The U-Shaped Impact of Manufacturing-Services Co-Agglomeration on Urban Green Efficiency: Evidence from the Yangtze River Delta

Abstract

Against the escalating challenges of global climate change and intensifying resource-environment constraints, exploring the green effects of industrial spatial organization has become crucial. Utilizing panel data from the Yangtze River Delta cities spanning 2011–2023, this study empirically examines the nonlinear impact of manufacturing-producer services co-agglomeration on urban green efficiency. The results reveal a significant U-shaped relationship: co-agglomeration initially suppresses efficiency due to coordination costs and congestion effects, but after crossing a specific threshold, the resulting scale economies and knowledge spillovers dominate and begin to promote green enhancement. Mechanism tests indicate that industrial upgrading serves as a direct mediating channel, while the mediating effect of green technological innovation exhibits a time lag. Further heterogeneity analysis shows that this U-shaped pattern is particularly pronounced in cities with low agglomeration levels, those not designated as low-carbon pilots, and non-resource-based cities. This study uncovers the nonlinear dynamics and key boundary conditions of the green effects arising from industrial co-agglomeration, providing an empirical basis for implementing differentiated regional spatial coordination policies.

1. Introduction

The escalating challenges of global climate change and intensifying resource and environmental constraints have rendered the promotion of urban green transition a pivotal imperative for achieving sustainable development goals worldwide. As the world’s largest manufacturing economy, China is actively exploring development pathways that harmonize economic growth with ecological preservation, primarily through industrial structural optimization and spatial organization innovation [1]. Within this national context, the Yangtze River Delta region, distinguished as one of China’s most economically dynamic, open, and innovative urban agglomerations, holds exemplary and pioneering significance in its practices of industrial agglomeration and green development. Industrial agglomeration, a quintessential spatial manifestation of regional economic development, has long attracted extensive scholarly inquiry regarding its environmental ramifications [2].

Traditional research in this domain has predominantly concentrated on the environmental impacts of single-industry or specialized agglomeration. Consequently, there remains a paucity of systematic and in-depth investigation into how the coordinated spatial agglomeration of two pivotal yet heterogeneous industries—manufacturing and producer services—shapes urban green development efficiency (GDE) [3]. Theoretically, the geographical co-location and functional synergy between these sectors can potentially foster green innovation and enhance efficiency through mechanisms such as scale economies, structural upgrading, and knowledge spillovers [4]. However, the initial phases of such coordination may also incur inhibitory effects stemming from coordination costs, factor congestion, and pollution concentration, suggesting the existence of a potential nonlinear, dynamic relationship [5]. Despite this theoretical proposition, empirical validation of this posited U-shaped relationship in the extant literature remains notably insufficient. Furthermore, rigorous exploration of key mediating transmission pathways, particularly the roles of green technological innovation and industrial upgrading, is relatively limited [6].

To address these research gaps, this study leverages panel data from 41 prefecture-level cities in the Yangtze River Delta spanning 2011 to 2023. We construct a two-way fixed effects model to empirically test the U-shaped impact of manufacturing-producer services coordinated agglomeration on urban GDE. Subsequently, we employ a mediation effects model to identify and elucidate the transmission mechanisms facilitated by green technological innovation and industrial upgrading. Additionally, this paper conducts a multifaceted heterogeneity analysis, scrutinizing variations based on agglomeration degree, low-carbon pilot policy status, and city resource endowment. This approach aims to uncover the boundary conditions that modulate the green effects of industrial coordinated agglomeration across diverse urban contexts.

This study seeks to contribute to the existing literature in three primary dimensions. First, it aims to reveal and empirically substantiate the nonlinear relationship between industrial coordinated agglomeration and GDE, thereby providing novel evidence for understanding the dynamic evolution of agglomeration’s environmental externalities. Second, it endeavors to clarify the distinct mediating roles played by green technological innovation and industrial upgrading, thereby deepening the mechanistic comprehension of how “coordination drives green transition.” Third, grounded in the empirical findings from the Yangtze River Delta, it aspires to furnish a scientific basis for crafting differentiated regional green coordination policies and to offer valuable insights for industry–environment synergistic governance in analogous regions globally.

2. Literature Review

2.1. Industrial Agglomeration Theory

The theoretical lineage of industrial agglomeration originates with Marshall’s seminal concept of industrial districts, which emphasized external economies arising from the geographic concentration of specialized industries, such as shared labor pools, specialized inputs, and knowledge spillovers [7]. This foundational view of specialization-based agglomeration was later complemented and challenged by Jacobs, who argued that urban innovation and growth are primarily driven by diversification—the co-location and interaction of different industries that foster cross-sectoral knowledge recombination and breakthrough ideas [8]. A major formalization and expansion of agglomeration theory came with the advent of New Economic Geography. Krugman provided a rigorous, model-based explanation for the endogenous formation of agglomeration through the interplay of increasing returns to scale, transportation costs, and factor mobility. Building upon these pillars, the theoretical framework evolved to systematically compare and synthesize these agglomeration patterns [9]. This scholarly progression naturally paved the way for the contemporary analytical focus on coordinated agglomeration—the strategic spatial coupling of interdependent yet heterogeneous industries, such as manufacturing and producer services, which is seen as an advanced form of spatial organization leveraging the synergies of both specialization and diversification externalities [10]. This coordination not only reduces transaction costs through shared infrastructure but also promotes regional innovation systems and value chain upgrading through cross-industry knowledge recombination and network embedding. Recent studies further indicate that ‘related variety’ across industries, particularly the knowledge linkages between manufacturing and high-end producer services, constitutes a more effective spatial organizational form for stimulating breakthrough green innovation [11].

The methodologies for measuring industrial agglomeration have evolved considerably, progressing from simple descriptive indices towards more sophisticated explanatory metrics. Early measurements relied heavily on descriptive indicators. Tools borrowed from industrial organization analysis, such as the Concentration Ratio and the Herfindahl–Hirschman Index, were used to gauge market or regional concentration. While simple to calculate and interpret, their fundamental limitation is the complete detachment from geographical spatial information, rendering them unable to distinguish between a national-level monopoly and a genuine local agglomeration formed by numerous interacting small and medium-sized enterprises (SMEs) [3]. Indicators based on administrative boundaries and industry shares, like the Location Quotient (LQ) and the Spatial Gini Coefficient, could identify specialization trends but remained susceptible to modifiable areal unit problems and could not differentiate between concentration dominated by a few large firms and genuine agglomeration arising from SME interactions [12].

A significant breakthrough was the Ellison–Glaeser (E–G) index [13], which effectively isolated the influence of firm size distribution to measure “pure” agglomeration economies for the first time. For assessing coordinated agglomeration, subsequent studies have developed improved versions of the E–G index or constructed linkage indicators using input-output tables. More advanced techniques, such as the Duranton–Overman (D–O) index, abandon predefined geographical units altogether [14]. Instead, they utilize firm-level micro-geographic data and kernel density estimation combined with random simulation to precisely characterize the spatial proximity patterns between firms. Complementary perspectives are offered by spatial statistics-based measures like Moran’s I—assessing spatial autocorrelation and point pattern analysis-based measures like the M-function—capturing distance decay effects [15].

2.2. Green Development Theory

The theoretical lineage of green development is rooted in profound critiques of the ecological consequences of industrial civilization. Its connotation has evolved from broad ethical advocacy towards a more precise theoretical framework and actionable policy objectives. A crucial direction in this theoretical deepening has been the shift from normative posturing to efficiency assessment, giving rise to the core concept of Green Total Factor Productivity (GTFP). GTFP integrates undesirable outputs, such as energy consumption and pollutant emissions, into the conventional production analysis framework, thereby measuring economic output efficiency under binding environmental constraints. It has emerged as a pivotal benchmark for evaluating the quality of green economic growth [16].

The inherent complexity and multidimensionality of green development necessitate that its measurement methods progress beyond simplistic single-indicator ratios towards comprehensive efficiency evaluation systems. Early research often employed ratio-based indicators like energy consumption per unit of GDP or carbon emissions per unit of GDP. While intuitive, these metrics fail to capture the overall efficiency and interconnectedness of multiple inputs and outputs within the economy-environment nexus. The mainstream and cutting-edge approach in contemporary academic research involves comprehensive efficiency measurement models grounded in Data Envelopment Analysis (DEA) and production frontier theory. This paradigm conceptualizes economic activity as a process transforming factor inputs into both desirable and undesirable outputs. By constructing a production possibility frontier, it measures the distance of decision-making units from this frontier, which represents their environmental efficiency or changes in GTFP [17].

Traditional radial DEA models faced shortcomings in adequately handling undesirable outputs. A major methodological advancement was the introduction of the Directional Distance Function by Chung et al. and the subsequent derivation of the Malmquist-Luenberger index, which allows for the simultaneous expansion of desirable outputs and contraction of undesirable outputs [18]. Another significant innovation is the non-radial, non-oriented Slacks-Based Measure (SBM) model proposed by Tone [16]. The SBM model addresses slack improvement issues more effectively, enabling more precise measurement of efficiency values, and has been widely adopted in environmental efficiency research [19]. Furthermore, Meta-frontier analysis is employed to compare efficiency levels across different technology groups. The SBM model utilized in this study represents one of the mainstream and advanced methodologies in this field, ensuring the scientific rigor and comparability of the GDE measurements.

2.3. Industrial Coordinated Agglomeration and Regional Green Development

Early research often examined manufacturing agglomeration and producer services agglomeration as parallel phenomena, separately investigating their environmental impacts. However, the mere geographical concentration of either sector in isolation may encounter nonlinear bottlenecks or limited efficacy in generating green benefits. A growing consensus suggests that the organic spatial combination and functional complementarity between these two industries are key to driving regions towards high-quality green development. Consequently, recent scholarship has increasingly focused on the impact of industrial coordinated agglomeration—specifically the co-location and interaction of manufacturing and producer services—on regional green performance [3,20]. The theoretical mechanisms underpinning this relationship primarily revolve around three categories of externalities:

(1)

Scale Externalities: Coordinated agglomeration enables the large-scale, cost-effective provision of shared environmental protection infrastructure and specialized green services. This effectively transforms the high fixed costs of green transformation, often prohibitive for individual firms, into variable operating costs that can be shared across the entire industrial ecosystem, thereby reducing the average cost of pollution abatement [5].

(2)

Structural Externalities: The core of this mechanism lies in the green restructuring of the regional economic system’s operational logic. When high-end producer services deeply embed themselves into the manufacturing value chain as knowledge carriers and governance nodes, they propel the industrial structure towards a more advanced, high-value-added, and low-pollution configuration. This shift releases “structural dividends” by reallocating resources from environmentally intensive, low-productivity sectors to cleaner, high-productivity activities [1].

(3)

Technological Externalities: Industrial coordination can break down technological silos between manufacturing and services, stimulating green technological innovation. Geographic proximity and established industrial linkages facilitate cross-sectoral spillover and recombination of green knowledge. This “innovation compensation” effect can potentially offset the compliance costs associated with environmental regulations, aligning with the Porter Hypothesis [21].

Critically, the net impact of coordinated agglomeration on green efficiency is not necessarily linear or monotonically positive. In the initial stages, negative effects may dominate due to several factors: high upfront coordination and adaptation costs, rising factor prices triggered by agglomeration, and the immediate concentration of pollution before shared abatement facilities are fully utilized or efficient networks are established--a phenomenon often termed the “congestion effect” [22]. It is only when coordination reaches a sufficient depth, scale, and quality that the positive externalities begin to outweigh the initial costs, potentially giving rise to a nonlinear, U-shaped relationship. Moreover, the strength and even the shape of this relationship are significantly moderated by heterogeneous contextual conditions, including regional institutional environments (e.g., environmental regulation stringency), resource endowments, and stages of industrial development [23].

While the existing literature provides a rich foundation of mechanistic analysis and partial empirical evidence, several avenues for expansion remain. First, many studies focus on linear relationships or singular effects, with insufficient rigorous testing and precise identification of potential nonlinear dynamic relationships and their inflection points between coordinated agglomeration and green development. Second, empirical investigations into the mediating pathways—especially research that simultaneously examines and compares multiple parallel mechanisms such as industrial structure advancement and green technological innovation—are relatively scarce. Finally, systematic empirical analysis based on a frontier region like the Yangtze River Delta urban agglomeration, a national exemplar of integrated development, can help reveal the specific manifestations and critical boundary conditions of the aforementioned mechanisms under particular institutional settings and developmental stages.

Aiming to address these gaps, this paper employs rigorous econometric models to test the U-shaped impact of manufacturing-producer services coordinated agglomeration on urban GDE. It further seeks to reveal the mediating roles of industrial advancement and green technological innovation, and to conduct in-depth heterogeneity analysis across dimensions of agglomeration degree, policy pilot status, and resource endowments. By doing so, it aims to contribute new empirical evidence to both theoretical refinement and evidence-based policy formulation in the realm of sustainable regional development. The details of the theoretical framework are shown in Figure 1.

3. Materials and Methods

3.1. Model Specification

To empirically test the hypothesized nonlinear relationship between industrial coordinated agglomeration and urban green development efficiency (GDE), and to control for other potential influencing factors, we establish the following econometric model based on theoretical analysis and prior research [24,25]. A two-way fixed effects panel model is employed to account for unobserved city-specific heterogeneity and time-specific shocks.

The baseline model is specified as follows:

$$\log {\mathrm{GDE}}_{\mathrm{i},\mathrm{t}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1{\mathrm{coag}}_{\mathrm{i},\mathrm{t}}+{\mathsf{\alpha }}_2{\mathrm{coag}}^2+\sum {{\alpha }_k\mathrm{logControl}}_{\mathrm{k},\mathrm{i},\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\lambda }}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i},\mathrm{t}}$$

(1)

where i and t denote city and year; $GDE$ is the dependent variable, green development efficiency; $coag$ is the core explanatory variable, the level of coordinated agglomeration between manufacturing and producer services; ${coag}^2$ is its squared term, introduced to capture the potential U-shaped relationship; $Control$ is the set of control variables; ${\mu }_i$ and ${\mathsf{\lambda }}_{\mathrm{t}}$ denote city and year fixed effects, respectively; and ${\epsilon }_{it}$, the idiosyncratic error term.

To examine the transmission mechanisms, we employ a mediation effects model based on the established procedure of He et al. and Song et al. [1,26]. The analysis follows. First, we regress GDE on industrial coordinated agglomeration alongside the control variables using a two-way fixed effects model. Provided the baseline relationship is established, we then conduct two subsequent regressions:

$$\log {GDE}_{i,t}={\beta }_0+{\beta }_1{coag}_{i,t}+{\beta }_2{coag}^2+\sum {{\beta }_{k}logControl}_{k,i,t}+{\mu }_i+{\lambda }_t+{\epsilon }_{i,t}$$

(2)

$${MED}_{k,i,t}={\gamma }_0+{\gamma }_1{coag}_{i,t}+\sum {{\gamma }_{k}logX}_{k,i,t}+{\mu }_i+{\lambda }_t+{\epsilon }_{i,t}$$

(3)

MED represents the mediator variable. Bootstrap methods are further employed to test the significance of the indirect effect.

3.2. Variable Selection and Measurement

3.2.1. Dependent Variable

We measure GDE using a non-radial, non-oriented Slacks-Based Measure (SBM) model within the Data Envelopment Analysis (DEA) framework, which effectively handles undesirable outputs [27]. The input-output system is constructed as follows:

Inputs: (1) capital stock, estimated using the perpetual inventory method with 2010 as the base year; (2) labor, measured by the year-end number of employed persons in each city; (3) energy, represented by the total electricity consumption of the whole society.

Desirable Output: Real Gross Domestic Product (GDP), calculated using 2010 constant prices.

Undesirable Output: An integrated environmental pollution index. Following common practice [28], we select emissions of industrial sulfur dioxide, industrial soot (dust), and industrial wastewater. The entropy weight method is applied to these three indicators to construct a comprehensive pollution index.

3.2.2. Core Explanatory Variable

Drawing on the methodology of Ellison and Glaeser [29] and its adaptation for studying industrial coordination in China [5,30], we construct a location quotient-based coordinated agglomeration index. First, the location quotient ($LQ$) for manufacturing (M) and producer services (P: Transportation, Storage, and Postal Services; Leasing and Commercial Services; Wholesale and Retail Trade; Financial Sector; Information Transmission, Computer Services, and Software Industries; Scientific Research, Technical Services, and Geological Prospecting Industries; Water Resources, Environment, and Public Facilities Management Industries.) in city $i$ at time $t$ is calculated:

$$L{Q}_{i,j,t}={\displaystyle \frac{q_{i,j,t}/q_{i,t}}{Q_{j,t}/Q_t}}$$

(4)

Then, a coordinated agglomeration index is constructed, which comprehensively reflects the degree of coordination between manufacturing and producer services through difference correction and scale superposition. A larger index indicates a higher level of coordinated agglomeration:

$${coag}_{i,t}=1-\left|{\displaystyle \frac{L{Q}_{i,M,t}-L{Q}_{i,P,t}}{L{Q}_{i,M,t}+L{Q}_{i,P,t}}}\right|+{\displaystyle \frac{L{Q}_{i,M,t}+L{Q}_{i,P,t}}{2}}$$

(5)

and, M represents manufacturing, P represents producer services, is the number of employed persons in industry j in city i in year t, is the total number of employed persons in city i in year t, is the total number of employed persons in industry j in the Yangtze River Delta in year t, and $Q_t$ is the total number of employed persons in the Yangtze River Delta in year t.

3.2.3. Mediator Variables

Green Technological Innovation (inn): Technology is a fundamental driver for breaking resource and environmental constraints and achieving green development. Coordinated agglomeration, through knowledge spillovers and collective learning, can promote green technological innovation. Green invention patents are a key indicator for measuring the output of core technological innovation addressing environmental problems. Measured as the natural logarithm of the number of green invention patents granted per 10,000 people [25].

Industrial Advancement (ts): Measured as the ratio of the added value of the tertiary industry to that of the secondary industry, reflecting the shift towards a more service-oriented and potentially less pollution-intensive economic structure [6].

Energy Consumption Intensity (eff): This refers to energy consumption per unit of GDP. Coordinated agglomeration can directly enhance GDE by improving management efficiency and promoting clean energy substitution, thereby reducing energy consumption intensity. It directly measures the decoupling between economic growth and energy consumption, representing resource use efficiency in green development [31].

3.2.4. Control Variables

To isolate the net effect of coordinated agglomeration, we include several city-level control variables commonly used in the literature on environmental efficiency:

Economic Development Level (gdp): Measured by the natural logarithm of GDP per capita. According to the Environmental Kuznets Curve theory, there is a nonlinear relationship between economic development stages and environmental quality. A higher economic development level usually implies stronger financial capacity for environmental protection, more complete infrastructure, and higher public demand for environmental quality, all of which directly or indirectly affect green development efficiency [32].

Science and Technology Investment (tec): Government investment in science and technology drives the regional innovation system. Measured by the proportion of science and technology expenditure in the local general fiscal budget, following studies on R&D support and innovation [33].

Infrastructure Level (infra): Measured by road mileage per unit of administrative area. This controls for the facilitation of economic activity and potential “carbon lock-in” effects. A well-developed network can reduce logistics costs and affect economic efficiency [34,35].

Urbanization Rate (urb): As a core macroeconomic variable describing the spatial organization form of regional population and economic activities, it has a profound impact on environmental governance efficiency. On one hand, urbanization, through the geographical concentration of population and economic activities, can generate significant agglomeration green effects by sharing green infrastructure at scale to reduce per capita pollution control costs and promoting knowledge and technology spillovers to drive regional green transformation. On the other hand, excessively rapid or extensive urbanization patterns may trigger congestion and pollution effects. Following Wu et al., this paper uses the proportion of the urban resident population to the total regional population to measure the urbanization rate [12].

Foreign Direct Investment (fdi): The introduction of foreign capital may be accompanied by the pollution haven effect or technology spillover effects, having complex impacts on local environmental technologies and standards. Drawing on the research by Hsu et al. on foreign investment and green development efficiency, it is measured by the proportion of actually utilized foreign capital in the regional GDP of the current year [36].

3.2.5. Summary of Variables

To provide a clear overview of the variables used in this study,

Table 1. Summary of the variables.

Variable Type	Variable Name	Abbreviation	Measurement
Dependent Variable	Green Development Efficiency	GDE	Calculated using the non-radial, non-oriented SBM model under the DEA framework.
Core Explanatory Variable	Coordinated Agglomeration Index	coag	Constructed based on the location quotient (LQ) of manufacturing (M) and producer services (P)
Mediating Variables	Green Technological Innovation	inn	The natural logarithm of the number of green invention patents granted per 10,000 people.
	Industrial Advancement	ts	The ratio of the added value of the tertiary industry to that of the secondary industry.
	Energy Consumption Intensity	eff	Energy consumption per unit of GDP.
Control Variables	Economic Development Level	gdp	The natural logarithm of GDP per capita.
	Science and Technology Investment	tec	The proportion of science and technology expenditure in the local general fiscal budget.
	Infrastructure Level	infra	Road mileage per unit of administrative area.
	Urbanization Rate	urb	The proportion of the urban resident population to the total regional population.
	Foreign Direct Investment	fdi	The proportion of actually utilized foreign capital in the regional GDP of the current year.

summarizes their types, names, abbreviations, and measurement.

3.3. Data Sources and Processing

Our analysis is based on a balanced panel dataset of 41 prefecture-level cities in China’s Yangtze River Delta region covering the period 2011–2023. The primary data sources are: The China City Statistical Yearbook (2012–2024); The China Energy Statistical Yearbook (2012–2024); The provincial statistical yearbooks of Shanghai, Jiangsu, Zhejiang, and Anhui (2012–2024); The National Bureau of Statistics and the State Intellectual Property Office of China for patent data. Sub-industry employment data for manufacturing and producer services are compiled from city-level statistical bulletins and economic census reports.

To ensure data consistency and mitigate potential econometric issues, we performed the following procedures: (1) For a small number of missing values in economic indicators, linear interpolation was applied. (2) To reduce the influence of extreme values, all continuous variables were winsorized at the 1 st and 99th percentiles. (3) Variables with large scales or potential heteroskedasticity (e.g., GDP per capita, patent counts) were transformed using natural logarithms. (4) Infrastructure and investment ratio variables were standardized for comparability. All empirical analyses were conducted using Stata 17.0.

Table 2. Descriptive statistics of core variables.

Variable	N	Mean	SD.	Min	Max
logGDE	533	3.535	0.265	2.863	4.188
coag	533	1.637	0.254	0.982	3.068
inn	533	−0.863	0.677	−2.851	0.726
ts	533	1.037	0.365	0.313	2.956
eff	533	0.616	0.279	0.040	1.720
gdp	533	11.625	0.669	9.848	13.657
tec	533	0.666	0.227	0.041	1.338
infra	533	1.444	0.364	0.565	2.249
urb	533	0.626	0.117	0.313	0.896
fdi	533	11.442	1.233	8.396	14.542

shows the basic statement of core variables.

4. Results

4.1. Analysis of Baseline Regression Results

To examine the baseline impact of industrial coordinated agglomeration on GDE, with a focus on its linear and nonlinear relationships and the confounding effects of control variables, the optimal model was determined through multiple tests, drawing on previous research: (1) the F-statistic of 32.74 (p = 0.000) led to the rejection of the null hypothesis that pooled OLS is superior to the fixed effects model; (2) the Hausman test statistic of 89.82 (p = 0.000) led to the rejection of the null hypothesis that the random effects model is superior; (3) the time fixed effects test yielded an F-statistic of 8.33 (p = 0.016), indicating the need to control for time heterogeneity. Consequently, a two-way fixed effects model was selected for the baseline regression. The results are presented in

Table 3. Baseline regression results.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	logGDE	logGDE	logGDE	logGDE	logGDE	logGDE	logGDE	logGDE
coag	0.047 ***	0.298 ***	0.032 **	−0.219 ***	0.031 **	−0.095	0.032 **	−0.189 **
	(3.34)	(3.45)	(2.63)	(−2.93)	(2.63)	(−1.30)	(2.63)	(−2.40)
coag_sq		−0.047 **		0.045 ***		0.023		0.039 **
		(−2.95)		(3.41)		(1.75)		(2.80)
gdp	−0.077 ***	−0.085 ***	0.412 ***	0.415 ***	−0.033	−0.026	0.412 ***	0.156 ***
	(−4.97)	(−5.46)	(11.27)	(11.47)	(−1.00)	(−0.78)	(11.27)	(6.33)
tec	−0.282 ***	−0.275 ***	−0.162 ***	−0.156 ***	−0.235 ***	−0.231 ***	−0.162 ***	−0.192 ***
	(−5.89)	(−5.78)	(−3.45)	(−3.34)	(−5.07)	(−4.99)	(−3.45)	(−3.96)
urb	1.131 ***	1.160 ***	−0.487 *	−0.481*	0.441 **	0.449 **	−0.487 *	0.897 ***
	(11.25)	(11.56)	(−2.15)	(−2.15)	(2.80)	(2.85)	(−2.15)	(5.64)
infra	−0.084 **	−0.079 **	−0.228 ***	−0.199 ***	−0.199 ***	−0.187 ***	−0.228 ***	−0.126 **
	(−2.82)	(−2.66)	(−5.71)	(−4.91)	(−5.47)	(−5.04)	(−5.71)	(−3.22)
fdi	0.030 *	0.028 *	0.008	0.007	0.044 ***	0.044 ***	0.008	0.044 **
	(2.58)	(2.45)	(0.60)	(0.50)	(3.45)	(3.40)	(0.60)	(3.20)
_cons	3.563 ***	3.350 ***	0.478	0.689 **	3.348 ***	3.412 ***	−0.699 *	1.156 ***
	(15.58)	(14.05)	(1.90)	(2.69)	(8.53)	(8.68)	(−2.28)	(4.12)
N	533	533	533	533	533	533	533	533
R2	0.375	0.385	0.412	0.426	0.363	0.412	0.435	0.440
Control city	NO	NO	YES	YES	NO	NO	YES	YES
Control year	NO	NO	NO	NO	YES	YES	YES	YES

Notes: Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

:

The findings reveal a significant U-shaped relationship between industrial coordinated agglomeration and GDE, characterized by initial suppression followed by subsequent promotion.

In the early stages of coordinated agglomeration, the geographical proximity between manufacturing and producer services does not immediately translate into green benefits. Instead, inhibitory effects may arise due to structural friction and coordination costs. This is specifically manifested in: (1) High costs of green industrial chain integration: Enterprises need to undertake production process reengineering and environmental equipment upgrades to adapt to the coordinated system, incurring substantial sunk costs. (2) Time lag and mismatch in knowledge spillovers: despite reduced geographical distance, the transfer of green technological knowledge across industries requires a shared cognitive base and trust mechanisms, with initial deficiencies in absorptive capacity [37]. (3) Prominent congestion effects: rising factor costs and the instantaneous intensification of environmental regulations in localized areas may increase the burden on enterprises in the short term, offsetting the scale economies brought by agglomeration [15].

Once coordinated agglomeration surpasses a critical threshold, its inherent green enabling mechanisms begin to dominate: (1) Scalable green governance: shared centralized pollution treatment facilities and specialized environmental consulting services achieve economies of scale, significantly reducing the average pollution control cost per enterprise. This essentially minimizes the cost of internalizing environmental negative externalities through agglomeration effects [38]. (2) Deep knowledge recombination and green innovation: stable innovation networks are established between manufacturing and producer services, facilitating cross-domain knowledge integration. As knowledge carriers, producer services not only provide energy-saving solutions but also embed themselves into manufacturing value chains, promoting green design throughout the product lifecycle, thereby pushing the green production possibility frontier outward [39]. (3) Market recognition and feedback for green products: coordinated agglomeration areas are more prone to forming green brands and standards, with market demand signals more clearly transmitted to the production end, incentivizing continuous green innovation.

Positive effect of economic development level (gdp): This not only confirms a regional manifestation of the Environmental Kuznets Curve (EKC) but, more crucially, reveals the “capital threshold” for green transition. A higher level of economic development implies stronger public fiscal capacity and private investment capability, forming the economic foundation for bringing green technologies from the laboratory to the market [21].

Negative effect of science and technology investment (tec): this counter-intuitive result may reveal a structural bias in current S&T resource allocation, where R&D investment might still flow heavily towards traditional technological paths promising short-term economic growth, rather than towards long-cycle, high-risk green disruptive technologies [23,40]. An alternative explanation is the immediate cost of R&D versus its lagged benefits; the panel data captures the current input costs, while the output of green technological innovation and the ultimate efficiency improvement exhibit significant time lags.

Negative effect of infrastructure level (infra): this suggests the potential carbon lock-in effect of current infrastructure. Infrastructure measured by road mileage density, if coupled with a fuel-based transportation system and traditional logistics networks, may solidify high-carbon travel and transport modes upon its expansion, hindering green transition [34,35].

Positive effect of foreign direct investment (fdi): this supports the applicability of the “Pollution Halo Hypothesis” in the Yangtze River Delta region. FDI entering this region likely possesses higher environmental standards and technological levels, bringing not only capital but also advanced environmental management practices and green technology spillovers. Through demonstration and competition effects, it elevates the local baseline of green efficiency [41].

4.2. Endogeneity Analysis

The potential presence of endogeneity might bias the regression results. Following previous research [42], this study addresses this by: (1) using geographic features: terrain ruggedness and lagged co-agglomeration as instruments. (2) using the one-period lagged terms of all explanatory variables for estimation; (3) based on diagnostic tests, comparing to System GMM, adopting Difference GMM (F = 23.89, p = 0.000; AR (2), p = 0.211). Considering that the two-way fixed effects model can effectively mitigate endogeneity caused by omitted variables, it was still employed to estimate the impact of manufacturing agglomeration on green development. The estimation results are shown in

Table 4. Results of endogeneity test.

Variable	(1)	(2)	(3)
coag	−0.419 ** (0.253)	−0.183 ** (−2.21)	−0.035 * (0.040)
coag_sq	0.040 ** (0.014)	0.0395 ** (2.63)	0.062 ** (0.027)
l_logGDE			0.265 ** (0.098)
gdp	0.297 *** (0.100)	0.150 *** (5.76)	0.458 *** (0.114)
tec	−0.073 *** (0.097)	−0.206 *** (−3.92)	−0.158 (0.097)
infra	−0.233 *** (0.069)	0.868 *** (5.21)	−0.330 *** (0.118)
urb	0.501 *** (0.322)	−0.119 ** (−2.87)	−0.874 (0.689)
fdi	0.045 ** (0.014)	0.0481 ** (3.108)	−0.006 (0.023)
_cons	1.156 *** (0.280)	1.194 *** (3.94)
N	533	492	451
R2	0.441	0.322
Control city	YES	YES	YES
Control year	YES	YES	YES
Model	IV-2SLS	IV-2SLS	Difference GMM

Notes: Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

. The endogeneity test results are consistent with the baseline regression results, indicating the robustness of the baseline estimates.

However, it is important to acknowledge that the primary strategy of using one-period lagged explanatory variables, while useful, may not fully resolve omitted variable bias or issues of reverse causation and simultaneity between GDE and coordinated agglomeration. Instrumental variable (IV) approaches, or dynamic panel methods, could offer stronger causal identification. The current strategy was chosen based on data availability and model specification tests, and its limitations should be considered when interpreting the results.

4.3. Mediation Effect Test

Based on the FE model, green technological innovation (inn), industrial advancement (ts), and energy consumption intensity (eff) were used as mediator variables to test the transmission mechanisms through which industrial coordinated agglomeration affects GDE. The Bootstrap test was combined to verify the significance of the mediation effects. The results are shown in

Table 5. Mediation effect test.

	(1)		(2)		(3)
	inn	logGED	ts	logGED	eff	logGED
coag	−0.339 **	−0.216 **	−0.0377 *	−0.242 **	−0.023 *	−0.212 **
	(−2.72)	(−2.87)	(−2.25)	(−3.29)	(−1.73)	(−2.83)
coag_sq	0.073 **	0.045 ***		0.050 ***		0.044 ***
	(3.28)	(3.33)		(3.85)		(3.33)
inn		0.007
		(0.29)
ts				0.147 ***
				(4.54)
eff						0.060
						(1.53)
gdp	0.976 ***	0.407 ***	0.508 ***	0.340 ***	−0.041	0.441 ***
	(16.21)	(9.06)	(10.22)	(8.71)	(−1.31)	(11.02)
tec	0.361 ***	−0.159 ***	0.374 ***	−0.210 ***	−0.0115	−0.151 **
	(4.64)	(−3.32)	(5.84)	(−4.45)	(−0.23)	(−3.25)
infra	0.133 *	−0.200 ***	0.332 ***	−0.244 ***	0.196 ***	−0.212 ***
	(1.97)	(−4.91)	(6.12)	(−5.98)	(5.00)	(−5.13)
urb	0.389	−0.484 *	−0.936 **	−0.342	−0.773 ***	−0.485 *
	(1.04)	(−2.16)	(−3.04)	(−1.55)	(−4.81)	(−2.17)
fdi	0.025	0.007	−0.069 ***	0.017	−0.058 ***	0.009
	(1.08)	(0.49)	(−3.51)	(1.21)	(−4.12)	(0.66)
_cons	−12.84 ***	−0.347	−4.109 ***	0.187	2.230 ***	−0.812 *
	(−24.66)	(−0.74)	(−9.84)	(0.56)	(5.62)	(−2.07)
N	533	533	533	533	533	533
R2	0.797	0.426	0.567	0.449	0.542	0.428
Control city	YES	YES	YES	YES	YES	YES
Control year	YES	YES	YES	YES	YES	YES

Notes: Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

.

(1)

Green Technological Innovation (inn): The mediating role is partially significant and exhibits a time-lag effect. The impact of coordinated agglomeration on green innovation itself is U-shaped, and its current-period contribution to GDE is not statistically significant. This suggests that coordinated agglomeration needs to reach a threshold to effectively stimulate green innovation, and the economic benefits of such innovation manifest with a lag, consistent with theories of technology diffusion and absorptive capacity [30].

(2)

Industrial Advancement (ts): The mediating effect is strong and significant. Coordinated agglomeration directly drives the regional economic structure to evolve from energy-intensive, high-emission manufacturing dominance towards a high-value-added, low-pollution integrated form of services and advanced manufacturing, by promoting the deep integration of high-end producer services and manufacturing. This structural change itself generates substantial structural dividends, as resources flow from sectors with low productivity and high environmental costs to those with high productivity and low environmental costs [26].

(3)

Energy Consumption Intensity (eff): The mediation effect was not statistically significant in this model. This may indicate that in the Yangtze River Delta context, the transmission mechanism through energy intensity is overshadowed by more dominant structural transformations or operates with a longer time lag than captured in the model [28].

4.4. Heterogeneity Analysis

Heterogeneity tests reveal that the U-shaped impact of industrial coordinated agglomeration on GDE is not universally applicable; its strength and form are profoundly constrained by a city’s existing structural conditions and development stage. Based on the FE model, the sample was grouped by agglomeration degree, low-carbon pilot status, and resource endowment to examine regional differences in the U-shaped effect. The heterogeneity test results are presented in

Table 6. Heterogeneity test results.

	(1)		(2)		(3)
	High-Level	Low-Level	Pilot	Non-Pilot	Resource	Non-Resource
coag	−0.120	−0.263 **	−0.071	−0.256 **	0.058	−0.292 **
	(−1.13)	(−2.73)	(−0.43)	(−3.05)	(0.47)	(−3.23)
coag_sq	0.030	0.053 **	0.026	0.0551 ***	0.002	0.056 ***
	(1.60)	(3.08)	(0.96)	(3.55)	(0.12)	(3.50)
gdp	0.544 ***	0.202 ***	0.657 ***	0.282 ***	0.267 ***	0.460 ***
	(8.75)	(4.16)	(9.31)	(6.56)	(4.68)	(9.77)
tec	−0.346 ***	−0.185 ***	−0.121	−0.250 ***	−0.152 *	−0.179 **
	(−4.36)	(−3.51)	(−1.78)	(−4.14)	(−2.19)	(−3.03)
infra	−0.150 *	−0.015	−0.259 **	−0.118 **	−0.184 **	−0.186 ***
	(−2.24)	(−0.33)	(−2.81)	(−2.67)	(−2.85)	(−3.75)
urb	−1.713 ***	0.121	−1.081	0.040	−0.183	−0.687 *
	(−4.07)	(0.51)	(−1.97)	(0.16)	(−0.64)	(−2.22)
fdi	0.006	−0.013	0.018	0.011	0.043	0.006
	(0.36)	(−0.61)	(0.68)	(0.60)	(1.42)	(0.38)
_cons	−0.881	1.454 ***	−2.998 ***	0.643	−0.096	−0.591
	(−1.75)	(3.47)	(−4.20)	(1.79)	(−0.20)	(−1.51)
N	277	256	155	378	117	416
R2	0.407	0.269	0.607	0.375	0.481	0.433
Control city	YES	YES	YES	YES	YES	YES
Control year	YES	YES	YES	YES	YES	YES

Notes: Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

.

(1)

Agglomeration Degree: The sample was divided into high and low agglomeration level groups based on the median of the coordinated agglomeration level. The results show that the U-shaped relationship is significant for cities with low initial agglomeration levels but flattens or becomes insignificant for high-agglomeration cities, possibly due to congestion effects or diminishing marginal green returns [26]. Physical proximity between industries is the main feature, but efficient knowledge exchange networks have not yet formed. At this point, marginal efforts to deepen agglomeration first face linear increases in static costs like organizational coordination and facility sharing, while dynamic, non-linear benefits from knowledge recombination are not yet apparent, resulting in an overall suppression of GDE. Once coordinated agglomeration crosses a threshold, sufficient interaction history and trust capital accumulate among firms and industries, stable innovation networks begin to form, and the positive externalities of knowledge spillovers start to increase non-linearly, gradually offsetting and surpassing coordination costs, propelling GDE onto an upward trajectory. For cities already at high agglomeration levels, their industrial networks may be relatively mature. Further deepening of agglomeration yields diminishing marginal green benefits, or may even produce negative effects due to excessive congestion and intensified competition, causing the U-shaped curve to flatten or become insignificant.

(2)

Low-carbon Pilot Cities: Based on the National Development and Reform Commission’s three batches of low-carbon city pilot lists, the starting year for pilots was identified and assigned a value of 1 in the dummy variable, with other cities assigned 0. A particularly noteworthy finding is the absence of a statistically significant U-shaped relationship in low-carbon pilot cities, while a clear U-shaped trajectory is observed in non-pilot cities. This may be attributed to the stringent and front-loaded environmental standards implemented in pilot cities, which raise initial coordination costs and delay the realization of green benefits—a phenomenon consistent with the “short-term cost” aspect of the Porter Hypothesis [43]. The pilot policy compels the industrial coordination system to embed green genes from its inception, potentially bypassing certain high-carbon lock-in stages and ultimately driving the green production possibility frontier toward a higher-order leap [44]. However, for low-carbon pilot cities, the stricter environmental standards and more rigorous assessment systems mandated by the state constitute a strong external institutional shock. In the early stages of coordinated agglomeration, enterprises not only need to bear conventional coordination costs but must also undertake additional expenditures for large-scale, high-standard green technological transformation and equipment upgrades to meet environmental requirements. This leads to a testable hypothesis for future research: whether the U-shaped relationship will eventually emerge over a longer time horizon as pilot cities undergo deeper structural adjustments.

(3)

Resource Endowment: With reference to the State Council’s National Sustainable Development Plan for Resource-based Cities (2013-2020), corresponding cities were defined as resource-based and assigned a value of 1 in the dummy variable, with others assigned 0. The U-shaped relationship is significant in non-resource-based cities but completely insignificant in resource-based cities. The economic structure, human capital, local finance, and even social perception of resource-based cities have long been highly self-contained systems revolving around resource extraction, primary processing, and related heavy or chemical industries. In this context, the coordination between manufacturing and producer services is more likely to reinforce the existing path: producer services primarily provide supporting services for resource-based manufacturing, and their knowledge spillovers are confined to improving traditional resource extraction efficiency rather than green technological innovation. Therefore, industrial coordinated agglomeration in resource-based cities, despite potentially high physical concentration, functionally falls into the trap of “pseudo-coordination.” Its agglomeration economies are more reflected in traditional scale economies, making it difficult to trigger structural and technological effects oriented towards green development [5].

4.5. Robustness Tests

To verify the reliability of the conclusions, robustness tests were conducted to examine whether the baseline U-shaped conclusion is robust. The results are shown in

Table 7. Robustness tests.

	(1)	(2)	(3)
	logS_GDE	logGDE	logGDE
coag	−0.173 **		−0.168 **
	(−2.38)		(−2.55)
coag_sq	0.039 **		0.034 **
	(3.05)		(2.90)
L.coag		−0.199 **
		(−2.49)
L.coag_sq		0.037 **
		(2.64)
gdp	0.334 ***	0.416 ***	0.408 ***
	(9.54)	(10.30)	(11.92)
tec	−0.204 ***	−0.145 **	−0.145 **
	(−4.50)	(−2.88)	(−3.17)
infra	−0.190 ***	−0.173 ***	−0.238 ***
	(−4.84)	(−3.93)	(−5.93)
urb	−0.378	−0.469	−0.399 *
	(−1.74)	(−1.87)	(−2.05)
fdi	−0.005	0.004	0.003
	(−0.38)	(0.31)	(0.25)
_cons	−3.578 ***	−0.477	−0.367
	(−11.80)	(−1.35)	(−1.23)
N	533	492	533
R2	0.332	0.398	0.440
Control city	YES	YES	YES
Control year	YES	YES	YES

Notes: Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

.

(1)

Replacing the Explained Variable: In measuring green development, the SBM-DEA model was replaced with the SBM-GML model. The U-shaped effect was fully replicated after replacement, indicating that the U-shaped relationship is robust to the measurement method of green development efficiency, and the nonlinear impact pattern of industrial coordinated agglomeration stably exists.

(2)

Lagged One-Period Regression: Considering the potential lagged effect of industrial coordinated agglomeration on GDE and to mitigate bidirectional causality bias, the one-period lagged explanatory variables (L.coag, L.coag_sq) were used for regression. The results remain robust. This not only alleviates potential reverse causality issues but also captures the dynamic cumulative effect of coordinated agglomeration, indicating that its shaping of green development is a continuous process whose benefits require time to materialize and accumulate, providing empirical support for the long-term nature of policies.

(3)

Winsorization: All variables were winsorized at the 5th percentile to correct for extreme values. The results show that the coefficient signs remain consistent with the baseline, proving that the baseline conclusions are not driven by extreme values and confirming that the core findings are not driven by a few outlier observations, enhancing the reliability of statistical inference.

5. Discussion

5.1. Key Findings and the U-Shaped Main Effect

Based on panel data from 41 cities in the Yangtze River Delta (2011–2023), this study identifies a significant U-shaped relationship between manufacturing–services co-agglomeration and urban Green Development Efficiency (GDE). The main findings are as follows:

(1)

Nonlinear impact: Co-agglomeration initially suppresses GDE, then promotes it after a threshold, confirming a U-shaped pattern.

(2)

Mediating mechanisms: Industrial advancement acts as a strong and direct mediator, while green technological innovation exhibits a significant but lagged mediating effect. Energy consumption intensity shows no statistically significant mediation in the current model.

(3)

Contextual heterogeneity: The U-shaped relationship is more pronounced in cities with lower agglomeration levels, non-low carbon pilot cities, and non-resource-based cities.

5.2. Interpretation of the U-Shaped Nexus

The U-shaped trajectory extends traditional agglomeration theory by highlighting a dynamic transition from cost-dominated to benefit-dominated phases. In the early stage, the geographical proximity of manufacturing and producer services does not immediately yield green benefits. Instead, coordination costs—such as the realignment of production processes, environmental standard harmonization, and the upfront investment in shared green infrastructure—create a net drag on GDE. This aligns with institutional and evolutionary economic geography, which emphasizes the initial adjustment burdens of spatial-industrial reorganization [37]. Concurrently, congestion effects—rising factor prices and localized pollution concentration before collective treatment facilities are fully operational—further suppress efficiency.

Once co-agglomeration crosses a critical threshold, synergistic green externalities begin to dominate. Scale economies in pollution control, deepened cross-sector knowledge recombination, and the emergence of green-oriented market signals collectively push GDE upward. This threshold effect underscores that the green dividends of industrial coordination are not automatic but require cumulative investment, time, and sufficient agglomeration density to materialize.

5.3. Mechanism Analysis

Our mediation tests reveal distinct transmission pathways, offering a more nuanced understanding of how co-agglomeration affects GDE.

(1)

Industrial advancement (ts) is the most robust mediator. Co-agglomeration drives the regional economic structure toward a more service-oriented, high-value-added composition, which inherently reduces pollution intensity and raises resource productivity. This finding is consistent with studies highlighting structural transformation as a primary engine of green growth [1,26].

(2)

Green technological innovation (inn) shows a significant but lagged mediating effect. The U-shaped impact on innovation itself suggests that a certain level of agglomeration maturity is needed to trigger knowledge spillovers. Moreover, the delayed translation of innovation into efficiency gains reflects the time required for technology diffusion and absorptive capacity building [3,37]. This aligns with the Porter Hypothesis, which posits that environmental regulation can spur innovation, yet its full benefits emerge over time.

(3)

Energy consumption intensity (eff) did not show a significant mediation effect in our model. This may indicate that in a developed region like the Yangtze River Delta, the dominant green transition mechanism has shifted from incremental energy savings to deeper structural and technological change. Alternatively, energy efficiency gains might be offset by rebound effects or captured through other channels such as industrial upgrading.

5.4. Contextual Heterogeneity and Boundary Conditions

The heterogeneity analysis reveals that the U-shaped relationship is not universal but is strongly conditioned by local context, which carries important theoretical and policy implications.

(1)

Agglomeration Degree: The attenuated or insignificant U-shape in high-agglomeration cities suggests the possibility of diminishing marginal green returns or the emergence of congestion effects, hinting at an optimal range for coordinated agglomeration [15].

(2)

Low-carbon Pilot Policy Status: The absence of a significant U-shaped relationship in pilot cities is a pivotal finding. This likely reflects the stringent, front-loaded environmental standards that raise initial coordination costs and delay the realization of green benefits—consistent with the “short-term cost” aspect of the Porter Hypothesis [43]. It does not imply policy failure but may indicate a different, potentially more profound restructuring trajectory compelled by the pilot policy, which embeds green criteria from the outset. This presents a testable hypothesis for future longitudinal research.

(3)

Resource Endowment: The complete insignificance of the U-shaped relationship in resource-based cities starkly illustrates the powerful “carbon lock-in” effect and path dependency. It demonstrates that without concerted efforts to diversify economic structures, spatial coordination alone may reinforce existing high-carbon industrial trajectories rather than catalyzing a green transition [5].

6. Conclusions and Policy Implications

6.1. Theoretical and Empirical Contributions

This study demonstrates that the impact of manufacturing–services co-agglomeration on urban Green Development Efficiency follows a nonlinear and context-dependent U-shaped pathway. The main theoretical contribution lies in moving beyond linear debates about agglomeration externalities, showing that the green effect evolves from a cost-dominated phase to a synergy-dominated one only after a critical threshold is reached.

Empirically, we highlight the distinct roles of industrial advancement as an immediate mediator and green technological innovation as a lagged one. Furthermore, we identify key boundary conditions: the U-shape is more evident in cities with lower agglomeration, non-pilot status, and non-resource-based profiles. These findings underscore that the same spatial industrial policy can yield divergent green outcomes depending on local structural and institutional conditions.

6.2. Differentiated Policy Recommendations

Given the heterogeneous effects, we propose the following targeted policy measures:

(1)

For cities on the left side of the U-curve (low agglomeration, initial stage): provide green coordination subsidies to offset upfront environmental adaptation costs; invest in shared green infrastructure to lower entry barriers; foster cross-industry collaboration platforms to accelerate learning and trust-building.

(2)

For cities on the right side of the U-curve (higher agglomeration, synergy stage): encourage the development of green technology alliances and joint R&D centers; promote green branding and certification schemes to enhance market recognition; optimize agglomeration quality to avoid congestion and over-intensification.

(3)

For low-carbon pilot cities (stricter environmental regulations): establish special green transformation funds to mitigate early-stage policy-induced costs; facilitate knowledge spillovers from pilot to non-pilot cities through regional cooperation networks; monitor long-term green performance to capture delayed benefits of structural restructuring.

(4)

For resource-based cities: prioritize industrial diversification policies alongside agglomeration strategies; offer fiscal incentives for green service industries and advanced manufacturing; implement tailored environmental standards that encourage gradual decarbonization.

(5)

To strengthen mediating mechanisms: create “industrial upgrading guidance funds” to support high-end green services; build regional green technology transfer platforms to shorten innovation diffusion lags; incorporate green innovation indicators into local government performance evaluations.

6.3. Limitations and Future Research

This study has several limitations that point to future research directions. First, while we employ lagged variables and instrumental-variable approaches, causal inference could be further strengthened by leveraging quasi-experimental designs or richer micro-level data. Second, the analysis focuses on prefecture-level cities; future work could examine county- or firm-level dynamics to uncover micro-mechanisms. Third, spatial spillover effects between cities are not modeled; spatial econometric techniques could reveal how co-agglomeration in core cities influences neighboring regions’ GDE. Finally, expanding the framework to include moderating variables—such as digital economy development, environmental regulation intensity, or financial support policies—could help explain variations in the U-curve’s slope and turning point across different institutional settings.