Assessment of Technical and Eco-Efficiency of Dairy Farms in the Republic of Serbia: Towards the Implementation of a Circular Economy

Abstract

Efforts to improve agricultural sustainability have increasingly focused on enhancing productivity while minimizing environmental impact. In the Republic of Serbia, dairy farming remains a critical sector due to its dual role in food production and environmental pressure. This study aims to evaluate the technical and eco-efficiency of dairy farms in the Republic of Serbia using FADN data and the Stochastic Frontier Analysis (SFA) method. Specifically, the SFA methodology was applied, which enables a separate assessment of time-invariant and time-variant efficiency, with the aim of clearly identifying the factors that shape milk production in the Republic of Serbia. It was found that the technical efficiency for the 2015–2023 period was at a level of 58.7%, while the eco-efficiency was estimated to be 13.1%. Observing the relationship between the estimated technical and eco-efficiency, it can be concluded that they share similar mechanisms for improvement. In both cases, time-invariant inefficiency dominated, indicating that factors under the control of farms, such as the characteristics of agricultural producers and farms, play a key role in shaping production efficiency. In this context, adopting circular economy principles, such as nutrient recycling, the use of renewable energy, and optimized input utilization, offers an additional opportunity to enhance both economic and environmental performance.

1. Introduction

The intensive exploitation of natural resources for industrial production, particularly in the second half of the 20th and the early 21st century, has led to severe environmental degradation while simultaneously endangering human and animal health. This situation has created a need for redefining business models on a global scale and developing an economic system that enables sustainable economic growth while ensuring environmental protection and the well-being of the population.

The first significant step in this direction was made through the concept of “sustainable development”, which has been widely accepted in both scientific and business circles. In the 1987 Report of the World Commission on Environment and Development, sustainable development was defined as “development that meets the needs of the present generation without compromising the ability of future generations to meet their own needs” [1]. According to this definition, sustainable development is shaped by three key dimensions: economic, environmental, and social. While the economic component plays a central role by generating the financial resources necessary for improving the other dimensions, a focus on economic growth alone is insufficient for achieving sustainability. Instead, synchronized action at all levels is required to balance economic development, environmental protection, and societal well-being [2]. While sustainable development encompasses three key dimensions, this study primarily focuses on the economic and environmental ones due to the nature of the available data and the applied methodological design. The social aspect, although acknowledged as essential, is not quantitatively assessed in this paper and represents an important avenue for future research.

The model that integrates the aforementioned principles is known as the green economy. This concept promotes sustainable economic development by reducing negative environmental impacts, fostering social equity, and improving quality of life [3]. In short, the green economy seeks to achieve balanced economic progress while ensuring responsible resource management and care for people and ecosystems. However, since every form of economic growth inevitably generates large amounts of waste, the question of its proper management arises. Modern consumerism, a characteristic of the 21st century, produces enormous quantities of waste that negatively impact the environment, human health, and animal well-being. It is evident that the consequences of excessive consumption often outweigh the benefits of economic growth. Therefore, it is essential to implement an economic model in which waste becomes a resource for further development.

In this context, the circular economy emerges as a logical extension of the green economy, offering a more specific model that focuses on closing resource loops, minimizing waste throughout the production and consumption cycle, and efficient resource utilization. This approach enables economic prosperity through effective material management and the minimization of negative environmental impacts [4]. Waste reduction, on the one hand, involves a wide range of actions such as eco-design, extended product lifespan, the use of by-products from one production process as inputs for another, recycling, and similar strategies. On the other hand, an equally important aspect is the efficient use of natural resources, which contributes to biodiversity conservation, carbon footprint reduction, long-term sustainability, slowing the rise in raw material prices, and fostering innovation [5].

Given the broad applicability of circular economy principles, it is essential to examine how these concepts are implemented across specific sectors of the economy. Among them, agriculture stands out as both a critical area for circular economy application and a sector with significant environmental impact.

The assessment of the implementation status of the circular economy concept is unique to each economic sector. Depending on the purpose of the final product, the focus may be directed toward different aspects of circular economy practices. In this regard, recent scientific literature has paid particular attention to the application of the circular economy concept in agricultural production [6]. This trend is not surprising, considering that agriculture ensures food security for the population, regardless of the overall economic strength of a country. Additionally, the importance of studying the circular economy in agriculture stems from the fact that agriculture is a major polluter [7]. Agriculture has a considerable environmental footprint, contributing to 25% of total greenhouse gas (GHG) emissions [8]. The main sources of agricultural pollution include emissions of ammonia (NH₃), methane (CH₄), and nitrous oxide (N₂O), which result from the use of pesticides and livestock manure, as well as from rice cultivation and emissions from ruminant livestock [9].

It is also essential to emphasize that the fundamental factor of agricultural production—land—is limited and is gaining increasing importance as the global population continues to grow. Consequently, from the perspective of implementing the circular economy concept and establishing long-term sustainable agricultural production, it is of particular importance to focus on the efficient use of natural resources. The significance of efficient resource utilization in agricultural production extends across all dimensions of sustainability. Effective resource management contributes to preserving natural ecosystems and reducing environmental impacts by minimizing soil and groundwater pollution, ensuring sustainable water management, and promoting biodiversity, thereby primarily achieving environmental sustainability. Furthermore, the efficient use of resources supports the establishment of economic sustainability by increasing productivity and reducing costs. Finally, ensuring food security, improving public health, and fostering rural development represent the social benefits of effective natural resource management [10].

Given the multifaceted significance of efficient natural resource utilization, the question arises how to assess the state of agricultural production while considering all three dimensions of sustainability and the need for implementing the circular economy concept. Furthermore, to obtain meaningful results, it is essential to analyze the performance of a representative sample of a large number of agricultural producers over time. If each new analysis required the repeated collection of necessary data, it would quickly become evident that such research is unsustainable. Therefore, it is particularly important to identify business performance indicators that encompass the aforementioned sustainability criteria and can be calculated using data from publicly available databases.

To adequately evaluate the current state of agricultural production in line with sustainability goals and the principles of the circular economy, it is necessary to identify performance indicators that integrate economic and environmental dimensions. Traditional indicators often capture only one aspect of sustainability, which limits their ability to inform comprehensive policy or managerial decisions. Therefore, an increasing number of studies have emphasized the importance of using integrated metrics. One such indicator that has gained significant traction in both academic research and policy discussions is eco-efficiency. This concept is widely applied in the assessment of agricultural system efficiency, with an increasing emphasis on environmental and social impacts [11]. In academic literature, eco-efficiency is regarded as a quantitative management approach that simultaneously addresses economic and environmental aspects, where the prefix “eco” signifies both ecological and economic dimensions of sustainability. This concept is typically described as a sustainability metric that links environmental impact with the value created within a production system [12]. The main objective of eco-efficiency is to optimize the benefits obtained from every unit of resource utilized. In practice, enhancing eco-efficiency involves maintaining the same level of production or service while minimizing resource use [13].

When it comes to agricultural production, eco-efficiency and the circular economy are often regarded as complementary approaches, as both aim to optimize resource use, minimize waste, and enhance sustainability. One important pathway through which these objectives intersect is operational eco-efficiency, which refers to strategies aimed at “narrowing resource flows”, i.e., reducing the amount of materials and energy used per unit of output [14]. This concept is fully aligned with the foundational goals of the circular economy, which strives to decouple economic growth from resource consumption by minimizing inputs, extending resource lifespan, and closing material loops. Enhancing operational efficiency in this way not only contributes to improved economic performance but also fosters regenerative systems where waste and emissions are significantly reduced.

However, operational efficiency represents only one dimension of eco-efficiency within circular agriculture. Other equally important aspects include the reuse of organic waste, nutrient recycling, the integration of renewable energy, and the reduction of greenhouse gas emissions. These practices are especially relevant in resource-intensive sectors such as dairy farming, where production systems can greatly benefit from more circular approaches. By integrating these broader dimensions of eco-efficiency, agricultural producers can significantly reduce their environmental impact while improving long-term resilience and sustainability, thus effectively contributing to the transition toward a circular agri-food system.

Recent scientific research has increasingly examined eco-efficiency as a key indicator of agricultural production performance, particularly in the context of sustainability, circular economy implementation, and the green transition [15]. Certain authors have highlighted this point by stating that “evaluating eco-efficiency can assist policymakers in designing agricultural policies that are more effective in achieving overall agricultural sustainability and, more specifically, the sustainability of particular agricultural systems”.

Beyond its role in shaping public policies, another frequently emphasized objective in high-impact research is the development of recommendations aimed at enhancing both economic and environmental efficiency within decision-making processes. Most studies have concentrated on crop and dairy farming [16,17,18,19,20,21,22,23,24,25,26,27], as well as olive production [28], the beef cattle sector [29], and agricultural production in general [12,30,31,32]. Moreover, extensive research has been conducted across different geographical regions and national contexts, exploring variations in eco-efficiency at a broader scale [33,34,35,36].

The concept of eco-efficiency has also become widely acknowledged in sustainable development research, as it is considered a crucial metric for evaluating sustainability [37]. This indicator is utilized at three distinct levels: the macro level, which assesses the national economy; the meso level, which focuses on regional analysis; and the micro level, which examines the performance of individual companies [38].

Recent studies have assessed eco-efficiency in agriculture using various methodological frameworks, primarily Data Envelopment Analysis (DEA), Life Cycle Assessment (LCA), and Stochastic Frontier Analysis (SFA) [15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38]. Each of these approaches brings specific strengths and limitations that influence the interpretation of results.

DEA, as a non-parametric method, has been widely used due to its ability to handle multiple inputs and outputs without requiring a predefined functional form [15,28,30]. It is especially common in farm-level assessments where data availability is limited. However, one of the main criticisms of DEA is its deterministic nature, as it does not account for random noise, an important factor in agricultural settings that are inherently influenced by weather conditions, pests, and policy fluctuations. For example, some authors acknowledge that ignoring statistical noise can lead to biased estimates of efficiency [15,23].

The LCA approach offers a more comprehensive environmental evaluation by considering the entire life cycle of products and processes [16,20,21]. These methods are powerful in capturing environmental impacts, but they require extensive and detailed data that are often unavailable in standard farm accounting systems such as FADN. Furthermore, LCA is primarily focused on environmental aspects and does not easily integrate economic performance indicators, which limits its application when eco-efficiency is defined as a ratio between economic output and environmental impact.

In contrast, parametric approaches such as SFA offer the ability to separate inefficiency from statistical noise and to explicitly model the production frontier. SFA has increasingly been used to assess eco-efficiency, particularly in studies aiming to account for heterogeneity across farms and external shocks [24,25,29]. For instance, some studies [24,25] have proposed a flexible parametric specification for eco-efficiency estimation and demonstrated the adaptability of the SFA approach by aligning results with greenhouse gas targets at the farm level.

Despite its advantages, traditional SFA methods have been criticized for their reliance on specific functional forms and distributional assumptions (e.g., half-normal or truncated-normal inefficiency terms). However, more recent extensions have addressed these limitations by decomposing inefficiency into persistent (structural) and time-varying (transient) components. This allows for a clearer distinction to be made between inefficiencies stemming from farm-specific characteristics and those caused by temporary factors. Such decomposition is particularly relevant for formulating targeted policy recommendations, as emphasized in [26,34].

Accordingly, our study contributes to the growing body of research that applies advanced SFA techniques to agricultural eco-efficiency analysis by using panel data from Serbian dairy farms. The chosen approach enables a more refined diagnosis of inefficiency sources, distinguishing between long-term structural issues and short-term inefficiencies, thereby aligning with contemporary methodological trends in the field.

However, for an analysis of agricultural production efficiency to be comprehensive, it is necessary to include an indicator that measures the effect of available resources in generating a specific output. The most commonly used indicator of agricultural production efficiency in scientific and professional literature, which takes production factors into account, is technical efficiency. Technical efficiency measures a manager’s ability to achieve the maximum output with a given level of input (or to minimize input use for a given level of output) under existing production conditions, i.e., for a given production technology [39]. As an indicator of agricultural production productivity, technical efficiency has been widely used in numerous scientific studies. As is the case with scientific studies on the assessment of eco-efficiency, the majority of research on technical efficiency evaluation focuses on crop production and dairy farming. Studies differ from one another depending on whether the assessment was conducted using the DEA (Data Envelopment Analysis) or SFA (Stochastic Frontier Analysis) method. It is also important to note that the majority of studies, particularly those from Europe, use the FADN (Farm Accountancy Data Network) database as their primary data source [40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64].

From a methodological perspective, both technical and eco-efficiency are used to evaluate the productivity of agricultural producers by comparing the level of outputs to the quantity of inputs used. The conceptual similarity between them lies in their shared foundation, assessing how efficiently a production unit utilizes available resources. However, the distinction is found in the type of inputs and impacts considered. Technical efficiency focuses exclusively on economic inputs and the physical production process, aiming to maximize output with minimal input. In contrast, eco-efficiency expands the analytical scope by incorporating environmental aspects, evaluating how efficiently economic value is generated while minimizing environmental pressures.

Assessing both indicators provides a more holistic understanding of production performance. Technical efficiency identifies the potential to improve resource allocation and reduce production costs, thereby serving as a valuable tool for internal operational improvements. Eco-efficiency, on the other hand, supports the transition toward sustainable production by revealing how input use affects environmental outcomes. It is particularly relevant in the context of evolving environmental regulations and market demands for sustainability. Accordingly, this study includes an analysis of both technical and eco-efficiency to capture the dual objective of enhancing productivity while reducing environmental impact.

Agricultural production in the Republic of Serbia is characterized by the dominance of crop production, which accounted for 68.1% of the total gross value added (GVA) of agriculture in the period from 2007 to 2022. This structure of Serbian agriculture primarily indicates a weak vertical integration between crop and livestock production, which consequently affects the financial performance of agricultural producers and leads to inefficient resource management. Specifically, livestock production as a whole contributes only 29.4% to the total agricultural GVA, raising the key question of which livestock production sectors in Serbia need improvement to achieve significant progress. In this regard, cattle farming, which includes milk and beef production, has traditionally stood out within Serbia’s livestock sector, accounting for 40.1% of the total value of livestock production and 11.8% of the total agricultural GVA [65].

In the dairy farming sector, the concept of circular economy encompasses a range of strategies aimed at optimizing resource utilization, reducing environmental burdens, and establishing closed production loops. Typical examples include the application of livestock manure as organic fertilizer, the integration of crop and livestock production to enhance nutrient cycling, and the use of waste materials for biogas generation. Additionally, the adoption of water-saving systems, renewable energy sources, and precision feeding techniques contributes to both environmental protection and economic resilience. Through such measures, dairy farms can reduce production costs, minimize emissions, and improve overall sustainability, aligning their operations with the principles of circular agriculture.

Therefore, the research presented in this study is based on the assumption that improving the cattle farming sector, particularly dairy farming, will enhance the overall agricultural sector of Serbia while strengthening the raw material base for the food industry. In this way, the focus of this research is on agricultural producers specialized in milk production. The primary goal of this study is to assess the technical and eco-efficiency of this agricultural sector. Additionally, the research aims to examine the relationship between the assessed eco-efficiency and the indicator of technical efficiency, determining the extent to which conclusions regarding the efficiency of dairy farming differ depending on the applied efficiency measure.

The paper is structured as follows. Section 2 describes the materials and methods. The third part shows the results, while the fourth part includes a discussion, divided into two subsections: on the technical efficiency of dairy farms and on the eco-efficiency of dairy farms in Serbia. The main conclusions are summarized in the final section.

2. Materials and Methods

In assessing eco-efficiency, researchers typically rely on three main methodological approaches: the ratio approach, life cycle assessment/material flow analysis, and the frontier approach [66]. The ratio method defines eco-efficiency as the relationship between the economic value of a product and its environmental impact. However, this approach can only be applied when both the economic and environmental aspects can be measured using a common, quantifiable unit [37]. On the other hand, the material flow analysis (MFA) approach often uses life cycle assessment (LCA), a well-established method for evaluating the environmental impact of a product throughout its entire life cycle, from raw material extraction to final disposal [67]. Although this method offers important insights, it requires large amounts of detailed data and frequently depends on estimates, given the challenges of collecting complete and accurate information [66].

The third commonly used method in eco-efficiency research is the frontier approach, which is considered the most comprehensive because it simultaneously takes into account both economic and environmental inputs and outputs [68,69,70]. This approach is divided into non-parametric and parametric methods. The non-parametric approach relies mainly on mathematical programming, with data envelopment analysis (DEA) being the most frequently applied technique. On the other hand, the parametric approach is grounded in econometric modeling and distinguishes between deterministic and stochastic frontier models. One of the main strengths of stochastic frontier models is their ability to include a composite error term, which separates random noise from inefficiency, allowing for the consideration of both internal and external factors that influence performance. This makes the method particularly suitable for eco-efficiency analysis in agriculture.

However, it is important to emphasize that the methodology based on stochastic frontier analysis (SFA) was originally used in scientific literature for the assessment of technical efficiency. Precisely for this reason, some authors define the methodology for evaluating eco-efficiency as “environmentally adjusted technical efficiency”, which clearly indicates the close relationship between technical and eco-efficiency.

In line with the above, this research applies the frontier approach, and the assessment of technical and eco-efficiency was conducted using a SFA model. Given that SFA has been widely used in numerous scientific studies primarily for evaluating technical efficiency, it is possible to identify various classes of models [60]. Among these, certain models stand out for their ability to not only assess overall technical efficiency but also its components, such as persistent technical efficiency (time-invariant component) and residual technical efficiency (time-varying component).

The advantage of such models lies in their ability to identify the factors influencing the achieved technical efficiency and to analyze the relationship among its components. Specifically, if persistent technical efficiency dominates (meaning that residual inefficiency is high), it suggests that the underlying causes of inefficiency may stem from factors beyond the control of farms, such as climatic conditions or agricultural policy measures. On the other hand, if residual technical efficiency dominates (indicating persistent inefficiency), then the influencing factors should be sought among those under the control of farmers, such as the characteristics of the farms or the attributes of the producers themselves.

In this regard, the SFA model applied in this research to assess technical and eco-efficiency follows the approach described by [71].

$$y_{it}={\beta }_0+{{\displaystyle \sum }}_n{\beta }_n\ln x_{nit}+{\epsilon }_{it}.$$

(1)

In this model, the error term ${\epsilon }_{it}$ can be decomposed into a stochastic component ($v_{it}$), which captures the impact of all random factors influencing the variability of the observed output, and a residual error term ($u_{it}$), which represents technical or eco-inefficiency and is affected by factors that are considered to be under the control of the production units. Accordingly, the relationship ${\epsilon }_{it}=v_{it}-u_{it}$ holds. The key feature of this model lies in the decomposition of inefficiency component $u_{it}$ into a time-invariant component ($u_i$), which reflects persistent technical or eco-inefficiency, and a time-varying component (${\tau }_{it}$), representing residual technical or eco inefficiency, so that: $u_{it}=u_i+{\tau }_{it}$. Therefore, it is evident from the given specification that this model is specifically adapted for the assessment of technical and eco-efficiency using panel data.

An additional advantage of the specified model lies in its ability to test for the presence of persistent technical or eco-inefficiency without imposing any restrictions regarding time dependence. By including a time variable as an independent variable in the production function model, it becomes possible to separately identify and analyze exogenous technical progress and time-invariant technical inefficiency.

Regardless of whether the assumption is that individual effects ${\alpha }_i$ follow a fixed or stochastic specification, the estimation of the model is conducted through a multi-step procedure. In the case of the SFA model with a fixed-effects specification, the estimation is carried out in four steps.

In the first step, it is necessary to apply the Least Squares Dummy Variable (LSDV) method or the within (covariance) transformation method to obtain estimates of unknown parameters ${\beta }_{n}s$. By applying the covariance method, individual effects ${\alpha }_i$ are eliminated from the model.

In the second step, based on the estimated values of ${\widehat{\beta }}_{n}s$, pseudo-residuals are calculated using the expression $r_{it}=y_{it}-x_{it}^{\prime }\widehat{\beta }$. These pseudo-residuals also can be expressed as ${\alpha }_i^{*}+{\omega }_{it}.$ Using the derived pseudo-residuals, the average value of $r_{it}$ over time for each observation unit i allows for the estimation of individual effects ${\alpha }_i^{*}$. Next, time-invariant inefficiency component $u_i$ is estimated using the formula: $\underset{i}{\max }\left\{{\overline{r}}_i\right\}-{\overline{r}}_i=\underset{i}{\max }{\widehat{\alpha }}_i-{\widehat{\alpha }}_i^{*}$, where ${\overline{r}}_i$ is the average value of $r_{it}$ over time for each unit i. It is important to emphasize that the estimation of time-invariant component $u_i$ is conducted relative to the most efficient observation unit in the sample.

In the third step, based on the estimated values of ${\widehat{\beta }}_{n}s$ and ${\widehat{u}}_i$, residuals ${\eta }_{it}=y_{it}-x_{it}^{\prime }\widehat{\beta }+{\widehat{u}}_i$. These residuals now contain the expression: ${\beta }_0+v_{it}-{\tau }_{it}$. Here, ${\beta }_0$ and ${\omega }_{it}$ were excluded in the previous step through the calculation of the average pseudo-residuals for each observation unit i. At this stage of model estimation, it is necessary to introduce assumptions regarding the distribution of components $v_{it}$ and ${\tau }_{it}$, specifically: $v_{it}~N\left(0,{\sigma }_v^2\right)$ and ${\tau }_{it}~N^{+}\left(0,{\sigma }_{\tau }^2\right)$. This means that time-varying inefficiency $v_{it}$ is assumed to follow a half-normal distribution. However, it is important to note that generalizing the distribution of ${\tau }_{it}$ to a truncated-normal distribution is not possible, which means that it is not possible to determine the influence of additional factors.

By treating residuals ${\eta }_{it}$ as the dependent variable, a regression model is defined through which intercept term ${\beta }_0$ is estimated. It is important to emphasize that by examining the distribution shape of residuals ${\eta }_{it}$, one can draw conclusions regarding the presence of time-varying inefficiency. Specifically, if the ${\eta }_{it}$ residuals exhibit negative skewness, it is meaningful to proceed with further analysis aimed at estimating the time-varying component of technical or eco-inefficiency.

On the other hand, if residuals ${\eta }_{it}$ are found to follow a normal distribution, then it is assumed that ${\tau }_{it}=0$, meaning that the model reduces to one with time-invariant technical or eco-inefficiency only. Finally, in the fourth step, the estimation of time-varying inefficiency component (${\tau }_{it}$) is carried out.

Given that the previously defined model is based on panel regression methodology, it is necessary to verify whether the fundamental econometric assumptions for panel data are satisfied, specifically, homoskedasticity, cross-sectional independence, absence of autocorrelation, and no harmful multicollinearity. In this research, the presence of homoskedastic variance was tested using the Modified Wald test for groupwise heteroskedasticity; cross-sectional dependence was assessed with Pesaran’s test of cross-sectional independence; and autocorrelation was examined using the Wooldridge test for autocorrelation in panel data.

Violations of key econometric assumptions in panel data models, such as heteroskedasticity, cross-sectional dependence, and autocorrelation, can significantly affect the reliability of regression results. Heteroskedasticity, or non-constant variance of the error term, leads to inefficient and biased estimates of standard errors, which, in turn, undermines the accuracy of hypothesis testing. Cross-sectional dependence implies that unobserved shocks may simultaneously influence multiple units, thereby violating the assumption of independence and inflating the risk of Type I errors. Autocorrelation, or the correlation of error terms over time within a unit, can also distort statistical inference by producing underestimated standard errors.

To address these issues, researchers commonly employ estimation techniques that adjust for such violations. One widely used solution is the application of robust standard errors, which are designed to yield consistent estimates even in the presence of heteroskedasticity or autocorrelation. In cases where cross-sectional dependence is a concern, methods such as Driscoll-Kraay or clustered standard errors are applied to obtain valid inference. These corrective measures help ensure that parameter estimates remain statistically reliable, even when ideal econometric conditions are not fully met.

Also, potential multicollinearity was evaluated based on the Variance Inflation Factor (VIF) values. Additionally, the appropriate specification of the panel model (fixed or random effects) was determined by applying the Hausman test.

In order to define a representative sample for assessing technical and eco-efficiency, it is essential to ensure the availability of high-quality, methodologically sound data. In this regard, specially designed surveys can serve as a primary data source, but they are often difficult to conduct (especially over longer periods of time) as they require adequate infrastructure and significant financial resources. Consequently, data from official institutions are of particular importance, as they are methodologically validated and comparable with data and analyses conducted at the international level.

The only database capable of meeting the aforementioned requirements is provided by the FADN system. FADN (Farm Accountancy Data Network) is a data collection and processing system based on the annual gathering of production, economic, and financial data from farms. Considering the available data from the FADN sample, both academic and professional literature recognize the eco-efficiency indicator as an effective tool for assessing farm performance, particularly in the context of circular economy implementation [17,23,27,29,41,72,73]. Similarly, there is a large number of studies in which the assessment of technical efficiency is carried out based on FADN data [57,74,75,76,77,78,79,80]. The number of farms included in the FADN sample in the Republic of Serbia has been steadily increasing, from 40 farms in 2011 to 1825 farms in 2023.

In the models of technical and eco-efficiency, the dependent variable used was the total value of agricultural production of the analyzed farms. In general, the formation of this variable in the FADN sample involves first calculating the total value of crop production, livestock production, and the value of other products and services. The sum of these three components represents the total agricultural production value for each observed farm. Considering that technical efficiency reflects the ratio between output and input, in the technical efficiency model, production value is expressed in EUR per livestock unit (LU). On the other hand, since eco-efficiency is defined as the ratio between output and environmental pressure, in the eco-efficiency model, the production value is expressed per hectare of utilized agricultural land.

In the technical efficiency model, the independent variables used represent key production factors:

• Labor input (AWU/LU)—this variable captures the total amount of work engaged in agricultural production, typically measured in Annual Work Units (AWU), representing the full-time equivalent of labor on the farm;
• Capital (€/LU)—this includes value of total assets such as machinery, buildings, and equipment, which reflects the level of long-term investments in the production process. Given that a separate variable referring to the utilized agricultural area has been extracted, the land value is not included in this variable;
• Intermediate consumption (intermediate production costs) (€/LU)—this variable includes the total value of inputs consumed during the production cycle, such as feed, seeds, fuel, maintenance costs, and other operational expenses;
• Total Utilized Agricultural Area (€/LU)—represents the total amount of utilized agricultural land, regardless of whether it is owned by the holding or leased.

In addition, to assess the presence of technical progress over time, a time variable in this case, the year of observation is included in the model.

In contrast, the eco-efficiency model uses variables that represent environmental pressure, which, in agricultural production, typically refers to the use of:

• Fertilizers (€/ha)—represented through monetary expenditures on mineral and organic fertilizers used on the farm;
• Plant protection products (€/ha)—captured as financial costs associated with pesticides and herbicides;
• Energy—includes expenditures for electricity, fuel, and other energy sources used in farm operations.

It is important to acknowledge that the variables representing environmental pressure, fertilizers, plant protection products, and energy, are expressed in monetary terms. While this approach is common in studies using FADN data, it has limitations, as monetary expenditures may not always reflect actual physical usage or the intensity of environmental impact. This constraint arises from the data structure of the FADN system, which does not include detailed physical input quantities or emission data in case of Serbian FADN data. As such, the results should be interpreted as relative measures of environmental input use rather than precise indicators of ecological burden.

3. Results

The descriptive statistics related to the data used for the assessment of technical and eco-efficiency are presented in

Table 1. Descriptive statistics for the used variables.

Variable	Range Interval		Interquartile Variation		Mean	Median	CV (%)	IQR (%)
	Min	Max	Q1	Q3
Total output (€/ha)	428.52	7201.45	1523.85	2846.15	2294.65	1935.39	50.71	30.26
Total output (€/LU)	393.94	38,447.33	1389.96	2751.27	2533.48	1921.53	111.25	32.87
Total labour input (AWU/LU)	0.01	1.01	0.10	0.23	0.18	0.16	67.61%	39.80
Capital (€/LU)	442.14	10,576.77	2247.97	4135.94	3388.28	3025.08	47.55	29.57
Total intermediate consumption (€/LU)	89.49	5350.84	839.44	1606.32	1330.28	1118.44	56.41	31.36
Total Utilised Agricutural Area (ha/LU)	0.02	3.64	0.82	1.40	1.12	1.05	41.37	26.55
Fertiliser (€/ha)	1.08	543.38	48.22	121.25	102.46	75.45	84.98	43.09
Crop protection (€/ha)	0.96	255.86	11.56	35.07	28.42	20.34	94.29	50.43
Energy (€/ha)	6.43	842.69	72.82	148.50	123.67	101.53	67.70	34.19

Source: authors’ calculation.

. Looking at the descriptive indicators for the output variables, a wide range of variation and a relatively high coefficient of variation (50.71%) are noticeable. Accordingly, it is important to highlight that the median values for total output (€/LU) and total output (€/ha) amount to 1935.39 and 1921.53, respectively. Regarding the variables used in the technical efficiency model, all variables showed a relatively high coefficient of variation, making the median values the most appropriate for describing the central tendency. The median value for total labor input (AWU/LU) was 0.16. The median value for the capital variable was €3025.08/LU, while for total intermediate consumption, it was €1118.44/LU. On the other hand, the independent variables used in the eco-efficiency model also indicated the presence of a relatively high coefficient of variation. The median values for the variables Fertilizer, Crop protection, and Energy were €75.45, €20.34, and €101.53 per hectare, respectively.

Given that the assessment of technical and eco-efficiency was conducted using an econometric model based on panel regression, it was necessary to verify the fulfillment of the key assumptions typical for panel regression models. The first step involved checking for the presence of harmful multicollinearity. It was found that there was no reason for concern regarding multicollinearity.

Table 2. Multicollinearity in the Technical Efficiency (TE) model.

Variable	VIF
Total labour input (AWU/LU)	1.06
Capital (€/LU)	1.49
Total intermediate consumption (€/LU)	2.03
Total Utilised Agricultural Area (ha/LU)	1.49
Time	1.96
Average	1.59

Source: authors’ calculation.

and

Table 3. Multicollinearity in the Eco-Efficiency (EE) model.

Variable	VIF
Fertiliser (€/ha)	1.75
Crop protection (€/ha)	1.54
Energy (€/ha)	1.27
Time	1.49
Average	1.51

Source: authors’ calculation.

show that there was no highly statistically significant correlation between the variables used in either model. The average value of the VIF (Variance Inflation Factor) for the indicators used in the technical efficiency model was 1.59, while in the eco-efficiency model, it was 1.51.

In the continuation of the analysis, the assumptions related to the presence of homoskedastic residual variance, panel independence, and the absence of autocorrelation were examined. The results of these tests are presented in

Table 4. Tests for checking assumptions in a fixed-effects panel model (TE model).

Test	Null Hypothesis (H0)	Test Statistic	p-Value	Result
Modified Wald test for groupwise heteroskedasticity	Homoskedastic variance of the model	${\chi }^2\left(76\right)=2310.30$	0.0000	H0 is rejected
Pesaran’s test of cross sectional independence	Independent panels	$CD=10.27$	0.0000	H0 is rejected
Wooldridge test for autocorrelation in panel data	No first-order autocorrelation	$F_{\left(1;75\right)}=9.44$	0.0030	H0 is rejected

Source: authors’ calculation.

and

Table 5. Tests for checking assumptions in a fixed-effects panel model (EE model).

Test	Null Hypothesis (H0)	Test Statistic	p-Value	Result
Modified Wald test for groupwise heteroskedasticity	Homoskedastic variance of the model	${\chi }^2\left(76\right)=9592.35$	0.0000	H0 is rejected
Pesaran’s test of cross sectional independence	Independent panels	$CD=30.56$	0.0000	H0 is rejected
Wooldridge test for autocorrelation in panel data	No first-order autocorrelation	$F_{\left(1;75\right)}=3.80$	0.0550	H0 is accepted

Source: authors’ calculation.

. In the technical efficiency model, all baseline assumptions were rejected, indicating the presence of harmful heteroskedasticity, cross-sectional dependence, and first-order autocorrelation. In the eco-efficiency model, the tests confirmed the presence of heteroskedasticity and panel dependence, while no autocorrelation was detected. Based on the obtained test results, in order to address the issues caused by the violation of these assumptions, the assessment of technical and eco-efficiency was carried out using panel regression models with robust standard errors.

Additionally, before estimating technical and eco-efficiency using stochastic frontier production function models based on panel regression, a test was conducted to determine whether the available data were more suitable for a fixed effects or random effects panel model. The selection of the appropriate model specification was made using the Hausman test. In both models, the null hypothesis was rejected, indicating that the fixed effects panel models were more appropriate for the analysis. The calculated χ² statistic for the technical efficiency model was 31.08, while for the eco-efficiency model, it was 53.29, with the p-values in both cases being less than 0.0000.

Table 6. Estimation of the Technical Efficiency (TE) and Eco-Efficiency (EE) Models Using Fixed Effects with Heteroskedastic Variance.

Technical Efficiency (TE) Model
Parameter	Variable	Coefficient	Robust Std. Err.
${\mathsf{\beta }}_{0.\mathrm{TE}}$	Constant	−113.8486 a	14.2459
${\mathsf{\beta }}_{1.\mathrm{TE}}$	ln_Total labour input	0.0724 b	0.0356
${\mathsf{\beta }}_{2.\mathrm{TE}}$	ln_Capital	0.2042 a	0.0628
${\mathsf{\beta }}_{3.\mathrm{TE}}$	ln_Total intermediate consumption	0.4873 a	0.0463
${\mathsf{\beta }}_{4.\mathrm{TE}}$	ln_Total Utilised Agricultural Area	0.0179	0.0630
${\mathsf{\gamma }}_{\mathrm{t}.\mathrm{TE}}$	Time	0.0577 a	0.0072
${\mathsf{\sigma }}_{\mathrm{u}.\mathrm{TE}}$		0.1748
${\mathsf{\sigma }}_{\mathrm{v}.\mathrm{TE}}$		0.1849
${\mathsf{\lambda }}_{\mathrm{TE}}={\mathsf{\sigma }}_{\mathrm{u}.\mathrm{TE}}/{\mathsf{\sigma }}_{\mathrm{v}.\mathrm{TE}}$		0.9454
${\mathsf{\rho }}_{\mathrm{TE}}={\mathsf{\sigma }}_{\mathrm{u}.\mathrm{TE}}^2/{\mathsf{\sigma }}_{\mathrm{TE}}^2$		0.4719
Eco-Efficiency (TE) Model
${\mathsf{\beta }}_{0.\mathrm{EE}}$	Constant	−174.5106 a	10.5873
${\mathsf{\beta }}_{1.\mathrm{EE}}$	ln_Fertiliser (€/ha)	0.0971 a	0.0196
${\mathsf{\beta }}_{2.\mathrm{EE}}$	ln_Crop protection (€/ha)	0.0636 a	0.0229
${\mathsf{\beta }}_{3.\mathrm{EE}}$	ln_Energy (€/ha)	0.0610 b	0.0242
${\mathsf{\gamma }}_{\mathrm{t}.\mathrm{EE}}$	Time	0.0898 a	0.0053
${\mathsf{\sigma }}_{\mathrm{u}.\mathrm{EE}}$		0.3895
${\mathsf{\sigma }}_{\mathrm{v}.\mathrm{EE}}$		0.2432
${\mathsf{\lambda }}_{\mathrm{EE}}={\mathsf{\sigma }}_{\mathrm{u}.\mathrm{EE}}/{\mathsf{\sigma }}_{\mathrm{v}.\mathrm{EE}}$		1.6016
${\mathsf{\rho }}_{\mathrm{EE}}={\mathsf{\sigma }}_{\mathrm{u}.\mathrm{EE}}^2/{\mathsf{\sigma }}_{\mathrm{EE}}^2$		0.7196
Number of observations		684
Number of farms		76

^a Statistical significance at the significance threshold α = 0.01. ^b Statistical significance at the significance threshold α = 0.05. Source: authors’ calculations.

, presented below, shows the results of the estimated models of technical efficiency (TE) and eco-efficiency (EE). As previously noted, the models were estimated using stochastic frontier production function models based on fixed effects panel regression with robust standard errors. In the TE model, it was evident that capital and intermediate consumption had a statistically significant impact on the output value at the α = 0.01 level. The labor input variable was statistically significant at the α = 0.05 level, while utilized agricultural land did not show a statistically significant effect on the output value. It is also important to note that technical progress could be observed, as the time variable was statistically significant (p < 0.0000). On the other hand, in the EE model, all variables were statistically significant at the α = 0.01 level, except for energy, which showed statistical significance only at the α = 0.05 level.

However, what is far more important for this research is the fulfillment of the assumptions required for estimating the stochastic frontier production function model. In the EE model, the indicators ${\lambda }_{EE}$ and ${\rho }_{EE}$ were greater than 1 and 0.5, respectively, which clearly confirmed the appropriateness of applying the stochastic frontier production function model. In the TE model, these values were close to the threshold values, so it was necessary to calculate the skewness measure of the technical inefficiency component ($u_i$). For consistency, the same test was also conducted for the EE model. In the TE model, the skewness was found to be negative (a desirable scenario), with a value of −0.1259. In the EE model, the skewness measure was 2.1488, which was expected, given the anticipated low level of the eco-efficiency score. In the continuation of the analysis, an assessment of technical and eco-efficiency was performed. As previously stated, the Kumbhakar and Heshmati (1995) [71] model enables the estimation of both overall residual and persistent efficiency, as presented in

Table 7. Estimation of Technical and Eco-Efficiency Based on the Kumbhakar and Heshmati (1995) [71] Fixed Effects Model.

Efficiency	Number of Observations	Mean	Standard Deviation	Minimum	Maximum
Residual TE	681	0.8913	0.0415	0.5612	0.9698
Persistent TE	681	0.6585	0.1135	0.4316	1.0000
Total TE	681	0.5870	0.1052	0.3162	0.9480
Residual EE	681	0.9018	0.0253	0.6860	0.9591
Persistent EE	681	0.1450	0.1097	0.0545	1.0000
Total EE	681	0.1307	0.0981	0.0467	0.9549

Source: authors’ calculations.

.

Regarding the technical efficiency of the analyzed farms, the Total TE was recorded at 0.5870, indicating that the same level of output could be achieved with a significant reduction in input use. Given that the persistent TE was considerably lower than the residual TE, it was justified to look for factors influencing performance improvement among those under the control of agricultural holdings. On the other hand, the Total EE was significantly lower than the Total TE, amounting to only 0.1307, which clearly indicated that eco-efficiency in dairy farms was at an extremely low level. In other words, there is substantial room for improvement from the standpoint of resource efficiency, and again, the influencing factors should primarily be sought within the scope of what farms can manage and control. This conclusion is supported by the relatively low persistent EE, which stood at 0.1450, in contrast to the residual EE, which was at a considerably higher level of 0.9018.

The notable discrepancy between the estimated technical efficiency (58.70%) and eco-efficiency (13.07%) suggests that, while farms are moderately effective in transforming inputs into outputs, they are substantially less efficient in minimizing environmental pressure per unit of production. Several factors may explain this gap. First, the prevailing production systems are still predominantly input-intensive, relying heavily on fertilizers, pesticides, and fossil energy sources, with limited implementation of resource-saving or circular practices. Second, the observed farms may lack the knowledge, capacity, or incentives to adopt agroecological methods or environmentally friendly technologies, which would contribute to higher eco-efficiency. Third, Serbia currently lacks a robust system of agri-environmental measures or targeted subsidies that directly promotes ecological sustainability. As a result, while farms may achieve reasonable productivity levels, their environmental performance remains weak, indicating that eco-efficiency is constrained more by systemic and institutional factors than by pure production inefficiency.

An illustrative representation of technical and eco-efficiency is shown in Figure 1 and Figure 2. Figure 1a displays the structure of total technical efficiency (TE), while Figure 1b shows the residual, persistent, and total TE over time.

The presented results are accompanied by an analysis of the structure of achieved technical efficiency (TE), shown in

Table 8. Structure of dairy farms based on achieved technical efficiency.

Technical Efficiency	0.01–25.00	25.01–50.00	50.01–75.00	75.01–100
Proportion of farms (%)	0.00	22.37	71.05	6.58

Source: authors’ calculations.

. It is evident that the majority of farms (71.05%) achieved TE levels between 50.01% and 75.00%, while the share of highly efficient farms with TE above 75% was only 6.58%. On the other hand, 22.37% of the farms fell into the low-efficiency category, with TE below 50%.

Figure 2a displays the structure of total eco-efficiency (EE), while Figure 2b shows the residual, persistent, and total EE over time.

Similar to Table 8 on technical efficiency,

Table 9. Structure of dairy farms based on achieved eco-efficiency.

Technical Efficiency	0.01–25.00	25.01–50.00	50.01–75.00	75.01–100
Proportion of farms (%)	0.00	22.37	71.05	6.58

Source: authors’ calculations.

presents the distribution of dairy farms according to their achieved total eco-efficiency (EE). It is noticeable that the vast majority of farms (as much as 94.74%) achieved EE levels below 25.00%. The proportion of farms with EE above 25% but below 50% was only 3.95%, while the share of farms achieving EE above 50% was as low as 1.31%.

4. Discussion

The average technical efficiency score of farms specialized in milk production in the Republic of Serbia for the 2015–2023 period was 58.70%. This result indicates that the use of inputs could potentially be reduced by as much as 41.30% without affecting the current level of output. Consequently, this raises the question of what these farms need to do to improve their production efficiency. In this regard, further analysis revealed that the persistent component of technical efficiency stood at 65.85%, while the residual component was at 89.13%. Given the dominance of persistent (time-invariant) technical inefficiency, it is clear that the sources of inefficiency should be sought among factors that are under the control of the farms themselves. On one hand, these influencing factors may relate to the structural characteristics of the farms, where it is reasonable to expect that by intensifying agricultural production, the available capacities can be better utilized, thereby achieving higher productivity [81,82].

Several mechanisms can be identified that should contribute to increasing production intensity in this case. First, genetic improvement of the dairy herd through better selection of dairy cows is essential. Additionally, improving farming conditions, which includes modernizing farm facilities, is expected to reduce long-term losses and thus ensure more efficient production. Moreover, more efficient land use would enhance the vertical integration of crop and livestock production, allowing for the introduction of balanced feeding rations and facilitating better production planning. In general, it appears logical that improvements in production technology (e.g., modernization of machinery, health monitoring of dairy cows, etc.) will also lead to increased productivity [83]. However, it must be acknowledged that any improvement in farm characteristics must be preceded by a proactive initiative from the decision-makers on the farm.

In line with the above, a second group of influencing factors emerges, namely, the characteristics of the producers themselves, whose improvement can lead to the desired progress. This primarily refers to the enhancement of managerial skills of the decision-makers on the farm, which may involve additional training or the involvement of young producers or educated household members in business decision-making processes. Engaging young producers in the decision-making process can bring multiple benefits. Most importantly, it is expected that young producers will be more motivated to adopt new technologies, which is essential for the modernization of agricultural production [84].

While intensification strategies such as genetic improvement, modernization of facilities, and better utilization of resources can lead to gains in technical efficiency, it is important to recognize that such measures may also result in increased use of external inputs and environmental degradation if not properly managed. Therefore, efforts to enhance productivity should be aligned with sustainability goals. This calls for a shift toward sustainable intensification or eco-intensification, which involves increasing output through a more efficient use of inputs, reducing waste, and implementing practices that minimize ecological harm. Integrating circular economy principles such as nutrient recycling, on-farm energy production, and closed-loop input systems can support this transition. In this way, technical efficiency improvements do not come at the expense of environmental performance, but rather, contribute to long-term resilience and sustainability of the agricultural system.

If the focus is shifted to the time-varying component of technical efficiency, it can be concluded that the observed farms are not significantly influenced by factors beyond their control, primarily referring to climatic conditions and administrative measures of agricultural policy. Previous research has shown that economic measures of agricultural policy do not have a statistically significant impact on the technical efficiency of dairy farms in the Republic of Serbia [85]. A similar conclusion may be drawn in this case, considering that the residual component of technical efficiency is at a relatively high level.

It is also important to emphasize that the time-related variable was found to be statistically significant, indicating that there has been some change in productivity, despite the relatively short time frame covered by the research. The analysis revealed that the average annual productivity growth rate amounts to 5.77%.

Productivity growth primarily depends on research and development activities within the observed sector. Similarly, several authors have highlighted investment in research as a key determinant of productivity improvement. This is particularly evident in the case of transition countries, where research and development activities effort are largely directed toward solving local challenges [86]. In this context, when considering the agricultural sector, it can be concluded that productivity growth in developing countries is largely the result of the continuous import of technological solutions in both crop and livestock production. In this process, agricultural advisory services can play a crucial role [87].

Therefore, further productivity growth can be expected in the upcoming period. However, it is essential to interpret this growth as a reflection of general technological advancement, rather than the result of individual efforts by farmers. This growth stems from ongoing global technological progress, such as the development of seed hybrids, agricultural machinery, and genetic improvements in dairy cattle. In line with the above, numerous scientific publications reviewed over the course of this research interpret total productivity growth in the dairy sector as a consequence of technological changes and the adaptation of dairy farmers to new production practices [71].

When it comes to the estimated eco-efficiency, it is evident that its level is significantly lower compared to technical efficiency. Specifically, the eco-efficiency score of dairy farms in the Republic of Serbia for the period 2015–2023 is only 13.07%. Such a low score indicates that the use of inputs could potentially be reduced by as much as 86.93% while maintaining the same level of output. In other words, the key takeaway here is that Serbian dairy producers are excessively using fertilizers, plant protection products, and energy, which not only leads to higher financial costs but also contributes to environmental pollution.

Similar to technical efficiency, an analysis of the components of overall eco-efficiency reveals that persistent eco-inefficiency is the dominant factor. Therefore, the answer to how eco-efficiency can be improved should be sought within the farms themselves. When it comes to the characteristics of agricultural holdings, the solution likely lies in better internal organization. In this context, it is expected that the implementation of the circular economy concept could significantly contribute to improving both ecological and economic sustainability of the observed farms [88].

Introducing circular economy principles into milk production can greatly enhance eco-efficiency by increasing resource efficiency and reducing waste generation. Several mechanisms can contribute to both the economic and environmental aspects of sustainability, including the use of agroecological methods, recycling and reuse of waste, improvement of energy efficiency, implementation of short supply chains, reduction of greenhouse gas emissions, and better market positioning of products in line with current trends and the Green Agenda.

Agroecological practices improve soil quality and reduce the use of synthetic fertilizers, which in the long term contributes to the production of higher quality livestock feed and reduces dependency on the commercial feed market for dairy cows. Additionally, using by-products of crop and livestock production for energy generation and organic fertilizer production can help achieve the desired level of eco-efficiency by lowering production costs, improving soil health, and enabling more precise planning. Furthermore, shortening the supply chain is expected to lower transportation costs and reduce emissions, which can be achieved through direct sales to end consumers [89].

The application of the circular economy in dairy farming contributes to waste reduction, better energy efficiency, resource recycling, and emission reduction. As a result, farms become more eco-efficient, enabling the production of milk with fewer inputs, lower costs, and a greater positive environmental impact. In the long term, this approach enhances the competitiveness and sustainability of dairy production.

On the other hand, when examining the time-varying component of eco-efficiency, it can be noted that it is at a relatively high level. This result is, in fact, expected, considering that agricultural policy in the Republic of Serbia lacks a developed system of subsidies aimed at environmental protection and efficient use of natural resources. In other words, it can be assumed that administrative agricultural policy measures in Serbia have little to no impact on the achieved eco-efficiency of agricultural holdings. This conclusion further highlights the fact that investing in the promotion and implementation of the circular economy concept could significantly improve the economic and environmental sustainability of farms specialized in milk production.

5. Conclusions

This study assessed both the technical and eco-efficiency of dairy farms in the Republic of Serbia, employing a panel SFA model that allowed for the separation of inefficiency into persistent (time-invariant) and residual (time-varying) components. This methodological framework offered a more nuanced understanding of performance dynamics by distinguishing between long-term structural limitations and short-term fluctuations. The empirical findings revealed that both technical and eco-efficiency levels are notably low, with persistent inefficiency accounting for the majority of the observed inefficiency. Such results indicate that systemic and enduring constraints, such as outdated infrastructure, limited access to advanced technologies, and insufficient knowledge transfer, pose more substantial barriers to improved farm performance than transitory influences like climate variability or agrarian policies.

The original contribution of this study lies in the application of a dual-efficiency framework that jointly examines economic productivity and environmental sustainability, which are often analyzed separately in the existing literature. Additionally, by decomposing inefficiencies, the study provides valuable insights into the structural versus operational origins of inefficiency, enabling more targeted and effective policy interventions. These findings highlight the necessity of implementing long-term strategies aimed at modernizing farm operations, promoting sustainable intensification, and enhancing managerial competencies, particularly through education, generational renewal, and stronger advisory support systems.

Given the extremely low level of eco-efficiency observed, the integration of circular economy principles emerges as a promising pathway for improving sustainability in the dairy sector. Practices such as nutrient recycling, on-farm renewable energy generation, precision input use, and the valorization of agricultural by-products can simultaneously enhance environmental performance and reduce production costs. By embracing such measures, Serbian dairy farms can improve their resource use efficiency, reduce ecological impact, and move toward a more resilient and competitive agricultural system aligned with the broader goals of green transition and climate neutrality.

Generally, observing the relationship between the estimated technical and eco-efficiency, it can be concluded that they share similar mechanisms for improvement. Improving one aspect of efficiency is expected to lead to an enhancement of the other.

Despite the relevance and robustness of the presented results, this study has several limitations that should be acknowledged. First, the analysis was based on data from a single country (Republic of Serbia), which may have limited the applicability of the findings to other geographical or institutional contexts. Second, the eco-efficiency assessment relied on proxy variables derived from monetary expenditures (e.g., fertilizers, energy use) which did not fully capture the physical or environmental impact of these inputs. Third, while the model distinguished between persistent and time-varying inefficiencies, it did not explicitly quantify the influence of individual structural or behavioral factors due to methodological constraints. Finally, the use of panel data over a relatively short time frame (2015–2023) may have restricted the ability to capture long-term trends and structural shifts in dairy farming practices. These limitations provide opportunities for future research employing alternative methodological approaches, incorporating a broader set of environmental indicators, and conducting comparative analyses across regions or production systems.

In line with these limitations, future research should focus on quantifying the impact of various structural, technological, and managerial factors on both the persistent and residual components of technical and eco-efficiency. To achieve this, it will be necessary to employ a different class of SFA models, particularly those adapted to two-step procedures that allow for the integration of additional explanatory variables. Such methodological advancements would provide a deeper understanding of the drivers of inefficiency and support the development of more targeted policy measures. Furthermore, expanding the scope of analysis to include other countries or production systems could enhance the external validity of findings and offer a broader perspective on sustainability transitions in the dairy sector. As the first study of its kind in the Republic of Serbia, this research lays the foundation for future work that could inform both producers and policymakers on how to foster a more efficient and circular agri-food system.