Environmental Efficiency of Agricultural Enterprises in Serbia: A Panel Regression Approach

Abstract

The agricultural sector is a cornerstone of Serbia’s economy, ensuring national food security and contributing significantly to GDP, but it also generates notable environmental pressures, particularly through air and water pollution. This paper investigates the impact of agricultural enterprises’ environmental pressures on their financial performance between 2011 and 2021. The sample comprises 52 of the 63 agricultural enterprises listed in the national PRTR register as major air polluters in Serbia. Using enterprise-level data, environmental performance is measured through air emissions relative to revenues, while profitability is captured by return on assets (ROA). Panel regression analysis is conducted with Dynamic Ordinary Least Squares (DOLS) and Fully Modified Ordinary Least Squares (FMOLS) estimators to assess the long-run relationship between eco-efficiency and financial outcomes. The results show that reductions in environmental pressure are associated with improved profitability, highlighting the trade-offs and synergies between ecological responsibility and economic performance. These findings underscore the importance of promoting eco-efficiency as both a managerial strategy and a public policy priority, offering evidence to support Serbia’s alignment with EU environmental and agricultural sustainability goals.

1. Introduction

The recent milestone of 8 billion people on Earth has intensified pressure on nature to provide water, food, and living space. At the same time, multiple crises in recent years, such as the COVID-19 pandemic and the Russo-Ukrainian and Israeli–Palestinian conflicts, have disrupted global supply chains, affecting the production, distribution, and availability of food. These disruptions have led to rising food prices, increased food insecurity, and heightened poverty. Although these challenges have brought renewed attention to the agricultural sector’s importance, the sector has always been a fundamental component of the broader socioeconomic system, playing a crucial role in national development [1]. This is particularly true in many less developed economies [2], where agriculture provides employment opportunities and sustains the livelihoods of billions of rural individuals worldwide [3,4]. Although the agricultural sector relies on environmental quality and is crucial for sustaining global food systems, it has also been implicated in various environmental challenges, contributing to negative externalities such as air and water pollution and land degradation. These negative aspects, referred to as agricultural pollution, can be defined as activities that harm, contaminate, and degrade the environment and ecosystems [5]. GHG emissions originating from the agricultural sector, notably carbon dioxide (CO₂), methane (CH₄), and nitrous oxide (N₂O), are on the rise worldwide due to human activities [6].

According to Silva and Magalhães [7], environmental efficiency, or eco-efficiency, refers to the capacity of businesses, sectors, or entire economies to generate goods and services with reduced environmental impact, since the environment can no longer be regarded as a cost-free resource. The concept of eco-efficiency can be analyzed at the macroeconomic (national economy), meso-economic (regional), and micro-economic (company) levels [8]. At the macro level, eco-efficiency highlights the need to decouple Gross Domestic Product (GDP) growth from its potential negative environmental impacts, while at the micro level, it means creating more value with less environmental harm [9].

Building on this micro-level perspective, this paper analyzes agricultural enterprises at the micro-economic level, focusing on eco-efficiency as an integral element of corporate sustainability [10]. Agricultural eco-efficiency (AEE) refers to the relationship between environmental performance and economic performance. At the enterprise level, environmental performance is measured using ratio coefficients that link physical data on emissions with financial data.

Therefore, this paper assesses the impact of environmental pressures from agricultural enterprises on their profitability over the period of 2011–2021. Environmental performance is measured using air emissions as environmental data and revenues as financial data, enabling the calculation of eco-efficiency at the enterprise level. The research is guided by two questions: (1) What is the relationship between environmental efficiency and return on assets (ROA) in agricultural enterprises?; and (2) does agricultural enterprise size significantly influence ROA? To address these questions, panel regression analysis is applied using Dynamic Least Squares (DOLS) and Fully Modified Least Squares (FMOLS) estimators over an eleven-year period. While most studies in this field apply non-parametric frontier methods such as Data Envelopment Analysis (DEA), this study employs cointegration techniques. Unlike DEA, which measures relative efficiency across units in a single period, cointegration allows for an assessment of the long-run dynamic relationship between environmental and financial performance.

The novelty of this study lies in its integrated approach to assessing eco-efficiency at the enterprise level in the Serbian agricultural sector, an area with limited empirical coverage, especially in widening countries. While most existing research focuses on developed economies, a comprehensive analysis from an EU accession country such as Serbia offers valuable new insights. Previous studies have examined agricultural eco-efficiency in Italy [8,11], Austria [10], Norway [12], Poland [13,14], Romania [15], the European Union [16], and Türkiye and 26 EU Member States [17]. By combining environmental and financial data over eleven years, this study provides a rare longitudinal view of eco-efficiency trends and their economic implications. Furthermore, the application of advanced cointegration techniques (DOLS and FMOLS) to analyze the environmental–financial nexus represents a novel contribution, as most prior research has relied on Data Envelopment Analysis (DEA), Life Cycle Assessment (LCA), or Stochastic Frontier Analysis (SFA) [18]. This methodological choice not only contributes to the eco-efficiency literature by offering an alternative to frontier approaches but also provides policy-relevant insights into how Serbia’s regulatory environment affects agricultural enterprise performance over time.

The paper is organized as follows. Section 2 presents the Serbian background on agriculture and environmental pressures. Section 3 reviews the theoretical background on agricultural eco-efficiency. Section 4 describes the data and methods. Section 5 reports the results. Section 6 discusses managerial and public policy implications. Section 7 concludes the study.

2. Serbian Background: Agriculture and Environmental Pressures

Serbia strives to balance economic growth and environmental protection to ensure a sustainable and resilient future [19]. Agriculture plays a significant role in Serbia’s economy and has historically been one of the key drivers of national development. According to the Statistical Office of the Republic of Serbia (SORS) [20], the share of agriculture in total Gross Value Added (GVA) averaged 7.6% between 2011 and 2023 (equivalent to about 5.4% of GDP). During this period, only manufacturing, wholesale and retail trade, vehicle and motorcycle repair, and real estate contributed more to overall GVA. This underlines the strategic importance of agriculture, both as a provider of food security and as a pillar of rural livelihoods.

From an environmental perspective, Serbia’s sustainability performance remains a challenge. The Environmental Performance Index (EPI) offers a comprehensive tool for benchmarking environmental policies and sustainability outcomes. Serbia’s national EPI score declined between 2018 and 2022, reflecting significant environmental degradation, although a modest improvement was recorded in 2024 (score of 49.8, ranking 62nd out of 180 countries). Serbian agriculture performs better than its overall EPI rank: in 2020, Serbia’s agricultural EPI ranked 8th globally, although this position dropped to 26th in 2024, suggesting that while agriculture remains comparatively strong, there is still room for substantial improvement. By contrast, Serbia’s EPI score for air quality remains low, at 31.6 in 2024, placing the country 123rd worldwide and 16th among 19 Eastern European countries (see

Table 1. EPI scores and ranks of Serbia.

	EPI for Country		EPI on Agriculture		EPI on Air Quality
	Score	Rank	Score	Rank	Score	Rank
2024	49.80	62	71.4	26	31.6	123
2022	43.90	79	45.3	51	29.4	116
2020	55.20	45	69.9	8	/	/
2018	57.49	84	/	/	/	/

Source: [21,22,23,24].

).

The sectoral breakdown of emissions further underscores agriculture’s role. In Serbia, the largest contributors to air pollutant emissions are households (42.3%), the electricity, gas, steam, and air conditioning supply sector (37%), and manufacturing (9.8%). Agriculture, forestry, and fisheries accounted for approximately 4% of total emissions during 2011–2021, while mining and quarrying contributed around 3%, and other sectors collectively up to 4% (SORS) [20]. Although agriculture’s share of total emissions may appear modest, the sector is the dominant source of ammonia emissions, which are closely linked to fertilizer use and livestock production [25]. In addition, agriculture contributes significantly to other pollutants such as non-methane volatile organic compounds, suspended particulate matter, nitrogen oxides, methane, and carbon monoxide, all of which degrade air quality and ecosystems.

3. Theoretical Background

The concept of agricultural eco-efficiency refers to the ability of enterprises to generate food and agricultural products while minimizing negative environmental impacts. In practice, it provides a useful tool for reducing the ecological pressures caused by agricultural activities without undermining their economic sustainability [11,26].

Eco-efficiency has been measured using a variety of methodological approaches. Non-parametric frontier models, such as Data Envelopment Analysis (DEA) and Slack-Based Models (SBM), dominate the literature, particularly in China and Europe [27,28,29]. These studies show wide regional variation in eco-efficiency scores, often reflecting technological progress, infrastructure, and human capital as positive drivers, while fertilizer use, livestock intensity, and resource inefficiency exert negative effects. Recent European studies, such as that of Cecchini et al. [11] in Italy, also confirm that farms can reduce emissions without decreasing production, but highlight that structural characteristics such as labor intensity and land area matter for performance.

Beyond DEA-type models, some scholars have adopted parametric approaches. For instance, Alem [12] incorporated methane emissions to estimate Norwegian dairy farms’ eco-efficiency, while Stevanović et al. [30] applied the Generalized Method of Moments to examine eco-efficiency–profitability linkages in Serbian polluting enterprises. Yet most of these studies remain static in nature, failing to capture long-run relationships.

But when we look at determinants of eco-efficiency across contexts, common determinants emerge. Technological progress, improved infrastructure, and higher incomes are generally associated with improved eco-efficiency [27,31]. By contrast, public investment in agricultural R&D, inefficient subsidy schemes, and structural distortions have been shown to undermine efficiency [11,31]. Policy measures such as environmental taxes [32] or subsidy design [12] are also key factors that can either incentivize or weaken eco-efficiency.

While a large body of evidence exists for China [29,31] and EU countries [16,17,33], such as Italy [11], Poland [13,14], Hungary [34], Romania [15], and Austria [10], research on Serbia and the Western Balkans is still limited. Existing studies in Serbia have focused primarily on industrial emissions, such as SOx and NOx [35,36], or on aggregate eco-efficiency scores across pollutants [37]. Stevanović et al. [38] analyze the transparency of enterprises in Serbia in relation to information on environmental pollution and environmental protection. Novaković et al. [18] emphasized that outdated infrastructure, limited access to advanced technologies, and insufficient knowledge transfer negatively affect eco-efficiency in Serbian farms specialized in milk production. A few studies link the environmental and financial performance of Serbian enterprises [30,39], but agricultural enterprises remain underexplored despite their dual role as major economic contributors and significant polluters.

Altogether, the literature suggests that eco-efficiency is a critical measure of sustainable agricultural development, influenced by technological, structural, and policy-related factors. However, most existing studies apply frontier methods that benchmark relative efficiency at a single point in time, offering limited insights into long-run dynamics. Moreover, evidence for Serbia remains scarce and fragmented, with little focus on agricultural enterprises. This study addresses these gaps by applying panel cointegration techniques (DOLS and FMOLS) to assess the long-term relationship between eco-efficiency, profitability, and enterprise size in Serbia’s agricultural sector.

4. Materials and Methods

There are many approaches to measuring enterprise eco-efficiency in the literature. In this paper, we adopt the ratio model, as it is widely applied in the literature [40] and offers a straightforward interpretation of the relationship between environmental pressures and economic outcomes [41]. Specifically, while DEA and SFA are commonly applied to measure eco-efficiency at the farm or enterprise level, our study employs a ratio-based approach, consistent with the Färell framework, and applies panel cointegration techniques (DOLS and FMOLS) to examine the long-run relationship between environmental pressures and profitability. This allows us to capture dynamic interactions and causal linkages that frontier methods cannot fully address.

The impact of agricultural enterprises’ environmental performance on their profitability is evaluated using panel cointegration techniques, specifically Dynamic Ordinary Least Squares (DOLS), as proposed by Kao and Chiang [42], and Fully Modified Ordinary Least Squares (FMOLS), as developed by Phillips and Moon [43] and Pedroni [44]. The analysis covers the period 2011–2021, providing an eleven-year panel of enterprise-level data. This time frame was selected because it represents the longest continuous period with reliable and comparable data on both financial performance and environmental indicators in Serbia, while also coinciding with the country’s progressive alignment to EU environmental and agricultural policies.

Yahyaoui and Bouchoucha [45] pointed out that DOLS outperforms FMOLS estimators in terms of mean biases, while Rahman et al. [46] revealed that DOLS has better sampling properties and is less biased than FMOLS. Li [47] found that both FMOLS and DOLS can effectively address heterogeneity and cointegration issues across different data structures. However, Li also argued that FMOLS may be superior because it can handle multiple data problems simultaneously, such as simultaneity and autocorrelation, and it produces reliable results even with smaller panel samples. Conversely, Ozdemir [48] emphasized that DOLS accounts for individual heterogeneity through short-run dynamics, individual-specific fixed effects, and time trends, while also avoiding endogeneity and serial correlation issues in heterogeneous cointegrated panels. Given these complementary strengths, both DOLS and FMOLS are applied in this study to ensure robustness of results. The use of these models allows a thorough investigation of the environmental efficiency of agricultural enterprises in Serbia, addressing the research questions from multiple methodological perspectives.

The regression panel analysis is based on environmental and financial data from 52 agricultural enterprises identified as significant polluters in the national PRTR register, covering the period 2011–2021, with a total of 572 observations. For data preparation and panel data analysis, EViews 14 (IHS Global Inc., Irvine, CA, USA), Stata 15 (StataCorp LLC., College Station, TX, USA), and MS Excel (Microsoft Corporation, Redmond, WA, USA) software packages were used. Serbia signed the Protocol on Pollutant Release and Transfer Registers in 2003, initiated the implementation of the Protocol, the E-PRTR, and the related Directive in 2008, and has voluntarily submitted data to the European Environment Agency since 2011. These enterprises are classified as significant polluters under the PRTR system because their annual emissions exceed regulatory thresholds for major air pollutants.

Environmental data consist of total air emissions reported by the 52 agricultural enterprises included in the National Register of Pollution Sources, maintained by the Serbian Environmental Protection Agency (SEPA). The national PRTR register lists 239 enterprises (as of 2022, the year in which the data were collected) whose plants are among the largest air polluters in Serbia, including 63 companies operating in the agricultural sector. Our sample consists of 52 agricultural enterprises, as environmental data were not available for 11 enterprises over the entire observation period [25]. In terms of their principal activities (according to the Regulation on the classification of activities), 26 enterprises (50%) were engaged in animal production, 24 enterprises (46%) in growing of non-perennial crops, one in mixed farming, and one in support activities for crop production [49].

Therefore, the same 52 enterprises were analyzed during the period 2011–2021. Reported pollutants include sulfur oxides (SOx), nitrogen oxides (NOx), carbon monoxide (CO), ammonia (NH₃), particulate matter (PM₁₀ and PM₂.₅), non-methane volatile organic compounds (NMVOCs), and methane (CH₄). These pollutants correspond to the categories defined in EU Regulation 691/2011 and are published by the Statistical Office of the Republic of Serbia. Financial data were obtained from the Register of Financial Statements of the Serbian Business Registers Agency (SBRA) [49].

The structure of the analyzed agricultural enterprises by size shows that in 2021, 50% were classified as medium-sized, 35% as small-sized, 10% as micro, and only 4% as large enterprises [49]. Medium-sized enterprises accounted for around 65% of total air emissions in the sample during the observed period, while small-sized enterprises contributed 17%. The remaining emissions originated from large (9%) and micro enterprises (2%) [25].

Descriptive statistics of the selected variables are presented in

Table 2. Descriptive statistics.

	Mean	St. Dev.	Range
ROA	2.386	14.020	305.306
AEE	0.228	0.609	12.306
SIZE	1,343,824	1,753,575	10,177,581

Source: Authors’ estimation.

. The average ROA indicates that enterprises generated around 2.4% profit from their total assets. However, the large standard deviation reveals substantial variability in profitability across enterprises, with the range pointing to extremely large differences between the lowest and highest ROA values. The mean AEE value suggests a relatively low overall level of eco-efficiency in the sample, with moderate variability between enterprises. The range indicates that most enterprises cluster around low eco-efficiency values, while a few exhibit significantly higher performance.

The average SIZE value is RSD 1,343,824 thousand (approx. EUR 11,466 thousand), with a standard deviation of RSD 1,753,575 thousand (approx. EUR 14,962 thousand), indicating wide dispersion in enterprise size. The range of RSD 10,177,581 thousand (approx. EUR 86,835 thousand) further highlights considerable differences among enterprises.

The following equation is estimated using DOLS and FMOLS:

$${\mathrm{R}\mathrm{O}\mathrm{A}}_{it}=\mathrm{C}+{{\mathsf{\lambda }}_1{\mathrm{A}\mathrm{E}\mathrm{E}}_{it}+\mathsf{\lambda }}_2{\mathrm{S}\mathrm{I}\mathrm{Z}\mathrm{E}}_{it}+{\mathsf{\epsilon }}_{it}$$

(1)

where i denotes the enterprise, t the time period, ${\mathsf{\epsilon }}_{it}$ the error term, and ${\mathsf{\lambda }}_1$ and ${\mathsf{\lambda }}_2$ the parameters to be estimated. C is a constant.

The agricultural eco-efficiency (AEE) indicator is defined as

$${\mathrm{A}\mathrm{E}\mathrm{E}}_{i,t}={\displaystyle \frac{{\mathrm{T}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l} \mathrm{a}\mathrm{i}\mathrm{r} \mathrm{e}\mathrm{m}\mathrm{i}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{s}}_{i,t}}{{\mathrm{T}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l} \mathrm{r}\mathrm{e}\mathrm{v}\mathrm{e}\mathrm{n}\mathrm{u}\mathrm{e}\mathrm{s}}_{i,t}}}$$

(2)

Three variables are included in the panel models. Enterprise profitability is measured by the Return on Assets (ROA) indicator. The second variable, agricultural eco-efficiency (AEE), is calculated as the ratio of total air emissions to total revenues. The third variable, enterprise size (SIZE), is included as a firm characteristic and as a determinant of financial performance. The total revenue is used as the SIZE measure (in RSD 000) and estimated the natural logarithm. Environmental data on air emissions are taken from the Serbian PRTR register, reported by agricultural enterprises, and expressed in kilograms per year.

5. Results

The agricultural eco-efficiency (AEE) indicator shows variation both across agricultural enterprises in Serbia and over time within the same enterprises. Analysis of total emissions and total revenues indicates that the aggregate AEE trend is more strongly influenced by emissions than by the revenue. The main drivers of this trend could be agricultural activities and related agri-technologies. Total air emissions (in tonnes per year) from the observed agricultural enterprises increased during 2011–2014 and rose sharply again in 2021. The sharp increase in 2021 may be linked to higher production intensity and slight improvements in reporting accuracy as Serbia aligned with EU environmental standards. It may also reflect post-COVID recovery effects, suggesting that both regulatory changes and renewed economic activity influenced emission dynamics. Total revenues from the analyzed agricultural enterprises increased during 2011–2015 and 2017–2021. A slight reduction in total revenues was noticed in 2016.

The regression panel analysis of the environmental-profitability nexus is conducted using DOLS and FMOLS models with three variables: ROA, AEE, and SIZE. The first step in the analysis involves testing for cross-sectional dependence (CD), followed by testing the stationarity of the observed variables.

To account for potential cross-sectional dependence among the units, the Pesaran CD test was applied [50]. This test is appropriate because, in our sample, the time dimension (T) is shorter than the number of agricultural enterprises (N). The null hypothesis assumes cross-sectional independence, while the alternative indicates cross-sectional dependence. The results, presented in

Table 3. Cross-sectional dependence test.

Variable	CD Test	p-Value
ROA	4.057	0.000
AEE	25.684	0.000
SIZE	22.110	0.000

Source: Authors’ estimation.

, confirm the presence of cross-sectional dependence in the data series. The cross-sectional dependence between ROA and AEE variables may be due to the fact that analyzed agriculture enterprises operate in the same economic sector and under the same macroeconomic conditions (e.g., the same environmental regulations). In addition, the enterprises could apply similar sustainability strategies to be competitive, to position themselves better and to be recognized as environmentally friendly enterprises.

Before estimating the models, the stationarity of the variables was tested. A second-generation unit root test was applied, as this approach accounts for cross-sectional dependence. For each series, the order of integration was determined using Pesaran’s CIPS panel unit root test [51], with the results presented in

Table 4. Second generation unit root test (CIPS).

CIPS	Level			First Difference
Variable	NCNT	Constant	Constant & Trend	NCNT	Constant	Constant and Trend
ROA	−2.044 *	−2.459 *	−2.884 *	−3.978 *	−4.017 *	−3.924 *
AEE	−2.477 *	−2.977 *	−3.225 *	−4.130 *	−3.964 *	−4.088 *
SIZE	−0.245	−1.167	−2.847 *	−2.509 *	−3.364 *	−3.144 *

Note: The symbol * presents level of significance at 1%. NCNT is no constant nor trend. Source: Authors’ estimation.

. Consistent with Meo et al. [52], the use of Pesaran’s second-generation test provides more reliable results than first-generation unit root tests, as it explicitly allows for cross-sectional dependence among the variables.

From Table 4, we can observe that the variables ROA and AEE are stationary at level under all specifications, while SIZE is non-stationary at level when no constant or just constant is included, but becomes stationary when a constant and trend are included. At first difference, all variables are stationary.

For robustness, first-generation panel unit root tests were also conducted, with results reported in

Table A1. The first-generation unit root tests.

Variables	Level		First Difference		Order of Integration
	Intercept	Intercept and Trend	Intercept	Intercept and Trend
Levin, Lin & Chu t *
ROA	−13.495	−17.209	−31.860	−30.323	I(0)
	(0.000)	(0.000)	(0.000)	(0.000)
EE	−116.133	−83.894	−26.292	−28.885	I(0)
	(0.000)	(0.000)	(0.000)	(0.000)
SIZE	−5.018	−9.895	−19.310	−20.441	I(0)
	(0.000)	(0.000)	(0.000)	(0.000)
Im, Pesaran and Shin W-stat
ROA	−8.408	−6.881	−20.327	−9.015	I(0)
	(0.000)	(0.000)	(0.000)	(0.000)
EE	−17.493	−11.896	−14.213	−7.564	I(0)
	(0.000)	(0.000)	(0.000)	(0.000)
SIZE	2.088	−0.079	−9.681	−4.011	I(1)
	(0.982)	(0.468)	(0.000)	(0.000)
ADF—Fisher chi-squared
ROA	258.544	235.699	524.814	392.862	I(0)
	(0.000)	(0.000)	(0.000)	(0.000)
EE	183.218	242.012	399.228	341.959	I(0)
	(0.000)	(0.000)	(0.000)	(0.000)
SIZE	120.762	140.848	313.516	242.450	I(1)
	(0.125)	(0.010)	(0.000)	(0.000)
PP—Fisher chi-squared
ROA	300.496	316.230	622.715	584.150	I(0)
	(0.000)	(0.000)	(0.000)	(0.000)
EE	195.533	236.502	432.748	435.114	I(0)
	(0.000)	(0.000)	(0.000)	(0.000)
SIZE	203.195	183.987	417.481	436.591	I(0)
	(0.000)	(0.000)	(0.000)	(0.000)

Note: p-value (given in the parentheses). Source: Authors’ calculation.

in Appendix A. The methods applied include Levin, Lin, and Chu t; Im, Pesaran, and Shin W-stat; ADF–Fisher chi-squared; and PP–Fisher chi-squared. According to Table A1, ROA and AEE are stationary, i.e., integrated in order zero, I(0). The SIZE variable is classified as non-stationary, I(1), according to the Im, Pesaran, and Shin W-stat and ADF–Fisher tests, but as stationary under the Levin, Lin, and Chu t and PP–Fisher tests.

As the next step, we tested whether the variables are cointegrated. The long-run relationship among variables was examined using the panel Westerlund cointegration test [53]. As noted by Meo et al. [52], most conventional cointegration tests may yield misleading results because they do not account for structural breaks. In contrast, the Westerlund and Edgerton [53] test provides robust results by addressing serial correlation, heteroskedasticity, cross-sectional dependence, and structural breaks [52]. The results, presented in

Table 5. Westerlund test for cointegration.

	Statistic	p Value
Variance ratio	−1.955 **	0.025

Note: The symbol ** presents level of significance at 5%. Source: Authors’ estimation.

, confirm the presence of cointegration among ROA, AEE, and SIZE. The reported probability value of 2.5% is below the 5% threshold, leading to rejection of the null hypothesis of no cointegration.

Additionally, the presence of a long-run relationship among the variables was tested using panel cointegration tests, specifically the Pedroni, Kao, and Johansen Fisher panel cointegration tests (see Appendix A,

Table A2. The cointegration tests.

Pedroni Residual Cointegration Test
	No deterministic trend		Intercept and trend		No intercept or trend
within-dimension
	Statistic	Weighted Statistic	Statistic	Weighted Statistic	Statistic	Weighted Statistic
Panel v-Statistic	−1.319 (0.906)	−1.663 (0.952)	−5.947 (1.000)	−5.278 (1.000)	0.042 (0.483)	−0.650 (0.742)
Panel rho-Statistic	−3.980 (0.000)	−0.911 (0.181)	0.499 (0.691)	2.668 (0.996)	−4.784 (0.000)	−1.900 (0.029)
Panel PP-Statistic	−21.999 (0.000)	−12.312 (0.000)	−29.071 (0.000)	−19.876 (0.000)	−11.496 (0.000)	−8.140 (0.000)
Panel ADF-Statistic	−19.060 (0.000)	−12.551 (0.000)	−20.990 (0.000)	−17.100 (0.000)	−11.409 (0.000)	−8.867 (0.000)
between-dimension
Group rho-Statistic	1.804 (0.964)		4.981 (1.000)		0.642 (0.740)
Group PP-Statistic	−18.414 (0.000)		−23.702 (0.000)		−14.009 (0.000)
Group ADF-Statistic	−13.477 (0.000)		−17.645 (0.000)		−11.245 (0.000)
Kao Residual Cointegration Test
	t-Statistic
ADF	−4.972 (0.000)
Residual variance	346.948
HAC variance	81.109
Johansen Fisher Panel Cointegration Test
Hypothesized No. of CE(s)	Fisher Stat.* (from trace test)		Fisher Stat.* (from max-eigen test)
None	480.9 (0.000)		480.9 (0.000)
At most 1	638.4 (0.000)		591.5 (0.000)
At most 2	249.9 (0.000)		249.9 (0.000)

Note: p-value (given in the parentheses). Source: Authors’ calculation.

). The results in Table A2 confirm that the variables are cointegrated, which justifies the application of FMOLS and DOLS regression analysis using Equation (1).

Table 6. The results of the regression analysis (the dependent variable is ROA).

Variable	DOLS	FMOLS
AEE	−4.694 * [−2.648]	−0.332 * [−6.856]
SIZE	0.093 ** [2.336]	−0.001 [−0.071]
R-squared	0.821	0.190
Adjusted R-squared	0.274	0.098

Note: In the brackets are t-Statistics. * and ** are p-values at 1%, and 5% confidence level, respectively. Source: Authors’ estimations based on SEPA and SBRA data.

displays a panel regression analysis with ROA as the dependent variable.

Pedroni [54] recommended the use of DOLS to ensure the accuracy of empirical findings. Within the specified parameters, this flexible method allows for the cointegration of heterogeneous vectors [55].

The results indicate that AEE significantly affects the profitability of Serbian agriculture-polluting enterprises. Specifically, profitability decreases as the AEE indicator increases. In both models, the independent variables are statistically significant, with coefficients at the 1% or 5% significance level. Enterprises with higher pollution per revenue have lower ROA and are generally less profitable. The reasons for this phenomenon are complex and include analysis of production costs, expenses, investments in new technologies, and specialization. Enterprises with higher pollution may incur additional production costs and expenses related to environmental regulations and investment in eco-technologies, all of which can reduce profitability. Due to the nature of their activities and their category of specialization, some enterprises can lead to higher environmental impact, impacting damage to reputation and profitability.

The results are analyzed along two main dimensions: (1) the impact of eco-efficiency on profitability and (2) the influence of enterprise size on profitability. The results from both models confirm that agricultural eco-efficiency (AEE) significantly affects the profitability of Serbian agriculture-polluting enterprises, thereby directly answering the paper’s first research question. The AEE variable has a negative coefficient in both models. The DOLS model shows that if the AEE indicator increases by 1%, ROA decreases by 4.7%, while the FMOLS model shows that if the AEE indicator increases by 1%, ROA decreases by 0.3%. Unlike the AEE variable, the SIZE variable has a positive coefficient in the DOLS model, which indicates that if SIZE increases by 1%, ROA also increases by 0.09%. However, the FMOLS model results suggest that SIZE has no significant impact on the profitability of agricultural enterprises. These findings partially address this paper’s second research question, as the DOLS results point to a size–profitability relationship, while FMOLS does not confirm this effect.

For comparative context, particularly in the Eastern European region, the results indicated that most of these countries recorded low eco-efficiency scores [16,17]. Yılmaz [17] highlighted that countries such as Estonia, Poland, the Czech Republic, Slovakia, Lithuania, and Latvia scored lower than Türkiye in terms of eco-efficiency due to their limited technological progress. Similar findings showed that Slovakia, Latvia, and Estonia had the lowest eco-efficiency scores among EU countries [16]. A study in Hungary [34] also identified comparable regional patterns, while Matsumoto et al. [33] emphasized the gap between Western and Eastern European countries in terms of eco-efficiency, favoring the Western region. Still, Liu et al. [27] analyzed eco-efficiency trends in China and found that AEE increased by around 76% (from 0.405 to 0.713) between 1978 and 2017, with income identified as a key driver of improved eco-efficiency. While their study focused on the determinants of eco-efficiency, our results examine the reverse relationship—how eco-efficiency itself influences profitability. This highlights the multidimensional nature of AEE, which can act both as an outcome shaped by economic development and as a factor influencing firm-level financial performance.

Sudha’s [56] results for India suggest that company SIZE does not affect ROA when using the Least Squares Dummy Variable (LSDV) model, which is consistent with our FMOLS findings. However, when employing a Random Effects (RE) model, Sudha [56] identified a statistically negative relationship between SIZE and ROA, which contrasts with our DOLS results. Furthermore, Sudha [56] found a positive long-run association between eco-efficiency-based corporate environmental performance and financial performance in Indian companies, a result that diverges from our findings for Serbian agricultural enterprises. Similarly, Savitri and Nik Abdullah [57] reported that in Indonesia, company profitability is influenced by eco-efficiency, with firms receiving stronger support from the environment, community, and society when they operate more efficiently.

In our study, higher AEE values represent lower eco-efficiency (i.e., greater emissions relative to revenue), while lower AEE values reflect higher eco-efficiency. Accordingly, our results demonstrate that environmental improvements, captured by lower AEE values, are associated with stronger financial performance in Serbian agricultural enterprises. Conversely, persistently high emission levels act as warning signals of environmental inefficiency, which can undermine both profitability and financial sustainability. This highlights the need for corporate polluters to mitigate and prevent environmental externalities by aligning business operations with environmental responsibility, thereby securing both long-term profitability and societal support.

6. Policy and Managerial Implications

6.1. Managerial Implications

Our findings reveal that agricultural enterprises with lower eco-efficiency (i.e., higher emissions relative to revenue) currently enjoy lower profitability in Serbia. This signals a short-term trade-off: firms may benefit financially from polluting practices but risk long-term sustainability as regulatory, market, and societal pressures intensify. Managers should not view emissions reduction merely as a compliance requirement but as a strategic investment in competitiveness. Integrating eco-efficiency into core decision-making can create operational savings, reduce resource dependency, and enhance market access [58,59]. Measuring eco-efficiency provides a basis for improving the development of policies aimed at sustainable management and efficient use of natural resources in agriculture [60].

Practical measures include the adoption of precision agriculture technologies such as smart irrigation and sensor-based crop monitoring, improving fertilizer management to optimize nutrient use efficiency, investing in renewable energy sources (e.g., solar and biogas) for farm operations, and enhancing livestock waste management and recycling systems. These practices can significantly reduce emission intensity while maintaining or even increasing output levels, a finding consistent with evidence that more eco-efficient farms achieve higher incomes and use inputs more efficiently without sacrificing production [14]. Moreover, eco-efficient enterprises are likely to benefit from stronger reputational capital, greater trust and acceptance by local communities, and preferential access to green financing and subsidy schemes [61]. As sustainability reporting becomes a growing requirement within global value chains, Serbian agricultural enterprises that strategically prioritize eco-efficiency will not only safeguard long-term competitiveness but also strengthen their integration into international markets, echoing results from Polish commercial farms, where higher eco-efficiency was associated with greater profitability and stronger sustainability performance [14].

6.2. Public Policy Implications

From a policy perspective, the negative AEE–profitability relationship underscores the need for targeted instruments that decouple profitability from pollution. Similar patterns have been documented in the recent literature [14,15,16,17], in which studies emphasize importance of proactive intervention. Left unaddressed, this pattern risks locking Serbian agriculture into unsustainable growth pathways, contrary to EU integration objectives. Policymakers should therefore take the following steps:

• Strengthen monitoring and enforcement. Improved accuracy and consistency in reporting [62,63] will ensure that eco-efficiency can be reliably tracked and benchmarked [13].
• Introduce financial incentives. Empirical studies [8,17] show that environmental investments become more attractive when supported by economic incentives such as subsidies [12], tax credits, or low-interest loans, which could encourage the adoption of cleaner technologies and renewable energy, similar to eco-scheme measures under the EU’s Common Agricultural Policy (CAP).
• Develop eco-efficiency benchmarks [13] and sectoral targets [11]. Policymakers can define performance standards [15], rewarding enterprises that outperform thresholds while penalizing persistent laggards.
• Mainstream eco-efficiency in agricultural development strategies. Serbia’s alignment with the EU Green Deal [19], the Green Agenda for the Western Balkans [60], and the Farm to Fork strategy [17] requires incorporating AEE metrics into national agricultural policy and rural development programs.
• Foster capacity-building and knowledge transfer [12]. Strong partnerships among government, universities, and agribusinesses can provide training and advisory services on sustainable farming practices, ensuring that smaller enterprises are not left behind. A key determinant of productivity improvement is investment in research and better internal organization. To modernize agricultural production, young producers should be included in the decision-making process because they are more inclined to embracing new technologies [18].

6.3. Broader Implications

By identifying high emissions as a “red flag” for both environmental and financial inefficiency, this study highlights eco-efficiency as a strategic performance metric. For managers, this means integrating AEE into corporate sustainability reporting and risk assessment. For policymakers, it implies that regulation should be complemented by market-based incentives that directly reward improvements in eco-efficiency. Together, these approaches can shift Serbian agriculture toward a development path in which environmental sustainability and financial performance reinforce each other, in line with EU climate-neutrality goals for 2050.

7. Conclusions

This paper analyzed the environmental and financial performance of Serbian agricultural enterprises by applying panel regression analysis to explore the relationship between agricultural eco-efficiency (AEE) and profitability, measured by return on assets (ROA). Using data from 52 polluting agricultural enterprises for the period 2011–2021, covering 572 observations, the analysis incorporated three key variables: ROA, AEE, and SIZE.

The findings demonstrate that environmental pressure, measured by air pollutant emissions, weakened between 2015 and 2020 but rose sharply in 2021. The conclusions can be structured along two main directions. First, regarding the impact of AEE on ROA, the regression analysis confirmed that AEE significantly affects profitability: enterprises become less profitable when air emissions increase, reflecting lower eco-efficiency. Conversely, improvements in environmental performance contribute positively to financial outcomes, underscoring the link between ecological responsibility and economic sustainability. Second, concerning the influence of enterprise size, the results are mixed. The DOLS model indicates a positive association between SIZE and profitability.

From a policy perspective, the results highlight the importance of promoting eco-efficiency in Serbian agriculture. Investments in cleaner technologies, improved resource management, and emission reduction can simultaneously enhance financial performance and environmental outcomes. By systematically monitoring eco-efficiency through indicators such as AEE, policymakers and stakeholders can better identify inefficiencies, set performance benchmarks, and design policies that align agricultural growth with sustainability goals.

For Serbian agriculture, which faces structural challenges in balancing competitiveness and sustainability, these findings provide actionable insights. They illustrate that addressing environmental pressures is not merely a compliance issue but a pathway toward greater long-term profitability and resilience. Policies that encourage eco-efficiency can therefore support Serbia’s broader alignment with EU environmental and agricultural standards while fostering sustainable rural development.

However, certain limitations should be acknowledged. The results rely on PRTR-reported air emissions, for which coverage and accuracy improved after 2011, yet residual reporting bias remains possible. In addition, the AEE indicator only captures air pollution, while the lack of soil and water pollution data may underestimate total environmental pressures. Although the use of DOLS and FMOLS mitigates endogeneity in cointegrated panels, reverse causality cannot be fully excluded. Future research could apply dynamic models, such as panel VAR or dynamic common correlated effects, to better capture feedback loops and causal dynamics. Robustness checks with alternative AEE definitions (e.g., emissions per asset value) and additional outlier treatment could also further strengthen the analysis.

Finally, this study opens avenues for future research. Comparative analyses across countries could reveal contextual differences in the eco-efficiency–profitability nexus, while the application of alternative methodologies such as Data Envelopment Analysis (DEA) could provide complementary assessments of agricultural eco-efficiency in Serbia. Extending the scope of the analysis to include other environmental pressures beyond air emissions could further enrich the evidence base and support more comprehensive sustainability-oriented policy design.