Place-Based Policy, Industrial Coagglomeration, and Urban Carbon Productivity: Evidence from the Establishment of China’s National New Zones (NNZs)

Abstract

Under the constraints of carbon peaking and carbon neutrality, in order to facilitate the realization of SDGs in China’s cities, place-based policy needs to strike a balance between “economic growth” and “carbon reduction”. This paper considers the establishment of NNZs as a quasi-natural experiment and constructs an asymptotic DID model based on the data of 283 Chinese cities in 2006~2021 to assess the causal effects of place-based policy on urban carbon productivity. It is found that the establishment of NNZs significantly enhances urban carbon productivity, and this conclusion still holds after considering the validity, endogeneity, and robustness of the model. Mechanism analysis shows that the policy preferences of tax incentive and improving transport and other infrastructural facilities, the policy supervision and the industrial coagglomeration are the positive moderating mechanisms of the establishment of NNZs to enhance urban carbon productivity, but in addition to the policy preference of financial subsidy. Moreover, under the moderating effect of industrial coagglomeration, the establishment of NNZs enhances urban carbon productivity through three mechanisms: deepening of the specialized division of labor, optimizing of industrial structure, and innovating synergistically. Heterogeneity analysis showed that the moderating effect of industrial coagglomeration on urban carbon productivity is heterogeneous depending on the spatial layout of NNZs, in terms of planning area, the effective range is 1000~2000 km² and the optimal range is 1500~2000 km², and as far as the layout pattern is concerned, the optimal pattern is the dual-city layout. The conclusions provide a realistic basis and direction of thinking for optimizing the policy design of NNZs and promoting the green transformation of place-based policy.

1. Introduction

The steady implementation of the SDGs set by the United Nations to strengthen global cooperation and promote sustained economic prosperity, social justice, and harmony, as well as global environmental security, is a world consensus and a need for China. By 2023, China’s carbon dioxide emissions will have risen from 1418.5 million tonnes in 1978 to 11,549.8 million tonnes (Data from the Statistical Yearbook of World Energy 2024). In order to reduce the heavy environmental costs behind China’s economic “growth miracle”, China made a solemn pledge at the 75th General Debate of the United Nations General Assembly to strive for “carbon peaking” by 2030 and “carbon neutrality” by 2060. The report of the 20th Party Congress also explicitly proposed to actively and steadily promote carbon peaking and carbon neutrality, and to coordinate the promotion of carbon reduction, pollution reduction, greening, and growth. Entering the new stage of development, effectively integrating the relationship between “economic growth” and “carbon reduction” is an inherent requirement for promoting high-quality development. Place-based policy implemented in a specific geographic space is an important basis for leading China to achieve a balance between economic growth and environmental protection in a “point to area” manner. As a high-level evolution of place-based policy, since the establishment of the first NNZ in 1992, the State Council has approved the establishment of 19 NNZs, aiming to create important growth poles to drive the high-quality development of the regional economics. In March 2024, the NDRC issued “Action Plan for High-Quality Construction of NNZs”, which calls for the creation of NNZs to be the leading areas for high-quality development. As a major place-based policy for economic growth and social welfare [1], can NNZs lead the “30·60” dual-carbon vision to be realized on schedule without depleting economic growth? The key lies in the institutional attributes of policy preference and policy supervision, as well as the spatial attributes of industrial coagglomeration. It is worth noting that industrial coagglomeration is different from industrial cluster that focuses on the spatial distribution pattern of individual industries and emphasizes the dependence, connection, and interaction between manufacturing and productive service industries under the premise of geographic proximity [2]. In December 2019, the General Office of the State Council issued the “Guiding Opinions on Supporting NNZs to Deepen Reform and Innovation and Accelerate the Promotion of High-Quality Development” and explicitly stated that it promotes the deep integration and development of advanced manufacturing and modern service industries and cultivates a number of characteristic industrial clusters. Therefore, effectively evaluating the policy effect of the establishment of NNZs on urban carbon productivity under the key moderating role of industrial coagglomeration can provide replicable and scalable schemes and experiences for other place-based policies to promote sustainability development in cities where economic growth is absolutely “decoupled” from carbon emissions.

Currently, the assessment of the carbon productivity enhancement effect of place-based policy is mainly based on pilot free trade zones and various types of development zones [3], neglecting to assess NNZs as a high-level place-based policy. Moreover, studies have focused on assessing the economic growth effect of the establishment of NNZs, arguing that the establishment of NNZs has continued to drive the economic growth of the local city. For example, Cao [4] finds that the establishment of NNZs can lead to a cumulative increase in the real GDP growth rate of a city by about 10.570%, while Zhang and Zhao [5] find that the lifting effect of the establishment of NNZs on the urban total factor productivity (TFP) can be strengthened over time, while the spillover effect on the neighboring urban TFP is immediate. Moreover, some existing studies have confirmed the economic growth effects of the establishment of NNZs on cities by identifying the policy effects of the establishment of NNZs on micro-enterprises, such as attracting firms to enter and balancing the layout of firms [6]. When analyzing the pathways through which the establishment of NNZs affects the economic growth of cities, scholars also found that the “policy dividend” released by the establishment of NNZs has remarkable effects on the factor allocation [7], industrial structure [8], technological innovation [7], income distribution [7], and energy consumption [9] of the local cities, and the above factors are also regarded as the key to curbing the carbon emissions of the enterprises in the zones, and the local cities [3,10], but there is little literature to extend the discussion in this direction and assess the effect of the establishment of NNZs on carbon reduction or the synergies of “economic growth” and “carbon reduction”. For example, one study found that the agglomeration and rational allocation of factors is an important mechanism for the establishment of NNZs to drive the economic growth of cities [7]; in fact, the agglomeration and rational allocation of innovation factors can also control the cost of green innovation through externalities such as knowledge diffusion, technology sharing, and resource compensation, which thereby can inhibit the urban carbon emissions. Another study evaluates the policy effect of the establishment of NNZs on green TFP, which can balance the relationship between economic growth and environmental protection [11] but does not further clarify its internal mechanism. NNZs as a spatial carrier of industrial geographic concentration and deep integration, industrial coagglomeration formed by manufacturing and productive service industries through forward, backward, and sideways linkages, is a key moderating mechanism for NNZs to enhance urban carbon productivity. However, Wang and Sun [12] found that the over-concentration of manufacturing and productive service industries in a limited space is prone to lead to negative industrial cohesion effects such as “crowding effect”, “locking effect”, and “rebound effect”. Then, does the moderating effect of industrial coagglomeration on urban carbon productivity enhancement vary according to the spatial layout of the NNZs? In the dynamic game of positive and negative externalities of industrial coagglomeration, according to spatial economics theory, a reasonable planning area can alleviate the above negative externalities to a certain extent. Furthermore, compared with “one city, one district”, the layout mode of “two cities, one district” can help to consolidate the synergistic relationship between cities, and to enhance the carbon productivity of cities. Furthermore, compared with the pattern of a single-city layout, the pattern of a dual-city layout can consolidate the synergistic relationship between cities [4], which can make the positive externality of industrial coagglomeration dominant in the dynamic game. Obviously, different spatial layouts of the NNZs may lead to heterogeneity of the moderating effect of industrial coagglomeration on the improvement of urban carbon productivity.

To sum up, the possible marginal contributions of this paper include the following: firstly, it evaluates the policy effects of the establishment of NNZs on urban carbon productivity that balance “economic growth” and “carbon reduction”, in an attempt to enrich the evidence for the evaluation of “economic–environmental synergies” of place-based policy. Moreover, as an important growth pole that radiates and drives the high-quality development of regional economics, the identification of the carbon productivity enhancement effect of NNZs emphasizes the evolution direction of place-based policy in the new development stage. Secondly, based on the perspective of institutional and spatial attributes, this article explores the moderating mechanisms of the establishment of NNZs to enhance urban carbon productivity: the policy preferential mechanism, policy supervision mechanism, and industrial coagglomeration mechanism, and further analyzes the deductive logic of the establishment of NNZs to strengthen the empowerment through industrial coagglomeration, including the effect of a specialized division of labor, the effect of the optimization of industrial structure, and the effect of synergistic innovation, providing empirical evidence for cracking the bottleneck of non-low-carbon economic growth of place-based policy. Thirdly, in view of the negative externalities of excessive geographical concentration of industries, the heterogeneity of the effects of industrial coagglomeration on urban carbon productivity enhancement in the NNZs with different spatial layouts is examined, which provides the basis of decision for the central government and local governments to issue and optimize place-based policy according to local conditions.

2. Theoretical Mechanisms and Research Hypotheses

As “national” place-based policy, the NNZs can empower cities to increase carbon productivity through prominent institutional and spatial attributes. The institutional attribute refers to the policy preferences of tax incentive, financial subsidy, and improving transport and other infrastructural facilities, as well as policy supervision. The spatial attribute refers to cross-industry coagglomeration under the premise of geographical proximity.

Firstly, there exists a moderating mechanism of policy preference. Taking tax incentive as an example, NNZs are given preferential treatment such as enterprise income tax exemption and VAT credit rebate, which reduces the tax wedge and eases the financing constraints. On the one hand, it provides “trial and error space” for enterprises to shift to low-carbon and environmentally friendly production chains, and it increases the possibilities of controlling carbon emissions at the source. On the other hand, it shares the cost of research and development of pollution treatment technologies; in other words, it reduces carbon emissions by means of end-of-pipe treatment. As a result, tax incentive is conducive to the enhancement of urban carbon productivity. Taking financial subsidy as an example, since it is essentially “unproductive rent-seeking behavior” [13], financial subsidy allocated by NNZs is prone to fail to enter or enter incorrectly into the low-carbon production of enterprises due to the possible bad motives of the entrepreneurs or asymmetry of information, which results in a distorted or wasteful allocation of green resources. In addition, additional financial subsidy may also induce or exacerbate the “inertia” of enterprises to independently research and develop low-carbon and environmentally friendly technologies, which thereby may weaken the positive effect of the establishment of NNZs on the urban carbon productivity. Taking the improvement of transport and other infrastructural facilities as an example, Xiong-an NNZ has planned to build a railway network of “four verticals and two horizontals” and a highway network of “four verticals and three horizontals” since 2018, which is not only conducive to cracking the spatial access bottleneck of the city, extending and accelerating the distance and speed of green knowledge spillover, and innovating energy-saving and emission-reduction technologies for enterprises, but also helping to reduce the cost of cooperation and transaction costs for enterprises, optimizing the flow of green factors, correcting the mismatch of green resources, and thus enhancing the effect of the establishment of NNZs in enhancing urban carbon productivity.

Secondly, there exists a moderating mechanism for policy supervision. Central supervision and autonomous decision-making are the prominent institutional designs for the policy of NNZs. As far as central supervision is concerned, as comprehensive functional zones that undertake major national development and reform and opening-up strategies, the economic activities of enterprises in NNZs must be premised on compliance with the central government’s will to regulate the environment. In terms of autonomous decision-making, the “right of early and pilot implementations” and sub-provincial economic management authority granted to NNZs have increased the space for local governments and management committees to make autonomous decisions according to local realities, which not only specifies the central government’s will to regulate the environment but also avoids friction and delays between the central and local governments, and is conducive to the reduction of urban carbon emissions. Therefore, the institutional design of central regulation and autonomous decision-making can not only enhance the firmness of enterprises in the region to implement environmental regulations but also stimulate their initiative to do so. Meanwhile, existing studies have verified that in the process of reducing carbon emissions per unit of output value of enterprises, environmental regulation also reduces carbon emissions per unit of output value of cities through spillover effects [3]. It can be seen that the policy supervision mechanism can positively moderate the enhancement effect of the establishment of NNZs on urban carbon productivity.

Finally, there exists a moderating mechanism for industrial coagglomeration. After the establishment of NNZs, the package of policy preference has encouraged a large number of enterprises to enter [6], which fulfils the precondition of industrial coagglomeration—geographical concentration—and Zhang and Jin [8] also point out that NNZs are the spatial carriers for the in-depth connection, interaction, and integration of manufacturing and service industries. According to the positive externality theory, industrial coagglomeration may play a moderating role through two aspects: enterprise productivity improvement and manufacturing value chain upgrading. Firstly, enterprise production efficiency is improved. In terms of cost effect, manufacturing enterprises can not only reduce the cost of searching for high-quality labor with the function of the labor reservoir of the productive service industry but also entrust non-core business to supporting the service industry to reduce the production cost of enterprises, so as to improve the marginal output of the core production link through factor reallocation. In terms of the learning effect, industrial cohesion provides manufacturing enterprises with specialized learning opportunities and improves the output of enterprises with the same amount of factor inputs through the “learning-by-doing” effect. As the productivity of manufacturing enterprises improves, the energy factor input per unit of output continues to decline [14], and funds reserved for clean technology research and development will continue to accumulate, which is conducive to unleashing the carbon reduction potential of enterprises, thus promoting urban carbon productivity. Second, manufacturing value chain upgrading: The forward, backward and sideways correlation mechanism of industrial coagglomeration can crack the risk of the value chain being “locked at the low end”, prompting the manufacturing and productive service industries to extend to the upper reaches of the value chain [15], forming low-carbon and environmentally friendly industrial clusters, and thus enhancing the effect of the establishment of NNZs in improving urban carbon productivity, such as Jinpu NNZ which has built a new energy automobile industry cluster under the role of industrial coagglomeration. To sum up, this paper infers the following:

H1. 

The policy preferences of tax incentive and improving transport and other infrastructural facilities, the policy supervision and the industrial coagglomeration positively moderated the effect of the establishment of NNZs on urban carbon productivity, while the policy preference of financial subsidy weakened the effect of the establishment of NNZs.

Unlike the natural evolution of the market mechanism, the industrial coagglomeration of NNZs is formed under the joint effect of “autonomous agglomeration” of enterprises and “guided agglomeration” of the government. Thus, this article focuses on the moderating role of industrial coagglomeration, and further analyzes how the establishment of NNZs empowers the enhancement of urban carbon productivity. Theoretically, it can be summarized by the following three effects:

Specialized division of labor effect: As a supplement to the old urban areas naturally formed under the traditional economy, NNZs are “new” in the sense that they expand the space for factor agglomeration, division of labor, and specialization. As a matter of fact, industrial coagglomeration is the key to releasing the optimal effect of specialized division of labor in NNZs. Under the role of manufacturing and productive service industry coagglomeration, NNZs will upgrade the original factors’ primary division of labor and specialization to semi-finished products or finished products’ intermediate and advanced division of labor and specialization, which can offset the transaction costs generated by the division of labor on the cannibalization of the division of labor economic utility and deepen the specialization of the division of labor, so as to improve the structure of the allocation of resources among enterprises and to eliminate the protection effect of allocation distortion on high-pollution, high-energy-consumption enterprises [16]. In other words, the deepening of the specialized division of labor in NNZs can guide the orderly flow of green factors while mitigating their degree of dispersion, correctly filling the allocation gaps of low-carbon environmental protection enterprises in terms of green knowledge, technology, capital, etc., and thus enhancing the carbon emission performance of the enterprises. Therefore, under the positive moderating effect of industrial coagglomeration, the establishment of NNZs can enhance urban carbon productivity through the correction of green resource mismatch by the specialized division of labor effect.

Industrial structure optimization effect: In order to prevent the elimination of the last position, the NNZs driven by the “superior complementary effect” tilt the policy preferences to technology-intensive, low-carbon, and environmentally friendly enterprises that receive key support from the central government, which attracts a large number of target enterprises to enter NNZs. At the same time, under the “survival of the fittest” competition mechanism, the target enterprises will have a “crowding-out effect” [17] on the polluting production enterprises, forcing them to carry out green technological innovation or green production chain innovation, and possibly forcing them to relocate or even withdraw from the market, thus boosting the optimization of the industrial structure of NNZs. In addition, the continuous improvement of industrial coagglomeration is conducive to further releasing the industrial structure optimization effect of the establishment of NNZs. For example, in the course of interaction between the manufacturing and financial industries, while financial institutions request enterprises to disclose environmental information, they will also make use of financial science and technology to monitor the flow of finance of enterprises in the green projects in real time, thus urging them to actively carry out green transformation and upgrading. Liu and Sun [3] as well as Jing [18] have confirmed that the advanced and rationalized industrial structure is a main mechanism to improve urban carbon productivity. Therefore, the optimization and upgrading of industrial structure by the establishment of NNZs can accelerate the substitution of low-carbon and environmentally friendly enterprises for polluting production enterprises, especially the landing of third-party enterprises such as pollution detection and pollution control, prompting manufacturing enterprises to focus more on their production and operation activities, and thus helping to improve urban carbon productivity by reducing the carbon emissions per unit of output value of enterprises.

Synergistic innovation effect: At present, the imbalance in the ratio of green patent output to conversion is one of the constraints to the improvement of urban carbon productivity. However, synergistic innovation among enterprises, universities, research institutes, and other innovation bodies is a good remedy to alleviate this problem, and NNZs are fertile ground for the establishment of such a synergistic innovation model. The “right of early and pilot implementations” given to NNZs breaks the spatial limitation of innovation factors gathering, and at the same time, it also cuts down the systematic transaction cost and spatial collaboration cost of innovation main bodies and strengthens the willingness of innovation main bodies to synergize and innovate. In addition, the coagglomeration of manufacturing and productive service industries has led to the emergence of high-quality industry–university–research platforms, where according to the theory of knowledge flow, industrial coagglomeration is able to open up the blockages and difficulties between knowledge output sides such as colleges and universities, scientific research institutes, etc. and the knowledge input side such as enterprises, which is conducive to closing the faults of the synergistic innovation mode, expanding the dimension of the synergistic innovation mode, and further releasing the synergistic innovation effect of the establishment of NNZs. The above encourages enterprises can obtain breakthrough green technological innovations on the basis of effectively linking “research” and “production”, and then curb carbon emissions per unit of output value by improving production efficiency and reducing energy losses [18], thus boosting the improvement of urban carbon productivity. To sum up, this paper infers the following:

H2. 

The establishment of NNZs can enhance urban carbon productivity through the specialized division of labor effect, industrial structure optimization effect, and synergistic innovation effect, and industrial coagglomeration has a positive moderating effect on the above three mediating mechanisms.

3. Research Design

3.1. Model Building

Since the approval of NNZs can lead to differences between cities that have established NNZs and those that have not established NNZs during the study period as well as differences between cities where NNZs are located before and after their establishment, and because NNZs are approved at different times, thus with reference to the studies of Beck et al. [19] and Wang [20], we considered the establishment of NNZs as a quasi-natural experiment and constructed an asymptotic DID model. The incremental difference in urban carbon productivity before and after the policy shock is captured by double differencing between NNZs (treatment group) and non-NNZs (control group), controlling for individual heterogeneous disturbances that do not vary over time and unobservable disturbances that vary over time, so as to cleanly assess the policy effect of the place-based policy of NNZs on urban carbon productivity. The baseline model is set as follows:

$$C{P}_{it}={\beta }_0+{\beta }_{1}DI{D}_{it}+\lambda X_{it}+{\mu }_i+{\eta }_t+{\delta }_{ct}+{\epsilon }_{it}$$

(1)

In Equation (1), the subscripts i, c, and t denote city, province, and time, CP_it is the carbon productivity of the city i in year t, DID_it is a dummy variable for whether the city i establishes an NNZ in year t, X_it is the set of city-level control variables, μ_i is the city fixed effects, η_t is the time fixed effects, δ_ct is the province–time fixed effects, ε_it is the random error term, and β₁ is the coefficient of the estimated policy effect.

3.2. Description of Variables

Urban carbon productivity (CP): Measured by the ratio of urban GDP to urban carbon emissions (CE), in which the measurement of urban carbon emissions refers to the method of Chen et al. [21]; specifically, on the basis of unifying the scales of DMSP/OLS and NPP/VIIRS satellite imagery by using the PSO-BP algorithm, the top-down inversion of the carbon emissions of Chinese cities is performed by matching CO₂ emission data of China’s whole area with nighttime lighting data.

Dummy variable for NNZs policy (DID): If the city i has established an NNZ in year t (treatment group), then DID_it = 1; conversely, if the city i has not established an NNZ in year t (control group), then DID_it = 0.

Control variables (X): These include the following: Population density (pop), measured as the ratio of urban resident population to the land area of the administrative region. Economic development level (gdp), measured by the logarithm of the urban per capita GDP. Industrial structure (ind), measured by the share of value added of the tertiary sector in the urban GDP. Human capital level (edu), measured by the logarithm of the number of university students per 10,000 people in the city. Level of openness to the outside world (fdi), measured by urban actual utilization of foreign capital as a share of GDP. Financial credit level (cre), measured by the ratio of urban year-end loan balance of financial institutions to GDP.

3.3. Data Sources

Taking 283 Chinese cities in 2006–2021 as the research sample, and excluding Shanghai (as China’s first NNZ to be approved for establishment, Shanghai Pudong NNZ is endowed with five functional policies that cannot be compared or replicated by other NNZs, so Shanghai is not included in the research sample in this paper) and cities with administrative planning changes, the data are mainly obtained from the China Urban Statistical Yearbook, China Industrial Statistical Yearbook, China Tertiary Industry Statistical Yearbook, as well as the CSMAR database and CEADs database, and the information of patent applicants is obtained from the search system of SIPO. In order to eliminate inflationary disturbances, this paper deflates the price-based indicators with 2006 as the base period. At the same time, in order to avoid the inefficiency of the estimation results caused by the direct elimination of missing data, this paper applies linear interpolation to fill in the missing data, and finally obtains the balanced panel data for 283 Chinese cities for the period of 2006~2021. In addition, it was statistically found that the VIF of the variables had a maximum value of 2.59 and a mean value of 1.52, which is in line with the rule of thumb (<10), so the possibility of multicollinearity of the variables can be ruled out.

4. Empirical Tests

4.1. Benchmark Model Test

Table 1. Benchmark model test.

Variable	(1)	(2)	(3)
DID	0.086 ***	0.083 ***	0.078 ***
	(0.032)	(0.029)	(0.029)
Controls	NO	STIRPAT variables	All variables
City FE	YES	YES	YES
Time FE	YES	YES	YES
Province–time FE	YES	YES	YES
Obs	4528	4528	4528
Adjusted R2	0.976	0.983	0.983

Note: Robust standard errors for clustering at the city level are shown in parentheses. *** denotes significance at 1% statistical levels.

reports the estimated results of the policy effects of the establishment of NNZs on urban carbon productivity, where column (1) does not incorporate the control variables, column (2) incorporates the key variables in the STIRPAT model—pop, gdp, and ind—and column (3) further incorporates other variables that affect urban carbon productivity—edu, fdi, and cre. The results reflect that the estimated coefficients of the variable DID are all significantly positive at the 1% level, which indicates that the establishment of NNZs has a remarkable effect on urban carbon productivity. Specifically, the estimated coefficient of the variable DID in column (3) is 0.078, that is, compared with non-NNZs, the urban carbon productivity of cities with the approval of the establishment of NNZs has increased by about 7.800%, so that H1 is partially verified. The promotion effect of policy preferences and benefit tilts on factor allocation and industrial coagglomeration enjoyed by NNZs suppresses urban carbon emissions by improving the factor utilization efficiency including energy and the enterprise production efficiency and can offset the economic constraint effect of carbon emission reduction to a certain extent, thus leading to the enhancement of urban carbon productivity.

4.2. Validity Tests

4.2.1. Parallel Trend Test and Dynamic Effect Identification

The construction of an asymptotic DID model to assess the policy effects presupposes that the parallel trend assumption is satisfied; based on this, the paper applies event analysis to test. On the one hand, it examines whether the trends in the treatment and control groups are similar before the policy shock, and on the other hand, it identifies the dynamic effects of the establishment of NNZs on urban carbon productivity. The model is set up as follows:

$$C{P}_{it}={\beta }_0+{\beta }_k{{\displaystyle \sum }}_{k=-10}^{10}DI{D}_{it}^k+\lambda X_{it}+{\mu }_i+{\eta }_t+{\delta }_{ct}+{\epsilon }_{it}$$

(2)

In Equation (2), $DI{D}_{it}^k$ is a dummy variable for whether the city i in year t is k years away from the establishment of NNZs, with the superscript k = t − T_i, where Ti is the time when the city i establishes the NNZs, and $DI{D}_{it}^k$ = 1 when t − T_i = k, or 0 otherwise, and $DI{D}_{it}^k$ = 1 when t − T_i ≤ −10 or t − T_i ≥ 10, or 0 otherwise. Meanwhile, in order to avoid the interference of multiple covariations, this paper takes the first 1 year before the establishment of NNZs (k = −1) as a base period; therefore, the dummy variable $DI{D}_{it}^{-1}$ is not included in Equation (2). As can be seen from Figure 1, before the establishment of NNZs (k < 0), the estimated coefficients β_k all oscillate around the value of 0 and are not significant at the 10% level, so the trend of urban carbon productivity in NNZs and non-NNZs is similar, which fulfills the hypothesis of an ex ante parallel trend, i.e., the effect of the establishment of NNZs on urban carbon productivity is not a result of the ex ante inter-group differences. Then, 9 years after the establishment of NNZs, the estimated coefficient β_k increases significantly and is significant at least at the 5% level, which indicates that the establishment of NNZs has a long-term dynamic enhancement effect on urban carbon productivity. However, from the 10th year after the establishment of NNZs, the enhancement effect may disappear. Therefore, China needs to give special attention to the sustainability of the policy effects of NNZs in the future and comply with the promotion principle of “improving quality and expanding capacity”, so as to avoid the bad impacts of “replicating incremental” on the “policy dividend” and the “reform dividend”.

Note: In Figure 1, the solid points are the estimated coefficients β_k, and the short vertical lines are 95% confidence intervals. In Figure 2, the hollow points are p-values and the curves are the probability density lines.

4.2.2. Placebo Test

In order to avoid the policy effect of the establishment of NNZs on urban carbon productivity being disturbed by unobservable omitted variables, this paper will conduct a placebo test, i.e., randomly matching the cities in the sample according to the number of NNZs established in each year and repeating the process 1000 times. From Figure 2, the estimated coefficients of the variable DID under the pseudo-policy shocks (p-values are mostly larger than 0.1) are mainly distributed around 0 and are quite different from the estimated coefficients under the actual policy shocks (0.078), which suggests that the effect of the establishment of NNZs on the improvement of the urban carbon productivity is not caused by the unobservable omitted variables.

4.3. Endogenous Treatment

Considering that the benchmark estimation results may be disturbed by endogeneity issues such as measurement error, omitted variables, two-way causation, and sample selection bias, this paper applies the PSM-DID model to mitigate the process. In this paper, all control variables are used as covariates, and the control group is re-matched to the treatment group according to one-to-one nearest-neighbor matching, kernel matching, and radius matching, and DID estimation is performed again while satisfying the premise of “data balance”. From the results reported in columns (1) to (3) of

Table 2. Endogenous treatment.

Variable	Nearest-Neighbor Matching	Kernel Matching	Radius Matching
	(1)	(2)	(3)
DID	0.069 **	0.073 ***	0.077 ***
	(0.030)	(0.020)	(0.029)
Obs	4440	2879	4462
Adjusted R2	0.983	0.987	0.983

Note: Each column incorporates all control variables, and each column controls for city FE, time FE, and province–time FE. The following tables are identical. Robust standard errors for clustering at the city level are shown in parentheses. ** and *** denote significance at 5% and 1% statistical levels.

, the estimated coefficients of the variable DID are significantly positive at least at the 5% level. This indicates that there are no serious systematic differences between the treatment and control groups; in other words, the benchmark estimation results are not distorted due to endogeneity problems caused by sample selection bias.

4.4. Robustness Analysis

4.4.1. Exclusion of Contemporaneous Policies

Liu et al. [10] have pointed out that innovation-driven policies have obvious constraints on urban carbon emissions. So as to “cleanly” assess the policy effects of the establishment of NNZs on urban carbon productivity, this paper treats the innovative cities (ICs) pilot policy implemented in 2008, along with the low-carbon cities (LCCs) pilot policy implemented in 2010 and the carbon emissions trading (CET) pilot policy implemented in 2013, as dummy variables in the benchmark model for separate estimation. From the results reported in columns (1) to (3) of

Table 3. Robustness tests.

Variable	Exclusion of Contemporaneous Policies			Mitigate the Non-Randomness of Policy Choices
	(1)	(2)	(3)	(4)	(5)	(6)
DID	0.069 **	0.076 ***	0.078 ***	0.048 *	0.071 **	0.078 ***
	(0.027)	(0.028)	(0.029)	(0.026)	(0.030)	(0.029)
ICs pilot policy	0.055 ***
	(0.016)
LCCs pilot policy		0.025
		(0.020)
CET pilot policy			0.157 ***
			(0.058)
L–M City	NO	NO	NO	YES	NO	NO
MP City	NO	NO	NO	NO	YES	NO
HHY-E City	NO	NO	NO	NO	NO	YES
Obs	4528	4528	4528	4528	4528	4528
Adjusted R2	0.983	0.983	0.983	0.983	0.983	0.983

Note: Robust standard errors for clustering at the city level are shown in parentheses. *, **, and *** denote significance at 10%, 5%, and 1% statistical levels.

, the estimated coefficients of the variable DID change but are still significant after the exclusion of contemporaneous interference policies, indicating that the interference of contemporaneous policies on the estimated coefficients is very limited, and that the effect of the establishment of NNZs on the enhancement of urban carbon productivity still exists significantly. Specifically, after excluding the effects of the ICs pilot policy and the LCCs pilot policy, the estimated coefficients of the variable DID decrease to 0.069 and 0.076, respectively, and the estimated coefficient of the variable DID is still 0.078 after excluding the effects of the CET pilot policy. This indicates that the implementation of the ICs pilot policy and the LCCs pilot policy on NNZs can enhance the effect of NNZs on urban carbon productivity, but the superimposed implementation of the CET pilot policy and NNZs policy does not produce obvious effects. Therefore, in the future, the government can consider implementing the ICs pilot policy or the LCCs pilot policy superimposed on the NNZs, so as to enhance the policy effect of the establishment of NNZs through the technological progress effect or the emission reduction effect.

4.4.2. Mitigate the Non-Randomness of Policy Choices

The ideal premise for constructing an asymptotic DID model to evaluate the policy effect is that the selection of the treatment and control groups is completely randomized. Although this paper has added observable city characteristic variables and controlled for fixed effects in the benchmark model, and passed the PSM-DID test, in order to further validate the reliability of the benchmark results, this paper employs the interaction term between the city non-random factors and the time trend and the IV-DID method to conduct a re-test.

Method 1: In the process of approving the establishment of NNZs, the State Council is bound to follow the principle of preferential selection considering that it is a comprehensive functional zone with “right of early and pilot implementations”. Therefore, this paper incorporates the interaction term between non-random urban factors and the time trend into the benchmark model to mitigate the estimation bias caused by the non-random selection of treatment groups, and the specific settings are as follows:

$$C{P}_{it}={\beta }_0+{\beta }_{1}DI{D}_{it}+{\beta }_2{Z}_i\times tren{d}_t+\lambda X_{it}+{\mu }_i+{\eta }_t+{\delta }_{ct}+{\epsilon }_{it}$$

(3)

In Equation (3), Z_i is a dummy variable for the non-random factors of city i, including whether it is a large or medium-sized city (L–M City), whether it is a modern port city (as a place where Western scientific ideas have been imported and the economy has been active since modern times, the ports of commerce have a good foundation for “right of early and pilot implementations”) (MP City), and whether it is a city on the east side of the Hu Huanyong Line (HHY-E City), and trend_t is a time trend term. The results in columns (4) to (6) of Table 3 reflect that the estimated coefficients of the variable DID are remarkably positive at least at the 10% level. This indicates that the benchmark estimates remain robust after controlling for the interference over time of non-random factors in policy choices.

Method 2: The inverse of the interaction term between urban topographic relief and the time trend is selected as the instrumental variable (IV), based on the following reasons: in the principle of exogeneity, topographic relief is a naturally occurring geographic factor, which does not have a direct effect on urban carbon productivity. In the principle of relevance, the cities with relatively flat topographic relief basically belong to regional center cities with high levels of resource concentration, industrial integration, and scientific and technological innovation. In addition, under the effect of the “preference effect”, the flatter the urban topography is, the higher the possibility that the city will be approved as an NNZ. The results (just the coefficients for the key variables) show that IV is significantly and positively correlated with DID, the F statistic (74.183) is greater than 10, which is in accordance with the rule of thumb, the K-P rk LM statistic is significant at the 1% level, which rejects the hypothesis of under-recognition, the C-D Wald F statistic (142.622) is greater than the Stock–Yogo critical value (16.380), which rejects the hypothesis of weak recognition, and the estimated coefficient of the variable DID remains significantly positive at the 1% level. Therefore, the benchmark results remain reliable.

4.4.3. Consider the Choice of Proxy Variables

So as to avoid estimation bias caused by the choice of proxy variables, this paper carries out the following robustness tests. In terms of the explanatory variables, the first is to replace the measurement method of CE, with reference to the research of Xu and Wang [22], and use the international statistical standards SNA and SEEA as the basis, dividing the urban carbon emissions into production carbon emissions, consumption carbon emissions, and income carbon emissions and measured by applying input–output tools. Secondly, correcting outliers, we shrink the sample for the top and bottom 1% of urban carbon productivity. In terms of explanatory variables, taking into account the fact that NNZs have experienced a long-term process of central–local gaming [23] before the State Council publicly approved their establishment, so the time of releasing the policy effect of NNZs is often earlier than the time of the State Council’s public approval of the establishment. Based on this, the NNZs approved by the State Council in the first half of the year are regarded as being set up in the previous year, and those approved by the State Council in the second half of the year are regarded as being established in the current year. The results in columns (1) to (3) of

Table 4. Robustness tests.

Variable	Consider the Choice of Proxy Variables			Consideration of Expected Effect
	Replace CE	Correcting Outliers	Change of Establishment Time	Previous Year	Previous Two Years
	(1)	(2)	(3)	(4)	(5)
DID	0.099 ***	0.082 ***	0.074 ***	0.062 **	0.065 **
	(0.019)	(0.031)	(0.028)	(0.025)	(0.026)
DID_one				0.020
				(0.012)
DID_two					0.020
					(0.012)
Obs	4528	4528	4528	4528	4528
Adjusted R2	0.987	0.982	0.983	0.983	0.983

Note: Robust standard errors for clustering at the city level are shown in parentheses. ** and *** denote significance at 5% and 1% statistical levels.

show that the estimated coefficients of the variable DID are all remarkably positive at the 1% level. This indicates that the choice of proxy variables does not interfere with the benchmark estimation results.

4.4.4. Consideration of Expected Effect

If before the State Council’s public approval of the establishment of NNZs, the local government has formed a certainty expectation of the public approval of the establishment of the behavior, then it will certainly interfere with the exogenous nature of the policy shock. Thus, this article, based on the consideration of the central–local game, advances the time of the policy shock by another one year and two years, and incorporates its dummy variables DID_one and DID_two into the benchmark model, respectively. The results reported in columns (4) and (5) of Table 4 reflect that the estimated coefficients of the variables DID_one and DID_two are insignificant, and the estimated coefficient of the variable DID is still remarkable positive, which suggests that the expected effect does not bias the benchmark estimation results.

4.4.5. Consideration of Heterogeneous Treatment Effect

Considering that the establishment of NNZs is in batches, when constructing an asymptotic DID model to evaluate its policy effect on urban carbon productivity, even if the ex ante parallel trend assumption is satisfied, there may be heterogeneous treatment effects depending on the time of establishment, which may lead to estimation bias. In this regard, firstly, this paper refers to Goodman-Bacon’s method to decompose the coefficients [24], and it can be seen from panel A in

Table 5. Robustness tests.

Panel A: Bacon Breakdown
Group Type	Estimated Coefficient	WEIGHTS
Time-varying treatment group	0.008	0.015
Never-treatment group vs. time-varying treatment group	0.122	0.970
Between the two	−3.190	0.011
Panel B: Callaway and Sant’Anna robust estimates
Variable	No policy shocks at all	No policy shocks at all times and no policy shocks yet
	(1)	(2)
DID	0.065 *	0.065 *
	(0.036)	(0.036)
Obs	4528	4528

Note: Robust standard errors for clustering at the city level are shown in parentheses. * denotes significance at 10% statistical levels.

that the weight of the time-varying treatment group is only about 1.500%, which suggests that the heterogeneous treatment effect interferes little with the benchmark estimation results. Secondly, it refers to Callaway and Sant’Anna’s method to perform robustness estimation [25], where panel B of Table 5 shows that the estimated coefficients of the variable DID are all remarkable positive at the 10% level, whether the sample with no policy shocks at all times is used as a control group or the sample with no policy shocks at all times and no policy shocks yet as a control group, which once again corroborates the robustness of the results of the benchmark estimation.

5. Mechanism Test

5.1. Model Building

5.1.1. Moderating Effect Model Building

In order to test whether the enhancement effect of the establishment of NNZs on urban carbon productivity is moderated by policy preference, policy supervision, and industrial coagglomeration, this paper incorporates the interaction term $DI{D}_{it}\times H_{it}$ into the benchmark model, which is set up as follows:

$$C{P}_{it}={\gamma }_0+{\gamma }_{1}DI{D}_{it}+{\gamma }_2{H}_{it}+{\gamma }_{3}DI{D}_{it}\times H_{it}+\lambda X_{it}+{\mu }_i+{\eta }_t+{\delta }_{ct}+{\epsilon }_{it}$$

(4)

In Equation (4), H_it is a proxy variable for the moderating mechanism, as described below:

The proxy variables for the policy preference moderating mechanism are tax incentive (TI), financial subsidy (FS), and road density (RD). TI is measured as the reciprocal of the sum of the urban main business taxes and surcharges and VAT payable as a share of fiscal revenues. FS is measured as the urban financial expenditures as a share of GDP. RD is measured as the sum of the city’s mileage of highways, railways, and inland waterways as a share of the land area of the administrative region.

The proxy variable for the policy supervision moderating mechanism is environmental regulation (ER), with reference to the study by Shen Kunrong et al. [26], as measured by the weighted average of the urban industrial SO₂ removal rate and the industrial soot removal rate, where the weight is the ratio of pollutant emissions per unit of industrial output value of the city to the China-wide average.

Industry coagglomeration (COA): Since industrial coagglomeration is not a unidirectional agglomeration of the same kind of industries or a spatial concentration of different industries, but an industrial cluster complex formed by the interdependence, association, and interaction of different industries under the premise of geographic proximity [2], therefore, referring to the research methods of Tang et al. [27] and Zhang et al. [28], we constructed an evaluation index system for the coagglomeration of the manufacturing industry and productive services (including transport, storage and postal industry, leasing and business services industry, wholesale and retail trade industry, finance industry, information transmission, computer services and software industry, scientific research and technical services industry, and environmental governance and public facilities management industry) from four aspects: industrial scale, economic performance, growth potential, and social contribution, and we use the gray GM (1, N) model to measure the index of urban industrial coagglomeration. Among them, the GM (1, N) model is a gray model established by applying first-order differential equations to N variables, which determines the degree of coagglomeration of industries by calculating the degree of geometric similarity between the family of curves composed of evaluation sequences and the curves composed of reference sequences. The GM (1, N) model mitigates the disturbances of randomness and uncertainty in the industrial interaction relationship. The specific calculation process refers to the research of Tang et al. [27].

5.1.2. Model Building of the Mediating Mechanism Under the Moderating Effect of Industrial Coagglomeration

Further, in order to test how industrial coagglomeration moderates the mediating mechanisms for the establishment of NNZs to enhance urban carbon productivity, such as the specialized division of labor effect, industrial structure optimization effect, and synergistic innovation effect, this paper refers to the studies by Liu and Mao [29] and Kong et al. [30] to constructing the model. This is in consideration of the fact that, on the one hand, the step-by-step test method may suffer from endogeneity bias and poor identification of the mechanism [29], and on the other hand, the theoretical mechanism part only explores the moderating effect of industry coagglomeration on the first half of the path of the mediating mechanism, and does not deal with the second half of the path. The model is set up as follows:

$$M_{it}={\theta }_0+{\theta }_{1}DI{D}_{it}+\lambda X_{it}+{\mu }_i+{\eta }_t+{\delta }_{ct}+{\epsilon }_{it }$$

(5)

$$M_{it}={\phi }_0+{\phi }_{1}DI{D}_{it}+{\phi }_{2}CO{A}_{it}+{\phi }_{3}DI{D}_{it}\times CO{A}_{it}+\lambda X_{it}+{\mu }_i+{\eta }_t+{\delta }_{ct}+{\epsilon }_{it}$$

(6)

In Equations (5) and (6), COA_it is the proxy variable for the industry coagglomeration moderating mechanism, and Mit is the proxy variable for the mediating mechanism, as explained below:

Specialized division of labor: Measured by the ratio of the “number of people employed in productive services in the city i in year t ÷ number of people employed in manufacturing” to the “total number of people employed in productive services in the 283 cities in year t ÷ total number of people employed in manufacturing”.

Optimization of industrial structure: Measured by the ratio of the value added from the tertiary industry to the sum of the value added from the primary and secondary industries in the city.

Synergistic innovation: Measured by the proportion of urban industry–university–research combination patent (the patent search formula for the combination of industry–academia–research is as follows: applicant = (company OR factory OR society) AND (university OR college OR school OR (research academy NOT company) OR (research institute NOT company) OR (design academy NOT company) OR (design institute NOT company) OR (engineering academy NOT company) OR (center NOT company) OR hospital OR workstation)) applications to the total number of patent applications. This is in consideration of the fact that the essence of synergistic innovation lies in knowledge spillover and technology diffusion among enterprises, universities, research institutions, and other related innovation subjects, so we referred to the study of Jin et al. [31] to measure.

5.2. Moderating Effect Test

Table 6. Moderating effect test.

Variable	Policy Preference			Policy Supervision	Industrial Coagglomeration
	(1)	(2)	(3)	(4)	(5)
DID × TI	0.014 ***
	(0.005)
DID × FS		−0.007
		(0.038)
DID × RD			0.017 ***
			(0.006)
DID × ER				0.255 **
				(0.128)
DID × COA					0.483 ***
					(0.138)
Obs	4528	4528	4528	4528	4528
Adjusted R2	0.983	0.983	0.983	0.983	0.983

Note: Robust standard errors for clustering at the city level are shown in parentheses. **, and *** denote significance at 5%, and 1% statistical levels.

reports the estimated results of the moderating effects of policy preference, policy supervision, and industry coagglomeration on the establishment of NNZs to enhance urban carbon productivity. The results show that the estimated coefficients of the interaction terms of the variable DID with TI, RD, ER, and COA are significantly positive at the 1%, 1%, 5%, and 1% levels, respectively, but the estimated coefficients of the interaction term of the variable DID with FS fail the 10% significance test. Therefore, the policy preference of tax incentive and improving transport and other infrastructural facilities, the policy supervision, and the industrial coagglomeration positively moderated the enhancement effect of the establishment of NNZs on urban carbon productivity, and the positive moderating effects of policy supervision and industrial coagglomeration are significantly better, while the policy preference of financial subsidy weakened the enhancement effect. To sum up, H1 is verified. The package of policy preferences enjoyed by the NNZs reduces the finance, transaction, and production costs of enterprises and leaves room for them to improve carbon productivity. However, the policy preference of financial subsidy may fail to positively moderate the carbon productivity of enterprises due to factor mismatches or wastage caused by biased selection of targets and lack of supervision of the process. Furthermore, the positive interaction between central supervision and local “early and pilot implementations” is conducive to the NNZs to bring the pressure of “stabilizing economic growth” and “reducing carbon emission” down to the smallest unit of economic activities—enterprises. Finally, the in-depth integration of manufacturing and production service industries in close geographical proximity can not only avoid the problem of excessive energy depletion caused by the concentration of homogeneous enterprises but also further enhance the urban carbon productivity by deepening the division of labor, facilitating transformation and promoting innovation.

5.3. Test of the Mediating Mechanism Under the Moderating Effect of Industrial Coagglomeration

According to Liu and Mao [29], columns (1), (3), and (5) of

Table 7. Mediating mechanism test under the moderating effect of industry coagglomeration.

Variable	Specialized Division of Labor			Optimization of Industrial Structure		Synergistic Innovation
	(1)	(2)	(3)	(4)	(5)	(6)
DID	0.152 **	0.123 **	0.268 ***	0.216 ***	0.054 ***	0.042 ***
	(0.064)	(0.062)	(0.062)	(0.064)	(0.013)	(0.013)
COA		−0.413		−1.066 **		−0.194 **
		(0.270)		(0.480)		(0.087)
DID × COA		0.750 **		1.104 **		0.281 ***
		(0.377)		(0.545)		(0.104)
Obs	4528	4528	4528	4528	4528	4528
Adjusted R2	0.874	0.875	0.888	0.889	0.961	0.961

Note: Robust standard errors for clustering at the city level are shown in parentheses. ** and *** denote significance at 5% and 1% statistical levels.

test the mediating mechanisms of the establishment of NNZs to enhance urban carbon productivity, including the specialized division of labor effect, industrial structure optimization effect, and synergistic innovation effect, and the results display that the estimated coefficients of the variable DID are remarkably positive at least at the 5% level. Meanwhile, columns (2), (4), and (6) of Table 7 test the moderating effect of industrial coagglomeration on the first half of the path of the mediating mechanisms of the establishment of NNZs to enhance urban carbon productivity, and the results show that the estimated coefficients of the interaction term between the variable DID and COA are remarkable positive at least at the 5% level. Thus, it can be seen that industrial coagglomeration significantly enhances the lifting effect of the establishment of NNZs on the deepening of the specialized division of labor, the optimization of industrial structure, and synergistic innovation in the city. As explained in the theoretical mechanism section, under the moderation of industrial coagglomeration, the three mediating mechanisms inhibit the carbon emission per unit of output value of enterprises by correcting the distortion of green resource allocation, accelerating the substitution of low-carbon and environmental protection enterprises to the polluting production enterprises, and solving the problem of “research and production disconnection” of the imbalance in the ratio of green patents’ output and transformation, thus promoting urban carbon productivity. Therefore, H2 is verified.

6. Further Discussion

The spatial attribute is a prominent feature that distinguishes place-based policy from other policies, which, while prompting a high degree of aggregation of economic resources in the target area, may also lead to an over-concentration of industries in a limited space and cause industrial coagglomeration negative externalities such as the crowding effect, locking effect, and rebound effect [12]. As a matter of fact, the planning area and layout pattern of NNZs can offset these negative externalities by alleviating the rigid constraints of resource and environmental endowments and weakening administrative barriers. Then, in what area range or layout pattern can the positive externalities of industrial coagglomeration prevail in the dynamic game, and thus enhance the urban carbon productivity lifting effect of the place-based policy of NNZs? In this regard, this paper further examines the heterogeneity of industrial coagglomeration in moderating the urban carbon productivity lifting effect in NNZs with different spatial layouts.

6.1. Heterogeneity Test of Moderating Effect of Industrial Coagglomeration in Different Planning Areas

Considering that among the NNZs involved in the study sample, the smallest land area of Ganjiang NNZ is 465 km² and the largest one is 2451 km², this paper takes 500 km² as the base unit and divides the 18 NNZs into the following five categories: <500 km², 500~1000 km², 1000~1500 km², 1500~2000 km², and >2000 km².

Table 8. Policy heterogeneity test.

Variable	<500 km2	500~1000 km2	1000~1500 km2	1500~2000 km2	>2000 km2
	(1)	(2)	(3)	(4)	(5)
DID	0.112 ***	0.175 *	0.055 ***	0.116 ***	0.095 ***
	(0.039)	(0.095)	(0.018)	(0.027)	(0.024)
Obs	4272	4224	4208	4288	4240
Adjusted R2	0.983	0.983	0.983	0.983	0.983

Note: Robust standard errors for clustering at the city level are shown in parentheses. * and *** denote significance at 10% and 1% statistical levels.

tests the differential impact of the establishment of NNZs on urban carbon productivity within the different planning areas, and

Table 9. Heterogeneity test for moderating effect of industry coagglomeration.

Variable	<500 km2	500~1000 km2	1000~1500 km2	1500~2000 km2	>2000 km2
	(1)	(2)	(3)	(4)	(5)
DID × COA	−0.029	0.673	0.298 **	1.912 **	1.262
	(0.163)	(0.464)	(0.122)	(0.859)	(1.303)
Obs	4272	4224	4208	4288	4240
Adjusted R2	0.983	0.983	0.983	0.983	0.983

Note: Robust standard errors for clustering at the city level are shown in parentheses. ** denotes significance at 5% statistical levels.

tests the heterogeneity of the moderating effect of industrial coagglomeration on the establishment of NNZs to enhance urban carbon productivity. Combining the estimation results reported in Table 8 and Table 9, it is evident that when the planning area is <1000 km², industrial coagglomeration weakens the effect of the establishment of NNZs on urban carbon productivity. The possible explanations are, firstly, the “knowledge convergence” and “technology lock-in” caused by the over-concentration of industries in a limited space seriously constraining the release of positive spillover effects of industrial coagglomeration. Secondly, constrained by the carrying capacity of resources and the environment, the scale effect of industrial coagglomeration is easily replaced by the congestion effect. Thirdly, the population density and production density brought by the excessive concentration of industries have led to the surge in local energy demand, inducing the rebound effect of “reducing carbon emission” [12]. Therefore, the combination of the above “neighborhood paradox” leads to a weakening of the lifting effect of the establishment of NNZs on urban carbon productivity by industrial coagglomeration. When the planning area is from 1000~2000 km², industrial coagglomeration positively moderates the enhancement effect, especially when the planning area is 1500~2000 km² (1.912 > 0.298). A reasonable planning area of NNZs can reduce the search cost, spillover cost, and cooperation cost of enterprises, consolidate the collaborative relationship between upstream and downstream industries, and promote the positive externality of industrial coagglomeration to prevail in the dynamic game. When the planning area is >2000 km², industrial coagglomeration does not effectively enhance the urban carbon productivity lifting effect of the establishment of NNZs. The possible explanation is that too large a planning area will dilute the positive externalities of industrial coagglomeration, such as the spatial spillover effect of knowledge and technology, which is easily diminished by the increase in geographic distance, and thus industrial coagglomeration will not be able to play the expected moderating role.

6.2. Heterogeneity Test of Moderating Effect of Industrial Coagglomeration in Different Layout Patterns

Considering that the layout patterns of Xixian NNZ, Guian NNZ, Tianfu NNZ, and Ganjiang NNZ are dual-city layout, this paper classifies the 18 NNZs into two categories: single-city layout and dual-city layout.

Table 10. Policy heterogeneity test.

Variable	Single-City Layout	Dual-City Layout
	(1)	(2)
DID	0.105 ***	0.137 ***
	(0.020)	(0.046)
Obs	4400	4304
Adjusted R2	0.983	0.983

Note: Robust standard errors for clustering at the city level are shown in parentheses. *** denotes significance at 1% statistical levels.

examines the differential impacts of the establishment of NNZs on urban carbon productivity in different layout patterns, while

Table 11. Heterogeneity test for moderating effect of industrial coagglomeration.

Variable	Single-City Layout	Dual-City Layout
	(1)	(2)
DID × COA	0.310 **	1.061 **
	(0.124)	(0.529)
Obs	4400	4304
Adjusted R2	0.983	0.983

Note: Robust standard errors for clustering at the city level are shown in parentheses. ** denotes significance at 5% statistical levels.

examines the heterogeneity of the moderating effect of industrial coagglomeration on the establishment of NNZs in enhancing urban carbon productivity. Combining the estimation results reported in Table 10 and Table 11, it can be known that industry coagglomeration positively moderates the urban carbon productivity enhancement effect of the establishment of NNZs regardless of whether it is in the single-city or dual-city layout pattern, but compared with the single-city layout pattern, it exerts a better positive moderating effect in the dual-city layout pattern (1.061 > 0.310), with the existence of a “1 + 1 > 2” effect. The possible explanation is that the NNZs with a dual-city layout are usually led and coordinated by the provincial government, which to a certain extent eliminates administrative barriers, breaks the market blockade and local protection, and is conducive to clearing the institutional obstacles to industrial coagglomeration. Moreover, under the effect of “mutual borrowing of scale”, the dual-city layout of NNZs can multiply the release of positive externalities of industrial coagglomeration and effectively offset the negative externalities of excessive geographical concentration of industries.

7. Research Conclusions and Policy Recommendations

This paper regards the establishment of NNZs as a quasi-natural experiment and constructs an asymptotic DID model based on the data of 283 Chinese cities in 2006~2021 to investigate the causal effect and mechanism of place-based policy on the urban carbon productivity that can balance “economic growth” and “carbon reduction”, as well as further discussing the heterogeneity of industrial coagglomeration moderating mechanisms in different spatial layouts. The study finds that, firstly, NNZs can significantly enhance urban carbon productivity, which is still valid after validity tests, endogenous treatment, and robustness tests, but the sustainability of the enhancement is a key issue for future attention. Secondly, the policy preferences of tax incentive and improving transport and other infrastructure facilities, the policy supervision, and the industrial coagglomeration moderating mechanisms enhance the lifting effect of NNZs on urban carbon productivity; however, the policy preference of financial subsidy may undermine the lifting effect for objective reasons such as bad motives of entrepreneurs, information asymmetry, and innovation inertia. However, NNZs still face the challenges of gradual dilution of “policy dividends”, imperfect top-level institutional design, and serious industrial isomorphism, so how to optimize the regulating effects of policy preference, policy supervision, and industrial coagglomeration is also a key issue to focus on in the future. In addition, the specialized division of labor effect, industrial structure optimization effect, and synergistic innovation effect are effective mediating mechanisms through which the establishment of NNZs can enhance urban carbon productivity, and the positive moderating effect of industrial coagglomeration on these three mediating mechanisms can further strengthen the enhancement of urban carbon productivity. Thirdly, the differential impact of the spatial layout of NNZs on the dominant problem in the game of positive and negative externalities of industrial coagglomeration leads to the heterogeneity of the moderating effect of industrial coagglomeration on urban carbon productivity. In terms of the planning area, the reasonable range of the positive moderating effect played by industrial coagglomeration is 1000~2000 km², and the optimal range is 1500~2000 km². In terms of the layout pattern, compared with the single-city layout pattern, the positive moderating effect of industrial coagglomeration is better under the dual-city layout pattern.

To sum up, we make the following policy recommendations:

Firstly, following the principle of “increasing quantity” and “improving quality” in tandem and promoting place-based policy give full play to their demonstration effect on the enhancement of urban carbon productivity. Taking the policy of NNZs as an example, we suggest that the State Council, based on the city cluster plan, give priority to supporting Wuhan, the center city of the city cluster in the middle reaches of the Yangtze River, and Zhengzhou and Hefei, the center cities of the city cluster in the Central Plains, to declare the establishment of NNZs, so as to increase the proportion of spatial coverage and at the same time solve the imbalance in spatial distribution, as well as to lead the whole region of China in the implementation of dual-carbon goals. In addition, considering the dilution of “policy dividends” by the promotion of NNZs, it is recommended to further improve the supervision and assessment mechanism. Specifically, on the one hand, the optimal utility threshold of the “policy dividend” is calculated through a combination of experience accumulation and data simulation, so as to avoid the diminishing marginal utility caused by “replicating incremental”. On the other hand, environmental indicators such as carbon “double control” are incorporated into the performance appraisal mechanism of local officials, and at the same time, dynamic entry and exit mechanisms are implemented based on the effectiveness of the establishment of NNZs, so as to fully mobilizes the local government’s ability to coordinate “stabilizing economic growth” and “reducing carbon emission” and effectively curb carbon emissions per unit of city output.

Secondly, it is advised to open up feasible pathways and promote place-based policy to realize green and high-quality construction that takes into account both “economic growth” and “carbon reduction”. Taking the policy of NNZs as an example, on the one hand, it is recommended that the “policy dividend” of tax incentive should continue to be released through increased tax exemptions for energy-saving and emission-reduction technologies and sustainable development projects, and that the design of the system of benign central-territorial interaction should be optimized through an appropriate degree of decentralization. In addition, it is recommended to focus on reforming the distribution of financial subsidy, so as to reduce the impact of resource mismatches or wastage on the urban carbon productivity enhancement through the introduction of competitive mechanisms caused. On the other hand, one should spare no effort to play the moderating role of industrial coagglomeration, specifically, at first, according to the pillar industries or leading industries in the local area, recruiting other industries that can be aggregated with them, so as to build a pattern of industrial coagglomeration with “staggered development and complementary functions”, and to avoid resource mismatches and vicious competition caused by excessive overlapping of industries. In addition, efforts should be made toward driving industrial coagglomeration to couple with digital technologies and guiding the deep integration of manufacturing and productive service industries by reconstructing the technology chain, integrating the value chain, and expanding the industrial chain. Finally, efforts should be made to cultivate a unified market, to crack the obstacles to the free flow and rational allocation of factors arising from market segmentation caused by administrative barriers and price competition by improving the construction of cross-regional transportation infrastructure, such as high-speed rail links, and by building a system of openness of public data, promoting the qualitative change of industrial coagglomeration “from nothing to something” or “from something to something better”. This wouls be to enhance the urban carbon productivity enhancement effect of the establishment of NNZs by offsetting the devouring of the economic utility of the division of labor by transaction costs, regulating the high-carbon production behaviors of the polluting enterprises, and clearing up the blockage of the knowledge flow among the innovation subjects.

Thirdly, the spatial layout of the cities that are the subject of place-based policy should be reasonably planned to ensure that the positive externalities of industrial coagglomeration will prevail in the process of the dynamic game with negative externalities. Taking the policy of NNZs as an example, in terms of planning area, it is suggested that the State Council should cautiously approve the establishment of NNZs below 1000 km² and above 2000 km². For the established NNZs below 1000 km², the local government can consider transferring industries and replacing part of the “spatially concentrated” geographic coagglomeration with “organizationally close” virtual coagglomeration, such as building an industrial internet platform. In terms of layout pattern, it is recommended that local governments join hands with neighboring cities that have been overlapped by a number of policies, such as the low-carbon cities pilot policy and the smart cities pilot policy, to jointly declare the establishment of NNZs, so as to induce industrial coagglomeration to positively moderate the lifting effect of NNZs on urban carbon productivity by suppressing negative externalities of industrial coagglomeration such as the crowding effect, locking effect, and rebound effect.