DO FDI DETERMINANTS AT AN INDUSTRY LEVEL DIFFER IN SOUTH-EAST ASIA?

(1)

DO FDI DETERMINANTS AT AN INDUSTRY

LEVEL DIFFER IN SOUTH-EAST ASIA?

by

Robin de Rooij

University of Groningen

Faculty of Economics and Business

Msc International Business & Management

April, 2011

Laan van Meerdervoort 428 2563 BE Den Haag

06 400 99 997 Robin.de.Rooij@gmail.com

(2)

ABSTRACT

This study aims to understand to what extent the determinants of United States FDI in South-East Asia differ at a sectoral level and if these differences are in line with those found in previous literature. The analysis focuses on manufacturing industries, classified according to OECD guidelines. Results show that determinants of FDI differ between sectors in ways that are largely in line with previous research. However, differences exist in terms of the effect of labor.

(3)

1 INTRODUCTION

In a quest to understand the world around us, researchers are constantly trying to model human behavior. Creating a model that accurately describes the patterns that can be seen in Foreign Direct Investment is only one of many examples. Since the majority of FDI occurs between developed countries, consequently the majority of research dealing with FDI has focused on these flows. One of the biggest strands of literature in this area deals with finding the determinants of FDI for which a consensus on the most important variables seems to have been found (Wheeler & Mody, 1992; Bloningen 2005).

However, understanding FDI flows to developing and transitioning economies provides a different challenge for researchers, due to their unique circumstances and the different variables at play. On the one hand, there is a trade-off between risks when investing in developing countries and the long-term potential in terms of market growth or short-term advantages of lower costs. As a result, in FDI research in developing countries, developing institutional quality is generally found to be more important (UNCTAD, 2000; Nonnemberg & Mendonça, 2004).

On the other hand, the stakes are thought to be higher for the host countries, as spillover effects from FDI are critical for their development towards a strong market economy. Although some research contradicts this conclusion (Hanson, 2001), under certain conditions, the positive spillover effects of FDI for transition economies are clear (Schoors & Tol, 2002; Alfaro, 2003; Crespo & Fontoura; 2006). One theory related to this area is that spillover effects are greater for sectors with higher technological complexity than for sectors with lower technological complexity as there is more potential for the transfer of knowledge.

(4)

finds that several differences exist at the sectoral level of FDI behavior for CEECs. The main differences identified by Resmini are:

1. The responsiveness of FDI to market size and openness is stronger in traditional low technology sectors;

2. A country’s development has a stronger influence on science-based and scale-intensive sectors;

3. Wage differentials are more important in the scale-intensive sectors;

Another major region of transitional economies is South-East Asia. In this region, just as in Eastern Europe, it is commonly believed that FDI has helped countries in this region rapidly develop into the economic powerhouses they are today. However, great differences also exist between the two regions. Apart from the obviously large cultural difference between the two, Asia is also unique in the extent of its highly efficient fragmentation of production and distribution networks (Thorbecke & Yoshitomi, 2006). This unique fragmentation may cause the differences found by Resmini to no longer be relevant as each country may be executing a different task in the process while still being active in the same sector. To test if the three findings mentioned above also hold for the Asian region, this research will add to the work done by Resmini, albeit with different data sources. The countries treated are: China, Hong Kong, Taiwan, Singapore, South-Korea, Thailand, Malaysia, the Philippines and Indonesia.

(5)

2 LITERATURE REVIEW

In this section, the existing empirical literature dealing with FDI determinants will be covered. Findings in this section will be used to determine which variables to include in the model to determine FDI in Asia. Firstly the variables proven to be of influence in general literature are considered. Secondly, literature on FDI in the Asian market is used to confirm if any specific variables differ for this region. Lastly, several articles that consider differences in determinants of FDI between industries are discussed.

Several different strands of literature have attempted to predict FDI based on a fixed set of determinants. On the one hand, there are those that look at the micro-level; since investment decisions differ for each company individually. Another group chooses to take a more macro-economic approach and considers total FDI flows to certain countries.

The first group considers firms individually and attempts to find all the determining factors in the investment decision. This is the ideal way of looking at the issue of where to invest. After all, all firms have their own specific capabilities, resources and needs, as well as different strategies. These capabilities, resources and strategies may call for different environments to invest in, which is why comparing the companies’

characteristics to country characteristics provides unique investment decisions each time.

(6)

A third method includes using micro-economic data such as industry and firm size as well as macro-economic data. The difference with the first method is that the micro-economic data in this case is used to create groups or clusters of companies that are then analyzed together. This way the FDI behavior of groups of companies, for instance in a certain industry, can be compared, often using macro-economic determinants. Instead of analyzing the investment decision on a case-by-case basis such as is the case in the first method; groups of companies are assumed to behave in a similar fashion. On the other hand, this method still allows for some differences to be found between different industries, as opposed to the second method mentioned. More on this strand of literature will follow later.

General FDI determinants

Over the years, many researchers created models that explain FDI flows based on a set of macro-economic determinants. Many of the determinants have been defined and consensus exists about which are the most important, but for some determinants that are not always found to be of influence, uncertainty still exists.

Wheeler and Mody (1992) for instance discuss international ‘location tournaments’ in which governments compete for foreign investments. They state that once companies have invested in a certain location, agglomeration effects such as better infrastructure and industrialization will make this location more and more attractive for future investors. Testing the determinants labor cost, market size, tax rates, infrastructure, industrialization, previous investments, relationship with neighboring countries, country risk and country openness, they find that agglomeration effects have a strong influence on FDI levels. They also find support for the importance of labor cost, market size and country openness.

(7)

Bloningen (2005) provides us with a more recent summary of the empirical literature on macroeconomic FDI determinants. He combines theoretical with empirical literature to form a comprehensive overview of recent research. He finds that the most important determinants are exchange rates, corruption, trade protection and trade effects. And even for these determinants, exceptions exist in which they are not found to matter. No clear empirical evidence is found on institutional variables and taxes.

Benassy-Quere, Coupet & Mayer (2005) provide a summary on literature dealing with institutional determinants of FDI. They go into much more detail on institutional determinants than Bloningen. Although it has long been difficult to determine the causal relationship between wealth, development and institutional quality, Benassy-Quere et al. show that recent literature points to the fact that better institutions lead to economic growth. They continue showing that institutions may be important determinants for FDI for several reasons. Firstly, by raising productivity prospects, good governance may attract FDI. Secondly, poor institutions bring additional costs to foreign investors. Thirdly, FDI is very vulnerable to uncertainty, due to the high sunk costs. Recent developments in institutional data and the further development of gravity models in FDI literature allow Benassy-Quere et al. (2005) to construct a gravity model including institutional distance between countries. They find that institutions matter regardless of wealth, so institutional variables are important determinants of FDI.

(8)

However, many of the researches still contradict each other to some extent about other variables. A simple explanation for this may lie in the inherent variations of FDI. FDI occurs in different countries, in different industries, and for different reasons, thus making it extremely difficult to explain total FDI for all countries based on a fixed set of determinants.

FDI determinants in Asia

Instead of trying to explain all FDI for all countries in one model, it may be useful to only consider FDI determinants that have been found to be of influence for the Asian region in previous studies, which is the region that this research will be focused on. Findings may be more robust for only a specific region and it may help complete our list of determinants tested in this study.

One such research that focuses only on Asia is the article by Kinoshita (1998); he uses data from a survey of 173 Japanese manufacturing firms choosing whether or not to invest in China, India, Indonesia, Malaysia, the Philippines, Thailand and Vietnam. He uses both micro and macro-level data and finds that rivalry, market size, infrastructure and policy environment all influence FDI, whereas labor cost is not found to be of influence. Distinguishing between small and large companies, Kinoshita finds that for small companies, low wages and good infrastructure are more important, whereas for large companies, market-seeking and rivalry are more important.

(9)

Quazi (2004) tests for determinants of FDI for South Asia specifically and finds that economic freedom, economic openness, economic prosperity, human capital and incremental lagged changes in FDI significantly increase FDI inflows, whereas political instability lowers it considerably.

Much of the recent literature on FDI in Asia has been focused on the giant attracting huge amounts of FDI; China. Because of its low cost labor and large potential market (Hong & Chen, 2001), it is considered to be one of the most favorable countries in the region to invest in. Foreign investors are also attracted to China because of the possibility to use internal markets to set up a network of international production. The disadvantage of only researching one country at a time lies in the fact that it reduces comparability between countries. The determinants are no longer considered to be the only factors explaining FDI, as the author assumes that FDI differs between countries for reasons not included in the determinants. However, this can also be considered an advantage, since it may be considered impossible to create a model that incorporates all determinants of FDI. Some countries may attract more FDI due to factors that have not been researched yet or because of factors that are impossible or very difficult to measure. By only considering one country at a time, one can consider the effect the individual determinants have over time, given the country, so results should be more robust.

(10)

Zhang (2001) finds that market size, liberalized FDI regime and improving infrastructure attract FDI. Sun, Tong, Yu (2002) look at the determinants of FDI in China over a time period from 1986 to 1998, they compare FDI determinants for different regions before and after 1991. They find that labor quality, infrastructure, political stability and openness to the foreign world are important determinants of FDI. They also note that wage has a positive relationship before 1991, but a negative one after 1991, similarly, GDP showed no significant relationship before 1991, but is positively related after 1991, indicating an increase in market-seeking investments.

Comparing these results to determinants in general, some determinants are once again the same, with the exception that country risk is not of influence. However, it must be noted that risk was often captured to some extent in the political quality measures in these studies. 1. Market size 2. Political quality 3. Labor cost 4. Infrastructure 5. Openness

FDI determinants per sector

Although research in the area of FDI determinants is scarce, several authors have attempted to create similar models that include different determinants between industries. Early research by Broadman & Sun (1997) indicates that determinants for FDI in various sectors may differ, although they do not test this empirically.

(11)

In the model, Resmini takes into account both macro-economic and institutional variables. Specifically: GDP per capita, population, distance, labor costs, risk, openness and manufacturing industry size. As mentioned in the introduction, she finds that the responsiveness of FDI to market variables and openness is stronger in traditional sectors, such as the manufacturing of food products and metal fabrication. Similarly, the country’s progress towards a market economy has a stronger influence on science-based and capital- or scale-intensive sectors such as manufacture of machinery, electrical machinery, etc. Wage differentials are found to be more important in the scale-intensive sectors.

Overall, Resmini’s model shows very strong fit, indicating that determinants do differ considerable between sectors. Unfortunately, the number of observations is too low to draw reliable conclusions for some of the sectors tested, especially considering the number of determinants included, increasing the need for additional research.

Eiras & Prado (2007) consider capital intensity and technology content of FDI in various sectors. They analyze sectoral industry data on U.S. investments abroad from the Bureau of Economic Analysis (BEA). They only consider institutional variables and find that better protection of property rights cause higher FDI in R&D, but not on capital intensive investment. Increased worker bargaining power has a negative impact on both R&D and capital intensive FDI. However, they fail to include other potentially important determinants of FDI to determine the strength of the effect of property rights compared to other measures.

Overall there is some evidence that differences in determinants exist between sectors in which FDI occurs. But further research is needed to determine if findings are robust and consistent across regions and studies.

Implications

(12)

FDI to Asia, most of the determinants that will be considered coincide with those found to be of influence for Asia before, with the exception that a measure on education is added, which is identified as an important measure for developing countries by UNCTAD.

The determinants treated are as follows: 1. Market size

2. Political quality & risk 3. Labor cost

4. Infrastructure 5. Openness 6. Education

The exact data used in this research to represent the determinants above will be explained in the data section. To find the relationship between these determinants to FDI levels across sectors, a regression analysis will be applied. More about the regression analysis will follow in the next section.

3 METHODOLOGY

(13)

The important thing to note is that the effect on FDI of each of the determinants identified from previous research needs to be estimated in parallel to each other. Regression analysis is one of the most well-known approaches to estimate the type of model described above. It has some inherent strengths and limitations that we will discuss here.

Regression analysis

Regression analysis is a form of correlation analysis in which one variable is considered to be the dependent variable and one or more other variables independent variables. In other words, the behavior of one variable is explained based on one or several other variables. Regression analysis therefore assumes – but cannot prove – a causal relationship. Since the goal in this study is not to prove causality directly, but rather to identify whether a difference exists between industries, causality is assumed as defined and proven in previous research dealing with FDI determinants.

As shown in the Literature section of this research, many determinants that may impact FDI have been identified already, but not all are always found to be significant. To determine which determinants truly impact FDI in this case, a stepwise approach is used. In a stepwise regression, variables can be added or removed from the model to test whether the model is improved by including or excluding it. This stepwise approach is mainly important because the goal is to compare determinants across sectors. If a determinants is not found to be of influence in one sector, but is found to be of influence in another sector, this could prove to be an important finding.

(14)

Requirements and Limitations

When the sample size is small and the number of independent variables is large, stepwise regression may lead to models that seem to be both highly reliable and highly significant. However, these results can occur simply because of the large number of variables available. There will always be some variables that predict the behavior of the dependent variable well. To avoid this problem, the rule of thumb is to always have around 20 times as many observations as there are independent variables. With the dataset containing 8 independent variables, this means a minimum number of 160 observations are needed. However, as can be seen in the Data collection section, collecting 160 unique observations with data for all of the key determinants has turned out to be outside the scope of this research.

Another requirement for effective regression analysis is that correlation between included variables cannot be too high. If variables are greatly correlated this will lead to multicollinearity issues, making it impossible to see which of the independent variables affects the dependent variable. As will be discussed in the data considerations section, applying a factor analysis solves both of these problems, making the results of the regression analysis more reliable.

Approach

In this research, all the variables that are included were found to be of influence in previous literature, so the initial assumption will be that all variables (or factors) will be of influence in the model. If the R2 is low, the model may be improved by removing variables, applying backward elimination. When variables with low individual significance levels are removed, the model may be improved in terms of its R2 and overall significance levels. This is a process of trial and error that will be explained in more detail in the results section.

(15)

4 DATA

Data collection is done in the following steps: Firstly, collection of FDI data is discussed; afterwards, the collection of data for the included determinants is covered, then the different steps used to make the data suitable for regression analysis are discussed.

FDI data

Obtaining FDI data that distinguishes between investments in different industries is crucial for this research, but world-wide FDI data at this level of detail has proven elusive. As an alternative, The Bureau of Economic Analysis (BEA) of the US Department of Commerce keeps a detailed account of US investment for each year.

US-based FDI data is available for both FDI flows and FDI stocks. FDI stock refer to the overall value of capital held abroad by companies, whereas FDI flow refers to capital transferred by a foreign direct investor to an FDI enterprise in a given year. Although FDI flows are the most commonly used measure of FDI in this type of research, FDI flows from the BEA show very erratic behavior for some countries, as individual investments into specific countries have a large impact on the figures.

This type of erratic data makes it difficult to estimate reliable models based on regression analysis as the effects of determinants are clouded out by ‘random’ elements. For this reason, this study makes use of FDI stock figures instead. Although this approach is not unprecedented (Benassy-Quere, Coupet & Mayer, 2007), it is noted that this reduces comparability with other studies, including that of Resmini (2000).

(16)

Due to time constraints, it is unfortunately not possible to determine separate models for each of the industries for which FDI data is provided by the BEA. Instead, only Manufacturing industries will be taken into account. In order to maximize the extent to which it is possible to generalize results and compare outcomes to existing research, a split in manufacturing industries in terms of their technological complexity is made.

TABLE 1

BEA Industries for FDI flows and stocks Primary sector

Petroleum Mining

Utilities

Manufacturing

Food and kindred products Chemicals and allied products Primary and fabricated metals Industrial machinery and equipment Transportation equipment

Electronic and other electric equipment Computer and electric products

Electrical equipment, appliances and components Other manufacturing

Services

Wholesale trade Information

Depository institutions Finance, insurance, real estate Holding companies (non-bank) Other services

Other industries

(Source: Bureau of Economic Analysis)

Section on the right indicates that one category was replaced by two separate categories between 1994 and 2006

(17)

intensity, as can be seen in the first column of table 2 on next page. This split allows for more granularity technological intensity than the NACE approach used by Resmini (2000).

Although the industries provided by the BEA do not fit the ISIC categories exactly, some overlap can be seen. To get a clear picture of how to use the OECD classification to group the BEA industries, the first step is to remove all non-manufacturing industries, as well as ‘other manufacturing’ and ‘other industries’, as these cannot be grouped in technological complexity. The remaining industries are shown in table 2.

Starting at the top of the list, is the ‘Electronic and other electric equipment’ industry. For the years 1999 through 2006, the industries ‘Computer and electric products’ and ‘Electric equipment, appliances and components’ are used (see table 1 in the Data collection section), which would make a distinction between medium-high-technology and high-technology sectors possible. However, since this distinction is not made for all years, all electric and electronic activities must be grouped in the medium-high technology sector.

TABLE 2

Grouping industries into categories

OECD categories BEA Industries Average BEA FDI

High-technology industries

Aircraft and spacecraft Pharmaceuticals

Office, accounting and computing machinery Radio, TV and communications equipment Medical, precision and optical instruments

Medium-high technology industries

Electrical machinery and apparatus Electronic and other electric equipment 1.970.000 Motor vehicles, trailers and semi-trailers

Chemicals excluding pharmaceuticals Chemicals and allied products

(18)

Medium-low technology industries

Building and repairing of ships and boats Rubber and plastics products

Coke, refined petroleum products and nuclear fuel Other non-metallic mineral products

Basic metals and fabricated metal products Primary and fabricated metals 84.000

Low-technology industries

Manufacturing, recycling

Wood, pulp, paper, paper products and printing

Food products, beverages and tobacco Food and kindred products 154.000 Textiles, textile products, leather and footwear

(Source: OECD Science, Technology and Industry Scoreboard 2005)

The ‘Chemicals and allied products’ category in the BEA database includes products ranging from pharmaceuticals to plastics, fuel and other non-mineral products (BEA). As such, it includes operations in high-technology, high-technology and medium-low-technology industries, making it difficult to classify it in any of the OECD groups. Since no classification can be made, the category is excluded from the sample.

The ‘Transportation equipment’ category coincides with the OECD category ‘Railroad equipment and transport equipment’ and can be grouped in the medium-high technology sector. ‘Industrial machinery and equipment’ is also part of the medium-high-technology sector, since it matches the ‘Machinery and equipment’ category of the OECD.

The ‘Primary and fabricated metals’ industry is part of the medium-low technology sector, since this category corresponds with the ‘Basic metals and fabricated metal products’ category in the OECD classification.

The category at the bottom of the list ‘Food and kindred products’ should be grouped in the low-technology sector of the OECD classification, as it corresponds to the OECD ‘Food products, beverages and tobacco’ category.

(19)

whereas there is only one observation for both the low- and medium-low-technology industries each. Since having one group which is by far larger than others drastically reduces comparability, creating more evenly sized groups is desirable.

The first step to create more similar groups is to group the low- and medium-low-technology industries together, creating a larger medium-low-medium-low-technology sector. These are the smallest industries in terms of the FDI they attract (see the column on the right in table 2).

As a next step, a separate group consisting of only electric and electronic products is created. Not only is the electronics industry the biggest of the industries taken into account, it has been identified in previous literature as one of the key drivers of development in the region, warranting some special attention. Furthermore, it is the only industry considered in this research that consists of medium-high-technology as well as high-technology activities. The activities may therefore be more technology-intensive than some of the other medium-high-technology industries included.

TABLE 3

Final classification of BEA industries based on OECD guidelines Electronics and other electric equipment Electronic and other electric equipment

Medium-high-technology industries Industrial machinery and equipment Transportation equipment

Medium-low-technology industries Primary and fabricated metals Food and kindred products

(20)

The next step is to identify the exact measures for determinants to use to explain the differences in FDI levels across countries and years. Just as the FDI data was put together for the single purpose of this research, data on FDI determinants was also gathered exclusively for this research.

Determinants

The data used to represent the determinants of FDI as discussed and defined in the Literature Review section is discussed here. Each of the following determinants is discussed in turn.

1. Market size

2. Political quality & risk 3. Labor cost

4. Infrastructure 5. Openness 6. Education

Market size. To measure total market size, three indicators will be used. Both the indicator GDP and population have been used to reflect market size. The data for both of these measures has been obtained from Datastream, where time series data from the Economist Intelligence Unit was available from 1987 until 2007, with estimates until 2030. All GDP data was provided in US dollars. The indicator GDP per capita is used, for which data was also readily available on Datastream for the period 1987 until 2007, provided by the Economist Intelligence Unit in US dollars.

(21)

Labor cost. Another determinant used is wage, which is expected to be important for labor-intensive manufacturing processes with low skill requirements. Wage data was obtained from the International Labour Organization’s database of labor statistics called LABORSTA. This website provides data for most of the countries in question for 1994 – 2006. Some data was missing for Malaysia, the Philippines and Indonesia. The available amount of data for the Philippines was acceptable, for Malaysia and Indonesia data could be supplemented with wage data from their national statistics offices.

Infrastructure. The infrastructure index combines measures on physical infrastructure and communications infrastructure. Three factors are taken into account: Road density, phone subscriptions per 100 inhabitants (both mobile and fixed), and number of internet users per 100 inhabitants. This approximation is roughly in line with measures used by Calderon & Chong (2004), using some different measures due to data limitations. The road density variable is calculated as follows: Total road network length in km / total country surface in km2 = road density variable. Some road density data is available from the IRF World Road Statistics 2006, as published by the International Road Federation (IRF). Data is available from 1999 until 2004, but for many countries data is not present for all of these years. Since the calculation of the road density indicator is very straightforward, the gaps could be easily filled by finding more data on total road lengths and country surface. Data on country surface was obtained from the CIA World Factbook. All data on total road lengths was obtained from national statistics bureaus and ministries of transportation.

(22)

Public Works and Highways and data was received for the years 1991 to 2006. For Korea, no additional statistics on road length were obtained. Malaysia’s Yearbook of Statistics contains road data for the years 2001 until 2005. For Indonesia, detailed road data is available from the National Statistics office for 1987 until 2005.

Data on number of internet users per 100 inhabitants and the number of phone subscriptions per 100 inhabitants was obtained from the United Nations database (UNdata). The figures are originally from the International Telecommunication Union (ITU) and are available for all countries except Taiwan from 1994 until 2004. For Taiwan, data from the National Statistical Bureau is used instead. All of the data is combined into a common infrastructure index, in which each of the three measures have an equal weight.

Education. The indicator for education combines data on literacy rates of the population and enrollment levels at schools. Data on literacy and enrollment in primary and secondary education is combined in the Human Development Report’s Education index (United Nations). This indicator is compiled as follows: Education index = 1/3 * literacy rate + 1/3 * enrollment rate primary education + 1/3 * enrollment rate secondary education. Data is available for the years 1997 until 2004; however, data for Taiwan is not included. Thankfully, data could be easily complemented using data for Taiwan from the National Statistical Bureau for the years 1994 until 2006. The education index was calculated manually using these figures.

Openness. The indicator for openness is calculated by adding total imports and exports and dividing this figure by GDP. Import and export data in US$ was obtained from Datastream, supplied by the IMF International Financial Statistics. Data was available for 1990 until 2006.

(23)

for several of the determinants included, in the next section; the limitations of the dataset will be discussed, as well as the methods in which they are dealt with.

Dataset for regression analysis

In this section all the specific steps taken to evaluate the quality of the dataset, as well as solve any problems to the extent that this is possible are treated. Firstly, there cannot be too many missing values in the dataset, as this will reduce the reliability of the model. Secondly, it is important for linear regression analysis that data is normally distributed. Thirdly, correlation of the included variables is tested. Lastly, the factor analysis required to deal with correlation is covered.

Missing values. When looking at the dataset closely, several years do not meet our needs in terms of the number of missing values. For the years 2005 and 2006, all countries except Taiwan have three missing values: the Infrastructure Index, Education Index and Openness measure. For the year 1994, six countries are missing 2 values and the three other countries are missing three values. These years with multiple missing values for each country are removed from the dataset. After removing this data, the remaining dataset contains 90 observations, with data for 10 years (1995-2004) and 9 countries.

Normal distribution. To check for normal distribution, histograms for all the variables used are provided in appendix IV with a normal distribution curve shown for comparison. Because most of the determinants differ greatly between countries, the data is not normally distributed. Especially for GDP and population, some countries have very high values whereas others have very low values.

(24)

Ln (FDI) Ln (Population) Ln (FDI Electronics) Ln (Wage)

Ln (FDI Medium-high tech) Ln (Infrastructure Index) Ln (FDI Medium-low tech) Ln (Institutional Quality)

Ln (GDP) Ln (Education Index)

Ln (GDP per capita) Ln (Openness)

Correlation. The next step is to perform a correlation test for all the variables. If variables are greatly correlated this will lead to multicollinearity issues, affecting the reliability of the regression analysis. Correlation of variables in this case makes it impossible to see which of the independent variables affects the dependent variable. Even if relationships are found, it is impossible to tell which variable has the greatest impact. For example, if GDP grows in a certain period, but infrastructure quality grows at a comparable rate, it is impossible to judge by means of multiple regression analysis which was responsible for increases in FDI, or to which extent both were responsible.

(25)

Although the correlation in itself may not have been unexpected, it is still a problem for regression analysis. The best solution is to perform factor analysis, which combines several independent variables into a new variable or factor. For example, a new factor that represents the variables GDP and population, because these variables correlate, may in turn be interpreted as a measure for market size. Variables can be combined this way until there is no longer any correlation between the factors left. Although this method makes it impossible to say anything about the individual variable’s part of the factor as a whole, it does provide a good way to check which factors may be of influence.

Factor analysis. The most common method of factor analysis is known as “principal component analysis”. With this method, a new set of factors is created that account for most of the variance observed in the original set of variables. Each factor represents the existing variables by means of a linear relationship as follows:

Factor 1 = a1 * variable 1 + b1 * variable 2 + … + x1 * variable x, Factor 2 = a2 * variable 1 + b2 * variable 2 + … + x2 * variable x, etc.

When performing factor analysis, the first factor is always created to account for as much variance of the original variables as possible. The second factor is created to account for as much of the left over variance as possible, etc. However, because of this method, the first factor will usually have much higher loadings for all variables than the other factors, which will have low loadings for all variables. A more meaningful distribution of loadings however, is obtained if each factor represents several specific variables (high loadings for these variables and low loadings for other variables). This distribution can be obtained by performing “rotation”. Rotation helps redistribute the loadings in such a way that each variable loads onto as few factors as possible. In other words, variables will be represented in fewer factors, this helps in the interpretation of the factors themselves.

(26)

Including a fourth factor would add even less to the factor analysis. By performing rotation on the two factor solution, the interpretation of the factors can be considered; most variables are more strongly represented in one factor than in the other after rotation.

The first factor can be interpreted to represent wealth and standard of living, since the variables GDP per capita, wage, infrastructure and education are strongly represented. The second factor seems to represent measure market size, since GDP and population have high loadings. The fact that these are negative in this case does not have any major implications for the rest of the analysis. Openness is also reflected very strongly in the second variable, probably because GDP is included in this measure. This is a good interpretation to work with as different aspects of the investment decision are represented.

Adding the third factor adds an additional 8% to variance explained; which may be useful for a more complete model, but only if it can also be interpreted in a meaningful fashion. When performing rotation again with three factors, the results are even more pronounced than when using two factors.

The first two factors are similar to the factors identified earlier. Although in the first factor, education no longer has such a high loading, whereas openness is more important. The negative signs for the second factor have switched so a high value for factor 2 now represents a larger market instead of a smaller one. The third factor mainly has a very high loading on education, but it can also be noted that its loading on wage is higher than on the other variables. For interpretation, it can be said to deal with labor quality and costs. Such a measure is a potentially important determinant for investment decisions and there may be great differences between sectors. Overall, the model with three factors is preferred for this research, since more potential differences can be identified between industries without risking multicollinearity issues.

(27)

the models based on FDI, both at the sector level and at the aggregate level. The results of these analyses can be found in the next section.

5 RESULTS

First, the regression analysis with three factors will be applied to Total US FDI stocks. If the determinants used in the study are found to be of influence on the total market, it can be assumed that the correct determinants were taken into account. If the model as a whole can predict a considerable amount of the variance in the FDI levels, the model is considered to be accurate in predicting FDI.

If the model can be effectively applied to the total market, the next step is to consider specific sectors of electronics, other medium-high tech sectors and medium-low tech sectors. By comparing the outcomes for each of the sectors, differences between high-tech vs. low-high-tech activities may be found. In the conclusions section, a comparison will be made between the models to determine if results are consistent with earlier findings as provided in the introduction.

Regression Total US FDI

Below, the outcomes of the first regression analysis results are provided.

As can be seen in the first table, the R2 is quite high, as 58.8% of variance in FDI levels can be explained based on this model. The individual determinants are also found to be significant, with the exception of the measure for market size. Since the method applied in this paper is stepwise regression using backward elimination, the market size measure is excluded from the model to test if the model’s predictive power is negatively impacted.

Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate 1 ,767(a) ,588 ,573 ,45746

(28)

Coefficients(a) Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 (Constant) 9,091 ,049 184,159 ,000 Development Factor ,522 ,050 ,746 10,522 ,000

Market size Factor ,002 ,049 ,003 ,049 ,961

Labor Factor -,111 ,049 -,160 -2,259 ,027

a Dependent Variable: Ln (Total FDI stock)

The results of the regression analysis without the market size variable are shown below. Since the R2 is at a comparable level to the R2 in the model in which the market size variable was included, market size is assumed not to be of influence on total FDI stocks in the market. This would imply that market-seeking is not one of the main goals of FDI in East-Asia; the effects of Development and Labor are much stronger.

a Predictors: (Constant), Labor Factor, Development Factor

Coefficients(a) Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 (Constant) 9,091 ,049 185,300 ,000 Development Factor ,523 ,049 ,746 10,587 ,000 Labor Factor -,111 ,049 -,160 -2,273 ,026

(29)

infrastructure, wealth and institutional quality has a significant and positive effect on FDI stock.

The labor factor is negatively related to FDI, indicating that, all else equal, FDI is driven by low labor costs. Although one may expect that higher education would lead to higher FDI levels, the education levels are also partially included in the development factor. So for a similar level of development, the country with lower labor costs would attract more investment.

Although the market size factor is not found to be of influence on FDI for the total market, the other factors do influence FDI. The model as a whole seems to predict variance quite well. To consider if differences exist between the various sectors, the model can now be used for each of the sectors individually. Even though market size was not found to be of influence for the market as a whole, it will still be considered in each of the following models, as it may be of influence in some sectors, but not in others.

Regression Electronics sector

When performing the same test for only the electronics sector, results are different. Neither labor, nor market size are found to have a significant influence on FDI levels1. The fact that market size does not influence FDI levels is most likely because market seeking is very limited in this sector, just as was seen in the model used to consider the total market.

A possible explanation why labor is not found to be of influence either is that the electronics sector is more capital intensive than some of the other industries. As the effects of development and wealth are far more important than that of labor costs and quality, the labor quality measure does not significantly influence FDI in the electronics sector.

1

(30)

For capital intensive industries, the requirements in terms of capital are often much more than simply the availability of money. A developed banking system, proper infrastructure, the availability of other companies and industries to support the capital intensive company in its activities; all of these are necessary for capital intensive industries to function effectively. All of these requirements are incorporated to some extent in the development factor in this study.

Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate 1 ,563(a) ,317 ,308 1,11842

a Predictors: (Constant), Development Factor

a Dependent Variable: Ln (FDI stock in Electronics sector)

Regression Medium-high tech sector

(31)

a Predictors: (Constant), Market size Factor, Development Factor

Market size Factor ,665 ,136 ,437 4,908 ,000

a Dependent Variable: Ln (FDI stock in Med-high technology sector)

Regression Medium-low tech sector

As opposed to the other sectors, development is not found to have a significant influence on FDI levels for the Medium-low tech sector. This difference may be explained based on the lower complexity and capital intensity of low tech sectors as industries with lower complexity generally do not need as many supporting services and manufacturers. Market size is significantly and positively related to FDI, indicating that products may be more focused on local markets in this industry. The labor factor is also significantly related to FDI levels, as opposed to in the other two sectors. This is most likely due to the higher labor intensity in this sector. Although one may have expected the labor factor to be negatively related to FDI as in the total market model, the relationship found is positive.

(32)

Coefficients(a) Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 (Constant) 4,749 ,110 43,064 ,000

Labor Factor ,560 ,111 ,436 5,060 ,000

a Dependent Variable: Ln (FDI stock in Med-low technology sector)

A possible explanation for the difference in the sign of the labor factor between the total FDI model and the Medium-low tech FDI model can be found in the type of education measure included in this study. The measure included represents the general quality of education for a large part of the population, without taking into account very high levels of education. A very high level of education may not be needed for low tech sectors, and a variable measuring a high level of education more accurately might be found to be negatively related to low tech FDI. However, one could claim that companies may require a certain minimum level of education of their workforce or potentially business partners, such as literacy.

Summary and Comparison

Summarizing the results in the following table, development is important for all sectors except for the medium-low sector. Market size is not found to be of influence for the Electronics sector, as investment in this sector is mainly intended for export. Medium-low and medium-high technology sectors are more likely to be market seeking. Labor cost and education were only found to impact investments in the medium-low technology sector.

TABLE 4

Summary of models of FDI determinants for different sectors

Total Electronics Medium-high Medium-low

Development

+

Market Size

+

(33)

This research was able to use a regression model to explain levels of FDI in South East Asia based on a set of determinants, but also to create different models based on investment levels in different sectors. This is in line with results by Resmini (2000), who also found significant differences between sectors.

Below the findings will be compared at an individual level. Although there was a difference in the specification of industries for the two studies, overlap can be found between the two.

TABLE 5

Comparison of sector specification

This research Resmini

Electronics and other electric equipment Scale intensive Specialized suppliers Science based

Medium-high-technology industries Scale intensive Specialized suppliers

Medium-low-technology industries Traditional

Firstly, Resmini found that the responsiveness of FDI to market size and openness is stronger in traditional sectors. In this research, the traditional sectors are represented by the medium-low technology sectors, so findings here are consistent. However, in this research, medium-high technology sectors are also found to respond to market size.

Secondly, Resmini found that a country’s development has a stronger influence on science-based and scale intensive sectors. These conclusions also hold in this research, with development being positively related to investments in electronics and medium-high technology sectors.

(34)

6 CONCLUSIONS

Despite an approach that was different in many ways to that used by Resmini (2000), this research has been able to validate some of the findings as proposed in earlier research. Differences can be seen to exist between determinants of FDI in different sectors, with more technologically intensive sectors more dependent on development and less technologically intensive sectors more dependent on market size. However, differences were found in the effects of labor.

The main limitation of this research has been its limited dataset; although reliable US investment data was found for a number of different categories, these categories differed for some of the years and were difficult to match to sectors of differing technological intensity. Furthermore, FDI flow data was so erratic, it was impossible to use for the purposes of this research, limiting comparability with other studies.

An additional problem was the missing data of determinants for some of the countries. Future research may be performed on ever expanding datasets, including more (or different) countries as well as a longer time span, in order to obtain more reliable results.

(35)

A persisting problem when performing this type of research is the multicollinearity inherent to many of the determinants of investments. This study has attempted to deal with this multicollinearity by means of factor analysis. Another possible way to cope with these issues would be to select countries in which the relationship between the determinants is less clear (for instance where regulatory development is not as directly related to wealth), although these will most likely be difficult to find.

(36)

(37)

REFERENCES

Alfaro, Foreign Direct Investment and Growth: Does the Sector Matter? (2003) Banga, Impact of Government Policies and Investment Agreements on FDI Inflows to Developing Countries: An Empirical Evidence (2003)

Benassy-Quere, Coupet & Mayer, Institutional Determinants of Foreign Direct Investment (2005)

Benassy-Quere, Coupet, Mayer, Institutional Determinants of Foreign Direct Investment (2007)

Bloningen, A Review of the Empirical Literature on FDI Determinants (2005) Broadman & Sun, The distribution of foreign direct investment in China (1997) Bureau of Economic Analysis: http://www.bea.gov/interactive.htm

Cheng & Kwan, What are the determinants of the location of foreign direct investment? The Chinese experience (1999)

Crespo & Fontoura, Determinant Factors of FDI Spillovers – What Do We Really Know? (2006)

Coughlin and Segev, Foreign Direct Investment in China: A Spatial Econometric Study (1999)

Department of Land Transport, Ministry of Transport Thailand Department of Public Works and Highways Philippines

Economist Intelligence Unit: http://www.eiu.com/

Eiras & Prado, Determinants of Capital and R&D Intensive Foreign Direct Investment (2007)

Hanson, Should Countries Promote Foreign Direct Investment? (2001)

Highways Department Hong Kong: http://www.hyd.gov.hk/eng/home/index.htm

(38)

LABORSTA: http://laborsta.ilo.org/

Lim, Determinants of, and the Relation Between, Foreign Direct Investment and Growth: A Summary of the Recent Literature (2001)

Ministry of Transportation and Communication Taiwan:

http://www.motc.gov.tw/mocwebGIP/wSite/np?ctNode=199&mp=2

National bureau of statistics of China: http://www.stats.gov.cn/tjsj/ndsj/2006/indexeh.htm

Nonnemberg & Mendonça, The Determinants of Foreign Direct Investment in Developing Countries (2004)

OECD, OECD Science, Technology and Industry Scoreboard (2005) Quazi, Determinants of Foreign Direct Investment in South Asia (2004)

Resmini, The determinants of foreign direct investment in the CEECs, new evidence from sectoral patterns (2000)

Schoors & v.d. Tol, Foreign direct investment spillovers within and between sectors: Evidence from Hungarian data (2002)

Singapore Department of Statistics: www.singstat.gov.sg

Statistics Indonesia: http://www.bps.go.id/

Sun, Tong, Yu, Determinants of foreign direct investment across China (2002) The Department of Statistics Malaysia: http://www.statistics.gov.my/eng/ The Heritage Foundation: http://www.heritage.org

The IRF World Road Statistics 2006, IRF (2006) http://www.irfnet.org/

Thorbecke & Yoshitomi, Trade-FDI-Technology Linkages in East Asia (2006) UN data: http://data.un.org/

UNCTAD, 2000. The Competitiveness Challenge: Transnational Corporations and Industrial Restructuring in Developing Countries. United Nations, Geneva.

(39)

APPENDICES

(40)

(41)

(42)

(43)

(44)

(45)

(46)

(47)

(48)

(49)

(50)

(51)

(52)

(53)

(54)

(55)

Appendix VII: Factor analysis Results

(56)

(57)

Appendix VIII: Regression Results Total FDI stock, step 1: initial model

a Predictors: (Constant), Labor Factor, Market size Factor, Development Factor

Coefficients(a) Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 (Constant) 9,091 ,049 184,159 ,000 Development Factor _,522 _,050 _,746 _10,522 _,000

Market size Factor ,002 ,049 ,003 ,049 ,961

Labor Factor _-,111 _,049 _-,160 _-2,259 _,027

Total FDI stock, step 2: model excluding Market size

Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate 1 _,767(a) _,588 _,579 _,45470

Coefficients(a) Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 (Constant) 9,091 ,049 185,300 ,000 Development Factor ,523 ,049 ,746 10,587 ,000 Labor Factor _-,111 _,049 _-,160 _-2,273 _,026

(58)

Electronics FDI stock, step 1: initial model Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate 1 ,567(a) ,321 ,297 1,12785

Market size Factor -,040 ,120 -,030 -,334 ,740

Labor Factor _,083 _,121 _,062 _,689 _,493

Electronics FDI stock, step 2: model excluding Market size

Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate 1 _,566(a) _,320 _,304 _1,12187

Coefficients(a) Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 (Constant) _6,926 _,120 _57,522 _,000 Development Factor ,761 ,121 ,564 6,271 ,000 Labor Factor _,083 _,120 _,062 _,691 _,491

Electronics FDI stock, step 3: model excluding Market size and Labor

a Predictors: (Constant), Development Factor

(59)

Medium-High FDI stock, step 1: initial model Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate 1 ,662(a) ,438 ,414 1,20393

Coefficients(a) Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 (Constant) _5,229 _,142 _36,790 _,000 Development Factor ,815 ,142 ,511 5,723 ,000

Labor Factor _-,135 _,139 _-,086 _-,965 _,338

a Dependent Variable: Ln (FDI stock in Med-high technology sector)

Medium-High FDI stock, step 2: model excluding Labor

a Predictors: (Constant), Market size Factor, Development Factor

Coefficients(a) Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 (Constant) _5,216 _,141 _36,877 _,000 Development Factor ,822 ,142 ,515 5,783 ,000

Market size Factor _,665 _,136 _,437 _4,908 _,000

(60)

Medium-Low FDI stock, step 1: initial model Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate 1 ,619(a) ,383 ,361 1,02968

Coefficients(a) Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 (Constant) _4,747 _,110 _42,966 _,000 Development Factor -,093 ,111 -,073 -,843 ,402

Labor Factor ,562 ,111 ,437 5,066 ,000

a Dependent Variable: Ln (FDI stock in Med-low technology sector)

Medium-Low FDI stock, step 2: model excluding Labor

a Predictors: (Constant), Labor Factor, Market size Factor

Coefficients(a) Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 (Constant) 4,749 ,110 43,064 ,000

Labor Factor ,560 ,111 ,436 5,060 ,000