Any questions related to this research can be sent to t.miedema.6@student.rug.nl
Cultural Distance and FDI: the Influence of
Sector-Specific Contact Intensity
Timme Miedema S2962292
University of Groningen – Faculty of Economics and Business Supervisor: K. M. Wacker
Co-assessor: Dr. P. Rao Sahib Master’s Thesis: June 2020
Keywords: cultural distance, FDI, contact-intensity JEL classification: F21, F23
Abstract
2
I – Introduction
With continued improvements in communication and transportation, the global economic market has become increasingly accessible to firms. The globalisation wave that started in the late 1990s sparked an interest in international expansion among company executives, boosting the rate at which corporations set up international subsidiaries, merge with foreign partners, or acquire foreign firms. Unfortunately, as indicated by the consulting firm KPMG, many of these foreign direct investment (FDI) efforts fail to meet pre-investment expectations (Kelly, Cook & Spitzer, 1999). The international business literature has explored whether part of the explanation might lie in the degree to which the cultural values of the multinational firm’s home country are different from the cultural values of the host country (the ‘cultural distance’ between the countries). As argued by Hofstede (1980), cultural differences between individuals or organisations increase the costs and risks associated with cross-cultural contact. Consequently, these higher risks and costs could impede the performance of international investments. Datta and Puia (1995) state that firms face larger integrational challenges when acquiring target firms in more culturally distant countries, resulting in a decrease in performance. Others point out that cultural differences between multinational firms and host countries create real liabilities in which foreign subsidiaries are often opposed to- or unfamiliar with the business practices imposed by the multinational (Bailey & Li, 2015; Flores & Aguilera, 2007). These are merely examples of how cultural distance can negatively impact location-decisions of multinational firms, thus exerting a negative influence on FDI stocks between two culturally distant countries. There is a substantial body of literature confirming this, with for instance Beugelsdijk, Kostova, Kunst, Spadafora and van Essen (2018) reporting that FDI is negatively associated with cultural distance.
3 zero, indicating that when the totality of these studies is observed, the literature fails to identify a conclusive significant effect. As a potential explanation for this lack of conclusiveness, they comment that the culture-FDI relationship is more complex than previously described in the literature. Most scholars and executive managers intuitively know that cultural differences are important factors in any FDI activity, but how they matter, and under which conditions they matter is poorly understood. To get a more complete overview of the relationship, new perspectives must be explored.
By taking a sector-specific perspective, this study attempts to address this hitherto undescribed complexity. The literature has not yet empirically explored the notion that cultural distance could have a heterogeneous effect depending on the business sector. Since some sectors require a multinational firm to more actively engage with suppliers or customers, cultural distance could play a larger role in the location-decisions of firms operating in these contact-intensive sectors. Using a dataset which includes the total FDI stock of US firms in 52 partner countries, disaggregated into nine manufacturing sectors, this study findssome initial evidence that, even though cultural distance, in general, does not exert a significant influence on FDI stocks, it does play a more important (negative) role in sectors with a higher level of contact-intensity. Even though the findings lack robustness to certain changes in variable- or model specification, this suggests that researchers should be aware of potential sector-dependent heterogeneous effects of cultural distance on FDI, and ought to be careful in generalising findings from studies which utilise data from a small number of sectors (e.g. Holburn and Zelner, 2010). Furthermore, the fact that cultural distance plays a larger role in certain sectors compared to others is an important realisation for policymakers aiming to attract FDI in a particular sector. Since contact-intensive sectors generally provide the largest scope for knowledge spill-overs to local firms, authorities that aim to attract foreign investors in these contact-intensive sectors should be aware that cultural differences could pose relatively larger barriers.
4 previously mentioned lack of robustness, the potential economic magnitude should warrant future research on the topic.
The remainder of this study is structured as follows: section II discusses previous literature on the topic of cultural distance and FDI, and develops a theoretical framework describing the potential existence of sector-dependent effects. Section III elaborates on the research method and data. Section IV reports the results of the analyses. Section V discusses the main conclusions. Lastly, section VI provides the implications and suggestions for future research.
II – Literature review and theoretical framework
The term cultural distance reflects the degree to which two countries differ on a cultural level (Shenkar, 2001), and has been utilised in an abundance of research in the field of FDI and trade. In general, a clear division can be made between research in the field of economics and research in the field of business, with the economics research focussing on general trends in international investments (stocks and flows), and the business research focussing more on the relationship between cultural distance and FDI performance. However, both approaches are indirectly linked, since performance potential of FDI (the topic of business research) is a major driver of the location decisions of multinational firms (Beugelsdijk et al., 2018), and will thus be captured in the overarching trends of investment flows and stocks (the topic of economics research). Hence, this study assumes that country-specific factors which increase (decrease) FDI performance potential will translate into higher (lower) levels of FDI stock invested in that country. This is also reflected in the work by Wacker (2016), who states that activities of multinational firms are adequately proxied by FDI stocks. Consequently, arguments concerning the relationship between cultural distance and FDI performance can also be utilised as arguments regarding the relationship between cultural distance and FDI stocks. The remainder of this section will discuss previous (economics) studies and their conclusions about the cultural distance – FDI paradigm, and will elaborate on the potential underlying reasons of their findings by focussing on performance-related arguments brought forward in the business literature. Lastly, the potential presence of sector dependent effects will be discussed.
The relationship between cultural distance and the level of FDI stock
5 156 papers from various fields of research (including economics, international business, strategy, and finance), Beugelsdijk et al. (2018) show that cultural distance has a heterogeneous effect depending on the stage of company internationalisation, and that a firm’s location decision (the first stage of internationalisation) is negatively affected by cultural distance. This finding is confirmed by Holburn and Zelner (2010), who report a strong negative relationship between cultural distance and multinational enterprise (MNE) location decisions, based on data concerning the electric power generation industry spanning the decade between 1990 and 1999. This indicates that, ceteris paribus, culturally distant countries will be characterised by a relatively lower level of (bilateral) FDI stock1. There are many underlying reasons for this finding, but the core of the argument revolves around the fact that a large cultural distance makes many aspects of doing business more complicated. In a way, this is also the most intuitive and straightforward way of looking at inter-cultural business. A lack of shared values, norms, and business approaches will disrupt effective communication and coordination, and will impede the performance of inter-cultural business efforts. As reported by Luo (2002), the very unique and rare characteristics of cultures complicate the process of acquiring, building, and exploiting resources in culturally distant countries. This originates from the fact that investing firms are more likely to be unaware of subtleties in local cultures and business practices. This lack of familiarity with the local culture reduces the ability of multinational firms to identify and understand the functional attributes and benefits of the local knowledge (Reus & Lamont, 2009). Gradually developing an understanding of these functional attributes and benefits, for example through interactions with local firms, is not likely to be facilitated by a large cultural distance either, since the understanding and dissemination of important knowledge throughout a network of firms is largely dependent on a shared set of values and a similarity in the approach to business (Dhanaraj, Lyles, Steensma & Tihanyi, 2004). Additionally, in line with arguments made by Lane, Greenberg and Berdrow (2004), employees of a firm generally prefer to interact and communicate with members from cultures that are similar to their own, lowering the possibility that multinationals can learn from local firms or individuals. This literature thus lends itself to the conclusion that doing business in culturally distant countries complicates the identification and exploitation of important knowledge, whilst also reducing the potential to learn these identification and exploitation skills over time.
6 Besides its negative influence on communication and understandability, a more complex cultural landscape will impede firm-wide efficiency by decreasing the effectiveness of company-wide implementation of certain technology or processes (Gomez-Mejia & Palich, 1997). To illustrate this, Heiko (1989) states that the inventory management method of Just-In-Time has been very popular and successful in Japan, but has failed to lead to significant gains in other countries. He attributes this to the fact that Japanese culture, with much concern for space and a preference for visual illustrations, provides a context that is perfect for the Just-In-Time approach. Implementation of this approach in other countries has been significantly less effective, stressing the role of cultural differences in the success of transferring internal approaches or processes to foreign subsidiaries. A larger cultural distance between countries will increase the likelihood that MNEs need to customise their technology and processes according to the host country’s cultural characteristics, and will lead to a lower level of company-wide efficiency, productivity, and performance. A similar argument can be made concerning certain aspects that are not internal to the firm, but are dictated by the market for the firm’s products. For instance, Bartlett (1986) indicates that cultural differences can create different consumer preferences, forcing firms to adapt their marketing and product promotion strategies, making it harder to share market activities and know-how amongst divisions in culturally distant countries. Entering culturally distant markets therefore decreases the possibility that firms can implement company-wide initiatives related to both internal- and market-focused processes, giving an incentive to avoid these culturally unrelated markets2.
Nevertheless, despite the arguments made before, many studies fail to find a significant negative effect of cultural distance on FDI, with a small sub-set of studies actually reporting a positive relationship (e.g. Delios, Gaur & Makino, 2008). As described by Tang (2012), it is not evident whether the notion that firms avoid culturally distant markets is a false stereotype or an actual managerial practice. By estimating an Ordinary Least Squares (OLS) regression based on bilateral FDI flow data concerning mainly OECD countries, Tang fails to find strong evidence for any effect of cultural distance on FDI. This is in line with conclusions drawn by Li and Guisinger (1992), who find that the relevance of cultural distance in FDI decisions is diminishing. By using data from two time periods, 1976-1980 and 1980-1986, they report that the initial negative effect of cultural distance on FDI is absent in the subsequent period. They
2 Even when firms do decide to confront the challenges described above, and enter culturally distant markets, their mode of
7 attribute this to the global trend towards more homogenous and integrated markets3, as well as to the acculturation process that is prevalent among managers. Training, experience, and improved technological communication processes have helped to break down some of the barriers to entering culturally different markets. A similar remark is made by Shenkar (2001), who states that easier communication methods and more international interaction bridge cultural distances, and encourage the convergence of cultural values4. This development illustrates that the relationship between cultural distance and FDI might be evolving, and that new research is necessary to empirically document this development.
With an abundance of literature failing to find a significant effect, there must be certain advantages for MNEs when locating in culturally distant markets that balance the negative implications described previously. Reus and Lamont (2009) find support for the hypothesis that cultural distance can result in significant performance gains if the firm is able to overcome the initial negative effects of complications in communication and knowledge identification. This is also stressed by Stahl and Voigt (2008), who report that differences in culture can lead to value creation and learning in international mergers and acquisitions. By locating in culturally distant countries, firms have the opportunity to learn from routines and approaches that are unique to that specific country (Kim, Hwang & Burgers, 1993; Reus & Lamont, 2009). However, this does not necessarily translate into immediate benefits for the firm, since the degree to which companies can benefit from this exposure largely depends on their integration capabilities. Nevertheless, for firms with adequate integration capabilities, increasing cultural distance will expand the ‘combination potential’ (Larsson & Finkelstein, 1999) of these firms, leading to significant performance gains. This is also mentioned by Beugelsdijk et al. (2018), who state that firms struggle more to transfer business practices to culturally distant markets, but that those which manage to do so experience significantly more benefits.
The learning benefits of locating in culturally distant countries might be particularly relevant for firms from emerging economies, which might actively target more developed markets to learn new and advanced business practices. This is illustrated by the observed trend where firms from emerging Asian countries invest heavily in Western markets in order to be surrounded by technology centres, demanding customers and strong competitors (Beugelsdijk et al., 2018;
3 Their research solely covers the triadic markets (North America, Europe and Japan), but their remarks concerning the trend
towards market homogeneity are very relevant in an increasingly globalised world.
4 Even though the studies by Li and Guisinger (1992) and Shenkar (2001) are relatively old, the trends of rapid innovations in
8 Guillén & García-Canal, 2009). Evidently, there is no indication that large cultural distances are the reason behind these investment decisions of Asian firms, but it does show that issues related to cultural distance can be outweighed by other considerations. Furthermore, Delios, Gaur and Makino (2008) find that for Japanese firms during the period 1980-2002, cultural distance was positively related to the location decisions of their international expansion, indicating that Japanese firms indeed targeted culturally distant locations during this period. However, since the current study takes a US perspective, the positive effect of cultural distance on FDI stocks is less likely to be found, since deliberate targeting of markets in order to learn advanced business practices is less relevant for firms already located in an advanced economy. With this in mind, and the fact that empirical evidence of a positive influence is largely absent, this study hypothesises that cultural distance mainly exerts a negative influence on US-owned FDI stocks, and that the potential positive effects merely counterbalance this to some degree. In line with this, hypothesis 1 is developed.
H1: Cultural distance negatively affects the location decision of multinational firms, and therefore, all else equal, countries that are culturally distant from the US attract a lower level of US-owned FDI stock.
9 inhabitants will consequently establish common organisational practices that characterise firms within this country (Beugelsdijk et al., 2018). This paper, therefore, argues that national culture will shape the organisational culture of firms within a specific country, and that using national cultural differences to explain organisational phenomena such as FDI, as done in the build-up to hypothesis 1, is thus a valid approach.
The case for sector dependent effects
So far, the literature has overlooked the possibility that cultural distance could have a heterogeneous effect on FDI stocks, depending on the sector. Intuitively, there are large differences in the way that business is conducted in different sectors, which evidently also holds for FDI activities. To illustrate this, a theoretical example will be proposed, which will then be supported by related literature.
10 Costa Rican market (Larrain, Lopez-Calva & Rodriguez-Clare, 2000). After the relocation of its production plant to the Central American country, it is stated that Intel maintained close relationships with over 200 local suppliers, that 35% of these suppliers engaged in interactive training programs organised by Intel, and that over 20 suppliers intensively customised their production processes to cater to Intel’s input needs. This clearly illustrates the fact that Intel’s production process is dependent on many relationship-specific inputs, and that the company therefore needs to develop and maintain strong relationships with many local actors. This need is much less pressing for ExxonMobil (in this particular example), since its activities are mostly conducted independently, without continuous inputs and customisation from suppliers. This hypothetical example therefore lends itself to the conclusion that Intel’s production process is substantially more intensive compared to ExxonMobil’s. Since a more contact-intensive production process intuitively requires more interaction with the local (business) culture, cultural distance might be a larger and more important factor in Intel’s location decision compared to ExxonMobil’s. Of course, there are many situations in which even firms like ExxonMobil require continuous customisation from local suppliers, but this example merely aims to point out that sectoral differences could play an important role in the cultural distance – FDI paradigm.
11 contact between distinct sectors, and that the influence of cultural distance could be more severe in the relatively contact-intensive sectors.
The existence of sector-dependent effects also aids in explaining why many studies have failed to find conclusive results regarding the relationship between cultural distance and FDI. Aggregating all sectors into one total FDI stock might hide some underlying effects, and even though there might be a significant influence of cultural distance on FDI in one particular sector, this can be disguised by the lack of effects or counterbalancing effects found for other sectors. Additionally, the existence of sector-heterogenous effects would mean that researchers should be careful when generalising findings from studies which utilise data from a small number of sectors (e.g. Holburn and Zelner, 2010), since a sector-heterogenous effect would indicate that these results might not be representative for all industries.In general, this study argues that the effect of cultural distance on FDI is amplified in the more contact-intensive sectors, and hence a sectoral approach to analysing FDI is needed. Consequently, hypothesis 2 is developed.
H2: The negative effects of cultural distance are amplified in contact-intensive sectors, and hence, all else equal, countries which are culturally distant from the US attract a relatively lower level of US-owned FDI stock in more contact-intensive sectors.
III – Research method and data Empirical basis: the Gravity Model
12 source country, the economic size of this country is a constant and is hence not included in the model. The gravity model which is relevant for this research can therefore be captured by the following equation:
𝐹𝐷𝐼𝑖= 𝛼0𝑌𝑖𝛼1𝐷𝑖𝛼2, (1)
where 𝐹𝐷𝐼𝑖 indicates FDI stock owned by US firms and invested in country i, 𝑌𝑖 is a measure of the economic size of country i, measured by its Gross Domestic Product, and 𝐷𝑖 indicates the geographical distance between the US and country i. Initially, the parameters 𝛼0, 𝛼1 and 𝛼2are unknown. However, as the aforementioned theory states, FDI stocks are proportional to the host country’s GDP, and inversely proportional to the distance between the US and that particular host country, indicating that 𝛼1is assumed to be positive and 𝛼2is assumed to be negative. In investigating the (sector-dependent) effect of cultural distance on FDI stocks, this study will solely use data regarding the year 2018. National cultures, sector-specific contact-intensity and FDI stocks are relatively stable over time, which means that expanding the dataset to cover multiple years will not greatly enhance the explanatory power of the model. To draw conclusions about the two hypotheses, the following equation is estimated5:
log 𝐹𝐷𝐼𝑖,𝑠 = 𝛼 + 𝛽1𝐶𝑢𝑙𝑡𝑢𝑟𝑎𝑙𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑖+ 𝛽2𝐶𝑜𝑛𝑡𝑎𝑐𝑡𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑖,𝑠+ 𝛽3𝐶𝑢𝑙𝑡𝑢𝑟𝑎𝑙𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑖× 𝐶𝑜𝑛𝑡𝑎𝑐𝑡𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑖,𝑠+ 𝛽4log 𝐺𝐷𝑃𝑖+ 𝛽5log 𝐺𝑒𝑜𝑔𝑟𝑎𝑝ℎ𝑖𝑐𝑎𝑙𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑖+ 𝛽6𝑇𝑎𝑥𝑖+ 𝛽7𝑃𝑜𝑙𝑖𝑡𝑖𝑐𝑠𝑖+ 𝛽8𝐿𝑎𝑛𝑔𝑢𝑎𝑔𝑒𝑖+ 𝛽9𝐴𝑑𝑚𝑖𝑛𝑖𝑠𝑡𝑟𝑎𝑡𝑖𝑣𝑒𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑖+ 𝛽10𝐹𝑟𝑒𝑒𝑇𝑟𝑎𝑑𝑒𝑖+ 𝛽11𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛𝑖+ 𝜀𝑖 (2)
The interaction term between contact-intensity and cultural distance reflects the main aim of this study. Contingent on the general variables of cultural distance and contact-intensity (who’s computations will be described in subsequent sections), the interaction variable will assess whether contact-intensity influences the relationship between cultural distance and FDI. As will be described later, each sector has a unique level of contact-intensity, hence this variable will also shed light on the potential existence of sector-dependent heterogeneous effects. The interaction variable will therefore be the basis of the conclusion whether cultural distance
5 Initially, the main estimation also included a measure of (log) host country population, similar to Tang (2012). However, this
13 matters more in more contact-intensive sectors, and in line with hypothesis 2, the coefficient 𝛽3 is predicted to be negative.
The dependent variable
The dependent variable of the estimation is the log of nominal FDI stock (USD millions) in sector s, owned by US firms, invested in host country i during 2018, calculated on a historical-cost basis6. The data was extracted from the Bureau of Economic Analysis (BEA), and covers nine manufacturing sectors, including mining, food, chemicals, primary and fabricated metals, machinery, computers and electronic products, electrical equipment and components, transportation equipment, and other manufacturing. All forms of FDI are included, and the dataset makes no distinction between mergers, acquisitions, joint ventures, or greenfield FDI. A list of host countries, 52 in total, is provided in appendix A. As the dataset only covers manufacturing sectors, there is a smaller scope for sector-specific contact-intensity to differ. However, as argued by Alfaro (2003), even within manufacturing there may be large differences in the level of linkages with local suppliers, indirectly suggesting that even within manufacturing, there might be substantial differences in contact-intensity. The total FDI dataset consists of 468 observations, but after accounting for the number of missing (confidential) values this drops down to 355. Unfortunately, a larger dataset which includes FDI stocks disaggregated on both a sector- and country-level is not readily available at this time.
Cultural distance
In order to quantify the cultural distance between the US and host country i, four of Hofstede’s cultural dimensions are used; power distance, individualism, masculinity, and uncertainty avoidance. In line with previous research (among others: Baily & Li, 2015; Tang, 2012), the remaining dimensions, long term orientation and indulgence, are not included due to the absence of many country-scores. Hofstede’s aggregation of cultural traits into a numerical index, combined with a focus on attitudes in a working environment, provides an excellent basis for economic- and business-related research. His dimensions have therefore been utilised in an incredible amount of culture-related research, with his 1980 study being one of the most frequently cited papers in the International Business field. Despite this widespread use, the cultural framework created by Hofstede has been criticised for a multitude of reasons. The main critique of Hofstede’s work is that it relies on data from employees of a singular MNE
6 Data marked by (*) was rounded to 0, since the BEA indicated that these values were very close to 0. Confidential data-points
14 (International Business Machines), thus neglecting the effect of organisational culture (McSweeney, 2002). In addition, Kogut and Singh (1988) point out that the method which was used for the construction of the scales was questionable. Nevertheless, both these critics subsequently proceed to stress the usefulness of Hofstede’s framework, and Drogendijk and Slangen (2006) state that the critique on Hofstede’s work might have been somewhat premature. Therefore, his cultural framework will form the basis of the empirical analysis of this study. As a robustness check, the GLOBE measures of cultural diversity (House et al., 1999) are included in addition. This adheres to comments by Beugelsdijk et al. (2018), who state that the effect of cultural distance on different stages of the firm internationalisation process can depend on how cultural distance is measured, indicating the need to use multiple measures. The GLOBE cultural dimensions include; uncertainty avoidance, future orientation, power distance, institutional collectivism, humane orientation, performance orientation, in-group collectivism, and gender egalitarianism. Country scores on these dimensions are calculated for two main cultural subsets; societal values and societal practices. However, this study solely uses scores related to the latter subset, for two reasons; (1) Tang (2012) finds that the societal values-based scores are very similar to Hofstede’s dimensions, and hence are unlikely to provide additional valuable information, and (2) since foreign investment is inherently a business-construct, societal practices are more relevant to the case at hand, since the largest business-related impact will originate from differences in business approaches and practices, rather than values.
In order to aggregate Hofstede’s four cultural dimensions into a single figure representing cultural distance, a composite index as described by Kogut and Singh (1988) is constructed (hereafter: KSI). This index uses the host country’s deviation from the US along all of the four cultural dimensions, corrected for differences in each dimension’s variance and arithmetically averaged to form a single number indicating country i’s cultural distance with respect to the US. This is algebraically described in the following equation:
𝐶𝑢𝑙𝑡𝑢𝑟𝑎𝑙 𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑖 = ∑4𝑑=1{(𝐼𝑖𝑑− 𝐼𝑢𝑑)2/𝑉𝑑}/4 , (3)
15 correlate with any of the independent variables. Nevertheless, the index has recently been the topic of some controversy due to a misspecification. It has been argued that the index fails to capture what it claims to, but instead describes the squared cultural distance between countries (Konara & Mohr, 2019), which might lead to potentially misleading conclusions. Since the KSI is very well established in the literature, this study does utilise it. However, to account for the potential erroneous results, a Euclidean Distance index will be constructed in addition, serving as a robustness check. This index is described by the following equation:
𝐶𝑢𝑙𝑡𝑢𝑟𝑎𝑙 𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑖 = √∑4𝑑=1{(𝐼𝑖𝑑− 𝐼𝑢𝑑)2 , (4)
where 𝐶𝑢𝑙𝑡𝑢𝑟𝑎𝑙 𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑖 is the cultural distance of country i from the United States, 𝐼𝑖𝑑 indicates the index for the dth dimension for country i, and u is the United States. This index is considered to be a more ‘pure’ measure of distance (Konara & Mohr, 2019). With regards to the GLOBE cultural distance measure, solely the Euclidean Distance formula is used. Table 1 portrays the correlation between the three measures of cultural distance. As can be expected due to their relative algebraic similarity, the correlation between both Hofstede measures is extremely high (0.9744). The correlation between the Hofstede measures and the GLOBE measure is very strong as well (0.570 and 0.597, both significant at the 0.1% level). Appendix A lists the host countries included in this study, and also reports the cultural distance(s) between the US and these host countries.
Table 1
Hofstede KSI Hofstede Euclidean GLOBE
Hofstede KSI 1
Hofstede Euclidean 0.975*** 1
GLOBE 0.570*** 0.597*** 1
note: this table shows the correlation between the 3 distinct measures of cultural distance (Hofstede dimensions using the Kogut and Singh index, Hofstede dimensions using the Euclidean Distance formula and the GLOBE dimensions using the Euclidean Distance formula). Significance is indicated by the asterisks (with *** indicating significance at the 0.1% level).
Individual cultural dimensions
16 can obscure some underlying mechanisms and results. Tang (2012) shows that refraining from aggregating culture into a singular construct of ‘distance’, and rather including the separate dimensions individually, may provide novel results. In order to capture a more complete picture of the effect of cultural differences on FDI, and the potential presence of sector dependent effects, Hofstede’s individual dimensions will also be included in a separate model.
Contact-intensity
The previously described hypothetical example regarding Intel and ExxonMobil illustrates that different sectors have a distinct level of contact-intensity, so in order to assess the potential heterogeneous, sector-dependent effect of cultural distance on FDI, a measure of contact-intensity needs to be constructed. Since some sectors (for instance Intel’s sector in the hypothetical example) require many distinct, customised inputs, firms operating in these sectors need to develop and maintain long-lasting relationships with many different (external) suppliers, increasing their exposure to local culture and business approaches. Consequently, it can be argued that multinational firms which operate in sectors that require many customised inputs from external suppliers have a more contact-intensive production process. In line with this, the dataset constructed by Nunn (2007) is used as a measure of contact-intensity. The dataset measures, on an industry level7, the proportion of intermediate inputs of that industry which require relationship-specific investments. Following the arguments made previously, it is argued that industries which require a large proportion of relationship-specific inputs are also characterised by a more contact-intensive production process, since a relatively larger amount of supplier-relationships need to be built and maintained. Hence, Nunn’s measure of relationship-specificity can be used as a measure of the contact-intensity of an industry’s production process.
Nunn identifies three different types of inputs; (1) inputs sold on an exchange, (2) inputs not sold on an exchange but reference priced in trade publications, and (3) inputs not sold on an exchange and not reference priced. These three types represent increasingly more relationship-specificity. If a specific input is sold on an exchange, there are many potential buyers and sellers, and the value of the input is the same inside the buyer-supplier relationship as it is outside it. Hence, the input is not relationship-specific and the buyer (the multinational firm)
7 Nunn’s dataset includes relatively specific 6-digit industry classifications. These industries are more specific than the
17 does not need to engage in relationship-building with any specific supplier. If an input is not sold on an exchange, but is reference priced in trade publications, it is considered to be an intermediate relationship-specific investment. Lastly, if an input is sold on neither an exchange, nor is it reference priced in trade publications, it concerns a relationship-specific input. In this case, the buyer needs to develop and maintain a relationship with the supplier in order to maximise value creation. The resulting measure (labelled by Nunn as: 𝑍𝑟𝑠1)8 indicates, on a specific industry level, the proportion of inputs which require relationship-specific investments. Each input is either considered to be relationship-specific or not (hence this is a 0 or 1 measure). Inputs that are not sold on an exchange and not reference priced in trade publications are identified as being relationship-specific (a value of 1), while any other inputs are classified as non-relationship-specific (a value of 0). Industries which require many inputs classified as 1 are hence characterised by a relatively contact-intensive production process.
Since Nunn’s dataset reports 6-digit industry classifications, it needs to be aggregated to an overarching sectoral level to match the FDI data. This is done using the BEA ‘the use of commodities by industry’ table9. Nunn’s original dataset utilised BEA 1997 I-O industry classifications (6-digit classifications), while this BEA table portrays 2012 NAICS industry codes (2-digit classifications). Aggregation of industries into sectors was thus done based on matching each industry ‘title’ in the Nunn data to the overarching sector which contains that specific industry title in the BEA table. A specific example of this aggregation is given in Appendix B. Since each industry has a distinct level of contact-intensity, aggregating them into sectors requires putting weights on each industry. Because there is no matching data concerning the total level of US-owned FDI on an industry-specific level, another method of weighing each industry’s ‘importance’ within the sector is needed. To this end, the total value of (host-country) exports of each industry as a fraction of that specific country’s total exports is used. This method is far from perfect, since there are many reasons why an industry’s importance in a country’s inward FDI might differ from its importance in a country’s exports. However, it does provide a relatively sensible way to aggregate the more specific industries into sectors, according to meaningful weights. An overall measure of contact-intensity per sector is now available, with values ranging from zero to one. A value of zero indicates an extremely low level of contact-intensity, where none of the industries contained in this sector require relationship-specific investments for any of their inputs, and a value of one indicates an exceptionally high level of
8rs stands for relationship-specific.
18 contact-intensity, where all of the industries contained in this sector require relationship-specific investments for all their inputs. Graph 1 shows the sectors and the corresponding average levels of contact-intensity.
Graph 1
note: due to ‘weighing’ each industry’s importance in a sector by using a country’s exports in that industry relative to its total exports, every country has a unique level of contact-intensity for every sector. This graph shows averages of this, thus portraying the overall average intensity per sector. Scores range from 0 (low intensity) to 1 (high
contact-intensity).
The results portrayed in Graph 1 appear to be intuitive. Based on Nunn’s data,the least contact-intensive sectors are the chemicals10, primary & fabricated metals, mining and food sectors, whereas transportation equipment and computers & electronic products have the highest degree of contact-intensity. Machinery, electrical equipment & components and other manufacturing (which mainly consists of farming, textiles and fabrics manufacturing) are moderately contact-intensive. This is in line with previous remarks, and supports the notion that industries with more linkages, like the ICT/microprocessor sectors, are generally more contact-intensive compared to low-linkage sectors like the primary & fabricated metals industry. A general variable for contact-intensity (CI) is included in the model to capture the potential bias of US firms towards sectors with certain levels of contact-intensity. Accounting for this potential bias will aid in adequately isolating the (sector-dependent) joint effect of cultural distance and contact-intensity.
10The fact that the chemicals industry has such a low level of contact-intensity might be somewhat surprising, but the largest
proportion of production in this industry concerns relatively ‘simple’ plastics, paints, detergents, and perfumes, which require very few specialised inputs and rely mainly on internal mass production.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Chemicals Primary & fabricated metals Mining Food Other manufacturing Electrical equipment & components Machinery Computers & electronic products Transportation equipment
19
Control variables
As proposed by the gravity model, the economic size of the source and host country should be included. However, since the source country (the US) is constant, only the economic size of the host country is relevant. To account for this, data on the host country’s (log) GDP in millions of US dollars is included. Due to the market-seeking behaviour of FDI, the effect of host country GDP should be positive (Davidson, 1980). Data on GDP was extracted from the World Bank Development Indicators. To account for the geographical distance between the countries, the (log) great circle distance (km) between Washington D.C. and the host country’s capital city is used as a control variable11. As indicated by the typical gravity model, the sign of this estimator should be negative. To control for the potential effect of the corporate tax rate, the top tax rate of the host country, during the year 2018, is included. A higher corporate tax rate is expected to have a negative effect on the FDI stock in that particular host country. Even though FDI stocks are relatively stable over time, and unlikely to be influenced greatly by the tax rate in a single year, top tax rates are stable over time in the majority of the countries, justifying the use of solely the 2018 rate. Data was collected from both the OECD and the KPMG tax database. Furthermore, Henisz’s (2002) political constraints index (POLCON) is used to control for the effect of political risk in the host country. The index is forward-looking, and indicates the likelihood of a drastic change in policy in the host country, based on the characteristics of the country’s political system. As argued by Busse, Königer and Nunnenkamp (2010), a high likelihood of policy changes increases the uncertainty associated with FDI, thus reducing the incentive to invest in that particular country. Since the measure ranges from zero (total political stability) to one (total political instability), a negative sign is expected. To account for the fact that FDI stocks develop over time, the average level of political risk over the years 2007-2017 was taken. Speaking a common language eases the business process, and is a strong predictor of MNE location decisions (Brewer, 2007). To capture this, the host country’s score on the 2018 English Proficiency Index, constructed by Europeiska Ferieskolan (EF), is included12. This index uses results on an English proficiency test taken by 2.3 million adults, and ranks countries based on these results. The scores of countries where English is the primary language (Australia, Canada, Ireland, New Zealand, and the United Kingdom) are not available and were hence equated to the highest scoring country in the dataset, The Netherlands. Administrative distance between countries tends to negatively affect investment, as pointed out by Berry,
11 Data was extracted from: https://www.gpsvisualizer.com/calculators on 13/02/2020.
12 EF is the world’s largest educational organisation, specialising in language training. For more information, see:
20 Guillén, and Zhou (2010). To control for administrative distance between the US and the host country, a dummy variable is included which indicates whether the US has an agreement with the host country on taxation of income and capital. The existence of an agreement incentivises FDI by avoiding ‘double taxation’, hence a positive sign is expected. Data were extracted from the IRS income tax treaties website. A dummy variable indicating whether the US has a free trade agreement with the host country is included, for which data was taken from the Office of the United States Trade Representative website. The existence of a free trade agreement points to the presence of strong ties between the US and the host country, whilst also easing business for multinational firms which set up foreign subsidiaries with the aim of importing goods back to the US (intra-firm imports). Therefore, a positive sign is expected. Lastly, to control for the influence of the level of human capital in the host country, the average years of schooling is included. Even though there are more determinants of human capital (for instance the actual quality of the educational system), the average years of schooling is a good proxy. The sign of this coefficient will indicate whether US multinational firms are biased towards countries with a highly educated workforce (capable labour) or towards countries with a relatively poorly educated workforce (cheap labour). The educational attainment dataset constructed by Barro and Lee (2013) provides this information, and the latest available year (2010) is used.
Descriptive statistics
Table 2 summarises the main characteristics of FDI and the control variables of interest. The FDI stock, owned by US firms and invested in a specific host country (measured in 2018), is around $2.5 billion on average, but with a large spread (a minimum of minus $599 million, and a maximum of over $48 billion).
Table 2 Descriptive Statistics
Variable Obs Mean Std.Dev. Min Max
FDI (millions) 355 2521.994 5105.481 -599 48162
Cultural Distance (Hofstede KSI) 468 2.355 1.309 0.015 5.150
GDP (millions) 468 1130000 2020000 23969.89 1.36e+07
Geographical Distance (km) 468 8362 3775.949 734.09 16356.59
Corporate Tax Rate (%) 468 24.74 7.396 8.5 55
Political Risk 468 .389 .156 0 .707
English Proficiency 459 57.801 7.916 41.6 70.27
Administrative Distance 468 .419 .494 0 1
Free Trade Agreement 468 .14 .347 0 1
21 Other averages can be readily observed in the table. Since Administrative Distance and Free Trade Agreement are 0/1 variables, their means can be interpreted as the proportion of host countries which have a treaty on taxation of income and a free trade agreement with the US, which is 41.9% and 14%, respectively.
Model specifications
To test the main hypothesis about the existence of a sector-dependent effect of cultural distance on FDI, and the additional hypothesis about the general influence of cultural distance on FDI, multiple sets of Ordinary Least Squares (OLS) models are utilised, each with their own set of robustness checks. The models are described below.
The main aim of this study is to illuminate the potential sector-dependent effect of cultural distance on FDI. As discussed in previous sections, every (manufacturing) sector has a distinct level of contact-intensity, and it is argued that cultural distance is more influential in sectors with a higher level of contact-intensity. Consequently, the interaction term consisting of the cultural distance between the US and a particular host country and the level of contact-intensity in a specific sector, captures this effect. As pointed out by Rajan and Zingales (1996), who analyse the interaction between a sector’s dependence on external finance and local financial market development, the fact that the main variable of interest is an interaction term simplifies the process of accounting for country and sector characteristics. By implementing both country and sector fixed effects, all unobserved heterogeneity stemming from country or sector characteristics is absorbed, and solely the interaction term ‘survives’. By isolating the interaction term, this model thus provides important insights regarding the existence of sector-dependent effects. The main advantage of this procedure is that it is unlikely to be criticised based on omitted variable bias or model specification, since both country- and sector-level effects are fully accounted for (Rajan & Zingales, 1996).
22 check the robustness of excluding fixed effects, an additional model including solely sector fixed effects (and thus omitting the sector-level variable of contact-intensity), is estimated.
It should be noted that the use of logarithmic models is subject to some potential drawbacks. The most prominent drawback relates to the zero (or negative) observations that are present in almost any international trade and investment dataset. There are many phenomena which explain the existence of these zero-valued investment positions, with studies pointing to a lack of cultural and historical links between countries (Rauch, 1999) and restrictive investment policies (in line with remarks from Burger, Van Oort & Linders, 2009) as some main reasons. The use of the natural logarithm of FDI stocks as the dependent variable creates a situation in which these zero observations are undefined, and are therefore excluded from the sample. This is a questionable procedure, since the observations could convey important information for the issues that are analysed (Yeyati, Stein & Daude, 2003). It is unlikely that the zero values are randomly distributed, so excluding them from the sample introduces a selection bias (Bergeijk & Brakman, 2010). Different models have been developed as a response to this ‘zero-observations’ problem, with the Poisson pseudo-maximum likelihood model (PPML), specified by Silva and Tenreyro (2006), being a prime example. However, there are two main reasons why the problem of zero-valued observations is less prevalent in this study: (1) because the sample does not cover bilateral data, but solely FDI stocks originating from the US, thus avoiding ‘double zero-entries’, and (2) because this study uses FDI stocks instead of flows. Stocks are the result of developments over time, and in order for a zero-valued observation to be found, investments from the US into a particular host country must not have happened at all, or disinvestments have occurred that cancel out all previous investments. The dataset used for this study contains 30 zero-valued observations (and 20 additional negative observations), accounting for 8.5% (or 14%) of the total useable dataset. As expected, the proportion of zero values is considerably lower compared to samples that cover trade and investment flows, with the related papers often stating that around half the dataset consists of zero values (e.g. Helpman, Melitz & Rubinstein, 2008). The use of specialised models, like the aforementioned PPML model, is therefore unwarranted, since these have been designed for cases where the proportion of zero-observations is considerably higher (Silva & Tenreyro, 2006).
23 and Manzocchi (2009) and Yeyati, Panizza and Stein (2003), a transformation of the dependent variable is employed13.
log𝐹𝐷𝐼𝑖,𝑠= log (𝐹𝐷𝐼𝑖,𝑠+ 1) (5)
The dependent variable log𝐹𝐷𝐼𝑖,𝑠 is replaced by log (𝐹𝐷𝐼𝑖,𝑠+ 1), thereby making sure that the zero-observations can still be defined. Because the values of 𝐹𝐷𝐼𝑖,𝑠are large (on average around 2500), the new regression coefficients of the log-log estimation can still be interpreted as elasticities, since log𝐹𝐷𝐼𝑖,𝑠 ≈ log (𝐹𝐷𝐼𝑖,𝑠+ 1). It should be noted that transforming data points is perhaps not the most usual or straightforward way to conduct research, hence this model will merely serve as a robustness check, and will give an indication as to how sensitive the main model is to inclusion or exclusion of the zero values14.
IV – Results
Before elaborating on the empirical findings of this study, a quick overview of the relationship of interest is provided. Appendix C shows a graph which plots log(FDI) against cultural distance. The graph appears to show a negative relationship15, suggesting that cultural distance is indeed negatively related to FDI. Evidently, this correlation does not necessarily imply causation. Therefore, this relationship, as well as the notion that the negative effect of cultural distance on FDI is more prominent in contact-intensive sectors, is more elaborately investigated with the use of the previously described empirical models. If not indicated otherwise, the empirical results all include robust standard errors in parentheses, accounting for unobserved heteroskedasticity. Significance levels are indicated by the asterisks (*** p<0.01, ** p<0.05, * p<0.1).
As an initial test of the main research question, Table 3 portrays the results of the model including both country and sector fixed effects. As stated in the previous section, by accounting
13 The transformation is done after the negative FDI observations are set to zero, making sure that both zero- and negative
observations are included in the post-transformation sample.
14 As a final robustness check of the full model, a model including FDI stock relative to host country GDP is estimated. Since
this model avoids logarithms, it includes the zero (and negative) values, whilst simultaneously dealing with FDI-outliers (see; Choi, 2009; Li, 2009). However, this model appears to be a poor fit for the data, since the r-squared drops drastically (from 0.4024 when using the logarithm of FDI to 0.1012 when using FDI divided by GDP). Additionally, most control variables lose their significance, indicating that there might be unobserved issues when using FDI divided by GDP. A potential cause could be that a sector-specific variable (FDI stock) is divided by a country-specific variable (GDP), but uncovering the underlying reasons behind the failure of this model is not the main aim of this study, hence this model is simply excluded from the results.
24 for all unobserved heterogeneity on both a country and a sector level, the interaction term between cultural distance and contact intensity can be adequately isolated (Rajan and Zingales, 1996). The results indicate that there is evidence of a sector-dependent relationship between cultural distance and FDI, but solely when Hofstede’s dimensions of culture are used. For both models utilising Hofstede’s framework, either using the KSI or the Euclidean distance formula, a significant negative coefficient is found (at the 5% and 10% level, respectively). This suggests that cultural distance is a more important factor in sectors which are relatively more contact-intensive, and that US firms operating in these contact-intensive sectors tend to avoid culturally distant markets. The slight difference in significance levels when using either the KSI or the Euclidean formula (with p-values of 0.022 and 0.061, respectively), suggests that there might indeed be empirical differences between these measures, lending support to the conclusions of Konara and Mohr (2019). Future research is needed to establish which measure is favoured. Additionally, the use of GLOBE cultural dimensions fails to show a significant sector-dependent effect of cultural distance, indicating that the results are not robust to the use of different cultural measures. The tentative conclusion that support for hypothesis 2 is found should thus be interpreted with caution.
Table 3
Initial test for sector-dependent effects using country & sector fixed effects
(1) (2) (3) Hofstede KSI Hofstede Euclidean GLOBE Euclidean Cultural Distance * -0.499** -0.0189* 0.107
Contact Intensity (CI) (0.216) (0.0101) (0.448)
Observations 305 305 269
R-squared 0.710 0.708 0.708
Country & Sector FE YES YES YES
note: the table includes robust standard errors in parentheses. Significance levels are indicated by the asterisks (*** p<0.01, ** p<0.05, * p<0.1).
Cultural distance
25 literature (e.g. Shenkar, 2001; Li & Guisinger, 1992; Tang, 2012; Stahl & Voigt, 2008), but contrasts the empirical findings of for instance Beugelsdijk et al. (2018) and Holburn and Zelner (2010). The lack of significant influence can be the result of a diverse set of underlying mechanisms. For instance, the trend towards integrated, homogenous markets, and the acculturation process of managers, could have diminished the effect of cultural distance on the location decisions of multinational firms (which is in line with findings from Li and Guisinger, 1992). Alternatively, the difficulties in communication and integration, which arise as a result of increased cultural distance, could be balanced out by the learning opportunities provided by culturally distant markets (in line with remarks from Reus and Lamont, 2009). Interestingly, model 1, which excludes the sector-dependent variables (Contact Intensity and the interaction term), shows a slightly negative (but insignificant) relationship between cultural distance and FDI. However, when including the sector-dependent variables (in models 2 and 3), the relationship switches sign, now being positive (but still insignificant). It is thus possible that previous studies, by failing to include sector-characteristics, have found incomplete results when it concerns the general effect of cultural distance on FDI. However, since the sign-change is not empirically significant, a deeper look into the sector-dependent effects is needed.
Cultural Distance and Contact Intensity
26
Table 4
Results: effect of cultural distance on FDI and potential sector-dependent effects
(1) (2) (3) (4) (5) Hofstede KSI Basic Hofstede KSI (Main Model) Hofstede Euclidean GLOBE Euclidean Hofstede KSI - Sector FE Cultural Distance -0.010 0.225 0.0105 -0.253 0.00658 (0.106) (0.185) (0.00936) (0.253) (0.158) Contact Intensity 1.231* 1.471* (0.667) (0.864) Cultural Distance * -0.627** -0.0257* 0.095 -0.295 Contact Intensity (0.278) (0.0133) (0.216) (0.225) GDP 1.091*** 1.095*** 1.111*** 1.092*** 1.120*** (0.0996) (0.0980) (0.0979) (0.1038) (0.0853) Geographical Distance -0.103 -0.0975 -0.103 -0.113 -0.0938 (0.153) (0.155) (0.157) (0.153) (0.126)
Corporate Tax Rate -0.0202 -0.0196 -0.0211 -0.0579** -0.0199
(0.0207) (0.0200) (0.0202) (0.0222) (0.0169) Political Risk 0.0220 0.0673 0.164 0.467 0.00958 (0.825) (0.802) (0.802) (0.782) (0.692) English Proficiency 0.0407** 0.0449** 0.0467** 0.0369** 0.0363** (0.0194) (0.0195) (0.0206) (0.0171) (0.0169) Administrative Distance -0.691** -0.692** -0.698** -0.726** -0.724*** (0.290) (0.288) (0.287) (0.287) (0.242)
Free Trade Agreement 0.473* 0.455* 0.418 0.565** 0.542**
(0.262) (0.264) (0.261) (0.263) (0.216) Educational level 0.0411 0.0367 0.0387 -0.0328 0.0796 (0.0657) (0.0650) (0.0659) (0.0690) (0.0554) Constant -8.942*** -9.923*** -10.36*** -6.996*** -8.727*** (2.387) (2.462) (2.633) (2.674) (2.068) Observations 298 298 298 262 298 Adjusted R-squared 0.374 0.379 0.398 0.394 0.540 Sector FE NO NO NO NO YES VIF (average) 1.97 3.16 3.73 1.81 2.84 VIF (maximum) 2.62 8.02 10.91 2.73 6.92
note: all models include robust standard errors in parentheses. Significance levels are indicated by the asterisks (*** p<0.01, ** p<0.05, * p<0.1).
27 the interaction term, and in model 5, contact intensity is already captured by the sector fixed effects16.
Interestingly, measuring (Hofstede) cultural distance using the Euclidean Distance formula instead of the Kogut and Singh index increases the explanatory power of the otherwise identical model by more than 5% (with the adjusted R-squared increasing from 0.379 to 0.398). This lends further support to Konara and Mohr’s (2019) conclusions that the KSI and the Euclidean Distance formula do indeed differ, which can have important consequences regarding the validity of research which has utilised the KSI. As mentioned before, new research should aim at investigating which measure is to be preferred.
Appendix D contains the outcomes of two additional robustness checks. Model 1 portrays the results of the main model in which the dependent variable is transformed {recall: log𝐹𝐷𝐼𝑖,𝑠 = log (𝐹𝐷𝐼𝑖,𝑠+ 1)}. The resulting model includes the zero-observations, reducing the potential sampling bias that is associated with the log-transformation of the main model described in Table 4. By comparing model 1 in appendix D to (the otherwise identical) model 2 in Table 4, the effect of this potential sampling bias can be deduced. Concerning the main variables cultural distance, contact intensity, and cultural distance * contact intensity, the signs of the coefficients do not change. However, the results indicate that the inclusion of the zero-observations is indeed influential, for two main reasons: (1) the significance of the interaction term disappears, and (2) the adjusted R-squared of the transformed model is 1.8% higher. This provides further evidence that the results found in the main model are not robust to changes in variable specifications.
Model 2 in appendix D provides one additional robustness check by estimating the main model, but with standard errors clustered on the country level. Clustering the standard errors accounts for autocorrelation, where unobserved components in individual outcomes are correlated within clusters (Abadie, Athey, Imbens & Wooldridge, 2017). The key assumption of no correlation of individual standard errors then changes to the fact that errors for observations within the same cluster may be correlated, but that errors are uncorrelated across clusters. In this study’s sample, clusters can be present at both a country- and sector level. However, Angrist and Pischke (2009) point out that the number of clusters should exceed 30 for the approach to be
16 With the cut-off value of 10 in mind, model 3 appears to show signs of multi-collinearity as well. However, after
28 econometrically valid. Clustering on a sector level is therefore not feasible, since the sample merely contains nine distinct sectors. Appendix D therefore solely portrays the results of the robustness check utilising country-clustered standard errors. This clustering does not affect the magnitude or interpretation of the coefficients, but merely increases the confidence intervals by allowing for some degree of correlation between observations. As can be observed in model 2 of appendix D, clustering standard errors on the country level does not change the results regarding the main variables. Similar to model 2 in Table 4, no significance is found for the general effect of cultural distance on FDI, but the interaction term is negative and significant at the 5% level, providing some additional evidence that cultural distance indeed matters more in sectors with a high level of contact-intensity.
Marginal effects and economic relevance
The (relatively non-robust) finding that cultural distance is more important in sectors which are relatively contact-intensive is useful from an academic perspective, since researchers ought to be aware of this when designing their studies. However, from a practical point of view, it is more useful to have explicit information regarding the implied magnitudes at distinct levels of contact-intensity. To assess this, the marginal effects of cultural distance at certain levels of contact-intensity should be distinguished. Graph 2 provides a visual representation of this, based on the main estimation (model 2 in Table 4). The graph clearly illustrates that cultural distance becomes a more important negative factor in FDI as we move towards sectors with higher levels of contact-intensity, judging by the downward sloping linear prediction line. The predicted effect becomes negative at contact-intensity levels exceeding 0.36. Based on the shaded area in the graph, which portrays the 95% confidence interval associated with the predicted values, cultural distance exerts a significant negative influence on FDI when contact-intensity exceeds 0.75. This number is of course specific to the sample that is used, but in this case, it indicates that cultural distance is a significant factor in the transportation equipment and computers & electronic products sectors, since these sectors have a contact-intensity exceeding 0.75.
29 between the US and Finland, would be associated with a predicted increase in FDI stock of 10.6% in the chemicals sector (although not significantly different from zero), and a predicted decrease in FDI stock of 39.2% in the transportation equipment sector (significantly different from zero at the 5% level). This clearly illustrates the potential economic relevance of the influence of contact-intensity on the cultural distance – FDI relationship17. Even though the findings are not particularly robust, the potential economic magnitude should motivate future research on the topic.
Graph 2
Marginal effect of Hofstede Cultural Distance at distinct levels of contact-intensity
note: this graph portrays the linear predicted marginal effects of Cultural Distance, based on certain levels of contact-intensity. The graph is based on the estimation of Model 2 in Table 4. The shaded area represents the 95% confidence
interval using robust standard errors.
Control variables
All five models in Table 4 report a strong, positive influence of host country GDP on the level of FDI stock, supporting the well-established theory that FDI is market-seeking. The sign of the corporate tax rate is negative in all models, as expected, but only significant in model 3, lending limited support to the conclusion that US multinationals tend to favour countries with
17 As a comparison, the effect of a one standard deviation increase in cultural distance on FDI, in the absence of the interaction
30 a low tax rate. Host country proficiency in the English language positively affects the level of US-owned FDI stock invested in that country, as indicated by the significance of the variable in all models. This supports the notion that speaking a common language eases the process of doing business. Furthermore, administrative distance exerts a significant, negative influence on the level of FDI stock, whilst the existence of a free trade agreement stimulates it. Three control variables show no significant effects in any of the models, namely; (1) geographical distance, (2) political risk, and (3) educational level. Geographical distance has a negative sign in all models, but the lack of significant findings shows that US firms do not drastically favour nearby countries when engaging in FDI activities. This contradicts the general principles of the Gravity Model, and might be a result of rapid advancements in communication technology, which reduce the need for spatial proximity. Alternatively, distance from the US might be poorly measured due to the fact that Washington DC is located on the Eastern end of this large country. For both the level of political risk, as well as the educational level, no effect which is constant across models (and no significance) is found.
Individual cultural dimensions
31
V – Conclusion
The results of this study provide some novel insights regarding the effect of cultural distance on FDI, specifically related to the influence of sector-specific contact-intensity. Table 5 provides a clear overview of the hypotheses, and states whether support is found based on the main models that are estimated. Firstly, by including multiple measures of cultural distance (Hofstede dimensions, GLOBE dimensions), and utilising distinct methods of calculating cultural distance, the empirical analysis fails to identify a significant effect of cultural distance on US-owned FDI stocks in manufacturing sectors. The apparent absence of effects fits well in the abundance of previous research, where no conclusive results have been established. There are multiple explanations for this lack of significance, of which at least a few are worth mentioning:
(1) The global trend towards more homogenous and integrated markets, and the acculturation process that is prevalent among managers of MNEs, could have diminished the potential barrier which is posed by a large cultural distance between source and host country (Li & Guisinger, 1992).
(2) More widespread use of advanced communication methods, and more international interaction between firms and individuals could have aided in bridging cultural distances, encouraging the convergence of cultural values (Shenkar, 2001).
(3) The increased difficulties in communication and business, which are associated with large cultural differences, could be off-set by certain advantages of expanding to culturally distant countries. For instance, by entering distant markets, firms have more opportunities to learn from routines and approaches which are unique to that specific country (Reus & Lamont, 2009; Beugelsdijk et al., 2018).
32 is a more important factor in sectors which are relatively contact-intensive, and that US firms operating in these contact-intensive sectors tend to avoid culturally distant markets. This sector dependent effect is robust to some changes in model- and variable specification (the use of clustered standard errors, and the use of different formulas for calculating cultural distance), but not to others (the use of GLOBE cultural dimensions, the inclusion of sector-fixed effects, and the inclusion of zero-observations for FDI). Further research, using more elaborate datasets, might establish a more robust effect. The existence of sector-dependent effects is relevant for both researchers (as will be discussed in the subsequent section) and policymakers. Since contact-intensive sectors generally provide the largest scope for knowledge and technology spill-overs to local firms (through the abundance of linkages in these sectors, see e.g. Edwards, Sengupta & Tsai, 2010; Alfaro, 2003; Alfaro & Charlton, 2007), policymakers might be incentivised to attract foreign MNEs in these contact-intensive sectors. This study points out that these policymakers should be aware that cultural distance could pose a relatively larger barrier for foreign MNEs operating in these sectors. Based on the Hofstede KSI model, the analysis of the marginal effects shows that the influence of contact-intensity on the cultural distance – FDI relationship could have large economic impacts. A one standard deviation increase in cultural distance results in a predicted increase in FDI stock of 10.6% in the least contact-intensive sector (not statistically significant) and a predicted decrease in FDI stock of 39.2% in the most contact-intensive sector (statistically significant). Evidently, these potential economic impacts are specific to the sample utilised in this study, but the potentially large economic magnitudes should warrant future research. Lastly, this study finds very limited support for Tang’s (2012) conclusion that differences in individual cultural dimensions (as opposed to composite indices) have a distinct, independent effect on FDI.
Table 5
Confirmation or rejection of the hypotheses
Hypothesis Hofstede (KSI) Hofstede (KSI) – clustered SE Hofstede (KSI) – country & sector FE Hofstede (KSI) – sector FE Hofstede (Euclidean) GLOBE (Euclidean) H1: cultural distance No No No No No No
H2: sector-dependent effect Yes Yes Yes No Yes No
