Knowledge Spillovers, Institutional Environment, and Entrepreneurship: Evidence from China

Abstract

The knowledge spillover theory of entrepreneurship (KSTE) predicts a positive relationship between knowledge creation and entrepreneurial activity. As a transitional economy, China exhibits great differences among regions in advancing market reforms and opening-up, largely due to the gradual nature of its economic transformation and opening-up. This situation provides a suitable setting for exploring the role of the institutional environment in the KSTE framework. In this study, we discuss the applicability of the KSTE in the Chinese context and theoretically analyze the role of the institutional environment from aspects of market reforms and opening-up. An empirical analysis based on the data of the Chinese manufacturing sector shows that the KSTE is applicable in China and it is applicable to industries with different technology levels and regions with different levels of economic development. More importantly, we find that both market reforms and opening-up strengthen the positive effect of knowledge creation on entrepreneurship. Our exploration in this field extends the KSTE.

1. Introduction

The view that entrepreneurship is critical to sustainable economic growth has been widely accepted [1,2,3] because the formation of new firms not only can promote economic growth and innovation but also provide employment opportunities for society [4,5]. Then, a question worth thinking about is what factors drive entrepreneurship?

A large amount of literature has investigated the effect of personal ability on entrepreneurship from the perspective of entrepreneurs, such as prior knowledge or experience and cognitive ability [6,7,8]. Other studies have suggested that wage level is an important factor for entrepreneurs to consider the opportunity cost of entrepreneurship [9,10]. In addition, some studies have explored the effects of regional characteristics, such as population, gross domestic product (GDP), and industrial agglomeration, on entrepreneurship [11,12,13]. Although these studies have explained the drivers of entrepreneurship from different perspectives, they fail to identify the source of entrepreneurial opportunities.

Traditionally, the two views on the source of entrepreneurial opportunities are as follows [14]. First is that entrepreneurial opportunities are objective realities that exist in the environment and can be “discovered” as a result of the unique features of individual entrepreneurs [6]. The second view argues that entrepreneurial opportunities are not in an exogenous fashion by the external environment but rather in an endogenous manner through the social skill and imagination of entrepreneurs [15]. Different from the above two views, the knowledge spillover theory of entrepreneurship (KSTE) proposed by Audretsch suggests that entrepreneurial opportunities come from new knowledge and ideas created by but not completely or exhaustively commercialized by incumbent firms and organizations, and the opportunities are endogenous response to regional knowledge contexts [16]. According to the theory, unused or underutilized new knowledge and ideas can be transferred to entrepreneurs and then used for commercialization. In short, the KSTE emphasizes a strong relationship between new knowledge and entrepreneurship.

Recently, much literature has empirically explored the KSTE in many countries. Particularly, most of the literature confirms the theory [17,18,19,20,21]. However, some studies have shown a negative relationship between new knowledge and entrepreneurial activity, and this relationship is related to human capital, related variety, and other factors [22,23]. Other studies have found that the KSTE is only applicable to certain industries. For example, Audretsch et al. [16] and Tsvetkova et al. [24] found that new knowledge only has a positive effect on entrepreneurship in high-tech industries. These studies have enriched our understanding of the KSTE, but few studies have focused on developing countries. Welter et al. [25] have argued that entrepreneurship is context specific. Some theories for entrepreneurship may hold for some national and institutional contexts but not for others [26]. Compared to developed countries, developing countries engage in less knowledge-intensive productive activities and lag further behind in terms of the business environment, frequency of innovation, and entrepreneurship, which poses some limitations in extending the KSTE to the developing country context. However, empirical attention to KSTE in the context of developing countries and their cities has not been abundant [26]. Therefore, whether this theory is applicable to developing countries is worthy of further exploration.

This study not only examines the KSTE in the context of China, the largest developing country, but also discusses and empirically tests the role of the institutional environment on the relationship between new knowledge and entrepreneurship. China has gradually established a market economic system in the past 40 years, and the market vitality has been stimulated, which has also brought remarkable economic growth [27]. In recent years, to realize the transformation of the economic growth pattern and sustainable economic growth, China has paid more attention to innovation and entrepreneurship. The Chinese government proposed the initiative of “mass entrepreneurship and innovation” in 2014 to promote technological innovation, the creation of new knowledge, and entrepreneurship, thereby contributing to the sustainable development of China’s economy. Innovation can provide the impetus for economic growth, and entrepreneurship represents economic vitality, both of which are important factors for sustainable economic development [28,29,30]. Therefore, studying the relationship between innovation and entrepreneurship in the KSTE framework has theoretical and practical significance for achieving sustainable economic development. Then, in China, can the creation of new knowledge promote entrepreneurship as predicted by the KSTE?

Although existing literature has examined the relationship between knowledge creation and entrepreneurship in the Chinese context, further expansions remain [31,32,33,34]. First, the focus of previous studies in the Chinese context is not on the relationship between knowledge creation and entrepreneurship, and they have not further discussed whether this relationship is widespread across industries and regions. Second, these studies have not dealt with endogeneity, which may lead to erroneous estimates. Given that firms are the main source of technological development, an increase in the number of new firms may increase the number of patents. Therefore, the exploration of the KSTE needs to solve reverse causation issues. In addition, omitted variables can also cause endogeneity. Third, most importantly, these studies have not considered the effect of the institutional environment on the relationship between knowledge creation and entrepreneurship. As a transitional economy, China is not only inferior to developed countries in terms of market economy system, but there also exist huge differences in the level of economic development and marketization among various regions within the country, mainly because China has adopted a gradual method in the process of market reforms and opening-up [35]. Prior to the implementation of the economic reforms in 1978, China’s economy was characterized by an introverted development strategy, and it rejected the entry of foreign capital [36]. Moreover, China adopted a planned economy where the production of goods was mainly done by state-owned enterprises (SOEs), and the distribution of resources and commodities was controlled by the government. With the passage of time, the planned economy has become an obstacle to economic development, and China needs to reintroduce market forces and build connections with the outside world. At the end of the 1970s, the Chinese government decided to adopt the policy of market reform and opening-up. The core of the transition from a planned economy to a market economy is to relax or cancel the government’s control over the private economy and promote the growth and development of a non-state-owned economy. To avoid economic and social instability caused by sudden and drastic changes, China adopted gradual reforms [37]. In the beginning, China granted special economic privileges to the southern coastal provinces of Guangdong and Fujian, allowing them to develop a market economy and open to the outside world on a trial-and-error basis. Shortly after, four special economic zones (SEZs) were opened in Shenzhen, Zhuhai, Shantou in Guangdong Province, and Xiamen in Fujian Province. Over time, China’s market reforms and opening-up have been accelerating. Among the world’s major economies, only China has undergone such a dramatic institutional transformation and retained the regional heterogeneity of the institutional environment over the past decades. This situation provides a suitable setting for us to study how the institutional environment affects the relationship between knowledge creation and entrepreneurial activity.

The contribution of this study is threefold. First, this study enriches the literature on the effect of the institutional environment on the relationship between knowledge creation and entrepreneurship. To the best of our knowledge, few studies focus on the role of the institutional environment in the relationship between new knowledge and entrepreneurship. Research has shown that the local institutional environment affects a firm’s behavior and profit expectations [38]. The generation and dissemination of entrepreneurial opportunities and the choice of entrepreneurs to start a business are all generated under the appropriate institutional environment. Therefore, research under the KSTE framework should pay attention to the role of the institutional environment. Our exploration in this field extends the KSTE. Second, this study explores whether the positive relationship between knowledge creation and entrepreneurship exists across industries with different technological content and regions with different levels of development, thus providing broader evidence for the applicability of the KSTE in China. Previous studies have shown that the relationship between knowledge creation and entrepreneurship may vary across industries [16,24]. We examine the industry heterogeneity of this relationship in the Chinese context. Moreover, given the huge regional development imbalance in China, this paper also examines whether there is regional heterogeneity in this relationship. Third, this paper constructs a Bartik instrumental variable (IV) for knowledge creation using China’s manufacturing patent application data and adopts the two-stage least squares approach to address the potential endogeneity problems.

The rest of the paper is organized as follows. Section 2 develops the hypotheses. Section 3 introduces the data and variables. Section 4 presents the econometric model and the results of the econometric tests. Finally, Section 5 summarizes the results of the analysis and explores the implications.

2. Hypothesis Development

2.1. KSTE

The KSTE connects the new knowledge with entrepreneurship based on the theories of knowledge uncertainty and knowledge spillover. Knowledge uncertainty refers to the commercial value of new products brought by innovation that cannot be accurately estimated because of the uncertainty of knowledge-based economic activities [39]. Confronted with uncertainty, decision-makers tend to maintain the status quo rather than choose to act upon new ideas for which no expected value and commensurate probability distribution corresponding to possible outcomes can be estimated. Hence, new knowledge and ideas may not be commercialized by incumbent firms [16]. This un-commercialized new knowledge and ideas as the source of entrepreneurial opportunities can be transferred to entrepreneurs through knowledge spillover.

Studies have shown that knowledge spillover, especially its tacit component, is spatially bounded [40,41,42]. Given this, knowledge spillover entrepreneurship is likely to be spatially bounded. In the context of Germany, Audretsch et al. [43] confirmed that knowledge spillover entrepreneurship is also limited by the geographical scope of the origin of knowledge. They believed that universities, as one of the main production institutions of knowledge, can transfer knowledge to enterprises through academic research and university graduates. With universities’ increasing knowledge investment and academic achievements, the number of knowledge-based startups around universities has increased significantly, which indicates that knowledge-based startups are located within geographic proximity to knowledge sources. Lee et al. [44] indicated that although a positive effect of new knowledge could be observed on new firms within and across the regional boundaries in Korea, they found that the intra-regional effect is stronger than the inter-regional effect. Most studies on the KSTE have only explored the intra-regional effect. Audretsch et al. [16] found a positive relationship between the amount of investment in knowledge in the region (measured by regional R&D intensity) and entrepreneurship in Germany, but the evidence is limited to high-technology entrepreneurship.

Although a large body of research gives evidence supporting the KSTE, exceptions remain. (Many studies have used data from various countries to find evidence supporting KSTE, such as Audretsch et al. [16] for Germany, Lee et al. [44] for South Korea, Qian et al. [17], Plummer et al. [18] for the US, Colombelli et al. [21], and Kanellopoulos et al. [20] for Italy). Tsvetkova et al. [24] found that regional new knowledge measured by the number of patents per 1000 residents has a negative effect on entrepreneurship in the high-tech goods-producing sector in the US. In this regard, their explanation is that in the high-tech goods-producing sector, the decreased entry as a result of more intensive patenting is likely to stem from intensified competition with more technologically advanced incumbents, increasing the costs of creating a new firm. Moreover, incumbents that have continuous R&D investment tend to operate well, pay high wages, and offer stable employment conditions, thereby hindering employees’ entrepreneurial behavior [9]. In addition, Qian et al. [22] indicated that although new knowledge contains entrepreneurial opportunities, large incumbents have advantages in appropriating their market value over new firms. Therefore, a higher level of new knowledge production does not necessarily lead to more dynamic entrepreneurial activity. Ejdemo et al. [23] also found that the number of patents per capita negatively affects entrepreneurship in Sweden. They suggested that when incumbents efficiently appropriate new knowledge by patenting, knowledge spillover entrepreneurship will be limited.

Previous research has contributed to the improvement of the KSTE in several ways, such as entrepreneurial capabilities, localized competition, and the characteristics of local knowledge bases. Studies have confirmed that personal abilities, including prior knowledge or experience [6,7,8] and cognitive ability [45,46], affect entrepreneurs to discover entrepreneurial opportunities [47]. The asymmetries in knowledge across individuals also create asymmetries in opportunities across individuals [48]. The commercial value of new knowledge and ideas will vary across individuals due to differences in personal knowledge and experience among those individuals. Qian et al. [48] defined entrepreneurial absorptive capacity as the ability of an entrepreneur to understand new knowledge, recognize its value, and subsequently commercialize it by starting a business. They also confirmed that entrepreneurial absorptive capacity is a critical determinant of knowledge spillover entrepreneurial activity. In addition, Plummer et al. [18] extended the KSTE to contend that localized competition hampers entrepreneurial activities by reducing the incentive to exploit new knowledge. They tested this conjecture in the contexts of Colorado and California. They suggested that localized competition has a two-fold effect on knowledge spillover entrepreneurship. That is, it not only increases the pool of opportunities available for discovery by entrepreneurs but also reduces the share of the opportunities entrepreneurs exploit. From the perspective of the characteristics of the local knowledge bases, Colombelli [19] and Colombelli et al. [21] investigated the effects of knowledge variety and knowledge similarity on the creation of new firms in Italy. The authors confirmed that the characteristics of local knowledge bases play a key role in shaping the creation of new firms.

To our knowledge, the majority of the KSTE empirical tests are for developed countries, with very limited evidence from developing countries [26]. (Except for a few studies in the context of China, only Iftikhar et al. (2020) have studied the KSTE in developing countries. They confirmed the KSTE in the Pakistani context.) In China, market mechanisms in many regions are still imperfect [35,49]. Imperfect market mechanisms can inhibit entrepreneurial activities, which may limit the applicability of the KSTE in China. However, as reform and opening-up progressed, a large number of market access restrictions were removed, and non-state enterprises continued to enter the market. This has unleashed China’s economic dynamism and greatly improved the working and living environment for workers. The environment where individuals work and live is considered to be directly related to the individual action of discovery and exploitation of new business opportunities [50,51]. The implementation of market reforms and opening-up provided the necessary living conditions and business environment for Chinese entrepreneurs to discover and utilize business opportunities. According to the KSTE, the improvement of an entrepreneur’s ability to discover and utilize business opportunities will promote the generation of knowledge spillover entrepreneurship [17]. Therefore, China’s market reforms and opening-up are conducive to the emergence of knowledge spillover entrepreneurship. In addition, the richness of the regional knowledge environment also promotes knowledge spillover entrepreneurship [48]. Over the past four decades, the Chinese government has been actively encouraging and supporting the technological innovation activities of enterprises or scientific research institutions, thus providing numerous sources of knowledge for knowledge spillover entrepreneurship. Therefore, we expect the KSTE to be applicable in China and propose the following.

Hypothesis A.

Knowledge creation spurs the emergence of new entrepreneurial activities in China.

The above hypothesis is consistent with the conclusions of some recent studies in the Chinese context [31,32,33,34]. Although these studies have found a positive effect of knowledge creation on entrepreneurship in the Chinese context, they do not further examine whether this positive effect exists in different industries and different regions. Previous studies have shown that the relationship between knowledge creation and entrepreneurship may vary across industries [16,24]. Therefore, the effect of knowledge on entrepreneurship also needs to consider industry heterogeneity. In addition, China has huge regional development imbalances. In comparison with the central and western regions, the eastern region has significant advantages in terms of economic development, innovation capabilities, and entrepreneurship vitality. Given the importance of the knowledge environment in the KSTE framework [48], the exploration of the applicability of the KSTE needs to be conducted in regions with different levels of development.

2.2. Role of Institutional Environment

A growing stream of research suggests that the quality of a region’s institutions plays an important role in accounting for disparities in rates of entrepreneurship and innovation across regions [52]. Anokhin et al. [53] argued that corruption undermines the foundations of institutional trust that are needed for the development of trade and entrepreneurial and innovative activities. Similarly, Boudreaux et al. [38] found that institutional constraints, such as cumbersome labor, credit, or business market regulations, can significantly hinder people’s ability to pursue entrepreneurship. They found that economic freedom not only channels individual effort to entrepreneurial action but also affects the extent to which individuals’ socio-cognitive resources are likely to mobilize and lead to high-growth entrepreneurship. Recently, Tavassoli et al. [53] suggested that a regional environment with more local openness can promote contact and interaction among people, thereby facilitating the generation of new knowledge. Given that the effect of institutions on entrepreneurial action has been widely confirmed by previous studies [54,55,56], we have reason to infer that the institutional environment of a region will significantly affect the knowledge spillover entrepreneurship in that region. However, the existing literature still lacks empirical exploration of this topic.

China’s special development path provides a suitable setting for us to study the role of the institutional environment under the KSTE framework. At the end of the 1970s, the Chinese government decided to adopt the policy of reform and opening-up. The core of the transition from a planned economy to a market economy is to relax or cancel the government’s control over the private economy and promote the growth and development of a non-state-owned economy. To avoid economic and social instability caused by sudden and drastic changes, China has adopted gradual reforms [37]. Although economic reforms have helped restructure China’s social, economic, and political institutions, they have created huge differences in the institutional environment among regions [35]. In the process of China’s transition to a market economy, market reforms and opening-up are the two most important aspects of a series of institutional changes. Next, we will demonstrate how the institutional environment affects the relationship between knowledge creation and entrepreneurial activity from two aspects: market reforms and opening-up.

2.2.1. Market Reforms

During the planned economy in China, the production of goods was mainly done by SOEs, and the distribution of resources and commodities was controlled by the government. Thus, the main purpose of market reforms in China is to relax or cancel the control of the government and state-owned enterprises over the market. Due to the gradual nature of China’s economic reforms, the depth of transition varies across regions leading to considerable regional differences in terms of local levels of market reforms [57]. In regions where the market reform process is lagging behind, the degree of marketization is low, the flow of resources and goods is under government control, and the boundary between the market and the government is blurred [58]. Local governments have the discretion to redistribute resources or enforce the rules and welfare that affect enterprises. This power can easily become a tool for government officials to seek personal gains, leading to corruption [59]. In addition, a notable feature of regions with a low level of market reforms is that state-owned enterprises (SOEs) occupy an important position in the economy [57]. Next, we discuss how the lag in market reforms can restrain the positive effect of knowledge creation on entrepreneurship from aspects of corruption and SOEs.

First, corruption can affect entrepreneurial profit expectations, increase the uncertainty of entrepreneurial activities, and distort the way firms improve competitiveness, thereby inhibiting knowledge spillover entrepreneurial activities. In the KSTE framework, the asymmetries in knowledge across individuals also create asymmetries in entrepreneurial opportunities across individuals [48]. For the same new knowledge, individuals with different knowledge reserves also have different expectations of their commercial profits. Corruption may reduce the profit expectations of entrepreneurs who want to start knowledge-driven entrepreneurship. The inventors or owners of new knowledge who expect low profitability of new knowledge may refuse to commercialize the knowledge. In regions with a high level of corruption, government officials may use their privileges to collect bribes from business owners or private agents [60,61]. This phenomenon brings additional operating costs to the business and reduces the expected profitability. Moreover, corruption lowers people’s expectations of public officials and others who directly or indirectly participate in transactions to comply with discipline and the law. The cost to offenders can be reduced or even offset by bribery, which increases the uncertainty of the transaction [52]. The increasing uncertainty of transactions will reduce the profit expectations of entrepreneurship, thereby reducing the desire of entrepreneurs to use new knowledge to start their own businesses. In addition, some firms may actively pay bribes to obtain bank financing and government contracts [61]. These rent-seeking behaviors enable incumbents to obtain market privileges, stabilize their monopoly position, and disrupt the marker mechanism [62]. Corruption also distorts the way firms enhance market competitiveness and increases the probability of firms investing in rent-seeking activities [63]. These effects of corruption undoubtedly reduce the desire of entrepreneurs to improve firm competitiveness through innovation and are adverse to the exchange and dissemination of knowledge, which is necessary for the exploration and development of knowledge [64]. Therefore, corruption inhibits knowledge spillover entrepreneurship.

Second, the close ties between SOEs and the government will reduce the willingness of entrepreneurs to use new knowledge to create new businesses. In China, in addition to controlling industries that are of strategic significance to national security and economic development, SOEs also bear many social burdens, such as increasing employment [65]. This phenomenon deepens the ties between SOEs and the government. In regions where the market reform process is lagging, SOEs still dominate the local market [57]. The close ties between SOEs and the government prompt local governments to formulate market rules and regulations that are conducive to SOEs and provide important market information and resources to SOEs [66,67]. However, this circumstance causes many negative effects. For instance, the behavior of state-owned banks in giving priority to the loan demands of SOEs disrupts the allocation of financial resources in the market and squeezes the resource of non-SOEs [68], thereby increasing entry barriers and difficulty of starting a business. Moreover, the privileges of SOEs have also led many firms to tend to sign commercial contracts with SOEs [57], which reduces the profitability of entrepreneurship and the potential entrepreneurs’ willingness to use un-commercialized knowledge to start a business.

Third, the lack of incentive mechanisms in SOEs affects the willingness of state-owned employees to innovate, thereby reducing the positive effect of knowledge creation on entrepreneurship. As SOEs are governed by administrative rather than economic imperatives [69], the motivation and ability of SOEs to innovate are weak [70]. Given the lack of incentives [71], obtaining sufficient benefits from innovation and opening up new markets is difficult for SOE managers. Under such circumstances, SOE managers are more inclined to complete their administrative tasks [72] rather than facing risks to engage in innovation and new business [68]. In addition, SOEs can provide employees with stable jobs [57,73]. A work environment that lacks innovation incentives weakens employees’ willingness to innovate, and stable work also reduces the probability of employees leaving to start a business and the flow of employees between different firms [9]. All of these phenomena reduce people’s pursuit of innovation and entrepreneurship and the flow of personnel among enterprises, hinder the exchange and dissemination of knowledge in the region, and ultimately reduce the positive effect of knowledge creation on entrepreneurship.

The situation is different in regions where the market reforms are relatively faster. In these regions, the boundary between the market and the government is no longer blurred, and the allocation of resources is mainly realized by the market mechanism, which reduces the probability of corruption, as well as the administrative and bureaucratic barriers to entrepreneurship. Moreover, in these regions, private enterprises have developed rapidly. In comparison with SOEs, the political ties between private enterprises and the government are weaker, which makes obtaining cheap resources through the government difficult for private enterprises. Therefore, the pressure of survival means that private enterprises usually have a stronger willingness to innovate than SOEs. In short, market-oriented reforms are conducive to raising potential entrepreneurs’ profit expectations for innovation and entrepreneurship, thereby promoting the development, exchange, and dissemination of knowledge and ultimately promoting the positive effect of knowledge creation on entrepreneurship.

On the basis of the discussion above, we propose the following.

Hypothesis B.

In China, knowledge creation promotes entrepreneurship more strongly in regions with a higher level of market reforms compared with regions with a lower level.

2.2.2. Opening-up

Prior to the implementation of the economic reforms in 1978, China’s economy was characterized by an introverted development strategy, and it rejected the entry of foreign capital [36]. With the passage of time, the planned economy has become an obstacle to economic development, and China needs to reintroduce market forces and build connections with the outside world. China first granted special economic privileges to the southern coastal provinces of Guangdong and Fujian, allowing them to develop a market economy and open to the outside world on a trial-and-error basis. Shortly after, four special economic zones (SEZs) were opened in Shenzhen, Zhuhai, Shantou in Guangdong Province, and Xiamen in Fujian Province. In 2001, China joined the WTO, which was a milestone in the process of China’s opening to the outside world and the globalization of the world economy. Although China’s opening-up has been greatly improved overall, enormous gaps still exist among regions [57], which may be an important reason for the differences in knowledge spillover entrepreneurship in various regions. We demonstrate this view from the following two aspects.

First, opening-up has brought advanced production technology and management knowledge to China, thereby providing a new source of knowledge spillovers for domestic firms. In comparison with domestic firms in developing countries, firms in developed countries have more advanced production technology and management knowledge [74]. Domestic firms can learn foreign advanced technologies through imitation or reverse engineering [75,76]. Moreover, through labor mobility, knowledge can also be spilled from foreign-invested enterprises to local enterprises [77]. In addition, foreign-invested enterprises need to purchase intermediate goods from domestic suppliers. To meet the demand for high-quality intermediate goods, some foreign-invested enterprises will provide domestic suppliers with technical assistance and staff training to optimize supplier management and improve production efficiency and products [78]. Therefore, knowledge spillovers from foreign-invested enterprises can stimulate the desire of domestic firms to pursue knowledge and technology, thereby increasing the frequency of knowledge exchange and ultimately spurring the emergence of knowledge spillover entrepreneurship. Moreover, the spillover of foreign-invested enterprises’ advanced technology and management knowledge has also improved the entrepreneurial absorptive capacity, which helps increase entrepreneurship triggered by knowledge creation. (Entrepreneur absorptive capacity is the ability of entrepreneurs to understand new knowledge, recognize its business value, and have the ability to create new businesses [17].) The dissemination of technical and management knowledge enables entrepreneurs to have the ability to understand knowledge and its commercial value and provides the necessary management knowledge for entrepreneurs to create new firms.

Second, opening-up promotes the development of China’s foreign trade. The opening-up policy allows foreign capital to enter the Chinese market. In China, more than 60% of exports are made by foreign-invested enterprises [79]. In addition to the fact that FDI can bring new knowledge to local firms, trade activities among countries are conducive to the cross-border transmission of knowledge [80]. Therefore, opening-up can promote the spread of advanced foreign technology and knowledge to the host country’s local market through trade channels, which is conducive to knowledge spillovers, thereby increasing the probability of the emergence of knowledge spillover entrepreneurship.

On the basis of the abovementioned analysis, we propose the following.

Hypothesis C.

In China, knowledge creation has a stronger positive effect on entrepreneurship in regions with a higher level of opening-up compared with regions with a lower level of opening-up.

3. Data and Variables

3.1. Data Source

The city is the basic geographic unit in this study. This geographical level is more fine-grained than the provincial level, thereby allowing us to better measure the difference in the regional institutional environment. Knowledge creation is measured by the number of applications for invention and utility patents, and the level of entrepreneurship is measured by regional manufacturing new firm formation rates, both at the city level. (According to the Patent Statistics Annual Report published by the China National Intellectual Property Administration in 2008, in China, invention and utility patents are concentrated in the manufacturing sector, which is the most important sector and source of growth for Chinese economy. Available at https://www.cnipa.gov.cn/tjxx/jianbao/year2008/indexy.html (accessed on 15 March 2021).)

The city-level patent application data are from the website of the China National Intellectual Property Administration. The data on new manufacturing firms are from the Annual Survey of Industrial Enterprises (ASIE) between 2002 and 2013, conducted by the National Bureau of Statistics of China. It includes all SOEs and non-SOEs with revenues above RMB 5 million. (After 2010, the threshold for inclusion for non-SOEs was raised from RMB 5 million to 20 million of the total sales.) The firms recorded in ASIE account for about 90% of China’s total industrial output, about 70% of the number of employees, and about 97% of exports [81], which can well reflect the operating conditions of China’s industrial sector. Given that a few variables in ASIE are noisy and misleading due to misreporting by some firms, this study follows Feenstra et al. [82] to delete observations if any of the following rules are met: (1) the annual average number of employees is less than 8; (2) the total assets are less than or equal to 0; (3) the total industrial output is less than or equal to 0; (4) the total current assets are greater than total assets; (5) the total fixed assets are greater than total assets; (6) the total wages are less than or equal to 0.

Other economic data at the city level used in this study are from the China City Statistical Yearbook. Cities lacking a large amount of data were excluded, leaving 260 cities in the end. To eliminate the effect of data fluctuations on the estimation results, the three years moving average strategy is performed for the number of patent applications.

3.2. Dependent Variable

The dependent variable of this study is regional manufacturing new firm formation rates, usually measured by the “ecological” approach and the “labor force” approach [20,83]. The ecological approach standardizes the number of new firms in a city by dividing them by the number of firms in existence at the beginning of the corresponding year each time. This approach essentially assumes that new firms emanate from the stock of existing firms and does not consider the difference in sizes between new and existing firms [84]. Therefore, the new firm formation rate can be biased upward in cities dominated by fewer larger firms [83]. The labor force approach standardizes the number of new enterprises with respect to the size of the regional labor force. This approach appears to accord with the fact that it is ultimately real people who open businesses; thus, it is widely used to measure the level of regional entrepreneurship [16,17,18,24].

Following previous studies, we define the dependent variable as the logarithm of the number of new firms per 1000 employees in the manufacturing sector at the city level (FIRMS). Notably, although we can obtain data on Chinese industrial enterprises from 2009 to 2013, ASIE did not record the new firms in 2009, and the statistical caliber for non-SOEs has been adjusted from RMB 5 million to 20 million after 2010. Thus, we do not include new firms after 2008. To solve the possible endogeneity problems of knowledge creation, we construct a Bartik IV. Since the construction of the Bartik IV requires patent data by industry in 2000, the statistical period for explanatory variables is 2001–2007. In addition, since explanatory variables are lagged to avoid reverse causation, the statistical period for new firms is 2002–2008. (The earliest year of patent application data by industry in China’s official statistics is 2000. Given the construction characteristics of the Bartik IV, we set 2000 as the base year. In order to satisfy the exogeneity of instrumental variables, the statistical period for explanatory variables is 2001–2007.)

In addition, a non-SOE was recorded by the ASIE only when its total sales reached the threshold. Therefore, if only counting new firms that were first recorded in the database in the year of establishment, the number of new firms would be underestimated. Take a firm established in 2008 as an example. It may only reach the total sales threshold in 2010; thus, it was first recorded in the ASIE in 2010, not in 2008. ASIE records the year of establishment of the firm, which allows us to classify it as a new firm established in 2008. Although we only counted the number of new enterprises established in 2002–2008, we extended the observation period to 2013 to reduce the measurement error. In this manner, new firms that were established before 2008 but were first recorded in the database during the period 2009–2013 can be identified.

3.3. Explanatory and Instrumental Variables

According to the KSTE, un-commercialized new knowledge and ideas constitute the main source of entrepreneurial opportunities. Therefore, measuring the regional knowledge production or regional stock of new knowledge has become the basis for selecting the main explanatory variables in the KSTE research. Previous studies have often used the number of regional patents per capita to measure the regional stock of new knowledge [18,22,23]. Although patents do not capture non-technical knowledge [18], they reasonably estimate the new knowledge created by innovative activity [85]. We use the number of patent applications per 1000 manufacturing employees to measure the stock of new knowledge at the city level (PATENT). (In China, the three types of patents are invention patent, utility model patent, and appearance design. As appearance design patents do not relate to technical solutions, they are not included in this study.)

Using PATENT as an explanatory variable may cause endogeneity problems. Given that firms are the main economic organization for technological development, the increase in the number of new firms may increase the number of patents. Although lagged explanatory variables alleviate potential reverse causation issues [86], there may still be unobserved factors that affect the development of knowledge and the new firm formation simultaneously, thereby causing the omitted variables problem. To solve the possible endogeneity problems, we construct a Bartik IV (PATENTIV) of PATENT, which is calculated as follows: (Bartik combined the local industry composition with national growth in employment across industries to isolate local labor demand shocks [87]. The Bartik instrument is also called the shift-share instrument, which is formed by interacting national industry growth rates and local industry shares [87]. The Bartik approach and its variants have been used in labor economics, resource economics, regional economics, and other economic fields [88,89,90].)

$$PATENTI{V}_{rt}=\ln \left[\frac{paten{t}_{ro}{{\displaystyle \sum }}_i\left(e_{ri0}\ast g_{it}\right)}{E_{rt}}\right]$$

(1)

where r and t refer to city and year, respectively. o refers to the base year (2000), and i refers to industry at a two-digit level. $e_{ri0}$ is employment share of industry i in city r in the base year, as the measurement of the initial industrial structure of city r. $g_{it}$ is the patent growth index of industry i in year t relative to the base year at the national level. (The number of patent applications in the manufacturing sector at the national level is from the Statistics on Science and Technology Activities of Industrial Enterprises.) $E_{rt}$ is the manufacturing employment (in thousands) in city r in year t. By calculating the inner product of $e_{ri0}$ and $g_{it}$, that is, ${\displaystyle \sum }_i\left(e_{ri0}\ast g_{it}\right)$, we can obtain the estimated value of the patent growth index ($\widehat{g_{rt}}$). By multiplying $\widehat{g_{rt}}$ and $paten{t}_{r0}$, the number of patents of city r in the base year, we can obtain $\widehat{paten{t}_{rt}}$, which is the estimated value of the number of patent applications ($paten{t}_{rt}$). $\widehat{paten{t}_{rt}}$ and $paten{t}_{rt}$ are related. (We illustrate the correlation between $\widehat{paten{t}_{rt}}$ and $paten{t}_{rt}$ through the construction process of $\widehat{paten{t}_{rt}}$. First of all, $paten{t}_{rt}$ can be expressed as: $paten{t}_{rt}=paten{t}_{r0}\frac{paten{t}_{rt}}{paten{t}_{r0}}=paten{t}_{r0}\ast {\displaystyle \sum }_i\frac{paten{t}_{ri0}}{paten{t}_{r0}}\ast \frac{paten{t}_{rit}}{paten{t}_{ri0}}=paten{t}_{r0}\ast {\displaystyle \sum }_i{s}_{ri0}\ast g_{rit}=paten{t}_{r0}\ast \widehat{g_{rt}}$, where $g_{rit}$ is the patent growth index of industry i in city r, and $s_{ri0}$ is the patent share of industry i in city r in the base year. In theory, we can decompose $g_{rit}$ as $g_{rit}=g_{it}+g_{rt}+{\tilde{g}}_{rit}$, where $g_{it}$ is the growth index of industry i in patents at the national level, $g_{rt}$ is the growth index of city r in patents, and ${\tilde{g}}_{rit}$ is the idiosyncratic industry–city growth index in patents. We use $g_{it}$ instead of $g_{rit}$ to obtain $\widehat{g_{rt}}={\displaystyle \sum }_i{s}_{ri0}\ast g_{it}$,which is a hypothetical projection of $g_{rt}$. Multiplying $paten{t}_{r0}$ and $\widehat{g_{rt}}$ yields the instrumental variable ($\widehat{paten{t}_{rt}}$) of $paten{t}_{rt}$, which is the expected number of local patents for each industry in city r to grow according to its patent growth rate at the national level. Given that $g_{it}$ and $g_{rit}$ are related, $g_{rt}$ is related to $\widehat{g_{rt}}$, and $\widehat{paten{t}_{rt}}$ is then related to $paten{t}_{rt}$. Given that China does not have patent statistics by industry in region, $s_{ri0}$ cannot be calculated. As an alternative, we use $e_{ri0}$ instead of $s_{ri0}$, which will not make the correlation between $\widehat{paten{t}_{rt}}$ and $paten{t}_{rt}$ disappear.) In addition, $\widehat{paten{t}_{rt}}$ is exogenous by construction because it relies on national industry growth in patents.

3.4. Institutional Environment Variables

We use the share of state-owned paid-in capital (SOEI) in the total paid-in capital of a city’s manufacturing sector to reversely measure the city’s market reforms level. A smaller SOEI reflects a higher level of market reforms. Regions at the relatively earlier stages of market reforms are often dominated by a few large SOEs, which weakens inter-firm competition because SOEs often lack strong incentives to develop their firm-specific advantages [57]. In such regions, SOEs generally have closer ties with local governments [66]; thus, local governments have strong incentives to achieve their political goals through SOEs [65,91]. Local governments have greater control over the local market dominated by SOEs; therefore, the local market reform level is low [92].

Prior to 1978, China’s economy was characterized by an introverted development strategy and rejected the entry of foreign capital [36]. In 1978, China decided to open up to the outside world, which not only brought in a large amount of foreign capital but also promoted the development of foreign trade. Therefore, we use the average of the share of foreign paid-in capital in the total paid-in capital of a city’s manufacturing sector and the share of exports in the total output of a city’s manufacturing sector as an index to measure the level of opening-up of the city (OPEN). A larger OPEN means a higher level of openness.

3.5. Control Variables

We added regional economic variables that may affect knowledge spillovers into the control variables, such as specialization (RS), related variety (RV), unrelated variety (UV), population density (PD), and human capital (HC).

Specialization refers to the geographic concentrations of firms in the same industry that can bring specialized labor, specialized input, technology spillover, and high demand [93]. The spatial aggregation of firms in the same industry facilitates the dissemination of knowledge and technology within the industry and promotes innovation [94,95]. Following Kanellopoulos et al. [20], we calculate the specialization index as follows:

$$RS=\frac{1}{2}\ast {\displaystyle \sum }_j\left|\frac{E_{jr}}{E_r}-\frac{E_{jn}}{E_n}\right|$$

(2)

where $E_{jr}$ is the employment of the four-digit industry j in city r; $E_{jn}$ is the total national employment of industry j; and $E_r$ and $E_n$ represent all manufacturing employment in city r and all manufacturing employment in China, respectively. A larger RS means a higher level of specialization in the city.

Related variety is considered the best measure of Jacobs externalities, that is, knowledge spillovers between related industries [53], whereas unrelated variety provides the building blocks for technological “breakthroughs” caused by combinations across unrelated knowledge domains [96]. Frenken et al. [97] defined related variety as within-industry diversity and unrelated variety as between-industry diversity. They claimed that only related variety would enhance knowledge spillovers, whereas unrelated variety would produce a portfolio-like effect, thereby improving the ability of regional economies to resist external risks and protect firms from special demand shocks. Following Frenken et al. [97], RV and UV are calculated as follows:

$$RV={\displaystyle \sum }_{i=1}{P}_i{\displaystyle \sum }_{j\in i}\frac{P_{ij}}{P_i}lo{g}_2\left(\frac{1}{P_{ij}/P_i}\right)$$

(3)

$$UV={\displaystyle \sum }_{i=1}{P}_{i}lo{g}_2\left(\frac{1}{P_i}\right)$$

(4)

where $P_{ij}$ is the employment share of four-digit industry j in the two-digit industry i that industry j belongs to in a city. $P_i$ is the employment share of two-digit industry i in a city in the total manufacturing industry of the city.

Previous studies have shown that cities provide greater proximity between people, which promotes their more frequent interaction and knowledge spillovers [98]. Following Frenken et al. [97] and Boschma et al. [99], we use population density (thousand people/km²) to measure urbanization to control the effect of urbanization on entrepreneurship.

High regional human capital can create an environment rich in local knowledge spillovers [100,101]. This study measures the human capital of a city with the number of employees in the science and technology service industry per thousand manufacturing employees in the city. The increase in human resources in R&D is not only conducive to knowledge creation but also the exchange of knowledge.

In addition to the abovementioned variables, we also consider some city-level economic variables affecting entrepreneurship, such as per capita wages (SALARY), economic growth rate (GDPGR), unemployment rate (UNEMP), financial expenditures of local governments on scientific research (GOV), the growth of the service sector (SERVICE), and localized competition (COMP). Per capita income can measure the opportunity cost of regional entrepreneurship [9,10]. Audretsch et al. [2] showed that entrepreneurship is positively correlated with economic growth. Regional economic growth can also measure local demand growth [12], which affects the profits of start-ups. The existing literature does not have a unified conclusion on the effect of unemployment on entrepreneurship. Some studies have suggested that unemployment promotes entrepreneurship [101,102], but other studies have reached the opposite conclusion [12,103].

The degree of government support for scientific research and commercialization of scientific knowledge will affect knowledge-based innovation and entrepreneurship [104]. We use the share of R&D expenditures in local financial expenditures to measure the local government’s support for scientific research activities. The output share of the service sector in the local economy is added to the baseline regression to control the effect of the service sector growth on entrepreneurial activities in the manufacturing sector.

Finally, we control for the effect of localized competition (COMP) on entrepreneurship. Following Plummer et al. [18], COMP is calculated as follows:

$$COMP=\frac{F_r}{E_r}/\frac{F_n}{E_n}$$

(5)

where $F_r$ and $F_n$ refer to the number of manufacturing firms in city r and the number of manufacturing firms in the country, respectively. The meanings of $E_r$ and $E_n$ are as described above. Thus, COMP is defined as the number of manufacturing establishments in the city per worker in the city relative to the number of manufacturing establishments in the country per worker in the country. A higher COMP means higher competition among local manufacturing firms.

4. Empirical Analysis

4.1. Baseline Model

To test the effect of the institutional environment on the relationship between knowledge creation and entrepreneurship, we initially need to test whether Hypothesis A is supported, that is, whether evidence can be found to support the KSTE in China. On the basis of previous research on KSTE, we construct a baseline model estimated by the two-stage least squares (2SLS) method, as shown as follows:

$$\widehat{PATEN{T}_{rt}}=\alpha +\gamma PATENTI{V}_{rt}+X_r\phi +{\epsilon }_{rt}$$

(6)

$$FIRM{S}_{rt}=\alpha +{\beta }_1\widehat{PATENT{S}_{r t-1}}+{\beta }_{2}SO{E}_{r t-1}+{\beta }_{3}OPE{N}_{r t-1}+X_{r t-1}\rho +{\mu }_r+{\sigma }_t+{\epsilon }_{rt}$$

(7)

where subscript r refers to a city; subscript t refers to a year; $X$ is a vector of control variables; φ and ρ are the coefficient vectors of the control variables; ${\mu }_r$ and ${\sigma }_t$ refer to city and year fixed effects, respectively, and ${\epsilon }_{rt}$ is the error term. Equations (6) and (7) are the regressions of the first and second stages, respectively. All exogenous variables are considered in the first stage to obtain a consistent estimator. To save space, all the estimation results of the first stage of 2SLS in this study are not reported. (The results of the first stage are available at the request of readers.) All explanatory variables in Equation (7) are lagging by one period to alleviate the problem of reverse causation [86].

The descriptive statistics of each variable are shown

Table A1. Descriptive statistics.

Variables	Obs	Mean	Std. Dev.	Min	Max
FIRMS	1820	−0.5861	0.7162	−3.4982	1.4440
PATENT	1820	0.7870	0.8386	−2.6989	5.3216
PATENTIV	1820	1.4916	0.9503	−2.0339	4.9152
SOEI	1820	0.2625	0.2012	0.0004	0.9486
OPEN	1820	0.1014	0.1004	0.0000	0.5124
RS	1820	0.4281	0.0412	0.2956	0.5642
RV	1820	1.5901	0.5770	0.0280	3.0915
UV	1820	3.5958	0.5873	0.5777	4.4600
lnPD	1820	−1.1119	0.8044	−4.1997	0.9958
lnHC	1820	3.3383	0.9974	−0.2284	6.2787
lnSALARY	1820	2.5935	0.4454	−4.6244	3.8981
GDPGR	1820	0.1591	0.0898	−0.2741	0.9327
UNEMP	1820	0.0649	0.0392	0.0031	0.4792
GOV	1820	0.0159	0.0334	0.0000	0.5950
SERVICE	1820	36.1107	7.7606	8.6100	72.0900
COMP	1820	1.0463	0.7311	0.0148	7.8497

in Appendix A. According to

Table A2. Correlation matrix.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)	(15)	(16)
(1)FIRMS	1
(2)PATENT	−0.05	1
(3)PATENTIV	−0.07	0.79	1
(4)SOEI	−0.19	−0.06	−0.12	1
(5)OPEN	−0.02	0.03	−0.07	−0.44	1.00
(6)RS	−0.25	0.01	0.06	0.09	−0.09	1
(7)RV	0.09	0.21	0.18	−0.39	0.47	−0.34	1
(8)UV	0.23	0.13	0.12	−0.36	0.27	−0.50	0.69	1
(9)lnPD	0.07	0.04	−0.02	−0.26	0.35	−0.17	0.51	0.42	1
(10)lnHC	0.04	0.41	0.46	0.22	−0.38	−0.07	−0.19	0.03	−0.31	1
(11)lnSALARY	−0.21	0.39	0.43	−0.28	0.31	0.11	0.29	0.08	0.13	−0.06	1
(12)GDPGR	−0.13	0.12	0.23	−0.12	0.01	0.11	−0.01	−0.12	−0.07	0.03	0.32	1
(13)UNEMP	0.16	0.02	0.13	0.03	−0.11	−0.09	−0.10	0.01	−0.14	0.13	−0.07	−0.04	1
(14)GOV	−0.05	0.07	0.20	−0.16	−0.03	0.04	0.02	0.01	−0.07	0.02	0.15	0.11	0.01	1
(15)SERVICE	−0.09	0.30	0.27	−0.03	0.20	−0.17	0.28	0.33	0.10	0.36	0.12	−0.11	0.08	−0.04	1
(16)COMP	0.50	0.05	0.12	−0.36	0.06	−0.07	0.15	0.21	0.03	−0.01	0.10	0.03	0.13	0.09	−0.03	1

in Appendix A, the correlation coefficient between PATENTIV and PATENT is 0.79; thus, the weak instrument problem should not exist. UV and RV are highly positively correlated with a correlation coefficient of 0.6923, which has similar findings in existing studies [19]. Except for the absolute value of extremely few correlation coefficients exceeding 0.5, the absolute value of most correlation coefficients is less than 0.5. Therefore, regressions in this study do not have serious multicollinearity. For the sake of caution, in the regressions that follow, we check the variance inflation factor (VIF) for independent variables to discover any multicollinearity among them. We find that the VIF values of all variables are lower than 3, which is considerably lower than the cut-off value of 10 [105]. This finding rules out the multicollinearity problem in our regressions. Finally, to control the effect of heteroscedasticity, we use robust standard errors when performing 2SLS.

4.2. Applicability of the KSTE in China

The second-stage estimation results for all manufacturing industries are shown in the first columns of

Table 1. Regression results by industry and region (the second stage of 2SLS).

Variables	Aggregate Industry	High–Tech Industry	Low–Tech Industry	Eastern China	Central and Western China
PATENT	1.184 ***	1.246 ***	1.175 ***	0.763 ***	0.836 ***
	(0.201)	(0.199)	(0.212)	(0.206)	(0.190)
SOEI	−0.471 ***	−0.764 ***	−0.423 **	−0.338	−0.229 *
	(0.146)	(0.170)	(0.166)	(0.386)	(0.138)
OPEN	−2.027 ***	−2.377 ***	−2.014 ***	−2.367 **	−0.221
	(0.765)	(0.891)	(0.768)	(0.940)	(0.792)
RS	0.666	2.078	0.473	2.875	−2.143 *
	(1.295)	(1.401)	(1.326)	(1.885)	(1.213)
RV	0.229	0.494 ***	0.219	−0.057	0.199
	(0.156)	(0.182)	(0.161)	(0.263)	(0.168)
UV	−0.025	−0.124	−0.072	0.381	−0.179
	(0.184)	(0.178)	(0.192)	(0.335)	(0.152)
lnPD	0.012	0.277	−0.117	0.221	−0.494 **
	(0.382)	(0.416)	(0.362)	(0.153)	(0.222)
lnHC	0.048	−0.043	0.088	0.039	−0.020
	(0.062)	(0.069)	(0.063)	(0.097)	(0.059)
lnSALARY	−0.024	−0.019	−0.019	−0.052 *	−0.015
	(0.029)	(0.054)	(0.037)	(0.028)	(0.030)
GDPGR	0.047	0.110	0.026	0.140	−0.104
	(0.214)	(0.216)	(0.227)	(0.230)	(0.212)
UNEMP	0.802	0.754	0.906	3.134 **	0.548
	(0.552)	(0.729)	(0.553)	(1.485)	(0.509)
GOV	0.479	0.619	0.459	−0.081	0.307
	(0.498)	(0.608)	(0.509)	(1.655)	(0.429)
SERVICE	−0.017 **	−0.019 **	−0.015 **	−0.018	−0.009
	(0.007)	(0.008)	(0.008)	(0.011)	(0.008)
COMP	0.106 ***	0.001	0.102 ***	0.084 **	0.055 *
	(0.029)	(0.015)	(0.033)	(0.038)	(0.032)
R–squared	0.6438	0.6317	0.6415	0.7709	0.7458
Observations	1820	1820	1820	602	1218
Number of cities	260	260	260	86	174
Underidentification test	0.0000	0.0000	0.0000	0.0000	0.0000
Kleibergen–Paap rk Wald F statistic	62.609	62.937	62.809	44.698	46.584
Endogeneity test	0.0000	0.0000	0.0000	0.0002	0.0025

Note: Robust standard errors are in parentheses. To save space, the estimation results of city fixed effect and year dummy are not reported. ***, **, and * represent significance levels at 1%, 5%, and 10%, respectively.

. The endogenous test result (p-values) of PATENT indicates a possible endogeneity problem in the model. Therefore, using the 2SLS model is appropriate. The under-identification test (p-value of Kleibergen–Paap rk LM) rejects the null hypothesis; thus, the equation can be identified. The value of the Kleibergen–Paap rk Wald F test is above 60, which is considerably larger than the Stock–Yogo weak ID test critical values of 10% maximal IV size (16.38), implying that no weak instrument problem exists. The estimation results show that the coefficient of PATENT is significantly positive, suggesting that knowledge creation can promote entrepreneurship. Thus, Hypothesis A is confirmed.

Previous studies in developed countries have shown that the relationship between knowledge creation and entrepreneurship differs in various industries, and knowledge creation has a significant effect on entrepreneurship in high-tech industries but not in low-tech industries [16,24]. To explore the differences in the relationship in various industries, we perform regression by industry after dividing the manufacturing sector into high-tech and low-tech industries. (Manufacturing industries with a higher R&D intensity are considered high-tech industries. The classification of high-tech industries is based on the industry classification issued by the Chinese National Bureau of Statistics (including 123 four-digit manufacturing industries. Available at http://www.stats.gov.cn/tjsj/tjbz/201812/t20181218_1640081.html (accessed on 15 March 2021) and http://www.stats.gov.cn/tjgz/tzgb/201904/t20190409_1658542.html (accessed on 15 March 2021).) As the largest developing country, China has huge differences in the level of economic development among various regions within the country, which provides a suitable setting for us to test the applicability of the KSTE in various regions. In comparison with the central and western regions, the eastern region has significant advantages in terms of economic development, innovation capabilities, and entrepreneurship vitality. (During the study period, the land area of eastern China only account for about 10% of the country, but its GDP accounts for more than 50% of the country, the number of patent applications accounts for more than 60% of the country, and the number of new manufacturing firms accounts about 70% of the country.) Given the importance of the knowledge environment in the KSTE [22], we investigate whether the applicability of the KSTE in China is affected by regional differences. To this end, we divided the cities into two regions: Eastern China and Central and Western China and regressed by region. (The eastern region includes Beijing, Tianjin, Hebei, Shanghai, Jiangsu, Zhejiang, Fujian, Shandong, Guangdong, and Hainan. Others are classified as the central and western regions.) The second-stage estimation results of the regressions by industry and region are listed in the last four columns of Table 1. The results of the IV test show that the hypothesis that PATENTIV is not identifiable is rejected, and PATENTIV is not a weak IV. Moreover, the coefficients of PATENT are all significantly positive, which indicates that the applicability of the KSTE in China is extensive, and it is not affected by the technology of industry and development of the region.

The coefficients of SOEI and OPEN are all negative and pass the significance test except for a few cases, which implies that increasing the level of market reforms is conducive to entrepreneurship, whereas increasing opening-up is not conducive to entrepreneurship. Previous studies have shown that the entry of FDI has a twofold effect on the local enterprises in the host country [74,106]. On the one hand, the entry of FDI has improved the production efficiency of local firms [75,77,78]. On the other hand, FDI may negatively affect the entrance of domestic firms by raising technological barriers to entry [107]. Mencinger [108] reported that FDI may force small emerging local competitors out of business. De Backer et al. [109] also reported similar results on the entry and exit of Belgian manufacturing firms. Recently, Eren et al. [106] found that an increase in FDI decreases the average monthly rate of business creation and destruction.

Now we turn to control variables. Among the control variables, only COMP and SERVICE have relatively good results of significance test. As is shown in Table 1, the coefficients of COMP are always positive, suggesting that increasing localized competition is conducive to entrepreneurship. Plummer et al. [18] also found that localized competition has a significant positive effect on entrepreneurship. The coefficients of SERVICE are all negative in the regressions, indicating that in cities where the service industry economy is relatively developed, there exist relatively few entrepreneurial activities in the manufacturing sector.

4.3. Role of Regional Institutional Environment

To test whether the institutional environment moderates the relationship between new knowledge and entrepreneurship, two approaches can be adopted. One is regressing by group after splitting the research sample according to institutional environment variables, and the other is performing regression with interaction terms between institutional environment variables and the variable representing knowledge creation. In comparison, the assumptions of the second approach are relatively strict because it requires that the coefficients of all control variables are not different among groups with different institutional environments, whereas the first approach does not have this requirement [110]. In view of the different regional institutional environments, ensuring that the effects of the control variables do not have differences between groups is difficult. Therefore, following Cleary [111], we choose the first approach to empirically explore the moderating effect of the institutional environment.

Specifically, we initially calculate the annual average value of SOEI (A_SOEI) and the annual average value of OPEN (A_OPEN) for each city and then divide all cities into two groups based on A_SOEI and A_OPEN. If the A_SOEI of a city is less than or equal to the median, then it is classified into group SOEI_1; otherwise, it is classified into group SOEI_0. On this basis, cities in group SOEI_1 have a relatively higher level of market reforms. Similarly, if the A_OPEN of a city is greater than the median, then it is classified into group OPEN_1; otherwise, it is classified into group OPEN_0. Obviously, cities in group OPEN_1 have a relatively higher level of opening-up.

After grouping, we calculate the empirical p-value using a bootstrapping procedure. We take SOEI as an example, and the steps are as follows. First, we regress for Equation (7) and obtain the estimation results of groups SOEI_1 and SOEI_0. Second, we record the difference (D0) in the PATENT coefficient between the two groups. Third, we mix the cities in groups SOEI_1 and SOEI_0. Assuming that the number of cities from the two groups is N1 and N2, we randomly select N1 and N2 cities and assign them to groups SOEI_1 and SOEI_0, respectively. Then, we record the difference (Di) in the PATENT coefficient between the two groups after re-regressing by group. This procedure is re-executed 1000 times. Then, the number of times that Di is greater than D0 can be counted, and the empirical p-value (m/1000) can finally be calculated. The empirical p-value has the same meaning as the p-value in the traditional tests. For example, an empirical p-value of 0.01 indicates that 10 out of 1000 Di exceed D0, which implies that the coefficient difference of PATENT between the two groups is significant at the 1% significance level.

Table 2. The moderating effect of market reforms (estimation results of the second stage of 2SLS).

	Aggregate Industry		High–Tech Industry		Low–Tech Industry
Variables	SOEI_0	SOEI_1	SOEI_0	SOEI_1	SOEI_0	SOEI_1
PATENT	0.513 ***	1.522 ***	0.358 **	1.684 ***	0.563 ***	1.465 ***
	(0.178)	(0.399)	(0.177)	(0.414)	(0.202)	(0.406)
SOEI	−0.275 **	−0.318	−0.462 ***	−0.703 *	−0.239	−0.255
	(0.132)	(0.392)	(0.176)	(0.402)	(0.164)	(0.400)
OPEN	−0.199	−3.336 **	−1.643	−2.758 *	−0.052	−3.411 ***
	(0.765)	(1.323)	(1.308)	(1.429)	(0.803)	(1.276)
RS	−3.112 **	2.317	−2.254	3.617	−3.076 **	1.919
	(1.253)	(2.155)	(1.520)	(2.245)	(1.371)	(2.150)
RV	0.120	0.296	0.187	0.696 **	0.097	0.286
	(0.160)	(0.310)	(0.224)	(0.319)	(0.163)	(0.308)
UV	−0.082	0.264	−0.144	0.174	−0.151	0.233
	(0.136)	(0.305)	(0.168)	(0.304)	(0.147)	(0.310)
lnPD	−0.484 **	0.371	−0.209	0.605 **	−0.607 **	0.228
	(0.205)	(0.285)	(0.223)	(0.306)	(0.247)	(0.270)
lnHC	−0.034	0.039	−0.088	−0.127	0.012	0.087
	(0.050)	(0.135)	(0.090)	(0.153)	(0.049)	(0.136)
lnSALARY	−0.178	−0.038	−0.374	−0.026	−0.193	−0.032
	(0.252)	(0.031)	(0.249)	(0.041)	(0.271)	(0.051)
GDPGR	−0.138	0.708 *	−0.058	0.808 *	−0.168	0.669 *
	(0.207)	(0.396)	(0.229)	(0.423)	(0.218)	(0.398)
UNEMP	0.394	0.789	0.311	0.915	0.494	0.875
	(0.695)	(0.777)	(0.814)	(0.872)	(0.731)	(0.795)
GOV	0.589	0.568	1.425	−0.048	0.450	0.688
	(0.592)	(0.881)	(0.950)	(0.888)	(0.577)	(0.883)
SERVICE	−0.010	−0.015	−0.004	−0.023	−0.010	−0.012
	(0.007)	(0.015)	(0.009)	(0.016)	(0.008)	(0.015)
COMP	0.027	0.102 ***	−0.032 **	0.016	0.030	0.109 ***
	(0.046)	(0.031)	(0.015)	(0.022)	(0.042)	(0.032)
R–squared	0.7768	0.4287	0.7262	0.5010	0.7724	0.4349
Observations	910	910	910	910	910	910
Number of cities	130	130	130	130	130	130
Underidentification test	0.0000	0.0001	0.0000	0.0001	0.0000	0.0001
Kleibergen–Paap rk Wald F statistic	41.272	21.865	41.106	22.214	41.479	21.679
Endogeneity test	0.2980	0.0000	0.3864	0.0000	0.2784	0.0000
Empirical p-value	0.015		0.009		0.031

Note: Robust standard errors are in parentheses. To save space, the estimation results of city fixed effect and year dummy are not reported. ***, **, and * represent significance levels at 1%, 5%, and 10%, respectively.

lists the second-stage estimation results of the effect of market reforms is conducive to entrepreneurship on the relationship between knowledge creation and entrepreneurship. Although in group SOEI_0, the endogeneity tests of PATENT in regressions for all manufacturing industries, high-tech industries, and low-tech industries cannot reject the null hypothesis that PATENT is exogenous, the 2SLS estimation results of PATENT are still consistent estimators because no weak instrument problems exist. In group SOEI_1, the endogeneity test results of PATENT indicate that a possible endogeneity problem exists in regressions for all manufacturing industries, high-tech industries, and low-tech industries. Thus, we use 2SLS for both groups. As shown in Table 2, the coefficient of PATENT is always positively significant in any regression, and the coefficients of PATENT in group SOEI_1 are always larger than those in group SOEI_0. The empirical p-values are all less than 0.05, suggesting that the coefficients of PATENT in group SOEI_1 are significantly larger than those in group SOEI_0. In other words, in regions with a higher level of market reforms, knowledge creation has a greater positive effect on entrepreneurship. Thus, Hypothesis B is supported.

Table 3. The moderating effect of opening-up (estimation results of the second stage of 2SLS).

	Aggregate Industry		High–Tech Industry		Low–Tech Industry
Variables	OPEN_0	OPEN_1	OPEN_0	OPEN_1	OPEN_0	OPEN_1
PATENT	0.724 ***	1.352 ***	0.599 ***	1.557 ***	0.719 ***	1.358 ***
	(0.187)	(0.348)	(0.176)	(0.388)	(0.203)	(0.359)
SOEI	−0.217	−0.519 *	−0.522 ***	−0.687 *	−0.194	−0.418
	(0.153)	(0.299)	(0.180)	(0.355)	(0.187)	(0.309)
OPEN	−1.045	−1.576	−2.490 *	−1.802	−1.006	−1.555
	(1.280)	(1.001)	(1.486)	(1.154)	(1.307)	(1.012)
RS	−1.484	1.005	0.846	1.099	−1.975	1.054
	(1.442)	(1.973)	(1.443)	(2.166)	(1.542)	(2.022)
RV	0.031	0.280	0.285	0.531	−0.008	0.279
	(0.175)	(0.297)	(0.174)	(0.338)	(0.187)	(0.300)
UV	−0.094	0.233	−0.076	−0.038	−0.151	0.195
	(0.150)	(0.331)	(0.149)	(0.355)	(0.157)	(0.343)
lnPD	−0.548 **	0.153	−0.137	0.387	−0.706 **	0.023
	(0.247)	(0.308)	(0.242)	(0.438)	(0.284)	(0.262)
lnHC	−0.030	0.017	−0.166 **	−0.046	0.032	0.028
	(0.055)	(0.142)	(0.071)	(0.159)	(0.057)	(0.145)
lnSALARY	−0.147	−0.047 **	−0.259	−0.027	−0.162	−0.045
	(0.266)	(0.022)	(0.260)	(0.047)	(0.289)	(0.043)
GDPGR	−0.137	0.451	0.012	0.532 *	−0.175	0.448
	(0.231)	(0.300)	(0.255)	(0.318)	(0.246)	(0.312)
UNEMP	0.535	1.403 *	0.032	2.432 **	0.697	1.364 *
	(0.645)	(0.812)	(0.758)	(1.209)	(0.655)	(0.800)
GOV	0.413	−1.885	0.397	−0.386	0.461	−2.387
	(0.445)	(1.717)	(0.533)	(2.156)	(0.459)	(1.754)
SERVICE	−0.013	−0.008	−0.009	−0.019	−0.012	−0.005
	(0.009)	(0.013)	(0.009)	(0.015)	(0.009)	(0.013)
COMP	0.035	0.161 ***	−0.040 **	0.032 *	0.042	0.165 ***
	(0.035)	(0.036)	(0.018)	(0.018)	(0.036)	(0.037)
R–squared	0.7733	0.5402	0.7188	0.5705	0.7674	0.5306
Observations	910	910	910	910	910	910
Number of cities	130	130	130	130	130	130
Underidentification test	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000
Kleibergen–Paap rk Wald F statistic	38.976	38.515	39.317	38.092	39.051	38.418
Endogeneity test	0.0307	0.0000	0.1066	0.0000	0.0353	0.0000
Empirical p-value	0.088		0.022		0.087

Note: Robust standard errors are in parentheses. To save space, the estimation results of city fixed effect and year dummy are not reported. ***, **, and * represent significance levels at 1%, 5%, and 10%, respectively.

shows the second-stage estimation results of the effect of opening-up on the relationship between new knowledge and entrepreneurship. The endogeneity test of PATENT in any regression in Table 3 indicates that it is an endogeneity variable. The values of Kleibergen–Paap rk Wald F tests are always larger than the Stock–Yogo weak ID test critical values of 10% maximal IV size (16.38), indicating that no weak instrument problem exists. The coefficients of PATENT in group OPEN_1 are always larger than those in group OPEN _0. The empirical p-values are all less than 0.1, suggesting that the coefficients of PATENT in group OPEN_1 are significantly larger than those in group OPEN_0. In other words, in regions with a higher level of opening-up, knowledge creation has a greater positive effect on entrepreneurship. Thus, Hypothesis C is confirmed.

4.4. Robustness Tests

Prior to the implementation of the economic reforms, China’s economy was characterized by an introverted development strategy and rejected the entry of foreign capital [36]. After the reform and opening-up, China established Special Economic Zones (SEZs). The effectiveness of attracting foreign capital was specified as an integral component in evaluating the SEZs’ overall performance and openness [112]. Therefore, the establishment of foreign-invested enterprises may be related to the opening-up policy but has a weak connection with knowledge creation. Therefore, we recalculate the dependent variable after removing foreign-invested enterprises to test the robustness of the conclusions obtained earlier. The second stage estimation results for the effects of market reforms and opening-up with removing foreign-invested enterprises are shown in

Table 4. The moderating effect of market reforms (excluding foreign-invested enterprises, the second-stage estimation results of 2SLS).

	Aggregate Industry		High–Tech Industry		Low–Tech Industry
Variables	SOEI_0	SOEI_1	SOEI_0	SOEI_1	SOEI_1	SOEI_0
PATENT	0.513 ***	1.478 ***	0.347 **	1.671 ***	0.570 ***	1.428 ***
	(0.179)	(0.381)	(0.177)	(0.408)	(0.203)	(0.390)
SOEI	−0.273 **	−0.277	−0.463 **	−0.649 *	−0.238	−0.216
	(0.134)	(0.383)	(0.180)	(0.394)	(0.167)	(0.391)
OPEN	−0.032	−3.083 **	−1.620	−2.490 *	0.111	−3.167 **
	(0.757)	(1.292)	(1.289)	(1.411)	(0.798)	(1.246)
Control variables	Included	Included	Included	Included	Included	Included
R–squared	0.7754	0.4624	0.7281	0.5262	0.7695	0.4613
Observations	910	910	910	910	910	910
Number of cities	130	130	130	130	130	130
Underidentification test	0.0000	0.0001	0.0000	0.0001	0.0000	0.0001
Kleibergen–Paap rk Wald F statistic	41.272	21.865	41.106	22.214	41.479	21.679
Endogeneity test	0.3254	0.0000	0.4271	0.0000	0.2910	0.0000
Empirical p-value	0.013		0.011		0.037

Note: Robust standard errors are in parentheses. To save space, the estimation results of city fixed effect and year dummy are not reported. ***, **, and * represent significance levels at 1%, 5%, and 10%, respectively.

and

Table 5. The moderating effect of opening-up (excluding foreign-invested enterprises, the second-stage estimation results of 2SLS).

	Aggregate Industry		High–Tech Industry		Low–Tech Industry
Variables	OPEN_0	OPEN_1	OPEN_0	OPEN_1	OPEN_0	OPEN_1
PATENT	0.711 ***	1.329 ***	0.591 ***	1.514 ***	0.712 ***	1.348 ***
	(0.188)	(0.339)	(0.176)	(0.374)	(0.203)	(0.353)
SOEI	−0.215	−0.490 *	−0.529 ***	−0.629 *	−0.193	−0.392
	(0.155)	(0.296)	(0.182)	(0.352)	(0.190)	(0.307)
OPEN	−0.829	−1.387	−2.481 *	−1.631	−0.758	−1.375
	(1.278)	(0.985)	(1.506)	(1.120)	(1.306)	(0.998)
Control variables	Included	Included	Included	Included	Included	Included
R–squared	0.7727	0.5522	0.7214	0.5948	0.7656	0.5352
Observations	910	910	910	910	910	910
Number of cities	130	130	130	130	130	130
Underidentification test	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000
Kleibergen–Paap rk Wald F statistic	38.976	38.515	39.317	38.092	39.051	38.418
Endogeneity test	0.0433	0.0000	0.1155	0.0000	0.0452	0.0000
Empirical p-value	0.081		0.024		0.089

Note: Robust standard errors are in parentheses. To save space, the estimation results of city fixed effect and year dummy are not reported. ***, and * represent significance levels at 1%, and 10%, respectively.

.

Furthermore, compared with non-SOEs, SOEs often have the political purposes of stabilizing the economy and increasing employment, which may lead to SOEs not being created from incentives for knowledge creation [65]. Therefore, on the basis of removing foreign-invested enterprises, we recalculate the dependent variable after removing state-owned enterprises. The second stage estimation results for the effect of market reforms and opening-up with removing foreign-invested enterprises and SOEs are shown in

Table 6. The moderating effect of market reforms (excluding foreign-invested enterprises and state-owned enterprises, the second-stage estimation results of 2SLS).

	Aggregate Industry		High–Tech Industry		Low–Tech Industry
Variables	SOEI_0	SOEI_1	SOEI_0	SOEI_1	SOEI_1	SOEI_0
PATENT	0.507 ***	1.498 ***	0.339 *	1.710 ***	0.572 ***	1.438 ***
	(0.185)	(0.388)	(0.181)	(0.419)	(0.212)	(0.395)
SOEI	−0.236 *	−0.270	−0.476 ***	−0.595	−0.193	−0.224
	(0.140)	(0.395)	(0.179)	(0.404)	(0.167)	(0.402)
OPEN	0.070	−3.160 **	−1.513	−2.570 *	0.189	−3.233 **
	(0.782)	(1.306)	(1.303)	(1.432)	(0.821)	(1.257)
Control variables	Included	Included	Included	Included	Included	Included
R–squared	0.7777	0.4490	0.7341	0.5082	0.7727	0.4539
Observations	910	910	910	910	910	910
Number of cities	130	130	130	130	130	130
Underidentification test	0.0000	0.0001	0.0000	0.0000	0.0000	0.0001
Kleibergen–Paap rk Wald F statistic	41.272	21.865	41.106	22.214	41.479	21.679
Endogeneity test	0.4408	0.0000	0.4778	0.0000	0.3617	0.0000
Empirical p-value	0.013		0.009		0.035

Note: Robust standard errors are in parentheses. To save space, the estimation results of city fixed effect and year dummy are not reported. ***, **, and * represent significance levels at 1%, 5%, and 10%, respectively.

and

Table 7. The moderating effect of opening-up (excluding foreign-invested enterprises and state-owned enterprises, the second-stage estimation results of 2SLS).

	Aggregate Industry		High–Tech Industry		Low–Tech Industry
Variables	OPEN_0	OPEN_1	OPEN_0	OPEN_1	OPEN_0	OPEN_1
PATENT	0.722 ***	1.350 ***	0.615 ***	1.549 ***	0.721 ***	1.362 ***
	(0.196)	(0.341)	(0.183)	(0.374)	(0.215)	(0.354)
SOEI	−0.182	−0.463	−0.530 ***	−0.608 *	−0.155	−0.363
	(0.159)	(0.304)	(0.181)	(0.359)	(0.187)	(0.315)
OPEN	−0.515	−1.415	−1.933	−1.721	−0.587	−1.379
	(1.342)	(0.997)	(1.524)	(1.135)	(1.357)	(1.009)
Control variables	Included	Included	Included	Included	Included	Included
R–squared	0.7749	0.5432	0.7257	0.5877	0.7680	0.5290
Observations	910	910	910	910	910	910
Number of cities	130	130	130	130	130	130
Underidentification test	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000
Kleibergen–Paap rk Wald F statistic	38.976	38.515	39.317	38.092	39.051	38.418
Endogeneity test	0.0557	0.0000	0.0860	0.0000	0.0601	0.0000
Empirical p-value	0084		0.025		0.088

Note: Robust standard errors are in parentheses. To save space, the estimation results of city fixed effect and year dummy are not reported. ***, and * represent significance levels at 1% and 10%, respectively.

. In addition, excluding firms with fewer than 8 employees may lead to an underestimation of entrepreneurship. We reincorporate new firms with employment less than 8 into the calculation of the dependent variable, and the obtained estimation results are presented in

Table 8. The moderating effect of market reforms (retaining firms with less than 8 employees, the second-stage estimation results of 2SLS).

	Aggregate Industry		High–Tech Industry		Low–Tech Industry
Variables	SOEI_0	SOEI_1	SOEI_0	SOEI_1	SOEI_1	SOEI_0
PATENT	0.518 ***	1.487 ***	0.346 **	1.648 ***	0.565 ***	1.450 ***
	(0.176)	(0.394)	(0.175)	(0.409)	(0.199)	(0.402)
SOEI	−0.287 **	−0.359	−0.464 ***	−0.684 *	−0.257	−0.306
	(0.130)	(0.389)	(0.176)	(0.401)	(0.167)	(0.398)
OPEN	−0.362	−3.319 **	−1.786	−2.730 *	−0.268	−3.417 ***
	(0.757)	(1.300)	(1.330)	(1.411)	(0.793)	(1.262)
Control variables	Included	Included	Included	Included	Included	Included
R–squared	0.7829	0.4252	0.7283	0.4992	0.7759	0.4258
Observations	910	910	910	910	910	910
Number of cities	130	130	130	130	130	130
Underidentification test	0.0000	0.0001	0.0000	0.0000	0.0000	0.0001
Kleibergen–Paap rk Wald F statistic	41.272	21.865	41.106	22.214	41.479	21.679
Endogeneity test	0.1860	0.0000	0.3419	0.0000	0.1842	0.0000
Empirical p-value	0.016		0.011		0.037

Note: Robust standard errors are in parentheses. To save space, the estimation results of city fixed effect and year dummy are not reported. ***, **, and * represent significance levels at 1%, 5%, and 10%, respectively.

and

Table 9. The moderating effect of opening-up (retaining firms with less than 8 employees, the second-stage estimation results of 2SLS).

	Aggregate Industry		High–Tech Industry		Low–Tech Industry
Variables	OPEN_0	OPEN_1	OPEN_0	OPEN_1	OPEN_0	OPEN_1
PATENT	0.692 ***	1.380 ***	0.569 ***	1.587 ***	0.692 ***	1.380 ***
	(0.188)	(0.348)	(0.183)	(0.385)	(0.204)	(0.360)
SOEI	−0.247	−0.502 *	−0.547 ***	−0.638 *	−0.231	−0.410
	(0.152)	(0.300)	(0.180)	(0.357)	(0.189)	(0.310)
OPEN	−1.182	−1.601	−2.507 *	−1.898	−1.246	−1.576
	(1.246)	(1.008)	(1.486)	(1.168)	(1.280)	(1.015)
Control variables	Included	Included	Included	Included	Included	Included
R–squared	0.7827	0.5162	0.7203	0.5541	0.7743	0.5111
Observations	910	910	910	910	910	910
Number of cities	130	130	130	130	130	130
Underidentification test	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000
Kleibergen–Paap rk Wald F statistic	38.976	38.515	39.317	38.092	39.051	38.418
Endogeneity test	0.0231	0.0000	0.1056	0.0000	0.0267	0.0000
Empirical p-value	0.069		0.014		0.074

Note: Robust standard errors are in parentheses. To save space, the estimation results of city fixed effect and year dummy are not reported. ***, and * represent significance levels at 1% and 10%, respectively.

.

As shown in Table 4, Table 6 and Table 8, the coefficients of PATENT in group SOEI_1 are always larger than those in group SOEI_0. The empirical p-values are all less than 0.05, suggesting that the coefficients of PATENT in group SOEI_1 are significantly larger than those in group SOEI_0. Therefore, Hypothesis B is confirmed again. As shown in Table 5, Table 7 and Table 9, the coefficients of PATENT in group OPEN_1 are always larger than those in group OPEN_0. The empirical p-values are all less than 0.1. Therefore, Hypothesis C is confirmed again.

5. Conclusions

Innovation and entrepreneurship play an important role in sustainable development. The KSTE argues that un-commercialized new knowledge as the source of entrepreneurial opportunities can be transferred to potential entrepreneurs through knowledge spillovers and commercialized by them. Although a considerable amount of literature supports the theory [17,18,20,21], previous studies on the KSTE have not considered the effect of the institutional environment on the relationship between new knowledge and entrepreneurship. This study uses data from Chinese city-level patents and new manufacturing firms to find evidence confirming the applicability of the KSTE in China. More importantly, we explore the role of the regional institutional environment under the KSTE framework, thereby extending the theory.

The empirical results show that new knowledge can increase entrepreneurship in China for both high-tech industries and low-tech industries. This finding is different from the conclusion that the positive effect of new knowledge on entrepreneurship only exists in high-tech industries [16,24]. In addition, the empirical results show that whether it is in the central and western regions with relatively underdeveloped economies or the eastern region with a relatively developed economy, the positive effect of new knowledge on entrepreneurship exists. This finding further supports the applicability of the KSTE in China. We also confirm that the institutional environment affects the relationship between knowledge creation and entrepreneurship; that is, market reforms and opening-up can enhance the positive effect of knowledge creation on entrepreneurship.

This study provides several implications for policymaking. A suitable institutional environment will increase the profit expectations of innovation and entrepreneurship and facilitate the exchange and dissemination of knowledge, thereby increasing the possibility of knowledge spillover entrepreneurship. From the perspective of policymaking, when policymakers consider various supporting policies to promote knowledge spillover entrepreneurship, they need to consider how to build a local institutional environment that can support the successful implementation of these policies.

Specifically, we have several policy recommendations. First, market-oriented reforms should be continuously promoted to reduce the uncertainty of economic activities and increase the willingness of individuals to innovate and start businesses. Excessive government control in the local market leads to high levels of corruption, which raise levels of uncertainty and transaction costs and can thus make an otherwise promising innovative opportunity difficult to commercialize profitably [52]. In addition, political ties between enterprises and the government affect the allocation of key regulatory resources [66], thereby distorting the market’s competitive environment and reducing the willingness to innovate and start businesses. Therefore, the government should clarify the boundary between the government and the market and avoid improper intervention in the market, which is conducive to reducing the administrative and bureaucratic barriers to entrepreneurship.

Second, the government should also improve the rules and regulations for the production and trading of intellectual property to provide legal protection for the production and exchange of knowledge. Furthermore, the set of policies aimed at promoting the exchange and dissemination of knowledge should include actions such as the establishment of science parks, incubators, or technology transfer programs.

Third, policymakers should take measures to reduce the negative effect of opening-up on local entrepreneurship. Although opening-up can enhance the positive effect of knowledge creation on entrepreneurship, it also has a crowding-out effect on domestic entrepreneurship. The government should provide corresponding funds or channel support for entrepreneurs in scientific research and study to improve the entrepreneurial absorptive capacity such that entrepreneurs can absorb the new technology and management knowledge brought by foreign-invested enterprises. In addition, policymakers should make every reasonable effort to promote collaboration and knowledge sharing between foreign-invested enterprises and domestic enterprises, thereby reducing the crowding-out effect of foreign-invested enterprises on domestic enterprises.

This study is not free from limitations. First, given that this study only conducts an empirical exploration in the context of China, whether the KSTE is applicable to other emerging economies still needs more evidence. Second, we only explore the role of market reforms and opening-up under the KSTE framework. For a transitional economy such as China, although market reforms and opening-up are the two most representative reforms in the process of institutional transformation, they are not the full picture of institutional transformation. Third, given that this paper focuses on the effect of the regional institutional environment on the relationship between knowledge creation and entrepreneurship, we do not further explore the role of relevant specific supporting policies under the KSTE framework. Therefore, future research should further explore the role of other institutional factors and policies under the KSTE framework.