Dynamic Analysis of CO₂ Emissions and Their Determinants in the Countries of Central and Eastern Europe

Abstract

This study addresses the problem of the relationship between the adopted development path and the emission levels of greenhouse gases. The analysis includes the countries of Central and Eastern Europe that joined the EU in 2004–2018. This study used a dynamic analysis due to the nature of the changes taking place, which cannot be assessed in static terms. The results of the research conducted so far for this group of countries have been inconclusive. The countries studied have the aim of accelerating economic growth in order to reduce their distance from other EU countries on the one hand, while attempting to pursue a policy favoring the reduction of greenhouse gas emissions, on the other. The aim of this evaluation was to determine the relationship between GDPs per capita and CO₂ emissions and to establish the factors determining this relationship. The results for the whole group showed the presence of N-shaped EKCs. This study showed the importance of energy productivity and demographic factors as well as the pace of GDP growth. This research extended the scope of research on CO₂ emissions and their determinants through the use of dynamic methods, as well as the complex course of their relation to GDP per capita in Central Eastern European Countries.

1. Introduction

With the increase in environmental awareness, more and more people responsible for economic policy pay more attention not only to the rate of economic growth, but also to the progressive degradation of the environment. Observed climate changes, increases in greenhouse gas emissions and increasing air pollution make us consider the relationship between environmental degradation and the rate of economic growth. In the long term, it is important to ensure that future generations will be able to meet their environmental needs and live in the natural environment, which is in line with the concept of sustainable development. The aim of this manuscript is therefore to find out whether there is a relationship between economic development measured as GDP and carbon dioxide emissions in Eastern European countries. For this purpose, an econometric analysis based on the Kuznets curve environmental concept was used. This approach makes it possible to simultaneously check the relationship between greenhouse gas emissions and economic development for a larger group of countries. For this purpose, a dynamic panel model was used. The obtained results show that the relationship between CO₂eq and GDP per capita for Central Eastern European Countries (CEECs) is in the shape of an inverted N curve. Therefore, the studied relationships were not permanent and after exceeding a certain growth threshold, greenhouse gas emissions increased again. This indicates the need to apply a different economic policy in each country, due to the heterogeneous nature of the changes in each country.

With the development of civilization, more and more attention is paid to the problem of greenhouse gas emissions and their long-term consequences. Therefore, currently, various theoretical approaches to modeling the energy transformation towards a zero-emission economy are being considered, the aim of which is to eliminate a number of limitations, such as the lack of taking into account the non-equilibrium situation [1]. The issue of uncertainty inherent in the assumptions on which these models are built is also emphasized [2]. This problem becomes multidisciplinary and is visible in many different fields and studies. This has been used in the shaping of waste management policy [3], various studies on life cycle assessments [4,5,6] and has created research opportunities for economies and entire groups of countries. The observed climate changes, attested by numerous scientific studies, encourage the analysis of the relationship between CO₂eq emissions and the selected path of socio-economic development [7,8]. This allows for the assessment of both the effectiveness of pro-environmental policy and the existing structural and technological changes. While investigating this problem, attention is paid to the heterogeneity of the individual countries, both in terms of applied institutional solutions and broadly understood socio-economic structures constituting the conditions for the implemented growth path. The relationship between income levels and CO₂eq emissions can be represented as an inverted U-shape, i.e., EKC (environmental Kuznets curve) [9], indicating that environmental degradation initially increases with a growth in income levels to reach the peak point of maximum emissions, and then decreases with an increase in per capita income [10]. The indicated discrepancies in the research results lead to a reflection on the solutions applied in regulatory policy. So, is reaching a certain point on the EKC permanent or can we expect a multidirectional change? This requires an explanation of the identified differences and an indication of the prospects for changes in climate change policy. The presented results also indicate the regional differentiation of the transformations of the EKC, which indicates the scope of the analysis that was performed in this study. The consequences of introducing new technological solutions are often not known, and it is difficult to assess their long-term effects on the environment. This may initially give the wrong signals for the relationship under study.

The research results so far have been inconclusive [11], which encourages further continuing of the search for an explanation of the identified contradictions. Referring directly to the traditional EKC, it can be seen that in the long run, the increase in per capita income leads to a significant reduction in its impact on the environment [12]. Therefore, a question arises about the possibility of maintaining the current rate of growth with climate change advancing and the simultaneous reduction of pressure on the environment. There are serious discrepancies in the literature on this subject, which constitutes the essence of this research.

This study aims to verify the existing differences in the shape of the environmental Kuznets curve for the countries of Central and Eastern Europe. The leading convergence hypothesis may suggest that there is a hypothetical point at which a regression in CO₂ emissions will occur, which, using panel methods, may indicate potential directions for changes in the GDP–CO₂ emission relationship. Indicating a universal shape of the EKC for a given region may condition the adoption of an appropriate policy, consistent with the concept of sustainable development. In the case of diagnosing differences and obtaining different shapes of the EKC, a heterogeneous national policy will be recommended. The differences at the level of regions occurring at the research stage encourage reflection on the individual course of the EKC. So far, no taxonomy of similar areas for the EKC has been made, especially for a large set of the economies of Central and Eastern Europe. Moreover, the use of different selections of variables hindered a full comparability of the obtained results. This study makes a significant contribution to EKC analyses and provides the implications of national and European policies.

2. Literature Review

Research on CO₂eq emissions can be divided into three groups. The first one includes an assessment of the shape of the very curve characterizing the relationship between greenhouse gas emissions and the amount of GDP per capita. The second one concerns the factors influencing the aforementioned relationship. The third group of studies, in turn, concerns the specification of the objective and spatial scope of this relation. Numerous studies have considered the dependencies and causal relationships between economic development and the quality of the environment. The traditional approach searched for the confirmation of the environmental Kuznets curve (EKC). However, the obtained results are inconclusive for individual countries and groups of countries [11]. Some studies confirm the occurrence of the environmental Kuznets curve both for individual countries and for larger groups of countries: Cole et al. [13] included seven countries, York et al. [14] included one-hundred eleven countries, Faiz-Ur-Rehman et al. [15] included four countries, Iwata et al. [16] included France, Arouri et al. [17] included twelve countries, Bölük and Mert [18] included sixteen countries, Ahmad et al. [19] included India, Danish, and Ozcan, and Ulucak [20] included India, and Miranda et al. [21] included three countries. At the same time, in many cases, the existence of a dependence consistent with the EKC was rejected by Dogan and Seker [22] in their study of 23 countries, pointing to a different shape of the relationship or stating the lack thereof. Moreover, even a confirmation of dependence in the form of the EKC for a short period does not necessarily implicate the existence of this relationship in the long run, as such a relationship may only occur periodically, as a cyclical element [23]. The literature therfore indicates the ambiguous shapes of the EKC, and consequently potential differences in the application of national policies. However, this creates the possibility of grouping certain areas with similar EKCs and isolating factors that are conducive to emission reduction without disruptions in the pace of economic growth. The theoretical background clearly indicates difficulties in classifying areas, variables and adopted EKC functions. These problems, still visible in national economies, require constant monitoring and evaluation of the conducted research in order to improve the functioning of a given policy.

Also, technological changes resulting from adaptation to the adopted regulations will have their consequences in terms of greenhouse gas emissions with a certain delay. This effect may not always be beneficial, which is less noticed. Then the transformations in the form of the N curve, inverted N curve, M curve and other combinations were considered. Depending on the results obtained, it is possible to compile recommendations regarding pro-environmental policies in the field of CO₂eq reduction. In previous studies, it is more and more often pointed out that the shape of the EKC can be described more effectively by means of a polynomial (N curve or other more diverse shapes) as seen in the following: the Day and Grafton [24] study involving Canada, the Akbostancı et al. [25] study involving Turkey, the Lee et al. [26] study involving 89 countries, the Lopez-Menendez et al. [27] study involving 27 EU countries, the Yang et al. [28] study involving the M-shaped curve for East Asia and the Pacific, the Moghadam and Dehbashi [29] study involving Iran, the Shahbaz et al. [30] study involving BRICS countries, and the Wang et al. [31] study involving 62 underdeveloped countries. Such results reveal the difficulty in embarking on the paths of sustainable development and bring the question of the rate of growth itself under consideration. The presented research results indicate significant discrepancies between the authors. However, it is difficult to find an explanation for these differences. However, it is of significant importance in the process of creating regulations by state institutions, which are not so easily reversible. It can therefore lead to wrong actions in the long run. In order to stay on the long-term growth curve, it is therefore necessary to define the precise shape of the EKC and the position of a given country or region. Here, the bifurcation points of the existing theory become visible; the reasons for which can be found in the structures of economies, adopted national policies, social awareness or institutional factors. The revealed research gap is therefore also important from the point of view of EU policy in the context of reducing CO₂ emissions. The adopted form of estimation is also not without significance. Selecting the appropriate form of the model while indicating the appropriate EKC determinants is the main aspect of conducting research on the EKC. In this context, the literature provides numerous examples and applications.

Various methods have been used in the study of the environmental Kuznets curve. The environmental Kuznets curve hypothesis investigated with a dynamic panel model has found application in many studies of environmental impacts, which have become a tool for the evaluation and implementation of pro-environmental policy in various regions. Applying the dynamic panel model to the 28 provinces of China from 1996 to 2012 [32], using the generalized moment method (GMM) estimator and the autoregressive distributed delay (ARDL) model with alternative panel estimators, confirmed the existence of the environmental Kuznets curve. The EKC hypothesis was not confirmed in the studies of Sirag, Matemilola, Law and Bany-Ariffin [33] in countries developing within a non-linear structure.

In the studies on the EKC curve, classical regression methods, ratio analyses and, relatively often, static panel methods were used. However, dynamic panel models were relatively rarely used. Dynamic models estimated on panel data are characterized by the influence of the time factor on the dependent variable, which is taken into account by using the lagged values of selected variables as regressors. This is particularly important for demonstrating the transformations of the studied phenomenon (which was important in this study) and for forecasting. Dynamic panel models require specific estimation methods. In the case of a dynamic model, the estimator characteristic of the static methods loses its consistency, which introduces appropriate difficulties, a different way of approaching the problem and the use of other tests of consistency.

Central and Eastern European Countries have been studied in different periods and with different methods, taking into account different explanatory variables and the inconsistent results of the EKC (

Table 1. Descriptive statistics of CO₂eq emissions (tons of CO₂ equivalent per capita) in the examined countries.

Descriptive Statistics	Bulgaria	Croatia	Latvia	Lithuania	Hungary	Poland	Romania	Slovakia
Mean	43,791.35	18,352.81	7280.25	11,678.63	46,495.89	307,790.8	75,550.00	29,686.65
Median	43,769.33	18,617.19	7187.76	11,690.30	45,078.29	307,524.14	74,908.10	30,030.91
Std. Deviation	3101.56	2233.44	576,00	782.70	5394.35	8682.63	9194.61	3008.47
Minimum	38,945.46	15,634.59	6551.72	10,603.36	39,146.90	290,378.94	65,126.40	25,534.82
Maximum	49,146.88	21,991.29	8342.14	12,824.56	54,621.87	317,830.28	90,002.35	34,322.11
25th percentile	41,886.88	16,349.13	6773.11	11,001.57	42,281.92	303,804.52	66,955.88	27,305.10
50th percentile	43,769.33	18,617.19	7187.76	11,690.30	45,078.29	307,524.14	74,908.10	30,030.91
75th percentile	44,835.60	20,484.45	7731.11	12,515.83	51,761.68	315,982.01	84,679.35	32,255.46

). In Destek’s study [34], the impact of the dimensions of economic, social and political globalization on environmental pollution has been examined, taking into account real income and energy consumption. The data from the 1995 to 2015 period were analyzed here with the use of second-generation panel data methods for the CEECs. It was then shown that economic and social globalization increases carbon dioxide emissions, while progressive political globalization reduces environmental pollution. International agreements on the reduction of emissions worked effectively in the analyzed group of countries. The occurrence of the EKC was also confirmed. The inverted U shape of the EKC was also shown in Antici’s paper [35], where the impact of such factors as GDP per capita, energy consumption per capita and trade openness on CO₂eq emissions per capita in the countries of Central and Eastern Europe from 1980 to 2002 was analyzed. Per capita energy consumption has a significant worsening effect on emission levels. The growing demand for energy is met with pollution-generating technologies and not environment-friendly ones. As indicated, environmental awareness and the will to protect the environment starts at the turning point of USD 2077–3156 in real GDP per capita. This value means that environmental awareness in the region examined starts at an earlier stage than in developed countries, therefore some developing countries may have a lower environmental lag value than other developed countries. In Adedoyin et al. [36], the relationship between energy production and CO₂eq emissions in three blocks of countries was analyzed: Central and Eastern Europe Countries (CEECs), the Commonwealth of Independent States (CISs) and New Member States (NMSs) from 1992 to 2014. The study used the GMM model for variables such as CO₂eq emissions, real GDP, renewable energy production, non-renewable energy production and natural resource revenues. The empirical results show that with the increase in renewable energy production, CO₂eq emissions in the CISs and Central and Eastern European countries increased, but CO₂eq emissions decreased by 0.02 in the NMS countries. The results for NMSs and the CISs are consistent with the inverted U-shape for the EKC hypothesis. There was also a difference in the level of environmental degradation between the analyzed groups of countries. Therefore, there is a need to apply appropriate national and international policies to decarbonize the natural environment.

In light of the presented conditions, this study undertook to explain the discrepancies in the shape of the environmental Kuznets curve with the use of a dynamic panel analysis, which made it possible to consider better the changes taking place. The research concerned a group of Central and Eastern European countries in which there was not consistent research. This paper presents the alternation of the changes taking place and shows that the morphological differences of these changes in individual countries resulted from disproportions in the energy efficiency of the economic systems. The premises for conducting a regulatory policy appropriate for a given stage of the EKC were also indicated. This research took into account additional indicators not previously recognized, taking into account social conditions (Severe material deprivation rate) and changes in the structure creating an operating surplus in economic entities (Value added at factor cost and Gross operating surplus). These indicators illustrate the propensity to invest in new technologies in terms of the distribution of state income and changes in income generation of the entities. This creates opportunities to search for the optimal relationship between GDP and CO₂ emissions, to select the appropriate instruments for implementing long-term goals for sustainable development and the application of a taxonomy of similarly shaped economies at a later stage will create the possibility of determining the relationships between the development of an economy and the shape of its EKC. The analysis conducted aims to search for the appropriate model of the GDP–CO₂ relationship, taking into account the limitations in the growth potential in the countries of Central and Eastern Europe. Thus, it is possible to place economies on a long-term growth path while maintaining care for the environment. This study therefore aims to verify the research hypothesis about the existence of a relationship between greenhouse gas emissions and N-shaped GDP per capita for the CEECs. In addition, the factor of energy efficiency in the socio-economic system and the adopted path of economic growth will be verified.

3. Materials and Methods

There are numerous hypotheses in the literature today about the level of environmental degradation and per capita income that take the functional form of the inverted U [37,38], resembling income inequality and per capita income in the original Kuznets curve. In the case of CO₂eq emissions, the overall model record assumes the following form [38]:

$${CO}_{2}emission=Const+{\alpha }_1{GDP}_{pc}+{\alpha }_2{GDP}_{pc}^2+e,$$

(1)

where CO₂ emission—CO₂eq emission level, GDPpc—GDP per capita, GDPpc²—square of GDP per capita, α1 and α2 are the regression parameters, e is an error and Const—constants.

In this approach, a negative and statistically significant value of GDP² per capita will indicate the presence of the environmental Kuznets curve. It will therefore be possible to define a certain level of threshold emissions, after which GDP will continue to increase and emissions levels will decrease.

It is postulated that this relation, in a reduced form, has the functional form of an inverted U and is estimated for each environmental quality indicator based on panel data for different countries/regions. The processes taking place in the economy are dynamic. Nevertheless, there is still a tendency in research practice to use static analyses that do not take into account the changes in the phenomena under examination over time. In time series regression models, it is common practice to deal with them by including the delayed values of accompanying variables of the dependent variable or introducing delays for both types of variables in the specification [34]. The introduction of such a delayed explanatory variable, however, leads to its correlation with group effects that are constant over time. This is the main reason why traditional panel models examining phenomena in static terms cannot be used to build a dynamic panel model as the obtained estimators would be inconsistent and biased [36]. In addition, there is the problem of endogenous selection using panel data [39]. An alternative approach to dealing with dynamic processes is to use a method developed in the context of samples with a small number of time series observations. This requires an estimation equation and differentiates them to transform the country-specific effects, which enables the dynamic specification of the differences with a lagged dependent variable.

The general form of the dynamic panel model can be recorded as [40]:

$$y_{it}={\mathsf{\delta }\mathrm{y}}_{i,t-1}+x_{it}^{’}\beta +u_{it}, \left|\mathsf{\delta }\right|<1,$$

(2)

$$u_{it}={\mu }_i+v_{it},$$

(3)

where y_it is the realization of the explanatory variable for the i-th object in the t-th period, ${\mathit{\delta }\mathit{y}}_{\mathit{i},\mathit{t}-1}$ is the regression parameter (δ) delayed by 1 time period for the explained variable (y) of the i-th object in the panel, ${\mathit{x}}_{\mathit{i}\mathit{t}}^{\mathit{’}}$ is the vector of the coefficients of x for the object i at time t, β is the structural parameters of the model, u_it is the constituent u which reflects the group effect in a random or non-random form and the error criterion is treated as white noise, u_i is the individual effect that is unobservable and not included in the regression equation and is specific to a given test unit and v_it is the remainder, purely random part of the random component [41].

Dynamic panel data introduce a non-correlation condition. To achieve this, the autocorrelation should be evaluated, and a decision should be made whether to apply the Arellano–Bond test. The probability of Ar (2) (pr > z) should be expected to be insignificant at 5%. This rejection applies when the probability pr > z is greater than 0.05, i.e., the error term is not serially correlated. This is confirmed by the lack of serial autocorrelation in errors. Typically, Ar (1) should be significant at 5% (AR (1) pr > z < 0:05) [42].

In addition to the curve shape estimation for the respective countries, a two-stage analysis was carried out. The curve fitting coefficient (R-square) and the size of the standard error of the residuals were used for assessment purposes. The higher level of the R-square demonstrated the form of the equation in the first step, and the lower value of the standard error of the residuals demonstrated the value in the second step. In the event that both assessments suggested the same shape of the EKC, it was considered correct; otherwise, the result was deemed indecisive („no”). This made it possible to demonstrate the level of heterogeneity of the countries.

4. Results and Discussion

Based on the literature review, a dynamic panel model was used to study the impact of CO₂eq emissions on countries that had undergone a political transformation in Eastern Europe, namely Bulgaria, Romania, Croatia, Hungary, Poland, Lithuania, Latvia and Slovakia. The analysis was concerned with the 2004–2018 period. The dependent variable was CO₂eq emission (tons of CO₂eq equivalent per capita), while the group of independent variables was represented by GDP per capita (GDP), square GDP per capita (GDP²) and cubic GDP per capita (GDP³), Solid fossil fuels (Final consumption–Energy use), Primary Energy consumption, Population, Energy productivity, Severe material deprivation rate, Value added at factor cost and Gross operating surplus. Energy efficiency as an indicator measures the amount of economic product that is manufactured per unit of gross energy available. Gross energy available represents the quantity of energy products necessary to meet the entire demand of the entities within the geographical area under consideration. The economic result is reported as a euro unit in chain amounts up to the reference year 2010 at the 2010 exchange rates or in the PPS (Purchasing Power Standard) unit. The former is used to observe the evolution over time for a specific region while the latter allows the comparison of the Member States in a given year.

There is a significant variation between countries in the base period (2004). In recent years (2014–2018), an increase in greenhouse gas emissions was again observed in the surveyed countries (except Romania). The highest levels of CO₂eq emissions per capita were recorded in Romania and Hungary, and the lowest in Latvia and Lithuania (Figure 1). In the vast majority of the studied countries, there was a significant reduction in these emissions (except for Poland); however, the reduction path implemented was varied. The largest fluctuations were in Poland and Romania, and the lowest in Latvia and Lithuania (coefficient of variation is in Table 1). At the same time, a sustained improvement in energy productivity was observed in all countries. In the case of this group of countries, we are faced with the so-called catch-up effect, which may mean increasing CO₂eq emissions in order to accelerate the pace of economic growth. On the other hand, according to the growth models of Romer or Harrod–Domar, maintaining growth based on the intensity of using the quantitative factors encounters an efficiency barrier. Consequently, it requires greater consideration of innovations in manufacturing processes, including energy-saving technologies and the use of renewable energy sources.

In search of explanations for the demonstrated changes, a model equation was developed. Subsequently, all variables were log transformed and a panel model of the AR1 order was constructed. In each case, the obtained matrix was non-positively determined and a One Step analysis was applied. A similar approach was adopted, among others, in the study by Jalil [43]. Due to the lagged dependent variable, it is reasonable to use the Generalized Moment Method (GMM). It allows for providing solutions to the endogeneity error, as well as to control the individual and temporal effects [44]. In the analyzed case, the dependent variable was delayed by one period, and the model hypothesis assumed the following form:

$$\begin{array}{cc}\hfill {CO}_{2}emission=& {\propto }_0+{\mathsf{\delta }}_1{CO}_{2}emissio{n}_{-1}+{\beta }_{1}Solid fossil fuels\hfill \\ & +{\beta }_{2}Primary energy consumption+{\beta }_{3}Population\hfill \\ & +{\beta }_{4}Energy productivity\hfill \\ & +{\beta }_{5}Severe material deprivation rate\hfill \\ & +{\beta }_{6}Value added at factor cost\hfill \\ & +{\beta }_{7}Gross operating surplus+{{\beta }_{8}GDP}^2+{\beta }_9{GDP}^3 \hfill \end{array}$$

(4)

where CO₂eq emission—CO₂eq emission (tons of CO₂ equivalent per capita), CO₂ emission-1—CO₂eq variable delayed by one period, GDP²—square real GDP per capita and GDP³—cubic real GDP per capita (GDP³).

First, the RE model was tested and the occurrence of cross-sectional autocorrelation was assessed. In the panel adopted for the 84 observations and eight groups, the following results of the Pesaran test were obtained:

Pesaran’s test of cross-sectional independence = 0.133, Pr = 0.8940.

Average absolute value of the off-diagonal elements = 0.312.

A high probability value allows the exclusion of the hypothesis about the occurrence of cross-sectional autocorrelation. For the dynamic model, the delayed dependent variable for one period was used. There must be an autocorrelation in the first period. In order to make sure that there was no autocorrelation for the second order, the Arellano–Bond test was used:

Arellano–Bond test for AR (1) in first differences: z = −2.33 Pr > z = 0.020.

Arellano–Bond test for AR (2) in first differences: z = −0.88 Pr > z = 0.378.

As expected, introducing a delay of the CO₂eq variable by one period will cause autocorrelation, which is a consequence of the action in the dynamic panel. It is important, however, to exclude the occurrence of second-order autocorrelation. The Arellano–Bond for AR (2) test excludes the occurrence of second-order autocorrelation. The dynamic nature of the model was assumed, and the results are presented in

Table 2. Dynamic panel model of CO₂eq emissions.

Variable Name	Coef.	Std. Err.	z	p > |z|	[95% Coef.	Interval]
CO2eq energy
L1.	0.1574	0.0638	2.47	0.014	0.0234	0.2825
Solid fossil fuels (Final consumption–Energy use)	0.1006	0.0187	5.38	0.000	0.0640	0.1373
Primary energy consumption	0.3367	0.1381	2.44	0.015	0.0661	0.6074
Population	0.5076	0.1178	4.31	0.000	0.2767	0.7386
Energy productivity	−0.2871	0.1445	−1.98	0.047	−0.5709	−0.0034
Severe material deprivation rate	−0.0023	0.0013	−1.72	0.086	−0.0048	0.0003
Value added at factor cost	−0.2076	0.1079	−1.93	0.054	−0.4191	0.0038
Gross operating surplus	0.2152	0.0740	2.91	0.004	0.0701	0.3602
GDP2	0.1732	0.0876	1.98	0.048	0.0015	0.3448
GDP3	−0.1172	0.0061	−1.93	0.052	−0.0236	0.0002
cons	−5.8621	3.9081	−1.5	0.134	−13.5218	1.7977

L1 represents the dependent variable delayed by 1 period.

.

Comments and explanations:

The number of observations = 84, number of groups = eight, number of instruments = 86, Wald χ² (10) = 7694.22, Prob >χ² = 0.000. Additionally, the Sargan test was performed and the over-identification of Chi-square = 95.39113 [0.0562].

A low value of the estimated probability indicates the significance of the dynamic model. In addition, the Sargan test value exceeds the 5% threshold, which indicates the right amount of instruments was used and there was no over-identification problem. The issue of time series also includes the issue of panel stationarity. It is assumed that the model should be stationary, so that the dynamics of the variable in the long run tends to the equilibrium state. The Levin–Lin–Chu unit-root test for the dependent variable was used to assess the stationarity (

Table 3. Stationarity assessment.

Tests	Statistic	p-Value
Unadjusted t	−3.6403
Adjusted t	−1.6760	0.0469

).

The low p-value allows us to reject the hypothesis about the existence of the unit root. The resulting panel is stationary. In order to confirm the existence of a long-term relationship, the Kao test for the occurrence of cointegration was used. In this model, the following hypotheses were adopted: H0 would have no cointegration and in H1 all panels would be cointegrated. The results of the test are presented in

Table 4. Cointegration tests.

Tests	Statistic	p-Value
Modified Dickey–Fuller t	−4.1758	0.0000
Dickey–Fuller t	−5.8832	0.0000
Augmented Dickey–Fuller t	−1.9383	0.0263
Unadjusted modified Dickey–Fuller t	−4.9085	0.0000
Unadjusted Dickey–Fuller t	−6.0581	0.0000

.

The obtained results give grounds for rejecting H0 and assuming that the phenomenon of cointegration occurs.

The valid instruments are uncorrelated with the error term. In case we have more instruments than necessary, L > K, we can perform a so-called J-test for overidentifying restrictions. This tests whether the Instrumental Variables of all the instruments are exogenous, assuming that a least one of the instruments is exogenous [45].

Therefore, two tests were introduced: the J test (Davidson–MacKinnon) for the non-nested models and Cox–Pesaran test for the non-nested models. In order to run the test, it is required to create two model hypotheses based on an alternative model:

$$y=\left(1-\lambda \right)X\beta +\lambda \left(Z\gamma \right)+e$$

(5)

In this model, a test of λ = 0 would be a test against H1. The problem is that λ cannot be separately estimated in this model; it would amount to a redundant scaling of the regression coefficients. Davidson and MacKinnon’s J test consists of estimating γ by a least squares regression of y on Z followed by a least squares regression of y on X and Zγ, the fitted values in the first regression. A valid test, at least asymptotically, of H1 is to test H0: λ = 0. If H0 is true, then plim λ = 0 [46]. Model 1 contained a lagged GDP regressor, and Model 2 a lagged CO₂eq variable. In both tests, M1 is rejected in favor of M2 for the adopted confidence level of 0.05 (

Table 5. J test for non-nested models.

	Dist	Stat	p > |Stat|
H0:M1/H1:M2	t(72)	5.87	0.000
H0:M2/H1:M1	t(72)	−1.41	0.163

and

Table 6. Cox–Pesaran test for non-nested models.

	Dist	Stat	p > |Stat|
H0:M1/H1:M2	N(0,1)	−60.69	0.000
H0:M2/H1:M1	N(0,1)	1.02	0.153

).

The presented results indicate that the relationship between CO₂eq and GDP per capita for the CEECs is shaped as an inverted N curve. Therefore, one cannot speak of the durability of the ongoing transformations. After exceeding a certain growth threshold (different for each of the analyzed countries), CO₂eq emissions increased again. This confirms the first of the hypotheses for the entire panel. There is therefore no permanent equilibrium, but a transition between different periods. The ongoing structural, technological and institutional changes may gradually lessen CO₂eq reduction and reactivate the positive relationship between economic growth and CO₂eq emissions. As in the research by Shahbaz [47], a certain heterogeneity of the results can be seen. However, when analyzing individual countries, it can be read as non-synchronous passages through individual points, which points to the cyclical nature of the changes in the relationship between CO₂eq and GDP. On the basis of previous studies, one can notice the heterogeneity of the aforementioned relationship at the level of individual regions [36,47,48]. Thus, a universal policy addressed to the entire region may turn out to be ineffective. In each case, it is necessary to analyze the specific conditions and scope of the changes made as well as the social and economic expectations. The asymmetry of the changes indicated the results from structural, technological institutional and social were disproportionate. Demographic (population size) and technological changes (primary energy consumption and energy productivity) are of particular importance. An important supplement is the economic and social conditions represented by the other factors included in the model (Table 2).

In the case of the analyzed explanatory variables, it is possible to indicate the expected directions of their impact on CO₂eq emissions. The increase in energy productivity should neutralize CO₂eq emissions. The same assumption was made in the work of Ding et al. [49] and the econometric analysis confirmed the correctness of the assumption with high-quality estimators. An increase in energy efficiency causes a decrease in the unit amount of energy used for production purposes and also helps to reduce energy costs. This shows the great role of energy innovations, which must be supported by state regulations and fiscal policy. In addition, the role of human capital in overcoming technological barriers and implementing modern solutions in energy production is important.

From an international perspective, energy productivity reduces the import of fossil fuels, which in turn reduces CO₂eq emissions. This confirms the adopted hypothesis. In the analyzed countries, the improvement of energy efficiency decreased, which indicates difficulties in reducing CO₂eq emissions with economic growth. However, differentiation in the changes in economic structures resulted in the pace and path of changes in CO₂eq emissions being different in individual countries.

Demographic factors are of great importance here, the synthetic measurement of which is population changes. The positive expected value of the population coefficient is a logical consequence of the increase in population demand for the use of energy and fuels that increase CO₂eq emissions. This was confirmed in the work of, among others, Evrendilek and Ertekin [50]. Due to the demographic changes taking place, the role of this factor will decrease in the analyzed countries.

Table 7. Domestic issue models—N or U-shapes.

Country	Square Models	Cubic Models	R-Square		Standard Error of Residuals		Decision
			Square	Cubic	Square	Cubic
Bulgaria	Y = −2368.74 + 457.224GDP − 21.9645GDP2	Y = 148,978−43,113.6GDP + 4159.17GDP2 − 133.742GDP3	0.2250	0.2294	0.0670	0.0697	no
Croatia	Y = −205.168 + 46.6045GDP − 2.52564GDP2	Y = 212,435−68,645.6GDP + 7394.16GDP2 − 265.482GDP3	0.2013	0.1578	0.1295	0.1197	no
Latvia	Y = −19.8441 + 6.51252GDP − 0.368380GDP2	Y = −9.32016 + 3.10167GDP − 0.0132580GDP3	0.1788	0.2329	0.0764	0.0771	Inverted N-shaped
Lithuania	Y = −34.6567 + 9.81785GDP − 0.546836GDP2	Y = −4180.39 ** + 1363.13GDP ** − 147.767GDP2 ** + 5.33712GDP3 **	0.4804	0.6559	0.0520	0.0442	N-shaped
Hungary	Y = 449.649−93.8203GDP + 5.01286GDP2	Y = 4978.70−1555.68GDP + 162.289GDP2 − 5.63993GDP3	0.2454	0.2460	0.1079	0.1127	no
Poland	Y = 28.2025−3.38863GDP + 0.184371GDP2	Y = −3710.83 + 1221.69GDP ** − 133.585GDP2 ** + 4.86791GDP3 **	0.0343	0.4232	0.0301	0.0243	N-shaped
Romania	Y = 8.22458 + 1.24296GDP − 0.102537GDP2	Y = −4957.61 + 1698.40GDP − 193.393sqGDP + 7.33602GDP3	0.5775	0.6455	0.0842	0.0806	N-shaped
Slovakia	Y = −14.7208 + 5.93510GDP − 0.347902GDP2	Y= −8577.13 + 2738.11GDP − 290.901GDP2 + 10.2978GDP3	0.7656	0.8466	0.0527	0.0445	N-shaped

** p < 0.05.

presents the individual Kuznets curves for the analyzed countries.

The approximate shapes of the Kuznets curves are characterized by the N shape (for Lithuania, Poland, Romania, and Slovakia) and inverted N (for Latvia). Such a negative value for the time-delayed CO₂eq variable was also predicted in the work of Leitão and Shahbaz [51]. However, the changes occur as a function determined by the exponents of the GDP variables. A gradual decrease in the number of inhabitants was recorded (in 2018, compared to 2010, the decrease in the number of inhabitants was as follows: for Bulgaria 5.01%, for Croatia 4.59%, for Latvia 8.78%, for Lithuania 10.6%, for Poland 0.12%, for Romania 3.76% and for Slovakia + 0.98%) and according to long-term forecasts, the number of inhabitants will further decrease, which will be a factor limiting the level of CO₂eq emissions. In addition, due to the effects of convergence on the pursuit of accelerating the pace of economic growth [52] and trying to catch up with the European Union countries, higher energy demand will constitute a negative effect. In recent years, most of these countries have shown an above-average growth rate (even 2.9% in the case of Romania in 2016) [https://data.oecd.org/gdp/real-gdp-forecast.htm#indicator-chart accessed on 23 April 2023] and this is a factor that will have a negative impact on the level of CO₂eq emissions. This factor is currently a key barrier to achieving carbon neutrality in the surveyed countries. Economic aspirations and the nature of socio-economic growth are in conflict with environmental effects. This is a serious problem for the studied group of countries. It requires an acceleration of the structural changes related to the technologies for obtaining and using energy.

Based on the research conducted, it can be seen that in most of the studied countries the N shape relationship was revealed (as seen in Table 6 for Lithuania, Poland, Romania and Slovakia). The renewed increase in emissions was not high, still, it did occur. This was similar to the results of Saud et al. [53] but different from the results of Destek et al. [34] and Adedoyin et al. [36]. The opposite (N-inverted) shape was found only in Latvia. For the remaining countries, it was difficult to clearly estimate the shape of the EKC. Due to the significant burden of energy generation with mine raw materials, it was relatively easy to reduce emissions in the initial period, which resulted in a transition to the “falling” part of the EKC curve and an interpretation confirming the existence of this curve. In the longer term, this requires further improvement in the efficiency of energy use, investments in technological innovations and the high installation costs of pro-ecological low-emission solutions (e.g., heat pumps, photovoltaics). However, this requires higher initial costs and often, especially for households but also in the case of some enterprises, co-financing from the state budget. This stage is implemented differently, hence there is a change of direction in the relationship. Initial effects resulting from structural changes and foreign investment were gradually exhausted in most of the countries surveyed. The effects of globalization and integration were less important, while the economic and social costs of further emission reductions emerged (e.g., liquidation of lignite and hard coal mines and energy based on these fuels as well as the use of traditional furnaces in residential buildings) [54]. Achieving a further reduction of CO₂eq emissions requires the elimination of many energy sources based on fossil fuels and their replacement, which, in turn, is associated with the need for co-financing, due to the existing financial barriers and improving the efficiency of energy use [55]. Technological innovation is of key importance in the development of renewable energy sources in the context of threats resulting from the low-altitude emissions in the rural areas in Poland. When a country becomes suitable for foreign capital investments, which are the source of technological innovations, a growth path in line with the shape of the inverted letter U and the transition to its falling part is achieved. Joining the EU accelerated this process with the deeper economic crises constituting serious challenges. Then the importance of the economic objectives increased again at the expense of the environment, which was observed in many countries [56]. This was especially true of the 2007–2008 financial crisis. In countries reporting the N-shaped EKC, the mechanisms of change were very similar. Reduction of CO₂eq emissions during the financial crisis of 2007–2008 (Lithuania, Poland, and partly Slovakia—with the exception of Romania—experienced a permanent decline and only recently a reversal of this trend). Thus, external shocks strongly affect the relationship under study, and asymmetric shocks may be a problem. Then, the acceleration of GDP per capita growth from 2014 to 2015 increased energy demand (increased in emissions and often large fluctuations (in Poland and Slovakia) while slowing down the improvement of energy productivity (in Lithuania, Poland, Romania and Slovakia). Additionally, large fluctuations in this productivity were recorded in Poland. In contrast, in Latvia, a sustained increase in energy productivity was maintained, which led to an inverted N-shaped EKC. Thus, there are barriers that exhaust the previously achieved higher level of improvement in energy productivity. At the same time, a negative influence factor was the acceleration of the GDP growth rate per capita above the average value (increased the importance of the catch-up effect). High instability is related to technological changes and their implementation as well as the economic sensitivity of a society to changes in energy prices.

Models of the transition of the individual phases in the studied countries are slightly different. The final shape is similar, but the cycle has a more complex morphology due to its inherent ability to change and the diversified national policies as well as economic and social structures. This is indicated by the existence of the diversified growth paths and structural transformations taking place in the studied countries, although the economic and social situation is relatively similar. At the same time, it means that many of the existing stimuli ensuring structural changes have weakened.

5. Conclusions and Policy Implications

The research conducted for the CEECs confirmed the existence of the relationship between greenhouse gas emissions and N-shaped GDP per capita. This leads to conclusions about the alternation of this relationship for the analyzed countries and the existence of a number of turning points. It means that the classic EKC has only a short-term dimension. At the same time, it requires a continuous application of changes in both fiscal and regulatory policy in order to achieve carbon neutrality.

• The key factor in these countries is the efficiency of energy use in the socio-economic system and the adopted path of economic growth. The acceleration of economic growth and the observed slowdown in the improvement of energy productivity resulted in the N-shaped curve in some of the surveyed countries. At the present stage, it is precisely the improvement of energy efficiency and related energy innovations. Demographic factors will lead to the reduction of CO₂eq, while the above-average GDP growth rate will result in an increase in energy demand as it poses a threat to increasing CO₂eq emissions. This is due to the catch-up effect. This requires both institutional changes and statutory regulations, as well as financial support to ensure the quick pace of transformations in the structure of energy generation in the studied countries and to reduce the importance of fossil fuels. Changes in the level of human capital and their relationship with the improvement of living conditions are of significant importance. These changes are very important for the role of enterprises and the transformations taking place in them towards reducing CO₂eq emissions. This allows us to resolve the contradiction between the growth possibilities reflected in the gross operation surplus and changes in the structure of production factors leading to the implementation or creation of pro-environmental innovations in the studied area.
• This research showed that the alternation of relations between GDPs per capita and the amount of CO₂eq emissions in the analyzed period was different than the traditional EKC. However, the morphology of the changes in individual countries is different, which results from the demographic transformations and changes in the energy efficiency of the economies of the studied countries. At the same time, the studied countries are at different stages of the cycle, which means that a universal policy of reducing CO₂eq emissions will not be effective.
• The N-shaped EKC was found for Lithuania, Poland, Romania and Slovakia. In these countries, the effects of the structural changes made so far have been exhausted. In their case, it is necessary to increase the programs supporting the growth of innovation and combining public and private funds in order to accelerate technological changes and changes in economic structures, while the mere tightening of regulations limiting CO₂eq emissions may result in a slowdown in GDP growth.
• At the current stage of research, no universal approach or compromise has been developed as to the shape of the EKC. It seems that structural factors may still play a significant role. The differences in the studies of Antici [35], Adedoyin [36] and many other researchers on the shape of the EKC suggest the need for further analyses of the ability of regions to reduce CO₂ emissions while maintaining an increasing rate of economic growth. Moreover, as indicated by studies [24,25,26,27,28,29,30,31], there are still discrepancies as to the shape of the EKC, which indicate the risk of taking incorrect political actions in the long term.
• We are faced with emerging economic and social constraints on the path of reducing CO₂eq emissions. There is a serious problem with the middle-income trap characterized by a lower share of technological innovations and, consequently, income barriers, which may limit the favorable changes that occurred in the previous period. It also points to a certain cyclical nature of the changes, at least at the present stage of development. Re-breaking the identified direction of change requires increasing the role of innovation and human capital in improving energy efficiency and the faster growth of low-emission or zero-emission energy sources. This type of relationship suggests that GDP growth alone will not be a remedy for environmental pollution problems, at least for some of the countries surveyed, and it is not possible to adjust economic structures, as doing so will not keep pace with the expected and actual rate of economic growth, as many of the structures that still exist are heavily dependent on conventional energy sources (i.e., fossil fuels). In the long run, economic growth and the increased energy consumption associated with it threaten the environment. It will be difficult to achieve carbon neutrality without changes in the regulatory and fiscal policies of the state, as the current incentives to reduce CO₂eq emissions have been exhausted. Further research should focus on diagnosing the factors causing transitions between the individual phases in the presented cycle of transformations. Please note that these factors will vary from period to period. At the same time, it is worth paying more attention to the technological changes and their long-term effects on the economy.
• The differences in the shape of the EKC in different regions may facilitate the development of taxonomies and groupings of areas with a coherent EKC, which will contribute to the proper selection of variables included in the EKC and will determine the level of development of the economies, which will make it possible to apply a specific type of economic policy. Moreover, in order to take into account the heterogeneity at the level of the economies, it is worth considering political and cultural factors in the procedure of selecting variables, which can be an intellectual contribution to the further development of the theory and research on the EKC.