The Stability of the Economic Growth–Energy Consumption–Pollution Nexus in the European Union

Abstract

Considering the environmental objectives assumed by the European Commission through Cohesion Policy, the main aim of this study is to check the stability of the relationship between economic growth, GHG emissions, and energy consumption, also known as the 3E relationship, which also allows us to recreate EU policy recommendations. The analysis of EU member states using panel data and time series (1990–2023) showed significant shifts in the relationship between the three variables, especially due to the COVID-19 pandemic. Consequently, many countries successfully decoupled emissions from economic growth. Nevertheless, the persistent carbon-intensive energy use patterns in most EU countries remain a major obstacle, emphasizing the need for increased efforts in cleaner energy technologies and sustainable economic transitions.

1. Introduction

The intricate relationship between the economy, energy, and the environment (“3E”) forms a critical nexus that underpins sustainable development. Understanding the dynamics of this “3E nexus” is essential because changes in any one of these domains invariably impact the others. This study aims to explore these interdependencies, highlighting the challenges and opportunities that arise at their intersection.

Historically, economic growth has been closely linked to increasing energy consumption. As economies develop and industrialize, the demand for energy to power factories, transportation, and households rises significantly. However, an overreliance on fossil fuels to meet this growing energy demand presents considerable challenges [1].

One key challenge is the potential for resource depletion associated with fossil fuels, which can ultimately hinder future economic growth. Furthermore, the combustion of these fuels releases greenhouse gases, a primary driver of climate change and air pollution. These environmental consequences can have severe adverse effects on ecosystems, human health, and agricultural productivity.

The environmental impacts extend beyond emissions. The extraction and utilization of natural resources for energy production can lead to significant environmental degradation, including habitat destruction and water pollution. Addressing these environmental concerns often necessitates government regulations.

Government regulations, such as emissions standards and renewable energy mandates, play a crucial role in influencing economic activities related to energy. While sometimes perceived as constraints, these regulations can also stimulate innovation and investment in cleaner technologies, potentially unlocking new economic opportunities and fostering a green economy.

Ultimately, achieving sustainable development requires a balanced approach that integrates economic growth, social equity, and environmental protection. A key strategy in this pursuit is the transition to cleaner energy sources, such as renewable energy. The composition of a country’s energy mix significantly shapes its environmental footprint and its vulnerability to economic shocks [2].

Improving energy efficiency offers another crucial pathway towards sustainability. By reducing overall energy consumption, efficiency measures can mitigate environmental impacts and simultaneously enhance economic competitiveness [3]. Effective policies and robust international cooperation are indispensable for navigating the complex challenges and capitalizing on the opportunities presented by the interconnected economy, energy, and environment.

The significance of this topic is underscored by the United Nations’ seventh Sustainable Development Goal, which aims to ensure universal access to safe and clean energy. This objective is closely linked to SDG 13, focusing on urgent climate action and equitable energy access for all [4]. Consequently, there is growing interest among politicians, researchers, and academics in exploring the intricate relationships between these three key variables.

Recent years have witnessed a decoupling of global carbon dioxide emissions from economic growth. While the world’s GDP has continued to expand, CO₂ emissions have stagnated. This trend is primarily attributed to a shift towards cleaner energy sources, such as renewable energy and natural gas, in advanced economies. Even rapidly developing nations like India and China have managed to lower their per capita emissions while experiencing economic growth. However, this relative decoupling is not enough to achieve the ambitious goal of limiting global warming to 2 degrees Celsius, as outlined in the Paris Agreement [5]. Substantial reductions in absolute CO₂ emissions are still necessary.

Overall, this research contributes to the existing body of knowledge in two primary areas. First, it examines the interconnectedness of energy, the economy, and the environment, often referred to as the “3E” relationship. While previous studies have primarily explored the pairwise relationships between these elements, recent research suggests that a more comprehensive analysis of the triple relationship is necessary to avoid biased estimations.

Second, this study investigates the stability of the 3E relationship over time. Given the historical events that have significantly impacted economic trends, it is essential to re-evaluate this relationship by accounting for potential structural breaks. The framework developed by Karavias and Tzavalis [6] on panel data and by Bai and Perron [7] on time series is particularly suitable for this purpose, as it allows for the identification and estimation of multiple breaks in the data.

The rest of this research is constructed around a logical flow. The second section provides insights into the main literature. The next section outlines the data, methodology, and analytical framework. The next two parts of this paper consider the empirical findings regarding unit root testing and the 3E relationship. Finally, the last section of this paper concludes and provides potential areas for future research.

2. Literature Review

Recent research shows that growth and carbon emissions might be unlinked, potentially marking a new era where development is not synonymous with environmental harm. Critics argue that some developed nations have decreased their domestic carbon emissions by outsourcing pollution-intensive industries to other countries. This means that the carbon footprint of goods often lies with the country where they are manufactured. For developing nations, this is especially significant due to their rapidly increasing carbon emissions, which contrasts with the declining emissions of developed countries [8]. Additionally, global climate action is undermined if developed countries merely relocate their polluting industries to developing nations, as this does not address the overall problem. Moreover, Cohen et al. found that, as countries become wealthier, their carbon emissions tend to grow at a slower rate, especially when measured based on where goods are produced rather than consumed [9].

Research on this topic has expanded significantly in the last few years. Early studies primarily examined the interconnectedness of environmental impacts, economic growth, and energy use, often represented by CO₂ emissions, output, and energy consumption. The body of literature is diverse, covering a lot of countries, periods, and analytical methods. Notable contributions to this field include the works of Payne, which found no evidence of consensus between states [10], as did Narayan and Popp for 93 countries, but the last study proved that, in G6 countries, energy consumption indirectly causes growth [11]. No connection between energy use and economic growth was supported by Narayan for 90 developing states, but a causal connection from energy use per capita to output per capita in 32 lower-middle-income nations was found [12]. Starting from a sample of 164 countries during 1822–2018, Freire-González et al. provided interesting findings related to CO₂-growth decoupling: a positive association, weakening over time, with few high-income countries reversing the association and 49 states experiencing decoupling (economic growth occurs without an increase in emissions) [13]. Most of the nations located in America, Africa, and Asian have not decoupled, while this issue has been solved by a lot of states in Europe and Oceania.

While these studies have garnered significant attention, several researchers have questioned the adequacy of the bivariate model, suggesting that additional variables may be necessary to mitigate potential biases arising from model misspecification. Notably, studies made by Apergis et al., Farhani et al., Zhang et al., Munir et al., and others have echoed this sentiment [14,15,16,17]. Employing a panel error correction model for 19 developed and developing nations from 1984 to 2007, Apergis et al. obtained a significant inverse long-term connection in the nuclear energy-pollution link [14]. Conversely, renewable energy consumption is directly correlated with emissions in the long term. Short-term analysis indicates that nuclear energy plays a crucial role in curbing CO₂ emissions, while renewable energy’s impact on emissions reduction is limited. This discrepancy may be attributed to the current state of storage technology, which struggles to address the type of renewable energy sources. Consequently, electricity producers may turn to more polluting energy sources during peak demand periods [18]. Panel cointegration methods were also employed by Zhang et al. to explore the long-term relationship between carbon emissions and five key factors (per capita GDP, primary energy consumption, international trade, fossil energy share, and quadratic per capita GDP) across 50 developing countries from 1995 to 2017 [17]. Our empirical results confirm the existence of a long-term equilibrium relationship. The fully-modified OLS (FMOLS) regression coefficients indicate that (a) the inverted U-shaped curve hypothesized by the Environmental Kuznets Curve (known as EKC) is evident in states like including Mexico, Croatia, Kazakhstan, Iran, Algeria, Indonesia, and Thailand; (b) energy consumption has a statistically significant and positive impact on carbon emissions; (c) international trade exhibits a negative effect in developing nations with trade surpluses; and (d) the impact of fossil energy share is mixed. The CO₂ emissions–energy use-growth nexus was revisited by Munir et al. (2020) across the five major ASEAN-5 countries during the period 1980–2016 [16]. The authors found unidirectional Granger causality running from GDP to CO₂ in Malaysia, the Philippines, Singapore, and Thailand. In Indonesia, Malaysia, and Thailand, GDP causes energy use unidirectionally. Singapore exhibits unidirectional causality from energy consumption to GDP, while the Philippines demonstrates bidirectional causality between GDP and energy consumption. Furthermore, the results provided support for the EKC hypothesis in the ASEAN-5 region.

Besides the panel data approach for a group of countries, there are also studies for a specific country, like Tunisia, for which Farhani et al. examined the dynamic interplay between carbon dioxide emissions, economic output, energy consumption, and trade from 1971 to 2008. Utilizing the bounds testing approach to cointegration and the ARDL methodology, the authors identified two long-term causal relationships among these variables. Additionally, the short-term analysis reveals three unidirectional Granger causality relationships, flowing from economic output, its square, and energy consumption to carbon dioxide emissions [15]. In recent research, Rehman et al. examined the impact of urbanization, energy use, fossil fuel consumption, per capita GDP growth, and CO₂ emissions on China’s economic growth. Their analysis, which employed unit root tests and explored asymmetric short- and long-run effects, highlighted the significance of these variables [19]. These studies collectively suggested that a tripartite analysis encompassing energy, the economy, and the environment can uncover biases present in earlier bivariate models. Moreover, Ehigiamusoe et al. specifically demonstrated that the estimated elasticities derived from the tripartite model differ from those obtained using bivariate specifications. These findings underscore the value of adopting a three-dimensional framework, encompassing energy, the economy, and the environment, as a foundational approach for such investigations [20].

The majority of studies on this topic have assumed a constant relationship between the variables, either in a bivariate or tripartite model. Several studies, including those by Balcilar et al., Cai et al., and Churchill et al., have questioned this assumption [21,22,23]. These researchers identified structural breaks in the trends of the variables within the tripartite relationship. Given these findings, it is advisable to re-examine the tripartite relationship by incorporating multiple structural breaks, with particular emphasis on the potential influence of the Great Recession. The Great Recession of 2008 triggered a series of financial and economic disruptions that significantly impacted key macroeconomic indicators worldwide. Numerous studies have explored the causes and consequences of the Great Recession. As the tripartite relationship may have been influenced by the Great Recession, it is crucial to assess the extent of its impact. It is essential to differentiate between cyclical fluctuations and fundamental changes in trends. Additionally, it is important to distinguish between decoupling, where resource growth rates lag behind economic growth, and absolute reductions in resource consumption. Decoupling occurs when the growth rate of environmental indicators, such as resource consumption or environmental impact, is slower than the growth rate of economic indicators like GDP. While decoupling is relatively common, it does not automatically result in an absolute decline in resource consumption. An absolute reduction in resource consumption will only materialize if resource productivity increases at a faster rate than the economy [4].

There are studies made for panels of countries. The dynamic interplay between energy consumption, CO₂, and output in the G7 economies since 1960 has been shaped by variations in economic expansion, the implementation of different regulations, and advancements in technology, according to findings from a time-varying vector autoregressive model [24]. Countries with underlying policy frameworks that more strongly support renewable energy and climate change mitigation efforts tend to exhibit greater decoupling between trend emissions and trend GDP, regardless of whether emissions are measured on a production or consumption basis. For 31 countries in the period 1974–2018, the study by González-Álvarez and Montañés demonstrated that the Great Recession was a key factor in causing structural breaks within the relationship between CO₂ emissions and economic growth [4]. Following the incorporation of these breaks into their model, they found that many nations, with a stronger trend observed in developed economies, exhibited a decoupling of CO₂ emissions from economic growth.

There are also papers that focus on a single country. Multiple studies have investigated decoupling in the EU countries. However, their findings have been inconsistent, with evidence of both strong and weak decoupling. In some cases, researchers have identified coupling during specific time periods. For example, Roinioti and Koroneos showed that Greece experienced a notable decline in CO₂ emissions from energy use between 2003 and 2013, coinciding with a reduction in energy consumption, particularly during the economic recession [25]. Moreover, Kriström and Lundgren analyzed CO₂ emissions in Sweden from 1900 to 2010, identifying changes in their patterns over time due to the implementation of policy proposals, such as an information package, pollution taxes, and various subsidies [26]. There are other studies for non-EU states. For example, Rehman et al. demonstrated that industrialization positively impacts CO₂ emissions in Pakistan [19], while Bekun and Agboola employed Maki co-integration to establish a long-run equilibrium relationship between electricity consumption, GDP, and CO₂ emissions in Nigeria [27]. Similar findings were reported by Samu et al. for Zimbabwe [28].

We also explored the environmental Kuznets curve (EKC) as an alternative model for capturing potential non-linear relationships. The EKC hypothesis has been extensively studied in research examining the relationship between economic growth and the environment. Additionally, we compared the results obtained from the EKC analysis to those derived from Bai and Perron’s methodology.

3. Data and Methodology

To analyze the stability of the 3E connection, three variables are used to proxy economic growth, energy and environment: GDP per capita, PPP (constant 2017 international $) from World Bank database, per-capita greenhouse gas emissions in CO₂ equivalents from Our World in Data (https://ourworldindata.org/grapher/per-capita-ghg-emissions) (accessed on 10 December 2024) and primary energy consumption per capita (kWh/person) from Our World in Data (https://ourworldindata.org/grapher/per-capita-energy-use?region=Europe) (accessed on 12 December 2024). The data cover the 1990–2023 period and the EU-27 member states. Compared to previous studies that consider per capita CO₂ emissions as a proxy for pollution [4,20], this paper uses per capita GHG emissions. This indicator is based on CO₂, nitrous oxide, and methane from all types of sources. The values of the variables in natural logarithms are used in the regression models. We have no missing data or outliers in the series.

A key characteristic of time-series data is their stationarity. When time series data are stationary, meaning their statistical properties (like mean and variance) remain constant over time, traditional statistical methods can be effectively employed for tasks such as regression analysis, mean and variance estimation, and future prediction. This allows for reliable inference and forecasting based on consistent patterns in the data. However, many time series exhibit non-stationarity, often referred to as unit-root processes, which necessitate distinct analytical approaches. Failing to address non-stationarity can lead to various challenges in statistical inference and prediction, particularly the issue of spurious regression, as demonstrated by Granger and Newbold [29]. Spurious regression can produce misleading results, such as high R-squared values and statistically significant coefficients, even when there is no genuine relationship between the variables. Consequently, assessing the stationarity of a time series is a crucial initial step in both time-series and panel-data analysis, ensuring that subsequent analyses are based on meaningful relationships rather than artificial correlations. Unit-root tests are statistical methods employed to determine if a time series is stationary or exhibits a unit-root property, as described by Dickey and Fuller [30]. These tests help researchers identify whether a series requires differencing or other transformations to achieve stationarity for valid statistical inference.

Perron’s research highlighted that structural breaks, exogenous shocks that alter model parameters, can significantly affect the accuracy of unit-root tests [31]. These breaks, often caused by major events like wars or economic crises, can make stationary data appear non-stationary, leading to incorrect conclusions about the underlying properties of the time series. To address this issue, Perron proposed new unit-root tests that account for potential structural breaks [31]. By incorporating the possibility of sudden shifts in the data’s behavior, these tests provide a more accurate assessment of stationarity in the presence of known disruptions. While Perron assumed the break’s date was known, subsequent studies by Zivot and Andrews [32] and Banerjee et al. introduced methods to endogenously determine the break date [33]. These advancements allow researchers to identify when significant shifts occurred within the data without prior knowledge, making the analysis more data-driven and robust to unknown events. Lumsdaine and Papell further extended this to handle cases with two unknown break points, providing a more comprehensive approach for series experiencing multiple significant shifts [34]. Karavias and Tzavalis addressed the impact of structural breaks on panel data unit-root tests [6]. They proposed new tests that allow for breaks in either the intercepts or both the intercepts and linear trends of the series. While the breaks are assumed to occur at the same time for all series, their magnitude can vary. The null hypothesis of these tests is that the panel series are unit-root processes without breaks, while the alternative hypothesis suggests they are stationary around breaking means or trends. These tests are crucial for analyzing panel datasets where common shocks might affect all units simultaneously, ensuring that the stationarity assessment accounts for these shared disruptions.

Karavias and Tzavalis’ tests are versatile and have specific advantages. Their suitability for both small and large datasets and their ability to handle multiple common breaks, whether their dates are known or unknown, make them widely applicable. The tests’ accommodation of non-normal errors with cross-sectional heteroskedasticity and dependence enhances their robustness in real-world applications where these issues are common in panel data [6]. Allowing for both homogeneous and heterogeneous autoregressive coefficients provides flexibility in modeling the dynamic relationships within the panel. Regarding their optimality, the invariance of these tests to initial conditions eliminates the need for potentially restrictive assumptions about the starting points of the series. Additionally, their insensitivity to the coefficients of deterministic components and strong power in the presence of linear trends contribute to more reliable stationarity assessments. The fact that structural breaks in the Karavias and Tzavalis tests can affect either the intercepts or both the intercepts and linear trends of the series allows for capturing different types of shifts in the data. When the break dates are unknown, a bootstrap method outlined by Karavias and Tzavalis is used to calculate the critical value and p-value, providing a data-driven approach to inference in the presence of unknown breaks [6]. Furthermore, the tests’ ability to accommodate cross-sectional heteroskedasticity, cross-sectional dependence (similar to O’Connell’s approach), and normal errors, as well as supporting unbalanced panel data, makes them a powerful tool for a wide range of panel data analyses.

Instability in model parameters can undermine the reliability of estimation and inference, leading to costly errors in decision-making. The moments when these parameters shift are known as “change points” in statistics and “structural breaks” in economics. Identifying and dating these breaks is crucial not only for accurate estimation, as models assuming constant parameters will be misspecified during periods of change, but also for comprehending the factors driving change and their impact on relationships. By pinpointing when these shifts occur, researchers can investigate the potential causes and consequences of these instabilities, leading to a deeper understanding of the underlying economic or social processes. In this paper, the authors first checked for the presence of breaks in the series when breakdates are not known using the Ditzen, Karavias, and Westerlund test [35]. This preliminary step helps to confirm whether structural breaks are indeed present in the data, justifying the use of methods that account for them. Subsequently, they applied the Karavias and Tzavalis panel unit root test, leveraging its ability to handle structural breaks in panel data to obtain a more accurate assessment of the stationarity properties of the series in the presence of potential disruptions [6].

Karavias and Tzavalis proposed two models for analyzing panel data with a single common break. The first model is designed to test whether a series is a random walk or a stationary series with a structural break in its mean. This model is particularly useful for panels with N cross-sectional units and T time-series observations. It allows researchers to examine the presence of a break in the series’ mean and assess its impact on the overall behavior of the data.

M₁:

$$H_0:y_{i,t}=y_{i,t-1}+{\epsilon }_{i,t}$$

$$H_1:y_{i,t}={\phi ·y}_{i,t-1}+\left(1-\phi \right)\left\{a_{1,i}I\left(t\le b\right)+a_{2,i}I(t>b)\right\}+{\epsilon }_{i,t}$$

(1)

$\phi$—autoregressive coefficient;

I(.)—indicator function;

$a_{1,i},a_{2,i}$—fixed effects before and after a certain break;

b—date when the break occurs;

i = 1, 2, …, N (index for cross-sections);

t = 1, 2, …, T (index for time).

The extension to the two breaks case could be performed. The second model tests whether a series follows a random walk with drift or a trend-stationary panel process with a structural break in both its mean and linear trend. This model is particularly useful for panels with N cross-sectional units and T time-series observations. It allows researchers to examine the presence of a break in both the mean and trend of the series and assess their impact on the overall behavior of the data. The representation of the second model is as follows:

M₂:

$$\begin{array}{c}{H}_0:y_{i,t}=y_{i,t-1}+{\beta }_i+{\epsilon }_{i,t}\hfill \\ H_1:y_{i,t}={\phi ·y}_{i,t-1}+\phi \left\{{\beta }_{1,i}I\left(t\le b\right)+{\beta }_{2,i}I(t>b)\right\}+\left(1-\phi \right)\left\{a_{1,i}I\left(t\le b\right)+a_{2,i}I\left(t>b\right)\right\}\hfill \\ +\left(1-\phi \right)\left\{{\beta }_{1,i}I\left(t\le b\right)+{\beta }_{2,i}I(t>b)\right\}+{\epsilon }_{i,t}\hfill \end{array}$$

(2)

${\beta }_i$—drift under $H_0$;

${\beta }_{1,i},{\beta }_{2,i}$—trend parameters under $H_1$;

$H_1$ is homogenous across various countries.

When H₁ in M₁ and M₂ considers two breaks, (1) and (2) become the following:

$$H_1:y_{i,t}={\phi ·y}_{i,t-1}+\left(1-\phi \right)\left\{a_{1,i}I\left(t\le b_1\right)+a_{2,i}I\left(b_1<t\le b_2\right)+a_{3,i}I(t>b_2)\right\}+{\epsilon }_{i,t}$$

(3)

$$\begin{array}{c}{H}_1:y_{i,t}={\phi ·y}_{i,t-1}+\phi \left\{{\beta }_{1,i}I\left(t\le b_1\right)+{\beta }_{2,i}I\left(b_1<t\le b_2\right)+{\beta }_{3,i}I(t>b_2)\right\}\hfill \\ +\left(1-\phi \right)\left\{a_{1,i}I\left(t\le b_1\right)+a_{2,i}I\left(b_1<t\le b_2\right)+a_{3,i}I(t>b_2)\right\}\hfill \\ +\left(1-\phi \right)\left\{{\beta }_{1,i}I\left(t\le b_1\right)+{\beta }_{2,i}I\left(b_1<t\le b_2\right)+{\beta }_{3,i}I(t>b_2)\right\}+{\epsilon }_{i,t}\hfill \end{array}$$

(4)

In the time series approach, for each country, we apply the test for multiple breaks at unknown breakdates of Bai and Perron before making the estimations. The Bai–Perron test starts from a model with more structural breaks [7] (m breaks, m + 1 regimes), where $y_t$ is the observed regresand at moment t; $y_t$ and $y_t$ are explanatory variables:

$$\begin{gathered} y_t=x_t^{\prime }\beta +z_t^{\prime }{\delta }_1+e_t, t=1,\dots ,T_1 \\ {y}_t=x_t^{\prime }\beta +z_t^{\prime }{\delta }_2+e_t, t=T_1+1,\dots ,T_2 \\ {y}_t=x_t^{\prime }\beta +z_t^{\prime }{\delta }_{m+1}+e_t, t=T_m+1,\dots ,T \end{gathered}$$

(5)

$e_t$—error at moment t;

$\beta , {\delta }_j$—vector of parameters, where j = 1, 2, …, m + 1;

($T_1,T_2,\dots ,T_m$)—break points.

Given T observations for dependent and explanatory variables, the main objective is to provide estimations for parameters and the break points.

The matrix form of the model (5) is given by the following:

$$Y=X\beta +\overline{Z}\delta +E$$

(6)

$$Y={(y_1,\dots ,y_T)}^{\prime }, X={(x_1,\dots ,x_T)}^{\prime }$$

$$\overline{Z}=diag\left(Z_1,\dots ,Z_{m+1}\right), \mathrm{Z}\text{-}\mathrm{m}\mathrm{a}\mathrm{t}\mathrm{r}\mathrm{i}\mathrm{x}$$

$$Z_i={(z_{T-i}+1,\dots ,z_{Ti})}^{\prime }$$

$$\delta ={({\delta }_1^{\prime },{\delta }_2^{\prime },\dots ,{\delta }_{m+1}^{\prime })}^{\prime }$$

$$E={(e_1,\dots ,e_T)}^{\prime }$$

Each break date is restricted to be asymptotic and limited to boundaries. For any partition, the least squares method is employed to provide an estimate for ${\delta }_j$ by minimizing the sum of the squared residuals.

After checking for the presence of the unit root in the series, specific regression models are built. In the case of stationary panel data, Pesaran and Smith use the Mean Group estimator [36] and system dynamic panel-data estimation to manage endogeneity. For stationary time series, linear/non-linear regressions with robust standard errors are used to explain GHG emissions in the EU member states (1990–2022). Moreover, the causal mediation analysis is made, considering that energy consumption might also indirectly affect pollution through GDP.

4. Results

The test for multiple breaks at unknown breakdates [35] assumes no break, while the alternative one states two breaks. The results of the test for multiple breaks at unknown breakdates in

Table 1. The results of test for multiple breaks at unknown breakdates.

Variable	Statistic	Estimated Break Dates:
GHG	122.45	1991, 2021
energy	178.45	1991, 1992
GDP	223.49	2020, 2021

Source: own calculations in Stata 18.

indicate that there are two breaks in each series. The 1% critical value, 5% critical value, and 10% critical value are 9.36, 7.22, and 6.28.

Under the null hypothesis, the Karavias and Tzavalis panel unit root test states that all panel time series are unit root processes [6], while the alternative hypothesis considers that some or all of the panel time series are stationary processes.

Table 2. The results of Karavias and Tzavalis panel unit root test.

Variable	Statistic	Bootstrap Critical-Value	p-Value	Estimated Break Dates:
GHG	−14.6842	2.4800	0.0000	1991, 2021
energy	−31.0468	4.0530	0.0000	1991, 1992
GDP	−11.4106	−4.3196	0.0000	2020, 2021

Source: own calculations in Stata 18.

suggests that the panel data for all variables are stationary, considering various estimated break dates. GDP has been deeply affected by the COVID-19 pandemic in 2020 and 2021, while pollution levels increased in 1991 when countries in Eastern Europe accelerated industrialization in the new political context, and reduced in 2021 when economic activities declined or many jobs moved online.

The results of the estimations in

Table 3. The results of the Pesaran and Smith Mean Group estimator and system dynamic panel-data estimation to explain GHG emissions in the EU (1990–2022).

Variable	Pesaran and Smith (1995) Mean Group Estimator		System Dynamic Panel-Data Estimation
	Coefficient	p-Value	Coefficient	p-Value
GHG in the previous year	-	-	0.927	0.000
energy	1.222	0.000	0.010	0.051
GDP	−0.415	0.000	1.361	0.049
GDP2	-	-	−0.067	0.05
constant	−6.627	0.000	−6.828	0.048
Root Mean Squared Error	0.0764	-	-	-

Source: own calculations in Stata 18.

suggest a direct impact of energy consumption on pollution. The inverted-U pattern was identified in the growth–pollution nexus, which supports the idea that, after a certain threshold of economic growth, the economic activity becomes beneficial for the environment due to technological progress. In other words, GDP tends to reduce GHG emissions in the long run, even if, in the short run, growth is not beneficial for the environment. Moreover, there is a clear tendency for an increase in GHG emissions according to the dynamic panel data model.

First, in the time series approach, the DF-GLS (Dickey and Fuller generalized least squares) test that does not take into account breaks in the series is considered, but the results are scarce and indicate that most of the series are non-stationary. For the DF-GLS test, the critical values at 1% critical value, 5% critical value, and 10% critical value are −3.770, −2.853, and −2.405.

The Bai and Perron test is applied to a time series for each country to establish the stationarity and to identify potential breaks in the series. The null hypothesis states no break(s), while the alternative specifies the existence of two breaks. The 1% critical value, 5% critical value, and 10% critical values are 9.36, 7.22, and 6.28. According to

Table 4. The results of the DF-GLS (Dickey and Fuller generalized least squares) test and of the test for multiple breaks at unknown breakdates.

Country	Variable	DF-GLS Test	Bai and Perron (1998) Test		Country	Variable	DF-GLS Test	Bai and Perron (1998) Test
			Statistic	Estimated break points				Statistic	Estimated break points
Austria	GHG	−1.822	38.64	2000   2011	Italy	GHG	−1.187	244.53	2008   2012
	energy	−2.088	36.89	1995   2013		energy	−1.928	65.85	1997   2011
	GDP	−1.462	212.39	1997   2005		GDP	−1.288	40.31	1997   2011
Belgium	GHG	−1.970	192.53	2008   2013	Latvia	GHG	−2.138	90.67	1993   2003
	energy	−1.737	74.18	1995   2010		energy	−2.206	27.52	1993   2003
	GDP	−1.641	155.65	1997   2004		GDP	−1.941	104.48	2002   2013
Bulgaria	GHG	−1.972	37.81	1996   2018	Lithuania	GHG	−2.339	20.16	1999   2003
	energy	−1.939	8.82	1998   2014		energy	−2.745	28.27	1993   2009
	GDP	−3.076	179.75	2004   2015		GDP	−1.606	116.1	2002   2012
Croatia	GHG	−2.350	14.91	1998   2003	Luxembourg	GHG	−1.542	91.85	1994   2013
	energy	−1.412	46.51	1995   2000		energy	−1.016	33.43	2001   2012
	GDP	−1.382	80.71	2001   2016		GDP	−0.676	233.62	1998   2003
Czechia	GHG	−3.547	111.49	1993   2008	Malta	GHG	−1.248	359.87	2010   2014
	energy	−2.907	16.28	2002   2012		energy	−1.391	3.63	1999   2018
	GDP	−2.325	147.5	2003   2014		GDP	−1.724	158.5	1999   2013
Cyprus	GHG	−1.271	30.92	1998   2010	Netherlands	GHG	−1.315	109	2010   2018
	energy	−1.665	47.63	1997   2011		energy	−1.876	46.21	2000   2013
	GDP	−1.560	78.88	1999   2017		GDP	−1.472	157.99	1997   2005
Denmark	GHG	−0.924	114.97	2007   2013	Poland	GHG	−1.341	141.8	1998   2008
	energy	−1.464	61.14	2008   2013		energy	−1.364	21.42	1999   2005
	GDP	−1.545	98.88	1997   2015		GDP	−2.871	111.17	1998   2009
Estonia	GHG	−1.836	11.66	1993   2003	Portugal	GHG	−1.949	51.83	1997   2008
	energy	−2.392	14.05	1993   2006		energy	−1.099	56.1	1994   1999
	GDP	−1.580	90.53	2002   2015		GDP	−2.328	130.9	1997   2016
Finland	GHG	−0.682	47.87	2000   2013	Romania	GHG	−3.171	71.61	1997   2012
	energy	−1.061	54.5	1996   2011		energy	−2.479	37.93	1997   2008
	GDP	−1.057	214.5	1997   2003		GDP	−2.358	184.96	2004   2016
France	GHG	−1.326	150.41	2006   2013	Slovakia	GHG	−1.699	77.75	1993   2011
	energy	−0.106	89.01	2008   2018		energy	−1.657	24.77	1993   2008
	GDP	−1.359	151.36	1997   2004		GDP	−2.894	70.78	2004   2016
Germany	GHG	−1.868	80.41	1998   2017	Slovenia	GHG	−1.007	38.08	2000   2013
	energy	−1.398	93.5	2008   2018		energy	−0.332	38.61	1995   2014
	GDP	−3.504	167.24	1999   2010		GDP	−1.377	78.64	2002   2016
Greece	GHG	−0.815	173.48	2012   2018	Spain	GHG	−1.376	73.01	1998   2008
	energy	−1.089	59.24	1997   2011		energy	−1.003	54.14	1998   2008
	GDP	−2.071	58.88	2000   2010		GDP	−1.379	132.13	1996   2000
Hungary	GHG	−2.151	100.1	1993   2008	Sweden	GHG	−0.327	51.82	1993   2012
	energy	−1.924	23.73	2011   2016		energy	−3.026	108.47	2005   2012
	GDP	−1.647	148.56	2001   2015		GDP	−1.474	166.41	1998   2005
Ireland	GHG	−1.511	30.45	1995   2009	-
	energy	−1.413	88.3	1997   2008
	GDP	−1.511	144.31	1997   2014

Source: own calculations in Stata 18.

, the null hypothesis was rejected for all the countries, which confirms the stationarity with the existence of two breaks in all the series and for all the countries at a 1% significance level. The descriptive statistics are reported in Appendix A.

Given the stationarity of the time series, the data at the level are used to construct linear/non-linear regressions to explain GHG emissions in each EU member state, and the results are reported in

Table 5. Linear/non-linear regressions to explain GHG emissions in the EU member states (1990–2022).

Country	Coefficient (Robust Standard Error in Brackets)
	Energy	GDP	GDP2	Constant
Austria	1.37 *** (0.052)	−0.465 *** (0.12)	-	−7.512 *** (1.267)
Belgium	0.997 ** (0.148)	30.014 ** (13.516)	−1.446 *** (0.633)	−164.043 (70.594)
Bulgaria	1.357 *** (0.143)	−0.2 *** (0.044)	-	−9.985 *** (1.558)
Croatia	−0.514 * (0.221)	18.227 *** (5.817)	−0.887 *** (0.287)	−86.693 *** (27.054)
Czechia	1.523 *** (0.18)	−19.374 *** (3.805)	0.916 *** (0.815)	88.354 *** (18.242)
Cyrus	0.852 *** (0.093)	0.335 *** (0.050)	-	−10.313 *** (1.135)
Denmark	1.255 *** (0.063)	−0.688 *** (0.071)	-	−3.428** (1.319)
Estonia	1.338 *** (0.118)	−0.38 *** (0.117)	0.038 *** (0.02)	−12.349 *** (0.87)
Finland	1.689 *** (0.09)	0.071 ** (0.032)	-	−16.631 *** (1.007)
France	1.114 ** (0.11)	−45.856 ** (20.332)	2.131 ** (0.962)	236.608 ** (106.926)
Germany	1.418 *** (0.129)	−20.391 *** (6.764)	0.926 *** (0.316)	99.317 ** (35.818)
Greece	3.411 *** (0.293)	−79.629 *** (12.27)	3.772 *** (0.591)	386.743 *** (62.776)
Hungary	1.455 *** (0.067)	−0.3685 *** (0.0189)	-	−9.3067 *** (0.7439)
Ireland	1.4 *** (0.124)	−8.296 *** (1.651)	0.374 *** (0.077)	33.715 *** (8.653)
Italy	1.953 *** (0.109)	−0.755 *** (0.139)	-	−10.236 *** (0.818)
Latvia	1.684 *** (0.549)	−10.347 *** (5.518)	0.551 *** (0.277)	33.318 *** (25.686)
Lithuania	−0.252 (0.273)	−6.521** (3.005)	0.337* (0.151)	36.054 *** (15.031)
Luxembourg	1.208 *** (0.028)	−21.887 *** (6.708)	0.924 *** (0.292)	118.774 *** (38.526)
Malta	−0.013 ** (0.005)	34.594 *** (3.495)	−1.722 *** (0.171)	−171.508 *** (17.792)
Netherlands	1.305 *** (0.130)	−0.688 *** (0.033)	-	−4.413 *** (1.491)
Poland	0.977 *** (0.074)	−0.296 *** (0.014)	-	−4.787 *** (0.713)
Portugal	1.458 *** (0.287)	−1.540 *** (0.237)	-	2.889 *** (1.342)
Romania	1.053 *** (0.070)	−0.118 *** (0.019)	-	−7.542 *** (0.818)
Slovakia	0.583 *** (0.108)	−0.222 *** (0.017)	-	−1.773 *** (1.190)
Slovenia	1.406 *** (0.132)	−0.006 (0.400)	-	−12.513 *** (1.572)
Spain	2.015 *** (0.074)	−73.861 *** (12.55)	3.481 *** (0.6)	372.645 *** (65.074)
Sweden	0.601 * (0.328)	70.411 *** (12.204)	−3.306 *** (0.576)	−379.249 *** (62.704)

Source: own calculations in Stata 18; * p-value < 0.1; ** p-value < 0.5; *** p-value < 0.01.

. Moreover, the role of mediator for GDP in the energy consumption–GHG emissions nexus is considered, and the evaluation of this indirect effect is made.

The linear connection between GDP and GHG emissions was supported for certain countries, as Table 5 indicates: Austria, Bulgaria, Cyprus, Denmark, Finland, Hungary, Italy, Netherlands, Poland, Romania, Portugal, Slovakia, and Slovenia. The non-linear relationship was checked for the rest of the countries. In the case of countries with the linear link, economic growth reduced pollution in Austria, Bulgaria, Denmark, Hungary, Italy, the Netherlands, Poland, Romania, Portugal, and Slovakia. There is no significant impact of GDP on pollution in Slovenia. In Croatia, Cyprus, and Finland, growth enhanced GHG emissions. The U pattern in the growth–pollution nexus was supported for the Czech Republic, Estonia, France, Germany, Greece, Ireland, Latvia, Lithuania, Luxembourg, and Spain, while the inverted pattern was met only in Belgium, Sweden, and Malta. Energy consumption plays a significant role in pollution control in all the EU member states, except for Lithuania. Energy consumption enhanced GHG emissions in Austria, Belgium, Bulgaria, Czechia, Cyprus, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Luxembourg, Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, Sweden, and Spain. Only in Croatia and Malta did more energy use reduce pollution.

As

Table 6. Causal mediation analysis.

Country	R-Squared (Coefficient of Determination)	Causal Mediation Analysis: Proportion
Austria	0.8437	0.3478
Belgium	0.9693	0.5885
Bulgaria	0.6864	0.6566
Croatia	0.3797	0.2913
Czechia	0.9538	0.3419
Cyrus	0.8279	0.2264
Denmark	0.9831	0.4097
Estonia	0.8942	0.2936
Finland	0.9132	0.4069
France	0.9716	0.4198
Germany	0.4335	0.9733
Greece	0.9031	0.3825
Hungary	0.9253	0.2619
Ireland	0.6765	0.5873
Italy	0.9615	0.4512
Latvia	0.8635	0.2516
Lithuania	0.378	0.3299
Luxembourg	0.9895	0.4062
Malta	0.8734	0.3762
Netherlands	0.9473	0.4578
Poland	0.9559	0.2055
Portugal	0.4757	0.9351
Romania	0.9308	0.624
Slovakia	0.8916	0.4662
Slovenia	0.7986	0.3750
Spain	0.9537	0.3864
Sweden	0.7217	0.4174

Note: Outcome equation includes treatment–mediator interaction.

suggests, in the EU member states, the indirect effect via energy accounts for 20.55% for Poland and 93.51% for Portugal of the effect of GDP on GHG emissions, and the remaining is due to other mechanisms.

Countries with a higher reliance on fossil fuels for energy generation (e.g., Poland) tend to have a lower indirect effect, as GDP growth is more directly linked to increased energy consumption and emissions. In contrast, countries with a more diversified energy mix, including renewable sources (e.g., Portugal), have a higher indirect effect, as GDP growth may lead to increased energy efficiency and reduced emissions intensity.

5. Discussion

It is necessary to understand how the specific ways the breaks influence the energy consumption–pollution emissions relationship. For example, the COVID-19 pandemic and the Ukraine war might have significantly altered outcomes through supply chain disruptions or energy price fluctuations. Some countries and sectors moved faster towards alternative energy sources like coal, nuclear, or natural gas (depending on availability and policy). This could have had mixed effects on emissions depending on the carbon intensity of the substitutes. Economic restructuring and decline in heavy industries in many Eastern European countries led to a significant reduction in energy consumption (industrial output plummeted, leading to lower energy demand), decrease in pollution emissions (reduced industrial activity directly translated to lower emissions), and shift in energy mix in some cases (gradual integration with Western European energy markets and adoption of different energy technologies) [37]. Our results are similar to those of Apergis et al., Farhani et al., Zhang et al., Munir et al., and others [14,15,16,17] for panel data, highlighting significant shifts in the 3E relationship. The findings are also in line with those of Roinioti and Koroneos, who showed that Greece experienced a notable decline in CO₂ emissions from energy use between 2003 and 2013, coinciding with a reduction in energy consumption, particularly during the economic recession [25].

5.1. Policy Recommendations

The findings of this study underscore the urgent need for a multifaceted and robust policy response at the EU level to decouple energy consumption from greenhouse gas emissions and address the observed instabilities in the 3E relationship. Given that energy consumption amplifies GHG emissions in the vast majority of EU-27 countries, a singular focus on energy efficiency will likely fall short of achieving meaningful reductions. Therefore, a central policy pillar must be a decisive and accelerated transition towards cleaner energy sources, coupled with effective mechanisms to internalize the environmental costs of carbon emissions. Implementing a comprehensive carbon pricing mechanism, such as a carbon tax or an enhanced emissions trading system, is crucial to incentivize businesses and individuals to reduce their carbon footprint across all sectors [38]. This should be complemented by the establishment and rigorous enforcement of stricter pollution standards for industries, particularly those identified as high emitters. Governments must actively support the research, development, and widespread adoption of cleaner industrial processes and technologies through targeted funding, tax breaks, and regulatory frameworks that favor sustainable innovation. Furthermore, recognizing the significant role of land use in carbon sequestration, policies promoting forest conservation and afforestation are essential. This includes implementing stronger measures to prevent deforestation, supporting sustainable forest management practices, and actively engaging in tree-planting initiatives to enhance carbon sinks and improve air quality. Finally, extending the scope of environmental considerations to the agricultural sector through the promotion of sustainable farming practices that minimize greenhouse gas emissions is vital for a holistic approach to pollution reduction [39].

Beyond direct emissions reduction, policies must address the underlying drivers of energy consumption and the observed growth–pollution nexus. This study’s indication of a potential rebound effect after periods of economic recession highlights the need for policies that promote sustainable consumption patterns and circular economy principles. This could involve incentivizing product longevity, repairability, and material recycling to reduce the demand for new production and its associated energy consumption. Moreover, given the varying impacts of economic growth on emissions across member states, policy responses may need to be tailored to specific national contexts while adhering to overarching EU-level goals. For countries exhibiting a U-shaped growth–pollution curve, it is crucial to proactively implement measures that prevent a resurgence of pollution as economic growth continues, ensuring that future development is inherently sustainable. Ultimately, a successful policy framework will require strong inter-ministerial coordination, engagement with stakeholders across industries and civil society, and a long-term commitment to achieving the Sustainable Development Goals through technological innovation and behavioral change [40].

For countries with high industrial emissions (e.g., Germany, Italy), an enhanced EU ETS should gradually reduce allowance caps and expand its scope to cover more sectors. Border carbon adjustment mechanisms should be implemented to prevent carbon leakage and ensure a level playing field. Revenue generated from carbon pricing should be reinvested in green technologies and supporting industries in their decarbonization efforts.

In the case of countries with a larger share of emissions from transport and buildings (e.g., Ireland, France), carbon taxes could be more directly applied to fuels and heating, with careful consideration for social equity through compensatory measures for vulnerable households. Expanding the EU ETS to cover these sectors or implementing comparable carbon pricing mechanisms is crucial.

It is necessary to provide financial and technical assistance for upgrading industrial facilities for countries with older industrial infrastructure (e.g., some Eastern European nations) to meet higher environmental standards. Implement a phased approach to stricter regulations, allowing time for adaptation while maintaining clear long-term goals. Countries with more modern industrial sectors should focus on pushing the boundaries of environmental performance through ambitious standards and incentives for innovation in pollution prevention and control [41].

In conclusion, specific policy recommendations for different country profiles can be proposed for the following:

• The following applies to countries with high levels of industrial emissions (e.g., Germany, Italy):

  ○

  Gradually reduce allowance caps and expand the scope of the EU ETS to cover more sectors.

  ○

  Implement border carbon adjustment mechanisms to prevent carbon leakage and ensure a level playing field.

  ○

  Reinvest revenue generated from carbon pricing in green technologies and supporting industries’ decarbonization efforts.

• The following applies to countries with a larger share of emissions from transport and buildings (e.g., Ireland, France):

  ○

  Apply carbon taxes more directly to fuels and heating, with careful consideration for social equity through compensatory measures for vulnerable households.

  ○

  Expand the EU ETS to cover these sectors or implement comparable carbon pricing mechanisms.

• The following applies to countries with older industrial infrastructure (e.g., some Eastern European nations):

  ○

  Provide financial and technical assistance for upgrading industrial facilities to meet higher environmental standards.

  ○

  Implement a phased approach to stricter regulations, allowing time for adaptation while maintaining clear long-term goals.

• The following applies to countries with more modern industrial sectors:

  ○

  Focus on pushing the boundaries of environmental performance through ambitious standards and incentives for innovation in pollution prevention and control.

5.2. Implications for Renewable Energy

This study’s findings carry significant implications for the trajectory and importance of renewable energy deployment within the European Union. The clear link between energy consumption and increased GHG emissions in the majority of member states underscores the critical role that renewable energy sources must play in decoupling economic activity from environmental degradation. As the elasticity of this relationship is greater than one in many countries, simply reducing overall energy consumption, while necessary, may not be sufficient to meet ambitious climate targets. A substantial and rapid shift towards renewable energy technologies—including solar, wind, hydro, and geothermal—becomes paramount to mitigate the adverse environmental consequences of energy use. Policy frameworks must, therefore, prioritize the acceleration of renewable energy deployment through a combination of supportive measures. This includes streamlining permitting processes, providing long-term investment signals through stable feed-in tariffs or contracts for difference, and investing in grid infrastructure to accommodate the variable nature of some renewable sources.

Furthermore, this study’s acknowledgement that, while renewable energy consumption can reduce CO₂ emissions compared to fossil fuels, the overall effect can vary, highlights the need for strategic and well-integrated renewable energy policies. This includes considering the specific characteristics of different renewable energy sources, their regional suitability, and the need for complementary technologies such as energy storage solutions to ensure grid stability and reliability. Investing in research and development to improve the efficiency and cost-effectiveness of renewable energy technologies, as well as exploring innovative solutions like green hydrogen, will be crucial for a successful energy transition. The policy recommendations for carbon pricing and stricter pollution standards will also indirectly incentivize the adoption of renewable energy by making fossil fuel-based energy more expensive and less environmentally attractive. Ultimately, the transition to a predominantly renewable energy system is not just an environmental imperative but also an opportunity for economic growth, job creation, and enhanced energy security within the European Union. The findings of this study reinforce the urgency and importance of creating a policy environment that strongly supports and accelerates this transition.

For fossil fuel-reliant nations like Poland, policies should prioritize a just transition that includes significant investment in retraining and creating new jobs in renewable energy sectors. They should also focus on phasing out coal and other high-carbon fuels with clear timelines and interim targets, coupled with financial support for diversifying energy mixes towards renewables and improving energy efficiency in existing infrastructure.

For countries dominated by renewables or with strong renewable potential, like Nordic countries, Spain, and Portugal, policies should focus on accelerating the deployment of advanced renewable technologies (e.g., offshore wind, green hydrogen), enhancing grid interconnectivity to facilitate energy sharing, and investing in energy storage solutions to ensure grid stability.

The causal mediation analysis revealed that, in countries like Sweden, Slovakia, Portugal, Romania, Netherlands, Luxembourg, Italy, Ireland, Germany, France, Finland, Denmark, Bulgaria, and Belgium, more than 40% of the effect of GDP on GHG emissions is due to energy consumption. In Finland, Germany, Ireland, and Luxembourg, the GDP growth enhanced pollution in the long run, and the policy measures should consistently target energy consumption. The implementation of more energy efficiency measures is required in these countries: higher standards for buildings, appliances, and vehicles to reduce energy consumption, educate the public on energy-saving behaviors, provide incentives for energy-efficient upgrades, and upgrade the energy grid to enable more efficient energy distribution and management [37]. The government could increase funding for solar, wind, and other renewable energy projects, offer tax breaks, subsidies, or feed-in tariffs to encourage businesses and households to adopt renewable energy, and support research into new renewable energy technologies and their integration into existing energy grids [38].

6. Conclusions

This research explored the connection between GHG emissions, economic growth, and energy consumption in the EU countries, given the objective of the European Commission to achieve a green transition. The goal was to determine if countries were decoupling emissions from economic growth and shifting towards cleaner energy sources. Our findings contribute to the ongoing discussion about the relationship between these three variables.

Previous studies often assumed a stable relationship between these factors. However, we employed econometric methods that allowed for changes in these relationships over time. Our results support our initial hypothesis and indicate that the relationship is not static. The COVID-19 pandemic played a significant role in this instability.

After accounting for structural breaks, we analyzed how the relationship between these variables evolved in different countries. We found that many countries successfully decoupled emissions from economic growth, achieving negative elasticities, but there are still a few countries that need to make more efforts for decoupling. The major problem remains the fact that energy consumption has enhanced pollution in 25 out of 27 EU countries, and the European Commission should reshape its policies to decouple energy consumption from growth. Moreover, the study of the mechanisms between the variables in the 3E relationship allows us to conclude that, in almost half of the countries, there is a considerable effect of GDP on GHG emissions due to energy consumption.

Globalization likely intensified the link between economic growth and energy consumption, with the impact on emissions being more complex and potentially leading to increased overall emissions in some phases. The 2007–2008 financial crisis caused a temporary decoupling driven by economic recession, highlighting the vulnerability of emissions to economic downturns rather than structural low-carbon transitions. The COVID-19 pandemic led to the most significant and rapid decoupling, offering a glimpse of lower emissions. However, the sustainability of this decoupling depends on the long-term economic recovery strategies and the success of policies promoting cleaner energy and sustainable practices. The persistent carbon-intensive energy use noted suggests that, while decoupling from growth might have occurred, the fundamental energy system transformation is still underway and faces significant hurdles.

Despite these insights on the 3E relationship in the EU member states, this study also presents limitations. We returned to the hypothesis of two structural breaks in the series since, for most of the countries, this hypothesis was supported, but the analysis could be extended to fewer or more breaks. Moreover, the analysis was reduced to the entire period (1990–2023) without any research on specific periods. Furthermore, this study did not include a detailed breakdown of energy source data. Analyzing the specific contributions of different energy sources (e.g., coal, oil, natural gas, nuclear, solar, wind) to both energy consumption and GHG emissions could provide a more granular understanding of the drivers of pollution. Additionally, the analysis did not disaggregate GHG emissions by specific industry sectors (e.g., manufacturing, transportation, agriculture, energy production). Such a breakdown could reveal which sectors are the most significant contributors to emissions and inform more targeted policy interventions.

Therefore, a future analysis might consider the stability of the 3E connection during the COVID-19 pandemic. Building upon the identified limitations, future research should aim to incorporate a detailed analysis of energy source composition to assess the impact of specific energy transitions on the 3E relationship. This could involve examining the elasticity of emissions with respect to consumption from different energy categories. Moreover, future studies should disaggregate GHG emissions data by industry sector to pinpoint the primary sources of pollution and evaluate the effectiveness of sector-specific policies. Investigating the 3E relationship across different sub-periods within the 1990–2023 timeframe could also reveal nuanced dynamics and the impact of specific policy changes or economic events beyond the identified structural breaks. Finally, exploring the potential for varying numbers of structural breaks (three structural breaks or more) for individual countries could offer a more tailored understanding of the evolving relationships at the member state level. Future work could employ complex models like multiple-break unit root tests to verify additional turning points, offering a fuller picture of the data’s characteristics. Incorporating machine learning techniques, such as clustering or anomaly detection, to automatically pinpoint potential breaks could minimize human bias and advance both the sophistication and accuracy of the approach. Multiple unit root tests were employed, reflecting rigor, but these traditional methods may falter when data exhibit nonlinear traits or complex trends. Testing methods better suited to nonlinear time series in a future study would help ensure the robustness of the results. Compared to previous papers, like that of González-Álvarez and Montañés, this study also takes into account the years of the pandemic that proved to be essential in explaining the instability of the 3E relationship in the EU, being the first analysis made only for the EU countries. Expanding the scope to non-EU countries in future work could reveal broader patterns and provide a more comprehensive perspective.