The Latent Heterogeneity in Investment Location Choices of Multinational Enterprises

(1)

DEPARTMENT OF ECONOMICS

DISCUSSION PAPER SERIES

The Latent Heterogeneity in Investment Location

Choices of Multinational Enterprises

Simona Rasciute and Eric J. Pentecost

WP 2008 - 16

Dept Economics

Loughborough University Loughborough

LE11 3TU United Kingdom Tel: + 44 (0) 1509 222701 Fax: + 44 (0) 1509 223910

(2)

The Latent Heterogeneity in Investment Location Choices of

Multinational Enterprises

Simona Rasciute1 and Eric J. Pentecost

Department of Economics Loughborough University Loughborough Leicestershire LE11 3TU UK December 2008 Abstract

The heterogeneity of investing firms is an important determinant of the distribution of foreign direct investment (FDI) location decisions. This paper, for the first time, explicitly allows for firms’ heterogeneity by using a latent class discrete choice model and a new multi-level data set to examine over 1100 individual firm FDI-location decisions over an 11-year period. The highly significant empirical results show that the responsiveness of the probabilities of choices to invest in a particular country location to country-level variables differs both across sectors and across firms of different sizes and profitability. Therefore, controlling for investing firms’ heterogeneity is important if robust estimates are to be obtained.

JEL classification Nos: F23, P33

Keywords: Latent Class model, firm heterogeneity, foreign direct investment, multi-level data

_____________________________________

Acknowledgement: The first named author is grateful to Professor William Greene for valuable advice on latent class modelling techniques.

1

(3)

1. Introduction

In 1989 when the Berlin Wall collapsed there was very little foreign direct investment (FDI) in Central and Eastern Europe, but more than 15 years later there is on average about 90 billion US dollars per annum flowing into the Central and Eastern European Countries (CEECs) and the total stock of FDI outstanding is about 643 billion US dollars in current prices. Consequently, there has been a significant increase in the empirical literature on the determinants and effects of FDI in CEECs (see, for example, Barrell and Pain, 1999; Carstensen and Toubal, 2004; and Benassy-Quere et al., 2007), although few empirical studies have specifically investigated the location choices of multinational enterprises (MNEs). The studies by Becker et al., (2005), Crozet et al., 2004) and Disdier and Mayer, (2004) that do attempt to investigate location choices rely either on the Multinomial logit (MNL) or the Nested logit (NL) models2. The MNL model, however, is subject to restrictive assumptions regarding the substitution patterns across alternative investment locations, while the NL model partially relaxes the independence from irrelevant alternatives (IIA) assumption in order to allow some substitution across alternative investment locations, neither of these models allow for heterogeneity between investing firms, which is potentially very important for the choice of investment location, as Nocke and Yeaple (2007) have demonstrated.

The principal contribution of this paper is therefore, to allow for source firm heterogeneity, by applying a Latent Class (LC) model to investigate investment location choices by MNEs in CEECs. The LC model allows for investing firms’ heterogeneity by the segmentation of investing firms into a predetermined number of classes for which parameters are estimated separately. The investment location choices of MNEs will not only depend on observed attributes, but also on latent heterogeneity that varies with unobserved factors. The LC model is superior to the Mixed logit (ML) model in that it does not require alternative distributions to be chosen for the random parameters, but instead captures investing firm heterogeneity with a discrete distribution. Furthermore, the majority of empirical literature that applies both models report that the LC model

2

The exception to this is Basile et al. (2008), who apply the Mixed logit model in the context of EU cohesion policy but, unfortunately, the interpretation of results and policy implications are based on the estimated coefficients. However, neither the sign nor the magnitude of the coefficients are informative and further estimation of elasticities and marginal effects is needed.

(4)

performs better than the ML model (see, for example, Greene and Hensher 2003 and Shen et al., 2006).

In addition, to be able to include investing firm heterogeneity a new data set has been constructed, which has investing firm characteristics, rather than just country-level and industry-level data. This multi-level data set, including firm, industry (or sector) and country effects has been complied from individual firm data supplied by the Bureau van Dijk, to simultaneously determine the firm-level FDI location decisions. The data set covers 1,108 FDI location choices of firms in the USA, the EU15, Norway, Switzerland, Russia and Japan into 13 Central and Eastern European Counties (CEECs) – the 12 recent EU member states, excluding Cyprus and Malta, but including Croatia, Russia and Ukraine - over an eleven year period from 1997 to 2007. The estimation results show that the investing firm’s characteristics have a significant role to play in the specific choice of investment location and therefore alternative locations tend to attract different types of FDI. It is therefore important to acknowledge that investing firm characteristics are possibly as important, if not more important, than the host country or industry characteristics of the investment receiving country.

The rest of the paper is set out as follows. Section 2 presents the LC model, Section 3 discusses the dataset, Section 4 describes the construction of explanatory variables used in the model and Section 5 presents the econometric results and policy implications. Finally, Section 6 concludes.

2. The Latent Class Model

The Latent Class (LC) model is a semi-parametric extension of the Mixed Logit model (random parameter model), which does not require the researcher to make specific assumptions about the distribution of random parameters across investing firms, as parameter heterogeneity across individual firms is modelled with a discrete distribution. The LC model approximates the unknown distribution of random coefficients by a finite number of mass points; therefore, simulation is not needed in the estimation process (Meijer and Rouwendal, 2006). Investing firms are implicitly divided into a number of classes Q, although it is not known which class contains a particular firm. Investing firm

(5)

behaviour depends on observable attributes and on latent heterogeneity that varies with factors that are unobserved (Greene and Hensher, 2003).

The profit parameters become class specific such that πisc|q = βqzisc + εisc|q, where

π is potential profit of foreign investment in country c, βq are class specific estimated profit parameters and z_isc, is a vector of country characteristics, which have different

values for firms investing in sector, s, and for firms of different characteristics. The probability that an investing firm i in class q chooses among C alternatives to locate its investment is given as:

(

)

(

)

∑

₌ ′ ′ = = = C c q isc isc q isc z z q class c Y 1exp exp ) Pr(

β

(1)

Within each class, the choice probabilities are assumed to be generated by the Multinomial logit (MNL) model.

The class membership, however, is not observed and class probabilities are specified by the MNL form. Therefore, the probability of investing firm i belonging to class q can be expressed as:

(

)

(

)

, 1,..., , 0, exp exp 1 = = ′ ′ =

∑

Q₌ Q q q i i q iq q Q h h H

θ

(2)

where hi denotes a set of observed investing firms’ characteristics, for example, investing firms size and profitability. The LC model estimates the probabilities of an investing firm belonging to each class and each investing firm is assigned to one of the classes on the basis of the largest probability. Due to the identification problem the Qth

parameter vector is normalised to zero, and, as a result, all other coefficients are interpreted relatively to the normalised class.

Combining the conditional choice equation (1) and membership classification equation (2), the joint probability that investing firm i belongs to class q and chooses alternative c can be written:

(6)

(

)

(

)

(

)

(

)













′













′

=

∑

₌ = = = Q q q i i q Q q Q q C c q isc isc q q ic iq ic

h

z

P

H

P

1 1 1 1

exp

θ

β

(3)

The parameter vectors βq and θq are simultaneously estimated by the maximum likelihood method and the log likelihood (LL) for the sample is defined as:

[

]

∑

₌ = ₌ ₌ = I i Q q iq icq I i Pi H P LL 1 1 1ln ln (4) The log likelihood is maximised with respect to the Q structural parameter vectors, βq, and the Q-1 latent class parameter vectors, θq. The issue in the estimation process is the choice of the number of classes, Q, as the comparison of the log likelihoods of models with a different number of classes is not appropriate. While increasing the number of classes increases the fit of the model, it may lead to some coefficients having very large standard errors. The trade-off between the goodness of fit and the precision of the parameter estimates can be found with the help of information criteria summarised by Shen and Saijo (2007), which could help determine the optimal number of classes, Q:

Akaike Information Criterion: AIC =−2

(

LL*_Q−K_Q

)

Akaike’s ρ2: ρ_Q2 =1−

[

AIC_Q

(

2LL₀

)

]

Bozdogan Akaike Information Criterion: AIC3=−2LL*_Q +3K_Q

Bayesian Information Criterion: BIC=−LL*_Q+

(

K_QlogN

)

2

where LLQ* is the log likelihood at convergence with Q classes, KQ is the number of

parameters in the model with Q classes, LL0 is the log likelihood of the sample with

equal choice probabilities, and N is the sample size. The “optimal” number of latent classes is indicated by the minimum (maximum) values of AIC, AIC3 and BIC (ρ_Q2).

(7)

3. The Data Set

The sample consists of 1,108 firm-level data observations on FDI location choices by firms of 20 market economies (EU15 countries, USA, Japan, Russia, Norway and Switzerland) to firms in 13 transition economies (12 new EU member states (except for Malta and Cyprus) plus Croatia, Russia and Ukraine) from 1997 to 2007. Most of the empirical literature on FDI focuses on Greenfield investment, excluding other entry modes such as mergers and acquisitions (M&A), joint ventures and institutional buy-outs (Brownfield investment), which are the most important in the CEECs (Head and Ries, 2008). For example, in 2005 the share of cross-border M&As in FDI was about 96 percent in Czech Republic, 84 percent in Estonia, 82 percent in Ukraine and 68 percent in Bulgaria (UNCTAD3 statistics)4. In order to test for the location determinants of Brownfield investment, the MNEs in the sample enter a foreign market via M&A, joint ventures or institutional buy-outs.

Of all 13 host CEECs in the sample, Poland has received the largest share of foreign capital allocations by MNEs to locate their investment (about 21 percent) and it was followed by Russia with approximately 17 percent of foreign investment location choices. Slovenia and Latvia, on the other hand, have received the smallest share foreign capital allocations (approximately two and three percent respectively). The two major investors in the CEECs in the sample are Finland and the UK with the shares of approximately 12 and 11 percent respectively. MNEs from Japan and Ireland were at the other end of the scale regarding investment location choices in the CEECs with approximately one percent each. The number of foreign capital allocations from each source country to each host country are summarised in Table 1.

The largest number of foreign capital allocation in the CEECs took place in the

traditional sectors (approximately 36 percent), followed by scale-intensive industries (about 24 percent) and service sectors (nearly 23 percent). Science-based industries have received the smallest share of FDI (approximately 18 percent) in the sample. However, when looking at the distribution of investment location choices among the four groups of

3

United Nations Conference on Trade and Development 4

It is difficult to estimate precisely what share of FDI flows is accounted for by cross-border mergers and acquisitions (M&A) because the values of cross-border M&As cannot be directly compared with FDI flows registered in the balance of payments.

(8)

industries in separate countries, traditional sectors have not necessarily attracted most foreign capital allocations. For example, in the Czech Republic, Hungary and Slovakia

scale-intensive industries have received the largest share of FDI (approximately 30 percent, 26 percent and 47 percent respectively), while in Estonia and Lithuania the

service sector has attracted most foreign capital allocations (about 35 percent and 33 percent respectively). The rest of the CEECs in the sample have the biggest share of foreign investment allocations in their traditional sectors.

The investing firms in the sample can be divided into four groups in respect to their size: small (up to EUR100,000 turnover), medium (from EUR100,000 to EUR1million turnover), large (from EUR1 million to EUR10 million turnover) and very large firms (above EUR10 million turnover). The largest share (49 percent) of very large firms in the sample have invested in scale-intensive industries, while service, traditional and science-based industries received approximately 20 percent, 18 percent and 12 percent of foreign capital allocations respectively by very large firms in the sample. Medium and large firms in the sample have selected traditional sectors to locate most of their investment (approximately 41 percent and 43 percent respectively). The second largest share of investment allocations by large investing firms has taken place in scale-intensive sectors (about 22 percent) and by medium firms in service sectors (approximately 26 percent). Small investing MNEs have chosen the traditional sector in which to locate the largest share of their investment (about 33 percent), while service

sector and science-based industries received about 26 percent and 21 percent of foreign capital allocations respectively. Scale-intensive industries have received the smallest amount (approximately 20 percent) of foreign capital allocations by small firms in the sample.

Investing firms in the sample can divided into three groups with respect to their profitability (proxied by earnings before interest and tax): firms that incur loss, firms that earn profit up to EUR50,000 and firms that earn profit of more than EUR50,000. Regardless of the investing firms’ profitability traditional sectors received most investment allocations: approximately 37 percent in the case of firms that incurred loss, about 39 percent in the case of firms that earned profit of up to EUR50,000 and about 36 percent in the case of firms that earned profit of more than EUR50,000.

(9)

4. Variable Specifications

Table 2 gives a summary of variable definitions and sources. The traditional determinants of FDI, are the market size of the host country, the cost of capital in the host country and the distance between investing and investment receiving countries. As Table 2 shows, market size is measured as the real GDP of the host country. The host country cost of capital is measured as the real discount (interest) rate. Both of these variables are expected to be positively associated with FDI inflows. Distance can be considered as a measure of the transaction costs of undertaking foreign activities, such as the costs of transport and communication, the costs of dealing with cultural and language differences, the costs of sending personnel abroad and the informational costs of institutional and legal factors, e.g., local property rights, regulations and tax systems. These kinds of costs are all assumed to increase with distance.

In addition to the above mentioned factors, three other country-specific factors are included in the empirical model: the national rate of unemployment and two dummy variables, one for European Union membership and another for the presence of a common border between the investing and the investment receiving country. A dummy variable for a common border between the source and the host country is included, as it is expected that the host country is more likely to be chosen to receive investment if it shares a border with the source country. Usually countries sharing the same border have similar cultures and language and stronger historical ties.

Countries that joined the EU by January 2007 had to satisfy the economic (market economy), political (democracy and human rights) and administrative (well-functioning institutions) criteria set at the Copenhagen European Council in 1993. The accession of a CEEC into the EU meant free trade with EU member states and the adoption of a Western type business and legal environment, which provided foreign investors with confidence in the completion and success of each country’s reforms. As a result, the parameter on the EU dummy variable is expected to have a positive sign.

The rate of unemployment in the host country, on the other hand, can be used as an indicator of labour market flexibility and availability of labour. Countries with high local demand for goods and services and high labour market flexibility are likely to face relatively low rates of unemployment, which may encourage firms to invest in a

(10)

particular host country. On the other hand, a high unemployment rate may mean that although it is easy to recruit labour, there is low demand locally and likely labour market rigidities. The impact of unemployment on the investment location decision is therefore strictly ambiguous and it may have a different effect on firms investing in different industries. For example, firms investing in traditional sectors employ less skilled labour and may be more concerned about the availability of workers, while firms investing in science-based industries, which employ more skilled labour, may be discouraged by higher unemployment, as unemployed people loose their skills through time.

Industry-level real wage rates are included as a proxy for the average variable costs of firms (it is implicitly assumed that workers are not fully mobile across sectors, at least in the short run). The profitability of the firm investing abroad is expected to be higher if the labour costs are lower in the chosen country than in the rest of the destination countries (Barrell and Pain, 1999). On the other hand, however, higher wages may reflect higher skills and, therefore, may have a positive effect for firms investing in science-based industries, where more skilled labour is employed as compared to other industries. As a result, the sign of the parameter on industry level real wage rates is ambiguous.

The industries that have received foreign capital can be divided into four groups: scale-intensive sectors, science-based industries, traditional sectors and service sectors. Scale-intensive sectors are typically oligopolistic, large firm industries, with high capital intensity, extensive economies of scale and learning, high technical and managerial complexity, for example, automobiles, aircrafts, chemicals, petrol and coal products, shipbuilding, industrial chemicals, drugs and medicines, petrol refineries, non-ferrous metals and railroad equipment. Science-based sectors, on the other hand, are characterised by innovative activities directly linked to high R&D expenditures, for example, fine chemicals, electronic components, telecommunications, and aerospace (Midelfart-Knarvik et al., 2000). Traditional (supplier-dominated) sectors include such industries as textiles, clothing, furniture, leather and shoes, ceramics, and the simplest metal products. Finally, banking insurance and retail are examples of service sector industries.

(11)

The country-level variables, however, may not only have a different value for firms investing in different sectors, but also for firms with different characteristics. Therefore, the firm-level variables included in the model are the turnover of the investing firm, as a proxy for its size and earnings before interest and tax, as a proxy for its profitability. Firms of different sizes and profitability possess different resources and capabilities. Small firms are assumed to be characterised by speed, flexibility and niche-filling capabilities due to their structural simplicity and faster decision making, entrepreneurial-orientation and less risk aversion (Woo, 1987). As a result, smaller firms respond quicker to the dynamics of the industry environment. Larger firms, which are usually more profitable, are able to acquire larger market shares by exploiting scale economies, bargaining power, patents, reputation and they have more financial resources to deal with shocks and business downturns (Dean et al., 1998).

Larger firms are expected to invest in countries with larger markets in order to exploit their economies of scale, while more profitable firms are expected to be less discouraged to invest in remote countries, as more financial resources are available to cover transaction costs, such as costs of transport and communication, the costs of dealing with cultural and linguistic differences and information costs of institutional and legal factors.

5. Estimation and Results

Before the LC model is estimated, the number of latent classes has to be determined. Although the log likelihood values increase with the number of latent classes showing the improvement in model fit, when an additional latent class is added, the number of parameters that has to be estimated increases and some coefficients can have large standard errors. As a result, the measures AIC, ρ_Q2, AIC3 and BIC, discussed in Section 2, are calculated for 1, 2, 3, 4 and 5 classes in order to determine the “optimal” number of latent classes (Table 3).

Three out of four criteria support five latent classes as the optimal solution for the data, as the calculated values of AIC and AIC3 (_ρ2_{) are minimum (maximum) for}

(12)

the number of classes, the LC model is, therefore, estimated with five latent classes. The estimation results with parameters estimated for each latent class are presented in Table 45.

The different estimated parameters for each latent class reveal five groups of investing firms in the sample. None of the groups has a clearly-defined type of investor who is concerned about particular, exclusive host country characteristics. The LC model allows for the preferences to overlap, as investing firms are heterogeneous and complex entities. Furthermore, the absolute values of the estimated parameter are not comparable across the latent classes due to scale differences and therefore are not informative. In order to be able to compare the results across different classes and to analyse to what degree investors with different characteristics benefit from different country-level variables, the marginal rates of substitution between two factors should be calculated for each class with at least the attribute in the denominator of the ratio being measured in monetary units6. Furthermore, both attributes used in the calculation of the ratio have to be statistically significant; otherwise, no meaningful measure can be established.

The calculated ratios with the parameter of the market size variable being in the denominator show how much investing firms are willing to “pay” in host country’s GDP in billion of EUR for the decrease (increase if the ratio is positive) in the distance between investing and investment receiving country by one kilometre, for the decrease in the hourly wage rate by one EUR, for the decrease (increase if the ratio is positive) in the unemployment rate by one percentage point, for the increase in the return on capital by one percentage point, for the host country being a EU member as compared to non-EU members and for the host country having a common border with the source country (Table 5). However, it is not the absolute value of the ratio itself that is of interest but its absolute value in relation to the equivalent ratios from other latent classes. The comparison of the ratios across different groups reveals the extent of the sensitivity of foreign investors in one class to country-level factors as compared to the foreign

5

The results of the Multinomial Logit (MNL) model estimation are available upon request. The fit of the model indicated by the Log-Likelihood function, Chi-squared and Pseudo R-squared is much better for the LC model as compared to the MNL model implying the superiority of the LC model.

6

In transport economics willingness to pay measures are used, for example, the value of travel time savings, defined as the amount of money an individual is willing to outlay in order to save a unit of time spent travelling, ceteris paribus.

(13)

investors from other latent classes. So, for example, firms in Class III are highly sensitive to distance and those in Class V to the wage rate.

The direction and the magnitude of the effect can also be revealed with elasticities and marginal effects. However, elasticities and marginal effects cannot be estimated for each class separately, as it is not known which class contains which investing firm, therefore, in the previous empirical literature that has applied the LC model, only an average effect over all classes is given. In contrast to the previous empirical literature, however, in this paper posterior probabilities are used to identify which firm belongs to which class, and, therefore, elasticities and marginal effects are estimated for each class separately and given in Table 6.

Following the results presented in Tables 4, 5 and 6 the first class contains foreign investors who choose to locate their capital abroad to access foreign markets (the estimated coefficients in Table 4 and estimated elasticities in Table 6 are positive for the market size variable for the investors in the first class) preferably in the EU member states (the estimated marginal effects in Table 6 and estimated coefficients in Table 4 are positive for the EU dummy variable for the investors in the first class) and they are not discouraged to invest in more remote countries in order to access those markets (the estimated elasticities in Table 6 and estimated coefficients in Table 4 are positive for the distance variables for the investors in the first class). Regarding individual investment receiving countries, market size appears to be especially important in attracting FDI in Russia, Czech Republic and Hungary (Table 6). The estimated marginal effects for the EU dummy variable for foreign investors in the first class are very high, especially for Hungary and Czech Republic (Table 6), indicating that EU membership is the key factor driving foreign capital allocations by foreign investors in the first class. Unemployment, on the other hand, has a negative effect on foreign capital allocation of firms in the first class, and the effect is the strongest for Croatia and Slovakia (Table 6).

The marginal rates of substitution cannot be calculated for the second class because the market size variable is not statistically significant for that class. However, based on the elasticities and marginal effects presented in Table 6 and the estimated coefficients in Table 4, it can be concluded that foreign investors in the second class are small firms who are concerned about the availability of a cheap labour force and they

(14)

prefer to invest in nearer neighbouring countries, preferably EU members, which have similar culture and traditions. Investors in the second class have the highest negative elasticities for the labour cost variable (Table 6) indicating that the higher labour costs in the host country the less likely that country will be chosen by foreign firms from the second class and this effect is strongest for Slovenia and Poland. The availability of the labour force for the investors in the second class appears to be of most importance in Poland, while a common border with the investing country is especially beneficial for Romania, Bulgaria and Ukraine (Table 6). Ukraine, for example, is a neighbour of Russia, which also appears to be its major source of foreign investment.

The third class contains larger, but less profitable, foreign investors with the highest negative sensitivity to the remoteness of the host country (especially Bulgaria and Ukraine) to the source country, as compared to foreign investors from other latent classes. The MNEs in the third class are not only discouraged to invest in more remote countries, but also in countries with higher wages (the effect is especially strong for Slovenia), as lower profitability means less resources available to cover higher labour costs and transactions costs associated with investment in more remote countries. Furthermore, these MNEs prefer to invest in countries with excess labour supply, which is probably low-skilled, as they also prefer host countries with lower labour costs. The higher the unemployment rate and the lower the wages in the host country, the more likely the country is to be chosen by foreign investors in the third class to locate their capital. Poland appears to be the country where foreign capital allocations have the highest positive sensitivity to unemployment for investors in the third class.

The fourth group of foreign investors contains profitable, but small investing firms in the sample, who choose to locate their capital in non-traditional sectors. Foreign firms in the fourth latent class have negative elasticities (Table 6) and the lowest overall marginal rate of substitution for the unemployment variable (Table 5) and the highest overall marginal rate of substitution for the labour cost variable. This can be explained by the fact that investors choosing to locate their capital in the non-traditional sectors (science-based industries, service sectors and scale-intensive industries) usually employ more-skilled labour and pay higher wages that reflect skill a premium as compared to MNEs investing in traditional sectors. Therefore, a higher unemployment rate in the host

(15)

country (especially in Croatia and Poland) will have a negative effect on the probability of foreign capital allocations in that country by foreign firms in the fourth class, while these firms will be less discouraged to invest in host countries with higher wage rates.

The investment by foreign firms in the fourth class is also driven by larger and closer neighbouring foreign markets, preferably EU members. In Russia and Poland market size has a stronger effect in attracting foreign capital, as compared to other CEECs. If the source country shares a common border with the host country, the host country is more likely to be chosen as an investment location by foreign firms in the fourth class, especially when the host countries are Estonia and Czech Republic (Table 6). Although, common border has a positive effect on the probability on selecting a host country for foreign capital allocations, the elasticities (Table 6) are quite low.

The decision to invest abroad by foreign firms in all four classes are positively affected by the cost of capital7 in the host country, with investing firms in the fourth latent class having the highest overall marginal rate of substitution for the cost of capital (Table 5). In the fifth latent class, the parameters of investing firm-specific variables are normalised and their values are set to zero (Table 4). Foreign investors in the fifth latent class choose to locate their capital in countries with large foreign markets; however, they are discouraged by higher labour costs and unemployment in the host country. Slovakia and Poland appear to have the highest elasticities for the market size variable, while foreign capital allocations by investors in the fifth class appear to be most discouraged by higher labour costs and unemployment in Croatia and Slovakia, as compared to other CEECs.

6. Conclusions

This paper applies the Latent Class (LC) model, which is one of the most flexible discrete choice models, for the first time, to investigate investment location choices of MNEs. It also makes use of a novel multi-level data set – allowing firm, industry (or sector) and country effects to simultaneously determine the firm-level FDI location decisions. The highly significant empirical results support the presence of heterogeneity in the investment location decisions, which is revealed by statistically significant

7

(16)

specific parameters. The model reveals five classes of foreign investors depending on investment receiving industry and country factors, investing firms’ individual characteristics and latent heterogeneity that varies with factors that are unobserved by the analyst. The results show that firms investing in different sectors and firms of different size and profitability benefit from host country factors to differing degrees. Firms investing in the non-traditional sectors are more likely to invest in countries with lower unemployment rates, but are less likely to be discouraged by higher wage rates as compared to MNEs that invest in traditional sectors. Investors choosing to locate their capital in non-traditional sectors usually employ more-skilled labour and pay higher wages that reflect a skill premium. The more profitable firms, on the other hand, are less likely to be discouraged to invest in more remote countries and pay higher wages, as compared to less profitable firms, as they have more funds to cover higher labour costs and transaction costs that arise from investment in more distant countries. This more general approach to the FDI location decision shows that to allow for firm heterogeneity is important if robust estimates are to be found for their complex effects.

(17)

Table 1: The Shares of Investing and Investment Receiving Countries BG CZ EE CR HU LT LV PL RO RU SI SK UA Austria 5 5 0 7 18 0 0 10 10 1 5 5 1 Belgium 0 4 0 1 3 0 0 6 3 7 3 4 0 Denmark 2 3 1 0 1 8 4 15 2 0 1 0 0 Germany 4 20 3 6 12 1 1 11 14 7 3 6 0 Finland 0 5 32 5 0 23 10 21 2 24 0 2 7 France 2 6 1 2 11 1 2 40 9 13 4 2 8 Greece 19 0 0 1 7 0 0 0 19 1 0 0 4 Ireland 1 2 0 0 0 0 0 3 0 3 0 1 3 Italy 1 2 0 0 4 0 0 8 7 7 1 1 0 Japan 0 1 0 0 1 0 0 2 0 5 0 0 0 Luxembourg 3 4 0 4 0 0 0 9 0 1 0 0 0 Netherlands 3 13 0 3 3 0 0 18 5 17 0 6 3 Norway 0 5 4 0 1 3 2 2 1 10 0 1 2 Russia 0 7 1 0 2 4 5 1 3 0 0 0 30 Spain 2 7 2 1 1 0 0 7 6 6 0 0 0 Switzerland 0 4 0 1 4 0 0 2 3 9 3 0 0 Sweden 3 4 14 2 1 16 7 13 0 24 2 0 2 UK 5 15 2 2 19 0 1 23 8 37 1 6 3 US 3 15 0 0 13 1 0 32 12 20 3 3 3

(18)

Table 2: List of variables, definitions and sources

Variable Definition Source

Choice a CEEC, in which firm n chooses to locate its investment over the period of time from 1997 to 2007 (it gets the value of 1 if the country received investment and 0 otherwise)

Bureau van Dijk Zephyr database

Distance Real GDP of the host country c of the year investment took place

IFS

Unemployment unemployment rate of country c (percentage per annum) of the year investment took place

IFS

Interest the real discount (interest) rate IFS

Border a dummy variable that takes a value 1 if both source country d and host country c share a border, and 0 otherwise

constructed

EU dummy variable that takes value 1 if country c joined EU before January 2007, and 0 otherwise

constructed

Traditional dummy variable that takes a value 1 if industry s is a traditional industry, and 0 otherwise

constructed

Wage hourly real wage rates in the industry s in the country c of the year investment took place

International Labour Organisation

Size turnover of the investing firm i in Euros of the year investment took place

Profit earnings before interest and taxes of the investing firm i in Euros of the year investment took place

Table 3: The AIC, ρ2Q , AIC3 and BIC Measures for 2, 3, 4 and 5 Classes

2 3 4 5 LogL*Q -2467.643 -2426.882 -2380.229 -2351.094 Kq 18 29 40 51 AIC 4971.286 4911.764 4840.458 4804.188 ρ2Q 0.1254 0.1359 0.1484 0.1548 AIC3 4987.286 4940.764 4880.458 4855.188 BIC 2530.7357 2528.5314 2520.435 2529.8567

The optimal number of latent classes is indicated by the minimum value of AIC, AIC3 and BIC and by the maximum value of ρQ2.

(19)

T ab le 4: T h e E st im at ed C oe ff ic ie n ts of t h e L at en t C las s M od el C las s 1 C las s 2 C las s 3 C las s 4 C las s 5 C o ef z-st at s C o ef z-st at s C o ef z-st at s C o ef z-st at s C o ef z-st at s Wa g e 0 .0 0 2 8 { 0 .0 4 2 } -3 .8 6 3 1 * * {-1 0 .7 1 8 } -0 .2 5 1 6 * * {-2 .2 2 0 } -0 .0 9 1 8 * * {-2 .2 5 7 } -2 .6 2 3 7 * {-3 .5 7 1 } D is ta n ce 3 .5 9 8 8 * * {2 .5 8 1 } -1 .2 8 7 7 * * {-4 .4 8 6 } -4 .6 8 1 4 * {-1 2 .0 6 7 } -1 .4 0 6 9 * {-1 3 .3 2 3 } 3 .7 4 5 6 { 1 .1 6 4 } G D P 4 .4 8 7 7 * {3 .9 0 7 } -0 .0 8 8 8 { -0 .7 9 0 } 0 .4 2 6 8 * {3 .9 3 4 } 0 .2 2 8 8 * {3 .1 6 5 } 3 .6 9 1 5 * {3 .2 5 2 } U n e m p lo y m e n t -1 7 .2 1 3 0 * * {-2 .1 5 4 } 6 .7 6 3 8 * * * {1 .6 4 3 } 1 0 .2 6 0 3 * {4 .3 6 6 } -8 .1 1 9 9 * {-5 .8 9 5 } -7 5 .7 7 9 9 * {-4 .1 7 2 } B o rd er -4 .2 3 2 2 * {-3 .2 0 7 } 1 .3 4 9 4 * {4 .1 1 1 } -2 9 .7 8 3 6 { 0 .0 0 0 } 0 .5 3 7 1 * {3 .0 6 8 } 5 6 .0 9 6 2 { .0 0 0 1 } E U 1 1 .5 2 4 9 * {3 .3 3 4 } 1 .0 6 4 5 * {5 .1 9 3 } 0 .0 4 3 2 { 0 .1 8 7 } 1 .9 6 9 8 * {7 .5 1 5 } -3 3 .9 3 3 5 { .0 0 0 1 } In te re st 2 3 .4 5 5 2 * {3 .0 5 0 } 4 .1 6 7 8 * {2 .1 4 9 } 6 .7 2 6 4 * * {2 .0 6 7 } 3 .7 8 0 9 * * {2 .2 1 7 } 1 6 .9 2 4 7 { 1 .3 5 4 } C o n st a n t 0 .0 4 9 8 { 0 .2 3 6 } -0 .3 3 5 3 { -1 .2 9 3 } -0 .3 1 9 8 { -1 .0 2 5 } 1 .2 8 1 4 { 7 .4 4 2 } F ix ed p ar a m et er S iz e -0 .0 2 3 6 { -0 .1 6 0 } -0 .7 7 0 4 * * {-1 .9 6 0 } 0 .6 9 1 9 * * * {1 .8 5 9 } -1 .8 3 8 5 * {-2 .9 6 7 } F ix ed p ar a m et er T ra d it io n a l 0 .0 6 9 9 { 0 .2 6 3 } 0 .3 5 1 8 { 1 .1 0 6 } 0 .5 6 0 9 { 1 .5 3 0 } -1 .1 1 8 0 * {-3 .7 6 6 } F ix ed p ar a m et er P ro fi t -0 .0 5 2 5 { -0 .3 7 9 } 0 .4 2 0 9 { 1 .5 9 3 } -1 .7 4 5 2 * * {-2 .1 4 7 } 0 .8 6 2 5 * {2 .7 3 3 } F ix ed p ar a m et er t-st a ti st ic s in p ar en th e si s * S ig n if ic a n t at 1 p er ce n t le v e l * * S ig n if ic an t at 5 p er ce n t le v el * * * S ig n if ic an t at 1 0 p er ce n t le v el T h e es ti m at ed c o ef fi ci e n ts i n d if fe re n t la te n t cl as se s ar e n o t co m p ar ab le d u e to s ca le d if fe re n ce s. T ab le 5: R at ios of t h e e st im a te d p ar a m e te rs w it h t h e p ar a m et er of t h e G D P var iab le b ei n g i n t h e d en om in at or V ar iab le s in th e n o m in a tor C las s I C las s II C las s II I C las s IV C las s V D is tan ce 0.8019 ns s -10.500 0 -6.1490 ns s Wage ns s ns s -0.5895 -0.4012 -0.7107 U n em p lo ym en t -3.8356 ns s 24.0401 -35.489 1 -20.5282 In te re st 5.2266 ns s 15.7601 16.5249 ns s E U 2.5681 ns s ns s 8.6093 ns s B or d er -0.9431 ns s ns s 2.3475 ns s T rad it ion a l ns s ns s ns s -4.8864 ns s S iz e ns s ns s 1.6211 -8.0354 ns s P ro fi tab il it y ns s ns s -4.0890 3.7710 ns s n ss = n o t st at is ti ca ll y s ig n if ic an t. I n o rd er t o c o m p ar e th e es ti m at ed c o ef fi ci en ts a cr o ss l at en t cl a ss es p re se n te d i n T ab le 9 .2 , m ar g in al r at e s o f su b st it u ti o d er iv ed , w h er e th e ra ti o s o f tw o e st im at ed p ar a m et er s in e ac h l at en t cl as s ar e ca lc u la te d , w it h t h e p ar a m et er o f th e m ar k e t si ze v ar ia b le b ei n g i n t h e d en o m in

(20)

T ab le 6 : C las s-S p ec if ic E las ti ci ti es an d M ar g in a l E ff ec ts Wa g e D is ta n ce C la ss I C la ss I I C la ss I II C la ss I V C la ss V C la ss I C la ss I I C la ss I II C la ss I V C la B u lg ar ia n ss -1 .7 4 6 6 -0 .1 8 0 6 -0 .0 6 8 8 -0 .1 1 0 5 1 5 .3 8 4 2 -1 .8 4 8 9 -6 .8 8 0 6 -2 .0 0 4 4 n C ze ch R ep n ss -7 .4 1 9 9 -0 .3 8 7 2 -0 .2 2 0 4 -1 .9 4 6 6 1 2 .2 1 7 5 -1 .4 7 4 9 -4 .6 4 2 1 -1 .7 1 7 5 n E st o n ia n ss -2 .2 5 0 9 -0 .2 1 0 3 -0 .0 9 1 9 -2 .0 1 1 7 1 1 .5 1 7 9 -1 .8 1 0 0 -1 .6 9 5 0 -1 .5 9 0 4 n C ro at ia n ss -3 .6 0 5 5 -0 .3 0 7 5 -0 .1 8 2 3 -5 .3 3 0 0 3 3 .4 4 0 0 -1 .4 9 5 0 -4 .0 1 5 1 -2 .3 2 1 2 n H u n g ar y n ss -5 .6 0 9 0 -0 .3 8 3 8 -0 .1 9 3 2 -0 .4 2 8 7 6 .9 9 5 3 -1 .3 9 8 8 -5 .0 6 3 7 -1 .9 5 8 2 n L it h u an ia n ss -4 .9 4 2 1 -0 .3 3 8 5 -0 .1 3 1 6 -3 .6 3 9 6 1 8 .5 0 5 6 -2 .6 7 2 3 -4 .8 8 6 9 -2 .1 1 2 9 n L at v ia n ss -4 .5 2 9 5 -0 .2 7 6 9 -0 .1 2 0 7 -3 .1 7 0 6 1 8 .2 6 0 3 -3 .2 0 1 9 -3 .5 7 4 7 -1 .9 0 5 5 n P o la n d n ss -7 .5 5 8 0 -0 .3 6 7 2 -0 .2 0 7 5 -1 .6 8 7 7 1 .1 0 4 7 -1 .8 9 8 8 -4 .8 5 5 1 -1 .6 7 4 6 n R o m a n ia n ss -0 .9 9 8 5 -0 .0 4 6 4 -0 .0 5 8 0 -1 .3 9 6 2 1 3 .5 2 6 8 -1 .6 9 3 4 -1 .9 9 8 0 -2 .4 4 5 7 n R u ss ia n ss -1 .3 2 5 1 -0 .1 3 1 7 -0 .0 7 9 4 -0 .0 0 8 6 6 .7 9 7 7 -2 .6 4 5 1 -4 .7 3 8 6 -2 .3 1 9 6 n S lo v e n ia n ss -1 1 .1 5 1 8 -0 .8 5 1 4 -0 .3 8 0 1 -5 .6 5 0 2 1 7 .5 1 9 2 -1 .1 9 9 5 -5 .7 2 3 8 -1 .9 0 7 6 n S lo v a k ia n ss -6 .4 0 4 9 -0 .3 2 3 5 -0 .1 7 5 2 -4 .9 0 1 0 2 1 .4 7 1 6 -1 .0 9 2 6 -4 .1 8 0 0 -1 .9 7 7 8 n U k ra in e n ss -0 .7 7 3 5 -0 .0 9 1 2 -0 .0 3 8 6 -0 .0 0 2 1 2 6 .1 9 4 0 -2 .0 3 1 3 -6 .9 1 5 5 -2 .5 8 6 9 n G D P U n e m p lo y m e n t C la ss I C la ss I I C la ss I II C la ss I V C la ss V C la ss I C la ss I I C la ss I II C la ss I V C la B u lg ar ia 0 .7 2 5 9 n ss 0 .0 7 6 7 0 .0 3 6 0 0 .0 5 7 2 -1 .8 5 1 3 0 .4 7 7 2 1 .1 8 6 1 -0 .7 6 4 3 -0 .5 C ze ch R ep 3 .1 2 1 1 n ss 0 .2 4 1 6 0 .1 5 9 9 0 .8 8 9 4 -1 .2 3 9 8 0 .6 2 0 3 0 .5 9 7 2 -0 .5 5 6 6 -1 .8 E st o n ia 0 .3 4 7 3 n ss 0 .0 1 9 5 0 .0 1 5 4 0 .3 8 7 3 -0 .6 2 0 9 0 .1 5 3 6 0 .2 0 8 9 -0 .2 0 7 1 -1 .5 C ro at ia 1 .1 9 8 4 n ss 0 .0 7 9 7 0 .0 5 7 8 0 .9 0 3 7 -2 .1 8 3 6 0 .6 9 9 2 1 .0 0 9 5 -1 .0 6 5 1 -1 0 H u n g ar y 2 .0 2 3 9 n ss 0 .2 5 2 0 0 .1 2 9 0 0 .2 3 3 7 -0 .7 8 1 5 0 .4 4 6 2 0 .6 2 8 7 -0 .4 7 9 2 -0 .5 L it h u an ia 0 .7 3 6 5 n ss 0 .0 6 7 5 0 .0 3 8 0 0 .6 4 9 8 -1 .2 2 9 5 0 .4 2 4 5 0 .8 0 1 6 -0 .4 9 3 5 -6 .7 L at v ia 0 .4 3 6 3 n ss 0 .0 3 3 7 0 .0 2 0 5 0 .4 1 8 8 -1 .3 6 0 7 0 .5 0 4 3 0 .6 5 7 4 -0 .5 6 1 4 -5 .4 P o la n d 1 .5 0 6 3 n ss 0 .5 7 2 7 0 .4 0 1 2 1 .8 5 4 4 -0 .4 7 3 2 1 .1 9 1 8 1 .1 5 3 4 -1 .0 8 5 3 -2 .9 R o m a n ia 1 .1 1 3 4 n ss 0 .0 6 3 3 0 .0 6 7 2 0 .9 5 1 9 -1 .0 6 4 7 0 .2 8 0 6 0 .3 6 5 4 -0 .5 3 0 4 -5 .4 R u ss ia 8 .0 6 4 6 n ss 1 .5 9 2 1 0 .8 7 0 6 0 .0 8 3 2 -0 .6 9 3 0 0 .4 0 9 4 0 .6 1 7 5 -0 .6 3 3 3 -0 .0 S lo v e n ia 0 .9 0 5 5 n ss 0 .0 8 4 1 0 .0 4 5 0 0 .5 5 6 8 -1 .5 6 4 2 0 .6 1 7 3 0 .8 9 5 5 -0 .6 5 1 8 -4 .9 S lo v a k ia 1 .2 5 9 0 n ss 0 .0 8 8 7 0 .0 6 2 9 1 .1 9 9 1 -2 .0 8 9 1 0 .8 7 6 7 1 .0 1 8 3 -0 .9 3 4 0 -8 .3 U k ra in e 1 .7 3 5 4 n ss 0 .1 8 0 4 0 .0 9 8 0 0 .0 0 3 3 -1 .6 3 2 6 0 .3 9 3 7 0 .9 3 3 5 -0 .7 3 1 4 -0 .0

(21)

In te re st E U C la ss I C la ss I I C la ss I II C la ss I V C la ss V C la ss I C la ss I I C la ss I II C la ss I V C la ss V B u lg ar ia 0 .8 5 5 6 0 .0 5 5 4 0 .2 1 5 0 0 .0 8 8 6 n ss 1 .0 5 7 9 0 .1 8 8 6 n ss 0 .2 2 1 5 n ss C ze ch R ep 0 .7 8 4 1 0 .1 5 1 2 0 .1 8 9 3 0 .1 3 2 9 n ss 1 .3 6 6 6 0 .0 1 0 1 n ss 0 .2 8 9 4 n ss E st o n ia 0 .4 1 0 0 0 .0 1 3 1 0 .0 5 7 4 0 .0 2 2 1 n ss 0 .7 2 4 8 0 .1 4 7 0 n ss 0 .3 4 4 1 n ss C ro at ia 1 .3 3 5 0 0 .1 8 5 5 0 .2 8 1 6 0 .2 1 9 8 n ss 0 .0 2 5 2 0 .1 2 4 7 n ss 0 .0 2 2 8 n ss H u n g ar y 0 .9 5 9 2 0 .1 9 6 9 0 .2 7 8 2 0 .1 3 4 6 n ss 1 .9 3 5 6 0 .0 3 9 0 n ss 0 .2 2 2 8 n ss L it h u an ia 0 .6 9 3 8 0 .0 6 5 1 0 .2 6 6 6 0 .0 7 1 2 n ss 0 .9 3 5 6 0 .0 2 3 3 n ss 0 .1 8 9 9 n ss L at v ia 0 .2 1 4 4 0 .0 0 7 6 0 .0 7 7 9 0 .0 2 7 2 n ss 0 .3 8 8 6 0 .0 5 0 1 n ss 0 .2 1 7 5 n ss P o la n d 0 .1 3 7 1 0 .1 4 3 4 0 .2 1 2 0 0 .1 4 5 6 n ss 1 .1 1 3 5 0 .0 1 1 9 n ss 0 .2 3 5 0 n ss R o m a n ia 0 .2 2 4 6 0 .0 2 7 4 0 .0 3 8 9 0 .0 3 6 1 n ss 1 .1 6 8 7 0 .2 4 0 8 n ss 0 .1 4 2 3 n ss R u ss ia -0 .2 4 8 5 -0 .0 8 3 4 0 .1 7 0 6 0 .0 0 9 7 n ss 1 .3 2 4 4 0 .1 4 2 2 n ss 0 .0 8 8 4 n ss S lo v e n ia 1 .0 2 1 9 0 .1 8 9 0 0 .2 8 2 6 0 .1 3 2 9 n ss 0 .7 1 3 6 0 .0 0 0 3 n ss 0 .1 9 2 6 n ss S lo v a k ia 0 .7 7 0 4 0 .1 5 1 3 0 .2 0 2 0 0 .1 2 0 2 n ss 0 .6 2 3 4 0 .0 3 1 2 n ss 0 .1 6 2 2 n ss U k ra in e 3 .9 6 0 6 0 .0 3 5 6 0 .6 0 9 1 0 .2 8 7 9 n ss 0 .0 0 6 2 0 .1 8 9 6 n ss 0 .0 2 7 8 n ss B o rd er C la ss I C la ss I I C la ss I II C la ss I V C la ss V B u lg ar ia -0 .3 8 8 5 0 .2 3 9 1 n ss 0 .0 6 0 4 n ss C ze ch R ep -0 .5 0 1 9 0 .0 1 2 8 n ss 0 .0 7 8 9 n ss E st o n ia -0 .2 6 6 2 0 .1 8 6 4 n ss 0 .0 9 3 8 n ss C ro at ia -0 .0 0 9 2 0 .1 5 8 1 n ss 0 .0 0 6 2 n ss H u n g ar y -0 .7 1 0 8 0 .0 4 9 4 n ss 0 .0 6 0 7 n ss L it h u an ia -0 .3 4 3 6 0 .0 2 9 6 n ss 0 .0 5 1 8 n ss L at v ia -0 .1 4 2 7 0 .0 6 3 5 n ss 0 .0 5 9 3 n ss P o la n d -0 .4 0 8 9 0 .0 1 5 0 n ss 0 .0 6 4 1 n ss R o m a n ia -0 .4 2 9 2 0 .3 0 5 3 n ss 0 .0 3 8 8 n ss R u ss ia -0 .4 8 6 3 0 .1 8 0 3 n ss 0 .0 2 4 1 n ss S lo v e n ia -0 .2 6 2 1 0 .0 0 0 4 n ss 0 .0 5 2 5 n ss S lo v a k ia -0 .2 2 8 9 0 .0 3 9 6 n ss 0 .0 4 4 2 n ss U k ra in e -0 .0 0 2 3 0 .2 4 0 4 n ss 0 .0 0 7 6 n ss n ss = n o t st at is ti ca ll y s ig n if ic an t. E la st ic it ie s ar e e st im at ed f o r th e co n ti n u o u s v ar ia b le s an d m ar g in al e ff ec ts a re e st im at ed f o r th e d u m m y v ar ia b le s. A d el a st ic it y m ea su re s th e p er ce n ta g e ch a n g e in t h e p ro b ab il it y o f ch o o si n g a p ar ti cu la r al te rn at iv e in t h e ch o ic e se t w it h r es p ec t to a g iv en p er ce n ta g e ch an g e at tr ib u te o f th at s a m e al te rn at iv e. D ir ec t m a rg in a l ef fe ct s r ep re se n t th e c h a n g e in t h e ch o ic e p ro b ab il it y f o r an a lt er n at iv e g iv e n a u n it c h a n g e in a v ar ia b le r el to t h at a lt er n at iv e

(22)

References

Barrell, R., Pain, N.,1997. Foreign direct investment, technological change, and economic growth within Europe. Economic Journal 107, 1770-1786.

Barrell, R., Pain, N.,1999. Domestic institutions, agglomerations and foreign direct investment in Europe. European Economic Review 43, 925-934.

Basile, R., Castellani, D., Zanfei, A., 2008. Location choices of multinational firms in Europe: The role of EU cohesion policy. Journal of International Economics 74, 328-340.

Becker, S. O., Ekholm, K., Jackle, R., Muendler, M. A., 2005. Location choice and employment decisions: A comparison of German and Swedish multinationals. Review of World Economics 141, 693-731.

Bénassy-Quéré, A., Coupet, M., Mayer, T., 2007. Institutional Determinants of Foreign Direct Investment. The World Economy, 30 (5), 764-782.

Buch, C. M., Kleinert, J., Lipponer, A., Toubal, F., 2005. Determinants and effects of foreign direct investment: Evidence from German firm-level data. Economic Policy, 53-109.

Carstensen, K., Toubal, F., 2004. Foreign direct investment in Central and Eastern European countries: a dynamic panel analysis. Journal of Comparative Economics 32, 3-22.

Clausing, K. A., Dorobantu, C. L., 2005. Re-entering Europe: Does European Union candidacy boost foreign direct investment? Economics of Transition 13, 77-103. Crozet, M., Mayer, T., Mucchielli, J. L., 2004. How do firms agglomerate? A study of

FDI in France. Regional Science and Urban Economics 34, 27-54.

Dean, T. J., Brown, R. L., Bamford, C. E., 1998. Differences in large and small firm responses to environmental context: Strategic implications from a comparative analysis of business formations. Strategic Management Journal 19, 709-728. Disdier, A. C., Mayer, T., 2004. How different is Eastern Europe? Structure and

determinants of location choices by French firms in Eastern and Western Europe. Journal of Comparative Economics 32, 280-296.

Greene, W. H., Hensher, D. A., 2003. A latent class model for discrete choice analysis: contrasts with mixed logit. Transportation Research Part B-Methodological 37, 681-698.

Head, K. Ries, J., 2008. FDI as an outcome of the market for corporate control: Theory and evidence. Journal of International Economics 74(1), 2-20.

Meijer, E., Rouwendal, J., 2006. Measuring welfare effects in models with random coefficients. Journal of Applied Econometrics 21, 227-244.

Midelfart-Knarvik, K. H., Overman, H. G., Redding, S. J., Venables, A. J., 2000. The Location of European Industry. European Economy - Economic Papers 142, Commission of the EC, Directorate-General for Economic and Financial Affairs (DG ECFIN).

Nocke, V., Yeaple, S., 2007. Cross-border mergers and acquisitions vs. greenfield foreign direct investment: The role of firm heterogeneity. Journal of International Economics 72(2), 336-365.

(23)

Shen, J., Saijo, T. (2007). Does energy efficiency label alter consumers' purchase decision? A latent class approach on Shanghai data. Osaka School of International Public Policy, Osaka University, Toyonaka, Osaka.

Shen, J., Sakata, Y., Hashimoto, Y. (2006). A Comparison between Latent Class Model and Mixed Logit Model for Transport Mode Choice: Evidences from Two Datasets in Japan. Graduate School of Economics and Osaka School of International Public Policy, Osaka University, Toyonaka, Osaka.

Wei, S. J. (2000). How taxing is corruption on international investors? Review of Economics and Statistics 82, 1-11.

Woo, C. Y. (1987). Path-Analysis of the Relationship between Market Share, Business-Level Conduct and Risk. Strategic Management Journal 8, 149-168.