What matters for environmental quality in the Next Eleven Countries: economic growth or income inequality?

Abstract

This study uses panel data for the period 1971–2013 to explore the implications of growth, wealth disparities, and per capita energy consumption on carbon emissions in a sample of Next Eleven (N-11) countries. It uses the first-generation (Pedroni and Kao) and second-generation (Westerlund) cointegration techniques to highlight a long-run interplay between the selected variables in carbon emission functions for all the N-11 countries. It also analyzes the long-run interactions among the series. Contrastingly, it also shows that economic growth, income inequalities, and per capita energy consumption accelerate CO₂ emissions. Besides examining the effects of wealth disparity square, the study also uses the environmental Kuznets curve hypothesis in the context of the N-11 countries and discusses the policy implications of its findings.

Introduction

As emerging economies grow at the expense of massive energy consumption (EC), they face several difficult challenges in various areas. Among these, environmental quality is one of the biggest concerns as it impacts climate change and poverty levels through a range of effects on agriculture productivity and people’s health (Hallegatte and Rozenberg 2017). Countries require greater volumes of energy to sustain their economic growth and development. Given the inelastic nature of the supply of natural resources such as land, drinking water, and clean air, as human activities increase with increasing populations, the accumulated goods and services and externalities produced in the form of pollution and greenhouse gas emissions (GHGs) grow disproportionately in relation to the planet’s capacity. An increase in economic activities is reflected in higher growth rates and gross domestic product (GDP) per capita which leads to additional environmental concerns as emission levels increase along with increasing demands for energy. The environmental repercussions of these activities are not only a matter of concern for domestic economies but also for the world at large, as all the countries are interconnected due to globalization; hence, all countries are likely to be affected by GHGs and face the risk of climate change although with differing impacts in different regions.¹

Going by theoretical discourses in economic theory on poverty alleviation, it is apparent that the developing world needs significant government efforts along with sustainable economic progress for poverty reduction policies to be efficient and successful. In practice, most of the industrialized and developing countries are expanding their economic activities and output levels. At the same time, they are also aspiring to reduce their carbon emissions. As production and income differences are profoundly intertwined, these in turn will lead to large wealth disparities both in the short and long run. Thus, the developing world is faced with various challenges. Climate change (for example, rising sea levels, storms, droughts, and floods) is one of the most severe of these; it is driven by substantial CO₂ emissions that lead to global warming. Environmental degradation because of rising carbon emissions and climate change is also a pressing problem as it threatens sustainable economic progress in the long-term and also the quality of living standards. It is generally recognized that climate change is a vital issue that needs to be addressed in energy and ecological economics.

Recent available data from the UN climate summit in Poland shows that CO₂ emissions were estimated to rise by 2.7% in 2018. Specialists noted that global carbon emissions in 2018 were expected to reach an all-time high level of 37.1 billion tonnes (Le Quéré et al. 2018). A mix of fossil fuels, deforestation, and other factors contributed substantially to these rising levels. Hence, rising per capita CO₂ emissions are commonly used as proxy for environmental pollutants often linked with higher per capita incomes. As suggested by various scholars (inter alia, Holtz-Eakin and Selden 1995; Kijima et al. 2010; Raza et al. 2015), carbon emissions are a main source of global warming and climate change. Changes in climatic conditions have risen to alarming levels raising governments' attention across the world. The governments all across the globe should work to protect environmental health by implementing a number of policy tools such as taxes and reliance on renewable energy. Some guidelines proposed by policymakers have generated extensive intergovernmental debates, particularly in developing nations. The Kyoto Protocol, which was adopted in Kyoto, Japan on 11 December 1997, is a relevant example of the efforts being made to reduce the carbon footprint that leads to global warming. The protocol entered into force on 16 February 2005 and represents a binding agreement to the UN Framework Convention on Climate Change (UNFCCC).

Given the fact that global warming is on the rise and there are increasing concerns about scarce energy sources, we believe that exploring the interplay between carbon emissions, progress, and wealth disparities in the context of N-11 countries² is a scientific endeavor worth undertaking by both scholars and policymakers. It is important to empirically validate the causality, if any, between economic development and income differences on the one hand and environmental degradation on the other in the N-11 countries. This is essential because understanding the direction of causality will provide valuable insights into the best ways of preserving environmental health; such an exercise will also offer examples of best practices for other developing economies to follow. If environmental degradation persists in N-11 countries, amid increased production levels and the associated massive energy consumption, it will have a domino effect throughout the globe. Hence, N-11 and other developing economies seeking to mitigate climate change will need to strengthen collaborations to address the implications of higher progress and higher energy consumption domestically.

Our methodological approach examines the interaction between the different economic series. It is based on an innovative model of the panel cointegration test developed by Pedroni (2004), the panel cointegration test developed by Kao (1999), the fully modified OLS (FMOLS) proposed by Phillips and Hansen (1990), and the dynamic OLS (DOLS) created by Stock and Watson (1993). Existing literature does not apply the panel cointegration, FMOLS, or DOLS methodologies in the context of environmental quality in relation to economic growth and income inequalities in the Next-11 countries. This is a major limitation of these studies because scholars only elude to empirical investigations of the relationships between carbon emissions, growth, and wealth disparities which can lead to inaccurate results on the environmental Kuznets curve’s (EKC) assumptions and misguide policymakers whose aim is to protect environmental health worldwide.

The economic rationale behind using these models is that the interplay between selected variables fluctuates as a result of changing economic parameters, natural calamities, energy and environmental strategies, and regulatory and technological innovations.

Our study uses annual data from 1971 to 2013 on a per capita basis for wealth disparities, growth, and carbon emissions in the context of N-11 countries because based on their contributions to global GDP, share of energy demand, and CO₂ emissions to world energy demand and their carbon footprint, they have the potential of becoming some of the largest economies in the twenty-first century. The N-11 states are increasingly recognized as major influencers in the global open economy and environmental policies next to Brazil, Russia, India, China, and South Africa (BRICS) economies but dissimilar to these in terms of their economic growth patterns accompanied by a greater degree of trade and financial openness. N-11 countries could surpass their rivals and become major market participants despite being exposed to a larger number of challenges relative to BRICS nations as a result of their strong economic reforms targeted at sustainable economic growth in the long term.

N-11 countries are enjoying rapid growth and are participating in global trade and investment projects (except for Iran which is a closed economy impacted the USA’s imposed sanctions). They are faced with rising energy demands triggered by investment and industrialization activities that use less energy-efficient technologies to boost economic progress; this is a major reason for environmental degradation. To limit their carbon footprint, Mexico and Nigeria introduced incentives for businesses to enhance national production via more efficient energy technologies. The dynamics of GDP per capita in the N-11 countries in relation to global trends between 1980 and 2013 shows that the average GDP per capita for the N-11 group was lower than the worldwide average (Esfahani and Rasoulinezhad 2015); however, during this period, the growth rate of GDP per capita in the N-11 nations was higher at 4.9% than the aggregate world growth rate (at around 3.2%).

Looking beyond the N-11 individual countries, it is important for academics and business professionals to understand that the N-11 group had a growth rate of nearly 4.5% during the last decade compared to almost 4.0% over the previous decade (data as of 2018) (O’Neill 2018). Considering the volume of its output, it is obvious that the N-11 group makes a significant contribution to the world economy. Nevertheless, despite substantial potential for growth, there are still some issues that can prevent N-11 countries from enjoying BRICS nations’ development trends. One factor is represented by the swings in global commodity prices, which effect N-11 producers. Another is political events that may also distress growth prospects. Although N-11 countries cannot rival the BRICS nations in scale, estimates reveal that by 2050, N-11 countries’ GDP could reach two-third the size of the Group of Seven (G7) countries, meaning that N-11 nations could have a major impact on the global political, economic, energy, and environmental landscape (Sachs et al. 2007; ALOnaizi and Gadhoum 2017). Given stable parameters for progress, the N-11 countries have the ability to grow at a rate of 4.0% over the next two decades; in addition, incremental demand for these countries could exceed that of G7 and be two times that of the G7 countries by 2050 (ALOnaizi and Gadhoum 2017). Hence, the N-11 group’s contribution to global GDP will increase faster.

In 1980, the N-11 group was responsible for over 6.3% of the global carbon emissions because of their petroleum consumption (Esfahani and Rasoulinezhad 2015). By 2013, this had increased to 12.0%, driven by higher population density, heavy reliance of the domestic economies on the manufacturing sector, and substantial share of fossil fuels in the electricity generation mix. The upward trend in economic progress increased the per capita energy consumption’s share to 11.0% of global consumption which further aggravated environmental degradation (Yıldırım et al. 2014). CO₂ emissions measured in kilogram per capita and income inequality measured as the Gini coefficient for the N-11 nations from 1971 to 2013 is presented in Figs. 1 and 2, respectively. The pattern differs among the nations more in the case of income inequalities than in the case of CO₂ emissions (Figs. 3 and 4 of the Appendix).

Fig. 1

Theoretical analysis of the income inequality-CO₂ emission nexus. Source: authorial adaptation after Liu et al. (2018)

Fig. 2

Dumitrescu-Hurlin (D-H) Granger causality test

The rest of this study is organized as follows. Section 2 reviews major scholarly work in this field. Section 3 details the methodology and the data used for our analysis. Section 4 gives the results of our study and discusses their significance. Section 5 provides concluding remarks and policy implications and indicates future research directions.

Review of related studies

Kuznets’ (1955) landmark study linking the inverted U-shaped interplay between wealth disparities and progress³ prompted many researchers to empirically investigate the role of growth in income inequality leading to many cross and individual country studies.

Dollar and Kraay (2002) argue that growth is good for the poor as there is evidence of a trickledown effect of the production process that not only creates employment opportunities and increases agriculture productivity but also reduces income inequalities by improving income distribution among the poorest. However, Kashwan’s (2017) findings contradict this. Kashwan is of the view that development does not favor the poor because it does not benefit the entire population (haves and have-nots) equally. As a consequence, preference for environmental quality declines over time. In contrast, Sachs (2015) postulates that rising concerns about the impact of economic growth on environmental quality are driven by higher income inequalities. This implies that economies in the globalized world may be good at achieving higher progress, but they fail to maintain an equitable distribution of income with sustainable environmental quality. Therefore, some scholars (for example, Boyce 1994; Magnani 2000; Jorgenson et al. 2017) stress that challenges of environmental quality are a result of social issues mainly generated by wealth disparities and power inequalities. Hence, it is essential to explore major academic achievements on the effects of income inequalities, energy consumption, progress, and urbanization on CO₂ emissions. Table 1 provides a summary of the findings of previous studies on the link between income inequalities and economic growth.

Table 1 Summary of findings from the previous studies on the link between income inequality and economic growth

CO₂ emissions and the income inequality nexus

The interplay between income inequality and climate change has received a great deal of attention from the international community in recent years (Jorgenson et al. 2017). Practical evidence indicates that growth fails to ensure environmental sustainability and instead diminishes it. This is a topic extensively debated by the academia (Wolde-Rufael and Idowu 2017; Kashwan 2017). The nexus between wealth inequality and environmental degradation has been examined by a number of theoretical and empirical analyses (Torras and Boyce 1998; Laurent 2015). As of date, theoretical studies on this relationship are grounded in three distinct approaches (Fig. 1).

The first approach provides a political explanation for the impact of income inequalities on environmental degradation. It relies on the distribution of wealth and power between those who have what is needed to better shield themselves from pollution activities and those who cannot exert any influence on such processes. The rich cohorts are also more predisposed towards higher levels of degradation as they possess resources that lead to greater energy consumption and environmental damage (Torras and Boyce 1998). This group is also more reluctant to engage in environmental protection, emphasizing the attached significant costs of such measures and their own capacity to avoid environmental hazards, thus shifting this responsibility to the poor (Scruggs 1998). Hence, the prevalence of one group or the other influences the degree of observed environmental damage.

The second approach gravitates around the concept of marginal propensity to emit (MPE), which states that pollution levels swing with changes in income distribution (Ravallion et al. 2000; Borghesi 2006). Ravallion et al. (2000) were the first to explain how low emissions are triggered by severe domestic inequalities. This was a very controversial finding and was heavily disputed by many academics, who argued that MPE actually diminished with income (Heil and Selden 2001; Schmalensee et al. 2006). These scholars also suggested that MPE’s dominant component was represented by consumption needs. This assumption ignores the Keynesian impact that points to lower income groups having a higher MPE compared to higher income groups. An increase in income inequality makes the poor damage the environment to meet their living standards (Boyce 1994). Deforestation by the poor in the emerging world is a relevant example as this has reduced the forest area and subsequently its ability to absorb CO₂. Jorgenson et al. (2017) advocated that an increase in incomes for the poor will most likely reduce environmental degradation.

The third theoretical perspective is rooted in advancing individual economic behavior (Liu et al. 2018). This stream seeks to prove that greater inequalities generate augmented energy consumption levels, which enhance CO₂ emissions (Schor 1999). Some authors (for example, Schor 1998; Veblen 2009) explain the presence of a Veblen effect according to which the rich engage in expensive consumption to satisfy their living standards. According to Bowles and Park (2005), income inequalities lead to working long hours, resulting in significant energy consumption and degradation of the environment (Fitzgerald et al. 2015).

In addition to these theoretical analyses, more and more empirical studies use the Gini coefficient to examine the interplay between income disparities and CO₂ emissions. However, their conclusions are conflicting (Liu et al. 2018). Ambiguities in Boyce (1994) arguments and other authors’ findings (for example, Ravallion et al. 2000; Borghesi 2006; Grunewald et al. 2017) led to a dialogue on the empirical validity of a significant relationship between inequality and environmental damage. Table 2 presents summary findings of previous studies worldwide on the link between CO₂ emissions and income inequalities.

Table 2 Summary of findings from the previous studies worldwide on the link between CO₂ emissions and the income inequality

The interplay between CO₂ emissions and energy consumption

Global energy consumption and environmental degradation have been rising constantly, and understanding the mechanisms and drivers of these emissions is important for guiding the process of policymaking and forecasting to mitigate their effects. In parallel, energy demand, similar to that of most other goods, shows significant fluctuations across income categories. For instance, emerging nations are expected to register a rapid increase in energy consumption as households enjoy middle-income levels (Wolfram et al. 2012). However, empirical arguments from high- and middle-income states seem to point the other way, as the rich spend a smaller income share on energy.

As of date, there are no clear images of the aggregate consequences of per capita energy consumption and the aggregate interplay between income, consumption trends, and environmental degradation (Caron and Fally 2018). This nexus directly influences the GDP’s CO₂ intensity which is an essential factor in estimating emissions and also understanding the magnitude of climate change. With the quality of the environment constantly deteriorating, understanding the determinants of CO₂ emissions and energy use is as relevant as ever. According to Caron and Fally (2018), spending on energy and energy-intensive goods reduces with income across a large sample of states. Their projection models point out that income increase changes the patterns of consumption in a manner that generally reduces environmental damage. However, increasing emissions in emerging nations, coupled with a shift from direct to indirect energy consumption, means that their impact on global emissions is only modest. EKC-based analyses have been attacked for lacking structure or causal interpretations (Levinson and O’Brien 2019; Caron and Fally (2018) and for identifying the impact that income has environmental degradation via changes in consumption. Table 3 gives a summary of the findings of previous studies on the link between CO₂ emissions and energy consumption.

Table 3 Summary of findings from previous studies worldwide on the link between CO₂ emissions and energy consumption

The nexus between CO₂ emissions and economic growth

Nnaji et al. (2013) reported a positive impact of fossil fuel consumption and development on CO₂ emissions in Nigeria. Wang (2013) found a reducing effect of differentiated output growth on CO₂ emissions in the USA and China. Salahuddin and Gow (2014) claim that progress had no long-term implications for environmental degradation in the Gulf Cooperation Council countries. Kivyiro and Arminen (2014) emphasize the long-run economic growth-CO₂ emissions nexus for six sub-Saharan African states, while Lau et al. (2014) found that CO₂ emissions stimulated economic growth in Malaysia and Allali et al. (2015) showed that development had a positive impact on CO₂ emissions in Algeria. Similarly, Abid (2015) observed both a short-run and a long-run interplay between growth and the carbon footprint in Tunisia, in addition to unidirectional causality running from progress to CO₂ emissions. Moreover, based on data covering the period 1960–2005 and a bounds test for the cointegration procedure, Halicioglu (2009) found two forms of long-run interplay between carbon emissions, energy consumption, income, and foreign trade in the context of Turkey. The first one indicates that CO₂ emissions are driven by energy consumption, income, and foreign trade and the second one highlights that income is driven by CO₂ emissions, energy consumption, and foreign trade. Esso and Keho (2016) also identified a positive long-term relationship among CO₂ emissions and progress and a bidirectional causal link between the variables in the Nigerian economy. Table 4 provides a summary of the findings of previous studies on the link between CO₂ emissions and economic growth.

Table 4 Summary of findings from previous studies worldwide on the link between CO₂ emissions and economic growth

Model building and data description

Model

Given our research objective and the context of theoretical and empirical literature discussed earlier, we specified the basic CO₂ emission function as noted below to understand the nature of the interaction and the effects of the key variables on the CO₂ function (CO₂ emissions). We estimated different versions of our basic models to avoid estimation problems as the addition of a larger number of variables in the same model could result in over parameterization, that is, consumption of degrees of freedom on the one hand and the inclusion of related variables that would trigger multicollinearity issues on the other:

$$ \ln {\mathrm{co}}_{2t}={\alpha}_0+{\beta}_1\ln {\mathrm{energy}}_t+{\beta}_2\ln {\mathrm{gdp}}_t+{\beta}_3\ln {\mathrm{gini}}_t+{\mu}_t $$

(1)

By adding GDP square in Eq. 2 and Gini square in Eq. 3, the EKC functional form can be represented as:

$$ \ln {\mathrm{co}}_{2t}={\alpha}_0+{\beta}_1\ln {\mathrm{energy}}_t+{\beta}_2\ln {\mathrm{gdp}}_t+{\beta}_3\ln {\mathrm{gini}}_t+{\beta}_4\ln {\mathrm{gdp}}_t^2+{\mu}_t $$

(2)

$$ \ln {\mathrm{co}}_{2t}={\alpha}_0+{\beta}_1\ln {\mathrm{energy}}_t+{\beta}_2\ln {\mathrm{gdp}}_t+{\beta}_3\ln {\mathrm{gini}}_t+{\beta}_4\ln {\mathrm{gini}}_t^2+{\mu}_t $$

(3)

where lnco₂ emissions represent CO₂ emissions per capita as a proxy of environmental quality, lnenergy denotes per capita energy consumption, lngdp is gross domestic product per capita as a proxy of economic development, lngdp² is the gross domestic product squared, lngini is the net Gini coefficient proxy of income inequality, lngini² is the Gini coefficient squared, which is used to understand whether there is an inverted U-shaped relationship or not, and μ_t is the error term. In addition, for the smoothness of the variables, we used the natural logarithmic for all variables like lnco₂, lnenergy, lngdp, lngdp², and lngini². α₀ is the fixed effect and β₁, β₂, β₃, and β₄ are slope coefficients. β₁ represents the elasticity of carbon emissions with respect to per capita energy consumption, that is, for every 1% increase in per capita energy consumption, the per capita carbon emissions increase by β₁ percent. Similarly, β₂ and β₃ in Eqs. 2 and 3 measure the elasticity of carbon emissions with respect to income and the elasticity of carbon emissions with respect to income inequality, respectively. The elasticity of emissions with respect to income in Eq. 2 can be formulated as ∂ ln co2/∂lngdp = β₂ + 2β₄ ln gdp which can be verified by the existence of the EKC hypothesis. The elasticity of emissions with respect to income inequality can be formulated as ∂ ln co2/∂ ln gini = β₃ + 2β₄ ln gini. The turning point⁴ of the EKC occurs at a level of income inequality in Eq. 3.

According to the EKC hypothesis, the long-term interplay between per capita energy consumption, wealth disparities, and economic growth on CO₂ emissions can be captured by Eqs. 2 and 3. The EKC hypothesis describes an inverted U-shaped link among environmental degradation and economic growth. We are interested in identifying an inverted U-shaped connection between environmental degradation and income inequalities which can be obtained mathematically by embedding the squared value of the Gini coefficient in the regressors.

Methodological approach

We seek to explore the causal interactions among carbon CO₂ emissions, per capita energy consumption, income inequalities, and economic development using modern econometric techniques. This analysis involves a three-step scientific approach. First, we determine the integration order of the series based on panel unit root tests. Second, we apply panel cointegration tests to verify the existence of any long-term relationships. Finally, we examine the size and direction of any potential causal interactions among our series.

Panel unit root tests/stationarity tests

We apply standard unit root tests to the determinants of CO₂ emissions, total per capita energy consumption, income inequalities, and per capita GDP. Narayan and Smyth (2009) argue that the Augmented Dickey-Fuller (ADF) test has low power in rejecting the null hypothesis of stationarity, particularly for short periods. Hence, recent academic works claim that panel stationarity tests are more powerful compared to individual time series ones (for example, Al-Iriani 2006). We use the panel unit root tests suggested by Levin et al. (2002), Im, Pesaran, and Shin (IPS), Im et al. (2003), Hadri (2002), and Beitung (2001). These are generally more robust than the first generation of panel tests (Narayan and Smyth 2009).

Moreover, the first-generation panel unit root tests are applied to panel data which neglects both structural breaks and cross-sectional dependence. These are commonly used in the carbon emission-per capita energy consumption literature. These are similar to the ADF-based IPS tests which assume a heterogeneous unit root (Im et al. 2003). In contrast, Breitung and Das (2005) and Levin et al. (2002) point out a homogenous unit autoregressive root. Levin et al. (2002) and IPS test the null hypothesis of time series integration. Hadri (2002) suggests a residual-based Lagrange multiplier test for the null of level or trend stationarity that includes heterogeneous disturbance terms.

Panel cointegration tests and long-run estimations

Using the panel cointegration technique is more powerful than time series models because models estimated from cross-sections of time series have more degrees of freedom and are more efficient than models estimated from individual time series. The panel cointegration techniques are particularly useful when the time series dimension of each cross-section is short.

If all the variables are found to be stationary after taking the first difference, then we proceed to test Pedroni (1999, 2004) and Kao’s (1999) panel cointegration tests. To test the existence of cointegration within a heterogeneous panel, Pedroni (1999, 2004) proposed two categories of cointegration tests and seven statistics. The first category is based on four statistics (panel statistics)—the v statistic, rho statistic, the Phillips and Perron (PP) statistic, and the ADF statistic. These statistics are classified in the within-dimension and take into account common autoregressive coefficients across the countries. The second category is based on three statistics (group statistics)—the rho statistic, the PP statistic, and the ADF statistic. These tests are classified in the between-dimension and are based on individual autoregressive coefficients for each country in the panel. The null hypothesis is that there is no cointegration and the alternative hypothesis is that there is cointegration between the variables. Pedroni (1999, 2004) and Kao’s (1999) panel cointegration tests are based on the residuals. Deviations from the long-run equilibrium relationship are represented by the estimated residuals. The null hypothesis, of no cointegration, ρ_i = 1, is tested via the following unit root test of the residuals:

$$ {\varepsilon}_{it}={\rho}_i{\varepsilon}_{it-1}+{\omega}_{it} $$

(4)

where ε_it is defined as the autoregressive process.

Panel cointegration test allowing for cross-sectional dependence

After having confirmed the non-stationarity of the variables for the panel data as a whole, it is natural to test the existence of a structural long-run relationship among the variables. To assess the robustness of our findings, we also implement the bootstrap panel cointegration test proposed by Westerlund (2007)⁵:

$$ \varDelta {y}_{it}={\phi}_i^{\hbox{'}}\;{d}_t+{\alpha}_i{y}_{i,t-1}+{\lambda}_i^{\hbox{'}}{x}_{i,t-1}+\sum \limits_{j=0}^{\rho_i}{\alpha}_{ij}\varDelta {y}_{i,t-j}+\sum \limits_{j=0}^{\rho_i}{\gamma}_{ij}\varDelta {x}_{i,t-j}+{e}_{it}\kern0.36em $$

(5)

where ϕ_i = (ϕ_0i, ϕ_1i )^',  d_t = (1, t)^', λ_i =  − α_iβ_i,. To derive the two-panel statistics, first the lag order ρ_i for each cross-section should be determined and the common error correction parameter α and its standard error σ_α are estimated. Hence, the estimates of α and σ_α can be estimated as:

$$ \hat{\alpha}={\left(\sum \limits_{i=1}^N\sum \limits_{t=2}^T{\tilde{y}}_{i,t-1}\right)}^{-1}\;\sum \limits_{i=1}^N\sum \limits_{t=2}^T\frac{1}{{\hat{\alpha}}_i(1)}{\tilde{y}}_{i,t-1}\varDelta {\tilde{y}}_{i,t}\kern0.6em $$

(6)

$$ {\hat{\alpha}}_{\hat{\alpha}}={\left[\;{\left(\frac{1}{N}\sum \limits_{i=1}^N{\left(\frac{{\hat{\alpha}}_i}{{\hat{\alpha}}_i(1)}\right)}^2\right)}^{-1}\;\sum \limits_{i=1}^N\sum \limits_{t=2}^T{{\tilde{y}}^2}_{i,t-1}\kern0.36em \right]}^{-1/2}\kern0.6em $$

(7)

The term ${\hat{\alpha}}_i$ is the estimated standard error obtained by applying OLS to Eq. 5. Thus, the panel statistics are:

$$ {P}_{\tau }=\frac{\hat{\alpha}}{{\hat{\alpha}}_{\hat{\alpha}}}\kern1.56em $$

(8)

$$ {P}_{\alpha }=T\hat{\alpha}\kern0.36em $$

(9)

Further, Westerlund (2007) formulated the group mean test statistics as:

$$ {G}_{\tau }=\frac{1}{N}\sum \limits_{i=1}^N\frac{{\hat{\alpha}}_i}{{\hat{\alpha}}_{\hat{\alpha}}}\kern0.48em $$

(10)

$$ {G}_{\alpha }=\frac{1}{N}\sum \limits_{i=1}^N\frac{T{\hat{\alpha}}_i}{{\hat{\alpha}}_i(1)}\kern0.6em $$

(11)

where ${\hat{\alpha}}_{{\hat{\alpha}}_i}$ is the standard error of ${\hat{\alpha}}_i$. Under the null hypothesis, all P_τ, P_α, G_τ, and G_α test statistics have limiting normal distribution as T and N go to infinity sequentially. In other words, the statistics are standard and normally distributed when standardized with appropriate moments. Hence, the asymptotic distributions and the moments are dependent on the deterministic terms and the number of regressors included in the regression model. Under the alternative hypothesis, the group mean statistics G_τ and G_α and the panel statistics P_τ and P_α diverge to negative infinity, which means that the test decision is made on the left tail of the standard normal distribution. Additionally, Westerlund (2007) proposed using Chang’s (2004) bootstrap approach for panel cointegration testing which accommodates for cross-sectional dependence.

The Dumitrescu-Hurlin panel Granger causality test

It is important for policymakers to know the directions of causality among the variables so that they can formulate appropriate policies. Therefore, this study employs the Granger causality test developed by Dumitrescu and Hurlin (2012) to demonstrate the causality relationship. This test has very flexible use both in case of T > N and T < N as well as for heterogeneous panels. One of the major advantages of this test is that it can be used for panel data in the presence of cross-sectional dependence. Moreover, this test solves the problems posed by the homogeneity assumption in the standard Granger causality test (Engle and Granger 1987). This test considers the following linear heterogeneous model:

$$ {y}_{it}={\alpha}_i+\sum \limits_{i=1}^k{\gamma}_i^{(k)}{y}_{i,t-k}+\sum \limits_{i=1}^k{\beta}_i^{(k)}{x}_{i,t-j}+{\varepsilon}_{it}\kern0.48em $$

(12)

where k ∈ N⁺ and k ∈ N^∗, β_i = (β_i⁽¹⁾, .........., β_i^(k)). α_i , ${\gamma}_i^{(k)}$, and ${\beta}_i^{(k)}$ indicate a constant term, lag parameters, and slope coefficients, respectively.

Data description

We worked with annual data covering the period 1971–2013 for N-11 countries (Bangladesh, Egypt, Indonesia, Iran, Mexico, Nigeria, Pakistan, the Philippines, Turkey, South Korea, and Vietnam). Our study includes CO₂ emissions (CO₂) in kilogram per capita, per capita energy consumption in kilogram of oil equivalent, and real GDP per capita in constant 2010 US dollars used as a proxy of economic growth borrowed from the World Development Indicators (WDI 2014). The Gini coefficient which measures income distribution is obtained from SWIID (2015).

Table 5 gives the mean values and standard deviations of the data series for N-11 countries.⁶ The descriptive statistics show that the data series is fairly dispersed, but the standard deviations are homogeneous. Table 5 also shows that South Korea had the highest means of CO₂ emissions (6677.444) and per capita energy consumption (2670.633), Turkey had the highest GDP mean (6953.612), and Mexico registered the highest mean of the Gini coefficient (47.381). The lowest means of CO₂ emissions (184.668) and per capita energy consumption (131.617) were in Bangladesh and the lowest Gini coefficient (31.088) and GDP were (757.444) in Pakistan. South Korea had the highest volatility in the case of CO₂ emissions (3273.488), energy use (1617.452), and GDP per capita (7000.265). The highest volatility of the Gini coefficient was in Nigeria (3.788).

Table 5 Descriptive statistics and pairwise correlation of the variables used (before taking logarithm) 1971–2013

Results and discussion

Cross-sectional dependence test

This study applied the cross-sectional dependence (CD) test developed by Pesaran (2004) to verify the cross-section dependence across the 11 countries under investigation. Table 6 gives the results of the test applied to CO₂ emissions, per capita energy consumption, income inequalities, and GDP per capita variables. We reject the null hypothesis of cross-sectional independence for all the variables, except for income inequality.

Table 6 Cross-sectional dependence test

Unit root tests’ results

We used the unit root tests developed by LLC, IPS, and Breitung and Das (2005) to verify the existence of unit roots. Levin et al. (2002) proposed the panel-based ADF test and assumed homogeneity in the dynamics of the autoregressive coefficients for all panel units. Under the LLC (2002) unit root test, the null hypothesis indicates the presence of a unit root; the alternative hypothesis claims there is no unit root. The IPS (2003) test which solves the serial correlation problem by accepting heterogeneity between units in a dynamic panel framework shows that under the null hypothesis of non-stationarity the statistics follow the standard normal distribution asymptotically. Table 7 gives the panel unit root tests’ results.

Table 7 Panel unit root tests

Panel cointegration results

After applying panel unit root tests, we proceeded to panel cointegration tests. Cointegration can be described as a systematic long-term interplay among two or more economic variables (Yoo and Goldsmith 2006). Based on the Pedroni test’s findings, Table 8 gives the cointegration between selected variables at the 5% level of significance for both the frameworks. Therefore, we have long-run interactions among the series. The Kao (1999) residual cointegration results also support cointegration at the 5% level of significance. Thus, there is strong evidence of a long-term relationship between the series. This is consistent with Lee (2005) and Sadorsky (2009). We used the Westeland test cointegration (2007) to establish a long-run relationship among the series. This also confirms cointegration among selected variables.

Table 8 Panel cointegration test results

Long-run elasticity

We first present the results of the empirical interplay between per capita energy consumption, economic growth, income inequalities, and CO₂ emissions. Pedroni (2004) recommended a superior test relative to single equation methods that can be used for exploring the cointegration vector. We seek to identify a strong interaction between total per capita energy consumption, economic development, income inequalities, and CO₂ emissions for all panel countries. Equations 2 and 3 give the regression between these four factors. The dependent variables are carbon emissions, a function of total per capita energy consumption, GDP per capita, and income inequalities.

Table 9 presents the findings of the FMOLS, DOLS, and random effects estimation methods which highlight whether per capita energy consumption, wealth disparities, and level of per capita income stimulate CO₂ emissions in the N-11 countries.⁷ The Hausman test confirms the random effects model as a preferred model over the fixed effects model as it helps avoid the cross-sectional effects and also allows us to cross-examine the long-run outcomes returned by FMOLS and DOLS. Both FMOLS and DOLS tests show consistent results that demonstrate a positive interplay running from per capita energy consumption, income inequalities, and level of income to CO₂ emissions at the 1% level of significance.

Table 9 Panel long-run results

The long-run elasticity of CO₂ emissions with respect to total per capita energy consumption is estimated at 0.589, 0.626, and 0.584 in model 1 (without EKC) and 0.594, 0.586, and 0.57 in model 2 (with EKC). This means that a 1% increase in per capita energy consumption leads to a 0.58% to 0.59% increase in CO₂ emissions. These findings indicate a monotonic relationship between CO₂ emissions and per capita energy consumption for all N-11 countries. There is no doubt that CO₂ emissions are driven by unsustainable per capita energy consumption patterns.

Our results confirm the presence of an inverted U-shaped relationship between income inequalities and carbon emissions for N-11 countries. Initially, CO₂ emissions increase and they start declining after a threshold level of income inequality has been reached. We identified an environmental EKC interplay among income inequalities and CO₂ emissions. The threshold Gini coefficient level of wealth disparities for all N-11 nations varied from 41 to 45. If a country has a level of inequality below 40, reducing the gap will trigger a decrease in carbon emissions. If inequality is higher than 45, reducing inequality will have no beneficial effects on environmental quality.

Regarding the long-run elasticity of CO₂ emissions with respect to wealth disparity, we found inelasticity coefficients in model 1 (without EKC) but elasticity in model 2 (with EKC) when including the Gini coefficient squared. The coefficient of income inequality is 0.862, 0.437 (insignificant), and 0.78 in model 1. A 1% increase in income inequality drives a 0.78% to 0.86% growth in carbon emissions. Our results illustrate that rising income inequalities force poor people to use low-cost fuels which release more carbon emissions. By contrast, falling income inequalities improve environmental quality. The coefficients for wealth disparity and income inequality square in model 2 are 21.267, 16.94, and 20.00 and − 2.798, − 2.276, and − 2.636, respectively. Further, the estimated threshold levels are 44.66, 41.27, and 44.41.

Our study found an inverted “U”-shaped relation between the level of income inequality and CO₂ emissions. This result shows that wealth disparities decreased CO₂ emissions when the level of income inequalities was more than 45% and increased CO₂ emissions when income inequalities were less than 40%. In countries with more equal distribution of income, an increase in inequalities has a bad effect on the environment. Similarly, countries with more unequal income distribution policies for lowering inequalities will lead to environment degradation. Hence, in both the cases, the level of wealth distribution plays a critical role in environment quality. This indicates that the rich people are more responsible for the degradation of environmental quality as compared to poor people because when the rich become richer, they try to invest their wealth in leading industries to get higher returns without caring about their impact on environmental quality. But in the case of poor people where environment forms a part of their livelihood strategy, they try and protect their environment.

The estimated long-run elasticity of CO₂ emissions with respect to GDP per capita is 0.456, 0.401, and 0.453 for model 1 (without EKC) and 0.449, 0.381, and 0.453 for model 2. This implies that a 1% increase in per capita GDP induces CO₂ emissions of 0.38% to 0.44%. We found a positive long-run interaction among CO₂ emissions and GDP per capita. This shows that an increase in disposable incomes motivates households to consume pollution-free fuels (renewable energy) for their activities.

Finally, the results of the long-run panel Granger causality test are reported in Table 10. There is evidence of bidirectional Granger causality between lngdp and lngdp² with lnco₂. However, lnco₂, lnenergy, lngini, and lngini² provide a unidirectional causality with lnco₂. Figure 2 gives a graphical illustration of lead-lag causality linkages summarized from the Dumitrescu-Hurlin (D-H) Granger causality test.

Table 10 Dumitrescu-Hurlin Granger causality test

Concluding remarks, policy implications, and future research directions

This research explained the nexus between CO₂ emissions, per capita energy consumption, income inequalities, and GDP per capita for the N-11 countries over the period 1971–2013. It investigated whether income levels or income inequalities matter in CO₂ emissions. We considered three models: one linearly and two non-linearly specified to verify the environmental Kuznets curve hypothesis for the N-11 countries. Our results show that in the long-run, an increase in income inequalities, economic development, and per capita energy consumption increases CO₂ emissions. This is happening in the N-11 states because for the sake of economic growth and poverty reduction, their governments are racing to the bottom leading to weakened environmental regulation policies that follow growth without much attention to the health of the environment.

Our findings also highlight that wealth disparities had both positive and negative effects in the countries that we studied. In countries with equal income distribution, an increase in inequalities had a deteriorating effect on the environment. Similarly, in countries with more unequal income distribution, policies for lowering inequalities and poverty could lead to environment degradation. So, in both cases, the level of wealth distribution played a critical role in environment quality. This indicates that the rich countries are more responsible for deterioration in environmental quality as compared to poor countries because when the rich become richer, they invest their wealth in leading industries to get higher returns without caring for their impact on environmental quality. But in the case of poor countries where the environment forms a major livelihood source and strategy, people try and protect their environment. In most of the sampled N-11 countries, the GDP per capita was below the turning point in development where countries had started investing in the environment.

Therefore, policymakers and governments in N-11 nations can increase environmental protection by applying carbon taxes, emission trading schemes, and various subsidy and technology-based incentive programs. Moreover, governments can promote the development and transfer of clean technology and also encourage both national and foreign investors to employ energy-efficient technologies when increasing production levels. In addition, in an effort to gain a double dividend, N-11 governments should adopt energy use reduction policies not only to preserve their environmental quality and for reducing energy costs but also to distribute the already scarce resources to sectors with high growth potential that are using resource-saving technologies. A shift in taxes from labor to capital under a revenue neutral tax policy will incentivize clean technology development and green jobs.

Future analyses are warranted in terms of applying the quantile-on-quantile regression (Q-Q-R) approach developed by Sim and Zhou (2015), the non-linear autoregressive distributed lag framework detailed by Pesaran et al. (2001), who postulate the possibility of an empirical examination of non-linear cointegration, asymmetric dynamic relationships, use of flexible adjustment smooth transition autoregressive (STAR) models (Chen 2003), and flexible policy-based adjustments to achieve targets (Khayyat and Heshmati 2019). Based on these methodologies, academics can examine which groups are the main influencers of environmental quality and assess their capacity to adjust towards targeted development and environmental outcomes.

Our study also provides reliable and consistent empirical findings that can be valuable for policymakers when implementing comprehensive environmental strategies for inclusive and sustainable economic development and growth.

Appendix

Fig. 3

CO₂ emissions (kg per capita)

Fig. 4

Income inequality (Gini coefficient)