Types Of Firms Generating Network Ex Tern Ali Ties And Mncs' Co-location Decisions

Strategic Management Journal Strat. Mgmt. J., 26: 595–615 (2005) Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/smj.464

TYPES OF FIRMS GENERATING NETWORK EXTERNALITIES AND MNCs’ CO-LOCATION DECISIONS SEA-JIN CHANG1 * and SEKEUN PARK2 1 2

School of Business Administration, Korea University, Seoul, Korea The Export–Import Bank of Korea, Seoul, Korea

This study identifies and examines sources of network externalities that influence MNCs to agglomerate their foreign operations in specific regions. Using data for Korean firms that invested in China, this study found that network externalities were sensitive to the types of firms constituting a regional network. It also found stronger network externalities within firms than across firms, from firms of the same nationality than from those of different nationalities, and from firms in the same industry than from those of different industries. As we defined the types of firms more precisely, distinctive curvilinear relationships between network externalities and the likelihood of co-location emerged. Copyright  2005 John Wiley & Sons, Ltd.

Why do multinational corporations decide to locate in one area rather than another? To date, research on this question has examined firms’ motivations for these decisions, the modes they use when entering an area, and the sequences of their entry decisions in that area. This work has been guided by the theme of foreign direct investment (Hymer, 1960; Dunning, 1988; Hennart and Park, 1994; Chang, 1995; Kogut and Chang, 1996). It has also considered such decisions at the national level, rather than assessing why a firm might enter one region within a nation rather than another. We argue that international business scholars should explore regional location decisions much more extensively, as these decisions shed light on MNCs’ foreign entry strategies. When MNCs announce they are investing in a country, they often specify a location they have decided upon prior to the announcement. Most Keywords: network externalities; agglomeration; co-location; foreign direct investment

∗ Correspondence to: Sea-Jin Chang, School of Business Administration, Korea University, Seoul, Korea 136-701. E-mail: [email protected]

Copyright  2005 John Wiley & Sons, Ltd.

countries consist of many regions, which differ greatly from each other in terms of prevailing wages, populations, technology bases, and infrastructures. Since MNCs presumably choose locations that seem to fit best with their strategic goals, the location decision within a country (e.g., Shanghai or Beijing) may be more important than the decision at the country level is (e.g., opening a factory in China). Furthermore, one firm’s location decisions could be influenced by the presence of other firms in a region. Foreign and local incumbents within a region can pose great threats and challenges to new entrants. At the same time, they can be great sources for complementary resources and learning. In a country such as China, which comprises vastly heterogeneous regions, it is crucial to examine location decisions at the regional level in order to understand MNCs’ entry strategies. Recently, several studies have focused on location decisions within a country. Head, Ries, and Swenson (1995), Shaver and Flyer (2000), and Chung and Song (2004) studied how Japanese firms chose manufacturing locations in the United

Received 4 June 2003 Final revision received 15 December 2004

596

S.-J. Chang and S. Park

States. They found that Japanese firms located their manufacturing facilities in states where many other Japanese firms had been located, although this tendency depended upon these firms’ relative resource strengths and their prior investments. These researchers suggested that positive network externalities might explain this pattern of agglomeration. A network externality occurs when the benefit or surplus that an economic agent derives from a good depends in part on changes in the number of other agents consuming the same kind of good (Katz and Shapiro, 1985; Arthur, 1990; Liebowitz and Margolis, 1995). It thus denotes situations when a product or service becomes more valuable as more people or firms use it. For example, when General Motors entered China in 1997 via a joint venture with Shanghai Automotive Industry Corporation (SAIC), it was able to capitalize on the infrastructure of qualified managers, laborers, and suppliers that Volkswagen, SAIC’s other joint venture partner, had developed since 1984. MNCs can also learn from earlier entrants’ experiences and avoid making similar mistakes. This recent work has, however, been limited in three important ways. First, in focusing exclusively on economic reasons for agglomeration, it has not considered the possibility that agglomeration might occur even without obvious economic reasons. Firms might, for instance, imitate other firms in order to gain legitimacy or reduce uncertainty (DiMaggio and Powell, 1983; Levitt and March, 1988). Foreign investments, especially in countries where the culture and language are distinct from that of an MNC’s home country, carry risks that are captured by the term ‘liabilities of foreignness.’ Such risks are greater in transitional economies, such as China. It is possible that foreign firms investing in China flocked to Shanghai or Beijing only because many other foreign firms had done so already. Second, most of this work has focused only on positive forms of network externalities. When agglomeration occurs, competition in factor and product markets increases costs. For instance, in Shanghai, foreign firms now have to pay top salaries to attract local managers, and housing for expatriates is extremely expensive. Local firms located near an MNC may be able to access technology or know-how by hiring local managers and engineers away. Organizational ecologists have long argued that increases in population density diminish new entrants’ survival rates (Hannan and Copyright  2005 John Wiley & Sons, Ltd.

Carroll, 1992). Thus, MNCs should evaluate the costs of both negative and positive externalities. Third, these studies have ignored variations among the types of firms that make up a regional network. Although some studies considered the heterogeneity of investing firms (Shaver and Flyer, 2000; Chung and Song, 2004), most have examined only a subset of all the firms in a region (e.g., Japanese investors and local firms in the United States) and often considered only one industry (e.g., electronics). We argue that the degree of externalities is contingent upon the composition of regional networks. For instance, Korean firms investing in China may derive stronger network externalities from other Korean firms than they can from non-Korean firms, and from firms in the same industries than they can from firms in other industries. This paper considers two empirical questions. First, it uses data for Korean firms that invested in China to examine whether positive or negative network externalities are larger. Although we cannot distinguish empirically between network externalities derived from real economic gains and those derived from legitimacy, we examine various arguments for network externalities and develop a hypothesis that argues for a curvilinear relationship. We expect that the likelihood of experiencing negative network externalities is more substantial when firms’ agglomerative behaviors are motivated by a desire to gain legitimacy rather than by real economic gains. Second, this study examines how network externalities vary according to the types of firms within a regional network. In other words, this study examines to what degree network externalities are firm specific, nation specific, or industry specific. This study also explores what choices by firms might maximize the net benefits of network externalities.

NETWORK EXTERNALITIES AND LOCATION DECISIONS Network externalities Economists have long emphasized the importance of network externalities (Marshall, 1920). Porter (1998) summarizes the potential benefits of agglomeration: (1) it improves accessibility to specialized factors and workers; (2) it improves access to information about market and technology trends; (3) it promotes complementarities Strat. Mgmt. J., 26: 595–615 (2005)

Types of Firms Generating Network Externalities among firms and promotes cooperation among firms; (4) it improves access to infrastructure and public goods; and (5) it increases competitive pressure among firms. Henderson (1986) empirically demonstrated that agglomeration increases factor productivity. Saxenian (1994) documented how microelectronics firms clustered in Silicon Valley. Krugman (1991) developed a formal model in which agglomeration results from manufacturing firms’ desire to locate in a place of larger demand in order to exploit scale economies and minimize transportation costs, while the location of demand depends on the location of manufacturers. Several studies of MNCs’ regional agglomeration patterns were based upon this economic rationale.1 Smith and Florida (1994) and Head et al. (1995) observed that Japanese firms co-located with other Japanese firms. They pointed to technological spillovers, specialized labor, and other inputs as the main reasons for agglomeration. Chung and Song (2004) found that Japanese electronics firms in the United States tended to colocate with other Japanese firms when they had less prior experience. At the country level, Song (2002) showed how Japanese firms’ prior investment in technological and sourcing capabilities led to subsequent investment in the same countries. Chung and Alcacer (2002) also found that firms in research-intensive industries are more likely to locate in regions with high R&D intensities. Although economists generally attribute regional agglomeration to real economic gains, organizational theorists have argued that agglomeration might occur for non-economic reasons. Agglomeration might occur, for instance, when firms wish to improve their legitimacy in order to access resources they need for survival and growth (DiMaggio and Powell, 1983; Suchman, 1995). A firm might locate in a popular place simply because so many other firms have located there already, thus legitimizing the location. This mimetic behavior has been observed in various contexts: adoption of the M-form organization structure (Fligstein, 1985) and the poison pill (Davis, 1991), and acquisition decisions (Haunschild, 1993). Further, the risk and uncertainty of venturing into a foreign country could increase firms’ imitation of 1 Early work in economic geography studied the impact of income, tax incentives, wage, and unionization on attracting more foreign direct investment (Coughlin, Terza, and Arromdee, 1991; Wheeler and Mody, 1992; Friedman, Gerlowski, and Silberman, 1992).

Copyright  2005 John Wiley & Sons, Ltd.

597

other firms (Levitt and March, 1988). Empirical work has indicated that legitimacy and uncertainty influence MNCs’ expansions into foreign countries. Guillen (2002) found that emerging multinationals that were in the early stages of internationalization imitated other firms. Henisz and Delios (2001) demonstrated that Japanese firms that lacked international experience relied more heavily on the past international expansion decisions of other firms in their reference group as cues for their own entry decisions. Knickerbocker (1973) also pointed out that MNCs in oligopolistic industries tend to imitate each other when they expand into foreign markets. In addition, some research has found that agglomeration can lead to negative externalities. For example, firms can benefit from the spillover of other firms’ knowledge and technologies, but their own knowledge and technologies can spill over to other firms. Appold (1995) found that agglomeration was negatively associated with performance in the U.S. metalworking sector. Shaver and Flyer (2000) argue that benefits and costs firms derived from co-location could differ according to their own core competences. They contend that firms with relatively more resources avoid agglomeration because, for them, the potential costs of spillovers are greater than the potential benefits. Agglomeration can also lead to intensified competition in both product and factor markets among adjacently located firms. Baum and Mezias (1992) demonstrated that hotels located in Manhattan that were similar in terms of location, price, and size posed greater threats to each other and reduced each other’s chances of survival as the area became more crowded. Agglomeration also drives up the costs of locally sourced inputs, such as wages of local managers and engineers and housing expenses for expatriates. Agglomeration can also reduce innovation via groupthink (Porter, 1998), thereby creating negative externalities. It can make firms in a regional cluster look only inward and reject ideas from other areas. Detroit’s attachment to gas-guzzling automobiles in the 1970s amidst oil shortages is a prime example of such rigidity. We expect that multinational corporations will consider both positive and negative network externalities. Population ecologists have long argued that population density influences startups’ performance. They have found that an increase in population density is initially positively correlated with Strat. Mgmt. J., 26: 595–615 (2005)

598

S.-J. Chang and S. Park

the founding rate of startups because it provides legitimacy and acceptance. After a certain level, however, population density is negatively associated with the founding rate due to increased competition in that particular niche (Hannan and Carroll, 1992). By a similar logic, we expect that firms will agglomerate up to a point, but that increases in negative network externalities outweigh any increase in positive ones after that level is reached. We expect that marginal benefits from agglomeration will decline since knowledge or experience spillovers and legitimacy gained from an additional firm in a region would become redundant and negligible. On the other hand, marginal costs from agglomeration will increase since the competition in both product and factor markets will become more severe and potential hazards from groupthink would become larger. Above that level, marginal costs exceed marginal benefits and the likelihood of co-location will decline (see Figure 1). The likelihood that marginal costs exceed marginal benefits is greater when firms decide to colocate in order to gain legitimacy rather than for economic reasons. When there are no real economic gains, the marginal benefits of legitimacy drop quickly and will be completely offset by the marginal costs of both intensified competition and the hazard of groupthink. Since Korean firms were at an early stage of internationalization and China posed great uncertainty as a transitional economy during the period we studied, their agglomeration was substantially motivated by the desire to gain legitimacy and avoid uncertainty. Given this assumption, we also believe it is more likely that Net of positive & negative network externalities

Positive externalities Negative externalities

MC

MB Density

Figure 1.

Rationale for a curvilinear relationship

Copyright  2005 John Wiley & Sons, Ltd.

network externalities were curvilinear as a function of the number of firms. There is, however, a possibility that marginal costs from agglomeration might not increase when an additional firm locates in a region. Porter (1998) and Porter and Stern (2001) argue that intensified competition in a regional cluster can promote innovation, which may lower the marginal cost curve in the long run. Some regional clusters, such as Silicon Valley, have been extremely innovative and cost efficient despite intense competition and high levels of agglomeration may neutralize any cost increasing pressures. Similarly, some firms can manage better the hazard of groupthink. In our study, we test for this relationship by inserting both a monotonic and a quadratic term for the intertemporal population of firms constituting a regional network. An alternative hypothesis is that network externalities are monotonically positive or negative. Hypothesis 1: The likelihood that a firm chooses a location increases as the number of firms already present in the same region increases to a certain point, and declines after that point. Types of firms generating network externalities Most of the agglomeration benefits identified by Porter (1998) originate from flows of experiencebased knowledge among firms. By hiring specialized workers and purchasing other inputs, a firm can tap the knowledge embedded in such resources. Firms can also share knowledge by detecting market and technological trends and promoting cooperation. The international business literature has stressed that it is important for firms to transfer and seek knowledge or experience when they invest in foreign countries and decide what entry mode to use. Johanson and Vahlne (1977), Kogut (1983), Barkema and Vermeulen (1998), Chang (1995), Kogut and Chang (1996), and Chang and Rosenzweig (2001) demonstrated how the knowledge or experience firms gained from their prior entries affected their subsequent entries and their choices of entry mode. Shaver, Mitchell, and Yeung (1997) argue that subsequent foreign entrants observe and learn from the success or failure of earlier entrants. Such benefits are, however, neither automatic nor guaranteed. Kogut and Zander (1992) and Zander and Kogut (1995) showed that knowledge that Strat. Mgmt. J., 26: 595–615 (2005)

Types of Firms Generating Network Externalities is less codifiable and teachable and more complex is more difficult to transfer even within a firm. They also demonstrated that, due to this difficulty, firms often hire workers away from the target firm to facilitate knowledge diffusion. Szulanski (1996) similarly observes that causal ambiguity and the absorptive capacity of the recipient, among other factors, lead to ‘stickiness’ and inhibit the transfer of knowledge to other parts of an organization. Hansen (1999) also shows that weak inter-unit ties among various product development teams in a firm help a project team search for useful knowledge in other subunits, but strong ties enable the transfer of complex knowledge. According to these studies, knowledge transfer and sharing can be difficult even within a firm or among affiliates within the same business group. This study thus treats a firm’s own prior experience, which has been conventionally taken for granted, as a source of network externalities for itself. Agglomeration may facilitate flows of knowledge among firms or within a firm, as it increases a firm’s ability to communicate with or hire people working for other firms or other divisions. The possibility of these flows might also be contingent on other factors, such as the identities of the firms targeted as sources of knowledge. We expect that it is easier to transfer knowledge among firms of similar background since there is less causal ambiguity and higher levels of absorptive capacity among such firms. Chung and Kalnins (2001) found that hotels that were similar in size and were affiliated with a chain derived greater agglomeration benefits. The extent to which firms co-locate to gain legitimacy also depends upon the types of firms that it can imitate (Haunschild and Miner, 1997). For example, trait-based imitation occurs when a firm imitates the behavior of high-status organizations (Fombrun and Shanley, 1990; Haveman, 1993). Guillen (2002) argues that firms expanding into foreign markets tend to imitate other firms whose experience, history, or location is relevant to their own situation. As a consequence, emerging multinationals tend to imitate other firms with which they are familiar because they are in the same industry or part of the same business group. In this study, we test whether network externalities are firm specific, nation specific, and industry specific. First, we expect that knowledge sharing and transfer are easiest and imitation is most prevalent when they occur inside a firm. A firm may Copyright  2005 John Wiley & Sons, Ltd.

599

invest more than once in a foreign country. It may locate its investments in the same region to benefit from the experience it has gained. Since many firms are diversified and are organized into product divisions, co-locating investments for multiple divisions helps firms share plants, equipment, and workers. Expatriate managers can add more businesses at the same location without hiring more managers. Business groups in many countries, where many legally independent firms are under the same ownership and administrative control, facilitate resource sharing among member firms (Granovetter, 1995). For instance, Samsung Group uses regional organizations to facilitate knowledge sharing by its affiliates. Samsung China coordinates various activities by the subsidiaries of its affiliate firms in China. In turn, these affiliates frequently emulate other affiliates when they expand into a new region. For instance, when Samsung Electronics is located in Tianjin, other affiliates of the Samsung Group such as Samsung Corporation and Samsung SDI are more likely to locate in Tianjin than they are in other regions since they can learn from Samsung Electronics’ experience in the same location and since Samsung Electronics’ presence legitimizes their own location choices. For his sample of Korean firms that invested in China, Guillen (2002) found strong evidence that a firm’s rate of entry increased as its affiliates set up their own plants. Martin, Swaminathan, and Mitchell (1998) showed that Japanese automobile suppliers followed their buyers, competitors, and suppliers into the United States. Bastos and Greve (2003) also found that industry affiliation and board interlocking ties were strong predictors for Japanese firms’ mimetic entry behaviors in Europe. We thus expect that marginal benefits from an additional firm in a region will be greater if the new firm is in the same boundary than they are if the new firm is unrelated. On the other hand, the marginal cost increase associated with agglomeration will be smaller when the same firm or affiliated firms are colocating investments than it is when unaffiliated firms co-locate, since a firm or a business group should be more able to coordinate its location decisions to avoid any competition among its own foreign operations. A firm or a group’s deliberate effort to avoid any pitfalls of groupthink can also lower the marginal costs of agglomeration. For instance, the LG Group deliberately spread the operations of its individual affiliates Strat. Mgmt. J., 26: 595–615 (2005)

600

S.-J. Chang and S. Park

over many regions, including several remote areas, while avoiding too much concentration in a few popular coastal areas. By doing so, it achieved balanced growth in the Chinese markets and maximized knowledge sharing at the group level. We therefore expect: Hypothesis 2: Network externalities are stronger within firms and among firms associated with the same business groups than they are for unaffiliated firms. We also expect that a firm’s country of origin significantly affects the degree of network externalities it can derive from other firms. Since each country has its own culture, national origin may affect the types of experiential knowledge a firm creates (Hofstede, 1980). As a consequence, some knowledge or experience may be nation specific, and firms may be able to learn more from the experience of firms from the same nation than they can from firms that are from different nations. For instance, a Japanese firm that transfers labor relationship practices commonly used in Japan may be uniquely qualified to learn from the past experience of other Japanese firms that implemented similar practices in the same foreign location. A Korean firm may be relatively more able to recruit a local manager who can speak Korean from a subsidiary of another Korean firm. Also, when there is a large community of firms from the same country, these firms often create country-specific infrastructures such as a Swedish school for Swedish expatriates’ children. A firm may also pay more attention to the actions of other firms from the same nation than it does to those of different nations, revealing another type of trait-based imitation. Such imitative behavior among firms from the same nation might be even stronger for firms from countries that have strong ethnocentric orientations, such as Korea and Japan. Various researchers have shown that Korean and Japanese firms frequently emulate each other (Guillen, 2002; Head et al., 1995; Henisz and Delios, 2001; Chung and Song, 2004; Alcacer, 2004). We thus expect the marginal benefits from agglomeration to be higher when firms of the same nationality co-locate investments than they are when firms of different nationalities colocate. It is also possible that an agglomeration of firms from the same nation might increase marginal Copyright  2005 John Wiley & Sons, Ltd.

costs, such as the costs of nation-specific inputs and the hazards of groupthink, than would an agglomeration of firms from several nations. We expect, however, that the proportion of nationspecific inputs to overall costs is low relative to the potential benefits from spillovers of nation-specific experience or knowledge, especially for firms in which ethnocentrism is strong. Since Korean firms are as ethnocentric as Japanese firms are, we hypothesize:2 Hypothesis 3: Network externalities are stronger for firms from the same nation than they are for firms from different nations. In addition, some network externalities may be industry specific. Although components of infrastructure, such as roads, transportation, and housing for expatriates, are shared by all firms, it may be easier to share other resources, like specialized suppliers and workers, within an industry boundary. Shanghai General Motors probably benefited more from Shanghai Volkswagen than it did from any other company in Shanghai. An industry also provides a frame of reference to all firms in it. It is a social structure that affects the flow of information and legitimacy from one firm to another (Guillen, 2002). Firms also measure their internal processes and performance against others in the same industry (Porac and Rosa, 1996). Therefore, firms tend to imitate other firms in the same industry to gain legitimacy or to reduce uncertainty, revealing another type of trait-based imitation. Guillen (2002) found that Korean firms are more likely to invest in China when more of their domestic competitors have already invested in China. Henisz and Delios (2001) showed similar results for Japanese firms. We thus expect that marginal benefits from agglomeration will be higher when firms in the same industry co-locate investments than they are when firms in different industries co-locate. 2 Henisz and Delios (2001), Guillen (2002), and Chung and Song (2004) argue that network externalities could be stronger for firms that had little or no international investment experience. They propose that network externalities from other firms would be weaker or non-existent after firms had such experience. We experimented with various interaction effects between a firm or group’s prior entry experience and the count of other types of firms. The interaction terms were generally insignificant, suggesting that our sample firms derived substantial network externalities from other firms even after they had investment experience.

Strat. Mgmt. J., 26: 595–615 (2005)

Types of Firms Generating Network Externalities On the other hand, negative network externalities, such as groupthink and competition in product and factor markets, may be also industry specific, and thus shift marginal costs upward. It is not clear whether the marginal benefits and marginal costs schedules would intersect at a higher level of agglomeration for firms in the same industry or firms in different industries. Porter (1998) and Porter and Stern (2001) argue that intensified rivalry among firms in the same region actually promotes innovation, and cite regional clusters such as Silicon Valley for semiconductors as an example. We argue that the innovationpromoting effects of intensified rivalry among firms of the same industry within a regional cluster are greater than the negative consequences of intensified competition in the factor and product markets are. We thus propose: Hypothesis 4: Network externalities are stronger among firms in the same industry than they are across industries.

RESEARCH METHODS Sample Korean firms’ investment activities in China provide an interesting empirical setting to study our research questions. First, Korea ranked as the fifth largest investor in China, after the United States, Japan, Taiwan, and Singapore.3 Second, compared to U.S., European and Japanese multinationals, Korean firms entered China relatively late. Most Korean investments in China took place after China and Korea established a diplomatic relationship in 1992. As a late entrant, Korean firms had many opportunities to evaluate network externalities when they decided which locations to enter. Third, Korean firms have a reputation for being ethnocentric and often emulating each other (Guillen, 2002). Thus, China provides a good setting to test whether network externalities are stronger among Korean firms than they are for other foreign firms. Fourth, many Korean firms 3 According to the official statistics, investments from Hong Kong and the U.S. Virgin Islands exceeded those from Korea. A large portion of these countries’ investments, however, actually came from firms in other nations that used Hong Kong and U.S. Virgin Islands as tax havens or as a beachhead for entry into China.

Copyright  2005 John Wiley & Sons, Ltd.

601

are associated with business groups, also known as chaebols. Within chaebols, individual affiliates share considerable know-how and expertise. Thus, Korean firms provide an interesting setting to gauge how strong network externalities are within the boundaries of a firm or a group. The sample for this study consists of Korean firms’ direct investments in China between 1988 and September 2002. Korean firms started to invest in China through Hong Kong in 1988. Over time, Korean firms have increased their investments in China. By September 2002, Korean firms had invested 6000 times in China, for a total of $4.1 billion in the manufacturing sector.4 When Korean firms invest overseas, they are required by law to report their investments to the government-owned Import–Export Bank of Korea, which maintains a database on the names of investors, dates, amounts, and locations of investing firms’ activities. For our sample, we selected Korean firms’ investments in China in the manufacturing sector whose declared investment amount exceeded $1 million. We focused only on manufacturing investments since manufacturing and non-manufacturing sectors require rather different types of experience, knowledge, workers, and other inputs. We also dropped small investments (i.e., those less than $1 million) since we wanted to have investments of comparable size by Korean firms and the other foreign firms, as well as local firms in our database. During 1988–2002, there were 661 investments that met our criteria. Among these cases, we dropped 121 cases from our sample since the identities of investors were either individuals, rather than firms, or could not be confirmed due to bankruptcies or closures. We also deleted cases where the intended investment amount was reported but did not materialize into an actual investment by September 2002. The 540 investments in our sample were worth $3.2 billion, representing more than 78 percent of Korean firms’ total investment in the manufacturing sector in China, as reported to the Import–Export Bank of Korea. Several firms in our sample invested in China at least twice during our time study period. Samsung Electronics and LG Electronics each invested 10 times, LG Chemical invested nine times, Hyundai 4 This figure excludes announced but unrealized investments and those that were liquidated or bankrupt from the Export-Import Bank of Korea statistics, as of September 2002.

Strat. Mgmt. J., 26: 595–615 (2005)

602

S.-J. Chang and S. Park

Motors invested six times, etc. Many of the firms in our sample were associated with large business groups (Chang, 2003). The Export–Import Bank of Korea uses the Korean Fair Trade Commission’s definition of affiliation with a business group to determine whether individual firms belonged to such groups.5 If we use this definition, for instance, affiliates of the LG Group, such as LG Electronics and LG Chemical, made a total of 27 investments in China. China is officially organized into 22 provinces, four special cities (Beijing, Tianjin, Shanghai, and Chongqing), and five autonomous regions, such as Inner Mongolia and Xinjiang. We adopted China’s classification of regions since all vital statistics in China are organized by region. Our sample of firms invested in 18 of these 31 regions. Figure 2 shows the geographic distribution of Korean firms’ investments in China. Korean firms show some agglomeration in regions such as Beijing, Tianjin, Liaoning, Shangdong, Jiangsu, Guangdong, and Shanghai. Compared to other foreign firms, which concentrated heavily in Shanghai, Guangdong, Beijing, and Tianjin, Korean firms dispersed more into the northern provinces and other inner regions. Although Koreans and Chinese are culturally similar, Korean firms are not free from the liabilities of foreignness. These firms reported various difficulties, such as their relationships with the government, motivating workers, and protecting their intangible resources (Export–Import Bank of Korea, 2002). Therefore, transfers and spillovers of experience-based knowledge were critical to improving the performance of these firms’ operations. Measures In order to reflect the location choices faced by Korean firms at the time of investment, we collected descriptive data for each region. We measured the population of each region (in increments of 10 million), expecting that firms would prefer locations with large markets. We included the average wage in the manufacturing sector (in thousands RMB) to reflect the relative attractiveness 5 The Korea Fair Trade Commission legally defines a business group as ‘a group of companies, more than 30% of whose shares are owned by some individuals or by companies controlled by those individuals, or those that are practically controlled by them despite lower ownership control.’

Copyright  2005 John Wiley & Sons, Ltd.

of a region as a manufacturing base. Although we acknowledge that highly skilled workers can be expensive in China and that wages are less important in the capital and technology-intensive industries, we assume that wages of skilled labors will correlate with those of unskilled labors across regions. We thus expect that average wages are a proxy, albeit an imperfect one, for the relative cost of regional labor in China. Our Highway/area variable denotes the length of the highways in a region in km divided by the size of the region in km2 in order to gauge the quality of physical infrastructure. We expect the more well developed the highway system is, the more likely it is that a firm prefers the location. The number of patents (in thousands) granted by the Chinese patent office was included to reflect the technical skills of organizations in a region. As documented by Sun (2000), state and local research institutes as well as corporations in China have aggressively filed for local patents. Foreign firms are more likely to favor a location where there are a large number of local patents. We collected all vital social statistics for the above variables from the China Statistical Yearbook, 1988–2001, and matched these data with the year that each investment occurred. Since there are many ethnic Koreans in China who have worked as translators and middle-level managers for the Chinese subsidiaries of Korean firms, we included the number of ethnic Koreans in a region (in thousands). We expect that the more ethnic Koreans there are in a region, the more attractive those regions are to Korean firms. We collected the distribution of ethnic Koreans from the Population Survey. Not all statistics were available for every year. The length of highway was not available until 1995, and the number of ethnic Koreans was available only for 1990 and 1998. We substituted data for these variables with data from the closest year. In order to gauge the degree of network externalities, we measured the cumulative counts of prior investments by different types of firms at the time of each investment. Prior studies using count variables noted that the level of locally accumulated experience, not just the binary variable noting the mere presence of experience, was more critical to subsequent location decisions (Song, 2002). Count of a firm’s own prior entry measured the number of a focal firm’s own prior entries into each region up to the time of an entry event. Count of entry by Strat. Mgmt. J., 26: 595–615 (2005)

603

Figure 2.

Distribution of Korean, foreign, and local firms in China

Types of Firms Generating Network Externalities

Copyright  2005 John Wiley & Sons, Ltd.

Strat. Mgmt. J., 26: 595–615 (2005)

604

S.-J. Chang and S. Park

firms within the same group reflected the cumulative count of entries by firms associated with the same business groups in each region up to the time of each entry event. For firms with no group affiliation, this variable is coded as zero. Count of other Korean firms was measured as the number of Korean firms not associated with the same business groups that were operating in each region at the time of each entry. It is hard to keep track of how many foreign and local Chinese firms existed in each year. During the last decade, the number of foreign firms setting up their operations in China exploded. There was also substantial restructuring among local firms in China. Many formerly state-owned firms and cooperatives were transformed into joint stock companies and were privatized. During this transformation process, there were many mergers and acquisitions among local firms. At the same time, the Chinese government encouraged foreign firms to form joint ventures with local firms in order to transfer technology or know-how and to maintain employment. We collected data for prior entries by non-Korean foreign firms and local firms from Dun & Bradstreet’s Major Corporations in P.R. China 2001. This directory consists of two volumes. Volume 1 contains a total of 2953 major foreign firms’ own subsidiaries and their joint ventures in the manufacturing sector in China. Volume 2 contains information on a total of 1341 major Chinese firms in the manufacturing sector. The directory does not supply any other information apart from the number of employees, the names of parent firms, the year of entry, and the addresses and contact information of operations.6 The average number of employees of foreign firms in this directory was 246, while it was 1580 for local firms. The average number of employees for the 540 Korean firms in our sample was 817, suggesting that the Korean firms in our database were somewhat larger than the foreign firms but smaller than the Chinese firms listed in Dun & Bradstreet’s directory. Count of other foreign firms and count of local firms were measured as the number of manufacturing operations of non-Korean multinationals and Chinese local firms, respectively, operating in each region at the time of each entry. 6 According to our telephone interview with managers at Dun & Bradstreet, this publication used various criteria such as sales and number of employees to select ‘major’ foreign and local firms to be included in the directory.

Copyright  2005 John Wiley & Sons, Ltd.

We further classified our count measures into two parts, reflecting the number of firms in the same industry and those in different industries. We used the Korean 2-digit Standard Industry Classification (SIC) code to determine whether firms were in the same industry as a focal firm. Dun & Bradstreet classified foreign and local firms in China according to the 2-digit U.S. SIC, which we converted to the 2-digit Korean SIC. In some models, we used variables that aggregated the various count variables defined above. Count of all unrelated firms is defined as the sum of count variables for other Korean firms, other foreign firms, and local Chinese firms in a region. Count of all firms in a regional network adds a firm’s own entries and entries by firms affiliated with the same group on top of count of all unrelated firms, thus comprising all types of firms in a regional network. In order to measure the joint effects of positive and negative network externalities, we measured the monotonic and squared terms of these count variables. If a squared term has a negative coefficient and a monotonic term has a positive one, that discrepancy may indicate an inverted U-shaped relationship. Methodology This study uses a conditional logit model to test our hypotheses (McFadden, 1974). Our sample firm faces a set of location choices, each of which has different attributes. Since the conditional logit model requires all choices to be selected at least once, we included only 18 regions, all of which were invested in at least once.7 The conditional logit model has been widely used to analyze how agents choose from a large set of alternatives. It is relevant for location choices in foreign direct investment (Head et al., 1995; Shaver and Flyer, 2000). It focuses on the attributes of each location in the choice set. Attributes for each region, including population, average wage, highway/area, ethnic Koreans, and number of patents, were the same for all Korean firms investing in China at time t. Count variables reflecting the number of incumbent firms that were sources of network externalities were measured according to the identities of investing firms. This 7 Refer to our discussion of the robustness tests in which we examined the independence of irrelevant alternatives in the results section.

Strat. Mgmt. J., 26: 595–615 (2005)

Types of Firms Generating Network Externalities model estimates how each attribute increases or decreases the chance that a location will be chosen rather than all other potential locations. Let us define an underlying latent variable, Vij t , to represent the utility a firm i derives from opening a manufacturing operation in region j at time t. Since there were m regions where firms could enter during our time study period, each observation (i.e., a firm choosing a location) has m rows of data, each corresponding to a specific region. Assuming a linear relationship with the latent variable, we can write: Vij t = βXij t + eij t

(1)

X is a vector of independent variables that affect choices. If a firm makes choice j in particular, then we assume that Vij t is the maximum among the m utilities. Hence, the statistical model is driven by the probability that choice j is made, which is: Prob(Vij t > Vikt ) for all other k = j

(2)

Let Yit be a random variable that indicates a choice made by firm i at time t. Assuming independence of irrelevant alternatives, the probability that a firm i chooses j location at time t can be written as follows: Prob(Yit = j ) = exp(βXij t )/ [k=1...m exp(βXikt )]

(3)

The maximum likelihood method is used to estimate β, which we can use to test whether various independent variables significantly affect the probability that one region will be chosen among all the regions in the choice set. In this conditional logit model, β cannot be interpreted as marginal effects as it could be in a linear regression. The marginal effects can be derived by differentiating Equation 3 with respect to the independent variables, X. In order to interpret the magnitude of the coefficient, we can calculate the ‘average probabilities elasticity,’ as reported in Head et al. (1995), which referred to an independent variable’s probability elasticity for the average option in the choice set.8 The elasticity of the probability of a particular firm i choosing region j with respect to an 8 See the appendix of Head et al. (1995) for more details about how to derive for the average probability elasticity.

Copyright  2005 John Wiley & Sons, Ltd.

605

independent variable, Xl , can be calculated by differentiating Equation 3: ∂ Prob (Yit = j ) Xl ∂Xl Prob (Yit = j ) (4) = βl [1 − Prob (Yit = j )]

Elasticity1ij t =

Summing over all firms and choices, the relationship between the average probability elasticity and the coefficient estimate, βl , is:




m−1 m (5) The above expression discounts the coefficient by (m − 1)/m. Since we have 18 regions in our dataset, we have to multiply individual coefficients by 0.94 (=17/18) to calculate the average probabilities elasticity. Elasticity1 =

i

j

t

Elasticity1ij t = βl

RESULTS Tables 1 and 2 show, respectively, descriptive statistics and results from conditional logit models. The odd-numbered models in Table 2 include only monotonic variables, while the even-numbered models include both monotonic and squared terms to test for curvilinear relationships. Model 1 shows a baseline model, in which the count of all firms in a regional network was included in addition to regional attributes such as population, average wage, highway/area, number of ethnic Koreans, and patents. The results showed that Korean firms preferred regions that were characterized by large populations, low wages, a well-developed highway system, a large population of ethnic Koreans, and a large number of patents. The count of all firms in a region was positive and significant, suggesting the existence of positive network externalities. The chi-square statistic shows the model to be highly significant, with p < 0.001. The pseudo R 2 is 0.1826, suggesting reasonable model fit.9 In Model 2, we added a squared term of the count of all types of firms in a region to test for a curvilinear relationship. When we added this term, both the monotonic and the squared terms of the 9 Pseudo R-squared is defined as one minus the ratio of the maximum likelihood functions of a model without any explanatory variable divided by the model that includes all explanatory variables.

Strat. Mgmt. J., 26: 595–615 (2005)

606

S.-J. Chang and S. Park

Table 1.

Descriptive statistics

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18)

Population (10 million) Average wage (thousands RMB) Highway (km)/area (km2 ) Ethnic Koreans (thousands) Patents (thousands) Count of all firms in a regional network Count of a firm’s own prior entry Count of entry by firms within the same group Count of all unrelated firms Count of other Korean firms Count of other Korean firms in the same industry Count of other Korean firms in different industries Count of other foreign firms Count of other foreign firms in the same industry Count of other foreign firms in different industries Count of local firms Count of local firms in the same industry Count of local firms in different industries

Mean

S.D.

Minimum

Maximum

4.05 5.68 40.38 105.03 2.13 182.96 0.002 0.01 182.85 13.28 1.06 12.22 117.25 6.71 110.53 52.31 3.24 49.07

2.32 3.05 48.35 284.45 2.32 334.97 0.19 0.45 334.91 26.53 3.11 24.54 292.71 21.44 277.04 56.59 5.74 53.43

0.64 1.34 1.06 0.07 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

9.04 20.41 277.27 1181.96 18.26 1965.00 4.00 10.00 1963.00 185.00 34.00 184.00 1651.00 206.00 1651.00 281.00 55.00 281.00

Correlation matrix (1) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10) (11) (12) (13) (14) (15) (16) (17) (18)

1.00 −0.13 1.00 −0.30 0.65 1.00 −0.15 −0.14 −0.20 1.00 0.32 0.54 0.32 −0.11 1.00 −0.16 0.63 0.62 −0.17 0.35 1.00 0.05 0.08 0.09 −0.02 0.08 0.06 1.00 0.04 0.14 0.16 −0.03 0.14 0.08 0.61 1.00 −0.16 0.63 0.62 −0.17 0.35 0.99 0.06 0.08 1.00 0.38 0.26 0.34 −0.07 0.37 0.24 0.13 0.15 0.24 1.00 0.23 0.20 0.28 −0.06 0.28 0.18 0.18 0.26 0.18 0.67 1.00 0.38 0.25 0.33 −0.07 0.36 0.24 0.12 0.13 0.24 0.99 0.60 1.00 −0.30 0.59 0.60 −0.14 0.26 0.98 0.03 0.06 0.98 0.07 0.07 0.07 1.00 −0.23 0.45 0.46 −0.11 0.20 0.73 0.06 0.08 0.73 0.05 0.13 0.04 0.74 1.00 −0.30 0.59 0.60 −0.14 0.26 0.97 0.03 0.06 0.97 0.07 0.06 0.07 0.99 0.71 1.00 0.40 0.51 0.42 −0.22 0.57 0.73 0.11 0.12 0.73 0.57 0.37 0.57 0.59 0.43 0.59 1.00 0.25 0.30 0.22 −0.13 0.38 0.41 0.11 0.14 0.41 0.31 0.34 0.29 0.33 0.38 0.32 0.58 1.00 0.39 0.51 0.42 −0.22 0.56 0.73 0.10 0.11 0.73 0.58 0.36 0.58 0.59 0.42 0.59 0.99 0.51 1.00

N = 9720 (= 540 × 18).

count variable lost significance. We performed a log-likelihood test by taking the ratio of the maximum probability under the constraint of the null hypothesis (i.e., the coefficient for the squared term equals zero) to the maximum likelihood with that constraint relaxed. The −2 log-likelihood has an asymptotic distribution of chi-square. The difference in chi-square at the bottom of Table 2 Copyright  2005 John Wiley & Sons, Ltd.

was very small, indicating that we could not reject the null hypothesis. Thus, Hypothesis 1 was not supported when we lumped all different types of firms into a single category and measured network externalities. The network externalities from all types of firms combined into a single measure seemed monotonic rather than curvilinear. Strat. Mgmt. J., 26: 595–615 (2005)

Types of Firms Generating Network Externalities Models 1 and 2 assumed that the firms constituting a regional network were homogeneous. In subsequent models, we relaxed this assumption by disaggregating firms into different types and testing hypotheses that corresponded to these types. Models 3 and 4 tested Hypothesis 2 by identifying within a region a focal firm’s own prior entries, entries by affiliated firms, and entries by firms not related to a focal firm in any way. In Model 3, which assumes linear relationships, all three count variables were significantly positive, suggesting positive network externalities. The coefficients for a firm’s own entry and the count of firms affiliated with the same business group were much greater than was the count of all unrelated firms. When we calculate the average probability elasticity, a 100 percent increase in the number of a firm’s prior entries in a region increases the likelihood of choosing that region by 46.6 percent (i.e., 0.496 × 0.94 × 100). Similarly, a 100 percent increase in the number of prior entries within a region by affiliate firms increases the likelihood of choosing that region by 47.7 percent (0.508 × 0.94 × 100). In contrast, a 100 percent increase in the number of unrelated firms in that region increases the likelihood that a focal firm chooses this region by only 0.1 percent (= 0.001 × 0.94 × 100). We performed a log-likelihood test to determine whether these differences in the size of coefficients were statistically significant by positing Model 1 as a null hypothesis, where coefficients for these three types of firms were constrained to be equal. The log-likelihood test comparing Models 1 and 3 rejected the null hypothesis at p < 0.001, suggesting that the differences in coefficients were significant. Thus, we found strong support for Hypothesis 2. Model (4) added the squared terms of three count variables to test for curvilinear relationships. The results for it showed that the coefficient for count of entries by affiliate firms was positive and significant, while the coefficient for its squared term was negative and significant, suggesting a curvilinear relationship. Both the monotonic and the squared terms for the count of a firm’s own prior entries and the count of unrelated firms in that region were insignificant in Model 4, suggesting monotonic rather than curvilinear relationships. The log-likelihood test comparing Models 3 and 4 rejected the null hypothesis that coefficients for all squared terms were zero. Thus, Model 4 provided some support for Hypothesis 1. It is Copyright  2005 John Wiley & Sons, Ltd.

607

somewhat tricky to test Hypothesis 2 in a nonlinear model such as Model 4. Since only the count of entry by firms affiliated with the same business groups turned significant in a curvilinear model, we found some support for Hypothesis 2 in Model 4. It is worth noting that network externalities from a firm’s own prior entries seemed monotonically positive, while the impact of entries by firms affiliated with the same business groups was inverted. The results may suggest that a firm does not expect much conflict or competition among its own investments in the same region. In contrast, affiliated firms, which transfer their own local managers to other affiliates or share other resources with each other, may incur some costs. It is also possible that business groups discourage too much agglomeration from occurring in any one region in order to guard against the harmful effects of agglomeration and achieve a well-balanced approach to the entire Chinese market. LG Group’s actions provide some anecdotal evidence for this conjecture. Models 5 and 6 subdivided the category of unrelated firms by the national origin of firms in order to test Hypothesis 3. Model 5 showed that the counts of other Korean firms and local firms in a region were positively related to the likelihood that a firm located in this region, suggesting positive network externalities. In contrast, the count of other foreign firms was negative and significant, suggesting negative network externalities. When we calculate the average probability elasticities, a 100 percent increase in the number of other Korean firms in a region increases the likelihood that a firm located in this region by 0.8 percent, while the same increase in the number of foreign firms decreases the likelihood that a firm located in this region by 0.1 percent. The log-likelihood test comparing Models 3 and 5 rejected the null hypothesis that coefficients for these three count variables were equal. Thus, we found support for Hypothesis 3. Model 6 added the squared terms of count variables. Compared to Model 4, curvilinear patterns emerged more strongly when we disaggregated the category of unrelated firms. Both the count of other Korean firms and local Chinese firms showed inverted U-shaped relationships, consistent with our expectation, while the count of other foreign firms showed a U-shaped relationship, contrary to our expectation. The loglikelihood test comparing Models 5 and 6 rejected Strat. Mgmt. J., 26: 595–615 (2005)

Copyright  2005 John Wiley & Sons, Ltd.

(1)

(2)

(3)

Conditional logit model of location choices by Korean firms in China (4)

(5)

Population (10 million) 0.406 (0.027)∗∗∗ 0.404 (0.029)∗∗∗ 0.383 (0.028)∗∗∗ 0.382 (0.029)∗∗∗ 0.149 (0.043)∗∗∗ Average wage (thousands −0.156 (0.043)∗∗∗ −0.156 (0.043)∗∗∗ −0.195 (0.045)∗∗∗ −0.194 (0.045)∗∗∗ −0.027 (0.055) RMB) 0.025 (0.002)∗∗∗ 0.025 (0.001)∗∗∗ 0.022 (0.002)∗∗∗ 0.023 (0.002)∗∗∗ 0.015 (0.002)∗∗∗ Highway (km)/area (km2 ) Number of ethnic Koreans 0.001 (0.000)∗∗∗ 0.001 (0.000)∗∗∗ 0.001 (0.000)∗∗∗ 0.001 (0.000)∗∗∗ 0.001 (0.000)∗∗ (thousands) Patents (thousands) 0.136 (0.027)∗∗∗ 0.132 (0.033)∗∗∗ 0.138 (0.027)∗∗∗ 0.135 (0.033)∗∗∗ 0.101 (0.028)∗∗∗ Count of all firms in a 0.001 (0.000)∗ 0.001 (0.001) regional network 0.571 (0.369) 0.387 (0.197)∗ Count of a firm’s own prior 0.496 (0.195)∗ entries Count of entry by firms in 0.508 (0.108)∗∗∗ 0.899 (0.192)∗∗∗ 0.542 (0.108)∗∗∗ the same group Count of all unrelated firms 0.001 (0.000)∗∗∗ 0.001 (0.000) (other Korean, foreign, locals) Count of other Korean firms 0.009 (0.002)∗∗∗ —in the same industry —in different industries Count of other foreign firms −0.001 (0.000)∗ —in the same industry —in different industries Count of local firms 0.008 (0.002)∗∗∗ —in the same industry —in different industries Count of all firms in a 0.000 (0.000) regional network2 −0.085 (0.131) Count of a firm’s own prior entries2 −0.064 (0.026)∗ Count of entry by affiliate firms2 Count of all unrelated firms2 0.000 (0.000)

Variables

Table 2.

(8)

−0.075 (0.026)∗∗

−0.069 (0.026)∗∗

0.024 (0.026) 0.015 (0.005)∗∗ −0.008 (0.126)

0.010 (0.009) 0.008 (0.002)∗∗∗

0.006 (0.003)∗ 0.012 (0.007)† −0.001 (0.000)∗∗ −0.003 (0.001)∗∗

0.127 (0.030)∗∗∗ 0.025 (0.004)∗∗∗

−0.035 (0.130)

0.015 (0.004)∗∗∗

−0.002 (0.001)∗

0.062 (0.014)∗∗∗ 0.005 (0.002)∗

0.495 (0.106)∗∗∗

0.941 (0.193)∗∗∗

0.033 (0.004)∗∗∗

0.239 (0.368)

0.393 (0.198)∗ 0.320 (0.371)

0.920 (0.193)∗∗∗

0.132 (0.034)∗∗∗

0.098 (0.028)∗∗

0.130 (0.034)∗∗∗

0.013 (0.002)∗∗∗ 0.001 (0.000)∗

0.014 (0.002)∗∗∗ 0.001 (0.000)∗∗

0.139 (0.043)∗∗ 0.059 (0.047) 0.019 (0.056) −0.079 (0.063)

(7)

0.013 (0.002)∗∗∗ 0.001 (0.000)∗

0.060 (0.047) −0.085 (0.063)

(6)

608 S.-J. Chang and S. Park

Strat. Mgmt. J., 26: 595–615 (2005)

Copyright  2005 John Wiley & Sons, Ltd.

∗∗∗

p < 0.001;

∗∗

−1275.8 569.9(6)∗∗∗ 0.1826

−1240.2 641.3(11)∗∗∗ 0.2054 (4) vs. (3) 10.1 (3)∗

−1245.2 631.2(8)∗∗∗ 0.2022 (3) vs. (1) 61.3 (2)∗∗∗

−1125.8 570.0(7)∗∗∗ 0.1826 (2) vs. (1) 0.1 (1)

p < 0.01; ∗ p < 0.05; † p < 0.10. Standard deviations are in parentheses. N = 9720

Count of other Korean firms2 /1000 —in the same industry2 /1000 —in different industries2 /1000 Count of other foreign firms2 /1000 —in the same industry2 /1000 —in different industries2 /1000 Count of local firms2 /1000 —in the same industry2 /1000 —in different industries2 /1000 Log-likelihood Chi-square (d.f.) Pseudo R 2 Log-likelihood test: Comparing models Differences in chi-squares (d.f.) 7.3 (2)∗

−1219.0 683.5(10)∗∗∗ 0.2190 (5) vs. (3) 61.8 (5)∗∗∗

−1188.1 745.3(15)∗∗∗ 0.2388 (6) vs. (5)

−0.046 (0.020)∗

0.001 (0.001)∗

−0.133 (0.025)∗∗∗

18.0 (3)∗∗∗

−1210.0 701.5(13)∗∗∗ 0.2247 (7) vs. (5)

64.6 (8)∗∗∗

−1175.7 770.1(21)∗∗∗ 0.2467 (8) vs. (7)

−0.048 (0.022)∗

−0.872 (0.765)

0.002 (0.001)∗∗

−0.031 (0.038)

−0.123 (0.029)∗∗∗

−3.085 (1.103)∗∗∗

Types of Firms Generating Network Externalities 609

Strat. Mgmt. J., 26: 595–615 (2005)

Copyright  2005 John Wiley & Sons, Ltd.

∗∗∗

p < 0.001;

∗∗

p < 0.01; ∗ p < 0.05; † p < 0.10. Standard deviations are in parentheses.

−1081.8 720.6(13)∗∗∗ 0.2498 8653

(0.048) (0.063) (0.003)∗∗∗ (0.000) (0.034)∗∗∗ (0.368) (0.193)∗∗∗ (0.030)∗∗∗

−1038.8 806.6(21)∗∗∗ 0.2797 8653

−1162.0 579.1(13)∗∗∗ 0.1995 8040

−1131.8 639.4(21)∗∗∗ 0.2467 8040

−0.669 (0.756) −0.035 (0.022)

0.002 (0.001)∗

−0.009 (0.006) −1.510 (0.957) −0.121 (0.029)∗∗∗

−0.028 (0.038)

(0.026) (0.004)∗ (0.126) (0.026)∗∗ (1.112)∗∗

0.270 (0.344)

0.018 0.011 0.001 −0.075 −2.845

−0.003 (0.001)∗

−0.001 (0.000)∗ 0.009 (0.009) 0.007 (0.002)∗∗∗

0.012 (0.007)†

0.007 (0.003)∗

0.024 (0.005)∗∗∗

0.066 −0.101 0.011 0.000 0.122 0.218 0.913 0.125

(4)

−0.121 (0.029)∗∗∗

(0.031) (0.007)∗∗∗ (0.126) (0.027)∗∗ (1.103)∗∗∗

(0.044)∗∗ (0.056) (0.002)∗∗∗ (0.000)† (0.029)∗∗ (0.197)† (0.106)∗∗∗ (0.014)∗∗∗

0.004 (0.002)†

0.130 −0.015 0.012 0.000 0.081 0.369 0.491 0.067

(3)

Excluding regions with less than 3 entries

−0.123 (0.029)∗∗∗

0.036 0.035 −0.001 −0.071 −3.085

0.003 (0.003)

−0.002 (0.001)∗ 0.006 (0.010) 0.009 (0.003)∗∗∗

0.002 (0.019)

(0.055) (0.092)∗∗∗ (0.003)∗∗∗ (0.000)∗ (0.038)∗∗ (0.375) (0.197)∗∗∗ (0.031)∗∗∗

0.013 (0.006)∗

−0.042 −0.327 0.011 0.001 0.107 0.264 0.876 0.130 0.020 (0.005)∗∗∗

(0.047)∗∗ (0.084) (0.003)∗∗∗ (0.000)∗∗ (0.042)∗∗ (0.203)∗ (0.112)∗∗∗ (0.015)∗∗∗

(2)

0.003 (0.003)

0.123 −0.044 0.018 0.001 0.134 0.432 0.502 0.058

(1)

Excluding Shanghai

Robustness tests of the independence from irrelevant alternatives

Population (10 million) Average wage (thousands RMB) Highway (km)/area (km2 ) Number of ethnic Koreans (thousands) Patents Count of a firm’s own prior entry Count of entry by firms in the same group Count of other Korean firms in the same industry Count of other Korean firms in different industries Count of other foreign firms in the same industry Count of other foreign firms in different industries Count of local firms in the same industry Count of local firms in different industries Count of a firm’s own prior entry2 Count of entry by the same group firms2 Count of other Korean firms in the same industry2 /1000 Count of other Korean firms in different industries2 /1000 Count of other foreign firms in the same industry2 /1000 Count of other foreign firms in different industries2 /1000 Count of local firms in the same industry2 /1000 Count of local firms in different industries2 /1000 Log-likelihood Chi-square (d.f.) Pseudo R 2 N

Variable

Table 3.

610 S.-J. Chang and S. Park

Strat. Mgmt. J., 26: 595–615 (2005)

Types of Firms Generating Network Externalities the null hypothesis that all the coefficients for squared terms equaled zero. Model 6 provided support for Hypothesis 1. The results for Model 6 showed that the number of unrelated Korean firms initially increased the likelihood of co-location up to a certain level (the estimated inflection point seemed to be around 124 entries) but depressed it after that level. Korean firms may compete with each other for some Korean-specific inputs. For instance, Korean firms may compete to hire managers who understand the local culture and speak Korean. The intensified competition in factor markets associated with increased co-location could outweigh any additional positive externalities derived from agglomeration. On the other hand, the results for foreign firms were initially negative, but turned positive after a certain level (the estimated inflection point seemed to be around 1000 entries). Again, it is hard to test Hypothesis 3 in a non-linear model. Yet the fact that network externalities from other Korean firms were positive up to 124 entries but were negative for other foreign firms up until 1000 entries provides some support for Hypothesis 3. This result seemed consistent with the distribution of Korean and foreign firms. Korean firms entered both regions where many foreign firms were present and regions where very few foreign firms entered. Models 7 and 8 used industry classification to distinguish firms. For Model 7, where monotonic relationships are assumed, the coefficient for the count of Korean firms in the same industry was much larger than that of Korean firms in different industries was. If we calculate the average probability elasticities, a 100 percent increase of other Korean firms in the same industry in a region increases the likelihood that a firm locates in this region by 5.8 percent. On the other hand, the same increase in other Korean firms in different industries increases the likelihood a firm locates in that region by only 0.5 percent. This result suggests that Korean firms in the same industries created large network externalities among themselves by sharing technology or know-how and providing legitimacy. The results for Model 7 also demonstrated that the negative coefficient for the count of other foreign firms in Model 5 was driven by foreign firms in different industries. Korean firms derived positive network externalities from foreign firms in the same industries. These firms derived negative Copyright  2005 John Wiley & Sons, Ltd.

611

network externalities by co-locating with foreign firms from different industries, perhaps because of increased competition in factor markets. In contrast, the presence of local firms in the same industry did not seem to attract Korean firms in China. Since many local Chinese firms trailed Korean firms in terms of technological know-how, the net flow of knowledge was from Korean firms to local firms. A Korean firm might derive more benefits from locating in a region where local firms in industries different from the Korean firm’s had already created an industrial infrastructure and an ample supply of workers. The log-likelihood test comparing Models 5 and 7 rejected the null hypothesis that the coefficients for count variables were the same regardless of industry. Thus, we found support for Hypothesis 4. Model 8 tested for curvilinear relationships by adding the squared terms of the count variables in Model 7. All these variables showed inverted U-shaped relationships in Model 6 except for the count of foreign firms in different industries, which showed a U-shaped relationship, the count of foreign firms in the same industry, which was monotonically positive, and the count of local firms in the same industry, which was insignificant. The log-likelihood test comparing Models 7 and 8 rejected the null hypothesis that the coefficients for all squared terms were zero, providing additional support for Hypothesis 1. In order to check the robustness of our results, we implemented several alternative formulations. The conditional logit model relies on the assumption of independence of irrelevant alternatives, which means that the relative probability of choosing two alternatives does not depend on the availability of other alternatives (McFadden, 1974; Hausman and McFadden, 1984). To demonstrate robustness regarding the assumption of independence from irrelevant alternatives, we experimented with different subsamples. Models 1 and 2 of Table 3 showed the same specifications as Models 7 and 8 of Table 2, while we dropped Shanghai from our choice set. Shanghai received a total of 1651 foreign entries, which represented more than 55 percent of all foreign entries. We dropped Shanghai from the choice set in order to see whether negative network externalities from foreign firms in different industries might disappear. As in Model 1 of Table 3, the negative impact of foreign firms in different industries was sustained even after we dropped Shanghai from the Strat. Mgmt. J., 26: 595–615 (2005)

612

S.-J. Chang and S. Park

choice set. The relationship between the count of foreign firms in different industries and the likelihood of co-location seemed to be monotonic rather than curvilinear after we dropped Shanghai. We also experimented with different cut-off points in determining the number of regions to be included in the choice set. For instance, when we dropped regions that received fewer than three entries, Xinjian, Anhui, and Hubei were excluded from this choice set. Models 3 and 4 showed results with only 15 choice sets. These results seemed consistent with the ones in Table 2. We also tested whether multiple entries by the same firms could affect our estimations. In a linear regression, it is relatively easy to include either fixed effects or random effects to account for such panel data. Including such effects is trickier, however, for the conditional logit model, as this model calculates the conditional likelihood for each choice set. We therefore experimented by dropping firms that invested more than once in China. The results of this test were consistent with those reported in this study. We also experimented with a model where we grouped multiple entries and their other alternatives in the choice sets by the same firm (or group) into a single choice set and calculated the conditional likelihood. For example, when a firm (or group) invested twice during the time study period, there were two regions that the firm (group) entered and 34 (i.e., 2 × 17) regions that it did not enter in the new choice set. Such specification generated results very similar to those reported in this study.10

DISCUSSION Multinational corporations’ location choices within a foreign country have only recently received 10 We also experimented with alternative specifications of attributes. Several prior works that studied agglomeration patterns, such as Head et al. (1995) and Chung and Song (2004), included ‘alternative specific constants’ instead of various attributes for each region, which we used in this study. Alternative specific constants are equivalent to region-specific dummy variables. Although such regional dummy variables may be able to capture unspecified regional attributes other than the five attribute variables specified in this study, they have the great limitation of being time-invariant. In a dynamic country such as China, such variables seemed inappropriate. When we added these region-specific dummy variables to the regional attributes in our study, however, most of these regional attribute variables turned insignificant. Since we were interested in observing which attributes explained location choices, we preferred using attribute variables rather than regional dummy variables.

Copyright  2005 John Wiley & Sons, Ltd.

attention. Our study confirmed a regional agglomeration pattern for our sample of Korean firms in China. It is consistent with prior research that observes the agglomeration of Japanese firms in the United States (Head et al., 1995; Shaver and Flyer, 2000; Chung and Song, 2004), and the innovation-promoting aspects of regional clusters (Porter, 1998; Porter and Stern, 2001). Those works treated network externalities as location specific but not as firm specific. By doing so, they might have erred in treating all firms within the network as homogeneous. This study enhances our theoretical understanding by defining and measuring network externalities as specific to a focal firm in question. It highlights the fact that investing firms derive levels of network externalities that vary according to firm boundaries, nationality, and industry affiliation. This new perspective on network externalities as firm specific generates several implications for further research. First, we empirically confirmed that the composition of a network influences what network externalities exist. The types of firms that constituted a regional network were heterogeneous and created varying degrees of network externalities. We found that network externalities were stronger among firms in the same business group, among firms of the same nationalities, and among firms in the same industries. Second, while prior work portrayed network externalities as monotonic, we found they were curvilinear, especially when we defined network externalities as firm specific and classified the types of firms constituting a regional network more precisely. We argued that negative network externalities occur when firms have spillovers of their own technology or knowledge, when there is intensified competition in factor markets, and when the potential for groupthink exists, and that these costs outweighed any positive network externalities after a certain level. It is interesting to note that agglomeration of firms of the same type is subject to a curvilinear relationship, suggesting that agglomeration of the same type of firms is not beneficial beyond a certain point. On the other hand, agglomeration of other foreign firms in the same industry shows a monotonically increasing pattern. This finding is consistent with Porter and Stern’s (2001) argument that agglomeration of firms with diverse backgrounds facilitates innovation. Third, this study found that a firm’s own prior entry, as well as entries by firms associated with Strat. Mgmt. J., 26: 595–615 (2005)

Types of Firms Generating Network Externalities the same business groups, created the greatest network externalities. The impact of a firm’s own prior entry or entries by firms in the same business groups, gauged by average elasticity, was 470 times higher than that of an unrelated firm’s entry (i.e., 46.7–47.7% vs. 0.1%). This result suggests that experience-based knowledge is best transferred or spilled over within the boundaries of a firm, confirming that it is difficult to transfer tacit knowledge across firm boundaries (Kogut and Zander, 1992; Szulanski, 1996). The results may also indicate that firms imitate other firms in the same business group to gain legitimacy or reduce uncertainty. This study therefore provides a rationale for the regional patterns of agglomeration by business groups that have been observed for Japanese vertical keiretsu in the United States (Head et al., 1995). This study also reconfirms the findings of Chung and Song (2004) that agglomeration occurs more within firms than between firms. Fourth, it advances our understanding of network externalities by integrating literature from organizational theory and economics. Our hypotheses reflected economists’ arguments that agglomeration occurs because of phenomena like knowledge spillovers and the sharing of various infrastructures (Marshall, 1920; Krugman, 1991). Our hypotheses were also consistent with organizational theory, namely in arguing that network externalities occurred because firms wished to gain legitimacy and reduce uncertainty through imitation (DiMaggio and Powell, 1983; Levitt and March, 1988; Guillen, 2002; Henisz and Delios, 2001). We argued that the curvilinear relationship between the number of firms and the likelihood of co-location would be more apparent when agglomeration was motivated by a desire to gain legitimacy rather than by real economic gains. In demonstrating this relationship, our study improves understanding of co-location. This study has several limitations. First, we could not empirically distinguish economic reasons for agglomeration from reasons proposed by organizational theorists, such as uncertainty avoidance and legitimacy. Researchers should develop measures that can empirically distinguish these two sets of reasons other than simple count-based measures. Second, it is possible that the results of this study may reflect idiosyncrasies of Korean firms. Korea is closely located to China, and Koreans use Chinese characters, so it is easier for them to Copyright  2005 John Wiley & Sons, Ltd.

613

learn Chinese and absorb Chinese culture. Koreans also share some cultural heritage with China. Korean firms might thus feel more confident about operating in China relative to firms from other countries, which may explain why Korean firms spread into many geographic regions while foreign firms tended to concentrate in a few locations. Third, there might be strong network externalities among Korean firms because of ethnocentrism. Future research should examine whether MNCs originating in other countries show similar patterns of strong network externalities with firms of the same nationalities. Fourth, this study lumped all other foreign companies into one category. MNCs from other nations may derive stronger network externalities from firms of similar culture than they do from those of dissimilar ones. Disaggregating the ‘foreign firms’ category into subgroups according to cultural similarities may show much stronger relationships from foreign firms than the ones we documented. Similarly, this study did not explicitly consider relatedness/complementarities of industries. Considering industries at a more disaggregated level and examining relatedness among them would help capture network externalities more precisely. Fifth, this study focused only on the heterogeneity of firms that were already in China and had created network externalities, but it did not address the heterogeneity of investing firms themselves. Shaver and Flyer (2000) and Chung and Song (2004) argued that investing firms’ heterogeneity also affects co-location decisions. Further studies to assess the effects of such heterogeneity are warranted. This study demonstrated that firms tended to colocate with others to benefit from network externalities. Our results suggest, however, that this generalization is misleading. Because network externalities showed curvilinear patterns as we defined types of firms within the network more narrowly, above a certain threshold, negative network externalities could outweigh positive ones. Since a regional network comprises many types of firms, managers who make location decisions should balance between positive and negative network externalities. Furthermore, this study showed that some firms coordinated location decisions to hedge against the possible negative impacts of agglomeration in a few locations. LG Group’s strategy to spread the operations of its individual affiliates over many Strat. Mgmt. J., 26: 595–615 (2005)

614

S.-J. Chang and S. Park

regions is a good example. Business divisions in an affiliate may be attracted to the immediate benefits from locating near its own previous investment. Since most diversified MNCs are organized into semi-autonomous business divisions that make their own foreign entry decisions, the hazards of negative network externalities that come from too much agglomeration can be substantial. Thus, our results suggest that managers of MNCs should develop an overall foreign entry strategy to guard against the hazard of too much agglomeration in a few popular locations.

ACKNOWLEDGEMENTS We thank Jemo Chung, John Lafkas, Panseop Lee, Chunkyu Park, Hweonjung Park, Jaehyuk Rhee, seminar participants at National University of Singapore, New York University, London Business School, Copenhagen Business School and Norwegian School of Management, and two anonymous reviewers for helpful comments and suggestions. Financial support from the Korea Research Foundation (2004-041-B00246) is gratefully acknowledged. Additional support from Korea University Business School through an SK Distinguished Research Award is also gratefully acknowledged.

REFERENCES Alcacer J. 2004. Location choice across the value chain: how activity and capability influence co-location. Working paper, Stern School of Business, New York University. Appold S. 1995. Agglomeration, interorganizational networks and competitive performance in the U.S. metalworking sector. Economic Geography 71: 27–54. Arthur B. 1990. Positive feedback in the economy. Scientific American, February: 92–99. Barkema H, Vermeulen F. 1998. International expansion strategy through start-up or acquisition: a learning perspective. Academy of Management Journal 41: 7–27. Bastos, P, Greve, H. 2003. Interorganizational learning and the location of manufacturing subsidiaries: is chain migration also a corporate behavior? Advances in Strategic Management, Vol. 20, Baum JC, Sorenson O (eds). JAI Press: Greenwich, CT; 159–191. Baum J, Mezias S. 1992. Localized competition and organizational failure in the Manhattan hotel industry. Administrative Science Quarterly 37: 580–604. Chang S. 1995. International expansion strategy of Japanese firms: capability building through sequential entry. Academy of Management Journal 38: 383–407. Copyright  2005 John Wiley & Sons, Ltd.

Chang S. 2003. Financial Crisis and Transformation of Korean Business Groups: The Rise and Fall of Chaebols. Cambridge University Press: New York. Chang S, Rosenzweig P. 2001. The choice of entry mode in sequential foreign direct investment. Strategic Management Journal 22(8): 747–776. Chung W, Alcacer J. 2002. Knowledge seeking and location choice of foreign direct investment in the United States. Management Science 48: 1534–1554. Chung W, Kalnins A. 2001. Agglomeration effects and performance: a test of the Texas loading industry. Strategic Management Journal 22(8): 969–988. Chung W, Song J. 2004. Sequential investment, firm motivation and agglomeration of Japanese electronics firms in the United States. Journal of Economics and Management Strategy 13: 539–560. Coughlin C, Terza J, Arromdee V. 1991. State characteristics and the location of foreign direct investment within the United States. Review of Economics and Statistics 73: 675–673. Davis G. 1991. Agents without principles? The spread of the poison pill through the inter-corporate network. Administrative Science Quarterly 36: 569–596. DiMaggio P, Powell W. 1983. The iron cage revisited: institutional isomorphism and collective rationality in organizational fields. American Sociological Review 48: 147–160. Dunning J. 1988. Explaining International Production. Unwin Hyman: London. Export–Import Bank of Korea. 2002. Environments and Cases of Investment in China (in Korean). Export–Import Bank of Korea: Seoul, Korea. Fligstein N. 1985. The spread of the multidivisional form among large firms, 1919–1979. American Sociological Review 50: 377–391. Fombrun C, Shanley M. 1990. What’s in a name? Reputation building and corporate strategy. Academy of Management Journal 33: 233–258. Friedman J, Gerlowski D, Silberman, J. 1992. What attracts foreign multinational corporations? Evidence from branch plant location in the United States. Journal of Regional Science 32: 403–418. Granovetter M. 1995. Coase revisited: business groups in the modern economy. Industrial and Corporate Change 4: 93–130. Guillen M. 2002. Structural inertia, imitation, and foreign expansion: South Korean firms and business group in China, 1987–95. Academy of Management Journal 45: 509–525. Hannan M, Carroll G. 1992. The Dynamics of Organizational Populations. Oxford University Press: New York. Hansen M. 1999. The search-transfer problem: the role of weak ties in sharing knowledge across organizational subunit. Administrative Science Quarterly 44(1): 82–111. Haunschild P. 1993. Interorganizational imitation: the impact of interlocks on corporate acquisition activity. Administrative Science Quarterly 38: 564–592. Haunschild P, Miner A. 1997. Models of interorganizational imitation: the effects of outcome salience Strat. Mgmt. J., 26: 595–615 (2005)

Types of Firms Generating Network Externalities and uncertainty. Administrative Science Quarterly 42: 472–500. Hausman J, McFadden D. 1984. Specification tests for the multinomial logit model. Econometica 52: 1219–1240. Haveman H. 1993. Follow the leader: mimetic isomorphism and entry into new markets. Administrative Science Quarterly 38: 593–627. Head K, Ries J, Swenson D. 1995. Agglomeration benefits and location choice: evidence from Japanese manufacturing investments in the United States. Journal of International Economics 38: 223–247. Henderson V. 1986. Efficiency of resource usage and city size. Journal of Urban Economics 19: 47–70. Henisz W, Delios A. 2001. Uncertainty, imitation, and plant location: Japanese multinational corporations, 1990–1996. Administrative Science Quarterly 46: 443–475. Hennart J-F, Park Y. 1994. Location, governance and strategic determinants of Japanese manufacturing investment in the United States. Strategic Management Journal 15(6): 419–436. Hofstede G. 1980. Culture’s Consequence: International Differences in Work-Related Values. Sage: Beverly Hills, CA. Hymer S. 1960. The international operations of national firms: a study of direct foreign investment. PhD dissertation, Massachusetts Institute of Technology (published by MIT Press, 1996). Johanson J, Vahlne J. 1977. The internationalization process of the firm: a model of knowledge development and increasing foreign market commitment. Journal of International Business Studies 8: 23–32. Katz M, Shapiro C. 1985. Network externalities, competition, and compatibility. American Economic Review 75: 424–440. Knickerbocker F. 1973. Oligopolistic Reaction and the Multinational Enterprise. Harvard University Press: Cambridge, MA. Kogut B. 1983. Foreign direct investment as a sequential process. In The Multinational Corporation in the 1980s, Kindleberger CP, Audretsch D (eds). MIT Press: Cambridge, MA; 38–56. Kogut B, Chang S. 1996. Platform investment and volatile exchange rates. Review of Economics and Statistics 78: 221–231. Kogut B, Zander U. 1992. Knowledge of the firm, combinative capabilities and the replication of technology. Organization Science 3: 383–397. Krugman P. 1991. Increasing returns and economic geography. Journal of Political Economy 99: 483–499. Levitt B, March J. 1988. Organizational learning. Annual Review of Sociology 14: 319–340. Liebowitz S, Margolis S. 1995. Path dependence, lockin, and history. Journal of Law, Economics, and Organization 22: 1–26.

Copyright  2005 John Wiley & Sons, Ltd.

615

Marshall A. 1920. Principles of Economies (8th edn). Macmillan: London. Martin X, Swaminathan A, Mitchell W. 1998. Organizational evolution in the interorganizational environment: incentives and constraints on international expansion strategy. Administrative Science Quarterly 43: 566–601. McFadden D. 1974. Conditional logit analysis of qualitative choice behavior. In Frontiers in Econometrics, Zarembka P (ed). Academic Press: New York; 105–142. Porac J, Rosa J. 1996. Rivalry, industry models, and the cognitive embeddedness of the comparable firm. In Advances in Strategic Management , Vol. 13, Shrivastava P, Huff A, Dutton J (eds). JAI Press: Greenwich, CT; 363–388. Porter M. 1998. Clustering and the new economics of competition. Harvard Business Review 76(6): 77–90. Porter M, Stern S. 2001. Innovation: location matters. Sloan Management Review 42(2): 28–36. Saxenian A. 1994. Regional Advantage. Harvard University Press: Cambridge, MA. Shaver J, Flyer F. 2000. Agglomeration economics, firm heterogeneity, and foreign direct investment in the United States. Strategic Management Journal 21(12): 1175–1193. Shaver J, Mitchell W, Yeung B. 1997. The effect of own-firm and other-firm experience on foreign direct investment survival in the United States, 1987–92. Strategic Management Journal 18(10): 811–824. Smith D, Florida R. 1994. Agglomeration and industry location: an econometric analysis of Japaneseaffiliated manufacturing establishments in automotiverelated industries. Journal of Urban Economics 36: 23–41. Song J. 2002. Firm capabilities and technology ladders: sequential foreign direct investments of Japanese electronics firms in East Asia. Strategic Management Journal 21(3): 191–210. Suchman M. 1995. Managing legitimacy: strategic and institutional approaches. Academy of Management Review 20: 571–610. Sun Y. 2000. Spatial distribution of patent in China. Regional Studies 34: 441–454. Szulanski G. 1996. Exploring internal stickness: impediments to the transfer of best practice within the firm. Strategic Management Journal , Winter Special Issue 17: 27–43. Wheeler D, Mody A. 1991. International investment location decisions: the case of U.S. firms. Journal of International Economics 33: 57–76. Zander U, Kogut B. 1995. Knowledge and the speed of the transfer and imitation of organizational capabilities: an empirical test. Organizational Science 6: 76–92.

Strat. Mgmt. J., 26: 595–615 (2005)