The determinants of Chinese Outward Foreign
Direct Investments in Europe
Name: Vincent Waterman
Student number: 6066798
Supervisor: Dr. K.B.T. Thio
Table of content
1. Introduction ... 3
2. Literature review ... 5
2.1 Theoretical FDI literature ... 5
2.2 Empirical FDI literature ... 6
3. Empirical model and data ... 8
3.1. The model ... 8 3.2. Variable selection ... 10 3.2.1. Main variables ... 10 3.2.2. Control variables ... 11 3.3. Data ... 11 3.4. Regression methods ... 12 4. Results ... 14
4.1. Regression model evaluation ... 14
4.2. Regression results ... 16
4.3. Summary and Comparison ... 19
6. Limitations and future research ... 21
7. References ... 21
8. Appendix ... 23
Statement of Originality
This document is written by Vincent Waterman who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in
creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
1. Introduction
China opened up its borders after reforms in 1979 and its trade has been growing ever since. The opening of the markets also led to a modest FDI outflow. Deng Xiaoping’s tour of South China in 1992 and the Go Global strategy in 1999, two strategies to enhance China’s trade, led to a boost in outward FDI (Kolstad & Wiig, 2012). The FDI stock into China has grown with a factor of 1.5 from 2003 to 2012 according to data of the United Nations Conference on Trade and Development (UNCTAD, 2015). However, the outward foreign direct investment (OFDI) stock has grown with an astonishing factor of 15 in the same time period. Where China was only the 17th largest country in terms of size in outward FDI flows in 1996, it is now ranked third with 7.6% of the world total in 2013 (Li, Liu, & Jiang, 2015).
Whereas in the early years of Chinese outward FDI most flows went towards neighboring and resource rich countries, nowadays we see a major increase in flows of FDI toward developed countries. The outward FDI flows from China to the European Union (EU) grew with a factor of 50 between 2003 and 2012 from 113 million to 6120 million, as can been seen in figure 1 (UNCTAD, 2015). The EU is not only China’s largest trading partner, but also an attractive destination for Chinese outward FDI. By entering one single EU member country, Chinese companies can access the entire EU market (Dreger, Schüler-Zhou, & Schüller, 2015). However, in absolute value the Chinese FDI stock is still rather small in comparison to the total FDI stock in Europe. In 2012 only 1 percent of the total FDI stock originated from China. The differences between EU countries are large. In 2012 only 1 million in FDI flowed to Finland, whereas 2.775 billion went to the UK. This indicates heterogeneity.
Figure 1. Source: UNCTAD (2015)
-1 000 2 000 3 000 4 000 5 000 6 000 7 000 8 000 M ill io ns of US D ol la rs Years
Chinese FDI flows into the EU
The share of Chinese OFDI stock in developed countries is growing fast between 2003 and 2012. Figure 2 and 3 visualize the allocation of the Chinese OFDI stock to different parts of the world in 2003 and 2012. The allocation of Chinese OFDI stock in Europe has grown from 1% to 6% in only 10 years.
Figure 2. Source: UNCTAD (2015)
Figure 3. Source: UNCTAD (2015)
Europe 1% North America 2% Other developed countries 2% Africa 1% Asia 80% Latin America 14%
Chinese OFDI stock 2003
Europe 6% North America 4% Other developed countries 4% Africa 4% Asia 67% Latin America 13% CIS 2%
In recent years literature related to outward Chinese FDI flows has grown fast, mostly regarding flows towards developing countries. The main conclusion in most research is that Chinese firms invest in natural resource rich countries. Little research has been done so far on Chinese FDI flows towards developed nations. Since the flows into developed nations are growing exponentially, it is interesting to look into the country-specific determinants of Chinese FDI inflows in Europe. In previous research, the natural research endowment of a host country was the most important determinant of Chinese FDI flows. This thesis will investigate if this also holds for European
countries.
The research question is: What are the determinants of Chinese outward FDI flows
to developed European countries between 2003 and 2012 and how do they differ from flows into developing nations?
In the empirical part of this paper data of UNCTAD, the World Bank, the World Intellectual Property Organization and the National Bureau of Statistics of China will be used. A pooled OLS, a fixed effects and a random effects regression will be performed using panel data on the different European countries. The goal of this regression is to answer the research question.
This paper is structured as follows; in chapter 2 an overview is given of general theories on FDI and empirical literature regarding Chinese OFDI. In chapter 3 the empirical model is discussed with the model, variable selection, data and regression methods. In chapter 4 a selection of regression techniques will be discussed as well as the results. In chapter 5 the conclusion of this paper can be found after which the limitations will be discussed in chapter 6.
2. Literature review
2.1 Theoretical FDI literature
The study of Faeth (2009) summarizes nine theoretical models and econometric models of FDI. She argues that FDI should not be explained with a single theory because they do not exclude each other, but can be used as complements. Her paper is used as guidance to understand the different theories on FDI. Only the most important theories will be discussed in this chapter.
The newest method to describe FDI is based on game theory. An example of such a game is one with two players, the MNE and the host government, with different stages in the investment decision process. The investment decision process is affected by government policies and incentives. Faeth (2009) gives three different incentives to invest according to the game theoretical model. There are fiscal incentives (tax
advantages), financial incentives (government grants and credits) and other incentives (subsidies for dedicated infrastructure and subsidies for services). Although game theory is one way of describing FDI, the empirical model in this paper will be solely based on traditional FDI models.
The first theoretical model trying to explain FDI is the Heckscher-Ohlin model. This classical model assumes two countries, home and foreign. Both countries have two production factors (capital and labor) and two goods. The most important assumptions are perfectly competitive goods, identical constant returns to scale and no transaction costs. The Heckscher-Ohlin model is based on the idea that both countries have different factor endowments. Each country exports the factor abundant good since its internal relative price is cheaper compared to the other country. Thus in the absence of
commodity trade, a capital abundant country would move capital to the foreign country where the rates of return are higher than in the home country (Faeth, 2009). This neoclassical model is heavily criticized due to the assumption of perfect competition.
Hymer (1976) and Kindelberger (1969) criticized this approach and argued that the assumption of perfect competition could not explain FDI. They stated that market imperfections like monopolistic advantages were needed to explain FDI by a
multinational enterprise (MNE). They stated that MNEs only did foreign direct
investments if the advantages they would get from the investment were greater than the disadvantages from entering a foreign market. These monopolistic ‘ownership
advantages’ consisted of managerial expertise, new technology or product
differentiation. Disadvantages included more uncertainty, physical distance and a different culture and legal system. Buckley and Casson (1976) argued that firms internalize transactions through FDI since the market was often inefficient. Thus they assumed persistent market failure. Firms were better off internalizing those
transactions and the decision to internalize depended on industry-specific factors, region-specific factors, nation-specific factors and firm-specific factors.
Dunning (1979) brought these theories together in his OLI framework combining motivations for firms to operate internationally. His OLI framework consisted of
ownership, location and internalization advantages. Besides the already discussed ownership and internalization advantages, location advantages could be access to protected markets, favorable tax treatments and lower production and transport costs. For the internalization advantages Dunning (1979) also added minimizing technology imitation and maintaining the firm’s reputation through effective management and quality control. The different advantages allowed for a variety of determinants for FDI (Faeth, 2009). Dunning (1979) claimed firms could have three different motives to perform FDI, namely the resource-seeking, the market-seeking, and the efficiency-seeking motive. Most empirical research on Chinese OFDI is based on Dunning’s motives.
2.2 Empirical FDI literature
Besides theoretical literature on general FDI theories more and more research is done specifically on Chinese OFDI. This empirical literature tries to model the choices Chinese firms makes when they select countries to invest in. The model used in this thesis is mainly based on the four most important empirical studies investigating the determinants of Chinese FDI. For each paper the used sample, the methodology including the variables used and the variables that were found significant will be presented. This chapter ends with a summary of the variables taken from the literature that will be used in the methodology of this paper.
The first to investigate the determinants of Chinese OFDI flows are Buckley et al. (2007), who discuss Dunning’s (1979) three FDI motives mentioned in chapter 2.1. Resource-seeking FDI from developing countries is undertaken to acquire or secure raw materials and energy sources. In case of China these investments can also take place in developed countries with a resource surplus such as Canada and Australia. Market-seeking FDI takes place for trade support reasons, such as the access to distribution networks, facilitation of exports of domestic firms and to export to other large and fast growing markets. Efficiency-seeking FDI takes place when firms seek for lower
production costs, especially cheap labor. The authors argue this is unlikely in case of China given its relatively cheap labor.
Buckley et al. (2007) investigate the determinants of Chinese OFDI in the period 1984 to 2001 towards 49 worldwide countries. To do this they use a pooled ordinary least squares (POLS) and a random effects (RE) generalized least squares regression and do not control for endogeneity. They include several main variables linked to the FDI motives described by Dunning (1979). For the ownership advantages a variable that measures the number of patent registrations in de host country is added, for the resource seeking motive the ratio of ore and metal exports to merchandise exports is added and the market seeking motive is measured by the host country’s GDP, GDP per capita and annual GDP growth rate. Also the host country’s imports and exports from and to China capture the market-seeking motive. The authors state that Chinese OFDI can take place to provide local support for domestic Chinese exporters and to increase their hard currency earning. Furthermore, the political risk index rating, a dummy variable that is 1 when more than 1% of the host population is Chinese for cultural proximity and a dummy variable for the Chinese government policy that is 0 before Deng’s South China tour and 1 afterwards are added as main variables. The control variables the authors use are the host country inflation rate, the annual average exchange rate and the geographic distance to Beijing.
The significant variables Buckley et al. (2007) find are the GDP variables, the ore and metal exports, the political risk index rating, the cultural proximity and government policy dummies, the inflation rate and the exports and imports ratios to GDP. They conclude that high levels of political risk and cultural proximity to host countries initially determine Chinese FDI and in later years the host’s natural resource
endowments. However, since this thesis investigates OFDI flows into developed nations, the political risk and cultural proximity will not be added to this paper’s empirical model.
Zhang & Daly (2011) also study the determinants of Chinese FDI towards 23 host countries in the period 2003-2009 with a POLS regression. They use GDP per capita, the GDP growth rate, the share of ores and metals to merchandise exports, the host’s
inflation rate, the annual average exchange rate, openness to FDI of the host country measured by the ratio of inward FDI to GDP and exports and imports from China to the host country separately. The authors do not control for possible endogeneity problems. Zang & Daly (2011) find that high volumes of exports from China attract FDI for the host country. Also, openness to FDI, market size, economic growth and natural resource endowments promote Chinese FDI.
Kolstad & Wiig (2012) investigate the determinants of Chinese outward FDI in the period 2003 to 2006. The authors add an interaction variable to their regression that captures the interaction of the level of institutions and natural resource
endowments, the World Bank rule of law index, and the share of fuels, ores and metals exports of GDP. The interaction term can be used to test if Chinese FDI is attracted more to resource rich countries when the institutions are worse. Furthermore they add UNCTAD FDI data, host’s GDP, total imports and exports as share of GDP, inflation, distance between capitals, and the separate variables of the interaction term as main variables. To test for robustness they add additional control variables but none of these were significant and were left out of the main regression. Kolstad & Wiig (2012) use a regular OLS and a POLS regression, which give similar results. The Fixed effect (FE) did not provide significant results. To address endogeneity they lag explanatory variables. They expect however that endogeneity is unlikely to matter since Chinese FDI is still too minor in most countries to affect institutions. The authors conclude that market size and the presence of natural resources coupled with poor institutions are determinants of Chinese FDI given their significance. The sign of the interaction term is negative, which indicates that more Chinese FDI flows towards countries with natural resources the weaker institutions are. However, it is credible to assume that the level of institutions will not be a determinant of OFDI in Europe since most trade and investment
institutions are pan-European. This is why this interaction variable is not added to this paper’s model.
Chang (2014) uses an augmented gravity model and panel data from 2003 to 2009 to investigate the determinants of Chinese FDI in 138 countries. The difference in variables used, compared to previous research, is that the author uses a variable that is measured as the distance-weighted average real GDP of other host countries in the sample. Chang (2014) uses the number of patents of a host country to capture the technical standards of that country. He finds that Chinese firms prefer to invest in high-tech advanced countries. However, they also tend to focus on natural resources in developed and developing countries.
To summarize the empirical research, it seems that the resource-seeking motive, the market-seeking motive, economic growth, the ownership advantage, openness, the exchange rate and inflation are determinants of general Chinese FDI. All of them are added this paper’s empirical model. It is credible to assume that political risk is not a determinant for Chinese FDI in Europe since that risk is expected to be similar across European countries.
3. Empirical model and data
3.1. The model
In the literature different models are used to test for determinants of Chinese FDI. The previous section summarized the variables that were found significant.
OFDIit= α +β1Resourcesit + β2GDPCit + β3EXit + β4IMit + β5GDPGit + β6Patentit + β7Openit + β8Inflit + β9Exchit + εit (Model 1)
Where OFDI is the amount of annual Chinese FDI received at time t (t=1,….T) by country i (i=1,….N). α is the intercept and βk are the estimated coefficients of the
variables. Resources is the ratio of ore and metal exports to merchandise exports of the host country, GDPC is the real GDP per capita and GDPG the real GDP growth both in constant (2005) US$, EX and IM are the host country’s imports and exports from and to China and Patent is the number of total annual patent registrations (residents and non residents) in the host country. As control variables Open is the ratio of inward FDI stock to host country’s GDP, Infl is the host country’s inflation rate and Exch is the annual average exchange rate against the RMB fixed to the US$. Table 1 shows the connection of all variables to the literature, the expected sign and the source of the data.
Motive Variable Expected Sign Source
OFDI OFDI Dependent var UNCTAD
Resource-seeking motive Resources + World Bank
Market-seeking motive GDPC + World Bank
Market-seeking motive IM + EX + National Bureau of Statistics of China
Economic growth GDPG + World Bank
Ownership advantage Patent + World Intellectual Property Organization
Openness Open + UNCTAD
Inflation Infl - World Bank
Exchange rate Exch - UNCTAD
Table 1
Where some studies use logarithmic transformations of the variables, others do not. Buckley et al. (2007) use logarithmic transformations of all used variables but do not explain why. They only state that their data is transformed into natural logarithms as they expect non-linearities in the relationships on the basis of theory and previous empirical work without any further explanation. Kolstad & Wiig (2012) and Zhang & Daly (2011) do not use any logarithmic transformations, although they both state their model is based on the model of Buckley et al. (2007). Because of the lack of explanation on logarithmic transformations in the literature, this research will use a model with logarithmic transformations and one without. The models are expected to have different outcomes because of the negative and zero values of some of the variables, which
changes the number of observations used in the panel dataset. Model 2 is given by:
lnOFDIit= α +β1lnResourcesit + β2lnGDPCit + β3lnEXit + β4lnIMit + β5GDPGit + β6Patentit + β7Openit β8Inflit + β9Exchit + εit (Model 2)
All the variables are the same as in model 1, except logarithmic transformations have been made of the OFDI, Resources, GDPC, EX and IM variables.
3.2. Variable selection
3.2.1. Main variables
The literature review in this thesis shows that different sets of variables are used as determinants for Chinese OFDI. Model 1 and model 2 show the variables that will be used for this research. This section will firstly describe the main variables used after which the control variables will be presented. The main variables represent the motives for Chinese OFDI while the control variables represent the differences between host countries. All data used is annual data. The data used for the dependent variable is the Chinese outward flow of FDI.
The variable used for the resource-seeking motive is the share of ores and metals to merchandise exports (Buckley et al., 2007; Zhang & Daly, 2011). The share of ores and metals is used as proxy for the host country’s natural resource endowments. Kolstad & Wigg (2012) use the share of total fuel, ores and metal exports of host country’s GDP as a proxy for the resource-seeking motive, which is also found significant. Because two different variables are used for the natural resource
endowments it can be concluded that there is not one best proxy variable. This research will use the proxy that is used most in literature, the share of ores and metals to
merchandise exports. According to the literature previous mentioned, the resource-seeking motive is expected to have a positive relation with Chinese FDI.
The second main variable is a proxy for the market-seeking motive. The research mentioned in this paper’s literature review uses both absolute real GDP and real GDP per capita to proxy the market-seeking motive. Chakrabarti (2001) states that that absolute GDP is a relatively poor indicator of market potential for the products of
foreign investors. He summarizes different studies on FDI and most of them use GDP per capita as a main variable. GDP per capita is seen as a measurement of the host country’s income level while absolute GDP is seen as a measurement of the market size of a country (Chakrabarti, 2001). Since in this research all host countries are part of the European common market, an investment in one of the host countries gives direct access to the entire European market. This is an additional reason for this research to choose real GDP per capita as a proxy for the market-seeking motive over absolute real GDP. The variable, in constant (2005) US$ prices, is expected to have a positive relation with the inflows of Chinese FDI (Buckley et al., 2007).
Two trade variables are also used as a proxy for the market-seeking motive, the host country’s imports from China and the exports to China are measured per 10,000 US$. Buckley et al. (2007) expect the host country’s imports to have a positive relation with Chinese FDI inflows.
Buckley et al. (2007) state that OFDI is associated positively with host market growth. In his growth hypothesis he states that rapidly growing economies provide more profitable investment opportunities than slower growing economies. Therefore, also real GDP growth in constant (2005) US$ prices will be added to the regression model as the fifth main variable. For completeness purposes, this paper also tried out a model without GDP growth since it was not found significant in all literature and because of its negative correlation with GDP per capita. However, the results were
almost identical to the model with GDP growth included so it is added to model 1 and model 2. An explanation for the negative correlation can be found in the growth theory where countries with a low GDP per capita are expected to have a higher growth rate.
Monopolistic ownership advantages like managerial expertise, new technology or product differentiation, can be reasons to undertake OFDI (Dunning, 1979). Host
countries are able to attract FDI when these ownership advantages are present. Buckley et al. (2007) and Chang (2014) use the number of patent registrations as a proxy for these advantages. This research also adds the number of patent registrations as the last main variable. Chinese OFDI is associated positively with host country’s endowments of ownership advantages.
3.2.2. Control variables
The first control variable that will be added to the regression model is the host country’s openness to OFDI. This variable is measured by the ratio of inward FDI stock as percentage of GDP. The more open a country is to international investment, the more attractive it is to be a destination for OFDI (Chakrabarti, 2001). Buckley et al. (2007) expect openness to have a positive relation with Chinese FDI inflows.
When a country’s exchange rate appreciates, foreign assets become cheaper and more profitable opportunities for OFDI will arise. For this reason the annual average exchange rate of the Renminbi (RMB) to the home country’s currency fixed to the US$ is added as a control variable. The variable is expected to have a negative relation with de dependent variable. The same holds for the inflation rate of the host country. A rise in the inflation rate and the exchange rate leads to higher investment costs.
3.3. Data
It is important to have a clear definition of foreign direct investment. The Organization for Economic Co-operation and Development (2009) gives the following definition, which is also used by UNCTAD (2015):
“The statistics on direct investment include cross-border investments made with the objective of establishing a lasting interest to exercise an influence in the
management of the direct investment enterprise (the target entity) which is evidenced by the ownership of at least 10% of the voting power by the direct investor.”
The data sample used for the empirical analysis will consist of annual panel data from 2003 to 2012 for 28 European countries1. The 28 countries are all developed European countries where any amount of inward Chinese FDI took place between 2003 and 2012 (UNCTAD, 2015). A sample period of more than ten years is not possible since China didn’t start with specifying FDI data by destination and not according to
international standards until after 2003 (Kolstad & Wiig, 2012). The data is only
1Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, the Netherlands, Norway, Poland, Portugal, Romania, Slovakia, Spain, Sweden, Switzerland and the United Kingdom.
available until 2012. Besides the OFDI data, also the Openness data (the inward FDI stock as percentage of host country’s GDP) and the host country’s official annual average exchange rate against the RMB come from UNCTAD(2015). The GDP variables, the ratio of ore and metal exports to merchandise exports and the inflation rate have the World Bank (2016) as its source. The number of annual patent registrations (resident plus non-resident) is reported by the World Intellectual Property Organization (2016). The amount of imports and exports from and to China in 10,000 US$ comes from the
National Bureau of Statistics of China (2016). Table 1 shows the sources of the variables used in this thesis.
Variable Obs Mean Std. Dev. Min Max
OFDI 280 88.80 388.90 -58.91 3482.32 Resources 280 3.87 3.39 0.27 20.19 GDPC 280 30102.86 19497.50 3335.88 87772.69 Ex 280 490719.50 1134533.00 111.90 9274397.00 Im 280 837317.90 1335168.00 11062.20 7640005.00 GDPG 280 1.92 3.81 -14.81 11.90 Patent 278 5275.39 11984.61 8 62417 Open 278 84.36 204.41 9.92 1809.92 Infl 280 2.83 2.29 -4.48 15.43 Exch 280 1.34 5.19 0.07 35.66 Table 2
Table 2 shows the summary statistics for the dataset. For the openness two Belgian and for the number of patents two Maltese observations are missing. The panel dataset is not perfectly balanced, however STATA labels the data as strongly balanced and so can be used for panel regression models. As is shown in the summary statistics, there is a lot of variation in all variables.
3.4. Regression methods
As mentioned before, the dataset in this research consists of panel data. The advantage of panel data is that it increases the precision of the estimation because of the pooling of time periods for each country. Three of the main empirical papers discussed in the empirical literature review also use panel data for their analysis (Buckley et al., 2007; Kolstad & Wiig, 2012; Zhang & Daly, 2011).
In this thesis different regression techniques will be considered. A pooled ordinary least squares (POLS) regression, a fixed effects (FE) regression and a random effects (RE) regression are used. A FE and RE regression can control for some forms of the omitted variables (Stock, Watson, & Addison-Wesley, 2007). Since the literature uses different regression methods, it can be concluded there is not one clear best method. This is why all three regression methods will be performed. With a F-test it can be tested if a FE regression gives better results than a POLS regression. With the Haussmann test it can be tested whether there is statistical evidence to choose the FE regression over the RE regression. Finally, the Breusch and Pagan Lagrangian multiplier (LM) test (Breusch & Pagan, 1980) can be used to test whether the RE or the POLS regression is preferred.
Although panel data can control for some forms of the omitted variable bias, it does not control for endogeneity problems. Endogeneity can take place because of simultaneous causality. There is simultaneous causality when causality runs from the dependent variable to an independent variable and backwards, causing the OLS
estimator to the biased and inconsistent (Stock et al., 2007). In both model 1 and model 2 there could be a simultaneity bias when there is strong correlation between Chinese FDI and one of the independent variables.
Theoretically it could be possible for simultaneous causality to arise between OFDI and the GDP per capita, GDP growth, the trade variables, the number of patent registrations and the exchange rate. Having a high GDP per capita or high GDP growth can lead to more inflows of FDI. However, more FDI inflows can lead to more
employment that leads to a higher income and GDP. A higher trade intensity between countries leads to more FDI flows. The other way around more OFDI flows can raise the supply of goods in the host country, which could raise trade. The number of patent registrations is expected to have a positive effect on FDI, but FDI could also have a positive effect on the number of patent registrations. New investors can incite firms to innovate. These innovations could raise the number of patent registrations of that firm. The same holds for the simultaneous causality between FDI and the exchange rate. When the FDI flows are large enough, the investments raise the demand for the host country’s currency. This risen demand could cause the currency to appreciate.
Although simultaneous causality could theoretically be possible, the two
correlation matrices in tables 8 and 9 in the appendix show that the correlation between the dependent en independent variable is low. The only correlation with a different sign than expected is the correlation between Chinese OFDI and GDP growth, which is found negative for both models. A possible explanation could be that countries with a low GDP growth are actively seeking for foreign investments, which could lead to more FDI inflows. Low correlations alone do not exclude the simultaneity bias. Kolstad & Wiig (2012) argue in their paper that Chinese FDI is too small in most African countries to affect their main variables. It is credible to assume, since the Chinese FDI flows into Europe are also relatively small, that endogeneity problems do not affect the regression results of this thesis.
To test for multicollinearity the Variance Inflation Factor (VIF) test can be performed. The VIF test and tolerance, 1/VIF, are both widely used measures of the degree of multicollinearity of an independent variable with other independent variables in a regression model (O’brien, 2007). O’brien (2007) gives an overview of previous research and rules of thumb for the VIF test values. According to the author two commonly used threshold for the VIF test are VIF>5 and VIF>10. These rule of thumbs do not automatically exclude multicollinearity but are good indicators. In table 3 and 4 the results of the VIF tests for respectively model 1 and model 2 are shown. Since none of the VIF values exceed the lowest threshold it is safe to assume there is no
Variable VIF 1/VIF Ex 4.3 0.232404 Patent 3.6 0.277733 Im 2.66 0.37658 Infl 1.36 0.734843 GDPC 1.28 0.778921 Resources 1.13 0.884117 Exch 1.12 0.891702 GDPG 1.1 0.91173 Open 1.04 0.963337 Mean VIF 1.95 Table 4 (Model 1)
Variable VIF 1/VIF
ln_Ex 3.49 0.286356 ln_Im 2.98 0.335426 ln_GDPC 2.45 0.408524 Infl 1.79 0.559113 Patent 1.73 0.576797 Open 1.53 0.653428 Resources 1.46 0.684309 Exch 1.37 0.728955 GDPG 1.17 0.852854 Mean VIF 2 Table 3 (Model 2)
4. Results
4.1. Regression model evaluation
Three different types of regressions will be performed for both models. Three tests will be conducted to test if the fixed effects, the random effects or the POLS regression provides the best results. The three tests will be discussed first, after which the results of the regression models will be discussed.
With a F-test it can be tested if a FE or a POLS regression is preferred. The null hypothesis states that all intercepts are equal. This would mean that all country specific effects are the same and implies that the POLS regression would be appropriate. The alternative hypothesis entails that at least one of the countries has a different intercept, which implies the FE regression is preferred. Model 1 gives a F-value of F(27,239)=2.15 with a p-value of 0.0013 and model 2 gives a F-value of F(26,117)=1.91 with a p-value of 0.0107. This means that with a 5% significance level the null hypothesis is rejected in both cases. This means that for model 1 and model 2 the FE regression is preferred to the POLS regression.
The Hausman test can be used to test whether the FE or the RE regression is preferred. The Hausman test identifies if there is a systematic difference between the coefficients of the FE and RE regression. It tests whether the unique errors (country specific) are correlated with the independent variables. The null hypothesis states that the RE regression is preferred and the alternative hypothesis states that the FE
regression is preferred. The Hausman test for model 1 gives a chi squared of 7.78 with a p-value of 0.3523. This means that for model 1 the RE regression is preferred over the FE regression. The Hausman test for model 2 does not present a value for chi squared because the between-entity error term has a value of zero. The between-entity error term measures the difference between the average OFDI in countryi and the average
OFDI in the sample. When this error term is zero it basically means none of the variance in OFDI is caused by the differences across panels, so there is no panel data effect in the RE regression. This would imply that a RE regression would give the same results as a POLS regression. The results in table 6 show that this holds. A possible explanation could be that the many zero values of the dependent variable are not taken into account in model 2 because of the logarithmic transformations of the data. The loss of data in the panel data sample could explain the loss of the panel data effect in the RE regression. This means that for model 2 only the FE and POLS regression have to be compared,
which is done with the F-test.
The Breusch and Pagan Lagrangian multiplier (LM) test (Breusch & Pagan, 1980) can be used to test whether the RE or the POLS regression is preferred. The null
hypothesis in the LM test is that variances across entities are zero. This means that there is no significant difference across units, so no panel effect, and the POLS regression is preferred. The RE regression is preferred under the alternative hypothesis. The results after the regression of model 1 are a chi squared of 5.13 with a p-value of 0.0117. This means that for model 1 the RE regression is preferred.
After running the three tests it is clear that for model 1 the RE regression is the preferred regression technique since the RE regression is preferred to the POLS and to the FE regression. For model 2 the FE regression is the preferred regression technique since the F-test prefers the FE to the POLS regression. For completeness purposes for both models all regression results are shown.
4.2. Regression results
Regressions Model 1 with dependent variable OFDI
FE RE POLS Resources -2.56867 10.55487 10.97412 (19.25477) (8.10090) (6.97831) GDPC -0.01419 0.00482*** 0.00511*** (0.01681) (0.00152) (0.00129) Ex 0.00005 0.00003 0.00001 (0.00006) (0.00004) (0.00004) Im 0.00014** 0.00009*** 0.00008*** (0.00006) (0.00003) (0.00003) GDPG 2.08212 1.52887 2.12116 (6.11352) (5.99917) (6.08690) Patent -0.03449 -0.00341 -0.00108 (0.02508) (0.00377) (0.00352) Open 0.12868 0.15582 0.17080 (0.17327) (0.11973) (0.11066) Infl -0.35553 6.57081 9.73986 (12.60618) (11.48121) (11.29976) Exch 9.19432 4.61196 4.33889 (30.51544) (5.39002) (4.51965) _cons 539.38622 -206.25544** -222.78506*** (505.82395) (80.45096) (71.76921) N 276 276 276 R2 0.105 0.139
Standard errors in parentheses * p < .10, ** p < .05, *** p < .01
Table 5
Table 5 presents the results of the regressions of model 1. The first number shows the coefficient and the number between brackets shows the standard error. The number of observations for model 2 is smaller than the number of observations for model 1 because of the zero values for the dependent variable. The natural logarithm of zero does not exist and STATA leaves out these observations.
As tested in chapter 4.1, only the results of the RE regression have to be considered. The first strongly significant variable is the GDP per capita, the market-seeking motive. A 1-unit rise in the GDP per capita leads to a 0.00482-unit rise in FDI inflows.
The second strongly significant variable is the host country’s imports from China, another proxy for the market-seeking motive. A 1-unit rise in the imports from China gives a 0.00009-unit rise in the amount of Chinese OFDI. Again, this is line with the research of Zhang & Daly (2011) and Kolstad & Wiig (2012) and with the theory that better trade relations encourage more FDI inflows. This result is in line with the theory that the market-seeking motive is a determinant of Chinese OFDI and was also found significant in previous research (Buckley et al., 2007; Kolstad & Wiig, 2012; Zhang & Daly, 2011).
Surprisingly, the resource-seeking motive is not found significant in the regression analysis of this model. Another significance level that stands out is the
coefficient of the ownership advantage. The number of patent registrations seems not to significantly affect Chinese OFDI in Europe. The low value of the R2, a measure of how much of the variance in OFDI is explained by the model, can be explained by the high amount of zero values for the dependent variable.
Regressions Model 2 with dependent variable lnOFDI FE RE POLS Resources 0.11551 0.09546** 0.09546* (0.30960) (0.04846) (0.04846) ln_GDPC -0.79670 0.14417 0.14417 (2.91442) (0.24675) (0.24675) ln_Ex 0.35911 0.15849 0.15849 (0.31333) (0.13841) (0.13841) ln_Im 0.89659** 0.57475*** 0.57475*** (0.39307) (0.16040) (0.16040) GDPG -0.01898 -0.04286 -0.04286 (0.04956) (0.04013) (0.04013) Patent -0.00024** 0.00003*** 0.00003*** (0.00011) (0.00001) (0.00001) Open 0.02985*** 0.01569*** 0.01569*** (0.00766) (0.00241) (0.00241) Infl 0.10945 0.10495 0.10495 (0.10024) (0.07794) (0.07794) Exch 0.16563 0.02701 0.02701 (0.13224) (0.02222) (0.02222) _cons -6.30025 -10.17258*** -10.17258*** (27.22080) (2.98893) (2.98893) N 153 153 153 R2 0.384 0.486
Standard errors in parentheses * p < .10, ** p < .05, *** p < .01
Table 6
Table 6 shows the results of the regressions of model 2. As tested in chapter 4.1, only the results of the FE regression have to be considered. The first significant variable is the proxy for the ownership advantage, the number of patent registrations. The ownership advantage is significant and surprisingly it is negative, although the value is small. This means the ownership advantage is a determinant of Chinese OFDI. The more patent registrations in a country, the less a country will receive Chinese FDI inflows. The sign is opposite of the expected positive sign. The variable is not significant in the paper of Buckley et al. (2007). It is credible to assume that the value is too small to have a real effect on the OFDI inflows.
The second strongly significant variable is the amount of imports from China by the host country. The FE regression indicates that a 1% increase in the host country’s imports from China raise Chinese outward FDI with 0.89659%. This result is in line with
the theory that the market-seeking motive is a determinant of Chinese OFDI.
The third and last strongly significant variable is the openness variable. Theory suggests that the more open a country is to FDI, so when FDI stock is present in a country, the easier it will receive FDI inflows. For the FE regression this means that a rise in the ratio of inward FDI stock to host GDP of 1 percentage point will raise the OFDI inflows with 0.02985%. The significance level of the openness variable is in line with the research of Zhang & Daly (2011).
Surprisingly, the resource-seeking motive, the share of ores and metal exports to merchandise exports, is only significant at a 5% level in the RE regression an not in the FE regression. This means we cannot conclude that the natural resource endowments determine Chinese FDI inflows. Buckley et al. (2007), who also use logarithmic
transformations, only find a significant result for this variable in their POLS regression including all countries and not for their RE regression only including OECD countries. This could imply that the natural resource endowments only determine Chinese OFDI in developing countries and not in developed countries.
Also the GDP variables are not significant in the regression. This means that according to these variables the market-seeking motive of Dunning (1979) seems not be a determinant of Chinese FDI. At last, the R2 of 0.384seems to be in line with the values found in previous research.
4.3. Summary and Comparison
Table 7 shows the signs of the significant results found after the regressions of model 1 and model 2. It also shows if the result was found with a FE, RE of POLS regression.
Variable Model 1 Model 2
Resource-seeking motive
Market-seeking motive GDPC + & IM + (RE) ln_IM + (FE)
Economic growth
Ownership advantage Patent + (FE)
Openness Open - (FE)
Inflation
Exchange rate
Table 7
Table 7 shows a summary of the results of the regression models. The first four motives consist of main variable and the three variables under the empty line are the control variable. Taking logarithmic transformations of certain variables in the model changes the significance level of main variables. Depending on the chosen model, the market-seeking motive, the ownership advantage and openness to FDI all have an effect on Chinese OFDI. Although the ownership advantage is significant, is does not have the
expected sign. A possible explanation could be that the coefficient is too small to matter. This would imply that there the number of patents actually does not effect Chinese OFDI in Europe. Another explanation could be that the number of patent registrations is not a good proxy for the ownership advantage. The only motive that affects Chinese OFDI flows in both models is the market-seeking motive. The resource-seeking motive it not significant in any of the two models. It cannot be concluded that the resource-seeking motive is a determinant for Chinese FDI in Europe.
5. Conclusion
Between 2003 and 2012 China’s OFDI stock grew with a factor 15. Because of this growth China is now ranked third globally in OFDI flows. Where in the early years of Chinese outward FDI most flows went towards neighboring and resource rich countries, nowadays we see a great increase in flows of FDI toward developed countries. Previous research mainly concludes that the natural resource endowment of a country is one of the main reasons to receive Chinese FDI.
If Dunning’s OLI model were correct, one would expect the resource-seeking motive to be a significant determinant in Chinese FDI flows to developed European countries. On the other hand, one could expect ownership advantages to be more
important when investing in developed countries than in developing countries. This had led to the following research question: What are the determinants of Chinese outward FDI
flows to developed European countries between 2003 and 2012 and how do they differ from flows into developing nations?
To be able to answer this question two regression models were constructed. The models consist of 7 main and 2 control variables. The panel data sample consist of 28 developed European country in the period 2003 to 2012. The data used comes from UNCTAD, the World Bank, the World Intellectual Property Organization and the National Bureau of Statistics of China.
Three tests to see which regression technique is most appropriate were
conducted. The test results of model 1 show that a RE regression is most appropriate. The market-seeking motive measured by real GDP per capita and the host country’s imports from China were measured strongly significant. This means the market-seeking motive is tested as a determinant of Chinese OFDI flows to European developed
countries. It indicates that having a high income and having good trade relations with China has a positive effect on receiving Chinese FDI inflows.
Model 2 showed that the FE regression is most appropriate and was performed. There were three variables that were strongly significant in the regression, namely the market-seeking motive measured by host country’s imports from China, ownership advantage measured in number patent registrations and the openness to FDI measured by the ratio of inward FDI stock as percentage of GDP. These variables are tested as determinants of Chinese OFDI to European developed countries. However, since the ownership advantage has the opposite effect of what was expected and is close to zero, it cannot be concluded that is a determinant of OFDI flows. The other significant variables
indicate that the trade relation intensity with China and being a being open to foreign investments have a positive effect on receiving Chinese FDI inflows.
It is interesting that logarithmic transformations of some of the variables give different results. Since the literature is not consistent in the models used, different results have to be considered. The market-seeking motive and the openness to FDI can be considered determinants of Chinese FDI flows to European developed countries, where only the market-seeking motive is significant in both models. It can be concluded that China undertakes OFDI in Europe for trade support reasons. Considering both models, it can be concluded that the resource-seeking motive is not a determinant of Chinese OFDI flows to developed European countries. A possible reason could be the lack of abundance in natural resources in Europe because of the high internal demand. This means the determinants of Chinese OFDI flows differ for European and developing countries.
6. Limitations and future research
The most important limitations of this research are related to the data availability of the dependent variable. A sample period of more than ten years is not possible since China didn’t start with specifying FDI data until after 2003, and the data is only available until 2012. Also, the dependent variable has a lot of values of zero since not every country received Chinese FDI inflows every year. Especially in model 2 with logarithmic transformations, this causes the number of observations to be limited. The number of observations is reduced from 276 to 153 compared to the results of model 1. In model 1 the values of zero for the dependent variable result in a very low R2.
Another limitation is that there is no clear empirical FDI model for Chinese FDI in the literature. The models in the literature differ in variables used, but also in the use of logarithmic transformations. Future research could therefore use a data sample
consisting of a longer time period. The zero values are expected to become less since the Chinese OFDI flows are expected to keep rising. This will probably result in more
significant determinants and a higher R2. Future research could also experiment with different models and variables. At last, when data is available over a long time period, future research could test if the determinants of Chinese FDI in Europe change over time. A possible research question could be if the determinants for Chinese OFDI flows changed after the 2008 economic crisis.
7. References
Breusch, T. S., & Pagan, A. R. (1980). The lagrange multiplier test and its applications to model specification in econometrics. The Review of Economic Studies, 47(1), 239-253.
Buckley, P. J., & Casson, M. (1976). The future of the multinational enterprise Macmillan London.
Buckley, P. J., Clegg, L. J., Cross, A. R., Liu, X., Voss, H., & Zheng, P. (2007). The
determinants of chinese outward foreign direct investment. Journal of International
Business Studies, 38(4), 499-518.
Chakrabarti, A. (2001). The determinants of foreign direct investments: Sensitivity analyses of cross-country regressions. Kyklos, 54(1), 89-114.
Chang, S. (2014). The determinants and motivations of china's outward foreign direct investment: A spatial gravity model approach. Global Economic Review, 43(3), 244-268.
Dreger, C., Schüler-Zhou, Y., & Schüller, M. (2015). Determinants of chinese direct investments in the european union.
Dunning, J. H. (1979). Explaining changing patterns of international production: In defence of the eclectic theory. Oxford Bulletin of Economics and Statistics, 41(4), 269-295.
Faeth, I. (2009). Determinants of foreign direct investment–a tale of nine theoretical models. Journal of Economic Surveys, 23(1), 165-196.
Hymer, S. H. (1976). The international operations of national firms: A study of direct
foreign investment MIT press Cambridge, MA.
Kindelberger, C. P. (1969). American business abroad: Six lectures on direct investment.
New Haven, CT: Yale UP,
Kolstad, I., & Wiig, A. (2012). What determines chinese outward FDI? Journal of World
Business, 47(1), 26-34.
Li, C., Liu, H., & Jiang, Y. (2015). Exchange rate risk, political environment and chinese outward FDI in emerging economies: A panel data analysis. Economics, 3(5-6), 145-155.
National Bureau of Statistics of China. (2016). Statistical database. Retrieved from http://www.stats.gov.cn/english/Statisticaldata/AnnualData/
O’brien, R. M. (2007). A caution regarding rules of thumb for variance inflation factors.
Quality & Quantity, 41(5), 673-690.
Organisation for Economic Co-operation and Development. (2009). OECD benchmark
definition of foreign direct investment 2008 OECD Publishing.
Stock, J. H., Watson, M. W., & Addison-Wesley, P. (2007). Introduction to econometrics. The World Bank. (2016). World bank development indicators. Retrieved from
UNCTAD. (2015). Retrieved from
http://unctad.org/Sections/dite_fdistat/docs/webdiaeia2014d3_CHN.pdf
World Intellectual Property Organization. (2016). Patent registrations. Retrieved from http://ipstats.wipo.int/ipstatv2/index.htm
Zhang, X., & Daly, K. (2011). The determinants of china's outward foreign direct investment. Emerging Markets Review, 12(4), 389-398.
8. Appendix
Model 1 OFDI Resources GDPC Ex Im GDPG Patent Open Infl Exch
OFDI 1 Resources 0.0149 1 GDPC 0.2535 -0.1789 1 Ex 0.2102 -0.1058 0.1542 1 Im 0.2747 -0.1224 0.1961 0.7719 1 GDPG -0.0267 0.0695 -0.1238 -0.0794 -0.1267 1 Patent 0.1756 -0.125 0.1378 0.8399 0.7147 -0.0641 1 Open 0.0631 -0.0999 0.0219 -0.0601 -0.0794 -0.0161 -0.1175 1 Infl -0.0558 0.2101 -0.4222 -0.1622 -0.1475 0.258 -0.1586 -0.0354 1 Exch -0.027 -0.14 -0.1974 -0.0882 -0.1088 -0.0411 -0.0771 -0.0248 0.1837 1 Table 8
Model 2 ln_OFDI Resources ln_GDPC ln_Ex ln_Im GDPG Patent Open Infl Exch
ln_OFDI 1 Resources 0.062 1 ln_GDPC 0.2893 -0.3136 1 ln_Ex 0.3775 -0.2945 0.3653 1 ln_Im 0.4676 -0.2146 0.3312 0.8024 1 GDPG -0.1427 0.0606 -0.2063 -0.1628 -0.1662 1 Patent 0.362 -0.13 0.1803 0.6036 0.5688 -0.0235 1 Open 0.3322 0.1922 0.2909 -0.2772 -0.1975 -0.0164 -0.2993 1 Infl -0.155 0.1173 -0.6164 -0.3355 -0.3005 0.298 -0.1837 -0.0978 1 Exch -0.0816 -0.1661 -0.2899 -0.2608 -0.2726 -0.0754 -0.1308 0.012 0.275 1 Table 9
