Skip to main content
Log in

Entrepreneurship and industrial clusters: evidence from China industrial census

  • Published:
Small Business Economics Aims and scope Submit manuscript

Abstract

This article studies the synergy effect of entrepreneurship on China’s industrial cluster. We propose an extension to Duranton and Overman’s (The Review of Economic Studies, 72(4), 1077–1106. https://doi.org/10.1111/0034-6527.00362, 2005) method which enables us to delimit industrial cluster in space. A cross-sectional model is identified with historical measures of local entrepreneur potential in the spirit of Chinitz (The American Economic Review, 51(2), 279–289, 1961). Alternatively, we use lagged cluster and entrepreneurship variables to mitigate the endogeneity problem. We find that measures of entrepreneurship contribute significantly to the cluster formation, the cluster size, and the cluster strength. The causality remains strong even when we control for lagged cluster variables. Out of the factors proposed by the NEG theory, access to sea ports, light industry, localization economies, urban population density, and market potential are generally found to benefit cluster. Most of the results are robust to alternative measures of entrepreneurship and instrumental variable.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Fig. 1
Fig. 2

Similar content being viewed by others

Notes

  1. The trend of increasing industrial concentration was observed ever since the mid 1980s and did not slow down until 2005 when rebalancing policies intervened (Wu and Zhu 2017).

  2. Data in Table 1 are taken from the 2007 Large-Scale Survey on Private Enterprises in China. The survey was conducted by the China Private Enterprise Committee that is affiliated to the All-China Federation of Industry and Commerce. The survey was conducted every 2 years since 1993. Because industrial employment outside the public sector was quite small in the 1980s, most of the fourth category (other enterprise staff) actually belonged to the public sector.

  3. Puga (2010) provides an excellent review of these mechanisms.

  4. Their original argument refers to entrepreneurship and any conceptual measure of urban success, which includes industrial clustering. They criticized a few influential empirical studies for failing to address this type of endogeneity.

  5. According to the NEG theory, agglomeration affects entrepreneurship by altering the cost of entry or operation. Such an effect can be interpreted as a movement along the supply curve (Glaeser et al. 2010a, b).

  6. The 2004 census data contain more than 1.2 million firms. In comparison, the 2004 survey data has only 276,826 firms. Thus, 80% of the census data are non-state small firms. The survey data are less appealing because these firms are all excluded.

  7. The secondary industry comprises mining, manufacturing, and utility supply. The majority of the firms in the data belong to manufacturing.

  8. The China Industry Census provides information on the starting time of each establishment.

  9. Our definition of small enterprises follows that of the Interim Provisions on the Standards for Medium and Small Enterprises, which was released by the State Economics and Trade Commission in 2003. A manufacturing firm is small if it meets any of the three criteria: (1) annual sales revenue is under 30 million RMB, (2) total assets are under 40 million RMB, and (3) employment is under 300 persons.

  10. Details of the algorithm are explained in Section A of the Online Resource.

  11. Chinese studies usually report a lower level of industrial concentration than in the west (He et al. 2008; Lu and Tao 2009; Lu 2010). Remarkably, all these studies used data that exclude non-state small firms. When we apply the same algorithm to the China Industry Survey data, only 36% of the industries (172) were found to be clustered. Therefore, data censoring may explain the low level of concentration reported in the literature.

  12. Out of the 12,646 industry-province pairs, most (10,701 or 85%) have no cluster, 1293 (10%) host a single cluster, and 652 (5%) host multiple clusters. Close inspection of the multi-cluster cases usually shows a dominant cluster with one or two satellite clusters that are much smaller in size.

  13. We employ two averaging strategies. The first is the simple average of the radii of all clusters in the industry-province pair, and the second one is weighted by either firm density or employment density. As we explain in the Online Resource (Section A), the algorithm employs one of the density criteria in identifying clusters. Here, we use the same weighting strategy as the one chosen by the algorithm.

  14. The ten largest sea ports of China, in descending order of cargo handling capacity, are Shanghai Port, Shenzhen Port, Ningbo-Zhoushan Port, Qingdao Port, Guangzhou Port, Tianjin Port, Dalian Port, Xiamen Port, Yingkou Port, and Lianyungang Port.

  15. Practically, we divide the area of arable land by the non-urban (agrarian) population.

  16. Officially the 31 provinces are grouped into three regions: East China, Central China, and West China. Empirical studies using provincial data usually control for regional disparities in economic development with these dummies.

  17. We report robust standard errors in all specifications. Even though the robust standard error is preferred in many situations, Greene (2003) cast serious doubt on its validity in maximum likelihood estimation. We also computed the conventional standard errors, but they barely change the inferences. These results are available from the authors upon request.

  18. Equivalently, we test whether the error terms of the structural equation are correlated with the error terms of the first-stage regression. A large test statistic suggests correlation, hence endogeneity. See Wooldridge (2010, p. 665) for details.

  19. The first-stage estimates are presented in Table S3 (Online Resource).

  20. A small value of vary1995 means a more balanced industry mix. Hence, we expect a negative sign on vary1995 if cluster benefits from diversification.

  21. We use the full-sample standard deviations in Table 4.

  22. Following the suggestion of an anonymous referee, we estimated the cluster_bin equation with a linear probability model (LPM), using the same instrumental strategy. This model estimates the marginal effects directly. Empirically, the marginal effect of entrepreneurship is larger if estimated by LPM (0.076). Both LPM and IV-probit yield quantitatively similar marginal effects for port, light, lq1995, and edu1982, but the estimates of other regressors, including land1987, vary1987, lab1995, mp, and soe are very different either in size or level of significance (results available from the authors upon request). The inferences are based on IV-probit estimates because Horrace and Oaxaca (2006) demonstrated that the LPM estimator is inconsistent unless the data satisfy a restrictive condition, which is not in our case.

  23. The standard deviations are taken from the last column of Table 4, i.e., they are computed for the sub-sample of industry-province pairs that host at least one cluster.

  24. We thank an anonymous referee for proposing this empirical strategy.

  25. With the 2004 survey data, we identify 1057 clusters in 172 industries, fewer than what we reported in Sect. 3. In an earlier version of this article, we used this cluster measure to construct the cluster variables in Eq. (1). Most of the qualitative results in Table 5 were retained, but the effects of entrepreneurship were found to be larger in size. These results are available from the authors upon request.

  26. We tried both instrumental valriables (E1985 and TVE1995) on all three measures of entreprenurship (NPE, SPE, PE). Compared to E1985, instrumenting with TVE1995 always generates larger estimated coefficients for the entrepreneurship variable, and we fail to reject the null hypothesis of exogeneity in the cluster_bin equations. Therefore, E1985 may do a better job than TVE1995 in correcting the endogeneity bias. These results are available from the authors upon request.

References

Download references

Acknowledgements

We are grateful to the guest editor Prof. Canfei He and three anonymous referees for their valuable comments and suggestions. The authors acknowledge the financial support from the National Natural Science Foundation of China (Grant Nos. 71373235, 71773112, and 71503232), the National Social Science Fund of China (Grant Nos. 15BJL051 and 13AZD082), the NSFC-ESRC Joint Research Project (Grant No. 71661137002), the Chinese Ministry of Education (Grant Nos. 16JJD790051 and 16JJD790052), Academy of Financial Research at Zhejiang University (Grant No. XK17002), the Cyrus Tang Foundation for CTF Young Scholar, and the Fundamental Research Funds for the Central Universities.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Yiyun Wu.

Additional information

Submit to the special issue on Entrepreneurship in China at Small Business Economics: An Entrepreneurship Journal

Electronic supplementary material

ESM 1

(PDF 637 kb)

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Zhu, X., Liu, Y., He, M. et al. Entrepreneurship and industrial clusters: evidence from China industrial census. Small Bus Econ 52, 595–616 (2019). https://doi.org/10.1007/s11187-017-9974-3

Download citation

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s11187-017-9974-3

Keywords

JEL classification

Navigation