Development zone policy and high-quality economic growth: quasi-natural experimental evidence from China

ABSTRACT Improving total factor productivity (TFP) promotes high-quality development. This study estimates the TFP of 284 cities in China for the period 2006–18, and constructs a time-varying difference-in-differences (DID) model of the impact of national development zones (DZs) on regional TFP. The findings are as follows: (1) although DZs improve economic performance, they have not created high-quality growth; (2) despite heterogeneous effects on high-quality development, the overall effect remains negative; and (3) the suppression effect is greater in central and western than in eastern cities, and in underdeveloped than in developed cities. Finally, policies should consider the quality of growth, and a framework of classified policy instruments is needed.


INTRODUCTION
Establishing development zones (DZs) and attracting foreign capital are common measures used by developing countries to increase employment, exports and economic growth. However, rapid expansion creates the dilemma of economic growth and environmental protection. Solving the problem of sluggish growth under high-quality development and environmental constraints has become an urgent issue at the political and academic interface. Since the reform and opening up, DZs have played an important role in rapid economic growth as an important institutional arrangement for the government to promote regional development, system reform and industrial layout. The early DZ policy is the experimental field of reform and opening up, undertaking the tasks of reform and opening up and of spurring institutional innovation. With economic development and the deepening of opening up to the outside world, DZs have become the main way to attract foreign investment and reasonably arrange different types of investment. Finally, they are also regarded as an important medium for promoting the free flow of factor capital and sustainable development.
To judge the effect of the economic growth of a country or region, we should not only focus on the growth rate, but also pay attention to its growth quality. The report of the 19th National Congress of the Communist Party of China states that we must adhere to quality first and prioritize efficiency; take structural reforms on the supply side as the main line; promote quality, efficiency and power change in economic development; and improve total factor productivity (TFP). The TFP level represents the efficiency of the comprehensive use of input factors and is an important indicator of economic quality. The effect of resource reallocation and capital remuneration gained through China's innate advantage in labour has gradually diminished, and it is difficult to sustain economic growth driven by capital and labour factor inputs. Therefore, as China enters a stage of high-quality economic development, it should further promote the three major changes in quality, efficiency and dynamics; support and lead industrial structure upgrading with scientific and technological innovation; increase the strength of independent innovation; make innovation a powerful engine for driving economic growth under the new normal; make innovation an important driving force for development; and realize the transformation from capital, land, labour and other production factors to efficiency-driven TFP.
In this context, it is of great theoretical value and practical significance to scientifically evaluate the policy effect and influence mechanism of DZs on regional development quality and to promote coordinated development. Therefore, the marginal contribution lies in the fact that from the perspective of research, the existing research on the impact of DZs on economic development and productivity has not reached a consistent conclusion. Based on the data of cities from 2006 to 2018 and DZs established during this period, this study examines the relationship between DZs and regional high-quality development and analyses the heterogeneity of policy effects to supplement the research on economic performance of DZs at the level of sustainable development. Second, limited to the availability of data, most existing studies on the policy evaluation of DZs are based on the China Development Zone Audit Announcement Catalogue (2006 Edition), which does not include the DZs established after 2006. This study expanded this sample to 2018 to obtain more timely conclusions. In terms of research methods, the difference-in-difference (DID) method can overcome the endogeneity problems caused by the close relationship with regional economy and geographical location and eliminate the interference of other factors, accurately identify the net effect of DZ policy, and effectively solve the two difficulties of effective measurement of industrial policy and causal relationship identification in the evaluation of industrial policy effects.

POLICY BACKGROUND AND THEORETICAL MECHANISMS
2.1. Policy background After more than 30 years of development, DZs coexist in various forms. There are 552 national DZs of a total of five medium types: 219 economic and technological DZs, 156 high-tech industrial DZs, 135 special customs supervision zones, 19 border/cross-border economic cooperation zones and 23 other types of DZs. Taking national-level economic and technological DZs as an example, in terms of economic growth contribution, their total imports and exports in 2019 were RMB 10.5 trillion, accounting for 10.6% of the total imports and exports of the country, which shows that DZs have played a significant role in driving economic growth.
Throughout China's development history, the establishment of DZs has experienced two climaxes. The first peak occurred in 1992, when 70 national DZs and 144 provincial DZs were established. However, before 2003, there were some problems in the construction of DZs, such as uneven development levels and repeated construction. In response to this problem, the state implemented a clean-up and verification of the establishment of DZs in 2003. In the process of rectification, compared with the central and western regions, the DZs in the eastern region are significantly compressed, and the DZ policy has a trend of priority development from the eastern region to the central and western regions. From 2003 to 2006, after several stages of consolidation, such as centralized rectification and planning review, the number of DZs was reduced, the planning area of DZs was reduced, the industrial characteristics of DZs were highlighted and the industrial layout of DZs was optimized. After 2006, the establishment of DZs ushered in a second development climax (Figure 1). After the previous round of rectification, the layout planning of the DZs at this stage was more reasonable, the industrial structure was more optimized and the driving effect on economic development was more significant. At this stage, the government's approval of DZs mainly adhered to the principles of sustainable development, intensive land use and industrial agglomeration. Construction avoids the problems of low-level repeated construction, waste of resources caused by large-scale development, and the introduction of preferential policies beyond the competent body's authority. DZs attract enterprises mainly through appropriate fiscal and tax subsidies and infrastructure to attract multinational companies to transfer high-tech and high-value-added processing and manufacturing links to undertake research and development (R&D) centres and their service outsourcing business. Since entering the new era, the government has adhered to the principles of opening-up guidance, reform and innovation, quality first, benefit priority, market leading and government guidance in the construction of DZs, focusing on promoting opening-up, scientific, and technological and system innovation, improving foreign cooperation and economic development quality, and creating a new highland of reform and opening-up. Therefore, this study selects DZs established after 2006 for the empirical analysis.

Theoretical mechanism analysis
According to the above literature review, the effects of DZs in China on high-quality economic development has long been a controversial topic. Therefore, from this point of view, the theory of this study is based on the following research. The results are shown in Figure 2.
DZs help to promote and strengthen the formation of industrial agglomeration, provide an important carrier for the agglomeration of target industries and related industries, give full play to the scale effect, spillover effect and competitive effect, and are conducive to promoting highquality development. The spatial concentration of related economic entities is conducive to the formation of a more complete industrial chain within the DZs; that is, the collaboration of business operations between up-and downstream enterprises in the region, and the resulting shortening of the spatial distance between enterprises can bring about economies of scale. This is mainly because the concentration of economic activities helps to reduce search costs, transaction costs, transport costs, equipment usage costs and labour costs, enabling enterprises to obtain higher economic returns and devote more resources to the development of new products and technologies. On the other hand, from the perspective of economic and knowledge production, DZs can bring about spatial reallocation of human, financial, scientific and technological resources. By sharing researchers, infrastructure and R&D knowledge, enterprises are encouraged to engage in green technological innovation to improve their production efficiency. The agglomeration of enterprises in DZs provides a good platform for the spillover of knowledge and technology, which not only reduces the cost of learning Development zone policy and high-quality economic growth: quasi-natural experimental evidence from China 591 and communication but also enhances their incentive to search, evaluate, integrate and use external knowledge and information, and encourages enterprises to innovate. The transfer of intermediate products or technologies is an important carrier of knowledge and technology spillover. Owing to the limited space-carrying capacity of the DZs, the industry and innovation elements associated with the enterprises in the DZs are not only concentrated in their interior but also distributed around the zones. When the enterprises in the zones cooperate and transfer technology with related industries and innovation subjects, a regional industrial network is formed which promotes the regional economy. Simultaneously, technology communication based on personnel flow in the DZs will strengthen the driving and communication effects on regional economic growth. DZs are often gathering areas for applied talent. The flow of human capital will remove the tacit knowledge and technology of enterprises at the same time, promote the diffusion of basic knowledge and technological innovation on the basis of interaction, and further play the leading role of industrial clusters in promoting regional technological progress and economic growth. Another path by which DZs influence regional economic development through the agglomeration effect is the competition effect. In a highly competitive market environment, the core competitiveness generated by internal innovation and technology accumulation raises barriers to entry and exit in DZs so that enterprises can spontaneously engage in healthy competition. First, the competitive effect of DZs can encourage enterprises to engage in research on cleaner production technologies and to invest more in green R&D, thereby improving economic efficiency. Second, the competition effect can encourage enterprises to learn from each other and increase their R&D expenditure through the cohort effect brought about by the technology spillover from enterprises in the zone, thus forming healthy competition among enterprises. DZs can affect high-quality economic development through policy effects, of which the positive impact is mainly reflected in tax incentives and government subsidies and the negative impact is mainly reflected in resource mismatch and path dependence. Among them, tax preference helps ease the tax burden of enterprises and reduces the marginal cost of enterprise innovation activities, thereby improving the output of enterprise innovation input and promoting enterprise capital accumulation. On the other hand, consistent with tax preference, government subsidy, as another kind of transfer payment, affects the innovation decision of enterprises by increasing capital ownership, so as to improve the enterprise's innovation investment willingness and help enterprises solve the problem of infrastructure coordination. At the same time, enterprises in the DZs also prioritize project approval, employment organization, bank loans, and industrial land. This policy support can directly or indirectly provide financial and policy support for enterprises' innovation activities with long cycles, high risks and large investments, and can effectively promote enterprises' innovation behaviour and enhance their growth mode to tend more to connotative growth so as to promote sustainable development. However, the flows of technology, capital and labour are not completely controlled by the market. Driven by the pursuit of regional economic growth goals and the alleviation of policy burden, local governments may set up DZs through land transfer, infrastructure investment, and ecological investment to carry out regional industry and investment competition, leading to intervention in resource allocation. Another potential problem is that the purpose of enterprises entering DZs is to obtain more rent-seeking policies involving government subsidies, tax incentives, land price incentives and many other such measures. This kind of policy rent-seeking effect leads to the aggravation of regional resource mismatches. The essence of DZ fever is that the local government, in order to compete with large enterprises, aggravates the degree of regional resource mismatch and reduces highquality development. On the other hand, if DZs give enterprises too much preferential treatment, it will lead to a lack of innovation power and reduce the production efficiency of enterprises. Excessive dependence on state subsidies leads to path dependence, which further worsens the improvement of high-quality development. Based on the above analysis, this study posits that DZs not only have a positive impact on tax preference and financial subsidies in agglomeration and policy effects, but also have a negative impact on resource mismatch and path dependence in the policy effect. Therefore, this study proposes the following hypotheses: Hypothesis 1a: DZs policy can promote high-quality economic development.
Hypothesis 1b: DZs policy can inhibit high-quality economic development.
DZs can influence high-quality development through governmental competition. The main paths include local championships and protectionism. Specifically, the purpose of DZs is to maximize local gross domestic product (GDP) growth and fiscal revenue, while ignoring whether industrial relevance and investment behaviour can meet the local comparative advantage. Different regions form tournaments, and local competition tournaments further strengthen the protection of local enterprises to inhibit high-quality local development. Under the constraints of Development zone policy and high-quality economic growth: quasi-natural experimental evidence from China regional economic growth objectives, local governments expand the number of DZs in the region through resource allocation and project approval. However, limited by the overall market scale and regional resource endowment, the policy intensity is not as high as possible. Local protectionism also strengthens the vicious competition between DZs, which leads to rent-seeking behaviour. This study holds that the differences in local government behaviour and regional endowment will make the policy effects heterogeneous. Specifically, a high policy intensity of DZs does not necessarily lead to a higher level of high-quality development. The policy effects in different regions will also differ because of different location conditions and factor endowments. Therefore, this study proposes the following hypotheses: Hypothesis 2: The policy intensity effect of DZs is heterogeneous.
Hypothesis 3: The policy effect of DZs is regionally heterogeneous.

LITERATURE REVIEW
Since the 1980s, the research on economic and social effects of DZs has been enriched. There are three aspects that are closely related to this paper: the agglomeration effect of DZs; the policy effect of DZs; and the government competition effect of DZs.

Agglomeration effect of DZs
Scholars have mostly studied the socio-economic impact of DZs from the perspectives of economic development, employment, taxation and land-use efficiency, but have not reached consistent conclusions.
Some scholars believe that DZs positively affect regional economic performance. Wang (2013) argues that the special economic zone (SEZ) policy has produced an agglomeration effect, which has brought foreign direct investment (FDI) and TFP growth to the target cities, and that there is a superposition effect of SEZ projects. Zhuang and Ye (2020) found that high-tech zones in China exhibited a rapid but unstable agglomeration trend from 1988 to 2018. Lu et al. (2019) showed that DZ policy can exert an agglomeration effect, attract more enterprises and provide more wages, thereby increasing output and productivity. Jiang et al. (2021) and Zhou (2020) provide explanations of how DZs improve the TFP of local listed companies from the aspects of technological progress, innovation level and resource allocation ability. Tan and Zhang (2018) found that high-tech zones had a significant positive impact on TFP in 277 cities, and this impact mainly depended on technological progress. Zheng et al. (2017) used the survey data of 110 industrial parks in China, and the results showed that industrial parks can have a spillover effect through human capital, foreign investment, share of state-owned enterprises and synergistic cooperation with adjacent enterprises that serves to improve productivity. Huang et al. (2017) found that DZs can improve land-use efficiency through cumulative and agglomeration effects, thus promoting enterprise performance. Luo et al. (2015) found that the spatial spillover effect of DZs can drive the productivity of enterprises in neighbouring areas, which is positively correlated with enterprise density and negatively correlated with distance. Alder et al. (2016) find that DZs can promote local economic growth by improving TFP, human capital and physical capital accumulation. Using data from 379 counties in Poland, Ciżkowicz et al. (2016) found that SEZs have a significant positive impact on employment in local and adjacent areas. Criscuolo et al. (2019) reached the same conclusion, but believed that it had no significant impact on TFP.
Some scholars believe that different types of DZs have different effects on regional performance. Luo et al. (2015) proved that the higher the DZ level, the greater the driving effect on productivity. Howell (2019) concluded that DZs in China from 1998 to 2007 promoted enterprise productivity through the spillover effect, and that the driving effect of enterprise productivity by high-tech zones is higher than that of SEZs. Ambroziak and Hartwell (2018) argued that city heterogeneity exists in the contribution of the SEZ to the Polish economy. Li et al. (2020) drew similar conclusions regarding regional and city-size heterogeneity. Jensen (2018) found that DZs in Poland significantly promote employment in adjacent areas at the initial stage, but show negative spillover effects after rapid expansion. Xi et al. (2021) found that DZs can significantly improve productivity through the agglomeration effect, but preferential policies reduce the entry barriers of enterprises, thereby reducing productivity.
However, some scholars believe that the DZ policy is not effective and may even create new economic problems. Crane et al. (2018) argued that DZs can significantly promote regional economics, but the agglomeration effect also exacerbates the wealth gap between regions. Bai et al. (2015) analysed the performance of China's high-tech zones using the dynamic network slack-based measure (SBM) model. The results show that the disharmony between production and R&D departments is the main reason for low efficiency. Hardaker (2020) argued that the enclave economy in Myanmar has limited employment quality and a low degree of sustainable development. Frick and Rodríguez-Pose (2022) found that DZs had no significant impact on the economic development of emerging countries from an international perspective. Schminke and Van Biesebroeck (2013) found that DZs and industrial parks could benefit from tax relief, preferential policies and advanced infrastructure to improve the quality of manufacturing enterprises' export products. Wu et al. (2021) found that SEZs can promote innovation through tax incentives, technology subsidies and enterprise agglomeration. However, Frick et al. (2019) found that preferential measures such as tax exemptions and subsidies in emerging countries have little effect on DZs. Sun et al. (2020) argued that tax incentives in high-tech zones in China have increased the financial burden on local governments, leading to a reduction in land-use efficiency and the quality of economic development. Grant (2020) argued that SEZs can significantly affect the choice of tariffs by local governments and avoid the impact of tariff distortion by changing the intermediate tariff level. Xi et al. (2021) found that the preferential policies of DZs reduce the entry barriers of enterprises, thus reducing the TFP of firms in the service sector. Guner et al. (2008) found that policy distortions lead to inefficient allocation of input factors and reduce enterprise output. Jedwab et al. (2017) found that developing countries have path dependence in the construction process, resulting in a low level of economic development. Chen et al. (2019) used the rectification of DZs in 2004 to analyse its impact on enterprise productivity. The results show that the closure of DZs leads to a decline in enterprise TFP due to scale non-competition, and the policy itself leads to spatial dislocation. Brandt et al. (2012) found that the government's policy bias leads inefficient state-owned enterprises to obtain more credit rationing, resulting in resource mismatches. Zhuang and Ye (2020) found that the preferential policies of high-tech zones make them the goal of local governments, which are greatly affected by the power of governments at all levels. Alkon (2018) studied the impact of the SEZ in India on the regional economy and found that rent-seeking weakens its development potential. Levien (2012) argued that the SEZ is a profit-eater designated by the state that acquires land through dispossession rather than the market, causing the rural population to fall into a state of continuous poverty. Sosnovskikh (2017) believes that government actions seriously hinder the economic efficiency of enterprises in the park. Chen et al. (2017a) found that exportprocessing zones can significantly increase total exports, and local governments prefer industrial sectors that are more closely related to the local industrial structure. Kahn et al. (2021) found that political preferences lead to different productivity levels in DZs in different regions. Giannecchini and Taylor (2018) found that learning from China's industrial park experience in Ethiopia cannot effectively promote the local economy, and that the policy effect is significantly related to its political and economic structure.

Government competition effect of DZs
The above analysis reveals that there is a wealth of research on the relationship between DZs and TFP, but no consistent conclusion has been reached. Most quantitative studies on DZs have focused on a quantitative assessment of the performance of the zones themselves, while relatively few studies have examined the macroeconomic impact of the establishment of DZs as a policy

Empirical model
The core issue is to explore the impact of DZs on highquality regional development. By 2018, China had 186 cities that had set up 552 DZs in different years, which naturally forms a multi-stage staggered 'quasi natural experiment'. Owing to the different time of setting up DZs in different cities, this study uses Beck et al. (2010) for reference. Cities with DZs established in the period 2006-18 were taken as the experimental group, while the cities without DZs were taken as the control group to construct a two-way fixed multi-period DID. The specific models are as follows: where i and t represent city and year; TFP it represents the TFP of city i in year t; DZs it represents the establishment of DZs in city i in year t; the cities that set up DZs in year t and later are assigned a value of 1, and other assignments 0; and Control it are control variables, including economic development level (GDP), industrial structure (Secind, Terind), social consumption level (Cons), regional financing level (Fin), foreign direct investment (FDI), human capital levels (HC), informatization level (Internet), government development level (Gov) and innovation level (Inno). d i and g t represent city-and time-fixed effects; and m it is a random-error term.

Variable description 4.2.1. Explained variable
The central question of this paper is to explore how DZs affect regional TFP; therefore, this study measures the TFP of 284 prefecture cities in China using the data envelopment analysis (DEA)-Malmquist productivity index method as follows: The index reflects the improvement of TFP from period t to t +1 in decision-making units (DMUs) under constant return to scale (CRS). When M > 1, productivity presents an upward trend; when M < 1, productivity presents a downward trend; when M ¼ 0, productivity remains unchanged. The Malmquist productivity index can be decomposed into technical efficiency change (EC) and technical change (TC). Therefore, (2) can be further Development zone policy and high-quality economic growth: quasi-natural experimental evidence from China written as (3): In the input indicators, according to the Cobb-Douglas (C-D) production function, labour and capital inputs were selected in this study. Labour input is expressed by the number of people employed at the end of the year, and capital input is expressed by the capital stock calculated using the perpetual inventory method. For output indicators, this study selected the real regional GDP deflated by the base year of 2000.

Control variable
The choice of control variables can effectively eliminate the influence of other factors on regional economic growth to better determine the net effect of DZs on TFP. In the selection of control variables, following Zheng et al. (2016) and Alder et al. (2016), this study selects the GDP of each prefecture city to characterize the level of regional economic development (GDP). With reference to Kim et al. (2021) and Chen et al. (2017b), this study measures the industrial structure of the prefecture city in terms of the proportion of secondary and tertiary industries in GDP, selects the proportion of total retail sales of social consumer goods in GDP to measure the consumption level (Cons), selects the proportion of the loan balance of financial institutions in GDP at the end of the year to depict the regional financing level (Fin), and selects the proportion of FDI in GDP to measure foreign capital introduction capacity (FDI). With reference to Zheng et al. (2016) and Jiang et al. (2021), this study selects the number of university students to measure the human capital of each prefecture (HC), and selects the number of Internet users to describe the level of regional informatization (Internet). Based on Alder et al. (2016) and Lu et al. (2019), the proportion of government expenditure in the GDP of each prefecture is used to measure the government's capacity to intervene (Gov) and the number of patent applications in a city to measure regional innovation (Inno).  Table A1 in the supplemental data online for the complete results).

Benchmark regression results
According to formula (1), this study estimates the DZ policy effect on high-quality development (Table 2). Model (1) is the result after controlling for the fixed effects of city and time, and model (2) is the regression result after adding control variables. The coefficient of model (1) is negative and significant at the 10% level, which indicates that DZs inhibit the growth of high-quality development.
The regression coefficient of model (2) is slightly larger than that of model (1), but it is still negative and significant at the 5% level. This shows that after controlling for a series of variables, DZs still fail to promote regional TFP growth, and Hypothesis 1b is verified. This seems to be inconsistent with the intuitive empirical judgment, and the reason for this is that, according to Figure 1: through the path of influence of DZs on regional TFP, it can be seen that, on the one hand, DZs can increase high-quality development through the agglomeration effect and the policy effect of tax incentives and financial subsidies, but at the same time, DZs can also lead to resource mismatch and industrial path dependence, as well as intensifying the competition among local governments, local political promotion tournaments and local protectionism. However, DZs can also lead to resource misallocation and industrial path dependence, as well as increased competition between local governments, local political promotion tournaments and local protectionism, which can lead to the establishment of DZs that inhibit high-quality development.
Specifically, in the operational process of DZs, the flows of technology, capital and labour are not completely 596 Tao Ma et al.
dominated by the market. Local governments rely on land transfer, infrastructure construction and ecological investment to promote economic development. Therefore, government behaviour hinders the free flow of elements. The traditional assessment method of local government officials' performance based on GDP growth has also formed the so-called 'development zone fever' to a certain extent, resulting in excessive concentration of resources and economic activities and aggravating the degree of regional resource mismatch. Compared with other levels of DZs, national DZs receive more financial subsidies and preferential policies, which leads to a more serious policy rentseeking effect. To sum up, it can be found that in current development stage, the negative impact of DZs is greater than its positive impact, making its net effect on highquality development negative. The regression coefficients of the control variables are also in line with expectations. It is worth mentioning that the coefficient of GDP is significantly positive, which indicates that although the establishment of DZs can significantly improve regional economic development, it cannot promote sustainable development. The central government put forward the urgent requirement of improving high-quality development, which should be the goal of local governments to actively improve the efficiency of resource allocation of DZs. This study also calculated the impact of the establishment of DZs on EC and TC after decomposition (Table 2, models 3-6). The empirical results show that although not statistically significant, the coefficient is negative, which also indicates that DZs hinder the change in technical efficiency and technical change.

Heterogeneity of policy intensity in DZs
The policy effect estimated in the benchmark regression refers to the effect of multiple intensities (economic and technological DZs, high-tech industrial DZs, and customs special supervision areas) on different cities. However, some cities set up only one type of DZ, while others set up different types of DZs at the same time. Howell (2019) showed that the simultaneous establishment of economic DZs and high-tech zones has a masking effect on productivity. Therefore, this study further subdivided the sample. SEZ 1 , SEZ 2 and SEZ 3 mean that city A has only one DZ. SEZ 12 , SEZ 13 , SEZ 23 and SEZ 123 indicate that a city has different types of DZ. For example, SEZ 12 indicates that the city establishes economic and technological DZs and high-tech industrial DZs, and SEZ 123 indicates that the city establishes economic and technological DZs, high-tech industrial DZs, and customs special supervision areas. This study re-estimates the effect of the DZ policy under different policy intensities. The results are presented in Table 3 (see Table A2 in the supplemental data online for the complete results).
Models (1) to (3) show the results for a single type of high-quality DZ development. The establishment of a single type of DZ inhibits high-quality development, Development zone policy and high-quality economic growth: quasi-natural experimental evidence from China with high-tech industrial DZs having the most significant inhibitory effect on high-quality development, followed by customs special supervision zones, and economic and technological DZs last. Models (4) to (7) in Table 3 show the results of different combinations of the three types of DZs on TFP. Additionally, the inhibitory effect of the two types of DZs on high-quality development is lower than the sum of the two types of DZs. The results of the three types of DZs are the same. Combined with previous theoretical analysis, this paper believes that reason for this phenomenon is that different combinations of DZs can realize the complementary advantages of technology and management level by strengthening the forward and backward links to a certain extent, so as to play an obstacle role in the decline of high-quality development. On the other hand, in order to attract high-quality enterprises, different DZs usually set a higher threshold or match the industrial transformation, which intensifies the competition between DZs. The 'survival of the fittest' mechanism caused by competition has transferred resources from low-efficiency DZs to high-efficiency DZs, effectively improved the regional resource allocation and saved the decline of TFP as a whole.

Regional heterogeneity test
Considering the vast territory of China, the impact of DZs on high-quality development may vary across cities in different regions in terms of their initial factor endowment, economic base and policy support. In this study, the 284 sample cities were divided into two categories according to different criteria: first, the sample cities were divided into eastern, central and western regions; second, the sample cities were divided into developed cities and underdeveloped cities according to the newly released City Business Charisma Ranking (2020 Edition). Developed cities include first-tier cities, new first-tier cities and second-tier cities, while the others are underdeveloped cities. The results of the regional heterogeneity tests are presented in Table 4 (see Table A3 in the supplemental data online for the complete results).
Models (1) to (3) in Table 4 show the heterogeneity test results. The establishment of DZs in the eastern region showed the strongest inhibitory effect, which was significant at the 1% level. A possible reason is that the eastern region itself has great advantages over the central and western regions in terms of geographical location, infrastructure, human capital and policy environment, so the vigorous implementation of DZs in the east will lead to excessive accumulation of input factors and finally inhibit high-quality development. In the central and western regions, although the regression coefficient of the model is not significant, it can be seen from the size of the coefficient that the central and western DZs have a much lower inhibitory effect on high-quality development than the eastern region. Models (4) and (5) in Table 4 show the results of the city-level heterogeneity test, in which the effect of DZs in developed cities on high-quality development is positive but insignificant, while the effect in underdeveloped cities is significantly negative, for reasons that may be related to the governmental behaviour of the cities where DZs are  Development zone policy and high-quality economic growth: quasi-natural experimental evidence from China 599 located. As an important means of government intervention in economic activities, DZs require the government to take advantage of the aggregate information to guide industrial development in the absence of market mechanisms. Compared to underdeveloped cities, developed cities have significant advantages in terms of financial resources, administrative approvals, tax incentives and other key resources that influence the establishment of DZs; therefore, DZs under the guidance of governments in developed cities operate more efficiently and can improve quality growth, whereby Hypothesis 3 is verified.

Parallel trend test
An important prerequisite for multi-period DID is that the trends in the experimental and control groups are consistent or do not fluctuate significantly prior to the implementation of the policy; in this case, before the establishment of DID, the trend in TFP in cities with DZs (experimental group) is consistent with those without (control group). Therefore, this study conducted a parallel trend test, whose results are shown in Figure 3. The analysis shows that before the policy shock (d0), the TFP of cities with DZs (experimental group) and cities without DZs (control group) basically maintain the same trend, and there was no significant difference. After the policy shock (d0), the TFP trend of the experimental group and the control group changed significantly, which can be judged to meet the parallel trend hypothesis.

Placebo test
To avoid bias in the estimated results of the experimental and control groups due to other policy changes or random factors, we conducted a placebo test by selecting a fictitious experimental group of municipalities that are known to be unaffected by policy shocks. First, a certain number of samples from cities without DZs are taken as a new experimental group, and the corresponding regressions are carried out to obtain the corresponding estimated coefficients to complete a placebo test. After repeated sampling 1000 times and regression, 1000 corresponding estimation coefficients can be obtained, and their means compared with the benchmark regression results, that is, whether the experimental and control groups are affected by other policy changes or random factors. According to the placebo results shown in Figure 4, it can be judged that no other policy change or random factors affect the results. The DID estimation results in this study are robust.

Propensity score-matching test (PSM)
In the process of using DID, the regression results may be selectively biased owing to self-selection problems, so the  PSM method can be used to reduce such endogenous problems. In this study, by matching the regions with the same urban characteristics to eliminate the factors that interfere with the establishment of DZs, this study adopts the year-on-year matching method to estimate the tendency score using the logit model, and then makes a new one-to-one matching. The estimated results are shown in Table 4 (model 6). The results show that the establishment of DZs still significantly inhibits regional TFP growth, proving the robustness of the results.

Replacing core variables
The DEA-Malmquist index method may have some bias in estimating TFP: the decomposition of the index is based on the assumption of CRS, and CRS is based on the assumption of perfect competition, but under the realistic conditions of imperfect competitive market and limited production technology, CRS is unrealistic, which in turn affects the reliability of the results. Thus, this study re-estimates the core variables in the following three respects: . Referring to Battese and Coelli's (1995) model, this study recalculates TFP using the stochastic frontier analysis (SFA) approach to ensure the maximum accuracy of the results. SFA sets the form of the production function as a trans-log production function, which is more flexible than the C-D function. At the same time, it relaxes the assumption of CRS and technology neutrality, and allows the underutilization of labour and capital. In addition, SFA considers the impact of the random error term on the TFP. . Under current constraints of resources and environment, this study takes natural resource input and environmental pollution into account in the calculation of TFP, which helps to more comprehensively and accurately consider the performance of economic development. Referring to the possibility set containing desired and undesired outputs constructed by Färe et al. (2007), this study uses the Malmquist-Luenberger index based on non-radial SBM directional distances to measure the dynamic changes in green total factor productivity (GTFP) in prefecture-level cities. . Referring to the research of Zheng et al. (2017) and Jia et al. (2020), this paper uses macroeconomic variables to represent economic development for a robustness test, including employment (Jensen, 2018), tax and per capita disposable income (Frick et al., 2019;Zheng et al., 2016). The results are presented in Table 5 (see Table A4 in the supplemental data online for the complete results). Table 5 (model 1) shows that the regression result is significantly negative, which indicates that the effect of DZs on TFP is still negative under the different methods, thus verifying the robustness of the benchmark regression. Although the regression results of model (2) are not significant, the coefficient is negative, which indicates that DZs still inhibit GTFP after considering environmental factors. Models (3) to (5) show that DZs can contribute significantly to higher levels of employment, tax and per capita income. This is consistent with the conclusions of Jia et al. (2020). A possible explanation is that the DZ policy mainly relies on large-scale physical investment to promote the growth of regional employment, tax and income, but does not improve production efficiency represented by the level of innovation and technology.
6.5. Generalized method of moments (GMM) estimation considering serial correlation As Bertrand et al. (2004) pointed out, the data used in DID estimation generally have a serial correlation problem, which causes the standard error of the estimator to significantly underestimate the standard deviation of the estimator, resulting in the value of the t-statistic being too large, leading to the excessive rejection of the original hypothesis. There are three sources of sequence correlation in the estimation: First, the data used in DID estimation usually cover a long time span. Second, the dependent variables commonly used in DID estimations generally have a high positive serial correlation. Finally, for all samples of a cross-sectional unit, the value of the core Development zone policy and high-quality economic growth: quasi-natural experimental evidence from China 601 explanatory variable policy × group rarely changes with time and thus has a high positive serial correlation.
The sample period of this study was 13 periods. To solve the possible serial correlation problem, this study constructs a GMM model for empirical analysis. The GMM model includes the differential GMM (DGMM) and system GMM (SGMM). This study uses both the DGMM and SGMM for empirical analysis (Table 6; see Table A5 in the supplemental data online for the complete results). The results show that there was no serial correlation problem in this study. The coefficient of L.TFP is significantly positive, indicating that TFP is significantly affected by the previous period. The coefficient of the DZ is significantly negative, which further verifies the benchmark results.
6.6. Considering potential anomalies of the model Considering that the time-varying DID deviates from the setting of the standard, this study uses the unified framework proposed by Callaway and Sant'Anna (2021) to analyse how to estimate a reliable treatment effect when there are different intervention groups and time points. Therefore, we re-estimated the results by using different aggregation strategies (Table 7).
Under different aggregation strategies, the policy effects are still similar to those in Tables 2 and 5, indicating that the policies of DZs can improve employment, tax, and per capita disposable income but have a negative impact on productivity, which further verifies the previous empirical results.

DISCUSSION
Based on the panel data of China from 2006 to 2018, this study estimates the TFP of 284 prefecture-level cities, and then uses the quasi-natural experiment of DZs to build a multi-period DID model to analyse DZs policy on highquality development. The results show that DZs significantly promote regional economic performance, but inhibit high-quality development. A certain degree of heterogeneity exists in the impact of policy intensity on high-quality development. At the urban level, the inhibition effect of central and western cities was more obvious than that of eastern cities, and the inhibition effect of underdeveloped cities was more obvious than that of developed cities. Finally, a series of robustness tests was conducted, and the regression results remained robust. Based on the above empirical conclusions, this study proposes the following policy recommendations: . Further expand the promotion effect of the economic performance of DZs and incorporate the quality constraints of economic growth into development planning and management.
At the initial stage of DZs, the government should focus on its subjective initiative, overcome the problems of imperfect market and asymmetric information, improve the performance evaluation in the process of policy implementation, formulate reasonable tax policies and financial subsidy policies, give full play to the policy effect of DZs, support the development of advanced manufacturing industries such as big data centres, artificial intelligence (AI), and industrial Internet, improve the efficiency of high-quality development of DZs through scale effect, spillover effect and competition effect and use existing policies and capital channels to support the development of living services such as medical health and community services, as well as productive services such as industrial design, logistics and exhibitions in DZs. Local governments should improve the subsidy standards for enterprises in DZs to cultivate high-skilled talents in key industries in short supply and provide support for the  joint construction of a talent training base and entrepreneurship incubation base between the DZs and vocational colleges. In addition, policies should be implemented to adhere to the core position of innovation in the overall construction of DZs; pay equal attention to innovation and application transformation; accelerate the construction of a technological collaborative innovation system with enterprises; promote the development of growth modes driven by innovation; and maintain sustainable growth.
Meanwhile, this study brings the quality constraints of economic growth into the development planning and management of DZs. While the speed of economic development, local governments should focus on the quality of economic development. At present, DZs still depend mainly on the extensive development model, which can no longer meet the requirements of the new era. Therefore, this paper proposes to consider the quality of economic growth as a clear binding index in the development planning of DZs and as a rigid reference index for DZs approved by the state and later policy support, so as to effectively avoid ineffective competition and path dependence of local governments. At the same time, priority should be given to improving the normal management and assessment system with economic quality as the core, establishing a dynamic entry and exit mechanism for later subsidy policies, forming effective constraints on the development direction of DZs, and actively guiding the transformation of DZs to high-quality development.
. Build a policy tool framework for classified policy implementation, and adhere to problem orientation, classified guidance, and accurate policy implementation for DZs in different regions, urban levels, and policy intensities.
Considering the heterogeneity of policy intensity, local governments should scientifically grasp the functional positioning of various DZs within their jurisdiction, make rational planning and guidance, optimize resource allocation, and develop leading and characteristic industries with local comparative advantages in combination with their own regional advantages, industrial development objectives, technological development level, and industrial relevance. To explore the path of differentiated development of DZs, moderately tilt the areas with a low development degree of DZs. For enterprises engaged in encouraged projects in DZs of central and western regions and improving the industrial chain, the government can coordinate the transfer payment funds and their own funds from higher authorities and provide corresponding support for qualified projects such as infrastructure construction, logistics, and transportation, undertaking industrial transfer, and optimizing the investment environment in the DZs of less developed central and western regions. The DZs in developed coastal areas should reasonably control the development scale and intensity and strengthen scientific and technological innovation and institutional innovation to speed up the transformation of old and new kinetic energy. At the same time, to break the segmented market environment, DZs cannot simply rely on tangible hands to intervene in the flow of production factors. DZs also need to break some low efficiency bias policies, cultivate a perfect and fair competition mechanism, break labour market segmentation, and promote the economic growth of DZs by bringing the effect of increasing total productivity.

CONCLUSIONS
Based on the quasi-natural experiment of DZs, this study examines the impact and mechanism of China's DZ policy effect. The results showed that DZs inhibit high-quality development. Different types of policy mixes have certain heterogeneity. At the city level, the inhibition effect of central and western cities is more obvious than that of eastern cities, and that of underdeveloped cities is more obvious than that of developed cities. However, this study still has the following deficiencies. First, the sample data only includes 284 prefecture-level cities. Therefore, future research should examine the TFP of enterprises in the DZs at a more micro level and then measure the impact of DZ policies. Second, it has been shown that the DZ policy has a significant spillover effect (Lu et al., 2020;Wu et al., 2020), but this study only considers the impact of DZs on the TFP of the region, so the impact of DZs on the spillover effect of surrounding cities and the trend of this impact change with the change in geographical distance, zone size and city level requires further investigation.