<Psychic Distance and Subsidiary Performance: an empirical research in the manufacturing sector and the non-manufacturing sector>

(1)

<Psychic Distance and Subsidiary Performance:

an empirical research in the manufacturing

sector and the non-manufacturing sector>

Faculty of Economics and Business

University of Groningen

Master Thesis for International Economics and Business

2009-2010

Student: Jun Yang Student ID: 1521136

Thesis Supervisor: Dr. Robbert Maseland

Methodology Supervisor: Prof. dr. Erik Dietzenbacher E-mail: s1521136@student.rug.nl

(2)

ABSTRACT

As the Central and Eastern European (CEE) countries opened up after the collapse of communism in early 90’s, these countries commenced a transition process towards the EU and more and more multinationals from the Western Europe attempted to set up subsidiaries and operate in the local markets. In this paper, I study the relationship between psychic distance and MNEs’ subsidiary performance measured by ROTA (return on total assets). These indicators are limited to the cultural and economic aspects. Using one sample of 19 manufacturing firms and another sample of 23 non-manufacturing firms over a period from 2000 to 2007, I test whether psychic distance in the cultural, economic and institutional perspective of the host countries do have an impact on MNEs’ subsidiary performance.

(3)

INDEX 1. INTRODUCTION...5 2. LITERATURE REVIEW...7 2.1. Economic Distance ...8 Hypothesis 1 ...9 Hypothesis 2 ...9 2.2. Cultural Distance ...10 Hypothesis 3 ...12 Hypothesis 4 ...12 2.3. Institutional Distance...12 Hypothesis 5 ...12 2.4. Firm-level variables ...12 Hypothesis 6 ...13 Hypothesis 7 ...13 Figure 1 ...14 3. METHODOLOGY...15 3.1. Data Collection ...15 Table 1...16 3.2. Selection bias ...17 3.3. Variables...18 3.3.1. Dependent variable ...18 3.3.2. Independent variables...19 Table 2...21 Table 3...24 Table 4...25 3.3.3. Control variables...26 3.4. Diagnostic Testing ...27

3.4.1. Normality of the residuals ...27

3.4.2. Heteroskedasticity ...30 Table 5...30 3.4.3. Autocorrelation...31 3.4.4. Multicollinearity...32 Table 6...32 3.5. Data Analysis ...33

3.5.1. Panel Data Estimation Methods...33

3.5.2. Fixed and Random Effects Estimators...34

3.5.3. Specific Statistical Model...36

3.5.4. Hausman-test...36

3.5.5. Robust estimate of variance ...37

3.6. Results ...38

3.6.1. Manufacturing sector...38

(4)

(5)

1. INTRODUCTION

In 1989, the collapse of the Berlin Wall indicated the end for the Soviet Union to the whole world. Since then, a new political and economic situation formed in the Central and Eastern Europe (CEE). The CEE countries thus commenced a transition process which contains reforms towards political democracy, private firms and a market economy open to international trade and foreign direct investment.

Although the starting point for the transition process differed among the countries within the Central and Eastern Europe (CEE), all the countries engaged in the process of implementing transition reforms mainly on changes in microeconomic, macroeconomic and institutional domains. These reforms differed in design and degree of implementation. Countries like Czech Republic, Hungary, Poland and Slovakia initiated some reforms even before the fall of the Berlin Wall. These countries are known as Viségrad countries. They were the frontiers among the CEE countries to liberate themselves from the old communist system and to start with the transition process peacefully. The Baltic countries soon followed, however due to the civil political instability and other cases, the transition was slower.

All in all, it was evident that a trend of transition among all the CEE countries started. The trend was characterized by a systematic shift from a centrally planned to a free market oriented economy (Naor, 1990). The change in institutions which occurred has been argued to be a unique type of change in modern time (Kornai, 2006). The access towards the European Union that occurred in the later 1990s is considered as a major indicator of the existence of such change.

According to Campos and Coricelli (2002), the reforms can be characterized by a major contraction of industrial output due to industrial restructuration and consequently led to a widespread recession and unemployment across the CEE region. One of the impediments during the process of transition was the social heritage since it might led to the presence of macroeconomic imbalance, distortion of prices, presence of a capital stock which was inappropriate for an open market economy. The opening-up of the markets, privatization of ownership and consequently the increase in Foreign Direct Investment (FDI) was seen as a key way to overcome this problem. As a matter of fact, FDI is widely considered as an impulse for economic change in financial resources, technology, management techniques and links to foreign country markets from the perspective of the CEE host countries.

(6)

multinational enterprises (MNEs) into the specific host countries with distinct backgrounds, difference may occur in economic, political, social, cultural and legal systems from the same systems of the home countries for the MNEs. Therefore, in order to enter and operate in those new country markets successfully, multinational enterprises have to analyze the distance in various aspects between the home and host countries, and thus adapt their strategies to the host country environment for achievement of satisfying firm performance.

In recent years, significant attention has been paid to the psychic distance with regard to international business research (Dow and Karunaratha, 2006). In fact, it is one of the most commonly applied theoretical constructs from a holistic perspective of both national and individual dimensions in analyzing MNEs’ internationalization activities (Sivakumar and Cheryl, 2001). Compared to other prevailing theoretical frameworks merely focusing on one of the aspects (e.g. cultural or institutional), psychic distance concept is considered comprehensive and convincing when analyzing firm-level performance. The original concept was designed by Johanson and Vahlne that refers to “the sum of factors preventing the flow of information to and from the market” (Johanson and Vahlne, 1977:24). Nordstrom and Vahlne (1994) broaden the theory who suggest that psychic distance indicates factors which prevent or disturb firm’s learning about and understanding of a foreign environment. Dow and Karunaratha (2006) further discuss the psychic distance stimuli which range from aspects of culture, language, education, economic development, political system, religion, and time zone. In other words, the degree of psychic distance is not simply determined by the presence of external environmental factors, it is also influenced by the mind’s processing in term of perception and understanding (O'Grady and Lane, 1996; Lee, 1998; Evans et al., 2000). Thus psychic distance reflects the perceived degree of similarities or differences between the home and foreign country environment.

(7)

In this paper, I intend to examine the relationship between psychic distance and firm performance by following the framework of Dow and Karunaratha (2006) in regard to psychic distance. However unlike many previous studies which only focus on cultural distance as the main variable for determining firm performance, I also include economic and institutional indicators in order to provide a more broad and comprehensive explanation on the issue of psychic distance and organizational performance because economic and institutional indicators are also essential instruments when measuring psychic distance. Furthermore, the research targets for performance analysis are 42 selected multinational firms from west-European countries that own subsidiaries in the CEE (Central and Eastern European) region between 1999 and 2008. In other words, the home and host countries of my selected firms are west-European and CEE countries respectively. The details of the selection of my sample and the possible occurrence of selection bias are discussed in the

section 3.1 namely “data collection” in the methodology section.

The main research question is “Does psychic distance between the host and home

country affect the subsidiary performance of MNEs?”. In the following sections, I

discuss the theoretical background and propose my hypotheses. Moreover, I describe the empirical analysis and interpretation of the results. Finally, I provide a conclusion with both the implications and limitation of the research paper.

2. LITERATURE REVIEW

(8)

distance which is defined as “international marketer's perceived socio-cultural distance between the home and target country in terms of language, business practices, legal and political systems and marketing infrastructure” (Lee, 1998:9). O'Grady and Lane (1996) and Lee (1998) successfully incorporate perception into their respective definitions of psychic distance, which overcomes the shortcoming from the previous studies. Moreover, their research also highlights the need to specify the factors for determining psychic distance. According to O'Grady and Lane (1996), psychic distance should incorporate both business differences and cultural differences. Based on this review, psychic distance can thus be comprehensively defined as the distance between the home and foreign markets resulting from the perception and understanding of cultural and business differences. Such business differences should include the legal and political environment, economic environment, business practices, language and industry or market sector structure (O'Grady and Lane, 1996; Evans et al., 2000). In other words, business differences contain distance in both institutional (legal and political environment) and economic environment. Thus psychic distance can be regarded as differences mainly in cultural, economic and institutional perspectives. In order to be more specific, I focus on these three factors as the main components of psychic distance in regard to their impacts on subsidiary performance.

2.1. Economic Distance

Ghemawat (2001) defines economic distance as an indicator to reflect the level of economic development of the host country relative to that of the home country. The difference in the levels of economic development between two countries often indicates the difference in factor costs (such as the wage rate) which is considered as an important factor influencing FDI decisions and firm performance of MNEs (Ghemawat, 2001).

(9)

referred to as the “vertical FDI” when the gaps on factor costs between the host and home countries are large; the second type investment is made when MNEs have a market-seeking emphasis which is also known as the “horizontal FDI” (Dunning, 1993).

Thus in manufacturing sector which emphasizes more on efficient-seeking approach regarding production activities, MNEs from developed economies tend to enter and operate in less developed countries with a large (economic) distance with regard to labor costs in concern of cost reduction for achieving efficiency in order to enhance the firm performance (Dunning, 1993; Fussell et al., 2006). According to Radosevic (1999), new production and innovation networks in the transition economies are mostly foreign-led. Evans et al. (2002) further point out that firms based in developed economies are likely to perform well in developing and newly-industrialized countries due to the first-mover advantages on establishing a presence in such markets. The perception of economic differences can thus improve organizational performance. However in non-manufacturing sector that focuses more on market-seeking approach in regard to provision of technology or service, MNEs from developed countries are likely to achieve better performance in host countries with a relatively low degree of economic distance. Dunning (2000) argues that from the market-seeking perspective, it has been increasingly important for MNEs to acquire and sustain its strategic assets over time for gaining and maintaining their dynamic ownership advantages in the knowledge-based global economy. In other words, MNEs tend to maintain a (technological) competitive edge over local firms so as to operate profitably. As technology which is dependent on location-specific factors1 differs across countries and also the technological level of a country is generally linked to its economic development, strategic assets are more likely to be acquired in countries with proximate economic development (Cantwell, 1989; Tsang et al., 2007). Thus economic distance may cause barriers for MNEs to explore and sustain strategic assets, and ultimately lead to loss of dynamic ownership advantages as well as poor firm performance.

Hypothesis 1: Economic distance between the home and host country has a positive effect on subsidiary performance of MNEs in the manufacturing sector.

Hypothesis 2: Economic distance between the home and host country has a negative effect on subsidiary performance of MNEs in the non-manufacturing sector.

1 _{These factors include past technological innovations, education system, and linkages between scientific}

(10)

2.2. Cultural Distance

Cultural distance has received a significant attention in the international business literature (e.g. Kogut and Singh, 1988; O'Grady and Lane, 1996; Lee, 1998). The research is mostly based on Hofstede's (1983) dimensions of national culture.

Hofstede (1983) initially introduces four dimensions to describe national cultures. These different dimensions help to explain different cultures. The first dimension is “Individualism versus Collectivism” which describes the relationship between individuals of whether individuals care for themselves or if they look after the interests of their in-group. The second dimension is “Power Distance”, which deals with the fact that power is distributed unequally. High power distance cultures allow inequalities grow over time. The third dimension “Uncertainty Avoidance” deals with the need to control and prevent uncertainty or to accept it. The fourth dimension is “Masculinity versus Femininity” which deals with if masculine or feminine values are valued more in the society (Hofstede, 1983). The score of 100 represents a strongly individualistic society and 0 as a strongly collectivistic society. The power distance scale runs from 0 (low power distance) to 104 (high power distance). The uncertainty avoidance scale also runs from 0 (weak uncertainty avoidance) to 112 (strong uncertainty avoidance). A high score up to 110 on the masculinity index indicates that it’s a masculine society while a low score down to 0 indicates that it’s a feminine society (Hofstede, 1983 and 2001).

Using Hofstede’s results of these dimensions, Kogut and Singh (1988) developed a formula index to determine the cultural distance between two countries. The formula is mainly used to reflect differences on national culture among countries based on the scores of Hofstede’s 4 cultural dimensions. Thus many studies have subsequently used the Kogut and Singh (1988) formula as a measure of cultural distance (e.g. Padmanabhan and Cho, 1996).

Differences in national cultures create cultural distance among different countries (Tihany et al., 2005) since differences in national cultures have been proved to result in different organizational and administrative practices and employee expectations which can be expected as “the larger the cultural distance between two countries occurs, the larger the distance between their organizational characteristics on average are (Kogut and Singh, 1988:414).

(11)

market. According to O'Grady and Lane (1996), operating in a culturally (or psychically) close country does not necessarily result in superior performance as expected. Evans et al. (2000) further argue that firms are likely to conduct more extensive market research when perceiving a high level of cultural distance between the home and the host country market. Thus, it improves the comprehensive understanding of the foreign market which can be used to make more informed decisions for enhancing firm (or subsidiary) performance which implies a positive relationship between cultural distance and firm performance. Such an assumption is more applicable to the situation when MNEs adopt horizontal FDI for market-seeking concern that aims to achieve abundant market access. In other words, in non-manufacturing sector which concerns more on market-seeking approach in regard to provision of technology or service, cultural distance is likely to promote the subsidiary performance.

(12)

Hypothesis 3: Cultural distance between the home and host country has a negative effect on subsidiary performance in the manufacturing sector.

Hypothesis 4: Cultural distance between the home and host country has a positive effect on subsidiary performance in the non-manufacturing sector.

2.3. Institutional Distance

Institutional distance is widely considered as an important determinant of firm structure and performance since in order to survive, firms must conform to the rules and belief systems prevailing in the environment (Scott, 1995; Kostova, 1996; Kostova and Zaheer, 1999). Gaur et al. (2007) argue that dissimilarities within the regulative, normative, and cultural-cognitive aspects of national institutional environments (institutional distance) between the home and host country do influence subsidiary legitimacy, coordination, control, and knowledge management and thus have a significant impact on subsidiary performance. According to Kostova and Zaheer (1999), it appears that the large institutional distance between the host and home country is, the more difficult it is for an MNE to establish legitimacy in the host country as well as to transfer strategic routines to foreign subsidiaries. Bartlett and Ghoshal (1989) further suggest that a large institutional distance may lead to the conflicting demands for local responsiveness in the host country and global integration within the organizational system of the MNE which may hinder the overall firm performance.

Hypothesis 5: Institutional distance between the home and host country has a negative effect on subsidiary performance.

Except for the cultural, economic and institutional distance indicators, there are also other important firm-level indicators which should not be neglected in regard to this research.

2.4. Firm-level variables

(13)

subsidiary’s experience in the local environment can generate valuable knowledge to its parent. Therefore with greater subsidiary experience, there is a larger local knowledge base for a subsidiary to share with its peers and corporate headquarters in order to achieve a higher level of organizational performance (Makino and Delios, 1996).

Last but not least, many previous studies have shown a positive relationship between firm size and performance. A firm’s performance is highly dependent on its profit rate and thus the larger the firm size, the higher is the profit rate which leads to a more satisfying firm performance (Hall and Weiss, 1967; Shepherd, 1972; Punnose, 2008).

Hypothesis 6: Subsidiary experience in a host country has a positive effect on subsidiary performance.

Hypothesis 7: Firm size of a subsidiary has a positive effect on its performance.

(14)

Figure 1. Relationship between Psychic distance and subsidiary performance of MNEs. Subsidiary performance of MNEs Difference on embeddedness: informal institutions, customs, traditions, norms, religions

Cultural distance Economic distance

Difference on factor costs Institutional

distance

(15)

3. METHODOLOGY

3.1. Data Collection

In order to test the above hypotheses, a sample containing a set of balanced panel data from year 2000 to 2007 in regard to companies related to my study is selected in 2010 among internationally experienced Western European enterprises that have invested in Central and Eastern European (CEE) countries. From the AMADEUS database, initially registered firms based in the 14 original member countries2 _{of the European}

Union which their ultimate owners reported at least a 10% ownership3 _{stake in a}

subsidiary and also operate in any of the following transition economies that include: Bulgaria, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, Slovakia and Slovenia are targeted. These countries were chosen for my research because they had all experienced various series of transformation from centrally-planned to market economies since 2003 and sequentially formed accession to the European Union from 2004 to 2007. Also I dedicate to search for companies with different industry scope in order to increase the representativeness of my sample. Thus among thousands of the companies that meet the above-mentioned conditions, 42 firms are randomly selected for my research.

Since cultural and economic distance as well as the specific firm-level variables in this study may posit different impacts on firm performance of subsidiaries related to different industry sectors, I classify these initially selected firms into two main categories: manufacturing and non-manufacturing type. As manufacturing subsidiaries are generally more involved in efficient-seeking approach than market-seeking approach compare to non-manufacturing firms and vice versa, it is quite necessary to separate these firms of different industrial types into two samples. In total, there are 19 manufacturing subsidiaries varying from 7 different industry scopes while the rest 23 firms are labeled as non-manufacturing sectors from 5 industry scopes. Hence in this case, there are two specific samples consist of panel datasets for manufacturing and non-manufacturing subsidiaries respectively from 2000 to 2007. Specifically in each sample, I control for no occurrence of two or more

2 _{These countries include: Austria, Belgium, Denmark, Finland, France, Germany, Greece, Italy, Luxemburg,}

Netherlands, Spain, Sweden, Portugal and the United Kingdom.

3 _{According to most of the empirical research, a minimum stake of ownership for firms to qualify as a foreign}

(16)

firms with the same host and home country. In this case, I am able to avoid the statistical results to be problematic. The reasons and details are explained in the

section 3.5.3 namely “specific statistical model” in the following paragraph. Table 1

summarizes the number of selected companies differed by host countries and presents the number and percentage of subsidiaries per industry scope.

Table 1

Panel A - Number of selected companies by host countries

Host country Total Poland 11 Czech Republic 6 Bulgaria 5 Estonia 4 Hungary 4 Romania 4 Latvia 2 Slovakia 3 Lithuania 1 Slovenia 2 Total 42

Panel B - Number and percentage of subsidiaries per industry scope

Manufacturing Number of subsidiaries Percentage

Motor-vehicles

5 26.3%

Wood, pulp and

(17)

mineral products 2 10.5% Machinery and electronic equipment 5 26.3% Chemical products (no pharmaceuticals) 1 5.3%

Gas and refined petroleum oil products 4 21% Total 19 100% Non-manufacturin g Wholesale and retail 12 52.2% Telecommunicatio n 5 21.7% Transportation support service 2 8.7% Construction of roads and motorways 3 13.1% Other 1 4.3% Total 23 100% Sub-total 42 3.2. Selection bias

In most studies that require data selection for construction of samples, it is possible for selection bias to occur. According to Heckman (1979), selection bias refers to a statistical bias occurs when there is an error in selecting the individuals or groups that may distort a statistical analysis in a scientific research. In other words, selection bias mainly results from method of sample collection and reduces the representativeness of the research findings.

(18)

all, as these companies are selected from the Amadeus database with a potential selection effect on profitability level and firm size. In detailed, the Amadeus database does provide firm-level data mostly on relatively large and successful enterprises instead of SMEs and bankrupted ones. Therefore it may allow selection effect to bias the representativeness of my statistical findings to companies in the real business field as all of the above-mentioned types of enterprises should be taken into account in the reality. Secondly, due to the limitation on data availability of the general industry characteristics in both manufacturing and non-manufacturing sectors4, it is possible that the percentage of industry scopes shown in Table 1 for both sectors regarding my selected samples does not preciously reflect the real industry situation. For instance, subsidiaries focusing on wholesale and retail scope count for 52.2% in the sample of non-manufacturing firms in my research. However it may be overestimated or underestimated in the real cases. Thus the representativeness of my samples may be affected by selection bias under such situations.

3.3. Variables

3.3.1. Dependent variable

The dependent variable is the subsidiary performance of MNEs in the host countries. Returns on total assets (ROTA) is used as the subsidiary performance measure in this study since it has been widely suggested or used as a measure of firm performance from many previous studies (e.g. Staw and Epstein, 2000; Wan and Hoskisson, 2003; van Dyck et al., 2005). Roughly, return on total assets refers to the ratio of a company’s earnings before interest and taxes (EBIT) divided by the total net assets (TNA) over the year. According to van Dyck et al., (2005), return on total assets is a measure of operating efficiency of a firm which reflects the long term financial strength of the organization. The reason I use earnings before interest and taxes (EBIT) instead of net income divided by total net assets as an indicator of firm performance is because taxation rules as well as capital structure are likely to differ among countries, and thus EBIT/TNA is more ideally used as a performance measure in studies involving cross-country analysis. The ROTA of each subsidiary firm relevant to the study is adopted from the AMADEUS database from 2000 to 2007. In case of the occurrence of missing data, the missing value for an item will be adjusted by adopting the index from the previous or latter year within the same variable.

4 _{These sectors include subsidiaries of MNEs from 14 EU member countries include: Austria, Belgium, Denmark,}

(19)

3.3.2. Independent variables

The two main independent variables for my study are cultural distance and economic distance.

Cultural distance

Cultural distance is measured by the Kogut and Singh (1988) index, which is based on Hofstede’s (1980, 2001) country scores of national culture. Hofstede (2001) was primarily concentrating on uncovering differences in work-related values across countries. His research is based on questionnaires administered to thousands of employees from 53 worldwide subsidiaries of IBM. Hofstede originally identified four dimensions of culture, which are individualism, power distance, uncertainty avoidance, and masculinity which discriminate between national cultures in the workplace. These dimensions largely and broadly account for cross-cultural differences in people’s belief systems and behavior patterns around the globe.

Individualism describes the relationship between an individual and his or her fellow individuals in society. It indicates itself in the way people live together and has a great variety of value implications. At one end of the spectrum are societies with very loose ties between individuals. Such societies allow a large degree of freedom and everyone is expected to look after his or her own self-interest and possibly that of the immediate family. Societies of this type refer to high individualism and display loose integration. At the other end are societies with very strong ties between individuals. Everyone is expected to look after the interests of their in-group and to hold only those opinions and beliefs sanctioned by the in-group which in turn, protects the individual. These societies can be described as “collective” which show tight integration. Hofstede (1980) found out that there is a strong correlation between a country’s degree of individualism and its prosperity. In general, wealthy countries with relatively high GNP per capita are individualistic while poor countries are collectivistic. Exceptions include the East-Asian countries like Japan, South Korea and Singapore whose cultures are much more collectivistic than those of most Western countries, but whose living standards are relatively high.

(20)

distance with high individualism, while relatively poor developing countries tend to rate high on both collectivism and power distance. Hofstede (1980) found that Latin European, Latin American, Asian and African countries generally had a large power distance, while Northern Europe and Anglo-Saxon countries, in contrast, scored low on this dimension.

Uncertainty avoidance reflects how a society deals with uncertainty about the future which is a fundamental fact of human existence. At one extreme, cultures which are considered as weak uncertainty avoidance socialize members to accept and handle uncertainty. People in such cultures tend to take risks rather easily and do not tolerate opinions and behavior different from their own. To prevent uncertainty societies set up laws and rules like companies do. Duties and rights (internal and external) are controlled by authorities. Some cultures need to have strong uncertainty avoidance like France. In France many strict regulations are used and tasks are heavily centralized in companies. For meetings it is important to consider that. There will be a much higher demand for details when creating a contract. This is to avoid any circumstances which could cause any kind of uncertainty for French business people. Organizing is therefore rather inflexible concerning changes which occur in business life. Hofstede (1980) found that Latin European, Latin American and Southeastern countries had particularly high uncertainty avoidance scores, while German-speaking countries had moderately high scores, and Asian, African, Anglo Saxon and European countries generally scored medium to low on this dimension.

Masculinity deals with the degree to which societies subscribe to the typical stereotypes which are associated with men and women. Masculine values stress making money and the pursuit of visible achievements. Such societies admire individual brilliance and idolize the successful achievement. These traditional masculine social values permeate the thinking of the entire society, women as well as men. Hofstede’s research indicated that within his sample, Japan, the U.S., the Germanic countries ranked highest in masculinity, while the Scandinavian countries and the Netherlands are predominantly feminine (Hofstede, 1980).

Using a multi-stage procedure, Hofstede (1983) assigned each country a score on each cultural dimension that varied between 0 and 118 while zero refers to the lowest degree and 118 refers to the highest degree of each dimension. I was able to obtain the scores for both the home and host countries relevant to my research. These countries and their correspondent scores are listed in Table 2.

(21)

,

where CDj stands for the cultural distance between the jth country and the host country, Iij is the index of the ith cultural dimension (e.g., individuality, power distance, masculinity–femininity, and uncertainty avoidance), Iih is the cultural dimension index for the multinational firm’s home country, and Vi is the variance of the index in the ith dimension for all home and host countries relevant to the study. With a combination of the data in Table 2 and the formula, I was able to derive the cultural distance index between the selected home and host country. As the Hofstede’s national cultural index for each country are constant over time, cultural distance that based on Hofstede’s cultural dimensions in order to reflect the differences on national culture between two countries is also regarded time-invariant. In other words, the cultural distance index for each country relevant to the research is constant from 2000 to 2007.

Table 2

Cultural indices5 by country

Country Power Distance Uncertainty Avoidance Individualis m Masculinity Austria 11 70 55 79 Belgium 65 94 75 54 Denmark 18 23 74 16 Finland 33 59 63 26 France 68 86 71 43 Germany 35 65 67 66 Greece 60 112 35 57 Italy 50 75 76 70 Luxemburg 40 70 61 50 Netherlands 38 53 80 14 Portugal 63 31 27 104 Spain 57 86 51 42 Sweden 31 29 71 5 United Kingdom 35 35 89 66

5 _{Source: Hofstede, G. (2001), Culture’s Consequences: International Differences in Work Related Values, 2001}

(22)

Bulgaria 70 85 30 40 Czech Republic 57 74 58 57 Estonia 40 60 60 30 Hungary 46 82 80 88 Latvia6 ₄₄ ₆₃ ₇₀ ₉ Lithuania7 45 67 50 65 Poland 68 93 60 64 Romania 90 90 30 42 Slovakia 104 51 52 110 Slovenia8 ₇₁ ₈₈ ₂₇ ₁₉ Economic distance

Ghemawat (2001) defines economic distance as the level of economic development of the host country relative to that of the home country. He further argues that differences in the levels of economic development between two countries often reflect gaps in factor costs influencing firm performance of MNEs. According to Makino et al. (2002), MNEs tend to choose their FDI location between developed and developing countries from the perspective of resource-exploitation mainly in labor concern. Such perspective enables MNEs to possess firm-specific advantages so as to operate abroad successfully depending on these advantages in regard to market-seeking and efficient seeking (or achieving cheap labor resources) concerns. Therefore, host countries which are less developed than MNEs’ home countries are more likely to provide the opportunity to exploit their firm-specific advantages. Compare to the home countries, these transition economies are likely to have much lower factor costs, such as wage rate and rent (Dunning, 1993). Hence it is not so difficult to understand that MNEs from developed economies are likely to enter and operate in less developed countries with a large (economic) distance with regard to labor costs for ultimately maximizing their operating profitability.

According to Almeida and Carneiro (2006), labor costs reflect the expenditure incurred by employers in order to employ personnel which include employee compensation (e.g.: wages, salaries, social security contributions etc.), vocational costs and other expenditure such as recruitment costs. Thus labor cost statistics

6 _{Source: Huettinger, M. (2008). Cultural dimensions in business life: Hofstede's indices for Latvia and Lithuania,}

Baltic Journal of Management, 3(3), 359-376.

7 _{Source: Huettinger, M. (2008). Cultural dimensions in business life: Hofstede's indices for Latvia and Lithuania,}

Baltic Journal of Management, 3(3), 359-376.

8 _{Source: Zagorsek, H., Jaklic, M. and Stough, S. (2004). Comparing Leadership Practices Between the United}

(23)

demonstrate a detailed explanation on the level, structure and short-term development of labor costs.

In this study, economic distance is measured by average hourly labor costs as the main indicator which refers to total labor costs divided by the corresponding number of hours worked (Eurostat, 2009) and is defined as the nominal differences on average hourly labor costs between the home country and the host country in regard to an MNE. The data is provided by European Union based on a four-yearly Labor Cost Survey (LCS) that indicates detailed information and data on the structure and the level of labor costs, hours worked and hours paid over a 12-year period from 1997 to 2008 in regard to the EU member countries. Initially I attempt to make a distinction between average hourly labor costs between manufacturing and non-manufacturing sectors. However due to the limitation on finding the relevant data, I adopt the data provided by Eurostat (2009). As the data classified by economic activity and enterprise size enable the LCS to cover the economic activities of both industry and service sectors (excluding public administration), no further distinction will be made for average hourly labor costs between the manufacturing and non-manufacturing sectors in this research. Table 3 summarizes the data on average hourly labor costs by country targeted for my study over the period from 2000 to 2007.

(24)

Table 3

Panel A – Average hourly labor costs by EU countries (in Euros)

EU Country 2000 2001 2002 2003 2004 2005 2006 2007 Austria 23.05 23.65 24.13 24.98 25.32 26.23 26.96 27.61 Belgium 26.61 27.89 29.17 29.58 30.3 30.62 31.43 32.56 Denmark 26.53 28.54 29.06 30.3 30.7 31.98 33.09 34.74 Finland 22.1 23.59 23.82 24.78 25.34 26.7 27.2 27.87 France 24.84 26 27.04 27.68 28.46 29.29 30.25 31.24 Germany 25 25.6 26.2 26.8 26.9 27.1 27.6 27.8 Greece 10.98 11.62 12.46 13.37 14.28* 15.25* 16.29* 17.39* Italy 18.99 19.27 19.99 20.64 21.39 22.04* 22.7* 23.38* Luxembu rg 24.48 25.39 26.21 27.02 29.97 31.1 31.98 33 Netherla nds 22.31 23.88 25.19 26.45 27.23 27.41 28.57* 29.77* Portugal 8.13 8.6 9.1 9.6 10.2 10.6 10.97 11.32 Spain 14.22 13.07 13.63 14.21 14.76 15.22 15.77 16.39 Sweden 28.56 27.41 28.73 30.43 31.08 31.55 32.16 33.3 United Kingdom 23.71 24.51 25.24 23.56 24.71 24.47 25.51 26.39

Note: * adjusted value to the missing data

Panel B – Average hourly labor costs by CEE countries (in Euros)

(25)

Institutional distance

Institutional distance is a composite measure obtained from the World Bank’s Governance Indicators which are also known as Kaufmann governance indices. Based on separate sources of subjective data on perception of governance, Kaufmann (2005) has constructed six indicators which provide a score on items such as “voice and accountability” (measuring the political process and includes the independence of media), “political stability” (measuring the likelihood that the government will be overthrown by unconstitutional interference and the stability of the economic environment), “government effectiveness” (measuring the quality of the public service provision, the bureaucracy, the competence of civil servants and the ability of the government to formulate and implement good policies), “regulatory quality” (measuring the degree of overall regulation of business development and the incidence of market-unfriendly policies), “rule of law” (measuring the quality of the legal system and the enforceability of contracts) and “control of corruption” (measuring the degree to which public power is exercised for private gain). The six governance indicators are measured in units ranging from about -2.5 to 2.5, with higher values corresponding to higher institutional advancement. These six indicators provide a very good sketch of the quality of national institutions. Also, the indicators allow testing of hypotheses considering cross-country differences in governance in CEE region. Moreover, these indicators encompass the broadest range of institutional issues and years of measurement and the governance indicators measure six institutional dimensions, cover the 1996–2008 period of observation, and are updated once every two years before 2002 and once a year after 2002 (Kaufmann et al., 2009). This implies that these aggregate estimates are informative about changes over time in the relative institutional positions of individual countries. In order to provide a score on institutional distance for a specific host country, I first sum up all scores in six institutional dimensions for the specific host country and derive the average score out of six and adopt it as the standard score for evaluation and analysis. Since the Kaufmann governance index (Kaufmann et al., 2009) does not provide data on the six indicators for each country in 2001 which is relevant to my research, I refer to the scores for 2001 as the mean of scores for 2000 and 2002. In this study, institutional distance is defined as the nominal differences on the average scores of Kaufmann governance index from 6 dimensions regarding the home and host countries in a particular year. Table 4 summarizes the average scores for each country relevant to the research from 2000 to 2007.

Table 4

(26)

EU Country 2000 2001 2002 2003 2004 2005 2006 2007 Austria 1.65 1.66 1.66 1.6 1.61 1.57 1.59 1.63 Belgium 1.37 1.44 1.51 1.48 1.43 1.35 1.35 1.36 Denmark 1.73 1.76 1.8 1.83 1.88 1.8 1.82 1.85 Finland 1.87 1.9 1.93 1.95 1.96 1.92 1.92 1.85 France 1.24 1.23 1.21 1.22 1.28 1.23 1.21 1.16 Germany 1.63 1.61 1.58 1.47 1.47 1.48 1.52 1.52 Greece 0.79 0.8 0.81 0.8 0.78 0.69 0.66 0.61 Italy 0.89 0.86 0.84 0.8 0.75 0.61 0.57 0.56 Luxembu rg 1.83 1.86 1.89 1.81 1.84 1.74 1.73 1.8 Netherla nds 1.85 1.81 1.77 1.72 1.73 1.65 1.62 1.67 Portugal 1.17 1.23 1.29 1.26 1.18 1.14 1.02 1.01 Spain 1.32 1.3 1.27 1.24 1.14 1.1 0.92 0.92 Sweden 1.77 1.77 1.78 1.82 1.83 1.7 1.71 1.8 United Kingdom 1.65 1.62 1.58 1.53 1.56 1.43 1.55 1.52

Panel B – World Bank’s Governance Indices by CEE countries

CEE Country 2000 2001 2002 2003 2004 2005 2006 2007 Bulgaria 0.14 0.21 0.27 0.23 0.25 0.24 0.22 0.23 Czech Republic 0.65 0.76 0.87 0.86 0.78 0.85 0.87 0.81 Estonia 0.84 0.89 0.94 1.03 1.03 0.99 1.05 1.03 Hungary 0.91 0.96 1 0.95 0.95 0.9 0.9 0.81 Latvia 0.44 0.52 0.61 0.74 0.66 0.67 0.71 0.66 Lithuania 0.51 0.6 0.7 0.81 0.76 0.74 0.69 0.69 Poland 0.65 0.66 0.66 0.65 0.51 0.54 0.47 0.49 Romania -0.09 -0.04 0.01 -0.02 0.03 0.04 0.14 0.13 Slovakia 0.46 0.53 0.59 0.69 0.72 0.77 0.75 0.68 Slovenia 0.88 0.93 0.99 1 0.99 0.92 0.96 0.94 3.3.3. Control variables

(27)

Firm size has been broadly argued to have a positive impact on performance from previous studies. For instance, Baumol (1959) initially hypothesizes that the rate of return is positively related to the size of the firm. Later other researchers also observed such relationship by arguing that larger the firm size, higher is the profit rate (Hall and Weiss, 1967; Gale, 1972; Punnose, 2008). Hence, a control variable firm

size is included and is measured by the total number of employees expressed in

full-time equivalents within the subsidiary (Power and Reid, 2003).

According to Oliver (1997), years of experience in a specific subsidiary sector from a host country is expected to exert a substantial impact on subsidiary performance. In order to control for such experience, a variable subsidiary experience is created by summing up the years of existence since the establishment of the subsidiary.

Besides experience in host countries, international experience of MNEs has also an influence on performance. Padmanabhan and Cho (1999) argue that the lack of international experience in regard to a multinational firm will posit a negative effect on its subsidiary performance. Thus international experience is also included as a control variable measured by the number of subsidiaries that have been established in countries worldwide in regard to the MNEs.

Last but not least, all the data regarding all the control variables are achieved from the AMADEUS database in regard to the relevant subsidiary from 2000 to 2007. The value of missing data for each variable is adjusted by adopting the index from the previous year or latter year.

3.4. Diagnostic Testing

As an integral part of model specification in econometrics, diagnostic testing is very important and should not be neglected in regard to accuracy and reliability of statistical results (Wooldridge, 2002). Thus several relevant diagnostic tests are conducted with results presented in this section. Starting with tests of normality of the residuals, heteroskedasticity, autocorrelation and multicollinearity, the remainder of the section further explains the statistical results from running these tests.

3.4.1. Normality of the residuals

(28)

which the error terms have an expectation of zero, and are uncorrelated as well as with equal variances, the best linear unbiased estimator (BLUE) of the coefficients is provided by the OLS estimator (Wooldridge, 2002). In other words, unbiased coefficient estimations in a linear regression model should be based on the one of the assumptions that regression errors as well as values for the dependent variable are normally distributed. The Central Limit theorem further suggests that if Gauss-Markov assumptions hold, and if the sample size is sufficiently large, then the least squares estimators have an approximate normal distribution (Hill et al., 2007). Given only the fact that both the two samples in my study contain over 150 observations which are sufficiently large, it is difficult to judge whether non-normality problem of the error terms occurs without a statistical test. Hence statistical software Stata is used to provide a statistical check on normality of error terms regarding my samples. The test is conducted with two steps.

In the first step, two histograms of residuals are created for both samples regarding manufacturing and non-manufacturing firms respectively.

Histogram 1 – Residuals of sample for manufacturing firms

0 1 0 2 0 3 0 P e rc e n t -40 -20 0 20 40 Residuals

(29)

0 5 1 0 1 5 2 0 2 5 P e rc e n t -20 0 20 40 Residuals

As we can see from both diagrams, the residuals for both samples are centered at zero due to the fact that mean of the least square residuals should be equal to zero so that the regression has an intercept (Wooldridge, 2002). Also the statistical results are reasonably bell-shaped which may suggest a normal distribution.

In the second step, the Jarque-Bera test (which is actually a chi-square test) is conducted for formally checking the normality of residuals. The null hypothesis (H0)

is formed as the residuals are normally distributed. The Jarque-Bera test is based on measures of two indicators namely skewness and kurtosis. If the value of skewness is reported zero, it suggests perfect symmetry of residuals. When the value of kurtosis is equal to 3, it then indicates perfect normal distribution of error terms. For the sample of manufacturing firms, the estimated value of skewness is -0.16 and the value of kurtosis is 5.06. While for the sample of non-manufacturing firms, the estimated value of skewness is 0.66 and the kurtosis index is shown as 5.30. In this case, the Jarque-Bera statistic for the sample of manufacturing firms is calculated to be 26.23 and the one for the sample of non-manufacturing firms is 53.91 which do not suggest normal distribution of the error terms for both samples. In both cases, the value of the calculated Jarque-Bera statistic is greater than the critical value (5.99) of chi-squared distribution with two degrees of freedom obtained at a 5% significant level. Thus there is sufficient evidence from the residuals to conclude that the normal distribution assumption is unreasonable at the 5% significant level and the null hypothesis should be rejected in regard to both two samples.

(30)

the coefficients in relate to this study. As non-normality of residuals is mainly resulted from issues of heteroskedasticity and autocorrelation, in the following section, I conduct tests to check whether these problems do exist regarding my datasets.

3.4.2. Heteroskedasticity

In statistics, heteroskedasticity occurs when a sequence of random variables have different variance. One of the assumptions within the ordinary least square (OLS) estimator suggests a constant variance to the error terms which is also known as homoscedasticity (Wooldridge, 2002). Heteroskedasticity often occurs in panel data regression analysis which contains both cross-sectional and time-series measurements as the error terms could vary or increase with each observation (Wooldridge, 2002). Although heteroskedasticity may not bias coefficient estimates of OLS, it can lead to an underestimation of the variance and thus of the standard errors of the coefficients. In this case, a regression analysis will still derive valid estimates of the relationship between the predictors and the outcome while it may judge the relationship to be statistically significant when it is actually insignificant. Thus it is quite necessary to detect whether the heteroskedasticity problem exists or it may occur and bias the statistical results to the research.

In this research, the Breusch–Pagan test is used for testing the heteroskedasticity regarding all of the explanatory variables in the linear regression model by conducting a chi-square test in regard to Lagrange multiplier value to decide whether or not to reject the null hypothesis. The null hypothesis (H0) assumes a constant variance for

the error terms. The software Stata is used for providing statistical supports for running the test. The statistical results9 for all regressors in regard to two samples relevant to the research are summarized in Table 5.

Table 5

Panel A – Results for the sample of manufacturing firms

Variable P-value LM value

cultural distance 0.248 1.34 economic distance 0.098 2.74 institutional distance 0.136 2.22 international experience 0.049 3.88 subsidiary experience 0.010 6.61

(31)

firm size 0.307 1.04

Panel B - Results for the sample of non-manufacturing firms

Variable P-value LM value

cultural distance 0.427 0.63 economic distance 0.711 0.14 institutional distance 0.862 0.03 international experience 0.390 0.91 subsidiary experience 0.225 1.47 firm size 0.613 0.25

For the sample of manufacturing firms, the null hypothesis should not be rejected to all of the explanatory variables except for the variable “international experience” and “subsidiary experience” with the p-value less than 0.05 (0.049<0.05; 0.01<0.05). For the sample of non-manufacturing firms, as the p-value of all explanatory regressors exceeds 0.05, the null hypothesis should not be rejected for all of the explanatory variables to the sample. In other words, heteroskedasticity does occur in the sample of manufacturing firms while does not occur in the sample of non-manufacturing firms. As my samples contain both cross-sectional and time-series data, the occurrence of heteroskedasticity is possible. Thus I attempt to correct for heteroskedasticity among the explanatory variables “international experience” and “subsidiary experience” in the sample of manufacturing firms relying on the statistical software Stata. I further discuss it in the section 3.5.5. namely “robust estimate of variance”.

3.4.3. Autocorrelation

In statistics, the term “autocorrelation” defines the situation when the error terms are correlated which occurs a lot in time-series or panel data regression analysis. Under an OLS model, autocorrelation problem does not bias the coefficient estimates but may lead to an underestimation of standard errors when the autocorrelations of the error terms at low lags are positive (Priestley, 1982). Thus I should ensure the independence of the error terms of the observations in my samples. In order to examine for autocorrelation, I follow a test for panel data discussed by Wooldridge (2002) which performs a Wald test of the null hypothesis (H0) of no autocorrelation.

(32)

0.05 which rejects the null hypothesis of no autocorrelation. Therefore I try to correct for autocorrelation among the error terms with the statistical support from the software Stata. It is further discussed in the section 3.5.5. called “robust estimate of variance”.

3.4.4. Multicollinearity

Multicollinearity phenomenon occurs when there are two or more predictor variables linearly related to each other at a high degree within a multiple regression model. In this case, the coefficient estimates are likely to change unpredictably regarding small changes in the regression model or data (Hill et al., 2007). Instead of reducing the reliability of the general model, multicollinearity among regressors tend to affect calculations in regard to individual variables. In other words, multicollinearity still indicates how valid the entire set of explanatory variables predicts the dependent variable, but it may fail to provide valid results about any individual predictor (Lipovestky and Conklin, 2001). In order to detect the possible multicollinearity, a correlation analysis of the explanatory variables is presented in Table 6.

Table 6

Panel A – Correlations between variables for the sample of manufacturing firms

rota cul_dis eco_dis ins_dis exp_yrs No_sub No_em

ployee rota 1 cul_dis 0.0136 1 eco_dis -0.1333 0.2886 1 ins_dis 0.1794 0.3745 0.3728 1 exp_yrs 0.0468 -0.096 -0.4271 -0.0165 1 No_sub 0.2963 0.1719 0.1755 0.3219 -0.1716 1 No_employee -0.097 0.0432 0.2299 0.0785 -0.1467 -0.0925 1

Panel B – Correlations between variables for the sample of non-manufacturing firms

rota cul_dis eco_dis ins_dis exp_yrs No_sub No_em

ployee

(33)

cul_dis 0.0637 1 eco_dis 0.1565 0.1521 1 ins_dis 0.0521 0.3390 0.6563 1 exp_yrs 0.1815 0.0652 0.1897 -0.0138 1 No_sub -0.0123 0.0540 0.0887 0.1843 0.0603 1 No_employee -0.1100 0.0749 -0.3787 -0.0656 -0.0976 -0.0939 1

The obtained correlations are relatively low in both samples regarding manufacturing and non-manufacturing firms which suggest that the explanatory variables do not linearly relate to each other at a high degree within a multiple regression model. Thus multicollinearity problem does not occur in both of the two samples regarding my research.

3.5. Data Analysis

3.5.1. Panel Data Estimation Methods

In order to explore the relationship between the independent or other explanatory variables with the dependent variable which in this case is subsidiary performance, a proper statistical regression analysis is required. In this research, I include two samples which both contain a sequence of cross-sectional units (selected firms) observed in a number of time periods (from year 2000 to 2007) and therefore a panel data regression method should be implemented for the statistical analysis.

(34)

which are all considered as if the quantities are not random. The random effects model however is totally the opposite which treats either the entire or part of the explanatory variables as if they are random (Wooldridge, 2002).

Theoretically, the OLS estimator appears to be inconsistent when the phenomenon of multicollinearity, heteroskedasticity, autocorrelation and non-normality of the residuals as these terms violate the basic assumptions of the OLS model which have been mentioned in the section 3.4 which is the diagnostic testing section. Empirically, many previous studies have suggested the pitfalls for the OLS model to deal with panel data analysis; for instance, the presence of heteroskedasticity will cause inefficient and biased estimates of the variance covariance matrix within the OLS framework (e.g. Arellano et al., 1991; Wooldrigde, 2002). As the previous section in regard to the diagnostic check has indicated the occurrence of heteroskedasticity, autocorrelation and non-normality of the residuals regarding my samples to the research, the OLS estimator seems not to be the most ideal regression model for this research. An alternate solution is to use an estimator for controlling bilateral specific effects implemented by a fixed effects model (FEM) or a random effects model (REM). Statistically choosing between a fixed effects model and a random effects model, two common assumptions should be made in regard to the individual specific effect: the fixed effects assumption and the random effects assumption. The former term suggests that the individual specific effect is correlated with the independent variables and the latter term indicates that the individual specific effects are uncorrelated with the independent variables. If the fixed effects assumption holds, then the fixed effects model is more efficient than the random effects model. But if this assumption does not hold, the fixed effects model is inconsistent and thus the random effects model is more proper. The following section discusses the two estimators and provides a statistical test (Hausman-test) to determine which estimator to adopt for this research.

3.5.2. Fixed and Random Effects Estimators

(35)

company dummy and the time dummy into the model, I only preserve the time dummy which should generate the same estimates as from the general fixed effects model with the company dummy excluding the time-invariant variables.

The revised fixed effects model can be presented as:

y

ti

=

β

k

X

kti

+ α

t

+ u

ti

(1)

) + (u

ti

-u

_ ti

) (4)

As we can see from equation (4), after the fixed effect transformation, the unobserved individual effect αtfalls out and thus it should yield unbiased and consistent results. The random effects model can be presented as same as the equation (1) assuming αt uncorrelated with each explanatory variable:

(36)

3.5.3. Specific Statistical Model

The specific model in regard to this study for exploring the impact of the independent and control variables on subsidiary performance in a given host countries is:

Y

ti

where i refers to the cross-sectional dummy (company), t refers to the time-series dummy (year), tis the intercept, β₁…β₆are the regression coefficients, ε_itis the error

term. In addition Y, X1…X6denote the dependent, independent and control variables.

In detailed:

Dependent variable

Y: subsidiary performance (rota)

Independent and control variables

X1: cultural distance (cul_dis)

X2: economic distance (eco_dis)

X3: institutional distance (ins_dis)

X4: subsidiary experience (No_sub)

X5: international experience (exp_yrs)

X6: firm size (No_employee)

As I mentioned in the section 3.1 namely “data collection”, I ensure each firm to have identical host and home country in both of the two samples. Related to the equation (6), suppose there are two firms (i and j) with the same host and home country, then the value of cultural distance, economic distance and institutional distance are identical for the two different firms. Then it occurs the situation when

X

1it =

X

1jt,

X

2it

=

X

2jt,

X

3it =

X

3jt for all the t periods and thus biases the statistical results of

coefficients of the X1, X2and X3variables.

3.5.4. Hausman-test

(37)

hypothesis (H0) is not rejected, then in large samples the random and fixed effects

estimates should be similar. Otherwise, rejection of the null hypothesis claims that there is a correlation between the error term and at least one of the explanatory variables. Under such circumstance, the random effects estimator is inconsistent while the fixed effects estimator is still consistent and efficient.

By using the software Stata, the Hausman-test is conducted. The results are shown in the Appendix IV. For the sample of manufacturing firms, the value of the test statistic F= 17.32 which yields a p-value of 0.0017; and for the sample of non-manufacturing firms, the value of the test statistic F = 12.72 yields a p-value of 0.0127. Hence the null hypothesis which assumes no correlation between the error term and any of the explanatory variables should be rejected due to the significant p-values (0.0017 < 0.05; 0.0127 < 0.05) in regard to both of my samples. In sum, judging from the results of the Hausman-test, the fixed effects estimator should be adopted since it is consistent and reliable for the statistical analysis for both of the two samples in regard to manufacturing and non-manufacturing firms respectively.

3.5.5. Robust estimate of variance

(38)

outcome of the cluster-level scores and within each cluster, the cluster-level score is achieved by the sum of the entire observation-level scores. In other words, even when the errors are both heteroscedastic and autocorrelated, the cluster robust estimator upgraded by Rogers (1993) is still consistent. As both heteroskedasticity and autocorrelation occur to my datasets, the cluster robust estimator is adopted to robust standard errors. In Stata, the command vce (cluster clustvar) is implemented for conducting the cluster robust estimation in regard to my research.

3.6. Results

In order to correct for heteroskedasticity and autocorrelation problems, robust statistical results are adopted in regard to both the two samples in this research.

3.6.1. Manufacturing sector

The statistical results10 for manufacturing firms are summarized in Table 7. The result indicates a significant coefficient for the independent variable “cultural distance” (cul_dis). The score of the parameter is -0.68 which suggests a negative relationship between the cultural distance and subsidiary performance. Also the coefficient of the independent variable “economic distance” appears to be significant and is with a value of -13.19 which posits a negative impact on subsidiary performance. In other words, both cultural distance and economic distance between the host and home country are likely to result in barriers for achieving satisfying subsidiary performance in regard to manufacturing firms. Hence Hypothesis 3 is confirmed which suggests a negative relationship between the cultural distance and subsidiary performance in regard to manufacturing firms; while Hypothesis 1 that posits a positive relationship between economic distance and subsidiary performance regarding manufacturing sector is rejected. In addition, the coefficient for the explanatory variable “institutional distance” is significantly high which appears as 5.71. The value indicates a positive effect of institutional distance on influencing subsidiary performance which indicates the larger the distance between the host and home country’s institutional environment, the better is subsidiary performance of an MNE in the host country. Thus Hypothesis

5 which assumes institutional distance to posit a negative impact on subsidiary

performance is rejected. For the control variables, all of the coefficients appear to be insignificant and thus none of the remaining hypotheses can be confirmed.

(39)

Table 7

The fixed effects estimation (FE) robust results – subsidiary performance

Variable Value Intercept 24.76 (2.689)** Cultural distance -0.68 (0.265)* Economic distance -13.19 (1.608)** Institutional distance 5.71 (0.725)** International experience -0.07 (0.040) Subsidiary experience 0.01 (0.005) Firm size -0.09 (0.154)

Cases in the analysis 152

Overall R-square 0.148

R-square within 0.182

R-square between 0.844

Note: * significant at P=0.05; **significant at P=0.01; standard errors are remarked in parenthesis. F (6, 138) = 25.33

Prob > F = 0.0002

3.6.2. Non-manufacturing sector

For non-manufacturing firms, the FE regression results11 are listed in Table 8.The results provide no significant coefficient for either the independent variables or the explanatory variables except for the variable “firm size” (No_employee). Moreover the intercept which is the time dummy variable is also insignificant. As we can see from the results, the coefficient for “No_employee” is -0.23 which suggests a negative impact on subsidiary performance from the control variable. Thus Hypothesis 7 which suggests a positive relationship between the firm size and subsidiary performance should be rejected. In general, the statistical results show that there is no significant impact of psychic distance from the cultural, economic and institutional aspects on subsidiary performance in regard to non-manufacturing firms. Hence none of the remaining hypotheses can be confirmed in this research judging from the robust statistical results.

(40)

Table 8

The fixed effects estimation (FE) robust results – subsidiary performance

Variable Value Intercept 1.09 (3.871) Cultural distance 0.53 (1.141) Economic distance 1.33 (1.695) Institutional distance 1.31 (1.002) International experience 0.04 (0.058) Subsidiary experience -0.001 (0.001) Firm size -0.23 (0.046)**

Cases in the analysis 184

Overall R-square 0.0297

R-square within 0.0399

R-square between 0.7327

Note: * significant at P=0.05, **significant at P=0.01; standard errors are remarked in parenthesis. F (6, 7) = 27.17

Prob > F = 0.0002

(41)

performance which is the opposite as I expected. Generally it is assumed that MNEs are discouraged from operating in distant country markets with differences in legal and political (institutional) environment in regard to possibility to achieve poor performance (Alexander, 1995). In other words, firms are likely experience a high level of uncertainty when entering and operating in a foreign country with a different institutional environment. In order to reduce the uncertainty, multinationals tend to be more cautious and aim to improve their knowledge of the local foreign market which may yield a positive effect on the subsidiary performance in the long run (Evans and Mavondo, 2000).

However in the non-manufacturing sector, psychic distance in cultural, economic and institutional aspects is not found to be statistically related to subsidiary performance. Instead, only one of the control variables namely “firm size” is shown to be negatively related to subsidiary performance. It can be explained as larger firms may experience lower profit rates owing to diminishing returns to the economies of scale in marketing processing (Audretsch et al., 2002).

4. CONCLUSION