}{x}_{it})+{\displaystyle \sum }_{j=1}^{p-1} {\lambda }_{ij} \Delta Y_{it-j}+{\displaystyle \sum }_{j=0}^q {\delta }_{ij}^{{*}^{\prime }} \Delta x_{it-j}+{\mu }_i+{\epsilon }_{it}$$

(7)

where ${\theta }_i=-\left(\frac{{\beta }_i}{{\varnothing }_i}\right)$indicates the long-run or the equilibrium relation between EF or EFP and the independent variables $x_{it}$. ${\lambda }_{ij}^{*}$ and ${\delta }_{ij}^{*}$ are short-run coefficients connected to its lag value and $x_{it}$. ${\varnothing }_i$ is the error-correction coefficient that estimates the speed of adjustment of EF or EFP into its long-run equilibrium. The error correction speed of adjustment (ec) should be between zero and a negative value under the prior assumption that the variable shows a return to long-run equilibrium. If ec is positive and larger than one, there will be no equilibrium. If it is positive and less than one, there will be no correction to the equilibrium. The hypothesis of homogeneity of the long-run parameters cannot be assumed a priori, but it is tested empirically by a Hausman-type test, as proposed by [84]. The Hausman test can determine the effect of heterogeneity on the means of the coefficients. If the parameters are homogenous, the PMG estimates are more efficient than MG and vice versa. The given random error (Ԑ) is assumed to be normally distributed at zero mean value and constant variance [85].

4. Results and Discussion

Table 2. Pesaran (2004) CD test for cross-sectional dependency.

	All		HI		UM		LM		LI
Variables	Statistic		Statistic		Statistic		Statistic		Statistic
ef	12.3	***	21.7	***	11.9	***	4.34	***	2.49	**
efp	4.89	***	16.00	***	10.9	***	7.16	***	0.84
lgdp	201	***	70.8	***	71.9	***	58	***	17.1	***
lsqgdp	201	***	70.7	***	72	***	58.1	***	17.3	***
fdv	112	***	71.7	***	41.3	***	21.4	***	5.94	***
es	8.84	***	39	***	7.33	***	19	***	28.8	***
lein	140	***	58.7	***	36.1	***	36.3	***	11	***
to	67.2	***	58.6	***	13.4	***	1.26		16.4	***
ur	185	***	40.5	***	61.7	***	19.3	***	64.1	***
liva	98.5	***	23.7	***	48.3	***	32.7	***	14.2	***
lsqiva	98.6	***	23.7	***	48.3	***	32.9	***	14.6	***
lava	42.4	***	0.67		28.7	***	26.3	***	3.15	***
lsqava	42.4	***	0.64		28.7	***	26.4	***	3.11	***
lsva	212	***	69	***	72.4	***	56.5	***	24.7	***
lsqsva	212	***	68.9	***	72.5	***	56.7	***	24.8	***

Note: *** p < 0.01, and ** p < 0.05.

presents the results of the Pesaran (2004) CD test. It shows a cross-sectional dependency, thus rejecting the null hypothesis for most variables except for some variables, indicating the need for both first and second-generation unit root tests.

Table 3. Im–Pesaran–Shin unit-root test.

	All				HI				UM				LM				LI
Variables	At Level		At First Difference		At Level		At First Difference		At Level		At First Difference		At Level		At First Difference		At Level		At First Difference
ef	−1.39	*	−25.38	***	−0.89		−14.51	***	−2.661	***	−12.8	***	1.5599		−11.14	***	−0.521		−12.05	***
efp	0.9		−25.03	***	2.457		−14.27	***	−0.403		−12.4	***	0.7353		−11.51	***	−1.306	*	−11.72	***
lgdp	8.37		−17.29	***	−1.79	*	−8.471	***	5.762		−7.92	***	10.845		−8.331	***	3.218		−10.09	***
lsqgdp	9.72		−17.23	***	−1.47	*	−8.549	***	6.679		−7.87	***	11.94		−8.249	***	3.667		−9.989	***
fdv	−2.49	***	−23.58	***	−4.42	***	−12.28	***	2.451		−12.4	***	0.6946		−10.32	***	−3.388	***	−12.09	***
es	5.72		−22.22	***	8.684		−11.99	***	−2.129	**	−11.5	***	1.2947		−10.94	***	2.899		−9.926	***
lein	7.69		−23.53	***	6.327		−14.36	***	−14.36		−12.1	***	6.4422		−10.56	***	0.025		−9.628	***
to	−1.48	*	−23.45	***	0.673		−13.07	***	−2.021	**	−11.7	***	0.3812		−10.35	***	−2.142	**	−11.62	***
ur	2.82		−2.902	***	0.026		−1.336	*	1.955		−1.68	**	−0.686		−6.579	***	−0.337		−2.773	***
iva	3.14		−17.31	***	−0.45		−10.24	***	1.958		−9.07	***	3.802		−7.743	***	1.414		−7.297	***
ava	1.66		−24.74	***	−3.06	***	−13.73	***	0.928		−12.4	***	4.2396		−10.81	***	2.069		−12.41	***
sva	9.38		−15.91	***	−2.71	***	−7.105	***	7.448		−7.84	***	10.433		−7.631	***	5.041		−9.543	***
lsqiva	4.23		−17.19	***	−0.3		−10.27	***	2.657		−8.98	***	4.5889		−7.744	***	2.057		−7.134	***
lsqava	1.97		−24.68	***	−3.1	***	−13.73	***	1.223		−12.4	***	4.5488		−10.77	***	2.161		−12.35	***
lsqsva	11		−15.75	***	−2.32	**	−7.192	***	8.666		−7.75	***	11.72		−7.383	***	5.509		−9.42740.0000	***

Note: *** p < 0.01, ** p < 0.05, * p < 0.1.

and

Table 4. Pesaran’s CADF test.

	All				HI				UM				LM				LI
Variables	At Level		At First Difference		At Level		At First Difference		At Level		At First Difference		At Level		At First Difference		At Level		At First Difference
ef	4.04		−16.744	***	1.552		−10.84	***	−1.718	**	−8.15	***	−0.697		−6.733	***	1.022		−8.683	***
efp	4.92		−16.369	***	1.771		−8.569	***	−1.202		−8.78	***	−1.382	*	−6.137	***	−0.221		−8.03	***
lgdp	−3.53	***	−10.219	***	−2.91	***	−5.565	***	−1.59	*	−5.5	***	−2.923	***	−6.062	***	−0.075		−7.026	***
lsqgdp	−3.13	***	−9.405	***	−2.65	***	−5.443	***	−1.425	*	−5.3	***	−2.738	***	−6.027	***	−0.025		−7.1	***
fdv	−2.67	***	−16.289	***	−2.28	**	−8.603	***	−0.928		−7.59	***	−1.049		−8.543	***	−4.111	***	−8.873	***
es	1.55		−12.389	***	1.72		−7.613	***	−2.121	**	−7.67	***	−0.202		−4.793	***	−1.059		−6.73	***
lein	−4.02	***	−15.217	***	−0.38		−8.578	***	−0.873		−6.57	***	−1.973	**	−7.420	***	−1.225		−5.961	***
to	−3.58	***	−13.412	***	−1.62	*	−5.666	***	−1.009		−7.6	***	1.666		−5.795	***	−1.661	**	−7.756	***
ur	−1.17		−12.26	***	2.837		−1.313	*	2.698		−2.15	**	4.27		−1.984	**	2.733		−2.267	***
iva	−0.98		−9.670	***	−0.63		−6.849	***	−1.899	**	−5.12	***	−4.201	***	−5.236	***	−0.04		−6.217	***
ava	−2.02	**	−14.141	***	−0.09		−9.03	***	−2.637	***	−7.79	***	−1.208		−8.039	***	1.169		−8.323	***
sva	0.22		−9.015	***	−0.88		−5.602	***	−0.875		−4.4	***	−3.013	***	−6.404	***	1.19		−7.153	***
lsqiva	−0.9		−9.264	***	−0.36		−6.475	***	−1.588	*	−4.69	***	−3.95	***	−5.295	***	−0.253		−6.735	***
lsqava	−2.24	**	−14.109	***	0.176		−8.998	***	−2.572	***	−7.71	***	−1.159		−8.013	***	1.216		−8.304	***
lsqsva	0.56		−8.084	***	−0.72		−5.518	***	−0.806		−4.48	***	−2.793	***	−6.203	***	1.254		−6.986	***

Note: *** p < 0.01, ** p < 0.05, * p < 0.1.

provide the Im–Pesaran–Shin [80] unit-root test and Pesaran’s CADF test [81]. Both tests indicate that some variables are stationary at the level, and all variables are stationary at the first difference.

Table 5. Hausman-type test.

Model	chi2	prob>chi2
EF	0.37	0.999
EFP	0.65	0.996

presents the findings of the Hausman-type test. Under the null hypothesis, the difference in the estimated coefficients between the MG and PMG is not significant. Accordingly, PMG is more efficient in our case. This implies that the impact of the independent variables follows a heterogeneous pattern in the short run and a homogenous one in the long run.

For all the five models, the results were estimated under pooled PMG following [27]. For optimal lag selection, several models were examined (represented in

Table A2. Model 1.

EF	1		2		3		4
	aic	bic	aic	bic	aic	bic	aic	bic
overall	−4547.130	−4457.090	−4494.880	−4404.840	−3790.428	−3700.387	−4098.921	−4008.881
HI	−349.703	−259.663	−385.223	−295.182	−106.853	−16.812	−238.354	−148.313
UM	−957.151	−867.110	−946.852	−856.812	−682.590	−592.549	−858.326	−768.285
LM	−1461.190	−1371.150	−1430.552	−1340.511	−1279.302	−1189.261	−1318.290	−1228.249
LI	−1875.313	−1809.565	−1847.981	−1757.940	−1726.203	−1636.162	−1734.708	−1644.668

Lag-selection decision, for ARDL models: 1:(10000000), 2:(11111111), 3:(10010010), 4:(11100011).

,

Table A3. Model 2.

EFP	1		2		3		4		5		6
	aic	bic	aic	bic	aic	bic	aic	bic	aic	bic	aic	bic
overall	−6184.207	−6094.166	-	-	-	-	−5389.307	−5299.266	−5394.248	−5304.207	−5465.327	−5375.286
HI	−1061.481	−971.440	−1002.278	−912.238	−674.569	−584.529	−691.592	−601.551	-	-	-	-
UM	−1377.139	−1287.098	-	-	−1104.300	−1014.260	-	-	−1172.490	−1082.450	−1212.656	−1122.615
LM	−1762.140	−1697.611	−1741.859	−1678.036	-	-	−1636.106	−1572.283	-	-	−1657.505	−1593.682
LI	−2088.319	−2024.496	−1997.142	−1933.319	−1857.874	−1794.051	−1873.631	−1809.808	-	-	-	-

Lag-selection decision, for ARDL models: 1:(10000000), 2:(11111111), 3:(10010010), 4:(11100011), 5:(10011000), 6:(11110000).

,

Table A4. Model 3.

iva Model	1		2		3		4
	aic	bic	aic	bic	aic	bic	aic	bic
overall	−5883.845	−5820.021	-	-	−5302.86	−5239.037	−5340.661	−5276.837
HI	−994.687	−930.864	−720.5491	−656.7257	-	-	−706.6225	−642.7991
UM	−1335.834	−1272.011	-	-	−1156.692	−1092.869	−1146.048	−1082.224
LM	-	-	−1547.138	−1483.315	−1576.151	−1512.328	−1607.273	−1543.45
LI	−1954.009	−1890.186	−1807.652	−1743.829	−1864.631	−1800.807	-	-

Lag-selection decision, for ARDL models: 1:(10000000), 2:(10001111), 3:(10011000), 4:(10010101).

,

Table A5. Model 4.

ava Model	1		2		3		4
	aic	bic	aic	bic	aic	bic	aic	bic
overall	−6150.76	−6086.94	−5534.9	−5445.85	−5358.87	−5269.82	-	-
HI	−976.947	−913.124	−765.32	−676.268	−733.674	−644.621	−893.988	−804.936
UM	−1455.48	−1391.65	−1209.78	−1120.73	−1167.51	−1078.46	−1388.51	−1299.46
LM	-	-	−1660.87	−1597.04	−1669.68	−1605.85	−1610.26	−1544.52
LI	−2074.59	−2010.77	−1916.39	−1827.33	−1874.33	−1785.28	-	-

Lag-selection decision, for ARDL models: 1:(10000000), 2:(10011111), 3:(10011000), 4:(10001111).

and

Table A6. Model 5.

sva Model	1		2		3		4
	aic	bic	aic	bic	aic	bic	aic	bic
overall	−6025.240	−5961.417	−5070.869	−4981.816	−5704.886	−5615.834	−5506.444	−5417.392
HI	−1045.653	−981.830	−652.971	−563.919	−885.970	−796.918	−807.9554	−718.9027
UM	−1459.932	−1396.109	-	-	−1271.066	−1182.013	−1192.105	−1103.052
LM	-	-	−1559.797	−1470.744	-	-	−1569.446	−1505.623
LI	−1973.585	−1909.762	−1834.122	−1745.069	−1903.019	−1813.966	−1827.627	−1738.575

Lag-selection decision, for ARDL models: 1:(10000000), 2:(11110000), 3:(11111111), 4:(10001111).

in Appendix A). The decision was based on the lowest Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC).

The error correction coefficient results are negative and less than one, and significant for all five models, suggesting a significant long-run dynamic relationship between EF and EFP and all explanatory variables.

Table 6. (Model 1): The long and short-run relation between EF and ln GDP, ln GDP square, FDV, ES ln ein, TO, UR.

EF	Overall		HI		UM		LM		LI
	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value
Long run
ln GDP	−2.470	0.000	−19.124	0.000	−6.687	0.000	−3.481	0.000	−3.688	0.012
ln sqGDP	0.216	0.000	1.213	0.000	0.545	0.000	0.313	0.000	0.283	0.015
FDV	0.281	0.000	−0.387	0.106	−2.885	0.000	−0.00006	0.002	1.042	0.000
ES	−0.004	0.000	−0.032	0.001	−0.018	0.000	0.000	0.970	0.000	0.898
ln ein	0.433	0.000	3.874	0.000	0.665	0.000	0.743	0.000	0.297	0.000
TO	0.001	0.056	−0.009	0.000	0.000	0.552	0.002	0.000	−0.001	0.148
UR	−0.007	0.002	−0.007	0.661	−0.050	0.000	−0.012	0.001	0.022	0.000
Short run
ec	−0.446	0.000	−0.575	0.000	−0.641	0.000	−0.568	0.000	−0.386	0.000
∆ln GDP	44.374	0.169	150.315	0.152	2.384	0.926	13.631	0.417	−17.028	0.054
∆ln sqGDP	−1.707	0.258	−6.412	0.183	−0.023	0.987	−0.860	0.427	1.313	0.069
∆FDV	−0.488	0.223	−1.154	0.417	−0.023	0.975	0.080	0.855	0.008	0.982
∆ES	0.093	0.655	0.353	0.590	−0.012	0.596	−0.020	0.002	0.001	0.700
∆ln ein	0.940	0.001	1.949	0.018	0.244	0.359	−0.463	0.101	−0.059	0.577
∆TO	0.001	0.715	0.008	0.040	0.001	0.731	−0.001	0.157	0.000	0.552
∆UR	−0.521	0.388	−1.113	0.324	−1.342	0.040	0.037	0.818	−0.747	0.075
α	3.539	0.000	40.382	0.000	15.286	0.000	5.111	0.000	4.799	0.000

Note: HI = high-income countries; UM = upper-middle-income countries; LM = lower-middle-income countries; LI = low-income countries. Ln GDP = natural logarithm of GDP. For all models PMG ARDL(1,0,0,0,0,0,0,0) is used.

and

Table 7. (Model 2): The long and short-run relation between EFP and ln GDP, ln GDP square, FDV, ES ln ein, TO, UR.

EFP	Overall		HI		UM		LM		LI
	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value
Long run
ln GDP	−1.650	0.000	−5.925	0.011	−7.172	0.000	−0.899	0.020	7.215	0.000
ln sqGDP	0.165	0.000	0.536	0.000	0.495	0.000	0.119	0.000	−0.547	0.000
FDV	−0.038	0.330	−1.538	0.000	−0.967	0.000	−0.055	0.176	2.309	0.000
ES	−0.007	0.000	−0.065	0.000	−0.047	0.000	−0.006	0.000	−0.002	0.011
ln ein	0.516	0.000	3.279	0.000	0.558	0.000	0.615	0.000	0.206	0.000
TO	0.002	0.000	0.000	0.795	0.002	0.002	0.001	0.000	−0.002	0.000
UR	−0.013	0.000	−0.057	0.002	−0.011	0.011	−0.013	0.000	0.000	0.784
Short run
ec	−0.347	0.000	−0.395	0.000	−0.330	0.000	−0.492	0.000	−0.455	0.000
∆ln GDP	23.437	0.348	82.101	0.262	2.907	0.833	−12.665	0.545	3.220	0.829
∆ln sqGDP	−1.023	0.391	−3.688	0.279	−0.087	0.910	0.749	0.564	−0.289	0.810
∆FDV	0.039	0.848	1.026	0.024	−0.345	0.422	−0.062	0.876	−0.425	0.023
∆ES	−0.067	0.624	0.089	0.730	−0.017	0.299	−0.010	0.004	0.001	0.783
∆ln ein	0.889	0.000	1.679	0.002	0.517	0.025	−0.233	0.091	0.063	0.614
∆TO	0.002	0.091	0.001	0.708	0.003	0.197	−0.001	0.016	0.001	0.162
∆UR	−0.073	0.623	0.654	0.503	0.107	0.402	−0.009	0.906	−0.391	0.011
α	1.703	0.000	4.058	0.000	9.416	0.000	0.507	0.001	−10.359	0.000

Note: HI = high-income countries; UM = upper-middle-income countries; LM = lower-middle-income countries; LI = low-income countries. Ln GDP = natural logarithm of GDP. For all models PMG ARDL(1,0,0,0,0,0,0,0) is used.

summarized the findings of the first and second models. Both tables show that most variables are significant across all income groups in the long run but not significant in the short run. In the long run, GDP and GDP squared have negative and positive signs, respectively, in both tables except for LI countries in the EFP model. Hence, a U-shape relationship exists between income and both EF and EFP in HI, UM, and LM countries, which means no support for the EKC in these groups. However, for LI countries the relationship with EF was U-shaped, and with the EFP model it was inverse U-shaped, supporting the EKC hypothesis.

For HI-countries, a 1% increase in income reduces EF and EFP by 19 and 7 gha, respectively, but a further increase in income will reduce EF and EFP by 1.2 and 0.5, respectively. Hence, the effect of income is larger in reducing the consumption effect on the environment than the effect of the production. In UM countries, a 1% increase in income reduces EF and EFP by 6.7 and 7.3 gha, respectively. A further increase in income increases EF and EFP by 0.5 gha. Hence, in those countries, an increase in income has similar effects on reducing the consumption and production consequences. An increase in income in LM countries has less impact on reducing EF and EFP (3.5 and 0.9 gha, respectively). However, for LI countries, a 1% increase in income will reduce EF by 3.5 but increase EFP by 7 gha. This shows that the increase in income in those countries contributes to increase the degradation coming from the production side.

The findings of the existence of a U-shape in some income groups are consistent with [86] for OECD countries, Ref. [18] for west, south, and southeast Asian countries, and with [46] for LI and LM countries. However, the findings contradict those of [50] for some Asian countries, Ref. [19] for 11 newly industrialized countries, Ref. [42] for MINT countries, Ref. [87] for Asia Pacific Economic Cooperation (APEC), Ref. [77] for LI, UM, LM, and HI countries, Ref. [46] for HI, UM, and LM countries, and [88] for Malaysia, who found support for EKC.

To summarize the finding regarding the income effect on the environment, both models exhibit similar behavior for the variables, except for LI countries and the magnitude of income that affects each measurement. For example, income in HI countries reduces the ecological footprint of consumption substantially (about threefold) compared to the ecological footprint of production, suggesting a presence of awareness or regulation imposed against polluted imports. For LI countries, an increase in income will lead to a better quality of imports and an increase in environmental degradation from exports. Part of the U-shape curve found in this study may indicate that some countries are already on the downward slope of the inverse U-shape curve proposed by EKC, while other countries are yet to catch up. However, at the same time, it contradicts the hypothesis which indicates that an increase in income will guarantee an environmental improvement as the relationship will be inversed beyond a certain threshold in our model.

The coefficient for financial development is significant across all income levels except for HI countries in the EF model. The results indicate that a 1% increase in FDV reduces EFP by 1.5 gha in HI countries and reduces EF and EFP in UM countries by 3 and 1 gha, respectively. A similar pattern occurs for LM countries but with a lower magnitude. These results are in line with [12,60,61,89], who found that FDV improves the environment. In contrast, FDV significantly increases EF and EFP in LI countries by 1 and 2 agh, respectively. Similar results were found by [55,62,63,88]. These findings suggest that investment in HI, UM, and LM countries leads to green projects. However, in LI countries, investment is toward intensive consumption of environmental resources and higher pollution, thus indicating the pollution-haven hypothesis in LI countries, where dirty production is moved from higher-income to lower-income countries.

Energy structure (i.e., renewable energy share) is significant and negative in HI and UM in both models, with a value of around .04 gha. In LM and LI, renewable energy consumption is only significant in the EFP model with minimal effect. These results concur with a major part of the literature that supports the mitigation effects of renewable energy [50,51,55,86,90]. The authors found that renewable energy reduces environmental degradation. The results indicate that the strategies in HI and UM countries to use renewable energy reduce the negative effects of economic growth, but their contribution is still small. However, renewable energy utilization was not very useful in LM and LI countries.

As expected, energy intensity has a significant positive impact on both models across income groups. The highest impact is in HI countries, where a 1% increase in energy used per unit of GDP increases the EF and EFP by around 4 and 3 gha, respectively. In other income groups, EF and EFP increased by less than 1 gha. Similar results were found by [24,25,91]. The findings indicate that energy efficiency is more critical in HI countries, given its high impact on both ecological footprints. Also, these results show that energy efficiency is a crucial factor for environmental control efforts as it has significant positive effect on all countries. Hence, spreading new technologies that reduce energy use is a fast solution to overcome environmental challenges.

Trade openness brings mixed results in both models with minimal consequences for environmental degradation (around 0.002 gha). The results suggest that TO reduces EF in HI, increases EFP in UM and LM, and decreases EFP in LI countries. Refs. [46,48,62] also found that trade openness increases environmental degradation. In contrast, Refs. [16,92,93] found that TO improves environmental quality.

Lastly, the urbanization coefficient is statistically significant in reducing EF and EFP in UM and LM countries, but in HI countries it reduces EFP only and does not affect EF. However, in LI countries, urbanization increases EF. Refs. [9,90] found that urbanization decreases environmental quality, as did [71] for middle-high and middle-low-income countries. This contradicts the findings of [93] who found that urbanization improves environmental quality globally. The reason behind this finding could be that urbanization in UM, LM, and HI countries results in better technology and ICT infrastructure and a shift to cleaner income-generating activities, such as service activities that reduce energy usage [68]. Urbanization in LI countries could be associated with higher energy usage by shifting from agrarian to industrial activities in the cities.

As stated earlier, the short-run relationship is only significant for some variables. For example, FDV has a short-run effect on HI countries by increasing EFP, but FDV reduces EFP in LI countries in the short run. This suggests that for HI countries, FDV leads to investment in economic activities that initially degrade the environment but become sustainable in the long run. However, in LI countries, FDV reduces EFP at the onset by providing better alternatives to the old production practices. Nevertheless, the structural effect begins to produce more pollution and exert environmental pressure due to the shift to industrial projects in the long run. Renewable energy, however, has only a short-run impact in LM countries on EF and EFP that reduces them by 0.02 and 0.01 gha, respectively, which suggests that the shift from non-renewable to renewable energy has an immediate impact on these countries. Moreover, it is confirmed that energy intensity is a critical factor in HI countries as it increases both EF and EFP by around 1.7 gha in the short run, which indicates an immediate effect. Similarly, in UM countries, energy intensity increases EFP by 0.5 gha. However, in LM countries, more energy per unit of GDP reduces EFP, thus indicating that with old technology more energy usage by a unit of GDP leads to higher production costs, leading some businesses in LM countries to be forced out of the market. The speed of adjustment indicates that the variables converged to a long-run relationship after 1.6 to 2.6 years for the EF model and after 2 to 3 years for the EFP model.

Table 8. The impact of the industrial sector (iva) on EFP.

EFP	Overall		HI		UM		LM		LI
	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value
Long run
ln iva	−1.417	0.000	−8.225	0.000	−5.681	0.000	−0.870	0.054	−0.030	0.782
ln sqiva	0.205	0.000	0.608	0.000	0.458	0.000	0.104	0.007	0.009	0.441
FDV	−0.244	0.043	−0.264	0.072	−0.958	0.000	0.430	0.002	0.397	0.063
ES	−0.008	0.000	−0.049	0.000	−0.039	0.000	0.005	0.001	−0.004	0.000
ln ein	0.595	0.000	1.811	0.000	0.564	0.000	0.433	0.000	0.080	0.023
TO	0.000	0.570	0.002	0.074	0.002	0.001	0.000	0.929	0.001	0.008
UR	−0.025	0.000	0.031	0.037	−0.010	0.004	0.013	0.000	−0.014	0.000
Short run
ec	−0.281	0.000	−0.398	0.000	−0.340	0.000	−0.481	0.000	−0.426	0.000
∆ln iva	30.635	0.071	111.051	0.024	8.716	0.663	−20.657	0.347	−1.455	0.557
∆ln sqiva	−1.520	0.108	−5.770	0.027	−0.397	0.748	1.410	0.354	0.156	0.541
∆FDV	0.246	0.387	0.959	0.111	−0.421	0.314	−0.114	0.757	0.327	0.422
∆ES	−0.026	0.810	0.091	0.729	−0.022	0.176	−0.003	0.629	0.001	0.825
∆ln ein	0.580	0.005	1.534	0.005	0.388	0.079	0.095	0.243	−0.118	0.394
∆TO	0.003	0.005	0.006	0.019	0.003	0.099	0.001	0.330	0.000	0.694
∆UR	−0.071	0.724	0.123	0.863	0.259	0.407	0.226	0.069	−0.190	0.561
α	0.876	0.000	10.290	0.000	6.826	0.000	0.491	0.006	0.606	0.021

Note: HI = high-income countries; UM = upper-middle-income countries; LM = lower-middle-income countries; LI = low-income countries. Ln iva = natural logarithm of industrial value-added, ln sqiva = square log of industrial value added. For all models PMG ARDL(1,0,0,0,0,0,0,0) is used, except for LM (ARDL(1,0,0,0,1,1,1,1)).

,

Table 9. The impact of the agriculture sector (ava) on EFP.

EFP	Overall		HI		UM		LM		LI
	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value
Long run
ln ava	−10.186	0.000	−5.006	0.104	−13.043	0.000	6.306	0.000	−0.218	0.185
ln sqava	0.977	0.000	0.640	0.011	1.263	0.000	−0.545	0.000	0.054	0.002
FDV	0.215	0.000	0.355	0.321	0.272	0.256	0.430	0.000	0.454	0.000
ES	−0.003	0.000	−0.161	0.000	0.008	0.032	0.002	0.231	0.000	0.574
ln ein	0.218	0.000	1.783	0.000	0.286	0.008	0.183	0.001	0.112	0.000
TO	0.002	0.000	0.014	0.000	0.002	0.045	−0.001	0.018	0.001	0.000
UR	−0.002	0.161	0.073	0.000	−0.022	0.000	0.017	0.000	−0.001	0.431
Short run
ec	−0.277	0.000	−0.200	0.002	−0.338	0.000	−0.471	0.000	−0.391	0.000
∆ln ava	−0.997	0.897	−18.353	0.304	12.441	0.420	−5.041	0.428	6.206	0.373
∆ln sqava	0.029	0.962	1.388	0.305	−1.053	0.392	0.443	0.428	−0.557	0.404
∆FDV	0.144	0.595	0.682	0.378	−0.517	0.331	0.224	0.541	0.135	0.513
∆ES	0.068	0.436	0.290	0.280	−0.060	0.003	−0.005	0.272	0.004	0.179
∆ln ein	0.816	0.000	2.173	0.000	0.327	0.124	0.067	0.501	0.010	0.915
∆TO	0.004	0.026	0.010	0.105	0.001	0.131	0.001	0.199	0.000	0.552
∆UR	−0.032	0.854	0.288	0.462	0.120	0.547	0.075	0.508	−0.365	0.023
α	8.006	0.000	0.843	0.003	12.412	0.000	−8.421	0.000	0.299	0.001

Note: HI = high-income countries; UM = upper-middle-income countries; LM = lower-middle-income countries; LI = low-income countries. Ln ava = natural logarithm of agriculture value-added, ln sqava = square log of agriculture value added. For all models PMG ARDL(1,0,0,0,0,0,0,0) is used, except for LM (ARDL(1,0,0,0,1,1,1,1)).

and

Table 10. The impact of the service sector (sva) on EFP.

EFP	Overall		HI		UM		LM		LI
	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value	Coefficient	p-Value
Long run
ln sva	0.265	0.000	3.921	0.000	0.414	0.000	−4.755	0.000	−0.829	0.001
ln sqsva	0.135	0.000	0.082	0.000	0.133	0.000	0.339	0.000	0.079	0.001
FDV	0.497	0.000	−1.907	0.000	0.621	0.001	0.722	0.003	−0.381	0.064
ES	0.000	0.716	−0.084	0.000	−0.029	0.000	0.002	0.141	−0.006	0.000
ln ein	0.690	0.000	3.358	0.000	0.328	0.003	0.094	0.219	0.288	0.000
TO	0.002	0.000	0.006	0.000	0.002	0.054	0.000	0.808	−0.002	0.000
UR	−0.040	0.000	0.055	0.000	−0.050	0.000	0.009	0.047	0.004	0.068
Short run
ec	−0.296	0.000	−0.445	0.000	−0.432	0.000	−0.448	0.000	−0.390	0.000
∆ln sva	33.679	0.249	80.326	0.386	−5.053	0.674	−30.778	0.339	0.607	0.951
∆ln sqsva	−1.741	0.236	−3.677	0.408	0.355	0.619	2.110	0.334	0.053	0.951
∆FDV	0.126	0.633	1.420	0.069	−0.639	0.249	0.552	0.165	0.790	0.105
∆ES	0.108	0.282	0.412	0.191	−0.027	0.163	−0.002	0.528	0.007	0.077
∆ln ein	0.786	0.000	1.536	0.004	0.438	0.044	0.130	0.100	−0.152	0.149
∆TO	0.002	0.075	0.001	0.888	0.001	0.704	0.001	0.099	0.000	0.533
∆UR	0.401	0.072	1.116	0.227	0.162	0.260	0.267	0.001	−0.465	0.018
α	−0.657	0.000	−19.763	0.000	−0.687	0.000	7.821	0.000	1.338	0.000

Note: HI = high-income countries; UM = upper-middle-income countries; LM = lower-middle-income countries; LI = low-income countries. Ln sva = natural logarithm of service value-added, ln sqsva = square log of service value-added. For all models PMG ARDL(1,0,0,0,0,0,0,0) is used, except for LM (ARDL(1,0,0,0,1,1,1,1)).

present the relationship between industrial, agriculture, and service sectors with EFP, respectively. In the long run, a U-shape relationship is established between industrial growth and EFP in HI countries, where a 1% increase in industrial GDP reduces EFP by 8.2 gha. A further increase will increase EFP by 0.6 gha. However, no significant effect was found on EFP in the agriculture sector, but a further expansion of the sector will increase EFP. In contrast, the service sector has a monotonic relationship with EFP in HI countries, where a 1% increase in service GDP increases EFP by 4 gha and a further rise in service GDP increases EFP by 0.08 gha. In UM countries, a U-shape relationship is also found between the industrial sector and EFP, where a 1% increase in industrial GDP reduces EFP by 5.7 gha, followed by an increase of 0.5 gha. Unlike HI countries, the agriculture sector has a significant effect on UM countries, where a 1% increase in agriculture reduces the EFP significantly by 13 gha. Similar to HI countries, the service sector has a monotonic relationship with EFP.

However, different patterns appear in LM countries, where a U-shape relationship has been established between the industrial sector and the service sector, but an inverse U-shape is found for the agriculture sector, suggesting support for the EKC hypothesis. In LI countries, the industrial sector has no significant effect on EFP in the long run. However, the service sector has a U-shape relationship with EFP, where a 1% increase in service GDP reduces EFP by 0.8 gha. In LI countries the agriculture sector has no significant effect on EFP, but a further rise in agriculture outputs will increase EFP.

The PMG results suggest that each economic sector has different consequences across regions, which indicates an underlying difference in technical and socio-economic factors across regions. For example, in HI and UM (with higher magnitude) and LM countries (with lower magnitude), technical advancement in the industrial sector between 1992 and 2015 could be the reason for improving environmental quality in these countries. This finding is in line with [25] for selected European countries who also found a U-shape relationship between the industrial sector and pollution. Furthermore, Ref. [9] found that the manufacturing value-added degrades the environment. They suggested that access to modern, cleaner, and more efficient technologies promotes environment-friendly behavior. However, the findings for the service sector in HI and UM countries contradict the expectation that the shift from the industrial sector to the service sector will cause environmental improvement. Refs. [23,94] discovered similar findings and suggested that the transportation sub-sector, within the service sector, is the main factor for increasing pollution in recent decades. Similarly, Ref. [90] found a positive long-run relationship between the transportation sector and environmental degradation.

For LI countries, however, the service sector improves environmental quality. This finding indicates a shift toward other service sub-sectors in LI countries, other than transportation, such as tourism, health, education, and telecommunication, that exert less pressure on land regenerative capacity. Moreover, the inverse U-shape found only in LM countries in the agriculture sector model signifies that, in these countries, growth in agriculture is associated with old technology that exerts intensive pressure on some environmental assets such as forests and pasture land. However, with further growth in agricultural GDP, these countries will have improved experience in increasing efficiency, reducing waste, and improving production technology and managerial practice.

In the long run, financial development reduces EFP in HI countries, confirming the results for the aggregate models. Interestingly, in UM countries, both the aggregate and industrial models report similar values for the FDV coefficient. Energy structure also reports a robust negative effect on EFP across regions in most sectors. Moreover, across all models and regions, energy intensity was confirmed to be a critical factor in the relationship between economic activities and environmental degradation. However, mixed results were obtained for trade openness and urbanization.

For the short-run dynamics, few coefficients are significant. However, for HI countries, the coefficient in the industrial sector raises a warning sign that industrial growth in HI countries at the initial stage since 1992 has been associated with high environmental degradation. An initial increase in industrial growth by 1% at the beginning of the period showed an increase in EFP by 111 gha. A negative relationship, however, appears with further industrial growth, suggesting an inverse U-shape pattern in the short run (supporting EKC). Nevertheless, after approximately 2.5 years, as indicated by the error correction speed, a long-run relationship is established, but with a U-shape relationship between industrial GDP and EFP. This finding suggests that industrial expansion in HI countries was associated with a high environmental cost at first. This may include building new infrastructure, intensive manufacturing facilities that reduce the area of natural resources such as forests, etc., large-scale extraction of natural resources, and intensive use of fossil fuel energy. However, after the initial stage, technical advancement could be the reason for such fast change in the long run. Nevertheless, the more logical reason for such a sharp change in our case could be due to imposed regulations and restrictions and institutional effects rather than the income effect. Accordingly, this interpretation further emphasizes that the increase in income will not automatically reduce environmental degradation but improving institutional quality is needed. Regarding the speed of adjustment, the variables took approximately two to three years to converge into long-run equilibrium, except for HI countries in the agriculture sector which took five years. The environment can thus be regarded as very sensitive to short-run shocks spanning less than three years, which may produce long-term impacts on most income groups.

Overall, the U-shape pattern seen in the aggregated model in HI countries could have emerged from industrial sector development since the agriculture sector has no significant effect, while the service sector increases the EFP. In UM countries, the observed U-shape may emerge from both industrial and agriculture sectors, since the service sector increases degradation. In LM countries, the U-shape in the aggregated model is explained by both the industrial and service sectors, while agriculture has an inversed U-shape. In LI countries, the inverted U-shape found in the aggregate model cannot be explained since the service sector has a U-shape association. The industrial and agriculture sectors showed no significant effect on EFP, suggesting that other factors may have affected such a relationship.

These results, however, do not correspond with the initial EKC hypothesis, which suggests that an increase in income after a threshold will be associated with an improvement in environmental quality. The interpretation of EKC is that society will start valuing the environment after having first satisfied basic needs; i.e., after they are both willing and able to afford better environmental quality. However, we argue that this interpretation may be appropriate for CO₂ emission since it can be directly related to reducing the quality of life. Nevertheless, a broader view of environmental degradation, such as the measure of multiple dimensions by EF and EFP, is not easily realized. Another aspect is that the intensive pressure on earth’s resources beyond its regenerative capacity may have an irreversible effect that manifests itself in the long run. Other interpretations, however, need to be developed to interpret the relationship between income and ecological footprint, or with other factors that may affect environmental degradation. For example, economic sub-sector analysis suggests that economic sectors behave differently in the ecological footprint of production. In some countries, the industrial sector is a crucial factor in reducing environmental degradation due to the associated efficiency and technology. In other countries, though, the service sector is identified as the critical sector.

5. Summary and Conclusions

Becoming a sustainable economy in the presence of climate change, biocapacity loss, food insecurity, inequality, and other environmental issues has posed challenges to countries globally. Given the complexity of the economic-ecosystem nexus, new holistic measurements and more in-depth analyses are needed to guide economic policy. One of the relatively new holistic measures of ecosystem quality is the ecological footprint, which is increasingly used instead of CO₂ emission. Unlike most studies, this study aims to provide a more in-depth analysis by investigating the impact of the aggregated and disaggregated GDP on environmental degradation measured by the ecological footprint of both production and consumption. To achieve this goal, the study analyzes panel data of 92 countries classified by income spanning the period 1992 to 2015 by using the pooled mean group estimation developed by Pesaran et al. [27]. Five models were examined under the framework of EKC. In the first and second models, the aggregated GDP was examined against consumption and the ecological footprint of production. In the third, fourth, and fifth models, the industrial, agriculture, and service sub-sectors were examined against the footprint of production. The study incorporated financial development, energy structure, energy intensity, trade openness, and urbanization as additional important factors impacting the environment.

The main findings suggest that a U-shape relationship exists in both the consumption and the production side for all income groups except for LI countries (inverse U-shape curve on the production side). Interestingly, a U-shape curve was found in HI and UM countries in the industrial sector but a monotonic relationship in the service sector. A U-shape curve was found for the industrial and service sector in LM and LI countries, but an inverse U-shape curve for agriculture outputs in LM countries. Moreover, we found that financial development promotes green investment in HI and UM countries but increases degradation in LM and LI countries. Energy structure in terms of renewable energy improves the environment in general. Energy intensity is a critical factor in the short and long run for all models. However, mixed results were obtained for trade openness and urbanization.

The findings suggest that the economic sector has different consequences on the environment across regions, indicating an underlying difference in technical and socio-economic factors. Generally, a threshold may exist in the relationship between economic sub-sectors and environmental degradation in terms of EFP for each region, depending on the nature of production practice. Further research should reveal the appropriate degree of expansion for each sector in each region to grow in harmony within the earth’s regenerative capacity.

Another suggestion this study brings is that it may be more appropriate for countries to depend on the diversification of economic activities along their growth path to allow for biocapacity to recover and sustain the natural capital in the region. Accordingly, economic policies should shift from focusing on faster growth at first and then clean later, as suggested by the EKC hypothesis, to a policy that plans for the degree of expansion of each economic sub-sector growth. This could happen by taking into account the economic sub-sectors’ behavior and their pressure on the country’s resources. Such a shift could lead to more sustainable growth.

Based on this study’s findings, future research can further investigate the EFP measurement, given that it is rarely examined in the literature. These studies can provide an insightful view of the impact that production activities may exert on a country’s resources. This study could assist in better policy planning and guide specialization choices while accounting for sustainable growth. Future research can also investigate the nature of the threshold between economic sub-sectors and environmental degradation, which could lead to an exciting interpretation of the relationship between economic growth and environmental degradation in parallel with the EKC hypothesis. Future research can also be conducted at a more disaggregated level, such as at the country level and the EFP’s components analysis, to provide more specific policy recommendations.