Determinants of Chinese Outward Foreign Direct Investment in developing host countries

(1)

Determinants of Chinese Outward Foreign Direct Investment

in developing host countries

Master Thesis, Semester 2 - 2018

Author: Philippa Hungar (s3493768) Email: p.s.hungar@student.rug.nl Words: 15,134 Supervisor: Dr. R.W. de Vries Co-assessor: Dr. E. Mendiratta University of Groningen Duisenberg Building, Nettelbosje 2, 9747 AE Groningen, The Netherlands

www.rug.nl/feb

(2)

ABSTRACT

(3)

TABLE OF CONTENT

1. Introduction page 8

2. Literature Review page 10

2.1. FDI Theory Review page 10

2.2. Institutional Theory and Chinese MNEs page 12

2.3. Institutional Distance and Risk page 15

2.4. Foreign-Market-Seeking FDI page 18

2.5. Resource-Seeking FDI page 19

2.6. Efficiency-Seeking FDI page 20

2.7. Strategic-Asset-Seeking FDI page 20

2.8. Conceptual Model page 21

3. Methodology and Research design page 23

3.1. Model page 23

3.2. Data page 23

3.3. Dependent Variable page 24

3.4. Independent Variables page 24

3.5. Control Variables page 27

3.6. Data Quality and Descriptive Findings page 29

4. Findings page 32

4.1. Model Specifications page 32

4.2. FE Findings – Hypothesis 1 page 34

4.7. Robustness Check: Comparison of FE and RE Findings page 39

4.7.1. RE Model page 39

(4)

4.7.5. Comparison of FE and RE Findings – Hypothesis 4 page 41 4.7.6. Comparison of FE and RE Findings – Hypothesis 5 page 41 4.7.7. Comparison of FE and RE Findings – Control Variables page 42

5. Discussion page 44

5.1. Discussion of Hypothesis 1 page 44

6. Conclusion page 53

6.1. Conclusions page 53

6.2. Implications page 55

6.3. Limitations and Future Research page 56

References page 59

(5)

LIST OF TABLES

Table 1. Variables 28

Table 2. Correlation Matrix 31

Table 3. Regression Analysis – Fixed Effects Model: Determinants for Chinese FDI 38

Table 4. Regression Analysis – Random Effects Model: Determinants for Chinese FDI 43

Table i. Country Set i

Table ii. Descriptive Statistics – Original Variables ii

Table iii. Descriptive Statistics – Transformed Variables ii

Table iv. Variables in logarithmic transformation iii

Table v. Variance Inflation Factor test iii

(6)

LIST OF FIGURES

Figure 1. China’s FDI flows 8

Figure 2. Structure of the SASAC 14

Figure 3. Conceptual Model 22

Figure i. Kernel density estimate v

Figure ii. Q-Q Plot v

Figure iii. Standardized Normal Probability Plot v

(7)

LIST OF ACRONYMS

ASEAN Association of Southeast Asian Nations

FDI Foreign Direct Investment

FE Fixed Effects

GDP Gross Domestic Product

MNE Multinational Enterprise

MOFCOM Ministry of Commerce People’s Republic of China

OECD Organisation for Economic Co-Operation and Development

OFDI Outward Foreign Direct Investment

OLS Ordinary Least Squares

POE Private-owned Enterprise

R&D Research and Development

RE Random Effects

SASAC State-owned Assets Supervision and Administration Commission

SOE State-owned Enterprise

(8)

1. INTRODUCTION

Since the enforcement of the “open door” policy and the opening up of the Peoples’ Republic of China (China), China is especially known as a huge recipient of foreign direct investment (FDI). In the 1980s outward foreign direct investment (OFDI) was minimal and only state-owned enterprises (SOEs) were authorized to invest in foreign markets. The year 1985 marks a turning point for private Chinese enterprises as they were granted to perform FDI too, if they received the authorization of the government. In 2002, the “going global” policy was promoted allowing all Chinese enterprises to perform OFDI (Buckley, Clegg, Cross, Liu, Voss, & Zheng, 2007). Chinese foreign investment developed a more market-oriented approach compared to the 1980s when OFDI followed mainly political considerations (Cheung, & Qian, 2009). Since the “going global” policy Chinese FDI has increased rapidly. In 2016, China was the largest overseas investor among the developing countries and the second largest worldwide with an OFDI of US $ 183 billion (UNCTAD, 2017). With the “One Belt, One Road” Initiative proposed by the Chinese government in 2014, the globalization of Chinese enterprises is expected to further prosper (Yao, Zhang, Wang, & Luo, 2017) and OFDI will increase briskly. Moreover, in 2016 China’s OFDI surpassed for the first time China’s inward FDI (Figure 1).

FIGURE 1. China’s FDI flows (million US Dollars)

(9)

Chinese FDI developed significantly in the past fifteen years and because of this short period Chinese OFDI was not studied as detailed as FDI into China.

In the international business field most FDI theories focus on FDI originating from developed countries. Therefore, most studies which explain Chinas OFDI motivation and determinants, use classical theories, e.g. the resource-based view, the institution-based view and Dunning’s OLI (Ownership, Location, Internationalization) paradigm; other researchers believe modifications of the classical theories are necessary which leads to inconsistent findings. Another issue is the predominant use of traditional determinants in research, whereby researchers neglect the shift of importance in the set of determinants due to globalization. In many cases, assumptions for the motivation and determinants of Chinese OFDI are contradicting, for example: Buckley et al. (2007) found that Chinese SOEs invest into riskier countries which is explained by provided access to capital fund below market rates by the government. Also, Rudy, Miller and Wang (2016) assume that SOEs have a bigger risk appetite than private enterprises. In contrast, Peng, Tan and Tong (2004) declare that SOE have ownership advantages in the domestic market and a decreased willingness to take risky options of OFDI. This example of a contradiction shows that the evidence for the host-country-level determinants attracting Chinese OFDI is very limited.

In this research I focus on the research question: “What host-country-level determinants drive

the foreign direct investment decision of Chinese multinational enterprises into developing economies?”

(10)

interpretation of my findings followed by the conclusion where I also present implications, limitations and suggestions for future research.

2. LITERATURE REVIEW

2.1. FDI Theory Review

The basic principles of the FDI theory are built on the perspective of competitive advantages of firms: Companies internationalize depending on their competitive advantage at specific locations, minimizing their overall costs - with the return exceeding the additional costs and risks of operating abroad (Buckley, & Ghauri, 1999).

Dunning (2000, 2001) developed the OLI paradigm explaining FDI activities: First, every firm must have ownership-specific advantages to be successful investing abroad; second, countries offer location-specific advantages consisting of specific assets and further location benefits e.g. taxes and raw materials; third, internalization advantage is dependent on the market entry mode (Rugman, 2009).

The OLI advantages are considered as motivations for internationalization and FDI. By combining the factors, Dunning’s eclectic paradigm explains FDI by distinguishing four different types: foreign-market-seeking FDI, efficiency-seeking FDI, resource-seeking FDI, strategic-asset-seeking FDI (Dunning, 1993, 2000, 2001).

Market-seeking FDI by companies from emerging markets have trade supporting reasons: FDI aims to improve distribution networks and facilitate and enhance exports (Dunning, 2001). Furthermore, market-seeking FDI fosters growth by improving knowledge of business opportunities (Drogendijk, & Blomkvist, 2013).

Resource-seeking FDI is defined as FDI to acquire resources and raw materials, which are scarce, unavailable or inefficient in the home country (Dunning, 2001).

(11)

costs. Efficiency-seeking FDI also contains the intention to obtain an optimal structure of networks and efficient knowledge exchange by using location-specific advantages (Dunning, 2001). Moreover, FDI aims to achieve greater economies of scale and scope (Dunning, & Lundan, 2008). Former research, such as the research by Buckley et al. (2007) explained that due to the low labour costs in China it is unlikely that Chinese MNEs invest abroad to improve efficiency.

Strategic-asset-seeking FDI is asset exploring in nature, seeking to acquire and strengthen strategic assets (Dunning, & Lundan, 2008). It primarily deals with value-added activities, e.g. R&D operations, technologies, reputation and brand names (Dunning, 2001). Chinese enterprises are latecomers in international trade and therefore, are assumed to be especially motivated in strategic-asset seeking FDI to catch up globally and to gain competitive advantages. Especially by investing into developed markets Chinese enterprises seek knowledge and new capabilities (Gugler, & Brunner, 2007).

These four motivation types are compatible and based on the resource-based view, which explains that firms invest internationally for resource exploration and exploitation, e.g. resources such as technological and scientific knowledge, brand names and natural resources (Wang, Hong, Kafouros, & Boateng, 2012).

(12)

Dunning’s four FDI types can still be transferred on Chinese OFDI as FDI is used as a tool to become globally competitive and is motivated by either type the company is lacking in.

2.2. Institutional theory and Chinese MNEs

FDI decisions of multinational enterprises are not only based on resources that might be explored and exploited in potential host countries (the location specific advantages), but FDI decisions are also driven by the home institutional environment the firm is embedded in. The institution-based view explains that strategic choices, hence also FDI choices, result from the interaction of organizations and institutions (Peng, 2002). The institution-based view considers a wide institutional environment effecting embedded firms and not only ownership as the OLI paradigm does. Nevertheless, depending on the ownership, enterprises face a different institutional environment. On the one hand, home country governments might support FDI, especially if the government is an ally to the enterprise (Luo, Xue, & Han, 2010), which is particularly relevant for SOEs. On the other hand, an unsupportive institutional environment, which could emerge through governmental interference, corruption and uncertainty etc., drives enterprises to host countries with a more suitable institutional environment (Luo, Xue, & Han, 2010). In China SOEs face a different and more supportive institutional environment for OFDI compared to private-owned enterprises (POEs) (Voss, Buckley, & Cross, 2010). Therefore, to integrate the home institutional environment into my research, I will consider Chinese enterprises in general and will not just base my hypotheses on SOEs and their home institutional environment as e.g. Buckley et al. (2007) did.

(13)

(14)

FIGURE 2. Structure of the SASAC

Source: Deng et al., 2011

In general, SOEs are considered to not only pursue economic but also political goals. SOEs are used by the government to strengthen the national economy by accelerating the development and by acquiring added values. Furthermore, SOEs follow political goals like maximizing the employment rate and social welfare (Lin, Cai, & Li, 1998).

Most Chinese POEs are small and medium sized enterprises and are active in more competitive industries, e.g. textiles, electronics and machinery.

(15)

(Buckley et al., 2007). In contrast, due to the difficulties of getting financial support from banks, POEs mainly have to invest internationally with their own capital (Huang, & Renyong, 2014). Due to Chinese policies SOEs were able to perform FDI decades earlier than POEs. Since 2002, POEs have been allowed to invest abroad (Buckley et al., 2007) and are therefore even more latecomers in FDI than SOEs. The Chinese government still has to approve OFDI in a long process with several ministries, taking considerably longer time for POEs than for Chinese SOEs (Luo, Xue, & Han, 2010). In conclusion, SOEs own monopolistic advantages in the domestic market and also in regard to OFDI. As a result of this, it is difficult for POEs to reach the same status as SOEs have, even though private firms are found to be more productive (Dougherty, Herd, & He, 2007). In 2015, still 70 % of Chinese OFDI into Europe was performed by SOEs (Hanemann, & Huotari, 2016).

2.3. Institutional Distance and Risk

Not only the home country institutions as mentioned in section 2.2., determine OFDI, but also the host country institutional environment along with political risk is an important host-country level determinant for FDI.

One simple definition of political risk is the risk that the “rules of the game” are changed by the host country’s government (Butler, & Joaquin, 1998). “Rules of the game” are also a common transcription for institutions. This emphasizes the relation and correlation of institutions and their quality and political risk.

(16)

domestic market. In contrast, Rudy, Miller and Wang (2016) assume that SOEs have a more pronounced risk appetite than private enterprises. With regards to the risk aversion of Chinese SOEs former research does not have the same results. However, it was found that Chinese FDI is attracted by countries with weak institutions (Kolstad, & Wiig, 2012). As explained above, weak institutions are correlated with higher political risk. Assuming that Chinese FDI is attracted by countries with weak institutions, it appears that Chinese FDI is positively related with political risk. Buckley et al. (2007: 510) found that “Chinese foreign investors seem not to perceive risk in the same way as industrialized country firms”. This can be explained in various ways: Through the embeddedness in the Chinese institutional environment Chinese enterprises have an ownership advantage in dealing with similar risky environments. Moreover, due to the high political and ideological influence on Chinese enterprises, especially on SOEs, they might be attracted by similar ideological countries, which are often developing economies with higher political risk. In addition, Chinese enterprises are latecomers in FDI, especially POEs, and therefore less experienced which could have “led to FDI projects undertaken with insufficient due diligence and attention to associated risks” (Buckley et al., 2007: 510). Therefore, it can be assumed that Chinese FDI is positively related to a host country’s high political risk.

Hypothesis 1a: A developing host country’s political risk has a positive effect on FDI from Chinese MNEs.

(17)

regulations and these adaption costs depend on the host country’s institutions (Cezar, & Escobar, 2015). Experience in the domestic institutional environment reduce the adaption costs if the institutional frameworks of the home and host country are comparable (Habib, & Zurawicki, 2002). As China itself is considered to have weak institutions – the values of the six Worldwide Governance Indicators issued by the World Bank (further explained in section 3.4.) lie on average under 0 for China - it is therefore most likely also attracted by countries with weak institutions. Buckley et al. (2007) explained that Chinese enterprises are drawn by similar ideological countries and have an ownership advantage dealing with similar environments. Consequently, Chinese enterprises have an advantage investing into countries where the institutional distance is low. In contrast, higher institutional distance between the home and the host country is related to higher adaption costs, thus, it is riskier for multinational enterprises operating in these institutional distant environments, which leads to the hypothesis that high institutional distance between the host country and China has a negative effect on Chinese FDI.

Hypothesis 1b: A developing host country’s institutional distance to China has a negative effect on FDI from Chinese MNEs.

(18)

Hypothesis 1c: A developing host country’s openness to FDI has a positive effect on FDI from Chinese MNEs.

2.4. Foreign-Market-Seeking FDI

In general, market size and market growth are characteristics of host-markets that are significant determinants of FDI (Buckley et al., 2007). Large markets attract more FDI compared to smaller ones and market growth pictures more opportunities of generating profits (Lim, 1983). Developing markets have a smaller market size and are thus not a target region for market-seeking FDI compared to developed markets. Moreover, Chinas market itself is considered to be a big market, which could result in low foreign market-seeking by Chinese MNEs. Chinese SOEs especially face a plethora of benefits and have a competitive advantage in the large domestic market, therefore there is a decreasing willingness to take risky options of OFDI and SOEs managers are less motivated to invest abroad (Peng, Tan, & Tong, 2004). Consequently, Chinese FDI is, due to their large home market, less attracted by countries with a small market size, measured through gross domestic product (GDP) and GDP per capita, and their market growth, measured by the growth rate of GDP. Developing markets have a relatively small market size which is why their GDP and market growth are assumed to have no effect on FDI from Chinese enterprises.

Hypothesis 2a: A developing host country’s GDP (per capita) and market growth has no positive effect on FDI from Chinese MNEs.

Hypothesis 2b: A developing host country’s GDP (per capita) and market growth has no negative effect on FDI from Chinese MNEs.1

(19)

When analyzing market-seeking FDI it should also be considered, that Chinese firms might already have trade ties with the host country in form of networks or export connections. These trade ties would be advantageous for Chinese MNE investing into these countries, even if the host country has a relatively small domestic market. Therefore, a host country that is open to Chinese imports would most likely be attractive for market-seeking FDI of Chinese MNE.

Hypothesis 2c: China’s export to a developing host country has a positive effect on FDI from Chinese MNEs.

2.5. Resource-Seeking FDI

The endowment of natural resources is a host-country-level determinant of FDI (Buckley et al., 2007). In 2014, most of the 15 largest host countries for Chinese OFDI were resource-abundant countries, e.g. Kazakhstan, South Africa and Russia (Yao et al., 2017). As the acquisition of natural resources usually takes a large investment, only SOEs tend to afford resource-seeking FDI (Huang, & Renyong, 2014) because of the financial support by Chinese banks. Also, the mining and the energy sector are dominated by state-owned enterprises.

As China is increasingly relying on natural resources, besides supply reasons China aims to secure price stability by investing into resource-rich countries (Yao et al., 2017). According to Lieberthal and Herberg (2006), by internationally investing into resources, China is less threatened by price volatilities and can “lock in” prices.

(20)

2.6. Efficiency-Seeking FDI

Buckley et al. (2007) argued that due to the low labour costs in China, it is unlikely that Chinese MNEs will invest abroad to improve efficiency. Since 2007 China’s average monthly earnings have increased significantly. According to the International Labour Organization, the annual growth of the mean real monthly earnings of employees in 2006 was 12.9 % and in 2015 still 6.9%. Many developing countries have lower income levels and their monthly earnings do not increase as fast. Therefore, it can be assumed that China invests into developing countries to reduce the overall costs and this way seeks efficiency.

Hypothesis 4: A developing host country’s income level has a positive effect on FDI from Chinese MNEs.

2.7. Strategic-Asset-Seeking FDI

Chinese MNEs, as latecomers in the global market, do all seek strategic assets. They aim to acquire knowledge, technologies and brand recognition as well as to overcome the liability of “made in china” (Yuefang, Liefner, & Wang, 2013). SOEs, same as POEs, seek acquisition of soft skills, development of capabilities and competency, attraction of human resources and improvement of brand image (Oliveira, Menzies, Borgia, & Figueira, 2017). Strategic assets are mainly implemented in developed economies. Developing countries - as latecomers - lack in competitiveness in strategic assets, thus, in strategic-asset-seeking determinants. Therefore, Chinese FDI into developing countries is not driven by strategic assets.

(21)

Hypothesis 5b: A developing host country’s R&D activity and availability of skilled workforce has no negative effect on FDI from Chinese MNEs.2

2.8. Conceptual Model

As explained in 2.2. a firms’ behavior and ability to perform FDI is affected by institutional forces, e.g. government involvement and (state) ownership (Huang, & Renyong, 2014). Therefore, it can be assumed that Chinese MNEs have different motivations to invest abroad and are attracted differently by host-country-level determinants than MNEs from different countries facing another institutional environment. This results into a range of FDI amount per FDI type (foreign-market-seeking, resource-seeking, efficiency-seeking, strategic-asset-seeking FDI) originating from Chinese multinational enterprises.

Figure 3 illustrates the conceptual model of this research, with the amount of FDI being the dependent variable and the host-country-level determinants being the independent variables. The two control variables will be further explained in section 3.5..

(22)

(23)

3. METHODOLOGY AND RESEARCH DESIGN

3.1. Model

In accordance to Buckley et al. (2007), I verify the hypotheses with an Ordinary Least Squares (OLS) regression analysis of the following log-log model (Formula 1):

(1) 𝑙𝑜𝑔𝑦_𝑖𝑡 = 𝛽₀+ 𝛽₁𝑙𝑜𝑔𝑥_𝑖𝑡1+ 𝛽₂𝑙𝑜𝑔𝑥_𝑖𝑡2+ ⋯ + 𝛽_𝑘𝑙𝑜𝑔𝑥_𝑖𝑡𝑘+ 𝛼_𝑖𝑡+ 𝑢_𝑖𝑡

Where 𝑦_𝑖𝑡 is the total amount of Chinese FDI in the developing country 𝑖 at the period 𝑡, 𝑥₁ to 𝑥_𝑘 are the independent variables for the developing country 𝑖 at the period 𝑡. 𝑢_𝑖𝑡 is the error term and 𝛼𝑖𝑡 are all unobserved factors that influence the dependent variable 𝑦𝑖𝑡. I use a log-log

model as a linear relationship between the non-transformed dependent and independent variables can not be assumed. With the transformation the model is transferred to a linear, normal distributed and homoscedastic one. The logarithmic transformation of the model sustains the ordering between the dependent and the independent variables and offers a possibility of clean interpretation which is for example not the case after performing a square root transformation (e.g. large negative values turn to large positive values after taking the square root) (Gelman, & Hill, 2007).

3.2. Data

(24)

137). As some countries are defined as developing countries by the United Nations (e.g. Hong Kong, Israel, Korea Rep, Singapore, Taiwan), but as developed countries by the International Monetary Fund, these countries are excluded from the dataset. Solely countries which were considered as a transition (emerging) or developing country by the United Nations as well as the International Monetary Fund in 2010 are included in my dataset. The specific country register can be found in the Appendix (Table i).

3.3. Dependent Variable

The dependent variable is the total amount of Chinese FDI into developing markets. Chinese OFDI data by country is published in the “2010 Statistical Bulletin of China's Outward Foreign Direct Investment” by the Ministry of Commerce People’s Republic of China (MOFCOM). The Chinese FDI outflows provided by the UNCTAD are also adopted from the MOFCOM. After 2010 there is no data of Chinese FDI differentiated by country available. I excluded the year 2005 in my analysis as the data in the “2010 Statistical Bulletin of China's Outward Foreign Direct Investment” for 2005 is not consistent. Additionally, the total Chinese OFDI per year given by the MOFCOM is not identical with the total Chinese OFDI rates published by the OECD and the World Bank. These inconsistencies show the general limitations of the OFDI data. China’s outward as well as inward FDI data does not reach the quality of international standards due to collection problems and varying statistical methodologies (UNCTAD, 2007).

3.4. Independent Variables

The independent variables - the host-country-level determinants - are tested through 14 variables listed in Table 1.

(25)

The Worldwide Governance Indicators from the World Bank are based on these dimensions. The indicators are arranged on a scale from -2.5, indicating weak governance, to +2.5, indicating strong governance. According to Tomio and Amal (2015), the Formula 2 based on Kogut and Singh (1988) can be used to estimate the institutional distance between China and the host countries:

(2) 𝐼𝐷𝑗̇ = ∑ {

(𝑊𝐺𝐼𝑖𝑗−𝑊𝐺𝐼𝑖𝑢)2

𝑉𝑖 }

6

𝑖=1 /6

where 𝐼𝐷𝑗̇ is the expressed institutional distance between the host country 𝑗 and China, 𝑊𝐺𝐼𝑖𝑗 is

the dimension 𝑖 for country 𝑗, 𝑊𝐺𝐼_𝑖𝑢 is the dimension 𝑖 for China and 𝑉_𝑖 is the variance of the index of the dimension 𝑖.

The independent variable “political risk” is calculated by Formula 3 taking the average of the six Worldwide Governance Indicators and, in order to exclude negative values, adding 2.5 to change the scale, ranging from -2.5 to 2.5, to a scale from 0 to 5:

(3) 𝑃𝑅𝑗 =

𝛴𝑊𝐺𝐼𝑖𝑗

6 + 2.5

𝑃𝑅_𝑗 is the political risk for the host country 𝑗, 𝑊𝐺𝐼_𝑖𝑗 is the dimension 𝑖 for country 𝑗.

As already explained in section 2.3. political risk is the risk that the institutional environment changes and is not stable. Therefore, the governance indicators reflecting institutional and political quality and stability on six dimensions is an appropriate tool to measure the political risk.

The variable “openness to FDI” is measured as the ratio of the inward FDI to the host country’s GDP (Buckley et al., 2007). This data was extracted from the data center of the United Nations Conference on Trade and Development (UNCTAD).

(26)

variable “China’s exports” to the respective host country is included. The value of exports in US Dollar from China to the respective countries is offered by the World Integrated Trade Solution.

To measure the endowment of natural resources, the variable “ores’ and metals’ exports”, measuring the host country’s ores’ and metals’ exports to total merchandise exports, and the variable “fuel exports”, standing for the host country’s fuel export to total merchandise exports, are included in my model. The data is extracted from the World Development Indicators issued by the World Bank, which includes crude fertilizer, minerals, metalliferous ores, scrap and non-ferrous metals defined by the Standard International Trade Classifications under the term “ores and metals”. “Fuel” is defined as the Standard International Trade Classifications of mineral fuels, lubricants and related materials by the World Bank.

The independent variables designated to control for the host country’s income level are “average earnings”, to measure the average monthly earnings of employees in US Dollar, and “growth of earnings”, representing the annual growth of mean real monthly earnings of employees measured in percent. I extracted the corresponding data from the database of the International Labour Organization. This data is unfortunately not consistent as it is a compilation of data from various data sources and estimations done by the International Labour Organization. Therefore, the quality of the data varies from country to country and is highly dependent on the quality of each data collection method. Data collection methods are most likely not equal between countries, especially with regards to developing countries.

(27)

“compulsory education” – the duration in years that children are obligated to receive education – is gathered from the World Development Indicators.

3.5. Control variables

In line with Chang (2014), I add one dummy variable in the regression for control purposes: “Dummy foreign trade agreement”, as trade agreements simplify investments and can be considered as a manifestation of good economic relations and a trade-supportive environment. The MOFCOM names the following free trade agreements China has signed and implemented: The China-Peru Free Trade Agreement, signed in 2009, the China-Chile Free Trade Agreement, becoming effective in 2006, the China-Pakistan Free Trade Agreement, came into force in 2007 and most importantly the ASEAN Free Trade Agreement, signed in 2002. The China-ASEAN Free Trade Agreement is an agreement between China and the China-ASEAN members, consisting of Brunei (not included in the model), Cambodia, Indonesia, Laos, Malaysia, Myanmar, Philippines, Singapore (not included in the model), Thailand and Vietnam (MOFCOM, 2018).

Supporting the dummy variable, a second control variable “geographic distance”, measuring the geographic distance between China and the host country is added. According to the “2010 Statistical Bulletin of China's Outward Foreign Direct Investment”, China’s total FDI into Asia amounted to US $ 44.89 billion, into Africa US $ 2.11 billion and into Latin America US $ 10.54 billion in 2010. As a multiple of Chinese FDI is directed to Asia compared to more distant continents, it is appropriate to control for geographical distance as a further factor influencing Chinese FDI.

(28)

Table 1 Variables

HYPOTHESIS / FDI TYPE VARIABLE LABEL / PROXY

HOST-COUNTRY-LEVEL

DETERMINANT DATA SOURCE

all Dependent variable

Chinese FDI - Annual net overseas direct investment

(China to host country) in USD FDI from Chinese MNEs

2010 Statistical Bulletin of China's Outward Foreign Direct Investment

Institutional Distance & Risk for calculation Host country’s Control of Corruption

Institutional distance (ID) / Political Risk (PR)

WGI (Worldwide Governance Indicators by the World Bank)

Institutional Distance & Risk for calculation Host country’s Government Effectiveness ID / PR WGI

Institutional Distance & Risk for calculation

Host country’s Political Stability and Absence of

Violence/Terrorism ID / PR WGI

Institutional Distance & Risk for calculation Host country’s Regulatory Quality ID / PR WGI

Institutional Distance & Risk for calculation Host country’s Rule of Law ID / PR WGI

Institutional Distance & Risk for calculation Host country’s Voice and Accountability ID / PR WGI

Institutional Distance & Risk Independent variable

Institutional distance - Institutional distance between

China and host country Institutional distance WGI (own calculation)

Institutional Distance & Risk Independent variable Political risk - Host country’s Political Risk Political Risk WGI (own calculation)

Institutional Distance & Risk Independent variable Openness to FDI - Inward FDI to host country’s GDP Openness to FDI UNCTAD database

Market-seeking FDI Independent variable GDP - Host country’s Gross Domestic Product Market size

WDI (World Development Indicator by the World Bank)

Market-seeking FDI Independent variable GDP per capita - Host country’s GDP per capita Market size WDI

Market-seeking FDI Independent variable

GDP growth - Host country’s annual increase (%) in

GDP Market growth WDI

Market-seeking FDI Independent variable China’s exports - China’s export to host country Trade ties World Integrated Trade Solution

Resource-seeking FDI Independent variable

Ores’ and metals’ exports - Host country’s ores’ and

metals’ exports to total merchandise exports Endowment of natural resources WDI

Resource-seeking FDI Independent variable

Fuel exports - Host country’s fuel export to total

merchandise exports Endowment of natural resources WDI

Efficiency-seeking FDI Independent variable

Average earnings - Host country’s Average monthly

earnings of employees income level International Labour Organization

Efficiency-seeking FDI Independent variable

Growth of earnings - Host country’s Annual growth of

mean real monthly earnings of employees (%) income level International Labour Organization

Strategic-asset-seeking FDI Independent variable

R&D expenditure - Host country’s Research &

Development expenditure (% of GDP) R&D activity WDI

Patent applications - Host country’s Number of patent

applications R&D activity WDI

Compulsory education - Host country’s Compulsory education, duration Availability of skilled workforce WDI ALL Dummy / Control variable

dummy foreign trade agreement - Regional trade

agreements between China and host country Regional trade agreements

Ministry of Commerce People‘s Republic of China (MOFCOM)

ALL Control variable

Geographic distance - Distance Beijing to host country

(29)

3.6. Data Quality and Descriptive Findings

Before the linear regression analysis can be performed the variables need to be tested for plausibility, along with a test for certain assumptions of an OLS regression, which have to be met to achieve sensible results. The assumptions of a linear regression are: linearity of the model, normal distribution of the errors, homoscedasticity, no (or little) correlation and multicollinearity (Chatterjee, & Simonoff, 2013).

Analyzing the descriptive statistics (Table ii, appendix) of my variables, it appears, that my data is not normally distributed, as the values for standard deviation, skewness and kurtosis are relatively high for most variables. In addition to this, a non-linear relation between the dependent and independent variable can be surmised (explained in 3.1.), thus a transformation of the variables to a natural logarithmic variable is performed. The dummy variable of the free trade agreements and the independent variable “compulsory education” stay in their original form. According to Wooldridge (2013), it is uncommon to transform variables, expressed in years, into a logarithmic form. Table iv in the appendix displays the transformed variables with their respective new labels. The descriptive statistics after the transformation of the variables in the logarithmic form are shown in Table iii in the appendix. The observations of the independent variable “log average earnings”, defining the average monthly earnings of employees with a number of just 108, stands out. To gain explanatory value in my model and in the regression analysis, I exclude the variable “log average earnings” from my analysis, because it extremely limits the total numbers of observations.

(30)

create a null hypothesis, assuming, that the residuals are normally distributed (Sharpio, & Francia, 1972; Sharpio, & Wilk, 1965). In both tests, the detected p-value was large enough, that the null-hypothesis could not be rejected, thus, there is no reason to surmise, that the residuals are not normally distributed.

Another assumption of linear regressions is homoscedasticity, hence the model is tested for heteroscedasticity. If the variance of the error term in a model is not constant the model has problems with heteroscedasticity (Wooldridge, 2013). The White’s test and the Breusch-Pagan test both operate with a null-hypothesis, assuming that the residuals are homoscedastic, implying a homogeneous variance of the residuals (Wooldridge, 2013). For this model the p-value for both tests is not significant. As a result of this, the null-hypothesis cannot be rejected and it can be assumed that the model has no problem of heteroscedasticity. To verify this outcome, I graphically tested for homoscedasticity too (Figure iv, appendix): The distribution of the plotted residual against the fitted values seems not to follow a specific pattern, which also indicates homoscedasticity.

(31)

Table 2 Correlation Matrix log institutional distance log political risk log openness to FDI log GDP log GDP per capita log GDP growth log China's exports log ores' and metals' exports log fuel exports log growth of earnings log R&D expenditure log patent applications compulsory education log geographic distance dummy foreign trade agreement log institutional distance 1

log political risk -0.0388 1

log openness to FDI -0.0206 0.2735* 1

log GDP -0.2776* 0.0626 -0.2629* 1

log GDP per capita 0.1959* 0.5789* 0.1400* 0.3321* 1

log GDP growth -0.0011 -0.059 0.0455 0.0309 -0.0271 1

log China's exports -0.2608* 0.0833 -0.1630* 0.8177* 0.2209* 0.009 1

log ores' and metals' exports -0.0695 0.2219* 0.066 0.0229 -0.1784* -0.0217 0.0118 1

log fuel exports -0.0551 -0.2135* -0.0864 0.4824* 0.3664* -0.0471 0.3585* -0.2382* 1

log growth of earnings -0.0268 -0.146 0.1482 -0.1574* -0.1479 0.2315* -0.0433 0.0767 0.0413 1

log R&D expenditure -0.1125 0.2846* -0.0121 0.4734* 0.0604 -0.0267 0.4241* 0.3969* -0.1129 -0.1359 1

log patent applications -0.0556 0.116 -0.2621* 0.8424* 0.4840* -0.0412 0.7592* 0.1225 0.3627* -0.2569* 0.5947* 1

compulsory education 0.1498* 0.3234* 0.0859 0.0898 0.3809* -0.0969 0.0704 -0.0427 0.0652 -0.0198 0.2090* 0.3067* 1

log geographic distance 0.0883 0.1879* 0.1164* -0.0266 0.1231* -0.1599* -0.2813* 0.1820* -0.0141 -0.2084* 0.2625* -0.0398 0.1378* 1

dummy foreign trade agreement -0.0062 0.0682 0.0283 0.1765* -0.0031 0.0179 0.3238* 0.0332 -0.0353 -0.0555 -0.0872 0.1312 -0.0733 -0.3901* 1

(32)

4. FINDINGS

4.1. Model Specifications

To test my hypotheses, I perform a regression of my model described in Formula 1. After the exclusion of some variables described in the descriptive findings, my model can be described as followed (Formula 4):

(4) log 𝐶ℎ𝑖𝑛𝑒𝑠𝑒 𝐹𝐷𝐼𝑖𝑡 = 𝛽0+ 𝛽1log 𝑖𝑛𝑠𝑡𝑖𝑡𝑢𝑡𝑖𝑜𝑛𝑎𝑙 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑖𝑡1+ 𝛽2log 𝑝𝑜𝑙𝑖𝑡𝑖𝑐𝑎𝑙 𝑟𝑖𝑠𝑘𝑖𝑡2+ 𝛽3log 𝑜𝑝𝑒𝑛𝑛𝑒𝑠𝑠 𝑡𝑜 𝐹𝐷𝐼𝑖𝑡3+ 𝛽4log 𝐺𝐷𝑃 𝑝𝑒𝑟 𝑐𝑎𝑝𝑖𝑡𝑎𝑖𝑡4+ 𝛽5log 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ𝑖𝑡5+ 𝛽6log 𝐶ℎ𝑖𝑛𝑎′𝑠 𝑒𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑡6+ 𝛽7log 𝑜𝑟𝑒𝑠′𝑎𝑛𝑑 𝑚𝑒𝑡𝑎𝑙𝑠′𝑒𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑡7+ 𝛽8log 𝑓𝑢𝑒𝑙 𝑒𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑡8+

[𝛽9log 𝑔𝑟𝑜𝑤𝑡ℎ 𝑜𝑓 𝑒𝑎𝑟𝑛𝑖𝑛𝑔𝑠𝑖𝑡9 + 𝛽10log 𝑅&𝐷 𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒𝑖𝑡10] + 𝛽11log 𝑝𝑎𝑡𝑒𝑛𝑡 𝑎𝑝𝑝𝑙𝑖𝑐𝑎𝑡𝑖𝑜𝑛𝑠𝑖𝑡11+ 𝛽12𝑐𝑜𝑚𝑝𝑢𝑙𝑠𝑜𝑟𝑦 𝑒𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛𝑖𝑡12+

𝛽13log 𝑔𝑒𝑜𝑔𝑟𝑎𝑝ℎ𝑖𝑐 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑖13+ 𝛽14𝑑𝑢𝑚𝑚𝑦 𝑓𝑜𝑟𝑒𝑖𝑔𝑛 𝑡𝑟𝑎𝑑𝑒 𝑎𝑔𝑟𝑒𝑒𝑚𝑒𝑛𝑡𝑖𝑡14+ 𝛼𝑖𝑡+ 𝑢𝑖𝑡

To estimate Formula 4, two statistical models are considered: the fixed effects (FE) OLS model and the random effects (RE) generalized least squares model. The FE model complies better with my analysis. The FE model is based on the presumption, that within the country, some unobserved effect 𝛼_𝑖𝑡 might impact or bias the independent or dependent variables. These unobserved effects are country-specific and time-invariant - accordingly also called “individual fixed effect” - and are allowed to be correlated with the observed independent variables (Wooldridge, 2010, 2013). In contrast, in the RE model the unobserved effects 𝛼_𝑖𝑡 are defined as a “random” variable and no correlation is allowed between the unobserved effects 𝛼_𝑖𝑡 and

the independent variables (Wooldridge, 2010, 2013). It should not be confused that in the FE model 𝛼_𝑖𝑡 is “being treated as nonrandom; rather, it means that one is allowing for arbitrary dependence between the unobserved effect […] and the observed explanatory variables” (Wooldridge, 2013: 286).

(33)

institutions in a country, hence it is prone to be correlated to institutional distance. This describes only one of many imaginable unobserved effects and therefore I have to expect correlations between 𝛼_𝑖𝑡 and my variables, which makes the FE model a better fit for my analysis.

Even though Wooldridge (2013) recommends to use the FE model instead of the RE model in most cases, the FE model also has its disadvantages. The FE model can not be used to analyze time-invariant variables. The control variable “log geographic distance” is consequently excluded in my main model, the FE model. The RE model in contrast contains the advantage of being able to investigate time-invariant variables (Wooldridge, 2013). Furthermore, the Hausman test – testing for the “statistically significant differences in the coefficients on the time-varying explanatory variables” (Wooldridge, 2013: 496) – indicated, that the RE model should be used instead of the FE model. Some researchers (e.g. Gelman, & Hill, 2007) share the opinion, that the RE model should always be favored. As already explained above, theoretically, the FE model aligns better with my research. This choice could be questioned taking into account the Hausman test, but according to Wooldridge (2013: 496), the result of the Hausman test preferring the RE model “means either that the RE and FE estimates are sufficiently close so that it does not matter which is used, or the sampling variation is so large in the FE estimates that one cannot conclude practically significant differences are statistically significant “. Still, I include the RE model in my research for comparison reasons and to check the robustness of the results (Table 4).

(34)

depicts no significance of the FE Model 8. As a result of this, I created the FE Model 7 (Table 3) to test my hypotheses. In this Model I excluded the variables “log growth of earnings” and “log R&D expenditure” as these two variables have the lowest numbers of observations with 174 and 175 observations each in the dataset (see Table iii in the appendix). In the FE Model 7 the control variable “dummy foreign trade agreement” suffers under no correlation problems – which can be attributed to the exclusion of “log R&D expenditure” – and it is included in the model. The Model consists of 142 observations and the average observation per group (country) with 3.5 – meaning that for each country on average 3.5 years are observed - are explanatory. This is also proven by the F statistic, which indicates the significance of the FE Model 7 at the .01 significance level. The R-squared within reveals the accountability of the model for 37,24% of the variance in the country units. In the FE models the R-squared within is the ordinary R-squared. The reported R-squared between and overall are “correlations squared” (StataCorp, 2017: 425). Before interpreting the impact of each individual variable and rejecting or accepting my hypotheses, it has to be pointed out, that this study makes use of a log-log model. The coefficients of the explanatory variables are referred to as elasticities, meaning that “the coefficient can be interpreted as the expected proportional change in [“log Chinese FDI”] per proportional change in x” – the explanatory variable (Gelman, & Hill, 2007: 64). Excluded from this interpretation scheme are the variables “compulsory education” and “dummy foreign trade agreements”, as these variables are not transformed into the logarithmic variables.

4.2. FE Findings - Hypothesis 1

(35)

Chinese FDI. The variable “log institutional distance” has a negative effect on Chinese FDI, but this impact is not significant in the FE Model 7. In the FE Model 2 (Table 3), where only the institutional variables and control variables are included in the regression, the variable “log institutional distance” shows a significant negative effect at the .05 level. The significance level depends on the specification of the model, thus the findings are ambiguous. This could be explained by the correlation of “log institutional distance” with “log GDP per capita”, “log China’s exports” and “compulsory education” at a significance level of .05 (Table 2). These three variables apparently have an overlapping explanatory value with “log institutional distance”. The correlations are in fact significant, but not extraordinary high, which is why “log institutional distance” loses its significance in the main model FE Model 7. Therefore, hypothesis 1b cannot be explicitly supported.

Hypothesis 1c is supported: The explanatory variable “log openness to FDI” effects Chinese FDI positively at a .05 significance level. This finding is consistent with the FE Model 2, where “log openness to FDI” is even significant at the .01 level. With a 1% increase in a host country’s openness to FDI, Chinese FDI increases 0.49%.

(36)

both variables are included (FE Model 7) “log China’s exports” loses significance. The Correlation matrix (Table 2) illustrates that these two variables are highly correlated. As “log China’s exports” is not significant in FE Model 7, hypothesis 2c can not be supported.

Hypothesis 3 is not supported as the independent variables “log ores’ and metals’ exports” and “log fuel exports”, controlling for the endowment of natural resources, are both not significant in the FE Model 7. In the FE Model 4 (Table 3), the variable “log ores’ and metals’ exports” is significant and the analysis indicates, that Chinese FDI increases 0.42% with every 1% increase in the host country’s ores’ and metals’ exports to total merchandise exports. The FE Model 7, where “log ores’ and metals’ exports” is not significant, reveals that the host country’s ores’ and metals’ exports to total merchandise exports have a negative effect on Chinese FDI. These contradicting findings will be further discussed in section 5.3.

The variable “log growth of earnings” does not attain significance in any model. It can be justified to exclude this explanatory variable from the FE Model 7, to gain explanatory significance in this model. Moreover, as “log growth of earnings” is not significant in the FE Model 5 or the FE Model 8 (Table 3) , the hypothesis 4 can not be supported. The F statistic of the FE Model 5 and 8 is not significant, indicating that the models themselves are not significant.

(37)

(38)

Table 3 Regression Analysis - Fixed Effects Model: Determinants for Chinese FDI

Constant 16.2092*** 16.3577*** -21.4956** 15.6617 16.5367*** 11.5004** -52.8968*** -119.6494*

(0.1879) (1.0655) (8.8617) (0.3154) (0.3431) (5.2343) (19.2703) (62.8540)

Independent Variables

log institutional distance -1.3041** -0.9494 -0.5780

(0.5592) (0.7581) (1.7764)

log political risk -0.5094 -5.6876** -11.2069*

(1.6924) (2.8036) (6.3901)

log openness to FDI 0.3004*** 0.4854** 0.3040

(0.1099) (0.2151) (0.3963)

log GDP per capita 2.2828* 7.1298** 19.8976*

(1.3757) (3.4233) (9.9671)

log GDP growth -0.1603 -0.0030 -0.3160

(0.1071) (0.1617) (0.4357)

log China's exports 0.9695*** 0.7028 -1.2003

(0.2079) (0.4971) (1.1381)

log ores' and metals' exports 0.4173** -0.2097 1.7890

(0.1843) (0.4653) (1.3382)

log fuel exports 0.0856 -0.1167 -0.9639

(0.0959) (0.2999) (1.0121)

log growth of earnings 0.0206 -0.0576

(0.1265) (0.2977)

log R&D expenditure 0.1220 1.5209

(0.4638) (0.9892)

log patent applications 0.1795 -0.4085 -0.0772

(0.6429) (0.2970) (0.9237)

compulsory education 0.4658* 0.3792 0.4596

(0.2604) (0.2391) (0.3070)

Control Variables log geographic distance

dummy foreign trade agreement 2.2382* 2.1821* 1.3619 2.2686** 1.8035 omitted 0.6742 omitted

(1.2535) (1.2087) (1.1023) (1.1464) (1.3683) (1.0640)

Number of Observations 388 368 335 286 157 113 142 52

Average observation per group 4.5 4.3 4 3.9 3 3.3 3.5 2.1

Model F statistic 3.19* 4.31*** 15.21*** 3.11** 0.88 1.15 4.91*** 1.89

R-squared: within 0.0105 0.0584 0.1977 0.0426 0.0169 0.0434 0.3724 0.6019

R-squared: between 0.1589 0.0966 0.1148 0.1474 0.1751 0.0007 0.0018 0.0052

R-squared: overall 0.1021 0.0984 0.1011 0.0804 0.0999 0.0001 0.0088 0.0085

Note: Standard Errors are reported in parentheses.

P-Values smaller than 0.01, 0.05, and 0.10 are indicated by ***,** and *, respectively.

FE (7) FE (8) ---omitted---FE (5) efficiency FE (6) strategic-asset Dependent Variable Log Chinese FDI

(39)

4.7. Robustness Check: Comparison of FE and RE Findings 4.7.1. RE Model

The findings of the RE Model are shown in Table 4. The RE models are structured like the FE models. The RE Model 7 (Table 4) is the main model testing the hypotheses. The FE Model 1 to 6 (Table 4) are the models testing each of the variable groups. All RE models persist of the same number of observations as the respective FE models. The RE Model 8 (Table 4) is included for reasons of completeness, just like the FE Model 8 in my previous testing.

The Wald chi-squared statistic of the RE Model 7 is highly significant at the .01 level. The R-squared overall value of the RE Model 7 indicates that 25.87% of the variance of the independent variable “log Chinese FDI” can be predicted from the variables included in the model. 70.62% (rho (𝜌)=0.7062) of the total variance in the RE Model 7 is explained by differences between countries. The remaining 30% are attributable to change over time within the countries. In the RE Model 1 to RE Model 6 above 50% of the total variance can be attributed to variation between countries. Around 50% is assignable to variation within countries over time. This does make the RE models adequate models to check for robustness by comparing the findings with the FE model findings.

4.7.2. Comparison of FE and RE Findings - Hypothesis 1

(40)

The RE Model 7 shows no significant impact of “log institutional distance” on “log Chinese FDI”. In contrast to the FE Model 2, the RE Model 2 still does not reveal importance of a host country’s institutional distance to China, as an indicator for Chinese FDI. In the RE models, the significance of “log institutional distance” does not depend on the specifications of the model, as it does in the FE Models. These findings confirm the rejection of hypothesis 1b.

Hypothesis 1c is also supported throughout all RE models. The independent variable “log openness to FDI” has a positive effect on Chinese FDI at a .01 significance level in RE Model 2. Also, in RE Model 7, “log openness to FDI” has a significant positive coefficient, meaning that a 1% increase of a host country’s openness to FDI is associated with a 0.35% increase of Chinese FDI. The positive effect of the host country’s openness to FDI on FDI from Chinese MNEs, is the most solid finding in the whole data analysis.

The variables “log GDP per capita” and “log GDP growth” are not significant in the RE Model 3 and the RE Model 7. The finding of the FE models, that a host country’s GDP per capita has a positive effect on Chinese FDI could not be supported by the RE Models, thus is not robust. In the RE Model 1 as well as in the RE Model 3, the variable “log China’s exports” is significant at the .01 level. A 1% increase of China’s exports to a developing host country is associated with a 1.08% increase of FDI from China. These finding would suggest a clear support of hypothesis 2c, which is not in accordance with the ambiguous findings of the FE models. The findings are hence not clear and have to be further discussed.

(41)

within the country units. This indicates that the market variables have an important explanatory value regarding Chinese FDI.

In line with the FE Model 7, the RE Model 7 shows no support for hypothesis 3, as the variables “log ores’ and metals’ exports” and “log fuel exports” are both insignificant. In contrast, in the RE Model 4 these variables are both significant at the .05 level, indicating a positive effect on Chinese FDI. The positive effect of the variable “log ores’ and metals’ exports” on Chinese FDI was also found in the FE Model 4. These findings are consistent revealing that this variable gains significance in a model with more observations, a smaller number of variables, thus less variance. The results of the the RE Model 7 can not support hypothesis 3. The significance of the variables depend on the specifications of the model, hence the findings remain ambiguous.

4.7.5. Comparison of FE and RE findings - Hypothesis 4

The variable “log growth of earnings”, used to test hypothesis 4, is not significant in neither RE Model 5 nor RE Model 8. As already explained in section 4.1., the variable is excluded from the FE Model 7, due to a low number of observations. For the same reason it is also excluded from the RE Model 7. Contrary to FE Model 5, RE Model 5 is, as a model, significant as shown with the .01 significance level of the Wald chi-squared statistic. Still, most of the models, which test the efficiency variable’s effect on Chinese FDI are not significant. For of this reason, I will further discuss these findings in the discussion section 5.4.

4.7.6. Comparison of FE and RE findings - Hypothesis 5

(42)

of a host country’s patent applications, a 0.48% decrease of Chinese FDI is expected. The FE Model 7 calculated a similar negative effect of “log patent applications” – a 1% increase of a host country’s patent applications is associated with a 0.41% decrease of Chinese FDI – but this finding is not significant and therefore, considering the FE models, both hypotheses are supported. According to the RE Model 7, only hypothesis 5a could be supported. The RE Model 6, which solely includes the strategic-asset variables and the control variables, is, as a model, not significant, because the Wald chi-squared statistic attained no significance. Furthermore, in this model, none of the strategic-asset variables are significant.

4.7.7. Comparison of FE and RE findings - Control Variables

Unfortunately, the FE model could not test the impact of the control variable “log geographic distance” on FDI from Chinese FDI, as it is a time invariant variable. In the RE models, this control variable is under most of the different model specifications insignificant and just for the RE Model 4 and 6 slightly significant. The dummy variable, controlling for free trade agreements, does fulfill the aimed at purpose: In the FE Models 1, 2 and 4 and the RE Model 1 to 5 the free trade agreements have a significant positive effect on FDI from Chinese enterprises. The significance of the positive effect is again dependent on the model specification.

(43)

Table 4 Regression Analysis - Random Effects Model: Determinants for Chinese FDI

Constant 19.1620*** 18.9631*** 5.5438 21.1973*** 19.8682*** 23.5731*** -5.1646 1.5739 (2.6231) (2.7431) (3.4533) (3.1912) (3.3037) (4.4367) (6.4653) (10.5895) Independent Variables

log institutional distance -0.3449 -0.0662 -0.0523

(0.2251) (0.3857) (0.6268)

log political risk -0.3895 -2.7391* -5.6424**

(0.4803) (1.5024) (2.7123)

log openness to FDI 0.2934*** 0.3470* -0.0990

(0.0950) (0.1895) (0.3242)

log GDP per capita -0.0103 0.3550 0.6150

(0.1251) (0.4415) (0.6672)

log GDP growth -0.1018 -0.0664 -0.1107

(0.1050) (0.1554) (0.3797)

log China's exports 0.5707*** 1.0767*** 0.8883**

(0.0863) (0.2230) (0.4216)

log ores' and metals' exports 0.1835** 0.3006 0.7754**

(0.0827) (0.1856) (0.3163)

log fuel exports 0.1113** 0.0746 -0.0661

(0.0490) (0.1508) (0.3099)

log growth of earnings 0.1476 -0.0339

(0.1195) (0.2811)

log R&D expenditure 0.2383 0.5704

(0.3374) (0.5284)

log patent applications 0.0772 -0.4821*** -0.4317

(0.1499) (0.1702) (0.3686)

compulsory education 0.0837 0.1718 0.2559

(0.1436) (0.1452) (0.1813) Control Variables

log geographic distance -0.3316 -0.3110 -0.0875 -0.5850* -0.4067 -0.8709* -0.1739 -0.5280 (0.2914) (0.3113) (0.3074) (0.3533) (0.3679) (0.4719) (0.5462) (0.6977) dummy foreign trade agreement 1.8177*** 1.7999*** 1.0468** 1.7268*** 1.6066*** 0.5188 0.6337 1.5139 (0.4983) (0.5120) (0.4778) (0.5327) (0.5442) (0.7927) (0.6634) (1.1327)

Number of Observations 388 368 335 286 157 113 142 52

Average observation per group 4.5 4.3 4.0 3.9 3.0 3.3 3.5 2.1

𝜒2 20.45*** 31.35*** 72.01*** 27.17*** 15.73*** 5.9000 48.41*** 19.7700 R-squared: within 0.0105 0.0491 0.1868 0.0358 0.0106 0.0199 0.3099 0.3860 R-squared: between 0.1724 0.1749 0.3035 0.2410 0.2288 0.1423 0.2563 0.3686 R-squared: overall 0.1114 0.1396 0.2453 0.1420 0.1410 0.0874 0.2587 0.3609 𝜎_𝑢 1.4143 1.4666 1.2443 1.3920 1.3219 1.4305 1.6206 1.5536 𝜎_𝑒 1.3732 1.3238 1.1950 1.2557 1.1850 1.4245 1.0452 0.9400 𝜌 0.5147 0.5510 0.5202 0.5513 0.5545 0.5021 0.7062 0.7320

Note: Standard Errors are reported in parentheses.

P-Values smaller than 0.01, 0.05, and 0.10 are indicated by ***,** and *, respectively.

RE (7) RE (8) RE (3) market RE (4) resources RE (5) efficiency RE (6) strategic-asset Dependent Variable Log Chinese FDI

(44)

5. DISCUSSION

5.1. Discussion of Hypothesis 1

In this section, I will discuss the findings explained in section 4 as some of the findings were contradictory to my hypothesis and because the statistical robustness of these findings is not consistent.

(45)

(46)

different industries would be an interesting topic for future research. Moreover, to further analyze the negative effect of political risk on Chinese FDI, other measurements of political risk should be used. I also propose to divide political risk into its different elements and to analyze how Chinese FDI is related to each of these risk elements.

The regression analysis including all the relevant independent variables (FE Model 7, Table 3) could not support hypothesis 1b, which conjectured that a host country’s institutional distance to China has a negative effect on Chinese FDI. This finding was supported by the RE models. Only the FE Model 2 (Table 3) revealed a significant negative effect of a host country’s institutional distance on Chinese FDI. The assumption, that a host country’s institutional distance to China can have a negative effect on FDI from Chinese MNEs seems valid and theoretically stable, because the coefficients of “log institutional distance” are negative in all models. However, the regression analysis does not prove that the institutional distance between China and a host country is a significant determinant of FDI from Chinese MNEs: The significance of “log institutional distance” depends on the model choice and its specifications, hence, the findings are ambiguous and further testing is needed.

(47)

relationship. Furthermore, the measurement of the variable “openness to FDI” contains the received FDI from all MNEs, regardless of the ownership and location of these MNEs. The host country’s openness to Chinese FDI in specific, is not tested. Host country’s governments perceive FDI from China in different ways – seeing threats or opportunities in FDI from China - and are consequently not equally open to Chinese FDI. Host countries of Chinese resource-seeking FDI are often concerned that they do not equally benefit from the investment projects, because the working conditions may be of low standard and Chinese labour is often engaged in the projects, instead of employing local workers (Sauvant, & Nolan, 2015). The raise of the minimum wage by the Zambian government, after Chinese enterprises entered the market (Sauvant, & Nolan, 2015), is one example for a host country’s reaction to Chinese FDI. Furthermore, obtaining FDI from Chinese enterprises also raises the concern of “national security”, e.g. the members of parliament of the Indian parliament submitted 38 of a total of 146 state-specific questions addressing FDI from China between 2009 and 2014. These questions primarily dealt with the concern of national security (Nazareth Satyanand, 2015). Other host countries seek to attract Chinese FDI, e.g. Chile and Russia. Moreover, many South-East-Asian countries mainly receive FDI from Chinese enterprises through linkages in global value chains and recognize Chinese investments as promoting development (Sauvant, & Nolan, 2015).

(48)

measurement methods to control for openness to FDI, e.g. investment friendly/unfriendly laws and regulations, low tax environment supporting investments etc.

I created the hypothesis 2a and 2b with the intention to suggest, that neither GDP nor market growth has a significant effect on Chinese FDI. At first, it might seem unnecessary to include variables in a study, where no significant effect is expected and normally, no explanatory value is gained for the study. In this case, it does make sense to include two hypotheses which together suggest no significant effect, because this suggestion contradicts Dunning’s FDI theory - with the background that I am analyzing Chinese FDI behavior, which hypothetically is different to investment behavior from industrial country’s companies. The entire research would be incomplete, if I had dropped variables explaining for market-seeking FDI. By including these variables (GDP per capita and GDP growth), Dunning’s theoretical framework is fully tested, my suggestions of no significant effect of GDP per capita on Chinese FDI is proven wrong and Dunning’s theoretical framework of the four FDI motivation types is proven right.

(49)

facing many benefits in China. This argumentation can as well exactly explain the contrary effect - that Chinese MNEs are market-seeking. Due to the large domestic market, Chinese MNEs, especially Chinese POEs, face a lot of competition. To enter developing markets, where the pressure of competition is not as high, opens the opportunity for an attractive sales and production market. The sales markets attractivity of a developing country is most likely going to depend on the industry and product type. The fact that Chinese FDI is attracted by markets with comparably higher GDP per capita rejects hypothesis 2a and supports hypothesis 2b. The host country’s market growth does not have a significant effect on Chinese FDI. Considering that GDP per capita does have a positive effect on Chines FDI and, accordingly, Chinese enterprises do perform market-seeking FDI, the non-significance of GDP growth could have two possible reasons: First, Chinese investment projects are not future oriented or not planned for several years. Therefore, Chinese enterprises are not interested in how the market grows and which size it might reach in the future. Second, GDP growth is not a valuable indicator for the attractiveness of the market environment. Countries with high GDP growth are mostly countries with a small market size (low GDP per capita). This also explains the negative - though not significant - impact of GDP growth on Chinese FDI. Accordingly, Chinese enterprises would prioritize many factors of a host country, e.g. the market size, over the market growth in their investment decision.

(50)

“log patent applications” (Table 2) could explain the insignificance of “log China’s exports” in the main model FE Model 7. One could try to explain the correlation with the Chinese enterprises preference to export into countries with more patents as this could be an indicator of stronger intellectual property rights. It seems that China’s export to a host country is a determinant of Chinese FDI. The FE Model 7, seems to contain too many limitations, with regard to the analysis of the variable “log China’s exports”, to prove it. A host country, which is an important destination for China’s exports, can also expect to be an interesting destination for investments from Chinese enterprises. It would also be interesting to research, whether the host country’s exports to China does have the same impact on Chinese FDI as it has on exports from China to the host country because they also are a manifestation of trade ties between China and the host country.

(51)

77 percent of the global oil deposits were controlled by ten companies and in 2006, 33 percent of the world’s non-energy mineral value was accounted by only ten enterprises. Therefore, Yao et al. (2017) claimed that Chinese enterprises invest into natural resources to ensure supply and secure price-stability. This is supported by the Chinese FDI flows: Table vi (appendix) indicates that the FDI flows in the mining industry are among the fourth biggest Chinese FDI flows in 2006 to 2010. In 2006, the mining industry accounted for the largest amount of Chinese FDI flows with US $ 8,539.51 and in 2009, the FDI flows of the mining industry took the second position with US $ 13,343.09 million of a total of US $ 56,528.99 million.

In their argumentation, Yao et al. (2017) did not differentiate between resource types. To supply the demand of fuels, it might be more profitable to import a higher amount of fuels, instead of investing into fuel exploitation. The natural resources extraction is investment intense and risk intense, which would explain the insignificance of the positive effect of the variable “log fuel exports” on FDI from Chinese enterprises. The oil demand of China has been met by over fifty percent through imports since 2007 (Yao et al., 2017). How much of these imports are related to Chinese FDI is hard to determine.

Even though the regression analysis could not solidly support hypothesis 3, the economic development over the past twenty years, facts about the Chinese natural resource consumption as well as theory sufficiently support hypothesis 3, which surmises that a host country’s natural resource endowment is an important determinant of Chinese FDI. It has to be taken into consideration, that former research (Yao et al., 2017; Lieberthal, & Herberg, 2006) does not differentiate between developing and developed host countries. It is not clear whether a host country’s resource endowment is an equally important determinant of Chinese FDI in developing host countries as it is in developed host countries and vice versa.

(52)

host country’s geographical distance to Beijing is associated with a decrease of 0.59% of Chinese FDI. As the transportation of natural resources is expensive, long distances are not attractive for investments into natural resources.