Eco-Efficiency of Crop Production in the European Union and Serbia

Abstract

This paper evaluates the eco-efficiency of crop production in the European Union (EU) and the Republic of Serbia for the period 2015–2023, using a stochastic frontier analysis (SFA) model based on panel data. Eco-efficiency was assessed as the ratio of agricultural output to key environmental pressures, with expenditures on fertilizers, plant protection products, and energy serving as proxies for ecological burden. The analysis shows that the average eco-efficiency score (Total EE) across the sample is 59.26%, implying that nearly 41% of inputs could be reduced without decreasing output. Decomposition reveals high residual eco-efficiency (93.62%) and lower persistent eco-efficiency (63.30%), suggesting that systematic inefficiencies dominate and are primarily linked to internal farm-level factors such as management practices, organizational structures, and technology adoption. Serbia’s total eco-efficiency score of 63.0% places it close to the EU average, confirming structural similarities with Southern and Eastern European countries. Eco-efficiency scores exhibit notable cross-country variation, ranging from approximately 35% to 96%. About 59% of countries fall within the 50–75% interval, while roughly 11% exceed 75%, indicating considerable scope for further improvement. Cluster analysis further indicates that while Serbia belongs to the lower-intensity group, it has significant potential to converge toward EU frontrunners through farm-level improvements. The findings highlight the importance of targeting internal determinants of efficiency, while recognizing that policy measures can provide enabling conditions and long-term incentives for the green transition. A coherent policy for the green transition should prioritize farm-level structural upgrades, such as technology adoption, advisory and knowledge transfer, and sustainable nutrient and soil management, supported by enabling CAP instruments (eco-schemes and GAEC) and IPARD measures to accelerate improvements in resource efficiency and environmental performance.

1. Introduction

Accelerated and long-term exploitation of natural resources for industrial production, particularly in the 21st century, has resulted in significant environmental degradation and increased risks to human and animal health. These trends highlight the need to reassess prevailing business models and to establish an economic system that promotes sustainable economic growth while safeguarding ecosystems and enhancing social well-being.

The first substantial step toward a new understanding of the relationship between the economy and the environment was achieved through the concept of sustainable development, which rapidly entered both scientific literature and business practice. In the report of the World Commission on Environment and Development (1987), sustainable development is defined as progress that meets the needs of the present without compromising the ability of future generations to meet their own needs [1]. This definition clearly points to three inseparable dimensions: economic, ecological, and social. Although the economic sphere provides the means for improving the other two, economic growth alone is not a sufficient condition for sustainability. Instead, coordinated action is required to achieve a balance between economic development, ecosystem protection, and social well-being [2]. The model that operationalizes these principles is commonly referred to as the green economy.

The green economy implies a low-carbon, resource-efficient, and inclusive growth trajectory that aligns economic development with the planet’s resource constraints. Its essence lies in directing technological change, pricing mechanisms, and policy instruments (e.g., pollution taxes, incentives for clean technologies, green finance) toward investments that generate higher added value while minimizing the ecological footprint. In this way, the concept of the green economy simultaneously enhances competitiveness (through lower energy and input costs, innovation), reduces risks (climatic, regulatory, and market-related), and creates “green” jobs, while accelerating progress toward the Sustainable Development Goals [3,4].

Agriculture represents a critical testbed for operationalizing the principles of the green economy. As a sector highly dependent on natural capital (land, water, biodiversity) and at the same time a source of substantial externalities (N₂O and NH₃ emissions, eutrophication, soil degradation), agriculture illustrates both the vulnerabilities and opportunities of green transition. Here, the principles of the green economy translate into tangible practices: improving nutrient and water efficiency, reducing dependence on energy and pesticides, preserving soil health and ecosystem services, and enhancing the resilience of food supply chains while fostering rural inclusion.

Furthermore, the importance of studying the concept of the green economy through the lens of agriculture arises from the fact that agricultural production is a major polluter, often generating tensions between the imperative of increasing food output and the need to protect the environment [5,6,7,8]. Agriculture and land use have a significant ecological footprint, contributing approximately one-quarter to one-third of global greenhouse gas emissions, depending on the scope of the assessment [9]. The main sources include methane (CH₄) from enteric fermentation in ruminants and rice paddies, nitrous oxide (N₂O) from mineral and organic fertilizers, and ammonia (NH₃) released during fertilizer application and manure management [9,10,11,12].

In this context, within agriculture, the green economy means producing more—and more reliably—while consuming fewer resources and reducing negative impacts on ecosystems, with farmers’ well-being and rural development considered integral objectives. The goal is to decouple production growth from increased resource use through precise management of fertilizers, water, and energy. Central to this approach is soil quality, which calls for regenerative practices such as preserving soil organic matter, reducing tillage intensity, introducing cover crops, and diversifying crop rotations. Complementary measures include biodiversity and ecosystem service preservation (e.g., shelterbelts, agroforestry, pollinator habitats, anti-erosion practices) as well as responsible water management, involving efficient irrigation technologies, planning in accordance with soil conditions, and protecting the quality of surface and groundwater.

Taken together, these measures position agriculture not only as a driver of environmental pressures but also as a pivotal arena for advancing the principles of the green economy and accelerating the transition toward sustainable development.

The second dimension encompasses energy use and climate aspects, which include the introduction of renewable energy sources on farms, reducing N₂O emissions through improved nitrogen management, and methane capture where applicable. Plant protection relies on integrated pest management (IPM), biocontrol, and risk reduction from pesticide use. Additionally, digitalization provides an operational backbone, as digital records of inputs and outputs enable more informed business decision-making.

The third component refers to institutional and socio-economic conditions, which entail decent labor standards, the empowerment of small farms and cooperatives, and access to green finance and environmentally demanding markets. In this regard, policies and their instruments (e.g., agri-environmental schemes and conditionality under the CAP, payments for ecosystem services, “carbon/eco-schemes,” green credit lines, and guarantees) channel investments toward sustainable practices [13,14,15,16].

Building on the principles of the green economy in agriculture, a crucial aspect is reliable analysis aimed at objectively assessing the state of the agricultural sector. In this context, one of the most significant operational metrics is eco-efficiency, which measures the relationship between economic performance (yield, production value) and the use of resources and environmental pressures (nutrients, water, energy, emissions). Measuring eco-efficiency allows for comparable benchmarking across countries over time, the identification of best practices and technological gaps, and the precise targeting of policies (CAP/IPARD) through investments in practices that simultaneously enhance productivity and reduce the ecological footprint. Yet, despite its importance, eco-efficiency remains underexplored in the agricultural context of many transition economies, including Serbia, where evidence-based insights are urgently needed to support the green transition.

Although the scientific literature recognizes eco-efficiency assessment at different levels and across various types of agricultural production, crop production stands out in particular. Assessing the eco-efficiency of crop producers is especially important as it clearly demonstrates how much economic performance is achieved per unit of key resources (nitrogen, water, energy, land), while simultaneously limiting negative impacts (N₂O/NH₃ emissions, eutrophication, soil degradation, pesticide-related risks). Moreover, the importance of field crop production is reflected in the fact that it accounts for about 60% of the total utilized agricultural area (≈100 million ha), while its share in the total value of agricultural production in EU countries in 2024 amounted to around 30% (≈160 billion EUR) [17,18]. In the case of the Republic of Serbia, crop production is of even greater importance, representing almost 50% of the total value of agricultural output (≈3 billion EUR) and covering around 78% of the total utilized agricultural area (≈2.5 million ha) [19].

In the Serbian context, several sectoral and regional studies provide solid evidence on crop performance and input use. The energetic and economic efficiency of wheat and sugar beet production in Serbia has been analyzed with particular emphasis on the dominant contribution of fertilizers to total energy inputs and overall efficiency performance [20]. Regional analyses for northern Serbia show that drought significantly affects the yields of major field crops, underscoring the role of climatic variability in performance differentials [21]. Recent multi-location trials on small grains further document how nitrogen fertilization shapes key yield components, explaining cross-farm differences in input efficiency [22]. In regard to this, to the authors’ knowledge, this paper provides the first harmonized eco-efficiency assessment of crop farms in Serbia using FADN data, allowing for a direct comparison with EU Member States, whereas previous studies have focused mainly on technical or cost efficiency aspects.

In line with the above, the aim of this study is to provide a comparative assessment of the eco-efficiency of crop production in the EU member states and the Republic of Serbia, at the national level, with particular attention to the situation in Serbia. Comparing the eco-efficiency of crop production in Serbia, as a candidate country for EU membership, with that of the member states represents a key analytical step for objectively positioning the sector in relation to the European “best-practice” frontier. Such comparison enables the quantification of technological and managerial gaps, provides a basis for targeted investment from EU funds (IPARD/IPA III), and facilitates alignment with the objectives of the CAP (GAEC/SMR, agri-environmental schemes, eco-schemes, etc.), thus ensuring that Serbia’s crop sector can not only converge with EU standards but also strengthen its competitiveness in the green transition.

This study offers several distinctive contributions to the literature on agricultural eco-efficiency. First, it provides a comprehensive EU-wide benchmark of crop sector eco-efficiency that explicitly includes Serbia and spans the period 2015–2023, relying on harmonized FADN data to ensure full cross-country comparability at the national level of crop specialized farms. Second, the paper applies a stochastic frontier model with Kumbhakar & Heshmati decomposition, which allows for a clear distinction between persistent (structural or managerial) and residual (transitory) inefficiencies, thereby identifying internal, farm-level constraints versus external shocks with direct policy relevance. Third, a rigorous panel data estimation framework is implemented, including diagnostic tests for heteroskedasticity, cross-sectional dependence, and autocorrelation, supported by Hausman’s guided fixed-effects specification and robust standard errors, ensuring statistical consistency and reliability. Finally, the analysis extends beyond efficiency estimation by incorporating a PCA-based cluster analysis that classifies EU countries, positions Serbia among its peers, and illustrates convergence trajectories toward the eco-efficiency frontier. Collectively, these elements advance existing research by moving beyond single index comparisons toward a more diagnostic and policy-oriented understanding of agricultural eco-efficiency, offering practical insights for targeted investment, advisory support, and alignment with EU sustainability goals.

Building upon the outlined contributions, this study further establishes a clear analytical framework to empirically test the proposed research objectives. Since eco-efficiency is shaped by both structural (persistent) and time-varying (residual) factors, the following hypotheses are formulated to capture its structural, temporal, and spatial dimensions within the EU-26 and Serbia:

H1. 

Residual eco-efficiency (EE) is higher than persistent EE across EU-26 and Serbia, indicating that structural and managerial factors are the main source of inefficiency.

H2. 

Agricultural eco-efficiency shows a positive trend from 2015 to 2023, reflecting gradual technological and managerial improvements.

H3. 

Countries in the high-efficiency cluster achieve higher persistent EE than low-efficiency ones, while Serbia (though belonging to the latter) exhibits improving residual EE, signaling convergence toward the EU frontier.

The article proceeds as follows: Section 2 surveys the relevant literature. Section 3 details the data, variables, and methodological framework. Section 4 reports the empirical results, Section 5 discusses their implications, and Section 6 concludes with key insights and policy-relevant takeaways.

2. Literature Review

Eco-efficiency is widely recognized in contemporary research as an approach for assessing the performance of agricultural systems, with increasing emphasis on their ecological and social outcomes [23]. In the literature, it is treated as a quantitative managerial tool that simultaneously considers economic and ecological aspects, where the prefix “eco” encompasses both dimensions of sustainability [24]. In this sense, eco-efficiency represents a measure that links the ecological dimension with the value created within a product system [25]. Given that production efficiency implies maximizing benefits per unit of input, improving eco-efficiency can be understood as achieving the same level of production or services with lower input use [26].

The concept of eco-efficiency has become central in assessing the sustainability of agricultural systems. It refers to the ability to create more value while reducing environmental impacts, thereby linking economic performance with ecological responsibility. As Keating et al. [27] emphasized, eco-efficient agriculture is critical for balancing food security, environmental protection, and resource efficiency. More recent reviews demonstrate that research on agricultural eco-efficiency has expanded rapidly, with China, Brazil, and Europe leading in publications, and methodologies such as data envelopment analysis (DEA), stochastic frontier analysis (SFA), life cycle assessment (LCA), and material flow analysis (MFA) dominating the field [28,29]. Nevertheless, there remains a lack of a unified methodological framework and standardized indicators for measuring eco-efficiency across countries and production systems [28].

China has become one of the most intensively studied cases due to the size of its agricultural sector and strong regional heterogeneity. Using DEA-SBM combined with the Theil index, Pang et al. [30] analyzed 31 provinces between 2003 and 2013. Their model incorporated sown area, agricultural labor, water use, machinery power, fertilizer application, and plastic film inputs as inputs, while outputs included the added value of agriculture and undesirable outcomes such as nitrogen and phosphorus surpluses and plastic waste. Results showed that the average eco-efficiency was 0.69, with only a few provinces achieving full efficiency. Coastal and economically advanced provinces were found to be leaders, while traditional grain-producing regions lagged behind. A broader temporal perspective was offered by Yang et al. [8], who examined 31 provinces during 2001–2018 using a DEA–Malmquist–Luenberger index complemented with panel regression. Inputs included land, labor, machinery, fertilizers, pesticides, irrigation water, and plastic films, with agricultural carbon emissions as the undesirable output. Eco-efficiency increased from 0.49 in 2003 to 0.78 in 2018, driven mainly by technological progress. However, structural factors such as public investment in agricultural research and the industrial composition of agriculture negatively affected efficiency, suggesting that policy design matters as much as technology. Another important contribution is from Liao et al. [31], who applied a Super-SBM DEA model and explicitly integrated ecosystem service values (ESVs) as desirable outputs alongside agricultural production. Inputs included land, labor, machinery, water, fertilizers, pesticides, plastic film, and energy, while CO₂ emissions were considered undesirable. Their study, covering 31 provinces from 2006 to 2018, revealed an average eco-efficiency close to 1.0, but with a downward trend after 2010, highlighting sustainability challenges as input intensification grew. At a more localized scale, Zhang & Jin [32] focused on 44 counties in Liaoning Province (2014–2020) using the Super-SBM DEA model with undesirable outputs. Inputs included land, labor, fertilizers, pesticides, machinery, irrigation, and plastic films, with outputs measured as gross agricultural production and grain yields. Undesirable outputs were carbon emissions and diffuse pollution. Their findings show that eco-efficiency increased from 0.716 to 0.813 during the study period, but spatial disparities persisted, with northwestern and central counties performing better than southern and coastal areas. Overall, Chinese studies consistently emphasize that while eco-efficiency has improved over time, progress is uneven across regions. Technological progress, structural transformation, and targeted policies are repeatedly identified as decisive factors.

In Europe, research has heavily relied on FADN microdata, which allows for detailed farm- and region-level analyses of both economic and environmental performance. Coluccia et al. [33] evaluated eco-efficiency across 21 Italian regions (2004–2017) using a DEA model. Inputs included labor (AWU), capital, land, fertilizer use, and irrigated area, while the output was agricultural production value. The results highlighted a dual structure: northern regions were more productive, while southern regions were relatively more resource-efficient. Expanding on this, Fusco et al. [34] assessed Italian agriculture for 2017 using DEA with net added income per hectare as the economic output, and water, fertilizer, and energy use (proxy for GHG emissions) as environmental pressures. The average eco-efficiency was only 0.58, indicating that environmental burdens could be reduced by 42% without harming economic results. At the farm level, Gołaś et al. [35] analyzed 601 Polish farms in 2017 using DEA with land, labor, capital, indirect costs, nitrogen and phosphorus surplus, and GHG emissions (IPCC methodology) as inputs, and production value as the output. Average eco-efficiency was 0.76, meaning that farms could reduce input use and emissions by a quarter without lowering output. Larger farms were more efficient, but livestock farms—particularly cattle and pig farms—exerted the strongest environmental pressures. Importantly, more eco-efficient farms achieved higher income per hectare, supporting the notion of sustainable intensification.

Alem [36] assessed 692 Norwegian dairy farms (1991–2020) using dynamic and static frontier analysis. The dynamic model estimated average eco-efficiency at 0.94, compared to 0.90 in the static model, suggesting a consistent potential for 6–10% improvement. These findings emphasize the importance of accounting for intertemporal technological adjustments when measuring eco-efficiency. On a broader EU scale, Pishgar-Komleh et al. [37] used Window-SBM DEA on FADN and Eurostat data (2008–2017), with land, labor, costs, and depreciation as inputs and agricultural production as output, while GHG emissions were included as undesirable outputs. Their findings show that older EU member states consistently outperformed newer ones: countries like the Netherlands and Belgium scored above 0.90, while Slovakia and Latvia remained below 0.60.

Expanding the literature to a broader scope, Rybaczewska-Błażejowska and Gierulski [25] applied an LCA + DEA framework to assess eco-efficiency in the EU-28 agricultural sector. Inputs derived from LCA included energy consumption, fertilizers, pesticides, emissions, and waste, while outputs were measured as agricultural GDP. Results showed that only ten countries (e.g., Belgium, Italy, the Netherlands, Sweden) achieved relative eco-efficiency, while most member states were eco-inefficient, primarily due to excessive fertilizer and energy use. Magrini [38] applied a panel translog SFA model to 40 European countries (1990–2019). Inputs included land, labor, livestock, machinery, and fertilizers, with outputs as gross production. Albania, Croatia, Iceland, and Portugal recorded stable growth, while Cyprus, Greece, Hungary, and Romania declined. Most countries maintained stable efficiency (0.93–0.95), though Denmark, Italy, and Slovenia saw declines in the last decade, whereas Ireland and Latvia improved. European studies thus underscore significant regional and structural disparities in eco-efficiency, strongly influenced by farm size, production orientation, and CAP support schemes. Finally, Yılmaz [39] extended this macro perspective to 26 EU states and Türkiye (2003–2022) using an input-oriented DEA model with VRS. Inputs included employment, energy consumption, and capital formation, normalized by land area, while outputs were agricultural GDP (desirable) and GHG emissions (undesirable). Results revealed that Türkiye, Italy, and the Netherlands frequently reached the efficiency frontier, while several Eastern European states lagged behind, reflecting stark eco-efficiency gaps across Europe.

Beyond China and Europe, studies from other regions highlight the broader relevance of eco-efficiency analysis. Grassauer et al. [40] combined LCA and DEA to assess 44 Austrian dairy farms, using indicators such as cumulative exergy demand, global warming potential, eutrophication potential, and aquatic ecotoxicity as inputs, with farm income, human-edible energy and protein, and high-nature-value farmland as outputs. Results showed higher eco-efficiency in organic farms (0.92) compared to conventional farms (0.81), and identified feed concentrates as the main driver of inefficiency. In Iran, Saber et al. [41] studied 200 rice farms (137 conventional, 47 low-input, 16 organic) through LCA (ReCiPe2016) and regression analysis. Inputs included diesel, fertilizers, pesticides, electricity, seeds, and manure, while impacts covered damage to ecosystems and human health. Eco-efficiency, defined as profit per unit of impact, was systematically higher in organic systems, largely due to better prices and lower ecological burdens.

Although research on agricultural eco-efficiency has expanded considerably, important gaps remain. The majority of studies focus on China and Western Europe, while Central and Eastern European countries are still underrepresented, despite their strong reliance on agriculture and specific structural challenges. This lack of empirical evidence limits the understanding of how transitional economies perform in terms of resource efficiency and environmental pressures. Furthermore, the absence of harmonized indicators and comparable methodologies makes it difficult to benchmark eco-efficiency across all European regions.

Previous studies focusing on agricultural efficiency in the EU and Serbia rarely provide harmonized cross-country comparisons, limiting the understanding of structural and managerial sources of inefficiency. Most existing analyses rely on aggregate or country-specific indicators, which overlook environmental dimensions and treat efficiency primarily through technical or cost aspects. Moreover, conventional efficiency indicators do not explicitly capture ecological pressures related to fertilizers, pesticides, or energy use, thus underestimating sustainability challenges in crop production. This study addresses these limitations by employing a stochastic frontier model with Kumbhakar & Heshmati decomposition, which distinguishes persistent (structural) from residual (transitory) inefficiency while incorporating environmental inputs. In doing so, it provides a more comprehensive and policy-relevant assessment of agricultural eco-efficiency across the EU and Serbia.

In addition, eco-efficiency is increasingly used as an indicator of the performance of agricultural production, not only from the perspective of sustainability but also in the context of introducing circular economy principles and advancing the green transition. Picazo-Tadeo et al. [42] emphasize that the assessment of eco-efficiency can help agricultural policymakers design instruments capable of achieving both the general objectives of agricultural sustainability and the sustainability of specific production systems. Beyond contributions to public policy, research often formulates recommendations for improving the economic and ecological performance of decision-makers.

Most studies focus on crop production and dairy farming [43,44,45,46,47,48,49,50,51,52,53,54], olive production [55], as well as agriculture more broadly [33,38,56,57]. At the same time, numerous studies cover different geographical areas and national contexts [30,58,59,60]. Finally, the concept of eco-efficiency also occupies an important place within the broader body of scientific literature on sustainable development, where it is considered a valuable indicator at three analytical levels: macro (national economies), meso (regions), and micro (individual enterprises) [61,62].

This multi-level relevance of eco-efficiency highlights its potential as a unifying metric for evaluating agricultural sustainability across diverse contexts, including Serbia’s crop sector, thereby reinforcing the importance of the comparative analysis undertaken in this study.

3. Materials and Methods

In the literature on eco-efficiency assessment, three main methodological approaches are commonly distinguished: the ratio approach, material flow analysis, and the frontier approach [63]. The ratio approach views eco-efficiency as the relationship between the economic value of a product and its environmental burden, but it is feasible only when both components can be expressed in a single comparable measure [61]. Material flow analysis most often relies on life cycle assessment (LCA), which covers the entire life cycle of a product—from resource extraction to disposal—thereby providing a detailed insight into the ecological footprint. However, it is highly demanding and often depends on approximations due to the complexity of collecting complete information [64,65].

The third, frontier approach, is the most widely applied in empirical studies because it simultaneously accounts for both economic and ecological inputs and outputs [66,67,68]. This approach is further divided into non-parametric techniques, based on mathematical programming such as data envelopment analysis (DEA), and parametric techniques rooted in econometric modeling, within which deterministic and stochastic frontier production functions can be distinguished. The key advantage of stochastic models lies in the use of a composite random error term, which enables the separation of inefficiency from internal and external random disturbances, thereby allowing for a more precise analytical examination of eco-efficiency determinants. Since agricultural production is strongly influenced by natural factors and shaped by agricultural policy, its exposure to external shocks is significant. Therefore, the following analysis places emphasis on stochastic frontier analysis (SFA) as a robust framework that explicitly integrates external influences into the assessment of eco-efficiency.

In line with the above considerations, this study applies the frontier approach, with eco-efficiency assessed using a stochastic frontier production (SFA) function. Although SFA was initially developed for measuring technical efficiency, extensions of the model have been introduced that allow not only for the estimation of overall efficiency but also for its decomposition into constituent parts: the persistent (time-invariant) component and the residual (time-varying) component. The advantage of these models is that, alongside efficiency assessment, they enable the identification of determinants underlying the observed patterns. If the persistent component dominates (or, equivalently, if residual inefficiency is small), the interpretation points to external factors beyond the control of farms, such as climatic conditions or the regulatory and administrative framework of agricultural policy. Conversely, the dominance of the residual component (or consistently low persistent efficiency) suggests that the main causes of inefficiency lie within the control of producers, such as farm characteristics, choice of technologies, and management practices—thus directing recommendations toward managerial and technological interventions.

Accordingly, this study estimates an SFA model as described in Kumbhakar & Heshmati [65]:

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

(1)

In this model, the error term ${\epsilon }_{it}$ can be decomposed into a stochastic component ($v_{it}$), which captures the influence of random factors affecting the variability of the observed output, and the remaining error term $u_{it}$, which represents inefficiency and is influenced by factors that can be considered under the control of production units. Accordingly, it holds that ${\epsilon }_{it}=v_{it}-u_{it}$.

The key feature of this model lies in the decomposition of the inefficiency term $u_{it}$ into a time-invariant component $u_i$, which reflects persistent eco-inefficiency, and a time-varying component ${\tau }_{it}$, which captures residual eco-inefficiency over time. Thus, it holds that $u_{it}=u_i+{\tau }_{it}$.

It is evident from this specification that the model is well-suited for assessing eco-efficiency using panel data.

Regardless of whether the specification assumes individual effects ${\alpha }_i$ to be fixed or stochastic, the estimation of the model is carried out through a multi-step procedure. In the case of stochastic frontier production function models with a fixed-effects specification, the estimation process typically consists of four steps.

In the first step, it is necessary to apply either the LSDV (least squares dummy variable) method or the covariance method in order to obtain estimates of the unknown parameters ${\beta }_{n}s$. By applying the covariance method, the individual effects ${\alpha }_i$ are eliminated from the model.

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

In the third step, based on the estimated values ${\widehat{\beta }}_{n}s$ and ${\widehat{u}}_i$, residuals are calculated as ${\eta }_{it}=y_{it}-x_{it}^{\prime }\widehat{\beta }+{\widehat{u}}_i$, which now include ${\beta }_0+v_{it}-{\tau }_{it}$ (with ${\beta }_0$ and ${\omega }_{it}$ being excluded from the model in the previous step by averaging the pseudo-residuals for each observational unit i). At this stage of model estimation, certain assumptions about the distribution of the components $v_{it}$ and ${\tau }_{it}$, must be introduced: $v_{it}~N(0,{\sigma }_v^2)$ and ${\tau }_{it}~N^{+}(0,{\sigma }_{\tau }^2)$. Thus, it is assumed that time-varying eco-inefficiency follows a half-normal distribution. However, the generalization of the distribution for the ${\tau }_{it}$, component, i.e., the introduction of a truncated-normal distribution, is not possible within this class of models.

By treating the residuals ${\eta }_{it}$ as the dependent variable, a regression model is defined to estimate the intercept term ${\beta }_0$. It is important to note that by examining the distribution of residuals ${\eta }_{it}$, conclusions can be drawn regarding the presence of time-varying eco-inefficiency. If the residuals ${\eta }_{it}$ are negatively skewed, it is reasonable to proceed with the analysis of time-varying eco-inefficiency. On the other hand, if the residuals are normally distributed, then it holds that ${\tau }_{it}=0,$ and the model reduces to one with time-invariant eco-inefficiency. Finally, in the fourth step, the estimation of the time-varying eco-inefficiency component ${\tau }_{it}$ is carried out.

Given that the specified model is based on a panel regression framework, it was necessary to conduct standard diagnostic tests to ensure the validity of inference (homoskedasticity, panel independence, autocorrelation, and multicollinearity). Specifically, homoskedasticity was tested using the modified Wald test for groupwise heteroskedasticity; panel independence was verified with Pesaran’s test of cross-sectional dependence; and autocorrelation was tested using the Wooldridge test for autocorrelation in panel data. Potentially harmful multicollinearity was examined through the values of the VIF indicator. The choice between fixed and random effects was verified using the Hausman test to ensure appropriate model specification.

This approach to assessing eco-efficiency allows for certain adjustments to obtain the desired productivity indicator. Since eco-efficiency is defined as the ratio of achieved output to environmental pressure, introducing variables that represent the ecological footprint on the right-hand side of the equation makes it possible to estimate eco-efficiency using the SFA model.

In order to construct a representative sample for the assessment of eco-efficiency, it is crucial to rely on methodologically grounded and internationally comparable data. Ad hoc surveys may be useful, but they are costly, logistically demanding, and difficult to implement over long time series. Consequently, they often fail to provide the coverage and consistency required for cross-country comparisons. By contrast, statistical data from official institutions have the advantage of standardized definitions, harmonized collection procedures, and international comparability of results.

In this context, the only source that simultaneously meets the requirements of representativeness, harmonization, and temporal consistency at the farm level is the Farm Accountancy Data Network (FADN), a system for the continuous collection and processing of annual production, economic, and financial data from farms in the EU. Based on FADN microdata, the scientific literature has widely constructed and applied eco-efficiency indicators to assess agricultural performance; see, e.g., [69,70,71,72,73,74]. In line with this, the present study uses weighted FADN data obtained from the official website of the European Commission (FADN Public Database).

The analysis covers a total of 26 EU countries (Malta excluded due to missing data) for the period 2015–2023. These data were complemented with data relating to the Republic of Serbia, obtained from the official FADN office within the Ministry of Agriculture of the Republic of Serbia. The selected period was determined by the fact that available Serbian data of sufficient quality are only available from 2015 onward, while 2023 represents the most recent year for which data are accessible.

As the dependent variable in the eco-efficiency model, we use the weighted average value of agricultural production in the analyzed countries, expressed in EUR per farm. More specifically, the construction of this variable for farms specialized in crop production within the FADN sample is based on the total value of crop output as well as the value of other plant products.

As independent variables in the eco-efficiency model, we used variables representing environmental pressures, which in agricultural production correspond to the use of fertilizers, plant protection products, and energy. Based on the available data from the FADN sample, the following proxies were applied:

• Fertilizers (EUR/farm): harmonized monetary expenditures on mineral and organic fertilizers applied on farms;
• Plant protection products (EUR/farm): harmonized expenditures on pesticides and herbicides;
• Energy (EUR/farm): harmonized spending on electricity, fuels, and other energy sources used in farm operations.

These proxies for environmental pressure (fertilizers, plant protection products, and energy) are expressed in monetary terms. Although this is standard in FADN-based analyses, it has clear limitations: expenditures do not necessarily reflect physical input use or the true intensity of environmental impacts. This constraint stems from the structure of FADN data, which—particularly in the case of Serbian FADN—does not provide detailed physical quantities or emissions. Consequently, the indicators should be interpreted as relative measures of environmental input use, rather than as precise measures of ecological burden.

It is important to note that all monetary variables, including total output and input costs, were expressed in real terms and deflated to constant 2015 prices using Eurostat’s agricultural input and output price indices (i.e., the index of producer prices of agricultural products for output and the index of purchase prices of the means of agricultural production for inputs). This procedure eliminates the effects of inflation and cross-country price fluctuations, ensuring temporal comparability of efficiency estimates.

Following the assessment of eco-efficiency, a cluster analysis was performed at the country level (EU-26 [EU-27 excluding Malta] + Serbia) using a set of standardized eco-efficiency indicators (economic output and key resource/environmental pressures expressed in a common unit, predominantly per hectare). All numerical variables were standardized to conform to a normal distribution (μ = 0, σ = 1). For ease of interpretation of the clusters, mean values of the variables were recalculated in their original scale, and visualization was conducted using a two-dimensional PCA projection (PC1–PC2). The resulting plot displays the countries as points, together with centroids and cluster contour lines, while the axes indicate the proportion of explained variance.

4. Results

Descriptive statistics for the data used in the eco-efficiency assessment are presented in

Table 1. Descriptive statistics for the variables used.

Variable	Range Interval		Interquartile Variation		Mean	Median	CV (%)	Viqr (%)
	Minimum	Maximum	Q1	Q3
Total output (EUR/farm)	13,643.2	723,032.8	38,437.1	178,127.8	128,144.7	86,100.8	97.8	64.5
Fertiliser (EUR/farm)	1281.9	140,135.3	4721.3	22,593.0	17,158.7	11,313.5	111.8	65.4
Crop protection (EUR/farm)	647.4	74,793.8	2390.2	16,467.2	11,253.0	6842.4	108.5	74.7
Energy (EUR/farm)	1585.6	83,328.4	4414.5	15,432.5	12,194.5	7387.2	104.1	55.5

. The table reports basic statistics for the period 2015–2023, in line with the variables included in the analysis. Examining the descriptive statistics for the output variable, a wide range of variation and a relatively high coefficient of variation (97.8%) are observed. It is therefore important to note that the median value of total output (EUR/farm) amounts to EUR 86,100.8.

Regarding the independent variables used in the eco-efficiency model, all variables exhibit relatively high coefficients of variation, which makes the median the most informative indicator of central tendency. The median value of total fertilizer expenditures is EUR 11,313.5 per farm. For crop protection, the median is EUR 6842.4 per farm, while for energy (EUR/farm), the median is EUR 7387.2 per farm. What is particularly noteworthy is that European farmers allocate significant resources to fertilizers, which suggests a preliminary conclusion that the degradation of agricultural land results in increased costs associated with restoring soil quality.

Continuing the descriptive analysis, Figure 1 presents the average production values per hectare by country included in the study. As expected, farmers in Western and Northern European countries achieve the highest results. The Netherlands stands out in particular, with farms reporting an average output of EUR 5789.7 per hectare according to FADN data. On the other hand, Finland, Latvia, and Estonia are the only countries with an output value per hectare of utilized agricultural land below EUR 1000. Serbia has a relatively high average of EUR 1767.0 for the observed period.

In a similar manner, Figure 2 presents the average expenditures per hectare on fertilizers, plant protection products, and energy. It is evident that farmers in countries achieving the highest average output also tend to spend the most on inputs.

However, it is worth noting that the Netherlands and Belgium, despite being among the countries with the highest output per hectare, show only average expenditure levels for the observed inputs. Likewise, countries with the lowest output also record the lowest input expenditures. Specifically, in the Republic of Serbia, average expenditures amount to around EUR 70.0 per hectare for fertilizers and energy, while spending on crop protection averages about EUR 15.0 per ha. These expenditures are considerably lower than the EU average, but it should be emphasized that production results are correspondingly modest compared to EU countries.

Given that the eco-efficiency assessment was conducted using an econometric model based on a panel regression framework, it was necessary to verify the standard assumptions characteristic of panel regression analysis. The first step involved testing for the presence of harmful multicollinearity. The results indicate no concern regarding multicollinearity.

Table 2. Multicollinearity in the eco-efficiency (EE) model.

Variable	VIF
Fertilizer (EUR/farm)	8.74
Crop protection (EUR/farm)	7.19
Energy (EUR/farm)	6.42
Time	1.21
Average	5.89

shows that there is no highly statistically significant correlation among the variables used, as the VIF values remain below the critical threshold of 10. The average VIF value for the indicators included in the eco-efficiency assessment is 5.89.

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

Table 3. 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	${\chi }^2\left(27\right)=438.77$	0.0000	H0 is rejected
Pesaran’s test of cross-sectional independence	Independent panels	$CD=13.95$	0.0000	H0 is rejected
Wooldridge test for autocorrelation in panel data	No first-order autocorrelation	$F_{(1;26)}=9.86$	0.0042	H0 is rejected

. It is evident that all baseline assumptions were rejected, indicating the presence of harmful heteroskedasticity, panel dependence, and first-order autocorrelation. In line with these findings, and in order to address the issues arising from the violation of assumptions, the eco-efficiency assessment was conducted using a panel regression model with robust standard errors.

${\chi }^2\left(\left(27\right)\right)=438.77$$CD=13.95$$F_{(1;26)}=9.86$ Additionally, prior to estimating eco-efficiency using the stochastic frontier production function model based on panel regression, a test was conducted to determine whether the available data were better suited to a fixed-effects or random-effects specification. The appropriate model specification was selected using the Hausman test. The null hypothesis was rejected, and therefore, the subsequent analysis was carried out using fixed-effects panel models. The estimated χ² statistic for the model was 23.10, with a p-value of 0.0001.

Table 4. Estimation of the eco-efficiency (EE) models using fixed effects with heteroskedastic variance.

Eco-Efficiency (TE) Model
Parameter	Variable	Coefficient	Robust Std. Err.
${\mathit{\beta }}_{0.\mathit{E}\mathit{E}}$	Constant	3.1237 a	0.2563
${\mathit{\beta }}_{1.\mathit{E}\mathit{E}}$	ln_Fertiliser (EUR/farm)	−0.0619 a	0.0678
${\mathit{\beta }}_{2.\mathit{E}\mathit{E}}$	ln_Crop protection (EUR/farm)	0.3941 a	0.0762
${\mathit{\beta }}_{3.\mathit{E}\mathit{E}}$	ln_Energy (EUR/farm)	0.5749	0.0542
${\mathit{\gamma }}_{\mathit{t}.\mathit{E}\mathit{E}}$	Time	0.0278 a	0.0032
${\mathit{\sigma }}_{\mathit{u}.\mathit{E}\mathit{E}}$		0.2174
${\mathit{\sigma }}_{\mathit{v}.\mathit{E}\mathit{E}}$		0.1072
${\mathit{\lambda }}_{\mathit{E}\mathit{E}}={\mathit{\sigma }}_{\mathit{u}.\mathit{E}\mathit{E}}/{\mathit{\sigma }}_{\mathit{v}.\mathit{E}\mathit{E}}$		2.0280
${\mathit{\rho }}_{\mathit{E}\mathit{E}}={\mathit{\sigma }}_{\mathit{u}.\mathit{E}\mathit{E}}^2/{\mathit{\sigma }}_{\mathit{E}\mathit{E}}^2$		0.8043
Number of observations		243
Number of countries		27

^a Statistical significance at the significance threshold α = 0.01.

below presents the results of the estimated eco-efficiency (EE) model. As noted, the estimated model is a stochastic frontier production function based on a fixed-effects panel regression with robust standard errors. In the eco-efficiency model, all variables are statistically significant at the 1% significance level (α = 0.01), with the exception of energy, which was not found to be statistically significant.

Although the energy variable was not statistically significant at conventional levels, it was retained in the model because it represents an essential production input in crop farming. Excluding it could bias the estimated coefficients of other inputs due to omitted variable effects. Moreover, energy use captures differences in mechanization intensity, fuel consumption, and operational efficiency across countries, which are important dimensions of eco-efficiency.

What is far more important for this study, however, is the fulfillment of the assumptions required for estimating the stochastic frontier production function. In the EE model, the indicators ${\lambda }_{EE}$ and ${\rho }_{EE}$ are greater than 1 and 0.5, respectively, which clearly confirms the validity of applying the stochastic frontier production function model.

In the subsequent stage of the analysis, the final estimation of eco-efficiency was performed. As noted, the Kumbhakar & Heshmati [68] model provides an assessment of both residual and persistent efficiency, as presented in

Table 5. Estimation of eco-efficiency based on the Kumbhakar & Heshmati (1995) [68] fixed-effects model.

Efficiency	Number of Observations	Mean	Standard Deviation	Standard Error	Minimum	Maximum
Residual EE	243	0.9362	0.0245	0.0016	0.8248	0.9754
Persistent EE	243	0.6330	0.1403	0.0090	0.4087	1.0000
Total EE	243	0.5926	0.1325	0.0085	0.3496	0.9552

.

Regarding the assessment of total eco-efficiency (total EE) in the analyzed countries, it is evident that Total EE stands at 0.5926, indicating that the same level of output could be achieved with a substantial reduction in input use. Given that persistent EE is significantly lower than residual EE, it is reasonable to focus on those factors under the control of farms when seeking improvements in economic and ecological performance. Conversely, the relatively high value of residual eco-efficiency suggests that external factors—such as agricultural policy measures or climatic conditions—do not exert a decisive influence on the observed level of efficiency. In simple terms, enhancing the economic and ecological performance of agricultural production in EU countries can primarily be achieved by focusing on the characteristics of producers and farms themselves.

An illustrative representation of eco-efficiency is provided in Figure 3 and Figure 4. Figure 3 shows the structure of overall EE, while Figure 4 depicts residual, persistent, and total EE across the time dimension.

When analyzing eco-efficiency in the Republic of Serbia separately, the results are somewhat unexpected. Specifically, total eco-efficiency for Serbia, based on the analyzed sample, amounts to 63.0%, placing the country within the largest group of nations. Furthermore, residual eco-efficiency is at 93.5%, while persistent eco-efficiency is at 67.4%, which does not deviate significantly from the EU average.

These results are complemented by an analysis of the structure of achieved EE (

Table 6. Structure of crop production on achieved eco-efficiency.

Eco-Efficiency	0.01–25.00	25.01–50.00	50.01–75.00	75.01–100
Proportion of countries (%)	0.00	29.6	59.3	11.1

). It is evident that the largest share of countries (59.3%) achieve EE between 50.01% and 75.00%, while the proportion of highly efficient countries (above 75%) is only 11.1%. Conversely, 29.6% of the countries record EE levels below 50%.

The temporal dimension of the analysis reveals substantial variation in eco-efficiency trajectories among EU Member States and Serbia. While most Western and Northern European countries (e.g., France, Denmark, the Netherlands, Germany, and Ireland) exhibit a relatively stable and high level of eco-efficiency, Central and Eastern European countries, including Serbia, display more pronounced fluctuations over time. In Serbia, the gradual improvement observed after 2019 reflects the combined effect of the gradual modernization of machinery and improved access to knowledge and advisory services through IPARD support.

In addition, the research results were complemented by a cluster analysis aimed at grouping countries according to their estimated eco-efficiency (Figure 5). The cluster analysis with k = 3 in the PCA plane (PC1 = 71.3%, PC2 = 25.4% of variance) identified three segments: (i) a cluster with negative PC1 values, composed mainly of Southern and Eastern European countries, characterized by lower intensity of the main indicators; (ii) a cluster around PC1 ≈ 1 (Belgium, Ireland, Denmark, France, the Netherlands, and Germany), with higher values on PC1 and internal differences along PC2; and (iii) a right-side “extreme” cluster (e.g., Czechia and Slovakia) with atypically high values on PC1.

Serbia belongs to the first cluster (PC1 < 0; PC2 slightly > 0), which implies scope for increasing economic performance per unit of resources while controlling for the pressures that differentiate countries along PC2.

The cluster analysis further highlights three distinct performance groups: (1) high-efficiency frontrunners, predominantly in Western Europe; (2) moderately efficient countries, such as Croatia, Hungary, and Poland, which are converging toward the frontier; and (3) lower-efficiency performers, including Serbia, Romania, and Bulgaria, characterized by structural constraints and limited technological adoption. Despite these gaps, the positive trend in Serbia’s residual eco-efficiency suggests an ongoing adjustment process, indicating that internal improvements are occurring even under external constraints.

Across the EU sample, the dynamics of eco-efficiency change align closely with the pace of implementing eco-schemes, digitalization, and knowledge-transfer measures under the Common Agricultural Policy (CAP). Countries that have integrated these instruments more effectively tend to demonstrate both higher efficiency levels and stronger year-to-year consistency. This heterogeneity underscores the need for targeted interventions that address structural weaknesses and enhance resilience at the farm level.

5. Discussion

The results of this study reveal that the average eco-efficiency of crop production in the EU and Serbia remains relatively modest, at 59.26%. This implies that nearly 40% of environmental pressures from fertilizers, plant protection products, and energy could, in theory, be reduced without compromising current levels of output. Similar findings are reported in previous studies that emphasize the existence of substantial efficiency reserves within European agriculture [42,43,72]. The persistence of such inefficiencies underscores the urgency of implementing farm-level improvements that can simultaneously enhance productivity and reduce ecological burdens.

Decomposition of the efficiency scores provides further insight. The high residual eco-efficiency (93.62%) suggests that transitory shocks—such as climatic fluctuations or short-term changes in agricultural policy—play only a minor role in determining performance. Conversely, the lower level of persistent eco-efficiency (63.30%) indicates that systematic inefficiencies dominate, which are closely linked to internal factors at the farm level, such as farm structure, management practices, and technology adoption. This aligns with findings from studies showing that managerial capacity, skills, and production strategies are critical determinants of long-term efficiency in agriculture [75,76,77].

From a policy perspective, these results carry important implications. The relatively limited impact of external factors such as climate conditions or short-term policy interventions suggests that agricultural policies may only influence eco-efficiency indirectly and over longer time horizons [15,16]. While measures under the CAP—such as agri-environmental schemes, eco-schemes, and conditionality—can provide enabling frameworks and financial incentives, their effectiveness ultimately depends on how individual farmers adapt their practices. In this sense, the literature emphasizes that policy support should be accompanied by investments in knowledge transfer, advisory services, and digital technologies that can strengthen farmers’ decision-making capacity [13,14,71].

For Serbia, the findings are particularly noteworthy. With a total eco-efficiency score of 63.0%, Serbia is broadly aligned with the EU average, indicating that inefficiency is not primarily a function of its candidate status but reflects structural similarities with other Southern and Eastern European countries. Cluster analysis confirms this, placing Serbia among the group of lower-intensity performers, while highlighting its potential to converge toward EU frontrunners through improvements in farm-level resource management. Similar patterns have been reported in comparative studies of transition economies, where structural adjustments at the farm level played a decisive role in bridging the eco-efficiency gap [58,74].

Taken together, the results suggest that the main leverage points for improving eco-efficiency in both the EU and Serbia lie within farms themselves. Persistent inefficiency highlights the need for technological upgrading, more sustainable nutrient management, improved soil practices, and organizational changes at the farm level. External factors—such as climate variability or agricultural policy frameworks—appear to have only limited influence in the short run, consistent with the view that policy impacts on eco-efficiency materialize primarily over the long term through structural incentives and institutional support [5,6,16]. Thus, strategies to enhance eco-efficiency should focus first on internal drivers, while policies should aim to enable and sustain these farm-level transformations over time.

The decomposition of eco-efficiency into persistent and residual components provides valuable insights into the nature of inefficiency across countries. The persistent component (persistent EE) largely captures structural and managerial features of crop production systems that remain stable over time. These include farm size, land-use intensity, machinery capital per hectare, crop diversification, and the adoption of modern cultivation practices. In countries such as France, Germany, and the Netherlands, high persistent EE is associated with advanced technology use, efficient nutrient management, and well-established advisory systems that enhance managerial performance at the farm level. Conversely, lower persistent EE in Serbia, Romania, and Bulgaria reflects smaller average farm size, limited mechanization, and lower availability of knowledge transfer services.

The residual component (residual EE) reflects the capacity of farms to adapt to short-term changes in production and environmental conditions. Higher residual EE, observed in countries like Ireland and Denmark, indicates resilience and flexibility in managing input fluctuations and climate variability. In contrast, lower residual efficiency in several Central and Eastern European countries can be attributed to the stronger influence of weather variability, unstable market conditions, and less flexible production structures.

Climate variability plays a crucial role in shaping short-term fluctuations in agricultural eco-efficiency, particularly through its effects on yields, input productivity, and resource use stability. Countries exposed to frequent droughts or temperature extremes, such as those in Southern and Eastern Europe, often exhibit greater variability in residual efficiency. At the same time, it is important to acknowledge the limitations of using monetary expenditures as proxies for environmental pressures. While this approach ensures data harmonization and cross-country comparability, it may not fully reflect the physical intensity or environmental footprint of input use, especially for fertilizers and energy.

Taken together, these results suggest that persistent EE is predominantly influenced by farm management and structural factors, while residual EE depends more on natural and policy-related variability. This distinction underscores the importance of integrating both long-term structural measures (investment in technology, education, and advisory services) and short-term adaptive instruments (climate risk management, input subsidies, and market stabilization tools) into policy frameworks that aim to improve agricultural eco-efficiency.

6. Conclusions

This study provides new insights into the eco-efficiency of crop production across EU member states and the Republic of Serbia, applying a stochastic frontier approach to panel data for the period 2015–2023. The results demonstrate that average eco-efficiency in the analyzed countries is relatively modest, at 59.26%. This suggests that nearly 41% of environmental pressures associated with key inputs—fertilizers, plant protection products, and energy—could potentially be reduced without sacrificing output. Such findings highlight both the untapped potential for resource conservation and the scope for aligning agricultural practices more closely with sustainability objectives.

Decomposition of the eco-efficiency scores further revealed that residual efficiency is high, while persistent efficiency remains relatively low. This indicates that systematic inefficiencies, rather than transitory shocks, are the primary constraint on eco-efficiency performance. Put differently, external factors such as climate variability or short-term policy measures appear to exert limited influence, whereas internal farm-level characteristics—such as management practices, organizational structures, and technology adoption—emerge as the decisive drivers.

For Serbia, the results are particularly significant. With a total eco-efficiency score of 63.0%, Serbia does not deviate from the EU average, suggesting that its challenges and opportunities largely mirror those of other European countries, especially in Southern and Eastern Europe. The cluster analysis corroborates this interpretation, positioning Serbia among the group of lower-intensity performers but simultaneously highlighting considerable room for convergence toward frontrunners such as Belgium, Denmark, France, Germany, Ireland, and the Netherlands. These results reinforce the argument that, even for a candidate country, the key pathways to improvement are not primarily external alignment with EU standards, but rather internal reforms at the farm level.

From a policy perspective, the findings suggest that while the Common Agricultural Policy (CAP) and related instruments—such as eco-schemes, agri-environmental measures, and payments for ecosystem services—provide a necessary framework, their impact on eco-efficiency is likely to be long-term and indirect. Policy support must therefore be complemented by targeted investments in knowledge transfer, advisory services, green finance, and digital technologies that empower farmers to adopt more sustainable production strategies. In the Serbian context, the integration of such measures through IPARD and future CAP alignment will be critical in enabling farm-level transformations.

A limitation of this study lies in the use of national-level FADN averages representing typical crop farms. While this approach enables cross-country comparability, it may mask significant within-country differences and introduce the risk of ecological fallacy when interpreting results at the micro (farm) level. Future research based on disaggregated farm-level data could provide a more detailed assessment of eco-efficiency patterns and drivers.

Overall, the results underscore that improving eco-efficiency in crop production is primarily a matter of addressing persistent inefficiencies through farm-level interventions. Strategies should prioritize enhancing farmers’ technical skills, promoting sustainable nutrient and soil management, improving organizational efficiency, and encouraging the adoption of innovative technologies. At the same time, external policy measures should continue to provide the enabling environment and long-term incentives necessary to sustain these changes. By focusing on this dual approach—farm-level improvements supported by policy frameworks—both the EU and Serbia can make meaningful progress toward reducing agricultural ecological pressures while safeguarding competitiveness and ensuring food security in the context of the green transition.

The obtained results also have strong implications for the ongoing monitoring of the CAP. In particular, the estimated eco-efficiency scores for fertilizers and plant protection products correspond closely with the harmonized risk indicators (HRI 1 and HRI 2), which measure progress in reducing the use and risk of chemical pesticides. These findings underline the potential of the eco-efficiency framework to complement existing CAP monitoring tools by providing a quantitative assessment of resource use and environmental performance at the farm level. Consequently, the proposed approach can support the evaluation of eco-schemes, GAEC standards, and IPARD measures aimed at promoting sustainable input management and improving the environmental outcomes of European agriculture.

From a policy perspective, future strategies should focus on strengthening data collection systems to include physical indicators of input use and emission intensity, enabling a more precise link between eco-efficiency and environmental performance. In addition, integrating farm-level monitoring of climate and resource variables could improve the accuracy of cross-country benchmarking. The main limitation of this study lies in the use of monetary proxies for environmental pressures and in the national level aggregation of FADN data, which may obscure regional heterogeneity. Future research should therefore aim to extend the analysis to micro-level datasets, combine physical and economic indicators, and explore the dynamic effects of green investments and digital technologies on eco-efficiency improvement.