Environmental Kuznets Curve and Sustainable Development in OECD Countries: A Panel Data Analysis

The study investigated the hypothesis of the Environmental Kuznets’ Curve (EKC) in the selected 24 OECD countries using annual panel data from 1971 to 1916. The cointegration analysis of the variables found a cointegration among GDP, square of GDP, energy use, each in per capita terms and CO₂ emissions. The Fully Modified Ordinary Least Square model results support the EKC hypothesis. Furthermore, the study also found a positive association between CO2 emissions and energy use per capita in the samples of 24 OECD countries. Therefore, for sustainable development, OECD countries need to aim for higher economic growth and development, improve the efficiency of energy production, usage, and consumption, and replace non-renewable energy with renewable energy sources. Alongside this, OECD nations should also have effective environmental regulations and efficient incentive systems and deploy sustainable and green technologies to lower CO₂ emissions.

Introduction

Globally, pollution in its many forms—primarily from human economic activity—is becoming a greater danger to sustainable development. The rising emissions of greenhouse gases, including CO₂, are considered to be the principal cause of global warming and climate change. This is a major contributor to GHG emissions and is likely to cause significant disruptions to the climate, which will harm food production and, consequently, food security, particularly in less developed nations where agriculture is primarily dependent on rainfall due to an increase in the frequency of sparse and irregular rainfall.

The EKC hypothesis postulates that environmental degradation gradually improves at a certain point of economic growth but initially worsens as economic development happens. It bears the name Kuznets (1955) after the economist postulated such a connection between economic growth and income disparity. The EKC hypothesis states that there is a non-linear linkage between environmental quality and economic growth, which is commonly presented as an inverted U-shaped curve. Factors such as urbanisation, industrialisation, and increased use of natural resources are thought to lead to environmental degradation, which first increases as a country’s economy grows. After a threshold level of economic growth, it takes a downturn, resulting in an inverted U-shaped curve.

In this stage, resource depletion and increased pollution may result from prioritizing economic development over environmental concerns.

However, the theory predicts that environmental deterioration will start to decline as a nation reaches a particular level of prosperity or development. Numerous variables, including increased environmental awareness and regulation, technological innovation, and structural changes, might be blamed for this reduction.

The Environmental Kuznet’s Curve (EKC) theory attempts to explain why environmental quality will initially deteriorate with a rise in income and then improve beyond a certain income level. It has drawn the attention of researchers worldwide and suggests that the quality of a nation’s environment and economic growth have an inverse U-shaped relationship.

This change in the relationship between income and environmental degradation in the latter phases of economic development is explained by the theory that higher economic growth will raise environmental awareness and regulation. It will also promote increased spending on environmental preservation and increase the accessibility and adoption of clean technologies. According to Arrowet al. (1995), economies also shift from dirty rural to polluting industrial to, ultimately, reasonably clean service economies. Because of all of this, the ecosystem will not deteriorate as much.

Thus, there is a solid theoretical foundation for the inverted U-shaped relationships that exist between economic progress and the environment. Behavioral and structural factors promote this relationship. One behavioural aspect driving the demand for environmental services is income elasticity, which raises people’s willingness to pay for improved environmental services (Lekakis & Kousis, 2001; McConnell, 1997; Roca, 2003). The degree of economic activity, the output’s sectoral structure, and technological innovation are a few examples of structural pressures (Grossman & Krueger, 1995; Kaufmannet al., 1998). Factors supporting this association include patterns of international trade, industrial operations flow, family wealth distribution, and demographic concerns (Heenrinket al., 2001; Heil & Selden, 2001; Magnani, 2000; Shi, 2003).

The EKC hypothesis raises the possibility that, following a slow increase and peak in environmental degradation, economic growth and prosperity will continue sustainably. In other words, it makes the implausible prediction that increased income alone will create forces that will address environmental degradation, given the mounting evidence of global warming, consequent climate change, and the growing effects these changes are having on the global community. Raising income on its own, however, might not be enough; to stop environmental damage and advance sustainable development, it must be combined with other measures.

Many attempts have been made in the past to test the EKC hypothesis with panel data as well as time-series data, but the majority of these have methodological flaws. This is particularly true when using panel data for investigation, which frequently has cross-sectional dependence and is disregarded by researchers. Because they use CO₂ emissions per capita rather than CO₂ emissions as the dependent variable, several of these studies also fall short of conventional inquiry and, as a result, lose some of their credibility. Since environmental pollution is a stock rather than a flow variable, substantial environmental degradation is not always present in a nation or region with high CO₂ emissions per person.

Therefore, this research investigates the environmental Kuznet’s curve hypothesis and its implications for sustainable development in OECD nations. It will use suitable econometric models to resolve the problem of cross-sectional dependence present in the panel data for reliably estimating the parameters of the econometric model.

Materials and Methods

In this study, the four variables used in log forms were the log of CO₂ emissions (co2), the log of per capita real GDP (gdp), the square of the log of per capita real GDP (gdp²), and the log of per capita energy use (eu). The panel data for 24 OECD countries was balanced and collected from the World Bank Development Indicators between 1971 and 2016.

The econometric model to be estimated is specified below:

(1) $$\begin{array}{r}co{2}_{it}={\alpha }_{0i}+{\alpha }_{1}gd{p}_{it}+{\alpha }_{2}gd{p}_{it}^2+{\alpha }_{3}e{u}_{it}+u_{it}\end{array}$$

α₀, α₁, α₂, α₃ are four parameters and u_it is the error term in the above equation.

The following is our expectation for the signs of the three parameters, β1, β2, and β3: α₁ $>$ 0, α₂ $<$ 0 and α₃ $>$ 0.

The EKC hypothesis will be validated if we find values of coefficients as, α₁ $>$ 0 and α₂ $<$ 0 while α₃ $>$ 0. This will also lead us to conclude that energy use per capita is positively related to CO₂ emissions.

The data for each variable was gathered from the World Bank’s World Development Indicators, which are available on their website ( www.worldbank.org). Below are the variables included, along with an explanation of each (Table I).

Table I. Description of the Variables

Variables	Description	Variable	Description
CO2	CO2 emissions in Kt.	co2	Natural log of CO2
GDPPC	GDP per capita (at a constant price of year 2010 USD)	gdp	Natural log of GDP
(GDPPC)2	Square of GDPPC	(gdp)2	Square of gdp
EUPC	Energy use Per Capita (kg of oil equivalent per capita)	eu	Energy use per capita

Literature Review

There is much empirical literature on studying the environmental Kuznet’s curve (EKC). Plenty of empirical literature has attempted to test the EKC hypothesis by analysing time series and panel data covering different countries and regions.

Stern (1998) examined the advances in our understanding of the EKC, criticizing some of the scientific findings and the arguments put up by the policy literature to support them. He discovered that the accuracy of the econometric techniques used to analyse the correlations had increased. He came to the conclusion that there has not been any systematic testing of the numerous related hypotheses, and the empirical breakdowns of the EKC into direct or underlying causes are non-systematic and have a limited scope.

Between 1990 and 2011, Apergis and Ozturk (2015) looked into the EKC theory using the data on 14 Asian nations. They applied the GMM technique to analyse panel data. They found evidence in favour of an inverted U-shaped relationship between emissions and per capita income, supporting the EKC theory. Using panel data from 25 OECD countries, Mehdiet al. (2016) analysed the relationships between GDP and CO₂ emissions per capita as well as non-renewable and renewable energy use between 1980 and 2010. Using both FMOLS and DOLS estimators, they found evidence supporting the EKC theory in the OECD countries.

Alam (2019) investigated how economic development affected India’s environmental quality. According to his research, CO₂ emissions fall when per capita income rises as long as industrial value-added growth is stagnant. Sisay and Balázs (2020) examined the EKC theory for 12 East African countries from 1990 to 2013 using the Pooled Mean Group (PMG) technique. They confirmed the EKC theory by discovering a bell-shaped link between CO₂ emissions and per capita income.

In their research, Onganet al. (2021) retested the environmental Kuznet’s curve theory for the United States. Two distinct time series showing increases and decreases were created from the per-capita income series, and only one series with income gains was used. In their cointegration technique, they applied the ARDL between 1990M1 and 2019M7. They argued that the empirical results of the undecomposed model are inconsistent with those of the decomposed model. They concluded that whereas the decomposed model substantially supported the EKC hypothesis in the US, the undecomposed model did not.

Holtz-Eakin and Selden (1995) looked at 130 nations between 1951 and 1986. They discovered proof that the EKC theory is correct. Coleet al. (1997) looked at seven different world regions between 1960 and 1991 and discovered evidence favouring the EKC. Schmalenseeet al. (1998) looked into 47 different countries between 1950 and 1990. Their research validated the EKC theory. Schmalenseeet al. (1998) examined data spanning 47 nations between 1950 and 1990. They discovered proof in favour of the EKC theory.

Agras and Chapman (1999) analysed the data of 34 Countries from 1971 to 1989. They discovered evidence in favour of the EKC. From 1971 to 1996, Galeotti and Lanza (1999) examined data from 110 different countries and discovered that the EKC theory was correct. Wanget al. (2011) analysed the data on Chinese provinces and found evidence in support of the EKC. Data from 148 countries were examined by Neumayer (2002) between 1960 and 1988. The evidence supported the EKC. Yanget al. (2015) examined information spanning 67 countries between 1971 and 2010. Their study also backed up the EKC. Other investigations by Halkos and Tsionas (2001), Richmond and Kaufmann (2006), Azomahouet al. (2006), Aslanidis and Iranzo (2009), Akbostanciet al. (2009), Villanthenkodathet al. (2021), Aruga (2019), and Ganda (2023) do not support the EKC theory.

Most prior empirical studies found evidence supporting the inverted U-shaped association between GDP per capita and CO₂ emissions, which are used to measure a nation’s environmental state. However, the review of the empirical literature above shows that researchers continue to differ on the nature of the relationships between income growth and CO₂ emissions as a measure of environmental degradation.

Results

Unit Root Test

We used four types of unit root tests (Breitung, 2000; Imet al., 2003; Levinet al., 2002); Fisher tests using the Augmented Dickey-Fuller and Phillips-Peron tests (Choi, 2001; Maddala & Wu, 1999) for analysing the properties of the included variables. The results of these unit root tests are presented in the table below. The results show that variables co2, gdp, gdp², and eu are I(0) in the first difference form, but I(0) in the level. Hence, we find that none of these variables in the level form is integrated of an order higher than 1 (Table II).

Table II. Panel Unit Root Test

Level
Method	co2	gdp	(gdp)2	eu
LLC	−5.3506* (0.0000)	−5.2738* (0.0000)	−4.6567* (0.0000)	−5.0567* (0.0000)
IPS	−0.9796 (0.1636)	1.6635 (0.9519)	2.2235 (0.9869)	−1.2209 (0.1111)
ADF-FS	65.3286** (0.0487)	44.8391 (0.6031)	38.8406 (0.8245)	60.4684 (0.1069)
PP-FS	86.0468* (0.0006)	84.2362* (0.0010)	72.9686 ** (0.0116)	83.3598* (0.0012)
First Difference
Method	D(co2)	D(gdp)	D(gdp2 )	D(eu)
LLC	−15.9959* (0.0000)	−13.7614* (0.0000)	−13.7849* (0.0000)	−11.8563* (0.0000)
IPS	−17.4289* (0.0000)	−14.5552* (0.0000)	−14.4756* (0.0000)	−16.9424* (0.0000)
ADF-FS	374.6220* (0.0000)	301.8750* (0.0000)	299.9950* (0.0000)	362.2020* (0.0000)
PP-FS	644.2030* (0.0000)	358.7820* (0.0000)	359.2170* (0.0000)	618.4920* (0.0000)

The first-generation unit root tests are no longer viable in the presence of cross-sectional dependence. Previous empirical investigations that have analysed panel data have ignored this issue and simply employed Pedroni (2001)’s panel co-integration test. The cross-sectional dependence test must be employed in addition to the second-generation unit root and cointegration tests to confirm the findings of the first-generation unit root tests. At the 1% significance level, the four cross-sectional dependence tests reject the null hypothesis that no cross-sectional dependency exists in any of the four variables. The table below illustrates that each of the four-panel variables is cross-sectionally dependent, necessitating the application of the second-generation unit root and co-integration tests (see Table III).

Table III. Test of Cross-sectional Dependence

Test	co2	gdp	gdp2	eu
Breusch-Pagan LM	5677.9110* (0.0000)	11249.5000* (0.0000)	11237.4600* (0.0000)	6142.0790* (0.0000)
Pesaran scaled LM	229.9206* (0.0000)	467.0632* (0.0000)	466.5506* (0.0000)	249.6769* (0.0000)
Bias-corrected scaled LM	229.6479* (0.0000)	466.7904* (0.0000)	466.2778* (0.0000)	249.4042* (0.0000)
Pesaran CD	26.4678* (0.0000)	105.9595* (0.0000)	105.8990* (0.0000)	51.5280* (0.0000)

We used the Pesaran (2007) CIPS stationary test to account for the cross-sectional dependence in a heterogeneous panel. The CIPS test does not reject the alternative hypothesis of heterogeneity and stationarity as it rejects the null that each variable is homogeneous and non-stationary.

We then used Pesaran (2007)’s CADF test, a t-test for the panel unit root with cross-sectional dependence in heterogeneous panels. Under the null hypothesis for the CADF test, all series are assumed to be non-stationary.

The Pesaran (2007) CIPS test allows an unobserved common factor with variable factor loadings in the data, and the autoregressive coefficient of the Dickey-Fuller (DF) regression may be heterogeneous. The test statistic is obtained using DF or ADF regressions that are particular to each panel member. These regressions employ the CADF regression model and the cross-sectional averages of both independent and dependent variables. The test protocols described by Imet al. (2003) are used to average the data particular to each group. There is a non-standard distribution in the test statistic under the null.

The second-generation unit root test results of each variable are shown in the following table. The CIPS and CADF test results show that co2, gdp, gdp2, and eu are all I(1) series (see Table IV).

Table IV. Results of the Second-generation Unit-root

Variable	CIPS-Test	t-bar value of CADF-Test
co2	−1.881	−1.604
gdp	−2.036	−1.866
gdp2	−2.025	−1.887
eu	−2.282	−1.626
D(co2)	–5.983*	–3.470*
D(gdp)	–4.465*	–3.089*
D(gdp2)	–4.539*	–3.093*
D(eu)	–5.914*	–3.548*

Cointegration Test

Pedroni (1999, 2001)’s test of cointegration uses four different panel statistics (v-Statistic, PP-Statistic, rho-Statistic, and ADF-Statistic) and three different group statistics (PP-Statistic, rho-Statistic, and ADF-Statistic). Each panel statistic rejects the null hypothesis of no cointegration in the variables. Additionally, at a 1% significance, one of the three group statistics rejects the null hypothesis (Table V).

Table V. Results of the Panel Cointegration Test Applied on Variables co2, gdp, gdp² and eu

Panel statistic	Value	Group statistic	Value
v-Statistic	1.887232** (0.0296)
rho-Statistic	−2.345281* (0.0095)	rho-Statistic	−0.643501 (0.2599)
ADF-Statistic	−1.966208** (0.0246)	ADF-Statistic	−0.728868 (0.2330)
PP-Statistic	−3.649726* (0.0001)	PP-Statistic	−2.845690* (0.0022)

However, we further employ the second cointegration test (Westerlund, 2007) due to the problem of cross-sectional dependence. Westerlund suggests the second-generation cointegration test to account for cross-sectional dependence in the variables when the panel data demonstrate such dependence.

This cointegration test aims to ascertain if each panel variable is error-correcting. Four different types of statistics, G_t, G_a, P_t, and P_a are used under the null of no cointegration. While the P_t and P_a test statistics provide evidence for cointegration throughout the panel, the G_t and G_a test statistics reject H₀, showing cointegration in at least one cross-sectional unit.

The statistical significance of the Z-values for the statistic G_t, G_a, P_t, and P_a is 5%, 1%, 10%, and 1%, respectively. Thus, these test statistics reject the null hypothesis of no cointegration (Table VI).

Table VI. Results of the Westerlund Cointegration Test

Statistic	Value	Z-value	P-value
Gt	−2.558	−1.685	0.046**
Ga	−17.764	−4.734	0.000*
Pt	−10.801	−1.317	0.094***
Pa	−16.537	−6.781	0.000*

Westerlund’s cointegration test suggests that the four variables co2, gdp, gdp² and eu are cointegrated.

Cointegrating Regression

Since the four variables are found to be cointegrated, we estimate the cointegrating regression as specified in (1) with the panel Fully Modified Ordinary Least Squares model taking co2 as the dependent variable and gdp, gdp², and eu as explanatory variables. The table below reports the results of the panel FMOLS. As anticipated, the model parameters with estimated values of α₁ = 0.8156, α₂ = −0.0933, and α₃ = 1.6271 are all statistically significant at 1% and α₁ > 0, α₂ < 0, and α₃ > 0 (Table VII).

Table VII. Pooled Estimation of the Panel Fully Modified Least Squares (FMOLS): Dependent Variable: co2

Variables	Coefficients	t-Statistic	Prob.
gdp	0.8156	93.7606*	0.0000
gdp2	−0.0933	−230.1115*	0.0000
eu	1.6271	165.1307*	0.0000
R-squared	0.1832
Adjusted R-squared	0.1816
Jarque-Bera	3.1224 (0.2098)

Discussion

The results of the OECD countries that were sampled corroborate the EKC hypothesis. The findings also show a positive correlation between energy usage and CO₂ emissions in the sampled OECD countries. Thus, the study finds evidence supporting energy consumption-led environmental degradation in the OECD countries.

Consistent with several prior empirical investigations (Galeottiet al., 2006; Mehdiet al., 2016; Moomaw & Unruh, 1997; Pauli, 2003), our panel data analysis of OECD countries furnishes evidence that supports the EKC hypothesis among the OECD nations. However, the inverted U-shaped or bell-shaped relationship between income per capita and CO₂ emissions is conditional upon keeping the per capita energy consumption level constant. Therefore, to reduce CO₂ emissions and improve environmental quality, relying solely on economic expansion and raising per capita income will likely be fruitless in the fight against environmental deterioration.

Thus, to achieve sustainable development, OECD nations should prioritize efficient energy production, usage, and consumption and implement economic policies that raise income and spur economic growth. Additionally, to combat environmental degradation, the OECD countries must either directly reduce CO2 emissions by increasing energy efficiency, reducing the amount of non-renewable energy, particularly fossil fuels, and increasing the amount of clean energy used overall, or indirectly through the implementation of appropriate environmental regulations, efficient incentive programs, and the adoption of sustainable or green technology to realize an inverted U-shaped environmental Kuznet’s curve and, consequently, sustainable development.

Conclusion

We analysed the balanced panel data of 24 OECD countries between 1971 and 2016. The four variables included in the study and expressed in logarithms, were CO₂ emissions, GDP per capita, GDP per capita squared, and energy use per capita. For every panel variable, we conducted a cross-sectional dependence test. Panel variables were determined to be cross-sectionally dependent by the cross-sectional dependence test. As a result, we used both the first-generation unit root test, which ignored the variables’ cross-sectional dependence, and the second-generation unit root test, which took it into consideration. The panel unit root test indicates that the variables are either integrated of order 0 or 1.

The first-generation cointegration test ignores cross-sectional dependency among the variables, while the second-generation cointegration test takes cross-sectional dependence in the panel data into consideration. These tests’ outcomes support the reliability of the regression model we employed to examine the connections between the four cointegrated variables.

Ceteris paribus, the EKC theory predicts a bell-shaped or an inverted U-shaped correlation between income per capita and CO₂ emissions, which is supported by the panel Fully Modified OLS model cointegrating regression results. Furthermore, energy use per capita and CO₂ emissions have a strong positive correlation. Thus, the OECD countries should implement economic policies to raise per capita income and economic growth and enhance energy consumption efficiency to reduce environmental degradation and achieve sustainable development. To lower CO₂ emissions and slow down environmental deterioration, OECD nations should also concentrate on improving energy production, use, and consumption efficiency. This can be achieved by replacing non-renewable energy with renewable energy. OECD nations should also implement environmental regulations, strong incentive programs, and other measures to minimize CO₂ emissions to follow the path of sustainable growth and development.