Agrifood Efficiency: DEA Evidence for Rural Competitiveness in Bulgaria

Abstract

This study evaluates the productive efficiency in the agrifood sector of 21 rural Bulgarian districts as a proxy for territorial competitiveness. Output-oriented Data Envelopment Analysis (DEA) was performed using district-level data from 2022 to 2024. The analysis incorporates five inputs related to labor, land, and capital and three economic outputs from agriculture and food processing. Results indicate substantial variation in efficiency among rural districts. Twelve districts form the efficiency frontier, with effective resource use and diverse structures; nine are inefficient due to scale or organizational/technological constraints. Bootstrap bias correction revealed standard DEA underestimates efficiency gaps. Frontier districts include large plains, mountainous regions and smaller, specialized systems, indicating diverse paths to competitiveness. A composite Territorial Competitiveness Index (TCI) showed frontier status does not guarantee efficiency, often due to underused manufacturing capital. Cluster analysis identified four performance groups needing different policy support, ranging from near-frontier territories that need knowledge transfer to deeply underperforming districts that require restructuring. No geographic clustering of efficiency was found, pointing to structural and institutional, rather than geographic, drivers. These results highlight the need for territorially tailored rural policies within the Common Agricultural Policy (CAP) and offer an empirical basis for diagnosing regional agrifood efficiency gaps.

1. Introduction

The rural economy and agrifood systems are undergoing major changes due to global market integration, technological progress, climate pressure, and shifting consumer demand for sustainable, high-value foods. Agricultural production is increasingly embedded in global value chains, while the EU’s Common Agricultural Policy (CAP) is reshaping economic and environmental conditions in rural regions. External shocks, such as geopolitical tensions, supply chain disruptions, energy price volatility, and climate variability, underscore the need for resilient, competitive agrifood systems [1,2]. Sustainability transitions, climate adaptation, and the use of environmental indicators in policy are also shaping the competitiveness of agrifood systems in Europe [3,4]. Thus, improving efficiency and competitiveness in agrifood and rural economies is now a key policy objective at all levels.

In the EU, the agrifood sector is crucial for regional development, job creation, and food security. However, significant disparities persist across countries and regions due to differences in resources, technology adoption, institutions, and market integration [5,6]. Rural areas face additional challenges, such as low population density, aging populations, limited infrastructure, and reduced investment access [7,8]. These differences highlight the need to evaluate how efficiently rural economies use resources and generate value. Regional competitiveness depends on the spatial organization of economic activity and the effective use of resources, shaped by local endowments, innovation, and production system efficiency [9,10]. Analyzing how regions convert resources into economic outputs provides insight into their competitiveness. Empirical studies confirm persistent regional differences in efficiency and productivity across the EU, despite some convergence in certain indicators [11].

Bulgaria is a pertinent case for such analysis. Despite some modernization in its agrifood sector, rural regions in Bulgaria remain highly heterogeneous in structure, productivity, and development potential [12,13,14]. Much of the country is classified as predominantly rural, and agrifood production is a key source of employment and income. Bulgarian agriculture faces the competitive pressures of the EU single market and must adapt to global trends in sustainability, digitalization, and consumption. These factors raise questions about the efficiency and competitiveness of Bulgarian regional agrifood systems in both European and global contexts.

Competitiveness is a concept that is often operationalized through efficiency indicators analyzed at national, regional, sectoral, and firm levels [15]. In agrifood economics, it is linked to the ability to achieve sustainable income and maintain market positions under competition. In rural areas, competitiveness also involves the capacity to attract residents, workers, and visitors, supporting long-term socio-economic vitality. The concept is complex, reflecting the interplay of economic, natural, technological, and institutional factors. There is no single accepted indicator; studies often use productivity, efficiency, trade, and structural measures [16]. Productivity and efficiency are core, as they show how resources are transformed into outputs. Efficiency-based approaches are valuable for assessing regional agricultural competitiveness, allowing for the simultaneous consideration of inputs, outputs, and environmental constraints [17]. Competitiveness is also measured using indicators like productivity, export performance, or costs. However, these may not capture the multidimensional nature of rural agrifood systems, which involve multiple inputs and outputs and are influenced by environmental, social and policy factors. Frontier efficiency techniques, increasingly used in rural economics, evaluate performance in such complex systems [18]. Recently, these methods have also been applied to assess eco-efficiency and sustainability in European agriculture [3,4,19].

Data Envelopment Analysis (DEA) is a widely used non-parametric method for measuring productive efficiency. DEA constructs an efficiency frontier from observed input–output combinations, assessing the relative performance of decision-making units (DMUs) against best-practice benchmarks [20]. Unlike parametric methods, DEA does not assume a specific production function and can handle multiple inputs and outputs, making it well-suited to the diverse agrifood sector. DEA is extensively used in research on farm efficiency, regional agricultural performance, and competitiveness [16,18].

Despite extensive research on agricultural productivity and efficiency in Europe, little attention has been paid to regional disparities within member states, particularly those dominated by rural areas. Most studies focus on farm-level or national efficiency, leaving a gap in regional analysis. Therefore, this study aims to evaluate the productive efficiency of Bulgaria’s rural agrifood sector using a DEA-based framework. By applying DEA to Bulgarian NUTS-3 regions, this study provides a comparative assessment of regional agrifood efficiency and identifies best-practice areas for informing rural development policy and competitiveness strategies. This analysis enhances our understanding of rural competitiveness and informs rural development policy and planning in Bulgaria.

The paper is structured as follows: Section 2 presents the conceptual framework and literature review; Section 3 details the methodology; Section 4 reports and discusses empirical results; and Section 5 and Section 6 conclude with policy implications and future research directions.

2. Literature Review

2.1. Agrifood Sector Efficiency as a Driver of Rural Competitiveness

Rural competitiveness is a region’s ability to achieve sustainable economic growth and wellbeing while attracting businesses and investment [9]. This concept is grounded in regional development theory, emphasizing productivity, innovation, and efficient resource use as foundations for long-term success [10,21,22]. In rural areas, where the agrifood sector drives economic activity, employment, and land use, productive efficiency is central to competitiveness. Rural regions with inefficient agrifood enterprises tend to lag behind in income, employment, investment, and demographic retention [22,23]. This study views agrifood productive efficiency as necessary but not sufficient for rural territorial competitiveness. It measures a region’s ability to convert resources into output, following Porter [9] and Camagni [21], but does not include factors like market access, innovation, or institutions, which are not available in district-level data. The DEA-based framework thus focuses on resource transformation, acknowledging that a fuller competitiveness assessment would require broader indicators.

Traditional competitiveness indicators often fail to capture the structural performance of agricultural systems, as they focus on trade outcomes or partial productivity rather than resource-use efficiency [16,24,25,26]. Standard approaches, such as total factor productivity, domestic resource cost ratios, and financial indicators, typically operate at the firm or national level and do not fully reflect territorial competitive performance [15,16]. Partial productivity measures ignore the multi-input, multi-output nature of agrifood systems, while profitability indicators are sensitive to price and subsidy shifts, making them unreliable proxies for territorial competitiveness [16,23].

Competitiveness is multidimensional and cannot be captured by a single variable [27]. In regional development, it is increasingly seen as a systemic property resulting from interactions among firms, institutions, infrastructure, and governance [21,28]. At the regional level, competitiveness depends on firm efficiency, institutional capacity, infrastructure, and value chains. Even efficient agrifood enterprises may underperform if not integrated into supply chains or lacking processing and marketing infrastructure. Agrifood value chains help rural regions to capture economic value, especially when local processing, logistics, and cooperatives allow participation in higher value-added segments [29,30]. Thus, rural competitive advantage relies on both individual efficiency and the sector’s ability to create value from regional resources.

Multidimensional frameworks for agribusiness competitiveness, such as price and cost competitiveness, product quality, delivery reliability, product innovation, and market responsiveness, capture this broader view [29,31]. At the regional level, these dimensions suggest examination of whether rural regions can maintain cost-competitive production, diversify into higher-value products, ensure reliable supply, and respond to evolving demands. Empirical evidence from EU regional analysis demonstrates that the answers to these questions depend significantly on the production structure of the regional agrifood system. Regions specializing in crop production, livestock, or mixed systems display different efficiency profiles, even when resources are controlled. This finding indicates that the composition of production types may constitute a structural determinant of regional competitiveness [32]. At the territorial level, these dimensions raise questions about whether rural regions can sustain cost-competitive production, diversify into higher-value products, ensure reliable supply, and adapt to changing demands. In Central and Eastern Europe (CEE), rural areas where small producers lack scale, organization, or market access remain disadvantaged unless supported by cooperatives, value chain integration, and targeted public investment [23,33]. Studies reveal persistent disparities in farm productivity, market integration, and value chain participation despite some convergence in macroeconomic indicators [23,24,34,35]. For Bulgaria, slow labor productivity convergence and a sharp decline in agricultural labor highlight the need for district-level efficiency analysis to identify sources of underperformance [14,36]. This aligns with the sustainability dimension of regional competitiveness, which emphasizes that competitive rural regions must generate economic value efficiently over the long run [9,21].

Research by Kimhi et al. [37] confirms persistent structural disparities in CEE agriculture, with ongoing productivity differences across EU states, including Bulgaria [38,39,40,41]. Kryszak et al. [17] identifies institutional quality and farm structure as key drivers of these gaps. At the EU level, Nowak et al. [42] document substantial diversity in agricultural technical efficiency across all 27 member states using DEA. The difference between the highest and lowest performers reaches 40%. They identify land quality, farm manager age, and investment subsidies as key determinants. Regarding the CEE context, Bojnec et al. [43] applied DEA to ten new EU member states from 2001 to 2006. They found all studied countries operated below the efficiency frontier. Bulgaria and Slovakia achieved the highest scores among the group. Institutional reforms and farm structure were identified as the main drivers of variation in efficiency. At the sub-national level, Lazíková et al. [44] provide a methodological precedent relevant to our study. They applied DEA at both the NUTS-3 region and LAU 1 district level for Slovak agricultural holdings, finding significant regional disparities in efficiency that aggregated national indicators could not reveal. For Bulgaria, Galluzzo [45] performed a DEA-based efficiency study, assessing Bulgarian farm performance using FADN data from 2007 to 2015. The main finding is that specialized farms outperformed mixed farms. CAP subsidies positively influenced both efficiency and rural socio-economic conditions. However, that study operates at the farm level and covers only primary agriculture, leaving unassessed the territorial dimension of the full agrifood system, including agricultural production and food processing. Koteva et al. [12] and Bachev et al. [39] document structural heterogeneity and sustainability gaps across Bulgarian regions using partial productivity and structural indicators. Koteva et al. [12] provide a comprehensive assessment of farm competitiveness across Bulgarian regions, identifying persistent disparities in productivity, investment capacity, and market integration. Bachev et al. [39] examine the economic, social, and ecological dimensions of agrarian sustainability across Bulgarian districts, confirming the multidimensional character of rural performance gaps. These studies establish the empirical context for the present analysis but rely on structural and partial productivity indicators rather than frontier efficiency methods. Therefore, by applying DEA to all 21 predominantly rural NUTS-3 districts simultaneously, the present study is the first that provides a spatially comprehensive efficiency benchmark for Bulgaria’s rural agrifood sector, an analytical dimension absent from the existing Bulgarian regional literature.

The present study focuses specifically on the productive efficiency dimension of competitiveness. Efficiency reflects the capacity of regional agrifood systems to transform available resources into economic outputs and therefore represents a foundational component of economic performance. By analyzing how districts convert labor, land, and capital into agricultural and food-industry output, the DEA framework captures the core production efficiency that underpins broader competitiveness dynamics. Other dimensions of regional competitiveness are acknowledged but cannot be incorporated directly in the model due to the absence of comparable district-level indicators. Trade-based metrics such as Revealed Comparative Advantage and domestic resource cost ratios are unsuitable at the district level due to a lack of disaggregated trade data. While total factor productivity indices are useful, they require long-time-series data not available in the NSI database. Partial productivity measures, like gross value added per hectare or per employee, reflect only single-input, single-output relationships and not the complex structure of regional agrifood systems. DEA addresses these issues: it works with cross-sectional data, accommodates multiple inputs and outputs, and produces spatially comparable efficiency scores as competitive benchmarks [18,20]. Thus, frontier efficiency analysis is an appropriate tool for assessing the agrifood competitiveness of Bulgarian rural districts.

2.2. Data Envelopment Analysis as a Tool for Rural Competitiveness Assessment

Given the limitations of single-indicator approaches and the inherently multidimensional nature of territorial agrifood competitiveness, a methodology is needed that can simultaneously handle multiple inputs and outputs, operate without restrictive assumptions about the functional form of production, and generate spatially comparable results suitable for policy interpretation. DEA, first introduced by Charnes et al. [20], meets all three requirements. It has become one of the most widely used efficiency analysis techniques in agricultural economics because of its ability to handle multiple inputs and outputs without assuming specific production functions [18,41]. As a non-parametric linear programming method, DEA constructs an empirical production frontier based on observed input–output data and measures each DMU’s efficiency by its distance from this frontier. Weights are determined endogenously to provide each DMU with its most favorable evaluation, ensuring that no unit is penalized for a particular input–output combination that may reflect legitimate structural conditions.

In the context of territorial agrifood efficiency, the DMUs are spatial units, districts, regions, or states, and the efficiency score assigned to each reflects how effectively that territory converts its agrifood sector inputs (land, labor, capital) into valued outputs (gross agricultural product, food-processing value added, export revenues). This approach makes DEA an analytical link between firm-level efficiency and territorial competitiveness: it combines the performance of enterprises within a territory into a single score that is directly comparable across territories and easily interpretable in terms of competitive positioning. Regional DEA studies have been used to assess agricultural productivity across countries, regions, and farm systems, offering comparative efficiency benchmarks for policy analysis [33,40,46,47].

Two standard model specifications form the basis for most DEA territorial applications. The CCR model [20] assumes constant returns to scale (CRS) and provides an overall technical efficiency score, while the BCC model [48] relaxes this assumption to accommodate variable returns to scale (VRS), distinguishing pure technical efficiency from scale efficiency. The difference between constant and variable returns to scale is especially significant in agricultural contexts, where farm size, regional production structure, and technological adoption can notably impact scale efficiency [18,46]. This decomposition is particularly valuable for rural territorial analysis, where differences in district size, population density, and the scale of the agrifood sector may systematically influence whether observed inefficiency is due to managerial and organizational factors or merely a suboptimal scale of operation relative to the efficient frontier.

The application of DEA to assess agrifood competitiveness at territorial and sector levels is well established in the European context [19,49,50,51,52]. Mavrommati et al. [49] developed a two-stage composite DEA model to evaluate operational and investment efficiency across Greek dairy enterprises, finding significant performance heterogeneity and demonstrating that slack analysis can identify precise, quantified targets for input reduction and output improvement. The study’s use of a composite DEA framework, integrating both operational and investment dimensions, illustrates how the method can capture the multi-faceted nature of agrifood sector performance within a single analytical structure, rather than relying on partial indicators. This is precisely the property required when agrifood efficiency is interpreted as a driver of territorial competitive advantage rather than as a stand-alone firm-level metric. At the territorial level, Marjanović et al. [50] use DEA to rank the circular economy efficiency of all 27 EU member states based on Eurostat-derived composite indicators. They demonstrate that the method can handle diverse spatial units and generate policy-relevant efficiency gradients across an entire institutional geography. Their discovery of significant cross-territorial variation in efficiency emphasizes both the diagnostic value of DEA for regional benchmarking and the importance of identifying structural factors that explain why some territories perform better than others on the efficiency frontier.

Altogether, these contributions establish DEA as a methodologically appropriate and empirically validated tool for assessing efficiency in the agrifood sector’s territorial dimension [53,54]. When applied to the Bulgarian context, the method allows for a systematic comparison of how effectively agrifood sectors in predominantly rural districts convert their resource endowments into economic output and how variations in that efficiency influence territorial competitive positioning.

In contrast to the studies discussed in the literature review above, the novelty of the present study is threefold. First, it presents the first DEA-based frontier efficiency assessment covering all predominantly rural NUTS-3 districts in Bulgaria, advancing beyond the farm-level focus of Galluzzo [45] and the structural indicators used by Koteva et al. [12] and Bachev et al. [39] to provide spatially comparable territorial efficiency scores. Second, it combines five analytical components (CRS/VRS efficiency decomposition, slack analysis, super-efficiency scoring, bootstrap bias correction, and a composite TCI) within a single diagnostic framework. This approach distinguishes between scale-driven and technically driven underperformance at the district level, a level of analysis not previously achieved in CEE regional studies [32,43]. Third, by including the food-processing and manufacturing sector alongside primary agriculture, the study captures the agrifood value chain, which has not been addressed in earlier Bulgarian efficiency research. It also examines whether efficiency variation aligns with geographic patterns using spatial autocorrelation analysis, providing direct insights for the spatial design of CAP interventions.

3. Materials and Methods

3.1. Study Design and Territorial Scope

This study employs a quantitative, non-parametric frontier approach to assess the competitiveness of rural agrifood systems in Bulgaria. The methodological framework is informed by the concept of territorial competitiveness developed in Section 2. The analysis is conducted at the NUTS-3 territorial level, treating each Bulgarian district as a DMU that transforms a set of regional agrifood inputs into multiple economic, social, and environmental outputs.

The territorial classification underpinning the study follows the OECD two-step typology [55], chosen for its methodological transparency, compatibility with Eurostat spatial frameworks, and established use in comparative rural research across EU member states. In the first step, municipalities (LAU 2 units) with a population density below 150 inhabitants per square kilometer are classified as rural. In the second step, NUTS-3 districts are assigned to one of three categories, predominantly rural, intermediate, or predominantly urban, based on the proportion of the district’s population residing in rural municipalities. This procedure results in the classification shown in

Table 1. Classification of Bulgarian districts by OECD territorial typology.

Typology Class	Districts	Count (N = 28)
Predominantly Rural	Vidin, Vratsa, Lovech, Montana, Pleven, Veliko Tarnovo, Gabrovo, Razgrad, Silistra, Dobrich, Targovishte, Shumen, Sliven, Stara Zagora, Blagoevgrad, Kyustendil, Sofia, Kardzhali, Pazardzhik, Smolyan, Haskovo	21 (51% of the national population)
Intermediate	Ruse, Varna, Burgas, Yambol, Pernik, Plovdiv	6 (29% of the national population)
Predominantly Urban	Sofia-Capital	1 (20% of the national population)

Source: own calculations based on data from National Statistical Institute (NSI). Note. Classification based on OECD [50] two-step territorial typology applied to Bulgarian LAU 2 (municipality) and NUTS-3 (district) units.

.

The analysis is restricted only to these 21 predominantly rural districts, ensuring analytical consistency. Intermediate districts (Ruse, Varna, Burgas, Yambol, Pernik, and Plovdiv) and the predominantly urban Sofia-Capital district are also excluded from the study as a methodological requirement: DEA benchmarks each unit against a common frontier, a process that is valid only when DMUs are homogeneous. Including territories with fundamentally different economic profiles, such as intermediate and urban, would result in a frontier shaped by urban agrifood systems, making it inappropriate for benchmarking rural districts [56,57]. Highly urbanized districts possess economic structures dominated by services and non-agricultural sectors, which would distort comparisons of agrifood resource-use efficiency across territorially comparable rural regions. Excluding intermediate and predominantly urban districts is therefore necessary to maintain analytical consistency with the study’s focus on rural agrifood competitiveness.

The OECD typology is preferred over other frameworks, such as the Eurostat degree of urbanization typology or Bulgarian national classifications, for three main reasons. First, it operates at the municipal level, the most detailed spatial scale where population density data are reliably available in Bulgaria, thus avoiding misleading conclusions that can result from district-level average densities. Second, it maintains methodological consistency with a broad range of comparative EU rural research, allowing Bulgarian findings to be contextualized within wider European patterns. Third, its criteria are transparent, publicly documented, and reproducible, based on publicly available data and supporting future longitudinal and cross-country analyses.

3.2. Data Sources and Variable Selection

Data was sourced from the National Statistical Institute (NSI) regional databases for district-level information, available for free, covering the period 2022–2024, for all 21 predominantly rural districts. The arithmetic means of the three years are used as the analytical values to reduce the impact of single-year distortions. It is particularly important for the period of 2022–2024, which encompasses post-pandemic economic recovery, energy price volatility, and supply chain disruptions arising from the conflict in Ukraine, all of which affected agrifood input costs and output values unevenly across years. The three-year mean therefore captures structural efficiency relationships rather than shock-driven fluctuations.

The eight variables within the DEA framework are chosen to represent the resource-transformation chain of the agrifood system and collectively serve as an operationally justified proxy for territorial competitiveness at the NUTS-3 level. On the input side, agricultural and food-industry employment represent the human capital and labor-market dimensions of competitiveness. Utilized agricultural area serves as a proxy for natural resource endowment, while capital investment in both the primary and processing sub-sectors reflects technological modernization and investment attractiveness. These dimensions are central to Porter’s [9] factor conditions and Camagni’s [21] territorial asset framework. On the output side, agricultural gross value added reflects primary sector productivity. Enterprise net sales reflect market integration and the commercial performance of the district’s non-financial enterprise base, and food-industry production value measures value chain depth and agro-industrial development. Collectively, these three outputs address the economic performance and value-creation dimensions identified in multidimensional competitiveness frameworks [15,16,29]. The DEA framework is thus positioned as a proxy for resource transformation in territorial competitiveness, measuring the effectiveness with which districts convert agrifood inputs into valued economic outputs, rather than serving as a comprehensive competitiveness scorecard. Dimensions such as institutional capacity, innovation systems, and environmental governance, identified as competitiveness drivers by Porter [9], Camagni [21], and Depperu and Cerrato [27], are acknowledged as being beyond the scope of available NUTS-3 data and are therefore not included in the present framework. The final variable framework, comprising five inputs and three outputs, is presented in

Table 2. DEA variable framework: inputs, outputs, and dimensions.

Role	Dimension	Indicator	Unit
INPUT	Labor in Agriculture	Employees in agriculture, forestry and fishing	Persons (annual)
	Labor in Food Industry	Employees in food and beverage manufacturing	Persons (annual)
	Land	Total utilized agricultural area	Hectares
	Capital in Agriculture	Expenditure on tangible fixed assets, agriculture	Thousand BGN
	Capital in Food Industry	Expenditure on tangible fixed assets, food processing	Thousand BGN
OUTPUT	Agricultural Value Added	Gross value added, agricultural sector	Million BGN
	Enterprise Performance	Net sales income, non-financial enterprises	Thousand BGN
	Food-Industry Output	Production value of food and beverage industry	Thousand BGN

Source: own elaboration.

.

The analysis does not fully satisfy the parsimony guideline N ≥ 3 × (m + s) = 24. Nevertheless, the analysis is conducted with explicit recognition of the potential reduction in the model’s discriminatory power. This limitation arises because the study examines the entire population of predominantly rural districts in Bulgaria rather than a sample. As a result, the relatively small number of DMUs may reduce the model’s discriminatory power and increase the likelihood of identifying multiple frontier units. To address this issue, a reduced-specification robustness model is estimated in Section 4, and the resulting rankings are compared with those of the full model. Reducing the variable set to improve the N/(m + s) ratio was considered but rejected, because omitting food-industry employment and capital would exclude core inputs for the food-processing sub-sector, central to the territorial competitiveness framework. Similarly, reducing the output set would require dropping either gross value added, a welfare-relevant output for agricultural policy, or food-industry production value, the main output of the processing sub-sector. Such reductions would reduce the intended multi-sector territorial assessment to a single-sector model, thus misrepresenting the agrifood system as conceptualized in Section 2.

Three additional considerations underpin the retention of the full variable set. First, each variable is theoretically justified within the competitiveness framework, and omitting any would compromise essential conceptual dimensions. Second, all input–output pairs satisfy positive isotonicity [56,57], meeting DEA’s monotonicity requirement. Third, regional DEA studies with similar or lower N/(m + s) ratios are well documented: Toma et al. [53] applied a 5-input, 2-output CRS DEA to 42 Romanian rural counties, Marjanović et al. [50] used a composite DEA for 27 EU member states, and Liu et al. [54] show that theoretically grounded selection offsets modest sample sizes when the full DMU population is included, as in this study of all 21 rural districts.

Nevertheless, the discrimination limitation of the VRS model is acknowledged; the high proportion of VRS-efficient units is due to the modest sample size relative to the number of variables. Therefore, VRS results are employed only for diagnostic decomposition rather than primary efficiency ranking. To address finite-sample concerns, a reduced-specification robustness model (CRS DEA with three inputs and two outputs) is estimated in Section 4.4 and its rank-order results compared to those of the full model.

3.3. DEA Model Specification

The DEA model captures the multidimensionality of rural agrifood competitiveness using an output-oriented specification aligned with the study’s territorial competitiveness framework. The analysis estimates the potential increase in economic, social, and environmental outputs that each district could achieve if its agrifood sector operated on the efficient frontier, given existing resources. Variable returns to scale (VRS) are adopted throughout, following the BCC model [48], to address structural heterogeneity across districts. A supplementary constant returns-to-scale (CRS) model [20] is also estimated to decompose technical efficiency into pure technical and scale efficiency, distinguishing managerial and organizational inefficiency from suboptimal scale.

The BCC output-oriented DEA model for each DMU is formulated as follows:

max θ subject to: Yλ ≥ θy₀, Xλ ≤ x₀, e^Tλ = 1, λ ≥ 0

(1)

where Y and X are the output and input matrices across DMUs; y₀ and x₀ are observed vectors for each DMU; λ is the intensity vector; θ is the proportional output expansion factor (efficiency score, θ = 1 indicates the frontier); and the constraint e^Tλ = 1 imposes VRS. Efficiency scores start at 1 (efficient), with higher values denoting greater inefficiency.

After computing efficiency scores, slack-based analysis identifies input surpluses and output shortfalls for each inefficient district, generating benchmarking targets [58]. Super-efficiency scores are calculated using the Andersen–Petersen model [59], excluding each DMU from its own reference set to differentiate among efficient units. Bootstrap DEA, following Simar and Wilson [60], yields bias-corrected efficiency estimates and 95% confidence intervals (B = 2000 replications). This corrects finite-sample bias, as demonstrated by Kryszak [35] for EU NUTS-2 regions. Beyond frontier estimation, three interpretive extensions support policy-relevant synthesis of DEA outputs.

First, a composite Territorial Competitiveness Index (TCI) aggregates three components, each normalized: CRS efficiency (w₁ = 0.50), scale efficiency (w₂ = 0.30), and slack-adjusted resource utilization (w₃ = 0.20). This hierarchy reflects the territorial competitiveness framework (Section 2), with productive efficiency as the core measure [18,58], scale optimality as a secondary factor [18], and resource utilization as a finer adjustment. The TCI synthesizes DEA outputs into a single, policy-relevant ranking reflecting frontier distance, scale, and input waste, aligning with multidimensional frameworks [15,16,21,31]. Sensitivity analysis with alternative weights shows ranking robustness (Spearman ρ > 0.94 in all scenarios). However, the TCI does not capture market access, product quality, or institutional governance, as identified by Porter [9], Depperu and Cerrato [27], and Camagni [21], due to NUTS-3 data constraints. Thus, it should be regarded as a resource-transformation index rather than a full competitiveness scorecard.

Second, an efficiency–structure typology cross-classifies districts by CRS frontier status and returns to scale, yielding four groups with distinct policy implications: frontier consolidators (CRS = 1), scale-up candidates (IRS), overextended producers (DRS), and technically constrained underperformers.

Third, k-means cluster analysis (k = 4, based on silhouette coefficient optimization) is applied to CRS and scale efficiency, producing data-driven groupings. The optimal four-cluster solution has a silhouette coefficient of 0.822, indicating strong cohesion and separation.

All analyses were performed in R version 4.4.2 (R Foundation for Statistical Computing, Vienna, Austria): CRS/VRS estimation, slack analysis, super-efficiency scoring, bootstrap bias correction, TCI construction, k-means clustering, and spatial autocorrelation, following best practices [50,51].

4. Results and Discussion

4.1. DEA Efficiency Results: CRS and VRS Models

Table 3. DEA Efficiency Results for 21 Rural Districts of Bulgaria (2022–2024 Average).

District	CRS Rank	CRS Efficiency	VRS Efficiency	Scale Efficiency	RTS	Quartile (CRS)
Vidin	1	1.0000	1.0000	1.0000	CRS	Q1—High
Vratsa	2	1.0000	1.0000	1.0000	CRS	Q1—High
Montana	3	1.0000	1.0000	1.0000	CRS	Q1—High
Gabrovo	4	1.0000	1.0000	1.0000	CRS	Q1—High
Razgrad	5	1.0000	1.0000	1.0000	CRS	Q1—High
Silistra	6	1.0000	1.0000	1.0000	CRS	Q2—Upper-Middle
Dobrich	7	1.0000	1.0000	1.0000	CRS	Q2—Upper-Middle
Blagoevgrad	8	1.0000	1.0000	1.0000	CRS	Q2—Upper-Middle
Sofia	9	1.0000	1.0000	1.0000	CRS	Q2—Upper-Middle
Kardzhali	10	1.0000	1.0000	1.0000	CRS	Q2—Upper-Middle
Pazardzhik	11	1.0000	1.0000	1.0000	CRS	Q3—Lower-Middle
Smolyan	12	1.0000	1.0000	1.0000	CRS	Q3—Lower-Middle
Targovishte	13	0.9797	0.9798	0.9999	IRS	Q3—Lower-Middle
Lovech	14	0.8842	0.9155	0.9658	IRS	Q3—Lower-Middle
Shumen	15	0.8811	0.9984	0.8825	DRS	Q3—Lower-Middle
Pleven	16	0.8622	1.0000	0.8622	DRS	Q3—Lower-Middle
Haskovo	17	0.8404	0.8523	0.9860	DRS	Q4—Low
Veliko Tarnovo	18	0.7880	1.0000	0.7880	DRS	Q4—Low
Stara Zagora	19	0.7710	1.0000	0.7710	DRS	Q4—Low
Sliven	20	0.7435	0.7460	0.9966	DRS	Q4—Low
Kyustendil	21	0.7368	1.0000	0.7368	IRS	Q4—Low
Mean/Count	—	0.9279	0.9758	0.9518	CRS:12 DRS:6 IRS:3	N = 21

Source: own calculation. Note: scale efficiency = CRS efficiency/VRS efficiency; RTS = returns-to-scale classification (CRS = optimal scale; DRS = decreasing returns; IRS = increasing returns). Mean values are arithmetic means across all 21 districts.

presents the full results of the output-oriented DEA under both the CRS [20] and VRS [48] specifications, together with scale efficiency scores, returns-to-scale classifications, and competitive performance quartiles based on CRS ranking.

The mean CRS efficiency across all districts is 0.9279. This result suggests that, on average, districts could increase agrifood output by approximately 7.8% without additional inputs if they operated at frontier efficiency. Under the CRS specification, 12 of the 21 districts (57.1%) achieve full efficiency (score = 1.000), thereby forming the efficient frontier against which all other districts are benchmarked. These districts include Vidin, Vratsa, Montana, Gabrovo, Razgrad, Silistra, Dobrich, Blagoevgrad, Sofia, Kardzhali, Pazardzhik, and Smolyan. Nine districts record a CRS efficiency below a score of 1.000, with six scoring below 0.90: Kyustendil, Sliven, Stara Zagora, Veliko Tarnovo, Haskovo, and Pleven. The results indicate the most pronounced gaps between actual and potential performance. The six districts identified have the most significant potential for efficiency improvements and merit further examination of the structural, resource, or managerial factors limiting their agrifood productivity.

Under the VRS specification, which accounts for structural size differences, 16 out of 21 districts achieve full efficiency. The higher proportion of VRS-efficient units (76.2%) reflects the established tendency of VRS models to identify more efficient DMUs when the sample size is modest relative to the number of variables, as is the case in this analysis (N = 21, m + s = 8) [51]. Consequently, the CRS model serves as the primary basis for competitive ranking and quartile classification, while the VRS model and the resulting scale efficiency decomposition are used for diagnostic purposes. The mean VRS efficiency is 0.9758, and the mean scale efficiency is 0.9518.

4.2. Scale Efficiency and Returns to Scale

Decomposing overall technical efficiency into pure technical efficiency (VRS) and scale efficiency (SE = CRS/VRS) demonstrates that scale-related factors significantly contribute to underperformance in several districts. The average SE across all 21 districts is 0.9518. Within the Q4 districts, SE is especially evident: Kyustendil (SE = 0.7368, IRS) indicates that a substantial portion of its CRS inefficiency results from operating below optimal scale, rather than from managerial or organizational shortcomings. In contrast, Sliven (SE = 0.9966) demonstrates that its CRS inefficiency is almost entirely attributable to pure technical inefficiency, rather than scale inefficiency.

The returns-to-scale classification identifies three distinct groups. Twelve districts operate at CRS, indicating that they have achieved optimal production scale within the agrifood sector; any proportional increase in inputs would result in a proportional increase in output. Six districts demonstrate DRS: Stara Zagora, Pleven, Shumen, Sliven, Haskovo, and Veliko Tarnovo. DRS indicates that these districts have surpassed their optimal agrifood scale, so further investment in inputs yields diminishing returns. For these territories, policy should prioritize productivity improvement and value chain intensification rather than capacity expansion [18,33]. Three districts, Lovech, Targovishte, and Kyustendil, exhibit IRS, suggesting that they operate below optimal scale and would benefit from agrifood sector consolidation and investment to achieve a more productive configuration.

4.3. Slack Analysis and Input Surplus Diagnostics

Analysis of the 2022–2024 average data indicates that the VRS-inefficient districts are Lovech, Targovishte, Shumen, Sliven, and Haskovo. Subsequent slack analysis for them identifies specific areas of resource surplus or output shortfall that extend beyond the proportional efficiency gap indicated by theta score. Collectively, the findings offer a more detailed assessment of inefficiency to support the development of targeted recommendations instead of a standardized approach. The results are presented in

Table 4. Slack analysis for VRS-inefficient districts (2022–2024 average, original units).

District	VRS Eff.	Input Slack: Employed A (Persons)	Input Slack: Employed C (Persons)	Input Slack: Agric. Area (ha)	Input Slack: Invest. Agric. (th. BGN)	Input Slack: Invest. Manuf. (th. BGN)	Output Slack: GVA Agri. (m. BGN)	Output Slack: Food Output (th. BGN)
Targovishte	0.9798	84	34	0	0	0	0.00	437,130
Haskovo	0.8523	103	0	47,884	0	1820	0.00	0
Lovech	0.9155	0	0	43,619	1403	67,347	0.00	0
Shumen	0.9984	405	0	0	23,469	91,439	0.00	0
Sliven	0.7460	0	0	0	10,260	42,267	0.00	0

Source: own calculation. Note: slack values represent surpluses (inputs) or shortfalls (outputs) remaining after proportional efficiency adjustment.

.

The slack analysis identifies diverse sources of inefficiency among the five underperforming districts. Targovishte demonstrates the least inefficiency, with a VRS efficiency score of 0.9798 and a minor surplus of employed persons in agriculture. This indicates that inefficiency in Targovishte is marginal and primarily due to a small labor surplus in the primary sector. The district’s CRS score of 0.9797 and IRS classification further suggest operation slightly below the optimal production scale. In contrast, Haskovo exhibits more pronounced structural inefficiency. In addition to the proportional output expansion indicated by its theta score, Haskovo has surpluses of employed persons in agriculture and of agricultural land. This reflects extensive low-productivity land use that does not yield proportional agrifood output [23].

Lovech displays a unique efficiency profile, characterized by IRS classification and a VRS score of 0.9155. The district’s input surpluses are concentrated in agricultural land and manufacturing investment, indicating that capital allocated to food processing has not produced output increases comparable to those achieved by frontier benchmarks. Under CRS, Lovech’s benchmark peers are Razgrad, Gabrovo, and Kardzhali, each of which achieves a similar or higher food-processing output with fewer resources. Full peer intensity weights (λ) for all inefficient districts are reported in Appendix A,

Table A1. VRS peer intensity weights (λ) for CRS-inefficient districts (2022–2024 average). Only peers with λ > 0.01 are reported.

District	Primary Peer (λ)	Peer 2 (λ)	Peer 3 (λ)	Peer 4 (λ)	Peer 5 (λ)
Lovech	Razgrad (0.430)	Gabrovo (0.394)	Kyustendil (0.143)	Kardzhali (0.033)	—
Targovishte	Razgrad (0.512)	Vidin (0.206)	Montana (0.112)	Kardzhali (0.092)	Blagoevgrad (0.078)
Shumen	Blagoevgrad (0.307)	Kardzhali (0.264)	Razgrad (0.256)	Dobrich (0.139)	Pazardzhik (0.033)
Pleven	Razgrad (1.138)	Blagoevgrad (0.205)	Sofia (0.098)	Gabrovo (0.023)	—
Haskovo	Razgrad (0.438)	Vidin (0.414)	Blagoevgrad (0.323)	Sofia (0.050)	Dobrich (0.021)
Stara Zagora	Sofia (0.758)	Blagoevgrad (0.381)	Razgrad (0.258)	Gabrovo (0.213)	—
Veliko Tarnovo	Razgrad (0.886)	Blagoevgrad (0.304)	Sofia (0.169)	Gabrovo (0.021)	—
Kyustendil	Kardzhali (0.356)	Blagoevgrad (0.121)	Sofia (0.081)	Vidin (0.030)	—
Sliven	Razgrad (0.435)	Sofia (0.191)	Kardzhali (0.146)	Silistra (0.143)	Vidin (0.120)

Source: own calculation. Note: λ weights from output-oriented CRS DEA (CCR model [20]). Weights sum to approximately 1.0 under the VRS convexity constraint. VRS peers include all VRS-efficient districts (λ > 0.01 under the BCC model). Kyustendil (CRS = 0.737, VRS = 1.000) is VRS-efficient and therefore constitutes a valid peer reference unit, despite being CRS-inefficient due to scale suboptimality.

. Shumen demonstrates a different pattern. Despite achieving near-full VRS efficiency (0.9984), Shumen has surpluses of employed persons in agriculture, in agricultural capital, and in manufacturing capital. This suggests systemic over-capitalization relative to frontier benchmarks, consistent with its DRS classification.

Sliven exhibits the second-lowest CRS efficiency score in the sample (0.7435, rank 20), with a nearly identical VRS score of 0.7460. The lack of labor and land slack, alongside significant investment surpluses in agricultural capital and manufacturing capital, indicates that capital over-deployment, rather than workforce or land misallocation, is the primary source of inefficiency. The scale efficiency of 0.9966 is close to unity, and the minimal gap between CRS and VRS scores (0.0025) confirms that Sliven’s competitive underperformance is attributable almost entirely to pure technical inefficiency. The district does not achieve frontier-level output from its current resource endowment, regardless of scale. This finding contrasts sharply with Kyustendil (rank 21, SE = 0.7368, IRS), where scale suboptimality explains most of the CRS gap. Therefore, Sliven requires enterprise-level reorganization, technology adoption, and value chain integration, rather than structural rescaling. Its primary CRS benchmark peers are Razgrad, Sofia, Kardzhali, Silistra, and Vidin, all of which achieve substantially higher agrifood output from comparable or leaner resource configurations.

4.4. Robustness Check: Reduced Specification of DEA

To assess whether the efficiency rankings from the full model (five inputs, three outputs) were sensitive to variable specification, a reduced-specification CRS DEA model was estimated. This approach addresses concerns related to the modest N/(m + s) ratio. The reduced model omits capital investment in food-processing manufacturing, retaining four inputs (agricultural employment, food-industry employment, utilized agricultural area, and agricultural capital investment) and all three outputs (agricultural GVA, net sales income, and food-industry production value). Exclusion of this variable is justified for two reasons: it exhibits the highest coefficient of variation among the 21 districts (CV = 0.876), making it the main source of frontier sensitivity in the full model, and it is the only input not directly linked to the primary agricultural production stage of the agrifood chain. The reduced model meets the parsimony guideline N ≥ 3 × (m + s), with N = 3 × (4 + 3) = 21, precisely satisfying this requirement.

The reduced model is estimated using the same output-oriented CRS specification as the full model (CCR model [20]). Rank-order agreement between the two specifications is evaluated using Spearman’s rank correlation coefficient (ρ). A threshold of ρ ≥ 0.90 is set to determine whether the full-model rankings are robust to variable specification, aligning with the sensitivity analysis standard applied to the TCI weighting scenarios.

The results demonstrate strong rank-order consistency (ρ = 0.935, p < 0.001). All nine CRS-inefficient districts identified in the full model remain inefficient under the reduced specification, and seven frontier districts (Kardzhali, Razgrad, Sofia, Blagoevgrad, Gabrovo, Smolyan, Silistra) maintain CRS = 1.000 in both models. The most notable change within the efficient set pertains to Montana, which achieves CRS = 1.000 in the full model but decreases to CRS = 0.870 (rank 15) in the reduced model. This outcome indicates that, in the full model, Montana’s frontier status is partly dependent on its food-processing capital configuration. When this input is excluded, Montana’s agrifood output mix is no longer supported at the frontier level by the remaining resource endowment. This observation aligns with Montana’s DRS classification in the main model and supports the interpretation that its competitive position relies on capital intensity in the processing sub-sector rather than scale expansion. Full efficiency scores and rank comparisons for both specifications are provided in Appendix A,

Table A2. Robustness check: CRS efficiency score comparison between full model (5 inputs, 3 outputs) and reduced model (4 inputs, 3 outputs), 21 predominantly rural districts in Bulgaria (2022–2024 average).

District	Full Model		Reduced Model		Δ Rank	Both Efficient?	Rank
	CRS Score	Rank	CRS Score	Rank
Gabrovo	1.0000	1	1.0000	1	0	√	2
Razgrad	1.0000	1	1.0000	1	0	√	3
Silistra	1.0000	1	1.0000	1	0	√	4
Blagoevgrad	1.0000	1	1.0000	1	0	√	5
Sofia	1.0000	1	1.0000	1	0	√	6
Kardzhali	1.0000	1	1.0000	1	0	√	7
Smolyan	1.0000	1	1.0000	1	0	√	8
Vidin	1.0000	8	1.0000	10	+2	√	9
Dobrich	1.0000	8	1.0000	1	−7	√	10
Vratsa	1.0000	10	1.0000	10	0	√	11
Montana	1.0000	10	0.8704	15	+5	—	12
Pazardzhik	1.0000	10	1.0000	1	−9	√
Targovishte	0.9797	13	0.8756	14	+1	—
Lovech	0.8844	14	0.8844	12	−2	—
Shumen	0.8811	15	0.8811	13	−2	—
Pleven	0.8623	16	0.8108	16	0	—
Haskovo	0.8404	17	0.7386	19	+2	—
Veliko Tarnovo	0.7880	18	0.7584	17	−1	—
Stara Zagora	0.7710	19	0.7208	20	+1	—
Sliven	0.7435	20	0.7435	18	−2	—
Kyustendil	0.7368	21	0.6757	21	0	—

Source: own calculation. Note: output-oriented CRS DEA (CCR model); both models estimated on 2022–2024 three-year arithmetic means for 21 predominantly rural NUTS-3 districts. The reduced model excludes capital investment in manufacturing relative to the full model, satisfying the N ≥ 3 × (m + s) parsimony guideline (N = 21, 3 × (4 + 3) = 21). Δ Rank = reduced model rank minus full model rank (positive = downward shift; negative = upward shift). The ‘√’ symbol indicates districts that achieve CRS = 1.0000 in both model specifications. Rank correlation: Spearman ρ = 0.935 (p < 0.001), confirming strong rank-order consistency between the two model specifications.

.

The robustness check confirms that the qualitative findings of the study, including the identification of frontier districts, the four-cluster typology, the policy groups, and the spatial autocorrelation result, are not artifacts of the full-model variable specification. The reduced model is employed solely for this robustness assessment; all policy interpretations are based on the full five-input, three-output model.

4.5. Super-Efficiency, Composite TCI, Cluster Typology, and Spatial Analysis

Super-efficiency scores, calculated by excluding each evaluated district from its own reference set, enable more precise ranking among the 12 CRS-efficient frontier districts and highlight significant differences in the robustness of their frontier positions (Appendix A,

Table A3. Super-efficiency scores (Andersen–Petersen model) for the 12 CRS-efficient frontier districts (2022–2024 average).

Rank	District	Super-Efficiency Score	Frontier Tier
1	Kardzhali	2.9131	Deep frontier
2	Razgrad	2.3472	Deep frontier
3	Sofia	1.8806	Deep frontier
4	Blagoevgrad	1.8522	Deep frontier
5	Vidin	1.3951	Strong frontier
6	Dobrich	1.3121	Strong frontier
7	Gabrovo	1.2893	Strong frontier
8	Silistra	1.1270	Moderate frontier
9	Smolyan	1.0819	Moderate frontier
10	Pazardzhik	1.0723	Moderate frontier
11	Montana	1.0604	Moderate frontier
12	Vratsa	1.0456	Marginal frontier

Source: own calculation. Note: Andersen–Petersen [54] output-oriented super-efficiency model: each frontier DMU is excluded from its own reference set. Tier classification: deep frontier, ≥1.500; strong frontier, 1.200–1.499; moderate frontier, 1.050–1.199; marginal frontier, <1.050.

). Kardzhali has the highest super-efficiency score, followed by Razgrad, Sofia, and Blagoevgrad. These four comprise the “deep frontier”, with input–output configurations so efficient that their removal would substantially raise efficiency targets for neighboring inefficient districts. In contrast, Vratsa, Montana, Pazardzhik, and Smolyan have super-efficiency scores only slightly above unity, indicating robust but less dominant frontier positions; these districts are efficiently organized but can be replaced as reference peers. Gabrovo, Vidin, Dobrich, and Silistra are intermediate. Thus, super-efficiency ranking creates a more nuanced competitive hierarchy than standard DEA scores and directly identifies the districts best suited to be models for rural development policy.

The bootstrap bias correction [60] confirms both the direction and magnitude of efficiency scores but also exposes a systematic downward bias in the standard DEA estimator for inefficient districts. The mean bias-corrected CRS score for all 21 districts is 0.892, versus the original 0.928, showing that the true average efficiency is about 3.6 percentage points lower than the standard DEA estimate. Bias is most severe in the lowest-performing districts: Kyustendil (0.602), Sliven (0.612), and Stara Zagora (0.650). Efficient frontier districts show negligible bias (zero for Vidin, Razgrad, Blagoevgrad, Sofia, Kardzhali), in line with asymptotic DEA properties at the boundary. The reliability of bootstrap inference depends on sample size: with N = 21, the kernel density estimation in our bootstrap procedure (following Simar and Wilson [60]) may yield overly broad bandwidths, and all upper confidence bounds for inefficient districts reach 1.000, limiting statistical discrimination. Thus, bootstrap results are best viewed as bias-corrected point estimates rather than precise intervals. This limitation underscores the need for a panel extension to increase sample size. Full bias-corrected scores and 95% confidence intervals are given in Appendix A,

Table A4. Bootstrap bias-corrected CRS efficiency scores and 95% confidence intervals, all 21 predominantly rural districts (2022–2024 average).

District	CRS (Original)	CRS (Bias-Corrected)	Bias	95% CI Lower	95% CI Upper
Vidin	1.0000	1.0000	0.000	1.0000	1.0000
Vratsa	1.0000	1.0000	0.000	0.8033	1.0000
Montana	1.0000	1.0000	0.000	0.8080	1.0000
Gabrovo	1.0000	1.0000	0.000	0.9808	1.0000
Razgrad	1.0000	1.0000	0.000	1.0000	1.0000
Silistra	1.0000	1.0000	0.000	0.8674	1.0000
Dobrich	1.0000	1.0000	0.000	0.9974	1.0000
Blagoevgrad	1.0000	1.0000	0.000	1.0000	1.0000
Sofia	1.0000	1.0000	0.000	1.0000	1.0000
Kardzhali	1.0000	1.0000	0.000	1.0000	1.0000
Pazardzhik	1.0000	1.0000	0.000	0.8111	1.0000
Smolyan	1.0000	1.0000	0.000	0.8199	1.0000
Targovishte	0.9797	1.0000	0.020	0.7545	1.0000
Lovech	0.8842	0.8305	−0.054	0.7640	1.0000
Shumen	0.8811	0.8243	−0.057	0.7623	1.0000
Pleven	0.8622	0.7901	−0.072	0.7611	1.0000
Haskovo	0.8404	0.7537	−0.087	0.7682	1.0000
Stara Zagora	0.7710	0.6498	−0.121	0.7569	1.0000
Veliko Tarnovo	0.7880	0.6747	−0.113	0.7547	1.0000
Sliven	0.7435	0.6115	−0.132	0.7635	1.0000
Kyustendil	0.7368	0.6016	−0.135	0.7702	1.0000
Mean	0.9279	0.8922	−0.036	—	—

Source: own calculation. Note: Bootstrap procedure following Simar and Wilson [60]; reflected kernel density estimator; bandwidth h determined by Silverman’s rule on log-transformed efficiency scores (h = 0.089); B = 2000 replications. Confidence intervals (CIs) are percentile-based. Districts ordered by CRS score, descending.

.

The composite TCI, developed as a policy-oriented synthesis of the DEA components rather than an independent performance measure, yields a policy-relevant ranking that integrates all three dimensions of productive performance. The top five TCI positions are jointly occupied by Vidin, Sofia, Kardzhali, Silistra, and Montana, all of which are frontier districts with zero input slacks and optimal scale. A particularly notable finding concerns Pazardzhik: although it achieves CRS = 1.000 and SE = 1.000, it ranks only 17th on the TCI due to its substantial manufacturing capital investment (the highest in the sample), generating disproportionate slack relative to its food-industry output. This diagnostic highlights a structural tension between frontier status and resource utilization efficiency that the standard DEA score does not capture and identifies Pazardzhik as a district where further investment in food-processing capacity should be contingent upon demonstrable output improvements rather than additional capital allocation. Full TCI rankings and component scores for all 21 districts are provided in Appendix A,

Table A5. Composite TCI for full rankings and component scores, 21 predominantly rural districts (2022–2024 average).

TCI Rank	District	CRS Score	Scale Efficiency	Slack Adj. Score	TCI (Normalized)
1	Vidin	1.0000	1.0000	1.0000	1.000
1	Sofia	1.0000	1.0000	1.0000	1.000
1	Kardzhali	1.0000	1.0000	1.0000	1.000
1	Silistra	1.0000	1.0000	1.0000	1.000
1	Montana	1.0000	1.0000	1.0000	1.000
1	Smolyan	1.0000	1.0000	1.0000	1.000
7	Razgrad	1.0000	1.0000	0.979	0.986
8	Blagoevgrad	1.0000	1.0000	0.975	0.984
9	Dobrich	1.0000	1.0000	0.958	0.972
10	Gabrovo	1.0000	1.0000	0.936	0.957
11	Targovishte	0.9797	0.9999	0.874	0.881
12	Vratsa	1.0000	1.0000	0.786	0.857
13	Lovech	0.8842	0.9658	1.000	0.772
14	Haskovo	0.8404	0.9860	1.000	0.719
15	Shumen	0.8811	0.8825	0.998	0.681
16	Pleven	0.8622	0.8622	1.000	0.631
17	Pazardzhik	1.0000	1.0000	0.354	0.567
18	Sliven	0.7435	0.9966	0.989	0.560
19	Veliko Tarnovo	0.7880	0.7880	1.000	0.432
20	Kyustendil	0.7368	0.7368	1.000	0.294
21	Stara Zagora	0.7710	0.7710	0.424	0.000

Source: own calculation. Note: TCI = 0.50 × CRS efficiency + 0.30 × scale efficiency + 0.20 × slack-adjusted score, normalized to [0, 1].

.

K-means cluster analysis (k = 4, silhouette coefficient = 0.822) in the joint CRS–SE space yields four distinct performance groups, closely aligned with but analytically separate from quartiles based on CRS ranking. Cluster I (n = 13, mean CRS = 0.998, mean SE = 1.000) comprises all frontier districts and Targovishte (CRS = 0.980, SE = 1.000); its high internal homogeneity (silhouette 0.91) confirms a coherent, high-performing group despite structural diversity. Cluster II (n = 2: Pleven, Shumen) shows moderate CRS (mean 0.871) and below-median scale efficiency; both are DRS districts with scale-driven gaps. Cluster III (n = 3: Lovech, Sliven, Haskovo) exhibits moderate-to-low CRS but high scale efficiency (mean SE = 0.983), indicating mostly technical inefficiency and necessitating targeted policy. Cluster IV (n = 3: Veliko Tarnovo, Stara Zagora, Kyustendil) is the lowest-performing, with mean CRS and SE both at 0.765, confirming compounded scale and technical deficits. This four-cluster typology provides more precise operational segmentation for rural development policy than the quartile system, distinguishing scale- from technically driven underperformance.

The spatial autocorrelation analysis using Moran’s I yields a statistic of I = 0.016 (p = 0.198), indicating that CRS efficiency scores do not display statistically significant geographic clustering across the 21 predominantly rural districts. It is important to acknowledge that with N = 21 spatial units, the statistical power of Moran’s I test is inherently limited. The probability of detecting moderate spatial dependence (Moran’s I ≈ 0.15–0.20) at conventional significance levels is reduced, so the non-significant result should not be interpreted as definitive evidence of spatial randomness but rather as no strong spatial clustering detectable at this level of analysis. This absence of spatial autocorrelation aligns with the regional convergence literature for CEE: Egri and Lengyel [58] find that rural NUTS-3 regions in Bulgaria converge substantially more slowly than urban and intermediate regions and that the drivers of this divergence are structural and institutional rather than geographic, confirming that proximity to efficient regions does not automatically generate efficiency spillovers [61]. The present local spatial typology is consistent with this interpretation, as most frontier districts act as isolated HL outliers rather than diffusion cores. The Vidin–Montana pair is the only genuine high–high spatial cluster; Kyustendil is an LH outlier that underperforms despite proximity to efficient neighbors. Future research incorporating CAP absorption rates, cooperative density, or governance quality scores would enable more direct testing of these structural mechanisms. Local spatial statistics and spatial lag values are provided in Appendix A,

Table A6. Local spatial statistics for CRS efficiency scores and spatial lag values, 21 predominantly rural districts (2022–2024 average). Global Moran’s I = 0.016, p = 0.198.

District	CRS Score	Spatial Lag (W·z)	Local Type	Interpretation
Vidin	1.0000	1.0000	HH	Efficiency cluster
Montana	1.0000	1.0000	HH	Efficiency cluster
Kyustendil	0.7368	1.0000	LH	Spatial outlier (low, high neighbors)
Gabrovo	1.0000	0.826	HL	Spatial outlier (high, low neighbors)
Razgrad	1.0000	0.912	HL	Spatial outlier (high, low neighbors)
Silistra	1.0000	0.960	HL	Spatial outlier (high, low neighbors)
Dobrich	1.0000	0.940	HL	Spatial outlier (high, low neighbors)
Blagoevgrad	1.0000	0.934	HL	Spatial outlier (high, low neighbors)
Sofia	1.0000	0.937	HL	Spatial outlier (high, low neighbors)
Kardzhali	1.0000	0.947	HL	Spatial outlier (high, low neighbors)
Pazardzhik	1.0000	0.935	HL	Spatial outlier (high, low neighbors)
Vratsa	1.0000	0.949	HL	Spatial outlier (high, low neighbors)
Smolyan	1.0000	0.960	LL	Spatial outlier (high score, low-avg area)
Lovech	0.8842	0.927	LL	Low-efficiency zone
Pleven	0.8622	0.918	LL	Low-efficiency zone
Veliko Tarnovo	0.7880	0.923	LL	Low-efficiency zone
Targovishte	0.9797	0.837	LL	Low-efficiency zone
Shumen	0.8811	0.945	LL	Low-efficiency zone
Sliven	0.7435	0.868	LL	Low-efficiency zone
Stara Zagora	0.7710	0.891	LL	Low-efficiency zone
Haskovo	0.8404	0.903	LL	Low-efficiency zone

Source: own calculation. Note: HH = high CRS, high lag (efficiency cluster); LL = low-CRS or low-lag area; HL = high CRS surrounded by lower-scoring neighbors (isolated efficient district); LH = low CRS surrounded by higher-scoring neighbors (spatial underperformer).

.

Collectively, the five analytical layers, CRS efficiency, scale decomposition, slack diagnostics, super-efficiency scoring, and composite TCI, identify three principal empirical patterns that inform the interpretation in next section. First, frontier status within the Bulgarian rural agrifood sector exhibits structural diversity, arising from qualitatively distinct input–output configurations rather than a single prevailing agrifood organization model. Second, the observed divergence between CRS rank and TCI rank, particularly for Pazardzhik (CRS rank 11, TCI rank 17) and Vratsa (CRS rank 2, TCI rank 12), indicates that resource utilization efficiency and frontier positioning represent empirically distinct dimensions of competitive performance, which have direct implications for investment appraisal. Third, the four-cluster typology shows that the bottom quartile is not a uniform group of underperformers; rather, it comprises districts with distinct efficiency deficits that require differentiated policy interventions.

4.6. DEA Findings and Their Implications for Rural Competitiveness in Bulgaria

The empirical results from the five analytical components presented in Section 4.1, Section 4.2, Section 4.3, Section 4.4 and Section 4.5 support three interconnected interpretive arguments related to the conceptual framework outlined in Section 2. This framework positions agrifood sector efficiency as a foundational determinant of territorial rural competitiveness rather than as an isolated firm-level metric. The frontier efficiency in Bulgarian rural districts exhibits structural diversity, resulting from qualitatively distinct input–output configurations rather than a single prevailing agrifood model. The observed divergence between CRS and TCI ranks in several districts, most notably Pazardzhik, indicates that frontier status and resource utilization efficiency are empirically distinct aspects of competitive performance. The four-cluster typology distinguishes scale-driven from technically driven inefficiency, with direct implications for policy targeting. By identifying these patterns at the NUTS-3 level, this analysis extends existing studies that investigate agricultural efficiency at broader territorial scales. Each of these patterns is examined in relation to the regional competitiveness and DEA literature.

The first argument addresses the geography of efficiency at the frontier. Among the 12 CRS-efficient districts, there is notable geographic and structural diversity. These include both predominantly agricultural territories with extensive arable land (Dobrich, Razgrad, Silistra, Montana) and districts characterized by stronger agro-industrial profiles (Blagoevgrad, Pazardzhik, Sofia district). This heterogeneity is theoretically significant, as it demonstrates that frontier efficiency in the Bulgarian rural agrifood context is not limited to large-scale, grain-dominated plains agriculture. Instead, it can be achieved through structurally distinct input–output configurations, provided that resources are allocated in proportions that align with the district’s output mix. Kardzhali and Smolyan, the two mountainous districts with relatively small agricultural areas in the sample, also achieve CRS efficiency. This confirms that small-scale, high-intensity agrifood systems can be territorially competitive when inputs are appropriately matched to outputs [26,32]. The structural mechanisms underlying this pattern, such as terrain-adapted production specialization, traditional agrifood practices in mountainous areas, and the influence of CAP support for regions with natural constraints, constitute a substantive research question. The current quantitative framework is insufficient to fully address this issue, indicating the need for dedicated qualitative investigation in future studies. The finding aligns with the multidimensional competitive advantage framework: efficiency, as measured by DEA, operates at the intersection of cost-based and quality-oriented performance, and different districts reach the frontier through various combinations of these dimensions [15,31,62]. This structural diversity at the frontier is consistent with the EU-wide evidence presented by Nowak et al. [42], who find that technically efficient agricultural systems across member states are characterized by qualitatively distinct input configurations, rather than a single dominant production model. Similarly, Bojnec et al. [43] demonstrate that in CEE member states, institutional context and farm organization determine which territories reach the efficiency frontier, rather than resource endowments alone.

The second argument examines the Q4 group and distinguishes between scale-driven and technically driven inefficiencies. Sliven provides the most illustrative example. With an SE of 0.9966, which is effectively at constant returns, its CRS inefficiency of 24.9% is almost entirely attributable to pure technical inefficiency. This indicates that the district does not achieve frontier-level output from its current resource endowment, regardless of scale. In contrast, Kyustendil (SE = 0.7368, IRS) exhibits a different pattern, with approximately three quarters of its overall efficiency gap resulting from operating well below optimal scale. These distinctions have direct policy implications. Territories such as Sliven require interventions focused on production organization, technology adoption, and value chain integration, reflecting the supply chain management and product innovation dimensions of competitive advantage identified by Thatte [31]. Conversely, regions similar to Kyustendil would benefit from structural consolidation. Increasing the scale of agrifood operations through cooperative formation, enterprise clustering, or land consolidation schemes would facilitate movement toward the CRS frontier. This finding is consistent with Mavrommati et al. [49] in the Greek dairy sector, where decomposition of scale and pure technical efficiency similarly revealed that different enterprise groups require targeted policy responses. At the sub-national level, Lazíková et al. [44] identify a similar pattern in Slovak agricultural districts: scale-driven and technically driven inefficiencies are concentrated in distinct spatial clusters. This finding confirms that applying uniform policy instruments across all underperforming territories would result in resource misallocation. The conclusion directly supports the differentiated cluster-based approach proposed for Bulgaria.

The third argument addresses the group classified as DRS. Stara Zagora, Pleven, Shumen, Haskovo, and Veliko Tarnovo all operate beyond their optimal agrifood scale. Despite this, most achieve a VRS efficiency of 1.000, indicating that they perform as well as possible given their size but that size itself is a competitive disadvantage under the output-oriented CRS benchmark. This counterintuitive result requires careful interpretation. These districts are among Bulgaria’s most industrialized agricultural regions, with substantial food-processing capacity. Their DRS classification does not reflect mismanagement but rather that the current scale of resource deployment, particularly labor and land in Haskovo, and capital investment in Stara Zagora and Sliven, exceeds what the frontier reference set considers productively optimal for the observed output configuration. The structural characteristics identified by Bojnec et al. [43] for CEE agriculture, small farm scale, limited specialization, and underinvestment relative to EU-15 benchmarks, provide a structural explanation for the IRS classification of Lovech, Targovishte, and Kyustendil in the study. These districts operate below the optimal production scale not because of poor management but because the structural adjustment process in Bulgarian agriculture remains incomplete relative to CAP efficiency benchmarks. In terms of territorial competitiveness, this highlights the quality and value dimension: these districts mobilize large resource bases, but agrifood output, measured by GVA, net sales, and food-processing production, does not increase proportionally [63]. Upgrading to higher-value processed food output, improving capital allocation efficiency in food manufacturing, and strengthening agro-industrial linkages within the district would enable these territories to transition from DRS to CRS performance [29,30], consistent with CAP objectives and the broader sustainability targets for productive and sustainable agrifood systems. Galluzzo [45] further supports this interpretation through a farm-level DEA study of Bulgarian agriculture, which demonstrates that specialization, rather than scale, is the primary determinant of efficiency differences among Bulgarian farms. Specialized holdings consistently outperform mixed-production farms regardless of size. At the territorial level, this suggests that DRS districts would benefit more from intensifying agrifood specialization and reinforcing linkages with processing than from expanding overall production capacity.

A key methodological limitation is the reliance on NACE-classified employment data as proxies for labor inputs in the food-processing sub-sector. The NSI district-level database records food-industry employment under manufacture of food products, beverages, and tobacco products, capturing only formally registered enterprises that meet NSI thresholds. This approach may systematically understate labor inputs in districts dominated by small-scale, artisanal, or family-based food-processing, notably in mountainous regions such as Smolyan, Kardzhali, and the Rhodope area of Blagoevgrad, where informal agro-processing is significant but not reflected in NACE records [12,39]. Conversely, the data may overstate labor input in districts hosting large food-processing enterprises whose operations extend beyond district boundaries, such as Stara Zagora and Pleven, which have major grain and dairy facilities. These biases can marginally overstate CRS efficiency scores in mountain districts (due to understated labor input) and understate them in large-scale processing districts (due to overstated labor input). While these biases do not alter the main conclusion, that mountain districts attain genuine frontier efficiency through focused, high-intensity outputs [30], they indicate that slack estimates for food-industry labor should be seen as indicative rather than precise benchmarks where agro-processing coverage is incomplete. Future research using more precise data would allow a fuller characterization of food-industry labor endowments.

The slack analysis provides actionable specificity to reinforce these arguments. The concentration of investment slacks in the food manufacturing sub-sector of Lovech and Sliven, and the land surplus in Haskovo, indicate two distinct types of structural misalignment. In the first case, capital is allocated to manufacturing without generating proportional output, suggesting absorption capacity constraints or market access barriers for processed food. In the second, agricultural land is underutilized relative to frontier benchmarks, consistent with ongoing low-productivity farming in parts of the district. Both patterns align with structural vulnerabilities widely documented across CEE rural economies [22,33]. The benchmark peer analysis further clarifies the situation: inefficient districts are systematically compared to peers such as Razgrad, Gabrovo, Kardzhali, and Montana, which are smaller-to-medium districts achieving high output-to-input ratios through focused, efficient agrifood systems rather than scale. This indicates that improving territorial competitiveness in Bulgaria’s rural districts depends less on additional resource mobilization and more on reorganizing and intensifying existing inputs in line with best-practice configurations (see Appendix A, Table A1 for full peer intensity weights).

A comparison with Marjanovic et al.’s [50] EU-level territorial DEA, which also identified substantial variation in efficiency across spatial units and established best-practice benchmarks with distinct structural characteristics, confirms that territorial efficiency gradients are a consistent feature of agrifood systems analysis at any spatial scale [52]. The frontier status of Dobrich, Razgrad, and Silistra, all located in the Danube Plain, demonstrates that land consolidation and capital modernization have resulted in measurable efficiency gains at the NUTS-3 level. The current DEA results reveal this pattern for the first time at this spatial scale. The persistence of efficiency heterogeneity among Bulgarian rural districts reflects the uneven structural adjustment observed by Petrick and Kloss [64] in Polish regions, indicating that post-transition agrifood convergence remains incomplete across CEE regardless of EU accession timing. Notably, this heterogeneity is not apparent in aggregate national indicators. The district-level DEA framework fulfills the disaggregation role that Flegl et al. [52] and Ratner et al. [51] identify as the principal policy benefit of frontier methods in sustainability benchmarking, by converting aggregate performance data into spatially actionable diagnostics to inform targeted CAP investment.

The efficiency range identified in this study aligns with patterns observed in comparable regional DEA research across CEE. Toma et al. [53], utilizing a five-input, two-output constant returns-to-scale (CRS) DEA for Romanian rural counties, similarly reported that most units achieved frontier efficiency under the CRS specification, while a minority of districts demonstrated persistent and structurally distinct underperformance, a result also evident in Bulgaria’s four-cluster typology. The comparison with Romania is particularly informative due to the similar institutional context: both countries joined the EU in 2007 and have experienced analogous structural adjustments under the CAP. Nevertheless, both display significant intra-country rural efficiency heterogeneity that is not captured by aggregate national indicators. The decomposition of scale and technical inefficiency in the current analysis also corresponds with findings from Latruffe et al. [33] regarding Polish crop and livestock farms, where institutional and organizational factors, rather than resource endowments, accounted for the majority of efficiency variation across regions. This cross-country consistency supports the conclusion that Bulgaria’s district-level efficiency disparities are not unique but instead reflect broader structural dynamics characteristic of post-transition CEE rural economies.

5. Policy Implications

The empirical findings from Section 4.1, Section 4.2, Section 4.3, Section 4.4, Section 4.5 and Section 4.6 yield four interconnected policy implications for rural development within the CAP framework.

5.1. Spatially Differentiated Investment Based on the Four-Cluster Typology

The cluster analysis reveals that Bulgaria’s 21 predominantly rural districts are not a homogeneous group of underperformers requiring uniform support. Instead, they represent four structurally distinct performance groups, each with a unique efficiency profile and corresponding policy need [32]. This finding has a critical implication: a single national rural development program that applies the same instrument uniformly across all districts will misallocate resources. The undifferentiated policy not only fails to close efficiency gaps but also risks exacerbating them by reinforcing the structural factors that generate inefficiency.

Cluster IV districts (Veliko Tarnovo, Stara Zagora, and Kyustendil) experience deficits, exhibiting both scale suboptimality and technical inefficiency. For these territories, the primary objective should be structural transformation, such as consolidating fragmented landholdings, supporting cooperative formation among small producers, and modernizing technology at the enterprise level. Investment support in these districts should depend on demonstrated output improvement relative to the efficiency benchmarks rather than on the volume of investment. Without this conditionality, public funds sustain existing capital configurations that have already failed to generate proportional agrifood output.

Cluster III districts (Lovech, Sliven, and Haskovo) are scale-efficient but face technical constraints. Additional capital investment is unlikely to enhance their performance, as their inefficiency stems from organizational and technological factors rather than structural ones. Policy should therefore prioritize reorganizing production systems, strengthening connections between primary agriculture and local food processing, and facilitating access to higher-value markets. The goal is to improve the productivity of existing resources rather than to increase resource deployment.

Cluster II districts (Pleven and Shumen) already operate beyond their optimal agrifood scale. Further capacity expansion would exacerbate, rather than resolve, their inefficiency. The priority in these districts should be to upgrade the quality and value of existing output through product diversification and processing innovation that increases economic returns per unit of input without expanding the overall resource base.

Cluster I near-frontier districts operate close to the efficiency frontier and have demonstrated effective resource transformation. In these territories, direct investment support yields diminishing returns. The most effective policy instrument is knowledge exchange, including structured peer learning from benchmark districts, technical advisory support, and inter-district cooperation frameworks.

5.2. Efficiency-Based Knowledge Transfer

The super-efficiency scoring and peer analysis identify Kardzhali, Razgrad, Sofia district, and Blagoevgrad as the “deep frontier” districts. These territories demonstrate that high agrifood output can be achieved through structurally diverse and resource-efficient approaches, combining focused specialized production with agro-industrial integration. Their input–output configurations represent best-practice models that are, in principle, directly transferable.

However, the spatial autocorrelation result indicates that geographic proximity does not drive efficiency spillovers. Efficient districts are spatially isolated, surrounded by less efficient neighbors, rather than serving as regional diffusion cores. Consequently, proximity-based knowledge transfer cannot be expected to succeed. For example, districts with structurally more relevant efficiency models are better for transfer than geographic neighbors, as the two districts share a similar profile.

The implication is that knowledge transfer should be structured around efficiency-based district pairing rather than geographic clustering. Policy should establish thematic inter-district cooperation networks that connect districts with similar agrifood structures, regardless of location, to facilitate the diffusion of production organization models, food-processing practices, and value chain governance mechanisms. Study visits, shared technical advisory services, and joint demonstration programs should target districts identified as peers of the benchmark units in the DEA reference set.

5.3. Output-Conditioned Public Investment Assessment

The composite TCI reveals a dimension of competitive performance not captured by the standard CRS efficiency score: resource utilization efficiency. Pazardzhik, the most illustrative case, achieves full CRS efficiency (rank 11) but ranks only 17th on the TCI because its high manufacturing capital investment generates disproportionately low industry output. This divergence between frontier status and TCI ranking is a structural finding, not an anomaly, as it identifies districts where the current capital configuration exceeds what the output mix can productively absorb.

This finding has a direct and actionable implication for investment decisions. The standard project selection criterion, which prioritizes the largest investment and highest projected economic impact, is counterproductive for districts with existing capital slack. Allocating additional public investment to enterprises in these districts exacerbates structural misalignment. Public support should instead be conditional on demonstrated improvement in output per unit of capital. This approach shifts the function of public investment to output leverage, aligning incentives with efficiency objectives.

5.4. District-Level Planning as the Foundation for Evidence-Based Spatial Intervention

The absence of significant spatial clustering has important implications for the design of Bulgaria’s rural development planning. Bulgaria’s CAP Strategic Plan is currently designed and monitored primarily at national level. It aggregates multiple NUTS-3 districts of structurally heterogeneous types, combining efficient and inefficient districts as well as rural and urban/intermediate territories. As a result, the efficiency signal is diluted at the region planning level, making it an unreliable basis for targeted interventions. Agrifood efficiency is spatially random at the district level; geographically defined instruments targeting broad administrative zones will not reach the most inefficient territories. The DEA–TCI–cluster diagnostic framework developed in this study can be used to provide a real-time efficiency map of Bulgaria’s rural agrifood sector. This would enable allocation decisions to be guided by demonstrated performance gaps rather than by administrative geography or historical expenditure patterns. It would constitute a concrete and cost-effective step toward evidence-based spatial targeting of rural development support.

6. Conclusions

This study evaluated the productive efficiency of agrifood systems across 21 predominantly rural districts in Bulgaria as a measure of territorial competitive performance. The principal empirical findings are as follows.

Under the CRS model, 12 out of 21 districts achieved full efficiency (CRS = 1.000), establishing a geographically and structurally diverse efficiency frontier. The average CRS efficiency was 0.928, suggesting a mean output expansion potential of 7.8%. Nine districts were CRS-inefficient; six fell below 0.90 (Kyustendil, Sliven, Stara Zagora, Veliko Tarnovo, Haskovo, Pleven). Returns-to-scale analysis identified three IRS districts (Lovech, Targovishte, Kyustendil) operating below optimal scale, and six DRS districts (Stara Zagora, Pleven, Shumen, Sliven, Haskovo, Veliko Tarnovo) operating above it. Slack analysis showed capital investment surpluses in manufacturing for Lovech and Shumen, with notable agricultural land surpluses in Haskovo and Lovech. Bootstrap bias correction lowered the mean efficiency estimate to 0.892, confirming the standard model’s underestimation of inefficiency gaps, most pronounced in Kyustendil, Sliven, and Stara Zagora. Super-efficiency scoring identified Kardzhali, Razgrad, Sofia, and Blagoevgrad as the most robust and policy-relevant benchmarks. The composite TCI highlighted Pazardzhik (ranked 17th) as a frontier district with a structural capital absorption issue. Cluster analysis (k = 4, silhouette = 0.822) identified four groups: a near-frontier majority (Cluster I, n = 13), a scale-constrained pair (Cluster II: Pleven, Shumen), a technically weak group with near-optimal scale (Cluster III: Lovech, Sliven, Haskovo), and a deeply underperforming group with compounded scale and technical deficits (Cluster IV: Veliko Tarnovo, Stara Zagora, Kyustendil). Moran’s I = 0.016 (p = 0.198) confirmed no significant spatial clustering of efficiency, supporting the conclusion that structural and institutional factors, rather than geographic proximity, primarily drive inter-district competitive variation.

These findings carry direct implications for sustainable development planning and policy. The efficiency gaps documented across Cluster III and Cluster IV districts, affecting territories that collectively account for a substantial share of Bulgaria’s rural population, represent structural barriers to achieving targets on sustainable agricultural productivity and on decent work and economic growth in rural areas. The spatial randomness of efficiency scores further implies that sustainability progress in Bulgarian rural districts cannot rely on geographic spillovers or proximity-based policy instruments; it requires district-level diagnostics of the kind provided by the DEA–TCI–cluster framework developed here. The absence of a single agrifood efficiency model among frontier districts, evidenced by the structural diversity of CRS-efficient units ranging from plains grain districts to mountainous high-intensity systems, suggests that sustainability-aligned rural development must accommodate territorially differentiated pathways rather than uniform targets. The results confirm that rural development policies should move beyond uniform national interventions and instead adopt territorially differentiated strategies that address the specific structural constraints faced by different groups of rural districts.

This study has several limitations warranting attention in future research. The cross-sectional approach, based on three-year averages, does not capture dynamic efficiency changes. Extending the analysis to a panel framework using Malmquist productivity indices [65,66] would help determine whether districts on the 2022–2024 frontier demonstrate stable competitive advantages or only temporary configurations. A DEA–Malmquist extension would align with recent EU agricultural efficiency studies [66] and allow decomposition of productivity change into efficiency catch-up and technological shift. The output set, constrained by NSI district data, omits environmental or sustainability indicators. Including variables such as organic farming shares, greenhouse gas intensity, or nitrogen surplus as undesirable outputs would reflect eco-efficiency priorities and align with sustainability-oriented DEA research [50,52], yielding a more comprehensive competitiveness assessment under CAP greening. The sample size (N = 21) does not fully satisfy the parsimony guideline, which limits the VRS model’s discriminating power. Advanced regression [65] could identify contextual determinants (e.g., governance quality, cooperative density, CAP participation, remoteness index) of CRS efficiency variation. Network DEA modeling of the two-stage production-to-processing chain [3] would help disaggregate sources of inefficiency beyond the current single-stage model. Finally, integrating qualitative evidence on value chain governance, institutional capacity, and food system organization would provide the interpretive depth necessary for contextualized rural development recommendations and help explain the near-zero spatial autocorrelation in terms of institutional factors isolating efficient districts from their neighbors. A particularly important direction concerns the efficiency of mountainous districts, where a mixed-methods approach would provide the interpretive depth needed to translate efficiency benchmarks into contextually grounded rural development recommendations.