Regional coordination and industrial green total factor productivity: evidence from the yangtze river delta economic coordination council

1 Introduction

City-regional governance has become increasingly central to debates on industrial upgrading and environmental transition. Production networks, labour markets, transport systems, and environmental externalities routinely extend across municipal boundaries, whereas industrial policy, environmental governance, infrastructure provision, and land use regulation remain largely organized by separate local governments. This mismatch matters because fragmented urban governance can encourage duplicated industrial investment, inefficient allocation of capital and labour, and regulatory arbitrage across jurisdictions (Shen, 2007). Recent work on place-based transition governance similarly argues that decarbonization and industrial change depend on institutions that operate across functional economic areas rather than within isolated administrative units (Cai et al., 2024; Allan et al., 2025).

Against this background, intercity cooperation is often presented as a practical response to policy problems that no single city can address effectively on its own. Regional cooperation may improve policy coordination, reduce cross-jurisdictional frictions, and internalize spillovers that arise within an integrated urban region (Castanho et al., 2017). Yet whether institutionalized intercity cooperation can improve industrial green productivity remains an open empirical question. A large part of the existing literature still evaluates green effects using emissions-based indicators such as air pollution, wastewater, or carbon intensity (Li and Wu, 2024; Lv et al., 2024). These outcomes are important, but they do not necessarily capture productivity-based greening. Emissions may decline because of end-of-pipe treatment, output contraction, or the relocation of pollution-intensive production, rather than because of cleaner and more efficient industrial upgrading.

For this reason, the central question of this paper is not simply whether regional cooperation reduces pollution, but whether it improves industrial green total factor productivity (GreenGTFP). GreenGTFP captures the efficiency dimension of industrial green development. It does not represent the full process of green industrial transformation, which also involves industrial-structure adjustment, energy-structure upgrading, cleaner production, pollution-control capacity, and institutional changes in environmental governance. The empirical focus is therefore productivity-based industrial greening, reflected in more efficient industrial production after accounting for factor inputs and undesirable environmental outputs (Li and Wu, 2024; Zhao et al., 2025).

This paper studies that question in the Yangtze River Delta (YRD), China’s most economically advanced, manufacturing-intensive, and innovation-dense urban region. The institutional setting is the Yangtze River Delta Economic Coordination Council (YRDECC), a formal intergovernmental platform through which member cities coordinate industrial development, environmental governance, infrastructure connectivity, and innovation-related policies. Historically, YRDECC membership evolved over multiple stages, and the post-2004 treatment variation used in this paper comes from later documented expansion waves within that longer institutional process. This staged institutional evolution is important because it allows the paper to identify the effects of institutionalized regional cooperation using variation in treatment timing rather than relying on a simple before and after comparison. Related studies have also used YRDECC expansion or YRD integration as quasi-natural experiments to evaluate regional policy effects, which reinforces the relevance of this institutional setting for causal analysis.

The paper develops two arguments. First, institutionalized intercity cooperation can improve GreenGTFP by reducing governance frictions that limit productivity-based industrial greening. A formal coordination platform may strengthen regulatory stringency, narrow regulatory divergence across cities, reduce factor misallocation, and support green innovation. These channels are usually examined separately in the existing literature. This paper treats them as connected mechanisms through which regional coordination affects industrial green productivity.

Second, the gains from institutionalized cooperation may be uneven across space. Larger cities usually have deeper markets, stronger industrial bases, better infrastructure, and greater absorptive capacity. They may therefore be better positioned to convert regional coordination into GreenGTFP gains. Smaller cities may benefit less, even when the regional average improves. This unevenness is important because regional integration can raise aggregate performance while also reinforcing existing spatial hierarchies.

The effects of intercity cooperation may also extend beyond directly treated cities. If YRDECC changes environmental regulation, innovation exchange, and factor allocation within an integrated region, neighbouring cities may be affected through policy diffusion, knowledge spillovers, and production linkages. The empirical analysis therefore examines both local effects and neighbouring-city spillovers.

These arguments point to three gaps in the literature. First, many studies of regional integration and environmental performance still focus on emissions outcomes, while less attention has been paid to industrial green total factor productivity. Second, existing studies often examine environmental regulation, innovation, or factor allocation separately, leaving the connected mechanisms of regional coordination less fully understood. Third, less is known about whether the GreenGTFP gains from formal regional cooperation are concentrated in economically stronger cities and whether these gains spill over to neighbouring cities.

This study addresses these gaps using a prefecture-level city panel covering 2004 to 2024. GreenGTFP is used to capture the efficiency dimension of industrial green development, not the full process of green industrial transformation. The empirical design exploits staggered entry into YRDECC and estimates treatment effects using staggered-adoption difference-in-differences methods. This design is suitable for the YRDECC setting because treatment timing differs across cities and within-region untreated comparisons disappear after full regional coverage.

The paper contributes to the literature in three respects. First, it shifts the evaluation of regional coordination from emissions reduction alone to industrial green total factor productivity. Second, it uses a clearly defined regional coordination institution and its staged expansion to identify how formal intercity cooperation affects GreenGTFP. Third, it shows that the GreenGTFP effects of cooperation are heterogeneous across cities and spatially connected, linking regional governance to the uneven geography of industrial green productivity.

2 Analytical framework and hypotheses

2.1 Institutionalized intercity cooperation and industrial green total factor productivity

Intercity cooperation is often invoked as a response to policy problems that exceed the administrative capacity of any single city. In integrated city regions, production linkages, labour mobility, infrastructure networks, and environmental externalities routinely cross jurisdictional boundaries, yet policy authority remains fragmented across local governments. As a result, regional development may be constrained by duplicated industrial projects, inefficient factor allocation, and regulatory inconsistency rather than by local endowments alone (Shen., 2007; Allan et al., 2025). Recent work on place-based transition governance similarly argues that decarbonization and industrial change depend on institutions capable of operating across functional economic areas rather than within isolated municipal units (Allan et al., 2025).

This perspective is particularly relevant to productivity-based industrial greening. A city may reduce emissions through end-of-pipe treatment or output contraction, but such changes do not necessarily imply productivity-enhancing upgrading. For this reason, this paper focuses on industrial green total factor productivity (GreenGTFP), which captures whether industrial output is improved in a way that is compatible with lower undesirable environmental outputs. Existing empirical studies often rely on pollution indicators alone, but recent research has shown that environmental improvement and productivity-based greening should not be treated as equivalent outcomes (Li and Wu, 2024; Lv et al., 2024). The analytical question, therefore, is whether institutionalized intercity cooperation can improve GreenGTFP rather than merely reduce observed emissions.

A formal cooperation platform such as the Yangtze River Delta Economic Coordination Council (YRDECC) may affect GreenGTFP through several linked mechanisms. Some studies emphasize that intercity coordination can reduce cross-border policy frictions and improve the quality of regional governance (Castanho et al., 2017). Others suggest that the developmental effects of cooperation depend on the institutional conditions under which coordination is organized and sustained (Kwadwo, 2025). Building on this literature, this paper argues that YRDECC may influence productivity-based industrial greening through environmental governance, green innovation, and factor allocation, while also producing heterogeneous and spatially diffused outcomes.

2.2 Environmental governance: regulatory stringency and regulatory divergence

A first mechanism operates through environmental governance. One influential line of research argues that fragmented local governance can encourage regulatory competition and weaken environmental performance. In China, for example, fiscal and developmental pressure may induce local governments to relax environmental constraints to attract investment and industrial activity (Bai et al., 2023; Cui et al., 2025; Chen, 2025). From this perspective, institutionalized regional cooperation may matter because it can strengthen the credibility and consistency of environmental governance across jurisdictions.

However, stronger coordination does not automatically imply stronger upgrading. Environmental regulation may impose short-run compliance costs, and its productivity consequences depend on whether firms face stable and credible incentives to innovate rather than merely temporary pressure to abate (Feiock, 2009; Goodman-Bacon, 2021; Hu et al., 2022; Lv et al., 2024). This is why the present paper distinguishes between regulatory stringency and regulatory divergence. Regulatory stringency refers to the intensity of environmental governance effort. Regulatory divergence refers to the extent to which a city’s regulatory intensity departs from the contemporaneous regional benchmark. The first captures how strong regulation is; the second captures how consistent regulation is across cities.

This distinction is important because intercity cooperation may work not only by making regulation stricter, but also by making it more coordinated. If regulatory divergence declines, firms have fewer opportunities to engage in regulatory arbitrage across cities. If regulatory stringency rises at the same time, cleaner production and upgrading may become more attractive in relative and absolute terms. Recent studies on green regional development increasingly point to this interaction between governance quality and regional coordination, rather than treating regulation as a purely local attribute (Li and Wu, 2024; Østergaard et al., 2024).

Hypothesis 1 Intercity cooperation improves industrial green total factor productivity by strengthening regulatory stringency and reducing regulatory divergence across member cities.

2.3 Green innovation and diffusion

A second mechanism operates through green innovation. A long-standing literature on regional development suggests that interurban interaction can enhance knowledge exchange, learning, and innovation spillovers, especially when cities are linked through formal and repeated institutional channels (Boix and Trullén, 2007). From this perspective, a regional coordination platform may reduce information frictions and facilitate the circulation of policy experience, technical knowledge, and innovation resources across cities.

Recent evidence from the YRD points in a similar direction. Institutional and market integration appear to strengthen innovation incentives, although the benefits are not necessarily distributed evenly across space (Ma et al., 2024; Hu et al., 2025). More broadly, studies of green transition governance have argued that innovation should be understood not only as a firm-level response, but also as an outcome shaped by intercity coordination, institutional support, and regional policy alignment (Eadson and van Veelen, 2023; Østergaard et al., 2024). These insights suggest that institutionalized intercity cooperation may improve GreenGTFP by supporting the generation and diffusion of cleaner technologies.

At the same time, the innovation channel is unlikely to be spatially uniform. Cities differ in absorptive capacity, industrial composition, and innovation infrastructure. As a result, regional cooperation may expand the scope for green innovation while still concentrating the strongest gains in cities with stronger technological and institutional foundations (Ma et al., 2024; Wang et al., 2026; Yang et al., 2026). The relevant empirical question is therefore not simply whether green innovation rises, but whether changes in green innovation are consistent with the observed pattern of GreenGTFP improvement.

Hypothesis 2 Intercity cooperation improves industrial green total factor productivity by increasing green innovation and facilitating the diffusion of cleaner technologies across cities.

2.4 Factor allocation and industrial upgrading

A third mechanism operates through factor allocation. In fragmented regional settings, the movement and matching of capital and labour can be impeded by administrative barriers, local protection, and policy inconsistency. These frictions may reinforce factor misallocation and reduce productivity growth. Recent work has highlighted that regional integration can improve development performance by lowering such frictions and facilitating a more efficient allocation of resources across cities (Zhang et al., 2020; Feng et al., 2024).

This channel is especially important for productivity-based industrial greening because productivity-based greening depends not only on cleaner regulation or better technology, but also on whether productive and less polluting firms can gain access to capital, labour, infrastructure, and regional markets. Intercity cooperation may therefore improve GreenGTFP if it reduces factor misallocation and supports more efficient industrial organization. At the same time, the same process may intensify spatial sorting. Higher-value and cleaner segments may become concentrated in stronger cities, while smaller cities receive lower-value or more pollution-intensive activities through reorganization and relocation (Li et al., 2022; Oh, 2010; Pu, 2025; Ren et al., 2024).

This dual possibility matters conceptually. If cooperation improves regional efficiency but redistributes activities unevenly across space, then the aggregate effect on GreenGTFP may still mask unequal local outcomes. The factor allocation channel thus links cooperation to both average upgrading and uneven regional transition.

Hypothesis 3 Intercity cooperation improves industrial green total factor productivity by improving factor allocation and reducing capital and labour misallocation.

2.5 Heterogeneous and spatial effects

The mechanisms discussed above imply that institutionalized intercity cooperation can improve GreenGTFP on average, but they also suggest that the gains may not be evenly distributed across cities. Research on sustainable regional development increasingly stresses that competitiveness, transition capacity, and environmental improvement can coexist with substantial spatial unevenness (Sedita et al., 2025; Li, 2023; Zhou et al., 2026). In city-regions, larger and economically stronger cities typically possess broader markets, stronger innovation systems, and better infrastructure. These characteristics may allow them to convert coordination, innovation diffusion, and factor reallocation into productivity gains more effectively than smaller cities.

This possibility is particularly salient in the YRD context. If institutionalized cooperation lowers barriers to mobility and strengthens regional coordination, economically larger cities may capture a disproportionate share of the benefits. Smaller cities, by contrast, may face weaker absorptive capacity and may benefit less on average, even when the regional average rises. This is why the heterogeneity analysis in this paper is organized around pre-treatment economic scale measured by 2004 real GDP, rather than around contemporaneous or post-treatment city size.

Hypothesis 4 The effect of intercity cooperation on industrial green total factor productivity is heterogeneous across cities, with stronger gains in economically larger cities and weaker gains in economically smaller cities.

A related implication is that the effects of cooperation may extend beyond directly treated cities. If YRDECC influences regulation, innovation, and factor allocation across an integrated urban region, then neighbouring cities may also be affected through policy diffusion, knowledge spillovers, and supply chain linkages. Recent studies of regional development interventions have emphasized that spillovers are not peripheral to evaluation but can be part of the treatment process itself (Roth et al., 2023; Tone, 2001; Wu and Sun, 2023; Zhang et al., 2024). In the context of productivity-based industrial greening, such spillovers may be positive if cleaner technologies and coordinated policies diffuse across space, but they may also be weaker or more uneven if burden shifting dominates.

Hypothesis 5 Intercity cooperation generates spatial spillovers in industrial green total factor productivity, such that YRDECC entry in one city affects GreenGTFP in neighbouring cities.

Taken together, this analytical framework gives the paper a clear empirical structure. The baseline analysis evaluates whether YRDECC entry increases GreenGTFP on average. The mechanism analysis then examines whether the estimated effects are consistent with changes in regulatory stringency, regulatory divergence, green innovation, and factor misallocation. The heterogeneity analysis assesses whether the gains are stronger in cities with larger pre-treatment economic scale measured by 2004 real GDP. Finally, the spatial analysis examines whether institutionalized intercity cooperation reshapes productivity-based industrial greening beyond the directly treated city by generating spatial spillovers across the region.

3 Research data and methods

3.1 Institutional setting and identification strategy

This paper studies the Yangtze River Delta Economic Coordination Council (YRDECC), a formal intergovernmental platform through which cities in the Yangtze River Delta (YRD) coordinate industrial development, environmental governance, infrastructure connectivity, and innovation-related policies (Li et al., 2026). Intercity cooperation can take many forms, but most informal arrangements are difficult to observe consistently over time. YRDECC membership, by contrast, provides an institutionally meaningful and observable marker of formal regional cooperation.

Historically, YRDECC membership evolved in several stages. The full YRD institutional sample contains 41 cities. Of these, 16 had entered YRDECC before 2004 and are therefore pre-sample treated, while 25 entered during the sample window through the later documented expansion waves. Supplementary Table B1 reports the city-by-city treatment roster, first documented membership year, cohort assignment, and the documentary basis used for coding.

The full YRD sample consists of Shanghai Municipality and the prefecture-level cities in Jiangsu, Zhejiang, and Anhui, for a total of 41 cities. The treatment of interest is entry into the formal YRDECC coordination framework. Let G i denote city i's first documented year of YRDECC membership and let T i denote the effective treatment year used in the annual panel. For cities that entered before 2004, T i < 2004, so they are treated as pre-sample treated units. For the accession wave formally admitted in October 2019, the baseline coding sets T i = 2020, because the data are annual and most of calendar year 2019 preceded formal entry. A robustness check instead codes 2019 as the first treated year. The treatment indicator is defined as

D i t = 1 t ≥ T i . (3.1)

In Equation 3.1 indicator equals one from the effective treatment year onward and zero otherwise. The post-2004 entry cohorts provide the within-sample treatment timing variation used in the staggered-adoption design.

Because treatment timing differs across cities, the baseline dynamic specification is written as a cohort-specific event-study model:

y i t = α i + λ t + ∑ g ∈ G ∑ k ≠ − 1 β g k 1 G i = g 1 t − g = k + γ ′ X i t + ε i t (3.2)

In Equation 3.2, y i t = ln ⁡ GreenGTFP i t, α i and λ t denote city and year fixed effects, and X i t is a vector of time-varying controls. G i denotes the first effective year in which city i entered the YRDECC coordination framework, and g ∈ G indexes the post-2004 entry cohorts used for staggered-adoption identification. Event time is defined as k = t − g. The year immediately preceding entry, k = − 1, is omitted as the reference period. Never-treated donor cities do not have a finite treatment cohort and therefore do not receive event-time indicators; they enter the estimation as comparison units. Cities that entered YRDECC before 2004 are treated as pre-sample treated units and do not contribute event-time identification in the baseline staggered-adoption estimates.

The reported dynamic effects are obtained by aggregating the cohort-specific coefficients β g k across entry cohorts using cohort-size weights. This specification is consistent with the staggered-adoption event-study framework and avoids imposing a common event-time coefficient across cohorts when treatment effects may vary by entry cohort and exposure length.

In addition to the cohort-specific event-study estimates, the paper reports group-time average treatment effects following Callaway and Sant’Anna (2021). For each treated cohort g and post-entry year t ≥ g, the group-time effect compares the observed outcome of cohort g with the corresponding counterfactual outcome constructed from never-treated donor cities and, where available, not-yet-treated comparison cities. The average post-entry effect is then obtained by aggregating group-time effects across cohorts and post-treatment years. This complementary estimator is used to check whether the baseline results depend on a particular staggered-adoption implementation.

The baseline controls include economic development, openness, urbanization, city size, population density, transport accessibility, and agglomeration. Economic development is measured by log real GDP per capita. Openness is measured by FDI intensity. Urbanization is measured by the non-agricultural population share. City size is measured by log urban population. Population density is measured by log population per land area. Transport accessibility is measured by log road density. Agglomeration is measured by the manufacturing location quotient. Standard errors are clustered at the city level.

3.2 Measuring industrial green total factor productivity

The dependent variable is industrial green total factor productivity, hereafter GreenGTFP. In this paper, GreenGTFP is used to capture the productivity and efficiency dimension of industrial green development. It should not be read as a complete measure of green industrial transformation. Green industrial transformation is a broader process that may involve changes in industrial structure, energy structure, pollution-control systems, cleaner production practices, and firms’ organizational responses to environmental regulation (Cui et al., 2025). GreenGTFP is appropriate for the present study because the research question concerns whether institutionalized regional coordination improves industrial production efficiency after taking factor inputs and undesirable environmental outputs into account. In this sense, the indicator reflects productivity-based industrial greening rather than green industrial transformation in its full structural and institutional meaning.

GreenGTFP is constructed using a slack-based measure model with undesirable outputs and is linked across years using the global Malmquist Luenberger index. The input side includes industrial labour, industrial capital, and industrial energy use. Industrial labour is measured by employment in industrial enterprises above designated size. Industrial capital is measured by the real capital stock of industrial enterprises above designated size. The capital stock is constructed using the perpetual inventory method based on deflated fixed-asset investment. The base year, depreciation rate, and initial stock assumptions are reported in the Supplementary Data Appendix. Industrial energy use is proxied by electricity consumption in industrial enterprises above designated size. Although electricity consumption cannot fully represent total industrial energy use, it is the only consistently available city-level industrial energy indicator for the full sample period.

The desirable output is real industrial value added. Undesirable outputs include industrial SO₂ emissions, industrial wastewater discharge, and industrial smoke and dust emissions. These environmental indicators are measured at the city-year level for industrial enterprises above designated size. The baseline outcome variable in the econometric analysis is ln ⁡ GreenGTFP i t.

Robustness checks further use alternative transformations of GreenGTFP and complementary environmental performance outcomes. Price factors are addressed before the GreenGTFP index is constructed. Industrial value added is converted into constant 2004 prices using the relevant industrial value-added deflator. Fixed-asset investment used to construct the capital stock is also deflated to constant 2004 prices before applying the perpetual inventory method, so the capital input reflects real investment rather than nominal investment. Monetary control variables, including real GDP per capita and related economic indicators, are also converted into constant 2004 prices using deflators consistent with the corresponding statistical sources. Energy use and pollutant emissions are measured in physical quantities and therefore do not require price deflation. This treatment helps ensure that changes in GreenGTFP reflect real changes in industrial productivity and environmental performance rather than nominal price fluctuations.

Accordingly, GreenGTFP is interpreted as a city-level measure of productivity-based industrial greening under available data conditions. To avoid relying only on this constructed productivity index, the robustness analysis also reports complementary environmental performance outcomes.

3.3 Mechanism variables, controls, and statistical inference

The mechanism analysis follows the analytical framework developed in Section 2. Rather than treating post-treatment variables as formal causal mediators, the paper evaluates whether the estimated treatment effects are consistent with changes in regulatory stringency, regulatory divergence, green innovation, and factor misallocation. The aim is therefore to assess mechanism consistency rather than to identify causal mediation shares.

The baseline controls are chosen to capture factors plausibly correlated with both YRDECC entry and GreenGTFP while avoiding mechanical overlap with the inputs and outputs embedded in the construction of GreenGTFP. As noted above, the control vector includes economic development, openness, urbanization, city size, population density, transport accessibility, and agglomeration. All monetary variables are converted into constant-price terms using deflators consistent with the underlying statistical sources (Zhang et al., 2025).

Factor allocation is captured by capital misallocation and labour misallocation indices derived from wedges in factor use relative to output shares, following standard misallocation approaches adapted to the city industry setting. Environmental governance is represented by two related but distinct dimensions. The baseline mechanism variable for environmental governance is regulatory stringency, proxied by investment-based environmental-governance intensity, measured as environmental governance investment divided by total industrial output. The paper also considers regulatory divergence, which captures within-region policy inconsistency and is defined as the absolute deviation between a city’s regulatory intensity and the contemporaneous leave-one-out YRD average. Lower values of regulatory divergence indicate stronger within-region alignment. Green innovation is measured using green patents constructed from environment-related technology classes based on standard patent concordances. To make the governance mechanism explicit, regulatory divergence is defined as:

R D i t = R e g i t − R e g Y R D , − i , t (3.3)

In Equation 3.3, Reg _it denotes investment-based environmental-governance intensity in city i at time t, and R e g Y R D , − i , t denotes the contemporaneous YRD average excluding city i.

Mechanism-consistent evidence is presented in two steps. First, the paper estimates reduced-form effects of YRDECC entry on each mechanism variable using the same staggered-adoption design as in the baseline analysis. Second, it estimates separate fixed-effects associations between the outcome variable and lagged mechanism variables:

y i t = α i + λ t + β M i , t − 1 + γ ′ X i t + ε i t (3.4)

In Equation 3.4, y i t denotes the baseline outcome and M i , t − 1 denotes the lagged mechanism variable of interest. These regressions are reported as mechanism evidence rather than as formal causal mediation tests.

To formally examine the statistical relevance of the proposed channels, we further conduct mediation tests for each mechanism variable. For each mediator M i t, we estimate the following two equations:

M i t = α i + λ t + a D i t + γ ′ X i t + μ i t (3.5)

y i t = α i + λ t + c ′ D i t + b M i , t − 1 + γ ′ X i t + ε i t (3.6)

In Equations 3.5 and 3.6 indirect effect is calculated as a b. We report both the Sobel test and city-clustered bootstrap confidence intervals. The Sobel statistic is computed as:

z = a b b 2 S E a 2 + a 2 S E b 2 (3.7)

In Equation 3.7 bootstrap test, cities are resampled at the cluster level, and the indirect effect is re-estimated in each replication. The bootstrap confidence interval is obtained from the empirical distribution of the estimated indirect effects. Because the mechanism variables are post-treatment outcomes, the mediation tests are interpreted as formal statistical tests of indirect effects rather than as a complete causal decomposition of the total treatment effect.

Inference accounts for dependence structures that are common in regional integration settings. Baseline standard errors are clustered at the city level. Additional robustness checks report spatially robust inference using distance-based spatial HAC procedures and broader cross-sectional dependence using panel-robust approaches. Where appropriate, a wild cluster bootstrap is used to assess sensitivity to a limited number of clusters. Diagnostic tests motivating the spatial specifications reported later in the paper are presented in Supplementary Table B3.

3.4 Sample and data sources

The analysis uses a prefecture-level city panel from 2004 to 2024. The full YRD institutional sample consists of 41 cities. The baseline staggered-adoption estimation sample is based on the 25 post-2004 entry cohorts together with a never-treated donor pool. Treatment timing is assigned using documented first-year YRDECC membership, as described in Section 3.1 and reported in Supplementary Table B1. In the baseline annual-panel coding, cities formally admitted in October 2019 are treated as entering from 2020 onward.

The baseline staggered-adoption estimation sample uses the 25 YRD cities whose effective treatment years fall within the 2004–2024 sample window, together with a donor pool of never-treated prefecture-level cities outside Shanghai, Jiangsu, Zhejiang, and Anhui. The 16 YRD cities that entered before 2004 are pre-sample treated and are therefore retained for institutional background and descriptive context but do not contribute event-time identification in the baseline staggered-adoption estimates.

The comparison pool is drawn from prefecture-level cities outside Shanghai, Jiangsu, Zhejiang, and Anhui that never joined YRDECC during 2004–2024. Consistent with the identification strategy, the national pool is used as a donor pool rather than as a mechanically valid control group. Cities with substantial missing information in industrial accounts or pollutant indicators are excluded. The remaining candidates are restricted to common support and reweighted using cohort-specific inverse-probability weights to align baseline covariates and pre-entry GreenGTFP dynamics over the 5-year pre-entry window. Observations with limited common support are excluded, and extreme weights are trimmed. In spillover sensitivity checks, the paper re-estimates the baseline results after excluding comparison cities in adjacent provinces and after excluding Tibet from the comparison pool.

Socioeconomic and industrial variables are compiled primarily from the China City Statistical Yearbook and related provincial and municipal yearbooks. Pollution indicators are drawn from official environmental statistical sources consistent with city-year coverage. Patent-based measures are constructed from official patent records using standard concordances for environment-related technologies. Variable definitions and descriptive statistics are reported in Table 1. Balance diagnostics are reported in Supplementary Table A1, the city-by-city treatment roster in Supplementary Table B1, the GDP-based grouping rule and sample distribution in Supplementary Table B2, and spatial diagnostic tests in Supplementary Table B3.

TABLE 1. Sample means and covariate balance before and after cohort-specific IPW.

The baseline staggered-adoption estimation sample consists of 25 YRD, cities with within-sample treatment timing variation and a control pool of never-treated prefecture-level cities outside Shanghai, Jiangsu, Zhejiang, and Anhui, observed over 2004–2024. The 16 YRD, cities that entered before 2004 are listed in Supplementary Table B1 but do not contribute event-time identification in the baseline staggered-adoption estimates. The weighted control mean is computed using cohort-specific inverse-probability weights constructed on baseline covariates and pre-entry GreenGTFP, dynamics. Standardized differences are reported before and after weighting. Smaller absolute values indicate better balance.

4 Empirical results

4.1 Baseline effects of YRDECC entry on GreenGTFP

4.1.1 Event study evidence

The analysis begins with the event study results. Because YRDECC entry is staggered across cities, the paper uses the cohort specific estimator of Sun and Abraham (2021), which is appropriate when treatment timing differs across units. The dependent variable is ln ⁡ GreenGTFP i t, and the year immediately before entry is taken as the reference period.

Figure 1 shows two patterns. First, the lead coefficients in the 5 years before entry are small and do not show a systematic pattern. This is consistent with the view that treated cities and their comparison units followed broadly similar trajectories before treatment. Second, the post entry coefficients become positive and gradually increase over time. The effect is not confined to the year of entry. Instead, it appears to strengthen as regional coordination deepens and the institutional effects of YRDECC accumulate.

The delayed response is consistent with the institutional nature of regional coordination. If intercity cooperation works through better environmental governance, stronger green innovation, and improved factor allocation, one will not expect the full effect to appear immediately. The dynamic profile in Figure 1 is therefore in line with the analytical framework developed earlier.

4.1.2 Baseline treatment effects

Table 2 reports the baseline treatment effects. Columns (1) and (2) present estimates based on Sun and Abraham (2021). Columns (3) and (4) report group time average treatment effects following Callaway and Sant’Anna (2021). In the weighted specifications, cohort specific inverse probability weighting is used to improve comparability between treated cities and the external donor pool.

TABLE 2. Baseline treatment effects of YRDECC entry on GreenGTFP.

SA, refers to Sun and Abraham (2021); CS, refers to Callaway and Sant’Anna (2021). Weighted specifications use cohort-specific inverse-probability weighting. All specifications include city fixed effects, year fixed effects, and the baseline control set. Standard errors are clustered at the city level. Treated cities refer to the post-2004 YRDECC, entry cohorts used in the staggered-adoption identification. Control cities refer to the never-treated external donor pool outside Shanghai, Jiangsu, Zhejiang, and Anhui. *p < 0.10, p < 0.05, *p < 0.01.

Across all four columns, YRDECC entry is associated with a positive and statistically significant increase in GreenGTFP. The estimates are close in magnitude across specifications, which suggests that the baseline result does not depend on one estimator. Panel B shows a similar pattern over event time. The estimated effect is smaller in the first one to 2 years after entry and larger in later periods. Read together with Figure 1 and Table 2 suggests that institutionalized intercity cooperation improves industrial green total factor productivity on average and that the effect becomes more visible over time.

Figure 1 and Table 2 provide the baseline evidence. YRDECC entry is associated with a meaningful increase in GreenGTFP, and the positive effect does not disappear when alternative staggered adoption estimators are used.

4.2 Robustness, falsification, and sensitivity to the comparison group

4.2.1 Counterfactual construction and weighting choices

By October 2019, all 41 YRD cities had formally entered YRDECC. Accordingly, from 2020 onward, the credibility of the external comparison group becomes central to later-period identification.

Table 3 therefore examines whether the baseline result is sensitive to alternative comparison group constructions, weighting choices, and trimming rules.

TABLE 3. Sensitivity to counterfactual construction, weighting, and trimming.

Column (2) excludes donor-pool cities in provinces adjacent to the YRD, region and excludes Tibet. Column (3) uses an alternative balancing procedure. Columns (4) and (5) vary trimming thresholds for extreme weights. Standard errors are clustered at the city level. Treated cities refer to the post-2004 YRDECC, entry cohorts used in the staggered-adoption identification. Control cities refer to the never-treated external donor pool and therefore vary only when the donor pool is explicitly restricted. *p < 0.10, p < 0.05, *p < 0.01.

Column (1) reports the baseline weighted specification. Column (2) adopts a more conservative comparison group by excluding donor pool cities located in provinces adjacent to the YRD region and by removing Tibet. This reduces the possibility that untreated comparison cities are themselves indirectly affected by nearby integration spillovers. Column (3) replaces the baseline weighting approach with an alternative balancing procedure. Columns (4) and (5) vary the trimming rule applied to extreme weights.

The main result remains stable across these checks. The coefficient on YRDECC entry stays positive and remains close to the baseline magnitude. This stability suggests that it suggests that the estimated effect is not being driven by one weighting scheme or by a small number of heavily weighted controls. It also reinforces the interpretation that the baseline findings are not an artefact of how the external counterfactual is constructed.

4.2.2 Placebo timing tests

A second concern is that the baseline effect may reflect spurious timing correlations rather than the consequences of YRDECC entry. Figure 2 addresses this issue by repeatedly assigning pseudo treatment timing and re-estimating the baseline specification. The placebo estimates are centred near zero, and the associated p-values are mostly above conventional significance thresholds. This pattern is difficult to reconcile with a story in which the main result is generated by chance timing alone.

Supplementary Table A2 reports a related falsification exercise based on the treated-cohort estimation sample. In this test, each treated city is assigned a pseudo treatment year two or 3 years earlier than its actual effective treatment year in the annual panel, and the baseline specification is then re-estimated under the same weighting, trimming, and common-support rules. The pseudo-treatment coefficients are small and statistically insignificant, which is inconsistent with a strong anticipation explanation.

Supplementary Table A3 further examines whether the result is sensitive to distance-based restrictions on the external donor pool. Specifically, the table progressively excludes never-treated comparison cities located within 300 km and 500 km of the YRD cooperation area to reduce possible spillover contamination in the external counterfactual. The estimated coefficient remains positive and stable across these increasingly restrictive donor-pool definitions.

The observation counts reported in Supplementary Tables A2, A3 differ from those in the main tables because these falsification exercises are implemented on restricted cohort-specific estimation samples rather than on the full city-year panel reported in the baseline results. Taken together, these exercises support the interpretation that the baseline effect reflects YRDECC entry rather than coincident shocks, anticipation, or arbitrary timing.

4.2.3 Robustness to annual-panel treatment coding of the 2019 accession wave

The baseline annual-panel coding treats the cities formally admitted in October 2019 as entering from 2020 onward, because most of calendar year 2019 preceded formal entry. Although this coding rule is appropriate for annual data, it is still useful to examine whether the main results are sensitive to this timing convention. Table 4 therefore re-estimates the baseline treatment effects under an alternative coding rule that instead treats the October 2019 accession wave as entering from 2019 onward.

TABLE 4. Robustness to alternative treatment-year coding for the October 2019 accession wave.

This table examines whether the estimated treatment effects are sensitive to the annual-panel treatment coding of the October 2019 accession wave. Columns with baseline coding treat the cities formally admitted in October 2019 as entering from 2020 onward in the annual panel. Columns with alternative coding instead treat the same cities as entering from 2019 onward. All other treatment coding rules remain unchanged. SA, refers to Sun and Abraham (2021), and CS, refers to Callaway and Sant’Anna (2021). Treated cities refer to the post-2004 YRDECC, entry cohorts used in the staggered-adoption identification, and control cities refer to the never-treated external donor pool outside Shanghai, Jiangsu, Zhejiang, and Anhui. Standard errors are clustered at the city level. *p < 0.10, p < 0.05, *p < 0.01.

The results remain positive across all specifications and are close in magnitude under both coding rules. The alternative coding yields slightly smaller coefficients, which is consistent with the idea that assigning treatment from 2019 may partially dilute the estimated effect by treating the pre-accession portion of that year as exposed. However, the overall pattern of results does not change. This suggests that the main findings are not driven by one annual-panel convention for coding the October 2019 accession wave.

4.2.4 Robustness to stricter donor-pool construction

A further concern is that, once all 41 YRD cities had formally entered YRDECC by October 2019, identification in the later period depends critically on the construction of the external donor pool. The baseline design already restricts donor cities using data-availability criteria, common-support requirements, and cohort-specific inverse-probability weighting. To assess whether the results are sensitive to stricter comparison-group construction, Table 5 re-estimates the baseline specification using progressively tighter donor-pool definitions.

TABLE 5. Robustness to stricter donor-pool construction.

Column (1) reports the baseline donor-pool specification. Column (2) excludes donor-pool cities in provinces adjacent to the YRD, region and excludes Tibet. Column (3) restricts the donor pool to a tighter region of common support without inverse-probability weighting. Column (4) combines the tighter donor-pool restriction with cohort-specific inverse-probability weighting, and Column (5) additionally applies a tighter trimming rule for extreme weights. Treated cities refer to the post-2004 YRDECC, entry cohorts used in the staggered-adoption identification. Standard errors are clustered at the city level. *p < 0.10, p < 0.05, *p < 0.01.

The table reports three additional checks beyond the baseline donor pool. First, it excludes donor-pool cities in provinces adjacent to the YRD region and excludes Tibet. Second, it restricts the donor pool to a tighter region of common support. Third, it combines this tighter donor-pool construction with cohort-specific inverse-probability weighting and a tighter trimming rule for extreme weights. Across all specifications, the estimated treatment effect remains positive and broadly stable, although the coefficients become slightly smaller and the standard errors increase modestly as the donor pool is narrowed. This pattern is consistent with the interpretation that the baseline findings are not driven solely by one broad donor-pool definition, while also indicating that tighter restrictions naturally reduce estimation precision.

These checks improve confidence in the external counterfactual used in the later-period analysis, but they do not eliminate the possibility that unobserved differences remain between later-entry YRD cities and never-treated donor-pool cities. For this reason, the estimates are still interpreted as design-conditional average post-entry effects rather than as evidence of fully random treatment assignment.

4.2.5 Alternative outcome and environmental-performance robustness

The baseline analysis uses GreenGTFP because the paper focuses on productivity-based industrial greening rather than pollution reduction alone. Since GreenGTFP is a constructed productivity index, this section checks whether the result is also reflected in a more direct environmental-performance outcome. Table 6 replaces GreenGTFP with an industrial pollution-emissions index (Sun et al., 2024; Yang et al., 2026).

TABLE 6. Alternative environmental-performance outcome.

The dependent variable is an industrial pollution-emissions index, where higher values indicate worse environmental performance. Negative coefficients indicate improved environmental outcomes. SA, refers to Sun and Abraham (2021), and CS, refers to Callaway and Sant’Anna (2021). All specifications include the baseline controls, city fixed effects, and year fixed effects. Standard errors are clustered at the city level. Treated cities refer to the post-2004 YRDECC, entry cohorts used in the baseline identification. Control cities refer to the never-treated external donor pool. *p < 0.10, p < 0.05, *p < 0.01.

The coefficients are negative across all specifications. Since higher values of the index indicate worse environmental performance, the negative estimates imply that YRDECC entry is associated with lower industrial pollution emissions. This result does not replace the GreenGTFP analysis, but it reduces the concern that the baseline finding is driven only by one constructed productivity measure.

Taken together with the baseline estimates, the alternative-outcome results indicate that YRDECC entry is associated with both higher industrial green productivity and better environmental performance.

4.2.6 Robustness to alternative price deflation

Since GreenGTFP is constructed from real industrial output, capital input, energy use, and undesirable outputs, the treatment of price factors may affect the measured productivity index. The baseline measure deflates industrial value added and fixed-asset investment before constructing the SBM-GML index. Supplementary Table A4 examines whether the result changes under alternative deflation rules. The coefficient on YRDECC entry remains positive and statistically significant across all specifications. The estimates are close to the baseline magnitude, ranging from 0.0195 to 0.0211. This indicates that the main result is not driven by one specific price-deflation convention.

4.3 Mechanism and mediation evidence

Section 2 proposes that YRDECC entry may affect GreenGTFP through environmental governance, green innovation, and factor allocation. Table 7 reports the first set of mechanism results. Panel A estimates the effects of YRDECC entry on each mechanism variable. Panel B relates lagged mechanism variables to subsequent ln ⁡ GreenGTFP, conditional on city fixed effects, year fixed effects, and the baseline controls.

TABLE 7. Reduced form effects and outcome associations.

Panel A reports the effects of YRDECC entry on each proposed mechanism variable. Columns (1), (2), (3), and (5) use the baseline estimation sample. Column (4) focuses on within-YRD regulatory divergence and is estimated on the YRD balanced panel only. Regulatory stringency is measured by investment-based environmental governance intensity. Regulatory divergence is defined as the absolute deviation between a city’s regulatory intensity and the contemporaneous leave-one-out YRD average. Panel B reports fixed effects associations between ln GreenGTFP and lagged mechanism variables. All specifications include the baseline controls, city fixed effects, and year fixed effects. Standard errors are clustered at the city level. Formal mediation tests based on Sobel statistics and city-clustered bootstrap confidence intervals are reported in Supplementary Table A5. *p < 0.10, p < 0.05, *p < 0.01.

The factor allocation results are consistent with the proposed channel. YRDECC entry is associated with lower capital misallocation and lower labour misallocation. Both forms of misallocation are negatively related to subsequent GreenGTFP. This pattern suggests that regional coordination improves industrial green productivity partly by reducing factor allocation frictions.

The environmental governance results show a similar pattern. YRDECC entry is associated with stronger regulatory stringency, measured by investment-based environmental governance intensity, and with lower regulatory divergence within the YRD. Regulatory stringency is positively related to subsequent GreenGTFP, while regulatory divergence is negatively related to it. These results indicate that regional coordination matters not only through stronger environmental governance, but also through more consistent regulatory conditions across cities.

The green innovation results also support the analytical framework. YRDECC entry is associated with more green patenting, and lagged green innovation is positively related to subsequent GreenGTFP. This evidence is consistent with the view that regional coordination facilitates green knowledge exchange and cleaner technology diffusion.

Formal mediation tests are reported in Supplementary Table A5. For each mechanism variable, the indirect effect is calculated as the product of two coefficients. The first coefficient measures the effect of YRDECC entry on the mechanism variable. The second coefficient measures the association between the lagged mechanism variable and GreenGTFP after controlling for YRDECC entry, city fixed effects, year fixed effects, and the baseline controls. Supplementary Table A5 reports the indirect effects, Sobel statistics, Sobel p values, and city-clustered bootstrap confidence intervals.

The mediation results provide statistical support for the proposed channels. The indirect effects are positive in the expected directions. The evidence is strongest for capital misallocation and green innovation. Regulatory divergence is also supported at the margin, while labour misallocation and regulatory stringency show weaker Sobel results but remain directionally consistent under the bootstrap confidence intervals. These findings suggest that YRDECC entry improves GreenGTFP through better factor allocation, more coordinated environmental governance, and stronger green innovation.

The mediation results are interpreted with caution because the mechanism variables are post-treatment outcomes and may be affected by multiple adjustment processes after YRDECC entry. The Sobel and bootstrap tests are therefore used to assess the statistical relevance of the indirect effects, not to provide a full decomposition of the total treatment effect. Taken together, Table 6 and Supplementary Table A5 show that the Effect on GreenGTFP of YRDECC entry is closely linked to changes in factor allocation, environmental governance, and green innovation.

4.4 Heterogeneous effects by economic scale and coefficient-difference tests

Section 2 suggests that the effect of YRDECC entry may differ across cities with different initial economic conditions. Cities are therefore grouped by real GDP in 2004, before the main treatment variation occurs. The grouping is fixed over time, so that city types are not affected by later changes after entry. Large, medium, and small cities correspond to the top 20 percent, middle 30 percent, and bottom 50 percent of the treated-city distribution in 2004. Supplementary Table B2 reports the grouping rule and sample distribution.

Table 8 reports the results. Panel A shows a clear gradient. The estimated effect is largest for large cities, smaller for medium cities, and close to zero for small cities. Larger cities appear more able to translate regional coordination into GreenGTFP gains, probably because they have stronger industrial bases, deeper markets, and greater innovation capacity.

TABLE 8. Heterogeneous effects by pre-treatment economic scale and coefficient-difference tests.

City groups are defined among the 25 post-2004 YRDECC, entry-cohort cities. Large, medium, and small cities correspond to the top 20 percent, middle 30 percent, and bottom 50 percent of the treated-city distribution. Panel A reports group-specific estimates. Panel B reports coefficient-difference tests from a pooled interaction specification. Wald p values and Fisher permutation p values are reported. All specifications include the baseline controls, city fixed effects, and year fixed effects. Standard errors are clustered at the city level. *, , and * indicate significance at the 10 percent, 5 percent, and 1 percent levels.

Panel B reports coefficient-difference tests. The large-small difference is statistically significant under both the Wald test and the Fisher permutation test. The medium-small difference is also significant, while the large-medium difference is positive but weaker. These tests show that the heterogeneity pattern is not based only on separate grouped regressions. The gains from YRDECC entry are systematically stronger in cities with larger pre-treatment economic scale.

4.5 Spatial dependence and spillovers

The spatial analysis is intended as a supplementary assessment of dependence and spillovers rather than as the core identification strategy (Sun et al., 2024). The final step is to examine whether the effects of YRDECC extend beyond the treated city itself. Before estimating spatial panel models, the paper tests for spatial dependence using Moran’s I and LM diagnostics. Supplementary Table B3 reports these results. They show that GreenGTFP and the residuals from the non-spatial baseline model both display spatial dependence. This provides the basis for estimating spatial specifications.

Table 9 reports spatial autoregressive and spatial Durbin models under two alternative spatial weight matrices, one based on contiguity and the other based on inverse distance. Across specifications, the spatial dependence parameter is positive and statistically significant. This means that GreenGTFP does not evolve independently across cities. In the spatial Durbin models, the indirect effect of YRDECC entry is also positive and statistically significant. The gains linked to YRDECC are therefore not purely local. They also extend to neighbouring cities.

TABLE 9. Spatial estimates of YRDECC entry effects.

The dependent variable is ln (GreenGTFP). Columns report spatial autoregressive and spatial Durbin models under two spatial weight matrices. The contiguity matrix is based on shared borders, and the distance matrix is inverse distance. Both matrices are row-standardized. In the SDM, specifications, direct, indirect, and total effects refer to the impact decomposition of YRDECC, entry across cities. All specifications include city fixed effects, year fixed effects, and the baseline control set. Standard errors are clustered at the city level. Treated cities refer to the post-2004 YRDECC, entry cohorts used in the baseline identification, and control cities refer to the never-treated external donor pool. *p < 0.10, p < 0.05, *p < 0.01.

These results support Hypothesis 5. They suggest that institutionalized intercity cooperation affects productivity-based industrial greening through a wider regional process, rather than only through changes within the treated city. A plausible interpretation is that policy diffusion, knowledge exchange, and cross-city production linkages transmit part of the effect across space.

5 Conclusion

This paper examines whether institutionalized intercity cooperation improves industrial green total factor productivity in the Yangtze River Delta. Using the staggered expansion of the Yangtze River Delta Economic Coordination Council, the analysis estimates the effect of formal regional coordination on GreenGTFP for Chinese cities from 2004 to 2024. GreenGTFP is interpreted as the productivity and efficiency dimension of industrial green development, rather than as a complete measure of green industrial transformation.

The results show that YRDECC entry significantly improves GreenGTFP, and the effect becomes stronger over time. The finding remains stable across staggered-adoption estimators, alternative comparison groups, placebo timing tests, alternative treatment-year coding, and complementary environmental-performance outcomes. The result is consistent with the regional nature of industrial production, environmental externalities, and innovation flows in the YRD.

The mechanism and mediation results point to three channels. YRDECC entry is associated with lower factor misallocation, stronger and more coordinated environmental regulation, and greater green innovation. Sobel and city-clustered bootstrap tests further support the statistical relevance of the indirect effects, especially through capital misallocation and green innovation. These results connect the Effect on GreenGTFP to changes in factor allocation, environmental governance, and innovation activity.

The effect is not evenly distributed across cities. It is strongest in economically larger cities and weakest in smaller cities. Coefficient-difference tests confirm that the large-small gap is statistically meaningful. The spatial results also show that the effect of YRDECC entry extends to neighbouring cities. Regional coordination therefore operates through both local adjustment and spatially connected processes such as policy diffusion, knowledge exchange, and intercity production linkages.

The findings have two implications. First, formal regional coordination can produce measurable improvements in industrial green productivity when it affects regulatory consistency, innovation diffusion, and factor allocation. Second, regional integration policies need to pay closer attention to smaller cities with weaker absorptive capacity. Otherwise, coordination may raise average regional performance while leaving part of the region behind. Future research could further examine differences in cooperation intensity after formal entry and compare alternative regional coordination arrangements across urban agglomerations.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this 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.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenvs.2026.1871994/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Otilia Manta, Romanian Academy, Romania

Reviewed by: Hanying Zhang, Chinese Academy of Agricultural Sciences (CAAS), China

Svetlana Slusarenco, Moldova State University, Moldova