Abstract

Given the increasing significance of knowledge spillovers in innovation, this study investigates and compares knowledge spillovers from arm’s length firms in the industries (market) with those from sister firms in the same business group (network). By dividing the knowledge pool into pools within and outside a sector to which a firm belongs, we reexamine the ongoing debate on the relative size of the intra- versus inter-sector spillovers, and address a new question on the relative size of spillovers from networks compared with that from arm’s length industries. We find that, although both intra- and inter-sector spillovers are significant, no evidence proves the dominance of either type of spillover, whether the spillover is from industries or from networks. More importantly, we find that spillovers from networks are greater than those from industries, regardless of whether the comparison was made between intra- and intra-, inter- and inter-, intra- and inter-, or inter- and intra-sector spillovers. Results imply that knowledge spillover is not automatic, and knowledge can be transferred better through direct interaction, which is more prevalent within network organizations, such as business groups.

1. Introduction

Knowledge is a public good to a certain extent. Thus, an innovating firm usually cannot prevent other firms from using its new product or process inventions, leading other firms to achieve more innovations by exerting less effort ( Jaffe, 1986 ; Medda and Piga, 2007 ). Other companies can use the knowledge of a firm at lower (or zero) costs because such knowledge is usually shareable, inexhaustible, and reusable. Knowledge can spill over through various routes, including upstream and downstream linkages, learning by doing and observing, movement of workers involved in research and development (R&D) activities, and various local networks between scientific and engineering personnel from different organizations ( Hubert and Pain, 2001 ; Medda and Piga, 2007 ; Plunket, 2009 ; Desrochers and Leppälä, 2010; Dindaroglu, 2010 ). Firms located in a spatial neighborhood (i.e., in the same cluster) or in a conceptual neighborhood (i.e., in the same industry or network) benefit from the knowledge-creation activities of other firms.

Given its existence in diverse contexts and channels, knowledge spillover has been the subject of a large number of theoretical and empirical studies with different foci. One of the early theoretical studies conducted is that of d’Aspremont and Jacquemin (1988) , which finds that the optimal amount of R&D in cooperative R&D in a duopoly will be larger with knowledge spillover than without it. This research has led to several extensions. 1 Empirical studies on knowledge spillovers are well surveyed in the work of Griliches (1992) , and defining the spillover pool, both domestic and international, has been a key issue in empirical literature. Jaffe (1986) constructs the potential spillover pool of an industry as the sum of the innovative activities of other firms in the industry. Empirical research confirms the effects of knowledge spillover from both domestic and international pools ( Geroski, 1995 ; Coe and Helpman, 1995 ; Adams and Jaffe, 1996 ).

Diverse results from empirical analyses can be observed by examining further the details of a spillover. For instance, Wakelin (2001) finds that companies in innovation-using sectors appear to benefit more from the R&D of other firms in the sector than from their own, whereas innovation-producing sectors do not appear to benefit from their own R&D expenditure or from that of others. Given this situation, Kafouros and Buckley (2008) address the emerging issues, such as the conditions under which firms benefit from spillovers, and the relative size of inter- and intra-sector spillovers. Although Kafouros and Buckley (2008) search for such conditions (e.g., firm size, technological opportunities, and competition), other reasons also exist. The present study pays attention to the inherent tacitness of knowledge, which tends to restrict the degree of knowledge flows in terms of transferability to and learning by other firms. An argument is that spillover and transfer among sister firms affiliated with the network organizations, similar to business groups, may be less subject to such limitations. However, only a few studies analyze the spillover impact of a knowledge pool of a network consisting of affiliates in a business group and compare them with spillover among arm’s length firms in the industry.

Business groups, which consist of legally independent firms operating in multiple markets, are a common phenomenon in emerging markets and in several developed economies. This form of organization has attracted increasing interest in economic and business studies, as reviewed by an article in the Journal of Economic Literature ( Khanna and Yishay, 2007 ). Earlier literature on business groups has focused on their emergence in an environment with market failures or institutional voids. Therefore, a business group is perceived as an organizational device for internalizing transactions that are too costly to happen in the markets ( Leff, 1978 ; Goto, 1982 ; Khanna and Palepu, 1997 , 2000 ). A resource-based view of a business group stresses developing and sharing certain capabilities among firms affiliated with the group ( Kock and Guillen, 2001 ). Chang and Hong (2000) confirm the positive profitability impacts of group-level resource variables, such as advertisement, intra-group transactions, and R&D expenditure. Resource (including knowledge) sharing among affiliates of a business group is more logical when certain resources are not available in the markets or when the benefits from such sharing within a network are greater than the benefits from market transactions, regardless of market failure ( Cheong et al. , 2010 ). Thus, the purpose of the present study is to compare the size of spillover impacts from a network with that from arm’s length industries.

This study on spillover within a business group can be compared with studies on spillover within another form of a network or Multinational Enterprises (MNEs). In this regard, Almeida et al. (2002) demonstrated that MNEs are superior to arms-length contractual agreements in transferring knowledge across countries, possibly due to less “resource costs.” 2 An interesting issue in the studies on MNEs is the so-called competition effect, which may interfere with the spillover among subsidiaries of MNEs because subsidiaries do not want to lose their uniqueness and bargaining power within the MNE upon transferring their knowledge to other subsidiaries ( Levitt and March, 1988 ; Forsgren, 1997 ; Birkinshaw and Hood, 1998 ; Hansen et al. , 2005 ). In a more general context, competition among firms in the same sector has been regarded as a mechanism that counterbalances the intra-sector spillovers ( Aitken and Harrison, 1999 ; Javorcik, 2004 ). Consequently, an interesting question emerges; that is, whether or not such counterbalancing effects are smaller among firms belonging to the same controllership, such as the same MNEs or business groups, than among firms in an arm’s length relationship.

From this perspective, business groups and MNEs may be considered as sitting on the same side, but they also differ in important ways. First, affiliates in the business groups, especially those in Korea, are legally independent of each other and listed separately; thus, with their own shareholders. More importantly, these affiliates are quite diverse in terms of their sectoral orientations and generally belong to different sectors. In contrast, subsidiaries of a MNE tend to be in the same sector as the parent firm. Moreover, there is often no clear-cut parent firm in the business groups unless the groups belong to the simple vertical hierarchy or to a structure where one group is the holding company and others are subordinate affiliates. In this regard, a typical business group in Korea generally belongs to the relatively complicated, matrix-type cross-ownership among the affiliates and not to the simple pyramid structure. Furthermore, each affiliate is most likely an MNE with numerous overseas subsidiaries. 3

The current study focuses on the function of a business group as an effective organization in expanding the knowledge resources of a firm, thus boosting productivity. We construct the knowledge pool of a business group (network) as an analogy to that of an industry and then compare the relative size of these kinds of spillovers, namely, spillovers from networks and industries. We initially test the hypothesis that productivity impacts of knowledge spillover pools from business groups may be greater than those of industry spillover pools. Branstetter (2000) observes the vertical keiretsu of Japan as a valuable economic institution that internalizes knowledge spillovers and verifies spillover impacts on the basis of the weighted R&D expenditure of other firms as a knowledge pool available to a firm. However, Branstetter (2000) does not compare the relative size of the impacts from the network and that from the industry. Assessing the impact of intra- and inter-sector spillovers both at the network and industry levels is the concern of the second hypothesis of the present study. This issue is becoming more relevant in the current era characterized by an increasing trend in technology fusion, but it is not addressed in the literature on business groups.

Regarding inter- versus intra-sector spillovers, a number of scholars suggest that the former is more significant than the latter in explaining economic growth, social returns, or productivity ( Hubert and Pain, 2001 ; Harris and Robinson, 2004 ; Kafouros and Buckley, 2008 ) for firms in highly competitive environments. Moreover, recent studies provide limited evidence on intra-sector technology spillovers, whereas significant inter-sector spillovers are commonly reported ( van Stel and Nieuwenhuijsen, 2004 ; Javorcik, 2004 ; Bwalya, 2006 ; Kugler, 2006 ; Badinger and Egger, 2010). This reasoning is consistent with the idea of technology fusion proposed by Kodama (1992) and Suzuki and Kodama (2004) ; that is, persistent technological diversification is necessary for the survival and long-term growth of a technology-based firm. Furthermore, spillovers cannot be understood in a one-dimensional manner. A number of studies combine the geographical dimension with the sectoral dimension. Autant-Bernard and LeSage (2011) analyzes the nature of intra- and inter-sector knowledge spillovers both within regions and between regions, and compares the relative size of spillover impacts.

However, to our knowledge, no study has yet examined intra- versus inter-sector spillovers among firms affiliated with a business group. Thus, we test a hypothesis on whether or not affiliates in a business group obtain more spillovers from sister firms in different sectors than from those in the same sector. In the present study, we use sales per employee as a dependent variable, which is a definite and simple measure of productivity that is less subjected to data noise. This variable is used in existing literature, such as in the work of Kafouros and Buckley (2008) . Then, we regress this variable on the variables measuring the sizes of knowledge pools in networks and arm’s length industries, controlling other factors. Robustness tests are also conducted using total factor productivity (TFP).

The current study proceeds as follows. Section 2 reviews the related literature and derives the two hypotheses on the relative sizes of knowledge spillovers from which an affiliate of a business group benefits. Section 3 describes the methodology and data used in this study. Section 4 presents the main results on the relationship between knowledge spillover and firm productivity. Section 5 concludes the study.

2. Literature and hypotheses: knowledge spillovers and business groups

Knowledge can be used at a lower marginal cost or without any cost because it is reusable and nonexclusive. Consequently, knowledge accumulated by a firm can broaden the technology base of other firms without appropriate compensation for the former ( Glaeser et al. , 1992 ). Once developed by a firm, knowledge enhances the knowledge production of other firms because new knowledge can be built upon existing knowledge without exhaustion ( Laursen and Meliciani, 2000 ). Knowledge has the inherent properties (at least in a partial sense) of non-excludability and non-rivalry. Thus, the production function of a firm depends on the level of knowledge available in the economy and on its own inputs ( Jaffe, 1986 ; Medda and Piga, 2007 ). 4

According to the perspective of transaction cost economics, firms use internal capital or labor markets to reduce transaction costs resulting from market imperfections. An internal market is less costly than an external one in the presence of asymmetric information. Firms conduct their own R&D and consequently generate diverse technologies. Technology is an input factor of a firm that can be traded in the markets. Nevertheless, the tacitness of technological knowledge makes transferring it completely through the markets difficult. Therefore, a number of firms acquire technology from intra-firm markets. Conglomerates or business groups are good examples of internal markets. These organizations replace external markets with trades or transfers occurring within the boundaries of related firms or business groups. The economic logic behind business groups can be a starting point of an argument on the positive benefits for group-affiliated firms resulting from the knowledge pool of the mother group. A rapidly changing business environment tends to lead firms toward requiring and promoting more cross-fertilization among technological areas. If a firm truly enjoys cross-fertilization among the technologies within its boundary, then spillovers will probably occur across firms within a business group (widened boundary of a firm). Branstetter (2000) argues that the vertical keiretsu of Japan is a valuable economic institution that internalizes knowledge spillovers.

Diversified business groups are common in emerging and developed economies ( Khanna and Yishay, 2007 ). According to the transaction cost perspective, when the cost of acquiring inputs through markets is extremely high, firms extend their boundaries and internalize rather than externalize their functions. During the early period of industrialization, market factors, such as skilled labor, capital, or technology, are nonexistent or incomplete. Under the circumstance of market imperfections, business groups replace poorly performing or nonexistent economic institutions, such as banks and external labor markets ( Ghemawat and Kanna, 1998 ; Chang and Hong, 2000 ). For example, in Korean business groups or chaebols, certain key activities (e.g., recruiting, training, and promoting employees; advertising; and R&D) have been conducted at the group level until recently. The labor mobility of scientists and engineers is a conduit for knowledge spillovers among firms ( Kim and Marschke, 2005 ; Dindaroglu, 2010 ). The higher mobility of skilled labor among affiliates may facilitate the transfer of knowledge better within the business group than within the market.

According to the perspective of resource-based theory of the firm, business groups can be effective organizations that promote the more efficient use of knowledge resources of a firm. A resource-based view of a firm stresses the sharing of resources among affiliates within the group ( Chang and Hong, 2000 ). In particular, reusable and inexhaustible resources, such as technologies and managerial resources, are transferred to other affiliates to generate efficiency-enhancing results. More efficient resource utilization is possible by pooling resources at the business group level and sharing them among affiliates. Intra-group transactions among affiliates can offer lower prices because information asymmetry is less notable among firms affiliated with the same group. The redistribution of profits from good to bad performers by the group headquarters can incur costs that are unique to business group affiliates ( George et al. , 2004 ).

Firms generally do not use technologies solely for internal production purposes ( Cesaroni, 2004 ). A number of underutilized technologies may exist even in firms that are highly diversified across product lines ( Cheong et al. , 2010 ). Cesaroni (2004) points out that the possibility of recovering the costs of R&D activities through in-house exploitation is drastically reduced when technological diversification is directed toward fields that are marginal to core technological competencies and if the firm is not facing a sufficiently large demand. Nevertheless, the recovery of R&D costs at the group level is easier through intra-group technology transactions and transfers when confronted with an imperfect technology market. 5 The presence of business groups enables affiliates to more readily develop technologies that are marginal to their core competencies. In a technologically diversified group, the chances of finding affiliates within the group that are underutilizing their technologies are high. In addition, the potential for knowledge cross-fertilization is better exploited within a business group than in a stand-alone firm. This reasoning is logical because knowledge is not exhausted but improved by the learning process through repeated use. Moreover, intra-group transfers of technologies are more efficient and faster than transfers among arm’s length firms ( Granstrand, 1999 ).

Therefore, we propose that knowledge pools at the group level affect the innovations of affiliated firms and their productivity. Given that member firms of a business group are closely interconnected, the influence of spillover pools from the network may be stronger than that of broader spillover pools from arm’s length industries. Thus, the first main hypothesis (H1) of this study is that productivity impacts of spillovers from firms affiliated with the same business group (network) are greater than those of spillovers from other firms in industries. To our knowledge, no other research, except that of Branstetter (2000) , has focused on spillovers from both an affiliated business group and an industry. However, Branstetter (2000) , who uses weighted R&D expenditure rather than patent applications to measure knowledge pools, does not address the relative size of impacts arising from different types of spillover pools, such as networks (business groups) and arm’s length industries.

The next hypothesis deals with comparing spillovers from the same and from different sectors. This issue has not been addressed in the literature on business groups, but it is becoming more significant, given the increasing trend toward technology fusion ( Kodama, 1992 ). Ideas from a certain sector can induce new ideas in another sector. 6 This type of spillover can emerge from the mobility of R&D employees, upstream (downstream) relationships with suppliers (customers), and public information contained in patents, scientific journals, conferences, and so on ( Los, 2000 ). In literature concerning the spillover effects of knowledge, a number of studies have addressed the issue of the relative size of inter- and intra-sector spillovers ( Bernstein, 1988 ; Rouvinen, 2002 ; Kafouros and Buckley, 2008 ). However, no previous study has focused yet on the effects of inter- and intra-sector spillovers within a business group and has compared spillovers from affiliates in different sectors and spillovers from affiliates in the same sectors. Early research on spillovers has mainly focused on intra-sector effects, such as positive externalities from subsidiaries of multinational corporations (MNCs) to domestic firms in the same sector. By contrast, results of recent studies are not conclusive with regard to intra-sector spillovers (possibly because of the competition effect). This outcome has caused researchers to switch their attention from intra-sector to inter-sector spillovers ( Sasidharan, 2006 ).

Table 1 summarizes the studies on inter- or intra-sector spillovers from R&D or foreign direct investment. Hubert and Pain (2001) argue that spillovers across sectors should be considered in policy making as an additional channel of spillovers. Studies attempting to identify intra-sector spillovers may underestimate externalities from inward investment by foreign-owned companies ( Hubert and Pain, 2001 ). Several recent studies suggest that inter-sector spillovers are more prevalent than intra-sector spillovers in explaining economic growth or social returns ( Harris and Robinson, 2004 ; Javorcik, 2004 ; van Stel and Nieuwenhuijsen, 2004 ; Kugler, 2006 ; Jordaan, 2008 ; Badinger and Egger, 2010). Moreover, Bwalya (2006) finds limited evidence on intra-industry spillovers from foreign to local firms through horizontal channels but cites significant evidence on inter-industry spillovers through backward and forward linkages. Criscuolo (2005) discusses different forms of researchers’ mobility within MNEs in terms of the possibly different degrees of knowledge transfer among subsidiaries.

Table 1

Literature comparing intra-sector and inter-sector spillovers

Studies Results Data Types 
Bernstein (1988) Inter-industry spillover to the social return is virtually the same and small for all industries. / Differentials between social and private returns and between social returns across industries depend on the extent of the intra-industry spillovers. Canada R&D spillovers 
Hubert and Pain (2001) There is evidence of significant intra-industry and inter-industry spillovers. / Both intra- and inter-industry effects are significant, and although inter-industry spillovers are marginally larger than intra-industry spillovers, the hypothesis of common coefficients cannot be rejected. UK FDI Spillovers 
Harris and Robinson (2004) Inter-industry spillovers are generally more prevalent than intra-industry spillovers. UK FDI spillovers 
van Stel and Nieuwenhuijsen (2004) No empirical evidence for a positive relationship between intra-industry spillovers and value added growth / evidence for positive relationships between inter-industry spillovers and value added growth. The Netherlands Regional spillovers 
Javorcik (2004) No evidence of intra-industry spillovers / positive productivity spillovers from FDI taking place through contacts between foreign affiliates and their local suppliers in upstream sectors (evidence of inter-industry spillovers). Lithuania FDI spillovers 
Bwalya (2006) There is little evidence in support of intra-industry spillovers, but significant inter-industry spillovers. Zambia FDI spillovers 
Kugler (2006) Knowledge spills over between but not within industries. Colombia FDI spillovers 
Medda and Piga (2007) Firms also benefit from spillovers originating from their own industries, as well as innovative upstream sectors (inter-industry spillovers). Italy R&D spillovers 
Kafouros and Buckley (2008) Both inter- and intra-industry spillovers are significant, but the magnitude of spillovers is higher for the R&D undertaken by intra-industry firms. / For firms in the environments of high competition, inter-industry spillovers outweigh intra-industry spillovers. UK R&D spillovers 
Jordaan (2008) The presence of FDI creates negative externalities within industries and positive externalities between industries through backward linkages. Mexico FDI spillovers 
Badinger and Egger (2010) Inter-industry spillovers dominate intra-industry spillovers, which turn out much smaller and even insignificant in some specifications. 13 OECD countries R&D spillovers 
Autant-Bernard and LeSage (2011) The largest direct and indirect effects are associated with private R&D activity that spills across industry boundaries. / But, inter-industry spillovers decrease more drastically with distance than intra-industry spillovers. France Regional spillovers 
Studies Results Data Types 
Bernstein (1988) Inter-industry spillover to the social return is virtually the same and small for all industries. / Differentials between social and private returns and between social returns across industries depend on the extent of the intra-industry spillovers. Canada R&D spillovers 
Hubert and Pain (2001) There is evidence of significant intra-industry and inter-industry spillovers. / Both intra- and inter-industry effects are significant, and although inter-industry spillovers are marginally larger than intra-industry spillovers, the hypothesis of common coefficients cannot be rejected. UK FDI Spillovers 
Harris and Robinson (2004) Inter-industry spillovers are generally more prevalent than intra-industry spillovers. UK FDI spillovers 
van Stel and Nieuwenhuijsen (2004) No empirical evidence for a positive relationship between intra-industry spillovers and value added growth / evidence for positive relationships between inter-industry spillovers and value added growth. The Netherlands Regional spillovers 
Javorcik (2004) No evidence of intra-industry spillovers / positive productivity spillovers from FDI taking place through contacts between foreign affiliates and their local suppliers in upstream sectors (evidence of inter-industry spillovers). Lithuania FDI spillovers 
Bwalya (2006) There is little evidence in support of intra-industry spillovers, but significant inter-industry spillovers. Zambia FDI spillovers 
Kugler (2006) Knowledge spills over between but not within industries. Colombia FDI spillovers 
Medda and Piga (2007) Firms also benefit from spillovers originating from their own industries, as well as innovative upstream sectors (inter-industry spillovers). Italy R&D spillovers 
Kafouros and Buckley (2008) Both inter- and intra-industry spillovers are significant, but the magnitude of spillovers is higher for the R&D undertaken by intra-industry firms. / For firms in the environments of high competition, inter-industry spillovers outweigh intra-industry spillovers. UK R&D spillovers 
Jordaan (2008) The presence of FDI creates negative externalities within industries and positive externalities between industries through backward linkages. Mexico FDI spillovers 
Badinger and Egger (2010) Inter-industry spillovers dominate intra-industry spillovers, which turn out much smaller and even insignificant in some specifications. 13 OECD countries R&D spillovers 
Autant-Bernard and LeSage (2011) The largest direct and indirect effects are associated with private R&D activity that spills across industry boundaries. / But, inter-industry spillovers decrease more drastically with distance than intra-industry spillovers. France Regional spillovers 

The dominance of externalities across sectors over those within a sector can be explained as follows. Inter-sector spillovers occur through backward and forward linkages between buyers and sellers, whereas intra-sector spillovers occur through imitation, licensing, competition, or labor mobility ( Harris and Robinson, 2004 ). An innovative firm has an incentive to facilitate knowledge transfers to upstream or downstream firms, thus enabling recipients to produce intermediate inputs or equipment more efficiently. Therefore, adverse competition effects are more likely to happen within a sector ( Bwalya, 2006 ). Knowledge spillovers within a sector may be counterbalanced by competition effects, a situation that leads to a lack of intra-sector spillovers, according to several studies ( Aitken and Harrison, 1999 ; Javorcik, 2004 ). By contrast, a greater potential for spillovers exists through forward and backward linkages, given that supplier–buyer relationships have grown stronger because of technological complexity ( Kugler, 2006 ). Kugler (2006) also suggests that generic technology, which can be deployed easily in production across sectors, is more likely to be propagated and requires fewer absorptive capacities than sector-specific technology. Suzuki and Kodama (2004) argue that persistent technological diversification through fusion across internal or external technologies is necessary for the survival and long-term growth of technology-based firms. Many examples of the inter-sectoral spillover from the industry and the importance of fusing technologies from diverse sectors can be found. 7

In general, the relative importance of intra- and inter-sector spillovers should consider the possible different impacts of specialization versus diversity on productivity ( van Stel and Nieuwenhuijsen, 2004 ). Although specialization facilitates spillovers among firms in the same sector, either through direct contacts and the mobility of skilled labor, diversity fosters spillovers among firms in different sectors through cross-fertilization of ideas across different lines of work ( van Stel and Nieuwenhuijsen, 2004 ; Plunket, 2009 ; Desrochers and Leppälä, 2010). Several studies indicate that industries in a region grow faster when the region is less specialized ( Jacobs, 1969 ; Glaeser et al. , 1992 ). The specialization viewpoint asserts that spillovers are more likely to occur among similar firms that share common knowledge. By contrast, the diversification viewpoint emphasizes that cross-fertilization and complementarities among firms enhance knowledge spillovers ( Autant-Bernard and LeSage, 2011 ).

Another stream of theoretical literature posits the idea of organizational and cognitive distance ( Nooteboom et al. , 2007 ), and finds that such distance has an inverted U shape with respect to the value of learning ( Wuyts et al. , 2005 ). This finding implies that spillover or learning effects are highest when a certain (optimal) level of distance exists. This new theoretical argument on optimal distance can be considered a “third” view, which does not support either of the views on the relative size between inter-sector and intra-sector spillovers. 8

The preceding discussion suggests that formulating a theoretical argument in favor of the dominance of either inter- or intra-sector spillovers is not easy because it depends on the relative importance of diversity versus specialization or optimal distance. Thus, the second main hypothesis (H2) of this study is that a firm in a network (business group) will obtain spillovers from sister firms in different sectors as much as from those in the same sector within the business group.

The two main hypotheses explained above may lead to several specific sub-hypotheses, if we consider the four sources of knowledge spillover pools, namely, (i) intra-sector knowledge pool of arm’s length industries, (ii) intra-sector pool of a network consisting of affiliates of business groups, (iii) inter-sector knowledge pool of arm’s length industries, and (iv) inter-sector pool of a network consisting of affiliates of business groups. Through a diverse combination of these four pools, we generated the following four sub-hypotheses.

First, two sub-hypotheses were derived from the first main hypothesis (H1) on comparing the network and arm’s length industries, first, in terms of intra-sector spillovers (H1-A), and, second, in terms of the size of inter-sector spillovers (H1-B).

Two sub-hypotheses derived from the first main hypothesis are:

(H1-A) The size of intra-sector spillover from firms in a network (business group) is larger than the intra-sector spillover from firms in arm’s length relationships (industries).

(H1-B) The size of inter-sector spillover from firms in a network (business group) is larger than the inter-sector spillovers from firms in arm’s length relationships (industries).

Second, two sub-hypotheses were likewise derived from the second main hypothesis (H2) on comparing the intra- and inter-sector spillover, the first of which is in the context of business groups (H2-A), and, the second, in the context of arm’s length industries (H2-B).

Two sub-hypotheses derived from the second main hypothesis are:

(H2-A) The size of intra-sector spillover in the same network (business groups) may not be significantly different from that of inter-sector spillover in the same network.

(H2-B) The size of intra-sector spillover from firms in arm’s length relationships may not be significantly different from that of inter-sector spillovers from firms in arm’s length relationships.

Although these four hypotheses are the main focus of this study, various specifications in regressions allowed us to deal with two more possible comparisons on the side:

(H3-A) Intra-sector spillover within a network is larger (or smaller) than inter-sector spillovers from firms in arm’s length relationship.

(H3-B) Inter-sector spillover within a network is larger (or smaller) than intra-sector spillovers from firms in arm’s length relationship.

3. Data and research method

3.1 Business groups in Korea

The planned analysis requires data on firms interacting within a network, and we find Korean firms affiliated with business groups to be a good choice. Korea is a country where business groups are still extensively spread as a key form of firms in the national economy. Data on Korean firms, which clearly identify affiliates over time, are widely used in academic research published in international journals. Korea also fulfills another requirement, that is, the patent data for each affiliated firm should be available so that their knowledge spillover pools could be identified.

An institutional basis for knowledge spillover among affiliates in business groups is group-level or centralized personnel management, which includes hiring and training of new staff by the group headquarters rather than by each affiliate ( Chang and Hong, 2000 ; Chang, 2003 : 93–94). This practice was in effect until the mid-1990s in the case of the oldest and largest business groups, although certain business groups, such as STX, still maintain such a system. For example, in the case of Samsung Group, one of the oldest business groups in Korea, group-level hiring of workers and centralized training started as early as 1957 and was maintained until as late as 1994. 9

Group-level management of human resources includes, among others, group-level advertisement of recruitment and 3–4 weeks’ training of newly hired employees, during which they train, eat, and rest together. After this common training period, the newly recruited workers are assigned to different affiliates. Their job applications are not for specific affiliates but are considered for the entire group, and thus they should not complain about which affiliate they will be assigned to in their initial jobs. Furthermore, each group maintains group-level retraining sessions for newly promoted senior managers, and these officers, particularly the directors, are rotated among affiliates and are subjected to group-level personnel management. Samsung maintains a group-level retraining organization called the Samsung Human Capital Development Center in the outskirts of Seoul. This organization is responsible for training newly promoted senior managers from all affiliates.

The LG Group has a similar program called the LG Academy. Similar to Samsung, each group has a special personnel management committee at its headquarters to make decisions on job rotation, promotion, and discipline of senior managers from all affiliates. This committee communicates directly with the group chairman or the controlling families.

Although the practice of group-level or centralized recruitment of new staff has decreased, business groups still tend to form and manage task forces for specific matters, including R&D projects, entry into new business areas, and major overseas investments. These task forces are composed of the most talented or specialized employees from various affiliates.

The aforementioned data suggest that the staff of Korean business groups tend to know each other from the beginning of their employment and cross paths throughout their careers, regardless of their initial affiliation. They often work together and are rotated across boundaries of affiliates but within the boundary of the group. Korean business groups are different from American-style conglomerates or MNCs, which have clearly identified parent firm and affiliates. In Korean business groups, all affiliate firms are legally independent, although they are under a common controlling family. Moreover, although several core-business companies exist, no clear distinction exists between parent and subsidiary firms because the equity ties among them are typically not in the shape of a pyramid but of a matrix (or a circle) ( Chang, 2003 : Ch. 6; Choo et al. , 2009 ). Each affiliate in a business group has its own R&D unit, and each affiliate applies and files patents in the name of each affiliate, which is responsible for relevant R&D projects. A group-level, independent R&D organization also exists, such as the Samsung Advanced Technology Institute ( Chang, 2003 : 87). However, this R&D institute focuses on basic or fundamental research and is not concerned with short-term R&D applications, which are conducted by each affiliate for its own specific purposes.

3.2 Data sources

The Korean firm data used in this study consist of distinctive data sets from two different databases, namely, typical financial statement data and patent data of firms.

First, financial data on Korean firms obtained from the Korean Information Service (KIS)-Value, the database of the KIS Company, have been used in many empirical studies focusing on Korean firms ( Chang and Hong, 2000 ; Choo et al. , 2009 ; Cheong et al. , 2010 ). The database covers all detailed items reported in financial statements and includes firms listed in the Korea Stock Exchange, the Korea Securities Dealers Automated Quotation (the Korean version of the NASDAQ Stock Market), and all externally audited firms. The firms belonging to business groups are then identified. Information on business groups and their affiliates is obtained from Choo et al. (2009) . The definition of group used in this study is broader than that of the typical top 30 chaebols designated and monitored by the Korea Fair Trade Commission. The number of affiliates of a business group varies each year, depending on the exits from and entries to the group. Therefore, the number of group affiliates in the sample of the present study is not fixed. For a group to be included in the sample, it should have more than two affiliates each year during the 7-year sampling period (1991–1997). In addition, an eligible group should have more than two patent applications and should remain in the KIS data set throughout the sampling period; otherwise, the group is excluded. Thus, the sample includes 79 groups and 417 affiliated firms. The number of firm-year observations is 2,242. Table 2 presents the names of the groups included in the sample and the number of affiliates of each group.

Table 2

List of business groups included in the sample (as of 1997)

Group name  Number of affiliates
 
Group name  Number of affiliates
 
Manufacturing and externally audited firms a  All b  Manufacturing and externally audited firms a  All b 
Samsung 24 60 Seah 11 
Hyundai 17 52 Daenong 
LG 15 43 DPI 
Dongyang Chem. 13 18 Nongshim 
Hanwha 11 32 Woonsan 
Daewoo 10 28 Dongsung Chem. 
Tongil 10 14 Hankook Elecronics 
Lotte 29 Samyang Food 
Posco 14 Samlip Food 
Sinho 31 Byucksan 14 
Daesang 26 Sambo Computer 13 
Dongkuk Steel 16 Haitai 13 
Hyosung 16 Kangwon Ind. 13 
Hwaseung 13 Sungsin Cement 10 
Kapul 12 Samchully 
Aekyung 11 Woongjin 
SK 42 Iljin 
Ssangyong 26 Daewong Pharm. 
Kia 26 Kirin 
Kolon 23 Hwachun 
Daesung 16 Ilshin 
Koryeo 12 Poongsan 
Pacific Corp. 11 Dongoh 
Yuhan Dongwon 
Daelim 19 Taekwang 
Samyang Sindonga 
Sungwoo Hanil Cement 
Doosan 24 Kyesung 
Anam 19 Hite 
Jinro 19 Taihan Electric Wire 
Yoongpoong 18 Haesung 
Halla 16 Rocket 
Hanchang Crown 
Samwha Choongbang 
Kangnam Samick 
Chongkeundang Kohap 13 
Hanglas Sammi 
Daewon Steel Sindoh 
Kumho 26 Kookje Pharm. 
Hanil 11    
Group name  Number of affiliates
 
Group name  Number of affiliates
 
Manufacturing and externally audited firms a  All b  Manufacturing and externally audited firms a  All b 
Samsung 24 60 Seah 11 
Hyundai 17 52 Daenong 
LG 15 43 DPI 
Dongyang Chem. 13 18 Nongshim 
Hanwha 11 32 Woonsan 
Daewoo 10 28 Dongsung Chem. 
Tongil 10 14 Hankook Elecronics 
Lotte 29 Samyang Food 
Posco 14 Samlip Food 
Sinho 31 Byucksan 14 
Daesang 26 Sambo Computer 13 
Dongkuk Steel 16 Haitai 13 
Hyosung 16 Kangwon Ind. 13 
Hwaseung 13 Sungsin Cement 10 
Kapul 12 Samchully 
Aekyung 11 Woongjin 
SK 42 Iljin 
Ssangyong 26 Daewong Pharm. 
Kia 26 Kirin 
Kolon 23 Hwachun 
Daesung 16 Ilshin 
Koryeo 12 Poongsan 
Pacific Corp. 11 Dongoh 
Yuhan Dongwon 
Daelim 19 Taekwang 
Samyang Sindonga 
Sungwoo Hanil Cement 
Doosan 24 Kyesung 
Anam 19 Hite 
Jinro 19 Taihan Electric Wire 
Yoongpoong 18 Haesung 
Halla 16 Rocket 
Hanchang Crown 
Samwha Choongbang 
Kangnam Samick 
Chongkeundang Kohap 13 
Hanglas Sammi 
Daewon Steel Sindoh 
Kumho 26 Kookje Pharm. 
Hanil 11    

Note : a The figure includes only the firms that are under obligation to receive external auditing.

b The number of affiliates for each group is counted using the list of business groups and affiliates from Maekyoung Business Yearbook. The firms that are free from the burden of external auditing are also included in the figure.

The second data set consists of patent data of Korean firms. Patent data refer to the output of innovation activities of a firm and serve as proxy for a firm’s technological capabilities. The current study uses patent applications filed with the Korean Intellectual Property Office (KIPO) from 1989 to 1997. Patent data can be downloaded from the Korea Intellectual Property Rights Information Service ( http://www.kipris.or.kr ), a publicly accessible Web-based patent database supported by KIPO. The information contained in a patent application includes the name of the applicants, renewal fee status, final decision on patentability, International Patent Classification codes, inventors, and abstracts. We have downloaded approximately 10,000 text files and arranged them by variables using software such as SAS and Ultraedit. We then match the applicants in the patent data with the company names in the financial data. Table 3 shows the trend of patent applications filed at KIPO. The table indicates that the technological capabilities of Korean firms, as represented by patent stocks, have been accumulating rapidly. The number of patent applications by local firms exceeded that of foreign applications in 1992. In 1993, domestic corporations overtook foreign firms in the number of applications. By 1997, the number of patent applications by domestic corporations reached 67,346, a sevenfold increase from 9,082 in 1990. The number of applications by domestic corporations decreased by approximately 35% in 1998 as a result of the 1997 Asian financial crisis and did not fully recover even in 2002.

Table 3

Patent applications by applicant type and by year in Korea

Year All Korean Foreign Externally audited 
1985 10,587 2,703 7,884 978 
1986 12,759 3,641 9,118 1,610 
1987 17,062 4,871 12,191 2,312 
1988 20,051 5,696 14,355 3,288 
1989 23,315 7,021 16,294 4,880 
1990 25,820 9,082 16,738 5,955 
1991 28,132 13,253 14,879 9,210 
1992 31,073 15,952 15,121 11,426 
1993 36,491 21,459 15,032 15,259 
1994 45,712 28,564 17,148 20,757 
1995 78,499 59,236 19,263 49,488 
1996 90,326 68,413 21,913 57,227 
1997 92,734 67,346 25,388 54,105 
1998 75,188 50,596 24,592 34,558 
1999 80,642 55,970 24,672 33,528 
2000 102,010 72,831 29,179 36,096 
2001 104,612 73,714 30,898 39,688 
2002 106,136 76,570 29,566 41,598 
Year All Korean Foreign Externally audited 
1985 10,587 2,703 7,884 978 
1986 12,759 3,641 9,118 1,610 
1987 17,062 4,871 12,191 2,312 
1988 20,051 5,696 14,355 3,288 
1989 23,315 7,021 16,294 4,880 
1990 25,820 9,082 16,738 5,955 
1991 28,132 13,253 14,879 9,210 
1992 31,073 15,952 15,121 11,426 
1993 36,491 21,459 15,032 15,259 
1994 45,712 28,564 17,148 20,757 
1995 78,499 59,236 19,263 49,488 
1996 90,326 68,413 21,913 57,227 
1997 92,734 67,346 25,388 54,105 
1998 75,188 50,596 24,592 34,558 
1999 80,642 55,970 24,672 33,528 
2000 102,010 72,831 29,179 36,096 
2001 104,612 73,714 30,898 39,688 
2002 106,136 76,570 29,566 41,598 

Note: Source: The Korean Intellectual Property Office.

“Externally audited” (last column in the table) refers to the number of patents applied for by externally audited firms, which have generally more than KRW 7 billion in total assets as of 2002. The number of patent applications in this column is obtained from the authors’ own calculations after firm–applicant matching.

3.3 Key knowledge pool variables and regression models

The annual number of patent applications is susceptible to noise; thus, this study uses 3-year cumulative sums of patent applications. Specifically, the study sums up the number of patents applied for in the periods T − 2, T − 1, and T to obtain a proxy for the stock of knowledge of a firm for year T. The concerned explanatory variables, such as network and arm’s length industry patents, are then calculated based on these cumulative values. The firms and their patents are classified on the basis of the standard industry classification codes of the affiliated firms, thus indicating that the variations are at the sectoral level of the firms and not at the underlying technological level.

In the current study, we construct four variables representing the additional knowledge pool of a firm aside from its own patents: arm’s length industry pools divided into intra- and inter-sector spillover pools and network pools divided into intra- and inter-sector spillover pools based on patents by affiliated firms of a business group.

The inter-sector-network spillover pool of a firm is represented by the 3-year cumulative sum of the numbers of patents applied for by its sister firms in the same business group but from different sectors. That is, the inter-sector-network spillover pool of an affiliated firm (a focal subsidiary of a business group) k is defined as  

group_patent(inter)=jiPgi,
where g and i denote the group and the sector, respectively, with which firm k is affiliated, and Pgj denotes the 3-year cumulative sum of the numbers of patents applied for by all sister firms belonging to business group g and sector j.

The intra-sector-network spillover pool of firm k is  

group_patent(intra)=Pgipk,
where Pgi indicates the sum of the numbers of patent applications by firm k and other sister firms in sector i and business group g to which firm k belongs, and pk denotes the number of patents applied for by firm k. Therefore, we simply sum up the numbers of patent applications by all sister firms in the sector into which firm k is classified and exclude the patent applications by firm k.

By contrast, the inter-sector arm’s length industry spillover pool of a firm is proxied by the sum of the numbers of patents that independent firms in other sectors have applied for. The inter-sector industry spillover pool of firm k is calculated as follows:  

industry_patent(inter)=jiPjjiPgj, 
where Pj denotes the numbers of patents that all firms in sector j have applied for, and i is the industry to which firm k belongs. The intra-sector arm’s length industry spillover pool of firm k is calculated by  
industry_patent(intra)=PiPgi,
where Pi denotes the numbers of patents applied for by all firms in sector i.

Aside from the four knowledge pool variables, we also try one more variable to represent the knowledge pool combining intra- and inter-sector sources for affiliated firms. This variable is notated as “group patent (intra + inter),” which is the sum of patents applied for by all other sister affiliates, excluding a concerned firm, in a business group. In other words, this variable is a proxy for group-level knowledge stock that can spill over to an affiliate. Thus, in calculating the knowledge stock of a group that is available to a firm, we exclude the patents of the concerned firm (referred to as a focal subsidiary firm k ).

We do not discount patents in different sectors in terms of technological distances. Although several studies, such as that of Kafouros and Buckley (2008) , adopt this type of weighting method, we have decided not to do so because whether long- or short-distance knowledge is more significant for a firm in raising productivity remains uncertain, as discussed in the preceding section. Instead, we attempt to measure the relative size of the impact itself.

Subsequently, we adopt the following regression models: (1), (2), and (3). As for the dependent variables, we use labor productivity as a productivity measure, which is defined as sales divided by the number of employees (i.e., sales per employee).

Equation (1) , which is the first equation, serves as a benchmark that revisits the classical question of simple inter- versus intra-sector spillover without considering the network-based spillover pool. The hypothesis is about the relative size of the two coefficients, β2 and β3 , and we hypothesize that there would be no difference in the relative size.  

(1)
productivity=α+β1group_affiliates_patent+β2industry_patent(intra)+β3*industry_patent(inter)+δZ+ηi+uit
where Z is the vector of other control variables, and group affiliate’s patent is the number of patents by a concerned firm itself. Equation (2) is the first step to include the network (business group)-based knowledge pool, combining both the intra- and inter-sector pools through the variable of group_patent (intra + inter). We are interested in discovering whether coefficient β4 is greater than either β2 or β3 .  
(2)
productivity=α+β1group_affiliates_patent+β2industry_patent(intra)+β3industry_patent(inter)+β4group_patent(intra+inter)+δZ+ηi+uit

Equation (3) , which is the final equation, is the focus of the analysis because it introduces both the intra- and inter-sector knowledge pools facing a firm affiliated to a network (business group). Our hypothesis shows that β5 and β6 are larger than β2 and β3 , respectively. Additionally, β6 should not be different in size from β5 , which means that no difference in the size of the intra- and inter-sector spillover exists at the network level.  

(3)
productivity=α+β1group_affiliates_patent+β2industry_patent(intra)+β3industry_patent(inter)+β5group_patent(intra)+β6group_patent(inter)+δZ+ηi+uit

One may wonder why we still need Equation (2) when the hypothesis concerning Equation (2) is somewhat trivial if we turn out to be correct in terms of this hypothesis concerning Equation (3) . This reasoning would be right in the ex post sense, but at the research design stage, we were not sure whether the spillover impact from the separate (either intra- or inter-sector) or combined (sum of intra- and inter-sector) knowledge pool of a network would be larger than that of arm’s length industry.

The vector (Z) of other control variables includes the following variables:

Age is the logarithm of the age of firm k. The age of firm k is defined as (one plus) the number of years elapsed since the foundation of firm k.

Market share is the ratio of the gross output of firm k to the gross output of the sector. The gross output of the sector is defined at the three-digit sectoral level. 10

Export ratio11 is the export to sales ratio, that is, the share of exported output.

Table 4 presents the descriptive statistics for the variables used in knowledge spillover estimations. First, the mean of labor productivity (defined as sales per employee) in the sample is KRW 232 million. The 3-year cumulative sum of the numbers of patents by independent firms in the same sector as the focal firm (i.e., intra-sector arm’s length industry spillover pool) is 5,593 on average. In comparison, the 3-year cumulative sum of the numbers of patents by independent firms in different sectors from a focal firm (i.e., inter-sector arm’s length spillover pools) is as large as 67,790 on average. These numbers imply that the inter-sector-industry spillover pool is approximately 12 times larger than the intra-sector-industry spillover pool. The average number of intra-sector-group patents of a representative firm k , which is affiliated with a business group, is only 486. The average number of inter-sector-group patents is approximately four times as large as that of intra-sector group patents. The maximum value of the export ratio is 99.6%, which means that several firms are fully oriented toward foreign markets. As shown in Table 5 , the correlations among variables are not high. Therefore, multicollinearity is not an issue in our data. 12

Table 4

Descriptive statistics of the variables used in analysis

Variables Mean Standard deviation Minimum Maximum Observed 
Sales per employee (million won) 232 172 37 1682 2,242 
Group affiliate’s patent 156 1,503 39,326 2,242 
Industry patent (intra) 5,593 13,310 88,690 2,242 
Industry patent (inter) 67,790 40,885 6,954 153,000 2,242 
Group patent (intra) 486 2,764 42,358 2,242 
Group patent (inter) 1,974 6,156 45,783 2,242 
Group patent (intra + inter) 2,460 6,873 45,783 2,242 
Export ratio (% export/sales) 18.4 27.5 0.0 99.6 2,242 
Market share (% firm sales/industry sales) 5.8 12.0 0.0 95.0 2,242 
Firm age (logged) 18.3 13.3 0.9 74.0 2,242 
Variables Mean Standard deviation Minimum Maximum Observed 
Sales per employee (million won) 232 172 37 1682 2,242 
Group affiliate’s patent 156 1,503 39,326 2,242 
Industry patent (intra) 5,593 13,310 88,690 2,242 
Industry patent (inter) 67,790 40,885 6,954 153,000 2,242 
Group patent (intra) 486 2,764 42,358 2,242 
Group patent (inter) 1,974 6,156 45,783 2,242 
Group patent (intra + inter) 2,460 6,873 45,783 2,242 
Export ratio (% export/sales) 18.4 27.5 0.0 99.6 2,242 
Market share (% firm sales/industry sales) 5.8 12.0 0.0 95.0 2,242 
Firm age (logged) 18.3 13.3 0.9 74.0 2,242 

Note: Group affiliate’s patents refer to the number of patents applied for by each firm. Industry patent (intra) for each firm in this table and succeeding regression tables refers to the patents applied for by all firms in the same sector, except those by itself (each firm observation in the sample) and other affiliates in the same business group. Industry patent (inter) for each firm refers to the patents applied for by all firms in other sectors than the sector to which a firm belongs, excluding those by other affiliates in the same business group. Group patent (intra) for each firm is the number of patents applied for by all affiliates in the same business group and the same sector, excluding those by itself (each firm observation in the sample). Group patent (inter) for each firm is the number of patents applied for by all of its sister firms in other sectors. All variables using the number of patents are based on the cumulative sum of patents applied during the current year and the preceding 2 years, namely, years T, T − 1, and T − 2.

Table 5

Correlations among the variables used in analysis

Variables Sales per employee 
1. Group affiliate’s patent 0.03         
2. Industry patent (intra) −0.06 0.18        
3. Industry patent (inter) 0.16 −0.04 −0.07       
4. Group patent (intra) −0.01 0.07 0.29 −0.03      
5. Group patent (inter) 0.17 0.05 0.02 0.00 0.05     
6. Group patent (intra + inter) 0.15 0.07 0.13 −0.02 0.45 0.92    
7. Export ratio 0.10 0.08 0.19 −0.10 0.12 0.09 0.13   
8. Market share 0.09 0.31 −0.06 −0.08 0.03 0.23 0.22 0.17  
Firm age (logged) 0.00 0.06 −0.08 −0.07 −0.01 −0.10 −0.10 0.14 0.24 
Variables Sales per employee 
1. Group affiliate’s patent 0.03         
2. Industry patent (intra) −0.06 0.18        
3. Industry patent (inter) 0.16 −0.04 −0.07       
4. Group patent (intra) −0.01 0.07 0.29 −0.03      
5. Group patent (inter) 0.17 0.05 0.02 0.00 0.05     
6. Group patent (intra + inter) 0.15 0.07 0.13 −0.02 0.45 0.92    
7. Export ratio 0.10 0.08 0.19 −0.10 0.12 0.09 0.13   
8. Market share 0.09 0.31 −0.06 −0.08 0.03 0.23 0.22 0.17  
Firm age (logged) 0.00 0.06 −0.08 −0.07 −0.01 −0.10 −0.10 0.14 0.24 

4. Regression results: from spillovers to productivity

Using the previously described models, we first conduct the Hausman test to check the validity of the random effect models. Then, we first report the result supported by the Hausman test for each specification in Tables 6–8 , and, in hypothesis testing, we report the results both by fixed and random effect models to show robustness ( Table 9 and Table A1). In the results, group patent is a proxy for the spillover pool from the network (business group), and industry patent is a proxy for the arm’s length spillover pool from independent firms in the same sector or different sectors. Thus, the significant and positive coefficients of these variables confirm the spillover effects on an affiliated firm from the business groups or arm’s length firms in an industry.

Table 6 presents the baseline results using the group affiliate’s patents (patents by a concerned firm), intra-sector, and inter-sector industry pools of a firm. On the basis of the baseline regressions, we reexamine the ongoing debate on the relative size of intra- versus inter-sector spillovers in arm’s length industries. Models 2 and 3 demonstrate that intra-sector and inter-sector spillovers, respectively, exist and are both significant. Among the control variables, export ratio and market share exhibit significant positive effects, whereas age shows a negative and insignificant effect throughout the models. When we place both variables in Model 4, the sizes of the two coefficients do not significantly differ. A test on the significance of the gap (reported in Table 9 along with other test results) reveals that the two coefficients are not significantly different. That is, the results provide no evidence on the dominance of either intra- or inter-sector spillovers. Such results are different from those of existing literature that find dominance of either intra-sector ( Bernstein, 1988 ; Rouvinen, 2002 ) or inter-sector spillovers ( Harris and Robinson, 2004 ; Javorcik, 2004 ; van Stel and Nieuwenhuijsen, 2004 ; Bwalya, 2006 ; Kugler, 2006 ; Badinger and Egger, 2010; Autant-Bernard and LeSage, 2011 ) but are in agreement with the perspective of Hubert and Pain (2001) and Medda and Piga (2007) .

Table 6

Comparing the impacts of intra- and inter-sector spillovers

Dependent variable = sales per employee Model 1 Model 2 Model 3 Model 4 
RE RE FE FE 
Constant 240.27*** (7.67) 230.73*** (7.35) 134.3*** (17.62) 133.12*** (17.48) 
Group affiliate’s patent 52.92*** (3.03) 28.99* (1.65) 39.19** (2.41) 29.08* (1.76) 
Industry patent (intra)  15.68*** (7.17)  6.33*** (3.02) 
Industry patent (inter)   7.90*** (20.54) 7.62*** (19.25) 
Export ratio 0.39*** (3.07) 0.39*** (3.16) 0.28** (2.25) 0.28** (2.30) 
Market share 2.4*** (3.70) 2.56*** (3.96) 6.60*** (5.83) 6.53*** (5.79) 
Firm age −10.02 (−1.41) −7.52 (−1.05)   
Industry dummies Yes Yes No No 
R -squared  Within 0.015 0.042 0.201 0.205 
Between 0.287 0.286 0.016 0.013 
Overall 0.231 0.233 0.022 0.020 
Hausman test 3.84 (0.28) 4.17(0.38) 36.02 (0.00) 32.81 (0.00) 
Dependent variable = sales per employee Model 1 Model 2 Model 3 Model 4 
RE RE FE FE 
Constant 240.27*** (7.67) 230.73*** (7.35) 134.3*** (17.62) 133.12*** (17.48) 
Group affiliate’s patent 52.92*** (3.03) 28.99* (1.65) 39.19** (2.41) 29.08* (1.76) 
Industry patent (intra)  15.68*** (7.17)  6.33*** (3.02) 
Industry patent (inter)   7.90*** (20.54) 7.62*** (19.25) 
Export ratio 0.39*** (3.07) 0.39*** (3.16) 0.28** (2.25) 0.28** (2.30) 
Market share 2.4*** (3.70) 2.56*** (3.96) 6.60*** (5.83) 6.53*** (5.79) 
Firm age −10.02 (−1.41) −7.52 (−1.05)   
Industry dummies Yes Yes No No 
R -squared  Within 0.015 0.042 0.201 0.205 
Between 0.287 0.286 0.016 0.013 
Overall 0.231 0.233 0.022 0.020 
Hausman test 3.84 (0.28) 4.17(0.38) 36.02 (0.00) 32.81 (0.00) 

Note: t(Z)-values reported in parentheses. ***Significant at the 1% level; **significant at the 5% level; *significant at the 10% level.RE: random effect, FE: fixed effect.

Table 7 reports the results that compare the impact of spillover between networks (business group) and arm’s length industries. In the regression shown in Table 7 , we use the variable “ group patent (intra + inter) ” to measure the size of group-level knowledge spillovers, which is compared with those of intra- and inter-sector knowledge spillovers from an arm’s length industry (Models 1, 2, and 4, respectively) and with their combined pool represented by industry (intra + inter) in Model 3. The results of the four models indicate that the coefficients of the network spillovers are always greater than the spillovers from the industry (either from the same or different sector) and the sum of the intra- and inter-sector spillovers from the industry. The F -test results in Table 9 confirm that the differences are significant, and the relative sizes are consistent with the hypotheses. Therefore, we can conclude that the impact of network-based spillovers is more powerful than that of arm’s length relationship-based spillovers. This finding implies that group-affiliated firms obtain additional benefits from their sister firms in terms of spillovers. 13

Table 7

Impacts of spillovers from the network and the industry

Dependent variable: sales per employee Model 1 Model 2 Model 3 Model 4 
RE FE FE FE 
Constant 222.20*** (7.18) 133.53*** (17.61) 132.88*** (17.53) 132.70*** (17.50) 
Group affiliate’s patent 23.28 (1.35) 33.56** (2.07) 23.05 (1.42) 26.31 (1.59) 
Group patent (intra  +  inter) 35.11*** (8.27) 19.53*** (4.51) 17.21*** (3.95) 17.80*** (4.05) 
Industry patent (intra) 12.16*** (5.54)   4.86** (2.29) 
Industry patent (inter)  7.45*** (18.81)  7.27*** (18.02) 
Industry patent (intra  +  inter)   7.11*** (18.94)  
Export ratio 0.36*** (2.93) 0.26** (2.16) 0.27** (2.23) 0.27** (2.20) 
Market share 2.27*** (3.56) 6.49*** (5.77) 6.41*** (5.71) 6.45*** (5.74) 
Firm age −4.84 (−0.69)    
industry dummies Yes No No No 
R -squared  Within 0.070 0.210 0.212 0.212 
Between 0.310 0.021 0.016 0.018 
Overall 0.258 0.027 0.024 0.025 
Hausman test 4.67 (0.46) 34.41 (0.00) 28.85 (0.00) 34.47 (0.00) 
Dependent variable: sales per employee Model 1 Model 2 Model 3 Model 4 
RE FE FE FE 
Constant 222.20*** (7.18) 133.53*** (17.61) 132.88*** (17.53) 132.70*** (17.50) 
Group affiliate’s patent 23.28 (1.35) 33.56** (2.07) 23.05 (1.42) 26.31 (1.59) 
Group patent (intra  +  inter) 35.11*** (8.27) 19.53*** (4.51) 17.21*** (3.95) 17.80*** (4.05) 
Industry patent (intra) 12.16*** (5.54)   4.86** (2.29) 
Industry patent (inter)  7.45*** (18.81)  7.27*** (18.02) 
Industry patent (intra  +  inter)   7.11*** (18.94)  
Export ratio 0.36*** (2.93) 0.26** (2.16) 0.27** (2.23) 0.27** (2.20) 
Market share 2.27*** (3.56) 6.49*** (5.77) 6.41*** (5.71) 6.45*** (5.74) 
Firm age −4.84 (−0.69)    
industry dummies Yes No No No 
R -squared  Within 0.070 0.210 0.212 0.212 
Between 0.310 0.021 0.016 0.018 
Overall 0.258 0.027 0.024 0.025 
Hausman test 4.67 (0.46) 34.41 (0.00) 28.85 (0.00) 34.47 (0.00) 

Note: t(Z)-values reported in parentheses. ***Significant at the 1% level; **significant at the 5% level; *significant at the 10% level.

Now, Table 8 presents the results by dividing network-based knowledge pools into intra-sector and inter-sector pools, namely, spillovers from affiliates in the same sector and those from affiliates in different sectors. We attempt several combinations to check for robustness. In Model 1, we investigate spillovers in the same sector and compare the impact of sister firms and independent firms in the same industry. The spillovers from the network are greater than those from the market (arm’s length industries), consistent with the results in Table 7 . In Model 2, we focus on spillovers from different sectors and compare the impact of sister firms and independent firms in different sectors. Again, the spillovers from the network are greater than those from the industry.

Table 8

Spillovers from affiliates in the same and different sectors

Dependent variable: sales per employee Model 1 Model 2 Model 3 Model 4 Model 5 
RE FE RE FE FE 
Constant 231.35*** (7.36) 134.48*** (17.69) 221.40*** (7.16) 132.91*** (17.51) 132.39*** (17.44) 
Group affiliate’s patent 30.31* (1.73) 35.46** (2.18) 22.44 (1.30) 33.85** (2.09) 27.30* (1.65) 
Industry patent (intra) 13.54*** (5.94)  12.67*** (5.62)  4.31** (1.96) 
Industry patent (inter)  7.61*** (19.26)  7.46*** (18.85) 7.30*** (18.04) 
Group patent (intra) 30.37*** (3.22)  26.94*** (2.89) 31.01*** (3.54) 25.36*** (2.75) 
Group patent (inter)  16.20*** (3.16) 37.50*** (7.67) 15.43*** (3.02) 15.36*** (3.01) 
Export ratio 0.40*** (3.18) 0.26** (2.15) 0.36*** (2.91) 0.27** (2.20) 0.27** (2.22) 
Market share 2.60*** (4.02) 6.41*** (5.68) 2.23*** (3.50) 6.61*** (5.86) 6.53*** (5.80) 
Firm age −7.76 (−1.09)  −4.57 (−0.65)   
Industry dummies Yes No Yes No No 
R -squared  Within 0.048 0.206 0.070 0.211 0.213 
Between 0.286 0.021 0.312 0.018 0.017 
Overall 0.235 0.027 0.259 0.025 0.024 
Hausman test 5.01 (0.42) 38.88 (0.00) 5.80 (0.45) 43.81 (0.00) 38.99 (0.00) 
Dependent variable: sales per employee Model 1 Model 2 Model 3 Model 4 Model 5 
RE FE RE FE FE 
Constant 231.35*** (7.36) 134.48*** (17.69) 221.40*** (7.16) 132.91*** (17.51) 132.39*** (17.44) 
Group affiliate’s patent 30.31* (1.73) 35.46** (2.18) 22.44 (1.30) 33.85** (2.09) 27.30* (1.65) 
Industry patent (intra) 13.54*** (5.94)  12.67*** (5.62)  4.31** (1.96) 
Industry patent (inter)  7.61*** (19.26)  7.46*** (18.85) 7.30*** (18.04) 
Group patent (intra) 30.37*** (3.22)  26.94*** (2.89) 31.01*** (3.54) 25.36*** (2.75) 
Group patent (inter)  16.20*** (3.16) 37.50*** (7.67) 15.43*** (3.02) 15.36*** (3.01) 
Export ratio 0.40*** (3.18) 0.26** (2.15) 0.36*** (2.91) 0.27** (2.20) 0.27** (2.22) 
Market share 2.60*** (4.02) 6.41*** (5.68) 2.23*** (3.50) 6.61*** (5.86) 6.53*** (5.80) 
Firm age −7.76 (−1.09)  −4.57 (−0.65)   
Industry dummies Yes No Yes No No 
R -squared  Within 0.048 0.206 0.070 0.211 0.213 
Between 0.286 0.021 0.312 0.018 0.017 
Overall 0.235 0.027 0.259 0.025 0.024 
Hausman test 5.01 (0.42) 38.88 (0.00) 5.80 (0.45) 43.81 (0.00) 38.99 (0.00) 

Note: t(Z)-values reported in parentheses. ***Significant at the 1% level; **significant at the 5% level; *significant at the 10% level.

Table 9

Test of the difference in size of the estimated coefficients

Hypotheses tested  Model 4 in Table 6
 
Model 4 in Table 7
 
Model 5 in Table 8
 
Hypotheses supported? 
F value (Prob. > F) F value (Prob. > F) F value (Prob. > F) 
H1-A: group patent (intra) > industry patent (intra)   4.34** (0.038) Yes (>) 
H1-B: group patent (inter) > industry patent (inter)   2.39+ (0.122) Yes (>) (marginally) 
H2-A: group patent (intra) = group patent (inter)   0.87 (0.350) Yes 
H2-B: industry patent (intra) = industry patent (inter) 0.33 (0.563) 1.17 (0.280) 1.66 (0.197) Yes 
H3-A: group patent (intra) > industry patent (inter)   3.83** (0.050) Yes (>) 
H3-B: group patent (inter) > industry patent (intra)   3.93** (0.048) Yes (>) 
H1: group patent (intra + inter) > industry patent (intra + inter) (from model 3 in Table 6 )   5.09** (0.024)  Yes (>) 
(H1’) group patent (intra + inter) > industry patent (intra)  6.21** (0.013)  Yes (>) 
(H1’) group patent (intra + inter) > industry patent (inter)  5.49** (0.019)  Yes (>) 
Hypotheses tested  Model 4 in Table 6
 
Model 4 in Table 7
 
Model 5 in Table 8
 
Hypotheses supported? 
F value (Prob. > F) F value (Prob. > F) F value (Prob. > F) 
H1-A: group patent (intra) > industry patent (intra)   4.34** (0.038) Yes (>) 
H1-B: group patent (inter) > industry patent (inter)   2.39+ (0.122) Yes (>) (marginally) 
H2-A: group patent (intra) = group patent (inter)   0.87 (0.350) Yes 
H2-B: industry patent (intra) = industry patent (inter) 0.33 (0.563) 1.17 (0.280) 1.66 (0.197) Yes 
H3-A: group patent (intra) > industry patent (inter)   3.83** (0.050) Yes (>) 
H3-B: group patent (inter) > industry patent (intra)   3.93** (0.048) Yes (>) 
H1: group patent (intra + inter) > industry patent (intra + inter) (from model 3 in Table 6 )   5.09** (0.024)  Yes (>) 
(H1’) group patent (intra + inter) > industry patent (intra)  6.21** (0.013)  Yes (>) 
(H1’) group patent (intra + inter) > industry patent (inter)  5.49** (0.019)  Yes (>) 

Note: ***, **, *, +: significant at the 1%, 5%, 10%, 15%, respectively.

In the first column, the hypotheses are about comparing the relative size of the coefficients of the specified variables. For hypotheses (H3), we do not have the a priori expected signs. H1’ is a variation of the first main hypothesis (H1) noted in the text. In testing, typical null hypothesis is that the coefficients of the regressors, X i and X j , are of the same size (H 0 : βi=βj , test statistic (Rbr)[R(XX)1R]1(Rbr)/qee/(nk)F(q,nk) ). See Johnston and Dinardo (1997).

Model 3 is a variation of Model 1, and Model 4 is a variation of Model 2. The results consistently show that a stronger impact is observed from the network than from the arm’s length industry, regardless of the dimension of the sector (intra- or inter-). The final test is conducted on all intra- and inter-sector spillover variables, as reported in Model 5. All variables remain positive and significant, and the results are also consistent with regard to the relative impact of the network versus the industry. That is, spillovers from networks are greater than those from the industry in both intra- and inter-sector dimensions. The size of the spillover from the knowledge pool of a firm itself (own patents accumulated over time) tends to be similar to that of the spillover from the knowledge pools of affiliated firms in the same sector.

Using the four coefficients of Model 5 in Table 8 , we conduct significance tests on various pair-wise combinations of these coefficients. The test results in Table 9 suggest that network spillovers tend to be larger than those from industries in all four possible combinations (intra- versus intra-, inter- versus inter-, intra- versus inter-, and inter- versus intra-). Given that the tests in Table 9 are based on the fixed effect results, we also tried the same testing using the random effect results, and they are reported in the Table A1. All comparisons are significant at the 5% level, except inter- versus inter-, which is marginally significant at 12% in Table 9 . But, even this pair’s gap turns out to be significant at 1% when we tried the random effects estimation, reported in Table A1. Thus, by considering all the results from Tables 7 and 8 , we can conclude that spillovers from the network (business) are greater than those from the arm’s length industries (market).

Another consistent result is the relative size of intra- versus inter-sector spillovers. Significance tests in Table 9 on the relative size of the inter- versus intra-sector spillovers from the industry (row H2-B in columns 2, 3, and 4 of Table 9 using the coefficients from Tables 6–8 ) consistently indicate no difference in size. Furthermore, the significance test on the relative size of inter- versus intra-spillover from the network (from affiliated firms) indicates no significant difference (row H2-A in Table 9 ). Thus, no difference exists between inter- and intra-sector spillovers in both network and arm’s length industry dimensions. Given that the literature tends to compare spillovers only in the single dimension of the arm’s length industry, this finding provides stronger support to the argument that intra- and inter-industry spillovers are both significant ( Hubert and Pain, 2001 ; Medda and Piga, 2007 ) and may be indifferent in their sizes.

The following discussion focuses on the magnitude of the spillover estimated from the regressions. According to the coefficients from Model 5 in Table 8 , an additional patent applied for by sister firms in the same sector to which a firm belongs results in an increase of approximately USD 6.58 in sales per employee for the firm affiliated with the same business group. By contrast, a firm experiences an increase of only USD 1.12 in labor productivity from one patent of an unrelated firm in the same sector. 14 One more patent of affiliated firms in different sectors tends to increase sales per employee by USD 3.99, whereas an additional patent in any independent firm in different sectors produces an increase of only USD 1.89 in the labor productivity of the concerned firm. Given that a firm in the sample hires an average of 1,753 employees, one more patent from sister firms in the same group pools results in a corresponding increase of USD 11,534 and USD 6,986 in sales for an average firm affiliated with a business group from intra- and inter-sector spillovers. From the industry spillover pools, an increase of USD 3,320 in sales is accrued to a firm from the inter-sector spillover, and an increase of USD 1960 is accrued from the intra-sector spillover. By contrast, adding another patent in its own knowledge pool of firm results in an increase of USD 12,417 in the sales of the firm and in an increase of USD 7.08 in labor productivity of the firm.

However, the absolute sizes of industry spillover pools significantly outweigh those of network spillover pools, as shown in Table 4 . For example, the knowledge pool of sister firms in different sectors has only 1,974 patents, whereas the knowledge pool of arm’s length industry in different sectors has 67,790 patents. Thus, the absolute size of the total effects of spillover pools at a given time is larger in the arm’s length industry than in the network (business group). However, industry spillover pools are not under the control of a firm. Thus, firms should pay more attention to the level and growth of spillover pools from the network. Moreover, most economic decisions are made based on the marginal effect rather than on the total (or average) effect comparison.

The following discussion presents the result of several robustness tests. The test is conducted using the TFP estimate data available from Choo (2007) as a dependent variable rather than labor productivity and by conducting cross-section estimations rather than panel estimations. The results are reported in Tables 10 (estimations) and 11 (tests of the differences in the coefficients). The results are consistent between Tables 9 and 11 . An additional insight from the cross-section results in Table 10 is that the coefficient of the intra-sectoral knowledge pool of other firms in the same sector is negative. This result is logical because it clearly captures competition effects among firms in the same industry, which cannot be demonstrated by panel regressions.

Table 10

Robustness test using TFP in both panel and cross-section estimation

Dependent variable: TFP  Panel regression
 
Cross-sectional regression
 
Specification 1 Specification 2 Specification 3 1995 + 1996 1996 + 1997 
Constant −0.38*** (−13.96) −0.38*** (−14.09) −0.38*** (−14.10) −0.07 (−0.44) −0.74*** (−4.18) 
Group affiliate’s patent 0.03 (0.52) 0.02 (0.34) 0.02 (0.37) 0.32*** (2.70) 0.19** (2.03) 
Industry patent (intra) 0.04*** (5.01) 0.03*** (4.20) 0.03*** (3.90) −0.11*** (−7.04) −0.03** (−2.49) 
Industry patent (inter) 0.03*** (22.42) 0.03*** (21.07) 0.03*** (21.05) 0.01+ (1.56) 0.04*** (4.74) 
Group patent (intra)   0.08*** (2.59) 0.13*** (2.77) 0.07* (1.93) 
Group patent (inter)   0.06*** (3.48) 0.12** (2.44) 0.10*** (3.39) 
Group patent (intra  +  inter)  0.07*** (4.39)    
Export ratio 0.001* (1.87) 0.001* (1.76) 0.001* (1.78) 0.003 0.003*** (2.74) 
Market share 0.02*** (6.01) 0.02*** (5.96) 0.02*** (5.99) 0.01 0.01** (2.19) 
R -squared  Within 0.262 0.027 0.270 0.201 0.203 
Between 0.066 0.071 0.070 
Overall 0.100 0.107 0.107 
Dependent variable: TFP  Panel regression
 
Cross-sectional regression
 
Specification 1 Specification 2 Specification 3 1995 + 1996 1996 + 1997 
Constant −0.38*** (−13.96) −0.38*** (−14.09) −0.38*** (−14.10) −0.07 (−0.44) −0.74*** (−4.18) 
Group affiliate’s patent 0.03 (0.52) 0.02 (0.34) 0.02 (0.37) 0.32*** (2.70) 0.19** (2.03) 
Industry patent (intra) 0.04*** (5.01) 0.03*** (4.20) 0.03*** (3.90) −0.11*** (−7.04) −0.03** (−2.49) 
Industry patent (inter) 0.03*** (22.42) 0.03*** (21.07) 0.03*** (21.05) 0.01+ (1.56) 0.04*** (4.74) 
Group patent (intra)   0.08*** (2.59) 0.13*** (2.77) 0.07* (1.93) 
Group patent (inter)   0.06*** (3.48) 0.12** (2.44) 0.10*** (3.39) 
Group patent (intra  +  inter)  0.07*** (4.39)    
Export ratio 0.001* (1.87) 0.001* (1.76) 0.001* (1.78) 0.003 0.003*** (2.74) 
Market share 0.02*** (6.01) 0.02*** (5.96) 0.02*** (5.99) 0.01 0.01** (2.19) 
R -squared  Within 0.262 0.027 0.270 0.201 0.203 
Between 0.066 0.071 0.070 
Overall 0.100 0.107 0.107 

Note: The TFP estimates used here are from Choo (2007) and the residuals from the fixed effect models. They are expected to be better than either those from Levinsohn and Petrin's Methodology, which suffers from non-monotonicity and perfect collinearity, or the generalized method of moments (GMM) estimation, which does not satisfy the specification criterion. We have checked all three methods. See Choo (2007) for details. Cross-section estimations using 1 year is subject to the multicollinearity problem because of several observations with zero values for the number of patents. Thus, we merge the two most recent year observations into one cross-section observation and take the average of the 2-year values to obtain the results in the last two columns.***, **, *, +: significant at the 1%, 5%, 10%, 15%, respectively.

Table 11

Hypotheses test using the results in Table 10

Hypotheses/models  Panel regression
 
Cross-sectional regression
 
Hypotheses supported? (from panel regression) 
Model 1 Model 2 Model 3 1995 + 1996 1996 + 1997 
F value F value F value F value F value 
(Prob. > F) (Prob. > F) (Prob. > F) (Prob. > F) (Prob. > F) 
H1-A: group patent (intra) > industry patent (intra)   2.28 27.04 2.55 Yes (>) 
  (0.132)+ (0.000)*** (0.111)+ (marginally) 
H1-B: group patent (inter) > industry patent (inter)   3.17 8.60 5.83 Yes (>) 
  (0.075)* (0.004)*** (0.016)** (marginally) 
H2-A: group patent (intra) = group patent (inter)   0.32 0.02 0.77 Yes 
  (0.574) (0.889) (0.382) 
H2-B: industry patent (intra) = industry patent (inter) 0.53 0.03 0.00 76.49 53.98 Yes 
(0.466) (0.856) (0.980) (0.000)*** (0.000)*** 
H3-A: group patent (intra) > industry patent (inter)   2.75 9.36 0.06 Yes (>) 
  (0.097)* (0.002)*** (0.812) (marginally) 
H3-B: group patent (inter) > industry patent (intra)   2.74 32.91 25.08 Yes (>) 
  (0.098)* (0.000)*** (0.000)*** (marginally) 
(H1’) group patent (intra + inter) > industry patent (intra)  3.99    Yes (>) 
 (0.046)**    
(H1’) group patent (intra + inter) > industry patent (inter)  5.75    Yes (>) 
 (0.017)**    
Hypotheses/models  Panel regression
 
Cross-sectional regression
 
Hypotheses supported? (from panel regression) 
Model 1 Model 2 Model 3 1995 + 1996 1996 + 1997 
F value F value F value F value F value 
(Prob. > F) (Prob. > F) (Prob. > F) (Prob. > F) (Prob. > F) 
H1-A: group patent (intra) > industry patent (intra)   2.28 27.04 2.55 Yes (>) 
  (0.132)+ (0.000)*** (0.111)+ (marginally) 
H1-B: group patent (inter) > industry patent (inter)   3.17 8.60 5.83 Yes (>) 
  (0.075)* (0.004)*** (0.016)** (marginally) 
H2-A: group patent (intra) = group patent (inter)   0.32 0.02 0.77 Yes 
  (0.574) (0.889) (0.382) 
H2-B: industry patent (intra) = industry patent (inter) 0.53 0.03 0.00 76.49 53.98 Yes 
(0.466) (0.856) (0.980) (0.000)*** (0.000)*** 
H3-A: group patent (intra) > industry patent (inter)   2.75 9.36 0.06 Yes (>) 
  (0.097)* (0.002)*** (0.812) (marginally) 
H3-B: group patent (inter) > industry patent (intra)   2.74 32.91 25.08 Yes (>) 
  (0.098)* (0.000)*** (0.000)*** (marginally) 
(H1’) group patent (intra + inter) > industry patent (intra)  3.99    Yes (>) 
 (0.046)**    
(H1’) group patent (intra + inter) > industry patent (inter)  5.75    Yes (>) 
 (0.017)**    

Note: See the notes in Table 9 .

5. Concluding remarks

Given the increasing significance of knowledge spillovers and technological fusion in innovations, this study investigates the productivity impacts of knowledge spillovers from firms in an arm’s length industry and from a network consisting of affiliated firms in the same business group. To shed light on the debate on the relative sizes of intra- and inter-sector spillovers, we also divide knowledge pools into intra- and inter-sector pools. This structure enables us to compare the sizes of spillovers over two-by-two combinations (between the industry and the network, and between the inter-sector and intra-sector).

First, this study finds that both intra- and inter-spillovers are significant channels of knowledge spillovers, but no evidence is found on the dominance of either intra- or inter-sector spillovers, regardless of whether the spillover is from an arm’s length industry or from a network. This finding should be considered as a new contribution because the literature tends to focus on spillovers from the arm’s length industry only. Second, and more significantly, we find that spillovers from networks are greater than those from arm’s length industries, regardless of the comparison among intra- versus intra-, inter- versus inter-, intra- versus inter-, or inter- versus intra-sector spillovers.

These results imply that knowledge spillover is not automatic, and knowledge can be transferred better through direct interaction and experience, which are more prevalent within a network organization, such as a business group. The knowledge of a firm is either available to, or more effectively exploited by, the affiliates of a concerned business group. In this sense, a business group is an effective organization for internalizing knowledge spillovers or for promoting more widespread knowledge diffusion among affiliates wherein the R&D activities undertaken by one firm benefits other firms in the same business group. The results are consistent with the finding on the resource-sharing advantage of business groups, as verified by Cheong et al. (2010) and Chang and Hong (2000) .

Given such benefit and all other things being equal, a strategic implication is that a firm can try to promote more network-type organizations, such as business groups, so that it can expand its technological capabilities and achieve a higher level of productivity associated with knowledge spillovers. Also, firms are advised to pay attention to both intra- and inter-sector spillover, because they are equally important, as shown by this study. However, any strategy implication should be taken with caution as the results are based on the one-country data. Moreover, the size of spillover may be subject to other factors, such as internationalization of firms or cultural factors, which are not accounted for in this study. For example, one may reason that higher spillover effects may be more specific to an environment with less inter-firm mobility of staff members or less internationalization (high homogeneity of staffs), as experienced by Korea in the past. This qualification may also be a direction for future or new research. One can investigate the hypothesis that higher inter-firm mobility of employees can be a substitute for intra-group knowledge spillover. Future research topics may include the impact on intra-group knowledge spillover of alternative spillover channels, such as an increase in inter-firm employee mobility, international ties of group-affiliated firms, and possibly M&A. 15 Also, one can compare the relative size of the spillover among affiliates in the same business groups and among subsidiaries in the same MNEs.

Funding

The authors gratefully acknowledge the support provided by the National Research Foundation of Korea (Grant no. NRF-2013-S1A3A2053312). The revised version of this paper benefitted from the three rounds of valuable comments by two anonymous referees.

1 For instance, Suzumura (1992) extends the case to an oligopoly.
2 Teece (1977) points out the so-called “resource cost” as a factor that limits transfer of knowledge or technologies across boundaries where resource costs means the costs involved in making sure the successful transfer of knowledge.
3 For example, there are four pillar affiliates in the area of electronics business of Samsung, such as Samsung Electronics, Samsung SDI, SEM, and Samsung Corning. Each has its own subsidiaries in various parts of the world. For more details on how these affiliates and their foreign subsidiaries cooperate among themselves in forming vertical clusters in and out of Korea, refer to Lee and He (2009) .
4 Although R&D spillovers spur the diffusion of new knowledge, they may also create disincentives for firms to undertake their own R&D investments ( Bernstein, 2000 ).
5 Technology transfer is different from knowledge spillover, which R&D practitioners cannot appropriate. However, the benefits from a business group do not result only from technology spillovers per se but also from the speed and ease of technology transfer within the group. In fact, making a clear-cut distinction between the effects of knowledge spillovers and those of transfers is difficult.
6 Los (2000) refers to this type of knowledge spillover as an “idea-creating” spillover.
7 Kodama (1992) , who coined the term “technology fusion,” observed that between 1980 and 1986, the Japanese textile industry spent 70% of total R&D outside principal products. In turn, the technologies developed by the sector gained potential for other sectors. For example, new fibers have been used in making building materials and filtration systems for kidney dialysis machines ( Kodama, 1992 ). In Korea, Cheil Industries, the former leading textile company in the country, has evolved into a major chemical company by applying its fiber technologies to electronic materials and chemicals.
8 While this third view corresponds to the inverted U curve between the value of spillover and the technical distance, the view emphasizing intra-sector spillover corresponds to a downward-sloping line (more distance, less spillover), and the inter-sector spillover view, to a upward sloping line.
9 Based on an interview (conducted on February 2013) with KT Chung, Director General of the Samsung Economic Research Institute who was formerly responsible for the personnel management system of the Samsung Group, and Mr. C.H. Lee, a section chief of the LG Academy.
10 The current study adds the outputs of all externally audited firms within a sector in the KIS database to calculate the sector output.
11 KIS export data have missing values. This study fills up the missing values with data obtained from the TS2000 of the Korea Listed Companies Association.
12 The correlation between the undivided group patents [group patent (intra + inter)] and group patent (inter) is significantly high. However, these variables are not used together.
13 Although not reported in Table 7 , we have also checked the possibility of nonlinearity (or decreasing return to spillover) by adding the square term of the group patent pool. The square term was shown to be insignificant.
14 The coefficients of spillover variables represent the impacts per patent in million Korean won; thus, we convert Korean won to US dollar by using the average exchange rate for 2010 (KRW/USD = 1,156.26). The constructed patent variables are based on the 3-year cumulative sum of the numbers of patents. Therefore, the real impact of one patent is approximately three times as large as the value of each coefficient.
15 For instance, Arora et al. (2014) analyze the interface between the different forms of firms and M&A as a mean to access external knowledge.

References

Adams
J. D.
Jaffe
A. B.
(
1996
), ‘
Bounding the effects of R&D: an investigation using matched establishment-firm data
,’
The Rand Journal of Economics
  ,
27
,
700
721
.
Aitken
B. J.
Harrison
A. E.
(
1999
), ‘
Do domestic firms benefit from direct foreign investment? Evidence from Venezuela
,’
American Economic Review
  ,
89
,
605
618
.
Almeida
P.
Song
J.
Grant
R. M.
(
2002
), ‘
Are firms superior to alliances and markets? An empirical test of cross-border knowledge building
,’
Organization Science
  ,
13
(
2
),
147
161
.
Arora
A.
Belenzon
S.
Rios
L.
(
2014
), ‘
Make, buy, organize: the interplay between research, external knowledge, and firm structure
,’
Strategic Management Journal
  ,
35
,
317
337
.
Autant-Bernard
C.
LeSage
J. P.
(
2011
), ‘
Quantifying knowledge spillovers using spatial econometric models
,’
Journal of Regional Science
  ,
51
,
471
496
.
Badinger
H.
Egger
P.
(
2008
), ‘
Intra-and inter-industry productivity spillovers in OECD manufacturing: a spatial econometric perspective
,’
CESifo Working Paper Series
.
Bernstein
J. I.
(
1988
), ‘
Costs of production, intra-and interindustry R&D spillovers: Canadian evidence
,’
Canadian Journal of Economics
  ,
324
347
.
Bernstein
J. I.
(
2000
), ‘
Factor intensities, rates of return, and international R&D spillovers: the case of Canadian and US industries
,’ in
Encaoua
D. H.
Laisney
F.
Mairesse
J.
(eds),
The Economics and Econometrics of Innovation
  .
Kluwer Academic Publishers
:
Boston
, pp.
519
542
.
Birkinshaw
J.
Hood
N.
(
1998
), ‘
Multinational subsidiary evolution: capability and charter change in foreign-owned subsidiary companies
,’
Academy of Management Review
  ,
23
(
4
),
773
795
.
Branstetter
L.
(
2000
), ‘
Vertical keiretsu and knowledge spillovers in Japanese manufacturing: an empirical assessment
,’
Journal of the Japanese and International Economies
  ,
14
,
73
104
.
Bwalya
S. M.
(
2006
), ‘
Foreign direct investment and technology spillovers: evidence from panel data analysis of manufacturing firms in Zambia
,’
Journal of Development Economics
  ,
81
,
514
526
.
Cesaroni
F.
(
2004
), ‘
Technological outsourcing and product diversification: do markets for technology affect firms’ strategies?
,’ in
Cantwell
J. G. A.
Granstrand
O.
(eds),
The Economics and Management of Technological Diversification
  .
London, Routledge
.
Chang
S. J.
(
2003
),
Financial Crisis and Transformation of Korean Business Groups: The rise and Fall of Chaebols
  .
Cambridge University Press
:
Cambridge
.
Chang
S. J.
Hong
J.
(
2000
), ‘
Economic performance of group-affiliated companies in Korea: ontragroup resource sharing and internal business transactions
,’
Academy of Management Journal
  ,
43
,
429
448
.
Cheong
K.
Choo
K.
Lee
K.
(
2010
), ‘
Understanding the behavior of business groups: a dynamic model and empirical analysis
,’
Journal of Economic Behavior and Organization
  ,
76
,
141
152
.
Choo
K.
(
2007
),
Linking Technological Diversification to Corporate Performance: The Case of Korean Firms
  .
Department of Economics, Seoul National University
.
Choo
K.
Lee
K.
Ryu
K.
Yoon
J.
(
2009
), ‘
Changing performance of business groups over two decades: technological capabilities and investment inefficiency in Korean Chaebols
,’
Economic Development and Cultural Change
  ,
57
,
359
386
.
Coe
D. T.
Helpman
E.
(
1995
), ‘
International R&D spillovers
,’
European Economic Review
  ,
39
,
859
887
.
Criscuolo
P.
(
2005
), ‘
On the road again: researcher mobility inside the R&D network
,’
Research Policy
  ,
34
(
9
),
1350
1365
.
d'Aspremont
C.
Jacquemin
A.
(
1988
), ‘
Cooperative and noncooperative R & D in duopoly with spillovers
,’
The American Economic Review
  ,
78
,
1133
1137
.
Desrochers
P.
Leppälä
S.
(
2011
), ‘
Opening up the ‘Jacobs Spillovers’ black box: local diversity, creativity and the processes underlying new combinations
,’
Journal of Economic Geography
  ,
11
,
843
863
.
Dindaroglu
B.
(
2010
), ‘
Intra-Industry Knowledge Spillovers and Scientific Labor Mobility
,’
Discussion Paper. SUNY Albany
.
Forsgren
M.
(
1997
), ‘
The advantage paradox of the multinational corporation
,’ in
Bjorkman
I.
Forsgren
M.
(eds.)
The Nature of the International Firm: Nordic Contributions to International Business Research
  .
Copenhagen Business School Press
:
Copenhagen
. pp.
69
85
.
George
R.
Kabir
M.
Douma
S. W.
(
2004
),
Business Groups and Profit Redistribution: A Boon or Bane for Firms
  .
Center for Economic Research, Tilburg University
.
Geroski
P.
(
1995
),
Market for Technology: Knowledge, Innovation and Appropriability, Handbook of the Economics of Innovation and Technological Change
  .
Blackwell
:
Oxford
, pp.
90
131
.
Ghemawat
P.
Khanna
T.
(
1998
), ‘
The nature of diversified business groups: a research design and two case studies
,’
The Journal of Industrial Economics
  ,
46
,
35
61
.
Glaeser
E. L.
Kallal
H. D.
Scheinkman
J. A.
Shleifer
A.
(
1992
), ‘
Growth in cities
,’
Journal of Political Economy
  ,
100
,
1126
1152
.
Goto
A.
(
1982
), ‘
Business groups in a market economy
,’
European Economic Review
  ,
19
,
53
70
.
Granstrand
O.
(
1999
),
The Economics and Management of Intellectual Property: Towards Intellectual Capitalism
  .
Edward Elgar
:
Cheltenham
.
Griliches
Z.
(
1992
), ‘
The search for R&D spillovers
,’
The Scandinavian Journal of Economics
  ,
S29
S47
.
Hansen
M. T.
Mors amd
M. L.
Løvås
B.
(
2005
), ‘
Knowledge sharing in organizations: multiple networks, multiple phases
,’
Academy of Management Journal
  ,
48
(
5
),
776
793
.
Harris
R.
Robinson
C.
(
2004
), ‘
Productivity impacts and spillovers from foreign ownership in the United Kingdom
,’
National Institute Economic Review
  ,
187
,
58
75
.
Hubert
F.
Pain
N.
(
2001
), ‘
Inward investment and technical progress in the United Kingdom manufacturing sector
,’
Scottish Journal of Political Economy
  ,
48
,
134
147
.
Jacobs
J.
(
1969
),
The Economy of Cities
  .
Vintage
:
New York
.
Jaffe
A. B.
(
1986
), ‘
Technological opportunity and spillovers of R&D: evidence from firms' patents, profits, and market value
,’
The American Economic Review
  ,
76
,
984
1001
.
Javorcik
B. S.
(
2004
), ‘
Does foreign direct investment increase the productivity of domestic firms? In search of spillovers through backward linkages
,’
The American Economic Review
  ,
94
,
605
627
.
Johnston
J.
Dinardo
J.
(
1997
),
Econometric Methods
  ,
4th edn
.
McGraw-Hill/Irwin
:
New York
.
Jordaan
J. A.
(
2008
), ‘
Intra-and inter-industry externalities from foreign direct investment in the Mexican manufacturing sector: new evidence from Mexican regions
,’
World Development
  ,
36
,
2838
2854
.
Kafouros
M. I.
Buckley
P. J.
(
2008
), ‘
Under what conditions do firms benefit from the research efforts of other organizations?
,’
Research Policy
  ,
37
,
225
239
.
Khanna
T.
Palepu
K.
(
1997
), ‘
Why focused strategies may be wrong for emerging markets
,’
Harvard Business Review
  ,
75
,
41
48
.
Khanna
T.
Palepu
K.
(
2000
), ‘
Is group affiliation profitable in emerging markets? An analysis of diversified Indian business groups
,’
The Journal of Finance
  ,
55
,
867
891
.
Khanna
T.
Yafeh
Y.
(
2007
), ‘
Business groups in emerging markets: paragons or parasites?
,’
Journal of Economic Literature
  ,
45
,
331
372
.
Kim
J.
Marschke
G.
(
2005
), ‘
Labor mobility of scientists, technological diffusion, and the firm's patenting decision
,’
RAND Journal of Economics
  ,
36
,
298
317
.
Kock
C. J.
Guillén
M. F.
(
2001
), ‘
Strategy and structure in developing countries: business groups as an evolutionary response to opportunities for unrelated diversification
,’
Industrial and Corporate Change
  ,
10
,
77
113
.
Kodama
F.
(
1992
), ‘
Technology fusion and the new R&D
,’
Harvard Business Review
  ,
70
,
70
78
.
Kugler
M.
(
2006
), ‘
Spillovers from foreign direct investment: within or between industries?
,’
Journal of Development Economics
  ,
80
,
444
477
.
Laursen
K.
Meliciani
V.
(
2000
), ‘
The importance of technology-based intersectoral linkages for market share dynamics
,’
Weltwirtschaftliches Archiv
  ,
136
,
702
723
.
Lee
K.
He
X.
(
2009
), ‘
Project execution and vertical integration capability of business groups: Samsung created in Korea, replicated in China
,’
Asian Business and Management
  ,
8
(
3
),
277
299
.
Leff
N. H.
(
1978
), ‘
Industrial organization and entrepreneurship in the developing countries: the economic groups
,’
Economic Development and Cultural Change
  ,
26
,
661
675
.
Levitt
B.
March
J. G.
(
1988
), ‘
Organizational learning
,’
Annual Review of Sociology
  ,
14
,
319
340
.
Los
B.
(
2000
), ‘
The empirical performance of a new inter-industry technology spillover measure
,’ in
Saviotti
P. N. B.
(Ed.),
Technology and Knowledge: From the Firm to Innovation Systems
  .
Edward Elgar
:
Cheltenham, UK
, pp.
118
151
.
Medda
G.
Piga
C.
(
2007
), ‘
Technological spillovers and productivity in Italian manufacturing firms
,’
Working Paper, Loughborough University
.
Nooteboom
B.
Haverbeke
W. Van
Duysters
G.
Gilsing
V.
den Oord
A. Van
(
2007
), ‘
Optimal cognitive distance and absorptive capacity
,’
Research Policy
  ,
36
,
1016
1034
.
Plunket
A.
(
2009
), ‘
Firms' inventiveness and localized vertical R&D spillovers
,’
Journal of Innovation Economics
  ,
4
,
147
170
.
Rouvinen
P.
(
2002
), ‘
The existence of R&D spillovers: a cost function estimation with random coefficients
,’
Economics of Innovation and New Technology
  ,
11
,
525
541
.
Sasidharan
S.
(
2006
), ‘
Foreign direct investment and technology spillovers: evidence from the Indian manufacturing sector
,’
UNU-MERIT working paper
.
Suzuki
J.
Kodama
F.
(
2004
), ‘
Technological diversity of persistent innovators in Japan: two case studies of large Japanese firms
,’
Research Policy
  ,
33
,
531
549
.
Suzumura
K.
(
1992
), ‘
Cooperative and noncooperative R&D in an oligopoly with spillovers
,’
The American Economic Review
  ,
82
,
1307
1320
.
Teece
D.
(
1977
), ‘
Technology transfer by multinational firms: the resource cost of transferring technological know-how
,’
Economic Journal
  ,
87
,
242
261
.
van Stel
A. J.
Nieuwenhuijsen
H. R.
(
2004
), ‘
Knowledge spillovers and economic growth: an analysis using data of Dutch regions in the period 1987–1995
,’
Regional Studies
  ,
38
,
393
407
.
Wakelin
K.
(
2001
), ‘
Productivity growth and R&D expenditure in UK manufacturing firms
,’
Research Policy
  ,
30
,
1079
1090
.
Wuyts
S.
Colombo
M. G.
Dutta
S.
Nooteboom
B.
(
2005
), ‘
Empirical tests of optimal cognitive distance
,’
Journal of Economic Behavior and Organization
  ,
58
,
277
302
.

Appendix

Table A1

Test of the difference in size of the estimated coefficients (using the random effect results of the following models)

Hypotheses tested  Model 4 in Table 6  Model 4 in Table 7  Model 5 in Table 8 Hypotheses supported? 
χ2 (1) (Prob. >  χ2 )  χ2 (1) (Prob. >  χ2 )  χ2 (1) (Prob. >  χ2 )  
H1-A: group patent (intra) > industry patent (intra)   2.95* (0.086) Yes (>) 
H1-B: group patent (inter) > industry patent (inter)   8.63*** (0.003) Yes (>) 
H2-A: group patent (intra) = group patent (inter)   0.00 (0.975) Yes 
H2-B: industry patent (intra) = industry patent (inter) 0.14 (0.708) 0.85 (0.356) 0.82 (0.366) Yes 
H3-A: group patent (intra) > industry patent (inter)   2.69* (0.101) Yes (>) 
H3-B: group patent (inter) > industry patent (intra)   9.70*** (0.002) Yes (>) 
H1: group patent (intra + inter) > industry patent (intra + inter) (from model 3 in Table 6 )   11.16*** (0.001)  Yes (>) 
(H1’) group patent (intra + inter) > industry patent (intra)  11.16*** (0.001)  Yes (>) 
(H1’) group patent (intra + inter) > industry patent (inter)  11.53*** (0.001)  Yes (>) 
Hypotheses tested  Model 4 in Table 6  Model 4 in Table 7  Model 5 in Table 8 Hypotheses supported? 
χ2 (1) (Prob. >  χ2 )  χ2 (1) (Prob. >  χ2 )  χ2 (1) (Prob. >  χ2 )  
H1-A: group patent (intra) > industry patent (intra)   2.95* (0.086) Yes (>) 
H1-B: group patent (inter) > industry patent (inter)   8.63*** (0.003) Yes (>) 
H2-A: group patent (intra) = group patent (inter)   0.00 (0.975) Yes 
H2-B: industry patent (intra) = industry patent (inter) 0.14 (0.708) 0.85 (0.356) 0.82 (0.366) Yes 
H3-A: group patent (intra) > industry patent (inter)   2.69* (0.101) Yes (>) 
H3-B: group patent (inter) > industry patent (intra)   9.70*** (0.002) Yes (>) 
H1: group patent (intra + inter) > industry patent (intra + inter) (from model 3 in Table 6 )   11.16*** (0.001)  Yes (>) 
(H1’) group patent (intra + inter) > industry patent (intra)  11.16*** (0.001)  Yes (>) 
(H1’) group patent (intra + inter) > industry patent (inter)  11.53*** (0.001)  Yes (>) 

Note: Please refer to the notes in Table 9 .