Regulatory quality and the EU circularity gap: forecasting the circular material use rate to 2035

1 Introduction

Water and waste infrastructure has become one of the more revealing test beds for climate policy. Treatment plants, collection systems, sorting facilities and resource recovery units are long-lived assets; they lock in emissions and material flows for decades, and they are financed and permitted through decisions that mix engineering, economics and law. Over the past fifteen years, these decisions have increasingly leaned on a set of structured analytical methods that we group here as evidence-based decision tools. Life cycle assessment estimates the environmental burdens of competing options across their full value chain (Laurent et al., 2014a). Techno-economic assessment weighs those options against costs and revenues. Multi-criteria analysis brings several objectives and several stakeholders into a single comparison (Soltani et al., 2015). More recently, artificial intelligence and data-driven platforms have begun to model system behaviour and to screen investment choices in close to real time (Abdallah et al., 2020; Sun and Scanlon, 2019). These tools promise to replace intuition with transparent, quantified reasoning.

The circular economy gives this promise a concrete target. Instead of extracting, using and discarding materials in a straight line, a circular system keeps water, nutrients and materials in productive use for as long as possible (Kirchherr et al., 2017; Stahel, 2016). In the water sector, this means treating wastewater as a source of water, energy and nutrients rather than a nuisance to be discharged (Voulvoulis, 2018; Kakwani and Kalbar, 2020). In the waste sector, it means designing collection and recovery so that secondary materials re-enter the economy and displace primary extraction, across a family of value retention strategies that runs from reuse and repair through to recycling (Ghisellini et al., 2016; Reike et al., 2018). The uptake of reuse also depends on how circular values are understood and translated into everyday behaviour, not only on technical availability (Remsei et al., 2023). The circular economy overlaps with, but is not identical to, the green-economy and bioeconomy agendas, and clarifying those boundaries has been an active task for the field (D'Amato et al., 2017). Both shifts carry a climate dividend, because primary material production is energy-intensive and because avoided extraction reduces pressure on ecosystems (Geissdoerfer et al., 2017). The European Union has made circularity a headline ambition, most visibly through the 2020 Circular Economy Action Plan and the aspiration to double the circular material use rate by 2030 (European Commission, 2020).

The emphasis on water and waste is because both systems have disproportionate climate and resource implications and are managed by consequential local choices. The gap between a linear and a circular design is substantial and easy to measure: conventional wastewater treatment is energy-intensive and emits methane and nitrous oxide, whereas circular plants can instead recover biogas, heat, reclaimed water, and phosphorus. The same applies to solid waste, as landfilling and open dumping both create methane emissions, and material recovery reduces primary production. The most recent global assessment reports that municipal solid waste reached about 2.56 billion tonnes in 2022 and is expected to rise to roughly 3.86 billion tonnes per year by 2050 under business as usual (Cook et al., 2026). The 2022 level had not been projected to occur before 2030 in the preceding edition (Kaza et al., 2018). These flows are treated plant-by-plant and contract-by-contract, and thus this study asks: Does an evidence-based tool genuinely change what gets built?

There is, however, a disconnect between the sophistication of these tools and the outcomes observed on the ground. Research in policy studies and science and technology studies has shown that evidence is not left unaltered as it passes through a decision process. It is framed, contested, ignored or selectively used, depending on institutional rules, the distribution of power, and the way uncertainty is managed (Head, 2016; Cairney, 2016). A life cycle result for a resource recovery plant will only affect an infrastructure pathway if the legal and administrative framework around the plant allows it: if permits can be granted, if procurement rules allow it and if regulators can hold the operator to account. This is where law enters the story. Whether the steel, concrete and pipework implied by an assessment tool are actually built is likely to depend on the quality of the regulatory environment, defined as the ability of public authorities to design and enforce sound rules.

Although this statement is intuitively plausible, three things have not been explored. First, most research on decision tools is at the methodological level and does not ask whether a decision was made within a given governance context. Second, the empirical literature that has investigated the relationship between institutions and circular outcomes has focused on explaining the past, typically by running panel regressions on historical recycling data, and has been limited in its use of this evidence to project where current trends are heading (Pronti and Zoboli, 2024; Georgescu et al., 2025). Third, the legal dimension of circular infrastructure, such as procurement law, water reuse regulation, extended producer responsibility and the accountability of algorithmic evidence, is typically not part of the quantitative circularity literature, and the two conversations do not often intersect.

This paper connects those two conversations. We take regulatory quality as a central explanatory idea; we treat the European Union circular economy monitoring framework as the decision arena, and we combine forecasting with a cross-country association test. Our research questions are: where is the European Union circular material use rate heading by 2035 on current trends, and how far will that leave it from its own targets? Is regulatory quality associated with how circular a member state economy actually is, once structural differences in material throughput are taken into account? And what do the answers imply for the legal and institutional reforms needed if evidence-based tools are to bite? We are explicit that the design is descriptive and hypothesis-sharpening rather than causally identified: the forecast and the cross-section are two complementary readings of official monitoring data, and the causal weight of the governance mechanism rests on the panel literature we build on rather than on our own cross-section.

The contribution is threefold. The theoretical contribution is to reframe the influence of decision tools as a function of the legal-institutional environment, and to show why a headline circularity metric can mislead a decision-maker who reads it without that context. The empirical contribution is a transparent, fully reproducible forecast of the European Union circularity gap to 2035, grounded entirely in the official Eurostat series, together with an EU-27 cross-sectional test of the regulatory quality link that explicitly separates the governance signal from the structural confound in the indicator. We do not claim the forecasting techniques themselves as a methodological contribution, since they are standard; the value lies in applying them transparently to a policy target and in connecting the literature on legal mechanisms to a quantitative reading of the circularity indicators. The policy contribution is a set of legal and procedural recommendations, framed as a prospective agenda, that connect circular infrastructure to procurement law, water reuse regulation and the governance of algorithmic evidence. Throughout, the argument keeps one foot in law and one foot in statistics, on the view that the circular transition is neither a purely technical nor a purely legal problem but a question of whether good evidence and good rules reinforce each other.

The rest of the paper is organised as follows. Section 2 reviews the literature on decision tools, on evidence in governance, and on regulatory quality, and sets out a conceptual framework with three propositions. Section 3 describes the data and the forecasting and cross-country methods. Section 4 reports the results. Section 5 discusses what they mean for law, policy and practice, and acknowledges the limits of the exercise. Section 6 concludes.

2 Evidence, governance and the legal-institutional environment

2.1 Evidence-based decision tools in circular water and waste infrastructure

The methodological foundations of the tools used in this field are mature. Life cycle assessment has been applied to solid waste systems for more than two decades, and systematic reviews have both catalogued the studies and set out guidance for better practice (Laurent et al., 2014a; Laurent et al., 2014b). The method is now standard in comparing, for example, landfill against incineration against material recovery, and it is increasingly paired with techno-economic assessment so that environmental and financial results sit side by side. Multi-criteria analysis extends the comparison to objectives that do not reduce to a single unit, and reviews of its use in municipal solid waste management show that it is often the vehicle through which multiple stakeholders are brought into a decision (Soltani et al., 2015). Facility-level measurement studies also show how quantified loss data can identify prevention opportunities within production systems (Tóth et al., 2021). In the water sector, circular thinking has generated its own analytical vocabulary, from reduce, reuse and recover frameworks for urban water to indicators of water circularity (Kakwani and Kalbar, 2020; Voulvoulis, 2018).

Techno-economic assessment is the financial complement to life cycle assessment, and in practice, the two are often conducted simultaneously so that a proposed resource recovery unit can be evaluated for both its environmental burden and its cost and revenue profile. This combination is relevant to infrastructure because public authorities do not typically act solely on environmental considerations, and a circular solution that appears preferable in a life cycle assessment will not be chosen if the techno-economic case is not strong enough, or if the income from the secondary outputs is not guaranteed. What unites these methods is that they all produce a comparison, and a comparison only changes an outcome if some actor with authority is required, or at least permitted, to act on it.

Two developments have sharpened the debate. The first is the proliferation of circularity indicators. Reviews have counted dozens of competing metrics and have warned that the choice of indicator is not neutral, because different metrics reward different behaviours and can point decision makers in different directions (Saidani et al., 2019; Corona et al., 2019). EU-focused composite-indicator research likewise shows that circular-economy rankings depend on indicator construction and comparison choices (Kozma et al., 2022). The circular material use rate, which anchors much of the present analysis, is one such metric, and its construction as a ratio of recycled input to total material input has consequences that we return to below. The second development is the arrival of data-driven and algorithmic tools. Machine learning is now used to forecast waste generation, extending a longer tradition of statistical waste-generation modelling (Beigl et al., 2008), as well as to optimise collection and screen water-management options (Abdallah et al., 2020; Sun and Scanlon, 2019). Recent supply-chain research similarly links generative AI with circular design intelligence while emphasising ethical responsiveness (Lukács et al., 2026). These tools are powerful, but they also import new questions about transparency and accountability, because a recommendation produced by an opaque model is harder to contest inside a legal process than one produced by a documented assessment (Tsamados et al., 2022).

2.2 Why evidence does not act neutrally

If the tools are mature, why is progress slow? The answer, developed across evidence-based policy research, is that the relationship between evidence and decision is mediated rather than mechanical. Head (2016) outlines a shift from evidence-based to evidence-informed policy, arguing that the language of evidence-based policy exaggerates the direct connection between research and policy decisions. Cairney (2016) demonstrates how evidence can be constrained by values, institutional routines and political timing, and how the same study can be decisive in one context and irrelevant in another. This is not a counsel of despair. It is a reminder that an analytical tool will be productive only in the arena into which it is introduced.

Transitions research comes to the same conclusion from a different angle. The multi-level perspective conceptualises infrastructure change as a result of the interplay between niche innovations, existing socio-technical regimes and landscape pressures, and emphasises the co-evolution of regulation, user practices and industry networks (Geels, 2002; Markard et al., 2012). A technology that can be technically proven may not scale if the regime in which it operates, its rules and its financing, does not change. Decision science adds that real decision makers are boundedly rational; they will not optimise over all options but instead satisfice within the constraints they face (Simon, 1955). Combined, these literatures suggest that the marginal benefit of an improved assessment tool is low in weak institutional settings and high in strong institutional settings. This is the intuition that our empirical work seeks to illustrate.

2.3 Regulatory quality and the legal-institutional environment

The concept of regulatory quality is well defined and widely measured. In the World Bank’s Worldwide Governance Indicators, Regulatory Quality is defined specifically as perceptions of the ability of the government to formulate and implement sound policies and regulations that permit and promote private sector development (Kaufmann et al., 2011; World Bank, 2024). It is one of six governance dimensions; Rule of Law and Government Effectiveness are separate indicators. We use Regulatory Quality as our primary measure of the legal-institutional environment, and we treat Rule of Law and Government Effectiveness as conceptually related dimensions rather than folding them into the single indicator. We also hold the interpretation at the level of the published construct throughout. Regulatory Quality is a broad, perception-based summary of the design and implementation of policies and regulations that permit and promote private-sector development. It is not a measure of environmental permitting throughput, waste-sector enforcement activity, procurement practice or sectoral administrative capacity, and we do not read it as one at any point in the analysis or the discussion. We also note, with the dataset’s authors, that each governance estimate is published with a standard error and a 90 percent confidence interval and is too coarse to diagnose specific reforms on its own; we therefore read the indicator as a broad proxy and return to its uncertainty in the analysis and limitations.

Theoretically, there are strong reasons why the quality of regulation should matter for circular outcomes. Twenty years of experience have already provided qualified support for the Porter hypothesis, which suggests that environmental regulation can promote innovation and competitiveness, rather than only cost, if it is well designed and not excessively stringent (Porter and van der Linde, 1995; Ambec et al., 2013). National environmental ambition can be studied through the OECD environmental policy stringency index, which shows large and persistent differences in the stringency of national environmental policy (Kruse et al., 2022). Whereas stringency captures how demanding a rule is, regulatory quality is intended to capture how well rules are designed and implemented, although the WGI measure of it is a broad perception-based summary rather than an observation of implementation on the ground. In theory, a demanding recycling target will have limited effect where permitting processes are slow, enforcement is weak and secondary material markets remain underdeveloped; none of those three channels is observed in our data.

The circular economy literature has begun to make this connection explicit. Policy studies show that the European transition depends on active government intervention across regulation, infrastructure and market design rather than on voluntary action alone (Hartley et al., 2020; Milios, 2018). Analyses of European Union circular policy also warn that much of the apparent activity is layered onto pre-existing instruments, so the coherence and enforceability of the framework cannot be taken for granted (Fitch-Roy et al., 2020). Most directly, recent panel work finds that good governance, measured through regulatory quality, government effectiveness and rule of law, is positively related to municipal recycling performance across the European Union, and that institutional trust helps that effect take hold (Pronti and Zoboli, 2024). Complementary panel evidence identifies economic, technological and environmental drivers of circular material use across member states and points to institutional and market conditions as central (Georgescu et al., 2025; Afolabi, 2025).

A distinction is needed between the two contributions of regulation, because they are often confused. Stringency is about how demanding a rule is, and quality is about how well a rule is designed and enforced. The OECD index indicates that, even among high-income countries, the stringency of environmental policy varies considerably and has increased unevenly over time (Kruse et al., 2022). A country can have a very demanding target but not achieve it if regulatory quality is inadequate, because the target is not supported by effective permits, effective enforcement, and effective markets for the recovered material. That is why the Porter mechanism is conditional. A well-designed regulation can stimulate innovation and investment that makes circular options viable, whereas a poorly designed regulation will only increase costs without generating the corresponding benefits (Porter and van der Linde, 1995; Ambec et al., 2013). This motivates our argument: if ambition rises without matching implementation capacity, a widening gap between what is claimed and what is delivered would follow. The forecasts reported below are compatible with such a gap, but they cannot show that this is its cause.

Law enters this picture through several concrete channels. Waste law establishes binding targets and specifies when a material is no longer considered waste, as in the updated Waste Framework Directive (European Union, 2018). In the Regulation on minimum requirements for water reuse (European Union, 2020; Voulvoulis, 2018), water law now establishes minimum quality requirements, thereby converting water reuse from an informal practice into a regulated activity. The drive toward mandatory green and circular procurement is important for infrastructure (Pouikli, 2021; Sönnichsen and Clement, 2020) because it defines whether public buyers can or must favour circular solutions. Extended producer responsibility places the burden for end-of-life management on the producer, and therefore alters the economics assessed by the assessment tools (Milios, 2018). When algorithmic evidence is introduced into permitting and planning, the law of accountability, the explainability and contestability of the model output, becomes part of infrastructure governance (Tsamados et al., 2022). The enforceability of these rules and the availability of credible dispute resolution also influence whether investors will commit to long-lived assets financed through public-private arrangements, thereby linking regulatory quality to the arbitration and contract environment around circular infrastructure.

2.4 Conceptual framework and propositions

The analysis is organised around three propositions, each framed as an association to be examined rather than a causal claim to be tested. P1. Across member states, higher regulatory quality is associated with stronger circular performance, both in the headline ratio once structural differences in material throughput are taken into account and in a numerator-based measure that does not condition on the denominator (examined in Section 4.3). P2. The headline circular material use rate is confounded by economic structure, so that resource-intensive economies can post low circularity scores despite strong institutions; reading the indicator without this structural context can obscure the governance signal, although structure alone does not account for every low-scoring, high-governance case (examined in Sections 4.1 and 4.3). P3. On current trajectories the aggregate circular material use rate will fall far short of the 2030 doubling target, and the size of the gap is too large to be closed by marginal improvements in analytical tools. A shortfall of that size is compatible with an institutional constraint on the transition, although our data observe neither the use of analytical tools nor institutional practice, and therefore cannot identify the operative constraint or rank candidate explanations against one another (examined in Sections 4.2 and 4.6).

3 Materials and methods

3.1 Research design and decision arena

The study uses the European Union circular economy monitoring framework as its decision arena, the formal instrument through which the Union and its member states track circular progress, set targets, and hold themselves accountable. The circular material use rate is the headline indicator of that framework, and it is the metric against which the 2030 doubling ambition is judged (European Commission, 2020). Decisions in this arena are taken by national and subnational authorities that plan, permit, procure and regulate water and waste infrastructure, within rules set at Union level. Anchoring the analysis to this arena lets us ask a governance question rather than a purely technical one, namely whether the institutional strength of a jurisdiction is reflected in its circular results, and where the aggregate is heading. Two clarifications about scope follow directly. First, the circular material use rate is an economy-wide material-flow ratio, not a direct measure of infrastructure investment, permitting or project outcomes; we therefore treat it as an aggregate signal of the arena and do not infer project-level decisions from it. Second, we do not directly observe the adoption or use of individual decision tools; the claim that the constraint is institutional rather than analytical is advanced as an interpretation supported by the forecasts and the cross-section, not as a measured effect of tool use. The design combines two strands. The first is a forecast of the aggregate circular material use rate to 2035. The second is an EU-27 cross-country association analysis. A scenario analysis links the two by asking what an accelerated, illustrative pathway would look like against the target.

3.2 Data and variables

All variables are drawn from published official statistics. Table 1 lists each variable, its definition, source, reference year, publisher vintage, extraction date and coverage, and distinguishes the variables used in estimation from those that provide context. The circular material use rate for the European Union and its member states is taken directly from the official Eurostat series (dataset cei_srm030, source dataset env_ac_cur; data vintage 24 November 2025, extracted 31 July 2026), together with the European Environment Agency indicator that draws on it (Eurostat, 2025; European Environment Agency, 2025). We use the complete reported annual series and do not interpolate or otherwise generate any interior values. Two features of the series require explicit acknowledgement, because they bear directly on how the forecasts should be read. First, the indicator is not built from independently observed annual data throughout. Its waste component is collected under the Waste Statistics Regulation (EC) No 2150/2002, which operates on a two-yearly reporting cycle, and Eurostat states in the reference metadata for cei_srm030 that this source is available only every second year and that Eurostat estimated the data for the missing odd years (Eurostat, 2024; European Union, 2002). The even reference years from 2010 to 2022 therefore rest on reported waste returns, whereas the intervening odd years are gap-filled. Second, the 2024 values for the Union and for every member state carry the Eurostat status flag ‘i’, denoting a value imputed by Eurostat, 2024 is itself a reporting year under the Regulation, but the corresponding return had not been incorporated at the November 2025 vintage. We retain the Eurostat estimates, label them as such throughout, and report in Sections 4.1 and 4.2 how the results behave when the estimated years are removed. Regulatory quality comes from the World Bank Worldwide Governance Indicators, expressed on the standard scale from about −2.5 to about +2.5 (Kaufmann et al., 2011). We use the current release of the database, which incorporates the 2025 methodology revision and extends the estimates to reference year 2024 (World Bank, 2026); series GOV_WGI_RQ. EST was extracted on 31 July 2026 from a database last updated on 18 March 2026. Because that revision changed both the underlying sources and the aggregation, values for several member states differ from those in the earlier release, and we therefore report the cross-country results for reference years 2024 and 2023 side by side. The published standard errors of the estimates (series GOV_WGI_RQ. SE, reference year 2024) are used to gauge how much measurement error the coefficients carry. Domestic material consumption per capita, used both as the structural control and as the basis for the numerator-based outcome, is taken from Eurostat economy-wide material flow accounts (env_ac_mfa, indic_env DMC, material TOTAL, unit tonnes per capita, reference year 2024; vintage 2 July 2026). Gross domestic product per capita on a purchasing power parity basis (World Bank World Development Indicators, series NY. GDP. PCAP. PP. CD, current international dollars, reference year 2024, extracted 31 July 2026) is reported for context only. Environmental policy stringency (OECD Environmental Policy Stringency database, composite index on a 0 to 6 scale, reference years 1990 to 2020; Kruse et al., 2022; extracted 31 July 2026) is likewise contextual and is not entered into the estimation, because it measures ambition rather than the implementation that is our focus. Global municipal solid waste generation and its projections are taken from What a Waste 3.0 (Cook et al., 2026). Municipal recycling rates and the statutory recycling targets are taken from Eurostat (cei_wm011, vintage 30 March 2026) and the revised Waste Framework Directive (European Union, 2018).

Table 1. Variables, definitions, sources, reference years, extraction details and coverage.

3.3 Forecasting methods

We forecast the aggregate circular material use rate from 2025 to 2035 on the official annual series using three complementary models, which lets us judge whether the conclusion is stable across model choices rather than an artefact of one specification. The first is an ordinary least squares linear trend, which provides a transparent central estimate and a prediction interval. The second is Holt linear exponential smoothing in its additive-trend, non-seasonal form. Because the treatment of initial states materially affects the fitted smoothing parameters on a series this short, we state it explicitly: the level is initialised at the first observation and the trend at the first difference, the two smoothing parameters are estimated by minimising the sum of squared one-step-ahead errors from the second observation onwards, and prediction intervals are computed from the analytical error-propagation formula for the additive-error, additive-trend model rather than by simulation. The third is an autoregressive integrated moving average model. For the ARIMA model we searched a candidate set of orders (0,1,0), (1,1,0), (0,1,1), (1,0,0), (1,1,1), (2,1,0) and (0,2,1), with and without a drift term, estimated each by exact Gaussian maximum likelihood through a Kalman filter, and selected the preferred specification using the Akaike information criterion; the random-walk ARIMA(0,1,0) minimised the criterion, and we report the ARIMA(0,1,0) with drift as a sensitivity analysis. For every model we report numerical point forecasts and 95 percent prediction intervals for 2030 and 2035, an in-sample one-step-ahead root mean squared error (for the random walk this is the standard deviation of the annual first differences, which is the corresponding quantity), and, because in-sample fit alone can flatter a short series, a rolling-origin or leave-future-out evaluation in which each model is re-estimated on expanding windows beginning with the first eight observations and scored at horizons of one, two and three years. For every model we also report residual diagnostics, since these bear directly on whether the prediction intervals can be read at face value: the lag-one residual autocorrelation and the residual autocorrelation function, a Ljung-Box portmanteau test of residual independence, a Jarque-Bera test of residual normality, a comparison of residual variance between the first and second halves of the sample, and, where the model takes a regression form, a Breusch-Pagan test for heteroskedasticity. As sensitivity checks we re-estimate the trend on the longer official series available from 2004, on the biennial subsample of Waste Statistics Regulation reference years for which the waste component is reported rather than estimated, and on the series with the Eurostat-imputed 2024 observation removed.

We benchmark the forecasts against the policy target of doubling the rate by 2030. The doubling benchmark is defined as twice the 2020 value, but the exact figure depends on the data and policy vintage, and it is worth being precise about why three numbers circulate. The current European Environment Agency indicator states the target as 22.4 percent, which is twice the 2020 value of 11.2 percent in the current Eurostat series. The Eurostat databrowser states 23.2 percent, which is twice a 2020 value of 11.6 percent and reflects an earlier vintage of the material-flow and waste accounts, before subsequent revisions lowered the 2020 estimate. The Clean Industrial Deal sets a 24 percent key performance indicator, a rounded policy figure rather than a mechanical doubling of any published observation (Eurostat, 2025; European Environment Agency, 2025; European Commission, 2025). The three figures therefore differ because the base year has been revised and because one of them is a policy round number, not because the underlying ambition has changed. That objective is rooted in the 2020 Circular Economy Action Plan and tracked within the 8th Environment Action Programme (European Commission, 2020; European Union, 2022). We adopt the Eurostat figure of 23.2 percent as the primary benchmark and report the full 22.4 to 24 percent band as a sensitivity, because the substantive conclusion does not depend on which vintage is used. From the target and the latest observed value, we compute the annual increase required to reach it and compare it with the increase observed. Using the predictive distribution of the fitted trend model, we report how far each benchmark lies from the 2030 central forecast in units of the prediction standard error, and whether it falls inside or outside the model’s 95 percent prediction interval. We deliberately report this as a conditional statement about the fitted model rather than as a probability that the policy target will be attained. Finally, we define two trajectories to 2035. An accelerated policy scenario assumes the annual gain triples relative to the historical average; we label this as a simple illustrative arithmetic scenario, not a governance-estimated projection. A target pathway rises linearly to the benchmark by 2030. The distance between the central forecast and these trajectories is what we call the circularity gap.

3.4 Cross-country association

The cross-country analysis tests whether regulatory quality is reflected in circular performance across all 27 member states, using the 2024 circular material use rate (Eurostat, including the values it flags as estimated). The main explanatory variable is regulatory quality. Two outcome measures are used, and the reason for the second is methodological rather than presentational. The circular material use rate is a ratio whose denominator is overall material use, that is, domestic material consumption plus circular use. Controlling for domestic material consumption per capita is therefore a defensible way of removing a structural confound, but it is not innocent: the control enters the construction of the dependent variable, so part of any resulting association is arithmetic rather than behavioural. To separate the two, we also construct a numerator-based outcome. Because the published rate r and published domestic material consumption per capita D are both available, the numerator of the indicator can be recovered exactly as circular material use per capita = D × r / (1 − r), which is the quantity of secondary material fed back into the economy for each inhabitant. Regressing this quantity on regulatory quality involves no conditioning on the denominator at all, so a positive association there cannot be mechanical. Because circular material use per capita is a strictly positive quantity with a long right tail, running from 0.38 tonnes per inhabitant in Romania to 6.52 in Estonia, we designate the log-linear regression of that quantity on regulatory quality as the single primary numerator specification. We report its coefficient, HC3 standard error, 95 percent confidence interval and p-value consistently wherever the numerator result is stated, and we treat the level regression, the rank correlation, the Theil-Sen slope, the leave-one-out and influence analyses and the throughput-controlled model as sensitivity checks around it. That choice also matches the scale on which Figure 1b, is drawn and fitted, so the plotted relationship and the primary estimate are the same object. We report, for both outcomes and for the reference year 2024 and reference year 2023 vintages of the governance data, Pearson and Spearman correlations, a simple regression on regulatory quality, a multiple regression adding the log of domestic material consumption per capita (and, for comparison, the log of income), and the partial correlation of regulatory quality with circular performance controlling for material intensity. All regression inference is reported with HC3 heteroskedasticity-consistent standard errors alongside the classical ones, because a 27-observation cross-section with a long right tail in material throughput is precisely the setting in which the two can diverge. Every coefficient is reported with its 95 percent confidence interval alongside its standard error, and classical and HC3 inference are reported side by side wherever the two diverge. Influence is assessed with leverage, externally studentised (studentised deleted) residuals, Cook’s distance and DFBETAS for the regulatory quality coefficient, against the conventional cut-offs of 2 k/n, 4/n and 2/√n respectively, together with leave-one-out re-estimation of that coefficient and a Theil-Sen slope over the full sample. Because the governance estimates are published with standard errors, we also report the implied reliability ratio and the corresponding attenuation-corrected slope. We treat these estimates as a descriptive cross-section that complements, and does not substitute for, the larger panel evidence in the literature, on which the causal reading rests (Pronti and Zoboli, 2024; Georgescu et al., 2025).

3.5 Reproducibility

Analysis was performed in Python with standard scientific libraries. Table 1 records, for every variable, the exact dataset identifier, the dimension filters applied, the publisher’s data vintage and the date of extraction, so that each series can be retrieved as we retrieved it. Every input to the analysis is a published official series that any reader can retrieve directly from its publisher, and we have designed the reporting so that the paper is self-contained: Table 1 gives, for each variable, the dataset identifier, the dimension filters applied, the publisher’s vintage and our extraction date, which together form a complete retrieval recipe, while Table 2 prints the full forecasting sample and Table 3 the full cross-sectional sample. Every statistic reported in Sections 4.1 to 4.6 can therefore be recomputed from the printed tables without recourse to the authors. In addition, we have assembled the raw Eurostat and World Bank extracts exactly as downloaded, their query strings, a checksum manifest, the transformation and estimation code, and a single script that regenerates every number, figure and table in the paper. That package is available from the corresponding author on reasonable request, and we will deposit it in a public repository under a persistent identifier upon acceptance if the journal requires it. The data availability statement records exactly this position and no other. The approach deliberately prioritises transparency over complexity because, as with algorithmic evidence (Tsamados et al., 2022), a forecast whose assumptions are visible is more useful to a decision arena than a black box that is harder to contest.

Table 2. European Union circular material use rate, 2010 to 2024, and annual change.

All values are as reported in the official Eurostat series (cei_srm030 / env_ac_cur, vintage 24 November 2025, extracted 31 July 2026). Values marked (est.) are Eurostat estimates: the odd years are gap-filled because the underlying waste statistics are reported only every second year, and the 2024 value carries the Eurostat imputation flag. No values were generated or interpolated by the authors.

Table 3. Cross-country dataset for all EU-27, 2024.

Circular material use rate from Eurostat (cei_srm030; all 2024 values flagged by Eurostat as imputed); circular material use per capita derived as DMC per capita × r / (1 − r); regulatory quality from the World Bank Worldwide Governance Indicators, current release incorporating the 2025 methodology revision, reference years 2024 and 2023 (database last updated 18 March 2026); domestic material consumption per capita from Eurostat (env_ac_mfa, vintage 2 July 2026); GDP per capita, PPP from the World Bank/IMF (rounded). All series extracted 31 July 2026.

4 Results

4.1 The historical trajectory of circular material use

The starting point is a sobering one. The European Union circular material use rate is presented in Table 2 and Figure 2 from 2010 to 2024, as reported by Eurostat. Over these fourteen years the rate rose from 10.7 percent to 12.2 percent, a net increase of 1.5 percentage points, of which about 1.0 point has come since 2015. The average annual increase is 0.11 percentage points. The fitted linear trend is 0.085 percentage points a year (standard error 0.019; 95 percent confidence interval 0.043 to 0.127), statistically different from zero at conventional levels (p < 0.001), and accounts for about 60 percent of the variation in the series; the economic magnitude is very small. On the official series the path is visibly non-monotonic, with dips in 2011 and 2019 and the largest single-year increment in over a decade (0.7 point) between 2022 and 2023, so the underlying signal is a slow, noisy drift rather than the smooth climb that a fitted line alone might suggest. One qualification applies to the whole series and is easy to overlook. Because the waste component is reported only every second year, the odd-year observations are Eurostat estimates rather than independent measurements, and the 2024 figure is flagged as imputed for the Union and for every member state. The fifteen annual points are therefore not fifteen independent draws, and both the apparent smoothness of the series and the nominal precision of any interval estimated on it are to that extent flattering. Restricting the series to the seven biennial reporting years yields a slower fitted trend, 0.054 percentage points a year rather than 0.085 (95 percent confidence interval 0.003 to 0.104; p = 0.042), and we carry that subsample through the forecasts in Section 4.2.

This slow aggregate motion obscures two structural points that matter for the rest of the analysis. The first is that the indicator is a ratio: because its denominator is overall material use, a given rate can rise either because more secondary material is fed back in or because primary consumption falls, and the published aggregate does not by itself let us separate the two. We therefore avoid attributing the historical gains to any single cause and treat the decomposition as a question for the member-state accounts. The second is that the aggregate masks high dispersion across member states, which the cross-country section quantifies.

4.2 Forecasts to 2035 and the circularity gap

The forecasts are the core quantitative result. Table 4 and Figure 3 summarise them, and for each model we report the full set of quantities needed to reproduce it. On the official series the linear trend places the rate at about 12.4 percent in 2030 and 12.8 percent in 2035, with 95 percent prediction intervals of 11.5 to 13.3 and 11.8 to 13.8 respectively, and an in-sample one-step root mean squared error of 0.30 percentage points. The random-walk ARIMA(0,1,0) benchmark, which minimised the Akaike information criterion across the candidate set (AIC 13.55, against 14.29 for the same model with drift and 15.51 or higher for every other candidate), forecasts an essentially flat path at 12.2 percent, with intervals of 10.5 to 13.9 in 2030 and 9.8 to 14.6 in 2035; its one-step error, measured as the standard deviation of the annual differences, is 0.36 percentage points. Adding a drift term of 0.107 percentage points a year raises the path to 12.8 percent in 2030 and 13.4 percent in 2035, with intervals of 10.8 to 14.9 and 10.2 to 16.5. Holt exponential smoothing, estimated with the initialisation set out in Section 3.3, converges to smoothing parameters of 1.00 and 0.37, weights the recent uptick heavily, and is the most optimistic of the three: it reaches 13.6 percent in 2030 and 14.7 percent in 2035, with a one-step root mean squared error of 0.45 percentage points, but its prediction intervals are correspondingly wide, at 9.2 to 17.9 and 5.8 to 23.6. Re-estimating the trend on the longer official series available from 2004 raises the central path to 13.3 percent in 2030 and 14.2 percent in 2035. A rolling-origin evaluation, re-estimating each model on expanding windows, gives one-year-ahead root mean squared errors of 0.35 percentage points for the random walk, 0.38 for Holt and 0.46 for the trend, rising to 0.70, 0.93 and 0.66, respectively, at three years. The models are therefore neither over-fitted nor spuriously precise, and no specification dominates at every horizon.

Table 4. Forecasts of the European Union circular material use rate to 2030 and 2035, with 95% prediction intervals, in-sample one-step errors, Akaike information criterion where applicable, and the robustness samples that exclude Eurostat-estimated years.

Residual diagnostics are reported here in full, because they bear on whether the prediction intervals above can be read at face value. For the linear trend the lag-one residual autocorrelation is +0.33 and the Ljung-Box statistic on five lags is 7.46 (p = 0.19); the residuals are consistent with normality (Jarque-Bera 1.82, p = 0.40) and show no detectable heteroskedasticity (Breusch-Pagan p = 0.61; second-half to first-half variance ratio 1.72, p = 0.53). For the random walk and its drift variant, whose residuals are the annual first differences, the lag-one autocorrelation is −0.17 and the Ljung-Box statistic is 1.47 (p = 0.92 and p = 0.83 respectively), with Jarque-Bera p = 0.98 and a variance ratio of 0.88 (p = 0.88). For Holt the lag-one autocorrelation is −0.06 and the Ljung-Box statistic is 0.77 (p = 0.86), with a mild positive skew traceable to the 2022-to-2023 increment (Jarque-Bera 5.08, p = 0.079) and a variance ratio of 0.67 (p = 0.65). On the biennial subsample the trend residuals give a lag-one autocorrelation of +0.10, a Ljung-Box statistic of 2.65 on three lags (p = 0.45) and a Breusch-Pagan p of 0.59. No diagnostic rejects at conventional levels. The one feature worth naming is the positive lag-one autocorrelation in the trend residuals: it is not significant on fifteen observations, but it is what one would expect of a series whose odd years are interpolated between reported ones, and it is a further reason to treat the trend model’s interval as the most optimistic of those we report and to weigh the difference-based models and the biennial subsample alongside it.

The purpose of using three models is to stress-test the conclusion rather than to embellish it. The trend and smoothing models fit a slow, noisy series; the random walk deliberately imposes less structure and, by treating next year’s best guess as this year’s value, yields a flat projection with an interval that widens with the horizon. The important observation is that even the most optimistic of the three, Holt at about 14.7 percent by 2035, sits far below the doubling benchmark, and that the 2030 benchmark lies outside the 95 percent prediction interval of every model at the 2030 horizon, the highest upper bound being 17.9 percent. One qualification belongs here. At the 2035 horizon the Holt interval is wide enough that its upper edge, 23.6 percent, does reach into the benchmark band. That reflects the very large uncertainty attached to a smoothing model with a near-unit level parameter extrapolated eleven years from a fifteen-point series; it is a reason to read the Holt path with caution, not a route to the 2030 target, which the same model still misses by a wide margin. Restricting the estimation to the biennial reporting years, where the waste component is reported rather than estimated, moves the central forecast down rather than up, to 12.0 percent in 2030 and 12.2 percent in 2035; dropping the imputed 2024 observation altogether gives 12.2 and 12.6 percent. The central conclusion is therefore stable across model specifications and across the treatment of the estimated years: the highest 2030 central forecast produced by any model we estimated on the official series, 13.6 percent under Holt, still sits more than eight percentage points below the lowest published benchmark.

Against the doubling benchmark, all forecasts imply a very large shortfall. Taking the Eurostat figure of 23.2 percent, the central forecast leaves a gap of about 10.8 percentage points in 2030, close to the entire distance the Union has travelled since the indicator began. Reaching the target from the 2024 value would require the rate to rise by about 1.83 percentage points every year to 2030, against an observed average of 0.11 points, roughly seventeen times the historical pace; the corresponding requirements are 1.70 points a year for the 22.4 percent benchmark and 1.97 points for the 24 percent benchmark, about sixteen and eighteen times the observed pace. We report the size of this shortfall as a distance rather than as a probability, because a fifteen-point, partly estimated series will not support a probability statement of that precision. Under the fitted trend model the 2030 central forecast is 12.37 percent with a 95 percent prediction interval of 11.47 to 13.28, so the 22.4 percent benchmark lies about 24 prediction standard errors above the central forecast, the 23.2 percent benchmark about 26 and the 24.0 percent benchmark about 28; the entire published benchmark band therefore lies far outside the interval. The same holds when the model is estimated on the biennial reporting years alone, where the 2030 interval is 11.04 to 12.86. These are conditional statements about a fitted model on a fifteen-point, partly estimated series, not probabilities of policy-target attainment, and we no longer attach a numerical probability to the latter. The accelerated illustrative scenario, which assumes the annual gain triples, still only reaches about 14.1 percent by 2030 and 15.7 percent by 2035. On every extrapolation of current behaviour that we examined, the Union misses its circular material use target by a wide margin, and an optimistic acceleration does not close the gap by 2030.

Required annual increase from the 2024 value to reach the 23.2% benchmark: 1.83 pp. per year (1.70 pp. for 22.4%; 1.97 pp. for 24.0%). Observed average annual increase, 2010–2024: 0.11 pp. per year, about seventeen times slower. Distance of each benchmark from the 2030 central forecast under the fitted trend model, in prediction standard errors: about 24 (22.4%), 26 (23.2%) and 28 (24.0%); the whole benchmark band lies outside the 95% prediction interval on the full series (11.47–13.28) and on the biennial reporting years alone (11.04–12.86). These are conditional statements about the fitted model, not probabilities of target attainment. Central-forecast gap to the 23.2% benchmark in 2030: about 10.8 percentage points. Akaike information criterion across the ARIMA candidate set: (0,1,0) 13.55; (0,1,0) with drift 14.29; (0,1,1) 15.51; (1,1,0) 15.52; (0,2,1) 17.16; (2,1,0) 17.32; (1,1,1) 17.39; (1,0,0) 23.53. Rolling-origin root mean squared error at horizons of one, two and three years: random walk 0.35, 0.58, 0.70 pp.; Holt 0.38, 0.66, 0.93 pp.; OLS trend 0.46, 0.61, 0.66 pp. Residual diagnostics (see Section 4.2): lag-1 residual autocorrelation +0.33 (OLS trend), −0.17 (random walk and drift variant), −0.06 (Holt), +0.10 (biennial trend); Ljung-Box p = 0.19, 0.92, 0.83, 0.86 and 0.45 respectively; Jarque-Bera p = 0.40, 0.98, 0.98, 0.08 and 0.74; second-half to first-half residual variance ratio p = 0.53, 0.88, 0.88, 0.65 and 0.28; Breusch-Pagan p = 0.61 (OLS trend) and 0.59 (biennial trend). No diagnostic rejects at the 5% level.

4.3 Regulatory quality and circular performance across member states

The cross-country results, in Table 3 and Figure 1, cover all 27 member states and tell a subtle story that, read correctly, qualifies as well as supports the governance argument. Circular material use in 2024 ranges enormously, from 32.7 percent in the Netherlands and above 20 percent in Belgium, Italy and Estonia down to 2.0 percent in Finland and Ireland and 1.3 percent in Romania. Regulatory quality on the current release also varies widely, from about 1.96 in Finland to about 0.29 in Bulgaria, so there is ample governance variation to work with. Moving to the current Worldwide Governance Indicators release changes several member-state scores relative to earlier vintages, because the 2025 revision altered both the underlying sources and the aggregation. The reference year 2023 and reference year 2024 estimates within that release are themselves very close, correlating at 0.99 with a mean absolute difference of 0.05, and we report results on both throughout.

Taken at face value, the simple correlation between regulatory quality and the headline circular material use rate across all 27 states is positive but weak and not statistically significant (Pearson r = 0.22, p = 0.27; Spearman 0.20; on the 2023 estimates, r = 0.25, p = 0.20). Part of the reason lies in the structure of the indicator. The denominator is overall material use, so economies with exceptionally high material throughput are mechanically pushed down the ranking: domestic material consumption per capita is about 38.8 tonnes in Finland and 22.9 tonnes in Ireland against an EU average near 13.3 tonnes, and the correlation between the headline rate and log material intensity is −0.56 (p = 0.002). It is this material throughput, not gross domestic product as such, that depresses the score: income is essentially uncorrelated with the headline rate (r = 0.08) and does not enter the indicator’s denominator, whereas material consumption does. Adding the log of domestic material consumption per capita to the regression gives a positive and significant coefficient on regulatory quality of 6.49 percentage points of circular material use per unit (HC3 standard error 2.25; HC3 95 percent confidence interval 1.84 to 11.14; robust p = 0.008; classical interval 2.21 to 10.77, classical p = 0.005), alongside a strong negative coefficient on material intensity of −13.58 (HC3 95 percent confidence interval −21.28 to −5.87; robust p = 0.001), and explains about 51 percent of the cross-country variance. The partial correlation of regulatory quality with the headline rate, holding material intensity fixed, is 0.54 (p = 0.005). On the 2023 governance estimates the corresponding figures are 6.9 percentage points and 0.56.

That specification cannot carry the interpretive weight on its own. Domestic material consumption is a component of the denominator of the dependent variable, so conditioning on it removes a genuine structural confound but also introduces an arithmetic relationship between regressor and regressand; a coefficient obtained this way is in part a property of the indicator’s construction rather than of behaviour. We therefore re-run the test on the numerator, circular material use per capita, recovered exactly from the published rate and published material consumption as set out in Section 3.4. This measure involves no conditioning on the denominator, so a positive association here cannot be mechanical. Our primary numerator specification, designated in Section 3.4, is the log-linear regression of circular material use per capita on regulatory quality; it is chosen because the outcome is strictly positive and strongly right-skewed, running from 0.38 tonnes per inhabitant in Romania to 6.52 in Estonia, and because it is the scale on which Figure 1b, is drawn and fitted. In that specification the slope is +0.49 log points per unit of regulatory quality, with an HC3 standard error of 0.27, an HC3 95 percent confidence interval of −0.06 to 1.05 and a robust p-value of 0.079. Under classical inference the same slope carries an interval of 0.00 to 0.98 and a p-value of 0.048, and the model explains about 15 percent of the variance. We state the implication plainly rather than reporting whichever variance estimator is the more favourable: on the primary specification the bivariate numerator association is positive but modest, it is not robust to the choice of standard error, and its robust confidence interval includes zero. It should be read as exploratory, and Sections 5 and 6 are framed at that strength rather than at the strength of the most favourable specification below.

The sensitivity checks all run in the same direction, which is why we describe the association as modest rather than absent. The Pearson correlation on the logged measure is 0.38 (p = 0.048) and the Spearman correlation on the levels is 0.42 (p = 0.030). In levels the slope is 0.94 tonnes per capita per unit of regulatory quality, with an HC3 95 percent confidence interval of 0.03 to 1.86 and a robust p-value of 0.044, which excludes zero only narrowly and is itself an illustration of the sensitivity to scale that the primary specification makes explicit. A Theil-Sen slope over all 27 states is positive at 0.58, and the leave-one-out slope stays between 0.40 and 0.65 without the sign ever changing. Correcting for the measurement error implied by the published governance standard errors, which give a reliability ratio between 0.83 and 0.88 across the range of standard errors published for this sample, would raise rather than lower the estimate, to between 0.56 and 0.59. Adding log material intensity raises the regulatory quality coefficient to 0.66 log points, with an HC3 95 percent confidence interval of 0.22 to 1.11 and a robust p-value of 0.005, so the association is clearer once structural throughput is held constant than it is in the raw bivariate comparison. Results on the 2023 vintage are slightly stronger throughout (r = 0.42, p = 0.028), so the conclusion does not depend on which reference year is used.

The two cases that motivated the structural argument require closer examination, because material throughput does not fully account for them. In the primary numerator specification Finland and Ireland are the two most influential observations, with Cook’s distances of 0.24 and 0.33 against a 4/n cut-off of 0.15, externally studentised residuals of −1.75 and −2.60, and DFBETAS on the regulatory quality coefficient of −0.58 and −0.70 against a 2/√n cut-off of 0.39; both sit far below their predicted values. In the level specification the most influential observation is instead Estonia, at a Cook’s distance of 0.37, which is a further reason for not treating that specification as the primary one. The comparison that makes the point most sharply is Ireland against Sweden. The two have almost identical material throughput, 22.85 and 22.81 tonnes per capita, and comparable governance scores, 1.79 and 1.67, yet Sweden feeds back 2.65 tonnes of secondary material per inhabitant and Ireland 0.47, a factor of about 5.6. Denmark, at 23.97 tonnes per capita, returns 2.49 tonnes; Estonia, at 25.28 tonnes per capita and a lower governance score than either, returns 6.52 tonnes, the highest in the Union. High material throughput therefore explains why Finland and Ireland score low on the ratio, but it does not explain why they also lie low on the numerator, and describing their position simply as a statistical artefact would be premature. Something specific to the composition of these two economies is at work, plausibly the weight of forestry and mineral extraction in Finland and of a small, import-dependent materials base alongside a large multinational sector in Ireland, and we treat this as an open question for the member-state accounts rather than a settled point. What we can say is that the association does not depend on them: excluding both, the correlation of regulatory quality with the headline rate rises to 0.43 (p = 0.03) and with circular material use per capita to 0.63 (p = 0.001). In the ratio specification, by contrast, the most influential single observation is the Netherlands (Cook’s distance 0.41, DFBETAS 0.77), and removing it lowers the regulatory quality coefficient from 6.5 to 4.8 without changing its sign or significance.

The practical lesson is a narrower one than the headline metric first suggests. A decision maker who ranks Finland against Italy on the headline rate alone might conclude that institutions do not help circularity, or even hinder it. That conclusion would not be safe, because the metric is not neutral: its denominator carries the structure of an economy as much as the performance of its recycling system. Once that structure is accounted for, and once the numerator is examined on its own, regulatory quality tracks circular performance across the 27 states consistently in sign, though modestly in size, at the margin of conventional significance on the primary specification, and with identifiable exceptions. Evidence, in other words, has to be read within its institutional and structural context before it can inform a decision, and the reading has to be honest about how much of the pattern that context explains.

Association results (all EU-27, n = 27; WGI reference year 2024, with reference year 2023 in parentheses). Headline ratio outcome: Pearson r (regulatory quality, circular material use rate) = 0.220, p = 0.135 for the one-sided reading in the positive direction and p = 0.270 two-sided (0.255, p = 0.199); Spearman = 0.206 (0.252). Correlation of the rate with log domestic material consumption per capita = −0.56 (p = 0.002); with GDP per capita = 0.08 (n.s.). Multiple regression of the rate on regulatory quality and log DMC per capita: regulatory quality +6.49 pp. per unit, HC3 SE 2.25, HC3 95% CI 1.84 to 11.14, robust p = 0.008; classical SE 2.07, classical 95% CI 2.21 to 10.77, classical p = 0.005 (+6.89, robust p = 0.006); log DMC per capita −13.58, HC3 95% CI -21.28 to −5.87, robust p = 0.001; R-squared = 0.52. Partial correlation of regulatory quality with the rate, controlling for material intensity = 0.54, p = 0.005 (0.56, p = 0.003). Numerator outcome (circular material use per capita, tonnes). PRIMARY SPECIFICATION, logged bivariate regression: slope +0.493, HC3 SE 0.269, HC3 95% CI -0.061 to 1.046, robust p = 0.079; classical SE 0.237, classical 95% CI 0.005 to 0.981, classical p = 0.048; R-squared = 0.147. Sensitivity checks: Pearson r on the logged measure = 0.384, p = 0.048 (0.422, p = 0.028); Spearman on levels = 0.418, p = 0.030; slope in levels +0.943, HC3 SE 0.444, HC3 95% CI 0.030 to 1.857, robust p = 0.044; with log DMC per capita added to the logged specification, +0.662, HC3 95% CI 0.219 to 1.106, robust p = 0.005. Excluding the two high-throughput, high-governance economies (Finland, Ireland): Pearson = 0.43 on the rate (p = 0.03) and 0.63 on log circular use per capita (p = 0.001). Theil-Sen slope (all 27): +2.68 on the rate; +0.58 on log circular use per capita. Reliability ratio of regulatory quality implied by the published WGI standard errors (0.196 to 0.232), given a sample variance of 0.310: 0.83 to 0.88, giving an attenuation-corrected primary slope of 0.56 to 0.59.

4.4 Municipal recycling and the statutory target gap

The circular material use rate is a general indicator of secondary material use across the economy. Municipal recycling, by contrast, is a more limited and more directly regulated outcome, and because it is tied to specific waste law it offers a second lens on the same question. In 2024, the most recent year reported, the EU municipal recycling rate stood at about 48.1 percent, itself an imputed EU aggregate, against 48.0 percent in 2023 at the same 30 March 2026 vintage (an earlier vintage gave 47.9 percent, and we have aligned the figure to the current release) and the revised Waste Framework Directive targets of 55 percent by 2025, 60 percent by 2030 and 65 percent by 2035 (European Union, 2018; Eurostat, 2026). The 2025 deadline has now passed, and on this trajectory the Union was not on course to meet it: the 2023 early-warning assessment identified 18 member states as at risk of missing the 55 percent target (European Environment Agency, 2023). Dispersion across states is very large, with rates running from around two-thirds of municipal waste in the leading economies to little more than a tenth in the laggards. We describe this dispersion rather than model it, because at the March 2026 vintage seven member states had not yet reported a 2024 figure, so a country-level test on this indicator would rest on an incomplete and non-random subsample. The descriptive pattern is the same divide seen in the circular material use data, and the ordering is broadly consistent with the governance gradient, with higher regulatory quality economies grouped toward the top and lower ones toward the bottom. We describe that ordering rather than test it. Because municipal recycling is less exposed than the circular material use rate to the material-throughput confound and yet shows the same ordering, it is consistent with the central argument, and it aligns with the panel evidence of a positive effect of regulatory quality on recycling performance at the Union level (Pronti and Zoboli, 2024). The pattern is compatible with an account in which statutory recycling targets are missed for reasons other than the availability of technology. We do not observe technology adoption, deployment capacity or enforcement in any member state, however, and the descriptive rates cannot distinguish between a technological, a capacity-based and an enforcement-based explanation.

4.5 Waste infrastructure demand to mid-century

This section is context rather than a test of the research questions, and we now state it briefly. What a Waste 3.0 reports global municipal solid waste of about 2.56 billion tonnes in 2022, a level the preceding edition had not expected before 2030, and projects roughly 3.86 billion tonnes a year by 2050 under business as usual, with some US$556 billion of investment estimated to be needed in low- and lower-middle-income countries over that period (Cook et al., 2026; Kaza et al., 2018). Values interpolated between those two published points at the implied growth rate of about 1.5 percent a year are given in Table 5 (B), and Figure 4. These are the flows circular infrastructure exists to recover, so a stalling circularity rate against rising volumes implies a growing absolute loss of material from the economy.

Table 5. Scenario summary and global waste demand.

(A) Gives the EU circular material use rate under three trajectories, including the trend estimated on the biennial Waste Statistics Regulation reference years alone. (B) Gives global municipal solid waste, with 2022 and 2050 as published by Cook et al. (2026) and 2030, 2035 and 2040 interpolated at the implied growth rate.

4.6 Scenario summary

Bringing the strands together, Table 5 also compares the business-as-usual forecast, the illustrative accelerated scenario and the target pathway. Business as usual reaches about 12.4 percent in 2030 and 12.8 percent in 2035, or 12.0 and 12.2 percent when the model is estimated on the biennial reporting years alone. The accelerated scenario reaches about 14.1 percent and 15.7 percent. The target pathway requires 22.4 to 24 percent by 2030. The distance between the accelerated scenario and the target is the space that policy would have to fill, and it is large. Because the accelerated scenario already assumes a tripling of the historical pace, the remaining distance is large in absolute terms. What would close it is beyond the reach of this design to determine: our data speak to the size of the gap, not to the mix of rules, enforcement, market conditions or technology that closing it would require.

5 Discussion

5.1 Institutional capacity as an interpretation consistent with the evidence

The results support a reading, and it is worth stating its limits alongside its content. The tools for evaluating circular water and waste infrastructure are mature, yet the aggregate circular material use rate has barely moved and is forecast to miss its target by a wide margin. At the same time, regulatory quality is positively associated with circular performance across all 27 states, both in the headline metric once its structural confound is taken into account and, on the primary specification, in a numerator-based measure that does not condition on that confound, although there the association is modest and its robust confidence interval includes zero. Taken together, these findings are consistent with an institutional constraint operating on the circular transition. They do not establish that the constraint is institutional rather than analytical, and it is worth being precise about why. This study observes neither the adoption nor the quality of life cycle assessment, techno-economic assessment, multi-criteria analysis or AI-enabled tools, nor infrastructure investment, permitting throughput or project outcomes. Ranking the two candidate explanations would require exactly those observations. What we can say is that the association we find runs in the direction the institutional account predicts, that it is modest in size and exploratory in strength, and that the forecast gap is large. These observations are compatible with an institutional reading. They do not weigh that reading against a methodological or technological one, because the design contains no measure of either, and we frame the policy implications below accordingly. It is also worth restating what the governance measure is: Regulatory Quality is a broad perception-based indicator of the design and implementation of policies and regulations that permit and promote private-sector development, not an observation of permitting throughput, procurement practice, enforcement activity or sectoral legal capacity. The discussion that follows should therefore be read as an argument about mechanisms the indicator is compatible with, not mechanisms it measures. It is consistent with the evidence-based policy literature, which has long argued that the influence of analysis depends on the arena it enters (Head, 2016; Cairney, 2016), with transitions research, which stresses that regimes change only when rules, finance and practices move together (Geels, 2002; Markard et al., 2012), and with the panel evidence that good governance raises recycling performance (Pronti and Zoboli, 2024), on which the causal weight of the mechanism continues to rest. The premise of evidence-based decision making is that a good number leads to a good choice, but our findings illustrate that even a well-designed official indicator can mislead if it is read without regard to context. Comparing Finland and Italy on the headline rate alone could wrongly suggest that governance is unrelated to circularity; as Section 4.3 showed, the governance signal becomes visible only once material intensity is taken into account, and even then Finland and Ireland remain unexplained. For the arena we study, this poses a question rather than answering one: whether the binding constraint lies in the design of analytical models or in the institutional conditions under which their outputs are acted upon. Our data cannot settle it, because they observe neither. The proposition that analytical tools make little difference to what gets built where permitting is slow, enforcement is limited and markets for recycled materials are thin is drawn from the evidence-based policy and transitions literatures cited above, not from the estimates reported here. The Netherlands already exceeds the doubling benchmark and Belgium sits close to its lower edge, while much of the Union remains in the single digits or low teens; the present design describes that dispersion but does not explain it.

5.2 Legal and institutional mechanisms

Several legal mechanisms can turn regulatory quality into circular outcomes, and each is a lever that policy can pull. Waste law establishes a binding architecture. The revised Waste Framework Directive raised recycling targets to 55 percent by 2025, 60 percent by 2030 and 65 percent by 2035, and the end-of-waste criteria decide when a recovered material may legally re-enter the market (European Union, 2018). Where these provisions are applied, the literature reports that secondary markets strengthen and that the economics assessed by assessment tools improve (Milios, 2018; Hartley et al., 2020); we do not observe either outcome in our data. Water law functions similarly for reuse: the Regulation on minimum requirements for water reuse replaces an informal approach with quality classes and risk-management requirements, lowering the regulatory uncertainty that has long deterred investment in water circularity (European Union, 2020; Voulvoulis, 2018).

The end-of-waste question is especially important because it is where law bears most directly on the markets that assessment tools evaluate. A recovered material remains subject to heavier handling, permitting and liability requirements, which reduce its value and discourage its use, until it ceases to be legally designated as waste. Clear and consistent end-of-waste rules are argued to remove those burdens and to let secondary materials compete on a more level footing, on which most circular business cases are said to depend (Milios, 2018). Thin and uncertain secondary-material markets are repeatedly identified in the literature as a key barrier to circular policy, and they are tied precisely to legal definitions and their inconsistent implementation across jurisdictions (Hartley et al., 2020; Milios, 2018; Merli et al., 2018). This illustrates why the legal-institutional environment is better treated as a set of concrete and improvable rules than as a nebulous background condition, even though the indicator used in our estimation summarises that environment only at a high level of aggregation and does not measure any of these instruments individually. Procurement law is a particularly direct lever for infrastructure, because public buyers commission a large share of water and waste assets: mandatory green and circular public procurement would require, rather than merely permit, authorities to weigh circular criteria, giving the outputs of life cycle and techno-economic assessment a statutory route into contract award (Pouikli, 2021; Sönnichsen and Clement, 2020). Extended producer responsibility reshapes the cost structure that assessment tools model, internalising end-of-life costs and making circular options more competitive on the very metrics decision tools compute (Milios, 2018). Each mechanism is a candidate route by which the legal-institutional environment could come to affect circular outcomes. Our cross-country analysis measures none of them: it uses a broad perception-based governance indicator, and the step from that indicator to any specific instrument discussed here is an argument drawn from the literature rather than an estimate from our data.

5.3 Algorithmic evidence, accountability and dispute resolution

As artificial intelligence and digital platforms enter infrastructure decisions, the legal questions shift from the content of the evidence to its accountability. A recommendation produced by an opaque model is harder to explain, challenge and review than one produced by a documented assessment, and this creates a governance problem precisely at the intersection of algorithmic output and regulatory framework (Tsamados et al., 2022; Sun and Scanlon, 2019). If digital evidence is to strengthen rather than erode the legitimacy of infrastructure decisions, it needs to be auditable, contestable and consistent with administrative-law standards of reason-giving. Otherwise the transparency that decision tools were meant to bring is lost inside a model that no stakeholder can interrogate.

There is also a dispute-resolution dimension that is easy to miss. Circular water and waste assets are often financed and operated through long-term public-private partnerships, and the willingness of private parties to invest hinges on the predictability of the rules and the availability of credible, neutral mechanisms to resolve disputes over performance, tariffs and regulatory change. Regulatory quality and the strength of the arbitration and contract environment are therefore likely to be related, and both plausibly affect whether an analytically attractive project can close, although neither channel is observed in the present design. Even the best assessments and the best regulatory bargains will not produce circular infrastructure if investors doubt that the bargain will be honoured or that disputes will be resolved fairly. Improving the legal certainty of circular infrastructure contracts is thus part of improving regulatory quality.

5.4 Comparative decision arenas across water and waste

A comparative dimension can be read across the two sectors, though we offer it as a hypothesis for further work rather than a finding of the present cross-section. In the water sector, the opportunity for reuse-led circularity appears large in water-stressed member states, but the institutional capacity to realise it is uneven; the Regulation on minimum requirements for water reuse was designed to address exactly this tension through harmonised water-quality classes, and its effectiveness will indicate whether a coherent legal framework can lift water reuse beyond its current modest level (European Union, 2020; Voulvoulis, 2018). In the waste sector, the widest dispersion in our data lies between a group of western and northern member states and a group of central and eastern ones, where recycling rates and circular material use are in the single digits and where the governance scores are lowest. That grouping is descriptive and admits exceptions: Estonia records the highest circular material use per capita in the Union on a mid-range governance score. The same sorting and treatment technologies are available across the Union, which is one reason the dispersion has often been read as institutional rather than technological, but our design observes neither technology adoption nor administrative capacity nor enforcement, and cannot adjudicate between those readings.

Read together, the two sectors suggest a pattern worth testing rather than a finding: the jurisdictions that lead on these indicators are also, for the most part, those that score highest on the governance measure, which is compatible with an account in which ambitious rules and the capacity to implement them travel together. Whether lagging jurisdictions are held back by data, by technology, by institutions or by some combination of the three is not something the present design can establish, and we offer it as a hypothesis for the panel and project-level work proposed in Section 5.7. This also suggests, as a prospective mechanism, that weak national implementation can be partly offset by harmonised EU-level instruments, such as the Water Reuse Regulation or a mandatory circular-procurement rule. These are supranational instruments of the Union’s own legal order, not foreign regulation, and the precise terminology matters: the Union sets the common rule, and national authorities implement it.

5.5 Implications for policy and practice

Three implications follow for the decision arena, all framed by one point: the gap we document is large, and it would not be closed by presentational changes alone. We do not claim to have shown that analytical tools are adequate and institutions are not, nor that either explanation is the more likely of the two. We claim only that the patterns reported here are compatible with an institutional account, and the recommendations below should be read as a prospective research and policy agenda at that strength. First, the target regime should be aligned with institutional capacity: an ambitious doubling target is a useful signal, but our forecasts show that the observed trajectory falls far short of it. The literature on which we draw points to the permitting, enforcement and market-building functions as the place to look for an explanation; establishing that they are the explanation would require the project-level data this study does not have. Second, evidence should be read in context: that a high-governance economy can post a low headline circularity score is a caution against ranking jurisdictions on a single indicator and an argument for a small dashboard of complementary indicators read together, including at least one numerator-based measure alongside the ratio (Saidani et al., 2019; Corona et al., 2019). Third, the legal instruments most often proposed as routes to a stronger regulatory environment, circular procurement, water reuse regulation, end-of-waste definitions and extended producer responsibility, deserve empirical attention alongside the analytical tools, and their effects on what is actually built should be estimated rather than assumed.

5.6 Limitations

Several limitations qualify these conclusions, and we state them plainly because they bound the claims. First, the design is descriptive and hypothesis-sharpening, not causally identified: the forecast and the cross-section are two readings of official monitoring data, and neither observes tool use, permitting, procurement, investment or project outcomes, so the interpretation that the constraint is institutional rather than analytical is an inference consistent with the evidence and supported by the panel literature, not a measured effect, and the study cannot rank the two explanations against each other. Second, the forecast series is not fully independent year on year. The waste component underlying the circular material use rate is reported biennially and the intervening odd years are estimated by Eurostat, while the 2024 values are flagged as imputed throughout; nominal prediction intervals are therefore somewhat narrower than the true information content of the series would justify, which is why we report the biennial-subsample and exclude-2024 variants alongside the headline models. Third, the cross-section is a single year of Eurostat values, some of which Eurostat itself flags as estimated, so it should be read as a snapshot; the regressions, though informative, rest on 27 observations, and the numerator-based outcome is recovered arithmetically from the published ratio and published material consumption rather than published directly, so it inherits the rounding of the reported rate, which matters most for the lowest-scoring states. Fourth, regulatory quality is a perception-based proxy published with non-trivial standard errors, between 0.196 and 0.232 on this sample, and it is broader than any single environmental instrument, so it cannot by itself diagnose which specific reform matters most (Kaufmann et al., 2011). More fundamentally, the published construct concerns the perceived ability of government to formulate and implement sound policies and regulations that permit and promote private-sector development. It is therefore not a measure of environmental permitting, waste-sector enforcement, procurement practice or legal capacity, and none of the specific legal mechanisms discussed in Section 5.2 is observed by it; the mapping between the two is an argument, not a measurement. The 2025 methodology revision also means that estimates are not directly comparable with those in studies using earlier releases. Fifth, the primary numerator association is exploratory rather than established: its robust 95 percent confidence interval includes zero, the level specification clears conventional significance only narrowly, and the two differ enough that the result should be treated as a direction worth pursuing rather than as a measured effect. Sixth, structure does not explain every case: Finland and Ireland remain substantially below their predicted circular material use per capita even after material throughput is set aside, and we do not have the member-state material-flow detail to say why. Seventh, the forecasts extrapolate historical behaviour and cannot anticipate discontinuous policy change; a substantial change in enforcement could bend the curve, although the size of the gap makes a full recovery of the 2030 target very unlikely. Eighth, the legal analysis is descriptive rather than doctrinal or comparative, and the global-waste material is contextual; both are offered to frame the problem, not to test it. Finally, the analysis is confined to the European Union, so external validity to very different institutional settings is untested, and we make no claim that the pattern generalises beyond it.

5.7 Future research

These limits point to a research agenda. The most valuable next step would be a full member-state panel that pairs circular indicators with governance and explicit legal-implementation variables, such as Waste Framework Directive transposition, end-of-waste decisions, extended producer responsibility schemes, circular-procurement rules and water-reuse uptake, together with explicit forecasting, so that the association we illustrate can be examined with stronger causal designs and projected jurisdiction by jurisdiction. Such a panel would also allow the numerator of the circular material use rate to be modelled directly in physical units, which would remove the denominator problem that constrains a single cross-section, and would let the Finnish and Irish cases be decomposed by material category. A second avenue is project-level process tracing inside specific procurement or permitting decisions, following how a particular assessment did or did not change what was built, which would connect the statistical pattern to the mechanism and would supply the tool-use observations this study lacks. A third avenue is the governance of algorithmic evidence in infrastructure, including how administrative-law standards of transparency and contestability can be applied to model-based recommendations. Each would deepen the central claim of this paper, that the circular transition depends on the marriage of good evidence and good rules.

6 Conclusion

Circular water and waste infrastructure is essential to climate and resource strategy, and the tools for evaluating it are more sophisticated than ever. Yet the European Union circular material use rate has risen only from 10.7 percent in 2010 to 12.2 percent in 2024, and convergent forecasts place it near 12.8 percent in 2035, roughly eleven percentage points short of the doubling benchmark, with every published target vintage lying far outside the 95 percent prediction interval of every model we estimated; restricting the estimation to the biennial years for which the underlying waste statistics are reported rather than estimated moves that path down, not up. Regulatory quality is positively associated with circular performance across all 27 member states, in the headline metric once its structural confound is taken into account and in a numerator-based measure that does not condition on it, although on the primary specification the association is modest, its robust confidence interval includes zero, and two high-governance economies remain unexplained. These findings are compatible with an institutional constraint on the transition. They do not establish one, and because the study observes neither the use of analytical tools nor permitting, enforcement or investment, they cannot be weighed against an analytical or technological explanation; we therefore offer the institutional account as a reading the data do not contradict rather than as a finding. The proposition that evidence-based decision tools shape circular infrastructure only to the extent that the surrounding rules, enforcement and markets allow is one this study motivates but does not test. The agenda it points to is to establish whether circular procurement law, water reuse regulation, end-of-waste clarity, extended producer responsibility and accountable algorithmic evidence measurably change what gets built, so that the numbers produced by assessment can be connected to the infrastructure the climate transition requires.

Correction note

This article has been corrected with minor changes. These changes do not impact the scientific content of the article.

Generative AI statement

The author(s) declared that Generative AI was used in the creation of this manuscript. Grammarly Premium was used solely to check grammar and improve the clarity of the language in the course of writing this manuscript. No generative artificial-intelligence tool was used to produce the content, analysis, data or citations, all of which were prepared and verified by the authors, who take full responsibility for the content of the manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Kirti Goyal, Lovely Professional University, India

Reviewed by: Xinke Du, Shanghai Normal University, China

Khanh Huy Nguyen, University of Economics Ho Chi Minh City, Vietnam

Ionela Gavrila-Paven, 1 Decembrie 1918 University, Romania