Economic Growth and CO₂ Emissions in Croatia: An ARDL-Based Assessment of the EKC Hypothesis

Abstract

This paper examines the long-run relationship between economic growth and CO₂ emissions in Croatia over the period 1990–2023 using the ARDL bounds testing approach. The analysis aims to assess the presence of an Environmental Kuznets Curve (EKC) and to shed light on Croatia’s position along the growth–emissions trajectory, an issue that has remained inconclusive in earlier studies. The results provide evidence of an inverted U-shaped relationship between the GDP per capita and CO₂ emissions, consistent with the EKC hypothesis. The estimates of marginal effects suggest that the impact of income on emissions weakens and may eventually turn negative at higher income levels, although the precise income level at which this transition occurs is sensitive to model specification and sample composition. Energy consumption emerges as the strongest long-run driver of emissions, while a higher share of renewable energy contributes significantly to their reduction. Institutional quality is found to be positively associated with emissions in the long run, reflecting growth-enhancing effects during the post-transition period rather than immediate environmental improvements. The contribution of this study lies in the use of a longer time span and a dynamic empirical framework that allows for a more nuanced assessment of the growth–emissions relationship in Croatia. Overall, the findings point to a gradual decoupling of economic growth from carbon emissions while highlighting that the sustainability of this trajectory depends critically on continued progress in the energy transition and on the alignment of institutional development with climate and energy objectives.

1. Introduction

Over the past three decades, the topic of global warming and climate change has become increasingly prominent across political, educational, scientific, business, and private spheres. There is not a single segment of society that has remained unaffected by the changes and adaptations required in response to global warming and climate change. This is unsurprising, given that, according to the most recent report of the Intergovernmental Panel on Climate Change [1], humans have largely caused global warming through their activities, particularly through the emission of greenhouse gases, especially CO₂. Recent analyses show that all eleven warmest years on record occurred between 2014 and 2024, with 2024 being by far the warmest year and 2023 the second warmest [2]. Most of this warming and harmful emissions result from the burning of fossil fuels, which remain the most important and widely used source of energy at the global level [3].

At the international level, the response to environmental challenges, global warming, and climate change has been articulated within the United Nations framework through the adoption of 17 global Sustainable Development Goals and the UN’s 2030 Agenda [4]. At the European level, the European Union, through the European Green Deal [5], is already working toward making Europe the first climate-neutral continent by 2050, which implies that it aims to achieve net-zero emissions by then. This goal is intended to be achieved primarily through reducing greenhouse gas emissions, investing in green technologies, and protecting the natural environment.

To achieve these, countries are obliged to reduce their environmentally harmful emissions, but the emissions reductions should not damage their economic growth. Countries face a challenge of designing a growth trajectory that does not harm the environment. That is precisely why the relationship between CO₂ emissions (as the major pollutant) and economic activity remains a central research topic.

This paper examines whether the characteristics of the Environmental Kuznets Curve (EKC) hypothesis are present in Croatia. The EKC hypothesis posits that in the early stages of development, economic growth and emissions tend to increase simultaneously, while after a certain income threshold, emissions begin to decline even as growth continues. In other words, the goal is to assess whether Croatia is still located on the “rising” segment of the curve [2] or has already transitioned into a phase where growth no longer leads to higher emissions, the so-called “falling” phase of the curve.

In accordance with the above, two research questions are posited in this paper. The first research question is the following: in which phase of the Environmental Kuznets Curve (EKC) does Croatia belong to during the 1990–2023 period? The second research question is as follows: has Croatia already reached the income turning point after which CO₂ emissions decrease with positive economic growth rates?

Two hypotheses posed in this paper arise from the research questions:

H1: 

Croatia exhibits a nonlinear, inverted-U-shaped relationship between economic growth and CO₂ emissions, as predicted by the Environmental Kuznets Curve (EKC).

H2: 

Croatia reached an income turning point during 1990–2023, after which further GDP growth no longer increased emissions.

The EKC concept is rooted in the seminal paper of Simon Kuznets [6], who initially examined the relationship between economic growth and income inequality, and was later adapted to analyze the link between economic development and environmental degradation. In its simplest form, the concept suggests that environmental degradation increases in the early stages of industrialization and economic growth and subsequently declines once technological progress, improved institutions, and structural changes in the economy take effect. Energy consumption and CO₂ emissions are frequently used indicators in this context, especially in countries where fossil fuels dominate the energy mix.

Although the EKC hypothesis has been extensively explored in international research, the findings for Croatia remain inconclusive. Several studies (e.g., [2]) argue that Croatia has not yet entered the declining part of the curve and shows no apparent signs of decoupling. Jošić, Jošić and Janečić [7] and Škrinjarić [8] likewise fail to confirm the EKC. On the other hand, Ahmad et al. [9] find evidence of the EKC in Croatia for an earlier period. However, this leaves the question unresolved, as today’s energy and institutional conditions differ substantially from those of the 1990s and early 2000s.

Against this background of mixed and often inconclusive findings, this paper offers several contributions to the existing literature on the EKC in Croatia. First, it covers the longest time span analyzed so far (1990–2023), encompassing the entire transition period as well as more recent structural and energy-related shocks, including EU accession, the COVID-19 pandemic and the 2022–2023 energy crisis. Second, unlike earlier Croatian studies that rely on descriptive analysis, static panel models or simpler time-series techniques (e.g., [8,10]), this study applies a methodologically robust ARDL framework, which allows for the simultaneous examination of short- and long-run dynamics and provides a more reliable test of the EKC hypothesis. Third, the model extends beyond the basic income–emissions relationship by incorporating key structural, energy and institutional determinants, thereby offering a more comprehensive explanation of the drivers of CO₂ emissions. Finally, by explicitly estimating the EKC turning point and identifying the phase of the curve in which Croatia is currently located, this paper provides a more precise and policy-relevant assessment of Croatia’s growth–environment nexus than existing studies. Although this analysis focuses on Croatia, the results are relevant for other small EU transition economies facing similar growth–environment trade-offs.

For the purposes of this analysis, the data were obtained from the World Bank databases, including the World Development Indicators (WDI) and the Worldwide Governance Indicators (WGI). The model extends beyond the basic EKC specification (the GDP–emissions relationship) by incorporating the structure of the economy (share of industry in the GDP), the energy consumption, the share of renewable energy, and the institutional quality. This helps clarify what actually drives emissions in Croatia and what kind of development path the country is likely to follow.

A further research problem lies in the widespread public perception that economic growth necessarily leads to increased resource exploitation and higher levels of pollution, thereby implying that economic growth and environmental degradation are mutually exclusive. In addition, academic studies for Croatia do not provide clear or consistent answers. The aim of this paper is therefore to reexamine the EKC hypothesis using more up-to-date and comprehensive data and the latest statistical models, as well as to provide insight into how far Croatia is from a greener and more sustainable development model. The findings also serve as a basis for formulating policy guidelines within national and European climate frameworks.

This paper is structured into five sections. The Introduction outlines the topic, the relevance of the research, the research questions and the paper’s goals. The second Section presents the theoretical background and the literature review. The third Section of the paper describes the data and the variables used in the ARDL model and the methodological approach. The fourth Section presents the empirical results, the discussion, and a comparison with the findings from related studies. The final Section presents main conclusions of the paper and its policy implications.

2. Literature Review

The Environmental Kuznets Curve (EKC) hypothesis was developed on the foundations of Simon Kuznets’s original hypothesis from 1955 [6]. His paper suggests that income disparities increase during the initial phase of development, but after a certain level of per capita income, these disparities begin to decline. Following the initial paper, numerous researchers began to adapt that concept for its application in environmental economics. Grossman and Krueger [11,12] were among the first authors to apply this concept, and their studies laid the foundation for what is today known as the EKC hypothesis. Examining trade and development patterns, they observed that pollution tends to increase during the early industrial phase but gradually declines as technological capabilities and institutions advance, thus forming an inverted U-shaped curve.

After their initial contribution, the literature about the EKC grew rapidly in different directions. Early papers by Shafik and Bandyopadhyay [13] and Panayotou [14] expanded the concept and pointed out that the EKC does not necessarily appear in the same way for every pollutant or across every country. Later surveys, particularly those by Stern [15] and Dinda [16], helped consolidate the debate. They showed that the shape of the curve often depends on factors such as the energy structure of an economy, the level of technology, the institutional quality, and sometimes simply the policy choices governments make.

On the empirical side, many studies claim to find support for the EKC. Hassan et al. [17], for instance, used a panel ARDL model on 64 countries from 1970 to 2015 and found evidence that fits the inverted U-shaped pattern in both developed and developing economies. Comparable results appear in the research on China and India [18] as well as in studies for Kenya [19] and Nepal [20,21]. The findings for Turkey [22,23], Malaysia [24], and the United States [25] tell a similar story. Wang and King [25], who examine U.S. states individually, show that many states have experienced absolute decoupling between income growth and CO₂ emissions. Still, they note an interesting shift after 2015, where the curve in some states starts to resemble a U-shape again, suggesting that emissions could rise if energy or transport policies were to change. For the OECD group, Saucedo et al. [26] also observe an inverted U-shape between the GDP and CO₂ emissions between 1994 and 2014. Earlier classic work by Cole, Rayner and Bates [27] is often cited as well, though their conclusions are a bit more nuanced. They confirm the EKC for pollutants such as SO₂ and particulates, but not for NOx, and argue that policy and structural shifts, not income growth alone, explain the decline in emissions at higher income levels. A more recent study by Özmerdivanlı et al. [28] again supports the EKC, but only for advanced economies (G7), which is not the case for a group of developing countries (E7).

There is, however, an equally large body of literature that does not support the EKC. The classic paper by Holtz-Eakin and Selden [29] is often cited in this group. Using a sample of roughly 130 countries for the period 1951–1986, they found a statistically significant nonlinear (i.e., quadratic) relationship between per-capita income and CO₂ emissions. Still, the turning point for CO₂ occurs at such a high income level that most countries are nowhere near it, which essentially means that the EKC does not hold in practical terms. Wagner [30], analyzing a longer historical panel of 95 countries, also finds no evidence of an inverted U-shape and draws attention to methodological issues in earlier studies that may have produced false positives. Similar results, where the inverted U-curve does not appear, have been reported for Indonesia [31,32] and Egypt [33]. Harbaugh, Levinson and Wilson [34] likewise show that EKC findings are highly sensitive to model specification and variable selection, while Tucker [35] reports an entirely positive association between the GDP and emissions in a global sample of 137 countries across 21 years. Along similar lines, a recent study on E7 and G7 countries [28] suggests that while the EKC may hold for advanced economies (G7), developing economies (E7) have not yet reached the level of development at which environmental degradation begins to decline. In Turkey, Halicioglu [36] finds that FDI, energy consumption and income are the key drivers of CO₂ emissions over 1960–2005.

An increasing number of authors argue that income growth alone is rarely sufficient to reverse the upward trend in emissions, particularly during periods of energy insecurity or structural shocks. The recent energy-system research highlights that energy crises are often accompanied by heightened electricity price volatility and market instability, which can alter consumption patterns and temporarily intensify emissions pressures, especially in systems undergoing rapid structural change and renewable energy integration [37]. A concrete illustration of this mechanism is the temporary rebound in coal use observed in several European countries during the 2022 energy crisis, as documented by the International Energy Agency [38], which shows how quickly environmental progress can stall under conditions of acute energy-market stress.

Due to these conflicting findings in the international literature, it has become clear that the EKC is not a universal rule but rather a theoretical framework whose validity must be tested empirically for each country, taking into account its specific institutional, energy and structural characteristics. Huang et al. [39] show that the relationship between energy consumption and economic growth depends on a country’s level of development, with energy reductions potentially constraining growth in less developed economies, while advanced economies exhibit reversed causality and greater energy efficiency. This is also why a growing body of research has emerged for Croatia, showing that the results can differ quite substantially depending on the period examined, the environmental indicator used and the methodological approach.

Several studies have examined the EKC hypothesis in the context of Croatia, often reaching divergent conclusions depending on the level of aggregation and the methodological approach. Ziemblińska et al. [2], for example, analyze 11 EU transition countries over the period 1990–2023 and show that all of them, except Croatia, appear to be positioned on the downward-sloping segment of the EKC. However, their analysis relies primarily on descriptive statistics and graphical inspection rather than formal econometric modelling. A more disaggregated perspective is provided by Srdelić and Barišić [40], who examine emissions across six sectors of the Croatian economy—agriculture, industry, buildings, energy, transport, and waste—using ARDL and ECM models for the period 1995–2021. Their results indicate substantial sectoral heterogeneity: while buildings and energy-intensive industries exhibit an inverted U-shaped EKC pattern, transport-related CO₂ emissions increase almost linearly over time. Such findings are consistent with broader sector-specific evidence showing that transport emissions often follow markedly different trajectories than aggregate national emissions and frequently fail to exhibit EKC-type behavior, even in relatively advanced economies, including the EU member states [41]. This sectoral perspective highlights the limitations of national-level EKC analyses and underscores the importance of considering sector-specific dynamics when assessing the growth–emissions relationship.

Ahmad et al. [9] analyze the EKC for Croatia over 1992–2011 using quarterly data on per-capita CO₂ emissions, the GDP per capita and its squared term. Their study applies ARDL and VECM, along with additional checks such as DOLS and FMOLS. They conclude that Croatia lies on the downward portion of the EKC in the long run, while short-term links between the GDP and emissions are weaker and less stable. Since their model uses only the basic EKC structure (CO₂, GDP and GDP²), this paper goes a step further by adding energy-related, structural and institutional variables as well as dummy variables for EU accession, the COVID-19 crisis and the 2022 energy shock. The dataset also spans a much longer period (1990–2023), which captures economic, institutional and energy transitions that earlier studies could not.

Jošić et al. [7] examine the relationship between economic growth and CO₂ emissions from 1990 to 2013 using standard time-series techniques: ADF stationarity tests, Engle–Granger and Johansen cointegration tests, and linear, quadratic and cubic OLS models. They find no long-run link between the GDP and emissions, meaning the EKC is not confirmed. Relative to their paper, this study uses a more advanced methodological setup, including the ARDL approach, a broader set of variables, and additional structural breaks. It also covers more recent years up to 2023, allowing us to capture developments that older studies simply could not observe.

Škrinjarić [8] looks at Croatia’s EKC at the county level using classical panel techniques (pooled OLS and fixed and random effects) for 2008–2016 and five types of pollutants (CO₂, CO, NO₂, SO₂ and PM10). She tests linear, quadratic and cubic EKC forms and includes a simple EU-entry dummy. Her findings indicate that the EKC does not hold for any pollutant, although the EU accession led to a slight reduction in emissions. Relative to her approach, this paper goes substantially further, for the reasons previously outlined: a longer period, a more sophisticated time-series methodology (ARDL), a national level analysis, and a broader statistical model.

Zmajlović, Pavelić and Hajdas Dončić [10] examine the link between economic growth and municipal waste in Croatia from 2004 to 2017, using this as a basis for testing the EKC hypothesis for waste. Their results show a linear relationship, meaning that waste actually increases alongside the GDP, indicating that the EKC does not hold in this case. They note that the turning point for waste appears at much higher income levels than those observed, which implies that stronger waste-management policies would be needed to reverse the trend. Compared with that study, which relies on simple first-difference models and focuses mainly on short-run linkages, the present paper offers several advances. It covers a longer period and uses the ARDL approach, which allows for both short-run and long-run analyses and provides a reliable test of cointegration.

Overall, the EKC research for Croatia has produced a wide array of findings. Several studies, such as Škrinjarić [8], Jošić et al. [7] and Zmajlović et al. [10], do not find evidence of an inverted U-curve, whether for emissions, waste or other pollutants; instead, the relationship with the GDP is mostly linear and upward-sloping. Ziemblińska et al. [2] similarly report that Croatia is the only EU transition economy in their sample where emissions do not decline as income rises. On the other hand, Ahmad et al. [9] and, to some extent, Srdelić and Barišić [40] do find indications of the EKC, though either only in the long run or only for specific sectors. In short, the literature offers uneven and often partial conclusions, depending on the timeframe, the pollutant and the empirical method used.

This paper makes several contributions relative to the existing literature. It covers the longest period examined so far (1990–2023), capturing key economic, energy and institutional developments that older studies did not. It uses the ARDL approach, which allows for separate short-run and long-run analyses and robust testing for cointegration, while many previous studies rely on simpler models or a first-difference analysis. Finally, the model includes a wider set of emission determinants: energy use, renewables, industrial structure and institutional quality, as well as structural breaks for the EU accession, the pandemic and the energy crisis. These features make the present study the most comprehensive and methodologically robust assessment of Croatia’s EKC relationship to date.

3. Data and Methodology

This Section of the paper presents a description of the variables and methodology used to examine the short- and long-run relationship between CO₂ emissions and the GDP per capita and other control and dummy variables.

3.1. Description of Variables

The empirical analysis uses annual data covering 1990–2023, a timeframe that reflects the major economic, energy and institutional shifts Croatia experienced from the start of transition to the more recent climate–energy challenges. The year 1990 is taken as the starting point because it marks Croatia’s independence and the beginning of deep political and economic reforms, including the shift from a centrally planned system towards a gradual integration into the European Union.

The data used in the analysis were sourced from internationally recognized and methodologically reliable databases, primarily the World Bank’s World Development Indicators (WDI) and World Governance Indicators (WGI). The variables included in the empirical model are presented in

Table 1. Overview of the variables.

Symbol	Variable/Units	Source	Transformation	Expected Sign
lnCO2pc	CO2 emissions per capita (tons)	WDI	log	−
lnGDPpc	GDP per capita (USD, 2015 prices)	WDI	log	+
lnGDPpc2	Square of log GDP per capita	calculated	–	−
lnECpc	Energy consumption per capita	WDI	log	+
IND	Industry share in GDP (%)	WDI	levels	+
RENEW	Share of renewable energy (%)	WDI	levels	−
WGI	Institutional quality (−2.5 to +2.5)	WGI	levels	−
EU	EU accession (2013)	author	dummy	n/p
COVID	COVID-19 pandemic (2020)	author	dummy	n/p
ESOK	Energy shock 2022–2023	author	dummy	n/p

.

Table 1 provides an overview of the variables used in the empirical analysis, including their definitions, data transformations and expected coefficient signs. The dependent variable is the natural logarithm of CO₂ emissions excluding LULUCF per capita (lnCO₂pc), which represents a standard indicator of environmental pressure in the EKC literature. This measure is widely used in the EKC literature because CO₂ is the dominant greenhouse gas, accounting for almost 80% of global emissions [42].

The main explanatory variables are the natural logarithm of GDP per capita (in constant 2015 USD) (lnGDPpc) and its squared term (lnGDPpc²), which together allow for the examination of the potential non-linear relationship between economic growth and CO₂ emissions as proposed by the EKC hypothesis.

To improve the economic relevance of the model, several additional control variables were included:

• Energy use (kg of oil equivalent per capita) (lnECpc): This variable is included to control the energy intensity of the Croatian economy, or more precisely, the level of energy consumption per capita. Growth in energy per capita is expected to increase CO₂ emissions, regardless of the GDP, because higher energy consumption usually means higher use of fossil fuels. This variable is a better indicator than production-based emissions, because production can be outsourced, which may decrease a country’s emissions even while global emissions are increasing.
• Industry’s share in the GDP (IND): This indicator measures the share of the total economic activity in industry, including construction. It is included in the model to capture structural changes in the economy that may affect emissions levels. In fact, emissions reductions do not necessarily have to be the result of income growth or greater energy efficiency but can also result from changes in the structure of the economy. If the share of industry in the GDP decreases, for example, due to a strengthening of the service sector, emissions levels tend to decrease, regardless of the overall level of income.
• Share of renewable energy in the total final energy consumption (RENEW): This variable measures the transition towards cleaner energy sources. It is expected that a higher share of renewable energy in the total energy consumption decreases CO₂ emissions, regardless of the GDP. That is because renewable energy replaces fossil fuels.
• Institutional quality (WGI), measured as the average of the six World Governance Indicators, which range from −2.5 to +2.5: The use of the average WGI index is standard in the literature as a summary measure of governance effectiveness. Since stronger institutions typically enhance policy implementation, reduce corruption and foster sustainable development, a negative association between institutional quality and CO₂ emissions is expected.

Due to the limited data availability, the missing values for IND (1990–1994) and WGI (1990–1995) were estimated using linear interpolation, while the missing values for RENEW (2022–2023) were obtained through a linear extrapolation based on the most recent observed trend. All of the imputed values and applied procedures are explicitly reported in Appendix B to ensure full transparency and reproducibility. To assess whether these data treatments affect the main results, additional robustness checks excluding all of the imputed observations are conducted and reported in Appendix A.

Given that Croatia has undergone several major structural shifts over the past three decades, each affecting both emissions and economic activity, the model includes three dummy variables to capture the most important breaks. One reflects the EU accession (set to 1 from 2013 onward), another the COVID-19 shock in 2020, and the third captures the energy shock in 2022–2023. Including these variables helps prevent large external shocks from biasing the estimated long-run relationships among the model’s key indicators.

All of the continuous variables are expressed in natural logarithms to reduce heteroscedasticity, facilitate the interpretation of coefficients as elasticities, and linearize potentially nonlinear relationships in the data. The variables expressed as percentages (IND, RENEW) or as an index (WGI) remain at levels to preserve their straightforward substantive interpretation.

3.2. Methodology

This study examines the Environmental Kuznets Curve (EKC) hypothesis for Croatia, which posits a nonlinear, inverted U-shaped relationship between economic growth and CO₂ emissions. In the early stages of development, rising income tends to increase emissions, whereas after a certain threshold, known as the turning point, emissions begin to decline. To assess whether Croatia exhibits such a pattern and whether the turning point has already been reached, an econometric framework combining long-run and short-run dynamics was employed.

The methodological approach proceeds in several steps. First, the stationarity properties of all of the variables were tested using the Augmented Dickey–Fuller test. The results reveal mixed integration orders: the CO₂ emissions, the GDP per capita and the energy consumption are stationary in levels, while the industry share, the renewable energy share and the institutional quality become stationary only after first differencing. This I(0)/I(1) combination renders traditional cointegration techniques such as VECM inappropriate, as they require all of the variables to share the same order of integration. Consequently, the Autoregressive Distributed Lag (ARDL) model of Pesaran, Shin and Smith [43] was selected as the core estimation framework. The ARDL approach works well when the dataset is small and when the variables do not all have the same order of integration, which is why it fits this analysis. It has also been applied in several EKC-related papers, including Halicioglu’s [36] study on Turkey and the paper of Ahmad et al. [9] on Croatia.

The starting point of the model follows the usual EKC framework, but it is extended by adding a set of economic, energy and institutional indicators. In practical terms, the natural logarithm of CO₂ emissions per capita is modelled as a function of the logarithm of the real GDP per capita and its squared value, while additional controls account for the energy use, the role of industry, the share of renewable energy and the quality of institutions. The empirical specification is therefore written as follows:

$$\begin{gathered} {lnCO}_{2,t}= {\alpha }_0+ {\alpha }_1{lnGDPpc}_t+ {\alpha }_2{{(lnGDPpc}_t)}^2+ {\alpha }_3{lnECpc}_t+ {\alpha }_4{IND}_t+ \\ {\alpha }_5{RENEW}_t+ {\alpha }_6{WGI}_t+ {\delta D}_t+ {\epsilon }_t \end{gathered}$$

(1)

where

• lnCO_2,t—natural logarithm of CO₂ emissions per capita (dependent variable);
• lnGDPpc_t—natural logarithm of GDP per capita;
• (lnGDPpc_t)²—squared logarithm of GDP per capita, capturing the potential non-linear EKC relationship;
• lnECpc_t—natural logarithm of energy consumption per capita;
• IND_t—industry value added (% of GDP);
• RENEW_t—share of renewable energy in total final energy consumption;
• WGI_t—institutional quality index (average of six Worldwide Governance Indicators);
• D_t—dummy variables capturing major structural shifts (EU accession in 2013, COVID-19 pandemic in 2020, and the 2022–2023 energy shock);
• ε_t—error term.

A positive coefficient on α₁ combined with a negative coefficient on α₂ (α₁ > 0; α₂ < 0) would confirm an inverted U-shaped relationship consistent with the EKC hypothesis. By contrast, two positive coefficients would imply a monotonically increasing relationship, while two negative coefficients would indicate a monotonically decreasing one.

After specifying the model, the optimal lag structure was selected using the Akaike Information Criterion (AIC). The ARDL bounds testing procedure was then applied to examine whether a stable long-run cointegrating relationship exists between CO₂ emissions and the explanatory variables. Since the computed F-statistics exceeded the upper critical bounds in all of the estimated specifications, the presence of a long-run relationship was confirmed, allowing for the estimation of long-run coefficients relevant to the EKC hypothesis.

Short-run dynamics were modelled through the error-correction mechanism (ECM), which captures the speed at which emissions revert to their long-run equilibrium following a shock. A negative and statistically significant error-correction term indicates that deviations from equilibrium are corrected relatively quickly in Croatia, enhancing confidence in the estimated relationships.

The turning point of the Environmental Kuznets Curve (EKC) is calculated using the estimated long-run coefficients on income and squared income. If α₁ is the coefficient on lnGDPpc_t and α₂ the coefficient on (lnGDPpc_t)², the income level at which emissions reach their maximum is obtained as follows:

$$turning point=\exp \left({\displaystyle \frac{{-\alpha }_1}{{2\alpha }_2}}\right)$$

(2)

This formula converts the logged value back into real income (in constant 2015 USD) and follows the standard EKC procedure commonly used in empirical research, including the original approach proposed by Grossman and Krueger [12].

The main methodological contribution of this paper to the existing literature on Croatia is its broader and more tailored analytical framework, designed to capture the specific characteristics of the Croatian economy.

4. Results and Discussion

4.1. Results

Figure 1 presents the relationship between economic growth and CO₂ emissions in Croatia in the observed period 1990–2023, in what is commonly known as the Environmental Kuznets Curve (EKC). Both the GDP per capita and CO₂ emissions per capita are expressed in natural logarithms to make the data more stable, improve their distribution, and match the functional form typically used in EKC models.

Figure 1 shows how the growth of the GDP per capita was initially accompanied by the growth of CO₂ emissions. The peak was reached in 2007, after which there was a gradual decline and stabilization of emissions despite further GDP growth. This trend suggests a potential inverted U-shaped relationship between economic development and environmental degradation, providing partial support for the EKC hypothesis, according to which pollution levels rise at the early stages of economic growth but begin to decrease once a certain income threshold (turning point) is reached.

This visual impression is broadly in line with the results reported by Jošić, Jošić and Janečić [7], who report that Croatia reached its peak in CO₂ emissions in 2007, followed by a noticeable drop. Their interpretation centers mostly on the impact of the 2008–2013 recession. The data used in this paper, however, paint a slightly different story: emissions continue drifting downwards even after the economy recovers, which suggests something more persistent, i.e., a possible decoupling of growth and emissions that fits well with the EKC hypothesis.

Table 2. Descriptive statistics (variable in levels).

	CO2pc	GDPpc	IND	ECpc	RENEW	WGI
N	34	34	34	34	34	34
Mean	4.54	10,911	23.0	2026	29.3	0.329
Median	4.49	11,539	22.9	2070	29.8	0.592
Standard deviation	0.579	2724	2.65	213	3.54	0.431
Minimum	3.44	6341	19.3	1576	21.6	−0.490
Maximum	5.69	16,754	27.7	2335	35.1	0.794

summarizes the descriptive statistics for all of the variables in their original levels. Presenting them this way helps keep the scale intuitive before switching to the logarithmic transformations used later in the ARDL estimations.

According to the results from Table 2 and the data (Figure 2), Croatia’s per-capita CO₂ emissions averaged 4.54 metric tons over the period, with relatively small fluctuations from year to year. The GDP per capita, at roughly USD 10,900 on average (2015 prices), shows a long, steady upward trend. The industrial share of the GDP, around 23%, suggests a moderately industrialized structure that has gradually shifted toward services (also visible in Figure 2c). The energy use per capita, averaging 2026 kg of oil equivalent, remains extremely stable over the entire period. The share of renewable energy, with an average of 29.3%, indicates a steady progress in the country’s energy transition. The data for the WGI index (average of 0.33) suggest gradual improvements in institutional quality and overall governance in the observed period.

Overall, the results highlight that over the observed period, Croatia experienced almost continuous economic growth (with the exception of the crisis years) alongside an increasing share of renewable energy sources and relatively stable emissions, particularly in more recent years. This combination suggests that Croatia may be in the early stages of decoupling economic growth from environmental pressures, which is broadly consistent with the expectations of the EKC hypothesis.

To complement this overview,

Table 3. Correlation matrix (Pearson’s r).

	lnCO2pc	lnGDPpc	lnGDPpc2	IND	lnECpc	RENEW	WGI
lnCO2pc	—
lnGDPpc	0.605 ***	—
lnGDPpc2	0.597 ***	1.000 ***	—
IND	−0.362 **	−0.879 ***	−0.879 ***	—
lnECpc	0.904 ***	0.856 ***	0.851 ***	−0.673 ***	—
RENEW	−0.583 ***	0.228	0.236	−0.405 **	−0.221	—
WGI	0.618 ***	0.866 ***	0.863 ***	−0.900 ***	0.826 ***	0.098	—

Note: *** p < 0.01, ** p < 0.05 indicate significance levels.

presents a correlation matrix that offers a first look at how the variables relate to one another. Although descriptive, these correlations help identify any potential multicollinearity issues that could influence the econometric results discussed later.

Table 3 presents the correlation results for the variables. At first glance, most of the correlations are fairly strong and statistically significant, which is not surprising given how closely economic growth, energy use and institutional quality tend to move together. The high correlations between the GDP per capita, its squared term and energy consumption are quite expected from a theoretical standpoint; as income rises, energy demand usually increases as well, and with it, emissions. Renewable energy (RENEW), on the other hand, shows a negative correlation with CO₂ emissions. This suggests that a higher share of renewables is generally associated with lower emission levels, which fits well with the idea behind the EKC and the broader expectation that cleaner energy sources help ease environmental pressure. Similarly, the negative relationship between the industrial value added and both the GDP per capita and the governance quality suggests structural changes in the Croatian economy toward a more service-oriented and institutionally developed framework. Although some correlation coefficients exceed 0.80, these values are typical for macroeconomic time series and do not necessarily imply multicollinearity problems; this will be formally verified through diagnostic testing in the ARDL model. Overall, the correlation analysis supports the theoretical expectations of the EKC hypothesis and provides a sound basis for the subsequent econometric estimation.

Figure 2 shows a graphical representation of the time-series data in the period from 1990 to 2023.

A quick look at the time-series data (Figure 2) reveals distinct trends among the main variables over the period 1990–2023. Several of them, such as lnGDPpc, IND and WGI, show fairly clear upward or downward movements over time, which already hints that these series might not be stationary in their original levels. In contrast, the patterns for lnCO₂pc, lnECpc, and RENEW appear more stable, although their visual inspection alone does not allow for a definitive conclusion regarding stationarity. Given these observations, and in line with the requirements of the ARDL methodology, it is necessary to formally test the order of integration of all of the variables. Therefore, the Augmented Dickey–Fuller (ADF) test was applied to examine whether each series is stationary at the level or becomes stationary after first differencing. This step ensures that none of the variables are integrated of order two, I(2), which would violate the assumptions of the ARDL bounds testing approach.

The results of the ADF stationarity test presented in

Table 4. Augmented Dickey–Fuller (ADF) unit root test for stationarity of variables.

Variable	Model Specification	ADF Test Statistic	1% Crit.	5% Crit.	10% Crit.	p-Value	Stationarity
lnCO2pc	Intercept	−2.675	−3.661	−2.960	−2.619	0.090	I(0) (borderline)
lnECpc	Intercept	−2.887	−3.662	−2.960	−2.619	0.058	I(0) at 10%
lnGDPpc	Trend + Intercept	−3.744	−4.263	−3.553	−3.210	0.033	I(0) at 5%
lnGDPpc2	Trend + Intercept	−3.740	−4.263	−3.552	−3.210	0.033	I(0) at 5%
IND	Trend + Intercept	−1.481	−4.263	−3.553	−3.210	0.816	I(1)
RENEW	Trend + Intercept	−2.836	−4.263	−3.553	−3.210	0.195	I(1)
WGI	Trend + Intercept	−0.560	−4.273	−3.560	−3.212	0.975	I(1)

Note: The ADF stationarity test was performed with a constant and/or trend, depending on the visual characteristics of each series. The optimal lag length was automatically selected based on the Schwarz Information Criterion (SIC).

indicate that the variables lnCO₂pc, lnECpc, lnGDPpc, and (lnGDPpc)² are stationary at levels (I(0)). On the other hand, the variables IND, RENEW, and WGI are non-stationary at levels but become stationary after first differencing (I(1)). This satisfies one of the fundamental prerequisites for the application of the ARDL approach, since none of the variables were integrated of order two (I(2)).

To examine the existence of a long-run equilibrium relationship between CO₂ emissions and the selected explanatory variables, the ARDL bounds testing approach was employed. Three model specifications were estimated. The first model includes one lag and no dummy variables. The second model incorporates structural dummy variables capturing major shocks: Croatia’s EU accession, the COVID-19 pandemic, and the energy crisis of 2022–2023, using one lag. Finally, the third model includes the same dummy variables with two lags. The results of the ARDL bounds tests for all of the model specifications are presented in

Table 5. ARDL bounds test for cointegration (k = 6, n = 34).

Regression Function	Lag Structure	Dummy Variables Included	F-Statistic	I(0) Lower Bound	I(1) Upper Bound	Cointegration
ARDL(1,1,1,1,0,0,0)	1 lag	No	12.477	3.426	4.79	Yes
ARDL(1,1,1,1,1,0,1)	1 lag	YES	19.140	3.426	4.79	Yes
ARDL(1,2,1,2,2,0,0)	2 lags	YES	5.852	3.426	4.79	Yes

Note: Reported critical values correspond to the 5% significance level for finite samples (n = 34), based on Narayan [44], which are automatically applied by EViews in small-sample ARDL models. The F-statistics exceed the upper critical bound even at the 1% significance level (in Models 1 and 2), confirming the existence of a long-run cointegration relationship. The full sample covers the period 1990–2023 (n = 34). Minor differences in reported observation counts across tables reflect the loss of initial observations due to lag selection in the ARDL framework, as implemented in EViews.

.

The results confirm the presence of a statistically significant long-run relationship between CO₂ emissions and the selected explanatory variables. The strongest evidence of cointegration is observed in the model with one lag and included dummy variables, indicating that structural and external shocks play an important role in explaining the long-run dynamics. These findings justify the application of the ARDL approach for analyzing the long-run and short-run dynamics among the variables.

Our findings differ from those of Jošić, Jošić, and Janečić [7], whose results did not confirm cointegration between CO₂ emissions and the GDP per capita for the period 1990–2013. The difference between their results and the results in this study are already explained in the literature review part. In addition to a longer period of analysis and the inclusion of additional variables in the model, this paper also employs different methodological procedures. Their study relied on Engle–Granger and Johansen procedures, which are known to have weaker power in short or structurally unstable samples. On the other hand, our analysis employs the ARDL bounds testing approach, which is more robust to structural breaks and more effective in identifying long-run relationships. Taken together, these elements help explain why our study finds evidence of cointegration, whereas the earlier study did not.

Table 6. Diagnostic tests for the ARDL model.

Test	Statistic	Probability	Decision
Breusch–Godfrey LM (serial correlation)	0.045	0.835	No serial correlation
Breusch–Pagan–Godfrey (heteroscedasticity)	0.565	0.868	Homoscedasticity
Ramsey RESET (functional form)	0.061	0.808	Correct specification
Jarque–Bera (normality)	0.156	0.925	Normal distribution

Note: All tests fail to reject the null hypothesis at the 5% level, confirming model adequacy.

reports the results of diagnostic tests used to assess the adequacy and statistical validity of the ARDL model. Among the estimated specifications, the ARDL(1,1,1,1,1,0,1) model, which includes one lag and structural dummy variables, was selected as the preferred model. This specification provides the strongest statistical evidence of cointegration (F = 19.14) and demonstrates theoretical consistency with the observed structural changes.

Diagnostic tests were carried out to check whether the estimated ARDL(1,1,1,1,1,0,1) model is appropriate and statistically reliable. The results support the null hypothesis in all cases, as all of the p-values are above the 0.05 significance level. This means that the model’s residuals are normally distributed, show no signs of heteroscedasticity or serial correlation, and that the model is correctly specified. In other words, the estimated ARDL model is statistically sound and stable, making it suitable for interpreting both the long-run and short-run relationships in the analysis.

Following the confirmation of a long-run relationship among the variables, the ARDL(1,1,1,1,1,0,1) model was estimated to examine both the long-run and short-run dynamics between CO₂ emissions and its determinants.

Table 7. Estimated long-run and short-run coefficients of the ARDL(1,1,1,1,1,0,1) model (Dependent variable: ∆lnCO₂pc).

Long-Run Coefficients	Coefficient	Std. Error	t-Statistic	Prob.
lnGDPpc	6.603 **	2.248	2.937	0.0097
lnGDPpc2	−0.365 **	0.125	−2.923	0.0099
lnECpc	0.924 ***	0.211	4.375	0.0005
IND	0.019 *	0.010	1.832	0.0855
RENEW	−0.014 ***	0.003	−4.160	0.0007
WGI	0.091 **	0.037	2.492	0.0241
Short-Run Dynamics (Error Correction Model)
∆lnGDPpc	−0.332	1.510	0.220	0.829
∆lnGDPpc2	0.020	0.083	0.246	0.808
∆lnECpc	0.850 ***	0.101	8.375	0.000
∆IND	0.003	0.005	0.638	0.5325
∆WGI	0.019	0.024	0.789	0.442
EU	0.032 *	0.015	2.078	0.054
COVID	0.023	0.019	1.245	0.231
ESOK	0.053 ***	0.016	3.375	0.004
ECT(−1)	−0.738 ***	0.114	−6.473	0.000

Notes: *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively. ECT(−1) is the error correction term, whose negative and significant coefficient confirms convergence toward long-run equilibrium. Dummy variables: EU, COVID and ESOK. The turning point was calculated as exp(−α₁/2α₂), using long-run coefficients from the ARDL model.

presents the estimated long-run and short-run coefficients along with their corresponding significance levels.

The results of the ARDL(1,1,1,1,1,0,1) model presented in Table 7 provide evidence supporting the Environmental Kuznets Curve (EKC) hypothesis for Croatia during the period 1990–2023. The long-run coefficients show that the GDP per capita (lnGDPpc) has a positive and statistically significant effect on CO₂ emissions (α = 6.603, p < 0.01), while the squared GDP per capita term (lnGDPpc²) is negative and significant (α = −0.365, p < 0.01). This provides evidence of an inverted U-shaped relationship between economic growth and environmental degradation, suggesting that CO₂ emissions initially rise with economic development but begin to decline once income surpasses a certain turning point.

In a quadratic log EKC specification, the impact of income on emissions is not constant but varies with the level of income. The marginal effect of the GDP per capita on CO₂ emissions is given by the expression

$${\displaystyle \frac{\partial ln{CO}_{2,t}}{\partial lnGDP{pc}_t}}={\alpha }_1+2{\alpha }_{2}lnGDP{pc}_t$$

(3)

where

• ∂lnCO_2,t/∂lnGDPpc_t denotes the marginal effect (elasticity evaluated at time t) of GDP per capita on CO₂ emissions per capita;
• α₁ is the estimated long-run coefficient on lnGDPpc_t;
• α₂ is the estimated long-run coefficient on the squared income term (lnGDPpc_t)²;
• lnGDPpc_t is the natural logarithm of GDP per capita at time t, evaluated at a specific income level;
• the expression highlights that the elasticity of emissions with respect to income varies with the level of GDP per capita and is therefore not constant along the EKC.

Evaluating this expression at representative income levels shows that at lower income levels the effect of GDP growth on emissions is positive, it approaches zero around the estimated turning point and becomes negative at higher income levels.

The marginal effects reported in

Table 8. Marginal effects of GDP per capita on CO₂ emissions at representative income levels.

Year	GDP per Capita (USD, 2015)	lnGDPpc	Marginal Effect
1995	7137.12	8.873	+0.126
2000	8676.53	9.068	−0.017
2007 (emissions peak)	12,427.62	9.428	−0.279
2023	16,754.34	9.726	−0.497

Notes: Marginal effects are computed according to Equation (3), using long-run coefficients from the ARDL model.

indicate that the relationship between income and CO₂ emissions in Croatia has changed over time. During the mid-1990s, when the GDP per capita was relatively low, economic growth was associated with rising emissions, reflecting an energy-intensive growth pattern characteristic of the early transition period. As income increased, the marginal effect of the GDP per capita on emissions gradually weakened and approached zero, before turning negative at higher income levels.

Building on this pattern of marginal effects, the EKC turning point was calculated using the estimated long-run income coefficients from the preferred ARDL(1,1,1,1,1,0,1) model, following standard practice as exp(−α₁/2α₂). The resulting estimate of approximately USD 8550 per capita (2015 prices) suggests that Croatia may be operating around the income range associated with the EKC turning point. This estimate falls within the range commonly reported in the EKC literature (e.g., Grossman and Krueger [12]) and indicates that, after a period in which emissions increased alongside economic growth, further income growth was no longer associated with rising CO₂ emissions but instead coincided with their gradual stabilization and decline.

It is important to emphasize that the estimated turning point is defined in income space rather than in calendar time. While total CO₂ emissions in Croatia reached their empirical peak somewhat later, in 2007, the marginal effect of income on emissions was already negative by that stage. This suggests that income levels associated with declining marginal effects may have been reached earlier, with emissions adjusting only gradually over time. Such a lagged response is consistent with the presence of structural rigidities and delayed adjustments in the energy mix, whereby changes in production structures, technologies, and energy sources translate into lower emissions only with some delay.

At the same time, the estimated EKC turning point proves to be sensitive to sample composition. When the model is re-estimated on a reduced sample excluding all of the interpolated observations (1996–2021), the implied turning point increases substantially, to around USD 14,470 per capita (Appendix A). While this robustness check confirms the presence of an inverted U-shaped EKC relationship, it also indicates that the precise income level at which emissions begin to decline is not uniquely identified. Consequently, the timing of Croatia’s transition into the downward segment of the EKC should be interpreted with caution, with a greater emphasis placed on the overall pattern of decoupling rather than on a specific income threshold.

The energy consumption per capita (lnECpc) has a strong, positive and significant effect on emissions, which is in line with our expectations. The positive coefficient (lnECpc, 0.924) indicates that a 1% increase in energy consumption raises emissions by about 0.92% in the long run. Since the elasticity is so close to one, the results suggest that emissions move almost in step with energy consumption. That means that Croatia’s energy mix is still based on fossil fuels.

The coefficient for renewable energy consumption in the total final energy consumption (RENEW) is negative and significant, which implies that a higher share of renewable energy in the total energy consumption significantly reduces CO₂ emissions. Since this variable is measured in percentage points and CO₂ emissions in logarithms, the coefficient (−0.014) implies that a one-percentage-point increase in renewable energy consumption reduces emissions per capita in the long run by around 1.4%. These results are also in line with our expectations. Although the impact may appear modest at first glance, changes in the energy mix accumulate over time, so the long-run contribution to decarbonization can be quite meaningful.

The coefficient of industry share in the GDP (IND) is positive and marginally significant, suggesting that a higher share of industry in the GDP increases emissions. Since this variable is measured in percentage points and CO₂ emissions in logarithms, the coefficient of 0.019 means that if the industry’s share in GDP rises by one percentage point, emissions increase by about 1.9%.

The institutional quality (WGI) shows a positive and significant long-run link with emissions, which is not in line with our expectations. The coefficient of 0.091 means that a one-unit increase in the governance index raises emissions by approximately 9% in the long run. One possible interpretation is that improvements and reforms of the Croatian institutions after transition are accompanied by greater economic activity, which implies greater energy consumption and greater emissions. In many countries, including Croatia, the environmental benefits of stronger and more inclusive institutions take time to reduce emissions, especially in post-socialist countries. That means that Croatia is still in a stage of development where “better” institutions have not yet shifted towards green policies, a phase that is expected to come to an end in the near future as EU green policies are increasingly adopted. This result is consistent with the distinction between “strong” and “green” institutions often discussed in the environmental governance literature, where governance improvements initially stimulate economic activity before environmental regulation becomes binding.

The short-run results indicate that in the short term, most of the variables used in the model do not have a statistically significant influence on CO₂ emissions. The GDP per capita (∆lnGDPpc) and its squared form (∆lnGDPpc²) are statistically insignificant, meaning that fluctuations in economic activity do not immediately affect CO₂ emissions. This is not unexpected, because short-run emissions tend to be driven more by energy consumption than by the GDP itself.

In contrast to the GDP variables, energy consumption (∆lnECpc) is an exception, as its short-run coefficient is positive and highly significant. It indicates that a 1% increase in per-capita energy consumption raises emissions per capita by about 0.85% within the same period. The large value of the coefficient indicates that short-run emissions respond quickly to changes in energy consumption per capita. This suggests that the current energy mix in Croatia is still carbon-intensive.

Other structural and institutional variables used in this model, such as changes in the industry share and shifts in governance quality, do not show statistically significant short-run effects. Their impact appears to emerge more gradually, which is why it becomes evident primarily in the long-run relationship rather than in the short-run dynamics.

Dummy variables included in the model have different impacts on emissions. EU accession has a small but marginally significant positive effect. The short-run coefficient on the EU dummy variable (0.032) suggests that Croatian accession to the EU was associated with an immediate increase in CO₂ emissions of approximately 3.2%. This short-term rise likely reflects increased economic activity, inflow of investments and EU funds, and consequently an energy demand. That was before the gradual adoption of EU environmental and climate policies began to exert downward pressure on emissions. The ESOK dummy, which captures the energy shock of 2022–2023, shows a positive and significant short-run effect on emissions. This indicates that the energy crisis temporarily pushed emissions upward. The last dummy variable, COVID, is insignificant in this model.

The error correction term (ECT), which connects short-term dynamics (variable changes) with long-term equilibrium and shows whether the system returns to long-term equilibrium after a shock and at what speed, is negative and highly significant. Its value of −0.738 indicates a relatively fast correction of deviations from the long-run path. About 74% of any imbalance is adjusted within a single year. This confirms that the model is stable and that a valid long-run relationship exists among the variables.

Taken together, the results indicate that Croatia’s CO₂ emissions are shaped primarily by long-run structural and energy-related factors rather than by short-run economic fluctuations. Economic growth, changes in the production structure, energy consumption patterns, and the gradual expansion of renewable energy sources play a central role in determining emissions dynamics over time. In contrast, short-term variations in the GDP and institutional indicators exert only limited and mostly insignificant effects on emissions. This pattern underscores the importance of long-run adjustments in the energy system and economic structure for achieving sustained emissions reductions.

4.2. Discussion

The ARDL results provide strong evidence of a stable long-run relationship between economic growth and CO₂ emissions in Croatia. The findings support the presence of an inverted U-shaped relationship between the GDP per capita and emissions, as predicted by the Environmental Kuznets Curve (EKC), and are consistent with the first hypothesis of the paper. The results further suggest that Croatia is likely operating around, or possibly beyond, the income range associated with the EKC turning point, such that further income growth is accompanied by declining marginal effects on emissions. At the same time, the estimated turning point is sensitive to model specification and sample composition. While the baseline model implies a turning point at around USD 8550 per capita (2015 prices), the robustness checks excluding interpolated observations yield substantially higher estimates. This indicates that, although the EKC relationship itself appears robust, the precise income level at which the transition occurs should be interpreted with caution.

The estimated turning point from the baseline specification falls within the range reported in the seminal EKC literature. Grossman and Krueger [11] document turning points for local air pollutants at income levels generally below USD 8000 per capita, while a number of later studies also identify relatively low turning points for transition and middle-income economies. In the Croatian context, the results of this study reinforce and extend the findings of Ahmad et al. [9], who confirm an inverted U-shaped EKC for the period 1992–2011 using an ARDL framework, although their estimated turning point is based on a shorter sample and a more parsimonious specification. By contrast, studies that do not identify an EKC for Croatia, such as Jošić et al. [7] and Škrinjarić [8], rely on shorter time spans, static econometric techniques, or subnational data, which may limit their ability to capture long-run nonlinear dynamics and structural breaks. The use of a longer time series (1990–2023), an extended EKC specification, and a dynamic ARDL approach in this study helps explain why an EKC relationship is identified here but not in some earlier contributions. The results also differ from those of Ziemblińska et al. [2], who argue that Croatia has not yet entered the downward phase of the EKC; this divergence appears to stem primarily from methodological differences, as their analysis relies on descriptive and graphical methods without formally testing long-run relationships or controlling for additional covariates and structural breaks.

The energy use per capita shows a strong and positive long-run impact on emissions. This is consistent with the results of other similar studies, for example, Ang [24] for Malaysia, Halicioglu [36] for Turkey, and Hassan et al. [17] for a broad panel of developed and developing countries, all of whom identify energy use as the dominant driver of emissions. These findings suggest that income growth alone does not guarantee environmental improvement unless it is accompanied by changes in the energy mix and energy efficiency.

The negative and statistically significant coefficient on renewable energy consumption is consistent with a growing strand of the EKC-related literature emphasizing the role of clean energy in reducing emissions. Similar conclusions are reported by Zhang et al. [45] and Hassan et al. [17], who show that an increasing share of renewables contributes to long-run decarbonization, although the magnitude of the estimated effects is typically modest. In this context, the size of the renewable energy coefficient obtained for Croatia is broadly in line with the findings for other transition and emerging economies, suggesting that the impact of renewables on emissions is comparable rather than unusually weak or strong. This indicates that while renewable energy expansion is necessary, it may not be sufficient on its own without complementary policies aimed at reducing overall energy intensity. Nevertheless, the result clearly shows that Croatia’s shift toward cleaner energy sources is moving in the right direction.

In addition to long-run structural factors, the results also highlight the importance of short-run energy shocks. The positive and statistically significant coefficient of the energy shock dummy (ESOK) indicates that the 2022–2023 energy crisis was associated with a temporary increase in CO₂ emissions in Croatia. This suggests that periods of acute energy-market stress can disrupt decarbonization trends, even in economies where longer-run dynamics point toward declining marginal effects of income on emissions. Recent energy-system research emphasizes that energy crises are often accompanied by heightened electricity price volatility and market instability, which can alter consumption patterns and constrain short-term adjustment of energy systems, thereby intensifying emissions pressures [37]. In this sense, the Croatian case illustrates how external energy shocks can temporarily offset longer-run improvements driven by income growth and renewable energy expansion.

One of the more intriguing results concerns institutional quality, which exhibits a positive long-run association with emissions. Although this may initially seem counter-intuitive, it is compatible with parts of the literature suggesting that improvements in institutions in developing or transition countries often accompany periods of more intensive economic activity as well as higher energy use. Zhang et al. [45] find similar results for several emerging economies, arguing that improvements in governance often coincide with periods of accelerated economic activity and higher energy demand. In transition economies, institutional strengthening may initially support market expansion and industrial restructuring rather than immediate environmental protection. This result does not imply that better institutions necessarily worsen environmental outcomes but rather that their environmental effectiveness depends on whether governance improvements are explicitly aligned with climate and energy policies. In this sense, the Croatian case appears to reflect a transitional phase in which institutional development has not yet fully translated into lower emissions. More specifically, the results suggest that Croatia’s institutions have become stronger in terms of governance and administrative capacity, but not yet explicitly greener.

Overall, the findings confirm that the EKC is not an automatic outcome of income growth but a conditional relationship shaped by energy structure, institutional development, and policy choices. The differences between this study and earlier research on Croatia largely stem from variations in time coverage, econometric methodology, and model specification, reinforcing the importance of country-specific and long-run analyses when evaluating the EKC hypothesis.

5. Concluding Remarks and Policy Implications

The results of this study provide evidence of a stable long-run relationship between economic growth and CO₂ emissions in Croatia and support the presence of an inverted U-shaped Environmental Kuznets Curve (EKC). The findings suggest that Croatia is likely positioned around the income range associated with the EKC turning point, such that further economic growth is associated with declining marginal effects on emissions. This pattern is consistent with a gradual decoupling of emissions from income growth, although the precise timing and income level of the transition remain sensitive to model specification and data treatment. While the baseline ARDL model implies a turning point at around USD 8550 per capita (2015 prices), the robustness checks yield higher estimates, indicating that the conclusions regarding the exact timing of Croatia’s transition should be interpreted with caution.

Among the key determinants, energy consumption emerges as the strongest long-run driver of emissions, while an increasing share of renewable energy contributes significantly to their reduction. Institutional quality, somewhat unexpectedly, shows a positive long-run association with emissions—a result also observed in several transition economies. This finding likely reflects the fact that improvements in institutional performance during the transition period coincided with the shift toward a market economy and higher growth rates, which were accompanied by more energy-intensive economic activity.

Compared with previous studies, this paper offers a clearer and more robust identification of the EKC for Croatia by using a longer dataset, incorporating structural breaks, and extending the model to include a broader set of explanatory variables, such as institutional quality, industrial structure, and renewable energy. These extensions allow the analysis to capture long-run dynamics that earlier, more limited specifications were unable to detect.

The results also yield several important policy implications. Although Croatia appears to be positioned around or beyond the income range associated with the EKC turning point, the sustainability of this trajectory depends critically on continued progress in the energy transition. Energy consumption per capita remains the strongest long-run driver of emissions, indicating that policies aimed at expanding renewable energy sources and reducing reliance on fossil fuels are essential for maintaining and strengthening the decoupling between economic growth and emissions. The clear negative effect of renewable energy on emissions further underscores the importance of simplifying administrative procedures and accelerating investment in green energy projects. Moreover, the positive long-run relationship between institutional quality and emissions suggests that institutional reform alone does not automatically lead to environmental improvements. Instead, institutional strengthening needs to be explicitly aligned with climate and energy objectives. Accordingly, a consistent and well-coordinated climate and energy policy framework is necessary to support Croatia’s long-run decarbonization path. Overall, the findings imply that progress toward lower emissions will depend not only on technological and structural change but also on stable and credible policies capable of anchoring long-term climate goals.

Similar to most time-series analyses, this study has certain limitations. The interpolation of early institutional and industrial data introduces a degree of measurement uncertainty, while the relatively small sample size constrains the precision of the estimates. To address these concerns, all of the imputed values and the interpolation procedures are explicitly documented in Appendix B, and the robustness checks, excluding all imputed observations, are reported in Appendix A. These additional analyses confirm that the core EKC findings are stable and not driven by data interpolation. In addition, the national-level focus inevitably masks important heterogeneity across sectors and regions, where emission trajectories and adjustment mechanisms may differ substantially. In this context, the sensitivity of the EKC turning-point estimates to sample composition highlights the inherent uncertainty associated with identifying precise income thresholds in time-series analyses.

Future research would benefit from a more detailed sector-level perspective, as the national EKC relationship may obscure distinct dynamics in energy, industry, transport, and buildings. Comparative analyses across other transition economies could further clarify whether Croatia’s experience reflects broader regional patterns or country-specific developments. Although this study focuses on Croatia, the findings are informative for other small EU transition economies that may be approaching or entering the declining segment of the EKC.