Development zones and firms’ performance: the impact of development zones on firms’ performance for a Chinese industrial cluster

ABSTRACT This paper examines the policy effect of development zones on firms’ performance, excluding the agglomeration effect. We constructed an industrial cluster dataset on China’s manufacturing industries to identify the impacts on firms’ productivity and other performance indices after the establishment of development zones in 2006. Based on the estimated results obtained from a difference-in-differences analysis, development zones are conducive to promoting firms’ performance, and the policy effect is heterogeneous across industrial clusters, regions and firms. The findings drawn from this study can be beneficial to policymakers in their pursuit of promoting regional development through favourable industrial policies.


INTRODUCTION
Regional development strategies have far-reaching consequences in the process of China's industrialization. In China, a development zone policy is often viewed as an important place-based policy for stimulating economic reforms. The creation of development zones implies that a local government has been granted the power to implement preferential industrial policies and enhance regional competitiveness (Lu et al., 2019;Wang, 2013;Zheng et al., 2017). To increase investment and stimulate local industrial development, development zones are equipped with better infrastructure 1 and offer tax deductions, preferential land and financing policies to attract firms. The location of development zones is not chosen randomly. In particular, a government establishes development zones in accordance with regional industrial conditions, geographical locations and infrastructural facilities. Such site selection may have systematically biased the results because any positive effects could primarily reflect successful initial targeting of better-endowed areas which would be more responsive to treatment (Lu et al., 2019). Therefore, it is necessary to answer whether the policy would support regional performance rather than merely be good at 'backing winners'.
In China, development zones are closely related to the development of industrial clusters (Kline & Moretti, 2014;Wang, 2019). The location selection of development zones is similar to the incubation of industrial clusters. Specifically, when industrial clusters become an important driving force of regional growth, governments often apply them to establish development zones, taking advantage of their better industrial bases. This phenomenon is commonly seen in China's coastal provinces with highly mature industrial clusters, such as Zhejiang and Guangdong. Therefore, it is difficult to estimate regional productivity improvement because of the agglomeration effect or the policy effect separately. In other words, if we estimate the impact of development zone policy on firms' performance under the premise of the existence of industrial clusters, then we can solve the site selection issue mentioned above.
To address these issues, we followed the definition of regional-specific industrial clusters in the literature and extended the measurement method of industrial clusters  by using firm-level panel data on China's manufacturing industries during the period 2003-09.
Inspired by Delgado et al. (2016), a cluster is identified not only by highly geographically concentrated firms belonging to a core industry but also by firms up-and downstream in the same region. For counties hosting industrial clusters, on average, the output of clusters accounts for 78.3% of the total output of the manufacturing sector in China, showing that clusters have an unignorable impact on regional development. Therefore, when studying place-based policies, such as development zones, the agglomeration effect derived from firms' geographical concentration should not be neglected. If industrial clusters have been identified using the new method, it will enable researchers to focus on firms equally affected by the agglomeration effect. This is the tactic used in this paper. We combined the dataset of industrial clusters with the dataset of the geographical distribution of development zones to determine if setting up development zones has a statistically significant effect on firms' performance. More specifically, the samples used in this paper are strictly limited to firms belonging to industrial clusters that emerged earlier than the establishment of development zones, so the agglomeration effect on firms within the development zones can be controlled. Based on the firm-level panel data, the establishment of development zones relates positively to firms' productivity growth and performance improvements. While previewing some of the results, we found that the establishment of development zones will increase firms' output by US$9440 and improve firms' investment efficiency by 3.056%. Compared with figures in previous literature, these estimated coefficient effects are considerably large (Lu et al., 2019). We found that most of the performance indices, including productivity, employment and exports, were enhanced due to the creation of development zones. However, we also discovered that the development zone policy seems to have heterogeneous effects across industrial clusters, regions and firms.
Our findings are consistent with much of the existing literature but add new insights in two ways. First, in the field of industrial agglomeration, government behaviour regarding industrial concentration and development has long been an issue of interest (Cainelli & Lacobucci, 2012;Lu & Tao, 2009). Although establishing development zones is closely related to the prosperity of local industrial clusters in China, because of vague geographical boundaries and endogeneity problems, only a few studies based on firm-level data have analysed the impact of development zones on industrial clusters (Wang, 2019). However, the dataset of clustered manufacturing firms enabled us to evaluate the policy effect of development zones at the firm level while excluding the agglomeration effect caused by geographical concentration. Second, we adopted the difference-in-differences (DID) method to evaluate the policy effect of development zones on firms' performance. A few papers have used the DID method to estimate changes in firms' productivity after the establishment of development zones; however, it is necessary to exclude the impact of time-invariant influences (Beck & Levkov, 2010). In contrast to earlier studies, the data and methods used in this paper alleviate these problems to some extent. Thus, our main contribution to the literature is the empirical evidence we provide with the case of China and the method we used to confirm the accuracy of the empirical findings. This makes our findings play a necessary supplementary role in the existing literature.
The rest of this paper is organised as follows. The next section reviews the literature on development zones and industrial clusters, which are closely related to this paper, followed in the third section by a brief overview of China's development zone policy. The data and methodology used in this paper are discussed in the fourth section. The fifth section presents the estimated results of the baseline model and various robustness tests. The sixth section probes into the heterogeneity effects of development zones on firms' productivity. Finally, the key findings and policy implications are summarized.

LITERATURE REVIEW
While industrial clusters play a crucial role in regional prosperity (Ketels & Protsiv, 2021), regional policy also influence the long-term development of industrial clusters. In particular, the government will implement a range of policies in accordance with regional comparative advantages to provide extra resources for industrial agglomeration (Blonigen, 2016;Moretti & Thulin, 2013). In China, with the incubation and expansion of industrial clusters, severe competition causes some industrial clusters to encounter the dilemma of extensive development and insufficient innovation (Wang, 2019;Zhang et al., 2021). To boost the local economy, governments at all levels across the country have put forward supportive policies, such as establishing development zones.
The Chinese government has issued no specific policy to support industrial clusters, but the location decision of a new development zone is closely related to the performance of local industrial clusters. In particular, the government considers regional industrial bases before planning development zones, which means regions with better industrial bases are likelier to be selected as development zones (Kline & Moretti, 2014;Lu et al., 2019;Wang, 2019). At the same time, if both investment environments and industrial bases are favourable, a relatively mature and high-quality industrial cluster can be nurtured in a symbiotic interaction zone (Brenner & Greif, 2006). Such a phenomenon is observed in France's Sophia Antipolis industrial park and China's Xinzhu industrial park (Wang, 2019). Therefore, the role of industrial clusters should not be ignored when evaluating the policy effects of development zones.
The existing literature on the effects of establishing development zones has mainly focused on regional prosperity, technical innovation, industrial cooperation and social development (Alder et al., 2016;Bräutigam & Tang, 2014;Cheng & Kwan, 2000;Wang, 2013). Considering the role of development zones in regional economic growth, recent literature suggests that development zones are beneficial to gross output growth and firms' productivity; they also have a significant spillover effect on neighbouring regions (Greenstone et al., 2010;Lu et al., 2019). A heterogeneous effect exists across regions, The impact of development zones on firms' performance for a Chinese industrial cluster 869 industries and firms. In particular, the policy effect of development zones is more effective for new entrants and capital-intensive firms (Miguelez & Moreno, 2018;Lu et al., 2019). Although many studies have examined the relationship between firms' productivity and development zones, few have explored firms' productivity, excluding agglomeration effect. Given the simultaneous causality between the prosperity of industrial clusters and the performance of economic zones, this paper complements the previous literature with an empirical analysis. Taking advantage of firm-level data and the identification method, we explored the heterogeneous effect across industrial clusters, regions and firms, as suggested by the earlier literature. In summary, our findings constitute pieces of evidence that complement the general understanding of the relationship between industrial clusters and development zones and contribute to the evaluation of place-based policy for the government at all levels in China.
The next section summarizes the development zone policy in China and provides detailed descriptions of the data and empirical strategies used in the paper.

ECONOMIC ZONES IN CHINA: BACKGROUND AND DEVELOPMENT
Development zones are specific areas equipped with better infrastructure and offer a bundle of preferential policies including (Alder et al., 2016;Lu et al., 2019;Wang, 2013;Zheng et al., 2017): . Tax deductions: the most common tax preferential policies in development zones are 'two-year exemption and three-year half payment' of the corporate income tax, which resulting in the average corporate income tax rates of 15-24% for high-and-new companies compared with the 33% firms normally pay in China. . Discounted land-use fee and utility prices: firms in development zones enjoy lower land transfer fees, utility prices (electricity, water, etc.) and favourable payment methods. For example, land transfer fees are 20-35% lower than those outside zones in Guizhou province. . Special treatment in securing bank loans: banks put a priority on and offer favourable interest rate to the loan applications from in-zone firms. . Firms in zones receive faster and easier administrative approval for registration.
The first national development zone in China was established by the government in 1984, called the Dalian Economic and Technological Development Zone. Until January 2018, 552 national development zones and 1991 provincial development zones 2 have been established in China. According to the Catalogue of China's Development Zones (2018 Version), 3 there are five types of national development zones: economic and technological development zones; high-tech industrial development zones; bonded zones; export-processing zones; and border economic cooperation zones. The first two types accounted for 87.6% of the zones. As for provincial-level development zones, there are provincial economic zones, high-tech industry zones and special industrial parks, among which 90% are economic zones.
In the past 40 years, the policy on the development of special economic zones has been adjusted and can be divided into three phases: . Initiation and rapid spread (1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003): various national zones, especially economic and technological development zones and high-tech development zones, were set up during this period. At the same time, the establishment of industrial parks boomed to support from local governments. . Rectification and standardization (2004-09): during the boom, some firms and local governments colluded for subsidies or tax deductions from the central government, which led to efficiency losses in the zones. Therefore, central and local governments rectified development zones. A large number of lower administrative level (city or county) and inefficient development zones were closed in 2003-04. In 2006, for the sake of regional economic growth, local governments established several standardized new development zones. Standardization has since been required for provincial development zones all over the country. . Upgrade (2010-18): to enhance innovation capabilities, upgrading 4 development zones became the main instruments during this period, which means lower administrative level zones could have an opportunity to be designated to higher administrative levels. Since 2010, a few development zones have been upgraded to provincial and national development zones and have become important driving forces for regional innovation and economic growth. Figure B1 in Appendix B in the supplemental data online shows the number of newly established development zones during the period 1984-2018. The creation of development zones becomes relatively concentrated in a few years, which is conducive to eliminating time interference and provides a rare quasi-natural experiment for evaluating the policy effects of the development zones. Based on these facts, we took 2006 as the policy interference year and carried out an empirical analysis. 5 With this year as the focal point, we symmetrically extended the period to three years before and after the focal year and focused on the changes caused by the policy over the period 2003-09.

DATA
Our firm-level data are from the China Annual Industry Survey, which covers all establishments in the manufacturing industry from 2003 to 2009. The dataset provides a rich set of firm-level variables, such as a firm's opening date, address, ownership, output, employees, industrial classification and other individual features. Based on the dataset, we identified industrial clusters and firms within 870 Ye Liu et al.
the development zones. It is worth pointing out that the manufacturing sectors we focused on were about 480 four-digit-code industries 6 in the range of 1310-4320.

Industrial cluster
We used a firm-level dataset to measure regional-specific industrial clusters. Based on the method used by Zhu et al. (2019), there are three steps to detect the location of any industrial cluster and determine the firms belonging to it. 7 First, we identified the continuous geographical boundaries of a given industry. According to Duranton and Overman (2005), we calculated the spatial kernel density of manufacturing firms for each four-digit core industry and discovered localised industries. Second, we compiled an inter-industry linkage dataset for each localized industry. Following Delgado et al. (2016), a wide range of inter-industry linkages should be captured in the definition of clusters. Therefore, we matched each four-digit core industry with the most related up-and downstream three-digit code industries based on China's input-output tables. Each core industry and its related industries were then incorporated into an inter-industry linkage dataset. Third, we identified the firms that belong to the industrial cluster. Using the geographical boundaries we obtained in the first step we singled out firms within the core industries as the centre of a cluster and calculated the employment density through an inter-industry linkage dataset within the range of geographical boundaries. We then determined the critical value and found all clusters within a geographical boundary where the employment density is greater than the critical value. For more details, see Appendix A in the supplemental data online. Using this method, we identified 1,318,327 manufacturing firms within industrial clusters.

Development zones
To estimate the policy effect of development zones, we divided the firms within the industrial clusters into two groups: the treatment group and the control group. Following Lu et al. (2019), we identified development zones at the county level. Specifically, if firms belonging to an industrial cluster are in the county where a new development zone was established in 2006, then they were in the treatment group, or otherwise in the control group.
To exclude the effect of development zones on the formation of industrial clusters, firms not in any industrial clusters before initiating a new development zone were omitted. We also excluded a few firms located in development zones established in 2003 and 2005, which accounted for less than 5% of the total number of newly established zones, after we removed invalid samples. Finally, our firm-level panel data comprise 943,163 firms in 385 four-digit manufacturing industries. Figure 1 represents the distribution of the industrial clusters and development zones at the county level.
Since the firms were divided into two groups, we used both binary and continuous variables to define whether a firm belongs to a development zone. Here we used four variables to represent a development zone. In the baseline model, we used a binary variable, DZ, where a value of 1 reflects that a firm is located in a development zone, and 0 otherwise. Like Luo et al. (2015), we also used continuous variables to distinguish the size and intensity of the development zones. In the robustness checks, the area of the development zone was measured by the DZ_scale, which is defined as the radius of a development zone. Two variables measuring the intensity of development zones are DZ_num and DZ_main. DZ_num is defined as the number of development zones established locally in 2006 at the county level, while DZ_main is the percentage contribution to a county's total industrial output from pillar industries in the development zones. We also defined a binary variable (time variable) policy to distinguish between the period before and after the establishment of development zones. Therefore, the interaction term of the development zone variable and the time variable was used to evaluate whether development zones would have a significant impact on the performance of firms within clusters.

Dependent variables
We considered several variables to evaluate firms' performance before and after the establishment of development zones. First, we used two types of firms' productivity: labour productivity and investment efficiency. For labour productivity, we defined per_output as firms' output per worker (Ciccone & Hall, 1996). The definition of investment efficiency follows that of Rechner and Dalton (1991). Meanwhile, ROA is defined as the profit rate to net worth. We then explored changes in other performance indices after the establishment of development zones, including firms' output growth, job creation and exports. Based on Lang et al. (1996), we used variable growth as the growth rate of firms' output. Also, following Zhu et al. (2019), for a firm i in industry j located in county z at time t, we defined employment ijzt as the fraction of the total net employment of industry j generated by firm i at time t. Similarly, export ijzt represents the fraction of the total export value of industry j generated by firm i at time t. Table B1 in Appendix B in the supplemental data online shows that the mean and standard errors of firms' productivity and other variables differed before and after the establishment of development zones.

Control variables
We also controlled for other key determinants of firms' performance at the firm, industry and regional levels. First, firms' lifespan and size matter (Jensen et al., 2001;Greenstone et al., 2010), so we controlled for firms' age in the logarithm form (Age) and added the binary variable of large companies (Large). 8 Then, we introduced the Herfindahl-Hirschman index (HHI), measured as the industrial output from the industry-county pair. It captures the intensity of industry competition that varies over time (Schmitz & James, 2005). Meanwhile, the regional-level data considered here are the level of regional economic development, size of the labour pool and foreign investment (Combes et al., 2011;Tomiura, 2007).
The impact of development zones on firms' performance for a Chinese industrial cluster 871 Therefore, we controlled for the logarithm of a city's gross domestic product per capita (PGDP) and foreign direct investment (FDI), respectively. Since the area of counties varies widely in China, we used the counties' worker density (Density) as a proxy for the size of the local labour pool. The regional-level data are cited from the 2003-09 China City and County Statistical Yearbooks. Table B2 in Appendix B in the supplemental data online presents the descriptive statistics for the variables used in this paper.

METHODOLOGY
To identify the causal impact of establishing development zones on firms' performance within industrial clusters, we conducted a quasi-natural experiment based on newly established development zones in 2006 using a DID analysis. Our data cover the period 2003-09, which includes both the pre-(2003-06) and post-(2007-09) development zone periods. Using the DID method, we examined the differences between the changes in firms' performance with the set-up of development zones (the treatment group) and the changes in firms without policy support (the control group). In short, we set the basic econometric specifications as follows: where the dependent variable performance ijzt is one of the variables that measure firms' performance, including industrial output per worker (per_output), profit rate to net worth (ROA), output growth rate (growth), employment proportion (employment) and export proportion (export) of firm i in industry j at time t. The key explanatory variable DZ ijz indicates the presence of development zones in county z (the treatment group). If DZ ijz = 1, then firm i is located in the development zones, while DZ ijz = 0 otherwise. In the robustness checks section, we substitute DZ ijz with other alternative variables (DZ_scale, DZ_num and DZ_main). Policy t is the policy variable, which equals 1 after the development zones were established in 2006; otherwise Policy t = 0. That is, firms in the treatment group would be affected by a development zone during the postpolicy period, while firms in the control group would not be affected during the entire sample period. Therefore, the cross-term DZ ijz × Policy t in equation (1) was used to distinguish the differences in firms' performance after the If DZ ijz × Policy t = 1, firms would be affected by the development zone; otherwise they would not. X ijzt is a set of control variables including Age, Large, HHI, PGDP, FDI and Density. We also included five additional vectors: a i and b j represent individual and industrial fixed effects, respectively; g t is a vector of time fixed effects over 2003-09 across all firms; d z accounts for the region fixed effect; and 1 ijzt is the error term.
According to the DID model, we can identify changes in firms' performance before and after the establishment of development zones. In particular, we are interested in understanding how variations in establishing development zones affect firms' performance within industrial clusters. The estimated parameter of key interest is represented by b 1 ; if b 1 is positive, this means firms in development zones perform better. Table 1 reports the basic regression results obtained in this study. Each column lists the results from one regression using the DID estimation shown in equation (1). Table 1 shows that establishing development zones substantially improves firms' productivity and other performance indices consistently, which is in accordance with the previous literature (Lu et al., 2019). Specifically, column 1 presents statistically significant and positive estimators for DZ*Policy, indicating that the establishment of development zones leads to a 59,990 yuan (US$9440) increase in firms' output per worker compared with the predicted level given by the time trend and firms' output per worker without the presence of development zones. 9 Meanwhile, when we look at the investment efficiency in column 2, we find that the estimated result is consistent with the result of labour productivity. As can be seen, the establishment of development zones seems to exert a strong effect on firms' investment efficiency, with an estimated impact of 3.056. In other words, the development zone policy increases firms' investment efficiency by 3.056%.

Basic results
We continue to explore other outcomes of industrial cluster firms after the establishment of development zones. Column 3 represents the results of the firms' output growth. It is apparent that the establishment of development zones has a significantly positive impact on firms' growth. Numerically, firms' output growth increased by 2.012% after the establishment of development zones. In columns 4 and 5, we report a 0.027% and 0.007% increase in firms' net employment and export contributions to the industry after establishing development zones, respectively. When extending the firm-level estimation to the region, there would be a huge increase in job creation and export expansion for the region. For the control variables, large firms, the level of regional development and foreign investment almost significantly increase firms' productivity, which is consistent with the literature (Combes et al., 2011;Tomiura, 2007). It should be pointed out that The impact of development zones on firms' performance for a Chinese industrial cluster 873 the employment rate will be relatively reduced with age in column 4, which is similar to Duranton and Puga (2001). It is also worth noting that HHI negatively impacts firms' productivity and output growth in columns 1-3. A possible reason for this is that the market power within an industrial cluster is relatively weak, and fierce competition forces firms to improve their productivity to remain in a market. Meanwhile, the stronger the market power, the greater a firm's contribution to an industry (Schmitz & James, 2005;Tallman et al., 2004), which indicates HHI has a positive correlation in columns 4 and 5.

Validity of the DID specification
The reliability of the baseline results obtained so far depends on the validity of the DID specification, which indicates that firms in the treatment group and control group have a common trend during the pre-policy period. Therefore, we used several tests to examine the validity of the DID specification, such as checking the assumption of the DID specification and the influence of non-observable variables. Table 2 presents the results of the validity tests. The parallel trend is the most important premise for DID specification (Beck & Levkov, 2010). In this paper, the key assumption of the parallel trend was whether the firms' productivity and other performance variables in the treatment group and the control group manifested comparable trends before the establishment of development zones. We multiplied DZ by the dummy variable for each year (excluding the first year to avoid multicollinearity) to identify the differences before and after the establishment of development zones.
In columns 1 and 5 of Table 2, we find that the coefficients of yearly interaction terms are insignificant for all pre-treatment periods, whereas the estimated coefficients are statistically significant after the establishment of development zones. This suggests that the two groups are comparable when we use either per_output or ROA as a proxy for firms' productivity. We plotted parallel trend tests. Based on Figure 2, the coefficients of the estimator are almost zero, indicating that firms' productivity within industrial clusters was similar before the set-up of development zones. Meanwhile, when we exchange the dependent variable with other proxies (growth, employment and export), a parallel trend still holds (see Figure B2 in Appendix B in the supplemental data online). The serial correlation of error terms might be another concern in the DID estimation, which may overestimate the significance of the interaction term DZ*Policy. To deal with this, we followed Bertrand et al. (2004) and collapsed the data into two periods, pre-and post-treatment periods, to help alleviate the serial correlation problem. We used the mean value of each firm's output per worker (per_output), profit rate to net worth (ROA), and all explanatory variables in the pre-and post-development zones. Based on this new two-period panel data, we re-estimated our baseline specification and report the results in columns 2 and 6 of Table 2. The estimators of interest remain robust, and the magnitude of the estimated coefficient of DZ*Policy is very close to the baseline results.
Controlling for the influence of unobservable variables is also essential for unbiased estimations. For example, the dynamic adjustment of other industrial policies by local governments in different regions may affect firms' productivity, resulting in estimation errors. To account for the missing variables, we controlled for unobservable factors that vary by time and region. The results are reported in columns 3 and 7 and remain robust.
A placebo test is widely adopted in a DID estimation to determine whether unobservable variables could significantly affect the estimation results (La Ferrara et al., 2012;Liu & Lu, 2015). To deal with this, we randomly redistributed development zones in all counties and repeated the random process 500 times. For each random distribution, we re-ran our baseline specification and obtained a new coefficient (b random , p. 11 ) of DZ*Policy. Columns 4 and 8 report the mean values of b random 1 after 500 random processes as −0.112 and −0.003, respectively. The estimated values were close to zero and statistically insignificant compared with the baseline results. Meanwhile, as shown in Figure 3, the distribution of b random 1 is concentrated near zero in the random processes, indicating that unobserved factors have little effect on the estimated results.

Robustness checks
After testing the validity of the DID specification, we performed several more tests on the robustness of the results. First, we changed the definition of industrial clusters and neglected the role of related industries in industrial clusters. That is, we only chose core industries to define the firms within industrial clusters and re-estimated the regressions.
Apparently, the new estimated results are like those of the baseline regressions (see Table B3 in Appendix B in the supplemental data online). Second, we used continuous variables (DZ_num, DZ_scale and DZ_main) instead of dummy variable DZ to test whether firms' productivity increases with the size and intensity of development zones in general. Table 3 reports the estimated effects of development zones on firms' productivity using alternative continuous development zone variables. In general, these results are consistent with those of the DID estimation in Table 1. It is clear that the more intensive a development zone is, the larger the impact on firms' productivity. Specifically, based on columns 1-5, firms' productivity and other performance indices significantly increase with the size and intensity of development zones.

HETEROGENOUS ANALYSIS
7.1. Heterogeneous policy effects on industries n the baseline estimation, the establishment of development zones is usually beneficial for increasing firms' productivity. However, due to the heterogeneity of initial endowments and development strategies, the development zone policy may have a heterogeneous effect across industrial clusters. Therefore, we examined whether there is any evidence of a heterogeneous effect that the establishment of development zones can have on firms' productivity.
Recent studies have suggested that natural advantages can facilitate the implementation of industrial policies in development zones (Greenstone et al., 2010;Zheng et al., 2017). It is possible to predict different effects on firms' productivity when the intensity of industrial clusters differs (measured by a cluster's geographical scale and firm density). Also, firms in capital-intensive and tech-intensive industries can catch up with new technology and improve their productivity more easily (Miguelez & Moreno, 2018). Hence, we divided the samples into labourintensive, capital-intensive and tech-intensive groups to test industrial heterogeneity. We re-estimated the baseline The impact of development zones on firms' performance for a Chinese industrial cluster 875 model with firms' labour productivity, according to scale, density and industrial heterogeneity (Table 4). Columns 1-4 of Table 4 are regressed with a subsample of firms distinguished according to cluster heterogeneity, such as large radius versus small radius and high density versus low density. We defined large radius as industrial clusters with a radius larger than the median value, while high density includes clusters with a proportion of aggregated firms in the whole industry higher than the median level.
The results in Table 4 prove that a significant difference exists between the subsamples. According to the estimated results, the coefficients of the interaction term for the large radius group and the low-density group were much larger than those of the small radius group and the high-density group. That is, larger industrial clusters benefit more from development zones, and lower density clusters become more efficient after receiving a special economic zone policy. Columns 5-7 show that firms in capital-intensive industries benefit most from a zoning policy on productivity improvement, followed by tech-intensive industries and, finally, labourintensive industries.

Heterogeneous policy effects on firms
Firms greatly differ in their size, productivity and local business environment. The impact of development zones on firms may have a heterogeneous influence according to firm heterogeneity. Therefore, in this subsection we re-run a set of subsample regressions according to firms' productivity and size, including high productivity versus low productivity and large size versus small size. Our definitions of high productivity and large size are like those in Table 4, which means the per capita revenue and annual revenue of a firm is greater than the median value, respectively.
In Table 5, columns 1-4, we find that all subsamples are statistically significant, and higher productivity and larger firms benefit much more from development zones,  which is like the results in Table 4. Regional heterogeneity is considered in columns 5-8. In columns 5 and 6, based on a firm's location, we divided our samples into the eastern group and the non-eastern group. It is shown that firms in the non-eastern group benefit more from development zones. We choose firms within two typical city clusters from each regional group. 10 Similarly, after the establishment of development zones, firms in Central Plains city clusters had a larger increase in productivity than the Yangtze River Delta city clusters. We suppose that natural advantages make the eastern region less reliant on policy support.

CONCLUSIONS
This paper extends the scope of the literature on the effect of development zones on industrial clusters. The analysis contributes particularly to determining whether the presence of development zones is beneficial to firms' productivity, excluding agglomeration effect.
Because of obscure boundary and endogeneity problems in the previous literature, there are few formal discussions about the estimation bias of development zones on a firm's productivity and other performance indices caused by industrial clusters. Lu et al. (2019) used a DID estimation to determine that three years after the establishment of development zones in China, firms' output increased by 49%; however, it reduced to 17.1% after controlling for the effects of agglomeration, according to our findings. That is, a policy's effect on a development zone will shrink once the agglomeration effect is excluded, which verifies our conjecture.
Extending Zhu et al. (2019), we defined industrial clusters as a high density of firm concentration within core industries and down-and upstream industries and estimated the policy effect of the development zones within industrial clusters. Also, to mitigate endogeneity, we omitted firms located in development zones that established before the initiation of industrial clusters and focused on the development zones established in 2006.  The impact of development zones on firms' performance for a Chinese industrial cluster 877 The findings in this paper suggest that a clear positive relationship exists between firms' productivity and the establishment of development zones. We also studied the increase in firms' output growth, employment and exports after the establishment of development zones. We adopted continuous variables for robustness checks, and the main findings remained robust to the alternative measures of development zones' size and intensity. To ensure the validity of the empirical method used, we undertook several different empirical strategies to identify the validity of the DID specification. Considering firm, regional and industrial heterogeneity, we estimated the heterogeneous effect across industrial clusters, regions and firms after establishing development zones. Therefore, these findings underpin the importance of development zones in the process of regional economic development, indicating the validity of development zone policy from the perspective of industrial clusters.
Our findings have significant implications for China's industrial policies. First, industrial clusters are an important driving force for regional prosperity in China's industrialization. Considering the important role of industrial policy, the government can stimulate cluster development by formulating favourable industrial policies. Second, because of the heterogeneity of initial endowments and economic conditions, we should focus on uncoordinated development across regions and industries. Therefore, industrial policies, such as the development zone policy, must be designed in accordance with regional characteristics. This paper has been a first step toward understanding the micro-foundations of place-based policies in the context of industrial clusters. Much remains to be done in the future. For example, as the growth of industrial clusters is widely concerned in the context of specific policy, further research direction should explore whether the creation of a zone cultivating stronger clusters (density, breadth, etc.). Second, it is interesting to explore whether in-zone firms' performance would be better during the global COVID-19 pandemic. Such analyses would undoubtedly be of great benefit in understanding the significant role of development zone policy in China.