University of Groningen Faculty of Economics and Business
Master Thesis
The Impact of FDI on product upgrading
Name Student: Paul Berger Student ID number: s2155362
Student email: p.berger.2@student.rug.nl
Paul Berger s2155362 I
Abstract:
This paper evaluates whether inward FDI conditional upon the level of social capabilit y stimulates product upgrading. Special attention is paid in order to use and interpret the interaction between FDI and social capability properly. A fixed effects model is applied in this paper. In general the results suggest that FDI inflows do have a positive impact for poorer countries with relatively high levels of social capability while richer countries do experience a negative effect at very high levels of social capability. Also differences between the social capabilities of the sexes have been found to have an impact.
Keywords:
Paul Berger s2155362 II
Table of content:
I. Figures ... 3
II. Tables ... 3
III. Table of Abbreviations ... 4
1. Introduction ... 1 2. Literature Review ... 4 3. Methodology... 13 3.1. Model ... 17 3.2. Variables ... 17 3.3. Data ... 18 4. Results ... 21
4.1. Limitations and further research ... 29
4.2. Conclusion ... 30
5. Appendix ... 32
5.1. List of all countries in the sample ... 32
5.2. List of variables... 33
5.3. Leverage versus residual squared plot ... 34
5.4. Breush and Pagan Lagrangian multiplier test for ransom effects ... 34
5.5. Hausman test ... 35
5.6. Correlation Matrix... 35
5.7. VIF test... 35
5.8. Breusch-Pagan / Cook-Weisberg test for heteroskedasticity ... 36
Paul Berger s2155362 III
I. Figures
Figure 1, World FDI inflows 1995-2011 ... 10
Figure 2, Tasks i plotted against value added ... 11
Figure 3, ME conditional upon the years of male education. ... 23
Figure 4, ME conditional upon the years of male, female and total education. ... 24
Figure 5, ME for two groups of countries according to their income. ... 25
Figure 6, ME in poor countries for the male, the female and the total population. ... 26
Figure 7, ME rich countries for the male, the female and the total population. ... 26
II. Tables Table 1, Descriptives. ... 19
Table 2, Three different measures for education. ... 22
Table 3, Robustness checks and two groups of countries. ... 27
Table 4, Three measures of education and two groups of countries. ... 28
Paul Berger s2155362 IV
III. Table of abbreviations
EU European Union
FDI Foreign Direct Investment
FEM Fixed Effects Model
GDP Gross Domestic Product
GNI Gross National Income
GVC Global Value Chain
ICT Information and Communication Technology MNE Multinational Enterprise
NAFTA North American Free Trade Agreement
OECD Organisation for Economic Co-operation and Development
OLS Ordinary Least Squares
REM Random Effects Model
R&D Research and Development
US United States of America
UNCTAD United Nations Conference on Trade and Development
Paul Berger s2155362 1
1. Introduction
Some countries grow faster than others. The prime example of this statement today would probably be China with its tremendous growth rates over the past two decades. Before China the prime examples were the Asian Tigers. When looking at Gross Domestic Product (GDP) growth rates in Europe former communist Europe as well as Turkey with its Anatolian Tigers come to mind. I think it is save to state that parts of this GDP growth can certainly be attributed to inward Foreign Direct Investment (FDI) flows. On the other hand a statement about FDI and its impact on industrial upgrading might not so easily be made. It could be that FDI leads to industrialization when targeted e.g. at assembly which certainly does add to GDP growth numbers which could lead to convergence in terms of income until up to a certain point. But if a country remains at tasks at the low value added end of a Global Value Chain (GVC) this convergence will come to a hold. Simply because higher value added activities may still be performed in the FDI home country.
FDI flows amounted to about 1.6 trillion US Dollars in 2012. Almost 70 million workers are employed in foreign affiliates and generated a value added of 7 trillion US Dollars.1
In the light of these numbers it is not surprising that the field of FDI is a prominent one for research as well as subject to debates centring on the impact of FDI in- and outflows. Sturgeon and Gereffi (2009) mention local linkages of multinational enterprises (MNEs) and technology transfer as positive impacts of FDI while negative effects can include the crowding out of domestic firms and the practice of international transfer pricing. In general the impact of FDI on growth is quite intuitively a positive one. Furthermore FDI seems to spur growth more than domestic investment (Borensztein, De Gregorio, and Lee, 1998). Several studies have found that differences in the level of technology can explain large parts of the differences in per capita income between different countries (e.g. Abramovitz, 1986; Fagerberg, 1994 or Hall, and Jones, 1999). One way to close the technology gap could be through spill over, resulting from FDI (Aitken, Hanson, and Harrison, 1997). However, a host country’s ability to absorb
1
Paul Berger s2155362 2 technology spill over might depend on a certain level of social capabilities (Abramovitz, 1986) like a country’s average level of education (Barro, and Lee, 1994).
After the second unbundling (Baldwin, 2006) it became feasible to slice up the value chain and perform certain tasks abroad. This lead to a shift in trade patterns (Hanson 2012) and large “North-South” trade flows evolved. The second unbundling not only made industrialization “easier and faster but [also] less meaningful” (Baldwin 2011). This was shown impressively by Dedrick, Kraemer, and Linden (2009) who find that only a very small portion of the financial value of an iPod is actually added in China where the iPods are assembled. The lions’ share is added in the US where the iPod is designed and branded. In other words: it is not industrialization that matters per se for a country’s development but rather which tasks a country performs in a GVC. So the process of upgrading or moving up the GVC and the quality of growth are the focus of this paper.
This paper contributes to the existing literature as it investigates the impact of FDI on export upgrading. The sample used was kept as broad as possible and includes 115 developing, emerging and highly developed countries. This sample delivers interesting results since it also includes a number of developing and emerging markets on which debates about what shapes economic development usually focus. Discussions circle around topics such as sector specific industrial development policies, natural resource endowments or the role of technological learning and industrial upgrading (Sturgeon and Gereffi, 2009). The presumably positive effect FDI has on the receiving economy is investigated in this paper. Of interest here is the impact on export upgrading in the host economy. Furthermore this paper investigates the interaction between FDI and the level of education in the receiving economy.
A brief definition of FDI is given by UNCTAD (2007) which defines FDI as “an investment involving a long term relationship and reflecting a lasting interest and control by a resident entity in one economy … in an enterprise resident in an [another] economy”.
Paul Berger s2155362 3 importance of export upgrading. The main research question is one of the correlations between inflows of FDI and export upgrading:
Do FDI inflows enhance export upgrading?
Measuring upgrading on the country level can be quite difficult; A simplistic approach will be used that can hint towards an answer for this question. Answering the research questions can explain how export upgrading is influenced by FDI and its interaction with social capabilities. Measurement issues concerning industrial upgrading and also social capabilities are discussed in the methodology section.
Sturgeon and Gereffi (2009) discuss measurement issues concerning industrial upgrading in a GVC. They highlight the Lall (2000) approach of analyzing exports according to their level of sophistication. In this paper high tech exports according to OECD (2011) are analysed. Country level data on the share of high tech exports will be the focus of this analysis. This leads to an answer of a sub research question such as:
Do FDI inflows measured as percent of GDP into a host country lead to a higher share of high tech exports?
The second sub question analyzed is concerned with the interaction between FDI and the level of education in the receiving economy.
Does a positive relationship between FDI and the export upgrading exist, given a certain level of education in the host country?
The main findings can be summed up as follows:
Paul Berger s2155362 4 female part of the population for both groups of countries. Especially interesting is that the ME for lower middle and low income countries always remains positive for the female population while it remains negative for the male part of the population up to a level of more than 5.5 years of mean years of education.
The remainder of this paper is structured as follows : after this introduction a literature review follows in chapter 2. From this literature review several hypotheses are derived. Also an overview of FDI flows is given. In chapter 3 the methodology applied is developed. In chapter 4 the empirical findings are discussed and the data used described. Robustness tests are conducted and conclusions are drawn.
2. Lite rature review
A higher share of sophisticated exports is associated with higher growth levels. This ‘stylized fact’ has been described several times (Rodrik, 2006; or Hausman et. al., 2006). So upgrading in terms of moving to the production of more technology intensive goods does have a positive impact on growth figures. Lall (2000) also argues that countries that export higher shares of high technology goods grow faster. The reason being that high technology intensive products “tend to grow faster in trade [because] they tend to be highly income elastic, create new demand and substitute faster for older products”. The author shows that exports in high tech goods (the most technology intensive group of goods) grew fastest in all observed 5 year periods from 1985 onward; e.g. resource based exports from developing countries grew by 6 per cent from 1985 to 1998, while Automotive exports grew by more than 20 per cent or electronic exports grew by more than 22 per cent in the same time period.
Paul Berger s2155362 5 analyse all four types of upgrading. So due to the complexity of this topic but also due to the availability of data this paper focuses on the aspect of product upgrading.
So what is the role of FDI in the process of upgrading? To answer this question we have to remind ourselves that differences in income mainly stem from differences in the level of technology used in a certain country. This has been found by several studies such as Abramovitz, (1986); Fagerberg, (1994) or Hall, and Jones, (1999). Abramovitz (1986) also formulates what he calls the ‘Catch-Up’ hypothesis. It explains why a country that is more distanced from the technology frontier catches up quicker than a country relatively closer to the frontier. The basic thought is that advanced technology which is already available somewhere else is transferred to a country that lags behind: through the implementation of this new technology a country jumps up the ladder in terms of technology. This jump or leap is larger the further away the lagging countr y was initially from the technology frontier. Abramovitz (1986) does not go into detail on how this new technology is transferred but one possible channel is FDI.
If the level of technology matters, one way to grow economically is to close this technology gap. One of the positive impacts of FDI that can be found in the literature is its potential to increase the rate of technology transfer through spill over ( Aitken et al, 1997; Barrell, and Pain, 1997; Young, and Lan, 1997). Even though some studies fail to find the positive link between FDI and technology spill over (Mohnen 2001), Ford, Rork, and Elmslie (2008) state that these mixed findings could potentially stem from a failure to include absorptive capability in the model. As we will see further below this capability is crucial to determine the impact of FDI on upgrading.
Paul Berger s2155362 6 firms: we can expect them to be much larger than other firms in terms of shipments and employment. Also they pay higher wages and they are on average more productive. One way that FDI can have a positive impact on a receiving economy therefore is through knowledge and technology spill over from more productive foreign firms. Aitken et al. (1997) investigate spill over effects from the geographic concentration of MNEs and from domestic exporting firms. Besides through technology transfers the authors mention several other possible positive channels, among them specialized infrastructure or information about foreign markets and consumers. Kemeny (2000) mentions several other channels; on one hand he mentions spill over effects that stem from a MNE’s organizational advantages such as fruits of R&D, advanced machinery, or management know how. Even though MNEs might try to protect these advantages, spill overs through former employees or reverse engineering seem to be inevitable. But on the other hand Kemeny (2000) also mentions another channel through which MNEs can spur domestic productivity: the entry of an MNE increases the competitive pressure on domestic firms. Especially when keeping in mind that MNEs in a stylized world are more productive than domestic firms (they have substantial organizational advantages). This new pressure could force domestic firms to engage in one of the forms of upgrading or they could learn to use already employed technology more efficient. In general I think it is save to state that FDI inflows are associated with accelerated economic growth measured by GDP. This has been found by a number of studies (e.g. Borensztein et al., 1998; or Doraisami, and Leng, 1996).
Paul Berger s2155362 7 What are the main characteristics in order to maximise the potential of inward FDI? Quite intuitively and also according to Hausman and Rodrik (2003) upgrading is not a trivial task. The authors mention time, capital but also the skills of the labour force as necessary resources. Lall (2000) describes the notion that in order to use FDI to its full potential a certain level of social capability (being able to use new technology efficiently) is necessary. Lall (2000) describes an interesting thought; he criticises the assumption that imported technology (capacity) can be used efficiently without additional costs (capability). This would mean that comparative advantage would solely depend on factor endowments and attempts to change them since there is no difference between capacity and capability. Lall (2000) argues that this thought is oversimplified and that the more complex the activities are in industries targeted by FDI, the higher the level of capability needs to be. Abramovitz (1986) mentions technical and political competence, and also commercial, financial and industrial institutions as dimensions of social capability. Kemeny (2010) captures these dimensions by taking into account the average years of schooling as well as the number of telephone lines and measurements for e.g. political rights and civil liberty. As will be explained in the methodology section the measure for social capability in this paper is more basic than those used by Kemeny (2010).
Paul Berger s2155362 8 education is taken into account. Also the authors confirm the notion of Aitken (1997) that FDI also does attract domestic investment.
The most important finding of Aitken et al. (1997) for this paper is the positive relationship between the geographic proximity to multinational firms and the likelihood that a domestic firm engages in exporting. So the proximity of a domestic firm to a MNE increases the likelihood of the domestic firm to engage in exporting. On the other hand a concentration of domestic exporters is not correlated with the likely hood of exporting. So in other words, Aitken et al. (1997) confirm that spill overs usually stem from MNEs and not from domestic firms.
Similar conclusions have been drawn by Ford et al. (2008). They investigate the interaction effect of FDI and human capital on growth of per capita output of US states. The authors find that given a certain level of human capital, the impact of FDI on growth becomes bigger than the one of domestic investment; the spill over resulting from FDI exceeds that of domestic investment given a certain level of capabilities. Wang and Wong (2011) do get to conclusions similar to those of Borensztein et al. (1998). They find a positive relationship between FDI and growth. When looking at the interaction between FDI and schooling the authors use different measures of schooling to take into account not only quantity but also quality. Wang and Wong (2011) find similar relationships to those found by Borensztein et al. (1998) but the educational threshold calculated is lower.
Paul Berger s2155362 9 could be due to the notion described by Gauselmann, Knell and Stephan (2011) that FDI is seeking low wage costs paired with well educated and trained workforce.
Also the impact of FDI could differ depending on the level of education for the two sexes. Shu, Zhu and Zhang (2007) investigate the impact of inward FDI on the Chinese labour market. The authors conclude that employment patterns differ for male and female workers: male workers tend to move to new and better paying jobs created by MNEs while Chinese women tend to participate in global production by entering export-oriented manufacturing. In other words men tend to move to higher value added tasks while women generally enter the GVC on the level of assembly. Braunstein and Brenner (2007) find that since 2002 inward FDI to China lead to larger wage increases for men than for women. The authors argue that this is due to the fact that women are more likely to be employed in low-skilled, export oriented production while men a more likely to be employed in industries associated with higher productivity.
At the end of the literature review FDI flows are discussed in more detail but also issues concerning the interpretation of export statistics are highlighted.
Baldwin (2006) describes mechanisms that can help explain the structure of today’s FDI flows and can also help to interpret export statistics. FDI flows have been subject to tremendous fluctuations in the past two decades. FDI flows stared surging in the mid-1990s (in line with Bladwin 2006) and reached a first peak in the year 2000 just before the Dotcom bubble. After declining for three years, flows started to pick up again and peaked again in the fourth quarter of 2007 and after that dropped sharply again for two years. In 2010 world FDI flows started to recover and are expected to reach the pre-crises level in 2014.
When looking at the numbers FDI flows become impressive: FDI stock reached 20.4 Trillion US Dollars and flows were expected to reach 1.6 Trillion US Dollars in 2012. An estimated 69 million workers are employed in foreign affiliates of MNE who generated a value added of 7 trillion US Dollars.2
The following figure depicts FDI inflows in the period from 1995 to 2011. The most important trends described above can clearly be seen.
2
Paul Berger s2155362 10 Figure 1, World FDI inflows 1995-2011 in Billions of US dollars. S ource: UNCTAD (2012).
These developments are what one would expect from Baldwin (2006). Baldwin in his influential paper discusses two phenomena: the so called first and second unbundling. In other words, he discusses the emergence of what is often referred to as globalisation. In a nut shell the first unbundling describes the process of the separation between production and consumption mainly d ue to reduced trade costs. So in a stylized world, trade would mainly occur in final goods. This first unbundling was interrupted by the two world wars and in the 1980s the second unbundling evolved. A drop in the costs and an increase in quality of information and communication technology (ICT) changed the face of globalisation: trade in final goods gave way to trade in intermediate goods. Or put differently: instead of only separating production and consumption it became feasible to perform different tasks in different geographic locations. This second unbundling not only had an impact on the structure of trade and FDI, it also led to the emergence of so called GVCs. The timing of the emergence of new ICT in the 1980s just before a sharp rise of FDI flows shortly after can be seen as a result of this second unbundling.
Paul Berger s2155362 11 in the apparel GVC and plots them
against the value added in each task (figure 2). Of course this is just one example of a GVC but it captures an important intuition quite well: it does matter which task a certain country performs in a GVC. It can also explain why Baldwin (2011) states that the second unbundling not only
made industrialization “easier and faster but [also] less meaningful”.
Let us take the apparel GVC as an example: Bangladesh is well known for its apparel industry, the main task being performed is the production (trim and cut) of apparel. As indicated by figure 2 this is the activity associated with the lowest value added content. Activities with high value added content like design or R&D remain almost exclusively in highly developed countries. So assuming that Bangladesh performs all the tasks in a GVC up to the point of production can lead to a wrong picture. Most value is still added in ‘northern’ countries. An impressive example can be found in the literature (Linden, Kraemer, and Dedrick, 2009). Linden et al. (2009) show that most of the value added of an Apple iPod is captured in the US where the iPod is designed, retailed and branded. The activity of assembly (referred to as production in figure 2), a low value added activity, is performed in China. So drawing conclusions about China’s level of development by looking solely at its export structure can be misleading: exporting iPods does not mean that high value added intangible tasks are performed in the country of production. This is the point Baldwin (2011) refers to when he states that industrialization has become more meaningless.
Does this mean that looking at export structures has become meaningless? No, but drawing conclusions about a country’s level of development solely on exports is difficult. As mentioned above technology intensive exports are associated with higher growth rates (Rodrik, 2006).
Paul Berger s2155362 12 typically results in increased exports in the apparel sector. These additional trade flows certainly do have an impact on the composition of exports of a certain country and also leads to shifts in trade patterns. This is in line with the shift in trade patterns found by Hanson (2012) described above.
From what we have heard so far it is not surprising that after the second unbundling GVCs emerged and also that worldwide trade patterns changed and that they are still changing (e.g. Hanson, 2012; or Lall, 2000). While worldwide trade flows were dominated for a long time by north- north trade, great shifts towards more south-north as well as south-south trade have recently occurred. Hanson (2012) mentions the rise of India, China and Brazil as examples. So the second unbundling did not only have an impact on FDI flows but also, quite intuitively, on worldwide trade patterns. Also we know now that it does matter what kind of goods a certain country does export, high tech exports being associated with lo ng term economic growth.
This leads paired with the thought of expertise- and technology spill over to the assumption that one should be able to observe inward FDI leading to a shift in the trade structure. Assuming that FDI investments are executed by MNEs that do have a certain edge over domestic firms in terms of technology this should lead to some sort of spill over and therefore upgrading.
After highlighting the interesting effect social capability or human capital can have on the effect of FDI on growth and upgrading testable hypotheses are developed. The findings of the literature described above permit me to derive several testable hypotheses for this paper. Most important in the context of this paper is the relationship between inward FDI and export upgrading as well as the interaction between FDI and education in the same context.
H1: A positive relationship between inward FDI measured as part of GDP and
the share of high tech exports as per cent of manufacturing exports does exist.
H2: The marginal effect of inward FDI on export upgrading is positive beyond
Paul Berger s2155362 13
H3: The marginal effects of FDI on export upgrading conditional on the level
of social capabilities is different for countries from different income groups.
H4: The marginal effects of FDI on export upgrading conditional on the level
of social capabilities is different for the two sexes.
3. Methodology
To measure export upgrading the share of high tech exports as a per cent of manufacturing exports is used in this paper. This measure is used to keep the sample as broad as possible in terms of countries and years included. As described above the category of high tech exports is the one associated with the highest growth numbers. Furthermore this measure permits us to capture one of the four types of upgrading in the GVC framework Humphrey and Schmitz (2002): product upgrading (moving to the production of more sophisticated goods). High tech goods are the most sophisticated category of goods as defined by OECD (2011) in terms of technology intensity.3 Therefore a rise in the share of this category of exports is considered as the measure of product upgrading in this paper. One of the main advantages of the variable used in this paper is that it is easily available for a large number of countries and years.
To capture the level of social capability the mean years of education for different groups of the population and different levels of schooling are used. These measures of education are not optimal to capture social capability: Kemeny (2010) measures social capability not only by looking at years of education but also at other factors such as phone lines or political rights. Hanushek, and Woessmann (2012) find that educational attainment measured in years of education is an imperfect measure when it comes to explaining long term growth. They highlight the advantages of a measure for cognitive skills. Also Wang and Wong (2011) argue that not only the quantity of education matters (years of education) but also the quality (here the authors construct an index that captures a country’s performance in international tests) when assessing the impact of FDI, education and their interaction on growth. Borensztein et al. (1998) use average
3
Paul Berger s2155362 14 years of education as a measure of human capital which is a comparable measure to the one used in this paper. Nonetheless the directions of the relationships found by Borensztein et al. (1998) are also confirmed by Wang and Wong (2011). The main problems with all these measures are that data is no t available for most countries. So constructing a measure to capture social capabilities was not feasible. Given the broad sample used here and also looking at the number of developing and emerging economies in the sample it is not surprising that the choices for measures of social capability are quite limited. Due to limited data availability but also due to the scope of this paper, mean years of education are used. Therefore the measure used in this paper actually captures education which is only one facet of human capital which in turn captures only a part of social capability. Nonetheless this seems to be appropriate since the results found by Wang and Wong (2011) and Borensztein et al. (1998) and Ford et al. (2008) do not differ substantially in terms of the directions of the relationships found.
To deepen the analysis I compare the interaction of FDI with the mean years of total schooling of the male population to two other measures: the mean years of total schooling for the female and for the entire population. As described in the literature review there could be differences in the impact of educational achievements for the different sexes. We can expect differences in terms of jobs performed ; female workers are mainly employed in low value added tasks while male workers tend to move to newly created well paid jobs (Shu, et al. 2007). Also differences in terms of wages do exist (Braunstein and Brenner, 2007) These differences could have an impact on the interaction between education and FDI.
Paul Berger s2155362 15 educational measures for different parts of the population are taken into account for the two different groups of countries.
To address the second hypothesis an interaction model is created. This model includes FDI flows as well as educational attainments and several control variables as well as a multiplicative interaction term. A panel regression is used since panel data is analysed with N countries and T time periods. Several panel models could be considered as options for this analysis. They allow us to analyse observations across time and countries which is not possible in time series or cross section analyses.
The pooled Ordinary Least Squares (OLS) model can be seen as the most restrictive one, since this model neglects differences across countries so the effects unique to each country are summed up in the error term eit. The so called Fixed Effects Model (FEM)
assumes that there is unobserved heterogeneity present in the data, so the intercept as well as the coefficients for each country are different. The model can be written as follows:
yit = β1 + β2 xit + β3 zit + … + eit
A popular simplification of the FEM is to have different intercepts for each individual country. This can be written as follows:
yit = β1i + β2xit + β3zit + … + eit
Here the missing i subscript for all β except for β1 indicate that the coefficients are
constant for all individuals. The FEM is generally not suitable for short and wide panels which would make the estimates not precise enough.
In the Random Effects Model (REM) all individual differences are assumed to be captured by the intercept but the individual differences are treated as being random and not as beeing fixed. The REM can be seen as more precise and it is able to estimate the impact of time invariant variables.
Paul Berger s2155362 16 First a Breusch-Pagan Lagrange multiplier test is conducted to test the following hypothesis: H0: σ2e = 0. If H0 is rejected the REM is appropriate since significant
random effects are present, otherwise a pooled OLS model should be used.
To check if correlation between the error component ui and the regressors is present in a
REM a Hausman test is used. After running the FEM and the REM the Hausman test is conducted. Here it is tested if the estimators of a FEM and a REM are significantly different. If the null hypothesis, that no statistical relevant differences exist, is rejected using the FEM is appropriate. As expected the conducted tests confirm that a FEM model is appropriate.
Paul Berger s2155362 17
3.1. Model
The basic model for the regression with EXP as the left hand variable can be stated as follows:
LOG_EXPit = β1 + β2FDIGDPit + β3EDUMALEit +
β4FDI_EDUMALEit +β5itGDPCAP+
β6RDGDPit + eit
where the i subscript indicates a certain country while the t subscript is the time indicator.
3.2. Variables
High technology exports as per cent of manufacturing exports (EXP): This is the left hand variable in this model. A Shapiro-Wilk test confirms that this variable should be transformed into logs to fit the assumption of a normal distribution of the residuals better.
FDI inflows as part of GDP (FDIGD P): This variable can be drawn from the WBG homepage without further calculations. Since the effect of inward FDI might not show immediately robustness checks are carried out with FDIGDP lagged by 1 and 2 time periods. A one year lag seems to be standard in the literature (e.g. Kim and Hwang 2000; or Anayanwu 2012).
Mean years of schooling for the male population above 25 years of age (EDUMALE): Here the approach of Barro and Lee (1994) is followed who find that average years of educational achievement of men is the most important one in this context. Since average years of male schooling cannot be drawn from the WBG homepage mean years of male schooling above 25 years of age is used. The variable can be drawn from the WBG homepage without further calculations. Since it might take some time for changes in the level of education to have an impact robustness checks with EDUMALE lagged by 1 and 2 time periods is applied. Since the relationship between EDUMALE and the impact of FDI is in the focus here an interaction term is calculated:
FDI_EDUMALEit = FDIGDPit * EDUMALEit
Paul Berger s2155362 18 the total population. A regression is run with a variable for the mean years of total schooling for the total population of age 25 and above and the corresponding interaction term. A second regression is performed including the mean years of total schooling of the female population of age 25 and above. The great advantage of using these three variables is that they are easily comparable since they measure the same educational achievements for different parts of the same population.
GDP per capita (GDPCAP): As a control for the level of the development of a specific country GDP per capita is included. Harding and Javorcik (2012) find that GDP per capita partly determines the number as well as the composition of exports. Hummels and Klenow (2005) find that richer countries export higher quantities of goods at “modestly higher prices, consisted with higher quality”. GDP per capita is a good proxy for labour costs and for the level of development (Harding and Javorcik, 2012). The authors find that the level of GDP per capita does have an impact on the export structure of a country and should be controlled for.
R&D expenditures as part of GDP (RDGDP): Since industrial upgrading could stem from domestic R & D, independently from inward FDI, this control variable needs to be included. Bandick and Hanson (2009) investigate the impact of inward FDI on the demand for skilled labour. The mechanism being that not only inward FDI but also domestic R&D can lead to an increased demand for high skilled workers.
3.3. Data
The data needed can be retrieved from the WBG and can be accessed online. In total the sample consists of 115 countries and 24 years.4 The time period observed runs from 1988 to 2011. This time period was chosen due to a lack of data mainly for the variable of high tech exports. Data for this variable is not available before 1988. Nonetheless the period from the late eighties onward is the most interesting period to look at when it comes to FDI flows. As we have seen above FDI flows started surging in the nineties so all important developments are captured by this time period. Also more than two decades are enough to consider possible lagging effects of variables such as FDI or education. R&D expenditures with a total of 1218 observations proved to be a bottleneck to data collection so data was taken from two different WBG sources; the
4
Paul Berger s2155362 19 World Development Indicators databank and the Innovation and Development data base.5 In general it can be stated that variables where chosen in such a way to ensure a data set as wide and long as possible. This was done to ensure that empirical results are based on a solid dataset. Several countries were excluded from the sample as outliers using a leverage versus residual squared plot.6
Table 1, Descriptives.
The country with the lowest educated population in the sample is Ethiopia with 0.86 years of mean total schooling of the total population above 25 years of age in 1999 (1.4 years for the male population and 0.3 years for the female population). On the other side of the spectrum are highly developed countries. The highest mean of total schooling for the total population was observed for Canada in 2010 with slightly over 13 years of mean education (male 13.2 and female 12.9 years).
Also the differences in terms of inward FDI as well as differences in domestic R&D expenditures are quite substantially: Hungary is the country with the highest FDI inflows within the whole period, in 2007 almost 52 per cent of the total economy’s GDP. The lowest observed value was also recorded for Hungary, this time in 2010: -16 per cent of GDP.
Expenditures in the domestic R&D sector reach their highest level in 2007 in Israel with 4.8 per cent of GDP. In general the position of the different countries is not too surprising: The highest places are exclusively taken by highly developed countries while the bottom places are taken by poorly developed countries. The main exceptions seem to be Monaco, Kuwait and Saudi Arabia which all can be found in the lower
5
A list of all variables and the corresponding sources can be found in the appendix. 6
Paul Berger s2155362 20 quarter for some years. The last place in the list is taken by Zambia in 2002 with domestic R&D expenditures amounting to 0.006 per cent of GDP.
Quite counter intuitively the list of countries exporting the highest shares of high tech goods of total manufacturing exports is topped by a number of deve loping countries. The Philippines lead the ranking with a share of almost 75 per cent of high tech exports. The lowest shares, not too surprisingly 0 per cent, can be found in almost exclusively in developing countries. How can the counter intuitive finding that the Philippines top the list be explained? I think it is important to remind ourselves that the measure used as left hand variable captures the share of high tech exports of total manufacturing exports,
not total exports. So in countries with low overall manufacturing exports this
Paul Berger s2155362 21
4. Results
As expected a Breusch and Pagan Lagrangian multiplier test for random effects as well as a Hausman test confirmed the use of a fixed effects model. Furthermore slight multicollinearity between the interaction term and FDI seems to be present. As mentioned in Brambor et al. (2005) this should not be of further concern here.7 High multicollinearity could lead to large standard errors which is not too worrying here since I am not primarily interested in the significance of the different parameters but rather this analysis focuses on the interaction term between FDI and education.
The purpose of this paper is to investigate the effect of FDI on export upgrading. Also, as has been described above, I am particular interested in the interaction between FDI and social capability measured in years of education. In general it has to be stated that the explanatory power of this model does not seem to be very high. This is not surprising given the simplicity in terms of the number of variables included in the model.
Table 2 shows the regression results as well as the results of several robustness checks with different measures of education. Table 3 further below shows the results of two robustness checks run with lags. Also regression results from a sample split into high and upper middle income countries on one side and low and low middle income countries on the other are presented in table 3. As the reader will notice these regressions do not include the variable R&D. This is due to insufficient data availability and is discussed in more detail at the end of this chapter.
7
Paul Berger s2155362 22 Table 2, Regression table with three different measures for education.
Regression Number: 1 2 3 4 5 6 FDI 0.0135*(0.007) (0.008)0.0014 (0.035)0.0206 (0.035)0.0170 (0.035)0.0332 (0.035)0.0247 Education 0.2712***(0.074) 0.2718***(0.074) 0.2976***(0.086) 0.2473***(0.50) 0.2770***(0.071) Interaction term FDI*Education -0.0019 (0.003) -0.0015 (0.003) -0.0034 (0.003) -0.0024 (0.003) GDP per Capita (0.000)0.0000 (0.000)0.0000 (0.000)0.0000 R² 0.00 0.03 0.03 0.03 0.03 0.03 No. of Observations 2056 2056 2056 2056 2056 2056
Robust standard errors are reported in brackets. Significant (2-ta iled) at the *** 1%, ** 5% and * 10% levels. In regressions 1 to 4 the educational attainments of the male population are used as the variable ‘Education’, as well as in the interaction term. In regression 5 educational attainments for the fema le part of the population and in regression 6 educational attainments for the total population are used.
As indicated above the impact of FDI, social capability and their interaction are at the core of this paper. In the methodo logy section some special features were highlighted when it comes to interpreting interaction terms. The results from model 4 (base model) in table 2 suggest that both FDI as well as the average years of education for the male population above 25 years of age have a positive impact on export upgrading. However, this result has to be interpreted in an appropriate way, taking the interaction term into account (Brambor et al. 2005). As in this paper a conditional hypothesis is tested, this conditionality is taken into account by the mentioned interaction term. The interpretation of this interaction term differs from the interpretation of normal coefficients. What has to be kept in mind is that the conditionality prevents us to interpret the coefficients of FDI and education independently. Rather the marginal impact of the interaction term on upgrading has to be calculated as:
= β1 + β3 * Years of education
Paul Berger s2155362 23 -0,005 0,000 0,005 0,010 0,015 0,020 0 1 2 3 4 5 6 7 8 9 10 11 12 13 additional unit of FDI remains
positive. So from this finding we have to assume that for most countries the ME remains positive since the limit of 11 years of education is quite high. Only 20 countries out of 115 in the sample for 2011 do reach this 11 year level of education. This finding is quite surprising given the findings in the
literature. From Kemeny (2010) we could have expected this relationship to be upward sloping for the entire sample, meaning that any additional year of education adds to the positive relationship between FDI and upgrading. It has to be mentioned though that Kemeny (2010) uses different measures for social capabilities as well as a more sophisticated measure for upgrading.
Paul Berger s2155362 24 -0,015 -0,010 -0,005 0,000 0,005 0,010 0,015 0,020 0,025 0,030 0,035 0,040 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Ma le Tota l Fema le but discourages product upgrading. Further, the measure of inward FDI used here captures FDI targeted at all sectors, not only high tech sectors. The relationship found might be different if FDI flows were analysed according to the targeted industries. This was not possible here due to data limitations.
Quite similar results were obtained for the two comparable measures of education: education for the total population above 25 and for the female population above 25 years of age (both measured in mean years of total education). These two other measures of educational attainment are used to deepen the analysis of the effects of education.
Overall they do confirm the notions found for male education. The MEs of all three interaction terms are depicted in figure 4. Since the confidence intervals are not depicted here it is not possible to determine below or
above which level of education the ME becomes statistically significant (Brambor et al. 2005). As can be seen in figure 4 the slope for the male population is less steep than the ones for the female and total population. So any additional unit of FDI conditional on the years of education for the male population leads to smaller positive or negative impact than for the female population. So the ME of the female population is bigger than that for the male population, be it positive or negative.
Paul Berger s2155362 25 -0,04 -0,03 -0,02 -0,01 0 0,01 0,02 0,03 0,04 0,05 0,06 0 1 2 3 4 5 6 7 8 9 10 11 12 Lower middle a nd low income Upper middle a nd high income
indicates that the effect turns negative at a lower level of education for female population as well as the total population. It is interesting to see that this ME for the female population is negative for 30 countries out of 115 in 2011 compared to 20 countries that do reach this level for male education.
Figure 5 depicts the ME of FDI conditional on the level of education for the male population for two groups of countries; high income and upper middle income countries on the one hand and
low and lower middle income countries on the other hand. Here we can observe an interesting finding. The ME for richer
countries is
downward sloping while that of poorer
countries is upward sloping. This means that for poorer countries below 5.5 years of education the ME is negative while after that it turns positive. The opposite is the case for richer countries. For them the ME turns negative if a level of approximately 10.3 years of education is exceeded. So beyond a certain level of education poorer countries benefit more from FDI than richer countries in terms of product upgrading. Striking about this finding is that almost all lower middle and low income countries reach an educational threshold of 5.5 years beyond which the ME turns positive. Only six countries8 are below this educational threshold in 2011. In other words this means that the ME is positive for almost all lower middle and lower income countries. This indicates that a certain level of education is necessary for poorer countries to benefit fully from FDI in terms of upgrading. This could mean that this level of education is necessary to perform certain tasks efficiently. On the other al lot of upper middle and high income countries in the sample reach an educational level of more than 10.3 years
8
Burkina Faso, Honduras, Ethiopia Mali, Mo za mb ique, Tanzania .
Paul Berger s2155362 26 -0,04 -0,03 -0,02 -0,01 0 0,01 0,02 0,03 0,04 0 1 2 3 4 5 6 7 8 9 10 11 12 Ma le Fema le Tota l -0,02 0 0,02 0,04 0,06 0,08 0 1 2 3 4 5 6 7 8 9 10 11 12 Ma le Fema le Tota l of mean education. The ME is negative for almost half upper middle income countries and for around three quarters of all high income countries in 2011.
The next two figures depict the MEs for the three measures of education and the two groups of countries. Figure 6 depicts the MEs for lower middle and low income countries while figure 7 depicts upper middle and high income countries. The results obatianed for lower middle and low income countries show that the ME is
always positive for the female population while the ME for the male population turns positve only after reaching a threshold of 5.5 years. As can be seen the positive ME for the male popualtion only exceeds the ME for the female population after a bit more than 7.5 years of mean years of schooling. As one could have expected most countries (two thirds) remain below this level of male education. So in general (for poorer countries) the education of the female part of the labor force promises bigger benfits in terms of product upgrading than an aditional year of education for the male part of the population. This changes
only at a quite high level of education, which could be an indication that primarely female workers are employed by MNEs while male workers might mainly be employed for more complex tasks than e.g. ‘just’ assembly.
The results for upper midle and high income countries in figure 7 show that the ME for the female part of the popualtion are higher below a level of approximatly 9.5 years of Figure 7, Marginal effect for lower middle and low income countries for the male, the female and the total population.
Paul Berger s2155362 27 schooling. Also the ME turns negative slightly to the right of the ME of the male part of the popualtion.
These results suggest that for around two thirds of the richer countries the negative ME for the male popuation is smaller than that for the female population. In general the differnces between the sexes seem to be less important in upper middle and high income countires than in lower middle and low income countries.
Two variables that were included as control variables (R&D and GDP per capita) prove to be of limited importance for these analyses. As can be seen in tables 2, 3 and 4 the coefficient for GDP per capita proofs to be almost 0 and hardly ever becomes significant.9
Table 3, Regression table with robustness checks and two groups of countries.
Regression number: 1 2 3 4 5 FDI (0.035)0.0170 (0.035)0.0337 -0.0030(0.035) (0.073)0.0502 -0.0280(0.040) Education male 0.2976*** (0.086) 0.0829** (0.042) -0.0007 (0.027) 0.1132 (0.081) 0.3795*** (0.156) Interaction term FDI*Education male -0.0015 (0.003) -0.0034 (0.042) 0.0000 (0.003) -0.0049 (0.007) 0.0051 (0.006) GDP per Capita (0.000)0.0000 (0.000)0.0000 0.0000**(0.000) (0.000)0.0000 (0.000)0.0000 R² 0.03 0.00 0.00 0.00 0.03 No. of Observations 2056 2026 2000 983 1073
Robust standard errors are reported in brackets. Significant (2-tailed) at the *** 1%, ** 5% and * 10% levels. Regression 1 equals regression 4 in table 2. In regressions 2 and 3 lags of 1 and 2 time periods respectively were used for the variables ‘FDI’, ‘Education male’ and the corresponding in teraction term. Regression 4 shows the results of regression 1 only for high inco me and upper middle income countries. Regression 5 shows the results of regression 1 only for lo w income and lowe r middle inco me countries.
9
Paul Berger s2155362 28 Table 5, Regression table including R&D.
Regression Number: 1 2 3 FDI (0.035)0.0170 -0.0110(0.032) -0.0111(0.032) Education male 0.2976*** (0.086) 0.2312*** (0.060) 0.2316*** (0.058) Interaction term FDI*Education male -0.0015 (0.003) 0.0009 (0.003) 0.0010 (0.003) GDP per Capita (0.000)0.0000 (0.000)0.0000 (0.000)0.0000 R&D 0.004 (0.102) R² 0.03 0.05 0.05 No. of Obs. 2056 1096 1096
Robust standard errors are reported in brackets. Significant (2-tailed) at the *** 1%, ** 5% and * 10% leve ls.
Regression number 1 in table 3 is the basic regression and is similar to regression 4 in table 2. In regressions number 2 and 3 of table 3 lags of one and two time periods where applied for the variables ‘FDI’, ‘education male’ and the interaction term. A lag of 1
Regression Number: 1 2 3 4 5 6 FDI -0.0280(0.040) (0.034)0.0056 -0.0095(0.037) (0.073)0.0502 (0.078)0.0755 (0.078)0.0639 Education 0.3795*** (0.156) 0.3189*** (0.108) 0.3542*** (0.074) 0.1132 (0.081) 0.1012 (0.072) 0.1144 (0.077) Interaction term FDI*Education 0.0051 (0.006) 0.0006 (0.005) 0.0027 (0.005) -0.0049 (0.007) -0.0079 (0.008) -0.0064 (0.007) GDP per Capita (0.000)0.0000 (0.000)0.0000 (0.000)0.0000 (0.000)0.0000 (0.000)0.0000 (0.000)0.0000 R² 0.03 0.04 0.03 0.00 0.01 0.01 No. of Observations 1073 1073 1073 983 983 983
Robust standard errors are reported in brackets. Significant (2 -tailed) at the *** 1%, ** 5% and * 10% levels. In regressions 1, 2 and 3 the educational attainments of the male , the fe ma le and the total population respectively for low middle and low inco me countries are used. In regressions 4, 5 and 6 the educational attainments of the ma le, the fe ma le and the total population respectively for h igh middle and high inco me countries are used.
Paul Berger s2155362 29 year increases the coefficients for the interaction term as well as for ‘FDI’. The coefficient for ‘Education male’ drops substantially but the overall relationships remain as found in regression number 1. Applying a lag of two years does change the relationships as well as the size of the coefficients. The impact of ‘FDI’ as well as ‘education male’ turns negative while the sign in front of the interaction term turns positive. This finding is quite interesting and might require some further attention. Noteworthy are the results for the variables R&D in table 6. The thought behind including this variable was to control for domestic R&D activities that could contribute to export upgrading independently form FDI. The results might indicate that, as expected, a slightly positive relationship does exist, but to make any certain conclusion a more extensive dataset would be necessary. What is interesting when comparing model 1 to model 2 in table 6 is that the inclusion of domestic R&D seems to change the positive FDI coefficient into a negative one. As can be seen the number of observations does drop quite dramatically if R&D is included. When rerunning regression 1 with the same observations as in regression number 2 we can see that the differences do not stem from including R&D but rather from a reduced data set. This led to the exclusion of R&D of the regressions presented in table 2 and 3.
4.1. Limitations and further research
The main advantage of the model used in this paper is at the same time its biggest weakness: its simplicity in terms of how many and which variables are used. This was done to guarantee a broad and deep sample. The drawback is that sufficient data for most countries is only available for a restricted number of variables and a quite low explanatory power of the model.
Paul Berger s2155362 30 would be interesting to deepen the analysis in terms of which forms of upgrading are related to FDI. A similar advantage as well as drawback can be seen when looking at the measure of social capabilities. As has been described above in this paper I use mean years of education as a variable which can only capture a part of human capital which in turn is only a part of social capability. A more sophisticated measure, such as the one constructed in Kemeny (2010) would have caught different dimensions of capabilities. Again the measures used in this paper ensure a broad sample. Also a more detailed variable for inward FDI (e.g. groups according to the technology intensity of the targeted industries) would have limited the sample of countries significantly.
A suggestion for further research could be to conduct an analysis in this field on a more sophisticated data and variables structure. A further interesting feature to go into in more detail could be the impact on FDI on differe nt kinds of upgrading in the light of social capabilities. Also a more thorough analysis of the differences between ‘rich’ and ‘poor’ countries and the impact on the interaction between education and FDI could be of interest.
4.2. Conclusion
As discussed in the literature review there are good reasons to believe that FDI results in greater benefits than domestic investments; e.g.: Aitken (1997), or Ford et al. (2008). We can expect FDI to lead to increased technology transfers, increased competitive pressures, changes in trade patterns as well as export structures.
Paul Berger s2155362 31 middle and low income countries on the one hand and upper middle and high income countries on the other hand the results fit the expectations quite well: the results show that for poorer countries the ME remains negative up to 5.5 years of mean education for the male population. After this threshold the ME is positive for every group of the population. Almost all poorer countries reach this threshold. This is in line with what one could have expected form Gausselman et al. (2011) who state that FDI seeks a highly educated workforce paired with low wages. On the other hand the findings for richer countries are not what one could have expected form Kemeny (2010) since the ME is downward sloping. So in those countries given a certain level of education (around 10 years) the ME is actually negative.
Paul Berger s2155362 32
5. Appendix
5.1. List of all countries in the sample
Paul Berger s2155362 33
5.2. List of Variables
Variable Description Database URL
EDUFEM
Mean years of total education for the female population, IIASA/VID Projection Education Statistics http://databank.worldbank.org/ data/views/variableSelection/s electvariables.aspx?source=ed ucation-statistics-~-all-indicators EDUMALE
Mean years of total education for the male population, IIASA/VID Projection Education Statistics http://databank.worldbank.org/ data/views/variableSelection/s electvariables.aspx?source=ed ucation-statistics-~-all-indicators EDUTOT
Mean years of total education for the total population, IIASA/VID Projection Education Statistics http://databank.worldbank.org/ data/views/variableSelection/s electvariables.aspx?source=ed ucation-statistics-~-all-indicators EXP
High tech exports as percent of manufacturing exports World Development Indicators http://databank.worldbank.org/ data/views/variableSelection/s electvariables.aspx?source=wo rld-development-indicators
FDIGDP FDI as percent of GDP
World Development Indicators http://databank.worldbank.org/ data/views/variableSelection/s electvariables.aspx?source=wo rld-development-indicators
GDPCAP GDP per capita in current US Dollars World Development Indicators http://databank.worldbank.org/ data/views/variableSelection/s electvariables.aspx?source=wo rld-development-indicators GNI per Capita
GNI per capita in current US Dollars (Atlas method)
Paul Berger s2155362 34
5.3. Leverage versus residual squared plot
Figure 8, Leverage versus residual squared plot.
5.4. Breush and Pagan Lagrangian multiplier test for ransom effects
Var : Log_ Expor ts 2.9113
e 0.8144
u 2.0188
chibar2 5483.37 Prob. > chi bar2 0.000
8 88 8 8888889 98888888 9 9111199999 9 9 9 11 11 11 111111 1112121212121212121212111211 12 12 121212 12 1219191919191212121219 16 19 19 19 24 24 24242424 242424 24 2427 27 27 27 2727292929292929272727292727292927292727272727 2929 29292930 3030 303030 3030 30 30 32 34 34343434 34 3434 34 34 3434 3434343434343434 3440 404034404040 40414242414242424242414242424141414141414141414141414141 4242424242 464848 46 4646464646464646 46 48484848484848484848 48484849 494949 494949 49 51 5151 5151515251515152515151 51 5151 52 52 52 52 52 52525353535252525252525252 53 5353 5353 53 53 53 535353 535353535353 57 5757 5757575757575757 5759 585858 5858 58585858 5858 5858 59 59625962626262626262626262626362 62 63 636767666666666666666666 6666 6767676767676767676767676767676767676767697272726969 7272727272747272727272727272727472747272727274 73 7474747474747474 78 78 7878788383848483 83 83 84 84 84 8484 84 84 84 84 84 85 85 8585 85858585858585858585 85 85 8585 87 87 87 87 87 878787898787878787889187878787879187919189899191898988 89919191889191919188919191919191919191 94 94 94 949494 94 94 94 94 949494 94 94 94 949494949494 96 96 96 96 96 96 96 96 96 96 96 969696969696969696 97 9798 97 98991021021029810210210298102102102102102102102102989810210298989810210298 9898 9898 102102102102 105105 105 105105 105105105105105105105 105 107107107 107107107107107107107107107 107 107 107107 107 113113113113113113113113113113113113 113113113113113 114114 114 114 114 114 114 114 115 115115115115115 115 115 115 116117116117116116116116116116 116116116 117117117117 117 117117117117117117117 117 117 117 119 119 119119119119119119119119 121 122 122 122 122 122122 122 122 122 125 125125125125 125 125 125125130130 130130132132132126132132130126126126126126126126126126126126126126126126126 130130 130 130 130 130 133 133 133 137 137137137137137137137139137137137137139137137137137137139139137137137137139137 139139 139139 139139139140144144 142 140 144 144 144144144 144 144 144 144144 144 144 144 146 146 146 146 146 146 146146 146 148148148148148148148148148148148148148148148148 150150150150151151150150 151 151 151 151 151151153151151153153153153153153153153153153153153153153 152152 152152152152 153 153 154154154154154 154 154154154154154154154 154154154157154157 157157157157157157 157157157157157157157157 157158158157158158158158157158158158 158 158158158 158 158 163 163 163163 163163 163 168 168168168168168 168 168 168 168168 168 168 168 168 168 168170170168170170 170 170170170170 170 170 170170170 170 170 171171171171171171171171171171171171171171171171171 174174 174174 174 174 174174176176174176176176176 176176176176 176 176 176 176176186176177 177176186185185185185176176185185185177185185185185185177185185 186186189192 196196196196196196196196 190190190190190190190190190190190190190190190 196 196196196201200 200201200201201201201200 200200200196196196200196196196196196200 200 201201204204203204203204204204204201204204204204204203204204204204203203203203203201203203203203203201201203203203203201203203203203 204 204204205205205205205205205205205205205 205 205209213213 213213213213213 0 .1 .2 .3 .4 L e ve ra g e 0 .005 .01 .015 .02
Paul Berger s2155362 35
5.5. Haus man test
H0: The d ifference in coefficients is not systematic Coe fficients:
FE RE Difference
FDI -0.0110 -0.01156 0.0005
Paul Berger s2155362 36
5.8. Breusch-Pagan / Cook-Weisberg test for heteroskedasticity H0: Constant variance
Chi2 141.57
Paul Berger s2155362 37
5.9. Sources
Abramovitz, M. 1986. Catching up, forging ahead, and falling behind. The Journal of Economic History, 46(2): 385–406.
Aitken, B., Hanson, G., & Harrison, A. 1997. Spillovers, Foreign Investment, and Export Behaviour. Journal of International Economics, 43: 103–32.
Anyanwu, J. 2012. Why Does Foreign Investment Go Where It Goes?: New Evidence From African Countries. Annals of Economics and Finance, 13(2): 425–462. Baldwin, R. 2006. Globalisation: the great unbundling(s). Prime Minister’s Office
Economic Council of Finland.
Baldwin, R. 2011. Trade and Industrialisation after Globalisation’s 2nd Unbundling: How building and joining a Supply Chain are different and why it mat ters. NBER Working Paper Series, Cambridge, MA.
Bandick, R. & Hansson, P. 2009. Inward FDI and demand for skills in manufacturing firms in Sweden. Review of World Economics, 145: 111–131.
Barrell, R., & Pain, N. 1997. Foreign Direct Investment, Technological Change, and Economic Growth within Europe. Economic Journal, 107(445): 1770–1786. Barro, R., & Lee, J-W. 1994. Sources of economic growth. Carnegie Rochester
Conference Series on Public Policy, 40: 1–46.
Borensztein, E., De Gregorio, J., & Lee, L.-W. 1998. How does Foreign Direct Investment affect Economic Growth? Journal of International Economics, 45(1): 1115–1135.
Braunstein, E., & Brenner, M. 2007. Foreign Direct Investment and Gendered Wages in Urban China. Feminist Economics, 13(3-4): 213-237.
Dedrick, J., Kraemer, K. & Linden, G. 2009. Who profits from innovation in global value chains?: a study of the iPod and notebook PCs. Industrial and Corporate Change, 19(1): 81–116.
Doraisami, A., & Leng, G. 1996. Foreign Direct Investment and Economic Growth: Some Time Series Evidence of the Malaysian Experience. Asian Economies, 25(3): 45–54.
Fagerberg, J. 1994. Technology and international differences in growth rates. Journal of Economic Literature, 32(3), 1147–1175.
