Introduction The reform and opening up in the late

The reform and opening up in the late 1970s, known as the Chinese economic miracle,

triggered noteworthy Chinese economic growth in the following decades (Ray, 2002),

with an average gross domestic product (GDP) growth rate of 9.95% between 1979 and

2011 and reaching the highest GDP growth rate of 15.2% in 1984, according to National

Bureau of Statistics China. The process of reform and opening up is also the process of a

change in cooperate governance in China, particularly more state-owned enterprises

(SOEs) towards privatisation (Xu and Uddin, 2008). Nowadays, the Chinese economy is

experiencing its fifth great institutional change of SOE reform - mixed ownership reform

(Lin et al., 2020). Specifically, mixed-ownership reform aims to promote market reform

of SOEs and mitigate business dependence on government (Wang et al., 2021), and has

undergone five stages – decentralization (1978-1984), transformation of control rights

(1985-1992), separation of ownership and control of the company (1992-2002),

reorganisation of the ownership of SOEs (2002-2012), mixed ownership reform (2012 –

present) (Lin et al., 2020; Li et al., 2013). Since 2012, China's rapid economic growth has

reached a turning point due to the exhaustion of its late-stage advantage, and economic

growth has entered a slow-growing period (e.g. Lo, 2018; Wang, 2015). During this

period, the Chinese government has made several efforts to accelerate economic growth.

One of the most cited is innovation, which researchers believe is necessary at this stage of

the economic slowdown (e.g. Chen et al., 2018; Lo et al., 2022). The concept of

innovation is broad, and this thesis concentrates on innovation at the firm level.

Researchers usually categorised innovation measurement into innovation input and

innovation output. Innovation input is measured by R&D activities, for example R&D

expenditure and R&D intensity (e.g., Cavdar and Aydin, 2015; Kamien and Schwartz,

1982; Kurt et al., 2015). Innovation output is measured by patent data, for example the

number of patents (e.g., Di Vito et al., 2010; Decker and Günther, 2017; Griliches, 1990).

Firm innovation performance plays an important role in firm profitability, because

innovators are able to somehow shield themselves from market forces (Love et al., 2009).

In practice, many factors can drive firm innovation performance, for example information

and communication technology (He et al., 2023), corporate governance (e.g., Chi, 2023;

Fan et al., 2023; Shaikh and Randhawa, 2022), gender diversity

(Tonoyan and Boudreaux, 2023), investor sentiment (Lin, 2023), policies (Chen and Jin,

2023). This thesis is interested in ownership structure, which is one part of corporate

governance. Different corporate governance in different countries generates various

national corporate innovation and entrepreneurship systems, which, in turn, induce a

difference in firm competitiveness over the world. Hoskisson et al. (2004) conclude two

types of corporate governance which facilitate different types of innovation: (1) a market-

based governance system promotes explorative and revolutionary innovations by taking

advantage of its dynamism, flexibility and diversity; (2) a relationship-based governance

system supports exploitative and incremental innovations due to its continuity, stability

and commitment. Besides, internal control systems and managerial incentives determine

the allocation of R&D investments so as to influence firm innovation performance

significantly (Aaboen et al., 2006, Birkinshaw et al., 2008). As to narrowing down the

research area from corporate governance to the ownership structure, previous research in

emerging countries assesses how these firm ownership impact innovation in emerging

countries (e.g. Choi et al., 2011; Chen et al., 2016; Rong et al., 2017; Liu et al., 2017;

Zhou et al., 2017). Most studies, however, concentrate on the relationship between one

specific ownership structure and firm innovation performance, such as the research on

concentrated ownership (Nguyen et al., 2015; Clò et al., 2020), institutional ownership

(Rong et al., 2017; Li and Ji, 2021), state ownership (Zhou et al., 2017; Dong et al., 2022;

Lo et al., 2022), insider ownership (Chen et al., 2013; Liu et al., 2017; Cheng et al., 2021),

and foreign ownership (Chen et al., 2016; Dong et al., 2022). Those five ownerships are

what this essay focuses on. Only a few scholars are interested in comparatively analysing

the impact of two or more ownership on firms' innovation activities in emerging

economies (Choi et al., 2011; Jiang et al., 2013; Li et al., 2014). One empirical reason can

be that the more ownership structures considered, the more time consuming to collect

data. However, the advantage to have five ownership structures discussed in this thesis is

obvious, as those five ownership structures cover all ownership in Chinese firms. Thus,

we can have a comprehensive discussion about how five ownership structures affect firm

innovation using the most recent data, as shifts in the political environment, foreign trade

environment or business environment can change the relationship between firm

ownership and innovation.

Using data from the Shanghai Stock Exchange between 2013 and 2019, this thesis focuses

on the impact of five ownership structures on firm innovation. In this context, innovation

refers to firm performance innovation, specialised and diversified innovation. The

findings reveal that firm concentration of ownership is positively related to firm

innovation performance, while foreign ownership and firm innovation performance are

negatively related. Otherwise, there was no significant linear relationship between firm

ownership and innovation performance. Thus, a new sample period was applied in this

case.

However, the innovation performance of concentrated ownership does not differ

significantly between the different levels when a threshold is used. In contrast, the study

of insider ownership indicates a positive effect on innovation performance when insider

ownership is below 5% or above 20%. In particular, insider ownership above 20% has a

greater impact on innovation performance than insider ownership below 5%. This result is

partially consistent with the work of Song et al. (2015). Furthermore, state ownership is

positively associated with innovation performance only when it is less than 5%, which is

partially consistent with previous findings (Choi et al., 2011). In addition, the results find

that only firms with more than 20% foreign ownership damage innovation performance.

There is no significant difference between different levels of institutional ownership in

terms of firm innovation performance. Hence, it fills the research gap about thresholds of

ownership structure and offers a guidance for current mixownership reform in China.

Third, the findings suggest that firm ownership is not related to innovation specialisation

or diversification but more external factors impacting the firm's decision to specialise or

diversify. Therefore, it fills the research gap about the determinants of firm innovation

diversification and specialisation.

1.2 Research Aims

The Chinese government has launched a number of initiatives to foster innovation and

find new sources of economic growth amidst a deceleration in economic growth. Since

2012, the Chinese economy has entered sluggish growth from the rapid economic growth

associated with reform and opening up. In response to the notable change in the economy,

the Chinese government has adopted a series of measures to stimulate economic growth,

including mixed-ownership reforms and exploring opportunities for further cooperation

and development with other countries. This thesis aims to examine the relationship

between firm ownership and innovation, investigate the relationship between five

ownership structures and innovation performance in China, find the thresholds that

influence company ownership and innovation, and how different levels of company

ownership affect innovation specialisation and diversification.

1.3 Research Objectives

The first purpose of this study is to investigate the relationship between five ownership

structures and innovation performance in China. As the literature on innovation

performance emphasises country-specific institutional factors while interpreting each

country's innovation performance (e.g. Abdullah et al., 2002; Di et al., 2010), this thesis

focuses on the emerging institutional setting with a large and rapidly developing market.

Traditional Schumpeter-inspired economics of innovation, and recent advances in his

work, seem incapable of explaining why firms sharing similar exogenous circumstances

may behave very differently in terms of innovation. By contrast, the literature on

corporate governance offers several helpful perspectives for understanding firms'

innovative activities. Belloc (2012) suggests three main aspects of cooperate governance

concerning a firm's innovation performance – ownership structures, labour and cooperate

finance. In this thesis, ownership structures are the focus.

The second objective is to find the thresholds that influence company ownership and

innovation. Firm ownership will be divided into different levels, and this thesis examines

the relationship between different levels of firm ownership and innovation. Wright et al.

(2005) suggest that the fundamental of effective corporate governance in research can be

the integration of agency theory and institutional theory, as this contribution may offer a

novel framework. Thus, a firm, which can implement efficient management in changing

environment, relies on a suitable structure between ownership, control device and

monitoring mechanism. Therefore, in this thesis, innovation will be analysed using

different levels of firm ownership to identify the thresholds that affect the relationship

between firm ownership and innovation. It will allow shareholders and firm managers to

better understand the relationship between firm ownership and innovation, and to use the

thresholds flexibly to build mixed ownership structures that allow firms to benefit from

different firm ownership.

The third objective is to see how different levels of company ownership affect innovation

specialisation and diversification. This thesis utilises the concept of knowledge to analyse

innovation specialisation and innovation diversification.

Knowledge plays an important role in the survival, growth and innovation of a company

(Healey and Mintz, 2021; Yu and 2021). The varying firm performance depends on the

firm's own knowledge base (Grant, 1996). The accumulation of a knowledge base adds to

a firm's competitiveness, facilitates research and development (R&D), and ultimately

transforms knowledge into commercial products (Cohen and Levinthal, 1990; Arora and

Gambardella, 1994). The degree of knowledge accumulation can be divided into two

categories, depth of knowledge and breadth of knowledge. The depth of knowledge helps

to strengthen the firm's capacity to absorb new knowledge, integrate external knowledge

into the firm's knowledge base, use external knowledge proficiently and effectively, and

increase the firm's competitiveness (Yang et al., 2017). The breadth of knowledge

facilitates the expansion of the scope of knowledge to bring new inspiration for

innovation (Katila and Ahuja, 2002), and the firm's capacity in adapting to technological

changes in related fields (Volberda, 1996; Srivastava and Gnyawali,

2011).

1.4 Research Questions

In order to achieve the research objectives, the following research questions need to be

addressed:

1. What is the impact of firm ownership on innovation performance?

2. What is the impact of different levels of firm ownership on innovation

performance?

3. How do different levels of firm ownership affect innovation specialisation?

4. How do different levels of firm ownership affect innovation diversification?

1.5 Background

1.5.1 Chinese Economy

Since the reform and opening up, the Chinese economy has entered an era of rapid

growth. From 2003 to 2013, Figure 1.1 indicates that the GDP growth rate in China

remained above 10% except for 2009, followed by a decline in growth rate to below 10%.

Then, the growth rate hit rock bottom (around 7%) in 2015 and climbed steadily for three

years before dropping again due to China-US trade war in 2019 and COVID19 in 2020,

tumbling down from 10% to around 3%. Obviously, the Chinese economy has moved into

a period of slow growth since 2013. As a result, the average GDP growth rate between

2012 and 2021 was around 8%, revealing a downshift in China's economic growth

compared to the rapid economic surge of the past after the Reform and Opening Up

policy.

Figure 1.1 GDP Growth Rate and GDP per capita Growth rate in China over period 2013-2021

Source: National Bureau of Statistics

It is officially called the "new normal", and the reasons for this "new normal" are

manifold (Lo, 2018). For example, the slowdown mentioned in his paper is mainly due to

China's "late-stage advantage" depletion. Years ago, after the reform and opening up,

Chinese companies improved or innovated their products or services by absorbing and

refining imported technology, increasing their competitiveness in the international market

and thus leading to rapid national productivity growth. Nevertheless, the rapid economic

growth that followed the reform and opening up has ended. With sluggish economic

growth, encouraging local innovation has become a priority (Lo et al., 2022). The

Chinese government has undertaken a number of initiatives to promote innovation and to

find new sources of economic growth in the face of slowing economic growth, such as

mixed ownership reforms.

1.5.1.1 Mixed-Ownership Reforms

Since the reform and opening up, a large number of SOEs have been restructured into

private or mixed ownership. Five stages of SOE reform are proposed by (Lin et al., 2020):

1. 1978 to 1984, at the beginning of the reform and opening up, when

decisionmaking rights were transferred from the government to Soviet-style

socialist firms

2. 1985 to 1992, when control of the company was transferred to SOE managers by

contract

3. 1992 to 2002, when the separation of ownership and control of the company

4. The reorganisation of the ownership of SOEs between 2002 and 2012

5. The reform of the mixed ownership system after 2012

At the end of 2017, that type of mixed ownership had become predominant in SOEs. Of

these, 69% of SOEs with direct links to the national government had converted to mixed

ownership, while provincial SOEs accounted for 56% (Shen and Yang, 2019). In addition,

25 of the 115 Chinese companies that entered the Fortune Global 500 in 2018 had

undertaken mixed-ownership reforms to varying degrees.

Zhang et al. (2020) point out that the mixed-ownership reforms have had a positive effect

on innovation, not only in terms of increased investment in R&D but also in terms of an

increased number of patents. The increased investment in R&D is due to SOEs obtaining

more funds from private enterprises after being turned into mixed ownership (Chen et al.,

2018).

Furthermore, the positive correlation between mixed ownership and innovation is

magnified within firms in monopolistic industries compared to highly competitive

industries. In addition, innovation in mixed ownership is influenced by region. The

eastern part of China has a greater impact on innovation capacity than other regions

(Zhang et al., 2020)

1.5.2 China’s R&D Expenditure

During the downturn in economic growth, Figure 1.3 exhibits that China has continued to

ramp up R&D investment, rising from 1184.7 billion yuan in 2013 to 2786.4 billion yuan

in 2020. In contrast to the annual growth rate in R&D expenditure of around 20% before

2013, the annual growth rate in R&D expenditure after 2013 shrinks to around

10%.

Figure 1.2 R&D Expenditure in China from 2002 to 2021

Source: National Bureau of Statistics

Generally, R&D expenditure is classified into three types - basic research, applied

research and experimental development. From Figure 1.4, experimental development

R&D ranks the top, applied research is second, followed by basic research. As China's

high-tech technologies have been repeatedly "strangled" by Western countries in recent

years, R&D expenditure in basic research in 2020 has nearly tripled compared to 2013.

Even so, R&D expenditure in basic research is still far below the other two R&D

categories.

Besides, R&D funding is mainly from the government and firms. Figure 1.5 demonstrates

that government funding for R&D is between 60% and 70% of total R&D funding

between 2002 and 2003. From 2004 to 2015, the state share hovers between 70% and

80%. After 2015, the state share of R&D funding is as high as 80%, while firm

R&D funding accounted for only 20%.

Figure 1.3 Types of R&D expenditure in China from 2002 to 2020 (￥ billion)

Source: National Bureau of Statistics

Figure 1.4 R&D expenditure in China

from 2002 to 2020 (%)

Source: National Bureau of Statistics

1.5.3 China’s Patent Data

Figure 1.6 presents that the number of patent applications in 2020 has more than doubled

compared to 2013, during a period of slowing economic growth in China. However, the

post-2013 patent growth rate per year has decelerated markedly compared to the pre-2013

period. The total number of patents is perennially dominated by invention patents, which

fluctuate between 60% and 70%, as shown in Figure 1.7.

Figure 1.5 Patent Data in China from 2002 to 2020

Source:

National Bureau of Statistics

Figure 1.6 Types of Patent Data (%)

Source: National Bureau of Statistics

1.6 Contributions

This thesis confirms that, through economic theories and with the use of new data, firm

ownership remains an important consideration in constructing firms' innovation capabilities. It

clearly depicts the impact of firm ownership on innovation performance and finds that

different ownership structures have different effects on innovation performance. This is the

first contribution.

In particular, the use of two thresholds, 5% and 20%, further clarifies the different impacts of

ownership structures on innovation performance and helps to guide firms to establish a good

ownership structure and improve their innovation performance. In the fifth round of SOE

reform, the 5% and 20% thresholds can also give the government a reference for SOE reform,

so that SOE reform can lead to a better ownership structure and promote innovation

performance. This is the second contribution and fills the research gap.

In addition, this thesis also shows that there is no relationship between firm ownership and

innovation specialisation. Also, there is no relationship between firm ownership and

innovation diversification. It also fills the research gap focusing on innovation diversification

and specialisation. In China's unique economic environment dominated by state-owned

enterprises, firms' innovation specialisation and diversification are more likely to follow

policy support or the needs of the country.

Indeed, different countries have different corporate governance systems, mainly manifested in

ownership structures, board independence, CEO duality, the presence of an audit committee,

et cetera. The significant impacts of firm ownership on Chinese firms depend on the corporate

governance system in China, which is currently an ongoing mix-ownership reform.

1.7 Structure of Thesis

Chapter 1 is introduction. Chapter 2 is literature review. Chapter 3 is research design

containing hypotheses proposed. Chapter 4 is research methodology. Chapter 5 is data

analysis. Chapter 6 is conclusion.

Chapter 2 Literature Review

2.1 Introduction

This chapter reviews the concepts of innovation and firm ownership in the mainstream

literature and discusses the importance of knowledge for innovation. Section 2.2 is the

literature review about innovation, including the measurement of innovation used in the

research. Section 2.3 contains five different types of firm ownership in China and focuses

on the relevance of innovation and company ownership. Section 2.4 introduces the

importance of knowledge for innovation. The study treats the knowledge from two

perspectives – depth of knowledge and breadth of knowledge. Finally, section 2.5

explains fresh insights from past literature.

2.2 Innovation

2.2.1 Introduction

Innovation is referred to as a new method, idea, product, et cetera. Baregheh et al.

(2009) calculate the number of definitions of innovation in different scientific papers.

Amazingly, there are totally about 60 definitions. Therefore, they tried to merge and

summarise the definitions of innovation in multiple disciplines and made the following

conclusions: “Innovation is the multi-stage process whereby organisations transform ideas

into new/improved products, services or processes, in order to advance, compete and

differentiate themselves successfully in their marketplace. ” (Baregheh et al., 2009). As

people likely confuse innovation with creativity, Amabile and Pratt (2016) distinguish

creativity from innovation and give their respective definitions: (1)

creativity is a new and useful idea generated by an individual or teamwork; (2) innovation

is the successful implementation of creativity within the organisation.

2.2.2 Frameworks to Determine Types of Innovation

There are various frameworks to determine the types of innovation.

The first framework, which is created by HBS professor Clayton Christensen, contains

two types of innovation – sustaining innovation and disruptive innovation (Bower and

Christensen, 1995). Sustaining innovation is to innovate persistently existing products or

services based on the needs of current customers (Satell, 2017). For example, customers

prefer faster CPUs for electronic products or higher pixel sizes for mobile phones.

Disruptive innovation means new products, services, or business models disrupt the

market, eventually replacing other competitors as industry leaders (Christensen, 1997).

Amazon and Netflix are two examples of disruptive innovation. Disruptive innovation

plays a vital role in a successful business in the long term (Christensen and Overdorf,

2000).

The second framework is that Satell (2017) further subdivides the types of innovation

based on Christensen’s research (Bower and Christensen, 1995; Christensen, 1997) by

solving different types of problems. He believes that the core of innovation is to solve

problems and summarises four types of innovation:

a) sustaining innovation: it aims to refine capabilities in existing markets. In this

case, people have a clear understanding of what problems to solve and what skills

to use.

b) breakthrough innovation: the problem is well-defined, but is hard to resolve within

the area in which the problem arose. However, that problem may be solved

quickly within adjacent areas (Kuhn, 1970).

c) disruptive innovation: if the basic logic of market competition shifts, due to

technological changes or other changes in the market, firms’ products or services

will gradually be eliminated from the market, even if the products or services are

significantly refined. Product innovation will only worsen firm development when

it happens, and firms must innovate their business models instead (Christensen,

1997).

d) basic research: the definition of basic research is that it has the capability to form a

basis for the growth of global technology in the long term (Iansiti and Lakhani,

2017). In pathbreaking innovations, there is no boundary. People can always find

some new phenomena. Just as no one could know how the world would be shaped

by Einstein's discoveries, or no one could guess that Alan Turing's universal

computer would one day become a reality (Satell, 2017). From the perspective of

the history of science, a large number of basic research construct a scientific

knowledge system. It is not for immediate application, but will be found and

applied to benefit humanity after a long time period. Therefore, various countries

attach more and more importance to basic research, especially China. Facing the

technological blockade of western countries, China has swollen its investment in

basic research. Data from the National Bureau of Statistics in China show that the

investment in basic research reached 133.6 billion yuan in 2019, accounting for

6% of the total social R&D expenditure for the first time. This proportion hovered

around 5% for many years. In 2020, China's basic research funding was 146.7

billion yuan, an increase of 9.8% over the previous year. In 2021, China's basic

research investment reached 169.6 billion yuan, accounting for 6.09% of the total

social R&D investment. The third framework is McKinsey’s three horizons model

(Baghai et al., 2000). It helps firms think about their future, sort out their business

portfolio, and formulate their strategy for business coordination. Eventually, firms

achieve the goal of healthy firm growth. The model discloses how innovation

happens in those three-time horizons:

a) Horizon 1 is to maintain and strengthen the core business, thereby innovating

continuously within existing products, services, or business models. The delivery

time of Horizon 1 could be 3 to 12 months.

b) Horizon 2 is to explore and discover new expansion, extending the core business

to new markets, for example. The delivery time of Horizon 2 could be 24 to 36

months.

c) Horizon 3 is to create entirely new possibilities and competencies, thereby

responding to or utilising disruptive opportunities. The delivery time of Horizon 3

could be 36 to 72 months.

In modern society, Blank (2019) disagrees traditional delivery time of three horizons. In

the last century, some disruptive ideas required years of research, design and delivery.

Nonetheless, the delivery time of ideas for Horizon 3 can be as fast as the one for Horizon

1 today, so the delivery time of Horizon 3 products, strategies and capabilities has a

devastating impact on competitors. For example, Airbnb uses existing technologies (i.e.

mobile phone app and landlord) to create a new asset-light business model for tourist

accommodation, unlike hotels and deployed quickly in the short term. Bizarrely, the most

common user of rapid Horizon 3 disruption is not market leaders but challengers and new

entrants (e.g. China) (Blank, 2019).

2.2.3 Stages of Innovation

The early model comprehends three stages of innovation – idea generation,

problemsolving, and idea implementation (Utterback, 1971). Nevertheless, at that time,

the focus was on the manufacturing industry. Consequently, Utterback’s model for stages

of innovation is relatively inadequate.

Then, Mariello (2007) proposes a model with five stages of innovation. The first stage is

to generate the idea through free exploration or competition and hand it over to people

who can promote it. The second stage is to promote and screen the idea. The third stage is

to test the idea under specific conditions. The fourth stage is to commercialise the idea,

including cost and benefit evaluation. Finally, the fifth stage is to diffuse the idea in the

company, so the internal staff can accept it and spend resources to implement it

systematically.

2.2.4 Research and Development

Research and development (R&D) is a series of innovative activities by enterprises or

governments while innovating or introducing new products or services. Hence, R&D is

the first stage of developing potentially new products or services, whereas innovation is

the process of transforming ideas into new products or services. To put it differently, R&D

is a part of innovation in the definition, even though R&D and innovation are sometimes

interchangeable. As innovation is non-quantitative, R&D is applied instead to evaluate or

compare innovation. The proportion of R&D expenditure of gross domestic spending in

2020 for the G20 is displayed in Figure 2.1. Korea ranks first, with 4.8% of gross

domestic product (GDP). United States has the second highest percentage of R&D

expenditure over GDP (i.e. 3.5% of GDP). Japan follows it with 3.3% of GDP. R&D for

Germany is 3.1% of GDP, which are fourth. The fifth is China and France, with 2.4% of

GDP. Other countries are below 2% of GDP. Some G20 countries' data are not shown in

Figure 2.1 due to missing data.

Figure 2.1 The proportion of R&D Expenditure of Gross Domestic Spending

2.3 Firm Ownership

2.3.1 Introduction

Institutions change among countries (Vitols et al., 2001). Freeman (1995) stresses firm’s

R&D activities, ownership structures, and control are overwhelmingly dependent on the

domestic platform, and thus the country-specific institutional factors play a significant

role in determining the national innovation system. In other words, the crossnational

Japan 2020

3.3

% of GDP

Korea 2020

4.8

% of GDP

China 2020

2.4

% of GDP

Russia 2020

1.1

% of GDP

Turkey 2020

1.1

% of GDP

Italy 2020

1.5

% of GDP

Germany 2020

3.1

% of GDP

France 2020

2.4

% of GDP

Canada 2020

1.7

% of GDP

USA 2020

3.5

% of GDP

Mexico 2020

0.3

% of GDP

diversity of institutional settings can cause differences in national innovation systems

(Nelson ed., 1993). As a result, countries differ in terms of which innovation strategies

they apply and which innovation performance they produce. Also, the institutional

divergence explains how ownership constituents affect the firm’s innovation strategies

(Hoskisson et al.,2002). Anguilera and Jackson (2003) agree that this diversity is due to

institutional differences but also explain it through stakeholder interactions. The former

matters through their capacity to support the latter, and the latter generates different

demands on the national institutional configurations. In developed countries, there are

three models of corporate governance structure with respect to country-specific contexts:

(1) the Anglo-American model, which is featured by outside shareholders (which means

the separation of ownership and control), marketorientated corporate control, equity

finance, and state laws; (2) Japanese model, which is characterised by inside shareholders,

a long-term and strong relationship between companies and bank, corporation controlled

usually by insiders, and legal framework designed to stimulate groups affiliated by trading

relationships and cross-shareholdings; (3) German model, which highlights concentrated

ownership, the long-term linkage between bank and corporation, bank representatives as a

board of directors, and nonmarket orientated corporate control (Morck and Nakamura,

1999; Gibson 2000; Vitols et al., 2001; Aguilera and Jackson, 2003; Toms and Wright,

2005). Based on these models, scholars examine the causatives between corporate

governance and firm innovation performance (e.g. Francis and Smith, 1995; Guadalupe et

al., 2012; Aghion et al., 2013; Matzler et al., 2015).

However, many countries' contexts cannot simply be defined as those three models due to

institutional heterogeneity (Aguilera and Jackson, 2003). Especially, these established

models cannot describe corporate governance structures in developing countries. For

instance, because of reform and opening-up policy, Chinese firms' ownership constituents

have shifted from one dominant state ownership to five modern ownership - state

ownership (Lin et al., 2010; Cullinan et al., 2012; Song et al., 2015; Dong et al., 2022; Lo

et al., 2022), concentrated ownership (Ma et al., 2010; Zeng, 2010; Clò et al., 2020),

insider ownership (Su, 2004; Cheung and Wei, 2006; Cheng et al., 2021), institutional

ownership (Hadani, 2012; Schmidt and Fahlenbrach 2017; Lin et al., 2017; Li and Ji,

2021), and foreign ownership (Yoshikawa et al., 2010; Dachs and Peters, 2014; Kwon and

Park, 2018; Dong et al., 2022) - over last four decades. However, the diversity of Chinese

corporate ownership does not entirely deny the crucial role of state-owned firms in firm

performance and innovation. State ownership is still widely used in China and trusted by

ordinary people. Xie et al. (2022) observed that the corporate governance structure model

in China no longer converged with the AngloSaxon corporate governance structure

model. From 2015 to 2017, Chinese SOEs applied an alternative governance model that

legalises the governance authority of the CPC over the board of directors through a set of

regulatory attempts. The politicised governance model disrupted investor confidence in

firms and raised significant concerns about the state's power undermining other

shareholders' interests (Xie et al., 2022). Indeed, the politicisation of the governance

structure may be a genuine concern for foreign investors. Likewise, Cheng (2021) pointed

out that the US system is geared towards allowing businesses to affect the government,

while the Chinese system is geared towards aligning businesses with government

objectives.

In addition to the ordinary shareholders of listed companies, there is a special category of

shareholders in China known as legal persons, which is broadly similar to the status of

institutional shareholders recognised in the United States. Their shares are owned by

domestic institutions independent of, or partly owned by, the central or local government.

In other words, SOEs consist of state-owned and legal person shares. Indeed, a review of

the list of shareholders for all Chinese listed firms discloses that the majority of

shareholders classified as legal person is, in fact, owners with close ties to the state.

In short, this section deals with five ownership structures – concentrated ownership,

insider ownership, state ownership, institutional ownership and foreign ownership -

including the link between them and innovation.

2.3.2 Agency Theory for Ownership Structure

2.3.2.1 Introduction of Agency Theory

According to Jensen (2000)’s definition, agency theory assumes that a contractual

relationship exists between two conflicting parties: one party is the principal or supervisor

(i.e. the principal); the other one is the subordinate (i.e. the agent). The principal

delegates decision-making powers to the agent and expect the agent performs some action

in the principal’s favour. In contrast, the principal’s and the agent’s behaviours can be

motivated by their self-interest (i.e. preference, conviction and information) because they

are considered a rational economic person. Then, agency theory comprises the following

issues (Akerlof, 1970; Murthy and Jack, 2014):

1) Asymmetric information: principal-agent problem usually arises when the

interests of the two parties are divergent and information is asymmetric (the agent

has more information than the principle), so that the principal cannot directly

guarantee that the agent’s interests always align with the principal’s best interests.

2) Moral hazard: moral hazard is likely to happen under information asymmetry. One

example is that the agent may lack effort in performing the delegated task, but the

principal has difficulty in evaluating the level of effort actually spent by the agent.

3) Adverse selection: this is where an agent misrepresents their skills in performing a

task, and the principal is unable to thoroughly check this in advance of deciding to

recruit them. To avoid this, the principal may contact the person for whom the

agent previously provided services.

In the modern corporation, a firm owner or shareholder is the principal, and a firm

manager is the agent. The separation of ownership and corporate control generates agency

problems. Agency theory addresses this problem arising from the different objectives or

desires between the principal and agent (Berle and Means, 1932; Clark et al., 1985).

These differences generate agency cost such as residual loss

(Jensen and Meckling,1976). However, it can be mitigated or eliminated by monitoring

managers' behaviour or providing them with incentive schemes that reward them

financially for maximising shareholder benefits (Boučková, 2015; Feldman and

Montgomery, 2015). Typically, the schemes cover the incentive stock plan that managers

obtain shares, perhaps at a lower price. Accordingly, the objective of managers is in

alignment with one of the shareholders (Jensen and Meckling, 1976). The agency theory,

therefore, assumes that people are egoists and suggests that conflicts of interest between

principal and agent can be dissipated by providing appropriate incentives or monitoring

(Berle and Means, 1932; Jensen and Meckling, 1976).The agency theory therefore

assumes that people are egoists and suggests that conflicts of interest between principal

and agent can be dissipated by providing appropriate incentives or monitoring (Berle and

Means, 1932; Jensen and Meckling, 1976).

2.3.2.2 Problem of Agency Theory in Developing Countries

Traditional agency theory was established to build a foundation for the Anglo-American

model, so previous studies of agency theory have mainly focused on developed countries

(Young et al., 2008). The emphasis of corporate governance reforms has therefore been on

addressing the problems caused by the separation of ownership and control of companies.

Nevertheless, some argue that traditional agency models are not applicable to developing

countries (Yusuf et al., 2018). It is because firms in developing countries have a relatively

high level of concentrated ownership, high family control and mismanagement (Young et

al., 2008; Yusuf et al., 2018). Despite such shortcomings, corporate governance reforms

based on the agency theory have diffused in developing countries mainly due to the desire

of these firms for investments from foreign financial institutions. Accordingly, there is a

lack of widespread attention to the issue of inappropriateness arising from corporate

governance reforms based on agency theory in developing countries (Yusuf et al., 2018).

The inapplicability of agency theory leads to ineffective or inefficient corporate

governance reforms in developing countries (Reed, 2002; Uddin and Choudhury, 2008).

As an example, in Pakistan, agency theory is weakened by management's opportunism in

not being able to use financial incentive schemes to reconcile the interests of the principal

and the agent, and by the scarcity of independent directors (Yusuf et al., 2018). He,

therefore, recommends the use of rigorous external audits and the appointment of audit

firms by the regulator. The other method is to introduce legislation protecting the rights of

minority shareholders.

2.3.2.3 Agency Theory and Innovation

According to agency theory, shareholders (i.e. principals) hire managers (i.e. agents) to

operate the firm on behalf of the shareholders (Jensen and Meckling, 1976). The growth

of firm value is related to shareholders' control power over managers' behaviours and

strategies. It is because the agent shrinks his/her responsibilities to the principal whenever

the agent has the opportunity because of egoism (Eisenhardt, 1989a). The most capable

agent can mitigate a rise in agency risks owing to adverse selection and moral hazard

(Jensen, 1993).

Radical technological innovation (RI) may compound the agency risks of adverse

selection and moral hazard due to high uncertainty of the link between technology and

markets and ignorance of the relationship between technology and market outcome

(O'Connor and Rice, 2013). Furthermore, RI requires the pooling of multiple agents to

participate in R&D (Winter, 2013), further magnifying information asymmetries and

making it difficult to allocate decision-control rights, set appropriate rewards and screen

out R&D failures due to management egotism (Aghion and Tirole, 1994). As a result,

senior managers may be reluctant to carry out RI in their core business for the concern

about disrupting the core business, while creating apprehension among bottom-level

employees (Christensen, 1997). Indeed, technical management has always believed that

RI should be separated from existing operations and that it should offer risk-averse and

middle-level management the autonomy to explore the uncertain routes to growth.

(Burgelman, 1991).

In contrast to traditional studies, some studies consider the top manager the principal and

the middle manager the agent (Jones and Butler, 1992; Shaikh and O'Connor, 2020).

Contrary to agency theory, innovation research usually discourages the usage of financial

incentives. The reason is that it can lead to jealousy and enmity among team members,

which can ruin a good atmosphere of teamwork. Lazonick (2007) presents an opposite

view that most well-established R&D firms make reparation to top managers with options

and equity. Those compensation packages can be even higher than those of

Wall Street executives (Lazonick and Tulum, 2011). The two opposing views on which

Shaikh and O'Connor (2020) are based state that the senior management's use of financial

rewards can drive RI project teams to incremental innovation, although the optional non-

financial external rewards used can enhance the intrinsic motivation of individual agents

to engage in the pursuit of RI at the project level.

2.3.3 Resource-based Theory and Innovation

2.3.3.1 Introduction of Resource-based Theory

The idea of resource-based view was first emerged in Wernerfelt’s paper (1984) and

evolved into a matured theory about corporate governance in Barney et al.’s paper (2011).

The main prospective of resource-based theory is to explain why firms differ and how

they achieve sustainable completive advantage based on firm resources and capabilities

(Barney et al., 2001). The resources that the firms can apply for implementation of

business strategies can be tangible and intangible assets containing physical assets,

financial capital, human capital, organizational resources (Barney and Arikan, 2005). The

capabilities can be management skills and firms’ organisational processes for example

(Barney et al., 2011). Faced with rapid technological change, Teece et al. (1997) go

further to the idea of “dynamic capabilities” and confirm that firms’ specific

organisational processes, tangible and intangible assets (knowledge stock) and managerial

processes determine the firms’ profitability because those three factors significantly affect

firms’ competencies based on the level of difficulty of replicability (expanding internally)

and inimitability (barriers to imitation).

2.3.3.2 Resource-based Theory and Innovation

In developing counties, firms may face resource scarcities and obsolescence. A crucial

challenge is to comprehend what impedes firms to acquire resources and capabilities and

how they might resolve (Wright et al., 2005). Hence, resource-based theory can be

applied to investigate innovation activities in a firm that require valuable and specific

resources. Individual firms usually rely on external recourses for innovation as they do

not have all needed resources. According to this perspective, the assumption of outside

shareholders who have rich resources benefiting innovation activities and firm

performance is made. Consequently, firms with state ownership can be seen as boundary

spanner. Those firms in developing countries have political and financial privileges over

others (Vo, 2018) and provide policy and resource benefit which is crucial for firm

development (Zhou et al., 2017). Specifically, state-owned firms can obtain more

tradeable or non-tradeable resources - important infrastructure resources (Chang et al.,

2006; Siegel, 2007), land (Tan, 2006; Chen et al., 2014), policy support, natural link exists

among SOEs, universities and research institutes, or R&D resources

(Wang et al., 2017) - to invest in innovation activities.

2.3.4 Concentrated ownership

2.3.4.1 Introduction of Concentrated Ownership

Concentrated ownership means that majority of stock shares are owned by individuals,

non-institutional or institutional investors, who have voting rights and/or cash flow rights.

Traditionally, the only degree of equity concentration is considered an indicator of

concentrated ownership. Usually, investors hold at least 5% of shares. The higher the

shares owned by a few investors, the stronger the governance power. It is because larger-

block shareholders have strong monitoring power over incumbent managers. As argued

by agency theory, shareholders’ interests are alignment with managers’ interests (Ortega-

Argilés et al., 2005), alleviating agency conflicts and reducing agency costs (Jensen and

Meckling, 1976; Hill and Snell, 1988; Francis and Smith, 1995; Morck et al., 2005;

Belloc, 2012). Consequently, concentrated firms perform better financially than separated

ownership (Claessens and Djankov, 1999; Xu and Wang,

1999; Singal and Singal, 2011; Wang et al., 2012; Nguyen et al., 2015).

2.3.4.2 Concentrated Ownership and Innovation

Concentrated ownership has a positive impact on innovation as it lowers agency costs and

constrains the behaviour of managers (e.g., Di Vito et al., 2010; Hill and Snell, 1988;

Holmstrom,1989; Baysinger et al., 1991; Francis and Smith, 1995; Lacetera, 2001).

Specifically, Baysinger et al. (1991) study 176 Fortune 500 companies by using linear

regression with R&D spending per employee as a dependent variable and stock

concentration levels as an independent variable, meanwhile controlling for average

industry R&D intensity, diversification, and firm size. Lacetera (2001) focuses on 27

US pharmaceutical firms from 1994 to 1999 by using the cross-sectional time-series

FGLS method with R&D intensity as a dependent variable and ownership concentration,

insider shareholding, insider presence in the board of directors, the presence of scientists

in the board of directors as independent variables, controlling firm size and financial

stability. Di Vito et al. (2010) use panel regression to investigate 259 firms in Canada

from 1998 to 2007 with R&D intensity as a dependent variable, concentrated ownership –

the level of voting rights, the difference between voting and cash flow rights, heir-

controlled, founder-controlled, and family-controlled - as independent variable, and the

presence of institutional shareholding, firm age, firm size, firm growth, the long-term

debt, and industry dummies. Chang et al. (2006) compare innovation performance in

group-affiliated firms and independent firms by collecting the data from the top 500

Taiwanese and Korean manufacturing firms between 1991 and 1999 and performing the

Poisson regression and the negative binomial regression model with the number of patents

as a measurement of innovation performance. One reason is that controlled owners are

more willing to take innovation strategies with high risk even if the possibility of success

of an investment project is low, while managers within dispersed ownership prefer

imitation strategies with low risk as they undertake the cost of failure (Hill and Snell,

1988). In transition economies, a high degree of concentration provides an efficient

monitoring mechanism that is crucial to innovation activities in transition economies

(Nguyen et al., 2015). Moreover, shareholders pursue a high return by virtue of greater

investment in innovation even though a specific high risk is borne (Hill and Snell, 1988;

Baysinger et al., 1991).

In contrast, some studies argue that stock concentration has a negative influence on

innovation performance regardless of a country’s level of development (Ortega-Argilés et

al., 2005; Di Vito et al., 2010; Minetti et al., 2015; Shi and Xie, 2016; Steffen and Iuliia,

2016; Wan et al., 2021). In opposition to the R&D intensity as a measurement of

innovation performance (dependent variable), Di Vito et al. (2010) perform the number of

patents as the alternative measurement of innovation performance to get the negative

impact. Ortega-Argilés et al. (2005) apply the Tobit-type model and the Passion

regression model with R&D expenditure per employee and the number of patents as

dependent variables respectively to explore in Spanish industries during 2001. The

relevant main independent variable is the degree of separation of ownership and

management functions, and control variables contain the firm size, firm age, debt,

concentrated ownership, a dummy variable for a listed firm or not, technological

opportunity level, region in which the firms located, and the structure of the market in

which the firms operated. Steffen and Iuliia (2016) analyse 75214 firms from 24 emerging

market economies (e.g., Taiwan, Israel, Hungary, South Africa, Brazil, Thailand, Poland,

Chile, and Mexico) between 1998 and 2012 by using the fixed effects model with R&D

expense as a dependent variable, ownership concentration and the levels of shareholder

rights protection as independent variables, sensitivity of cash flow rights, inflation rate,

firm size, leverage, and a complementarity effect between the levels of capital and R&D

intensity as control variables. Minetti et al. (2015) investigate 20000 Italian manufacturers

by using OLS, Probit, two-stage least squares (2SLS) regressions with R&D expenditure,

R&D personnel, and the number of patents as main dependent variables, degrees of

ownership concentration as the main independent variable, firm age, firm size, number of

employees, credit rationing, credit relationship and the duration of the relationship with

the main lender (banks), region in which the firms located, number of branches,

provincial GDP growth, provincial Herfindahl, and local financial development. Wan et

al. (2021) study Chinese firms listed on the Shanghai and Shenzhen Stock Exchanges

from 2007 to 2018 by using the pooled crosssectional regression with R&D intensity as

the dependent variable, ownership concentration as the main independent variable, firm

size, leverage, firm performance, Tobin’s Q, CEO age, cash holdings, operating cash flow,

and the nature of the controlling shareholder as control variables. The reasons include

agency conflicts between large-block and minority stockholders, risk aversion caused by

an absence of diversification, increased risk undertaken by owners, liquidity constraints in

the market, and fewer opportunities for negotiation of firm value (Ortega-Argilés et al.,

2005; Choi et al., 2011; Minetti et al., 2015). Further, Wan et al. (2021) explain that the

tunnelling effect of controlling shareholders disincentives the investment in R&D. By

definition, the tunnelling effect is an immoral business behaviour in which large

shareholders transfer firm assets or profits to private-owned firms for their benefit at the

expense of minority shareholders. Accordingly, an improvement in the regional

governance environment can dampen the negative effect of concentrated ownership on

innovation performance caused by the tunnelling effect (Wan et al., 2021).

Besides, concentrated ownership and innovation performance are non-linearly related

(Abdullah et al., 2002; Gompers et al., 2004). Lee (2005) focuses on listed firms in the US

and Japan during 1995 across seven industries - automotive, chemicals, communication,

computers, electronics, pharmaceuticals, and power – by using the country-specific and

pooled regressions and proposes that the concentration-innovation relationship is not only

nonlinear but also nonmonotonic: in the US, ownership concentration is negatively

related to innovation outcomes when R&D investment is low, but positively related to

innovation when R&D investment is high; in Japan, ownership concentration is positively

related to innovation when R&D investment is low, but negatively related to innovation

when R&D investment is high. The relevant dependent variable is the number of patents,

the independent variable is stock concentration, and control variables consist of

contemporaneous R&D expenditures, firm size, market-to-book, leverage, and industry

dummies. Li et al. (2010) and Chen et al. (2014) also reveal an inverse U-shape

relationship between ownership concentration and innovation in emerging economies

such as China by using one-factor, two-factor, and correlated uniqueness models with a

seven-point differential scale for product innovation as the dependent variable,

concentrated ownership as the independent variable, firm size, region, a industry dummy

variable to identify a high-tech industry, sales growth and return on equity, a dummy

variable to identify a state-owned firm, firm development stage, and effective production

process as control variables. The relevant sample period is from 2002 to 2004. The other

reason of non-linear relationship is that the effect of board independence on firm

performance rises as the degree of concentration decreases (Li et al., 2015).

2.3.4.3 Concentrated Ownership and Hypothesis

Firms with concentrated ownership have agency costs created from principal-agent

objective conflict. The majority shareholders are likely to advance their interests by

expropriating from minority shareholders in developing countries (Su et al., 2008). It is

not feasible in developed countries where the concentration of ownership may lead to

more effective monitoring mechanisms (Fama, 1980; Zajac and Westphal, 1994), but it

may not be the case in developing countries. Aside from this, risk aversion due to lack of

diversification, increased risk to be borne by owners, liquidity constraints in the market,

and reduced opportunities to negotiate the value of the firm can all negatively affect

innovation performance for ownership concentration (Ortega-Argilés et al. 2005; Choi et

al. 2011; Minetti et al. 2015). In recent research, Wan et al. (2021) raise the issue of the

tunnelling effect in Chinese firms, which hurts minority shareholders’ interests and

undercuts R&D investments. Hence, the following hypothesis is proposed: H1.

Concentrated ownership and the firm’s innovation performance will be negatively related.

2.3.5 Insider Ownership

2.3.5.1 Introduction of Insider Ownership

Insider ownership is defined as shareholding by individuals closely related to the firm's

management and/or individuals with exclusive voting rights. It includes the firm's

founders and their descendants, subsidiaries, managers, executive directors and

employees (Chang et al., 2006). It then can be grouped into three types of insider

ownership – managerial ownership (e.g. Vijayakumaran, 2021), family ownership (e.g.

Delgado-García et al., 2022), and employee ownership (e.g. Hennig et al., 2022. ).

Firms with insider ownership demonstrate that managers' objectives are aligned with

insiders' interests rather than those of dispersed outsider ownership shareholders. Insiders

are more aware of the reality of the company than outsiders, especially individual

investors (Choi et al., 2012). As a result, a higher firm performance can be achieved for

firms with insider ownership (e.g. Drakos and Bekiris, 2010). In other words, insider

ownership – managerial ownership (e.g. Jensen and Mecking, 1976; Florackis et al.,

2009), family ownership (Chang, 2003; Morck et al., 1988), and employee ownership

(e.g. Nickel, 1990; Kim and Patel, 2017) – has a positive impact on company performance

owing to the lower agency costs.

Conversely, some studies propose that managerial ownership (e.g. Gomes, 2000; Acharya

and Bisin, 2009) and family ownership (e.g. Thomsen and Pedersen, 2000) are negatively

correlated to firm performance. Thomsen and Pedersen (2000) provide evidence of firm

performance with family ownership destroyed by firm owners who are risk averse.

2.3.5.2 Insider Ownership and Innovation

Those two opposite effects of insider ownership on firm performance also result in reverse

views of firm innovation performance.

On the one hand, there is a positive relationship between insider ownership and

innovation performance (e.g., Chang et al., 2006; Choi et al., 2011; Chen et al., 2013;

Lodh et al., 2014; Minetti, 2015). For example, Lodh et al. (2014) apply an unbalanced

panel regression to investigate 395 listed Indian firms on the Bombay Stock Exchange

from 2001 and 2008 with number of patents and innovation productivity as the dependent

variables, family ownership as the main independent variable, firm size, firm age,

knowledge stock, a dummy variable to determine whether the CEO is a member of a

founding family, foreign ownership, government ownership, wage intensity, employee

compensation, industry dummies, and business risk as control variables. Chen et al.

(2013) investigate Taiwanese listed firms from 1996 to 2007 by using the Tobit regression

with R&D expenditure and the number of patents as the dependent variables, a continuous

family ownership and a dummy variable to identify the presence of the members of the

founding family as the independent variables, firm growth, operating cash flow, the

volatility of cash flow, leverage, capital intensity, CEO overconfidence, past firm

performance, firm size, firm age, year dummies, and industry dummies as control

variables. Song et al. (2015) study 242 listed firms in China during 2009, by using the

OLS and Tobit regressions with the ratio of the revenue generated by new products to the

total revenue generated by all products as the dependent variable, market orientation,

identity of the dominant shareholder, management ownership, and ownership

concentration as the independent variable, firm size, firm age, marketing expenditure

intensity, R&D expenditure intensity, and industry dummies as control variables. One

reason of the positive impact is that the firm owner can have a long-term horizon that

increases investment in innovation, because the firm will be passed on to descendants

(Caselli and Gennaioli, 2013). Due to the greater concern about the longterm presence of

the firm, family ownership has a positive effect on firm innovation (Chen et al., 2013;

Lodh et al., 2014; Minetti, 2015). Regarding employee ownership, employees prefer

stable jobs by maximising long-term value, thereby seeking technological innovation to

increase firm value. Thus, it results in a positive firm innovation performance (Chang et

al., 2006). Also, talented human resources are required for innovation activities that

employees can participate. Consequently, knowledge diffusion occurs between those

employees and others in the firm (Choi, 2012). Finally, managerial ownership magnifies

the positive correlation between market orientation and firm innovation, as top managers

can change their preferences – time and risk preferences - to be the same as the

shareholders' preferences (Song et al.,

2015). In this context, the managers can invest in an R&D project that may ensure

future performance (Chang, 2003). Hence, managerial ownership benefits firm innovation

performance. Kurt et al. (2015) concentrate on 340 large German listed firms between

2000 and 2009 by using the 2SLS regression with the forward citations of patents

(innovation output) as the dependent variable, family ownership, family management, and

family governance as the independent variables, firm performance, firm size, firm age,

firm risk, firm leverage, firm capital intensity, firm intangible assets intensity, industry

R&D intensity, and industry investment share as control variables. On the other hand,

Decker and Günther (2017), Liu et al. (2017) and Chi (2023) suggest that family

ownership and firm innovation are negatively correlated. In contrast to innovation output

as the dependent variable used in Kurt et al. (2015)’s paper, the innovation input measured

by R&D intensity generates the negative impact of family ownership on innovation input.

Decker and Günther (2017) focus on German machine tool industry during 2000 and 2010

by using the Poisson regression and the negative binomial models with the number of

patents as the dependent variable, family ownership, family generation, personal family

ownership, institutionalised family ownership as the independent variables, firm age, firm

size, knowledge stock, population density, the number of potential cooperation partners,

the average number of universities and universities of applied sciences in the region

where the firm located, and the number of year that firms observed in the buyer’s guide as

control variables. Chi (2023) investigates 1391 listed firms which are non-financial and

non-statecontrolled in Taiwan from 2000-2017 by using the fixed effects model with

patent counts and citation counts as the dependent variables, family ownership and family

control as the independent variables, firm size, investment opportunity, leverage, capital

investments, R&D investments, and firm age as control variables. One interpretation of

the negative impact is that a family-owned firm's limitation of management capabilities

may diminish the company's R&D activities (Graves and Thomas, 2006; Bloom and Van

Reenen, 2010). The other one is that the behaviour of family managers is influenced by

informal institutional mechanisms when the goal of the family owner is socio-emotional

wealth (Gómez-Mejía et al., 2007; Berrone et al., 2012). As a result, it restricts R&D

investment in order to lower business risk (Liu et al., 2017). Thirdly, agency costs, which

causes the negative relationship between family ownership and insider ownership, rise

from excess control rights, and can alleviate by external shareholders who are also large

shareholders and highly expected financing costs (Chi,

2023).

2.3.5.3 Insider Ownership and Hypothesis

Agency theory suggests that insider ownership can lower managerial pressures to

maximise short-term values, thereby leading to enhance R&D investment (Choi et al.,

2012). In recent years, however, equity pledges have become an increasingly common

source of financing for company insiders. Andersonand Puleo (2015) report that between

2006 and 2011, at least one insider pledged their shares in US firms, accounting for 26%

of the total sample, with the average insider pledging 33.3% of total equity. Insiders

usually pledge equity for two reasons (Dou et al., 2019): (1) insiders obtain funds through

equity pledges for personal consumption or investment; and (2) insiders use the funds

from equity pledges to purchase shares in their own companies, increasing their control in

the firm. According to data from Wang et al. (2020)’s research, most insiders pledge

equity for personal investments in China. They find that pledging of equity has made

insiders more conservative towards innovative R&D and more reluctant to undertake the

cost of R&D failure. Then, the following hypothesis is suggested:

H2. Insider ownership and the firm’s innovation performance will be negatively

related.

2.3.6 State Ownership

2.3.6.1 Introduction of State Ownership

The definition of state ownership is the percentage of shares in a firm held by central or

regional governments and various entities connected with the government. The existence

of state ownership is imperative to maintain customers’ benefits with lower prices, adjust

anti-inflation or expansion (Marrelli et al., 1998; Willner, 2003), and substitute for

markets that lack private-owned firm capital (Marrelli et al., 1998).

Most studies reveal that state ownership is positively related to firm performance (Vernon-

Wortzel and Wortzel, 1989; Xu and Wang, 1999; Sun and Tong, 2003; Jiang et al., 2008;

Song et al., 2016). Hence, one interpretation is that market competitiveness is insufficient

in developing countries, which can be cured by the public intervention (Willner, 2003).

Further, multiple research argues for a less positive view of government-owned firms

during the privatisation process because of the divergence of objectives (e.g. Dewenter

and Malatesta, 2001; Cornett et al., 2010).

Boycko et al. (1996) refute that state ownership benefits firm performance. They point out

that state-owned firms seek social and political goals rather than profit maximisation.

Excess employment, for example, is one of the results if the managers comply with

politicians’ objectives. Additionally, jobs sourced from state-owned firms are given

priority to candidates who have a political connection (Krueger, 1990). Thus, problems

arising from the state-owned firm’s inefficiency (Ikenberry, 1990) and political failure

(Boycko et al., 1996) lower firm performance, which is supported by empirical studies

with regard to the negative (e.g. Lin et al., 2009) or non-linear relationship (e.g. Yu, 2013;

Hess et al., 2010) between state ownership and firm performance. Nonetheless, state-

owned firms do not always mean inefficiency (Kay and

Thompson, 1986; Vernon-Wortzel and Wortzel, 1989; Willner, 2003).

2.3.6.2 State ownership and Innovation

A critical role is played by the government in facilitating innovation capacity (Schaaper,

2009; Fan, 2011; Franco and Leoncini, 2013; Lo et al., 2022l; Wang and Jiang, 2021) and

developing a firm's innovation activities (Johnson, 1982; Amsden, 1989; Haggard, 1994).

Distinct institutional settings can generate different technological capabilities among

countries (Mahmood and Singh, 2003) and affect firm strategy and innovation process

(Hobday, 2005). For instance, a firm performs better in Korea as technology innovation is

specialised, while diversified technological innovation produces a better firm performance

in China (Bong Choi and Williams, 2013).

In developing countries such as China, governments develop innovation capabilities

indirectly via direct intervention (i.e. direct funding and tax incentives) and by facilitating

the interaction of key implementers (i.e. government research institutions, higher

education and the business sector) in science and technology activities (Schaaper,

2009).

At the firm level, the government is not only portrayed as an investor but also as a

resource allocation coordinator (Xu and Zhang, 2008). Some studies suggest a positive

relationship between state ownership and firm innovation (e.g., Chen et al., 2022. Choi et

al., 2011; Choi et al., 2012; Yi et al., 2017; Li and Xia, 2008; Lo et al., 2022; Xu and

Zhang, 2008). For example, Choi et al. (2011) focus on 548 Chinese listed firms across

eight industries - automotive, chemicals, communication, electronics, machinery,

pharmaceuticals, textiles, and power industries – from 2001 to 2004 by using the Poisson

and negative binomial models with the number of patents as the dependent variable, state

ownership as the main independent variable, firm size, firm profitability, sales growth,

firm age, leverage, long-term investment, knowledge stock, public Ashares, and sectoral

context as control variables. Choi et al. (2012) investigate 301

Korean listed firms from 2000 to 2003 by performing the negative binomial model with

patent data as the dependent variable, state ownership as the independent variable, firm

size, firm profitability, sales growth, firm age, leverage, R&D intensity, business groups,

and technology sector as control variables. Chen et al. (2022) take a sample of all Chinese

listed firms during a period of 2009-2017 by using the chain multiple of mediating effects

models with patent counts as the dependent variable, a continuous state ownership, a

dummy variable for state-owned firms, and the ratio of state ownership to non-state

ownership as the independent variables, government subsidies as mediating varaible, firm

characteristics and external environmental factors -firm size, firm age, profitability,

leverage, cash holding, tangible assets, executive shareholding ratio, board size, board

independence, cultural diversity of the board social trust and environmental regulations -

as control variables. Lo et al. (2022) concentrate on Chinese listed firms between 2007

and 2018 by using OLS, Tobit, PSM regression models with patent data as the dependent

variable, percentage of state ownership in the top three shareholders and a dummy

variable for state-owned enterprises (SOEs) as the independent variables, firm size,

political connection, firm age, leverage, firm profitability, the mode of assets utilisation,

the growth prospects, financing constraints, a dummy variable for the chairman of the

board as well as the top manager, management ownership, concentrated ownership,

institutional ownership, the intensity of research and development in the industry, and

industry dummies as control variables. The first advantages of SOEs in firms' innovation

activities are political, financial and resource-related privileges (Chen et al., 2023).

Specifically, the government offers exclusive endorsements and treatment to SOEs in the

context of weak intellectual property right (IPR) protection (Sheng et al., 2011), which

means SOEs obtain better IPR protection than other types of firms (Wang et al., 2012).

Secondly, SOEs are given priority over non-SOEs in allocating government R&D

expenditure toward developing China's national innovation, which plays an important

institutional role (Sun and Liu, 2014). Thirdly, innovation is full of uncertainty and high

risks. Long-term capital is necessary because innovation has a long payback period and a

high failure rate (Choi et al., 2011, 2012). SOEs can get support from government

subsidies or put administrative restrictions on rivals. It then encourages SOEs to take

more risks (Kornai et al., 2003; Chen et al., 2023). Fourthly, SOEs have access to

important infrastructure resources if policies encourage or discourage certain types of

development. SOEs then benefit from priority rights as well as being used to promote

government-initiated innovation (Chang et al., 2006; Siegel, 2007). Fifthly, the limited

availability of land and high real estate prices indicate significant constraints on

innovation activities that require large R&D centres, but this is not a problem for state-

owned enterprises (e.g. Tan, 2006). Sixthly, in China, tax incentives for product

innovation (i.e. 50%) and intangible asset inventions (i.e. 150%) can be obtained through

approval by the tax authorities. Due to political connections, SOEs are more likely to

receive tax incentives than other enterprises, and so that the former spend less on R&D

than the latter. It allows for higher profits. At the same time, the ample resources and

preferential treatment that SOEs receive from the government can lessen the risk aversion

of top management in their R&D investments and provide an incentive for them to invest

more in creating new knowledge. Most national and provincial research projects in China

are carried out by SOEs, universities or a combination of both. The government can easily

supervise the process of these research projects and assess innovation outputs and

economic benefits, which helps SOEs achieve higher performance (Ruiqi et al., 2017).

Seventhly, a majority of Chinese universities and research institutions are state-controlled

and governed, and so they have some natural links with SOEs. Motohashi and Yun (2007)

's research indicate that over 30% of surveyed SOEs actively outsource science and

technology activities to universities and public research institutions. SOEs can benefit

from university-industry collaboration to obtain complementary capabilities that can

diminish R&D risks, thereby enhancing innovation performance and firm performance

(Eom and Lee, 2010; George, Zahra and Wood, 2002). Eighthly, SOEs rely on their close

ties with the government to gain access to advanced technology and management

experience, as well as the scientific talent they need, as a way to increase the efficiency of

the use of R&D resources performance (Wang et al., 2015; Ruiqi et al., 2017). Ninthly,

SOEs are also more likely to receive purchase orders from the government, which greatly

assists in the commercialisation of R&D product performance (Ruiqi et al., 2017).

Conversely, some researchers find state ownership is negatively related to firm innovation

(e.g., Ayyagari et al., 2011; Vo, 2018). Vo (2018) study Vietnam-listed and non-financial

firms from 2007 to 2015 by using the linear regression with risk taking behaviour as the

dependent variable, sate ownership as the independent variable, firm size, firm fixed

assets, firm cash holdings, and firm age as control variables. Ayyagari et al. (2011)

investigate 19,000 firms across 47 developing economies between 2002 and

2004 by applying the logit probability model with the aggregate Innovation Index, Core

Innovation, or eight individual indicators of firm innovation as the dependent variable,

state ownership as the main independent variable, firm size dummies, firm age, legal

status, number of establishments, industry dummies, country dummies, and capacity

utilization as control variables. One reason is that while government subsidies enhance

firms' innovation, excessive government subsidies diminish the positive or even negative

impact on firms' R&D, implying a waste of resources (Yi et al., 2021). This is mainly due

to the fact that, in the interest of continued government resource support, firms with large

government subsidies prioritize their R&D strategies and resource allocation, including

specific technological and production goals, to meet governmental concerns rather than to

build up and improve their own innovative capabilities (Rhee and Leonardi, 2018).

2.3.6.3 State ownership and Hypothesis

In the latest research on SOEs, Lo et al. (2022) reaffirm the importance of state ownership

for firm innovation. At the firm level, the advantages of state ownership contain political,

financial and resource-related privileges (Zhou et al., 2017). Drawn on the literature

review in chapter 2, the following hypothesis is recommended: H3. State ownership and

the firm’s innovation performance will be positively related.

2.3.7 Institutional Ownership

2.3.7.1 Introduction of Institutional Ownership

Institutional ownership is defined as the proportion of ownership owned by financial

institutions. There are various types of institutional owners -mutual funds, hedge funds,

pension funds, investment advisers, bank trusts, insurance companies and venture capital

(Chen et al., 2007).

In general, institutional investors are mainly sophisticated professional investors whose

primary purpose is to earn long-term profits for their clients (Connelly et al., 2018).

Institutional owners often work directly or indirectly with their companies due to the size

of their holdings, their investment strategies, their influence on financial markets, and

their failure to sell underperforming firms (Edmans and Holderness, 2017). Direct

engagement refers to the direct involvement of the institution in the discussion of the

firm’s direction and strategy by the firm’s senior management, and such direct

engagement serves as a core competitive advantage for institutional investment services

(Healey and Mintz, 2021). Indirect engagement means that the ability of institutions to

motivate firms and enforce discipline is critical to stimulating and promoting strategic

actions and processes that they believe are beneficial to the company (Brav et al., 2008).

Healey and Mintz (2021, p.840) list five ways in which institutions are typically utilised

to indirectly influence firms: “the appointment of board members, (ii) risk oversight, (iii)

adjustment of executive compensation, (iv) implementation of corporate governance

structures, and (v) public criticism of the firm either via announcements in the media or

support of shareholder proposals”. In sum, institutional investors are enabled to leverage

decision-making in the business. Furthermore, institutional investors (as majority

shareholders) can monitor effectively at a lower agency cost than minority shareholders,

thus positively impacting firm performance (Pound, 1988; McConnell and Servaes, 1990;

Lin and Fu, 2017). As argued by the active monitoring view, managers have greater

pressure from institutional investors to maximise shareholder value (Shleifer and Vishny,

1986). Financial institutions have an incentive of large equity stakes to monitor for the

sake of remitting agency problems and lowering information asymmetries that hinder

innovation (Minetti et al., 2015). Accordingly, empirical studies demonstrate that

institutional ownership is positively correlated to firm value (McConnell and Servaes,

1990; Lin and Fu, 2017; Healey and

Mintz, 2021).

2.3.7.2 Institutional Ownership and Innovation

Likewise, the positive impact of institutional ownership on firm performance, institutional

ownership is also positively correlated to innovation performance (Berger et al., 2014;

Choi et al., 2011; Eng and Shackell, 2001; Fan et al., 2023; Miller et al.,

2022; Mishra, 2022; Opler and Sokobin, 1997; Rong et al., 2017). For example, Choi et

al. (2011) focus on 548 Chinese listed firms during 2001 and 2004 by using the Poisson

and negative binomial models with the number of patents as the dependent variable,

institutional ownership as the main independent variable, firm size, firm profitability,

sales growth, firm age, leverage, long-term investment, knowledge stock, public Ashares,

and sectoral context as control variables. Rong et al. (2017) study Chinese listed firms

from 2002 to 2011 by using the OLS, fixed effects, 2SLS, and Gaussian mixture models

with patent and citation counts as the dependent variables, institutional ownership as the

independent variable, Tobin’s Q, return on asset, leverage, year dummies, industry

dummies, and firm dummies as control variables. Miller et al. (2022) study the listed

firms in the US from 1991 to 2008 by applying the linear regression with patent counts

and citation counts as the dependent variables, monitoring institutional ownership as the

independent variables, firm size, firm age, capacity intensity, labour productivity and

quality, profitability, stock performance, growth opportunities, cash holdings, capital

structure (leverage), stock volatility, institutional block ownership, four-digit SIC

Herfindahl Index and its squared term as control

variables.

Chi et al. (2019) find that only mutual funds as one type of institutional investors

positively affect firm innovation, other types of institutional investors consisting of

insurance company, pension fund and Qualified Foreign Institutional Investor have less

positive or no impact on firm innovation. This paper contains a sample of non-financial

listed firms in China during a period of 2001-2004. It uses fixed effects models with

patent data as the dependent variables, a categorical variable for three types of

institutional ownership (i.e., mutual funds, insurance company and pension fund, and

foreign institutions) as the independent variables, ownership concentration, state control, a

dummy variable for the CEO position holding by the board of chair, board dependence,

the number of board meetings, profitability, sales growth rate, cash holding, leverage, firm

size, a dummy variable for a firm listed on the ChiNext board, and region as control

variables.

Mishra (2022) concludes a positive impact of institutional ownership on firm innovation if

the institutional ownership is below the threshold and turns into a negative impact if the

institutional ownership is above the threshold. The research scope is the number of firms

to the non-financial and non-utility firms listed in the Standard & Poor's (S&P) 1500 in

2005, and the sample period is from 2000 to 2018. It applies Tobit and crosslagged

structural models with R&D intensity and knowledge capital as the dependent variables,

institutional ownership as the independent variable, firm size, financial slack, financial

leverage, firm-specific risk, Tobin's q, CEO tenure, firm diversification, and market

concentration as control variables.

That kind of a positive effect of institutional ownership on firm innovation can be

attributed both to effective agency oversight (by agency theory) and to the role of the

agency in protecting managers from the risk of dismissal (by management career

considerations) (Bushee, 1998; Minetti et al., 2015; Miller et al., 2022). In addition to

measures such as monitoring mechanism, institutional investors in emerging markets can

even threaten invested firms to exit the firms to further minimise agency conflicts

compared to developed markets (Chi et al., 2019). In this case, managers are unlikely to

reduce R&D so as to turn around declining earnings, when the level of institutional

ownership is high (Bushee, 1998). Furthermore, institutional investor networks contribute

firm innovation and gain more patents compared to their counterparts (Fan et al., 2023).

Mishra (2022) disagrees with this view, arguing that managers scale back R&D

investment when institutions retain a high level of ownership. This behaviour of a

reduction in R&D investment depends on the proportion of institutional ownership. When

institutional ownership is below the threshold, R&D investment and institutional

ownership are positively correlated, but the opposite is true when it is above the threshold.

To elaborate, when institutional shareholding exceeds a certain threshold, managers will

succumb to pressure from institutions, resulting in management shortsightedness. Short-

sightedness refers to maximising the short-term value for managers rather than the long-

term value (Bushee, 1998). The cause may be that institutional investors only hold shares

for a short period, resulting in a short-term horizon (Minetti et al., 2015). In contrast,

long-term institutional ownership signals a shift from financial investment to an

investment in the firm's future, helping reassure managers. Directly, there is a positive

influence of the stability of institutional investors' shareholdings on the relationship

between institutional ownership and innovation (Sakaki and Jory, 2019). Moreover, active

or passive monitoring depends on the types of institutional investors, the shareholdings of

institutional investors, stress sensitivity and institutional investors from national or

international firms (Lin and Fu, 2017; Sakaki and Jory, 2019;

Mishra, 2022).

2.3.7.3 Institutional Ownership and Hypothesis

Institutional investors monitor the behaviour of managers either explicitly or implicitly

through governance activities or by gathering information on the quality of R&D

investments respectively (Bushee, 1998). Active monitors may help managers get rid of

concerns about R&D failure, motivate innovation and pursue long-term value (Monks and

Minow 1995). That kind of a monitoring role becomes even more critical as the

incumbent management is weakened (Fung, 2012). Even though financial institutions are

under-developed in emerging countries compared to those well-developed and dispersed

institutional ownership in US listed firms, Rong et al. (2017) state that institutional

investors have a positive impact on innovation performance, which comes mainly from

mutual funds. Thus, the following hypothesis is proposed:

H4. Institutional ownership and the firm’s innovation performance will be positively

related.

2.3.8 Foreign Ownership

2.3.8.1 Introduction of Foreign Ownership

Foreign ownership is defined as a firm owned or controlled by an individual who is not a

citizen of that country or by a firm not headquartered in that country. There are six types

of foreign investments in China: (1) Chinese-Foreign Equity Joint Ventures; (2) Chinese-

Foreign Contractual Joint Ventures; (3) Wholly Foreign-Owned Enterprise; (4)

Share Company With Foreign Investment; (5) Foreign Invested Holding Company; (6)

Joint Exploitation.

Most studies note that foreign ownership is positively correlated to firm performance and

productivity(e.g. Srholec, 2009; Yang and Tsou, 2020; Xu et al., 2022). It is primarily due

to foreign firms having larger equity stakes, higher commitment, and long-term

engagement (Douma et al., 2006). Furthermore, foreign direct investment (FDI)

contributes to domestic firms' productivity and indirectly affects local firms' productivity

in vertically related industries. It is facilitated by technological knowledge, management

practices and governance influences being transferred among foreign and domestic firms,

thereby creating new business linkages. This phenomenon is so-called the spillover

effects of FDI (Singhania et al., 2015).

2.3.8.2 Foreign Ownership and Innovation

Most papers discover a positive relationship between foreign ownership and innovation

performance (Srholec, M., 2009; Choi et al., 2012; Gu et al., 2020). For example, Choi et

al. (2012) investigate 301 Korean listed firms between 2000 and 2003 by using the

negative binomial model with patent data as the dependent variable, foreign ownership as

the independent variable, firm size, firm profitability, sales growth, firm age, leverage,

R&D intensity, business groups, and technology sector as control variables. Srholec

(2009) focuses on 46000 firms in industry and market services from 12

European Union members – Belgium, Bulgaria, Czech Republic, Estonia, Germany,

Latvia, Lithuania, Norway, Portugal, Romania, Slovakia and Spain – between 1998 and

2000 by using the probit regression model with a dummy variable for whether the firm

has innovation activity (e.g., new product innovation, process innovation) and dummy

variables for the firm cooperated with national, foreign partners or both as the dependent

variables, foreign ownership as the independent variable, firm size, a dummy variable for

whether a firm exports to the foreign market, a dummy variable for whether a firm was

established during the sample period, and industry dummies as control variables. The

main reason of the positive impact is the spillover effects of FDI. As previously discussed,

domestic firms can enhance their R&D capabilities by acquiring advanced foreign

knowledge through foreign investment (Singhania et al., 2015). In emerging economies,

knowledge transfer from multinational companies is a vital strategic resource for domestic

firms to improve their performance in technological innovation (Choi et al., 2012). Even

if foreign firms prevent the diffusion of knowledge in order to maintain their competitive

advantage (Jiang et al., 2013), it does not help (Srholec, M., 2009). Moreover, global

resources are another benefit of foreign investment. Many multinational companies often

attempt to strengthen their competitiveness in the global market by finding low-cost or

technologically complementary resources in foreign markets. Resources in markets are

equally important to local firms (Choi et al., 2012). Furthermore, Gu et al. (2020) argue

that foreign banks can enhance information environments and diminish agency problems

for firms, thereby stimulating firm innovation (e.g. increased quantity and quality of

patent applications) within an economy where government intervention is intense, and

investor protection is insufficient.

Dong et al. (2022) show that foreign ownership negative affects the relationship between

innovation and export by considering the sample of Chinese manufacturing firms with a

sample period of 2000-2007. They employ the linear regression with export performance

as the dependent variable, patent counts, patents adjusted by firm size, and the share of

new product sales in total sales as the independent variables, state ownership and foreign

ownership as moderators, firm size, firm age, total factor productivity, leverage,

marketing capability, tangible resources, international openness, marketization, regional

dummies, industry dummies, and time dummies as control variables. This negative

moderating effect is mainly because innovation in multinational companies strongly rely

on parent companies or research centres located in other country. In other words, those

foreign investments from multinational companies to the domestic country are principally

for production not innovation (Dong et al., 2022). 2.3.8.3 Foreign Ownership and

Hypothesis

Spillover effects contribute to domestic firms improving R&D capabilities through

foreign investment (Singhania et al., 2015). Knowledge transfer from multinational

companies is a vital strategic resource for domestic enterprises to improve their

technological innovation performance in emerging economies (Choi et al., 2012). The

diffusion of that kind of knowledge is difficult to restrict, even if foreign companies are

intent on obstructing it (Srholec, M., 2009). In particular, foreign investors can mitigate

agency problems for firms in markets where government intervention is intense and

investor protection is insufficient (Gu et al., 2020). Consequently, the following

hypothesis is proposed:

H5. Foreign ownership and the firm’s innovation performance will be positively

related.

2.4 Depth and Breadth of Knowledge

2.4.1 Introduction

Knowledge is crucial to the survival, growth and innovation of a business (Healey and

Mintz, 2021; Yu and 2021). On the basis of knowledge being scarce, non-tradable and

non-imitable, with no equivalent substitutes, a firm’s knowledge base becomes the

foundation for maintaining its competitive advantage (Grant, 1996). Firm performance

differences depend primarily on their knowledge base (Grant, 1996). In order to maintain

a competitive advantage in a knowledge-rich industry, firms must constantly increase

their knowledge base through substantial investments (Xu and Cavusgil, 2019). The

accumulation of a knowledge base depends on the evaluation of new alternative

knowledge and the assimilation of new knowledge with known knowledge, which

ultimately facilitates the successful creation of new products (Cohen and Levinthal, 1990;

Arora and Gambardella, 1994). So the quality of innovation depends on a firm’s

absorptive capacity (that is, its ability to assess, absorb and use new knowledge)

(Abecassis-Moedas and Mahmoud-Jouini 2008; Tortoriello 2015). Hence, a firm’s

knowledge base is the most valuable asset.

To build up the knowledge base, a firm can accumulate knowledge by exploring new

areas of knowledge (i.e. breadth of knowledge) or deepen understanding and enrich

knowledge by intensifying knowledge in known fields (i.e. depth of knowledge). By

definition, breadth is the wider range of innovation knowledge, and depth is the more

specialised innovation knowledge. The breadth of knowledge is related to innovation

diversity, while the depth of knowledge is related to innovation quality (Lodh and

Battaggion, 2015). Apart from internal R&D, firms can obtain knowledge from strategic

alliances and acquisitions. (Lin and Wu, 2010).

Research on innovation in recent years has placed an in-depth emphasis on the importance

of developing the technical breadth and depth of knowledge (e.g. Lin and

Wu, 2010; Lodh and Battaggion, 2015; Yu and Yan, 2021).

2.4.2 Depth of Knowledge

Conceptually, depth of knowledge encompasses two meanings that possess

complementary characteristics. One is the knowledge base owned by the firm itself, and

the other is the relative competitiveness of the firm's knowledge base against that of its

competitors (Lin and Wu, 2010). With regard to the three ways of acquiring knowledge,

Lin and Wu's (2010) research suggest a focus on internal R&D to build up knowledge in

the company's core areas when the depth of knowledge is low. Conversely, knowledge

acquisition should be shifted to cross-firm alliances and acquisitions when the firm's

depth of knowledge is high.

2.4.2.1 Advantages of Depth of Knowledge

Firstly, the exploration of the depth of knowledge helps firms to effectively identify the

value of new knowledge and improve their absorptive capacity (Yang et al., 2017). As a

result, it is easier to find paths to new product success (Caner and Tyler 2015;

FerrerasMéndez et al. 2015) to identify gaps between its own technology and the latest

technology in the industry (Zahra and George, 2002), and to be more aware of current

market trends (Roberts and Adams 2010; Tsai et al. 2013) than a firm with a weak

knowledge base, for example.

Secondly, if a company focuses on exploring the depth of knowledge, it can assimilate

and integrate external knowledge with a deep knowledge base (Yang et al., 2017). It is

because knowledge from external sources can be substantially different from the firm's

own knowledge, making it difficult to understand and assimilate external knowledge

(Jimenez-Castillo and Sanchez-Perez, 2013).

Thirdly, only those with a deep understanding and know-how can effectively use newly

acquired knowledge, as knowledge is intangible and difficult to document explicitly. A

deep knowledge base contributes to the commercialisation and exploitation of knowledge

(Zhou et al., 2009) and lays the foundation for product and process optimisation and

improvements (Healey and Mintz, 2021).

Fourthly, the depth of knowledge base can be a sustainable competitive advantage for the

firm because of the moat created by asset mass efficiency and time compression

diseconomies (Dierickx and Cool, 1989). Lin and Wu (2010, p.583) define asset mass

efficiency and time compression diseconomies: "Asset mass efficiencies come from the

effect that the more assets a firm has, the lower the marginal cost of producing further

additions to the asset stock. Time compression diseconomies come from the effect that

asset accumulation cannot be rushed." Although a company's knowledge base is an

intangible asset that is not documented on the balance sheet, it is known as a strategic

asset for successful companies.

2.4.2.2 Disadvantages of Depth of Knowledge

Firstly, the marginal benefit of exploring the depth of knowledge is diminishing (Katila

and Ahuja, 2002). If improvements along the technology trajectory reach their limits, the

benefits gained from subsequent product development will increase at a diminishing rate.

Further development based on the same knowledge gets more expensive, and the solution

becomes more complex, finally resulting in the costs of exploring the depth of knowledge

outweighing the benefits.

Secondly, too much focus on the depth of knowledge exploration may not only enable a

firm to be an unbeatable industry leader but also irreparably damage it (Katila and Ahuja,

2002). Knowledge gained from over-exploration of existing knowledge can become

obsolete when disruptive or radical technological breakthroughs emerge in the industry.

Technologies or strategies that once gave a company a competitive lead can become

problems that need to be solved (Leonard, 1995). Argyris and Schon (1978) point out that

companies may try to hide or hinder disruptive or radical innovation in order to maintain

the status quo.

2.4.3 Breadth of Knowledge

The exploration of the breadth of knowledge allows firms to acquire and assimilate

technical information from a variety of sources. In detail, a firm with a wide knowledge

domain is more aware of a diverse customer base and multiple market segments than a

firm with a narrow knowledge domain (De Luca and Atuahene-Gima 2007). In a highly

competitive industry, companies are used to learning and imitating the knowledge of other

companies for their R&D activities. The more complex the knowledge domain becomes,

the more product innovation relies on acquiring external knowledge

(Carayannopoulos and Auster, 2010).

2.4.3.1 Advantages of Breadth of Knowledge

Firstly, the exploration of the breadth of knowledge adds new elements to the firm’s

knowledge domain, and thus brings new inspiration for innovation (Katila and Ahuja,

2002) and enabling the firm to integrate relevant technologies in a more sophisticated way

(Bierly and Chakrabarti, 2009; Srivastava and Gnyawali, 2011).

Secondly, an increase in the knowledge domain increases a firm’s ability to adapt to

technological changes in relevant areas and enhances its flexibility to change its strategy

(Volberda, 1996; Srivastava and Gnyawali, 2011). It addresses the core rigidity problem

posed by knowledge over-exploration (i.e. the second disadvantage of knowledge

exploration discussed in sub-section 2.4.2.2).

2.4.3.2 Disadvantages of Breadth of Knowledge

Firstly, the limited resources of a firm constrain the exploration of the breadth of

knowledge. Excessive pursuit of knowledge breadth may suppress a firm's knowledge

accumulation and development in specific areas (Lausen and Salter, 2006; Zhou and Li,

2012). An increase in knowledge breadth may divert too many firm resources to different

areas, with only a tiny piece of resources allocated to each area. It results in inferior

innovation outcomes (Xu, 2015). Yang et al. (2017) suggest that the impact of the breadth

of knowledge on new product performance depends on the depth of knowledge. If the

depth of knowledge is high, the breadth of knowledge will have a positive impact on the

new product and vice versa.

Secondly, the cost and complexity of integrating new knowledge become higher as the

breadth of knowledge expands (Grant，1996).

2.4.4 Hypotheses for Innovation Specialisation and Innovation Diversification

2.4.4.1 Concentrated Ownership

If the level of concentrated ownership is high, the majority shareholder pursues a

diversification strategy to reduce the risk taken and to achieve personal goals rather than

value maximisation. When shareholder rights are not adequately protected, concentrated

ownership and diversification exhibit a quadratic U-shape, unlike the linear correlation

observed in legally sound markets (Del Brio et al., 2011). However, Parigi and Pelizzon

(2008) argue that when large shareholders are able to transfer profits, they will not be

interested in diversifying their investments and instead focus back on repurchasing firm

shares, and the firm's ownership structure becomes concentrated. However, a less

diversified investment also means less potential for innovative diversification. In China,

listed firms are characterised by a concentration of shareholdings, insufficient protection

of minority shareholders' rights, and tunnelling accordingly (Wang et al., 2020).

Therefore, the following hypothesis is proposed:

H6a: The higher the level of concentrated ownership, the higher the level of

innovation specialisation.

H6b: The higher the level of concentrated ownership, the lower the level of innovation

diversification.

2.4.4.2 Insider Ownership

Chen and Ho (2000) point out that a low level of insider ownership gives rise to agency

problems, which means that the interests of managers are not aligned with those of

insiders. It causes significant value loss due to diversity, whereas it is not found at a high

level of insider ownership.

Nonetheless, insider ownership contributes to diversification due to 1) agents seeking to

increase their reputation and value in the company (Shleifer and Vishny, 1989) and 2) the

second generation of family-owned firms having more resources to diversify than the

founders of the first generation of firms (Weng and Chi, 2019).

H7a: The higher the level of insider ownership, the lower the level of innovation

specialisation.

H7b: The higher the level of insider ownership, the higher the level of diversity of

innovation diversification.

2.4.4.3 State Ownership

In general, SOEs are more likely to favour diversification than non-SOEs (Guthrie, 1997;

Li et al., 1998). The first reason is that SOEs are more likely to have access to political

resources and financial support, which can contribute to the successful implementation of

a diversification strategy (Lu and Yao, 2006). The second point is that diversification

weakens the negative impact of the external market environment on the company (Lee

and Hooy, 2018) or helps the firm establish the resources and capabilities that are the

firm's core competencies (Keister, 1998). The third point is that the government generally

appoints the managers of SOEs with a high level of state ownership, and the political

connections of the managers also positively influence diversification to a large extent (Li

et al., 2012). Fourthly, managers appointed by the government do not have enough

incentives and professional knowledge to effectively supervise firm development or

pursue profit maximisation. In this situation, the diversification strategy allows that kind

of manager to pursue personal interests or other non-profit objectives (such as seeking a

large firm size) (Delios et al., 2008). Thus, the following hypothesis is proposed:

H8a: The higher the level of state ownership, the lower the level of innovation

specialisation.

H8b: The higher the level of state ownership, the higher the level of innovation

diversification.

2.4.4.4 Institutional Ownership

Institutional ownership has a positive impact on diversification for the following reasons:

1) long-term stable institutional investors (Jafarinejad et al., 2015); and 2) effective

monitoring mechanisms, especially active monitors, can mitigate diversification discounts

(Singh et al., 2004; Hartzell et al., 2014). Moreover, the lower the diversification discount,

the more willing institutional investors are to have extensive research, increase profits,

and remain competitive in an increasingly globalised market. Thus, the following

hypothesis is proposed:

H9a: The higher the level of institutional ownership, the lower the level of innovation

specialisation.

H9b: The higher the level of institutional ownership, the higher the level of innovation

diversification.

2.4.4.5 Foreign Ownership

On the one hand, foreign ownership plays a positive role in innovation specialisation. In

particular, the foreign parent company, through the rational deployment of resources,

allows different subsidiaries to innovate and specialise in a particular area, and ultimately

the subsidiary takes a firm foothold in the industry's value chain (Collinson and Wang,

2012).

One the other hand, firms face localisation issues when growing their business in other

countries (Hitt et al., 2000). Foreign firms are at a disadvantage compared to local firms

in terms of local resources, suppliers and end sellers (Gaur and Kumar, 2009).

Especially in the service sector, business operations rely heavily on social networks Yang

et al. (2017). As an example, they point out that foreign hotels in China are not as familiar

with Chinese business practices and their social networks are not as strong as local hotels,

resulting in higher transaction or operational costs associated with diversification. As a

result, firms are reluctant to undertake diversification when the level of foreign ownership

is high. The lower the level of foreign ownership, the more willing the company is to

diversify (Yang et al., 2017). This is primarily due to the local advantage. Hence, the

following hypothesis is suggested:

H10a: The higher the level of foreign ownership, the higher the level of innovation

specialisation.

H10b: The higher the level of foreign ownership, the lower the level of innovation

diversification.

2.5 Fresh Insights from Previous Literature

The previous sub-sectors have reviewed relevant theories and empirical literature about

the impact of firm ownership on innovation performance. First, this thesis examines the

relationship between firm ownership and innovation performance based on previous

literature (e.g., Chi, 2023), but with an updated sample period of 2013 – 2019. This

sample period is mainly because the year 2013 is the beginning of the fifth round of SOE

reform (Lin et al., 2020). Second, Mishra (2022) focuses on the threshold of institutional

ownership and concludes that intuitional ownership below or above the threshold is

positively or negatively related to firm innovation, respectively. This thesis then extends

the idea of threshold to all five ownerships. Compared to many previous literature

measures ownership structures (independent variable) as a continuous variable or a binary

variable for whether the firm belongs to a particular firm ownership (e.g., Chen et al.,

2013; Lo et al., 2022), firm ownership is a categorical variable with different levels of

percentage of firm ownership. Thus, it will show the different impacts of firm ownership

on innovation performance based on different levels of firm ownership. Third, Bong Choi

and Williams (2013) propose that a firm performs better in Korea as technology

innovation is specialised, while diversified technological innovation produces better firm

performance in China. Based on the concepts of innovation specialisation and innovation

diversification, this thesis concentrates on how firm ownership affects firm innovation

specialisation or innovation diversification.

2.6 Summary

This chapter reviews the literature with regard to the concepts of innovation and firm

ownership and considers the importance of knowledge for innovation. The impact of firm

ownership on innovation performance varies with political contexts, market

environments, corporate environments, internal and external resources, et cetera. The role

of knowledge in the innovative growth of firms comes from two aspects – depth of

knowledge and breadth of knowledge. Also, this section presents hypotheses related to the

relationship between firm ownership and innovation performance, firm ownership and

innovation specialisation, and firm ownership and innovation diversification, based

primarily on the literature review.

In sum, hypotheses on firm ownership and innovation performance will be tested first

with continuous and categorical ownership structures (i.e., first and the second

contribution) as follows:

(1) H1. Concentrated ownership and the firm’s innovation performance will be

negatively related.

(2) H2. Insider ownership and the firm’s innovation performance will be negatively

related.

(3) H3. State ownership and the firm’s innovation performance will be positively

related.

(4) H4. Institutional ownership and the firm’s innovation performance will be

positively related.

(5) H5. Foreign ownership and the firm’s innovation performance will be positively

related.

Then, five ownership structures will be tested against the depth of innovation and

diversity of innovation (i.e., the third contribution), respectively, as follows: H6a:

The higher the level of concentrated ownership, the higher the level of innovation

specialisation.

H6b: The higher the level of concentrated ownership, the lower the level of innovation

diversification.

H7a: The higher the level of insider ownership, the lower the level of innovation

specialisation.

H7b: The higher the level of insider ownership, the higher the level of diversity of

innovation diversification.

H8a: The higher the level of state ownership, the lower the level of innovation

specialisation.

H8b: The higher the level of state ownership, the higher the level of innovation

diversification.

H9a: The higher the level of institutional ownership, the lower the level of innovation

specialisation.

H9b: The higher the level of institutional ownership, the higher the level of innovation

diversification.

H10a: The higher the level of foreign ownership, the higher the level of innovation

specialisation.

H10b: The higher the level of foreign ownership, the lower the level of innovation

diversification.

Chapter 3 Data and Methodology

3.1 Introduction

Section 3.2 is about the sample and data. Sector 3.3 discusses all variables used in the

models. Section 3.4 contains the research models, methods for mitigating endogeneity,

and robustness tests.

3.2 Sample and Data

To practically evaluate the relationship between firm ownership and innovation

performance in China, this thesis collects data from all Chinese firms listed on the

Shanghai Stock Exchange between 2013 and 2019. This sample period is mainly because

the year 2013 is the beginning of the mixed ownership reform. Financial data and the data

on ownership structures are collected from firm annual reports and China Securities

Market. Patent data are obtained from China's State Intellectual Property Office.

However, some firms which do not have R&D expenditures or patent data are removed.

Besides, this thesis is interested in the non-financial firms only. Hence, it leaves 1409

firms and 8518 firm-year observations.

3.3 Variables

3.3.1 Dependent Variables

Dependent variable is innovation performance, which is difficult to quantify and compare,

yet it is crucial to choose appropriate measurement since reliable information about

innovation activities and outcomes is required for data analysis (Michie, 1998).

The conclusion might not be the same if different measures of innovation had been used.

Unfortunately, there are no widely accepted measures because of the various

dimensionality of innovation (Kline and Rosenberg, 1986; Smith, 2005). Kuznets (1962)

classifies the measurement of innovation into two groups: (1) measures of input;

(2) measures of output.

3.3.1.1 Measures of Input

For the investigation of the relationship between firm ownership and innovation, R&D

intensity is a widely acceptable indicator of innovation input (e.g., Abbas et al., 2022; Di

Vito et al., 2010; Kurt et al., 2015; Lacetera, 2001), as input measurement measures the

amount of innovation drive business invests in. In this thesis, R&D intensity is defined as

the ratio of R&D expenditure to the firm’s total sales. Since R&D investment is crucial

for firms to pile up greater technological and market capabilities and subsequently

increase innovation performance, R&D expenditure is a reasonable indicator of

innovation investment (Matzler et al., 2015). Moreover, R&D intensity is not only able to

control size effects and heteroscedasticity, but also easy to compare different firms’

innovation performance (Barker and Mueller, 2002; Chen and Hsu,

2009). Other popular R&D measurements contain R&D expenditure (Chen et al., 2013;

Minetti et al., 2015; Steffen and Iuliia, 2016), R&D expenditure per employee

(Baysinger et al., 1991), and R&D personnel (Minetti et al., 2015).

3.3.1.1.1 Advantages of R&D measurement

R&D data is publicly available and is regularly collected by the State. In addition to

sector-specific data, time series are also available. National researchers usually use the

data for cross-national, cross-sector and cross-firm comparisons (Kleinknecht et al.,

2002).

3.3.1.1.2 Disadvantages of R&D measurement

Aghion and Tirole (1994) argue that assessment of R&D activities may give a too limited

view of innovation, since the R&D has been represented as different actors with conflict

purposes in existing economic literature. Secondly, R&D is an innovation input that may

not be an essential factor for producing innovative products or processes, thereby

overestimating innovation intensity (Flor and Oltra, 2004; Becheikh et al., 2006). Thirdly,

not all innovations are generated from research laboratories, which means they can be

solutions to a specific challenge or ideas innovators suddenly got. As a result, R&D data

underestimates innovation activities, which excludes those that emerged from non-R&D

investments (Michie, 1998; Kleinknecht et al., 2002; Edquist and Zabala, 2015). Fourthly,

large firms have an advantage over small and medium enterprises (SMEs), because SMEs

tend to have informal and occasional R&D efforts compared to large companies, thereby

underestimating innovation intensity

(Kleinknecht, 1987; Kleinknecht et al., 2002).

3.3.1.2 Measures of Output

For the investigation of the relationship between firm ownership and innovation, patent

data measured by ln (1 + 'ℎ) *+,-). /0 12')*'3), due to the possibility of

highly skewed patent counts (e.g., Chen et al., 2022; Chi et al., 2019; Wan et al., 2021).

Output measurement measures the effectiveness of innovative investments in achieving

the desired results. Patent is a document issued by a sovereign state and granted to an

inventor if the invented product or process is a novelty and has potential utility. It

excludes rights for others except the inventor to use this product or process for a limited

period (Comanor, 1964). Patent data is usually applied to measure innovation output

(Griliches, 1990) since it is a proper proxy for innovation outcomes (Kamien and

Schwartz, 1982).

For the investigation of the relationship between firm ownership and innovation

specialisation, the depth of innovation (Depth) is measured by Herfindahl-Hirschman

Index (HHI). HHI is usually to measure the market concentration. The higher the HHI, the

higher the market concentration (Kvålseth, 2018). In this thesis, the depth of innovation

measured by HHI is to find out the concentration rate of patent in one category for a firm.

In this case, State Intellectual Property Office (SIPO) in China typically classifies patents

into eight groups (i.e. A – H), as shown in the following table.

Table 3.1 International Patent Classification (IPC)

As a result, the depth of innovation is calculated by HHI based on patent data:

5)1' :!!;#

(3.1) "

where

*% is the number of patents in the ith section of IPC for a firm; N is

the total number of patents for a firm.

The higher the HHI, the more concentrated the types of patents.

For the investigation of the relationship between firm ownership and innovation

diversification, the diversity of innovation (Diversity) is measured by the entropy. The

lower the entropy coefficient, the greater the diversified innovation. As we can see,

entropy is the other measurement of concentration. The reason for not using 1 minus HHI

to measure the diversity of innovation is that HHI has already been applied to measure the

depth of innovation, and hence, the statistical inferences of the diversity of innovation are

converse to the ones of the depth of innovation.

The Shannon entropy coefficient as an alternative measurement of concentration is

(Shannon, 1948):

(=*A1(B (3.2)

where 1( is the percentage of type j.

The properties of entropy coefficient are as follows (Hart, 1971; Bandt, 2020):

•<5 ≥ 0: ED is 0 when 1( = 1 (i.e., all products are concentrated in one type),

because 1 × log(1) = 0. It means entropy coefficient is 0 when there is high

concentration. Also, ED is 0 when 1( = 0 , becuase 0 × log(0) = 0 .

Consequently, ED is non-negative.

•: ED is ln (I) when 1( = "' for all j, which means the number of

products is equal in all types.

In order to avoid the ED of 0 if all products are in one type and use it for measuring

diversification, the entropy coefficient based on patent data is modified by exponential

function (Jost, 2006):

*(S

5JK).3J'L = M<5 = exp QR=* T UV

(3.3)

(&' S *(

where

*( is the number of patents in the jth section of IPC in Table 3.1 for a firm;

S is the total number of patents for a firm.

The higher the value of MED, the higher the diversification.

3.3.1.2.1 Advantages of Patent Data

Firstly, the patent database is rich in data and time-honoured. Secondly, the patent

database is open to the public and electronically categorised by field of technology,

allowing easy access to detailed information on patents and citation analysis to assess

their relative importance (Kleinknecht et al., 2002).

3.3.1.2.2 Disadvantages of Patent Data

The first shortcoming is that the patent system is not the same across countries. Second,

some innovations may not be patentable, especially for software (Griliches, 1990;

Michie, 1998). Third, the tendency to patent differs among sectors (Griliches, 1990;

Archibugi and Sirilli, 2001). The reasons can be high patenting expenses, complicated

patenting procedures, or the quick diffusion process of new technology (Mansfield, 1985;

Michie, 1998). Hence, some enterprises apply other proper approaches to protect their

innovation outcomes, for example, industrial secrecy (Archibugi and Pianta, 1996;

Michie, 1998; Kleinknecht et al., 2002). Fourth, patent quality varies significantly

(Kamien and Schwartz, 1982; Griliches, 1990; Griffith et al., 2006). Fifth, patent data is

more like a measure of invention instead of innovation (Coombs et al., 1996; OECD,

1997; Flor and Oltra, 2004). Becheikh et al. (2006) explain that innovation is the

transformation of an invention into a merchantable new or improved product or process.

Hence, measuring the number of patents granted may lead to an overestimation of

innovation outcome, if invented products or processes which have not been marketable

are included in patent data.

3.3.1.3 R&D Intensity vs Patent Data

Both R&D intensity and patent data are used as the dependent variable in the models. It is

because innovation covers, but is not limited to, research and development. R&D is only

the first step in the innovation process. Since innovation is hard to quantify, R&D and

patent data are used to measure and compare the innovation, giving investors and

researchers an intuitive sense of innovation performance. In other words, R&D intensity

is innovation input and patent data is innovation output. Nevertheless, Acharya and Xu

(2017) argue that the patent data can be more effectively reflect the actual innovation

output for a firm than R&D measurements. One reason is that the patents are listed on

China's State Intellectual Property Office when the firms have applied for them, and thus,

the number of patents for a firm each year is calculated based on the patent application

year. It is because the year of application is closer to when the innovation was made

(Griliches, 1990). Furthermore, innovation input has a higher likelihood of endogenous

problem than innovation output (Leten et al., 2007; Van de Vrande et al., 2011).

Additionally, innovation inputs (e.g., R&D intensity) do not always generate innovation

outputs (e.g., patents) (Dong et al., 2022; Tavassoli, 2018).

3.3.2 Independent Variables

To test Hypotheses 1 - 10, this thesis portrays the specific features of the firm's ownership

structure. The shares of Chinese listed firms are differentiated into A-shares, B-shares, H-

shares, N-shares and S-shares, due to where the shares are listed and the investors they are

exposed to. A-shares are issued by companies registered in China for subscription and

trading in RMB by domestic institutions, organizations or individuals (excluding Taiwan,

Hong Kong and Macau investors). B-shares are also issued in China by companies

registered in China, but unlike A-shares, they are issued with a nominal value in RMB but

subscribed and traded in foreign currencies, and listed and traded on Chinese stock

exchanges. H-shares, N-shares and S-shares refer to foreign stocks registered in mainland

China, but listed in Hong Kong, New York and Singapore respectively (Li et al., 2006;

Arslan-Ayaydin et al., 2022.).

By the specific characteristics of the firm's ownership structure in China, ownership

structures are measured in the following table. The reason to contain all types of shares

(i.e., A-shares, B-shares, H-shares, N-shares and S-shares) is to take individual investors

and organisations from foreign countries (Choi et al., 2011).

Table 3.2 Independent Variables

Independent

Variables

Description Literature Hypothesis

Concentrated

Ownership

Sum of squared of firm

shares owned by top 5 large

shareholders

Kvålseth (2018) H1, H6a and H6b

State Ownership Proportion of firm shares

owned by all levels of

government, its related

agencies and solely

stateowned enterprise in

top 10 largest shareholders

Chen et al. (2022);

Chi et al. (2019)

H3, H8a and H8b

Insider

Ownership

Proportion of firm shares

owned by managers,

directors, supervisory board

members and workers in

top 10 largest

shareholders

Chang et al. (2006);

Choi et al. (2011)

H2, H7a and H7b

Institutional

Ownership

Proportion of firm shares

owned by financial

institutions in top 10 largest

shareholders

Choi et al. (2011);

David et al. (2006)

H4, H9a and H9b

Foreign

Ownership

Proportion of firm shares

owned by foreign

corporation and

institutional investors (from

different types of shares

except A Share) in top 10

largest shareholders

Choi et al. (2011) H5, H10a and

H10b

3.3.3 Control Variables

There are several variables controlled in the thesis. Firm size is measured by total assets

of the firm to proxy for how firm size affects its innovation (e.g., Miller et al., 2022;

Minetti et al., 2015; Wan et al., 2021). Firm age is measured by the number of years that a

firm was established to account for the effect of a firm’s life cycle on firm innovation

(e.g., Chen et al., 2023; Miller et al., 2022; Decker and Günther, 2017). Knowledge stock

is measured by the total number of accumulated patents during the period from the year of

establishment to year 1 to capture how the number of patents accumulated affects

innovation ability for a firm (e.g., Choi et al., 2011; Decker and Günther, 2017; Lodh et

al., 2014). Leverage is estimated by the ratio of total debt to total assets (debt ratio) to

take the impact of capital structure on innovation into account (e.g., Steffen and Iuliia,

2016; Wan et al., 2021). Profitability is measured by the ratio of net income and total

assets (named as return on assets) to account for operating profitability (e.g., Choi et al.,

2011; Lo et al., 2022). R&D intensity is measured by ratio of R&D expenditure to the

firm’s total sales to take innovation input into account for innovation output (e.g., Steffen

and Iuliia, 2016), and hence, this control variable is only available for the patent data as

the dependent variable. Industry dummy variables in the following table distinguish

different sectors that firms belong to (e.g., Di Vito et al.,

2010; Lee, 2005; Lodh et al., 2014; Wan et al., 2021).

Table 3.3 Dummy Variables

Dummy Variable

Code

Dummy Variable Description Number of

Firms

1 Agriculture, forestry, animal husbandry and fishery 14

2 Mining industry 53

3 Manufacturing industry 1440

4 Electricity, heat, gas and water production and supply

industry

5 construction industry 55

6 Wholesale and retail trade 103

7 Transportation, warehousing and postal industry 80

8 Accommodation and catering industry 5

9 Information transmission, software and information

technology services

148

10 Real estate 61

11 Leasing and business services 21

Continued

12 Scientific research and technical service industry 40

13 Water conservancy, environment and public facilities

management industry

14 Education industry 5

15 Health and social work 2

16 Culture, sports and entertainment industry 31

17 Comprehensive industry 7

3.4 Research Model

3.4.1 Firm ownership and Innovation performance

If Y%* is the R&D intensity of firm i at time t and Z+%* is the kth explanatory variable of

firm i at time t, the model can be described as the following equation to test Hypotheses

H1-H5:

[&5 J*')*3J'L%* = ] + ^'/_*).3ℎJ1%,*-' + ^#ln (3J`))%,*-' + ^.2a)%,*-' +

^/b*/_c)da)%,*-' + ^0c)K).2a)%,*-' + ^1J*d+3'.L%,*-' + ^2e./0J'2-JcJ'L%,*-'

^$c*(J*K)3')%,*-' + f%* (3.4)

e2')*'%* = ] + ^'/_*).3ℎJ1%,*-' + ^#ln(3J`))%,*-' + ^.2a)%,*-' +

^/b*/_c)da)%,*-' + ^0c)K).2a)%,*-' + ^1J*d+3'.L%,*-' + ^2e./0J'2-JcJ'L%,*-'

^$c*(J*K)3')%,*-' + [&5 J*')*3J'L%,*-' + f%* (3.5)

where

•/_*).3ℎJ1%* is firm ownership for firm i at time t.

•ln (3J`))%* is log of firm size for firm i at time t. Since the value of firm size

is too large compared to other values of variables. For example, firm size of

Dongfeng Motor Corporation in 2013 was too big (i.e. ¥ 20191845033.17), but

the relatively concentrated ownership was too small (i.e. 61% in 2013). Then, the

estimated parameters ] and ^ from the model looks inharmonious. The estimated

parameters for firm size can be too small compared to others.

Consequently, the logarithm is used, that is ln (gJ`)).

•2a)%* is firm age for firm i at time t.

•b*/_c)da)%* is knowledge stock for firm i at time t.

•c)K).2a)%* is the total debt and firm size ratio for firm i at time t.

•J*d+3'.L%* is dummy variables for firm i at time t.

•e./0J'2-JcJ'L%* is profitability for firm i at time t.

•c*(J*K)3')%* is log of long-term investment. The reason of taking natural

logarithm is discussed above in ln (3J`))%*.

Since innovation input (i.e., R&D intensity) and innovation output (i.e., patent) are

applied in this thesis to examine the relationship between firm ownership and innovation,

equations (3.4) and (3.5) are both for the research question (1) in Chapter 1. For research

questions 1 and 2 in Chapter 1, both continuous and categorical variables for firm

ownership are used in equations (3.4) and (3.5). Most studies employ continuous variable

for firm ownership (e.g., Chen et al., 2022), but it may generate a problem if the

relationship between firm ownership and innovation is actually nonlinear or only has a

significant influence when the percentage of firm ownership is above or below the

threshold (e.g., Mishra, 2022). Hence, for the second contribution in this thesis, a

categorical variable of firm ownership is used to find out the threshold.

Firm ownership is grouped by:

•Zero Level: the percentage of firm ownership = 0%

•Low Level: 0% < the percentage of firm ownership < 5%

•Medium Level: 5% ＜ the percentage of firm ownership < 20%

•High Level: 20% ＜ the percentage of firm ownership < 100%

Two reasons for choosing 5% and 20% to be thresholds are as follows:

1) Appendix A displays histograms of five ownership structures. All five histograms

indicate a positive skewness, because the data is more often piled up below 20%.

Hence, 20% is taken as a threshold.

2) Listed Company Takeover Measures announced by the China Securities

Regulatory Commission (CSRC) mention that once an investor's shares reach 5%

of the issued shares of a listed company, each increase or decrease in shareholding

needs to be reported and announced to the CSRC. Therefore, 5% is taken as a

threshold.

In particular, concentrated ownership is measured by HHI, so that it is a non-zero

variable. Hence, only three levels of concentrated ownership exist – low level, medium

level, and high level. As opposed to other ownership structures, there are only three levels

of concentrated ownership.

3.4.2 Firm Ownership and Innovation Specialisation

The third research question is to investigate the relationship between the depth of

innovation and firm ownership at time t. The regression model is as follows to test

Hypotheses H6a – H6a:

5)1'ℎ%* = ] + ^'/_*).3ℎJ1%,*-' + ^#[&5 J*')*3J'L%,*-' + ^.ln (3J`))%,*-' +

^/2a)%,*-' + ^0b*/_c)da)%,*-' + ^1c)K).2a)%,*-' + ^2e./0J'2-JcJ'L%,*-' +

^$ln (J*K)3')%,*-' + [&5 J*')*3J'L%,*-' + f%* (3.6)

One thing that needs to be confirmed is that firm ownership (ownership) is a categorical

variable in this case, and the levels of firm ownership for this categorical variable are

shown in sub-sector 3.4.1. Apart from the concentrated ownership, there are four levels of

firm ownership – zero level, low level, medium level, and high level. The categorical

variable of concentrated ownership only has three levels – low, medium, and high. Also,

HHI measures the depth of innovation as discussed in sub-sector 3.3.1.2. Innovation is

more specialised as the level of firm ownership is higher than the baseline.

3.4.3 Firm ownership and Innovation Diversity

The fourth research question in Chapter 1 is to investigate the relationship between the

diversity of innovation and firm ownership at time t. The regression model is as follows to

test Hypotheses H6b - H10b:

5JK).3J'L%* = ] + ^'/_*).3ℎJ1%,*-' + ^#[&5 J*')*3J'L%,*-' + ^.ln (3J`))%,*-' +

^/2a)%,*-' + ^0b*/_c)da)%,*-' + ^1c)K).2a)%,*-' + ^2e./0J'2-JcJ'L%,*-' +

^$ln (J*K)3')%,*-' + [&5 J*')*3J'L%,*-' + f%* (3.7)

Again, ownership is a categorical variable as discussed in sub-sector 3.4.1. The diversity

of innovation is calculated by entropy coefficient as discussed in subsector 3.3.1.2.

Based on the equation 3.4, the higher the value of diversity is, the higher degree of

diversification is.

3.4.4 Controlling for Endogeneity

Since panel regression is used in this thesis, Hausman Specification (HS) test is applied to

determine fixed effects model or random effects model is preferred (Hausman, 1978). In

practice, endogeneity is a common problem for researchers (e.g., De Silva, 2023, Gao et

al., 2019; Tang et al., 2022), and may be caused by omitted variables, such as government

policies that facilitate firm innovation, R&D investment persistence and reverse

causality (Chi et al., 2019; Kang et al., 2017; Gao et al., 2019; Mishra, 2022). Lack of

recognition and treatment of endogeneity can lead to inconsistent and biased estimated

coefficients, incorrect interpretations, or even erroneous findings (Bascle, 2008). To

eliminate endogeneity in the regression models, this thesis lags firm ownership and

control variable by one period, referring to Chen et al. (2022), Gao and

Zheng (2020), Li et al. (2021) and Mishra (2022).

3.4.5 Robustness Tests

Robustness tests are performed to confirm the stability of the estimated coefficients in the

regression models, which would have changed if this had not been done. There are four

regression models (i.e., equations (3.4) – (3.7)) to test all hypotheses, and hence, there are

four robustness tests. First, the change in R&D intensity, using the difference in R&D

intensity between the current year and last year divided by R&D intensity in the last year

as the dependent variable, can be employed to test the relationship between firm

ownership and innovation performance (equation (3.4)). Second, the change in the

number of patents for a firm, which is estimated by the ratio of the difference in patent

counts between the current and last year to patent counts in the last year as the dependent

variable, can be used to test the relationship between firm ownership and innovation

performance (equation (3.5)) (Chi et al., 2019). Third, Shannon entropy shown in sub-

sector 3.3.1.2 is one of the measurements for concentration (Shannon, 1948), and hence,

can proxy for HHI to measure innovation specialisation so as to check the robustness of

the estimated coefficients in the equation (3.6). Fourth, 1h667 can be an alternative

measurement for the innovation diversification (Jost, 2006), as

HHI is a measurement for innovation specialisation as discussed in sub-sector 3.3.1.2.

3.5 Summary of Research Methodology

This chapter discusses sample and data, all variables, and research models. Endogeneity

may arise from empirical work and can be solved by lagged ownership and lagged control

variables. Besides, robustness tests make sure the robustly estimated coefficients in the

regression models.

Chapter 4 Empirical Investigation into Ownership -

Innovation Relationships

4.1 Introduction

This chapter performs empirical inquiries into the relationships between ownership

structure and innovation performance. It reports the empirical results from modelling the

relationships by testing the hypotheses developed in chapter 3. Specifically, it investigates

the effects of firm ownership on innovation performance. It goes further to examine the

impacts of firm performance on the depth of innovation and the diversification of

innovation. Empirical findings are summarised and discussed with implications for

management and research. In the following, section 4.2 reports descriptive statistics,

offering a general outlook of the sample companies in ownership structure and innovation

activities. Whereas section 4.3 present, analyse and discuss the modelling results.

4.2 Descriptive Statistics and Summary Data

4.2.1 Introduction of Descriptive Statistics and Summary Data

This thesis analyses data from firms listed on the Shanghai Stock Exchange between 2013

and 2019. The data were obtained from firm annual reports and Patent Search and

Analysis of State Intellectual Property Office. There are totally 1409 firms.

4.2.2 Descriptive Statistics and Summary Data of Variables

Table 4.1 expresses Descriptive Statistics and Summary Data for independent variables

and dependent variables. Independent variables contain five ownership structures.

The mean of state ownership is 24.58%, and the median is 16.28%. 75% of total

observations is about 47%, which is a substantially high percentage compared to other

firm ownership. It is almost twice as high as the second-highest third quartile (i.e.

concentrated ownership). Besides, state ownership has the highest mean and median

among the five ownership structures, which discloses the ongoing dominance of SOEs in

China.

The mean of insider ownership is 7.38%, and the median is 0.01%. The third quartile is

2.12%, which reveals a large number of firms in China without insider ownership or only

with a small piece of insider ownership.

The mean of foreign ownership is 4.70%, and the median is 0%. The upper quartile of

foreign ownership (1.06%) is even less than insider ownership. The fewest firms with

foreign ownership as opposed to other corporate ownership.

The mean of institutional ownership is 13.53%, and the median is 6.87%. 75% of data

points have a level less than 20%.

The mean of concentrated ownership is 18.10%, and the median is 15.26%. The upper

quartile is 25.29%, which exhibits that Chinese firms are highly concentrated to some

extent.

There are three dependent variables - R&D intensity, depth of innovation and diversity of

innovation.

The mean of R&D intensity is just 2.55%, and the median is 1.31%. The third quartile is

as small as 3.71%. The third quartile is as small as 3.71%, which signifies a low degree of

investment in R&D by listed companies.

The mean of the depth of innovation is 0.61, and the median is 0.54. The degree of

innovation specialisation ranges from 0 to 1, with the third quartile at 0.81 - a

considerable proportion of firms focus on a single area of R&D.

The mean of the diversity of innovation is 1.79, and the median is 1.69. The upper quartile

is 2.05.

Table 4.1 Descriptive Statistics and Summary Data for Independent and Dependent Variables

Mean Standard

Deviation

Minimum Maximum 25% 50% 75%

State Ownership 24.58% 25.72% 0% 95.26% 0% 16.28% 47.42%

Insider

Ownership

7.38% 16.48% 0% 89.99% 0% 0.01% 2.12%

Foreign

Ownership

4.70% 12.05% 0% 88.55% 0% 0% 1.06%

Institutional

Ownership

13.53% 16.89% 0% 90.99% 2.05% 6.87% 17.76%

Concentrated

Ownership

18.10% 12.93% 0.002% 79.42% 8.24% 15.26% 25.29%

R&D Intensity 2.55% 4.37% 0 169.43% 0.0004% 1.31% 3.71%

Depth of

Innovation

0.61 0.24 0.19 1 0.41 0.54 0.81

Diversity of

Innovation

1.70 0.46 1 2.76 1.37 1.69 2.05

4.2.3 Comparison of Means

As explained in sub-section 4.4.8, the firm ownership is assigned to four levels:

•Zero Level: the percentage of firm ownership = 0%

•Low Level: 0% < the percentage of firm ownership ≤ 5%

•Medium Level: 5% ＜ the percentage of firm ownership ≤ 20%

•High Level: 20% ＜ the percentage of firm ownership ≤ 100%

Table 4.2 indicates the means of R&D intensity, depth of innovation, and diversity of

innovation for each level of firm ownership.

Table 4.2 Means of Independent and Dependent Variables with Different Levels of Ownerships Structure

R&D

Intensity

Depth of

Innovation

Diversity of

Innovation

State

Ownership

Zero (0%) 0.0344 0.6472 1.6202

Low

(0% <

Own

ership

0.0317 0.6191 1.6724

(5% <

Ownership

≤20%)

0.0254 0.6294 1.6517

High

(≥20%) 0.0180 0.5643

1.7890

1.6972

Insider

Ownership

Zero (0%) 0.0170 0.6081

Low

(0% <

Ownership

0.0224 0.5830 1.7544

(5% <

Ownership

≤20%)

0.0418 0.6269 1.6571

High

(≥20%) 0.0468 0.6437

1.6224

1.6785

Foreign

Ownership

Zero (0%) 0.0257 0.6168

Low

(0% <

Own

ership

0.0266 0.5771 1.7662

(5% <

Ownership

≤20%)

0.0241 0.5965 1.7334

High

(≥20%)

0.0236 0.5964

1.7257

1.6672

Institutional

Ownership

Zero (0%) 0.0304 0.6204

Low

(0% <

0.0246 0.5822 1.7545

Own

ership

(5% <

Ownership

≤20%)

0.0272 0.6111 1.6907

High

(≥20%)

0.0222 0.6341 1.6494

Continued

Concentrated

Ownership

Zero (0%) - - -

Low

(0% <

Ownership

≤5%))

0.0292 0.6432 1.6249

Medium

(5% <

Ownership

≤20%)

0.0276 0.6203 1.6727

High

(≥20%)

0.0213 0.5784 1.7626

a) Comparison of Means for State Ownership

The mean of R&D intensity in Table 4.2 decreases as the level of state ownership

increases.

Non-state ownership has the highest value of innovation specialisation (i.e.

0.6472). The medium level comes next (i.e. 0.6294), followed by the low level

(i.e. 0.6191). The high level produces the least value for innovation specialisation

(i.e. 0.5643)

Conversely, the high level ranks first in innovation diversification (i.e. 1.7890).

The low level has the second largest value of innovation diversification (i.e.

1.6724), followed by the medium level (i.e. 1.6517). The lowest value of

innovation diversification is zero level (i.e. 1.6202). In terms of the mean value of

innovation diversification alone, SOEs are more diversely innovative than non-

SOEs.

b) Comparison of Means for Insider Ownership

Regarding insider ownership, the mean of R&D intensity increases as the level of

insider ownership increases in Table 4.2.

Apart from the zero-level, the higher the insider ownership, the more

specialisation in innovation. However, insiders at a low level are reluctant to

specialise in innovation, as opposed to non-insider ownership.

In turn, except for the zero level, the higher the insider ownership, the less innovative

diversification.

c) Comparison of Means for Foreign Ownership

The highest mean value of R&D intensity in Table 4.2 is the low-level foreign

ownership (i.e. 0.0266), followed by the zero-level (i.e. 0.0257). The third is the

medium-level (i.e. 0.0241), and the fourth is the high-level (i.e. 0.0236). At low

levels, foreign ownership recorded the highest mean in terms of innovation

diversification (i.e. 1.7662). The next highest is the medium level (i.e. 1.7334),

followed by the high level (i.e. 1.7257). Finally, the fourth is the zero level (i.e.

1.6785).

It suggests that firms prefer innovation diversification with foreign ownership rather

than those without foreign ownership.

d) Comparison of Means for Institutional Ownership

Non-institutional ownership ranks first in R&D intensity (i.e. 0.0304) in Table 4.2.

The medium level of institutional ownership has the second highest R&D intensity

(i.e. 0.0272). The next is the low level (i.e. 0.0246). The last one is the high-level (i.e.

0.0222). Interestingly, non-institutional investors are more willing to innovate than

institutional investors.

Excluding the zero-level, the higher the institutional ownership, the more the

innovation specialisation is preferred. Unless institutional ownership is above 20%,

non-institutional investors are more active in innovation than institutional investors.

Except for the zero-level, the higher the institutional ownership is, the lower the

innovation diversification is. A high-level institutional ownership is less devoted to

innovative diversity than non-institutional ownership

e) Comparison of Means for Concentrated Ownership

Since the HHI index measures concentrated ownership, it is a non-zero

measurement. Therefore, there is no value for R&D intensity, innovation

specialisation, and innovation diversification at zero level of concentrated

ownership in Table 4.2. Only three levels of concentrated ownership are available,

unlike other firm ownership.

The higher the level of concentrated ownership, the lower the R&D intensity, on

average.

The depth of innovation increases as the level of concentrated ownership

decreases.

In contrast, the greater the concentration of ownership, the greater the willingness

of firms to diversify in innovation.

4.2.4 Comparison of Means by ANOVA

Table 4.3 shows that all p-values from the ANOVA tests are less than 5%, thus rejecting

the null hypothesis at the 5% level. Hence, the means for all four levels of R&D intensity,

depth of innovation and diversity of innovation are significantly different, excluding

concentrated ownership. Moreover, three levels of concentrated ownership have also

significantly different means for R&D intensity, depth of innovation and diversity of

innovation.

Table 4.3 P-values from ANOVA

R&D Intensity Depth of Innovation Diversity of Innovation

State Ownership 0.0000 0.0000 0.0000

Insider Ownership 0.0000 0.0000 0.0000

Foreign Ownership 0.0000 0.0015 0.0001

Institutional Ownership 0.0000 0.0000 0.0000

Concentrated Ownership 0.0000 0.0000 0.0000

4.2.5 Comparison of Means by Post-Hoc Analysis

In order to know how three/four levels of ownership structures differ in means of three

variables (i.e. R&D intensity, the depth of innovation, and the diversity of innovation),

Post-Hoc analysis was applied with Tukey-Kramer because sample sizes for each level of

five ownership structures are wildly different as shown in Appendix B. In addition to

concentrated ownership, there are six pairwise comparisons for the other four firm

ownership (i.e. Low vs Zero, Medium vs Zero, High vs Zero, Medium vs Low, High vs

Low and High vs Medium). However, due to non-zero values of concentrated ownership,

it only has three pairwise comparisons (i.e. Medium vs Low, High vs Low and High vs

Medium).

The p-values of pairwise comparisons are indicated in Appendix C. Table 4.4 exposes

which pairwise comparison has significantly different means of three variables at the

5% level. 'Yes' indicates a significant difference, whereas 'No' is the opposite.

Table 4.4 Pairwise Comparisons (Significantly Different Means or Not)

R&D

Intensity

Depth of

Innovation

Diversity of

Innovation

State Ownership Low - Zero No No No

Medium -

Zero

Yes No No

High - Zero Yes Yes Yes

Medium –

Low

No No No

High – Low Yes Yes Yes

High -

Medium

Yes Yes Yes

Insider Ownership Low - Zero Yes Yes Yes

Medium -

Zero

Yes No No

High - Zero Yes Yes Yes

Medium –

Low

Yes Yes Yes

High – Low Yes Yes Yes

High -

Medium

No No No

Foreign Ownership Low - Zero No Yes Yes

Medium -

Zero

No No No

High - Zero No No No

Medium – No No No

Low

High – Low No No No

High -

Medium

No No No

Institutional

Ownership

Low - Zero Yes No Yes

Medium -

Zero

No No No

High - Zero Yes No No

Continued

Medium –

Low

No Yes Yes

High – Low No Yes Yes

High -

Medium

Yes No No

Concentrated

Ownership

Medium -

Low

No No No

High - Low No Yes Yes

High -

Medium

No Yes Yes

a) Post-Hoc Analysis for State Ownership

Table 4.4 demonstrates a significant difference in the mean R&D intensity of those

four pairwise comparisons: medium and zero, high and zero, high and low, and

high and medium. In addition, other variables have significantly different means

in those three pairwise comparisons: high and zero, high and low, and high and

medium. It discloses that the means of all variables for the high-level state

ownership are significantly different from the other levels.

The pairwise comparisons ' Low vs Zero' and 'Medium vs Low' do not exhibit any

significant difference in R&D intensity, innovation specialisation and

diversification.

b) Post-Hoc Analysis for Insider Ownership

For R&D intensity, only the comparison between the high-level and the medium-

level insider ownership says 'No' in Table 4.4, while all other pairwise

comparisons say 'Yes'. Accordingly, the means of R&D intensity for insider

ownership are significantly different from non-insider ownership, regardless of the

level of insider ownership.

The other two variables - the depth of innovation and the diversity of innovation

- have significantly different means in the following comparisons: (1) 'Low and

Zero'; (2) 'High and Zero'; and (3) 'Medium and Low'.

c) Post-Hoc Analysis for Foreign Ownership

None of the pairwise comparisons in Table 4.4 displays any difference in the mean

of R&D intensity for foreign ownership. Of the six pairwise comparisons, only the

low-level and the zero-level differed significantly in the mean values of

innovation specialisation and diversification.

d) Post-Hoc Analysis for Institutional Ownership

For institutional ownership, there is a significant difference in the mean R&D

intensity of those three pairwise comparisons in Table 4.4: low and zero, high and

zero, and high and medium. Moreover, both innovation specialisation and

diversification significantly differ in the means between medium and low levels as

well as high and low levels.

e) Post-Hoc Analysis for Concentrated Ownership

A significant difference in the means of all variables, except R&D intensity, was

found in comparing the high level of concentrated ownership to the other two

levels. On the other hand, R&D intensity does not differ significantly for the three

levels of concentrated ownership in Table 4.4.

4.2.6 Summary of Descriptive Statistics and Summary Data

This section compares the various levels of firm ownership across the three variables

involving R&D intensity, innovation specialisation and innovation diversification.

4.3 Empirical Results and Analysis

4.3.1 Introduction

This section focuses on testing the hypothesis tests presented in Chapter 4.

Before proceeding with the regression analysis, a boxplot for R&D intensity is drawn.

Figure 4.1 illustrates two influential outliers of interest in particular, which have been

circled in red. Upon searching the data, both outliers are from a firm called Caihong

Display Devices (hereinafter referred to as Caihong), caused by a spike in R&D intensity

between 2014 and 2015. So as not to let two outliers affect the regression model, the data

from Caihong will be first wiped out.

4.3.2 R&D Intensity

4.3.2.1 Concentrated Ownership

4.3.2.1.1 Model 1: Continuous Concentrated Ownership

i. Tests for Assumptions of Panel Data Regression

Figure 4.1 Boxplot for R&D Intensity

The LM test was applied to determine which type of regression to use. As the p-value for

the LM test was p <0.01, the null hypothesis of no panel effect was rejected by it at the

5% level. Accordingly, the panel data regression was better than the pooled regression.

Given that the p-value for the HS test was p <0.01, it rejected the null hypothesis of a

random effect at the 5% level. A panel data regression with fixed effects was, therefore, an

appropriate model.

Then, tests of the regression assumptions were then carried out to ensure that the

coefficients were inconsistent or biased.

First, the mean of residuals was 0, which meant that the assumption of linearity did not

violate (Assumption 1). Second, the assumption of homoscedasticity violated as the

modified Wald test had a p-value of p < 0.01 (Assumption 2). Third, the p-value for the

run test of randomness was p <0.01, which rejected the null hypothesis (Assumption 3).

In other words, residuals were autocorrelated. Fourth, the p-value of the Ramsey

RESET Test was p <0.01, which rejected the null hypothesis (Assumption 4).

Accordingly, misspecification was not a concern in this case. Fifth, the total sample size

was large enough to suggest that residual distribution was asymptotically normal

(Assumption 5). Sixth, the pairwise correlation matrix in Appendix D reveals that

multicollinearity is not a problem as the correlation is still low. Hence, the assumption of

no perfect multicollinearity held (Assumption 6).

In order to correct the violations of the assumptions, the fixed effects model was

reregressed by the robust standard error.

ii. Results

Table 4.5 Regression with Continuous Concentrated Ownership

(1)

Fixed Effects

VARIABLES Robust t-statistics p-values

Concentrated 0.0127* 1.7700 0.0770

(0.0072)

Firm Size -0.0024** -2.3700 0.0180

(0.0010)

Firm Age 0.0016*** 7.6600 0.0000

(0.0002)

Knowledge Stock 0.0000 1.1400 0.2530

(0.0000)

Leverage 0.0005 0.5100 0.6070

(0.0009)

Continued

Public A-shares -0.0069 0.2620 -0.0190

(0.0062)

Long-term Investment 0.0006** 0.0200 0.0001

(0.0003)

Industry 2 0.0059 0.6490 -0.0194

(0.0129)

Industry 3 0.0168** 0.0410 0.0007

(0.0082)

Industry 4 0.0021 0.8080 -0.0147

(0.0085)

Industry 5 -0.0016 0.9110 -0.0294

(0.0142)

Industry 6 -0.0047*** 0.0000 -0.0057

(0.0005)

Industry 7 0.0112 0.2810 -0.0092

(0.0104)

Industry 8 0.0119 0.3130 -0.0113

(0.0118)

Industry 9 0.0442** 0.0240 0.0059

(0.0196)

Industry 10 0.0126 0.1140 -0.0030

(0.0080)

Industry 11 0.0041 0.6390 -0.0131

(0.0088)

Industry 12 0.0112 0.3370 -0.0117

(0.0117)

3 0.0332* 0.0550 -0.0007

(0.0173)

Industry 14 0.0009 0.9400 -0.0214

(0.0113)

Industry 15 0.1320*** 0.0000 0.1179

(0.0072)

Industry 16 0.0588*** 0.0000 0.0394

(0.0099)

Industry 17 0.0082 0.5690 -0.0200

(0.0144)

Industry 18 0.0179 0.1560 -0.0069

(0.0126)

Constant 0.0249 0.2390 -0.0166

(0.0212)

Observations 8,286

Number of Firms 1,409

Adjusted R-squared 0.0806

FIRM FE YES

Log-likelihood 23946

INDUSTRY FE YES

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The table above signifies the panel data regression outputs.

For the independent variables, the p-value of concentrated ownership is p < 0.1, which

rejects the null hypothesis at the 10% level. Concentrated ownership and R&D intensity

are significantly and positively related at the 10% level. In other words, R&D intensity is

expected to increase by 0.0127 as the level of concentrated ownership increases by one

unit, while holding other things equal. It then is at odds with the hypothesis (i.e. H1).

For control variables, the p-value of ln(firm size) is p < 0.05, which rejects the null

hypothesis at the 5% level. Ln(firm size) and R&D intensity are significantly and

negatively related at the 5% level. If firm size increases by 1%, the mean of R&D

intensity decreases by 2.4e-5, ceteris paribus. The p-value of firm age is p < 0.01, which

rejects the null hypothesis. Firm age is significantly and positively related to R&D

intensity. The expected change in R&D intensity is 0.0016 for an additional one year

increase in firm age, while holding other variables constant. The p-value of ln(long-term

investment) is p < 0.05, rejecting the null hypothesis at the 5% level. If long-term

investment increases by 1%, the mean of R&D intensity increases by 6.322e-6, ceteris

paribus.

The baseline for industry dummy is ‘agriculture, forestry, animal husbandry and fishery’,

which is denoted as industry code 1. The category of all dummy variables’ codes and

descriptions are in Table 4.5. The p-value of industry code 3 is p < 0.05, which rejects the

null hypothesis at the 5% level. On average, the R&D intensity of the manufacturing

industry is 0.0168 higher than the baseline, ceteris paribus. The p-value of industry code 6

is p < 0.01, which rejects the null hypothesis at the 1% level. The mean R&D intensity of

wholesale and retail trade is 0.0047 lower than the baseline, ceteris paribus. The p-value

of industry code 9 is p < 0.05, rejecting the null hypothesis at the 5% level. The mean

R&D intensity of information transmission, software and information technology services

is 0.0442 higher than the baseline, ceteris paribus. The p-value of industry code 13 is p <

0.1, which rejects the null hypothesis at the 10% level. The mean R&D intensity of

scientific research and technical service industry is 0.0332 higher than the baseline,

ceteris paribus. Both industry codes 15 and 16 have pvalues of p < 0.01, which rejects the

null hypothesis. On average, the education industry and health and social work have an

R&D intensity of 0.1320 and 0.0588 higher than the baseline, respectively, ceteris

paribus.

In addition, the log-likelihood value is 23946.

4.3.2.1.2 Model 2: Categorical Concentrated Ownership

i. Tests for Assumptions of Panel Data Regression

Applying the LM test was to determine which type of regression to use. Due to the pvalue

of p < 0.01 for the LM test, it rejected the null hypothesis of no panel effect at the 5%

level. Panel data regression was, therefore, better than pooled regression. Due to the p-

value of the HS test being p < 0.01, the null hypothesis of a random effect was rejected at

the 5% level. Consequently, a panel data regression with fixed effects was an appropriate

model.

The regression assumptions were then tested to ensure that the coefficients were not

inconsistent or biased.

First, the mean of residuals was 0, implying the parameters' linearity (Assumption 1).

Second, the assumption of homoscedasticity violated by the modified Wald test as the p-

value was p < 0.01 (Assumption 2). Third, the p-value of the run test of randomness was p

< 0.01, which rejected the null hypothesis (Assumption 3). It meant that there was a serial

correlation. Fourth, the p-value of the Ramsey RESET Test was p < 0.01, which rejected

the null hypothesis of no omitted variable at the 1% level (Assumption 4). As a

consequence, there was no misspecification error. Fifth, residual distribution was

asymptotically normal because of a large sample size (Assumption 5). Sixth, Appendix D

indicates the pairwise correlation matrix, showing that correlations keep low. Then, the

assumption of no perfect multicollinearity held.

To sum up, the fixed effects model with the robust standard error was applied thereafter

because of the violations of assumptions.

ii. Results

Table 4.6 Regression with Categorical Concentrated Ownership

(2)

Fixed Effects

VARIABLES Robust t-statistics p-values

Medium -0.0015 -0.7700 0.4400

(0.0019)

High -0.0004 -0.2000 0.8430

(0.0021)

Firm Size -0.0025** -2.5100 0.0120

(0.0010)

Firm Age 0.0015*** 7.7400 0.0000

(0.0002)

Knowledge Stock 0.0000 1.0600 0.2900

(0.0000)

Leverage 0.0005 0.5200 0.6000

(0.0009)

Public A-shares -0.0024 -0.4500 0.6560

(0.0053)

Long-term Investment 0.0006** 2.3200 0.0210

(0.0003)

Industry 2 0.0066 0.5100 0.6130

(0.0131)

Industry 3 0.0173** 2.1200 0.0340

(0.0082)

Industry 4 0.0030 0.3600 0.7220

(0.0085)

Industry 5 -0.0007 -0.0500 0.9600

(0.0141)

Industry 6 -0.0045*** -8.4600 0.0000

(0.0005)

Industry 7 0.0114 1.0900 0.2760

(0.0105)

Industry 8 0.0113 0.9600 0.3390

(0.0118)

Industry 9 0.0448** 2.2900 0.0220

(0.0196)

Industry 10 0.0124 1.5200 0.1280

(0.0081)

Industry 11 0.0045 0.5100 0.6080

(0.0088)

Continued

Industry 12 0.0115 0.9800 0.3270

(0.0118)

Industry 13 0.0325* 1.8200 0.0690

(0.0179)

Industry 14 0.0018 0.1600 0.8720

(0.0113)

Industry 15 0.1320*** 17.7200 0.0000

(0.0074)

Industry 16 0.0593*** 5.9400 0.0000

(0.0100)

Industry 17 0.0084 0.5900 0.5570

(0.0143)

Industry 18 0.0185 1.4700 0.1420

(0.0126)

Constant 0.0281 1.3200 0.1860

(0.0212)

Observations 8,286

Number of Firms 1,409

Adjusted R-squared 0.0802

FIRM FE YES

INDUSTRY FE YES

Log-likelihood 23944

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The baseline is the firm with low level of concentrated ownership.

From the table above, the p-values of medium and high levels are both greater than 0.1,

respectively, so null hypotheses cannot be rejected. R&D intensity is not significantly

different among the three levels of concentrated ownership on average while holding

other variables constant.

For control variables, the p-value of ln(firm size) is p < 0.05, which rejects the null

hypothesis at the 5% level. Ln(firm size) and R&D intensity are significantly and

negatively related. The mean of R&D intensity decreases by 2.5e-5 if firm size increases

by 1%, ceteris paribus. The p-value of firm age is p < 0.01 which rejects the null

hypothesis at the 1% level. Firm age and R&D intensity are positively and significantly

related. The expected change in R&D intensity is p < 0.01 for an additional one year

increase in firm age, ceteris paribus. The p-value of ln(long-term investment)

is p < 0.05, which rejects the null hypothesis at the 5% level. If long-term investment

increases by 1%, the mean of R&D intensity increases by 6e-6, ceteris paribus. For

industry dummies, the p-value of industry code 3 is p < 0.05, which rejects the null

hypothesis at the 5% level. On average, the R&D intensity of the manufacturing industry

is 0.0173 higher than the one of agriculture, forestry, animal husbandry and fishery,

ceteris paribus. The p-value of industry code 6 is p < 0.01, which rejects the null

hypothesis. The mean R&D intensity of wholesale and retail trade is 0.0045 lower than

the baseline, ceteris paribus. The p-value of industry code 9 is p < 0.05, which rejects the

null hypothesis at the 5% level. The mean R&D intensity of information transmission,

software and information technology services is 0.0448 higher than the baseline, ceteris

paribus. The p-value of industry code 13 is less than 0.1, which rejects the null hypothesis

at the 10% level. The mean R&D intensity of scientific research and technical service

industry is 0.0325 higher than the baseline, ceteris paribus. The pvalues of industry codes

15 and 16 are both p < 0.01, so those two null hypotheses can be rejected. The means of

R&D intensity in the education industry and health and social work are 0.1320 and 0.0593

higher than the base level, respectively, ceteris paribus. Except variables mentioned

above, all other variables’ p-values are greater than 0.1, which cannot reject the null

hypotheses.

The log-likelihood value is 23944.

4.3.2.1.3 Discussions for Concentrated Ownership with R&D Intensity

i. Likelihood Ratio Test

Log-likelihood in model (1) with continuous concentred ownership (i.e. 23946) is

marginally higher than the one in model (2) with categorical concentred ownership (i.e.

23944). The p-value of the likelihood ratio test is greater than 0.1, which fails to reject the

null hypothesis. Accordingly, model (2) is not better than model (1).

ii. Continuous Concentrated Ownership

In Table 4.5, concentrated ownership has a significantly positive impact on R&D

intensity. As a consequence, concentrated ownership and a firm's innovation performance

are positively related, yet it is not consistent with the hypothesis in Chapter 3 suggesting

the negative correlation (H1).

The reason for that kind of positive impact could be that a high degree of concentration

generates an efficient monitoring mechanism that lowers agency costs and disciplines

managers' behaviour (Baysinger et al., 1991; Cho, 1992; Francis and Smith, 1995;

Holmstrom, 1989; Nguyen et al., 2015). Specifically, large shareholders may prefer

innovation strategies with high risk, even if the probability of success is low. Conversely,

managers from firms with a low degree of concentration may be more willing to take

imitation strategies with low risk, because they do not want to undertake the cost of

failure (Hill and Snell, 1988).

Another advantage of concentrated ownership is that large shareholders can optimise and

integrate all resources internally or externally to increase the probability of a successful

R&D project (Lacetera, 2001). In other words, larger shareholders enhance R&D

investment.

Furthermore, firms with concentrated ownership may be more interested in long-term

development in a business field rather than short-term profit maximisation (Mayer, 1997;

Miozzo and Dewick, 2002). On the one hand, those firms are willing to occupy a leading

position or even monopolise the market. Apart from political reasons, innovative

technologies beyond the times may help those firms to achieve their goals. Accordingly,

R&D investment plays a crucial role in this case. On the other hand, a few large

shareholders are more willing to invest in a long-term R&D project than to maximise the

firm profit in the short term, because they believe innovation's success may improve the

firm's survival in the market (Chang et al., 2006; Chang and Hong, 2000; Mahmood and

Mitchell, 2004; Rowley and Bae, 2004; Shleifer and Vishny, 1996). In the manager's

view, those few large shareholders' aims may relieve their managerial pressures to

maximise short-term profits, thereby boosting R&D investment

(Baysinger et al., 1991).

iii. Categorical Concentrated Ownership

R&D intensity in concentrated ownership does not vary across the three levels, in line

with the findings in Table 4.4. Thus, it also does not agree with hypothesis H1.

iv. Control Variables

It is not surprisingly that firm age has a significantly positive effect on innovation

performance, even if that impact is trivial. In view of the fact that old firms can benefit

from their business experience and foresight (Arrow, 1962; Sorensen and Stuart, 2000;

Chang et al., 2002), and enhancing R&D investment to improve the degree of firm growth

persistence and let firms survive in an increasingly competitive market.

In addition, the negative impact of firm size on innovation performance is too small to be

negligible. However, firms with concentrated ownership may be satisfied with firm scale

and are not willing to undertake the cost of R&D failure, if firm size upswings

dramatically.

Similarly, the positive impact of long-term investment on innovation performance is also

negligibly small. The performance of firm with concentrated ownership to increase their

R&D intensity is more evident if there is substantial long-term investment.

Regarding industry dummies, the baseline is agriculture, forestry, animal husbandry and

fishery. On average, firms with concentrated ownership within those five industries – (1)

manufacturing industry; (2) information transmission, software and information

technology services; (3) scientific research and technical service industry; (4) education

industry; and (5) health and social work - have a significantly higher R&D intensity

than the baseline. It may be that technological innovation in each of these sectors requires

considerable time, effort, capital, etc.

4.3.2.2 Insider Concentration

4.3.2.2.1 Model 1: Continuous Insider Ownership

i. Tests for Assumptions of Panel Data Regression

To begin with, the LM test was applied to define the type of regression to be utilised. The

LM test had a p-value of p < 0.01, given that it rejected the null hypothesis of no panel

effects at the 1% level. Panel data regression was, therefore, superior to pooled regression.

The HS test had a p-value of p < 0.01 as it rejected the null hypothesis of random effects

at the 1% level. Therefore, the panel data regression with fixed effects was an appropriate

model.

The regression assumptions were tested afterwards for consistent and unbiased

coefficients.

First, the mean of residuals was 0, which indicated that the assumption of linearity in

parameters holds (Assumption 1). Second, the assumption of homoscedasticity violated

due to the p-value of p < 0.01 by the modified Wald test (Assumption 2). Third, the pvalue

of the run test of randomness was p < 0.01, which rejected the null hypothesis

(Assumption 3). To put it differently, there was autocorrelation. Fourth, the p-value of the

Ramsey RESET Test was p < 0.01, which rejected the null hypothesis of no omitted

variable (Assumption 4). As a consequence, there was a misspecification error. Fifth,

residual distribution was asymptotically normal because of a large sample size

(Assumption 5). Sixth, the pairwise correlation matrix in Appendix D reveals that

multicollinearity is not a problem as the correlation is still low. Hence, the assumption of

no perfect multicollinearity held (Assumption 6).

To sum up, the fixed effects model with the robust standard error was applied thereafter to

fix the problems of violations of assumptions.

ii. Results

Table 4.7 Regression with Continuous Insider Ownership

(3)

Fixed Effects

VARIABLES Robust t-statistics p-values

insider 0.0133 0.9600 0.3380

(0.0138)

Firm Size -0.0025** -2.3900 0.0170

(0.0011)

Firm Age 0.0015*** 7.6800 0.0000

(0.0002)

Knowledge Stock 0.0000 1.0200 0.3070

(0.0000)

Leverage 0.0007 0.5700 0.5670

(0.0012)

Public A-shares -0.0037 -0.6400 0.5220

(0.0057)

Long-term Investment 0.0006** 2.0800 0.0370

(0.0003)

Industry 2 0.0071 0.5500 0.5840

(0.0129)

Industry 3 0.0169** 2.1200 0.0340

(0.0080)

Industry 4 0.0028 0.3400 0.7310

(0.0083)

Industry 5 -0.0011 -0.0700 0.9410

(0.0141)

Industry 6 -0.0044*** -8.1300 0.0000

(0.0005)

Industry 7 0.0109 1.0800 0.2790

(0.0101)

Industry 8 0.0105 0.9100 0.3630

(0.0116)

Industry 9 0.0441** 2.3300 0.0200

(0.0189)

Industry 10 0.0127 1.6200 0.1040

(0.0078)

Industry 11 0.0044 0.5200 0.6060

(0.0085)

Industry 12 0.0113 0.9900 0.3220

(0.0115)

Industry 13 0.0302* 1.8200 0.0690

(0.0166)

Industry 14 0.0013 0.1200 0.9070

(0.0110)

Industry 15 0.1322*** 19.1700 0.0000

(0.0069)

Industry 16 0.0586*** 6.0400 0.0000

(0.0097)

Continued

Industry 17 0.0080 0.5600 0.5770

(0.0143)

Industry 18 0.0182 1.4800 0.1380

(0.0123)

Constant 0.0287 1.2800 0.1990

(0.0224)

Observations 8,284

Number of Firms 1,409

Adjusted R-squared 0.0823

FIRM FE YES

Log-likelihood 23959

INDUSTRY FE YES

For the independent variables, the p-value of insider ownership in Table 4.9 is greater than

0.1, which fails to reject the null hypothesis at the 10% level. Insider ownership and R&D

intensity are insignificantly and positively related. It is in disagreement with the

hypothesis (i.e. H2).

For control variables, the p-value of ln(firm size) is p < 0.05, which rejects the null

hypothesis at the 5% level. Ln(firm size) and R&D intensity are significantly and

negatively related at the 5% level. If firm size increases by 1%, the mean of R&D

intensity decreases by 2.5e-5, ceteris paribus. It is a really minor impact. The p-value of

firm age is p < 0.01, which rejects the null hypothesis. Firm age is significantly and

positively related to R&D intensity. The expected change in R&D intensity is 0.0015 for

an additional one unit increased in firm age, while holding other variables constant. The

p-value of ln(long-term investment) is p < 0.05, which rejects the null hypothesis at the

5% level. If long-term investment increases by 1%, the mean of R&D intensity increases

by 6e-6, ceteris paribus.

For industry dummies, the p-value of industry code 3 is p < 0.05, which rejects the null

hypothesis at the 5% level. On average, the R&D intensity of the manufacturing industry

is 0.0169 higher than the one of agriculture, forestry, animal husbandry and fishery,

ceteris paribus. The p-value of industry code 6 is p < 0.01, which rejects the null

hypothesis. The mean R&D intensity of wholesale and retail trade is 0.0044 lower than

the baseline, ceteris paribus. The p-value of industry code 9 is 0.020, which rejects the

null hypothesis at the 5% level. The mean R&D intensity of information transmission,

software and information technology services is 0.0441 higher than the baseline, ceteris

paribus. The p-value of industry code 13 is p < 0.1, which rejects the null hypothesis at

the 10% level. The mean R&D intensity of scientific research and technical service

industry is 0.0302 higher than the baseline, ceteris paribus. The pvalues of industry codes

15 and 16 are both p < 0.01, so those two null hypotheses can be rejected. The means of

R&D intensity in the education industry and health and social work are 0.1322 and 0.0586

higher than the base level, respectively, ceteris paribus. Except variables mentioned

above, all other variables’ p-values are greater than 0.1, which cannot reject the null

hypotheses.

The log-likelihood value is 23959.

4.3.2.2.2 Model 2: Categorical Insider Ownership

i. Tests for Assumptions of Panel Data Regression

The LM test was applied to see the type of regression to be used. As the p-value for the

LM test is p < 0.01, it rejected the null hypothesis of no panel effect at the 1% level. Panel

data regression was, therefore, better than pooled regression. Given that the pvalue for the

HS test is p < 0.01, it rejected the null hypothesis of a random effect at the

1% level. Consequently, a panel data regression with fixed effects was a suitable model.

Regression assumptions were tested afterwards to ensure coefficients were consistent and

not biased.

First, the mean of residuals is 0, which implied the linearity in parameters (Assumption 1).

Second, the assumption of homoscedasticity violated by modified Wald test as the pvalue

was p < 0.01 (Assumption 2). Third, the p-value of run test of randomness was p < 0.01,

which rejected the null hypothesis (Assumption 3). It meant that there was serial

correlation. Fourth, the p-value of Ramsey RESET Test was p < 0.01, which rejected

the null hypothesis of no omit variable at the 5% level (Assumption 4). As a

consequence, there was a misspecification error. Fifth, residual distribution was

asymptotically normal because of large sample size (Assumption 5). Sixth, the

correlation matrix is display in Appendix E, which suggests that there is no violation of

the assumption of no perfect multicollinearity, due to low correlations (Assumption 6).

To sum up, the fixed effects with robust standard error was applied thereafter, because of

the violations of assumptions.

ii. Results

Table 4.8 Regression with Categorical Insider Ownership

(4)

Fixed Effects

VARIABLES Robust t-statistics p-values

Low 0.0025*** 2.8200 0.0050

(0.0009)

Medium 0.0017 0.6900 0.4900

(0.0022)

High 0.0103* 1.8100 0.0710

(0.0053)

Firm Size -0.0025** -2.4500 0.0140

(0.0010)

Firm Age 0.0015*** 7.5400 0.0000

(0.0002)

Knowledge Stock 0.0000 1.0600 0.2880

(0.0000)

Leverage 0.0005 0.5700 0.5670

(0.0010)

Public A-shares -0.0030 -0.5000 0.6170

(0.0056)

Long-term Investment 0.0006** 1.9800 0.0480

(0.0003)

Industry 2 0.0081 0.6500 0.5140

(0.0126)

Industry 3 0.0187** 2.3400 0.0200

(0.0080)

Industry 4 0.0048 0.6000 0.5490

(0.0082)

Industry 5 0.0012 0.0800 0.9380

(0.0139)

Industry 6 -0.0021* -2.0700 0.0380

(0.0011)

Industry 7 0.0129 1.2500 0.2100

Continued

(0.0102)

Industry 8 0.0108 0.9500 0.3400

(0.0114)

Industry 9 0.0456** 2.4400 0.0150

(0.0186)

Industry 10 0.0147* 1.8500 0.0650

(0.0079)

Industry 11 0.0065 0.7500 0.4540

(0.0086)

Industry 12 0.0135 1.1700 0.2420

(0.0115)

Industry 13 0.0335* 1.8500 0.0640

(0.0180)

Industry 14 0.0028 0.2600 0.7940

(0.0108)

Industry 15 0.1328*** 19.4900 0.0000

(0.0068)

Industry 16 0.0605*** 6.1500 0.0000

(0.0098)

Industry 17 0.0096 0.6900 0.4880

(0.0141)

Industry 18 0.0200 1.6200 0.1060

(0.0124)

Constant 0.0253 1.2500 0.2110

(0.0213)

Observations 8,286

Number of Firms 1,409

Adjusted R-squared 0.0846

FIRM FE YES

INDUSTRY FE YES

Log-likelihood 23965

The baseline for the independent variable is the firm without any insider ownership (i.e.

zero-level).

From the table above, the p-value of the low level is p < 0.01, which rejects the null

hypothesis at the 1% level. On average, R&D intensity is 0.0025 higher for lower levels

of insider ownership than for non-insider ownership, ceteris paribus. The p-value of the

medium level is greater than 0.1, which fails to reject the null hypothesis. The expected

R&D intensity between the medium-level and the zero-level is indifferent. The p-value of

the high level is p < 0.1, which rejects the null hypothesis at the 10% level. Highlevel

insider ownership benefits 0.0103 more on R&D intensity than non-insider ownership,

ceteris paribus.

For control variables, the p-value of ln(firm size) is p < 0.05, which rejects the null

hypothesis at the 5% level. Ln(firm size) and R&D intensity are significantly and

negatively related. The mean of R&D intensity decreases by 2.5e-5 if firm size increases

by 1%, ceteris paribus. The p-value of firm age is 0 p < 0.01, which rejects the null

hypothesis. Firm age and R&D intensity are positively and significantly related. The

expected change in R&D intensity is 0.0015 for an additional one year increase in firm

age, ceteris paribus. The p-value for long-term investment is p < 0.05, rejecting the null

hypothesis. Long-term investment and R&D intensity are positively and significantly

correlated. Other things being equal, a one-unit increase in long-term investment results in

an expected change in R&D intensity of 6e-5.

For industry dummies, the p-value for industry code 3 is p < 0.05, rejecting the null

hypothesis at the 5% level. On average, R&D intensity in manufacturing is 0.0187 higher

than that in agriculture, forestry, animal husbandry and fishing, ceteris paribus. A p-value

of p < 0.05 for industry code 6 rejects the null hypothesis. While holding other variables

constant, the average R&D intensity in wholesale and retail trade is 0.0021 below the

baseline. Industry code 9 has a p-value of p < 0.05, rejecting the null hypothesis at the 5%

level. The average R&D intensity for information transmission, software and information

technology services is 0.0456 above the baseline, ceteris paribus. The p-value for industry

code 13 is p < 0.1, rejecting the null hypothesis at the 10% level. Scientific research and

technology services have an average R&D intensity of 0.0335 above the baseline, ceteris

paribus. Industry 15 and 16 have a p-value of p < 0.01, so the null hypothesis is rejected

for these two. The mean values for R&D intensity regarding the education sector and

health and social work are 0.1328 and 0.0605 above the baseline level, respectively,

ceteris paribus. The null hypothesis

cannot be rejected with p-values greater than 0.1 for all variables except those mentioned

above.

The log-likelihood value was 23975.

4.3.2.2.3 Discussions for Insider Ownership with R&D Intensity

i. Likelihood Ratio Test

The log-likelihood value in Table 4.7 is 23959, which is smaller than the one in Table 4.8

(i.e. 23975). The p-value of the likelihood ratio test is close to 0, which rejects the null

hypothesis. Accordingly, the model with categorical insider ownership is better than the

model with continuous insider ownership.

ii. Continuous Insider Ownership

Table 4.9 demonstrates no relationship between insider ownership and R&D intensity, and

this finding does not agree with the previous hypothesis (H2).

The non-relationship between R&D intensity and insider ownership may be because that

insider ownership usually accounts for a small part of the total share capital in China.

Approximately 33% of firms have no insider ownership, 45% have less than 5%, and

7.5% have insider ownership between 5% and 20% (i.e. Appendix B). The mean insider

ownership in Table 4.1 is as small as 7.38%. In short, insider ownership is less developed

in China.

Secondly, in order to ensure that future position is secure, managers are more likely to put

more effort into improving public relations than enhancing the firm performance (Bisot

and Child, 1996; Xin and Pearce, 1996; Peng, 2000).

Thirdly, equity compensation for managers and boards of directors is less common in

China (Choi et al., 2011).

Table 4.10 indicate that insider ownership is more beneficial to R&D intensity than non-

insider ownership at all levels except the medium level. Insider ownership is more

beneficial to innovation performance when it is greater than 20% or less than 5%,

compared to firms without insider ownership. Further, firms with more than 20% insider

ownership have an advantage over firms with a proportion of less than 5% when it comes

to R&D intensity. Again, it disagrees the hypothesis H2.

The results partly agree with the agency theory that a rise in insider ownership causes

managers' interests to align with the shareholders' interests, owing to the efficient

monitoring scheme. It then favours innovation, as effective monitoring mechanisms allow

managers to mitigate concerns about the consequences of R&D failure on their careers.

Insiders with small shareholdings may be employees who have been rewarded with

shares. So, equity incentives also have a positive effect on corporate innovation in China.

Moreover, employees are more willing to innovate in order to achieve self-worth and have

stable employment (Chang et al., 2006).

Another reason is that the firm owner may focus on innovation in order to find longterm

competitiveness of products or services for a successful continuation of the firm in the

hands of future generations (Caselli and Gennaioli, 2013).

The fourth reason is that agency theory is not appropriate, as it advocates that insider

ownership can lower managerial pressures to maximise short-term values, thereby

enhancing R&D investment (Choi et al., 2012). Alternatively, the transaction cost of

economy (TCE) examines the relationship between insider ownership and innovation

performance more precisely than agency theory (Suk et al., 2012). TCE offers an

explanation of why companies exist, scale up or farm out activities to an external

environment. Firms aim to optimise the unnecessary transition costs incurred in the

exchange of resources within the business environment. A variety of factors -

opportunism, limited rationality, environmental uncertainty, limited information and asset

specificity - are noted by the theory as affecting the extent of transaction costs

(Williamson, 1965). These factors are all expected to worsen the firms' innovation

performance potentially. Compared to developed countries, emerging countries can be

seen as underdeveloped markets with uncertain business environments and a lack of or

expensive resources for innovation (Suk et al., 2012). In an uncertain business

environment, firms can safeguard themselves through political connections. It also brings

more policy-related information to firms, lowering transaction costs and promoting

technological innovation. In a family business, insider ownership may be beneficial for

communication between top management as a way of consolidating decision-making

power and thus saving transaction costs. Such situations allow firms to be sensitive to

changes in the external business environment and to allocate firm resources efficiently to

achieve innovative performance (Poza et al.,1997; Tagiuri and Davis, 1996).

In short, even though agency theory suggests that insider ownership contributes to firms’

innovation performance, the fact is that only 14.1% of observations have insider

ownership over 20%. A side reflection of the high-level insider ownership in Chinese

listed firms is relatively rare.

iii. Categorical Insider Ownership

Table 4.10 indicate that insider ownership is more beneficial to R&D intensity than non-

insider ownership at all levels except the medium level. Insider ownership is more

beneficial to innovation performance when it is greater than 20% or less than 5%,

compared to firms without insider ownership. Further, firms with more than 20% insider

ownership have an advantage over firms with a proportion of less than 5% when it comes

to R&D intensity. Again, it disagrees the hypothesis H2.

The results partly agree with the agency theory that a rise in insider ownership causes

managers' interests to align with the shareholders' interests, owing to the efficient

monitoring scheme. It then favours innovation, as effective monitoring mechanisms allow

managers to mitigate concerns about the consequences of R&D failure on their careers.

Insiders with small shareholdings may be employees who have been rewarded with

shares. So, equity incentives also have a positive effect on corporate innovation in China.

Moreover, employees are more willing to innovate in order to achieve self-worth and have

stable employment (Chang et al., 2006).

Another reason is that the firm owner may focus on innovation in order to find longterm

competitiveness of products or services for a successful continuation of the firm in the

hands of future generations (Caselli and Gennaioli, 2013).

iv. Control Variables

Firm age has a statistically and significantly positive impact on R&D intensity for firms

with insider ownership in Table 4.10, yet the impact is minimal. The learning-by-doing

model tells that the older the firm, the more business experience and foresight it has

(Arrow, 1962; Sorensen and Stuart, 2000; Chang et al., 2002). Accordingly, older firms

with insider ownership may encourage R&D activities to survive in an increasingly

competitive market.

Firm size significantly and negatively affect R&D intensity for firms with insider

ownership, but this impact is trivial in Table 4.10. If firm size upswings dramatically,

firms with insider ownership may be satisfied with firm scale and unwilling to undertake

the cost of R&D failure.

Regarding industry dummies, the baseline is agriculture, forestry, animal husbandry and

fishery. On average, firms with insider ownership within those six industries – (1)

manufacturing industry; (2) information transmission, software and information

technology services; (3) scientific research and technical service industry; (4) education

industry; and (5) health and social work - have a significantly higher R&D intensity than

the baseline as shown in Table 4.10. One interpretation is that an equitycompensation

scheme is prevalent in those industries to promote innovative performance. As with

concentrated ownership, insider ownership has a lower innovation performance in the

wholesale and retail trade than the baseline.

4.3.2.3 State Ownership

4.3.2.3.1 Model 1: Continuous State Ownership

i. Tests for Assumptions of Panel Data Regression

First, the LM test was employed to identify the type of regression to be used. Given that

the LM test has a p-value of p < 0.01, it rejects the null hypothesis of no panel effect at

the 5% level. Hence, the panel data regression is superior to the pooled regression. The p-

value of the HS test is p < 0.01 as it rejects the null hypothesis of random effects at the

1% level. As a result, the panel data regression with fixed effects is an appropriate model.

Then, the tests were applied for defining whether the regression assumptions violated.

First, the mean of residuals is 0, which indicates that the assumption of linearity in

parameters holds (Assumption 1). Second, the assumption of homoscedasticity violates by

modified Wald test, because the p-value of the test is p < 0.01 (Assumption 2). Third, the

p-value of run test of randomness is p < 0.01, which rejects the null hypothesis

(Assumption 3). To rephrase it, residuals are serially correlated. Fourth, the p-value of

Ramsey RESET Test is p < 0.01, which rejects the null hypothesis of no omit variable

(Assumption 4). As a result, there is a problem of model misspecification. Fifth, the total

sample size is large enough to say that residual distribution is asymptotically normal

(Assumption 5). Sixth, the pairwise correlation in Appendix D displays that assumption of

no perfect multicollinearity holds (Assumption 6), because correlations are pretty low.

In order to correct the violations of the assumptions, the fixed effects model was

reregressed by robust standard error.

ii. Results

Table 4.9 Regression with Continues State Ownership

(5)

Fixed Effects

VARIABLES Robust t-statistics p-value

state 0.0012 0.3700 0.7100

(0.0033)

Firm Size -0.0025** -2.4200 0.0160

(0.0010)

Firm Age 0.0015*** 7.6500 0.0000

(0.0002)

Knowledge Stock 0.0000 1.0700 0.2850

(0.0000)

Leverage 0.0005 0.5200 0.6010

(0.0009)

Public A-shares -0.0028 -0.5000 0.6170

(0.0057)

Long-term Investment 0.0006** 2.3300 0.0200

(0.0003)

Industry 2 0.0068 0.5300 0.5990

(0.0129)

Industry 3 0.0171** 2.0800 0.0370

(0.0082)

Industry 4 0.0029 0.3400 0.7360

(0.0085)

Industry 5 -0.0011 -0.0800 0.9390

(0.0142)

Industry 6 -0.0046*** -8.0100 0.0000

(0.0006)

Industry 7 0.0115 1.1100 0.2690

(0.0104)

Industry 8 0.0110 0.9300 0.3540

(0.0118)

Industry 9 0.0447** 2.2900 0.0220

(0.0195)

Industry 10 0.0127 1.5900 0.1120

(0.0080)

Industry 11 0.0046 0.5200 0.6010

(0.0088)

Industry 12 0.0115 0.9800 0.3250

(0.0117)

Industry 13 0.0322* 1.8100 0.0700

Continued

(0.0178)

Industry 14 0.0014 0.1300 0.9000

(0.0113)

Industry 15 0.1326*** 18.5000 0.0000

(0.0072)

Industry 16 0.0593*** 5.9600 0.0000

(0.0099)

Industry 17 0.0084 0.5900 0.5560

(0.0143)

Industry 18 0.0183 1.4500 0.1470

(0.0126)

Constant 0.0271 1.2600 0.2070

(0.0215)

Observations 8,286

Number of Firms 1,409

Adjusted R-squared 0.0799

FIRM FE YES

Log-likelihood 23942

INDUSTRY FE YES

Table 4.9 signifies the panel data regression outputs for state ownership. The p-value of

state ownership is greater than 0.1, which fails to reject the null hypothesis. Hence, state

ownership is insignificantly and negatively related to R&D intensity. The result is

different from the previous hypothesis (i.e. H3).

For control variables, the p-value of ln(firm size) is p < 0.05, which rejects the null

hypothesis at the 5% level. Ln(firm size) and R&D intensity are significantly and

negatively related at the 5% level. If firm size increases by 1%, the mean of R&D

intensity decreases by 2.5e-5, ceteris paribus. It is a really minor impact. The p-value of

firm age is p < 0.01, which rejects the null hypothesis. Firm age is significantly and

positively related to R&D intensity. The expected change in R&D intensity is 0.0015 for

an additional one year increase in firm age, while holding other variables constant. The p-

value of ln(long-term investment) is p < 0.05, which rejects the null hypothesis at the 5%

level. If long-term investment increases by 1%, the mean of R&D intensity increases by

6e-6, ceteris paribus.

For industry dummies, the p-value for industry code 3 is p < 0.05, rejecting the null

hypothesis at the 5% level. On average, the R&D intensity of the manufacturing industry

is 0.0171 higher than the one of agriculture, forestry, animal husbandry and fishery,

ceteris paribus. Industry code 6 has a p-value of p < 0.01, rejecting the null hypothesis.

The average R&D intensity in wholesale and retail trade is 0.0046 lower than the

baseline, all else equal. Industry code 9 has a p-value of p < 0.05, rejecting the null

hypothesis at the 5% level. The average R&D intensity for information transmission,

software and information technology services is 0.0447 above the baseline, all else equal.

The p-value of industry code 13 is p < 0.1, which rejects the null hypothesis at the 10%

level. The mean R&D intensity of scientific research and technical service industry is

0.0322 higher than the baseline, ceteris paribus. The pvalues of industry codes 15 and 16

are p < 0.01, so those two null hypotheses are rejected. The mean R&D intensity

regarding the education industry and health and social work is 0.1326 and 0.0593 above

the baseline level, respectively, all else equal. The p-values for all variables except those

mentioned above are greater than 0.1, and the null hypothesis cannot be rejected.

The log-likelihood value is 23942.

4.3.2.3.2 Model 2: Categorical State Ownership

i. Tests for Assumptions of Panel Data Regression

First, an LM test was applied to observe the type of regression to be used. As the LM test

has a p-value of p < 0.01, it rejects the null hypothesis of no panel effect at the 1% level.

Thus, the panel data regression is better than the pooled regression. Since the pvalue of

the HS test is p < 0.01, it rejects the null hypothesis of a random effect at the 1% level.

The panel data regression with fixed effects is, therefore, an appropriate model.

Afterwards, regression assumptions were tested to ensure the coefficients were unbiased

and consistent.

First, the mean of residuals is 0, which means that the assumption of linearity does not

violate (Assumption 1). Second, the assumption of homoscedasticity violates as the

modified Wald test has a p-value of p < 0.01 (Assumption 2). Third, the p-value of the

run test of randomness is p < 0.01, which rejects the null hypothesis (Assumption 3). In

other words, residuals are autocorrelated. Fourth, the p-value of the Ramsey RESET

Test is greater than 0.1, which fails to reject the null hypothesis (Assumption 4).

Accordingly, misspecification is not a concern in this case. Fifth, the total sample size is

large enough to suggest that residual distribution is asymptotically normal (Assumption

5). Sixth, the pairwise correlation matrix in Appendix D suggests no perfect

multicollinearity because of low correlations (Assumption 6).

In order to correct the violations of the assumptions, the fixed effects model was

reregressed by the robust standard error.

ii. Results

Table 4.10 Regression with Categorical State Ownership

(6)

Fixed Effects

VARIABLES Robust t-statistics p-values

Low 0.0031* 1.8300 0.0670

(0.0017)

Medium -0.0018 -0.6500 0.5150

(0.0028)

High -0.0008 -0.3800 0.7040

(0.0021)

Firm Size -0.0025** -2.4200 0.0160

(0.0010)

Firm Age 0.0015*** 7.8400 0.0000

(0.0002)

Knowledge Stock 0.0000 1.0800 0.2790

(0.0000)

Leverage 0.0005 0.5200 0.6020

(0.0009)

Public A-shares -0.0021 -0.3600 0.7170

(0.0057)

Continued

Long-term Investment 0.0006** 2.2900 0.0220

(0.0003)

Industry 2 0.0069 0.5300 0.5930

(0.0128)

Industry 3 0.0173** 2.1200 0.0340

(0.0081)

Industry 4 0.0024 0.2900 0.7720

(0.0084)

Industry 5 -0.0005 -0.0400 0.9700

(0.0142)

Industry 6 -0.0047*** -9.1200 0.0000

(0.0005)

Industry 7 0.0105 1.0000 0.3160

(0.0105)

Industry 8 0.0105 0.8900 0.3760

(0.0118)

Industry 9 0.0448** 2.2900 0.0220

(0.0195)

Industry 10 0.0124 1.5600 0.1200

(0.0079)

Industry 11 0.0040 0.4600 0.6440

(0.0087)

Industry 12 0.0114 0.9700 0.3310

(0.0117)

Industry 13 0.0323* 1.8300 0.0680

(0.0177)

Industry 14 0.0012 0.1100 0.9140

(0.0113)

Industry 15 0.1322*** 18.5200 0.0000

(0.0071)

Industry 16 0.0586*** 5.8500 0.0000

(0.0100)

Industry 17 0.0084 0.5900 0.5570

(0.0142)

Industry 18 0.0177 1.4200 0.1540

(0.0124)

Constant 0.0281 1.3000 0.1930

(0.0216)

Observations 8,286

Number of Firms 1,409

Adjusted R-squared 0.0820

FIRM FE YES

INDUSTRY FE YES

Log-likelihood 23953

The baseline is the zero level of state ownership.

The low level has a p-value of p < 0.1 in Table 4.10, rejecting the null hypothesis at the

10% level. On average, R&D intensity for the low-level is 0.0031 higher than the

zerolevel. The other two levels of state ownership do not differ significantly from non-

state ownership in R&D intensity since they have p-values greater than 0.1.

For control variables, the p-value of ln(firm size) is p < 0.05, which rejects the null

hypothesis at the 5% level. Ln(firm size) and R&D intensity are significantly and

negatively related. The mean of R&D intensity decreases by 2.5e-5 if firm size increases

by 1%, ceteris paribus. The p-value of firm age is p < 0.01, which rejects the null

hypothesis. Firm age and R&D intensity are positively and significantly related. The

expected change in R&D intensity is 0.0015 for an additional one year increase in firm

age, ceteris paribus. The p-value of ln(long-term investment) is p < 0.05, which rejects the

null hypothesis at the 5% level. If long-term investment increases by 1%, the mean of

R&D intensity increases by 6e-6, ceteris paribus.

The baseline for industry dummy is 'agriculture, forestry, animal husbandry and fishery',

which is denoted as industry code 1. The p-value of industry code 3 is p < 0.05, which

rejects the null hypothesis at the 5% level. On average, the R&D intensity of the

manufacturing industry is 0.0173 higher than the one of the baseline, ceteris paribus.

The p-value of industry code 6 is p < 0.01, which rejects the null hypothesis. The mean

R&D intensity of wholesale and retail trade is 0.0047 lower than the baseline, ceteris

paribus. The p-value of industry code 9 is p < 0.05, which rejects the null hypothesis at

the 5% level. The mean R&D intensity of information transmission, software and

information technology services is 0.0448 higher than the baseline, ceteris paribus. The p-

value of industry code 13 is p < 0.1, which rejects the null hypothesis at the 10% level.

The mean R&D intensity of scientific research and technical service industry is 0.0323

higher than the baseline, ceteris paribus. Both industry codes 15 and 16 have pvalues of p

< 0.01, and both hypotheses can be rejected. On average, the education industry and

health and social work have an R&D intensity of 0.1320 and 0.0586 higher than the

baseline, respectively, ceteris paribus. Other control variables' p-values are relatively

large, which fail to reject the null hypotheses at the 10% level.

The log-likelihood value is 23953.

4.3.2.3.3 Discussions for State Ownership with R&D Intensity

i. Likelihood Ratio Test

The log-likelihood value in Table 4.11 is 23942, which is lower than the one in Table 4.12

(i.e. 23953). Nonetheless, the p-value of the likelihood ratio test is p < 0.01, rejecting the

null hypothesis at the 1% level. The model with categorical state ownership (i.e. model 6)

performs better than the model with continuous categorical state ownership (i.e. model 5).

ii. Continuous State Ownership

There is no relationship between state ownership and innovation performance found in

Table 4.13, which rejects the hypothesis made in Chapter 3 (i.e. H3).

Though firms with state ownership have resource benefits, the importance of resource

allocation from the government drops when firms are listed in the exchange. Except for

financial support from the government, listed firms are more likely to raise funds in the

market (Zhou et al., 2017).

Another reason may be that highly sophisticated SOEs that have invested heavily in

innovation are not listed for technical confidentiality as well as political and commercial

reasons.

iii. Categorical State Ownership

Results such as those in Table 4.10 refute the conclusion that state ownership and

innovation performance are non-correlated, concluding that low-level state ownership has

a greater impact on R&D intensity than the zero-level. Innovation is facilitated when state

ownership is below 5% rather than for non-state ownership. The result is consistent with

hypothesis H3 only when state ownership is low.

One reason could be that the low-level state ownership takes advantage of state resources

without the fear of the state seizing the firm's control rights. In 2015, the CPC

promulgated regulations on the work of the party committees in the firm with the clear

intention of the firm's control rights (Xie et al., 2022). It may cause other investors

concerns about the firm's future and increase great pressure on managers. Accordingly,

Managers in a firm with a high level of state ownership may abate R&D investment, as

they are reluctant to bear the cost of R&D failure for the sake of their future careers. iv.

Control Variables

In spite of the fact that firm age positively affects R&D intensity in Table 4.14, such

influence is minor. Due to the learning-by-doing model, older firms take advantage of

business experience and foresight (Arrow, 1962; Sorensen and Stuart, 2000; Chang et al.,

2002). As a result, firms with state ownership favour R&D investment to let firms be alive

in an increasingly competitive market.

The negative impact of firm size on R&D intensity is paltry in Table 4.14. Nevertheless,

firms with state ownership may no longer be interested in R&D investment if the firm size

becomes extremely large.

The positive and slight effect of long-term investment on R&D intensity in Table 4.14

shows that state-owned firms’ interests in R&D rise as there is a surge in long-term

investment.

Regarding industry dummies, the baseline is agriculture, forestry, animal husbandry and

fishery. On average, firms with insider ownership within those five industries – (1)

manufacturing industry; (2) information transmission, software and information

technology services; (3) scientific research and technical service industry; (4) education

industry; and (5) health and social work - have a significantly higher R&D intensity than

the baseline. It includes highly sophisticated industries, such as manufacturing, where the

government invests enormously in scientific research in order to break through the

technological blockade from western countries as soon as possible. So, in contrast to

agriculture, forestry, animal husbandry and fishery, it is understandable for the

government to have a relatively high R&D intensity in these sectors.

4.3.2.4 Institutional Ownership

4.3.2.4.1 Model 1: Continuous Institutional Ownership

i. Tests for Assumptions of Panel Data Regression

Initially, the LM test was employed to decide the type of regression to be applied. As the

LM test had a p-value of p < 0.01, it rejected the null hypothesis of no panel effect at the

1% level. Consequently, the panel data regression was better than the pooled regression.

Since the p-value of the HS test was p < 0.01, it rejected the null hypothesis of a random

effect at the 1% level. A panel data regression with fixed effects was, therefore, an

appropriate model.

After choosing an appropriate model, assumptions should be tested to confirm the

regression estimates were meaningful (i.e. consistent and unbiased).

First, the mean of residuals was 0, which suggested that the assumption of linearity in

parameters held (Assumption 1). Second, the p-value of the modified Wald test was p <

0.01, rejecting the null hypothesis of homoscedastic errors (Assumption 2). Third, the

pvalue of the run test of randomness was p < 0.01, which rejected the null hypothesis

(Assumption 3). To put it in another way, residuals were serially correlated. Fourth, the p-

value of the Ramsey RESET Test was p < 0.01, which rejected the null hypothesis of no

omitted variable (Assumption 4). As a result, there was a problem with model

misspecification. Fifth, residuals were asymptotically normal due to a large sample size

(Assumption 5). Sixth, the assumption of no perfect multicollinearity held due to

correlations in Appendix D being pretty low (Assumption 6).

Then, the robust standard error was applied to correct the violations of the assumptions.

ii. Results

Table 4.11 Regression with Continuous Institutional Ownership

(7)

Fixed Effects

VARIABLES Robust t-statistics p-values

institutional 0.0001 0.0200 0.9840

(0.0042)

Firm Size -0.0025** -2.4200 0.0160

(0.0010)

Firm Age 0.0015*** 7.6400 0.0000

(0.0002)

Knowledge Stock 0.0000 1.0600 0.2890

(0.0000)

Leverage 0.0005 0.5200 0.6000

(0.0009)

Public A-shares -0.0026 -0.4000 0.6860

(0.0065)

Long-term Investment 0.0006** 2.3300 0.0200

(0.0003)

Industry 2 0.0069 0.5300 0.5930

(0.0129)

Industry 3 0.0171** 2.1100 0.0350

(0.0081)

Industry 4 0.0029 0.3500 0.7270

(0.0084)

Industry 5 -0.0010 -0.0700 0.9440

(0.0141)

Industry 6 -0.0045*** -8.5400 0.0000

(0.0005)

Industry 7 0.0115 1.0900 0.2770

(0.0105)

Industry 8 0.0107 0.9100 0.3660

(0.0118)

Industry 9 0.0447** 2.2900 0.0220

(0.0195)

Industry 10 0.0127 1.6100 0.1080

(0.0079)

Industry 11 0.0046 0.5300 0.5960

(0.0087)

Industry 12 0.0116 0.9900 0.3220

(0.0117)

Industry 13 0.0319* 1.8000 0.0720

(0.0177)

Industry 14 0.0014 0.1300 0.8960

(0.0111)

Industry 15 0.1326*** 18.5400 0.0000

(0.0072)

Industry 16 0.0590*** 5.6900 0.0000

(0.0104)

Industry 17 0.0085 0.5900 0.5530

(0.0143)

Continued

Industry 18 0.0183 1.4600 0.1440

(0.0125)

Constant 0.0270 1.2400 0.2140

(0.0217)

Observations 8,286

Number of Firms 1,409

Adjusted R-squared 0.0798

FIRM FE YES

Log-likelihood 23942

INDUSTRY FE YES

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The p-value of institutional ownership in the Table 4.11 is greater than 0.1, which fails to

reject the null hypothesis. Hence, institutional ownership is insignificantly and positively

related to R&D intensity.

For control variables, the p-value of ln(firm size) is p < 0.05, which rejects the null

hypothesis at the 5% level. Ln(firm size) and R&D intensity are significantly and

negatively related at the 5% level. If firm size increases by 1%, the mean of R&D

intensity decreases by 2.5e-5, ceteris paribus. The p-value of firm age is p < 0.01, which

rejects the null hypothesis. Firm age is significantly and positively related to R&D

intensity. The expected change in R&D intensity is 0.0015 for an additional one year

increase in firm age, while holding other variables constant. The p-value of ln(long-term

investment) is p < 0.01, which rejects the null hypothesis at the 1% level. If long-term

investment increases by 1%, the mean of R&D intensity increases by 6e-6, ceteris

paribus.

The baseline for industry dummy is ‘agriculture, forestry, animal husbandry and fishery’.

The p-value of industry code 3 is p < 0.05, which rejects the null hypothesis at the 5%

level. On average, the R&D intensity of the manufacturing industry is 0.0171 higher than

the one of the baseline, ceteris paribus. The p-value of industry code 6 is p < 0.01, which

rejects the null hypothesis. The mean R&D intensity of wholesale and retail trade is

0.0045 lower than the baseline, ceteris paribus. The p-value of industry code 9

is p < 0.05, which rejects the null hypothesis at the 5% level. The mean R&D intensity of

information transmission, software and information technology services is 0.0447 higher

than the baseline, ceteris paribus. The p-value of industry code 13 is p < 0.1, which rejects

the null hypothesis at the 10% level. The mean R&D intensity of scientific research and

technical service industry is 0.0319 higher than the baseline, ceteris paribus. Both industry

codes 15 and 16 have p-values of p < 0.01, so null hypotheses are rejected. On average,

the education industry and health and social work have an R&D intensity of 0.1326 and

0.0590 higher than the baseline, respectively, ceteris paribus. Other control variables and

dummy variables’ p-values are relatively large, which fail to reject the null hypotheses at

the 10% level.

In addition, the log-likelihood value is 23942.

4.3.2.4.2 Model 2: Categorical Institutional Ownership

i. Tests for Assumptions of Panel Data Regression

To start with, the LM test was employed to determine the type of regression to be used.

Due to the fact that the LM test had a p-value of p < 0.01, it rejected the null hypothesis of

no panel effect at the 1% level. Panel data regression was, therefore, better than pooled

regression. As the p-value of the HS test was p < 0.01, it rejected the null hypothesis of

random effects at the 1% level. Consequently, the panel data regression with fixed effects

was an appropriate model.

After that, regression assumptions should be tested due to consistent and unbiased

coefficients needed.

First, the mean of residuals was 0, which means that the assumption of linearity did not

violate (Assumption 1). Second, the assumption of homoscedasticity violated by the

modified Wald test with the p-value of p < 0.01 (Assumption 2). Third, the p-value of the

run test of randomness was p < 0.01, which rejected the null hypothesis (Assumption 3).

In other words, residuals were autocorrelated. Fourth, the p-value of the Ramsey RESET

Test was p < 0.01, which rejected the null hypothesis at the 1% level (Assumption 4).

Accordingly, misspecification was a concern in this case. Fifth, the total sample size was

large enough to suggest that residual distribution was asymptotically normal (Assumption

5). Sixth, the pairwise correlation was displayed in Appendix D and suggested that there

is no perfect multicollinearity due to relatively low values of correlations (Assumption 6).

The fixed effects model with robust standard error was then utilised to correct the

violations of the assumptions.

ii. Results

Table 4.12 Regression with Categorical Institutional Ownership

(8)

Fixed Effects t-statistics p-values

VARIABLES Robust

Low 0.0005 0.3700 0.7140

(0.0012)

Medium 0.0008 0.5400 0.5880

(0.0014)

High 0.0007 0.3900 0.6940

(0.0019)

Firm Size -0.0025** -2.4500 0.0140

(0.0010)

Firm Age 0.0015*** 7.6900 0.0000

(0.0002)

Knowledge Stock 0.0000 1.0500 0.2950

(0.0000)

Leverage 0.0005 0.5200 0.6010

(0.0009)

Public A-shares -0.0031 -0.4800 0.6310

(0.0065)

Long-term Investment 0.0006** 2.3100 0.0210

(0.0003)

Industry 2 0.0068 0.5200 0.6020

(0.0130)

Industry 3 0.0171** 2.1000 0.0360

(0.0082)

Industry 4 0.0030 0.3600 0.7180

(0.0084)

Industry 5 -0.0009 -0.0700 0.9480

(0.0141)

Industry 6 -0.0046*** -8.6300 0.0000

(0.0005)

Continued

Industry 7 0.0116 1.1200 0.2640

(0.0104)

Industry 8 0.0107 0.9100 0.3650

(0.0118)

Industry 9 0.0447** 2.2900 0.0220

(0.0195)

Industry 10 0.0128 1.6100 0.1070

(0.0079)

Industry 11 0.0046 0.5300 0.5950

(0.0087)

Industry 12 0.0117 1.0100 0.3130

(0.0116)

Industry 13 0.0318* 1.7800 0.0750

(0.0178)

Industry 14 0.0015 0.1400 0.8920

(0.0112)

Industry 15 0.1328*** 19.1100 0.0000

(0.0070)

Industry 16 0.0590*** 5.8200 0.0000

(0.0101)

Industry 17 0.0084 0.5800 0.5610

(0.0144)

Industry 18 0.0184 1.4700 0.1430

(0.0125)

Constant 0.0270 1.2500 0.2110

(0.0216)

Observations 8,286

Number of Firms 1,409

Adjusted R-squared 0.0797

FIRM FE YES

INDUSTRY FE YES

Log-likelihood 23942

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The baseline is zero-level institutional ownership.

All p-values for three levels of institutional ownership are greater than 0.1 in the table

above. No significant difference in institutional ownership compared to noninstitutional

ownership for all three levels of R&D intensity, on average, ceteris paribus. For control

variables, the p-value of ln(firm size) is p < 0.05, which rejects the null hypothesis at the

5% level. Ln(firm size) and R&D intensity are significantly and negatively related. The

mean of R&D intensity decreases by 2.5e-5 if firm size increases by 1%, ceteris paribus.

The p-value of firm age is p < 0.01, which rejects the null hypothesis. Firm age and R&D

intensity are positively and significantly related. The expected change in R&D intensity is

0.0015 for an additional one year increase in firm age, ceteris paribus. The p-value of

ln(long-term investment) is p < 0.05, which rejects the null hypothesis at the 5% level. If

long-term investment increases by 1%, the mean of R&D intensity increases by 6e-6,

ceteris paribus.

The baseline for industry dummy is ‘agriculture, forestry, animal husbandry and fishery’.

The p-value of industry code 3 is p < 0.05, which rejects the null hypothesis at the 5%

level. On average, the R&D intensity of the manufacturing industry is 0.0171 higher than

the one of the baseline, ceteris paribus. The p-value of industry code 6 is p < 0.01, which

rejects the null hypothesis. The mean R&D intensity of wholesale and retail trade is

0.0046 lower than the baseline, ceteris paribus. The p-value of industry code 9 is p < 0.05,

which rejects the null hypothesis at the 5% level. The mean R&D intensity of information

transmission, software and information technology services is 0.0447 higher than the

baseline, ceteris paribus. The p-value of industry code 13 is p < 0.1, which rejects the null

hypothesis at the 10% level. The mean R&D intensity of scientific research and technical

service industry is 0.0318 higher than the baseline, ceteris paribus. Both industry codes 15

and 16 have p-values of p < 0.01, which rejects the null hypothesis. On average, the

education industry and health and social work have an R&D intensity of 0.1328 and

0.0590 higher than the baseline, respectively, ceteris paribus. Other control variables’ p-

values are relatively large, which fail to reject the null hypotheses at the 10% level.

In addition, the log-likelihood value is 23942.

4.3.2.4.3 Discussions for Institutional Ownership with R&D Intensity

i. Likelihood Ratio Test

The log-likelihood value in Table 4.11 is 23942, which is same as the log-likelihood value

in Table 4.12 (i.e. 23942). Likelihood ratio test generates the p-value above 0.1, which

fails to reject the null hypothesis. The model with continuous institutional ownership is

better than the model with the categorical institutional ownership.

ii. Continuous Institutional Ownership

Table 4.17 reveals no relationship between institutional ownership and R&D intensity,

whereas this conclusion is inconsistent with the hypothesis made in Chapter 3 (i.e. H4).

Despite firms with institutional ownership favour efficient monitoring mechanisms,

particularly with weak incumbent management (Fung, 2012), it is underdeveloped in

developing countries compared to developed countries. Of the data, approximately 8% of

the sample have no institutional ownership, 34% have less than 5% (but higher than 0%),

and 36% have levels between 5% and 20%. Institutional investors usually regulate and

monitor managers via governance activities or gathering information about the quality of

R&D investments (Bushee, 1998). If the level of institutional ownership is low, active

monitoring by institutional ownership cannot eliminate managers’ concerns about R&D

failure and cannot relieve managers’ pressure to maximise short-term profits.

Additionally, the holding period of institutional investors may be too short for gaining

long-term benefits of R&D investment.

iii. Categorical Institutional Ownership

On average, institutional ownership is not significantly different across all three levels of

R&D intensity compared to non-institutional ownership. It is then in disagreement with

hypothesis H4.

iv. Control Variables

Firm age has a statistically and significantly positive impact on R&D intensity for firms

with institutional ownership in Table 4.17, yet the impact is paltry. A rise in firm age

accumulates business experience and business foresight, owing to the learning-by-doing

model foresight (Arrow, 1962; Sorensen and Stuart, 2000; Chang et al., 2002). Firms with

institutional ownership boost R&D investment to survive in an increasingly competitive

market.

Firm size significantly and negatively affect R&D intensity for firms with institutional

ownership, yet this impact is trivial. If firm size upswings dramatically, firms' interest in

R&D investment reduces as they are unwilling to undertake the cost of R&D failure.

The little and positive impact of long-term investment on R&D intensity discloses that

R&D intensity grows extensively only if there is a bulk investment in long-term assets.

With regard to industry data, the baseline is agriculture, forestry, animal husbandry and

fisheries. For those firms with concentrated ownership in those five sectors - (1)

manufacturing; (2) information transmission, software and information technology; (3)

scientific research and technical services; (4) education; and (5) health and social work -

the R&D intensity is, on average, significantly higher than the baseline. In contrast, the

R&D intensity of the wholesale and retail sector in institutional ownership is below the

baseline.

4.3.2.5 Foreign Ownership

4.3.2.5.1 Model 1: Continuous Foreign Ownership

i. Tests for Assumptions of Panel Data Regression

First of all, the LM test was applied to see which type of regression to be used. Since the

p-value of the LM test was p < 0.01, it rejected the null hypothesis of no panel effect at

the 1% level. Hence, panel data regression was better than pooled regression. As the p-

value of the HS test was p < 0.01, it rejected the null hypothesis of random effect at the

1% level. Therefore, panel data regression with fixed effects model was a proper model.

Then, regression assumptions should be tested to confirm coefficients that were consistent

and unbiased.

First, the mean of residuals was 0, which meant the parameters are linear (Assumption

1). Second, the variance of error terms was heteroscedastic because the p-value of the

modified Wald test was p < 0.01. The null hypothesis was then rejected (Assumption 2).

Third, the p-value of the run test of randomness was p < 0.01, which rejected the null

hypothesis (Assumption 3). To put it in another way, residuals were serially correlated.

Fourth, the p-value of the Ramsey RESET Test was p < 0.01, which rejected the null

hypothesis of no omitted variable (Assumption 3). As a result, there was a problem with

model misspecification. Fifth, residuals were asymptotically normal due to a large

sample size (Assumption 5). Sixth, Appendix D indicates multicollinearity is not an issue

as the correlation is quite low (Assumption 6).

In order to correct the violations of the assumptions, the model was re-regressed by the

robust standard error.

ii. Results

Table 4.13 Regression with Continuous Foreign Ownership

(9)

Fixed Effects t-statistics p-values

VARIABLES Robust

foreign -0.0246** -2.5000 0.0120

(0.0098)

Firm Size -0.0024** -2.3800 0.0170

(0.0010)

Firm Age 0.0015*** 7.6800 0.0000

(0.0002)

Knowledge Stock 0.0000 1.0900 0.2740

(0.0000)

Leverage 0.0005 0.5300 0.5940

(0.0009)

Public A-shares -0.0007 -0.1300 0.8960

(0.0057)

Long-term Investment 0.0006** 2.3700 0.0180

(0.0003)

Industry 2 0.0065 0.5000 0.6170

(0.0129)

Industry 3 0.0172** 2.1000 0.0360

(0.0082)

Industry 4 0.0028 0.3400 0.7360

(0.0084)

Industry 5 -0.0007 -0.0500 0.9590

Continued

(0.0141)

Industry 6 -0.0045*** -8.4100 0.0000

(0.0005)

Industry 7 0.0115 1.1000 0.2710

(0.0104)

Industry 8 0.0122 1.0300 0.3040

(0.0119)

Industry 9 0.0447** 2.3000 0.0220

(0.0195)

Industry 10 0.0120 1.5100 0.1300

(0.0079)

Industry 11 0.0046 0.5200 0.6010

(0.0087)

Industry 12 0.0116 0.9900 0.3200

(0.0117)

Industry 13 0.0321* 1.8000 0.0720

(0.0178)

Industry 14 0.0015 0.1300 0.8960

(0.0113)

Industry 15 0.1327*** 18.5600 0.0000

(0.0071)

Industry 16 0.0587*** 5.9200 0.0000

(0.0099)

Industry 17 0.0082 0.5700 0.5670

(0.0142)

Industry 18 0.0183 1.4500 0.1460

(0.0126)

Constant 0.0256 1.2000 0.2320

(0.0214)

Observations 8,286

Number of Firms 1,409

Adjusted R-squared 0.0815

FIRM FE YES

Log-likelihood 23949

INDUSTRY FE YES

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The p-value of foreign ownership in the table above is p < 0.05, rejecting the null

hypothesis at the 5% level. R&D intensity is expected to decrease by 0.0246 for an

additional one unit increase in foreign ownership, all other things being equal. For control

variables, the p-value of ln(firm size) is p < 0.05, which rejects the null hypothesis at the

5% level. Ln(firm size) and R&D intensity are significantly and negatively related at the

5% level. If firm size increases by 1%, the mean of R&D intensity decreases by 2.4e-5,

ceteris paribus. The p-value of firm age is p < 0.01, which rejects the null hypothesis.

Firm age is significantly and positively related to R&D intensity. The expected change in

R&D intensity is 0.0015 for an additional one year increase in firm age, while holding

other variables constant. The p-value of ln(long-term investment) is p < 0.05, which

rejects the null hypothesis at the 5% level. If long-term investment increases by 1%, the

mean value of R&D intensity increases by 6e-6, ceteris paribus.

The baseline for industry dummy is 'agriculture, forestry, animal husbandry and fishery'.

The p-value of industry code 3 is p < 0.05, which rejects the null hypothesis at the 5%

level. On average, the R&D intensity of the manufacturing industry is 0.0172 higher than

the one of the baseline, ceteris paribus. The p-value of industry code 6 is p < 0.01, which

rejects the null hypothesis. The mean R&D intensity of wholesale and retail trade is

0.0045 lower than the baseline, ceteris paribus. The p-value of industry code 9 is p < 0.05,

which rejects the null hypothesis at the 5% level. The mean R&D intensity of information

transmission, software and information technology services is 0.0447 higher than the

baseline, ceteris paribus. The p-value of industry code 13 is p < 0.1, which rejects the null

hypothesis at the 10% level. The mean R&D intensity of scientific research and technical

service industry is 0.0321 higher than the baseline, ceteris paribus. Both industry codes 15

and 16 have p-values of p < 0.01, which rejects the null hypothesis. On average, the

education industry and health and social work have an R&D intensity of 0.1327 and

0.0587 higher than the baseline, respectively, ceteris paribus. Other variables' p-values are

relatively large, which fail to reject the null hypotheses at the 10% level.

In addition, the log-likelihood value is 23949.

4.3.2.5.2 Model 2: Categorical Foreign Ownership

i. Tests for Assumptions of Panel Data Regression

An LM test was first applied to identify the type of regression to be used. Since the

pvalue of the LM test was p < 0.01, it rejected the null hypothesis of no panel effect at

the 1% level. The panel data regression was, therefore, better than the pooled regression.

Since the HS test had a p-value of p < 0.01, it rejected the null hypothesis of a random

effect at the 1% level. Then, A panel data regression with fixed effects was a suitable

model.

Then, all regression assumptions should be tested to get consistent and unbiased

estimators.

First, the mean of residuals was 0, which meant that the assumption of linearity did not

violate (Assumption 1). Second, the assumption of homoscedasticity violated due to the

p-value of p < 0.01 in the Wald test (Assumption 2). Third, the p-value of the run test of

randomness was p < 0.01, which rejected the null hypothesis (Assumption 3). It meant

that residuals were serially correlated. Fourth, the p-value of the Ramsey RESET Test

was p < 0.01, which rejected the null hypothesis at the 1% level (Assumption 4).

Accordingly, misspecification was a concern. Fifth, the total sample size was large

enough to suggest that residual distribution was asymptotically normal (Assumption 5).

Sixth, Appendix D demonstrates relatively low correlations, which means

multicollinearity is not a problem (Assumption 6).

The fixed effects model was then applied to correct the violations of the assumptions.

ii. Results

Table 4.14 Regression with Categorical Foreign Ownership

iii. (10)

Fixed Effects t-statistics p-values

VARIABLES Robust

Low -0.0006 -0.8300 0.4090

(0.0007)

Medium -0.0026 -1.3100 0.1900

(0.0020)

High -0.0080*** -2.8700 0.0040

(0.0028)

Continued

0.0170

0.0000

0.2970

0.5950

0.8320

2.4000 0.0170

0.5100 0.6120

2.1100 0.0350

0.3400 0.7310

-0.0500 0.9620

-7.9700 0.0000

1.1100 0.2670

1.0300 0.3050

2.3000 0.0220

1.5300 0.1260

0.5300 0.5940

1.0000 0.3180

1.8000 0.0720

0.1400 0.8910

18.5400 0.0000

5.9300 0.0000

0.5700 0.5670

1.4600 0.1450

1.1800 0.2390

Continued

FIRM FE YES

INDUSTRY FE YES

Log-likelihood 23950

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The baseline is non-foreign ownership.

From the table above, the p-value of the high-level foreign ownership is p < 0.01,

rejecting the null hypothesis at the 1% level. On average, R&D intensity for the highlevel

is 0.0080 lower than the zero-level while holding other variables constant. The other two

levels of R&D intensity are not significantly different compared to nonforeign ownership,

as their p-values are greater than 0.1.

For control variables, the p-value of ln(firm size) is p < 0.05, which rejects the null

hypothesis at the 5% level. Ln(firm size) and R&D intensity are significantly and

negatively related. The mean of R&D intensity decreases by 2.4e-5 if firm size increases

by 1%, ceteris paribus. The p-value of firm age is p < 0.01, which rejects the null

hypothesis. Firm age and R&D intensity are positively and significantly related. The

expected change in R&D intensity is 0.0015 for an additional one year increase in firm

age, ceteris paribus. The p-value of ln(long-term investment) is p < 0.05, which rejects the

null hypothesis at the 5% level. If long-term investment increases by 1%, the mean of

R&D intensity increases by 7e-6, ceteris paribus.

The baseline for industry dummy is ‘agriculture, forestry, animal husbandry and fishery’.

The p-value of industry code 3 is p < 0.05, which rejects the null hypothesis at the 5%

level. On average, the R&D intensity of the manufacturing industry is 0.0172 higher than

the one of the baseline, ceteris paribus. The p-value of industry code 6 is p < 0.01, which

rejects the null hypothesis. The mean R&D intensity of wholesale and retail trade is

0.0044 lower than the baseline, ceteris paribus. The p-value of industry code 9 is p < 0.05,

which rejects the null hypothesis at the 5% level. The mean R&D intensity of information

transmission, software and information technology services is 0.0448 higher than the

baseline, ceteris paribus. The p-value of industry code 13 is p < 0.1, which rejects the null

hypothesis at the 10% level. The mean R&D intensity of scientific research and technical

service industry is 0.0321 higher than the baseline, ceteris paribus. Both industry codes 15

and 16 have p-values of p < 0.01, which rejects the null hypothesis. On average, the

education industry and health and social work have an R&D intensity of 0.1327 and

0.0588 higher than the baseline, respectively, ceteris paribus. Other variables’ p-values are

relatively large, which fail to reject the null hypotheses at the 10% level.

In addition, the log-likelihood value is 23950.

4.3.2.5.3 Discussions for Foreign Ownership with R&D Intensity

i. Likelihood Ratio Test

The log-likelihood value in Table 4.13 is 23949, and the log-likelihood value in Table 4.14

is 23950. Then, the likelihood ratio test generates a p-value above 0.1, which fails to

reject the null hypothesis. Then, the model with continuous foreign ownership is better

than the model with categorical foreign ownership.

ii. Continuous Foreign Ownership

Table 4.13 discloses that foreign ownership and R&D intensity are negatively related,

which is not in line with the previous hypothesis (i.e. H5).

One interpretation is that many foreign firms have set up subsidiaries in China. In such

cases, foreign firms prefer to import R&D from the parent firms than to innovate locally

in the subsidiaries. It has the advantage of being more cost-effective for foreign firms by

cutting down on using a range of resources, including money, required for R&D

(Globerman and Meredith, 1984). High foreign ownership inhibits the incentive and the

ability to innovate in subsidiaries to some extent. However, indigenous innovation is

essential to China's new source of inclusive growth.

Furthermore, approximately 67% of the data does not contain foreign ownership.

Moreover, within the remaining 33%, financial investments in Chinese firms through the

Hong Kong Securities Clearing Company (HKSC) are also included. HKSC is the sole

subsidiary of the Hong Kong stock exchange. It is an authorised clearing institution that

operates the central clearing and settlement system in Hong Kong. Some Chinese firms

may be listed not only in Shanghai Stock Exchange, but also in Hong Kong stock

exchange. If firms are registered in mainland China, firm shares listed in Hong Kong are

called H shares. Foreign investors cannot directly buy shares in the stock exchange

located in mainland China, but they can indirectly buy H shares of Chinese listed firms on

the Hong Kong Stock Exchange. The firm's annual report does not show each investor's

name, but only HKSC and hence many foreign investors via H shares are interested in a

financial investment rather than supervision. Hence, it may result in a negative impact of

foreign ownership on innovation performance, as foreign investors pursue short-term

maximisation rather than long-term value.

iii. Categorical Foreign Ownership

Table 4.14 confirms the negative relationship between foreign ownership and R&D

intensity for the proportion of foreign ownership over 20%. Again, the result is not in line

with hypothesis H5.

High foreign ownership adds to the fact that the company is a subsidiary of a foreign

company. Due to the cost of innovation, foreign firms are more willing to take advanced

technology from the parent company to China, leaving the Chinese subsidiary to become

an original equipment manufacturer (OEM) and inhibiting its willingness to develop its

R&D capabilities (Globerman and Meredith, 1984). The parent company prefers the

subsidiary to exploit its advantages - a large pool of trained workers, relatively low

resource costs, and a vast Chinese market - to extract more benefits for the parent

company.

iv. Control Variables

The effect of firm age on R&D intensity is statistically positive but minor, as shown in

Table 4.13. Older firms benefit from business experience and foresight, according to the

learning-by-doing model (Arrow, 1962; Sorensen and Stuart, 2000; Chang et al., 2002).

The small and negative effect of firm size on R&D intensity exposes that the willingness

of firms with foreign ownership to invest in R&D reduces as firm size upswings

dramatically.

The long-term investment has a slight and positive influence on R&D intensity. It shows

that a growth of R&D intensity is obvious only when a huge sum of investment in long-

term assets.

Regarding industry dummies, the baseline is agriculture, forestry, animal husbandry and

fishery. On average, firms with foreign ownership within those five industries – (1)

manufacturing industry; (2) information transmission, software and information

technology services; (3) scientific research and technical service industry; (4) education

industry; and (5) health and social work - have a significantly higher R&D intensity than

the baseline.

4.3.3 Depth of Innovation and Diversity of Innovation

4.3.3.1 Concentrated Ownership

4.3.3.1.1 Depth of Innovation and Concentrated Ownerships

i. Tests for Assumptions of Panel Data Regression

The first step still is to define which model to be used. LM test was applied, resulting in a

p-value of p < 0.01 which rejected the null hypothesis of no panel effect at the 1% level.

Hence, panel data regression was better than pooled regression. As the p-value of the HS

test was p < 0.01, it rejected the null hypothesis of random effect at the 1% level.

Therefore, panel data regression with a fixed effects was a proper model. Then,

regression assumptions should be tested to obtain consistent and unbiased

estimators.

First, the zero mean residual implied that the assumption of linearity holds (Assumption

1). Second, the p-value of the modified Wald test was p < 0.01, which illustrated the

heteroscedastic residuals. Assumption 2 did not hold. Third, the p-value of the run test of

randomness was p < 0.01, which rejected the null hypothesis (Assumption 3). To rephrase

it, there was a serial correlation in residuals. Fourth, the p-value of the Ramsey

RESET Test was greater than 0.1, which failed to reject the null hypothesis (Assumption

4). It meant that there is no misspecification error. Fifth, the total sample size was large

enough to say that residual distribution is asymptotically normal (Assumption 5). Sixth,

the pairwise correlation matrix is in Appendix E. It exhibits low correlations except for

the correlation between depth and diversity. It was understandable that depth and diversity

are highly correlated, as they were two-sided of a mirror. It was good news that we could

know the other side's story if we know one side. Hence, the high correlation would not

bother the regression analysis since those two variables would not be used in one

regression. In other words, multicollinearity was not an issue in this situation, and

Assumption 6 held.

Due to heteroscedastic and non-independent residuals, the robust standard error was

applied to fix those problems.

ii. Results

Table 4.15 Depth of Innovation for Concentrated Ownership

(11)

H6a

Depth of Innovation t-statistics p-values

VARIABLES Robust

Continued

Medium -0.0048 -0.1800 0.8580

(0.0270)

High -0.0117 -0.3400 0.7360

(0.0348)

R&D Intensity 0.1254 0.6000 0.5460

(0.2075)

Firm Size -0.0008 -0.0500 0.9630

(0.0164)

Firm Age -0.0106*** -4.1500 0.0000

(0.0026)

Knowledge Stock -0.0001* -1.8600 0.0630

(0.0000)

Leverage -0.0071 -0.6700 0.5020

(0.0106)

Public A-shares -0.0930 -1.2100 0.2280

(0.0771)

Long-term Investment -0.0154*** -2.9800 0.0030

(0.0052)

Constant 1.1882*** 3.7700 0.0000

(0.3153)

Observations 4,074

Number of Firms 958

Adjusted R-squared 0.0192

FIRM FE YES

Log-likelihood 1977

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The baseline is low-level concentrated ownership. The p-value of the medium-level is

greater than 0.1, which fails to reject the null hypothesis. It has the same situation for the

high level, which contains a p-value above 0.1. The experimental results disagree with the

previous hypothesis (i.e. H6a).

For control variables, the p-value of firm age is p < 0.01, which rejects the null

hypothesis. There is a significantly negative relationship between firm age and the depth

of innovation. The depth of innovation is expected to decrease by 0.0106 for one

additional year increased in firm age, ceteris paribus. The p-value of knowledge stock is p

< 0.1, which rejects the null hypothesis at the 10% level. The depth of innovation is

expected to decrease by 0.0001 as the total number of patents increases by 1 unit while

holding other variables constant.

The p-value of long-term investment is p < 0.01, which rejects the null hypothesis at the

1% level—long-term investment and the depth of innovation in significantly and

negatively related. Expected change in depth of innovation is -0.000154, if long-term

investment increases by 1%, ceteris paribus. The p-values of other variables are all greater

than 0.1, which fails to reject the null hypothesis.

The log-likelihood value, in this case, is 1977.

4.3.3.1.2 Diversity of Innovation and Concentrated Ownerships

i. Tests for Assumptions of Panel Data Regression

The p-value from the LM test is p < 0.01, which rejects the null hypothesis of pooled

regression. The HS test was applied afterwards and showed a p-value of p < 0.01,

rejecting the null hypothesis of random effects. Consequently, the fixed effects model was

applied.

After that, regression assumptions should be tested for consistent and unbiased estimators.

First, the zero mean residual implied that the assumption of linearity held (Assumption 1).

Second, the p-value of the modified Wald test was p < 0.01, which showed the

heteroscedastic residuals. Assumption 2 was violated. Third, the p-value of the run test of

randomness was p < 0.01, which rejected the null hypothesis (Assumption 3). To rephrase

it, there was a serial correlation in residuals. Fourth, the p-value of the Ramsey RESET

Test was p < 0.05, which failed to reject the null hypothesis at the 5% level (Assumption

4). It meant there was not a problem of misspecification. Fifth, residuals were

asymptotically normal due to a large sample size (Assumption 5). Finally, Assumption 6

does not violate because most correlations matrix in Appendix E remains at a low level

apart from the correlation between diversity and depth. However, it did not raise a

concern for multicollinearity, as those two were dependent variables.

Since residuals were heteroscedastic and serially correlated, the robust standard error was

applied.

ii. Results

Table 4.16 Diversity of Innovation for Concentrated Ownership

(12)

H6b

Diversity of Innovation

t-statistics

p-values

VARIABLES Robust

Medium

-0.0006

(0.0480)

-0.0100 0.9900

High

0.0075

(0.0617)

0.1200 0.9040

R&D Intensity

-0.1640

(0.3483)

-0.4700 0.6380

Firm Size

0.0274

(0.0304)

0.9000 0.3670

Firm Age

0.0197***

(0.0047)

4.1700 0.0000

Knowledge Stock

0.0002**

(0.0001)

2.4500 0.0150

Leverage

0.0039

(0.0183)

0.2100 0.8330

Public A-shares

0.1794

(0.1359)

1.3200 0.1870

Long-term Investment

0.0266***

(0.0091)

2.9200 0.0040

Constant

0.0719

(0.5896)

0.1200

0.9030

Observations 4,074

Number of Firms 958

Adjusted R-squared 0.0286

FIRM FE YES

Log-likelihood -328.6

The p-values of the medium-level and the high-level are both greater than 0.1. Both

hypotheses cannot be rejected. There is not a significant difference in the diversity of

R&D for the three levels of concentrated ownership, on average, when other variables are

held constant. The experimental results are inconsistent with the previous hypothesis

(i.e. H6b).

For control variables, the p-value of firm age is p < 0.01, which rejects the null

hypothesis. There is a significantly positive relationship between firm age and the

diversity of innovation. The expected change in the diversity of innovation is 0.0197 for

one additional year increase in firm age, ceteris paribus. The p-value of knowledge stock

is p < 0.05, which rejects the null hypothesis at the 5% level. Knowledge stock and the

diversity of innovation are positively related. Excepted change in the diversity of

innovation is 0.0002 for an additional one unit increased in knowledge stock while

holding other variables constant. The p-value of ln(long-term investment) is p < 0.01,

which rejects the null hypothesis at the 5% level. Ln(long-term investment) and the

diversity of innovation are significantly and positively related. If long-term investment

increases by 1%, the diversity of innovation is expected to increase by 0.000266, ceteris

paribus.

The log-likelihood value is -328.5954.

4.3.3.2 Insider Ownership

4.3.3.2.1 Depth of Innovation and Insider Ownerships

i. Tests for Assumptions of Panel Data Regression

A p-value of p < 0.01 for the LM test rejects the null hypothesis of pooled regression. The

HS test was then applied, showing a p-value of p < 0.01, rejecting the null hypothesis of a

random effect. Therefore, the fixed effects model was applied.

First, the zero mean residual implied that the assumption of linearity held (Assumption 1).

Second, the p-value of the modified Wald test was p < 0.01, which suggested

heteroscedastic residuals. Assumption 2 violated due to non-constant variance. Third, the

p-value of the run test of randomness was p < 0.01, which rejected the null hypothesis

(Assumption 3). In other words, residuals were autocorrelated. Fourth, the pvalue of the

Ramsey RESET Test was greater than 0.1, which failed to reject the null hypothesis

(Assumption 4). It meant that there is no misspecification error. Fifth, the total sample size

was large enough to say that residual distribution was asymptotically normal (Assumption

5). Finally, Assumption 6 holds due to a low level of correlations in Appendix E.

Due to the violations of assumptions, the robust standard error was applied to fix these

problems.

ii. Results

Table 4.17 Depth of Innovation for Insider Ownership

(13)

H7a

Depth of Innovation

t-statistics

p-values

VARIABLES Robust

Low

0.0118

(0.0155)

0.7500

0.4510

Medium

-0.0239

(0.0299)

-0.8300

0.4070

High

0.0159

(0.0394)

0.3700

0.7100

R&D Intensity

0.1169

(0.2066)

0.3600

0.7190

Firm Size

0.0008

(0.0166)

0.2800

0.7760

Firm Age

-0.0107***

(0.0026)

-4.4700

0.0000

Knowledge Stock

-0.0001*

(0.0000)

-1.8000

0.0720

Leverage

-0.0061

(0.0111)

0.0300

0.9750

Public A-shares

-0.1019

(0.0727)

-1.2400

0.2150

Long-term Investment

-0.0156***

(0.0052)

-3.2100

0.0010

Constant

1.1506***

(0.3172)

3.6700

0.0000

Observations

4,074

Number of Firms 958

Adjusted R-squared 0.0200

FIRM FE YES

Log-likelihood 1980

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The baseline is zero-level insider ownership.

The p-values for all three levels in the table above are greater than 0.1, indicating that

there is no significant difference between the R&D intensity of insider ownership and

non-insider ownership for the three levels, on average, all else being equal. The

conclusions show inconsistency with hypothesis H7a.

For control variables, the p-value of firm age is p < 0.01, which rejects the null

hypothesis. There is a significantly negative relationship between firm age and the depth

of innovation. The depth of innovation is expected to decrease by 0.0107 for one

additional year increased in firm age, ceteris paribus. The p-value of knowledge stock is p

< 0.1, which rejects the null hypothesis at the 10% level. The mean depth of innovation

decreases by 0.0001, as knowledge increases by one unit, ceteris paribus.

The p-value of long-term investment is p < 0.01, which rejects the null hypothesis at the

1% level—long-term investment and the depth of innovation in significantly and

negatively related. Expected change in depth of innovation is -0.000156, if long-term

investment increases by 1%, ceteris paribus. The p-values of other variables are all greater

than 0.1, which fails to reject the null hypothesis.

The log-likelihood value is 1980.

4.3.3.2.2 Diversity of Innovation and Insider Ownerships

i. Tests for Assumptions of Panel Data Regression

The LM test showed a p-value of p < 0.01, rejecting the null hypothesis of pooled

regression. The HS test was then applied, showing a p-value of p < 0.01, rejecting the null

hypothesis of a random effect. Therefore, a fixed effects model was applied.

The regression assumptions were tested, and the results were as follows.

First, the zero mean residual implied that the assumption of linearity held (Assumption 1).

Second, the p-value of the modified Wald test was p < 0.01, which showed the

heteroscedastic residuals. It then violated Assumption 2. Third, the p-value of the run test

of randomness was p < 0.01, which rejected the null hypothesis (Assumption 3). To put it

differently, residuals were serially correlated. Fourth, the p-value of the Ramsey RESET

Test was greater than 0.05, which failed to reject the null hypothesis at the 5% level

(Assumption 4). It meant there was not a problem of misspecification. Fifth, residuals

were asymptotically normal due to a large sample size (Assumption 5). There was no

perfect multicollinearity due to pretty low correlations in Appendix E

(Assumption 6).

Then, the robust standard error was utilised for the fixed effects model.

ii. Results

Table 4.18 Diversity of Innovation for Insider Ownership

(14)

H7b

Diversity of Innovation

t-statistics

p-values

VARIABLES Robust

Low

-0.0206

(0.0281)

-0.7200

0.4690

Medium

0.0517

(0.0522)

1.0300

0.3050

High

-0.0145

(0.0668)

-0.1800

0.8550

R&D Intensity

-0.1400

(0.3484)

-0.1800

0.8540

Firm Size

0.0245

(0.0307)

0.6400

0.5220

Firm Age

0.0199***

(0.0047)

4.5300

0.0000

Knowledge Stock

0.0002**

(0.0001)

2.4100

0.0160

Leverage

0.0028

(0.0192)

-0.3700

0.7090

Public A-shares

0.1831

(0.1284)

1.2700

0.2040

Long-term Investment 0.0270*** 3.1500 0.0020

(0.0092) -0.7200 0.4690

Constant

0.1346

(0.5930)

Observations

4,074

Number of Firms 958

Adjusted R-squared 0.0296

FIRM FE YES

Log-likelihood -326.0

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The baseline is zero level of insider ownership.

The p-values for all three levels in the table above are all greater than 0.1, denoting that

the R&D intensity between insider ownership and non-insider ownership is not

significantly different at the three levels, on average, ceteris paribus. The conclusions are

not in agreement with hypothesis H7b.

For control variables, the p-value of firm age is p < 0.01, which rejects the null

hypothesis. There is a significantly positive relationship between firm age and the

diversity of innovation. The expected change in the diversity of innovation is 0.0199 for

one additional year increase in firm age, ceteris paribus. The p-value of knowledge stock

is p < 0.05, which rejects the null hypothesis at the 5% level. Knowledge stock and the

diversity of innovation are positively related. Excepted change in the diversity of

innovation is 0.0002 for an additional one unit increased in knowledge stock. The pvalue

of ln(long-term investment) is p < 0.01, which rejects the null hypothesis at the 1% level.

Ln(long-term investment) and the diversity of innovation are significantly and positively

related. If long-term investment increases by 1%, the diversity of innovation is expected

to increase by 0.000270, ceteris paribus.

The log-likelihood value is -326.

4.3.3.3 State Ownership

4.3.3.3.1 Depth of Innovation and State Ownerships

i. Tests for Assumptions of Panel Data Regression

The p-value of the LM test is p < 0.01, rejecting the null hypothesis of pooled regression.

The HS test is then applied, showing a p-value of p < 0.01, rejecting the null hypothesis of

a random effect. Therefore, a fixed effects model is applied.

After that, the tests of regression assumptions were applied.

First, the zero mean residual implied that the assumption of linearity held (Assumption 1).

Second, the p-value of the modified Wald test was p < 0.01, which suggested

heteroscedastic residuals. It then violates Assumption 2. Third, the p-value of the run test

of randomness was p < 0.01, rejecting the null hypothesis (Assumption 3). To rephrase it,

there was no serial correlation in residuals. Fourth, the p-value of the

Ramsey RESET Test was greater than 0.1, which failed to reject the null hypothesis

(Assumption 4). It meant that there was no misspecification error. Fifth, the total sample

size was large enough to say that residual distribution was asymptotically normal

(Assumption 5). Consequently, the assumption of normality held. Sixth, Appendix E

indicates no multicollinearity due to low correlations (Assumption 6).

Due to heteroscedastic and serially correlated residuals, the fixed effects model was

modified by robust standard error.

ii. Results

Table 4.19 Depth of Innovation for State Ownership

(15)

H8a

Depth of Innovation t-statistics p-values

VARIABLES Robust

Low -0.0135 -0.7800 0.4340

(0.0173)

Medium -0.0081 -0.2900 0.7740

(0.0281)

High -0.0242 -1.0300 0.3040

(0.0235)

R&D Intensity 0.1256 0.6000 0.5500

(0.2100)

Firm Size 0.0009 0.0500 0.9560

(0.0166)

Firm Age -0.0106*** -4.1800 0.0000

(0.0025)

Knowledge Stock -0.0001* -1.8700 0.0620

(0.0000)

Leverage -0.0059 -0.5600 0.5750

(0.0105)

Public A-shares -0.1043 -1.4300 0.1530

(0.0728)

Long-term Investment -0.0155*** -2.9900 0.0030

(0.0052)

Continued

Constant 1.1625*** 3.6900 0.0000

(0.3153)

Observations 4,074

Number of Firms 958

Adjusted R-squared 0.0193

FIRM FE YES

Log-likelihood 1978

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The baseline is zero-level state ownership.

All p-values for three levels of state ownership are greater than 0.1 in the table above. No

significant difference in state ownership compared to non-state ownership for all three

levels of innovation specialisation, on average, all else being equal. Thus, the findings are

not consistent with hypothesis H8a.

For control variables, the p-value of firm age is p < 0.01, which rejects the null

hypothesis. There is a significantly negative relationship between firm age and the depth

of innovation. The depth of innovation is expected to decrease by 0.0106 for one

additional year increased in firm age, ceteris paribus. The p-value of knowledge stock is p

< 0.1, which rejects the null hypothesis at the 10% level. The mean depth of innovation

decreases by 0.0001 as knowledge increases by one unit, ceteris paribus. The p-value of

long-term investment is p < 0.01, which rejects the null hypothesis at the 1% level. Long-

term investment and the depth of innovation in significantly and negatively related.

Expected change in depth of innovation is -0.000155 if long-term investment increases by

1%, ceteris paribus. The p-values of other variables are all greater than 0.1, which fails to

reject the null hypothesis.

The log-likelihood value is 1978.

4.3.3.3.2 Diversity of Innovation and State Ownerships

i. Tests for Assumptions of Panel Data Regression

The p-value for the LM test was p < 0.01, rejecting the null hypothesis of pooled

regression. The HS test was then applied, showing a p-value of p < 0.01, rejecting the null

hypothesis of a random effect. Therefore, a fixed effects model was applied.

Following this, a test of the regression hypothesis was applied.

First, the zero mean residual implied that the assumption of linearity holed (Assumption

1). Second, the p-value of the modified Wald test was p < 0.01, which suggested

heteroscedastic residuals. It violated Assumption 2. Third, the p-value of the run test of

randomness was p < 0.01, rejecting the null hypothesis (Assumption 3). There was a serial

correlation in residuals. Fourth, the p-value of the Ramsey RESET Test was greater than

0.01, which failed to reject the null hypothesis at the 1% level (Assumption

4). It meant there is not a problem of misspecification. Fifth, residuals were

asymptotically normal due to a large sample size (Assumption 5). Sixth, Assumption 6

holds due to low correlations in Appendix E.

Because of heteroscedastic residuals, the fixed effects model was re-regressed by the

robust standard error.

ii. Results

Table 4.20 Diversity of Innovation for State Ownership

(16)

H8b

Diversity of Innovation t-statistics p-values

VARIABLES Robust

Low 0.0224 0.7200 0.4700

(0.0309)

Medium 0.0316 0.6200 0.5320

(0.0506)

High 0.0466 1.1600 0.2480

(0.0403)

R&D Intensity -0.1480 -0.4200 0.6750

(0.3531)

Firm Size 0.0249 0.8100 0.4160

(0.0307)

Firm Age 0.0196*** 4.2000 0.0000

(0.0047)

Knowledge Stock 0.0002** 2.4500 0.0150

(0.0001)

Continued

Leverage 0.0012 0.0700 0.9460

(0.0183)

Public A-shares 0.1866 1.4500 0.1480

(0.1287)

Long-term Investment 0.0268*** 2.9100 0.0040

(0.0092)

Constant 0.1014 0.1700 0.8630

(0.5894)

Observations 4,074

Number of Firms 958

Adjusted R-squared 0.0288

FIRM FE YES

Log-likelihood -327.7

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The baseline is zero level of state ownership.

The p-values for three levels of state ownership are greater than 0.1, so null hypotheses

cannot be rejected. All three levels of state ownership show no significant difference in

R&D intensity from non-state ownership when comparing state and non-state ownership,

on average, ceteris paribus. The results are not in line with hypothesis H8b. For control

variables, the p-value of firm age is p < 0.01, which rejects the null hypothesis. There is a

significantly positive relationship between firm age and the diversity of innovation. The

expected change in the diversity of innovation is 0.0196 for one additional year increase

in firm age, ceteris paribus. The p-value of knowledge stock is p < 0.05, which rejects the

null hypothesis at the 5% level. Knowledge stock and the diversity of innovation are

positively related. Excepted change in the diversity of innovation is 0.0002 for an

additional one unit increased in knowledge stock. The pvalue of ln(long-term investment)

is p < 0.01, which rejects the null hypothesis at the 1% level. Ln(long-term investment)

and the diversity of innovation are significantly and positively related. If long-term

investment increases by 1%, the diversity of innovation is expected to increase by

0.000268, ceteris paribus.

The log-likelihood value is -327.7.

4.3.3.4 Institutional Ownership

4.3.3.4.1 Depth of Innovation and Institutional Ownerships

i. Tests for Assumptions of Panel Data Regression

A p-value of p < 0.01 for the LM test rejected the null hypothesis of pooled regression.

The HS test was then applied, showing a p-value of p < 0.01, rejecting the null hypothesis

of a random effect. Therefore, a fixed effects model was applied.

Afterwards, a test of the regression hypothesis was applied.

First, the zero mean residual implied that the assumption of linearity held (Assumption 1).

Second, the p-value of the modified Wald test was p < 0.01, which meant heteroscedastic

residuals(Assumption 2). Third, the p-value of the run test of randomness was p < 0.01,

which rejected the null hypothesis (Assumption 3). To rephrase it, there was

autocorrelation in residuals. Fourth, the p-value of the Ramsey

RESET Test was greater than 0.1, which failed to reject the null hypothesis (Assumption

4). Hence, there was no misspecification error. Fifth, the total sample size was large

enough to say that residual distribution was asymptotically normal (Assumption 5).

Finally, multicollinearity was not a concern due to relatively low correlations in Appendix

Since some assumptions of panel data regression were violated, the robust standard error

modified the fixed effects model.

ii. Results

Table 4.21 Depth of Innovation for Institutional Ownership

(17)

H9a

Depth of Innovation t-statistics p-values

VARIABLES Robust

Low 0.0083 0.5400 0.5920

(0.0154)

Medium 0.0012 0.0700 0.9450

(0.0172)

High -0.0098 -0.4000 0.6860

(0.0242)

Continued

R&D Intensity 0.1211 0.5800 0.5590

(0.2073)

Firm Size -0.0010 -0.0600 0.9540

(0.0166)

Firm Age -0.0104*** -4.0700 0.0000

(0.0026)

Knowledge Stock -0.0001* -1.8500 0.0650

(0.0000)

Leverage -0.0073 -0.7100 0.4790

(0.0103)

Public A-shares -0.0849 -1.1800 0.2360

(0.0716)

Long-term Investment -0.0154*** -2.9800 0.0030

(0.0052)

Constant 1.1750*** 3.7400 0.0000

(0.3145)

Observations 4,074

Number of Firms 958

Adjusted R-squared 0.0193

FIRM FE YES

Log-likelihood 1978

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The baseline is zero-level institutional ownership.

The p-values for all three levels in the table above are greater than 0.1, indicating no

significant difference between the R&D intensity of institutional ownership and

noninstitutional ownership for the three levels, on average, all else equal. The conclusions

show inconsistency with hypothesis H9a.

For control variables, the p-value of firm age is p < 0.01, which rejects the null

hypothesis. There is a significantly negative relationship between firm age and the depth

of innovation. The depth of innovation is expected to decrease by 0.0104 for one

additional year increased in firm age, ceteris paribus. The p-value of knowledge stock is p

< 0.1, which rejects the null hypothesis at the 10% level. The mean depth of innovation

decreases by 0.0001 as knowledge increases by one unit, ceteris paribus. The p-value of

long-term investment is p < 0.01, which rejects the null hypothesis at the 1% level—long-

term investment and the depth of innovation in significantly and negatively related.

Expected change in depth of innovation -0.000154, if long-term investment increases by

1%, ceteris paribus. The p-values of other variables are all greater than 0.1, which fails to

reject the null hypothesis.

The log-likelihood value is 1978.1874.

4.3.3.4.2 Diversity of Innovation and Institutional Ownerships

i. Tests for Assumptions of Panel Data Regression

LM tests showed a p-value of p < 0.01, rejecting the null hypothesis of pooled regression.

Then an HS test was applied, showing a p-value of p < 0.01, rejecting the null hypothesis

of a random effect. Thus, a fixed effects model was applied.

Subsequently, a test of the regression hypothesis was applied.

First, the zero mean residual implied that the assumption of linearity held (Assumption 1).

Second, the p-value of the modified Wald test was p < 0.01, which showed the

heteroscedastic residuals (Assumption 2). Third, the p-value of the run test of randomness

was p < 0.01, which rejected the null hypothesis (Assumption 3). There was a serial

correlation in residuals. Fourth, the p-value of the Ramsey RESET Test was greater than

0.01, which failed to reject the null hypothesis at the 1% level (Assumption 4). It meant

there was not a problem of misspecification. Fifth, residuals were asymptotically normal

due to the large sample size (Assumption 5). Sixth, Appendix E displayed correlations at a

low level, and hence there was no concern about perfect multicollinearity (Assumption 6).

The fixed effects model was then modified by the robust standard error.

ii. Results

Table 4.22 Diversity of Innovation for Institutional Ownership

(18)

H9b

Diversity of Innovation t-statistics p-values

VARIABLES Robust

Continued

Low -0.0161 -0.6000 0.5470

(0.0267)

Medium -0.0054 -0.1800 0.8590

(0.0302)

High 0.0164 0.3800 0.7030

(0.0430)

R&D Intensity -0.1527 -0.4400 0.6610

(0.3478)

Firm Size 0.0279 0.9100 0.3620

(0.0306)

Firm Age 0.0194*** 4.1100 0.0000

(0.0047)

Knowledge Stock 0.0002** 2.4500 0.0140

(0.0001)

Leverage 0.0038 0.2100 0.8320

(0.0180)

Public A-shares 0.1559 1.2300 0.2210

(0.1273)

Long-term Investment 0.0267*** 2.9200 0.0040

(0.0091)

Constant 0.0868 0.1500 0.8820

(0.5851)

Observations 4,074

Number of Firms 958

Adjusted R-squared 0.0287

FIRM FE YES

Log-likelihood -327.8

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The baseline is the zero level of institutional ownership.

All three levels in the table above have p-values greater than 0.1, indicating that, other

things being equal, there is no significant difference between the R&D intensity of

institutional and non-institutional ownership at the three levels, on average. The

conclusions are shown to be inconsistent with Hypothesis H9b.

For control variables, the p-value of firm age is p < 0.01, which rejects the null

hypothesis. There is a significantly positive relationship between firm age and the

diversity of innovation. The expected change in the diversity of innovation is 0.0194 for

one additional year increase in firm age, ceteris paribus. The p-value of knowledge stock

is p < 0.05, which rejects the null hypothesis at the 5% level. Knowledge stock and the

diversity of innovation are positively related. Excepted change in the diversity of

innovation is 0.0004 for an additional one unit increased in knowledge stock. The pvalue

of ln(long-term investment) is p < 0.01, which rejects the null hypothesis at the 1% level.

Ln(long-term investment) and the diversity of innovation are significantly and positively

related. If long-term investment increases by 1%, the diversity of innovation is expected

to increase by 0.000267, ceteris paribus.

The log-likelihood value is -327.78.

4.3.3.5 Foreign Ownership

4.3.3.5.1 Depth of Innovation and Foreign Ownerships

i. Tests for Assumptions of Panel Data Regression

The p-value for the LM test was p < 0.01, rejecting the null hypothesis of pooled

regression. This was followed by the application of the HS test, which showed a p-value

of p < 0.01, rejecting the null hypothesis of a random effect. Thus, a fixed effects model

was applied.

After that, a test of the regression hypothesis was applied.

First, the zero mean residual implied that the assumption of linearity held (Assumption 1).

Second, the p-value of the modified Wald test was p < 0.01, rejecting the null hypothesis

of homoscedasticity (Assumption 2). Third, the p-value of the run test of randomness was

p < 0.01, which rejected the null hypothesis (Assumption 3). To rephrase it, there was a

serial correlation in residuals. Fourth, the p-value of the Ramsey

RESET Test was greater than 0.1, which failed to reject the null hypothesis (Assumption

4). It meant that there was no misspecification error. Fifth, the total sample size was large

enough to say that residual distribution was asymptotically normal (Assumption

5). Sixth, the assumption of no perfect multicollinearity holds since correlations in

Appendix E remain low (Assumption 6).

Due to the violation of assumptions, the robust standard error was applied to fix these

problems.

ii. Results

Table 4.23 Depth of Innovation for Foreign Ownership

(1)

H10a

Depth of Innovation

t-statistics

p-values

VARIABLES Robust

Low

-0.0145

(0.0107)

-1.3500

0.1770

Medium

0.0139

(0.0251)

0.5500 0.5800

High

-0.0179

(0.0500)

-0.3600 0.7210

R&D Intensity

0.1077

(0.2109)

0.5100 0.6100

Firm Size

-0.0015

(0.0164)

-0.0900 0.9270

Firm Age

-0.0101***

(0.0026)

-3.9200 0.0000

Knowledge Stock

-0.0001*

(0.0000)

-1.8800 0.0610

Leverage

-0.0077

(0.0104)

-0.7300 0.4630

Public A-shares

-0.1097

(0.0755)

-1.4500 0.1470

Long-term Investment

-0.0152***

(0.0051)

-2.9600 0.0030

Constant

1.1986***

(0.3122)

3.8400

0.0000

Observations

4,074

Number of Firms 958

Adjusted R-squared 0.0200

FIRM FE YES

Log-likelihood 1980

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The baseline is zero-level foreign ownership. All p-values for three levels of foreign

ownership are greater than 0.1 in the table above. On average, there is no significant

difference in foreign ownership compared to non-foreign ownership for all three

innovation specialisation levels. Thus, the findings are not consistent with hypothesis

H10a.

For control variables, the p-value of firm age is p < 0.01, which rejects the null

hypothesis. There is a significantly negative relationship between firm age and the depth

of innovation. The depth of innovation is expected to decrease by 0.0101 for one

additional year increased in firm age, ceteris paribus. The p-value of knowledge stock is p

< 0.1, which rejects the null hypothesis at the 10% level. The mean depth of innovation

decreases by 0.0001, as knowledge increases by one unit, ceteris paribus.

The p-value of long-term investment is p < 0.01, which rejects the null hypothesis at the

1% level—long-term investment and the depth of innovation in significantly and

negatively related. The expected change in depth of innovation is -0.000152, if longterm

investment increases by 1%, ceteris paribus. The p-values of other variables are all greater

than 0.1, which fails to reject the null hypothesis.

The log-likelihood value is 1980.

4.3.3.5.2 Diversity of Innovation and Foreign Ownerships

i. Tests for Assumptions of Panel Data Regression

The LM test showed a p-value of p < 0.01, and the null hypothesis of pooled regression

was rejected. The HS test was then applied, showing a p-value of p < 0.01, rejecting the

null hypothesis of a random effect. Hence, a fixed effects model was applied.

This was followed by the application of a test of the regression hypothesis.

First, the zero mean residual implied that the assumption of linearity held (Assumption 1).

Second, the p-value of the modified Wald test was p < 0.01, rejecting the null hypothesis.

In other words, residuals were not constant (Assumption 2). Third, the pvalue of the run

test of randomness was p < 0.01, which rejected the null hypothesis

(Assumption 3). There was a serial correlation in residuals. Fourth, the p-value of the

Ramsey RESET Test was greater than 0.1, which failed to reject the null hypothesis at the

1% level (Assumption 4). It meant there was not a problem of misspecification.

Fifth, residuals were asymptotically normal due to the large sample size (Assumption 5).

Sixth, the pairwise correlation matrix in Appendix E indicates no perfect multicollinearity

(Assumption 6).

Since some assumptions of panel data regression were violated, the fixed effects model

was modified by the robust standard error.

ii. Results

Table 4.24 Diversity of Innovation for Foreign Ownership

(2)

H10b

Diversity of Innovation

t-statistics

p-values

VARIABLES Robust

Low

0.0287

(0.0194)

1.4800

0.1400

Medium

-0.0272

(0.0441)

-0.6200 0.5380

High

0.0024

(0.0889)

0.0300 0.9780

R&D Intensity

-0.1425

(0.3539)

-0.4000 0.6870

Firm Size

0.0279

(0.0301)

0.9300 0.3540

Firm Age

0.0187***

(0.0047)

3.9600 0.0000

Knowledge Stock

0.0002**

(0.0001)

2.4600 0.0140

Leverage

0.0048

(0.0180)

0.2700 0.7890

Public A-shares

0.2066

(0.1325)

1.5600 0.1190

Long-term Investment

0.0265***

(0.0091)

2.9200 0.0040

Constant

0.0627

(0.5789)

0.1100

0.9140

Observations 4,074

Number of Firms 958

Adjusted R-squared 0.0295

FIRM FE YES

Log-likelihood -326.2

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The baseline is non-foreign ownership.

The p-values for three levels of foreign ownership are higher than 0.1, so null hypotheses

cannot be rejected. All three levels of foreign ownership show no significant difference in

R&D intensity from non-foreign ownership when comparing foreign and non-foreign

ownership, on average, all other variables are equal. The results are not in line with

hypothesis H10b.

For control variables, the p-value of firm age is p < 0.01, which rejects the null

hypothesis. There is a significantly positive relationship between firm age and the

diversity of innovation. The expected change in the diversity of innovation is 0.0187 for

one additional year increased in firm age, ceteris paribus. The p-value of knowledge stock

is p < 0.05, which rejects the null hypothesis at the 5% level. Knowledge stock and the

diversity of innovation are positively related. Excepted change in the diversity of

innovation is 0.0002 for an additional one unit increased in knowledge stock. The pvalue

of ln(long-term investment) is p < 0.01, which rejects the null hypothesis at the 1% level.

Ln(long-term investment) and the diversity of innovation are significantly and positively

related. If long-term investment increases by 1%, the diversity of innovation is expected

to increase by 0.000265, ceteris paribus.

The log-likelihood value is -326.33393

4.3.3.6 Discussion of Depth of Innovation and Diversity of Innovation

None of the firm ownership shows a significant effect on innovation specialisation or

diversification. The findings are inconsistent with Hypotheses H6a – H10b. It reveals that

innovative specialisation or diversification is not determined by firm ownership but other

factors.

One interpretation is a national demand. SOEs, as state-controlled enterprises, undertake

to accept the tasks assigned to them by the state. State-owned enterprises operate in a

wide range of markets, including those with profound barriers to private companies

(Marrelli et al., 1998). These include knowledge barriers, trade barriers, political barriers,

et cetera. Because of those barriers, SOEs' task is to promote commercial or technological

developments in those areas, such as the nuclear power industry or outer space

exploration. The decision to specify or diversify innovation in SOEs is, therefore, likely to

be determined by the government, rather than by individuals or firm management.

Second, innovation specialisation or diversification depends on the inventor. Boh et al.

(2014) divide inventors into three categories - generalists, specialists and polymaths - and

point out that different types of inventors contribute differently to organisations.

Specialists have a depth of knowledge that allows them to specialise in innovation and

obtain technological innovations that have a high impact. On the other hand, the breadth

of knowledge of the generalists and polymaths allows them to have many ideas and to

explore and innovate in different fields, obtaining patents in several areas. Third, the

direction of innovation may also be influenced by national policy. For example,

companies may prefer to conduct R&D in industries that are strongly supported by the

state as a way to access better resources.

Fourth, the thresholds used for classifying different levels of firm ownership may not

provide a good explanation of the impact of different levels of ownership on innovation

specialisation or diversification. The results might be improved by other thresholds. Fifth,

the measurement of a firm's innovation specialisation and innovation diversity is based on

the total number of patents, whereas some firms may choose not to patent their advanced

technologies in order to safeguard them, with industrial secrecy as an example (Archibugi

and Pianta, 1996; Michie, 1998; Kleinknecht et al., 2002). So using the number of patents

to measure innovation and the firm's knowledge base may be biased.

4.4 Summary of Data Analysis

The relationship between concentrated ownership and R&D intensity is positive. The

higher the level of concentrated ownership, the higher the R&D intensity. It disagrees

with the hypothesis suggested in Chapter 3 (i.e. H1). Moreover, different levels of

concentrated ownership do not behave differently in terms of innovation specialisation

and innovation diversification. Hence, the results are inconsistent with hypotheses H6a

and H6b.

Regressions from continuous and categorical insider ownership generate results

inconsistent with H2. Despite the fact that insider ownership is less developed in China,

insider ownership is more conducive to innovation performance than in firms without

insider ownership, when insider ownership is greater than 20% or less than 5%. In terms

of R&D intensity, firms with insider ownership over 20% are more advantageous than

those with a proportion below 5%.

All three levels of insider ownership are indistinguishable from non-insider ownership

in specialised or diversified innovation. That is not following the previous hypotheses

H7a and H7b.

The result with continuous state ownership is not in line with hypothesis. Nevertheless,

the result with categorical state ownership partly agrees with hypothesis H3. It reveals

that there is a positive impact of state ownership on innovation performance when the

level of state ownership is below 5%. In terms of R&D intensity, none of the other three

levels of ownership differs significantly from non-state ownership. Hence, that is at odds

with the previous hypotheses H8a and H8b.

Institutional ownership and R&D intensity are not related. Therefore, both regression

results, including continuous or categorical institutional ownership, are not in accordance

with hypothesis H4. Furthermore, The other three levels of ownership are not

significantly different from non-institutional ownership in terms of R&D intensity.

This is at odds with the previous hypotheses H9a and H9b.

Regressions with continuous and categorical foreign ownership show that foreign

ownership and R&D intensity are negatively related, which is inconsistent with

hypothesis H5.

Moreover, there is no significant difference between foreign and non-foreign ownership in

innovation specialisation or diversification. Neither of those conclusions is compatible

with the hypothesis (H10a and H10b).

Chapter 5 Conclusion

5.1 Introduction

This chapter concludes main findings (i.e. Section 6.2), and provide contributions (i.e.

Section 6.3), limitations (i.e. Section 6.4) and recommendations (i.e. Section 6.5).

5.2 Main Findings

5.2.1 Introduction of Main Findings

This sub-section comprehensively summarises how firm ownership affects firm

innovation based on the data from the Shanghai Stock Exchange between 2013 and 2019.

The sub-section 6.2.2 focuses on the R&D intensity of a firm. The sub-section

6.2.3 deals with specialised and diversified innovation for a firm.

5.2.2 Innovation Performance

5.2.2.1 Concentrated Ownership

Concentrated ownership has a significant positive effect on R&D intensity. The more

concentrated ownership, the higher the R&D intensity. It follows agency theory when

high concentrations of ownership favour innovative performance. Such positive effect

benefits from efficient monitoring or incentive schemes by diminishing the conflict

between agents' and principals' interests and regulating agents' (managers') behaviour (e.g.

Berle and Means, 1932; Jensen and Meckling, 1976; Boučková, 2015; Feldman and

Montgomery, 2015).

While Yusuf et al. (2018) point out that the prevalence of high concentrated ownership,

strong family control and mismanagement in developing country firms weakens the

foundation of agency theory, the data from the study (Appendix B) indicates that only

about one-third of the sample size contains concentrated ownership above 20% and only

less than one-fifth of the sample size contains insider ownership (including family

ownership) above 20%. Indeed, highly concentrated ownership does exist in Chinese

listed firms, but in terms of data, they only account for approximately one-third of firms.

Moreover, the proportion of family ownership involved in insider ownership is even

smaller. Thus, the high concentration of ownership and strong family control proposed by

Yusuf et al. (2018) is less common in Chinese listed firms and does not affect the

conclusions of agency theory on the concentration of ownership in terms of innovation.

Another interpretation is that a high degree of concentrated ownership assists in the

optimal allocation and integration of internal or external resources (Lacetera, 2001). Even

though a change in innovation performance is not based on different levels of

concentrated ownership, it does not mean concentrated ownership has an indifferent

innovation performance with non-concentrated ownership.

In sum, concentred ownership benefits innovation performance, and that is inconsistent

with hypothesis H1.

5.2.2.2 Insider Ownership

Insider ownership and R&D intensity are not related. Reasons include insider ownership

being less developed in China (Choi et al., 2011), managers focusing on public relations

rather than performance (Bisot and Child, 1996; Xin and Pearce, 1996; Peng, 2000), and

the imprecision of agency theory in explaining the relationship between insider ownership

and innovation (Suk et al., 2012). The data (Appendix B) supports Choi et al. (2011)’s

claim that over one-third of the companies are non-insider owned. About half of the

observed data hold less than 20% insider ownership (excluding non-insider ownership). It

highlights that underdeveloped insider ownership is still a common phenomenon among

listed firms in China.

In addition, high and low levels of insider ownership are more conducive to innovation

than non-insider ownership, and even the high-level is more predominant than the

lowlevel. It may be caused by the fact that firm owners have long-term aspirations for the

company, and employees actively seek innovation in order to achieve self-worth and

obtain stable employment.

Another reason is that the low level of shareholdings may be held by employees who have

contributed significantly to the firm. Equity incentives also motivate other employees in

the company to innovate actively, thus creating a positive correlation between a low level

of insider ownership and innovation.

On the other hand, insider ownership positively affects innovation performance due to the

efficient monitoring scheme at the high level of firm ownership.

In short, the low and high levels of insider ownership are beneficial to innovation

performance, although insider ownership is relatively uncommon among Chinese listed

companies. The thresholds for the positive impact of insider ownership on innovation

performance are the level of insider ownership below 5% and greater than 20%. Both

regression results deny hypothesis H2.

5.2.2.3 State Ownership

Unexpectedly, there is a non-relationship between state ownership and R&D intensity.

One interpretation is that listed state-owned companies may raise funds from the market

and be less reliant on state resources than unlisted state-owned companies (Zhou et al.,

2017). It may also be explained by an absence from the list of highly sophisticated

stateowned enterprises that have invested heavily in innovation owing to technological

secrecy as well as political and commercial reasons.

In terms of different levels of state ownership, it offers a different perspective with the

low-level state ownership promoting innovation, unlike the medium-level and the

highlevel. A low level of state ownership (i.e. < 5%) can exploit resource benefits without

concern about the state seizing control of the firm.

In summary, regression results partly agree with hypothesis H3. When state ownership is

less than 5%, there is a positive influence of state ownership on innovation performance.

5.2.2.4 Institutional Ownership

Institutional ownership and R&D intensity are not related, as regression analyses indicate

it in both continuous and categorical institutional ownership. While almost all listed firms

are held by institutions, only about a fifth of the total sample size has more than 20%

institutional ownership. If institutional ownership is low, institutional investors will not be

able to actively monitor the firm to alleviate managers' concerns about R&D failure

(Bushee, 1998). In addition, most institutional investors are shortterm financial investors.

If the holding period is too short, institutional investors are unable to reap the long-term

benefits of their R&D investments.

To sum up, there is no relationship between institutional ownership and innovation

performance. Results from the regressions with continuous institutional ownership and

categorical institutional ownership are not in accordance with the hypothesis H4.

5.2.2.5 Foreign Ownership

Both regression analyses of continuous and categorical foreign ownership illustrate a

negative impact of foreign ownership on R&D intensity. However, there is no association

between the low and medium levels of foreign ownership with innovation intensity, but

the high level of foreign ownership has a significant negative impact on innovation

intensity.

Owing to the costs and resources associated with research and development, subsidiaries

invested by foreign firms are more likely to acquire advanced technologies directly from

foreign parent firms than innovating locally. Furthermore, most foreign investors keep a

low level of firm shares. As a consequence, those foreign investors may not offer critical

technology or other resources for innovation activities, unlike multinational firms (Teece,

1986).

Briefly speaking, regression results of the negative relationship between foreign

ownership and innovation performance disagree with hypothesis H5.

5.2.3 Depth of Innovation and Diversity of Innovation

The empirical results suggest that firm ownership is not related to innovation

specialisation or diversification. Knowledge acquisition may be dependent on firm

ownership, for example spillover effect of foreign direct investment. The decision to

specialise or diversify innovation may, however, be more of an external one. It could be

national demand, national policy, the knowledge base of the inventor, or inappropriate

thresholds used in classifying firm ownership.

5.2.4 Control Variables

5.2.4.1 Firm Size

Firm size plays a vital role in R&D intensity. The larger the firm, the less willing it is to

innovate. It is possible that large firms have complex internal organisational structures

and heavy bureaucracy, and react slowly when the market environment changes. If the

market does not allow outsiders or makes it difficult for them to gain a foothold in the

market, then the vested interests of the large companies drive them to maintain the status

quo and to stop focusing on the actual and potential interests of their customers.

5.2.4.2 Firm Age

Firm age plays a vital role in R&D intensity, the depth of innovation and the diversity of

innovation. Nevertheless, such effects of firm age are trivial.

R&D intensity increases as firm age increases. According to the learning-by-doing model,

the older a firm is, the more business experience and foresight it has (Arrow, 1962;

Sorensen and Stuart, 2000; Chang et al., 2002).

Firm age harms the depth of innovation but benefits the diversity of innovation. The

primary goal of young firms is to survive in the market, and so that they only can R&D in

a single field with limited resources. Over time, they grow from a small firm to a large

firm with sufficient resources, capability and capacity to develop in multiple areas.

5.2.4.3 Long-term Investment

The minor and positive effect of long-term investment on R&D intensity suggests that

innovation performance increases as long-term investment increases. The long-term

investment could be an R&D investment if investing in patents, research institutions and

experimental equipment, for example.

Regardless of ownership, long-term investment has a positive effect on diversified

innovation but rather a negative effect on specialised innovation. Even if the effect is

paltry, it reiterates that firms are willing to research in different fields. As a result, it may

benefit the firm and innovation performance in the long term.

5.2.4.4 Knowledge Stock

Knowledge stock is the total number of patents firms hold. An increase in the total

number of patents may foster firms to widen the type of patents.

5.2.4.5 Industry Dummies

Full names of the industry codes are described in the Research methodology (i.e. Table

3.3). The baseline is agriculture, forestry, animal husbandry and fisheries. No matter what

the firm ownership is, firms in the following fields have a greater R&D intensity than the

baseline: (1) manufacturing industry; (2) information transmission, software and

information technology services; (3) scientific research and technical service industry; (4)

education industry; and (5) health and social work. One interpretation is that an equity-

compensation scheme is prevalent in those industries to promote innovative performance.

R&D intensity in the wholesale and retail sector are below the baseline in all five

ownerships.

5.3 Contributions

Firstly, this study presents a comprehensive analysis of the impact of firm ownership on

innovation performance in the context of One Belt and One Road, using the most recent

data available.

Secondly, firm ownership is divided into different levels, and innovation performance

differences between different groups are examined to derive thresholds of firm ownership.

Based on the above two points, the results show that concentrated ownership has a

significantly positive impact on innovation, which is different from the results of previous

studies (Wen et al., 2016). Insider ownership below 5% or above 20% has a positive

impact on innovation performance. In particular, insider ownership over 20% has a higher

impact on innovation performance than insider ownership below 5%. The outcome partly

agrees with Song et al. (2015)’s work. Furthermore, state ownership is positively

associated with innovation performance only when it is less than 5%, which is partially in

line with previous research findings (Choi et al., 2011). In addition, the results show that

institutional ownership and innovation performance are not related, which is inconsistent

with the view of Rong et al. (2017). Finally, foreign ownership is unexpectedly negatively

related to innovation performance. However, in the analysis using different levels of

foreign ownership, it is found that only firms with foreign ownership over 20% have a

negative impact on innovation performance. The result is not the same as Choi et al.

(2011)’s findings.

Thirdly, the report examines how firm ownership affects innovation specialisation and

diversification. The results indicate that firm ownership is unrelated to innovation

specialisation or diversification. Instead, innovation specialisation or diversification is

more associated with external factors.

5.4 Limitations

First, this thesis does not include the performance of listed companies on the One Belt and

One Road. It is still vague on how Belt and Road affects the relationship between firm

ownership and innovation performance.

Second, the ownership of companies used in this article is only broadly classified into five

categories - concentrated ownership, insider ownership, state ownership, institutional

ownership and foreign ownership. In fact, however, firm ownership can be subdivided

into various categories. For example, insider ownership can be subdivided into

management ownership, employee ownership and family ownership. The different types

of insider ownership also have a different impact on innovation, as reviewed in Chapter 2.

Therefore, mixing them together might have made the results unreliable to a certain

extent.

Third, due to the paucity of literature on the link between firm ownership and innovation

specialisation, hypothesis testing of the link between firm ownership and innovation

specialisation is based on the link between firm ownership and innovation diversification

in the opposite direction. It also poses some problems in proposing hypothesis tests.

Fourth, the thresholds used in the paper (i.e. 5% and 20%) are not very helpful in

examining the effects of firm ownership on innovation specialisation and innovation

diversification.

5.5 Recommendations and Future Research

First, due to the lack of information on Belt and Road among listed companies, future

research could collect such relevant information and identify more intuitively the impact

of Belt and Road on the relationship between firm ownership and innovation

performance.

Secondly, future research could provide a detailed classification of firm ownership in

order to identify precisely the association between firm ownership and innovation

performance.

Thirdly, only one set of thresholds (5% and 20%) is used in the paper. While thresholds of

5% and 20% had a significant effect on the relationship between firm ownership and

innovation performance, they did not significantly affect the relationship between firm

ownership and specialised or diversified innovation. Future research could focus on

thresholds based on 5% and 20% to see if changes in thresholds have a significant impact

on innovation specialisation or innovation diversification.

As the two thresholds of 5% and 20% have implications for insider ownership and state

ownership, future research could design mixed ownership stakes in SOEs based on these

two thresholds. This would not only benefit multiple parties but would also allow the

resources of each party to be fully utilised.

Students also viewed