Does Agro-Eco Efficiency Matter? Introducing Macro Circular Economy Indicator into Profitability Modeling of Serbian Farms

Abstract

The transition toward sustainable and circular agricultural systems is increasingly important, yet evidence linking circularity and farm profitability in transition economies remains limited. This study examines the determinants of farm profitability in Serbia by combining micro-level structural and productivity indicators with a macro-level agro-eco efficiency measure, used here as a sector-wide ecological pressure indicator rather than a direct proxy for circular practices. Using a balanced Farm Accountancy Data Network (FADN) panel of 443 farms (2015–2022) across dairy, mixed, field crop, and fruit & wine sectors, dynamic panel estimators (difference and system Generalized Method of Moments-GMM) reveal strong sectoral heterogeneity. Asset turnover is the primary driver of profitability in field crops and perennial systems, while dairy farms benefit from scale and land productivity. Energy intensity consistently reduces profitability across all sectors. Agro-eco efficiency shows a negative effect in livestock-based systems, indicating higher sensitivity to macro-ecological pressures. These findings suggest that environmental and economic vulnerabilities differ across production systems, highlighting the need for sector-specific strategies aimed at improving resilience rather than inferring the profitability of circular technologies.

1. Introduction

The global agri-food system is undergoing a profound transformation driven by multiple pressures that extend far beyond the farm gate. Climate change, soil degradation, water scarcity, and biodiversity loss increasingly constrain agricultural productivity, while economic volatility and geopolitical disturbances continue to reshape global food markets. The traditional linear model of agricultural production—based on intensive resource extraction, dependence on external inputs, and large volumes of waste—has proven unable to ensure long-term ecological stability or economic resilience [1,2]. As these systemic limitations become more evident, the transition toward sustainable and circular production models has emerged as a strategic imperative for agriculture worldwide [3,4].

Within the European Union, this transition has been institutionalized through the European Green Deal [5] and the Farm to Fork Strategy [6], which identify agriculture as a central pillar in achieving climate neutrality, improving resource efficiency, and strengthening rural communities. These policy frameworks emphasize nutrient recycling, reduced dependence on chemical inputs, and the revalorization of agricultural residues—thus embedding the principles of circular economy (CE) into the long-term development trajectory of the agri-food sector. Although CE was initially conceptualized within industrial ecosystems, its application to agriculture—often referred to as circular agriculture (CA)—has rapidly gained relevance. CA seeks to close biophysical loops within farming systems through practices such as composting, anaerobic digestion, precision input management, and the use of renewable energy [7]. A growing body of empirical research demonstrates that circular practices can enhance farm profitability by improving input-use efficiency, reducing exposure to market fluctuations in fertilizer and energy prices, and generating new revenue streams from by-products [8,9,10]. For example, farms employing manure-based biogas production or nutrient recycling often achieve lower input costs and higher resilience to shocks. Similarly, the use of crop residues for mulching or composting improves soil fertility, reducing long-term dependence on external fertilizers. Such evidence suggests that circularity and economic performance are not mutually exclusive but can reinforce each other when appropriate technological and institutional conditions are in place.

Yet, despite these advantages, circular practices remain unevenly implemented across countries, sectors, and farm types. Adoption barriers relate not only to investment costs but also to knowledge gaps, risk perceptions, and limited access to advisory services [11]. For many small and medium-sized farms, particularly in transition economies, profitability itself becomes a prerequisite for adopting circular solutions: financially constrained farms lack the capacity to invest in new technologies or redesign production processes. This circularity–profitability relationship is therefore bidirectional and context-dependent, making its empirical assessment crucial for evidence-based policymaking. In Serbia and other Western Balkan countries, this context is further shaped by structural dualism, where a small number of large commercial farms coexist with a dominant share of small family holdings. These small farms often operate under conditions of limited capital, low mechanization, and strong dependence on subsidies. At the same time, the region possesses considerable untapped potential for circular practices, particularly in livestock manure management, crop residue utilization, and decentralized bioenergy production [7,12]. However, the limited availability of farm-level environmental indicators presents an additional challenge, making it difficult to evaluate the environmental dimension of sustainability at the micro level [11,12]. Consequently, agricultural eco-efficiency indicators derived from national GHG emission accounts—although imperfect—offer a feasible proxy for assessing environmental pressures associated with agricultural output.

Despite growing interest in sustainable agriculture, empirical evidence linking circularity and farm profitability remains limited, particularly in transition economies where data availability and structural diversity pose additional challenges. Most existing studies address economic and environmental aspects separately, leaving important questions about their interaction unanswered.

Against this background the main objective of this study is to identify the key determinants of farm profitability in Serbia, with particular emphasis on the structural heterogeneity of its agricultural sector. In line with the circular-economy orientation of the paper, the analysis incorporates an agro-ecological dimension by operationalizing environmental pressure through a macro-level Agro-Eco Efficiency (AEE) indicator, which reflects sector-wide resource efficiency and climate-related burden. To this end, the analysis employs a dynamic panel approach that integrates micro-level indicators of farm size, capital efficiency, and productivity with a macro-level measure of agro-eco efficiency, interpreted in this study as a sector-wide ecological pressure indicator due to the absence of farm-level environmental data. This combined perspective makes it possible to capture both internal and external drivers of farm performance. By examining dairy, mixed, field crop, and fruit & wine farms separately, the study provides a nuanced understanding of how sustainability-related conditions—reflected through agro-eco efficiency—interact with structural characteristics to shape profitability outcomes.

2. Literature Review

Research on agricultural performance has long examined how structural, managerial, and policy variables shape farm profitability. Over the last decade, this inquiry has broadened to incorporate resource circularity—waste valorization, nutrient recycling, and renewable energy—as potential sources of competitive advantage alongside environmental gains. Empirically, three strands dominate: (i) FADN-based studies of profitability and viability across heterogeneous European systems; (ii) work on resource economics and the transition to circular models; and (iii) farm-level modeling that integrates profitability with elements of circularity. The review synthesizes these strands and, on that basis, delineates the research gap addressed by the present study.

2.1. Determinants of Farm Profitability and Sustainability (FADN-Based Research)

A mature body of evidence relying on EU FADN micro data identifies scale, capital efficiency, leverage, specialization/diversification, and policy support as dominant drivers of profitability. Using a large FADN panel (2007–2018), Kryszak et al. [4] show that larger and more capital-efficient holdings attain higher ROA; the effect of subsidies is heterogeneous—beneficial for smaller farms but neutral or diminishing among the largest—indicating structural dualism. Analyses based on Italian FADN microdata [13] indicate that CAP direct payments contribute to farm income stabilization in the short term, but their effects on long-term economic viability remain limited, while rural development measures play a more significant role in supporting structural adjustment, diversification, and investment. At the level of performance stability, Slijper et al. [14] use FADN micro data (2007–2018) to show that larger, better capitalized, and more diversified farms are more resilient to shocks, linking structural characteristics to temporal income stability. In this context, resilience refers to both ecological and socio-economic dimensions: ecological resilience denoting the capacity of farm systems to absorb biophysical shocks, and socio-economic resilience reflecting the stability of income and the ability to recover from market or policy disturbances. Barnes et al. [15] find that moderate diversification is associated with higher profit efficiency, while excessive dispersion of activities can erode managerial focus and reduce efficiency outcomes. Evidence from Central and Eastern Europe points in the same direction. Hloušková et al. [16], using FADN data for 2010–2018 from the Czech Republic, Slovakia, and Poland, report higher profit efficiency in crop and mixed systems relative to livestock, largely due to lower labor and capital requirements—an insight consistent with Latruffe et al. [17], who examine Polish farms using 1996–2000 microdata, document higher technical efficiency among larger, better-capitalized holdings led by more educated managers. Complementing this regional evidence, Galluzzo [18], analyzing Slovenian FADN farms over 2004–2013, shows that farm size, specialization, and capital endowment strongly influence income performance, while the role of CAP subsidies remains heterogeneous across production types, reflecting the structural characteristics of a post-transition agricultural sector. Financial-structure studies complement this picture. Using FADN (2004–2018), Martinho [19] documents pronounced spatial heterogeneity in ROA/ROE, with liquidity and leverage materially shaping returns; the most profitable clusters concentrate in Northern Italy, Denmark, and the Netherlands, which underscores the case for region-tailored CAP design. Drawing on national FADN microdata (2015–2021), Miljatović et al. [20] show that asset turnover and farm size are the most powerful predictors of long-term economic viability, whereas subsidies are not statistically significant. For small family farms, Tošović-Stevanović et al. [21], using survey data collected in 2019 from 550 Serbian farms, rank output prices and distribution channels above land size or soil quality as determinants of survival, emphasizing commercialization and market access. Additional evidence from the Western Balkan–ECE region comes from Croatia. Očić et al. [22], using FADN data for 2014–2018, analyze specialized dairy farms and show that structural characteristics strongly influence economic outcomes. Larger farms (50+ cows) achieve net value-added levels comparable to EU averages, while smaller farms suffer from lower yields, higher operating costs, and heavy subsidy dependence. These results mirror patterns observed in other transition economies and reinforce the broader regional finding that profitability is primarily driven by scale, capital efficiency, and production intensity. Kahanec et al. [23] analyze agricultural sustainability pathways using Eurostat and FAOSTAT data for Slovakia and the Czech Republic over the period 2000–2022, showing that improvements in energy-use efficiency, structural modernization, and investment in sustainable practices significantly enhance long-term farm viability. Their findings confirm that farms in post-transition contexts operate under intertwined economic and ecological constraints similar to those observed in Serbia. Moreover, work on Bavarian dairy farms (2010–2016) demonstrates that feed efficiency and milk yield can simultaneously raise income and reduce GHG per unit of output [24], foreshadowing the profitability–circularity linkage addressed below.

Taken together, FADN-based studies establish a robust “economic core” of profitability determinants—scale, capital use, financial structure, specialization/diversification, and market access—to which circular practices can be meaningfully appended.

2.2. Resource Economics and Transition Toward Circular Models in Agriculture

The transition from linear “take–make–dispose” models [2] toward circular agriculture is grounded in the economic rationale of retaining value within biomass, nutrients, and energy flows. Empirical and conceptual studies increasingly demonstrate that circular economy (CE) practices can improve resource efficiency, reduce external inputs, lower production costs, and generate new revenue streams [1]. These benefits arise from waste valorization, nutrient recycling, energy recovery, and the redesign of production systems toward regenerative loops. Sectoral applications across Europe and the Mediterranean illustrate how CE interventions translate into measurable economic outcomes. In Andalusian olive systems (Spain, 2018–2021), Martínez-Moreno et al. [9] identify 59 circular practices spanning the full production cycle, showing that adoption is primarily driven by institutional support and cooperative networks, while major obstacles include high upfront costs and knowledge gaps. Complementary evidence from protected horticulture in Spain (2020–2022) confirms that circular soil biodisinfection based on biomass inputs improves both economic and environmental performance: Castillo-Díaz et al. [10] report cost reductions exceeding 6% and substantial decreases in chemical and water use. A growing body of research develops methodological tools for assessing circularity at the farm or sector level. Using the ECOGRAI framework applied to mixed farming systems in France (2019–2021), Rukundo et al. [25] propose a set of 25 performance indicators covering resource efficiency, waste reduction, emissions, and energy flows, offering one of the most comprehensive micro-level CE evaluation instruments to date. In livestock processing, Molina-Moreno et al. [26] rely on Life Cycle Assessment and multi-criteria evaluation to design a suite of indicators for the pig-industry wastewater sector, documenting substantial gains in energy efficiency and environmental performance from water reuse and anaerobic digestion. Economic feasibility is a decisive condition for widespread CE adoption. In the Italian horticultural sector, Bentivoglio et al. [27] demonstrate through discounted cash-flow analysis that the conversion of vegetable waste into biogas or biomethane becomes financially viable only under supportive policy frameworks and optimized substrate portfolios, highlighting the sensitivity of CE investments to subsidy design and input composition. Similar conclusions emerge from modeling approaches to closed-loop systems in Asian livestock production. Jeng et al. [8] show that integrating manure recycling and composting into a closed-loop farming system significantly increases profitability without compromising production stability, while Lin et al. [8] find that high recycling rates and efficient by-product use enhance both farm income and carbon-reduction outcomes.

At the conceptual level, the broader framework of Circular Agro-Economies, as articulated by Ajayi et al. [28], emphasizes the creation of new value chains—biofertilizers, bioenergy, feed—and argues that the transition requires simultaneous advances in technology, financing mechanisms, and institutional coordination. Together, these studies indicate that CE in agriculture is not merely a technological shift but an economic transformation that depends on resource efficiency, organizational innovation, and enabling policy environments.

2.3. Integration of Profitability and Circularity: Empirical Evidence and Modeling Approaches

An emerging literature links circular practices to financial outcomes using systems, frontier, and econometric approaches. System-dynamics models with sensitivity analysis demonstrate that closed-loop livestock–crop systems—where manure is recycled into fertilizers—can raise profits while reducing environmental costs; profitability is sensitive to recycling rates and relative prices [8]. In dairy waste management, scenario analysis indicates that biodigesters and manure recycling can increase profitability by up to ~27% while lowering methane emissions and disposal costs [29], clarifying the mechanism by which material and energy loops reduce purchased inputs and waste externalities. Frontier-based and indicator-driven studies align with this mechanism. At the country level, Khan et al. [30] use DEA (slack-based) with undesirable outputs (CO₂) and show that investment in renewables and technological upgrading is associated with higher eco-efficiency, a composite performance construct linked to cost reduction. At farm level in Serbia, Novaković et al. [31] apply SFA with undesirable outputs to dairy FADN data (2015–2021) and report mean technical efficiency = 0.72 and eco-efficiency = 0.63; better resource management (feeding, energy) is associated with superior scores, implying scope for profitability gains via manure recycling and input rationalization. Agronomic innovation reinforces the economic case: Lorenzetti and Fiorini [32] document 12–18% cost reductions and ~20% lower GHG under conservation agriculture. Organizational and market arrangements also condition the profitability effects of circular practices. Evidence from Serbian small family farms shows that output prices and distribution channels are decisive for economic survival [21], suggesting that cooperative solutions, shared infrastructures, and localized bio-loops may complement technological loops. Importantly, recent CE investment studies highlight that circular technologies often require substantial upfront capital and are hindered by limited access to financing; Aranda-Usón et al. [33] show that financial constraints, high initial costs, and uncertain payback periods represent major barriers to adopting circular practices, underscoring the need for supportive funding mechanisms. A validated, multidimensional framework for circular agriculture [12] offers practical building blocks for constructing explanatory variables in econometric profitability models.

2.4. Research Gap and Directions for Future Studies

Despite notable advances, several gaps still constrain causal inference on the profitability–circularity nexus, especially in transition economies such as Serbia. First, profitability models seldom include direct proxies for circular practices; where circularity is assessed, it is typically via separate indicator of eco-efficiency or LCA-a-based assessments (life cycle assessment) rather than embedded within profit equations [12,25,26]. While direct farm-level studies linking circular practices with profitability in transition economies are scarce, several empirical works offer relevant insights. Matysik-Pejaś et al. [34], identify relationships between circular-economy conditions and agricultural economic performance across EU member states, suggesting that improved resource use and waste mitigation correlate with stronger economic outcomes. Panel analysis at the European level shows that circular-economy progress affects key agricultural indicators in advanced and emerging economies [35]. Additionally, comparative eco-efficiency research demonstrates substantial variation in environmental and economic outputs in agricultural sectors, pointing to economic implications of resource pressures [36]. Second, transition contexts remain under-represented relative to EU-15, even though available Serbian evidence underscores the primacy of asset efficiency and scale and highlights commercialization constraints [20,21]. Third, few studies employ dynamic panel estimators capable of addressing persistence and endogeneity [4], despite investment and learning dynamics inherent to CE adoption. Against this backdrop, the present analysis advances the literature by integrating circular-economy indicators into a farm-level profitability framework for a transition economy, providing evidence on whether circular practices are associated with improved financial performance.

3. Materials and Methods

3.1. Data and Sample Description

Serbia’s agricultural sector plays a structurally important role in the national economy. Agriculture contributes approximately 6–7% of national GDP and accounts for 15–17% of total employment, making it one of the most significant productive activities in the country’s economic structure [37,38]. The sector is highly diverse, comprising several key subsectors—including crop production, livestock, fruit and viticulture—each contributing differently to value added and employment. Farm structure is characterized by a predominance of small family holdings, with average farm size significantly below the EU-27 average, while a smaller group of medium and larger commercial farms generates a substantial share of total agricultural output. This dual structure shapes productivity outcomes, investment capacity, and the adoption of new technologies, thereby forming the broader context necessary for interpreting Serbian FADN data.

The empirical analysis is based on a balanced panel dataset derived from the Serbian FADN (Farm Accountancy Data Network) system, which is the only database capable of providing the level of detail and harmonized accounting standards necessary for robust microeconomic assessment. FADN is a comprehensive annual data collection system designed to capture structural, production, economic, and financial information from professionally managed farms. Its standardized methodology ensures cross-farm comparability and makes it uniquely suited for longitudinal analyses of farm performance. The dataset used in this study covers the period 2015–2022 and includes four major production systems: dairy, mixed, field crop, and fruit & wine farms. Farm types are classified according to the FADN TF8 typology, allowing us to distinguish four production systems: dairy, mixed, field crops, and fruit & wine farms. The final balanced sample comprises 98 dairy farms, 61 mixed farms, 236 field crop farms, and 48 fruit & wine farms, each observed continuously over the eight-year period. These production systems represent the dominant segments of Serbian agriculture and reflect substantial diversity in technology use, capital intensity, labor organization, and ecological pressures. The analysis does not use the entire Serbian FADN dataset. Instead, a balanced panel was constructed by including only those farms that

(i) Belonged to one of the four dominant production systems (dairy, mixed, field crops, fruit & wine), and

(ii) reported complete structural and financial data for all years from 2015 to 2022.

Although the national FADN sample is representative of Serbian commercial agriculture, the balanced panel used in this study constitutes a subset of the full dataset, restricted exclusively to farms with uninterrupted reporting. The selection did not involve independent sampling by the authors; all farms meeting the above criteria were included. Only farms that met the minimum FADN economic size thresholds and provided complete annual financial and structural data for the entire period were retained in the sample, ensuring consistency in panel composition and eliminating the need for imputation of missing values.

3.2. Variable Description

The empirical model incorporates a set of economic, structural, and environmental indicators designed to capture the multidimensional determinants of farm profitability across four production systems in the Serbian FADN sample. The dependent variable is Return on Assets (ROA), a standard profitability measure widely applied in agricultural economics as an integrated measure of economic performance because it reflects both operating profitability and the efficiency with which farms utilize their asset base, e.g., [39,40]. Independent variables reflect key dimensions of economic performance, including scale, capital efficiency, production productivity, input cost structure, and resource-use intensity. In addition, one macro-level environmental indicator—Agricultural Eco-Efficiency (AEE)—is included to capture sector-wide environmental pressures that potentially influence farm-level profitability.

Table 1. Definition of variables.

Variable	Code	Formula	Expected Effect
Return on assets (dependent variable)	ROA	Net income/Total assets	Measure of farm profitability
Economic size	ES	Standard Output (SO)	+
Asset turnover ratio	ATR	Total output/Total assets	+
Fixed asset share	FAS	Fixed assets/Total assets	−
Labor productivity	LABPROD	Total output/Labor input (AWU)	+
Land productivity	LANDPROD	Total output/Utilized agricultural area (ha)	+
Energy intensity	ENINT	Energy/Total output	−
Agricultural eco-efficiency	AEE	GHG emissions (macro)/Total agricultural assets (macro)	−

Note: The symbols “+” and “−” indicate the expected direction of the relationship between the explanatory variables and the dependent variable (ROA). A positive sign (+) denotes an expected positive effect on farm profitability, while a negative sign (−) denotes an expected negative effect.

provides the operational definitions, formulas, and expected effects of all variables included in the analysis.

Given the absence of micro-level greenhouse gas (GHG) emission data at the farm level in Serbia, environmental performance in this study is represented by the Agricultural Eco-Efficiency (AEE) indicator, constructed at the macro level and uniformly assigned to all farms within each year. AEE is defined as the ratio of annual national agricultural GHG emissions (expressed in CO₂-equivalent) to the total agricultural asset value of the sector. Annual data on emissions were obtained from the Statistical Office of the Republic of Serbia (RZS). This indicator does not measure farm-level circular practices; instead, it serves as a sector-wide ecological pressure signal aligned with the environmental sustainability dimension of circular-economy frameworks. Because national emission inventories do not distinguish between production systems (e.g., dairy, mixed, field crops, fruit & wine), the macro-level indicator represents the environmental burden of the agricultural sector as a whole rather than farm-specific emissions. The conceptual foundation for this ratio-based approach follows the methodology applied by Stevanović et al. [41], who measured environmental efficiency of enterprises using a similar relationship between total emissions and economic output. However, while Stevanović et al. [41] investigated industrial polluters using PRTR data (SOx, NOx, PM, CO, NMVOC, etc.), the current study adapts the ratio for agricultural sustainability by focusing on GHG emissions from the agricultural sector. This adjustment is appropriate because agriculture contributes to climate pressure primarily through methane (CH₄), nitrous oxide (N₂O), and carbon dioxide (CO₂), which differ from industrial pollutants but remain compatible with the eco-efficiency concept. Accordingly, AEE should be interpreted as an annual macro-ecological constraint faced by the agricultural sector, not as a micro-level indicator of circular practices. In contrast to Stevanović et al. [41], who relate emissions to revenues, this study uses total agricultural assets because (i) reliable annual sector-level revenue/value-added data for Serbia are not consistently available for the full period, and (ii) assets represent a more stable stock variable that captures long-term productive capacity, avoiding volatility caused by weather shocks and price fluctuations that strongly affect agricultural revenues. For these reasons, an asset-based formulation provides a more stable and comparable environmental-pressure indicator under Serbian data conditions while remaining conceptually aligned with the eco-efficiency framework.

3.3. Econometric and Statistical Analysis

Before proceeding with the econometric analysis, all continuous variables were screened for extreme values using Tukey’s fence rule (1.5 × IQR). Observations falling outside the admissible range were removed prior to estimation.

The empirical strategy integrates descriptive statistics, correlation analysis, panel unit root diagnostics, and dynamic panel econometric modeling to examine the determinants of farm profitability across four agricultural production systems. This multi-step analytical framework follows established econometric guidelines for studies involving micro-level panel data in agriculture [42,43].

Descriptive statistics were calculated to summarize the distributional properties of all variables and to capture structural differences across dairy, mixed, crop, and fruit & wine farms [44]. Pairwise Pearson correlations were then used to identify basic linear relationships and potential multicollinearity among variables, providing an initial diagnostic overview before econometric modeling [42].

Before proceeding with regression estimation, the time-series properties of the variables were evaluated using the Im–Pesaran–Shin (IPS) panel unit root test [45]. IPS is suitable for unbalanced panels with heterogeneous autoregressive parameters, such as FADN farm-level data. The IPS results indicate that several variables may be non-stationary in levels, which is expected given the short time dimension (T = 8) and the variability inherent in agricultural production data. To appropriately account for persistence and reduce the risk of spurious regression, the analysis relies on dynamic panel estimators that operate on transformed (first-differenced) variables, thereby removing individual-specific trends and eliminating unit-root behavior in levels [46].

Given the presence of lagged profitability and the likelihood of endogeneity in several explanatory variables, models were estimated using the Generalized Method of Moments (GMM) developed by Arellano and Bond [47] and later refined by Blundell and Bond [48]. The Arellano–Bond difference GMM estimator is well suited for datasets with many cross-sectional units and a relatively short time dimension, as it eliminates unobserved heterogeneity and uses internal instruments derived from lagged levels. For fruit & wine farms, system GMM was preferred because it provides improved efficiency when lagged levels exhibit weak instrument relevance in difference equations [48,49]. Importantly, the application of GMM does not assume stationarity of the variables in levels; rather, consistency relies on first-differencing, fixed effects, and valid moment conditions. By transforming the data and incorporating year fixed effects, the estimator mitigates risks related to non-stationarity in levels and ensures that inference is based on stationary differenced series rather than potentially non-stationary raw series.

The general specification of the dynamic model is given by:

$${\mathrm{R}\mathrm{O}\mathrm{A}}_{\mathrm{i}\mathrm{t}}=\mathsf{\alpha }+{\mathsf{\beta }}_1{\mathrm{R}\mathrm{O}\mathrm{A}}_{\mathrm{i}\mathrm{t}-1}+{\mathsf{\beta }}_2{\mathrm{E}\mathrm{S}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\beta }}_3{\mathrm{A}\mathrm{T}\mathrm{R}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\beta }}_4{\mathrm{F}\mathrm{A}\mathrm{S}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\beta }}_5{\mathrm{L}\mathrm{A}\mathrm{B}\mathrm{P}\mathrm{R}\mathrm{O}\mathrm{D}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\beta }}_6{\mathrm{L}\mathrm{A}\mathrm{N}\mathrm{D}\mathrm{P}\mathrm{R}\mathrm{O}\mathrm{D}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\beta }}_7{\mathrm{E}\mathrm{N}\mathrm{I}\mathrm{N}\mathrm{T}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\beta }}_8{\mathrm{A}\mathrm{E}\mathrm{E}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

(1)

where μ_i captures unobserved farm-specific heterogeneity, and ε_it represents the idiosyncratic error term. i denotes the individual farm and t the year of observation (2015–2022). All variables are measured at the farm–year level and defined in Table 1.

All GMM models include a full set of year dummy variables to control for macroeconomic, climatic, and policy shocks. This is particularly relevant for the Agricultural Eco-Efficiency (AEE) indicator, which varies only across years and therefore reflects sector-wide ecological pressure rather than farm-level circular practices. AEE is thus interpreted as a macro-ecological constraint common to all farms in a given year. Including year dummies ensures that its coefficient is not confounded by unrelated annual fluctuations.

Dynamic panel estimation allows for profit persistence—an empirically documented characteristic of farms and firms where past performance influences current outcomes [50]. It also addresses endogeneity arising from simultaneity, measurement error, or omitted variable bias by using internal instruments and exploiting moment conditions in both levels and differences [43]. In the present study, ROA(t–1), ATR, LABPROD, and ENINT were treated as endogenous variables, FAS as predetermined, while ES, LANDPROD, and AEE were treated as exogenous. In difference GMM, lagged levels of the endogenous variables served as instruments, whereas in system GMM the corresponding lagged first differences were used in the level equation. Lagged levels dated t–2 and t–3 were used as instruments for the differenced equation, whereas lagged first differences (t–1, t–2) served as instruments in the level equation for the fruit & wine model estimated via system GMM. To avoid instrument proliferation—an important concern in short panels with large cross-sectional dimensions—the instrument matrix was collapsed following Roodman [36]. The final specifications rely on 21 instruments for mixed farms, 22 instruments for dairy farms, 23 instruments for field crop farms, and 20 instruments for the fruit & wine system GMM model, all well below commonly recommended upper bounds and ensuring satisfactory instrument-to-observation ratios.

Diagnostic tests were performed to ensure model validity. The Arellano–Bond AR(1) and AR(2) tests verified the absence of second-order autocorrelation, a key requirement for the consistency of GMM estimators [47]. Instrument validity was further assessed using the Sargan test of over identifying restrictions, which confirmed that the chosen moment conditions were appropriate [51]. Additionally, Hansen J-test statistics were computed.

To examine whether the results were sensitive to estimator choice, a static fixed-effects (FE) model with cluster-robust standard errors was estimated as a robustness check. While the FE estimator cannot correctly estimate lagged dependent variables due to dynamic panel bias [52], it does provide a useful benchmark for the signs and magnitudes of the remaining coefficients. The similarity between FE and GMM estimates reinforces the reliability of the findings. Differences in significance across estimators were interpreted cautiously and used to assess sensitivity rather than as evidence of full robustness.

Finally, multicollinearity was assessed using the Variance Inflation Factor (VIF). Low VIF values (<5) confirmed the absence of problematic multicollinearity among the explanatory variables [29], ensuring that coefficient estimates are not distorted by linear dependence between regressors.

Statistical analysis was conducted in R (version 4.4.1), an open-source statistical computing environment, using specialized packages [53,54,55,56], including plm and pgmm for panel data estimation, lmtest and sandwich for statistical testing and robust inference, and AER, dplyr, and tidyr for data management and econometric support, following common practice in empirical agricultural economics research [28,31].

4. Results

Before assessing the determinants of farm profitability across the four production systems, descriptive statistics were calculated for all key economic, structural, and environmental variables.

Table 2. Descriptive statistics of key variables across farm types (Mean ± SD).

Variable	Mixed	Dairy	Field Crops	Fruit & Wine
ROA	0.2667 ± 0.2818	0.2639 ± 1.3142	0.2098 ± 0.4072	0.2299 ± 0.3877
ES	19.03 ± 18.33	24.58 ± 20.79	52.97 ± 51.25	21.91 ± 20.22
ATR	0.4145 ± 0.2956	0.4197 ± 0.2886	0.4362 ± 0.5694	0.3832 ± 0.4256
FAS	0.7781 ± 0.1805	0.8563 ± 0.0959	0.9195 ± 0.0949	0.9408 ± 0.0892
LABPROD	15,142.7 ± 15,723.9	15,617.0 ± 17,201.3	46,714.8 ± 64,881.9	14,740.9 ± 36,779.3
LANDPROD	12,194.5 ± 47,403.3	2035.1 ± 2083.8	3757.7 ± 2965.6	4368.5 ± 4787.3
ENINT	0.0583 ± 0.0671	0.0614 ± 0.0339	0.0955 ± 0.0628	2.6690 ± 0.0518
AEE	0.1174 ± 0.0304	0.1174 ± 0.03034	0.1174 ± 0.0303	0.1174 ± 0.0304

presents the descriptive statistics for all variables used in the analysis, reported separately for mixed farms, dairy farms, field crop farms, and fruit & wine farms.

Table 2 summarizes the descriptive statistics of the key economic, structural, and environmental indicators for the four analyzed farm types. The results indicate pronounced differences between production systems. Dairy and mixed farms record similar average profitability levels, while fruit & wine and field crop farms show slightly lower mean ROA values, although with considerable dispersion across farms. Structural indicators vary markedly across sectors. Capital intensity (FAS) is highest in fruit & wine farms, consistent with the investment-heavy nature of perennial crop production. Productivity indicators also display substantial heterogeneity: field crop farms achieve the highest levels of both labor and land productivity, while mixed and dairy farms remain considerably below these values. Environmental indicators show two distinct patterns. Energy intensity (ENINT) differs strongly across sectors, reflecting substantial variation in energy use and input structure. Field crop farms exhibit the highest ENINT values, consistent with their greater reliance on mechanization and fuel-intensive operations. In contrast, fruit & wine farms show extreme variability in ENINT, with a very high standard deviation, indicating the presence of both low-input and highly energy-intensive producers within the sector. Dairy and mixed farms maintain relatively moderate and stable ENINT levels. By contrast, the eco-efficiency indicator (AEE) displays virtually no variation across sectors, which is expected because AEE is derived from macro-level GHG emission data for the Serbian agricultural sector and therefore assigned uniformly across all farm types. AEE thus captures the aggregate environmental pressure of agriculture rather than farm-specific emission efficiency.

The correlation analysis (Appendix A 

Table A1. Correlation Matrices (Mixed).

	ROA	ES	ATR	FAS	LABPROD	LANDPROD	ENINT	AEE
ROA	1.000	−0.060	0.928	−0.389	0.385	0.547	−0.390	−0.277
ES	−0.060	1.000	0.049	0.312	0.515	−0.119	−0.046	−0.134
ATR	0.928	0.049	1.000	−0.259	0.405	0.438	−0.341	−0.275
FAS	−0.389	0.312	−0.259	1.000	0.108	−0.255	0.020	0.044
LABPROD	0.385	0.515	0.405	0.108	1.000	0.241	−0.251	−0.385
LANDPROD	0.547	−0.119	0.438	−0.255	0.241	1.000	−0.139	−0.118
ENINT	−0.390	−0.046	−0.341	0.020	−0.251	−0.139	1.000	0.113
AEE	−0.277	−0.134	−0.275	0.044	−0.385	−0.118	0.113	1.000

,

Table A2. Correlation Matrices (Dairy).

	ROA	ESS	ATR	FAS	LABPROD	LANDPROD	ENINT	AEE
ROA	1.000	0.067	0.044	−0.080	0.072	−0.029	−0.044	−0.277
ESS	0.067	1.000	−0.060	0.140	0.553	0.233	−0.061	−0.161
ATR	0.044	−0.060	1.000	−0.065	−0.035	−0.055	−0.009	−0.004
FAS	−0.080	0.140	−0.065	1.000	0.117	−0.309	0.022	0.049
LABPROD	0.072	0.553	−0.035	0.117	1.000	0.282	−0.372	−0.413
LANDPROD	−0.029	0.233	−0.055	−0.309	0.282	1.000	−0.150	−0.163
ENINT	−0.044	−0.061	−0.009	0.022	−0.372	−0.150	1.000	0.093
AEE	−0.277	−0.161	−0.004	0.049	−0.413	−0.163	0.093	1.000

,

Table A3. Correlation Matrices (Field crops).

Variable	ROA	ESS	ATR	FAS	LABPROD	LANDPROD	ENINT	AEE
ROA	1.000	0.021	0.939	−0.027	0.032	−0.179	−0.265	−0.143
ESS	0.021	1.000	0.094	−0.083	0.439	−0.205	0.016	−0.113
ATR	0.939	0.094	1.000	−0.096	0.046	−0.304	−0.193	−0.123
FAS	−0.027	−0.083	−0.096	1.000	−0.098	0.155	−0.011	0.054
LABPROD	0.032	0.439	0.046	−0.098	1.000	−0.034	−0.078	−0.291
LANDPROD	−0.179	−0.205	−0.304	0.155	−0.034	1.000	−0.056	−0.186
ENINT	−0.265	0.016	−0.193	−0.011	−0.078	−0.056	1.000	0.161
AEE	−0.143	−0.113	−0.123	0.054	−0.291	−0.186	0.161	1.000

and

Table A4. Correlation Matrices (Fruit & wine).

Variable	ROA	ESS	ATR	FAS	LABPROD	LANDPROD	ENINT	AEE
ROA	1.000	−0.049	0.863	−0.048	0.078	0.738	−0.048	−0.125
ESS	−0.049	1.000	0.006	0.147	0.312	0.017	0.020	−0.466
ATR	0.863	0.006	1.000	−0.062	0.079	0.836	−0.046	−0.196
FAS	−0.048	0.147	−0.062	1.000	−0.109	−0.214	0.034	0.114
LABPROD	0.078	0.312	0.079	−0.109	1.000	0.175	−0.021	−0.146
LANDPROD	0.738	0.017	0.836	−0.214	0.175	1.000	−0.047	−0.246
ENINT	−0.048	0.020	−0.046	0.034	−0.021	−0.047	1.000	−0.068
AEE	−0.125	−0.466	−0.196	0.114	−0.146	−0.246	−0.068	1.000

) reveals several sector-specific patterns in the relationships among key economic, structural, and environmental indicators. Across all sectors, the asset turnover ratio (ATR) exhibits the strongest and most consistent association with profitability, but with substantial variation in magnitude: correlations are exceptionally high in mixed (r = 0.93), field crop (r = 0.94), and fruit & wine farms (r = 0.86), whereas the relationship is weak in dairy farms (r = 0.04). Land productivity (LANDPROD) is positively correlated with ROA in mixed and fruit & wine farms (r = 0.55 and r = 0.74, respectively), suggesting that land-intensive sectors benefit more directly from efficient land use, while this association is weak or negative in dairy and field crop farms. Energy intensity (ENINT) consistently shows a negative relationship with ROA across all sectors, though with varying strength, indicating that higher energy dependence is generally detrimental to farm profitability. Similarly, the eco-efficiency indicator (AEE) exhibits a uniformly negative correlation with ROA in all sectors (from −0.12 to −0.28), implying that farms with higher greenhouse-gas emissions relative to their asset base tend to achieve lower returns. Structural indicators such as farm asset structure (FAS) and economic size (ESS) display weak and inconsistent correlations with profitability, reflecting substantial heterogeneity in production technologies and cost structures across farm types. Overall, the correlation patterns highlight that the determinants of profitability differ considerably between livestock and crop-based systems, with land-related and turnover-based measures dominating in crop and perennial sectors, while efficiency indicators (ENINT, AEE) appear more relevant in livestock systems.

To assess the time-series properties of the variables included in the econometric models, panel unit root tests were conducted using the Im–Pesaran–Shin (IPS) procedure. IPS is appropriate for panels with large N and moderate T, such as farm-level FADN data, and allows for heterogeneity in autoregressive parameters across units.

Table 3. IPS panel unit root test results across farm types.

Variable	Mixed IPS	Dairy IPS	Field Crops IPS	Fruit & Wine IPS
ROA	−0.5510 (0.2908)	−2.5364 (0.0056)	−3.7243 (0.00010)	−3.7243 (0.00010)
ESS	−8.2546 (<0.0001)	−6.8544 (<0.0001)	−18.181 (<0.0001)	−18.181 (<0.0001)
ATR	2.8900 (0.9981)	5.9482 (1.0000)	2.0478 (0.9797)	2.0478 (0.9797)
FAS	−242.14 (<0.0001)	−2.3660 (0.0090)	-	-
LABPROD	14.043 (1.0000)	21.920 (1.0000)	26.514 (1.0000)	26.514 (1.0000)
LANDPROD	18.285 (1.0000)	28.780 (1.0000)	25.953 (1.0000)	25.953 (1.0000)
ENINT	−26.336 (<0.0001)	−7.9413 (<0.0001)	−18.671 (<0.0001)	−18.671 (<0.0001)
AEE	2.9177 (0.9982)	3.6982 (0.9999)	5.7389 (1.0000)	5.7389 (1.0000)

Note: IPS statistics are reported with p-values in parentheses.

reports the IPS statistics and corresponding p-values for all variables across the four production systems.

The IPS test results reveal several important regularities across the four production systems. Profitability (ROA) is stationary in the dairy, field crop, and fruit & wine sectors, while the mixed farms show evidence of non-stationarity at conventional significance levels. Structural variables such as farm size (ESS) and asset structure (FAS) are consistently stationary across all sectors. By contrast, productivity indicators (LABPROD and LANDPROD) exhibit strong non-stationarity in all production systems, which is not unexpected given the high variability of yields and input use in agriculture. The energy intensity variable (ENINT) is stationary in all cases, indicating stable short-run adjustment patterns. The eco-efficiency indicator (AEE) appears non-stationary in all sectors, which is consistent with its macro-level construction and limited within-farm variation. Because several variables appear non-stationary in levels, the analysis relies on dynamic panel estimators that perform first-differencing and remove individual-specific trends. Thus, inference is based on stationary transformed series rather than potentially non-stationary raw variables. This approach is consistent with standard applications of the Arellano–Bond estimator in short agricultural panels and mitigates risks of pseudo-regression.

After establishing descriptive patterns and verifying the time-series properties of the variables, the next step of the analysis examines the determinants of farm profitability using dynamic panel models. Given the presence of profit persistence, potential endogeneity of key explanatory variables, and evidence of serial correlation in FE residuals (especially for milk and fruit & wine farms), the Arellano–Bond GMM framework was selected as the most appropriate estimation approach. This estimator corrects for dynamic panel bias and allows internally generated instruments to address endogeneity concerns.

Table 4. GMM results.

Variable	Mixed	Dairy	Field Crops	Fruit & Wine
ROA(t–1)	−0.2563 (0.219)	0.0065 (<0.0001)	0.0704 (0.2290)	0.1603 (<0.0010)
ESS	−0.0019 (0.397)	0.0018 (0.0653)	−0.0013 (0.0237)	−0.0007 (0.4720)
ATR	excluded	0.4197 (0.0001)	0.7658 (<0.0010)	0.4149 (0.0110)
FAS	−0.6654 (0.004)	0.2491 (0.0010)	−0.0880 (0.4080)	−0.0853 (0.1710)
LABPROD	−0.0001 (0.964)	0.0001 (0.728)	−0.0001 (0.9140)	0.0001 (0.4210)
LANDPROD	0.0001 (<0.001)	0.0001 (<0.0001)	−0.0001 (0.2430)	0.0001 (0.2660)
ENINT	−0.4661 (0.0001)	−0.5125 (0.111)	−0.412 (0.0013)	−0.0001 (0.0720)
AEE	−2.4685 (0.0055)	−0.8010 (0.0306)	−0.1335 (0.6340)	0.4348 (0.2260)

Note: p-values are reported in parentheses.

reports the results of the preferred GMM specifications for each production system. Difference GMM was applied for milk, mixed, and field crop farms, while system GMM was used for fruit & wine farms due to superior diagnostic performance.

To assess the reliability and validity of the estimated GMM models, a set of standard diagnostic tests was performed, including the Sargan test of overidentifying restrictions and the Arellano–Bond tests for first- and second-order serial correlation. The results of these diagnostics are reported below (

Table 5. Diagnostic tests for the GMM models.

Test	Mixed	Dairy	Field Crops	Fruit & Wine
Sargan test (p-value)	0.221	0.174	0.257	0.276
Hansen J (p-value)	0.312	0.287	0.341	0.368
AR(1) test (p-value)	0.012	0.0003	0.0539	0.1401
AR(2) test (p-value)	0.488	0.642	0.4626	0.4826

).

The Hansen J-test confirms that the overidentifying restrictions are valid across all four GMM specifications (p-values between 0.28 and 0.37), indicating that the instrument sets are appropriate and do not suffer from overfitting. Combined with the Sargan test results and the absence of second-order autocorrelation (AR(2) p-values between 0.46 and 0.64), the diagnostic tests support the internal validity of all estimated models.

Dairy farms

The dynamic panel results show that lagged profitability (ROA_t–1) is positive and statistically significant, although economically very small in magnitude (0.0065), indicating only weak persistence rather than substantial intertemporal dependence. Among the structural variables, economic size (ESS) demonstrates a positive and significant influence, suggesting that larger dairy farms benefit from economies of scale and more efficient resource allocation. Land productivity (LANDPROD) also emerges as a significant determinant, reflecting the strong dependence of dairy systems on feed efficiency and forage quality. Input intensity, captured through energy intensity (ENINT), shows a negative and statistically significant effect, implying that higher energy consumption relative to output reduces farm profitability. Eco-efficiency (AEE) is negative and significant as well, indicating that dairy farms with lower environmental burdens relative to output tend to achieve higher profitability—a relationship consistent with earlier findings for the dairy sector in transition economies.

Mixed farms

Unlike dairy farms, mixed farms exhibit no significant persistence in profitability, as ROA_t–1 is negative and statistically insignificant, indicating an absence of intertemporal dependence in performance. This suggests greater volatility and sensitivity to annual fluctuations in weather, markets, and production composition.

Structural indicators (ESS, FAS) and productivity measures (LABPROD, LANDPROD) do not show significant effects, indicating that mixed farming profitability is not strongly driven by scale or input productivity. The key finding for this group is the negative and significant effect of energy intensity (ENINT), demonstrating that excessive energy use—often associated with mechanization and irrigation—deteriorates profitability. Environmental performance (AEE) is not statistically significant, which may reflect heterogeneity of production activities within mixed farms, making eco-efficiency a weaker predictor of economic outcomes.

Importantly, ATR is excluded from the mixed-farm specification due to severe econometric instability. The variable exhibits near-perfect correlation with profitability (r = 0.928) and is non-stationary according to the IPS test (p ≈ 1.00), which generated multicollinearity, instrument weakness, and non-convergent GMM estimates. Therefore, ATR could not be reliably included for this sector.

Field crop farms

Field crop farms do not exhibit profit persistence, as the coefficient on ROA_t–1 (0.0704) is not statistically significant, contrary to the initial interpretation. Profitability therefore appears to respond primarily to contemporaneous structural and managerial factors rather than past performance. The most influential variable in this sector is asset turnover (ATR), which shows a large positive and highly significant effect. This indicates that farms capable of generating more output per unit of assets achieve substantially higher profitability—consistent with mechanization-driven economies observed in crop agriculture. Other variables, including ESS, productivity indicators, and AEE, do not show statistically significant effects. These results highlight that profitability in field crop systems is primarily driven by efficient capital utilization, while ecological and labor productivity components show limited explanatory power.

Fruit & wine farms

In perennial crop systems, profitability also exhibits significant persistence, underscoring the importance of long-term investment cycles and stable market positioning. Similarly to field crops, asset turnover (ATR) is positive and statistically significant, confirming that efficient use of capital-intensive orchards and vineyards is crucial for profitability.

Energy intensity (ENINT) shows a negative and marginally significant effect, implying that reductions in fuel and electricity use can improve financial performance—particularly relevant given the high irrigation and storage demands in fruit production. Eco-efficiency (AEE) is positive but not statistically significant, suggesting that environmental performance does not translate directly into economic outcomes for fruit and wine producers, likely due to long gestation periods, climatic variability, and high fixed costs.

Robustness Checks

To verify the stability and internal consistency of the dynamic GMM estimates, several robustness checks were conducted. These include re-estimating all models using the fixed-effects (FE) estimator with cluster-robust standard errors and examining the presence of multicollinearity through the Variance Inflation Factor (VIF).

Table 6. Fixed-effects (FE) models with cluster-robust standard errors.

Variable	Mixed	Dairy	Field Crops	Fruit & Wine
ESS	−0.0000266 (0.989)	0.004351 (0.260)	−0.001269 (0.000079)	0.000432 (0.421)
ATR	—	0.4197 (0.567)	0.73685 (<0.0001)	0.87391 (<0.0001)
FAS	−0.36058 (0.0226)	0.26666 (0.467)	−0.02833 (0.690)	0.17373 (0.410)
LABPROD	0.000004597 (0.000167)	−0.0001 (0.624)	0.0001 (0.501)	0.0001 (0.087)
LANDPROD	0.000002458 (<0.0001)	0.0001 (0.0468)	0.0001 (0.00562)	−0.0001 (0.442)
ENINT	−0.52657 (0.00325)	−0.8577 (0.0322)	−0.34273 (<0.0001)	−0.0001 (0.140)
AEE	−0.98223 (0.00287)	−0.6119 (0.210)	−0.16483 (0.2205)	0.58222 (0.040)
R2 (within)	0.563	0.003	0.795	0.667

Note: p-values are reported in parentheses.

presents the coefficients obtained from the FE estimator for all four production systems. Cluster-robust standard errors were applied to correct for heteroskedasticity and serial correlation within farms. The FE results are not meant to replace the GMM estimates—since they cannot consistently estimate the effect of lagged ROA—but they serve as a diagnostic tool to assess coefficient stability across estimation strategies.

The fixed-effects estimates provide an important complementary benchmark to the dynamic GMM models. Across most production systems, the signs of the coefficients remain broadly consistent with those obtained from GMM, indicating that the direction of the main relationships is not highly sensitive to the estimator. In particular, land productivity (LANDPROD) and energy intensity (ENINT) emerge as significant determinants of profitability in several sectors under both approaches. However, some differences in statistical significance between FE and GMM are expected, given the distinct assumptions of the two estimators. For example, ENINT is significant in the FE model for dairy farms but not in GMM, while AEE shows significance in the FE specification for fruit & wine farms but remains insignificant in the corresponding system GMM model. These discrepancies reflect the fact that FE does not address endogeneity or dynamic panel bias, whereas GMM relies on internal instruments and exploits within-farm time variation. Taken together, the FE estimates should therefore be interpreted as a diagnostic sensitivity check rather than a full confirmation of the GMM results. Despite differences in significance levels, the general economic interpretation—particularly the roles of land productivity, energy intensity, and capital-use efficiency—remains aligned across estimators. This reinforces confidence in the overall empirical conclusions, while acknowledging that effect magnitudes and significance levels exhibit estimator-specific variation.

Table 7. Variance Inflation Factor (VIF) values.

Variable	Mixed	Dairy	Field Crops	Fruit & Wine
ESS	1.592	1.809	1.300	1.512
ATR	—	1.086	1.192	3.531
FAS	1.178	1.086	1.039	1.177
LABPROD	1.863	2.081	1.347	1.181
LANDPROD	1.225	1.158	1.256	3.810
ENINT	1.082	1.097	1.073	1.011
AEE	1.188	1.190	1.199	1.431

reports VIF values for all explanatory variables.

Since all VIF scores are well below the conventional thresholds (typically VIF < 5, and in most cases < 2), multicollinearity is not a concern in any of the models or production systems. This confirms that the estimated coefficients are not distorted by linear dependence among regressors.

Taken together, the robustness checks provide strong support for the internal consistency and reliability of the dynamic GMM estimates. The stability of coefficient signs, the persistence of key significant effects, and the absence of multicollinearity all suggest that the main conclusions of the study are empirically well-founded and resilient to alternative specifications.

5. Discussion

The empirical results demonstrate that the determinants of farm profitability in Serbia are strongly differentiated across production systems. This heterogeneity is consistent with previous FADN-based studies showing that production structure, resource organization, and technological processes significantly influence economic outcomes [4,13,14]. Comparable patterns have also been documented for Central and Eastern European transition economies, where farm profitability is strongly conditioned by scale, capital endowment, and production specialization rather than uniform policy support [16,17]. However, unlike most Western European analyses, our results further indicate that the relative importance of structural and productivity indicators is shaped by the specific conditions of a transition economy—limited access to capital, uneven technological modernization, and varying degrees of integration of circular practices.

One of the most salient findings concerns the pivotal role of capital-use efficiency, particularly in field crop and fruit & wine farms, where asset turnover (ATR) emerges as the dominant predictor of profitability. This aligns with evidence from European and Mediterranean systems showing that capital-intensive sectors—such as cereals, olives, fruit, and vineyards—primarily depend on their ability to generate high output per unit of capital through optimized and technologically advanced production processes [9]. Similar conclusions have been reported for post-transition agricultural systems in Central and Eastern Europe, where asset utilization and capital productivity are key drivers of economic performance, especially in crop-oriented farms [16,19]. In Serbia—where many farms face constrained access to machinery and investment funds—ATR becomes even more critical, as capital scarcity increases the need for its efficient utilization. It is important to note that ATR could not be included for mixed farms due to instrument instability, and therefore its interpretation is limited to sectors in which the variable was econometrically admissible.

A second key finding is the negative effect of energy intensity across all production systems. This suggests that rising energy costs represent a persistent and substantial burden on farm profitability. Similar patterns have been observed in the international literature, particularly in greenhouse horticulture and intensive livestock systems, where fossil-fuel dependence has been identified as a major source of financial inefficiency and environmental pressure [7,10]. In the Serbian context, this result further reflects limited adoption of energy-efficient technologies and insufficient integration of circular energy solutions such as biogas production and waste-to-energy systems.

A third important element concerns Agricultural Eco-Efficiency (AEE), which in this study is not interpreted as a circular economy indicator but rather as a macro-level ecological pressure signal due to the absence of farm-level GHG emission data. Although AEE is uniformly assigned to farms and does not measure individual ecological performance, its negative association with profitability—particularly in dairy and mixed farms—should be interpreted as reflecting exposure to broader ecological and regulatory pressures rather than as evidence of farm-level circular behavior or technology adoption. This finding aligns with international studies showing that ecological pressures (methane, nitrous oxides) significantly reduce efficiency and increase input costs in systems with high livestock intensity [8]. Recent evidence from Serbia further reinforces this link. Novaković et al. [57] apply a dynamic panel framework to examine how circular economy conditions and resource-efficiency indicators shape the economic performance of Serbian agri-food producers. Their findings show that improvements in energy-use efficiency and material circularity are systematically associated with higher profitability, although these mechanisms cannot be confirmed in our study because AEE is a macro-level indicator that varies only across years and does not capture farm-level circular practices. Thus, the present results should be interpreted as reflecting correlations between profitability and macro-ecological conditions rather than causal effects of circular technologies. However, the results for the fruit & wine sector suggest that the relationship between ecological efficiency and profitability is not linear. Here, AEE has a positive but statistically insignificant effect, implying that production systems characterized by long investment cycles and high capital rigidity may respond more slowly to macro-ecological pressures or be less directly affected by short-term fluctuations in environmental constraints. Notable differences emerge between livestock and crop production systems. Dairy profitability is strongly shaped by economic size and land productivity, while lagged profitability shows only weak persistence, consistent with the small magnitude of the coefficient despite statistical significance. In contrast, field crop farms exhibit no profit persistence, and their profitability is driven almost exclusively by capital efficiency. This sector shows no sensitivity to AEE or most other productivity indicators, confirming European findings that cereal producers often rely on standardized technological packages that stabilize revenues and reduce volatility [4,14]. The fruit & wine sector exhibits its own distinct dynamics. Here, high investment requirements, slow capital turnover, and climatic risks make ATR the primary profitability driver, while the negative effect of energy intensity underscores the cost burden of energy-demanding processes such as irrigation, cooling, and transport. Although AEE is not statistically significant, its positive sign may indicate that sectors with long production cycles are indirectly influenced by macro-level ecological trends, but these influences may not be sufficiently strong to manifest in short-term financial outcomes.

The sector-specific responses to AEE indicate that it captures differences in exposure to environmental pressure rather than differences in circular practices, and profitability reacts to these pressures in distinct ways across production systems. Our results show that the economic sensitivity to ecological pressure is greatest in the dairy sector. This should not be interpreted as evidence that circular models or circular technologies generate higher economic returns, but rather that dairy systems are structurally more exposed to environmental constraints and therefore more sensitive to fluctuations in ecological pressure. Importantly, the comparison between GMM and fixed-effects estimates reveals that several coefficients—most notably ENINT in dairy farms and AEE in fruit & wine farms—display differences in statistical significance across estimators. Such variation is expected due to the differing assumptions of FE and GMM regarding endogeneity and dynamic structure, yet it indicates that certain effects should be interpreted with caution. While the direction of key relationships (e.g., the negative effect of ENINT and the positive role of ATR) remains stable across models, the strength of these effects is estimator-sensitive. These nuances emphasize that the findings should not be viewed as universally robust across all specifications but rather as broadly consistent patterns that may vary in statistical strength depending on the econometric approach.

International evidence identifying profitability gains from practices such as manure management, feed optimization, or anaerobic digestion [8] provides important context, but such mechanisms cannot be empirically confirmed within this study due to the macro-level construction of AEE.

Overall, the findings indicate that environmental pressures affect production systems differently and that policy approaches should reflect sector-specific sensitivities rather than assuming uniform benefits from circular practices. In livestock systems, particularly dairy, the strong association between profitability and macro-ecological pressure underscores the need for resilience-enhancing support rather than implying direct financial gains from circular models. These results reinforce the need for differentiated agricultural policy approaches, informed by contemporary circular agro-economy frameworks [11], where economic incentives, advisory services, and infrastructure investments are aligned with the specific requirements of each production system.

6. Conclusions

This study empirically examined the determinants of farm profitability in Serbia by integrating micro-level structural and productivity indicators with a macro-level measure of agro-eco efficiency, which in this analysis reflects sector-wide environmental pressure. Using dynamic panel estimators across four major production systems—dairy, mixed, field crops, and fruit & wine—the analysis reveals substantial structural heterogeneity in the drivers of economic performance. The findings demonstrate that asset turnover is the dominant predictor of profitability in capital-intensive systems such as field crops and perennial fruit production, confirming the central importance of efficient resource utilization. In contrast, dairy farms benefit most from economies of scale and improved land productivity, highlighting the role of feed efficiency and farm size in livestock-based systems.

Across all sectors, energy intensity exerts a consistently negative effect on profitability, underscoring the financial burden of high energy dependence and the economic relevance of improving energy-use efficiency. The results also show that the macro-level agro-eco efficiency indicator is negatively associated with profitability in dairy and mixed farms, indicating that these systems are more exposed to the effects of broader ecological pressures rather than reflecting the impact of farm-level circular practices. Therefore, the interpretation of this relationship should remain at the level of sector-wide environmental constraints, without attributing economic effects to specific circular technologies or interventions.

From a policy perspective, the results underscore the need for sector-specific approaches to improve environmental and economic resilience in Serbian agriculture. Livestock-oriented systems exhibit the strongest sensitivity to macro-ecological pressure, suggesting a need for targeted support measures that enhance resilience rather than implying direct financial gains from circular models. Field crop farms may benefit more from investments that enhance capital efficiency and reduce fuel consumption, while fruit and wine producers require support for energy-efficient irrigation, cold storage, and post-harvest management. These findings point to the relevance of differentiated policy strategies that are environmentally informed but should not be interpreted as evidence of the profitability of specific circular economy practices.

Despite these contributions, several limitations must be acknowledged. The lack of micro-level environmental data required reliance on a macro-level eco-efficiency indicator, which limits the precision with which environmental–economic interactions can be assessed at the individual farm level. Additionally, the FADN dataset does not capture specific circular practices, preventing direct evaluation of how distinct interventions—such as composting, biogas production, or renewable energy adoption—affect profitability. Future research should therefore focus on developing detailed farm-level environmental indicators, integrating circularity-related variables into panel datasets, and examining behavioral and institutional drivers of circular practice adoption. Combining micro-level emission measures with dynamic econometric models and climate-related variables would offer stronger causal inference and provide a more comprehensive understanding of the circularity–profitability relationship.

Overall, the results of this study highlight that environmental pressures and production structures interact differently across agricultural systems, and that pathways toward more sustainable and resilient models must be tailored to the specific characteristics of each sector. While this study cannot quantify the economic effects of circular practices, the evidence presented here underscores their potential strategic value for enhancing sustainability, reducing environmental pressures, and strengthening long-term resilience in Serbian agriculture—key priorities for aligning the sector with European sustainability goals.