Curvature and turning point of the environmental Kuznets curve in a global economy: the role of governance

Abstract

This study explores the role of governance in sha** the environmental Kuznets curve (EKC), especially focusing on its curvature and turning point. The study highlights the role of various governance indicators in the formulation, implementation, and enforcement of environmental regulations. However, the study asserts that since in develo** countries poverty, infrastructure, and human development are valued above a clean environment, good governance is less likely to contribute to mitigating pollution in develo** countries. Using a panel of 160 countries, the study finds that better governance helps bring down the critical level of per capita income at which the relationship between income and pollution turns negative. Furthermore, the EKC can be shifted downward by improving governance in the respective countries. The study, however, reveals that the dividends of better governance are more pronounced at higher income levels. Since good governance works only when the desired regulations are in place, it is recommended that for poor countries aid packages for governance reforms need to emphasize the enactment of specific environmental regulations. Investment in institutions is expected to yield maximum dividends in such countries that have gained high-income status but are still lacking in institutional development.

Introduction

The inadvertent economic activities of humans on the planet are the main cause of producing by-products in the form of pollution, which has increased rapidly in recent decades (Budyko 1977). Pollution appears in many shapes like land, water, air, noise, light, and radiological pollution (CEF 2016). This study, however, mainly focuses on carbon dioxide emission, which contributes a major share to greenhouse gas (GHG) emissions. GHG is mainly caused by the consumption of fossil fuels and other anthropogenic human activities (Reay et al. 2007). It is a negative externality of modern economic growth that diminishes the waste recycling capacity of the environment. The unchecked GHG emissions have detrimental impacts in the form of global warming, melting of glaciers, increases in sea levels, worsening of human health, and threats to marine life and wildlife.

Various studies have been conducted to find out the nature of the relationship among the various indicators of environment and per capita income. Among these, CO₂ emission is the premier indicator, which is considered to be responsible for 76% of GHG emissions and the resultant increase in global temperature. In light of scientists’ recommendations for a sustainable environment, the Intergovernmental Panel on Climate Change (IPCC) vows to preserve the natural ecosystem and limit global warming to 1.5 °C rather than 2 °C (IPCC summary 2018; Burakov and Bass 2019). The voices for the protection of climate have led to the Kyoto Protocol 1997 agreement, which was the first breakthrough towards mitigating emissions by developed and develo** countries. The Kyoto Protocol puts constraints on the excessive use of fossil energy by countries. The reduction in GHG emissions is also included in the Millennium Development Goals (Tarverdi 2018).

The demand for a clean and safe environment plays the role of a normal good in consumer preferences. As the income level in developed countries increases, the demand for a cleaner environment increases, whereas low-income countries are not so much fearful about environmental degradation (Lucas et al. 1992). The environmental economics literature reveals that if the income elasticity of clean environment demand is positive, then as GDP per capita increases, the demand for the protection of the environment will also increase.

The above behavioral pattern is often represented by the so-called environmental Kuznets curve (EKC), a concept analogous to Simon Kuznets’ theory of the inverted U-shaped relationship between economic development and income inequality. According to the EKC hypothesis, emission of environmental pollutants follows an inverted U-shaped relationship with the level of economic activity. This relationship states that at the initial phase of economic growth increase in economic activity is associated with environmental degradation because of the increased use of fossil energy. However, after a certain threshold level as income levels increase further, the increase in demand for a clean environment offsets the increase in the supply of pollution, thereby resulting in an inverse relationship between income and pollution (Grossman and Krueger 1991; Panayotou 1993; Arrow et al. 1995; Cole et al. 1997; De Bruyn 2002; Majeed and Mazhar 2020; Farooq et al. 2022).

According to another explanation, in the early phase of industrialization, the environment is badly affected due to the increased level of pollution because more importance is given to the output of goods, employment, and income than a clean environment (Dinda 2004; Dasgupta et al. 2002). This process exhausts natural resources rapidly; and due to the low income of the people, they are unable to pay the cost of avoiding such environmental losses. With further economic growth achieved through industrialization, the living standards of people rise and they start fearing about environmental losses. Institutions tend to come forward to confront pollution issues and intervene through regulations. On behalf of the public interest, these regulations bring pollution to lower levels.

To what extent this realization of the importance of a clean environment translates into such practical steps that eventually result in a cleaner environment depends on a country’s ability to enact and implement effective regulations with consistency and accountability. These aspects of good governance are part and parcel of the very basic definition and concept of governance. A simple definition is an extent to which governments through their institutions legislate and implement policies in the best interest of citizens. According to Kaufmann et al. (2005), governance is the plural and inclusive concept, which is the outcome of joint activities of the non-profit sector and entrepreneurs and represents a plan for sustainable development and quality of life.

The role of institutions in the enforcement of environmental regulations has been discussed in various studies from the early literature (Panayotou 1997; López and Mitra 2000; Dasgupta et al. 2002; Sohail et al. 2022). Since individuals cannot enact or enforce any regulations for environmental protection, this responsibility falls on state institutions that can step forward in the public interest, kee** in view the social benefits and costs of their actions. Hamilton et al. (2004) have added governance in the EKC analysis of the environment and concluded that governance improvements can reduce pollution significantly before a country reaches the middle-income level. Kaufmann et al. (2005) studied the role of governance, considering its six individual indicators, in controlling pollution. The role of governance in environmental protection has also been confirmed recently for BRICS countries by Danish et al. (2019), for Middle-Eastern and North African countries by Farzanegan and Markwardt (2018), for 100 developed and develo** countries by Lau et al. (2018), and for Asian-Pacific countries by Ulucak (2020).

The present study contributes to the literature in the following ways: first, the study explores the role of governance in environmental protection following a different approach. Rather than just considering how governance can influence pollution, the study also allows governance to act as a moderating factor that can potentially alter the relationship between income and pollution. Thus, improvement in governance has assigned two roles. First, better governance can lead to a cleaner environment independent of income level. Second, better governance can also influence the nature of the relationship between income and pollution. In particular, the study explores whether better governance can make it possible to reverse the direction of a positive relationship between income and pollution at an earlier rather than later phase of economic growth. In other words, the study explores the possibility of avoiding the negative externality of growth in terms of pollution emissions at per capita incomes less than otherwise possible. Second, the study provides a global perspective employing sufficiently rich data containing broad variations in per capita income, emission level, and state of governance. All available panel data are used, which consist of 160 countries and 23 years (1996 to 2018).

This study attempts to answer the following questions. First, can pollution be reduced systematically by adopting good governance practices in general? Second, is the effectiveness of good governance in reducing environmental pollution uniform across the whole range of low to high incomes? In other words, do the marginal improvements in governance yield similar dividends among poor countries that tend to lie on the rising portion of EKC as in rich countries that are expected to lie on the declining portion of EKC? Third, do better governance practices enable various countries to bring the turning point of EKC to lower levels of per capita income? In other words, can the negative externality of rising per capita income in terms of environmental pollution be mitigated at an earlier stage of development by adopting good governance practices?

The rest of the paper is organized as follows: a brief review of literature in the “Literature review” section is followed by methodology in the “Methodology” section. Data, variable construction, and estimation procedure are described in the “Data and estimation procedure” section, while the results are presented in the “Results and discussion” section. Finally, the study is concluded in the “Conclusion and policy recommendations” section.

Literature review

The empirical relationship between environmental indicators and per capita income was first explored in three contemporary and independent studies by Grossman and Krueger (1991), Shafik and Bandyopadhyay (1992), and Panayotou (1993). These studies found an inverted U-shaped relationship between some proxy of environmental pollution and per capita income. This relationship was termed as environmental Kuznets curve (EKC) in Panayotou (1993) because of the analogy with Kuznets’ curve (Kuznets 1955).

Following this initial work, the possible existence of EKC in various sets of countries has been explored with a variety of data sets and econometric techniques. For example, Selden and Song (1994) employed panel as well as cross-sectional data and found evidence in support of the EKC hypotheses. However, they observed that the turning points of per capita income for certain categories of emission are higher than those found in previous studies and warned that the overall level of pollution is still expected to rise in future. Roberts and Grimes (1997a, b) studied the relationship of CO₂ emission and GDP growth for 147 countries for the period 1965–1990. While the study confirmed the presence of EKC for developed countries, it found monotonic positive relationship between income and pollution in develo** countries.

Research on the nature of relationship between emission and income (including EKC) has been extended in various directions such as the consideration of various pollution indicators and the roles of energy intensity, trade openness, foreign direct investment, sector-wise composition of the economy, and financial-sector development (see, for example, Omisakin and Olusegun 2009; Acaravci and Ozturk 2010; Gani 2012; Onafowora and Owoye 2014; Farhani et al. 2014; Abid 2016; Balogh and Jámbor 2017; Bokpin 2017; Ang 2007; Jabeur and Sghaier 2018; Ahmad and Majeed 2019; Ahmad and Majeed 2022; and Ullah et al. 2021). The empirical evidence in Bekun (2022) indicates that CO₂ pollution in India can be reduced significantly through the replacement of fossil fuel energy with renewable energy. The study also observed that the current investment in the energy sector appears significantly environmentally friendly.

Several studies have explored the role of governance in controlling environmental pollution with mixed findings. Using data for 16 Middle-Eastern and North-African countries, Jafari et al. (2012) observed that improvements in government effectiveness tend to reduce environmental degradation, while the impacts of regulatory quality and control of corruption are ambiguous. Similarly, Rockström et al. (2013) observed that the absence of rule of law, control of corruption, transparency, and accountability have contributed to environmental degradation. More recently, Swain et al. (2020) observed on the basis of evidence from 58 non-OECD countries that control of corruption is effective in improving environmental quality. This result confirms the earlier finding of Desai (1998), Fredriksson and Svensson (2003), Cole (2007), and Sahli and Rejeb (2015) that increase in the level of corruption leads to further degradation of the environment.

In another comprehensive study of 99 develo** countries, Gani (2012) concluded that good governance indicators, specifically, political stability, rule of law, and control of corruption, have negative impacts on CO₂ emission. Similar conclusions are drawn in Esfahbodi et al. (2017) for the UK and Bokpin (2017) for the African region and Jabeur and Sghaier (2018) for Middle-Eastern and North-African countries.

In a recent study based on panel data of 35 countries from different income groups, Güney (2022) found that good governance can be instrumental in reducing CO₂ emissions through the adoption of solar energy generation. Danish et al. (2019) also found support for the favorable effect of improvement in governance on CO₂ emission in BRICS countries. Similar results are reported for BRICS in Baloch and Wang (2019), for South Asian countries in Zakaria and Bibi (2019) and Ashraf et al. (2022), for 38 Sub-Saharan countries in Sarpong and Bein (2020), for 93 emerging and develo** countries in Wawrzyniak and Doryń (2020), and more recently for Saudi Arabia in Sarwar and Alsaggaf (2022).

The above discussion shows that good governance appears as an important tool for addressing the issue of environmental pollution. A few empirical studies (Laegreid and Povitkina 2018; Wang et al. 2018; and Wawrzyniak and Doryń 2020) have also analyzed the potential moderating effects of good governance on environmental pollution and found some encouraging results. However, these studies did not address the role of governance in sha** the EKC. Thus, the role of governance in this context still appears underexplored. In particular, it is not clear whether good governance is effective in controlling pollution at all levels of income and can good governance help reverse the rising trend of environmental pollution in relatively low-income countries or whether we still have to wait till such time these countries catch up with middle- to high-income countries. In other words, there is a need to explore the role of governance in changing the shapes of the EKC, in particular, the location of its turning point.

Methodology

Before bringing governance into the EKC equation, it is important to understand the operational concept of governance and its measurement. According to the World Bank reports (Kaufmann et al. 2011; Kaufmann and Kraay 2007), two indicators of governance namely “government effectiveness” and “regulatory quality” are developed to measure the government’s capacity to formulate and implement sound policies. Another two indicators, “rule of law” and “control of corruption,” are aimed at measuring respect for the institutions that govern economic and social interactions between citizens and the state. Finally, two more indicators “voice and accountability” and “political stability and absence of violence” measure the quality of process by which governments are selected, monitored, and replaced.

While government effectiveness is important for the formulation of environmental protection policies, regulatory quality, rule of law, and control of corruption are important for the implementation, enforcement, and acceptance/respect of these policies by public and state organs. However, any indicator of good governance does not by itself guarantee a cleaner environment unless a clean environment occupies an important place in the manifesto of a political party (or any other group or dictator) holding power to govern a country. In developed countries, a clean environment is valued more highly by citizens and, hence, by political parties (or any other aspirants of power), as compared to develo** countries where eradication of poverty, raising standards of living, and development of infrastructure and human development facilities are given prime importance. In other words, good governance is less likely to yield dividends in the context of the environment in develo** countries where the public policy does not critically focus on the environment and there are not many such policies in place that are to be implemented and enforced.

The above discussion provides a basis for introducing governance in EKC as a moderating factor that enters the equation in an interactive form with income. A few earlier attempts in this direction (e.g. Laegreid and Povitkina 2018; Wang et al. 2018; and Wawrzyniak and Doryń 2020) have allowed the moderating role of governance in the GDP-CO₂ relationship either in linear setting or by splitting date between sets of countries having high- and low-quality institutions on environmental pollution and found encouraging results. However, the present study addresses the role of governance in sha** the EKC by making all parameters of the curve to be potentially sensitive to the state of governance. Representing the state of the environment by per capita CO₂ emission and the level of economic activity by per capita income, denoted by CO₂ and Y, respectively, we start with the following standard quadratic equation in logs underlying the EKC equation (Eq. 1):

$$\mathrm{logCO}2=\alpha +\beta \mathrm{log}Y+\delta \mathrm{log}{Y}^{2}+U$$

(1)

The location and shape of EKC depend on the three parameters present in the equation. To bring in the role of governance, these parameters are allowed to vary systematically with the state of governance, denoted by G, as follows.

$$\alpha ={\alpha }_{0}+{\alpha }_{1} G$$

(2a)

$$\beta ={\beta }_{0}+{\beta }_{1} G$$

(2b)

$$\delta ={\delta }_{0}+{\delta }_{1} G$$

(2c)

Finally, substituting right-hand sides of Eqs. (2a), (2b), and (2c) in Eq. (1) yields:

$$\mathrm{logCO}2=\left({\alpha }_{0}+{\alpha }_{1} G\right)+\left({\beta }_{0}+{\beta }_{1} G\right)\mathrm{log}Y+\left({\delta }_{0}+{\delta }_{1} G\right)\mathrm{log}{Y}^{2}+U$$

(3)

EKC hypothesis is verified if the rate of change in CO₂ emission is positive up to a certain level of per capita income, say Y^* and negative thereafter. Further mathematical details to work out the role of governance in sha** EKC are provided in the Appendix.

Finally, to avoid possible specification bias, Eq. (3) is extended to include other (control) variables in the model. Although the literature proposes quite a few potential control variables, we consider four of them on the basis of overwhelming evidence. For example, using a dynamic global simulation model, Koch and Kaplan (2022) showed that restoring even half of the potential forest area can account for 56 to 69% of the carbon storage. In another global study, Lawrence et al. (2022) found that the net cooling effect of forests (the net of biophysical and biochemical effects) is substantial, which also includes favorable spatial spillover effects. Similar results are reported by Li et al. (2021) for 30 provinces in China.

Muhammad et al. (2020) analyzed the effects of trade on CO₂ emission and found that these effects are statistically significant, though their signs and magnitudes can vary across income groups. Meng et al. (2018) observed that export-led global CO₂ emission has surged because of the emerging global supply-chain trends that tend to rely on the develo** South for the production of energy-intensive raw materials and intermediate goods.

Literature also indicates that production activity in different sectors of the economy is not equally energy intensive and may face different sets of environmental regulations. For example, using a panel of 94 middle-income countries, Sebastian et al. (2022) found a negative relationship between CO₂ emissions with agricultural value added for any given level of GDP. Similar results are reported in Khurshid et al. (2022) for Pakistan. However, Eisen and Brown (2022) found that livestock production has a substantial adverse impact on the environment through methane and nitrous oxide. Similarly, the contribution of manufacturing output to pollution is well documental (see, for example, Liu et al. 2019 and Stavropoulos et al. 2022 for a recent account).

Based on the above discussion, Eq. (3) is written in the following general form for econometrics application with panel data, wherein Forest, Trade, Man, and Agr denote the proportion of geographic area covered under forests, trade openness (exports plus imports as ratio to GDP), and the shares of manufacturing and agricultural outputs in GDP, respectively. The subscripts i and t are country and period (year) identifiers (see Eq. 4).

$$\begin{array}{c}\mathrm{log}{\mathrm{CO}2}_{it}={\alpha }_{0}+{\beta }_{0}\mathrm{log}{Y}_{it}+ {\delta }_{0}\mathrm{log}{Y}_{it}^{2}+{\alpha }_{1} {G}_{it}+{\beta }_{1} {G}_{it}\mathit{log}{Y}_{it}+{\delta }_{1} {G}_{it}\mathrm{log}{Y}_{it}^{2}\\ + {\theta }_{1} {\mathrm{Forest}}_{it}+{\theta }_{2} {\mathrm{Trade}}_{it}+{\theta }_{3} {\mathrm{Man}}_{it}+{\theta }_{4} {\mathrm{Agr}}_{it}+{\lambda }_{i}+{\phi }_{t}+{\varepsilon }_{it}\end{array}$$

(4)

Note that ${\lambda }_{i}$ and ${\phi }_{t}$ represent the country and period effects that can be deterministic or stochastic. If ${\lambda }_{i}$ are treated as deterministic (stochastic) components, the model is referred to as country fixed (random) effects model. Similarly, deterministic (stochastic) treatment of the components ${\phi }_{t}$ is akin to period fixed (random) effects model.

Data and estimation procedure

Since data on governance indicators are not available for a sufficiently long period, the study relies on panel data for the period 1996 to 2018. A sample of 160 countries is selected for which the required data are available over almost the entire period. A few missing values are filled by inter-/extrapolation.

Environmental pollution is approximated by per capita CO₂ emission, while income is represented by per capita GDP. The data on CO₂ emission are in million tons, which are converted to kilograms and expressed in per capita terms. For valid international and over time comparison, the data on GDP are taken at constant PPP 2011 US dollars.

The governance index is computed based on four governance indicators: government effectiveness, regulatory quality, rule of law, and control of corruption, which are constructed to measure the government’s capacity to formulate and implement public policies and to ensure that their enforcement is respected by the public and state organs. The last two indicators, voice and accountability and political stability and absence of violence, are not included. The reason is that in many develo** countries, democracy has not taken roots and as a result, democratically elected governments may not necessarily show better governance records on effectiveness and implementation of policies even though they may show moderate governance scores on how governments are selected, monitored, and replaced.

The governance indicators are scaled − 2.5 (the worst state) to 2.5 (the best states). Since all four indicators are highly correlated, their direct inclusion in the regression equation will most probably distort parameter estimates due to multicollinearity. In such situation, it makes sense to combine all the indicators into a single index. This index is calculated using principal component analysis (PCA) and is linearly normalized on zero to one scale.

Data on CO₂ emissions are taken from Global Carbon Project, while the data on governance indicators are collected from Worldwide Governance Indicators (WGI). The data on all the remaining variables including GDP, population, forest area as a proportion to total area, trade openness, and the shares of agricultural and industrial sectors in GDP are taken from World Development Indicators (WDI).

A variety of empirical modeling and estimation strategies is available, and each one has its specific merits. Generally, a pooled least square model in which all the fixed and random effects are absent yields poor results unless the sample size including the cross-sectional and time dimensions is small. Fixed and random effects models are aimed to account for observational heterogeneity. In case where cross-sectional units are chosen randomly, the heterogeneity can be attributed to random sampling, and in this case, random effects models are preferred. However, when cross-sectional units are chosen systematically, as we have done by including all the countries on which data are available, fixed effects models have more appeal because data are expected to vary across countries systematically.

In the presence of inertia in the dependent variable and autocorrelation in errors, the above models produce biased results both in small and large samples, and the preferred estimation technique is the generalized method of moments (GMM).

Other modeling strategies motivated by time-series econometrics are suitable when the number of periods is large, if not greater than the number of cross-section units. These options are not considered here because of the small period of analysis (23 years).

The possible presence of fixed and random effects in Eq. (4) is tested using the Wald and Lagrange multiplier (LM) tests, respectively. In case the presence of both cannot be rejected, the choice between the two types of effects can be made on the basis of the Hausman and Wu test. While the model containing fixed effects in the absence of autocorrelated errors can be estimated by OLS, the one containing random effects is to be estimated by iterative feasible generalized least squares method.

It is highly probable that CO₂ follows time inertia, and therefore, 1-year lagged CO₂ may be included on right side of the equation. This variable suffers from endogeneity problem if the country effects ${\lambda }_{i}$ are treated as random variables or the general random error term ${U}_{it}$ is autocorrelated. If this happens, the appropriate estimation technique will be generalized method of moments (GMM).

As usual, the choice of instruments for the GMM estimation is a tricky matter, and at least two considerations are important in this respect. First, there should be sufficient number of instruments to gain incremental precision as compared to two-stage least squares estimates. But the use of too many instruments can create an artificially strong relationship. The standard practice is to use time-specific instruments proposed in Arellano and Bond (1991). These instruments are derived on the moment conditions that lagged dependent variable with two or higher period lags are orthogonal to random errors in each period available for estimation.¹ Unlike the single time series data, panel data provide as many observations as the number of cross-sectional units (countries in our case) for each single period. Therefore, as proposed in Arellano and Bond (1991), it is possible to generalize quite a few instruments if the orthogonality conditions are applied period by period. Additional instruments can also be made available by using similar dynamic structure or simple static structure on the moment conditions on other variables appearing on right-hand side of the equation.

The second consideration in the choice of instrumental variables concerns the validity of the instruments, which is addressed by applying Sargan’s test using J-statistic (based on the original work of Sargan 1958). If the number of instruments is barely minimum (equal to the number of parameters), the sample moment conditions are exactly satisfied, and GMM collapses to the (simple) method of moments, which leaves nothing to test for the validity of moment conditions. The advantage of GMM lies in having excess instruments or over-identifying moment conditions in which case all the sample moment conditions cannot hold simultaneously, and their validity is tested in a statistical sense.

Results and discussion

Before estimating the proposed regression model, PCA is used to construct the governance index. The analysis reveals that the first principal component alone accounts for 94.35% common variation in the four governance indicators. Therefore, the construction of a single index appears sufficient, and the other orthogonal dimensions of data can be safely ignored.

Tables 1 and 2 show the correlation matrix and variance inflation factors (VIF), which indicate that despite high correlation in two pairs of independent variables, there is no serious problem of multicollinearity in data as indicated by VIFs. This is because of the overall large sample size (3680 observations) used in the estimation.

Table 1 Correlation matrix

Table 2 Variance inflation factors

Estimation of Eq. (8) is done in various rounds to ensure robustness. Initially, it is estimated with country and period fixed effects, and Wald’s test is applied to test two null hypotheses one by one; that is (a) country fixed effects are jointly equal to zero and (b) period fixed effects are jointly equal to zero. The model is then estimated with country and period random effects, and LM test is applied to test null hypotheses that (a) country random effects are equal to zero and (b) period random effects are equal to zero. The two regression results are presented in the second and third columns of Table 3. In both cases, while the absence of period (fixed or random) effects cannot be rejected, the absence of country (fixed or random) effects is rejected. This result rules out pooled least square estimation on account of significant country effects. The two models are then restricted by excluding the period (fixed and random) effects, and the results are shown as the country fixed effects model and country random effects model in the fourth and fifth columns of Table 3.

Table 3 Regression results based on fixed and random effects models (dependent variable is log per capita CO₂)

At the next step, the Hausman and Wu test is applied to make a choice between fixed and random effects models, and the test results support the former model against the latter one. In all the estimated equations significant, first-order autocorrelation in errors is found to be present, which calls for the inclusion of lagged dependent variable on the right side of the equation. Since country random effects are time-invariant, these are by construction correlated with lagged dependent variable (just as they are correlated with the current dependent variable) and the most appropriate estimation technique, in this case, is GMM. Regarding fixed effects estimation, the results are valid if the inclusion of lagged dependent variable removes autocorrelation from errors. The last two columns of Table 3 show that this is not the case, and significant autocorrelation persists and the recommended estimation technique is again GMM.

Table 4 presents the results of two rounds of GMM estimation based on Eq. (8) converted to the first difference form (in order to drop out the fixed or random country effects). GMM-I employs Arellano and Bond’s (1991) set of time-specific instruments (that is, a two-period lag of dependent variable with orthogonality condition applied period by period). In GMM-II, in addition to Arellano and Bond’s instruments, two-period lags of all the variables appearing on the right-hand side of the equation are also used with an orthogonality condition applied to the entire sample rather than on a period-by-period basis.

Table 4 Regression results based on GMM (dependent variable is log per capita CO₂)

A few observations from Tables 3 and 4 are notable. First, the results seem remarkably robust across the alternative estimation techniques as far as signs of regression coefficients are concerned. In all the eight estimates presented in the two tables, the signs of regression coefficients of the five key variables under focus (log per capita GDP, square of log per capita GDP, governance index, governance index interacting with log per capita GDP, and governance index interacting with the square of log per capita GDP) remain the same with no exception. The signs of regression coefficient of the control variables also remain the same with two exceptions with the signs of the share of the manufacturing sector in GDP. Both the exceptions occur in the two fixed effects models with lagged dependent variable, which are in any case not as reliable as the GMM estimates.

Second, the statistical significance of the parameter estimates is also reasonably consistent. As expected, the most significant results are obtained under GMM estimation in which all the parameter estimates are highly significant. On the other hand, the least number of significant parameter estimates are found in fixed effects model with country and period fixed effects but no lagged dependent variable, followed by fixed effects models with lagged dependent variable. The possible reasons could be the presence of a large number of fixed effects and additionally the presence of lagged dependent variables, which collectively suck most of the variation in the dependent variable, thereby leaving not enough variation to be explained by other variables. The trade openness variable appears significant only in GMM estimates.

In any case, the consistency of results means that we can focus on and discuss in detail the estimation results obtained from the most appropriate technique, GMM, which also produces the most significant relationship. Specifically, the following discussion is based on the results of GMM-I model, which employs time-specific instruments proposed by Arellano and Bond (1991).

Since the relationship of CO₂ emission with income and governance variables is non-linear, there is no sense in interpreting each regression coefficient in a typical way to mean partial effect. All we can say is that the signs and significance of parameter estimates make sense. For example, if the governance level is held constant at the lowest level (equal to zero), CO₂ emission is observed to increase with per capita income at a diminishing rate up to some point after which the relationship turns negative. It is possible to draw a similar conclusion at another level of governance above the minimum level.

Figure 1a, b displays sets of EKCs at the extreme levels of governance index (zero and one) and at all its nine-decile points. In the first figure, variables are measured in logs, while in the second one, they are measured in levels. While drawing these figures, all the variables other than per capita income and governance score are set equal to their sample mean values. In each figure, the highest (lowest) curve corresponds to the lowest (highest) governance score, which is equal to zero (one), and the inverse relationship between governance and CO₂ emission remains monotonic throughout the range of governance index. This observation affirms our first research question, that is, does good governance contribute to reducing pollution systematically?

Fig. 1

a, b EKCs in log per capita terms at alternative governance levels (higher curves correspond to lower governance scores)

However, the two figures also show that the shifts in EKC resulting from better governance levels are neither uniform nor proportional to the location of the curve with respect to per capita income. Thus, the answer to our second research question, that is, does good governance help reduce pollution at all income levels in a uniform way, is negative. The figures show that the gap between the EKCs representing any two governance scores widens as we move along the curve from left to right, implying that good governance brings more fruitful results in high-income countries as compared to low-income countries. Figure 2 further illustrates this point by drawing the relationship between log per capita income and the percentage improvement in per capita CO₂ when the governance scores increased from 20 to 80 percentile points. The curve is almost flat in the low-income range that prevails in low-income countries, or the low-middle-income countries lying in the low-income range where the predicted reduction in CO₂ emission is very low, around 5%. However, the percentage reduction in pollution becomes larger at higher income levels, reaching up to 50% at the highest income level in the given data.

Fig. 2

Percentage reduction in per capita CO₂ emission resulting from improvement in governance score from 20 to 80 percentile

Regarding the third research question, it is confirmed from Fig. 1a, b that the critical level of per capita income at which the relationship between income and pollution turns negative is inversely correlated with the governance index. This relationship is further illustrated through Fig. 3, which shows that the increase in governance index from 20 to 80 percentage points is associated with about 45% reduction in the critical level of per capita income. This is overwhelming evidence to support the proposition that better governance practices enable various countries to bring the turning point of EKC to lower levels of per capita income. Thus, if most countries follow good governance standards, the world can reverse the negative spillover of economic growth on the environment at a lower level of income than otherwise predicted.

Fig. 3

Relationship between governance score and turning point of EKC

The effects of control variables on pollution are found to be according to expectations. The first variable, the share of geographic area covered under forests, is negatively and significantly correlated with pollution. Each one-percentage point reduction in forest area can be associated with 0.19% increase in per capita CO₂ emission. This is an expected result because deforestation is known to be a reason for CO₂ emissions and related environmental consequences. The next control variable is trade openness, which is also found to be negatively and significantly correlated to pollution as each one-percentage point increase in trade openness is associated with 0.14% reduction in per capita CO₂ emission. The countries with higher trade volumes have to adopt relatively more efficient modern production techniques that tend to conserve energy use and adopt environmental protection measures that in turn result in lower pollution (Majeed et al. 2022; Tahir et al. 2021).

Since the regression model includes the shares of both manufacturing and agricultural sectors in the equation, an increase in any of these shares will be at the cost of the reduced share of the rest of the economy, which means the share of services sectors. The results show that per capita CO₂ emission is positively correlated with the share of manufacturing and negatively correlated with the share of the agricultural sector, and both relationships are statistically significant. This essentially means that compared to services, manufacturing activity is more pollution-intensive, whereas agricultural activity is less pollution-intensive. This result is also according to expectations because the production activity in the manufacturing sector is by and large more energy-intensive and that in the agricultural sector is less energy-intensive.

Although no empirical work is found that explores the role of governance in the context of EKC the way the present study has done, it is still possible to compare our results with some of the earlier findings. Laegreid and Povitkina (2018) observed that the moderating role of the control of corruption is crucial. Wawrzyniak and Doryń (2020) could not confirm the moderating role of the control of corruption and observed a significant reduction in CO₂ emissions in the countries with high government effectiveness in some but not all specifications of the model. Laegreid and Povitkina (2018) also found that the adverse effect of GDP per capita on CO₂ emissions is not profound in rich well-governed countries but the moderating effect of institutions is small. The present study, on the other hand, finds overwhelming support for the moderating role of governance because it allows all parameters of the EKC to vary across different levels of the governance index. It is also noted that this relationship remains robust across eight different modeling options for panel data regressions.

Conclusion and policy recommendations

Using a panel data set for 160 countries and 23 years, this study analyzes the role of institutions in determining how pollution can be related to the pace of economic activity. The study finds that the level of pollution can be systematically reduced by adopting good governance practices. However, returns to better governance in terms of a cleaner environment are higher at higher income levels, increasing exponentially with the increase in per capita incomes. Finally, it is observed that better governance is effective in bringing down the critical level of per capita income at which the relationship between income and pollution turns negative.

The following conclusions are drawn from the above results. In low-income countries, better governance seems to yield negligible environmental outcomes not only because the direct impact of governance on pollution is very small but also because their per capita income levels remain well below the critical level at which the relationship between income and pollution turns negative even after governance score is increased up to the maximum possible level. Good governance remains ineffective largely because the focus of institutions in these countries is not as much on environment as on the other pressing issues like infrastructure, human development, and poverty. Good governance in develo** countries will be fruitful only when the desired regulations are in place.

Second, although rising per capita income in low-middle-income countries is expected to result in increased pollution in the coming years, this undesired consequence of economic growth can be counterbalanced by investment in institutions to yield a better state of governance. In such countries, the direct impact of better governance on the environment is small, but the main advantage comes from the role of governance in reversing the relationship between income and pollution.

Finally, in upper-middle and high-income countries, good governance is most productive in controlling pollution. Although most of these countries also score well on governance, yet there are exceptions. The countries that have gained the status of high-income countries in recent years because of the rental value of natural resources are not developed in the true sense and there is scope for improvements in institutions. Investment in institutions in such countries is expected to yield maximum dividends for a clean environment.

The study leads to a number of policy guidelines. Since dividends of good governance on the environment are negligible in poor countries where the environment is not prioritized in public policies, it is important to persuade these countries to reconsider their priorities with a long-term perspective. This persuasion can be in the form of aid packages for governance reforms that are made conditional upon the adoption and enforcement of specific environmental regulations. In middle-income countries, the role of good governance in improving the environment is found to be more fruitful. In these countries, it is pertinent to weigh the costs and benefits of alternative strategies for addressing the issue of environmental degradation. These strategies include, in particular, investment in environmental protection technology and investment in environment protection governance. Assessing the relative strengths of the two strategies would be useful for working out the optimal policy mix.

The present study provides important insights into the relevance of governance in sha** the growth-pollution relationship. However, the study has certain limitations. The study combines four indicators of governance into one index, whereas the role of each indicator may not necessarily be the same. In addition, all 160 countries are combined in a single model, whereas the shape of EKC may differ between developed and develo** countries. Future research may address along with these issues. With COVID-19 almost over, more usable data will hopefully be available to estimate the model using panel time-series econometric techniques.

Appendix

To draw implications of governance for the shape and turning point of the EKC, consider the first and second derivatives of the expected value of log CO₂ with respect to log Y:

$$\frac{\partial \left(\mathrm{logCO}2\right)}{\partial \mathrm{log}Y}={\beta }_{0}+{\beta }_{1} G+2 \left({\delta }_{0}+{\delta }_{1} G\right)\mathrm{log}Y$$

(A1)

$$\frac{{\partial }^{2}\left(\mathrm{logCO}2\right)}{\partial {\left(\mathrm{log}Y\right)}^{2}}=2{\delta }_{0}+2{\delta }_{1} G$$

(A2)

The turning point of EKC, given below, is the level of per capita income at which the first derivative is equal to zero.

$$\mathrm{log}{Y}^{*}=-\frac{1}{2}\frac{{\beta }_{0}+{\beta }_{1} G}{{\delta }_{0}+{\delta }_{1} G}$$

(A3)

Finally, the turning point responds to changes in the state of governance as follows:

$$\frac{\partial \mathrm{log}{Y}^{*}}{\partial G}=\frac{1}{2}\frac{{\beta }_{0}{\delta }_{1}-{\delta }_{0}{\beta }_{1}}{{\left({\delta }_{0}+{\delta }_{1} G\right)}^{2}}$$

(A4)

It can be seen from Eqs. (A1) to (A4) in the Appendix that the location, slope, and curvature of the EKC all depend on the state of governance. The EKC hypothesis will prevail if ${\beta }_{0}+{\beta }_{1} G>0$ and ${\delta }_{0}+{\delta }_{1} G<0$ at all levels of governance indicator. Furthermore, the turning point in EKC also depends on the state of governance and in what direction it varies with the improvement in governance depends on the signs and magnitudes of the four parameters appearing in Eq. (4). Based on discussion at the beginning of this section, the signs of various parameters are expected to be as follows:

$${\alpha }_{0}<0, {\beta }_{0}>0, {\beta }_{1}>0, {\delta }_{0}<0, {\delta }_{1}<0$$