The Impact of Income Inequality on Carbon Emissions in Asian Countries: Non-Parametric Panel Data Analysis

Widening income inequality and environmental degradation are two of the most important problems that threat the sustainable development. For that, this study aims to examine the effect of income distribution on CO2 emissions in seven Asian countries over the period 1971–2014 using a non-parametric panel estimation method and time-varying coefficients. Specifically, we apply the local linear dummy variable estimator (LLDVE) approach that allows evaluating the coefficients which vary over time for the panel data models. The results reveal that there is a strong non-linear correlation between income inequality and per capita CO2 emissions. The non-parametric model suggests that there is a negative relationship between income inequality and environment degradation over the whole study period expect for the period 1988–1997 which was positive. Our findings broadly support the existence of the "equity-pollution dilemma", whereby income redistribution induces environmental pollution. This dilemma has potential implications for policies designed to promote redistribution in the selected Asian countries.


Introduction
The concerns about sustainable development have increased since the environmental quality is increasingly affected due to economic activities. Global warming and climate change have been the most important environmental problems in our ages. The high carbon dioxide emission (CO 2 ) is the major contributor to the greenhouse effect. Since the industrial revolution, the majority of CO 2 emission has recorded in the industrialized countries. Due to the rapid economic growth, the shares of developing country emissions exceeded those of industrialized economies recently (IEA [44]).
From the beginning of industrialization, economic growth has been accompanied with high-energy consumption and high carbon dioxide emissions in many parts of the world. Therefore, the nexus between CO 2 emission, energy consumption, and economic growth has been the subject of much academic research. This nexus is closely related to verify the validity of the environmental Kuznets curve (EKC) hypothesis, which assumes an inverted U-shaped link between economic growth and environmental pollution. From this perspective, the environmental pollution level increases with per capita income during the early stages of economic growth and then declines after reaching a threshold of a certain per capita income due to high demand and affordability for better environmental quality (Grossman and Krueger [34]).
However, the current period is characterized not only by the growing environmental crises, but also by social crises caused by the unequal income distribution. Income inequality means that the income created in a country for a period, cannot be equitably shared between citizens, regions, or social classes. In both developed and developing countries, the inequitable income distribution contributes to economic and social problems (Piketty [76]; Piketty and Saez [77]; Stiglitz [90]). A rise in income inequality can depress the aggregate demand and contributes to economic stagnation (Fitoussi and Saraceno [28]). Besides, the unequal income distribution increases criminality, health inequalities, and high school dropout rate (Feldstein [27]; Wolde-Rufael and Idowu [102]).
Widening income inequality and growing environmental crises pushed the researchers to investigate the relationship between them. In the literature, there are various approaches through which income distribution can affect the environmental quality. The high level of income inequality can positively or negatively affect CO 2 emissions.
The first approach is based on power weighted social decision rule introduced by Boyce [16] to investigate the effect of income distribution on environmental degradation. He suggests that the environment quality depends on income and the distribution power between different social classes. In society with a high level of income inequality, the rich recklessly use the natural resources. Likewise, in favor of their interests, the rich can also often skew national decision-making which can be more aligned with harmful policies to the environment. Therefore, environmental decisions become dominated by the rich and that leads to environment pollution. Magnani [65] argued that the low level of income inequality can boost the voice of the powerless and allow them to be more productive within the political process. Moreover, egalitarian income distribution raises the demand for better environment by creating an ecological awareness. Therefore, income inequality increases the level of pollution than within a more egalitarian society.
The second approach that describes the relationship between income distribution and environment degradation is based on the marginal propensity to emit (MPE) which suggests that the environmental degradation varies with the equitable distribution of income and the economic behavior of households. Some studies (Ravallion et al. [81]; Heerink et al. [40]; Berthe and Elie [10]) suggest that the high level of income inequality can ameliorate the environmental quality through the marginal propensity to emit. Grunewald et al. [35] and Jorgenson et al. [52] noted that MPE varies according to income level, an increase in income decreases MPE. Likewise, Jorgenson et al. [52] pointed out that the MPE has been also affected by marginal propensity to consume (MPC). If high-income households have a lower MPC, redistribution of income from poor to rich will improve the environmental quality as suggested by the standard Keynesian model.
The third approach suggests that the increase of income inequality may affect negatively the environmental quality by raising the consumption competition which contributes to a rise in energy consumption and CO 2 emissions (Chao and Schor [20]). The rise of consumption is due to the Veblen effect which suggests that products are used not for their functional utility but to signal a social status desired by the consumer; the rich consume luxurious brands to show wealth. Besides, the additional working hours in society with high level of income inequality have increased the consumption of energy and CO 2 emissions due to economic growth and greater consumption by household (Bowles and Park [15], Fitzgerald et al. [29]; Knight et al. [57]; Jorgenson et al. [52]).
Growing income inequality and environmental pollution in relation to economic growth become more and more pressing issues especially in Asian countries. Asia is considered as the "growth center of the world". It has recorded the highest average GDP growth rate of all developing region in the world since 1990. For that this region includes three of the five biggest CO 2 emitters in the world (China, India, and Japan). The rapid economic growth experienced in Asia over the last 20 years is a remarkable success story in the fight against poverty. But this growth has also greatly widened the gap between the rich and the poor. Developing countries in Asia have recorded the highest level of income inequality among six world developing regions. Likewise, there is no study that estimates the time-varying correlation between income inequality and environmental pollution non-parametrically in Asian countries. To our knowledge, the only study that investigated the link between CO 2 emission and income inequality on G7 countries using non-parametric model was presented by Uddin et al. [92].
According to the existing environmental economics literature, the majority of subsequent empirical studies have investigated the environmental impact of income inequality with a parametric approach (see Table 7 in Appendix). Whereas the parametric model has a restrictive functional form that may result in a specification errors, however, the correlation between income inequality and environment degradation can change over time. This relationship can be influenced by many factors such as technological advances to protect the environment, awareness and attitudes towards climate change and preferences for a clean environment, and changes in the rate of urbanization, all of which are sources of non-linearities in the long run. The widespread rise of income inequality and environmental degradation are two serious threats for the sustainable development especially in Asian region. For that, the purpose of this paper is to investigate non-parametrical association between income inequality, economic growth, and environment degradation using the local linear dummy variable estimator (LLDVE) by focusing on seven Asian countries. This technique was invented by Li et al. [59]. It allows evaluating the coefficients which vary over time for the panel data models.
The rest of the paper is organized as follow: Sect. 2 describes the data. Sect. 3 presents empirical specification and methodology in both parametric and non-parametric modelling. The empirical results are summarized and discussed in Sect. 4. The conclusion is provided in Sect. 5.

Data Description
We employ annual dataset for seven Asian countries, consisting of India, Japan, Korea, Malaysia, Pakistan, the Philippines, and Sri Lanka, for the period spanning from 1971 to 2014. The choice of the sample and the time period is strictly based on the data availability. The data has been gathered from five databases. The per capita CO 2 emissions (expressed in metric tons) data for the whole period are extracted from the Carbon Dioxide Information and Analysis Centre (CDIAC) database (Boden et al. [11]). The Gini coefficient as measure of income inequality is extracted from the SWIID database (Solt [89]). In this database, a distinction is made by Solt [89] between pre-and post-tax Gini coefficients, given that post-tax Gini coefficients tend to be weak due to social re-distribution. For that, in this study, we have used the after-tax Gini coefficient data, given that our main objective is to verify whether income re-distribution has an effect on CO 2 emissions. The data of real GDP per capita (expressed in 2011$) and population are collected from Maddison Project Database (Maddison [64]). The data on trade openness (expressed in percentage of GDP) and data on financial development, measured by domestic credit to private sector (in percentage of GDP), are extracted from the World Development Indicators (World Bank [104]) database. Finally, the Economic Freedom of the World Index (EFWI) which is used as instrumental variable for income inequality is collected from the Fraser Institute (Gwartney et al. [36]). 1 Trends in income inequality and per capita CO 2 emissions in the considered Asian countries during the period 1971-2014 are illustrated in Fig. 1. This figure suggests that income inequality and carbon emissions are strongly correlated over this period. However, the trends in these two variables exhibit significant versatility. The trend in income inequality is decreasing, while the trend in carbon emissions is increasing, for the period from 1971 to the end of the 1980s in almost all countries except Japan and Sri Lanka, which are experiencing the opposite situation. Then, throughout the 1990s, the trends are moving in the same way. Afterwards, at the beginning of the 2000s, the situation is similar to that of the first period, but with a lower magnitude. Accordingly, in the overall picture, an inverse correlation between income inequality and CO 2 emissions predominated throughout the 1971-2014 period for these selected Asian countries.

Non-parametric Panel Data Specification
Following the modelling strategy used by Churchill et al. [23] and Uddin et al. [92], our econometric specification linking income inequality to CO 2 emissions takes the following form: (1) can be written under matrix representation such as: where Y it is the natural log of per capita CO 2 emissions and f t = f i ( t T ) are unknown economy-specific trend functions, while t,j = j t T = ( t,1 , … , t,5 ) T are an unknown timevarying coefficients. LnINEQ it , LnGDPPC it , LnPOP it , LnTR it , and LnFD it are the natural log of income inequality, real GDP per capita, population, trade openness and financial development, respectively. i and u it indicate the nonobserved individual effects and the disturbance term, respectively. The ∑ N i i is assumed equal to zero for the purpose of identification.
In our regression, we controlled for the real GDP per capita because it was a commonly used indicator of economic activities in empirical literature, and its effect on CO 2 emissions, on one hand, could be expected to be positive as greater economic activities lead to higher use of energy as well as other natural resources which in turn lead to higher environmental pollution (Bélaïd and Youssef [7]; Mardani et al. [67]). However, on the other hand, the quality of the environment can be improved if increased incomes induce the use of environmentally friendly production techniques and the consumption of cleaner energy (Ghazouani et al. [32]). Population has proven to be the most significant variable influencing CO 2 emissions in large body of empirical literature (Liddle and Lung [60]; Wang et al. [98]; Ahmed et al. [1]; among others). A higher population size may generate more energy consumption and therefore more CO 2 emissions from fossil fuels. It can also contribute to emissions through deforestation. Trade openness has been also a common routine in several empirical investigations. Its impact on CO 2 emissions can be positive or negative. Trade openness induces the flow of final and intermediate goods between countries for consumption or for onward production process. One source of environmental pollution is the greater consumption and production resulting from international trade (Shahbaz et al. [84]). In contrast, trade openness can improve environmental quality through the transfer of environmentally friendly technologies (Frankel and Rose [30]; Shahbaz et al. [85]). Accordingly, trade openness could be controlled in our above specific regression. Likewise, financial development can also be controlled in the regression due to the fact that it can assist firms in adopting enhanced environmentally friendly technologies in the energy industry (Ben Youssef et al. [8]; Boufateh and Saadaoui [14]). However, the financial sector can adversely affect the environment by increasing CO 2 emissions through the growth of the manufacturing sector (Shahbaz et al. [84]). Growing financial development may, therefore, induce a mitigated impact on the environment. Overall, the inclusion of these different variables in the regression may address the issue of omitted variable bias in econometric estimation. Following the example of Churchill et al. [23] and Uddin et al. [92] among others, the common trend f t and the timevarying coefficients t,j will be estimated in a non-parametric manner using the LLDVE method proposed by Li et al. [59]. Silvapulle et al. [87] and Uddin et al. [92] provide a welldetailed formulation of the LLDVE method; however, our study differs from that of Uddin et al. [92] concerning the selected country sample and the control variables used in the model. It should also be noted that the quadratic term of GDP per capita is not considered in our non-parametric modelling, as this method is specifically designed to capture non-linearities of the data, including non-linearities in the GDP-CO 2 emissions nexus across time.
Nevertheless, checking for the existence of non-linear trends in our series is a relevant step before the execution of LLDVE method. For that, we apply the Maan-Kendall trend test (M-K test). This test, proposed by Mann [66] then taken up by Kendall [56], is non-parametric and allows a priori the detection of trends that are not necessarily linear. With the advantage that the data are not necessarily normally distributed or linear, the M-K test tests the null hypothesis that there is no trend against the alternative hypothesis that there is a downward (or upward) trend in the one-tailed test. For the time series x, the M-K statistic (S) is defined as follows: where n is the length of the series, x k and x l two generic values of sequential data, and the function sign x l − x k is defined by: So, the S-statistic represents the gap between the number of positive differences and the number of negative differences found in the analyzed time series. If the series has a tendency to increase (decrease), the S-statistic should take on positive (negative) values. If the hypothesis is null, there is no trend in the correlation data between the considered variables and time, each order in the data set being equally alike. Under this assumption, the S-statistic is approximately a normal distribution with the mean E(S) = 0.

Parametric Panel Data Specification
Our main focus in this study is to examine the impact of income inequality on environmental degradation (which is measured by the per capita CO 2 emissions) in a nonparametric way. However, to provide a benchmark for the results obtained by this method, we will present first the parametric point estimations. In this case, our parametric model, derived from the original EKC model and including income inequality as another determinant of per capita CO 2 emissions, is as follows 2 : where LnGDPPC 2 it is the natural log of the quadratic term of GDP per capita included to capture the non-linear impact of the GDP per capita variable. u it = i + t + e it , where i indicates the individual fixed effect, t refers to the time effect, and e it is the specific-country disturbance term. In order to choose the appropriate method to estimate the parametric panel model (5), we should, first of all, implement some panel test, namely, cross-sectional dependency test, panel unit root test and panel cointegration test.
To test the presence or absence of cross sectional dependency (CSD) across countries, we use the Pesaran's [73] CD test. Indeed, the CD tests' hypotheses are as follows: H 0 : absence of cross sectional dependency across countries against H a : existence of cross-sectional dependency across countries. One conclusion of this test is that it allows us to choose which panel unit root test will be applied to examine the stationarity of the variables. However, if we reject H 0 , second-generation panel unit root tests will be used in this case. Outcomes of the CD test outlined in Table 1 does not accept the null hypothesis in 1% threshold of significance for all variable, with exception of the income inequality variable ( LnINEQ ) which reject H 0 at 10% level, which allows to conclude that our panel data are cross-sectional correlated.
In the presence of CSD, we employ the second-generation panel unit root test proposed by Pesaran [74], namely, crosssectional augmented IPS test (CISP), as it considers heterogeneity as well as cross-sectional dependency issues that exists in our panel data. The CIPS test tests the null hypothesis that the series for each country has a unit root against they do not has a unit root as an alternate hypothesis. Table 2 presents the results of the CIPS test with intercept and trend indicating that the null hypothesis is accepted at the level but rejected at the first difference. Therefore, based on these results, we suggest that our variables are stationary in their first difference.
As for the panel cointegration test, we use the bootstrap panel cointegration test proposed by Westerlund and Edgerton [100]. This test is considered suitable for CSD in the data by computing robust p-values from bootstrap replications. Westerlund and Edgerton [100] augmented four normally disturbed tests which are used to test the null hypothesis of no cointegration. Two, such as between groups (G t ) and among groups (G a ), present group-mean tests, while the other two, namely, between panels (P t ) and among panels (P a ), for the whole panel tests. We implement this test using 500 bootstrap replications to control CSD in order to compute robust critical values. Table 3 provides the findings of this test and shows that there is no evidence of cointegration among groups and among panels tests (i.e., G a and P a ), whereas between groups and between panels (i.e., G t and P t ) tests, panel cointegration relationship exists at the 1% and 5% threshold of significance, respectively. Then, the results suggest the existence of long-run associations among variables. Once the panel evidence of cointegration is confirmed, we can proceed to estimate the panel parametric model. Accordingly, we employ the common correlated effects mean group (CCEMG) estimator proposed by Pesaran [75] as it accounts for CSD, and also it accommodates for heterogeneity or endogeneity issues which can be observed in panel data.
The CCEMG method has been shown to be robust to the availability of a finite number of "strong" factors which may be linked to local spillover effects and an unlimited number of "weak" factors that represent global shocks (Pesaran [73]). The objective of the estimator is to get coherent estimates of the parameters tied to the observable variables. The CCEMG estimators consider non-observed common factors with heterogeneous factor loadings by employing cross-sectional panel averages of the dependent and independent variables as supplementary regressors. According to Kapetanios et al. [55], this provides a given level of flexibility, given that the impact of non-observed common factors may vary across nations, whereas their trend might be non-linear or possibly non-stationary. The CCEMG estimator solves the problem of cross-sectional dependence by including as additional regressors the cross-sectional panel averages of the independent and dependent variables. Pesaran [73] proves that these averages account for the non-observed common factors. Given that for each panel member the relationship is estimated on a separate basis, the heterogeneous impact is likewise provided by construction. Afterwards, the estimated coefficients are averaged across the members of the panel, whereby various weights may be used.

Parametric Specification Results
The baseline long-run results from the CCEMG estimator are given in Table 4. As our main objective is to examine the association between income inequality and CO 2 emissions under the EKC hypothesis, whereas to verify the robustness of our parametric analysis, we have attempted to expand the model by including further control variables, namely, trade openness and financial development. Thus, we have three models to estimate. In column 1, the model is without trade openness and financial development; in column 2, the model without trade openness; and finally, in column 3, the model with all variables. For all specifications, the results suggest that income inequality does not show any significant impact   [92], revealing that in their parametric analysis, income inequality does not affect CO 2 emissions in the G-7 countries. Findings from all specifications indicate also negative and positive coefficients for LnGDPPC and LnGDPPC 2 at different significance level, respectively, which is like a normal U-shaped relationship among CO 2 emissions and economic growth. Therefore, EKC's assumption does not valid. Besides, we find the population variable has a significant and positively signed coefficient, indicating that an increment of the population leads to an increase in CO 2 emissions. However, our parametric modelling fails to provide any significant impact of trade openness and financial development on CO 2 emissions in the panel of these Asian countries as the income inequality impact.
In addition, we have looked at the impact of income inequality along with the other control variables on CO 2 emissions for each of the 7 countries, using the CCEMG estimator. The results depicted in Table 5 show no strong difference from the finding obtained from the whole panel estimate. Especially, findings show that a 1% increase in income inequality is associated with a 1.96% and 6.63% reduction in CO 2 emissions in Japan and Malaysia, respectively. A 1% increase in income inequality leads to an increase in CO 2 emissions by 17.5% in Sri Lanka, while the income inequality impact in India, Korea, Pakistan, and the Philippines remains insignificant. We find evidence that trade openness is positively and significantly associated with CO 2 emissions only in Pakistan and the Philippines, while there is no significant impact for the other countries with the exception of Japan, which exhibits a significant negative impact regarding this relationship. Results show that financial development leads to decline in CO 2 emissions in India and Korea, while degrades the environment in Japan and Pakistan. In addition, our results exhibit that EKC does not verified for each country, whereas a normal U-shaped curve does manifests in Japan, Korea, and the Philippines.
It is important to note that a possible shortcoming of the parametric panel estimator is that estimates can be misleading if the model is incorrectly specified (Hailemariam et al. [37]). To overcome this problem, and to grasp the smooth time-varying association between income inequality and CO 2 emissions, we employ local linear estimates of a nonparametric panel data approach.

Non-parametric Specification Results
Before providing the results of the LLDVE technique, we apply the M-K trend test to verify whether or not our data exhibits a non-parametric trend. The results of M-K test are shown in Table 6. Table 6 suggests that a significant and positive trend in CO 2 emissions is evident in all countries except for the Philippines. Similarly, the global trend for the selected countries is positive and statically significant at the 1% level of significance. With the exception of the Pakistan, income inequality exhibits a positive trend in India, Japan, and Sri Lanka and a negative trend in Malaysia, Korea and the Philippines, while a positive significant global trend at the 5% level of significance. The trends in GDP and in population are positive and significant for each of the countries at the 1% level as well as the global trend for the seven countries. The trade openness shows a significant positive trend in India, Malaysia (at the 1% level), and in the Philippines (at the 5% level) with a positive significant global trend at the 5% level for whole panel. Finally, we prove a positive and significant trend for financial development for India, Japan, Korea, Malaysia (at the 1% level), and Sri Lanka (at the 10% level). The global trend is also statistically significant at the 1% level with a positive sign. In overall, the M-K test confirms the existence of trend in our series. Accordingly, we proceed to a rigorous non-parametric analysis using the LLDVE technique (presented in Sect. 2) to estimate the common trend f t and the time-varying coefficients t,j . Figure 2 depicts, therefore, the non-parametric results from the LLDVE method, along with the 90% confidence intervals. The non-parametric panel data estimates, contrarily to parametric point estimates, capture the time-varying association between income inequality and per capita CO 2 emissions. According to Fig. 2, the relationship between income inequality and per capita CO 2 emissions is significant and negative for almost the entire period, except for 1988 to 1997 when it was significant and positive.
Therefore, based on these results, we suggest that there is a reverse relationship between income inequality and per capita CO 2 emissions for the whole period, except for the period 1988-1997 when it is positive, whereas, because of the boundary effect that characterizes the non-parametric  3 The reverse relationship between income inequality and per capita CO 2 emissions during the period 1975-1988 is sustained by the income distribution patterns observed at the time. The period 1975 to 1988 was characterized by significant growing of income inequality. However, according to WIR [105], income inequality has been widening rapidly in Asia since 1980s. Although technological advances, globalization, and market-oriented reforms have been the main drivers of rapid growth in developing Asia over the past two decades, they have also had enormous distributional consequences (Kanbur et al. [53]). A commitment has been made by all Asian countries to become more integrated with the world economy in the decades to come. However, globalization is considered one of the factors behind the surge of income inequality in the region (Zhuang et al. [111]). For instance, trade integration might affect the relative demand and therefore the relative wages of skilled and un-skilled workers. It might equally change the distribution of income among capital and labor, since capital and skills frequently have a complementary relationship and often operate together, which depresses the demand for un-skilled labor (Wood [103]). For its part, financial integration can enhance poor people's access to finance, yet gains are likely to be elite-driven, which means that inequality rises at low-income levels and falls with rising income (Choi [22]; IMF [46]; Claessens and Perotti [24]). Beyond this, fiscal policy weaknesses are another factor driving up inequality in the region in this period (Zhuang [110]). For the majority of countries in the region, tax systems are heavily skewed towards consumption taxes, which impose a heavy burden on low-income and middleincome groups. Alternatively, the wealth of the upperincome groups is likely to be concentrated under this tax system because taxes are quite favorable to labor income instead of capital earnings and property. Therefore, this leads to falling in MPC from low-income and middleincome groups. For this reason, the opposite relationship between rising of income inequality and decline in per capita CO 2 emission may be explained by the falling of the MPC and also by the fewer emit upper-income earners in the period [1975][1976][1977][1978][1979][1980][1981][1982][1983][1984][1985][1986][1987][1988].
While all the flows that make up globalization can affect income inequality, the most important impact probably comes from flows related to technology and information. Technological change thus affects employment by devaluing and revaluing skills and, indeed, creating entirely new skills and jobs. For many, this relationship between skills and technology is an important factor in increasing income inequality (Goldin and Katz [33]). However, and according to Hüfner and Koske [43], the increasing income inequality may improve the marginal propensity to invest for the upperincome households as its inclined MPC. Therefore, through financial deregulation that facilitated corporate borrowing, income inequality has promoted capital accumulation, which has usually been a complement to productivity or productivity enhancement in high-income economies (Jones [51]; Uddin et al. [92]). At the same time, technology advances have brought about improvements in the energy efficiency of goods bought by upper-income households as well as by lower-income households (Uddin et al., [92]). Thus, this is another possible way that may explain the reverse relationship between per capita CO 2 emission and the higher-income inequality in the selected Asian countries for this period.
Income inequality continued to rise from 1988, but it reversed around the mid-1990s. While the relationship between income inequality and per capita CO 2 emission is significantly positive for the period 1988-1997, one explanation for the environmental degradation exacerbated by the growing income inequality between rich and poor could be that the rich because of their wealth have access to lifestyles which are energy intensive, commodity intensive, and manufactured goods (UN/ESCAP and ADB, [94]). Some of the most significant changes are the increasing usage of private automobiles and the growing use of a number of other consumer durables including televisions, air conditioners, and washing machines. There has also been a significant increase in the consumption of wood products, particularly paper. For instance, according to UNDP [93], the number of vehicles in East Asia multiplied by 14 between 1975 and 1993, over seven times the global average growth rate. Consequently, the energy use increased at an average annual rate of about 6% for the period 1980-1998, which is five times faster than other regions (Cleveland [25]).
The positive relationship between high-income inequality and per capita emission is also mainly related to overexploitation of natural resources in the Asian's region. Meanwhile, within high-income inequality societies, wealthy people have a tendency towards a powerful influence on policy-making. Inequalities in income may cause political instability leading the rich to favor a policy that overexploits natural local resources and exports the proceeds overseas (Boyce [16]). At the same time, the poor are denied vital resources and are compelled to resort to unsustainable exploitation of the natural resources at their immediate disposal to satisfy their livelihood needs (UN/ ESCAP and ADB [94]). This explains why, around 1990s, it was estimated that the countries of the region were losing between 70 and 90% of pre-existing natural habitat as a result of agriculture, as well as infrastructure development, deforestation, and land deterioration (MacKinnon and MacKinnon [63]). India, Indonesia, the Philippines, and Sri Lanka were among the countries that recorded the most serious losses. The Philippines, for example, experienced the loss of about 70% of its mangrove forests (AOE, 2001). About 40% of the land capable of supporting closed tropical forests was estimated to have become devoid of forest cover in the period, mainly as a resulting from human actions (Wilson [101]). Additionally, many countries in the region were confronted pressure on land. About 28% of the region's land area had some degree of land degradation in 1990s, among them about 350 million ha of degraded land are in India, Pakistan, and China (UNEP [95]). Therefore, the over-exploitation of natural resources, either by rich or poor, will intensify the energy use and hence will generated more CO 2 emissions (Wu et al. [106]). Moreover, with the depletion of natural resources, their role as CO 2 emission sinks will be destroyed.
After reaching a threshold level in the mid-1990s, income inequality shows a declining trend until the mid-2000s. The Asian financial crisis of 1997-1999 is considered to have been a key factor behind this trend. It is common for the rich to face significant capital losses during a crisis, which adversely affecting their incomes (Lopez-Acevedo and Salinas [62]). In addition, crises lead to greater income damage for employees in non-tradable sectors, including financial services, as well as for those in the top income deciles (Wan and Wang [97]). However, the decline in income inequality was accompanied by an increase in CO 2 emissions in these Asian countries, particularly between the late 1990s and the mid-2000s, before experiencing a slow downward tendency after this period. The increase in CO 2 emissions stems from an increase in the consumption of energy-intensive products, especially by low-income households. Where the increase in income equality has contributed to an improvement in the wealth of lowincome earners, which is reflected in an increase in their MPC more than those with high incomes, so, more consumption of energy-intensive products boosted the MPE and intensified CO 2 emissions.
The mid-to-late 2000s, the relationship between income inequality and per capita CO 2 emissions was negative. Although income inequality shows a slight increasing trend, CO 2 emissions started with a small decreasing trend and then became stable. This result partly reflects both the growing commitment of these Asian countries to environmental challenges from the early 2000s and the nature of income inequalities around that time. Following the growing concern in the 1990s regarding the effects of CO 2 emissions on climate change, countries in the Asian region have already undertaken significant measures to diversify their energy mix and have started to realize substantial socio-economic gains. National renewable energy targets have been set by all countries in the region. For example, in 2003, Japan's government stepped up its involvement in the energy market and introduced its Renewable Portfolio Standard Scheme in which electricity retailers were required to provide a portion of electricity generated from renewable resources to grid users (IEA [45]; Wang et al. [99]). The growth of non-fossil based energy has been prioritized in Malaysia since 1999 and has been emphasized as the 5th fuel in their energy mix with the target of providing around 5% of its power generation in by 2005 (Siva et al. [88]). Similarly, the renewable energy growth in Korea showed a rapid growth between 1990 and 2013, at about 46% (Boo [12]). Accordingly, by 2015, investment in renewable energy, excluding large hydropower projects, has reached $160.6 billion in Asia and the Pacific, among them about $103 billion in China and $10 billion in India (McCrone et al. [69]). That points to the significant expansion of the renewable energy sector since the beginning of the century. Subsequently, therefore, using one unit of electricity would necessarily lower, on average, CO 2 emissions. In brief, renewable energy development and deployment has brought about an overall decline in an individual entity's ability to emit. This result is consistent with Zheng et al. [108] who found that emissions have a downward trend in East Asia during the period 2005-2016. The upward trend in income inequality is mainly related to the rapid economic growth experienced by the region in the early 2000s (OECD [70]; Jain-Chandra et al. [49]). Rapid growth has contributed to a rise in income inequality because of rapid income growth among the high-income groups in urban areas (Shi and Sicula [86]). Therefore, one explanation for the negative relationship between income inequality and per capita CO 2 emissions may be a re-distribution of income in favor of higher incomes that have limited emission capacity, under the assumption that the MPE declines with income.
For the other variables in Fig. 2, the GDPpc plot looks roughly like a normal U-shape, whereby the relationship between GDP and CO 2 emission follows a downward trend from the mid-1980s to the mid-2000s and then starts to increase from that time onward. This result, which is consistent with the parametric estimation findings, does not yet validate the EKC hypothesis for our sample. Since the majority of the countries in our sample are developing countries, a rejection of the conventional EKC hypothesis is not surprising since they may not have reached the expected incomelevel to capture the inverted U-shaped relationship embodied in the EKC hypothesis. In fact, some existing studies have found the same outcome not only for the same countries, but also for the same study period, we worked on, such as Kanjilal and Ghosh [54] and Chandran and Tang [19] for India and Malaysia, respectively, for the period 1971-2008 and Saboori and Sulaiman [82] for the Philippines in the 1971-2009 period. We find that relationship between trade openness and CO 2 emissions is non-linear, despite the small magnitude of the effect. The time-varying coefficient of trade openness was negative for the 1980-1987 and 1993-2000 periods, while it was positive for the 1988-1992 period. We find that financial development has a non-linear association with CO 2 emissions. The financial development's timevarying coefficient is positive and has an inverted U-shape from 1976 to the late 1980s and becomes stable until 2000 before turning negative in the 2000-2010 period. The findings are coherent with the positive impact on CO 2 emissions during the early stages of financial development, while at later stages, after a certain level, financial development can contribute to the reduction of emissions (Javid and Sharif [50]; Ben Youssef et al. [8]; Uddin et al. [92]). Figure 3 plots both the country-specific CO 2 emissions trend estimations and the common trend function with 90% confidence intervals. In each box, the dark blue and the dashed red lines represent the common trend function and country-specific CO 2 emission trends, respectively, while the black dashed lines are the wild bootstrapped 90% confidence bands. The blue light lines are the actual CO 2 emissions recorded for the individual economy. The CO 2 emission individual trends for Japan, Korea, and Pakistan are above the common trend, although CO 2 emissions in Pakistan converge towards the common trend until the mid-1990s and then become slightly higher than the common trend. CO 2 emissions in Japan and Korea are very higher than the common trend, while the specific trend of the Korea is increasing considerably and that of Japan seems to have a horizontal shape. The individual trends of CO 2 emissions for India, Malaysia, and the Philippines are entirely below the common trend, although India's specific trend closely follows the common trend. Finally, CO 2 emissions in Sri Lanka largely follow the common trend until the early 1980s after which they fell below the current trend.

Common and Country-Specific CO 2 Emissions Trends
Overall, although the country-specific trend functions show significant differences in CO2 emissions across the considered Asian countries, they broadly follow the common trend function.

Robustness Cheek: Su and Ullah [91] Endogeneity Test
The endogeneity of income inequality is a potential concern as this variable may be correlated with another omitted variable, exposed to measurement error problems, or codetermined with another explanatory variable in the model. In order to address this issue, we carried out the endogeneity test suggested by Su and Ullah [91], which is relevant in a non-parametric set-up. 4 To implement this test, we employ the economic freedom as an instrument variable (IV). Economic freedom is generally linked to market-oriented institutions and policies that are supposed to contribute to long-run economic development which in turn may reduce income inequality (Acemoglu et al. [2]), while some believe that higher degrees of economic freedom are also at the cost of rising income inequality (Okun and Summers, [71]). 5 Accordingly, we believe the existence of strong correlation between economic freedom and income inequality; hence, the former can be treated as IV for the latter. The non-parametric estimator of Su and Ullah [91] builds on the same idea as the standard two-step least squares, whereas it operates in three steps. The first step consists in performing a non-parametric regression of the variable being endogenous on all regressors that are exogenous. In the second step, a non-parametric regression of the dependent variable will performed on each of the regressors, including the variable being endogenous and the residuals arising from the first step regression. The third, and final, step implies marginal integration to ascertain the validity of the zero mean error term hypothesis.
Following Henderson and Parmeter [41], we now conduct this endogeneity test involving income inequality as potential endogenous variable and trade openness as control variable. 6 The outcomes of this test are shown in Fig. 4. We compare the results from local linear estimates where endogeneity is ignored (left-hand diagram) with the IV local linear estimates of Su and Ullah [91] that considers endogeneity (right-hand diagram) by employing economic freedom as IV of income inequality. In Fig. 4, each point on the dark black mid-line corresponds to the estimated gradient for each observation, while each point on the outer light blue and light red lines refers, respectively, to the lower and upper limits of a 95% bootstrapped confidence band for the specific point estimates. When the upper and lower confidence intervals for this estimate both lie in the upper right quadrant, the estimated gradient for an observation is statistically significant with positive sign. If the confidence intervals for the estimated gradient both lie in the lower left quadrant, the given estimate for that observation is statistically significant with negative sign. However, the gradient estimate for an observation appears to be statistically non-significant in cases where the upper or lower confidence intervals overlap the horizontal axis (Henderson et al. [42]).
Based on the results reported in Fig. 4, we observe that the local linear estimate that ignores endogeneity (left panel) and the Su and Ullah IV local linear estimation that take endogeneity into account (right panel) show no significant difference. The statistical significance of the estimated gradients is proven at the 5% level of significance, which means that endogeneity does not bias the estimates. Furthermore, the estimated gradient is broadly distributed in the lower left quadrant, more so than in the upper right quadrant, suggesting that the relationship between income inequality and carbon emissions is mainly negative. This broadly confirms the LLDVE findings obtained in the previous section (Fig. 2).

Conclusion and Policy Implications
We have tested the relationship between income inequality and CO 2 emissions for a panel of seven Asian countries over the 1971-2014 period using a non-parametric procedure. Specifically, we empirically examined the time-varying trend and coefficient functions that govern the income inequality-carbon emissions relationship in these countries employing the local linear data-driven method -LLDVE technique. Carbon emissions, under this approach, are allowed to evolve over time into the form of unknown functional forms along with confidence bands generated by the wild bootstrap method.
Our non-parametric panel estimates reveal a strong non-linear relationship between income inequality and CO 2 emissions which is largely negative. The estimate coefficient of income inequality is negative and statically significant for the periods between 1975 and 1988 and between 1997 and 2010, while positive and statically significant for the period 1988 to 1997. We assign these findings to the main factors of inequality and CO 2 emissions over these different periods. Between 1975 and 1988, income inequality increased mainly due to globalization, which depressed the demand for unskilled labor and promoted financial gains in favor of the rich, consumption taxes that weighed heavily on the poor and technological changes that negatively affected employment. Higher-income inequality leads to lower MPC for low-and middle-income households, which in turn leads to weaker MPE. Falling in MPE and improvements in the energy efficiency of goods -bought by upper-income households -through technology advances are contributing in decline of CO 2 emissions during this period. Between 1988 and 1997, the growing in income inequality is due generally to the same above-mentioned drivers. The rise in CO 2 emissions over this period is induced by the growing consumption of energy-intensive goods by high-income households and the over-exploitation of natural resources by both rich and poor people. Between the end of the 1990s and the mid-2000s, the decline in income inequality is due in particular to the Asian crisis of the 1990s. Greater income equality led to an upsurge in MPE due to an increase in MPC of low-income households. Between the mid to late 2000s, the increase in income inequality is due to the rapid economic growth experienced by the region in the early 2000s. The slight decline in CO 2 emissions during this period highlights the growing commitment to environmental challenges, lower MPEs arising from increased income inequality, and improvements in green technologies.
Our findings with regard to the income inequality and CO 2 emissions linkage support the existence of a potential trade-off among them, commonly called the "equitypollution dilemma" whereby income redistribution induces environmental pollution (Sager, 2019). There are potential implications associated with this dilemma for policies designed to promote redistribution in our selected Asian countries. At first glance, as noted by Sager (2019), the "equity-pollution dilemma" does not necessarily mean that income redistribution is not desirable; rather, the optimal degree of redistributive policy requires an in-depth understanding of welfare economics and will depend on a number of assumptions about market structure, household well-being, and desired social outcomes. Another implication is that income redistribution policies in the presence of strict environmental protection policies, such as the promotion of renewable energies, will contribute to improving the quality of the environment. In addition, income redistribution-environmental linkage will provide an important platform for designing appropriate environmental policies that builds on each country's competitive advantage and resource endowments. In this regard, such a relationship suggests that there is a need to synchronize redistributive policies and environmental policies between Asian countries either in national and regional levels (see, e.g., Rasiah et al. [79]; Masud et al. [68]).
In conclusion, it noteworthy that our results suggest that the equity-pollution dilemma existed consistently from the mid-1970s to 2010 in the selected Asian countries; however, its inexistence afterwards could reflect the boundary conditions involved in this non-parametric analysis. While by observing trends in CO 2 emissions and income inequality from 2010 onwards, its existence is suspected. Therefore, that encourages us to further explore this relationship once aggregate emission data are available after 2014 for these and other countries in the Asian or other continents.