Lakhasly

factor sharing. Agglomeration enables enterprises to share production factors, such as specialized labor and services. Sharing production
factors can accelerate the flow of knowledge and promote innovation. Gerlach et al. (2009) found that compared with spatially
separated enterprises, those in agglomeration areas had more R&D investment and were more likely to make risky R&D decisions. They
also found that enterprises in agglomeration areas chose more diversified R&D projects. The second is the matching effect of the labor
market. Agglomeration improves the matching quality of the labor market, thereby reducing the costs and increasing the profits of
enterprises, so that enterprises are more capable of carrying out R&D investment. It makes it easier for enterprises to recruit R&D
personnel in the case of the resignation of the existing R&D personnel, which is conducive to the smooth running of R&D projects. The
third is the knowledge spillover effect. Innovation needs inspiration, and the geographical proximity of enterprises is conducive to the
exchange of ideas between technical personnel and promotes the generation of inspiration (Marshall, 1890). Agglomeration can also
promote the dissemination of tacit knowledge.3 Sharing tacit knowledge and the latest innovation among geographically adjacent
enterprises can reduce the uncertainty of enterprise R&D (Feldman, 1994). The fourth is the competition effect. Economic agglomer-
ation intensifies enterprise competition, thus prompting enterprises to accelerate the upgrade of products and promote innovation
(Richardson, 1996).
The existing research on the effect of China's SEZs mainly examines the impact of the SEZs on economic growth (Wang, 2013; Alder
et al., 2016; Liu and Zhao, 2015), export (Wu and Huang, 2012; Huang et al., 2013), industrial restructuring and manufacturing
upgrading (Li and Shen, 2015; Zhou et al., 2018), enterprise behavior (Lu et al., 2019; Wang and Zhang, 2016; Zheng et al., 2017), and
spillover effect (Zheng et al., 2017). There are very few empirical studies that directly investigate the impact of SEZs on innovation. Wu
and Li (2017) measured the innovation ability of enterprises by whether enterprises produced new products and the proportion of the
output of new products in the total output. They used the enterprise address in the Chinese Industrial Enterprise Database during
1998–2007 to identify whether an enterprise was in SEZs, and explored the impact of the SEZs on enterprises' innovation ability. The
estimated results showed that the SEZs inhibited enterprises' innovation ability. Compared with Wu and Li (2017), our study has two
advantages. First, we use the patent data to measure innovation, which can avoid the measurement errors caused by the unclear
definition of new products in the Chinese Industrial Enterprise Database. Secondly, we use the data during 1985–2011, covering the
whole process of the emerging and rising of SEZs, so our conclusion is more reliable