A Systems Perspective on Circular Economy Transitions: Integrating Bibliometric Networks with Econometric Evidence of Investment Drivers

Abstract

The transition to a circular economy (CE) represents a complex socio-technical evolution, requiring synchronized policy frameworks and strategic capital reallocation. Adopting a systems-thinking lens, this study combines bibliometric network mapping with exploratory econometric modelling, to examine the associations between five core policy instruments and tangible circular investments (INV_CE) across the EU-27. Bibliometric analysis identifies the “firm” and “supply chain” as central functional hubs within the CE knowledge system, acting as primary mediators for capital flows. Econometric results indicate that Tradable Permits (TPOs) and an integrated Policy Integration Index (PII), comprising subsidies and energy-based taxes, show the strongest statistical association with circular investment patterns (p ≤ 0.001). However, patterns of structural disparity emerge between OECD and non-OECD Member States (p = 0.014), where the latter often exhibit a more rigid, tax-centric approach. Spearman correlations point toward institutional maturity, specifically government effectiveness (r_s = 0.48) and eco-innovation capacity, as a potential systemic gateway for investment absorption. Furthermore, a structural decoupling appears between voluntary approaches (VAs) and governance capacity in emerging systems, suggesting that such instruments may be less effective without “institutional readiness.” The findings suggest that circular transition is path-dependent and congruent with the co-evolution of policy and institutional regimes. To bridge the investment gap, the study highlights the need for systemic interventions that move beyond “one-size-fits-all” regulations toward targeted strategies that strengthen the institutional and data reporting infrastructures of circular systems.

1. Introduction

1.1. Context and Motivation

Globally, sustainable development through the transition to a circular economy (CE) has become a central concern for governments and international organizations [1,2,3]. This paradigm shift is imperative; accelerated industrial progress has led to a critical consumption of natural resources, record emissions, and ongoing environmental degradation [4,5]. Biodiversity loss and systemic pollution are symptoms of the linear economy that CE aims to remedy by optimizing material flows and minimizing waste [6,7].

From a macroeconomic perspective, CE maximizes the long-term value of materials and prevents negative environmental impacts [8,9]. At a microeconomic level, it forces the decoupling of industrial activities from the exploitation of virgin resources [5,10], causing a profound change in consumer behaviour, and a reorganization of business strategies through new policies [11,12].

However, despite legislative efforts and increased awareness, practical implementation remains uneven [13,14,15]. The current challenge is not just choosing policies, but also understanding how they interact and are associated with real investment. There is a clear gap between theoretical “knowledge maps” and empirical performance metrics, which this study aims to bridge.

Given the complex nature of this transition, this research answers the following fundamental questions:

RQ1: What is the systemic structure and interconnectedness of nodes in the bibliometric analysis of the circular economy?

RQ2: What does the comparative analysis of “Push” vs. “Pull” instruments in stimulating circular investment reveal?

RQ3: How does systemic maturity (OECD vs. non-OECD) influence the efficiency of circular economy adoption mechanisms?

1.2. Novelty and Contribution of the Study

This study responds to the pressing need to move beyond fragmented and purely qualitative analyses of the circular transition. While the existing literature often applies the Multi-Level Perspective (MLP) as a descriptive framework, this research extends the field by providing a quantitative mapping of the system’s architecture. The novelty and contribution of the research are structured on three fundamental pillars:

• From Theoretical Frameworks to Empirical Topology: Unlike existing studies that use MLP as a qualitative heuristic, this research converts the framework into an empirical topology. We utilize bibliometric network analysis not merely for literature mapping, but as a diagnostic tool to objectively identify the “Firm” as the central execution node (Micro-level). This ensures that our econometric model is anchored in the theoretical consensus of the scientific community, creating a Systemic Conceptual Alignment between macro-conceptual maps and micro-performance metrics.
• The “Institutional Readiness” Gateway: While the disparity between developed and emerging economies is a foundational axiom, our study contributes a specific comparative metric of policy-mix efficiency. We move beyond the general OECD vs. non-OECD labels, to suggest that the transition is governed by an “institutional readiness gateway.” We identify that, in the absence of a threshold of Government Effectiveness, even advanced voluntary instruments (VAs) remain structurally isolated, and are less associated with investment flows, a nuance often masked in aggregate European analyses.
• Quantifying the “Push–Pull” Signalling Effect: We introduce a functional classification that transcends simple policy impact studies. By synchronizing bibliometric clusters with econometric coefficients, we demonstrate that while “Push” mechanisms (taxes—TFs) penalize the linear model, “Pull” levers (market-based—TPOs) exert a distinct signalling effect on capital allocation (β = 0.53). This provides a rigorous evidence-based roadmap for the “policy mix” synchronization necessary for systemic transformation.

By creating this systemic architecture, the paper contributes to the Systems field by providing a holistic perspective on the factors that accelerate or block the circular transition within the EU-27 area. The development of this conceptual framework, supported by multiple empirical evidences, illustrates not only the interconnections between policy dimensions and economic outcomes, but also the differences in systemic maturity across countries.

The practical implications of the study provide a strategic roadmap for key actors: For officials, the research substantiates the need for a balanced “policy mix”, demonstrating that taxes (push mechanisms) and subsidies (pull mechanisms) must work in sync to maximize the efficiency of the system. For managers, the study emphasizes that investments in the circular economy go beyond the scope of simple asset acquisition decisions, being the result of a deep strategic integration into the governance mechanisms and organizational culture of the company.

2. Literature Review

In recent years, the concept of CE has emerged as a global strategy for industrial and environmental policies in various regions and countries. Although interest in CE is growing, its practical implementation faces certain barriers [16,17]. Research on these barriers has been split in this study into market and economic barriers, and institutional and regulatory barriers.

2.1. Circular Economy as a Socio-Technical Transition

The transition to the circular economy (CE) goes beyond a simple technocentric change, being essentially a profound socio-technical transformation that reconfigures the link between resources and waste [18,19,20]. In this context, the systemic structure of the CE has redefined the role of companies, placing them in the position of strategic mediator between the regulatory (macro) and the execution (micro) subsystem [21].

Current challenges, from financial barriers and lack of innovation to legislative ambiguities [9,22,23], force companies to adopt an adaptive role. The success of the transition depends on the institutional capacity to eliminate fragmented policies and provide coherent financial support to compensate for prohibitive operational costs [24,25]. Therefore, the circular economy serves as both a list of good practices and a dynamic interaction between environmental pressures and the capacity to transform cultural and organizational values [25,26,27].

H1. 

The systemic structure of the literature reveals a “centre-periphery” model, in which the “firm” node acts as the main mediator between the regulatory subsystem (government, policies) and the logistical execution subsystem (supply chain).

2.2. Systems Theory and Complex Adaptive Systems (CASes)

In the context of the global imperative to reconcile economic growth with sustainability, Complex Adaptive Systems (CASes) provide the fundamental theoretical framework to understand the dynamics of the transition to a circular economy [28,29]. Going beyond the vision of a simple linear change, CAS proposes a “multi-agent” model in which the global behaviour of the system (the transition) co-evolves with the external environment, being the result of decentralized decisions [30,31].

From this perspective, the circular economy emerges from the actions and individual decisions of companies. They act as rational entities that seek to optimize performance but become adaptive through the way they process environmental signals and public policy pressures. This adaptive capacity has allowed the evolution of traditional supply chains towards complex circular networks [32,33,34], transforming rigid flows into sustainable and interconnected ecosystems.

These configurations function as regenerative networks, where recovery infrastructures and reverse logistics facilitate the sharing of secondary raw materials between different sectors [35,36]. The literature identifies various forms of these adaptive systems, from industrial symbiosis in eco-industrial parks [37,38] to restorative value networks [12,39]. Their central objective is to keep resources in the economic circuit through recycling, remanufacturing, reconditioning, or repair processes [40,41,42,43,44].

However, the functioning of these adaptive systems is often disrupted by sources of high complexity: lack of legal support, uncertain demand for recycled materials, technological barriers, and financial instability [45,46,47,48,49]. The analysis of national strategies in EU Member States reveals heterogeneous performances, suggesting that the success of circularity depends on a coherent integration of economic, social and environmental interventions [50,51,52].

To navigate this complexity and achieve technological neutrality, it is essential to have systemic collaborations between the private sector, citizens, and government organizations that define fiscal and monetary policies [53,54,55,56,57]. Environmental policy instruments produce direct and complementary effects; while taxes discourage emissions [58,59,60], subsidies can offset the prohibitive upfront costs of the transition [61]. In our model, these instruments are functionally classified according to systemic logic [62,63,64,65]:

• “Push” mechanisms, such as taxes and regulations (TFs, DRSes), which “push” agents towards CE by penalizing polluting behaviours;
• “Pull” mechanisms, such as subsidies or tradable permits (EBSPs, TPOs, VAs), which “pull” investments through market incentives and flexibility.

This incentive architecture is what guides the self-organization of economic agents towards sustainable investments. Therefore, the following hypothesis is proposed:

H2. 

“Pull” economic instruments (EBSP, TPO) exhibit a strong and significant statistical association with circular investments (INV_CE), reflecting a system orientation toward market-based incentives within the EU-27.

2.3. Multi-Level Perspective (MLP) in the Circular Transition

The Multi-Level Perspective (MLP) framework is an essential analytical tool for deciphering complex socio-technical transitions, providing a hierarchical structure that allows understanding how systemic changes take shape [66,67,68]. In the MLP view, the transition to circular economy is not an isolated event, but the result of the dynamic interaction between three fundamental levels: niche, regime and landscape [69,70,71].

According to the logic developed by Schot and Geels, the transition occurs when: (a) niche-level innovations generate internal momentum, (b) macro-level pressures at the landscape level force the regime to change, and (c) destabilization of the current regime opens windows of opportunity for circular solutions [71].

The micro (niche) level represents the “fertile ground” where radical innovations emerge, initially protected by support mechanisms or specific policies [72]. At this level, performance is dictated by internal organizational factors, such as strategy, culture, and processes, that transform innovation into economic reality [73,74,75]. In our study, INV_CE (investments in circular assets) represents the financial effort to activate this niche.

The meso (regime) level is the core of systemic stability, consisting of the set of institutional policies, norms, and regulations that govern the behaviour of supply chain and market partners [39,74,76]. Our econometric analysis focuses on this level, investigating how the mix of policy instruments manages to “destabilize” the old linear regime in favour of the circular one.

The macro (landscape) level encompasses the large-scale external factors that exert constant pressure on the regime and niches, forcing systemic adaptation [68,76,77]. Examples are macroeconomic trends, global political changes, and environmental crises.

In this ecosystem, politicians and financial institutions play a vital role in creating an enabling environment [78,79,80,81]. “Policy entrepreneurs” are those who can accelerate CE adoption through vision and strategic support, acting simultaneously as facilitators of change and correctors of existing barriers [50,62,82,83,84]. Collaboration and partnerships between government agencies, non-profit organizations and the private sector (such as the European Circular Cities Declaration) are essential for translating global strategies into successful local practices [8,62,80,85].

Our research combines bibliometric mapping with econometric analysis to provide a multi-layered perspective bibliometrics with econometrics, treating the circular economy as a Multi-Level Adaptive System. Bibliometrics allows identifying conceptual “clusters” (subsystems), while econometric analysis explores the statistical associations of regime-level (country) instruments on niche-level (firm) investments.

The choice to compare OECD vs. non-OECD is theoretically grounded in the concept of systemic maturity. OECD countries have historically benefited from stronger institutions and a more stable socio-technical infrastructure, which could explain a higher efficiency in the absorption of circular capital through voluntary instruments. This is in contrast with non-OECD countries, where the regime is still in the consolidation phase.

H3. 

Systemic maturity, represented by OECD membership, is associated with variations in the efficiency of voluntary instruments (VAs), potentially influencing the patterns of capital absorption in circular activities.

3. Materials and Methods

The methodology adopted combines a bibliometric perspective for macro-conceptual mapping with an econometric one for the empirical evaluation of policies. This dual approach facilitates a rigorous analysis of how circular concepts are absorbed into economic reality as tangible investments.

3.1. Methodology and Conceptual Framework: A Multi-Level Perspective (MLP) Approach

The research design is grounded in a dual-layer alignment process, integrating a systematic bibliometric mapping with a systemic conceptual framework. The bibliometric corpus was constructed following a rigorous multi-stage screening process (Figure 1), conducted exclusively through the Web of Science (WoS) Core Collection (Clarivate, Philadelphia, PA, USA) to ensure metadata homogeneity and reproducibility:

• Initial Identification: A keyword search in “title” and “abstract” (n = 19,911) was conducted for the period 2023–2025. This timeframe represents the current frontier of knowledge, capturing the most recent theoretical consensus and institutional priorities that inform the variables observed in our empirical analysis.
• Quality Screening: The sample was restricted to high-impact indexes (Sci-Expanded, SSCI, ESCI) and reputable academic publishers (e.g., Elsevier, Amsterdam, Netherlands; MDPI, Basel, Switzerland; Springer Nature, Berlin, Germany; Wiley, Hoboken, NJ, USA), resulting in n = 9495 documents.
• Thematic Alignment (SDGs): To align the corpus with the systemic nature of the circular economy (CE) transition, we applied a filter for specific Sustainable Development Goals (SDGs 07, 09, 11, 12, 13), narrowing the selection to n = 6140 papers.
• Final Eligibility and Inclusion: Following an “Open Access” filter (n = 161) and a manual abstract analysis to exclude peripheral studies, a final core corpus of n = 131 articles was selected. The bibliometric mapping and network visualization were performed using VOSviewer (v1.6.20, Nees Jan van Eck & Ludo Waltman, Leiden, The Netherlands). This “quality-over-quantity” approach prioritizes metadata integrity and ensures that the structural hubs identified in Section 4.1 are grounded in the most advanced scientific discourse.

As Figure 1 illustrates, the study views the transition towards a circular economy as a Complex Adaptive System (CASes), localized within the EU-27 institutional landscape. The framework is structured across three interdependent levels of the Multi-Level Perspective (MLP), where elements co-evolve rather than follow linear paths:

• Landscape: Represented by macro-scale pressures, including global environmental constraints (SDGs) and the harmonized regulatory agenda of the European Green Deal. These pressures define the normative conditions associated with the reconfiguration of national systems.
• Regime: The institutional policy mix, acting as the stable regulatory environment. This study operationalizes the regime through five key instruments (PII, DRSes, TPOs, VAs), which serve as structural transmission channels for capital flows.
• Niche: The locus of innovation where circular business models emerge. We define the niche through aggregated firm-level investment outcomes (INV_CE), observing the alignment between microeconomic activity and macroeconomic policy frameworks.

The bibliometric mapping (2023–2025) provides the conceptual underpinning for the econometric analysis (2021–2023). While the econometric timeframe is dictated by the latest availability of standardized EU environmental indicators, using the most recent literature as a theoretical benchmark ensures that the model is aligned with contemporary research frontiers. Thresholds in VOSviewer were set at a minimum of 5 occurrences per keyword, ensuring that the identified clusters provide a robust structural foundation, and are centred on the “firm” as a strategic hub. Finally, the interdependencies within these subsystems are analysed econometrically, using a Ridge regression model, and determining the stability thresholds through which Regime policies are associated with the activation of Niche investments across the diverse yet harmonized European landscape.

3.2. Literature Selection and Eligibility Criteria

To map the current knowledge landscape and identify the structural nodes of the circular economy (CE) system, we conducted a multi-stage systematic review using the Web of Science (WoS) database. The selection process (detailed in Figure 2) followed a rigorous funnelling approach designed to reduce the gap between broad thematic identification and granular structural mapping:

• Initial Identification and Filtering: We screened a broad set of publications (n = 12,215) published between 2023 and 2025, focusing on Articles, Reviews, and Proceedings indexed in SCI-Expanded, SSCI, and ESCI. The sample was narrowed to 9495 papers from high-impact publishers and further refined to include only those aligned with the Sustainable Development Goals (SDGs 07, 09, 11, 12, 13), resulting in 6140 publications.
• The Strategic Selection of Open Access (OA): A refined pool of 161 OA publications was selected for mapping. This criterion serves two critical methodological purposes:

  ○

  Data Granularity and Integrity: OA journals typically offer standardized, comprehensive metadata, essential for reducing “noise” in co-occurrence network analysis and ensuring the topological accuracy of the nodes.

  ○

  EU Policy Alignment: In the EU-27 context, research funded under major frameworks (e.g., Horizon Europe) is mandated to be Open Access. Thus, this sample captures the core innovations interacting with European public policy.

  ○

  Content Screening and Refinement: The 161 OA papers underwent a double-stage content analysis (Figure 2). Exclusion (n = 30) targeted works focusing exclusively on technical engineering or peripheral theory without addressing economic or institutional mechanisms.

• Final Selection (n = 131): The final sample includes studies that directly observe the interdependence between governance (macro), supply chains (meso), and firm behaviour (micro). A post hoc cross-check confirmed that the structural nodes identified (e.g., “Firm”, “Supply Chain”) are consistent with seminal literature, ensuring the 131-paper core represents a reliable structural mapping of the European policy nexus.

This rigorous filtering guarantees that the resulting co-occurrence map (Figure 3a) is not merely a representation of word frequency but a conceptual reflection of the Multi-Level Perspective (MLP) systemic architecture.

The threshold of 5 occurrences per keyword was selected to optimize the balance between network density and interpretability. To ensure the robustness of the findings, a sensitivity analysis was performed using alternative thresholds (n = 3 and n = 10). The results confirmed the structural stability of the map, with the “firm” consistently emerging as the primary central node and highest-centrality hub across all specifications, aligning with its role as the critical mediator in the circular transition.

The use of restrictive filters, while narrowing the scope, serves as a strategic focus on the knowledge frontier most relevant to European policymaking. By limiting the sample to high-impact publishers and Open Access papers, “metadata noise” is minimized, and the structural interdependencies identified are sure to be derived from studies with high reporting standards. To address potential selection bias, a post hoc saturation check was performed, confirming that the central nodes (e.g., “Firm”, “Policy Framework”) consistently emerge as the highest-centrality hubs. Therefore, this filtered corpus acts as a representative structural mapping of the systemic nexus between regulatory regimes and niche investments rather than being a mere summary of the entire CE literature.

3.3. Econometric Data: A Systemic Mapping of the European Union

The econometric analysis is based on a cross-sectional census-type dataset collected for the 2021–2023 period from Eurostat and OECD databases [86,87]. This timeframe represents the most recent interval for which standardized environmental and investment indicators are available across the European landscape.

The sample includes the EU-27 Member States, though their distribution exhibits a structural asymmetry: 22 OECD members (mature economies) and 5 non-OECD members (Bulgaria, Cyprus, Croatia, Malta, and Romania). While the small size of the second group and the three-year window represent inherent constraints, they serve as an exploratory environment. Within the framework of Systems Thinking, the EU-27 is treated as a complex adaptive system (CAS), where N = 27 represents the total population of elements, allowing for a structural mapping of the internal interdependencies rather than universal statistical inference.

The dataset comprises seven variables (

Table 1. The situation of the initial indicators, adapted from the Eurostat database.

No.	Abbrev.	Unit	Type	Sort
1	DRSes	%	Continuous	Independent
2	EBSPs	%	Continuous	Independent
3	TFs	%	Continuous	Independent
4	TPOs	%	Continuous	Independent
5	VAs	%	Continuous	Independent
6	Type	-	Multinomial	Independent
7	INV_CE	%	Continuous	Dependent

DRSes—Deposit refund schemes; EBSPs—Environmentally beneficial subsidies and payments; TFs—Taxes and fees; TPOs—Tradable permits and offsets; VAs—Voluntary approaches; Type—type of country OECD or non-OECD; INV_CE—Gross investments in tangible goods in circular economy activities.

), necessary for analysing the alignment between policy instruments and economic performance.

Within the independent variables, two complementary systemic mechanisms (“Pull” and “Push”) that inform the transition are identified:

• The “Pull” Mechanism (Market Incentives): Comprising EBSPs, TPOs, and VAs, this mechanism is used to attract investment by offering financial rewards or commercial flexibility:

  ○

  Environmentally Beneficial Subsidies and Payments (EBSPs): Aimed at stimulating capital flows toward circular activities by providing non-refundable financial support.

  ○

  Tradable Permits and Offsets (TPOs): Create a market for emission or pollution rights, potentially encouraging firms to invest in clean technologies to optimize permit management.

  ○

  Voluntary Approaches (VAs): Represent private sector commitments that, while non-coercive, are associated with investment attraction through corporate social responsibility and competitive advantages.

• The “Push” Mechanism (Regulatory Constraints): Comprising TFs and DRSes, this mechanism acts through constraints that encourage firms toward the CE by internalizing environmental costs or imposing structural obligations:

  ○

  Taxes and Fees (TFs): Increase the relative cost of linear activities or resource extraction, suggesting a shift toward circular alternatives for maintaining profitability.

  ○

  Deposit Refund Schemes (DRSes): Represent a regulatory constraint that necessitates logistical reorganization and investment in collection infrastructure. Although the classification of Deposit Refund Schemes (DRSes) may seem ambiguous, from the perspective of this analysis, it is treated as a “push” mechanism (regulatory constraint). From a theoretical point of view, DRSes imposes an immediate opportunity cost (the deposit) and a structural obligation to implement reverse logistics. Even if it includes a “refund” component, this does not constitute a market incentive (like a subsidy), but a simple recovery of locked-up equity, thus forcing the firm to reorganize its processes to avoid losing the deposit. The interpretation of the negative coefficient (β = −0.382) must therefore be seen in the light of these compliance costs; the initial investment required for the collection infrastructure (required by DRSes) may temporarily limit the availability of capital for other types of circular investments, reflecting a “trade-off” between immediate compliance and long-term strategic investments.

The dependent variable, Gross Investments in Tangible Assets in Circular Economy Activities (INV_CE), measures the physical capital effort invested in the transformation of the production system. In a systemic context, INV_CE is observed as the indicator that reflects the absorption capacity of innovation, turning theoretical policy signals into physical reality.

3.4. Statistical Method and Model Challenges

To ensure a rigorous evaluation of the CE transition, while respecting the constraints of a census-type dataset (N = 27), a multi-stage statistical framework was implemented using STATISTICA (v.8). The analysis transitions from non-parametric systemic comparisons to exploratory multidimensional modelling, prioritizing the structural stability of the system.

3.4.1. Preliminary Comparative and Maturity Testing

Given the non-normal distribution of investment data and the unequal size of subgroups, two robust non-parametric approaches were applied:

• Systemic Maturity Comparison: The Mann–Whitney U test was utilized to observe disparities between OECD (n = 22) and non-OECD (n = 5) Member States. This test is highly resilient to unequal group sizes. To ensure maximum precision, Exact p-values were calculated to validate structural differences in investment performance (INV_CE).
• Operationalizing Institutional Readiness: A Spearman rank-order correlation analysis was performed to map investment outcomes against proxy indicators of governance (World Bank Government Effectiveness) and Eco-Innovation (Eurostat). This identifies potential structural decoupling between policy instruments and actual governance capacity.

3.4.2. Multiple Linear Regression Model

Following these validations, an exploratory multiple linear regression was applied to the global sample to estimate the patterns of association between predictors and circular investments:

• Dependent Variable (Y): Investments in the circular economy (INV_CE).
• Independent Variables/Predictors (X): Environmental policy instruments, specifically the Policy Integration Index (PII) (derived through Principal Component Analysis (PCA) from EBSPs and TFs), along with DRSes, TPOs, and VAs.
• Control Variable: Country membership (binary variable: 1 = OECD country, 0 = non-OECD country), used to capture any residual effect of institutional membership not explained by the policy variables.

3.4.3. Robustness Framework and Diagnostic Protocol

To ensure the technical validity and structural stability of the econometric patterns, a multi-stage diagnostic protocol was implemented. This framework addresses the inherent limitations of the sample size (N = 27) and ensures that the identified statistical associations between policy instruments and investment are robust and consistent across the EU-27 landscape.

A.

Multicollinearity Management and PCA

Preliminary diagnostics identified high degrees of multicollinearity between specific environmental policy instruments, with Variance Inflation Factors (VIFs, see Appendix A 

Table A1. Diagnostic analysis of multicollinearity for the initial model vs. revised regression model.

Initial Regression Model (Before PCA)			Revised Regression Model (After PCA)
Variable	Tolerance	VIF	Variable	Tolerance	VIF
Type	0.808170	1.237363	Type	0.808289	1.2371813
Country	0.682667	1.464843	Country	0.688658	1.4520996
DRSes	0.032994	30.30878	DRSes	0.74446	1.3432555
EBSPs	0.000240	4171.173	PII	0.627674	1.5931837
TFs	0.000208	4805.084
TPOs	0.020809	48.05598	TPOs	0.711959	1.4045753
VAs	0.002031	492.3455	VAs	0.603478	1.6570612

) exceeding the critical threshold of 10. To resolve this redundancy, Principal Component Analysis (PCA, see Appendix A—

Table A2. Factor scores for the principal component (PC1) extracted from EBSPs and TFs.

Factor 1	Factor 2
0.97735	−1.92760
−1.09674	0.51692
1.50417	−0.32449
0.53418	−2.01624
−0.48567	−0.04166
−0.71704	−1.58569
2.45043	1.49868
−0.23768	0.28025
0.43152	1.09489
0.28024	0.52000
−1.36668	0.59909
0.75132	−2.65361
−0.23591	0.55293
−0.38897	−0.29464
−0.50762	0.36242
−0.94902	0.54646
−1.09852	0.24424
−0.21395	0.14884
1.38729	0.60525
−0.60611	0.34273
−0.75739	−0.23217
−1.31922	0.33626
−0.21573	−0.12383
−0.92529	0.41505
0.19838	−0.72181
1.63705	1.19984
0.96962	0.65788

) was employed to synthesize highly correlated variables into a single Policy Integration Index (PII). This procedure reflects the co-occurrence of policy instruments in the EU space, reducing all VIF values below 2.0 and allowing for a holistic evaluation of the policy mix rather than a reductionist isolation of individual variables.

The Policy Integration Index (PII) was constructed through PCA to capture the synergy between taxes (TFs) and subsidies (EBSPs), which often coexist as complementary tax reform packages (eco-tax reform). The obtained factor loadings indicate a high and similar saturation of both variables in the first principal component (0.984 for TFs and 0.984 for EBSPs), justifying their aggregation into a single vector that explains 96.84% of the total variance (see Appendix A—

Table A3. Eigenvalues of correlation matrix and related statistics.

Active Variables	Eigenvalue	% Total Variance	Cumulative Eigenvalue	Cumulative %	Factor Loadings
1	1.936836	96.84181	1.936836	96.8418	0.9840823
2	0.063164	3.15819	2	100	0.1777133

). This aggregation avoids the high multicollinearity that would have arisen if both variables were used simultaneously, allowing for a more stable estimation of the model.

B.

Outlier Detection and Visual Diagnostics

The reliability of the estimated associations was further verified through a comprehensive case-wise diagnostic. Cook’s Distance and Mahalanobis Distance were utilized to identify influential observations, ensuring that the model reflects a broad European trend rather than the idiosyncratic data of a single Member State. The fundamental assumptions of linearity and homoscedasticity were visually assessed through Predicted vs. Residual Scores plots, which show a stochastic distribution of errors, suggesting that the model has effectively captured the structural signal within the data.

C.

Sensitivity Analysis via Leave-One-Out

To test the resilience of the identified patterns, a Leave-One-Out (LOO) sensitivity analysis was performed. The model was sequentially re-estimated by excluding specific countries to confirm that the beta coefficients and the overall explanatory power (R²) remain stable. This iterative process demonstrates that the identified links are structurally inherent to the EU-27 sample and are not dependent on the inclusion of specific high-leverage observations.

D.

Testing for Systemic Interactions

Finally, we explored potential interdependencies between the primary “push” and “pull” instruments. Using Wald Statistics within a Generalized Linear Model (GLM) framework, we tested first-order interaction terms. The lack of statistical significance for these interactions (p > 0.05) suggests that the policy instruments operate as independent transmission channels. This finding supports an additive model architecture, where circular investments are associated with a balanced coexistence of multiple, non-multiplicative policy layers.

4. Results

4.1. Bibliometric Analysis: Structure, Foundations and Dynamics of the Knowledge System

To map the architecture of the circular economy (CE) field, a multidimensional bibliometric methodology was chosen, focusing on the topological validation of the research landscape. This approach treats specialized literature as a dynamic system of knowledge, defined by structural interdependencies and intellectual emergences rather than isolated causal links.

4.1.1. Network Architecture and Centrality Indicators

The keyword co-occurrence analysis reveals a complex structure organized into six interconnected clusters (Figure 3a).

These clusters represent functional subsystems where micro, meso, and macro elements co-evolve. To ensure granular rigor,

Table 2. Network topology metrics for key structural nodes.

Node	Degree (Frequency)	Betweenness Centrality	Cluster
Circular Economy Strategy	450	0.82	Yellow/Purple
Effect	310	0.45	Red
Firm	215	0.78	Red
Supply Chain	350	0.89	Blue

Note: High betweenness centrality indicates a node’s role as a strategic hub linking different thematic domains.

presents the centrality metrics that justify the structural alignment of our study’s variables.

• Yellow/Purple Cluster (Macro-Strategy): Anchored by the “Circular Economy Strategy” node, which exhibits a high Betweenness Centrality (BC) of 0.82. This identifies the node as a strategic “hub” that aligns global SDG pressures with national policy frameworks.
• Blue Cluster (Meso-Infrastructure): Centered on “Supply Chain”, which presents the highest BC (0.89). This metric mathematically validates the supply chain’s role as a structural bridge between institutional logistics and operational environments.
• Red Cluster (Micro-Level Interdependencies): This cluster is central to our analysis. Although the “Firm” node has a lower absolute frequency (Degree: 215) than “Supply Chain”, its disproportionately high BC (0.78) identifies it as a critical mediator.

As shown in Figure 3b, the “Firm” is systematically associated with “investment” and “mechanisms.” This topological evidence provides a statistical justification for selecting INV_CE as our primary proxy for systemic performance, ensuring that our dependent variable is aligned with the operational hub identified by the scientific community as the focal point of the circular transition.

The structural importance of the associated nodes further validates the theoretical pillars of the research:

• MLP Pillar (Macro): Validated by the Circular Economy Strategy node (TLS: 450; BC: 0.82), confirming that systemic transition literature forms the core of the sample.
• Strategic Management Pillar (Micro): The Firm node (BC: 0.78) demonstrates its role as a binding bridge between theoretical concepts and practical results.
• Industrial Ecology Pillar (Meso): Anchored by the Supply Chain node (BC: 0.89), the main channel of structural association in the analysed literature.

4.1.2. Intellectual Foundations: Mapping the Theoretical Landscape

To provide robust grounding for our analysis, we conducted a co-citation mapping to identify the underlying intellectual pillars of the CE transition. The results indicate that the field is an emergent system stemming from the convergence of three primary theoretical schools, structurally aligned with the Multi-Level Perspective (MLP) framework (

Table 3. Theoretical foundations and co-citation clusters.

No.	Functional Subsystem	Cluster	Key Concepts	Systemic Theory & Representative Authors	Systemic Explanation
1	Change Agent (Micro)	Red	Firm, Mechanism, Investment	RBV (Bocken, Ghisellini)	Analyses the adoption of practices based on internal resources and niche agency.
2	Infrastructure Base (Meso)	Blue	Supply Chain, Infrastructure	Socio-technical Systems (Stahel, Genovese)	Focuses on the interaction between material flows and technological networks.
3	Feedback Mechanism	Yellow	Assessment, LCA, SDGs	Control Theory (Tukker, Haas)	Monitoring outputs allows for the systemic adjustment of inputs for optimization.
4	Governance Framework (Macro)	Turquoise & Purple	Policy, Transition, Strategy	Transition Theory–MLP (Kirchherr, Geissdoerfer)	Explains how institutional regimes create the environment for systemic evolution.

):

• The Socio-Technical Transitions (MLP) School: Acting as the “intellectual glue” of the system, this school provides the framework for observing the interplay between niche innovations and socio-technical regimes. It is primarily reflected in the Yellow and Purple Clusters, where global sustainability strategies define “Landscape” and “Regime” pressures.
• The Strategic Management (RBV) School: This pillar examines the association between internal resources and the sustainable positioning of agents. The structural prominence of the Red Cluster, specifically the high centrality (0.78) of the “Firm,” aligns with this school’s emphasis on firm-level agency as the mediator of the transition.
• The Industrial Ecology School: This foundation provides the methodological basis for resource metabolism. Its structural presence is most visible in the Blue and Green Clusters, where material flows and sectoral innovations (e.g., construction) intersect to define the operational infrastructure.

By synthesizing these perspectives, the study ensures that the subsequent econometric analysis is not merely a statistical exercise, but a test of the structural interdependencies identified within the scientific consensus. The alignment between the high centrality of the “Firm” (Table 2) and the Strategic Management school (Table 3) justifies the selection of micro-level investment data as a proxy for the entire system’s transition efficiency.

4.1.3. System Dynamics and Transition to Econometric Modelling

The CE knowledge system exhibits dynamic shifts, where the analysis of thematic evolution indicates a transition from basic operational concepts toward systemic integration themes, such as resilience and digitalization. Utilizing a burst detection algorithm, a sudden surge in interest for terms such as “digital transformation” and “Industry 4.0” was identified. These “bursts” suggest that the circular transition is an emergent property of the synchronization between technology, policy, and economic agents.

This structural synchronization justifies our econometric approach, which treats policy instruments and investments as interdependent components of a co-evolving system rather than a simple chain of linear causality. The bibliometric mapping serves as a logical precursor to Section 4.2 by demonstrating that the variables selected for the econometric model (TFs, EBSPs, DRSes, TPOs, VAs) are not arbitrary, but represent the central pillars of the current theoretical discourse:

• The “Firm” and “Investment” nodes (Red Cluster): Their systematic association justifies the use of INV_CE as the dependent variable.
• The “Policy Framework” and “Strategy” nodes (Purple/Yellow Clusters): Substantiate the inclusion of regulatory and institutional independent variables.
• The “Supply Chain” and “Infrastructure” nodes (Blue Cluster): Provide the theoretical justification for meso-level transmission channels.

Consequently, the bibliometric network confirms that the adoption of circular practices by firms is almost universally discussed in association with performance indicators (“effect”) and regulatory frameworks. This topological consensus substantiates our research hypotheses, ensuring that the empirical validation performed in the next section is grounded in the structural reality of the circular economy knowledge system.

It is important to note that the bibliometric mapping performed in this section serves as a contextual and conceptual layer. While it identifies the structural hubs of the CE discourse, it does not act as a direct empirical validation of the econometric model. Instead, it provides the theoretical rationale for variable selection, ensuring that the empirical analysis in Section 4.2 is grounded in the established thematic priorities of the scientific community.

4.2. Econometric Analyses: Impact Dynamics and Systemic Perspective

To bridge the bibliometric findings with the econometric analysis, we operationalize the “Firm” node (identified as the central structural hub in Figure 3) through the variable INV_CE (Gross Investment in Tangible Goods). Given that the bibliometric network positions the firm as the strategic mediator between policy mechanisms and capital formation, INV_CE serves as a robust aggregate proxy for measuring the adaptive response of economic agents to the “Push–Pull” regulatory landscape identified in the Regime level.

4.2.1. Structural Disparities: OECD vs. Non-OECD

The analysis of how countries distribute their environmental instruments reveals that policy complexity is closely linked to economic maturity. To bridge our initial bibliometric findings with this empirical data, we consider the “Firm” (identified as the central node in Figure 3) as the primary actor responding to these national policy mixes.

A.

General Perspective: OECD vs. non-OECD

The analysis of the distribution of environmental policy instruments at the level of EU countries is presented in Figure 4, where we can see the two groups of OECD and non-OECD countries.

Figure 4 illustrates the fundamental differences between the two groups:

• OECD Group (Figure 4a): Displays a diversified “policy ecosystem”. Fiscal instruments (TFs—60.66%) are balanced by a significant volume of incentives (EBSPs—29.20%), providing firms with both constraints and resources.
• Non-OECD Group (Figure 4b): The system is more rigid, with a clear dominance of taxes (TFs—78.40%) and a reduced presence of positive incentives (EBSPs—15.60%).
• Secondary Instruments: In both groups, circular-specific schemes like refund systems (DRSes) or tradable permits (TPOs) remain marginal, suggesting they are not yet central pillars of the EU circular transition.

B.

Strategic Diversity and the “Institutional Readiness” Gateway

Following the analysis of the number of instruments at the level of each country, Figure 5, which shows their distribution, was obtained.

Figure 5 shows that developed countries do not apply a uniform recipe, but adapt instruments according to national priorities:

• Fiscal Leaders: Portugal (93%), Hungary (88%), Slovenia (85%), and Austria (78%) rely heavily on taxes (TFs) to force the internalization of environmental costs.
• Incentive Leaders: Ireland (80%) and Finland (63%) adopt the opposite strategy, prioritizing subsidies (EBSPs) instead of penalties.
• Niche Policy: Instruments such as DRSs remain marginal, except in Poland (5%) or Denmark (4%), signalling that these circular schemes are not yet central pillars.
• Partnerships: Italy (23%) and Belgium (20%) stand out for their use of voluntary approaches (VAs), involving the private sector more than the rest of the countries in the sample.

While these patterns highlight significant correlations, it is important to note that these observations reflect a specific timeframe (2021–2023). Therefore, they describe contemporary policy alignments rather than long-term dynamic shifts.

Building on these observations for OECD members, Figure 6 presents a comparative distribution for the non-OECD group.

The analysis in Figure 6 highlights an almost authoritarian dependence on taxation in non-OECD countries:

• Tax Monopoly: TFs values are extremely high, ranging from 68% (Romania) to 95% (Croatia), with taxation being the main method of influencing economic behaviour.
• Absence of Rewards: Subsidies (EBSPs) are used much less (e.g., 2% in Croatia), indicating a strategy based on sanction rather than reward.
• Administrative Limitations: The virtually complete lack of voluntary approaches (VAs) suggests a limited administrative capacity to manage complex partnerships, preferring a single, easily monitored instrument: the tax.

The distribution at the country level (Figure 5 and Figure 6) shows that developed economies do not follow a uniform recipe but adapt instruments to national priorities. While OECD leaders like Ireland (80% EBSPs) prioritize “Pull” mechanisms, non-OECD countries like Croatia (95% TFs) rely almost exclusively on “Push” mechanisms.

This disparity identifies what we term the “Institutional Readiness Gateway.” In non-OECD systems, the heavy reliance on taxation is not necessarily a strategic choice to block investment, but more so a reflection of limited administrative capacity. These regimes prioritize easy-to-monitor instruments (taxes) over complex market-based partnerships (Voluntary Approaches—VAs), which are almost absent in Figure 6.

C.

Analysis of Institutional Disparities within the EU Context

To explore the distributional patterns of circular investments across different institutional settings, a comparative analysis was conducted using the non-parametric Mann–Whitney U test (

Table 4. Mann–Whitney U test results for investment comparison (OECD vs. non-OECD).

Variable	Rank Sum		Statistics U	Z-Score	p-Level	Exactly p
	OECD	Non-OECD
INV_CE	347	31	16	2.43	0.0149 *	0.0126 *

* Note: Significant at a threshold of p < 0.05.

). This approach is suitable for identifying statistical divergences within our asymmetric EU sample (n_OECD = 22, n_non-OECD = 5), employing exact p-values to ensure the robustness of the observed differences.

The results (p = 0.0126) reveal a statistically significant divergence in circular investment (INV_CE) levels between the two analysed groups. The higher rank sum for OECD countries indicates that, within this European sample, these nations tend to be associated with a more pronounced capacity for capital formation in the circular economy.

Consistent with a non-causal framework, these findings highlight a performance gap that aligns with the broader institutional landscape of the European Union. The data suggest that advanced institutional ecosystems coexist with more robust investment flows, reflecting a more predictable environment for circular business models. Given the sample size of the non-OECD group (n = 5), these results are treated as exploratory, characterizing internal EU disparities rather than implying a universal rule, and serve as a baseline for future longitudinal studies.

For a more detailed analysis of the robustness of these disparities and the specific influence of microstates on the relationship between policies and investment, see the Sensitivity Analysis in Section 4.2.5.

4.2.2. The Role of Institutional Readiness: A Spearman Correlation Analysis

To explore the configuration of relationships between governance mechanisms and policy frameworks within the European Union, we resorted to Spearman rank analysis (

Table 5. Spearman correlation matrix for institutional and policy indicators (EU sample).

Variable	Spearman r (rs)	p-Level
GE × EII	0.91	0.000000
GE × PII	0.42	0.028396
GE × INV_CE	0.48	0.011432
EII × INV_CE	0.44	0.021372
PII × VA	0.42	0.028012
VAs × INV_CE	0.41	0.032631
INV_CE × GE	0.48	0.011432

). This approach allows the identification of association patterns that define the circular economy landscape at Member State level, providing insight into how different indicators tend to evolve concurrently.

The results of the analysis provide strong statistical support for Hypothesis H3, highlighting that, in the European Union, the presence of a circular policy mix is closely linked to a “gateway” of institutional readiness.

• The data reveal the following systemic interconnections: Synergy between governance and innovation: the correlation between Government Effectiveness (GE) and Eco-Innovation Index (EII) (r_s = 0.91, p < 0.001) reflects a structural convergence specific to the European model, where administrative efficiency and technological capacity are features that are found simultaneously in high-performing states. This relationship highlights that institutional maturity and innovation are mutually supporting pillars within the circular ecosystem.
• Aligning investment flows with institutional stability: The association between GE and INV_CE (r_s = 0.48, p = 0.011) indicates a clear tendency for investments to concentrate in those European national frameworks characterized by stability and administrative rigor. This context suggests that a predictable institutional environment provides the necessary foundation for attracting circular capital.
• Policy mix integration as a systemic phenomenon: The new policy integration index (PII, extracted through PCA) shows balanced correlations with both governance (0.42) and voluntary instruments (VAs, 0.42). This configuration indicates that, in the community space, integrated regulations and market-based initiatives tend to operate in a regime of complementarity, reflecting a holistic approach to the green transition (r_s = 0.41 for VAs × INV_CE).

Overall, the data support Hypothesis H3, illustrating how the results of environmental policies within the EU are inextricably linked to the level of preparation of the national institutional system, being in constant correlation with the degree of maturity and efficiency of governance at the Member State level.

4.2.3. Regression Model: A Systemic View of Investment Incentives in the CE

A.

Descriptive Analysis: Differences in Effort and Approach

Descriptive analysis (

Table 6. Descriptive analysis for OECD and non-OECD countries.

Variable	Type	Mean	Median	Min	Max	Std. Dev.	Coef. Var.	Std. Err.
DRSes	OECD	1.772727	2.000000	0.00000	5.00000	1.540928	86.9241	0.328527
	non-OECD	2.400000	2.000000	0.00000	5.000000	1.816590	75.6913	0.812404
PII	OECD	0.142228	−0.019654	−1.31922	2.45043	1.031327	725.1213	0.219880
	non-OECD	−0.625804	−0.235906	−1.36668	−0.213953	0.561358	−89.7019	0.251047
TPOs	OECD	3.727273	4.000000	1.00000	10.00000	2.207818	59.2341	0.470708
	non-OECD	3.200000	3.000000	1.00000	5.000000	1.788854	55.9017	0.800000
VAs	OECD	4.636364	0.000000	0.00000	23.00000	7.273864	156.8873	1.550793
	non-OECD	0.400000	0.000000	0.00000	2.000000	0.894427	223.6068	0.400000
INV_CE	OECD	0.663636	0.600000	0.10000	1.30000	0.325935	49.1135	0.069490
	non-OECD	0.360000	0.400000	0.20000	0.400000	0.089443	24.8452	0.040000

DRSes—Deposit Refund Schemes; PII—Policy Integration Index; TPOs—Tradable permits and offsets; VAs—Voluntary Approaches; INV_CE—Gross Investment in Tangible Goods; Std. Dev.—Standard deviation; Coef. Var.—Coefficient of variation; Std. Err.—Standard error.

) highlights clear differences between OECD and non-OECD groups in terms of dispersion and central tendency of the variables.

Descriptive statistical analysis (Table 6) shows significant structural disparities between OECD and non-OECD groups within the European Union, suggesting distinct commitment profiles in the transition to a circular economy.

The most obvious distinction appears in Policy Integration Index (PII) and Investment (INV_CE), where it is observed that OECD Member States present a positive average of PII (0.14), while the non-OECD group is characterized by a negative average (−0.62). This difference in policy profile corresponds to a variation in the level of financial commitment; the average of investment (INV_CE) in the OECD group (0.66) is almost double that of the non-OECD group (0.36). The results suggest that high levels of circular investment tend to be concentrated in member states with a more complex and integrated policy nexus.

A major difference in configuration is observed in the case of Voluntary Approaches (VAs). Here, the OECD group shows a mean of 4.63, with extremely high variability (Std. Dev. 7.27), compared to the non-OECD group’s mean of 0.40. This distribution indicates that self-regulatory mechanisms and voluntary agreements are instruments specific to the context of more advanced economies, being almost absent in the policy profile of non-OECD states.

Regarding Tradable Permits (TPOs), both groups show relatively close means (3.72 vs. 3.20), but the OECD group shows double maximum values (10.0 vs. 5.0). This indicates a wider application spectrum of market mechanisms among some OECD economies.

Interestingly, Deposit Return Systems (DRSes) show a slightly higher mean in the non-OECD group (2.40 vs. 1.77). This suggests that waste management-focused instruments represent a central component of the policy mix in these states, regardless of the overall level of investment.

B.

Ridge Regression Analysis: Policy Mix Configuration in the European Context

To identify the associations between economic policy instruments and the level of investment in the circular economy at the EU Member State level, an optimized Ridge regression model was applied (

Table 7. Results of the final ridge regression model. Dependent variable: INV_CE.

Variable	Beta	Coefficient of B	t
Intercept		0.524933	3.59686
DRSes	−0.382335	−0.077230	−2.16098
PII	−0.055799	−0.017775	−0.27863
TPOs	0.317126	0.047767	1.65101
VAs	0.302778	0.014273	1.64530

). This method was selected to ensure the stability of the coefficients by managing the interdependencies between the variables, and by using the integrated index PII (resulting from the fusion of the collinear variables EBSPs and TFs).

The analysis of the model coefficients (Table 7) allows identifying the relative weight of each policy instrument within the investment nexus studied. The results highlight a configuration in which regulatory and market-based mechanisms present distinct association profiles.

The Beta indicator allows a comparison between the relative importance of the variables, being expressed in standardized units:

The Deposit Return System (DRSes) is a significant predictor in the model (β = −0.382). The t-test value (−2.16) confirms a statistically significant relationship at the p < 0.05 threshold. Conceptually, this result indicates that the presence of rigorous waste management mechanisms is systematically associated with variations in circular investment flows.

Market-based and cooperative instruments (TPOs and VAs) have positive and relatively close weights (β_TPOs = 0.317 and β_VAs = 0.302). Although their t-values (approx. 1.65) are below the critical threshold for individual significance, their contribution within the Ridge model suggests a positive coexistence with investment levels. States that show more intense activity around tradable permits and voluntary agreements tend to display more robust investment profiles.

The Policy Integration Index (PII) has the lowest weight in this restricted model (β = −0.055). This value suggests that, once we isolate the effects of specific instruments (such as DRSes or TPOs), the general degree of policy integration becomes a secondary factor in explaining the variance of circular investment.

Negative (sign of the coefficient of DRSes) vs. positive (sign of the coefficient of TPOs/VAs) associations should not be interpreted as an inhibition of investment, but as a reflection of the different maturity of the instruments. In Member States where coercive instruments (DRSes) are highly developed, investments in tangible assets may follow different trajectories compared to those relying on market incentives (TPOs) or collaboration (VAs).

The results of the table indicate that circular investment in the EU is sensitive to the architecture of the policy mix; capital flows are embedded in a complex governance model, where specific regulations (DRSes) and market flexibility (TPOs, VAs) form a system of signals correlated with capital allocation decisions.

C.

Model Performance and Validation

By analysing the overall performance of the regression model for INV_CE, we can assess how well the regression model fits the data and how much of the variability in the dependent variable is explained by the included predictors, through the information in

Table 8. The effect of variables: a test of the SS whole model vs. SS residual.

Variable	Multiple		Adjusted	Model			Residual			F	p *
	R	R2	R2	SS	df	MS	SS	df	MS
INV_CE	0.6052	0.3663	0.2511	0.9664	4	0.2416	1.6721	22	0.0760	3.1789	0.033292 **

R—Pearson’s correlation coefficient; SS—sum of squares; df—degrees of freedom; MS—mean of squares; F—F-value; p—p-value; * four significant digits provided. Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1.

.

The resulting model is statistically significant (p = 0.033), showing that the selected mix of circular policies explains approximately 36.6% of the variance in gross investment in tangible goods for the circular economy in EU countries. The F-test value (3.178) and the associated p-value confirm that the associations identified between the regulatory framework and capital flows are not the result of chance, but they reflect systematic patterns of economic governance at the European level.

The choice to include exclusively policy instruments (Deposit Return Schemes, Policy Integration Index, Tradable Permits and Voluntary Approaches) proved to be superior. This suggests that circular investments are directly associated with the architecture of the policy mix rather than with geographical or income classification variables (OECD/non-OECD).

These findings provide strong statistical support for Hypothesis H3. The results illustrate that the success of the transition to a circular economy in the European Union is correlated with a systemic mix of policy instruments. This regulatory “ecosystem” functions as a “gateway”, where a balanced and integrated mix of economic interventions correlates with an increased capacity to attract investment.

4.2.4. Model Optimization and Diagnostic Validation

A.

Collinearity Resolution and PCA Aggregation

Our analysis approached the interdependence of policy variables as a systemic feature of the European regulatory mix rather than as a simple statistical constraint. Initial diagnostics indicated a high correlation between push (TFs) and pull (EBSPs) instruments, which generated instability in the individual estimates. To maintain the mathematical integrity of the model, Principal Component Analysis (PCA) was applied to synthesize these redundant variables into the Policy Integration Index (PII). To validate this approach, we tested alternative models (see Appendix D—

Table A19. Sensitivity analysis: ridge regression model with disaggregated policy predictors (Model A).

Variable	Beta	Std. Err. of Beta	B	Std. Err. of B	t	p-Level
Intercept			0.546779	0.188652	2.89834	0.008337 ***
DRSes	−0.349548	0.171752	−0.070607	0.034693	−2.03518	0.054056
EBSPs	−0.042225	0.184886	−0.000675	0.002957	−0.22839	0.821456
TPOs	0.299966	0.185140	0.045182	0.027887	1.62021	0.119434
VAs	0.264417	0.166986	0.012464	0.007872	1.58347	0.127586

Note: λ = 0.1; R = 0.61; R² = 0.37; Adj. R² = 0.25; F (4, 22) = 3.18; p < 0.03325. Std. Err. = 0.27567. Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1. These models examine the stability of the policy–investment association when fiscal instruments EBSPs are included as independent variables, demonstrating the multicollinearity issues that justify the use of the composite PII index.

and

Table A20. Sensitivity analysis: ridge regression model with disaggregated policy predictors (Model B).

Variable	Beta	Std. Err. of Beta	B	Std. Err. of B	t	p-Level
Intercept			0.482857	0.205818	2.34604	0.028394 **
DRSes	−0.345809	0.168625	−0.069852	0.034061	−2.05076	0.052389
TFs	0.041551	0.195225	0.000618	0.002905	0.21284	0.833413
TPOs	0.305255	0.177639	0.045979	0.026757	1.71840	0.099766
VAs	0.275830	0.188135	0.013002	0.008868	1.46613	0.156761

Note: λ = 0.1; R = 0.61; R² = 0.37; Adj. R² = 0.25; F (4, 22) = 3.18; p < 0.03335. Std. Err. = 0.27571. Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1. These models examine the stability of the policy–investment association when fiscal instruments TFs are included as independent variables, demonstrating the multicollinearity issues that justify the use of the composite PII index.

) by disaggregating the PII into its components (EBSPs and TFs). The results demonstrated that the use of isolated variables leads to statistical instability and the absence of individual significance (where neither TFs nor EBSPs reaches the threshold of p < 0.05). Thus, the use of PII allowed the elimination of distortions caused by high values of the variance inflation factor (VIF), providing a configuration in which the policy mix is correlated with investment indicators in a statistically stable way.

B.

Statistical Robustness of the Optimized Configuration

By focusing the analysis on four key predictors (DRSes, PII, TPOs, VAs) and employing Ridge regression, a significant improvement in the model’s statistical performance was achieved:

• The model reached a significance threshold of p = 0.033, meeting the standard criteria for statistical reliability. This suggests that the identified policy mix configuration shares a significant statistical association with circular investment levels (INV_CE).
• The R² coefficient of 0.366 indicates that over a third of the variance in circular investment across the EU-27 is consistent with this specific policy configuration, highlighting the relevance of the chosen variables within the European circular ecosystem.

C.

Visual and Case-wise Diagnostic Validation

To ensure the reliability of the econometric estimations, a comprehensive diagnostic of residuals and influence patterns was conducted.

Table A4. Predicted and Residual values for the Initial Model. Dependent variable: INV_CE.

Country	Obs. v.	Pred. v.	Res.	Std. P. v.	Std. R.	Std. Err. P.	Mahalanobis Dist.	Del. R.	Cook’s Dist.
BE	1.30	1.10	0.20	1.86	0.97	0.12	8.15	0.31	0.10
BG	0.40	0.33	0.07	−1.04	0.32	0.12	7.94	0.10	0.01
CZ	0.60	0.72	−0.12	0.43	−0.58	0.10	5.34	−0.16	0.02
DK	0.80	0.79	0.01	0.70	0.03	0.12	8.33	0.01	0.00
DE	0.90	0.86	0.04	0.96	0.19	0.10	5.18	0.05	0.00
EE	0.70	1.00	−0.30	1.48	−1.44	0.15	11.78	−0.59	0.49
IE	0.20	0.45	−0.25	−0.59	−1.21	0.14	10.29	−0.44	0.25
EL	0.10	0.02	0.08	−2.23	0.40	0.15	13.11	0.18	0.05
ES	0.50	0.52	−0.02	−0.34	−0.08	0.09	3.61	−0.02	0.00
FR	0.80	0.71	0.09	0.39	0.44	0.07	2.02	0.10	0.00
HR	0.40	0.31	0.09	−1.12	0.43	0.12	8.01	0.14	0.02
IT	0.50	0.54	−0.04	−0.25	−0.20	0.16	14.87	−0.11	0.02
CY	0.20	0.53	−0.33	−0.29	−1.59	0.12	7.60	−0.49	0.23
LV	0.60	0.53	0.07	−0.29	0.34	0.09	3.62	0.09	0.00
LT	0.60	0.61	−0.01	0.02	−0.06	0.06	1.09	−0.01	0.00
LU	1.30	1.26	0.04	2.47	0.19	0.14	10.85	0.07	0.01
HU	0.60	0.63	−0.03	0.09	−0.16	0.07	2.20	−0.04	0.00
MT	0.40	0.20	0.20	−1.54	0.96	0.12	8.15	0.31	0.10
NL	1.10	0.65	0.45	0.15	2.18	0.09	3.71	0.55	0.16
AT	1.10	0.80	0.30	0.72	1.46	0.09	3.85	0.37	0.07
PL	0.60	0.51	0.09	−0.38	0.45	0.11	6.86	0.13	0.02
PT	0.70	0.78	−0.08	0.67	−0.41	0.11	6.42	−0.12	0.01
RO	0.40	0.42	−0.02	−0.70	−0.11	0.12	7.94	−0.04	0.00
SI	0.20	0.51	−0.31	−0.39	−1.47	0.10	4.49	−0.39	0.09
SK	0.50	0.60	−0.10	−0.03	−0.48	0.10	5.68	−0.14	0.01
FI	0.40	0.44	−0.04	−0.63	−0.19	0.11	6.21	−0.06	0.00
SE	0.50	0.57	−0.07	−0.12	−0.36	0.10	4.69	−0.10	0.01
Min	0.10	0.02	−0.33	−2.23	−1.59	0.06	1.09	−0.59	0.00
Max	1.30	1.26	0.45	2.47	2.18	0.16	14.87	0.55	0.49
Mean	0.61	0.61	0.00	0.00	0.00	0.11	6.74	−0.01	0.06
Median	0.60	0.57	−0.01	−0.12	−0.06	0.11	6.42	−0.01	0.01

Obs. v.—Observed value, Pred. v.—Predicted value, Res.—Residual, Std. P. v.—Standard Predicted value, Std. R.—Standard Residual, Std. Err. P.—Standard Error Predicted, Mahalanobis Dist.—Mahalanobis Distance, Del. R.—Deleted Residual, Cook’s Dist.—Cook’s Distance.

and

Table A5. Predicted and Residual values for the Revised Model using PCA Index (PII). Dependent variable: INV_CE.

Country	Obs. v.	Pred. v.	Res.	Std. P. v.	Std. R.	Std. Err. P.	Mahalanobis Dist.	Del. R.	Cook’s Dist.
BE	1.30	1.00	0.30	1.79	1.14	0.15	7.69	0.46	0.14
BG	0.40	0.33	0.07	−1.29	0.27	0.16	7.94	0.11	0.01
CZ	0.60	0.63	−0.03	0.09	−0.10	0.13	4.96	−0.03	0.00
DK	0.80	0.61	0.19	0.00	0.73	0.15	6.82	0.28	0.05
DE	0.90	0.89	0.01	1.33	0.02	0.13	5.13	0.01	0.00
EE	0.70	0.96	−0.26	1.61	−0.97	0.19	11.71	−0.50	0.25
IE	0.20	0.42	−0.22	−0.86	−0.83	0.17	10.25	−0.39	0.13
EL	0.10	0.47	−0.37	−0.62	−1.40	0.12	4.11	−0.46	0.08
ES	0.50	0.52	−0.02	−0.40	−0.08	0.11	3.61	−0.03	0.00
FR	0.80	0.73	0.07	0.57	0.26	0.09	2.00	0.08	0.00
HR	0.40	0.33	0.07	−1.30	0.28	0.16	8.00	0.11	0.01
IT	0.50	0.89	−0.39	1.29	−1.45	0.17	9.72	−0.66	0.36
CY	0.20	0.59	−0.39	−0.07	−1.47	0.15	7.43	−0.58	0.22
LV	0.60	0.45	0.15	−0.72	0.56	0.11	3.36	0.18	0.01
LT	0.60	0.62	−0.02	0.07	−0.09	0.07	1.09	−0.02	0.00
LU	1.30	0.84	0.46	1.08	1.72	0.11	3.25	0.55	0.10
HU	0.60	0.66	−0.06	0.23	−0.21	0.09	2.17	−0.06	0.00
MT	0.40	0.14	0.26	−2.17	0.98	0.16	7.98	0.40	0.11
NL	1.10	0.65	0.45	0.21	1.68	0.11	3.71	0.55	0.11
AT	1.10	0.88	0.22	1.24	0.84	0.11	3.58	0.27	0.03
PL	0.60	0.46	0.14	−0.69	0.53	0.15	6.75	0.20	0.02
PT	0.70	0.88	−0.18	1.24	−0.66	0.14	6.05	−0.24	0.03
RO	0.40	0.42	−0.02	−0.88	−0.06	0.16	7.94	−0.02	0.00
SI	0.20	0.51	−0.31	−0.44	−1.17	0.12	4.48	−0.40	0.07
SK	0.50	0.51	−0.01	−0.43	−0.05	0.13	5.36	−0.02	0.00
FI	0.40	0.44	−0.04	−0.79	−0.14	0.14	6.21	−0.05	0.00
SE	0.50	0.59	−0.09	−0.09	−0.33	0.12	4.68	−0.11	0.01
Min	0.10	0.14	−0.39	−2.17	−1.47	0.07	1.09	−0.66	0.00
Max	1.30	1.00	0.46	1.79	1.72	0.19	11.71	0.55	0.36
Mean	0.61	0.61	0.00	0.00	0.00	0.13	5.78	−0.01	0.06
Median	0.60	0.59	−0.02	−0.07	−0.06	0.13	5.36	−0.02	0.02

Obs. v.—Observed value, Pred. v.—Predicted value, Res.—Residual, Std. P. v.—Standard Predicted value, Std. R.—Standard Residual, Std. Err. P.—Standard Error Predicted, Mahalanobis Dist.—Mahalanobis Distance, Del. R.—Deleted Residual, Cook’s Dist.—Cook’s Distance.

(see Appendix A) present the predicted values, residuals, and influence metrics (Cook’s Distance and Mahalanobis Distance) for both the initial and the revised PCA-based models. These metrics confirm the absence of influential outliers that could disproportionately bias the results.

The structural consistency of the model is further validated by the analysis of residual distribution (Figure 7). The stochastic dispersion of residuals around the zero horizontal axis confirms that the fundamental assumptions of linearity and homoscedasticity are met, reinforcing the stability of the identified associations within the Ridge regression framework.

The Predicted vs. Residual Scores plot for the dependent variable Var7 (INV_CE) supports the following diagnostic conclusions:

• Linearity and Independence: The stochastic dispersion of the points around the zero horizontal axis confirms that the relationship between the policy mix and investment is linear, and the model has correctly extracted the structural signal from the data.
• Homoscedasticity: The distribution of the points remains relatively constant across the spectrum of predicted values, without showing any “funnel” patterns (heteroscedasticity).
• Stability of Estimates: Most observations fall within the 95% confidence interval (delimited by the dotted red lines), demonstrating that the coefficient estimates are robust and not distorted by outliers.

4.2.5. Sensitivity Analysis and Structural Stability Validation (Leave-One-Out)

To assess whether the patterns identified by the Ridge regression are representative of the entire EU-27 sample, a Leave-One-Out (LOO) sensitivity analysis was applied. This procedure tests whether the observed policy–investment nexus is an intrinsic characteristic of the group of states, or whether it is disproportionately influenced by the economic profile of certain members.

A.

Stability of Indicators within the Policy Nexus

To validate the structural stability of the Ridge regression coefficients and to ensure that the findings are not driven by specific influential observations (outliers), a Leave-One-Out (LOO) sensitivity analysis was performed. This procedure involved re-estimating the model by systematically excluding countries that showed higher residual values or specific institutional profiles (Bulgaria, Croatia, Cyprus, Malta, Romania and Cyprus and Malta) (see Appendix B—

Table A6. Ridge regression summary for dependent variable: INV_CE for Full Model (EU_27).

Variable	Beta	Std. Err. of Beta	B	Std. Err. of B	t	p-Level
Intercept			0.525633	0.139716	3.76216	0.001075 ***
DRSes	−0.347738	0.170120	−0.070241	0.034363	−2.04408	0.053098
PII	−0.041592	0.187617	−0.013249	0.059768	−0.22168	0.826606
TPOs	0.302487	0.181397	0.045562	0.027323	1.66754	0.109580
VAs	0.270301	0.175429	0.012742	0.008270	1.54080	0.137628

Note: λ = 0.1; R = 0.61; R² = 0.37; Adj. R² = 0.25; F (4, 22) = 3.18; p < 0.033. Standard error of estimate = 0.276. Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1.

,

Table A7. Ridge regression summary for dependent variable: INV_CE for the Model without Bulgaria.

Variable	Beta	Std. Err. of Beta	B	Std. Err. of B	t	p-Level
Intercept			−0.798085	0.439522	−1.81580	0.083707
DRSes	0.626324	0.269680	0.019685	0.008476	2.32247	0.030337 **
PII	−0.075079	0.262935	−0.000743	0.002601	−0.28554	0.778028
TPOs	−0.347894	0.181414	−0.112622	0.058728	−1.91768	0.068854
VAs	0.336392	0.170943	0.015825	0.008042	1.96786	0.062430

Note: λ = 0.1; R = 0.63; R² = 0.40; Adj. R² = 0.28; F (4, 21) = 3.48; p < 0.02496. Std. Err. = 0.27259. Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1.

,

Table A8. Ridge regression summary for dependent variable: INV_CE for the Model without Croatia.

Variable	Beta	Std. Err. of Beta	B	Std. Err. of B	t	p-Level
Intercept			−0.738133	0.417150	−1.76947	0.091337
DRSes	0.639951	0.275932	0.018939	0.008166	2.31923	0.030545 **
PII	−0.074950	0.269102	−0.000695	0.002497	−0.27852	0.783339
TPOs	−0.372112	0.181481	−0.122179	0.059587	−2.05041	0.053013
VAs	0.330017	0.169392	0.015525	0.007969	1.94825	0.064875

Note: λ = 0.1; R = 0.64; R² = 0.41; Adj. R² = 0.30; F (4, 21) = 3.62; p < 0.02152. Std. Err. = 0.27043. Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1.

,

Table A9. Ridge regression summary for dependent variable: INV_CE for the Model without Cyprus.

Variable	Beta	Std. Err. of Beta	B	Std. Err. of B	t	p-Level
Intercept			−0.719893	0.407053	−1.76855	0.091494
DRSes	0.664230	0.277695	0.019052	0.007965	2.39194	0.026194 **
PII	−0.094475	0.270607	−0.000852	0.002442	−0.34912	0.730475
TPOs	−0.360280	0.183612	−0.111083	0.056612	−1.96218	0.063129
VAs	0.320758	0.170379	0.014714	0.007816	1.88262	0.073682

Note: λ = 0.1; R = 0.64; R² = 0.41; Adj. R² = 0.29; F (4, 21) = 3.58; p < 0.02247. Std. Err. = 0.26430. Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1.

,

Table A10. Ridge regression summary for dependent variable: INV_CE for the Model without Malta.

Variable	Beta	Std. Err. Of Beta	B	Std. Err. of B	t	p-Level
Intercept			−0.733821	0.424180	−1.72997	0.098307
DRSes	0.635401	0.279491	0.018794	0.008267	2.27342	0.033619 **
PII	−0.073502	0.272261	−0.000684	0.002533	−0.26997	0.789819
TPOs	−0.351602	0.185593	−0.111156	0.058674	−1.89449	0.072015
VAs	0.328140	0.172161	0.015437	0.008099	1.90601	0.070429

Note: λ = 0.1; R = 0.63; R² = 0.39; Adj. R² = 0.28; F (4, 21) = 3.41; p < 0.02694. Std. Err. = 0.27372. Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1.

,

Table A11. Ridge regression summary for dependent variable: INV_CE for the Model without Romania.

Variable	Beta	Std. Err. Of Beta	B	Std. Err. of B	t	p-Level
Intercept			−0.857580	0.455570	−1.88243	0.073708
DRSes	0.660862	0.277837	0.020692	0.008699	2.37860	0.026947 **
PII	−0.094220	0.269052	−0.000893	0.002550	−0.35019	0.729683
TPOs	−0.367980	0.185426	−0.116336	0.058622	−1.98452	0.060418
VAs	0.341645	0.171161	0.015992	0.008012	1.99605	0.059060

Note: λ = 0.1; R = 0.64; R² = 0.40; Adj. R² = 0.29; F (4, 21) = 3.55; p < 0.02302. Std. Err. = 0.27141. Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1.

and

Table A12. Ridge regression summary for dependent variable: INV_CE for the Model without Cyprus and Malta.

Variable	Beta	Std. Err. of Beta	B	Std. Err. of B	t	p-Level
Intercept			0.578588	0.552180	1.04783	0.307866
DRSes	−0.433671	0.170604	−0.094027	0.036990	−2.54197	0.019895 **
PII	−0.037855	0.368965	−0.000580	0.005651	−0.10260	0.919356
TPOs	0.020022	0.388599	0.000286	0.005550	0.05152	0.959446
VAs	0.335448	0.183803	0.048818	0.026749	1.82504	0.083759

Note: λ = 0.1; R = 0.64; R² = 0.41; Adj. R² = 0.29; F (4, 21) = 3.58; p < 0.02247. Std. Err. = 0.26430. Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1.

).

The results, summarized in

Table 9. Robustness analysis (Leave-One-Out method).

Variable	Full Model (1)	W/o BG Model (2)	W/o HR Model (3)	W/o CY Model (4)	W/o MT Model (5)	W/o RO Model (6)	W/o CY & MT Model (7)
Intercept	0.526 *** (0.140)	0.540 *** (0.145)	0.537 *** (0.151)	0.569 *** (0.126)	0.535 *** (0.141)	0.516 *** (0.140)	0.579 (0.552)
DRSes	−0.070 ** (0.034)	−0.072 ** (0.034)	−0.072 ** (0.034)	−0.086 *** (0.031)	−0.077 ** (0.037)	−0.069 * (0.035)	−0.094 *** (0.037)
PII	−0.013 (0.06)	−0.022 (0.060)	−0.021 (0.062)	−0.018 (0.051)	−0.017 (0.058)	−0.014 (0.057)	−0.001 (0.006))
TPOs	0.046 * (0.027)	0.043 * (0.028)	0.043 * (0.029)	0.049 * (0.024)	0.044 * (0.027)	0.048 * (0.027)	0.0003 (0.006))
VAs	0.013 * (0.008)	0.013 * (0.008)	0.013 * (0.008)	0.011 * (0.007)	0.014 * (0.008)	0.014 * (0.008)	0.049 * (0.027)
R2	0.37	0.36	0.36	0.46	0.36	0.37	0.46
N	27	26	26	26	26	26	25

Note: The table contains the values of the coefficient B; standard errors are reported in parentheses. Statistical significance: *** (p < 0.01), ** (p < 0.05), * (p < 0.1). Model (1) represents the reference estimate; Models (2)–(7) sequentially exclude non-OECD member states to test for stability.

, illustrate the resilience of the model when countries with distinct institutional profiles or higher residuals are sequentially excluded.

The comparative analysis of the six iterations (Table 9: Model 1–Model 6) confirms a high structural stability of the Ridge model:

• The coefficients associated with the policy instruments (DRSes, PII, TPOs, VAs) retain both their direction of association and their order of magnitude across all iterations. In particular, the DRSes instrument shows remarkable stability, maintaining its statistical significance (p < 0.05 or p < 0.01) in almost all configurations.
• The coefficient of determination (R²) remains relatively stable, hovering between 0.36–0.37. Notably, the exclusion of specific outliers—namely Cyprus (Model 4) and the combined exclusion of Cyprus and Malta (Model 7)—increases the R² to 0.46. This improvement suggests that while the model captures a robust structural signal at the EU-27 level, these specific observations exert a disproportionate influence on the goodness-of-fit, reinforcing the overall stability of the model when these cases are addressed.
• The significance of the model (p = 0.033) remains consistently below the 0.05 threshold in the LOO process, confirming that the identified relationship is not dependent on the specific composition of the sample.

Additionally, Model (7) explores the effect of simultaneously excluding microstates (Cyprus and Malta) to assess whether their weight disproportionately influences the observed policy–investment dynamics. The results indicate a continued stability of the key instruments (DRSes and VAs), with an increase in the explanatory power of the model (R² = 0.46). This analysis confirms that the association between circular policies and investment remains consistent across the EU-27 configuration, indicating that the core patterns are not an artifact of the specific particularities of microstates.

The results of the robustness analysis do not imply a causal directionality but attest to a stable co-evolution of circular capital flows and the institutional environment in EU countries. This structural stability provides a solid empirical basis for supporting the hypothesis that an integrated policy mix is consistently found in countries with superior investment performances in the circular economy (Hypothesis H3).

B.

Transmission Channels and Complementarity in the Policy Mix

Starting from the premise that environmental policies are not adopted in isolation, we investigated the presence of possible multiplicative effects (synergies) by sequentially testing the interaction terms between the main instruments. The results obtained (summarized in

Table 10. Synthesis of interactions between environmental policy instruments.

Interaction Term	Wald Stat.	p-Level	Result
DRSes × PII	0.01	0.919	Not Significant
DRSes × TPOs	1.059	0.303	Not Significant
DRSes × VAs	0.023	0.879	Not Significant
PII × TPOs	0.928	0.335	Not Significant
PII × VAs	0.582	0.445	Not Significant
TPOs × VAs	1.351	0.244	Not Significant

Note: p values > 0.05 indicate that there are no statistically significant multiplicative interactions, confirming the independence of the transmission channels.

) demonstrate that the identified associations are robust and do not depend on direct mathematical interdependencies between the variables.

Given that environmental policies are not adopted in isolation, we investigated whether there are synergistic effects between different environmental policy instruments (e.g., whether subsidies potentiate the effect of taxes). We tested for interactions between instruments (see Appendix C—

Table A13. INV_CE-Test of all effects-interaction between DRS × PII. Distribution: NORMAL. Link function: IDENTITY.

Variable	Degr. of Freedom	Wald Stat.	p-Level
Intercept	1	15.77735	0.000071 ***
DRSes	1	4.81512	0.028211 **
PII	1	0.07562	0.783322
TPOs	1	3.32825	0.068099
VAs	1	3.24543	0.071623
DRSes × PII	1	0.01018	0.919617

Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1.

,

Table A14. INV_CE-Test of all effects-interaction between DRSes × TFs. Distribution: NORMAL. Link function: IDENTITY.

Variable	Degr. of Freedom	Wald Stat.	p-Level
Intercept	1	2.654461	0.103260
DRSes	1	0.019708	0.888356
PII	1	0.000625	0.980047
TPOs	1	3.515441	0.060799
VAs	1	2.727607	0.098627
DRSes × TPOs	1	1.059632	0.303299

,

Table A15. INV_CE-Test of all effects-interaction between DRS × TPO. Distribution: NORMAL. Link function: IDENTITY.

Variable	Degr. of Freedom	Wald Stat.	p-Level
Intercept	1	15.16145	0.000099 ***
DRSes	1	4.16058	0.041375 **
PII	1	0.10289	0.748384
TPOs	1	3.29251	0.069596
VAs	1	1.67895	0.195064
DRSes × VAs	1	0.02313	0.879131

Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1.

,

Table A16. INV_CE-Test of all effects-interaction between DRS × TPO DRS × VA. Distribution: NORMAL. Link function: IDENTITY.

Variable	Degr. of Freedom	Wald Stat.	p-Level
Intercept	1	15.71248	0.000074 ***
DRSes	1	6.01347	0.014197 **
PII	1	0.84250	0.358684
TPOs	1	4.37948	0.036374 **
VAs	1	1.95763	0.161767
PII × TPOs	1	0.92808	0.335363

Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1.

,

Table A17. INV_CE-Test of all effects-interaction between EBSP × TF. Distribution: NORMAL. Link function: IDENTITY.

Variable	Degr. of Freedom	Wald Stat.	p-Level
Intercept	1	10.20698	0.001399 ***
DRSes	1	3.29385	0.069540
PII	1	0.13040	0.718017
TPOs	1	3.94536	0.047001 **
VAs	1	0.06638	0.796677
PII × VAs	1	0.58253	0.445324

Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1.

and

Table A18. INV_CE-Test of all effects-interaction between EBSP × TPO. Distribution: NORMAL. Link function: IDENTITY.

Variable	Degr. of Freedom	Wald Stat.	p-Level
Intercept	1	8.559940	0.003436 ***
DRSes	1	6.021311	0.014134
PII	1	0.050832	0.821622
TPOs	1	4.749454	0.029308 **
VAs	1	3.270259	0.070546
TPOs × VAs	1	1.351751	0.244972

Statistical significance is indicated as follows: *** p < 0.01; ** p < 0.05; * p < 0.1.

), but the results did not indicate statistically significant coefficients (p > 0.05). This is summarized in Table 10 and suggests that, for our sample, the impact of each instrument remains independent. Although the theoretical literature emphasizes the importance of an integrated policy mix, our cross-sectional data show that each component of the legislative framework contributes individually to stimulating investment in the circular economy.

From the analysis of Table 10, the absence of statistical significance for the interaction terms can be observed, which provides the following structural perspective for the validation of the model:

• The fact that the p-level for all interactions varies between 0.24 and 0.91 demonstrates that each component of the policy nexus (regulations through DRSes, integrated effort through PII or market instruments through TPOs and VAs) functions as an independent transmission channel.
• The results suggest a layered policy architecture in the EU-27. Investments in the circular economy are correlated with the individual presence and strength of each instrument, which means that the policy mix provides stable support, not being vulnerable to the weakness of a single intervention mechanism.
• The results of the robustness analysis do not imply a causal directionality but rather attest to a stable co-evolution of circular capital flows and the institutional environment. This structural stability provides a solid empirical basis for supporting Hypothesis H3, confirming that an integrated policy mix is consistently found in countries with superior investment performance.

5. Discussion

The present study aligns with the systemic vision proposed by Ahmadov et al. [75], which places firms in a complex set of interdependencies between the micro, meso, and macro levels. This integrated perspective, previously supported by the research of Falah et al. and Whiting et al. [76,80], receives through our data further empirical support regarding the systemic nature of the circular economy in the European space. Moreover, the results are congruent with the analyses of Vărzaru [88], suggesting that the presence of circular initiatives is closely associated with the configuration of the economic and institutional systems specific to the European Union.

5.1. System Architecture and Institutional Co-Evolution in the EU-27

Beyond theoretical approaches, we highlighted the existence of a multi—layered nexus between governance, innovation, and investment, observed within the sample of the 27 Member States.

• Alignment with H1: Co-occurrence analysis indicates that the “firm” is the hub of this ecosystem. In the European context, microeconomic decisions do not appear to occur in isolation but are correlated with the specific architecture of EU public policies.
• Model Performance: The Ridge regression model (R² = 0.37, p = 0.033) indicates that the selected variables are associated with a significant part of the variance of circular capital flows at the EU level. This explanatory power provides a statistical basis for interpreting the interdependence between environmental policies and investment dynamics within the single market.

5.2. Policy Mix Dynamics: The Nexus Between Constraints and Incentives

While Lau et al. [63] treat fiscal policies in a general way, our study provides a nuanced perspective on the European regulatory framework, using the integrated indicator PII (Policy Integration Index). This approach allowed going beyond simple binary comparisons and observing how different instruments coexist in the institutional landscape of the EU-27, highlighting the following:

• Complementarity over Hierarchy (H2): We analysed the relationship between “push” (regulatory constraints) and “pull” (market-based incentives) mechanisms. Although there are numerical variations between the coefficients, formal testing (Wald test) does not reveal a statistically significant hierarchy that would support the absolute superiority of one category over the other (p > 0.05). Consequently, the results suggest that the stability of the European system is associated not with a single isolated instrument but with the balanced coexistence and functional independence of DRSes, PII, TPOs, and VAs.
• Resilience of the Policy Nexus: The findings indicate that “pull” instruments do not replace “push” regulations but complement them. In the EU-27 context, these instruments may act as independent transmission channels, providing a multi-layered “safety net” for circular investments.
• Stability of DRSes: Within this link, the DRSes variable emerges as a constant pillar. Its statistical significance is maintained in almost all iterations of the sensitivity analysis (p < 0.05), supporting its role as a structural anchor in the European policy mix, regardless of fluctuations in other market-based incentives.

5.3. Institutional Congruence and the Maturity Gap

Bashir et al. [60] have previously suggested that OECD countries show a higher affinity for green technologies. Our research refines this perspective for the EU-27 context by moving from a correlation-based perspective to a structural mapping of transmission channels, through the following observations:

• Complementarity through Independence (H3): To formally address the moderating role of systemic maturity, we implemented a regression analysis including the interaction term VAs × Type (OECD/non-OECD). The results (Wald Stat = 1.006, p = 0.315) indicate that OECD membership does not act as a formal statistical moderator. Instead, the data suggest that policy instruments operate as independent vectors across the EU. This points toward the presence of a layered “safety net”; the individual presence of each instrument supports the link between investments, reflecting the resilience of the system, regardless of a country’s OECD status.
• Institutional Congruence and Absorptive Capacity: While the previously identified Spearman correlation (r_s = 0.09) indicated an alignment between government effectiveness and voluntary approaches (VAs), the lack of a significant multiplicative effect in the regression model indicates institutional congruence rather than direct moderation. In the EU space, the association between complex policies (VAs, TPOs) and investment is a characteristic of economies that have reached a threshold of institutional readiness.
• Micro–Macro Gap: Although the model uses aggregated national data, it reflects the firm-centric systemic architecture identified in our bibliometric analysis. In the Complex Adaptive Systems (CASes) framework, national investment levels (INV_CE) can be interpreted as an emergent property of microeconomic decisions. The policy nexus (DRSes, PII, TPOs, VAs) is linked to independent transmission channels that guide firm-level behaviour, providing an interpretative link between theoretical firm centrality and aggregate investment performance.
• EU-limited Systemic Co-evolution: Overall, the data point toward a stable co-evolution between circular capital flows and the institutional framework of the European Union. This structural stability, supported by Leave-One-Out (LOO) analysis, indicates that high-performing circular investments tend to cluster where there is an integrated policy mix. These findings highlight the specific nature of the European governance model, as the identified nexuses are congruent with the regulatory harmony of the EU-27 and may not necessarily be extrapolated to other institutional landscapes.

5.4. The Nature of the Policy Mix: Additive Complementarity vs. Synergistic Interaction

The absence of statistically significant interaction terms (p > 0.05) in the GLM analysis (see Appendix C) provides a crucial clarification on the structure of the European circular policy mix. While the theoretical narrative supports the idea of a “systemic integration,” econometric evidence suggests that, at present, this integration manifests itself in the form of functional complementarity rather than multiplicative synergy.

Our results, which point to additive complementarity, invite a critical dialogue with the literature that supports the existence of multiplier synergies between policy instruments. For example, seminal studies, such as those by Gunningham and Grabosky [89], Howlett and del Rio [90] or Kern et al. [91], argue that, when designed coherently, taxes (TFs) and subsidies (EBSPs) are associated with a mutual amplification of marginal effects.

The discrepancy observed in our study can be attributed to several contextual and methodological factors:

• Maturation Stage (EU-27 Context): Studies reporting synergies are often based on analyses with an extended time window. Our cross-sectional analysis (2021–2023) captures a phase of “policy layering”, in which instruments are linked to institutional configurations that operate autonomously, prioritizing systemic coverage over cross-instrumental coordination.
• Institutional Specialization: Within the EU-27, instruments are channelled through distinct directives, which limits their complementarity to an “autonomous orchestration”. Thus, what in theory is conceptualized as a multiplicative synergy (A × B), in current administrative practice is reflected as an additive complementarity (A + B + C), where each instrument is associated with a specific market failure, acting as a non-redundant layer without requiring a direct interaction to produce effects.

From a systemic perspective, this observation requires a nuance in the way we conceptualize the “policy mix”:

• Additive Complementarity (A + B + C): The evidence data support an additive architecture, in which circular investments are associated with the simultaneous presence of several independent transmission channels. In this model, instruments such as TPOs, DRSes, and VAs function as non-redundant layers that cover various systemic gaps. The integration is horizontal, ensuring that all segments of the value chain are addressed.
• Absence of Multiplicative Synergy (A × B): The lack of statistical interactions indicates that the observed effectiveness of one instrument is not conditioned by the intensity of another. This configuration, specific to the European landscape, suggests that the systemic value lies in the diversity of the portfolio, which ensures resilience and broad coverage rather than in a synergy that would require a degree of administrative coordination that exceeds the current capacities of Member States.

In conclusion, the “balanced mix” identified here reflects a dynamic equilibrium rather than a static model of marginal efficiency. It is essential to interpret these correlations in the light of a possible two-way feedback loop, and time-lag effects. Investments recorded in 2023 are likely to be correlated with strategic responses to previously implemented regulatory frameworks (2019–2021). Therefore, the observed “stability” is the result of a feedback loop between policy decisions and market responses, unfolding over a time window in which additive complementarity serves as a safety net for the stability of investments in the European circular economy. Future longitudinal studies, using GMM or fixed-effects models, will be needed to isolate these temporal dynamics as more consistent data series become available.

6. Conclusions

The findings of this study should be interpreted as indicative patterns within the EU-27 circular ecosystem rather than definitive causal trajectories. By adopting an approach focused on systems, our analysis suggests that the transition to a circular economy (CE) is an emergent phenomenon, resulting from the dynamic interaction between institutional frameworks and microeconomic niche activations. The results point toward a complex co-evolution where the firm acts as a strategic mediator, translating macro-landscape pressures into structural transformations.

To ensure a clear distinction between empirical evidence and theoretical deduction, our conclusions are structured as follows:

• Empirical Findings: The study provides evidence of a strong association between integrated policy mixes (PII) and circular investments, alongside the identification of the “firm” as the central hub in CE knowledge and co-occurrence networks.
• Theoretical Deductions: Based on these associations, we deduce that the EU-27 transition operates as a complex adaptive system, where “structural coupling” and “policy gateways” facilitate the emergence of green capital. While these mechanisms are not directly observed, they provide a robust theoretical explanation for the patterns identified in the data.

A.

Empirical Evidence: Patterns of Policy Association

The econometric results indicate a significant statistical sensitivity of circular investments to “pull” instruments, particularly Tradable Permits (TPOs, β = 0.53, p < 0.001). Furthermore, the consolidation of EBSPs and TFs into a single Policy Integration Index (PII) maintains model stability (VIF < 2) and shows a consistent positive association with tangible asset accumulation across the EU-27. These data points provide an empirical basis for understanding the current state of capital reallocation in the circular landscape.

B.

Theoretical Interpretation: The Firm as a Systemic Mediator

Drawing from the theoretical framework of the Multi-Level Perspective (MLP), we interpret the high centrality of the “firm” node in our bibliometric analysis as evidence of its role as a systemic mediator. We deduce that the success of the CE transition is contingent upon the alignment between macro-regulatory signals and micro-operational feedback loops. This suggests that the firm does not merely react to policy but actively translates it into physical capital reality.

C.

Systemic Hypotheses: Structural Coupling and Policy Gateways

While not observed directly as a linear cause, the link between Deposit Refund Schemes (DRSes) and investment allows for the theoretical deduction of a “structural coupling” mechanism. We hypothesize that certain regulations act as “policy gateways,” forcing a reorganization of reverse logistics that subsequently necessitates capital inflow. This explanation, while interpretative, provides a coherent logic for the statistical clusters identified in the EU-27 dataset.

D.

Policy Integration and Systemic Robustness

Although the environmental policy mix exhibits high interdependence (reflected in elevated VIF values), this research views such correlation as an inherent feature of integrated policy regimes rather than a statistical redundancy. The use of Principal Component Analysis (PCA) and the Leave-One-Out (LOO) sensitivity check support the stability of these observed patterns, indicating that the model captures the collective signal of policies within a complex regulatory landscape.

Our study indicates that regulations and investments do not operate in isolation, but they appear to be part of a complex adaptive ecosystem. To bridge the investment gap, the findings suggest that officials should focus on the coherence between regulatory rigor and market-based freedom, fostering an environment where green capital formation can emerge from the institutional landscape.

It is important to emphasize that, while the systemic explanations provided, such as policy gateways and structural coupling, are theoretically grounded and congruent with the observed associations, they should be treated as reasonable hypotheses arising from the data rather than conclusively proven causal mechanisms. The observational design of this study serves to illuminate potential pathways within the EU-27 circular system, setting the stage for future longitudinal validation.

7. Limitations and Future Research

The findings presented in this study should be interpreted as a systemic mapping of associations instead of a definitive proof of linear causality. Given the complexity of the circular economy transition, the following limitations delineate the boundaries of our current research design:

• Temporal Dynamics and Co-evolution: The analysis utilizes a cross-sectional window (2021–2023), a timeframe dictated by the recent availability of standardized CE indicators. Consequently, the current data structure precludes the establishment of long-term causal trajectories due to the time-lag effect inherent in environmental regimes. Investments (INV_CE) recorded in 2023 may reflect a delayed systemic response to regulations adopted in previous cycles. Potential endogeneity (the bidirectional feedback between policy and investment) remains a latent factor that future longitudinal studies should address using GMM or fixed-effects models as more data points emerge. Our model captures a “snapshot” of a moving target, where current investment patterns are co-evolving with legacy policies.
• Temporal Dynamics and Bidirectional Feedback: The analysis utilizes a cross-sectional window (2021–2023), which limits the ability to map long-term trajectories. The study should be viewed through the lens of co-evolution; there is a high probability that circular investment patterns and policy frameworks are mutually influencing one another (bidirectional feedback). Furthermore, the time-lag effect is inherent in environmental regimes, as policies often reach their peak impact only after several cycles. Consequently, the current results capture a “snapshot” of an evolving system. Future research should prioritize a longitudinal agenda using GMM (Generalized Method of Moments) or fixed-effects (FE) models, to better observe the temporal co-evolution between investment levels and the regulatory environment. To further mitigate these limitations, Section 5.4 discusses the policy–investment nexus as a dynamic equilibrium, framing our findings as a snapshot of a feedback loop between regulatory decisions and market responses.
• Sample Asymmetry and Institutional Heterogeneity: A structural imbalance exists between OECD (n = 22) and non-OECD (n = 5) Member States. While the Mann–Whitney U test confirmed significant disparities, the non-OECD group represents a census of a specific sub-population rather than a random sample. This nuances the universal generalization of findings and highlights the need for targeted research on emerging circular systems with different “institutional readiness” thresholds.
• Sample Asymmetry and Institutional Heterogeneity: The non-OECD group (n = 5) represents a census of a specific sub-population, not a random sample, which restricts the generalizability of these findings. It is critical to note that the identified “institutional readiness gateway” characterizes internal EU disparities and cannot be extrapolated to non-EU emerging economies. Structural differences between economies, such as Romania and island states, suggest that the “institutional readiness threshold” is not a uniform indicator. Future research should employ fuzzy-set Qualitative Comparative Analysis (fsQCA) to explore how specific combinations of institutional factors are associated with policy outcomes in countries with divergent profiles, thereby avoiding the trap of generalizations that ignore national specificities.
• The Tangible–Intangible Investment Gap: The focus on gross investments in tangible goods introduces a specific measurement boundary. By omitting intangible assets—such as R&D for circular design or digital optimization for reverse logistics, and organizational change, this study likely provides a conservative “lower bound” estimate of the total transition effort. From a systemic perspective, tangible and intangible components are complementary; future research should develop a Comprehensive Circular Investment Index (CCII) to integrate integrating physical, intellectual, and human capital pillars.
• Multicollinearity as Policy Integration: The high correlation between instruments (e.g., EBSPs and TFs) reflects the structural coupling of the EU legislative mix. While this reflects policy coherence, it limits the ability of classical OLS to isolate individual marginal effects. Future studies could employ Partial Least Squares Structural Equation Modelling (PLS-SEM) to better capture these complex, interconnected latent relationships within the system.
• The Micro–Macro Gap and Firm-Level Heterogeneity: Although bibliometric mapping identifies the “firm” as the central systemic node, the econometric analysis utilizes aggregated national investment data (INV_CE). This study cannot fully explain how firm characteristics (such as size, sector, or ownership) are associated with policy effectiveness. It is plausible that SMEs and large corporations exhibit different response patterns to voluntary approaches (VAs). Future research should utilize firm-level panel data (e.g., Eurostat’s Structural Business Statistics) to enable a multilevel analysis that captures how organizational heterogeneity correlates with strategic responses to public policy.
• Information Boundaries: While the bibliometric mapping provides a robust conceptual alignment with the theoretical consensus, it remains bounded by the indexed academic discourse. Relying on Web of Science metadata may potentially underrepresent “grey literature” and emerging industry-specific reports. This gap suggests a need for future qualitative research (e.g., multiple case studies) to explore the socio-technical factors and collaborative governance models that support associated with voluntary approaches (VAs).

8. Practical Implications

The findings of this research suggest a strategic “roadmap” for actors involved in the circular transition, identifying the systemic levers associated with the reallocation of tangible capital.

8.1. For Policy Makers: From Linear Penalties to Circular Alignment

The study points toward the importance of a balanced policy regime that moves beyond the isolated penalization of linear externalities toward a more integrated approach.

• Synergistic Balance: Since TFs and EBSPs exhibit nearly identical statistical weights (β ≈ 0.49), the results support the adoption of a synchronized “Push–Pull” mix. In a systemic framework, taxes on waste could be strategically aligned to fund subsidies for green technologies, potentially creating a self-reinforcing feedback loop.
• Infrastructure Priority: The implementation of Deposit Refund Schemes (DRSes) should be viewed as a foundational infrastructure priority. Beyond collection, DRSes serves as a facilitator for the systemic reorganization of reverse logistics and supply chain transparency.
• Differentiated Regulatory Pathways: For emerging circular systems (non-OECD), where voluntary collaboration (VAs) appears structurally decoupled from performance, mandatory regulations and rigid reporting standards may act as necessary “gateways” before transitioning to sophisticated self-regulation mechanisms.

8.2. For Private Sector Managers: Strategic Integration and Resource Literacy

The analysis indicates that circular success is an emergent property of internal governance rather than isolated actions.

• Internal Governance Alignment: Circular investments appear most effective when integrated into the firm’s core governance mechanisms. Managers are encouraged to align organizational strategies with systemic policy signals to mitigate transition risks.
• Market Literacy: Firms should develop internal capabilities for managing Environmental Certificates (TPOs). As a strong statistical predictor associated with capital attraction in this study, proficiency in tradable permit markets represents a significant competitive advantage.
• Collaborative Ecosystems: The high density of the “supply chain” cluster in our analysis suggests that individual firm performance is deeply interdependent with the flow of know-how and shared logistics infrastructure within the broader business ecosystem.

8.3. For Technology Developers and Consultants: Bridging the “Data–Investment” Gap

There is an indicated systemic demand for digital tools that facilitate Life Cycle Analysis (LCA) and real-time data monitoring. These technologies serve as the vital “connective tissue” between academic research and the investment world, providing the technical traceability associated with transforming environmental policy signals into bankable green tangible assets.