Does investing in human capital attract FDI
in East Asia?
Name: Leonard Knottnerus (11416238)
Supervisor: Drs. N. Leefmans Second reader: Prof. dr. F. Klaassen
Date: 14/07/2017
MSc thesis
University of Amsterdam
Faculty: Economics and Business
MSc Economics: International Economics and Globalization
Summary
This research studies the effect of human capital on FDI inflows into East Asian countries. Panel data regressions have been used to find out if human capital levels are of influence on the level of FDI inflows and if so, which type of human capital. To do so, 23 countries have been studied in the time period ranging from 1970-2015. The estimations show a significant positive relation between tertiary education and FDI, while the other human capital indicators are found to be insignificant in terms of impacting FDI inflows.
2 Statement of Originality
This document is written by Student [Leonard Knottnerus] 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.
3 Table of contents page 1. Introduction 4 2. Literature review 2.1 FDI determinants 7
2.2 Human capital as determinant for FDI 11
3. Data and research methods
3.1 The model 15
3.2 Selection of variables 16
4. Results and discussion 20
5. Conclusion 29
References 31
4 1. Introduction
Living standards and economic growth have increased dramatically in East Asia since the 1960s (Zhuang, 2016). This phenomenon is often referred to as the ‘East Asian miracle’. The East Asian miracle left scholars with the question as to what factors and policies contributed to this rapid growth of the region (Stiglitz, 1996). One characteristic of East Asia that stands out in recent decades is the high (growth) level of Foreign Direct Investment (Zhuang, 2016). These foreign direct investment (FDI) inflows have, arguably, contributed to growing wealth and development in the region (Zhang, 2001). Growth theory supports this statement because it indicates that FDI gives countries access to international capital markets, new technology and skills (Zhuang, 2016). These developments, in turn, contribute to technological innovation and economic growth. There is a large amount of literature about the possible effects of FDI on economic growth and productivity. See for example Lloyd (1996) and Alfaro, Chanda, Kalemli-Ozcan and Sayek (2010). Most of these papers point out that there is indeed a positive relation between FDI and economic growth. Another positive feature of FDI is its low volatility. FDI is less volatile than other types of capital flows, according to Chuhan, Perez-Quiros and Popper (1996). Thus, it seems that governments might have incentives to promote FDI inflows.
This raises the question as to what factors are of influence on the level of FDI inflows. There is a vast economic literature about the determinants of FDI. It is important to note that the determinants of FDI change over time and are in most cases country/region specific (Quazi, 2007). Quazi (2007) identifies the following determinants that have a positive effect on FDI inflows for East Asia: domestic investment climate, domestic market size, return on investment, lagged changes in FDI and political stability. Also, country specific factors can be of great importance. For instance, until 1998, South-Korea levied a high tax on FDI inflows which obviously has had a negative impact on FDI inflows during that period. A study by Vijayakumar, Sridharan and Rao (2010) identifies labour cost, infrastructure, currency value and gross capital formation as additional determinants for the BRIC countries. So the cost of labour (wages) appears to affect FDI inflows as well. However, there seems to be a trend from the importance of real wages as a determinant of FDI, towards education of the workforce as a main driver of FDI (Pfeffermann and Madarassy, 1992; Noorbakhsh, Paloni and Youssef ,2001; Faeth, 2009). FDI has shifted towards more knowledge- and skill intensive industries since the 1980s, thus it seems likely that the availability of skilled labour is becoming more
5 important as a pull-factor with respect to FDI. This provokes the question as to what the (strength of the) relationship is between FDI and human capital investments. Thus, the research question of this thesis is: “Does investing in human capital attract FDI for countries in East Asia?”
The economic literature points out that the relationship between FDI and human capital
appears to be bi-directional1. First, FDI can affect human capital development via technology
transfers. Literature about this channel of human capital development is quite abundant. For example, Zhuang (2016) finds that FDI, overall, enhances secondary education in developing countries, while it has a negative effect on tertiary education. However, FDI from ‘rich countries’ enhances both secondary and tertiary education. Thus it appears to be important to differentiate with respect to different source countries. The second relation between FDI and human capital is that human capital affects FDI, instead of the other way around. Countries can invest in human capital which might attract FDI. This latter effect is the one that this thesis is about. The empirical literature in support of this relationship between human capital enhancement and FDI is scant, especially when narrowed down to a specific group of countries, in this case East Asian countries.
A fairly recent and influential study that investigates the effect of human capital on FDI inflows is the paper by Noorbakhsh et al. (2001). They have some important findings. First, they find that human capital is a significant determinant of FDI. Second, they even find that human capital is amongst the most important determinants with respect to FDI. Third, the importance of human capital has increased over time. They base their findings on panel-data of 36 countries in Africa, Asia and Latin America in the period 1980-1994. Most of these countries lie in Africa and Latin America and the only ones in Asia are: Malaysia, The Philippines and Thailand. Furthermore, they use an overly restrictive empirical model in which they do not allow for country specific fixed effects. When they ran the model with country specific fixed effects, the results were poor.
This thesis will add to the economic literature because it will test whether the results found by Noorbakhsh et al. (2001) apply to countries in East Asia as well. The dataset of Noorbakhsh et al. (2001) only contains three countries in East Asia, while this research will study twenty-three countries in the region. Moreover, the countries selected by Noorbakhsh et al. (2001) are
1 To control for the endogeneity that results from this bi-directional relationship, the lagged values of human
capital are also examined. It is reasonable to expect that it takes a few years or more for FDI flows to adjust to changes in human capital levels.
6 mostly countries that are in the early stages of development, while this thesis will also include more developed and highly developed countries such as South-Korea and Japan. Furthermore, I shall test whether the results of Noorbakhsh et al. (2001) hold when different control variables are included, using three different empirical models: Ordinary Least Squares (OLS); Random Effects; and Fixed effects. Three models have been used in order to find out which one is the most appropriate and to test the robustness of the results. The final results are based on fixed effects estimations with heteroskedasticity-robust standard errors, while Noorbakhsh et al. (2001) have based their findings on the OLS model. Most importantly Noorbakhsh et al. (2001) use a dataset on education by Nehru, Swanson and Dubey (1995), which ranges from 1980-1994. My thesis shall rely upon more recent data, namely the period ranging from 1970-2015. This is relevant because there is reason to believe that human capital has become of growing importance in recent years since FDI has started to shift to the skill intensive industries since the 1980s (Arnal and Hijzen, 2008; UNCTAD, 2016). This shift in FDI will be explained more extensively in chapter 2. Also, in this thesis a distinction is made between two types of source countries, namely developed countries and the rest of the world. I expect that human capital has a larger effect on attracting FDI from developed countries because these countries produce more skill intensive goods and therefore are expected to react more strongly to changes in the skill levels of workers in recipient countries.
The thesis is structured in the following way: Chapter 2 provides a literature review, which consists of two parts. The first part discusses determinants of FDI (in general) and the second part discusses human capital, specifically, as determinant of FDI. Chapter 3 discusses the empirical model and the variables used. Chapter 4 states the empirical results of the research; Chapter 5 draws conclusions based on the results and provides a policy recommendation.
7 2. Literature review
2.1 FDI determinants
FDI inflows in East Asia have increased dramatically since the 1980s. Appendix 1 tells us that their share relative to Gross Domestic Product (GDP) has sharply increased as well, in the
period 1985-20102. The average share of FDI in GDP was approximately 1% in 1985, while
in 2010 this number has risen to over 7%. When we differentiate with respect to net FDI inflows for different recipient countries in East Asia one country in particular stands out, namely China. China is the largest FDI recipient developing-country, since the 1990s. The only country in the world that receives more FDI inflows is the United States (UNCTAD, 2016). The high growth of FDI in East Asia indicates that the region must have features that make it attractive to foreign firms. Especially firms from OECD countries seem to be interested in the region, given their high share in total FDI in East Asia (appendix 4). An average of over 338 billion dollar of annual FDI inflows (in the period 1970-2009) originates from OECD countries, which is roughly one-third of total FDI inflows in the region. To explain why (developed) countries are so interested in East Asia, it is of importance to differentiate with respect to different kinds of FDI.
An author that has done a lot of research on different sorts of FDI and the underlying
incentives that MNEs face is J.H. Dunning. Dunning (1980) states that in order to understand why firms engage in FDI one has to make a distinction between horizontal- and vertical FDI. Horizontal FDI refers to production of products and services similar to those produced in the source country. Vertical FDI refers to geographical fragmentation of the production process, meaning different stages of the production process are performed in different countries. With respect to the preconditions for the occurrence of FDI, Dunning (1980) developed the eclectic paradigm. This paradigm is also known as the Ownership, Internalization, and Location (OIL) framework. Initially MNEs are disadvantaged when compared with local firms: MNEs have to adapt to a different rule of law, different business culture, different consumers’ preferences and so on (Veestraeten, 2017). Therefore, MNEs must have comparative advantages in order for FDI to be an attractive business strategy to them.
Within the OIL-framework, having ownership advantages means that a firm has firm-specific assets that other companies do not have. Examples of firm-specific assets are: patents,
2 The list of East Asian countries Zhuang (2016) used is the following: Brunei, Cambodia, China, Hong Kong, Indonesia, Japan, Korea, Laos, Macau, Malaysia, Mongolia, Myanmar, Philippines, Singapore, Thailand and Vietnam.
8 differentiated product, superior labour skills, experience, and reputation. But having
ownership advantages alone is not enough for companies to pursue FDI because they can also exploit the market power that results from the ownership advantages by exporting their products or by giving a license to a firm in the host country. However, giving licenses to local producers is often not an option because this would result in high transaction costs. For instance, giving a license may entail costly training of the workers of the firm in the host country and may require the use of costly expats. The existence of large transaction costs may stimulate the innovating firm to exploit the competitive advantage within its own organisation (internalization). If high transaction costs rule out licensing, the only option that a firm has besides FDI, is exporting its product. Therefore, there also needs to be a locational advantage to induce the firm to firm to do FDI. Locational advantages can be things like access to protected markets, market size, market growth, (cheap) labour costs and low taxes abroad (Veestraeten, 2017). The OIL framework and the corresponding actions by firms are depicted in figure 1. It shows that, in theory, firms have to meet all three comparative advantages in order for them to participate in FDI.
Figure 1: The OIL-framework by Dunning (1980)
Picture obtained from: https://harmkuiper.wordpress.com/2011/03/08/eclectic-approach-john-dunning/
Dunning (1993) uses his OIL-framework to distinguish four different types of FDI. The ‘market seeking’ type refers to FDI aiming to sustain presence in foreign markets or to exploit new markets. The ‘resource seeking’ type of FDI is about the acquisition of particular
(natural) resources which are not available in the home country or those which are available but at a higher price. This type of FDI can be characterized as vertical FDI, whereas the market seeking strategy is mostly seen as horizontal FDI as it involves replicating production facilities in the host country (Franco, Rentocchini, and Vittucci Marzetti, 2008). The third is
9 the ‘strategic resource seeking’ type, which describes firms wanting to enhance their
international competitiveness by acquiring strategic assets (such as technological knowledge) of foreign firms. Because of the similarities with resource seeking FDI, some authors choose
to combine the two into one category3 (Campos and Kinoshita, 2003). The fourth type of FDI,
called ‘efficiency seeking FDI’, occurs when MNEs benefit from geographically dispersed production in order to reap the benefits of economies of scale or economies of scope. The idea is that companies will optimize and restructure their investments and the allocation of their international activities based on developments that are of influence on their businesses, such as changes in labour productivity. Therefore, efficiency seeking FDI is presumed to be most responsive to fluctuations in human capital levels. Zhou and Lall (2005) note that the
efficiency seeking type is influential in East Asia because the region is very heterogeneous with respect to skill levels and other factors that determine the production cost. Therefore, the expectation is that FDI inflows will be influenced (to some extent) by changes in human capital levels.
With respect to specific determinants of FDI in recent decades, certain trends are observable. The relative importance of some of the determinants seems to shift over time (Nunnenkamp, 2002). Thus, key pull-factors in the 1980s, for example, are not necessarily relevant pull factors nowadays. To control for this a time dummy is introduced, as will be explained in chapter 3. Faeth (2009) provides a literature overview with respect to nine theoretical models concerning FDI determinants nowadays. Faeth (2009) indicates that (a measure for) import tariffs is a commonly used determinant because tariffs make FDI more attractive relative to exports. High tariffs induce countries to establish production facilities in the host country in order to avoid the import tariff, this is often referred to as tariff-jumping and is embedded within the market seeking type of FDI. Other important FDI determinants are labour costs, market size, market growth, political stability, macro-economic stability and the lagged value of FDI. (UNTAD, 1998; Noorbakhsh et al., 2001; Quazi, 2007; Faeth, 2009). Market size enables companies to benefit from economies of scale, which points out that there is some overlap between market seeking- and efficiency seeking FDI. The lagged value of FDI is included for several reasons, one of them being the time-to-build argument. Establishing a factory takes place over a number of years and therefore FDI in the previous period is a determinant of FDI in this period.
3 In this thesis the two categories are also combined into one group, which is referred to as ‘resource seeking FDI’.
10 An important trend that started during the 1980s is that FDI started to shift towards the (high-end) service sector. This development is still observable today, according to recent World Investment Reports and OECD papers (Arnal and Hijzen, 2008; UNCTAD, 2016). This shift also means that FDI is concentrating in the knowledge- and skill-intensive industries
(Pfefferman and Mandarassy, 1992; Michie, 2001; Arnal and Hijzen, 2008). This shift in FDI towards industries that require skilled labour is illustrated in figure 2. The figure shows that (inward) FDI has shifted towards skilled- and medium skilled labour and away from low skilled labour in both developing countries and developed countries (between 1990 and 2005). This indicates that human capital may have become more important as a determinant of FDI relative to (cheap) labour costs.
Figure 2: FDI in host countries per skill level
Source: OECD paper by Arnal and Hijzen (2008)
Globalization has reduced boundaries/costs for companies to base themselves abroad, which is likely to enhance this effect. Companies nowadays can move certain branches abroad, towards locations that are most suitable for that particular activity (efficiency seeking FDI). Since most East Asian countries are rapidly enhancing their human capital levels (see appendix 2 and 3), this may have boosted FDI inflows to the region. I expect that human capital will be especially a significant determinant of FDI from developed source countries, that are characterized by the production of skill-intensive goods.
11 2.2 Human capital as determinant for FDI
The hypothesis of human capital as a determinant of FDI is embedded in the economic literature. Michie (2001) states that the major effect of FDI on human capital enhancement appears to be caused by governments trying to attract FDI by investing in schooling rather than by multinational enterprises (MNEs) investing in schooling for their employees. Michie (2001) comes to these conclusions by studying economic literature for different countries. Feenstra and Hanson (1996) find that FDI mostly consists of developed countries shifting their production towards developing countries and the production of developed countries (such as OECD countries) is found to be more skilled-labour intensive. As such, Feenstra and Hanson (1996) expect that countries that have a large supply of highly educated workers will attract more FDI because the relative wage of skilled labour is relatively low in those countries. Their paper relies on an empirical analysis of data for industries in multiple countries during the period 1972-1990. Blomstrom and Kokko (2003) state, on the basis of a literature overview, that MNEs have the potential to provide attractive employment opportunities for highly educated students in sciences and engineering, provided that governments invest sufficiently in higher education. Hoffmann (2003) provides a game-theoretical model which shows that, in a situation with FDI inflows, an education subsidy by the government can change a population which is originally abundant in low-skilled labour into an equilibrium where highly-skilled labour is abundant. Moreover, Zhang and Markusen (1999) by means of a theoretical econometric model, find that the availability of skilled labour is a requirement for MNEs to extend their activities abroad, which can have an impact on FDI inflows.
However, most of these studies provide only theoretical evidence of the causal relation between human capital and FDI and do not provide (extensive) empirical evidence. Empirical papers and books on this topic do exist but many date back to the 1990s or earlier and their findings are more often than not inconclusive. Examples of this are the book by Natarajan and Tan (1992) on Malaysia, Singapore and Thailand and the paper by Sibunruang and Brimble (1988) which focuses solely on Thailand. Natarajan and Tan (1992) have based their findings on questionnaire surveys to MNEs during the late 1980s, asking companies about their motivations for investing in FDI. They have extensively studied over 30 companies and governments to explore what determines FDI and human capital was often reported to be of influence. Sibunruang and Brimble (1988) have used a similar methodology. They have
12 studied 678 foreign firms that do business or have offices in Thailand, in the 1980s, using survey data combined with macro-economic data on the Thai economy.
A more recent paper that examines the effect of human capital on FDI is the one by Kinoshita and Campos (2003). They have studied 25 transition economies in Eastern Europe and former Soviet-states between 1990 and 1998. An interesting feature about this research is their choice
of variables. Besides the usual control variables4, the authors also include variables to
measure the rule of law, quality of bureaucracy, and an index of FDI restrictions. Unfortunately, data on these variables were not suitable for this thesis because of missing
data5. Kinoshita and Campos (2003) find that human capital is not a significant determinant of
FDI. This result is probably the result of their selection of countries, namely transition economies. It seems likely that there is too little cross-country variance in their study, as most of these economies show rather high levels of human capital and little variation compared to East Asian countries for instance. Also, the time period they use is very limited and as a result their number of total observations (119) is also small, meaning their results may be influenced by small sample bias. Kinoshita and Campos (2003) are aware of this problem and therefore use both a fixed effects regression and a generalized-method-of-moments regression (GMM). The GMM method is sensitive to small sample bias, therefore the GMM will not be used in this thesis (Bun and Windmeijer, 2010). Another difference in methodology is that Kinoshita and Campos (2003) have used annual data rather than averaged data which might result in measurement errors, but this will be explained more thoroughly in chapter 3. Empirical papers that date back to the 1980s and earlier also exist but these shall not be discussed in this thesis because, as indicated by Nunnenkamp (2002), the determinants of FDI change over time. A fairly recent and influential empirical study that investigates the effect of human capital on FDI inflows is the paper by Noorbakhsh et al. (2001). As mentioned in the introduction, Noorbakhsh et al. (2001) have studied 36 developing countries located in different continents between 1980 and 1994. They have used different estimation methods but eventually they base their findings on OLS regressions, corrected for heteroskedasticity. One of the estimation methods they tried out, was the random effects regression. However, this model was strongly rejected by the Hausman test in favour of the fixed effects model. They then ran the fixed effects regressions but it showed poor results. They argue that this is the case because the
4 The usual control variables refers to: market size, lagged FDI, natural resource rents, quality of infrastructure, inflation rate, and openness to trade.
5 The data were either unavailable for some of the countries in the sample or was unavailable for a large number of years.
13 fixed effects approach removes too much cross-country variation. This makes sense, given
their time-interval is somewhat limited, leaving only 4 time periods6. A fixed effects
regression is costly in terms of degrees of freedom when the number of countries in the dataset is much larger than the number of periods. However, the use of OLS rather than fixed effects is debatable because it has several shortcomings as is discussed in chapter 4.
Many of the variables Noorbakhsh et al. (2001) have used are also used in this thesis although there are some differences. Similar to what is done in this thesis, they use both flow- and stock indicators to measure human capital levels and they also differentiate between secondary- and tertiary schooling. They, like Kinoshita and Campos (2003), include variables to control for openness to trade, natural resource rents, and FDI in the previous period. However, unlike Kinoshita and Campos (2003) they control for market growth rather than
market size7. Moreover, they use credit to the private sector rather than inflation as proxy for
financial stability and they also include a measure that indicates whether wages are high in the host country compared to the rest of the world (efficiency wages). However, these two variables turn out to be highly insignificant which suggests that they may not be suitable for this type of research.
The outcome of the research by Noorbakhsh et al. (2001) is that secondary education and total years of secondary and tertiary schooling combined are significant determinants of FDI in developing countries. Furthermore, the coefficients of these human capital indicators seem to
have increased over time8. However, the drawbacks of this research (such as the short sample
period and debatable methodology) pointed out that more recent empirical studies are needed to test the robustness of the results obtained by Noorbakhsh et al. (2001). More specifically, research is needed to find out if the results still hold when different control variables, a different set of countries, a less restrictive empirical model, and a larger sample period are used.
An article that addresses some of these issues is the article by Cleeve, Debrah and Yiheyis (2015). They have assessed the role of human capital on influencing FDI inflows towards sub-Saharan Africa, using panel data on 35 countries ranging from 1980-2012. In terms of control
6 Noorbakhsh et al. (2001) construct time periods comprised of averaged data for every 3 years.
7 Market growth (measured as GDP growth) was found to be significant in every modification of the model, pointing out that market growth has to be controlled for in order to avoid omitted variable bias.
8 Noorbakhsh et al. (2001) tested this by splitting the sample into three time periods and testing each one individually.
14
variables, they use the ones customary in the literature9 (Noorbakhsh et al., 2001; Kinoshita
and Campos, 2003). However, Cleeve et al. (2015) use both market size and market growth as control variables rather than just one of the two. Furthermore, they use inflation rates instead of credit provided to the private sector as proxy for financial and macro-economic stability. Cleeve et al. (2015) have also included a measurement for the extent of political participation in their regressions. These control variables seem to be logical and embedded in the economic literature, as will be explained in chapter 3.
Cleeve at al. (2015) find, using different statistical methods, that both secondary- and tertiary
schooling positively influence FDI inflows in Sub-Saharan Africa10. The aim of this thesis is
to examine whether the same results hold for East Asian economies, using an even longer sample period (1970-2015) under different alterations of the model.
9 By which is meant: market size, lagged FDI, natural resource rents, quality of infrastructure, inflation rate, and openness to trade.
15 3. Data and research methods
3.1 The model
To investigate the effect of human capital on FDI an econometric model is used, which has the following form:
𝐹𝐷𝐼𝑖𝑡 = 𝛼 + 𝛽𝐻𝐶𝑖𝑡+ 𝛽′𝐶𝑉𝑖𝑡+ 𝜆𝑖+ 𝜀𝑖𝑡
The model consists of: the dependent variable which is FDI inflows; HC is the measure for human capital; CV is a vector of all control variables; λ is the country-fixed effect and ε is the
error term (Noorbakhsh et al., 2001). Subscript 𝑖 indicates the recipient country in question
and subscript 𝑡 indicates the year. The time period that is studied in this thesis is the period
ranging from 1970 until 2015. This thesis will rely on panel data regressions. Three conventional estimation methods have been used, namely: OLS; random effects; and fixed effects regressions (Noorbakhsh et al., 2001; Cleeve et al., 2015). The variables used in this research, largely coincide with those used by Cleeve et al. (2015). The reasoning behind the inclusion of the selected variables along with a precise definition for each variable, is provided in the second part of this chapter.
The data on FDI inflows is obtained from the database of the United Nations Conference on Trade and Development (UNCTAD) and from the World Bank. Data about human capital levels are obtained from the database constructed by Barro and Lee (2013) and by data from the World Bank. The database of Barro and Lee (2013) provides human capital data at five-year intervals. To alleviate this problem, the (mostly annual) data are converted to 5-five-year averaged time periods and as result 9 periods are constructed. This approach also mitigates
business-cycle effects, measurement errors, and it deals with the problem of missing data11.
This technique is widely used in the economic literature (Noorbakhsh et al., 2001; Kinoshita and Campos, 2003; Cleeve et al., 2015).
A separate regression shall be conducted to find out whether FDI originating from developed countries reacts more strongly to changes in human capital than FDI coming from the rest of
the world12. Developed countries produce more skill intensive goods and therefore are
expected to react more strongly to changes in the skill levels of workers in recipient countries.
11 To check whether the estimations are not heavily influenced by the choice for 5-year averaged rather than yearly data, the model was re-estimated using annual data and the results are reported in chapter 4.
12 The UNCTAD identifies developed countries on the basis of a certain thresholds with regard to per capita GNI, the human assets index and the economic vulnerability index. Source: http://unctadstat.unctad.org/EN/Classifications.html
16 However, note that this second regression is less accurate than the first one due to fewer observations, as UNCTAD data on this subject are only available for the period 2001-2012. The twenty-three countries that have been studied in this research are all considered East
Asian or South-East Asian13. I have chosen this region because many of these countries have
experienced high economic growth in recent decades and it is interesting to know whether human capital growth has been (partly) driving this process by increasing the amount of FDI inflows. Also, this region provides sufficient variation with respect to human capital growth. Most countries have experienced a rapid growth of human capital, but not all. Countries like Myanmar, Laos and Cambodia for instance have not invested in human capital as heavily as most countries have (appendix 5). The expectation is that these countries have also attracted less FDI inflows than other countries in the region, controlling for other variables.
3.2 Selection of variables
The dependent variable in this study is net FDI inflows in the host country, as a percentage of GDP. A relative measure is used to control for any large-country effects. This measure for FDI is widely used in the economic literature as it avoids that a large country, such as China, heavily influences the results. FDI flows are used rather than stocks because flows are less sensitive to the “book-value bias”. Data on capital stocks are expressed in book-values without adjustments for inflation and changes in exchange rates, making them less reliable and transparent than FDI flows (Root and Ahmed,1979; Noorbakhsh et al., 2001).
The human capital variables that have been selected for this study are the following: ENROLSEC, ENROLTER, YEARSTOT, YEARSSEC, YEARSTER, and NOSCHOOL. Here the variable ENROLSEC refers to the gross secondary school enrolment ratio in percentages and is commonly used in empirical literature (Root and Ahmed, 1979; Noorbakhsh et al., 2001; Cleeve et al., 2015). The variable ENROLTER indicates the enrolment ratio for tertiary education. It captures the higher skill levels such as management skills and technical expertise. The expectation is that both variables are positively correlated with FDI inflows.
13 The list of selected countries is the following: Bangladesh, Bhutan, Brunei, Cambodia, China, Hong Kong, India, Indonesia, Japan, South-Korea, Lao, Macao, Malaysia, Maldives, Mongolia, Myanmar, Nepal, Pakistan, Philippines, Singapore, Sri Lanka, Thailand, and Vietnam.
17 The variable YEARSTOT refers to the average years of total schooling attained by the population aged 15 and over. This variable is used in the empirical literature as a proxy for the accumulated stock of human capital in the economy (Nunnenkamp, 2002; Cleeve et al., 2015). YEARSSEC (average years of secondary education) and YEARSTER (average years of tertiary education) also serve as proxy’s for the education stock. All of these human capital variables are expected to be positively related to FDI inflows. The final human capital indicator that is included is NOSCHOOL, which measures the percentage of the population (age 15 and over) that have had no schooling at all. Hence, a negative coefficient is expected. Most of the control variables that have been selected in this study are common in the literature on FDI determinants, the first one being GDP per capita. GDP serves as a proxy for purchasing power and market size. More purchasing power and a larger domestic market are expected to be positively related to FDI inflows and this type can be characterized as market seeking FDI. However, GDP per capita also serves as a proxy for real wages in the host country. Higher real wages, given fixed labour productivity, are expected to discourage (efficiency seeking) FDI inflows. This means that the sign of this variable can be both positive or negative, depending on which effect dominates. Following the approach by Cleeve et al. (2015), the variable is measured by the logarithm of per capita GDP in constant US dollars.
Market growth is also generally included as a control variable (Faeth, 2009). The growth rate of the domestic market in host countries is typically perceived to be a major determinant of FDI inflows (Root and Ahmed, 1979; Noorbakhsh et al., 2001; Faeth, 2009). Economic growth increases the demand for goods and services and therefore attracts market seeking FDI. Therefore, a positive sign is expected. This variable is measured as annual GDP growth in percentages (GDPGROWTH).
The lagged value of FDI (LAGFDI) is included as well, for two reasons. First, the time-to-build argument states that establishing a (production) facility abroad takes place over a number of years and therefore FDI in the previous period is a determinant of FDI in this period. A second reasoning for the inclusion of lagged FDI has to do with signalling. The argument is that a (large) FDI flow to a certain country means that investors have confidence in this particular country and signal this to other investors, thereby influencing FDI flows in the next period. Consequently, the expectation is that this variable has a positive sign.
18 The fourth control variable, that is included, is trade openness (OPEN). Trade openness is measured as the ratio of total trade (exports plus imports) to GDP (Noorbakhsh et al., 2001). If a host country is heavily involved in trade, this creates investment opportunities for MNEs. This variable also serves as a proxy for free trade. A country with a high degree of openness is assumed to bolster free trade by pursuing free trade agreements and similar strategies. This arguably creates a favourable investment climate. Subsequently, a positive sign is expected. The control variable POLICY is also included in this study. It measures the extent of political participation and it can be seen as an indicator of democratic values. The index ranges from -10 (very autocratic) to +10 (very democratic). Data on this variable are obtained from the ‘POLITY IV PROJECT’ database and this variable is also used by Cleeve et al. (2015) as a determinant of FDI inflows. Faeth (2009) states that democratic regimes attract more FDI than autocratic regimes, implying that a positive sign is expected.
Faeth (2009) also provides evidence that countries that have good infrastructure are more successful in attracting FDI than countries with poor infrastructure. Good infrastructure decreases transportation costs, which stimulates investments (Biswas, 2002). For this reason, the expectation is that good infrastructure has a positive effect on FDI inflows (positive sign). The proxy INFRA is used to measure the quality of infrastructure. It measures the number of mainline telephones per hundred people. The choice for this proxy has to do with the availability of data. Other indicators, such as measurements for the quality of roads, are available but data on East Asian countries are often missing or only available for a limited time-interval.
A control variable that is often used in the economic literature, is the availability of natural resources in the host country (Noorbakhsh et al., 2001). The variable RESOURCES measures the natural resource rents as a percentage of GDP and is used to find out whether natural resources attract FDI inflows. However, there is evidence that the importance of this variable has diminished over time as a result of globalisation. Even so, it is used in most empirical papers on FDI and in some papers it appears to be a significant determinant of FDI (Kinoshita and Campos, 2003; Cleeve et al., 2015). Especially least developed countries are relatively dependent upon resource seeking FDI (Cleeve et al., 2015). The expectation is that this type of FDI will not be of (big) influence on FDI inflows in East Asia, given that East Asian countries are rather developed. But this variable has to be controlled for nonetheless. As a result, a (mildly) positive sign is to be expected.
19 Historically, a measurement for macroeconomic stability is also believed to be of importance as a determinant of FDI inflows. Following the methodology of Cleeve et al. (2015), INFLATION is used as a proxy for macroeconomic stability and it measures the inflation rate (based on the consumer price index). Successful inflation stabilizations are associated with fiscal adjustments and the ability of governments to manage the economy well (Noorbakhsh et al., 2001; Cleeve et al., 2015). Also, price fluctuations distort prices and are believed to harm FDI flows (Faeth, 2009). Therefore, the expectation is that an increase in the inflation rate has a negative effect on FDI inflows. Consequently, a negative sign is expected.
Because there is reason to believe (as explained earlier) that human capital has become more influential over time, the dummy TIME is added (Noorbakhsh et al., 2001; Nunnenkamp, 2002; Arnal and Hijzen, 2008). This dummy splits the study period in two. The period before 1990 is coded with a 0 and the period after (and including) 1990 is coded with a 114. This dummy is included independently as well as interactively with the human capital indicator to find out whether the effect of human capital has changed in recent decades. This dummy, used independently, captures the increase in global FDI flows in recent decades. The expectation is that the variables used to measure human capital, will have a bigger effect in the time period since 1990.
Periods of economic crises are typically characterized by low FDI flows worldwide. To control for this, the dummy CRISIS is added. This dummy captures major economic crises, being: the Asian financial crisis in 1997; and the “great recession” in 2007. The dummy is constructed in the following way: CRISIS=1 if period is 6 and 8; CRISIS=0 otherwise (Cleeve et al., 2015). Period 6 ranges from 1995-2000 and period 8 ranges from 2005-2010. This means that two years prior to the initial crisis and the two years after the crisis are also taken into account. Ideally, the years leading up to the crisis would be excluded from the CRISIS dummy. Therefore, the dummy is re-estimated using annual data and the years prior to the crisis are excluded from the dummy in order to find out whether or not the results concerning the CRISIS dummy are biased by the choice for 5-year averaged data. The expectation is that during periods of economic crisis, FDI inflows will be significantly lower. However, at first glance the data do not support this claim. FDI inflows in crisis years does not seem to be lower than in other years. The variable was added nonetheless and the results are described in chapter 4.
14 Other time intervals (post 1985 and post 1995) have also been tested but the results were very similar to the ones that are reported in chapter 4, thus only the ones that measure ante- and post-1990 have been reported.
20 4. Results and discussion
The time period that is studied in this thesis is the period ranging from 1970 until 2015, using nine non-overlapping 5-year averaged time periods. Appendix 5 provides a summary of the data with respect to human capital indicators and FDI flows, per country. It shows that there is sufficient variation with respect to human capital and FDI flows, amongst the countries in the dataset. The reported minimum- and maximum values are often far apart from each other for both human capital and FDI flows. The data also indicate that somewhat of a linear trend is observable for both human capital levels and FDI flows through time, similar to the graphs in
appendices 1,2 and 315. The human capital data also suggest that the countries in the dataset
are in different stages of development. Countries like Hong Kong and Singapore are highly developed (with 8,9 and 7,2 average years of total schooling respectively), while a country like Nepal has only 2,2 years of total schooling on average. The different control variables that have been used also show much variance over time as well as cross-country variance. Multiple estimation methods were used to find out which ones are the most appropriate and to test the robustness, the results are displayed in table 1. The three methods that have been used are: Pooled Ordinary Least Squares (OLS); random effects (RANDOM); and the fixed-effects model (FIXED). Each column in table 1 shows the results of a different method and the name of the method that has been used is depicted above each column. In each regression the
tertiary enrolment ratio (ENROLTER) was used as human capital indicator16. The OLS
regression assumes the intercept coefficient to be the same for all countries, thereby ignoring heterogeneity across countries (Stock and Watson, 2007). The heterogeneity across countries is of interest because it provides additional information. Therefore, at first glance, the OLS estimation seems not very appropriate but it is included nonetheless because Noorbakhsh et al. (2001) base their findings on this method and Cleeve et al. (2015) have also used it (along with other methods).
In the random effects model, the variation across entities is assumed to be random and uncorrelated with the explanatory variables included in the model. Like the OLS method, it is used by Noorbakhsh et al. (2001) and by Cleeve et al. (2015) to test the robustness of their
15 Therefore, the minimum stock values that are stated in appendix 5 can in most cases be linked to the early years in the dataset (around 1970), whereas the maximum value can be linked to more recent years (around 2015)
21 results. A Hausman test was conducted to test whether the errors are correlated with the regressors, the null hypothesis being that they are not (Stock and Watson, 2007). The results are reported in appendix 6 and provide evidence against the random effects test in favour of the fixed effects method. Therefore, both OLS and Random effects appear to be sub-optimal for this analysis because individual-specific effects are important in terms of controlling for omitted variable bias (Stock and Watson, 2007).
The fixed-effects model is used when variables that change over time are of interest. It explores the relationship between predictor- and outcome variables within an entity, in this case within individual countries. Therefore, the fixed-effects model appears to be most appropriate for this thesis. Table 1 shows that the tertiary enrolment ratio (ENROLTER) is highly significant in each of the three tests and the sign is positive, although the size of the coefficient differs in all three models. This gives support to the robustness of human capital as determinant of FDI inflows. The OLS estimates show a coefficient of 0,053 while the fixed effect estimates show a coefficient that is nearly twice as high (0,095). This difference could be due to omitted variable bias in the OLS estimates.
A test for heteroskedasticity was then used and the results are shown in appendix 7. The test pointed out that heteroskedasticity was indeed a problem and therefore heteroskedasticity-robust standard errors had to be used to avoid biased results (Stock and Watson, 2007).
Table 1: Results using different estimation methods
Dependent variable: FDI
Explanatory (1) (2) (3)
Variable OLS RANDOM FIXED
ENROLTER 0.0532*** 0.0726**** 0.0953**** (0.0173) (0.0193) (0.0266) GDPGROWTH 0.404**** 0.443**** 0.438**** (0.0931) (0.0979) (0.116) LOGGDP -0.929** -0.822* -0.369 (0.374) (0.444) (0.769) OPEN 0.0356**** 0.0333**** 0.0226** (0.00318) (0.00443) (0.0107) INFRA 0.0342 0.0144 -0.0248 (0.0351) (0.0397) (0.0491) INFLATION 0.0111 0.0104 0.0144 (0.0133) (0.0130) (0.0142) RESOURCES 0.125**** 0.158**** 0.206*** (0.0348) (0.0423) (0.0616) POLICY 0.0411 0.0248 0.0149 (0.0381) (0.0410) (0.0496) N 138 138 138 R2 0.656 0.421 adj. R2 0.635 0.286 F 30.80 10.10
Standard errors in parentheses
22 The fixed effects model was re-estimated using robust standard errors and taking into account the lagged value of FDI (LAGFDI), the results are reported in table 2. Each column in table 2 represents a different human capital indicator. The name corresponding to the human capital indicator that has been used, is displayed above each column. The lagged value of FDI was included, because autocorrelation tests have shown that the dependent variable is heavily
correlated with its lagged value17. Moreover, the explanatory power of the model is
significantly lower when LAGFDI is excluded, as indicated by significantly lower R-squared values reported in appendix 8. Therefore, omitting it would yield biased results (Stock and Watson, 2007). However, the inclusion of this variable reduces the explanatory power of other control variables. The variables indicating the growth rate of the gross domestic product (GDPGROWTH) and openness to trade (OPEN) were significant in all three models reported in table 1, but are no longer significant (at the 10%-level) in any of the regressions in table 218.
With respect to the different human capital indicators (HC) it can be noted that both average years of tertiary education (YEARSTER) and the enrolment ratio for tertiary education (ENROLTER) are significant, at the 1% and at the 10% level respectively. Although both have a positive sign (which is expected), the coefficients are quite different. This is due to the unit of measurement. The size of the coefficient of YEARSTER (4,643) is much higher than that of ENROLTER (0.0867) because the former is a stock variable and the latter is a flow variable. A stock variable takes into account accumulation over time (multiple years), whereas a flow measure only provides information for a certain year. Therefore, it makes sense that the coefficient of the stock measure is larger than that of the flow measure.
The other human capital indicators are not significant in table 2. This indicates that tertiary schooling levels appear to be a significant determinant of FDI in East Asian countries, both measured as a stock (YEARSTER) and as a flow measure (ENROLTER), whereas the other human capital indicators are not. The other two variables that are significant in almost all regressions in table 2, are the lagged value of FDI (LAGFDI) and natural resource rents (RESOURCES). Both variables show the expected (positive) signs and are highly significant in almost all regressions.
17 To test for autocorrelation a Woolridge test was conducted in STATA
18 The variable LOGGDP was only significant in the OLS model reported in table 1 and like ‘GDPGROWTH’ and ‘OPEN’, is now insignificant in table 2
23
Table 2: Results using different human capital indicators
Dependent variable: FDI
Explanatory (1) (2) (3) (4) (5) (6)
Variable HC=YEARSTER HC=ENROLTER HC=YEARSSEC HC=ENROLSEC HC=YEARSTOT HC=NOSCHOOL
HC 4.643*** 0.0867* 0.657 -0.0224 0.450 0.0318 (1.577) (0.0421) (0.541) (0.0323) (0.391) (0.0316) LAGFDI 0.564*** 0.546*** 0.661*** 0.515 0.648** 0.699*** (0.195) (0.186) (0.217) (0.302) (0.225) (0.226) GDPGROWTH 0.303 0.356 0.237 0.268 0.242 0.262 (0.192) (0.210) (0.188) (0.182) (0.189) (0.198) LOGGDP 0.593 0.553 0.390 1.670 0.0825 1.762 (0.837) (0.829) (1.152) (1.495) (1.447) (1.204) OPEN 0.00757 -0.000162 0.00688 0.0163 0.00601 0.00683 (0.00927) (0.0101) (0.00811) (0.01000) (0.00781) (0.00892) INFRA -0.0407 -0.0713 -0.00184 0.00129 0.00832 -0.00848 (0.0534) (0.0564) (0.0414) (0.0533) (0.0425) (0.0430) INFLATION 0.00101 0.00859 -0.00142 -0.00170 -0.00308 0.00144 (0.0122) (0.0157) (0.0129) (0.0139) (0.0129) (0.0132) RESOURCES 0.160*** 0.137*** 0.195*** 0.206*** 0.199*** 0.166*** (0.0548) (0.0436) (0.0530) (0.0671) (0.0540) (0.0515) POLICY 0.0146 0.00343 0.0160 0.0325 -0.00501 0.0390 (0.0350) (0.0316) (0.0344) (0.0467) (0.0428) (0.0473) N 121 119 121 116 121 121 R2 0.577 0.561 0.542 0.509 0.543 0.538 adj. R2 0.542 0.525 0.504 0.467 0.505 0.501 F 10.03 6.964 6.116 28.30 5.918 7.975
Standard errors in parentheses
* p < 0.10, ** p < 0.05, *** p < 0.01, **** p < 0.001
To check whether the estimations are not heavily influenced by the choice for 5-year averaged instead of yearly data, the model was re-estimated using annual data. This test is reported in appendix 9. Note that the outcomes of this test are less reliable than the ones in table two because of business cycle fluctuations amongst other things, as indicated in chapter 3. Therefore, this test is only used to see if there are no structural differences. The results seem to be largely in line with those reported in table 2, although in appendix 9 the only human capital indicator that is significant is average years of tertiary schooling (YEARSTER) and the enrolment ratio for tertiary schooling (ENROLTER) is no longer significant. However, this could very well be the result of business cycle fluctuations. Like in table 2, YEARSTER is highly significant, has a positive sign and a large coefficient. In sum, appendix 9 provides no evidence that opposes the previous findings.
Having established that YEARSTER is positively signed and significant, irrespective of the model used, this research continues by testing this human capital indicator under different
24 specifications of the model. The measurement for the quality of infrastructure (INFRA) and the control variable for democratic values in a country (POLICY) are both dropped out because they have been highly insignificant in every modification of the model so far. Also, the signs of these two variables have not been robust, meaning that they do not act as determinant of FDI in this research and can be excluded from the regressions. The insignificance of INFRA might have to do with the rather high level of development in most countries in our sample. A proxy that measures the number of mainline telephones per hundred people may not effectively capture the state of the infrastructure in East Asian economies, while in less developed countries this measurement is probably more adequate. The insignificance of the control variable POLICY can possibly be explained by missing data for several countries and periods.
In table 3, the variable LAGFDI is replaced by the lagged change in FDI (∆FDI-1) because this measurement embodies the same information as LAGFDI but has less of a downward bias towards the coefficients of other control variables in the model (Achen, 2000). The results are presented in column 1. Then, a the dummy for the period before and since 1990 (TIME) is added both independently as well as interactively with HC (YEARSTER) to find out whether the effect of human capital has changed in recent decades (since 1990) and to take into account the increase in FDI flows worldwide since (approximately) 1990, for reasons explained earlier. The estimations, inclusive of the TIME dummy, are shown in column 2. In column 3, a dummy indicating years of economic crisis (CRISIS) is added to see if periods of economic instability (period 6 and 8) are of influence on the results obtained earlier.
The changes in the model do not have a large impact on the explanatory power of the estimations as indicated by the R-squared and the adjusted R-squared. The significance of HC decreases from 1% to the 5% level, but like ∆FDI-1 it remains significant. Remarkably, the variables for openness to trade (OPEN) and the growth rate of GDP (GDPGROWTH) were insignificant in table 2 but they emerge as significant determinants in table 3. The opposite is true for the variable indicating resource rents (RESOURCES), which is now highly insignificant. This suggests that these variables were somehow connected to information embedded in either the control variable for democratic values in a country (POLICY) or the
25 control variable for quality of infrastructure (INFRA), although from a theoretical point of
view they would seem unconnected19.
Table 3: Regressions under different alterations of the model Dependent variable: FDI
(1) (2) (3) (4) YEARSTER 6.430** 8.930** 6.414** 6.788** (2.515) (3.362) (2.477) (2.635) ∆FDI (-1) 0.479**** 0.478**** 0.486**** 0.505**** (0.0757) (0.0763) (0.0779) (0.0901) GDPGROWTH 0.403** 0.397** 0.438** 0.387* (0.181) (0.179) (0.174) (0.193) LOGGDP -0.0226 -0.277 -0.200 (0.996) (1.020) (0.945) OPEN 0.0543** 0.0544** 0.0540** 0.0441** (0.0227) (0.0229) (0.0228) (0.0210) INFLATION 0.0531 0.0495 0.0435 (0.0325) (0.0320) (0.0341) RESOURCES 0.139 0.148 0.146 (0.156) (0.152) (0.148) TIME 0.567 (0.732) TIME x YEARSTER -2.271 (2.111) CRISIS 0.666 (0.526) N 130 130 130 135 R2 0.560 0.561 0.566 0.534 adj. R2 0.535 0.529 0.538 0.520 F 20.46 15.74 18.52 14.08
Standard errors in parentheses
* p < 0.10, ** p < 0.05, *** p < 0.01, **** p < 0.001
Table 3 (column 2) shows that the inclusion of a TIME dummy does not have a large effect on the coefficients of the model. The (independent) TIME dummy and the interaction term with HC are not significant. To check the robustness of this result regarding the TIME dummy, the regressions in table 3 were computed once again using annual as well as averaged data. Different time periods have been studied to check whether this would have an impact on
the TIME dummy, but the TIME dummy remained insignificant in every estimation20. The
insignificance of the TIME dummy contrasts the findings of Noorbakhsh et al. (2001), who found that human capital has become more influential over time. This difference in findings is likely due to the choice of countries in the sample. Noorbakhsh et al. (2001) have studied
19 OPEN and GDPGROWTH were not influenced by the inclusion of ∆FDI (-1) , because table 3 was also run with LAGFDI and the results were similar.
20 The sample periods post 1985 and post 1995 were also tried out, using both annual and averaged data, but the results were similar to the ones reported in table 3.
26 countries in the early stages of development, while East Asian economies are more developed, resulting in different types of FDI flows.
Table 3 (column 3) also shows that the dummy indicating crisis years (CRISIS) is insignificant (P-value = 0,22). However, ideally the years leading up to a crisis (1995, 1996, 2005, and 2006) would be excluded from the regression. For this reason, a robustness test is conducted, using the same method as the one reported in table 3 (column 3) but with different crisis years and on the basis of annual data. The dummy was constructed in the following way: CRISIS=1 if year is 1997, 1998, 1999, 2007, 2008, and 2009; CRISIS=0 otherwise. The years that have been included in the dummy refer to the Asian financial crisis (1997) and to the Financial Crisis of 2007. The two years after the initial crisis are included as well because the economy was still in recession during those years. This test indicates that the CRISIS
dummy is insignificant21. For the second robustness check, the years corresponding to Black
Monday (1987, 1988, and 1989) are also marked as crisis years. Again, the results do not
change much22. Therefore, the annual data in which the years prior to the CRISIS are
excluded provide no evidence that the coefficients reported in table 3 (column 3) are biased. The dummy indicating crisis years was not significant in any of the regressions discussed earlier. This is an interesting finding because the expectation is that FDI flows would decline during crises, as Cleeve et al. (2015) found to be true for countries in Sub-Saharan Africa. Moreover, the CRISIS coefficient (although insignificant) even emerged as positive in all regressions. This suggests that FDI flows towards East Asian countries might even be larger during periods of economic recession. A possible explanations could be that markets are overconfident in the months leading up to a crisis (the so called ‘boom’ in the boom and bust cycle), resulting in high FDI flows. Another explanation could be that Asian markets are somehow seen as a ‘safe heaven’ during times of economic downfall, but this topic surpasses the objectives of this thesis.
In the final step of the model (column 4 in table 3), the insignificant variables were all dropped and the explanatory power of the model only slightly decreased, as indicated by a lower R-squared. To check whether the results regarding the significance of the different HC indicators had changed by the alterations in the specifications of the model, model 4 in table 3 was re-estimated using different HC indicators and the results are reported in table 4. The findings are quite similar to those reported in table 2. More specifically, the only HC variables
21 The reported coefficient is 0,32 and the P-value is 0,41 22 The reported coefficient is 0,20 and the P-value is 0,47
27 that are found to be significant are the indicators for tertiary schooling (YEARSTER and ENROLTER), both significant at the 5% level and with positive coefficients. The other HC indicators do not have a significant effect on FDI inflows.
Table 4: Coefficients of different HC indicators
YEARSTER ENROLTER YEARSSEC ENROLSEC YEARSTOT NOSCHOOL 6.788** 0.0948** 1.170 0.0294 0.634 -0.0353
(2.635) (0.0414) (0.690) (0.0260) (0.372) (0.0347) Standard errors in parentheses
* p < 0.10, ** p < 0.05, *** p < 0.01, **** p < 0.001
To check whether the significance of the previous findings with respect to human capital are not biased by endogeneity, model 4 in table 3 was estimated again using the lagged values of ‘average years of tertiary education’ (YEARSTER-1) and ‘enrolment ratio of tertiary
education’ (ENROLTER-1). The results did not differ much from the original estimations23.
Interestingly enough, the coefficient of YEARSTER-1 is larger than that of YEARSTER. This could reflect a stronger effect of HC on FDI in the long term. However, for ENROLTER we do not observe such a trend so there is no strong evidence to support this statement. The other HC indicators were also tested in this manner, but the lagged values were all found to be insignificant. This gives support to the claim that tertiary education positively affects FDI inflows in East Asia, whereas secondary- and lower education do not.
Finally, a regression was set up to test whether FDI inflows from developed countries
(FDIDEV), as defined by the UNCTAD11, differs from FDI inflows in general. Unfortunately,
data on this matter is only available for certain countries for a very limited number of years. Therefore, the corresponding regression is unreliable and merely serves as a rough approximation. The estimations based on a unbalanced dataset, comprised of annual data, are reported in appendix 10. All limitations aside, YEARSTER emerges as the only significant HC indicator, giving support to tertiary education as determinant of FDI. The outcomes cannot be compared with those in table 2 and 3 because data on the variable FDIDEV are measured in absolute values rather than as a relative measure (per GDP). Converting it to a relative measure is problematic given that the data source is different and each institution uses
23 The coefficient of YEARSTER-1 is 7,86 and the P-value is 0,046; The coefficient of ENROLTER-1 is 0,08 and the P-value is 0,112.
28
different definitions, which would result in measurement errors24. Therefore, we cannot
conclude whether FDI from developed countries responds differently to changes in human capital levels. This could be a topic for future research if more bilateral FDI data would be made available.
The tertiary education measures used in this research have proven to be robust to different estimation methods and different alterations of the model specifications. The results point towards a positive relation between tertiary education and FDI inflows, taking into account other determinants of FDI. This finding is in line with that of other studies. However, the same cannot be said for secondary education. The measures for secondary education (YEARSSEC and ENROLSEC) were not found to have a significant effect on FDI, whereas other studies did find a positive relationship between the two. For instance, Noorbakhsh et al. (2001) and Cleeve et al., (2015) find a significant positive relation between secondary education and FDI inflows in developing countries. This gives support to the claim that the effect of secondary education is larger in developing countries and less of a determinant in more developed countries, such as the East Asian countries studied in this research.
The results in table 3 seem to indicate that both market seeking- and efficiency seeking FDI are of importance in East Asia. The growth rate of GDP was found to have a positive effect on
FDI inflows, which can be characterized as market seeking FDI25. The significance of tertiary
education and the insignificance of RESOURCES as determinant of FDI, give support to the claim that efficiency seeking FDI is more influential in East Asia than natural resource seeking FDI (Zhou and Lall, 2005). However, the minimum threshold for attracting efficiency seeking FDI appears to be tertiary schooling rather than secondary schooling.
24 Data on FDIDEV is provided by the UNCTAD whereas data on the GDP is taken from the World Bank database. The UNCTAD data is measured in US dollars (on the basis of the value of the US dollar per 2010), whereas data by the World Bank is provided on basis of the current value of the US dollar.
29 5. Conclusions and Recommendations
East Asian economies have attracted rather high levels of FDI in recent decades. This research has confirmed some of the usual FDI determinants while others appeared to be less influential (or not of influence at all) in this region. In line with the economic literature, openness to trade, lagged changes in FDI and growth of the domestic market were found to have a significant positive effect on FDI inflows. Other traditional determinants such as market size, natural resource rents and periods of economic crisis were not found to be significant determinants of FDI in this region. Finding out which determinants are valid and which ones are not is useful if countries want to promote FDI inflows. Countries might be inclined to promote FDI inflows, because FDI is associated with technological innovation and economic growth according to several studies.
The main purpose of this thesis, however, has been to examine whether human capital levels act as determinant of FDI inflows into East Asian economies. To test this, data for 23 East Asian countries were studied over the period 1970-2015. The countries in question and the broad time period that has been studied, are some of the contributions of this research to the economic literature. Panel data regressions have shown that tertiary education is a significant and highly influential determinant of FDI. The indicators used to measure tertiary education have been robustly positive and significant. Variables that were used to measure secondary schooling, total years of education, and the percentage of the population without schooling were not found to be significant determinants of FDI. This shows that East Asian countries, and possibly other countries in the same state of development, should stimulate tertiary education if they are aiming to attract FDI. Thus, the minimum threshold for attracting efficiency seeking FDI appears to be tertiary schooling. This differs from less developed parts of the world, where secondary schooling levels also act as determinant of FDI. The significance of the growth rate of GDP suggests that a portion of FDI inflows in East Asia is comprised of the market seeking type of FDI. The variable indicating natural resource rents as a percentage of GDP, was not found to have a significant effect on FDI inflows. Therefore, no evidence was found that a (large) share of FDI in East Asian countries is of the resource seeking type.
This research continued by splitting the sample period in two in order to find out whether tertiary education has become more influential over time. This appeared not to be the case.
30 This is in contrast with Noorbakhsh et al. (2001), who found that tertiary education had become a more important determinant of FDI in recent years. Also, a regression was conducted to examine whether FDI originating from developed countries responded differently to human capital levels than FDI in general. Unfortunately, these test results were not very useable owing to many missing data, but at the least the test results did not contradict earlier findings. Average years of tertiary education emerged as the only significant human capital indicator, giving support to tertiary education as a determinant of FDI inflows. Finding out what factors are of influence on attracting FDI originating from developed countries could be interesting for future research, if more data on bilateral FDI flows is made available.
