Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Three essays on supply chain networks and R&D investments
(USC Thesis Other)
Three essays on supply chain networks and R&D investments
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
THREE ESSAYS ON SUPPLY CHAIN NETWORKS AND R&D INVESTMENTS by Hyojin Song A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (ECONOMICS) August 2015 Copyright 2015 Hyojin Song Acknowledgments First, I would like to thank my advisor, Simon Wilkie for his invaluable guidance and encouragement. He is the most creative and innovative person I have ever seen. He motivates me all the time and allows me to think about the problem in a different way. While discussing with him, I could learn how to enjoy doing research. I would love to thank Cheng Hsiao, Yu-wei Hsieh, Hashem Peraran, Jian Lian and Grey Socic for being my committee of the qualifying/defense. Thanks to their help- ful comments and suggestions, I could think of the problem in various ways. I really appreciate Hangkoo Lee, KIET for allowing me to use invaluable data sets and advice. Without his help and advice, I would not research the Korean automobile industry and could not be here today. A special thank you goes to Sangtaek Kim. When I met him in 2008 as a master students, I could not imagine how much I can learn from you. For many projects and papers, I learned how to solve the problem. Whenever I face difficult situations during my Ph.D., his comments always helped me to get over the difficulties. I want to thank many good friends who have supported me. I really would like to thank Jay Gwon, Yusun Hwang and Younoh Kim. I remember that all their comments helped me to take each step from the core exam to the defense. I would like to thank my old friend, Hyewon Choi, Jihye Lee and Heeyoon Ko. Thanks to them, I could learn how to balance between research and regular life. ii Finally, but the most importantly, I thank my parents, Byungsub Song and Boeun Han, for their unconditional supports and love. Their constant support made this dis- sertation possible. I also thank all my Song family members. iii Table of Contents Acknowledgments ii List of Tables vii List of Figures viii Abstract ix Chapter 1: Introduction 1 Chapter 2: Does Sutton Explain the Korean Automobile Industry?: Exclusiveness and Partnerships with Hyundai and Kia 7 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.2 Korean Automobile Industry . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.2.1 Downstream Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.2.1.1 Market Overview . . . . . . . . . . . . . . . . . . . . . . 9 2.2.1.2 Hyundai-Kia vs. Non Hyundai-Kia . . . . . . . . . . . . 10 2.2.2 Upstream Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2.2.2.1 Market Overview . . . . . . . . . . . . . . . . . . . . . . 11 2.2.2.2 Exclusiveness vs. Non Exclusiveness . . . . . . . . . . . 12 2.2.3 Policy Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.3 Theoretical Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 2.3.1 Sunk Costs: R&D Investments . . . . . . . . . . . . . . . . . . . . 16 2.3.1.1 Exogenous Sunk Costs Case . . . . . . . . . . . . . . . . 16 2.3.1.2 Endogenous Sunk Costs Case . . . . . . . . . . . . . . . 17 2.3.2 Asymmetry of R&D behavior . . . . . . . . . . . . . . . . . . . . . 19 2.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 2.5 Empirical Patterns and Network Structures . . . . . . . . . . . . . . . . . 25 2.5.1 Knowledge Transferability . . . . . . . . . . . . . . . . . . . . . . 25 2.5.2 Market Size and Market Concentration . . . . . . . . . . . . . . . 27 2.5.3 R&D Patterns and Profit . . . . . . . . . . . . . . . . . . . . . . . . 30 2.5.4 Network Structures and Component Distribution . . . . . . . . . 33 2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 iv Chapter 3: Stability of the Supply Chain Networks : Evidence from the Korean Automobile Industry 38 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 3.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 3.3 Network Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 3.3.1 Information between upstream and downstream firms . . . . . . 45 3.3.1.1 Degree with length 1 . . . . . . . . . . . . . . . . . . . . 45 3.3.1.2 Exclusiveness . . . . . . . . . . . . . . . . . . . . . . . . . 46 3.3.2 Information flows among upstream firms . . . . . . . . . . . . . . 47 3.3.2.1 Degree with length 2 . . . . . . . . . . . . . . . . . . . . 47 3.3.2.2 Number of Competitors . . . . . . . . . . . . . . . . . . . 48 3.3.2.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 3.4 Stability of the Supply Chain . . . . . . . . . . . . . . . . . . . . . . . . . 50 3.4.1 Network Structures in 2008 and 2013 . . . . . . . . . . . . . . . . 51 3.4.2 Eigenvalue Approach . . . . . . . . . . . . . . . . . . . . . . . . . 55 3.4.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 3.4.2.2 Distance Measures . . . . . . . . . . . . . . . . . . . . . . 57 3.4.3 Evidence of Stability: Litigation . . . . . . . . . . . . . . . . . . . . 60 3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 Chapter 4: Estimation of Network Effects on Automotive Parts Suppliers’ R&D Investments: the Fixed Effects Filtered Estimation 62 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 4.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 4.1.2 Connections with Bain’s Paradigm . . . . . . . . . . . . . . . . . . 63 4.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 4.3 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 4.3.1 Econometric Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 4.3.2 Fixed Effects Filtered Estimator . . . . . . . . . . . . . . . . . . . . 69 4.3.3 Dependent Variable . . . . . . . . . . . . . . . . . . . . . . . . . . 73 4.3.4 Time-variant Regressors . . . . . . . . . . . . . . . . . . . . . . . . 73 4.3.5 Time-invariant Regressors . . . . . . . . . . . . . . . . . . . . . . . 74 4.3.5.1 Degree with length 1 . . . . . . . . . . . . . . . . . . . . 74 4.3.5.2 Exclusiveness . . . . . . . . . . . . . . . . . . . . . . . . . 75 4.3.5.3 Degree with Length 2 . . . . . . . . . . . . . . . . . . . . 75 4.3.5.4 Number of Competitors . . . . . . . . . . . . . . . . . . . 76 4.3.6 Characteristics of Components . . . . . . . . . . . . . . . . . . . . 77 4.3.6.1 Union . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 4.3.6.2 Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 4.4.1 Network Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 4.4.2 Exclusive Dealing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 4.5.1 Geographical Distribution . . . . . . . . . . . . . . . . . . . . . . . 88 4.5.2 Endogeneity Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 v 4.5.3 Determinats of Exclusiveness: Probit Estimation . . . . . . . . . . 91 4.5.3.1 R&D Investment . . . . . . . . . . . . . . . . . . . . . . . 93 4.5.3.2 Patents . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 4.5.3.3 Same Area . . . . . . . . . . . . . . . . . . . . . . . . . . 93 4.5.3.4 Number of Competitors . . . . . . . . . . . . . . . . . . . 93 4.5.4 Other Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 Chapter 5: Conclusion 95 Bibliography 98 Chapter A: Appendix to Chapter 2 101 Chapter B: Appendix 107 vi List of Tables 2.1 R&D Investments Conducted by Downstream Firms . . . . . . . . . . . . 11 2.2 R&D Investments Conducted by Upstream Firms . . . . . . . . . . . . . 12 2.3 Policy Changes in South Korea from 1962 to 2012 . . . . . . . . . . . . . . 14 2.4 Sales and R&D Investments . . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.5 Market Structure with respect to Partnership Structures . . . . . . . . . . 29 2.6 Evolutionary Patterns of R&D Expenditures . . . . . . . . . . . . . . . . . 31 2.7 Operating Profit Percentage with respect to Partnership Structures . . . 32 2.8 Distribution with respect to Product Categorization . . . . . . . . . . . . 35 2.9 Summary of Empirical Patterns with respect to Segment . . . . . . . . . 36 3.1 Summary Statistics of Network Measures . . . . . . . . . . . . . . . . . . 44 3.2 Network Measures (Example) . . . . . . . . . . . . . . . . . . . . . . . . . 50 3.3 Comparison between 2008 and 2013 Partnership Structures . . . . . . . . 54 3.4 Observations in Network Segments . . . . . . . . . . . . . . . . . . . . . 54 3.5 Distance Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 4.1 Number of Birth/Death . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 4.2 Number of Firms in the Sample . . . . . . . . . . . . . . . . . . . . . . . . 68 4.3 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 4.4 Estimation results: Network effects (A) . . . . . . . . . . . . . . . . . . . 82 4.5 Estimation results: Network effects (B) . . . . . . . . . . . . . . . . . . . . 83 4.6 Estimation results: Network effects (C) . . . . . . . . . . . . . . . . . . . . 86 4.7 Estimation results: Exclusiveness . . . . . . . . . . . . . . . . . . . . . . . 87 4.8 Distribution of Networks within Area . . . . . . . . . . . . . . . . . . . . 89 4.9 Estimation: Exclusiveness . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 vii List of Figures 2.1 Supply Chain Networks in the Korean Automobile Industry . . . . . . . 22 2.2 Vertical Structure of Air Cleaner (Press Type) . . . . . . . . . . . . . . . . 23 2.3 Mapping Characteristics of Components into Two Spaces . . . . . . . . . 24 2.4 Four Segments with respect to Exclusiveness and Identity of Partnerships 27 3.1 Number of Suppliers with respect to Partnerships . . . . . . . . . . . . . 42 3.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 3.3 Node Degree Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 4.1 Bain’s Paradigm and Circular Relationship . . . . . . . . . . . . . . . . . 64 4.2 Input-Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 4.3 Factories in South Korea . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 4.4 Number of Firms within Area . . . . . . . . . . . . . . . . . . . . . . . . . 80 B.1 Process of Supply Chain in the Automobile Industry . . . . . . . . . . . . 108 viii Abstract This dissertation evaluates how supply chain networks play a central role in the Korean automobile industry. The study consists of three essays. The first essay analyses the dynamic patterns of the Korean automobile industry focusing on the network mea- sures, the second essay tests the stability of network in the Korean automobile industry, and the third essay estimates the effects of network measures on automotive parts sup- pliers’ R&D investments. The first essay (Chapter 2) summarizes the Korean automobile industry and presents dynamic patterns. The dynamic patterns show that network structures should be con- sidered to connect theory with empirical data. The dynamic patterns of the Korean automobile industry are not consistent with the insights from Sutton (1991). As mar- ket size increased, the number of automotive parts suppliers remained almost the same from 1999 to 2010. However, the automotive parts market did not seem to be R&D intensive. To explain the contradiction, I define networks byexclusiveness and the iden- tity of partnerships (Hyundai-Kia and Non Hyundai-Kia) as network measures and divide the automotive parts suppliers into four sectors. When considering network measures, I find the consistent results with Sutton’s insights. The second essay (Chapter 3) shows the network measures and tests the stability of the supply chain in the Korean automobile industry. To test the stability, I use the eigen- value approaches of the Laplacian matrix. The network structure implies the partner- ship between automobile assemblies and automotive parts suppliers and the network structures are available in 2008 and 2013. Since network structures could be defined ix by adjacency matrices, distance measures would be based on graph spectra. I define the degree matrices for two matrices. The Laplacian matrix is the difference between the degree matrix and the adjacency matrix and the eigenvalue is used to calculate the distance measure. The measure shows that the supply chain networks in the Korean automobile industry is very stable. The third essay (Chapter 4) estimates the network effects on suppliers’ R&D deci- sions using panel data techniques. Due to the stability of the supply chain, network measures are time-invariant and thus cannot be estimated by the fixed effects estima- tor. Instead, the Fixed Effects Filtered estimator (Pesaran and Zhou, 2013) is used which can estimate both time-variant and time-invariant regressors consistently. The results suggest that the identity of partners,exclusiveness and information flows between com- petitors are important factors in suppliers’ R&D strategy. In addition, I find that internal R&D and external R&D are substitutes when supplier has an exclusive contract. x Chapter 1 Introduction Consisting of more than 30,000 parts, vehicles are one of the most complicated con- sumer products. As contributions to the supply chain account for roughly three quar- ters of the content of a vehicle, controlling the quality of vehicles requires supply chain management. To maintain and improve the quality of vehicles, both automobile assem- blies’ Research and Development (R&D) investments and automotive parts suppliers’ R&D are important. The objective of this dissertation is to investigate the role of supply chain networks in the automotive parts suppliers’ R&D investment behavior. One main feature of the supply chain is that information 1 , which is directly connected with R&D investment, flows as products and services move via the linkage. The hypothesis is that automotive parts suppliers have less internal R&D investments when knowledge is transferable from automobile assemblies to automotive parts suppliers because of the stability of the supply chain. If suppliers receive blueprints from Hyundai based on their long- term partnership, they might have less incentives to invest in R&D to minimize their sunk costs. Before testing the hypothesis using panel data techniques, I investigate the importance of networks structures and test the stability of the supply chain. In Chapter 2, I present dynamic patterns as evidence that networks structures have to be considered to connect theory with empirical data. Sutton’s (1991) endogenous sunk costs model shows that non-monotonic relationships between market size and market concentration can be found if endogenous elements in sunk costs play a role 1 Grossman and Helpman (1991) demonstrated that R&D is an input of technology and two main char- acteristics of technology are non-rivality and partial nonexcludability. Partial nonexcludability is that the owner of technological information may not prevent others from using it and this characteristic creates spillovers. Information flows can be interpreted as partial nonexcludability in that sense. 1 in enhancing consumers’ willingness to pay. From 1999 to 2010, the size of the Korean automobile market grew through implementation of various Free Trade Agreements (FTAs), 2 and the market size of the automotive parts industry increased because it is directly connected to the automobile industry. However, the number of automotive parts suppliers has not changed. There were 881 parts suppliers in 2001 and 898 in 2013 3 . This non-monotonic relationship implies that the Korean automobile market exhibits endogenous sunk costs. However, the Korean automotive parts industry does not seem to be R&D intensive. To be consistent with Sutton’s model, the Korean automotive parts industry should be R&D intensive. To explain this contradiction, I focus on information flow via net- work structures. Especially, the stability of the supply chain structure is an interesting characteristic that the Korean automobile industry has. The stability of supply chains has been selected as the strength of the Korean automobile market (Lee, 2010). The hypothesis is that automotive parts suppliers can have less internal R&D if knowledge is transferable between automotive parts suppliers and automobile assemblies and this relationship is stable over time. I divide automotive parts suppliers into four segments by the level of knowledge transferability. As a proxy for knowledge transferability, two network measures, exclu- siveness and the identity of the partnerships, are used. Exclusiveness is defined as suppli- ers who have one partner to trade. The four segments are theexclusivewithHyundai-Kia segment (Exc-HK), the exclusive with Non Hyundai-Kia segment (Exc-NonHK), the non- exclusivewithHyundai-Kia segment (NonExc-HK), thenon-exclusivewithNonHyundai-Kia segment (NonExc-NonHK). As the market size of all four increases, the market concentration moves in different directions for each segment. The data set shows a non-monotonic relationship for the 2 FTAs are in effect with Chile, Singapore, EFTA, ASEAN, India, Peru, the EU and the U.S. and under negotiation with Canada, Mexico, GCC, Australia, New Zealand, China, Vietnam and Indonesia. The main clauses in contracts pertain to taxes on vehicles and automotive parts. 3 The number of suppliers did not change over time. There were 878 suppliers in 2003 and 794 in 2008. 2 NonExc-HK segment and a monotonic relationship for the NonExc-NonHK segment. If the patterns were consistent with Sutton’s insights, suppliers in theNonExc-HK segment would focus more on the quality-sensitive products, and the Nonexc-NonHK segments are more likely to produce homogeneous products. The component-related data set allows me to capture the component distribution in each segment. I find the results are consistent with Sutton’s insights. Automotive parts suppliers in theNonExc-HK segment produce more technology-oriented products such as steering, and power generating systems. The NonExc-NonHK segment is more likely to produce fewer quality-sensitive products such as lamps and parts (aluminum). I also find that the exclusive segments produce more design-oriented products such as molding and leather. In Chapter 3, I test the stability of the supply chain networks by the observed net- works of the Korean automobile supply chain in two different time periods. Theoreti- cally, the substitute conditions are crucial to guarantee the existence of pair-wise stable allocations (Hatfiled and Milgrom, 2005; Halfield and Kominers 2010). However, the substitutes do not hold for automotive suppliers. Hatfield and Kojima (2008) show that stable allocation may exist even if contracts are not substitutes. A naturally arising question is then whether it is possible to test the stability of net- works empirically. The stability of the supply chain is based on the long-term relation- ship between automotive parts suppliers and automobile assemblies. However, there is no existing literature to test the stability of supply chain networks in the automobile industry due to the inability of data set. The observed network data from two time periods, 2008 and 2013 allow me to measure the distance of two networks and test the stability. Since networks structures can be defined by adjacency matrices, distance measures should be based on the graph spectra. The distance measures, which I use, are the eigenvalue approach of the Laplacian matrix. The Laplacian matrix is defined as the 3 difference between the degree matrix and the adjacency matrix. To test stability, I calcu- late eigenvalues of the normalized Laplacian matrix of two network structures in 2008 and in 2013 and then derive the distance measures. The measures show that the supply chain networks in the Korean automobile industry are very stable. In Chapter 4, I estimate the effects of network measures on automotive suppliers’ R&D investments using panel data. I define the network measures to capture the infor- mation flow between upstream and downstream firms and among upstream firms: the degree with length 1, the degree with length 2, the number of competitors, andexclusive- ness. In the panel data analysis, a main concern is the unobservable firm-heterogeneity. If the unobservable firm-heterogeneity were correlated with regressors, the OLS results would be biased and inconsistent. The fixed effects estimator cannot be used here since all the network measures are time-invariant due to the stability of network structures. I applied the Fixed Effects Filtered (FEF) estimator (Pesaran and Zhou, 2013) to estimate both time-variant and time-invariant regressors consistently with the correct covariance matrix. The Hausman and Taylor (HT) estimator is considered but cannot be identified due to the restrictions of the data set. The results show thatexclusiveness and the level of price competition play an important role concerning the level of investment committed to R&D. This dissertation contributes to three strands of the literature. The first contribution addresses the role of endogenous sunk costs on the relationship between market struc- ture and market concentration (Sutton, 1991). Empirical analysis has been done for dif- ferent industries: newspapers and restaurant industry (Berry and Waldfogel, 2010), the supermarket industry (Ellickson, 2007, 2013) and the mutual fund industry (Gavazza, 2011; Park, 2013). In the U.S. mutual fund industry, Park (2013) divides the exogenous sunk costs market and the endogenous sunk costs market by segments with and with- out loads and Gavazza (2011) uses the retail and institutional funds industry. 4 Second, this study contributes to empirical literature on networks. Estimating the network effects is related to measurement issues because it is not obvious how net- works should be measured. Typically defining the network is simply defining the neighborhood. For example, in the early development literature, the village level had been used as the best possible measure of networks (Munshi and Myaus, 2006, Mun- shi, 2004, Foster and Resenzweig, 1995) and specific questions to determine the rela- tionships are often included in the experiments (Conley and Udry, 2010). However, research on networks effects among firms has been limited due to the availability of data. This paper empirically estimates the network effects on suppliers’ R&D invest- ment decisions under the supply chain structure. The uniqueness of the data allow me to define various measures to capture the network effects between automotive parts suppliers and automobile assemblies as well as among parts suppliers. In addition, I directly test the sustainabilty of the supply chain. The theoretical analysis focuses on the conditions to obtain a pair-wise stability in the supply chain (Ostrovsky, 2008; Hat- field and Kominers, 2012). Same-side substitutability and cross-side complementarity are the main conditions to guarantee stable allocations and the automobile industry is a typical example that does not satisfy the substitutability condition. Hatfield and Kojima (2008) show that stable allocation may exist even if contracts are not substitutes but a weaker form of substitutability is necessary. Fox (2010)’s work uses the maximum score estimator. This study also contributes to the empirical literature on R&D investment. Belder- bos et al. (2006) test if R&D cooperations with different types of partners are comple- ments using the definition of complementarity by Milgrom and Roberts (1990). They find empirical evidence of complementarities from joint cooperation strategies with competitors and customers and with customers and universities. This dissertation organizes as follows: Chapter 2 analyses the dynamic patterns of the Korean automobile industry focusing on the network measures. In Chapter 3, I test the stability of network in the Korean automobile industry. Chapter 4 estimates the 5 effects of network measures on automotive parts suppliers’ R&D investments. Finally, Chapter 5 concludes. 6 Chapter 2 Does Sutton Explain the Korean Automobile Industry?: Exclusiveness and Partnerships with Hyundai and Kia 2.1 Introduction This paper presents the empirical evidence that supply chain networks play a cen- tral role in automotive parts suppliers’ Research and Development (R&D) investments. Specifically, I focus on the inconsistency between theoretical implications from Sutton (1991) and dynamic patterns in the Korean automobile industry. Suttons endogenous sunk cost model predicts that automotive parts suppliers are R&D intensive. However, a low level of Korean automotive parts suppliers R&D investments has been consid- ered as a significant problem. To explain this contradiction, I put weights on knowledge transferability based on the stability of the supply chain. I define knowledge transfer- ability by network structures (exclusiveness and the identity of partnerships, especially Hyundai-Kia and Non Hyundai-Kia). By analyzing component-related information, I 7 found that Suttons predictions are still consistent with empirical patterns when net- works structures are considered in empirical analysis in R&D investment. Supply chain networks consist of three elements: downstream firms, upstream firms and linkages between downstream firms and upstream firms. Downstream firms are automobile assemblers such as Hyundai and Kia and upstream firms are automotive parts suppliers who produce parts or components and sell them to downstream firms. Linkages between downstream firms and upstream firms can be defined as transac- tions. One main feature of the supply chain is that information flows as products and services move via the linkage . Since information and knowledge through linkages are directly related to R&D Investments, considering network structures has been more and more important in empirical analysis of R&D investments. In spite of growing interests, research is limited due to data inability. In this paper, I focus on the role of supply chain structures in empirical analysis using Sutton (1991)s endogenous sunk costs model. I present dynamic patterns as evi- dence that networks structures have to be considered to connect theory with empirical data. Sutton’s endogenous sunk costs model shows that non-monotonic relationships between market size and market concentration can be found if endogenous elements in sunk costs play a role in enhancing consumers’ willingness to pay. From 1999 to 2010, the size of the Korean automobile market grew through implementation of vari- ous Free Trade Agreements (FTAs) , and the market size of the automotive parts indus- try increased because it is directly connected to the automobile industry. However, the number of automotive parts suppliers has not changed. There were 881 parts suppli- ers in 2001 and 898 in 2013 . This non-monotonic relationship implies that the Korean automobile market exhibits endogenous sunk costs. However, the Korean automotive parts industry does not seem to be R&D intensive, which is not consistent with Sutton’s model. To explain this contradiction, I focus on special characteristics of the supply chain in the Korean automobile industry which have been derived by Korean govern- ment policy. 8 This Chapter consists of the following sections. Section 2.2 explains the Korean automobile industry focusing the supply chain. Section 2.3 discusses theoretical impli- cations of Sutton’s endogenous sunk costs model. Section 2.4 describes data sets. Sec- tion 2.5 presents empirical patterns with respect to network structures and Section 2.6 concludes. 2.2 Korean Automobile Industry 2.2.1 Downstream Firms 2.2.1.1 Market Overview Downstream firms in the automobile market are defined as automobile assemblers who buy parts or components from upstream firms, assemble them to produce vehicles, and sell to consumers. BMW, Toyota and Hyundai are examples of downstream firms. In the Korean automobile market, there are six automobile assemblers: Hyundai, Kia, GM Korea, Renault Samsung, Ssangyong, and Tata Daewoo. A distinguishing charac- teristic of the Korean domestic automobile market is that domestic auto brands dom- inate the market. By 2012, Hyundai accounted for 44.6%, Kia for 33.5%, GM Daewoo for 9.5%, Renault Samsung for 7.4% and Ssangyong for 2.6% of the market. In other words, all other foreign brands including BMW, Honda, Toyota, Mercedes, Audi, and Nissan accounted for only 2.4% of the Korean domestic market. The government had supported the domestic automobile market especially until 1980 in order to increase regional growth and employment of the middle class. The stable domestic market share has helped domestic automobile assemblers to enhance competitiveness in the interna- tional market. To analyze the downstream firms in Korea, historical facts are important. The cur- rent Korean automobile market structure changed tremendously after the 1997 Asian Financial Crisis. Before 1997, Hyundai, Kia, Daewoo, Samsung and Ssangyong were 9 domestic automobile companies. Kia had financial troubles, and Kia Motors declared bankruptcy in 1997 and acquired 51% of the assets. Currently, 32% of Kia is owned by Hyundai. Kia Motors is considered as a subsidiary of Hyundai Motors. Daewoo Motors ran into financial problems and sold to General Motors. Daewoo Commercial Vehicle Company was separated from parent Daewoo Motors and was acquired by Tata Motors. Samsung Motors has been a subsidiary of Renault and changed its name to Renault Samsung Motors from 2000. Ssangyong Motor was acquired by Tata Dae- woo in 1997, sold to Chinese automobile manufacturer SAIC in 2004 and then, to Indian Mahindra and Mahindra Limited in 2011. 2.2.1.2 Hyundai-Kia vs. Non Hyundai-Kia Korean automobile companies can be divided into two groups by two criteria: for- eign ownership and market power. The first group is Hyundai and Kia and the second group is Non Hyundai-Kia (GM Daewoo, Renault Samsung, SsangYong and Tata Dae- woo). First, the Korean automobile manufacturers except for Hyundai and Kia were acquired by foreign automobile manufacturers. Domestic or foreign ownership is important to decide the level of R&D investments conducted in Korea. Hyundai-Kia has to design and produce their own models while foreign-owned automobile com- panies do not need to and instead, focus on licensing vehicle models from GM and Renault. Second, there is a big difference between the two groups from the viewpoints of the market share. If Hyundai and Kia were considered as one company, the market share of Hyundai and Kia would be 78.1% and a summed market share of the four is less than 20%. Naturally, downstream firms have more power than upstream firms by the characteristics of the producer-driven supply chain in the automotive industry. The difference of the market share would affect the power in negotiations with automotive parts suppliers. Naturally, downstream firms have more power than upstream firms by the characteristics of the producer-driven supply chain in the automobile industry. 10 Table 2.1: R&D Investments Conducted by Downstream Firms 2.2.2 Upstream Firms 2.2.2.1 Market Overview Upstream firms are defined as automotive parts suppliers who produce parts or components such as airbags and brake pedals and sell them to downstream firms which are automobile assemblers. Historically, upstream firms were developed under verti- cally integrated frameworks when the automobile industry started. Automobile assem- blies provided automotive parts suppliers with the blueprint including the details of technically sensitive information. Thus many features of upstream firms have been developed according to the identity of partnerships. Automobile assemblies rather than automotive parts suppliers are the primary investors in R&D. $2.46 billion which accounts for 67.5% was invested by 5 automobile assemblies in 2008. 11 Table 2.2: R&D Investments Conducted by Upstream Firms 2.2.2.2 Exclusiveness vs. Non Exclusiveness Exclusiveness is defined as the suppliers having only one partnership with an auto- mobile assembly. Exclusiveness represents a higher level of knowledge transferability from automobile assemblies to automotive parts suppliers. This is because automo- bile assemblies tend to be less concerned that technologically sensitive information will spread to their rivals. Knowledge transferability based on exclusiveness allows internal R&D and external R&D to be substitutes. 2.2.3 Policy Changes Historical policy changes over the previous six decades explain the structural devel- opment of the Korean automobile industry. It is possible to divide this development 12 into 8 periods based on the political circumstances. Table 1 summarizes changes in the Korean automotive industry from 1962 to 2012. Lee (2012) divided this period into two identifiable time frames, Protection and rationalization from Period I to Period III and Globalization, innovation and regulation from the Period IV to Period VIII. Period I, 1962-71 was Localization. The goal of the Korean government is to protect the domestic market and nurture domestic auto makers. To overcome a lack of knowl- edge and know-how in producing vehicles, auto makers were encouraged to sign tech- nical licensing agreements with advanced foreign auto companies. As a tool of domes- tic market protection, operations of foreign auto makers were banned except for joint ventures. A low level of exchange rates, interest rates and oil prices had positive effects on the speed of localization. 13 Table 2.3: Policy Changes in South Korea from 1962 to 2012 14 Period II, from 1972 to 1976, was a time of Quantitative growth which focused on developing native models and production capability. Economy of scale and scope was used by the long-term growth plan. In addition, the government supported ver- tical integration and cooperation. In 1974, the government announced its Long-term automotive industry promotion plan, which promoted cooperation between parts and assembly industries, and in 1975, it designated 35 vertically integrated parts producing facilities. These systematic supports in favor of a stable relationship between automo- tive parts suppliers and automobile assemblies were the main source of the current stability of the supply chain in the Korean automotive market. Period III, 1977-1981 can be called Mass production and specialization; it established large-scale production and export strategies. From Period IV , the focus had moved from protection to liberalization. Period IV , from 1982 to 1991, was Liberalization to change from government-led growth to private-initiated growth with higher competition. Prior to 1987, foreign car makers were prohibited from selling vehicles in Korea. From 1987, they were permitted to sell vehicles larger than 2000 cc and from 1988 on, they were allowed to make all types of cars. The domestic market had been protected by a high level of tariffs, 50% by 1987, which gradually decreased to 20% in 1990 as a result of liberalization. Period V , 1992-1997, was Technology innovation in which Korean automakers expanded and became self-reliant in terms of technology. There were big changes before and after the 1997 Asian financial crisis. By the IMF’s instruction, globalization occurred rapidly during Period VI, 1998-2002, and qualitative growth was achieved. Period VII, 2003-2007, was represented by the environmental and safety related regulations and future-oriented vehicle development plans. The Green car project explains the Period VIII, 2008-2012, which developed new engines to stimulate growth. 15 2.3 Theoretical Implications In this section, I discuss how empirical patterns are consistent with Sutton (1991)’s endogenous sunk costs model. Corchon and Wilkie (1994) explain asymmetry of R&D behaviors. 2.3.1 Sunk Costs: R&D Investments Sutton (1991) studies two analytic frameworks 1 , exogenous sunk costs and endoge- nous sunk costs to show the relationship between market size and market concentra- tion. A major difference between the exogenous sunk costs model and the endogenous sunk costs model centers around the role of fixed outlays in sunk costs in determining the consumers’ willingness to pay. 2.3.1.1 Exogenous Sunk Costs Case The exogenous sunk cost model is the two-stage game. The first stage is the entry decision. Firms enter the market if the net profit, , is greater than the sunk costs for setting up,. The number of firms, N, is determined in this stage. 1 In considering the theoretical implications, I only include examples of the Cournot competition. Details and other cases such as the Bertrand competition are discussed in Sutton’s book. Both exogenous and endogenous sunk costs models are solved by backwards induction. 16 The second stage is the Cournot competition. Given the number of firms, N, the level of price and quantity are determined by maximizing their profit. 2 The model yields the following equation: N = r S The main implication of the exogenous sunk costs model is a monotonic relationship between the market size and the market concentration. As the market size increases, the market becomes more competitive. 2.3.1.2 Endogenous Sunk Costs Case The endogenous sunk costs model consists of three stages. The first stage is the entry decision. Firms determine to enter the market if the net profit exceeds the sunk costs. The difference is that the sunk costs contain two elements: the set up cost,, and R&D investment,A(u). (u)F (u) = +A(u): At the second stage, given the number of firms, firms determine the quality level,u, by the first-order condition. Then, the levels of R&D or advertising are determined by the response function,A(u) 3 2 The demand schedule isX =S=p, and firm i’s profit, =pi( P xj )xicxi 3 A(u) is a convex smooth function on the domainu 1 andA(1) = 1. In Sutton’s example,A(u) = a (u 1); > 1 17 d du j u= u dF du j u= u 0 and u 1 w:complementaryslackness The third stage is the Cournot Competition 4 . N + 1 N 2 = 2 [1 a= S N 2 ]: The main implications of the endogenous sunk costs model are that a non- monotonic relationship between the market size and the market concentration can be found according to the value of anda= . In the R&D or advertising-intensive industry in which endogenous elements in sunk costs are associated with consumers’ preference, the market concentration can even increase as the market size increases. From 1999 to 2010, the market size of the automotive parts industry increased as measured by sales. As the market size increased, the number of automotive parts sup- pliers remained the same. There were 881 parts suppliers in 2001 and 898 in 2013. This non-monotonic relationship implies that the Korean automobile market is more likely to be an instance of endogenous sunk costs. However, the R&D investments patterns of automotive parts suppliers are not consistent with Sutton’s model. The automotive parts industry does not seem to be R&D intensive. To think about this paradox, I will focus on knowledge transferability in Section 5 with empirical patterns. 4 ui=pi = uj=pj is derived by the consumer’s problem to maximizeU = (ux) z 1 subject topixi + pzz M wherex is the good in interest,z is outside good, andu is the index of perceived quality. The demand schedule isX =S=p and profit is =pi( P xj )xicxi. 18 2.3.2 Asymmetry of R&D behavior Corchon and Wilkie (1994) model the sources of the productivity paradox in which investments in a new technology have not led to a general increase in productivity. The model consists of two stages based on the traditional Cournot model with a dis- crete change of technology. Technology consists of a fixed capital cost,f, and a variable cost, c. The initial production technology is T 0 = (f 0 ;c 0 ) and the new technology is T 1 = (f 1 ;c 1 ), wheref 1 >f 0 ,c 1 <c 0 . The setting implies that more investments in R&D can decrease suppliers’ variable costs. By the Cournot competition 5 , the following three conditions can be derived: (1) A firm will innovate if and only if f a + (n 1)c 0 nc 1 n + 1 g 2 f ac 0 n + 1 g 2 >f 1 f 0 (2) Each firm invests in the new technology if f ac 1 n + 1 g 2 f a + (n 1)c 1 nc 0 n + 1 g 2 f 1 f 0 (3) A sufficient condition for productivity to fall is: f 1 f 0 > (c 0 c 1 )f ac 1 n + 1 g 5 Firms compete to maximize the profit. Profit is i(q;f0;c0) =D(Q)qic0qif0. The demand curve is normalized linear and total factor productivity is defined as the inverse of summation of average costs. 19 Corchon and Wilkie find that when ac 1 c 0 c 1 n 2 n1 , then for some values off 1 f 0 the productivity paradox will occur. They show that if the second equation holds, then the symmetric equilibrium is the unique subgame perfect equilibrium of the game (SPNE). Hyundai and Kia invest in R&D but the four other assemblies rarely commit to R&D investments in Korea. The current automobile market is far from being an instance of symmetric equilibrium. To understand an asymmetric case, I need to consider the meaning that equation (2) does not hold. Equation (2) is the condition that all firms invest in a new technology. When the leading firm introduces the new technology 6 , the firm can enjoy the profit and an increase in market share. Then as more firms adopt the new technology, a high level of competition erodes market share. Since fixed costs in investing in R&D have to be spread between firms, the productivity paradox can occur. In the Korean automobile industry, Hyundai and Kia adopted the technology earlier than the other assemblies and increased their market share. When the other assemblies adopt the new technology, their productivity may not increase due to increased com- petition and to a high level of fixed costs which spread to a small market share. Then, during the 1997 Asian financial crisis, they were acquired by foreign companies. 2.4 Data The main objective is to find empirical patterns with respect to network structures. The data set used in this Chapter is constructed by three data sets: annual financial data, network data and component-related information. 6 Their results show that a monopolist introduce the new technology if and only if the new technology raises total factor productivity and the equation (2) implies the equation (1) 20 The first data set contains 480 automotive parts suppliers from 1999 to 2010. The data set has been constructed from annual financial reports using the DART (Data Anal- ysis, Retrieval and Transfer) system of Financial Supervisory Service. It consists of the total revenue, profit, total liability, equity, capital, current assets, current liability, R&D investment, and number of employees. Filing disclosure documents on main events is mandatory for listed corporations in the KRX or the KOSDAQ or for companies whose capital assets exceed $7 million based on the previous year. The second data set is the network data of the Korean automotive industry in 2008. It reveals the linkages between 892 automotive parts suppliers and 6 automobile assem- bly companies in 2008. The information about the linkages allows researchers to define the networks measures between upstream and downstream firms and among upstream firms. Figure 2.1 shows the complexity of the network structures. 21 Figure 2.1: Supply Chain Networks in the Korean Automobile Industry 22 The third data set is component-related information which offers important infor- mation for two purposes. As Figure 2.1 shows, the supply chain structure is extremely complicated. The key trick to simplify the structure is to consider the supply chain with respect to each component. For example, there are 7 suppliers to produce the air cleaners press type. Figure 2.2: Vertical Structure of Air Cleaner (Press Type) As Figure 2.2 shows, the relationship is more simply observable and the market can be seen as a two-sided market. The data set includes which parts are produced by 480 suppliers. Since there are more than 50,000 components, the information is grouped into 378 different components. By combining this with the second data set, I can identify which components are produced by each supplier. I projected 378 parts on two different spaces. The first space is categorized by materials and the second space focuses on the use of the automotive parts in the vehicle. The first space contains 21 categories and the second does 11. For example, I can map oil pump as pump in the first standards of categorization and power generating and transfer system by the second standards as Figure 2.3 shows. 23 Figure 2.3: Mapping Characteristics of Components into Two Spaces 24 During the process to combine the first and second data set, I exclude parts suppli- ers who sell their parts to the other parts suppliers. In other words, the sample includes only first-tier suppliers who have partnerships with automobile assemblies. Lee (2010) said that there are more than 2000 second-tier suppliers but it is difficult to observe them by the data set. This is because the customers of second-tier suppliers vary across the industry. Products of second-tier suppliers are raw or intermediate material of auto- motive parts as well as materials in other industries. 2.5 Empirical Patterns and Network Structures 2.5.1 Knowledge Transferability Since knowledge is not visible, measuring the knowledge transferability is the first empirical issue. To find an adequate proxy of knowledge transferability, I used two measures: exclusiveness and the identity of partnerships. Exclusiveness is defined as the suppliers having only one partnership with an auto- mobile assembly. Exclusiveness represents a higher level of knowledge transferability from automobile assemblies to automotive parts suppliers. This is because automo- bile assemblies tend to be less concerned that technologically sensitive information will spread to their rivals. Knowledge transferability based on exclusiveness allows internal R&D and external R&D to be substitutes. The identity of partnerships is defined as Hyundai-Kia (HK) and Non Hyundai- Kia (NonHK). Hyundai and Kia have distinguishing characteristics from the four other assemblies (GM Korea, Renault Samsung, Ssangyong, and Tata Daewoo) based on two 25 criteria: foreign ownership and market power. Having a partnership with Hyundai and Kia determines a higher level of knowledge transferability because Hyundai and Kia are the only companies to conduct R&D in Korea. The four other automobile assem- blies were acquired by foreign automobile manufacturers and focus more on licensing vehicle models. For example, GM Korea licenses the model from General Motors and R&D of General Motors is conducted in the U.S. 73.35% of patents have been made by Hyundai and Kia in Korea, which is consistent with R&D patterns. The market share of Hyundai and Kia is 78.1% if the two companies are considered as one company and a summed market share of the four other assemblies is less than 20%. Theoretically, asymmetry of R&D behavior could be explained by Corchon and Wilkie (1994). Their model shows that a monopolist can enjoy the profit and an increased market share when he introduces the new technology. As more firms adopt new technology, a high level of competition erodes market share. Since fixed costs in investing in R&D have to be spread between firms, the productivity paradox can occur. According to two measures of the level of knowledge transferability, I divide auto- motive parts suppliers into four segments as follows: (1) Exclusive with Hyundai-Kia segment (Exc-HK): High-High (2) Exclusive with non Hyundai-Kia segment (Exc-NonHK): High-Low (3) Non-exclusive with Hyundai-Kia segment (NonExc-HK): Low-High (4) Non-exclusive with non Hyundai-Kia segment (NonExc-NonHK): Low-Low The Exc-HK segment has the highest level of knowledge transferability and the NonExc-NonHK segment has the lowest level of knowledge transferability. Exclusive- ness has a higher level of knowledge transferability than non-exclusiveness and having a partnership with Hyundai-Kia has a higher level than not having a partnership with Hyundai-Kia. Analysis with respect to segments provides us with interesting insights. 26 Figure 2.4: Four Segments with respect to Exclusiveness and Identity of Partnerships 2.5.2 Market Size and Market Concentration The market size can be measured by the average total sales of parts suppliers and the market concentration is measured by the 3-firm concentration ratio and the 5-firm concentration ratio. Table 2.2 and Table 2.3 describe the patterns of the market size and the market concentration for each segment. The market size increased for all four segments from 1999 to 2010. The market size increased from 50.1 billion won to 143.2 billion won and from 27.3 billion won to 77.9 billion won for the Exc-HK segment and for the Exc-NonHK segment respectively. 27 Table 2.4: Sales and R&D Investments 28 Table 2.5: Market Structure with respect to Partnership Structures 29 The market concentration of both the Exc-HK segment and the Exc-NonHK segment did not change over time. The NonExc-HK segment shows a non-monotonic relation- ship. As the market size increased from 78.5 billion won to 378.7 billion won, the 3-firm concentration ratio of the NonExc-HK segment increased from 30.28 to 53.45. A mono- tonic relationship is found for the NonExc-NonHK segment. The market size increased from 48.6 billion won to 99.6 billion won and the 3-firm concentration ratio decreased from 59.50 to 42.43. The data show that with the increase in the market size, the market became less concentrated. R&D increased. 2.5.3 R&D Patterns and Profit Interesting features of dynamic patterns of R&D are observed in the Exc-HK seg- ment and the NonExc-HK segment. Even if both segments have a contract with Hyundai-Kia, R&D behaviors are completely different. For 12 years, R&D expenditure of the Exc-HK segment did not increase. Since the market size increased, the R&D inten- sity decreased. In the case of the NonExc-HK segment, the average R&D expenditure increased more than 5 times. Both the Exc-NonExc segment and the NonExc-NonHK segment have twice increased R&D expenditure over 12 years. In addition, I checked the operating profit percentage to see the relationship between exclusiveness and the financial stability. Table 2.5 shows Operating profit per- centage with respect to year and partnership structure. The Exc-HK segment can be considered as having the most stable sources of profits from Hyundai-Kia since the standard error is the lowest among four segments. Suppliers in the Exc-NonHK seg- ment are most affected by financial fluctuations. When economic conditions are good, their operating profit percentages are greater than the profit percentage of suppliers in the Exc-HK segment. However, their profit percentages were even negative in 1999 and 2009. 30 Table 2.6: Evolutionary Patterns of R&D Expenditures 31 Table 2.7: Operating Profit Percentage with respect to Partnership Structures 32 2.5.4 Network Structures and Component Distribution Sutton’s model does not consider the exclusive contract case. Suppliers do not need to invest in R&D if a downstream firm provides important quality related information based on an exclusive contract. The NonExc-HK segment and the NonExc-NonHK seg- ment can be explained by Sutton’s model. Thus, I discuss the non-exclusive segments (Exc-HK and Exc-NonHK) and then deal with the exclusive segments (NonExc-HK and NonExc-NonHK). According to Sutton’s prediction, R&D intensive industry, which can be regarded as the endogenous sunk cost market, is more product-differentiated and their prod- ucts are more quality-sensitive. I find a monotonic relationship in the NonExc-NonHK segment and a non-monotonic relationship in the NonExc-HK segment. Since the qual- ity is used as a channel to derive the non-monotonic relationship in the case of the endogenous sunk costs model, I have the following predictions: If the non-exclusive with HK segment contains a higher level of endogenous sunk costs, components/parts produced in this segment are more likely to be product-differentiated and more sensi- tive to quality. Similarly, if the level of endogeneity in the sunk costs were smaller in the NonExc-NonHK segment, their products would more likely be homogeneous products and less sensitive to quality. The data set allows me to identify which components/parts suppliers are produc- ing. Since there are more than 300 categories of components/parts, I need to map these into smaller numbers of categories. I mapped 300 components/parts ids into two dif- ferent spaces and obtained the distribution of component in each segment. The distri- bution is described in Table 2.6. I can interpret the results of Table 2.6 by observing the frequency. The NonExc-HK segment produces more electronics, electric equipment, fabricated metal, and pumps 33 by the mapping 1 and steering, electronics (batteries), and power generating systems by the mapping 2. In the case of the NonExc-NonHK segment, I find evidence that sup- pliers concentrate on standardized products: by the mapping 1, aluminum, molding, and rubber and by the mapping 2, lamps. A major characteristic of the exclusive segments is the knowledge transferability from downstream firms to upstream firms. Both the exclusive with HK segment and the exclusive with non HK segment would be different from Sutton’s model. The com- ponent distribution shows which components are more often produced. The Exc-HK segment has the highest level of knowledge transferability from Hyundai-Kia and focuses on the design specific component. Products produced by suppliers in the exclusive with HK segment are leather, and molding by the mapping 1, and seats, and parts by the mapping 2. Suppliers in the Exc-NonHK segment are more likely to produce casting by the mapping 1 and parts by the mapping 2. 34 Table 2.8: Distribution with respect to Product Categorization 35 Table 2.9: Summary of Empirical Patterns with respect to Segment 36 2.6 Conclusion In this Chapter, I present dynamic patterns as evidence that networks structures have to be considered to interpret the theoretical implications from Sutton’s model. Without the network structures, patterns of the Korean automobile industry do not explain the insights from Sutton’s endogenous sunk costs model. As market size increased, the number of automotive parts remained almost the same from 1999 to 2010. However, the automotive parts market did not seem to be R&D intensive. To explain this contradiction, I focus on information flows from automobile assemblies to automotive parts suppliers. Historically, the Korean government encouraged automobile assemblies and auto- motive parts suppliers to be vertically integrated to maximize the efficiency. As a result, it was common that the Korean automobile assemblies provided blueprints of parts/components to their partners based on their long-term partnerships. However, it is not obvious to measure how knowledge is transferable from automo- bile assemblies to suppliers in the real world. For a proxy of the level of knowledge transferability, I use two measures, exclusiveness and the partnership with Hyundai- Kia and then define four different sectors: Exc-HK, Exc-NonHK, NonExc-HK and NonExc-NonHK. When networks structures are considered, consistent results with Sut- ton’s model are obtainable. 37 Chapter 3 Stability of the Supply Chain Networks : Evidence from the Korean Automobile Industry 3.1 Introduction There has been a large and growing literature on network from different viewpoints. Some focus on estimating how shocks diffuse through networks (Acemoglu et al., 2012; Cao, 2010; Allen and Gale, 2000; Kelly et al., 2012; Pistol, 2009) and evaluating the agents’ behavior compared with the optimal level of networks (Nagurney, 2009). How- ever, the results are still limited due to the complexity of network structures and case- specific situations. Theoretically, there are two crucial conditions to guarantee the existence of pair-wise stable allocations (Hatfiled and Milgrom, 2005; Halfield and Kominers, 2010; Ostrovsky, 2008). Two conditions are the same-side substitutability and the cross-side complemen- tarity. The cross-side complementarity is a mirror image of same-side substitutability. However, the supply chain networks in the automobile industry are a typical exam- ple that the same-side substitutability may not hold. There is a possibility to have a sta- ble allocation even if the same-side substitutability does not hold (Hatfield and Kojima, 2008). Fox (2010) estimates the parameters to maximize the number that matched pair’s 38 revenue function is greater than the counterpart pair’s revenue function using Man- ski’s maximum score estimator in the U.S. automobile industry. His work is based on the assumption that the network is pairwise stable. Even if the stability of the supply chain is an important theoretic issue, there is no clear approach to test the stability in an empirical sense. The main objective of this Chapter is to test the stability of networks empirically and evaluate the network structures in the Korean automobile industry. The stability of the supply chain is based on the long-term relationship between automotive parts suppliers and automobile assemblies. However, there is no existing literature to test the stability of supply chain networks in the automobile industry due to the inability of data set. The observed network data from two time periods, 2008 and 2013 allow me to measure the distance of two networks and test the stability. Testing the stability in the Korean automobile industry is an interesting and chal- lenging issue. The stability of the supply chain implies long-term contracts between upstream firms and downstream firms and has been selected as a main source to derive their rapid growth and success of the Korean automobile assemblies in the global com- petitive market. Historically, the stability was established by the Korean government’s policy. The Korean government supported vertical integration between upstream firms and down- stream firms. For example, in 1974, the government announced cooperation and in 1975, 35 vertically integrated parts producing facilities are constructed. Even if the sta- bility was a main characteristic of the Korean automobile industry, there is an argument that the stability has been weaken tremendously because of the competitiveness in the global market. However, there is no existing literature to test the stability in the auto- mobile market. In this Chapter, I have two approaches toward network structures. The first is calcu- lating various network measures such as degree to examine the network structure and 39 the second is the graph spectra approach. In both approaches, the role of linkages is important. I define the supply chain networks in the automobile industry as a system which contains three elements: downstream firms, upstream firms and linkages. Upstream firms are automobile companies who are purchasing components from automotive parts suppliers, produce vehicles and sell them to consumers. In the Korean automo- bile industry, there are six automobile assemblies: Hyundai, Kia, GM Korea, Renault Samsung, Ssangyong and Tata Daewoo. Downstream firms are suppliers who produce parts/components such as airbags and brake pedals and sell them to downstream firms. Linkages are transactions or partnerships between upstream firms and downstream firms. Linkages are the most important element in the supply chain since information flows via linkages. The first approach is to define network measures. Network implies how infor- mation flows via linkages and thus, defining networks is related to the neighbor- hood concept. I use two categories to define network measures: (1) information flows between upstream and downstream firms and (2) information flows among upstream firms. Degree with length 1 and exclusiveness capture how information flows between upstream firms and downstream firms via the quantity and quality of linkages. Degree with length 2 and the number of competitors measure information among upstream firms who produce the same component/part. The second is the graph spectra approach. Networks structures can be defined by adjacency matrices by the graph spectra. The distance measures, which I use, are the eigenvalue approach of the Laplacian matrix. The Laplacian matrix is defined as the difference between the degree matrix and the adjacency matrix. To test stability, I calcu- late eigenvalues of the normalized Laplacian matrix of two network structures in 2008 and in 2013. Then the distance measures can be calculated. I found the evidence that the supply chain networks in the Korean automobile industry are very stable. 40 The rest of the Chapter is organized as follows. Section 3.2 describes data sets. Sec- tion 3.3 defines network measures and present summary statistics. Section 3.4 explains the eigenvalue approach to test the stability of the supply chain. Section 3.5 concludes. 3.2 Data The data set used in this Chapter is the network data set. The network information was gathered by the Korean government for the entire automotive parts suppliers in Korea in 2008 and 2013. The data set identifies who the partner (partners) is for each supplier. In this Chapter, suppliers are defined as the first-tier supplier in the automo- bile industry. This is because of difficulties of capturing the second-tier suppliers. There are more than 2000 second-tier suppliers who are producing intermediate goods. The second-tier suppliers are selling products to not only automobile parts suppliers but also producers in other industries. In 2008, there are 892 automotive parts suppliers and they are selling their parts or components to 6 automobile assemblies (Hyundai, Kia, GM Korea, Renault Samsung, Ssangyong, and Tata Daewoo). In 2013, there are 898 suppliers and 6 automobile assem- blies. My interest is the linkages between automotive parts suppliers and automobile assemblies. In Chapter 2, I examined how complex the network structure is and it was shows in Figure 2.1. The question becomes how to simplify the complexity of the supply chain to understand the structure. Another way to think of the structure is to categorize parts suppliers by partnerships. Figure 3.1 shows how suppliers are distributed with respect to partnership structures. 41 Figure 3.1: Number of Suppliers with respect to Partnerships 42 The number within parentheses refers to how many suppliers exist in the group. For example, A(25) implies 25 suppliers having a partnership with Hyundai. In the case of DBA(90), there are 90 suppliers linked to both Hyundai and Kia. Similarly, SEXTA(6) shows that there are 6 suppliers selling their components to all six automobile assem- blies. It is natural to consider the differences of 22 groups by their linkage. TRPA(31) connecting with Hyundai, Kia and GM may be different from A(25), B(7) and C(62), which have a single linkage with automobile assemblies. The component-related information is combined to calculate network measures in Section 3.3. The component-related data directly shows which parts are produced by 480 suppliers. The number of competitors can be defined as the number of suppli- ers who produce the same part/component. The same part is considered based on 378 components. The degree with length 2 is the number of suppliers who could be reached within 2 walks. The degree with length 2 is also calculated to produce the same com- ponent and it is based on 378 component categories. 3.3 Network Measures How to measure networks is one of the fundamental issues to estimate the network effects in the supply chain. Network measures are defined to estimate to effects: how information flows between upstream firms and downstream firms via their partner- ships and between upstream firms who produce the same component. Table 3.1 summarizes networks measures of the Korean automobile industry. 43 Table 3.1: Summary Statistics of Network Measures 44 3.3.1 Information between upstream and downstream firms In this subsection, degree with length 1 and exclusiveness are defined to measure network effects between upstream firms and downstream firms. 3.3.1.1 Degree with length 1 The degree with length 1 is defined as the number of firms which can be reached within one walk. Jackson (2008) defines the degree of a node as the number of links that involve that node, which is the cardinality of the node’s neighborhood. d i (g) = #fj :g ji = 1g = #N i (g) The degree with length 1 can be interpreted as the number of partnerships because automobile assemblies are the firms which automotive parts can reach within one walk. The sample contains only first-tier suppliers; second-tier suppliers are excluded from the data set. There are 6 automobile assemblies: Hyundai, Kia, GM Korea, Renault Samsung, SSangyong and TaTa Deaewoo. Degree with length 1 can be defined in two ways: (A) Hyundai and Kia are two separate firms and (B) Hyundai and Kia are considered as a company. Table 3.1 shows that the mean of Degree with length 1 (A) is 2.63 and the standard deviation is 1.34. The mean with length 1 (B) is 2.01 and its standard deviation is 1.12. The degree with length 1 would capture the behavior of parts suppliers with respect to information with automobile assemblies. Information spreads via linkages between upstream firms and downstream firms in the supply chain. 45 The direction of information is not one way. From downstream firms to upstream firms, automobile assemblies can directly provide parts suppliers with technology- related information to maintain their vehicles’ qualities. As 65% of R&D investments are conducted by automobile assemblies, the amount of information flow from down- stream firms to upstream firms would not be negligible. The other direction is from upstream firms to downstream firms. Suppliers know more about detailed processes of how their components are produced in their fac- tory. Suppliers’ technology-specific information developed by their innovative activi- ties would also flow to automobile assemblies as the parts/components are sold. Asym- metric information issues are related with unbalanced information flows and Toyota’s brake pedal scandal is an example. 3.3.1.2 Exclusiveness Exclusiveness is defined as the suppliers with only one partnership with an automo- bile assembly. e i (g) = 1 ifd i (g) = 1 = 0 otherwise The data set shows that 45% of the suppliers are exclusive and 31% of suppliers have an exclusive partnership with Hyundai-Kia. 11 % of supplier have an exclusive partnership with GM Korea and there are only 3% of suppliers who have an exclusive contract with other three assemblies (Renault Samsung, Ssangyong, and Tata Daewoo). Exclusiveness is an important factor to determine the level of transferability of tech- nology. Exclusiveness is often regarded as a part of contract to prevent automotive parts suppliers from selling parts/components to other automobile assemblies. This is 46 because technology specific information is more likely to be developed by downstream firms and they do not want to spread their information to their rivals. 3.3.2 Information flows among upstream firms 3.3.2.1 Degree with length 2 The degree with length 2 is the number of firms which can be reached with two walks. Mathematically, I can write it with the concept of an extended neighborhood as follows: d 2 i (g) = #N 2 i (g) N 2 i (g) =[ j2N i (g) N j (g) The degree with length 2 is a variable to capture the effect of how information flows via linkage among automotive parts suppliers. The measure focuses on the network effect that sensitive information spreads to competitors via supplier’s partnerships. The degree with length 2 is a similar concept with the number of competitors which will be used as the network measure as well. However, the degree with length 2 is smaller than or equal to the number of competitors since the degree is the number of competitors which are connected with the supplier’s partners. Competitors are defined as the firms who produce the same component. Because 73.5% of suppliers produce more than one component, both the degree with length 2 and the number of competitors are likely to be measured more than once. To handle this case, I use minimum, average and maximum to define the number of 47 competitors and the degree with length 2. The summary statistics shows that minimum of degree with length 2 is 3.29, maximum is 9.30 and average is 5.72 respectively. 3.3.2.2 Number of Competitors The number of competitors is defined as the number of suppliers who produce the same component. As I define in the degree with length 2, the same component implies one of 378 components. Since many suppliers produce more than one component, min- imum, maximum and average are used. The minimum of the number of competitors is 4.06 and its maximum is 10.18. The mean is equal to 6.63. The number of competitors can be a proxy of the toughness of the price competi- tion. The entrants are less likely to enter the market when price competition is tougher. The toughness of the price competition is an important variable but a difficult factor to observe in empirical analysis. The number of competitors and the degree with length 2 are similar concepts but the difference is if the identity of partnerships plays a role to determine the real competitors or not. 3.3.2.3 Examples Suppose that there are 9 auto parts suppliers for three components. The networks structures are as follows. 48 Figure 3.2: Example Supplier 1 sells Component 1, 2, and 3 to Hyundai-Kia, GM Korea and Renault Samsung and the degree with length 1 is equal to 3. The number of competitors is 4, 2 and 4 respectively and then, the minimum, maximum and average measures are 2, 4, and 3. The degree with length 2 is calculated by the number of suppliers who are reachable with 2 walks. For Component 1, Supplier 1 can reach Supplier 2, 3 and 4 through Hyundai-Kia, GM Korea and Renault Samsung. Table 3.2 summarizes the 49 values of networks measures for the example. Supplier 3 and 5 have the same number of competitors but have a different values of the degree with length 2. Table 3.2: Network Measures (Example) Measures Supplier 1 2 3 4 5 6 7 8 9 Degree 1 3 1 3 3 2 5 1 1 3 Degree 2 Min 1 2 4 4 2 2 2 2 4 Max 3 3 4 4 3 2 2 2 4 Avr 2 2.5 4 4 2.5 2 2 2 4 Number of competitor Min 2 2 4 4 4 2 4 4 4 Max 4 4 4 4 4 2 4 4 4 Avr 3.3 3 4 4 4 2 4 4 4 Number of components 3 2 1 1 2 1 1 1 1 Exclusiveness General 0 1 0 0 0 0 1 1 0 Hyundai- Kia 0 1 0 0 0 0 0 0 0 GM Korea 0 0 0 0 0 0 0 0 0 Others 0 0 0 0 0 0 1 1 0 3.4 Stability of the Supply Chain In this section, I would like to compare the 2008 network structure with the 2013 network structure. The main objective of this section is to test the stability of the net- works. I examine the stability by various ways. I checked node degree distribution and compared the 2008 network structure with the 2013 network structure. The second 50 approach used in this section is based on the distance measure derived by the eigen- values of the Laplacian matrices. I also show the legal case to show the stability of the network structures in the Korean automobile industry. 3.4.1 Network Structures in 2008 and 2013 To understand the differences of the 2008 network and the 2013 network, I take three approaches. The first approach is to draw node distribution in 2008 and 2013, respectively. The second approach is summarizing the change of the partnerships for 5 years and analyzing the change with respect to the identity of the partnership. The third approach is using the network structure which has been used in Chapter 2. Thus, I will divide suppliers into four groups based on exclusiveness/non-exclusiveness and Hyundai-Kia/Non Hyundai-Kia and how the network changes from 2008 to 2013. Figure 3.3 describes node degree distributions in 2008 and 2013. The left hand side is drawn by the 2008 network data set and the right hand side is drawn by the 2013 network data. The distributions of the 2008 network and the 2013 network have a lot of similarities. Network structures are observed in 2008 and in 2013. A direct measure of the stabil- ity of the supply chain is a comparison between partnership structures of 643 matched pairs. Table 3.3 shows the comparison between the 2008 network and the 2013 network. Among 643 suppliers, 499 suppliers (77.60%) have exactly the same partnerships. 87 suppliers (13.53 %) have one more partnership in 2013 compared to 2008 and 45 suppli- ers (7%) lost one partnership in 2013. The right side of Table 3.3 explains who the identity of the change is. There are only 9 suppliers who had a partnership with Hyundai in 2008 but lost it in 2013. As a partner of Hyundai, 15 new suppliers are added. If Hyundai and Kia are considered as one merged company, the change is even smaller. The biggest change came from 51 Figure 3.3: Node Degree Distributions Renault Samsung. There are 41 suppliers who have a new partnership with Renault Samsung in 2013. Other three (GM Korea, Ssangyong and Tata Daewoo) has relatively more changes than Hyundai-Kia does. Hyundai-Kia is the most stable partner since there are only 5 new suppliers and 8 suppliers who stopped partnering with Hyundai-Kia. The results are consistent with 52 what I found in Chapter 2. In Chapter 2, I divide suppliers into two groups: Hyundai- Kia and Non Hyundai-Kia because of the foreign ownership and market share. The network patterns support that two groups behave in different ways and partnership structures are more stable when suppliers have partnerships with Hyundai-Kia. Table 3.4 shows the number of suppliers in each segment. I use the same four seg- ments which I used in Chapter 2. The four segments are divided by exclusiveness/non- exclusiveness and Hyundai-Kia/Non Hyundai-Kia. The number of suppliers in the Exc-HK segment decreased from 199 to 166 (by 33) and the number of suppliers in the NonExc-HK segment increased from 204 to 210 (by 6). In the case of Non-HK segment, the changes are much bigger. Suppliers in the Exc-NonHK segment increased from 303 to 414 and NonExc-NonHK increased from 88 to 109. The results show that suppliers who have partnerships with Hyundai-Kia are less likely to change partnerships. In addition, Table 3.4 provides the evidence that the Non- HK segment has a lowest level of entry barriers because entrants may enter the market through Non Hyundai-Kia as their first partner. 53 Table 3.3: Comparison between 2008 and 2013 Partnership Structures Table 3.4: Observations in Network Segments 54 3.4.2 Eigenvalue Approach The distance measures based on the graph spectra to calculate the stability (Lin, 1991; Jurman et al, 2011; Pincombe, 2007; Banerjee, 2009) are considered to be more sophisticated measures. 3.4.2.1 Definition Network structures can be written by the adjacency matrix. The adjacency matrix is a matrix to specify the network structure. In this case, the matrix is N+6 by N+6 where N is the number of automotive parts suppliers and 6 is the number of the automobile assemblies. The entry, a ij is equal to 1 if the supplier i has a partnership with j and zero otherwise. The adjacency matrix is symmetric because I only include the first-tier suppliers. The adjacency matrix,A N+6;N+6 can be written as follows: A N+6;N+6 = 0 B B B B B B B @ a 1;1 a 1;2 a 1;N+6 a 2;1 a 2;2 a 2;N+6 . . . . . . . . . . . . a N+6;1 a N+6;2 a N+6;N+6 1 C C C C C C C A 55 A N+6;N+6 = 0 B B B B B B B B B B B B B B @ 0 0 a 1;N+1 a 1;N+6 . . . . . . . . . . . . . . . . . . 0 0 a N;N+1 a N;N+6 a N+1;1 a N+1;N 0 0 . . . . . . . . . . . . . . . . . . a N+6;1 a N+6;N 0 0 1 C C C C C C C C C C C C C C A The degree matrixD N+6;N+6 is the diagonal matrix with the vertex degree and can be defined as follows: D N+6;N+6 = 0 B B B B B B B B B B B B B B @ deg(v 1 ) 0 0 0 0 0 deg(v 2 ) 0 . . . . . . 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0 0 0 deg(v N+5 ) 0 0 0 0 0 deg(v N+6 ) 1 C C C C C C C C C C C C C C A Entries on the diagonal are the degree with length 1 and entries off the diagonal are zero. The Laplacian matrixL N+6;N+6 is the difference between the degree matrix, D N+6;N+6 , and the adjacency matrix,A N+6;N+6 . The Laplacian matrix,L N+6;N+6 is: L N+6;N+6 =D N+6;N+6 A N+6;N+6 56 The entry of the Laplacian matrix,l i;j 1 can be written as follows: l i;j = 8 > > > > > > < > > > > > > : deg(v i ); ifi =j 1; ifi6=j 0; otherwise Distance measures can be computed by the eigenvalues of the normalized Laplacian matrix. The normalized Laplacian matrix,L , is: L N+6;N+6 =D 1=2 LD 1=2 =ID 1=2 AD 1=2 l i;j = 8 > > > > > > < > > > > > > : 1; ifi =j anddeg(v i )6= 0 1 p deg i deg j ; ifij is an edge 0; otherwise whereL N+6;N+6 = (l i;j ) The next step is to calculate eigenvalues of the normalized Laplacian matrix. All eigenvalues are between 0 and 2. 3.4.2.2 Distance Measures The first measure (M1) is used as an intra-graph measure to evaluate changes in the time-series of graphs. Two graphs, G and H, with N nodes have the following 1 LN+6;N+6 = (li;j )N+6;N+6 57 eigenvalues:f 0 1 ::: N1 g andf 0 1 ::: N1 g, respectively. For an integer k, the distance is defined as follows: d k (G;H) = 8 > > > > > < > > > > > : r P N1 i=Nk ( i i ) 2 P N1 i=Nk 2 i if P N1 i=Nk 2 i P N1 i=Nk 2 i r P N1 i=Nk ( i i ) 2 P N1 i=Nk 2 i if P N1 i=Nk 2 i P N1 i=Nk 2 i The M1 measure requires the same number of nodes and thus, the matched pairs of 2008 networks and 2013 network are used. The M1 measure is non-negative, symmetric, and satisfies the triangle inequality. According to the calculation, the M1 is equal to 0.069. To determine the level of stability, I compare the values with the results from Jurman et al. (2011). In their experiment, the measure 2 between a random network, A withA 5 which modifies 5% of nodes is 0:977 0:076 when the number of nodes is equal to 100. The comparison shows that the distance between 2008 networks and 2013 networks is very small. The second measure is the Jensen-Shannon measure based on the Kullback-Leibler divergence measure. Since the Kullback-Leibler divergence measure is not symmetric and does not satisfy the triangle inequality, the Jensen-Shannon measure was intro- ducted (Banerjee, 2009). The Kullback-Liebler divergence measure and the Jensen- Shannon measure can be defined as: KL(p1;p1) = X x2X p a (x)log p 1 (x) p 2 (x) : 2 In their paper, my first measure is equal to D1 and the second measure, the Jensen-Shannon measure is D6 58 JS(p1;p2) = 1 2 KL(p 1 ; p 1 +p 2 2 ) + 1 2 KL(p 2 ; p 1 +p 2 2 ) p 1 ;p 2 are two probability distributions of the random variable X. Using the spectral probability distribution, f of the normalized Laplacian, the distance can be defined as: d(G;H) = p JS(f G ;f H ) I compute the Jensen-Shannon measures for the matched 649 pairs as well as the unmatched two networks (800 nodes vs. 904 nodes). The Jensen-Shannon measures for matched 649 pairs and unmatched two networks (800 nodes vs. 904 nodes) are 0.034 and 0.040, respectively. The Jensen-Shannon measure for two networks, a random net- work, A with A 5 which modifies 5% of nodes is 1:102 0:074. The Jensen-Shannon measure also provides us with the evidence of stability of networks in the Korean auto- mobile industry. Three different approaches consistently conclude that the supply chain networks in the Korean automobile industry are very stable. Table 3.5: Distance Measures 59 3.4.3 Evidence of Stability: Litigation The case regarding the cartel between wiper system suppliers in January 2014 pro- vides us with evidence to show how the structures of the supply chain are stable and how information spreads to the competitors. The Korean Fair Trade Commission (FTC) penalized illegal restraints of trade and monopolies with $16.5 billion of fines for Denso Corporation, Denso Korea automotive and Bosch Korea. Denso and Bosch compete in the auction of wiper systems for several models of Hyundai and Kia including Sonata and Genesis. The level of price competition increased enormously in the wiper system market after wiper producers began making products that were all of the same qual- ity. In this circumstance, Denso and Bosch agreed to predetermine the winner of each model before the auction in order to maintain a high price for wiper systems, which violates the antitrust law. There are two implications from this case. First, partnership structures are very stable over time even if the auction-based system for each vehi- cle model changes the share of wiper systems for each automobile assembler. Second, there exists a high level of price competition among suppliers. Third, suppliers have full information about the price and technology of their competitors, which is direct evidence that information flows via the supply chain structure. 3.5 Conclusion In this Chapter, I empirically analyzed network structures in the Korean automobile industry and test the stability of the networks of the supply chain. Observed network structures in 2008 and in 2013 60 First, I define various network measures focusing on two information flows: (1) flows between upstream firms and downstream firms and (2) flows among upstream firms. To capture the information flows between upstream firms and downstream firms, I define degree with length 1 and exclusiveness. Degree with length 2 and the num- ber of competitors measure information among upstream firms who produce the same component/part. These network variables will be used in the next Chapter as main variables to estimate the network effects. Second, I test the stability of the network structures. By the graph spectra approach, networks structures can be defined by adjacency matrices. The distance measures, which I used in this Chapter, are based on the eigenvalue approach of the Laplacian matrix. I calculate the M1 measure and Jensen-Shannon measure and found the evi- dence that the supply chain networks in the Korean automobile industry are signifi- cantly stable. The node distribution in 2008 and 2013 and summary statistics of changed partnerships support the evidence that the supply chain is stable. The underlying concept of the Jensen-Shannon measure is the differences between eigenvalues of the two matrices. To test the sustainability in a rigorous way, I plan to apply the recently suggested methods (Peel and Clauset, 2014; Asta and Shalizi,2014). The pair-wise stability is an importance concept in the theory. However, testing the concept of the stability is still limited in the empirical analysis. This Chapter may suggest a way to think of testing the stability of the network data sets in a dynamic setting. 61 Chapter 4 Estimation of Network Effects on Automotive Parts Suppliers’ R&D Investments: the Fixed Effects Filtered Estimation 4.1 Introduction 4.1.1 Motivation It is widely known that private R&D spending is lower than the optimal level because of the nature of public goods. There are various government policies includ- ing subsidies to encourage firms to increase private R&D spending. However, it is less clear when the complexity of the supply chain is introduced. In the supply chain set- ting, each firm’s R&D investment decision is not only related to the firm’s innovation but also related to the partner’s R&D. The main goal of this chapter is to estimate the network effects on the investments committed by the Korean automotive parts suppliers to research and development (R&D). The question is as follows: Do suppliers change their R&D behavior based on network structures? 62 One of difficulties in network analysis is data availability. The data set is constructed by three different data sets: network data, accounting data, and component-related information data. Using the network data, I could reveal the linkages between 800 first-tier automotive parts suppliers and six automobile assemblies (Hyundia, Kia, GM Korea, Renault Samsung, Ssangyong and Tata Daewoo). The uniqueness of the data set allows me to define network measures. Network measures are defined to capture the effects of quantity/quality of linkages. The supply chain networks consists of downstream firms, upstream firms and linkages between them. I focus on the quantity and quality of the linkages. The identity of part- nerships (Hyundai-Kia and non Hyundai-Kia), and the exclusiveness (having only one partnership) are measures of The quality of linkages. Similarly, the number of partner- ships and the number of competitors are the qauntity of linkages. The econometric goal of this chapter is to estimate the effects of both time-variant and time-invariant regressors consistently. Because the supply chain networks do not change over time in the Korean automobile industry, network measures are time- invariant. When regressors and unobservable firm-heterogeneity are correlated, the results of OLS are biased/inconsistent. Although the Fixed Effects estimator is consis- tent, the time-demeaning procedure eliminates time-variant regressors. As a solution, I apply the Fixed Effects Filtered estimator (Pesaran and Zhou, 2013). 4.1.2 Connections with Bain’s Paradigm Traditionally, Industrial Organization literature has focused on the Bain Paradigm (1956), a chain between Structure (the level of concentration), Conduct (the degree of collusion) and Performance (Profitability). There have been numerous criticisms of the one-way flow from Structure to Conduct, and from Conduct to Performance. It is more reasonable to consider three elements based on a circular relationship. Figure 4.1(a) and 63 Figure 4.1 (b) graphically show the difference between the Bain Paradigm and the cir- cular relationship. In this Chapter, I would like to revisit three elements from the Bain Paradigm to estimate the effects of Conduct and Performance on the Structure control- ling all other directions within one specific industry, the Korean automobile industry. Figure 4.1: Bain’s Paradigm and Circular Relationship ”Structure” is barriers to entry and variables which are relatively stable over time and affect the behavior of sellers and buyers. Sutton mentioned that R&D intensity and advertising expenditure are appropriate variables for structure. Conduct is the level of collusion which represents firm’s strategic behavior. In this Chapter, the number of partnerships, the number of competitors, the characteristics of component, and the number of component are able to be interpreted as Conduct as well as Structure. Performance is the level of production efficiency which is clearer. The profit, debt, current asset, total capital, current capital and current liability indicate firm’s Perfor- mance. Since logarithm cannot be applied due to many of zero R&D expenditure, all performance measures are scaled by the total revenue or the number of employees. 64 As Figure 4.1 (b) shows, it is a typical simultaneous equations problem. Thus, one of the most important problems is how to deal with the endogeneity. If the error term were correlated with regressors, the OLS results would be inconsistent/biased. The rest of the Chapter organized as follows: Section 4.2 discusses Data and Section 4.3 explains variables which I included in the regression analysis. Section 4.4 shows empirical analysis and Section 4.5 presents results. Section 4.6 shows further issues and Section 4.7 concludes. 4.2 Data The panel data used in this Chapter have been constructed from three sources of data sets by matching automotive parts suppliers’ identity. The panel data consist of 334 first-tier automotive parts suppliers from 1999 to 2010. Thus, the total number of observations is 4008 (the balanced panel). Three data sets come from the Korean gov- ernment and I use three for a different purpose. The first data set is used for the net- work structures, the second is for performance measures and the third data set shows component-related information and is used to construct variables including the number of competitors and characteristics of components. The first data set allows me to define network related variables such as the degree with length 1 (the number of partnerships), the degree with length 2 and the number of competitors. To calculate the degree with length 2 and the number of competitors, the component-related information has been used. The data set shows the linkages between 892 automotive parts suppliers and 6 automobile assembly companies in 2008. It implies that I can identify who the partner (partners) is for each supplier. A percent- age share for each automobile assembly company for each automotive parts supplier can be identified but is not used in this Chapter. 65 The second data set contains 480 automotive parts suppliers from 1999 to 2010. The data set has been constructed from annual financial reports using the DART (Data Anal- ysis, Retrieval and Transfer) system of Financial Supervisory Service. Filing disclosure documents on main events is mandatory for listed corporations in the KRX or the KOS- DAQ or for companies whose capital assets exceed $7 million based on the previous year. Financial reports consist of the total revenue, profit, total liability, equity, capi- tal, current assets, current liability, R&D investment, and number of employees. Those variables could be considered as the variables to capture the profitability. In the esti- mation procedure, they are used to estimate the effects of profitability as well as their ability related to profitability. The third data set is component-related information. The data directly show which parts are produced by 480 suppliers. Since there are more than 30,000 components, the information is grouped into 378 different components. The data set is used for three purposes. First, component characteristics are used as control variables after mapping into two spaces. 378 parts are still too many in order to control the characteristics of components. I projected 378 parts on two different spaces. The first space is categorized by materials and the second space focuses on the use of the automotive parts in the vehicle. The first space contains 21 categories and the second does 11. For example, I can map oil pump as “pump” in the first standards of categorization and “power generating and transfer system” by the second standards as Figure * shows. Second, the component-related information is to calculate the number of competi- tors and the degree with length 2 by combining with the first data set. I define the number of competitors as the suppliers who produce the same part/component. The same part is considered based on 378 components. The degree with length 2 is the num- ber of suppliers who could be reached within 2 walks. The degree with length 2 is also calculated to produce the same component and it is based on 378 component categories. 66 Third, I can define the number of components that each supplier produces. Th data show that the mean of the number of components is 4.09 and the standard deviation is 4.38. The limitation is that we are unable to capture the financial variables such as profit by each part when a supplier produces more than one automotive part. During the process to combine the first and second data set, I exclude parts suppliers who sell their parts to the other parts suppliers. In other words, the sample includes only first- tier suppliers who have partnerships with automobile assemblies. Lee (2010) said that there are more than 2000 second-tier suppliers but it is difficult to observe them by the data set. This is because the customers of second-tier suppliers vary across the industry. Products of second-tier suppliers are raw or intermediate material of automotive parts as well as materials in other industries. Table 4.1: Number of Birth/Death After matching for three data sets, my data set consists of 334 first-tier automotive parts suppliers from 1999 to 2010. In the firm analysis, firms’ birth and death are one of the most important issues. In the automotive industry, it seems that the birth and death of suppliers would not be a problem. Table 4.1. shows the number of suppliers who enter and exit each year, and I have already excluded firms that do not sell parts to the automobile assembly companies. In the empirical analysis, I used the balanced panel data. The number of firms is 334 since it is calculated by (i)-(ii)-(iii)+(iv)-(v) in Table 4.2. 67 Table 4.2: Number of Firms in the Sample 4.3 Empirical Analysis 4.3.1 Econometric Issues I would like to estimate the effects of network measures on automotive parts sup- pliers’ R&D decisions. The static model is constructed as follows: y it = i +x it 0 +z i 0 +u it where i is the automotive part suppliers, i=1,2,...,N;t is year,t=1,2,...,T. The dependent variable,y it is each supplier’s R&D expenditure for individual firm i and time t. x it s are time-variant variables which represent the Performance in the Bain’s paradigm: Profit, debt, current asset, total capital, current capital and current liability. z i s are time-invariant variables: the degree with length 1, the degree with length 2, the number of competitors, the number of components,exclusiveness, the identity of the partnership, the characteristics of component, union and areas. Summary statistics of dependent variable, time-variant regressors and time-invariant regressors are given in Table 13. i is defined as i = + i and the unobservable firm-specific heterogeneity which does not vary over time. In general, T is small, the pooled OLS is biased and inconsis- tent (Hsiao, 2003). If i is correlated with regressors, POLS is biased and inconsistent. 68 The Fixed Effects estimator is still consistent when i is correlated with exogenous vari- ables,x it . However, the weakness of the Fixed Effects estimator is that the coefficients of time- invariant regressors are not estimable. The time-invariant variables are eliminated by the demeaning procedure. This is a serious problem since all the network measures are time-invariant due to the stability of the supply chain. The econometric question is the following: how can I estimate both time variant and time-invariant regressors consistently even if there exists the correlation between regressors and firm-specific heterogeneity? Hausman and Taylor estimator is a remedy to use the concept of the instrumental variables but cannot be applied to my study since the restrictions are not satisfied. I will consider the Fixed Effects Filtered (FEF) estimator in this Chapter which is shown to work well under various scenarios by Pesaran and Zhou (2013) 1 . 4.3.2 Fixed Effects Filtered Estimator To estimate the coefficient of time-invariant variables consistently even if the firm- specific heterogeneity is correlated with exogenous variables, I apply the Fixed Effects Filtered estimator. In this subsection, I discuss the fixed effects filtered estimator. The underlying idea of FEF is straightforward and can be implemented in two step. The first step of FEF estimator is to apply the FE estimation to get an estimator of, which is consistent when N is large. In the second, the FEF estimator uses the residuals from the first step as explained variable and run OLS to obtain an estimator of . Pesaran and Zhou (2013) show that this FEF estimator is consistent ifz i is uncorrelated with i and 1 There is also so called Fixed Effects Vector Decomposition (FEVD) in the political science, as shown by Pesaran and Zhou (2013), this FEVD estimator is identical to FEF estimator if an intercept is included in the regressor, otherwise it is inconsistent in general. Refer to Pesaran and Zhou (2013) and Plumber and Troeger (2007) for more details. 69 u it . This FEF estimator works well in various situations including heteroskedasiticity and serial correlation. Moreover, a corresponding variance estimator of is also pro- posed in their paper. Formally, for the FEF estimator, can be estimated by usual FE estimator, denoted by ^ FE and then calculate the residual as e it =y it x it 0 ^ FE ; By averaging over t, I have the following: e i = 1 T T X t=1 e it ; Then the FEF estimator of can be estimated by using OLS to the following model: e i = +z i 0 + i + u i The covariance matrix is ^ FEF = ( N X i=1 (z i z)(z i z) 0 ) 1 N X i=1 (z i z)( e i e); 70 On the other hand, even if z it is correlated with i andu it and given the existence of valid instruments, the FEF estimator can be valid. 2 2 It should be noted that it could be very difficult to find valid instruments in the firm level data and so, this Chapter does not consider the instrumental variables approach for eample, Hausman and Taylor (1981)and FEF-IV(Pesaran and Zhou (2013)) of estimating . 71 Table 4.3: Summary Statistics 72 4.3.3 Dependent Variable The dependent variable,y it is each supplier’s R&D expenditure for individual firm i and time t. y it = J X j=1 K X k=1 y k ijt j is the automobile company; k is the automotive part. To control the size effects, R&D expenditure is divided by two variables respectively: the total revenue or the number of employees. In the previous literature, taking a logarithm is common to adjust the size effect. Since 70% of R&D spending is zero, the logarithm could not be used. The mean of R&D is 737.74 and R&D divided by total revenue (R&D intensity) is 0.69. R&D labor ratio is 1.61. 4.3.4 Time-variant Regressors Time-variant regressors are Profit, Debt, Current asset, total capital, current capital, and current liability. Those are measures of ”Performance”. In the case of profit, the pos- itive coefficient is more plausible. The correlation with i is more likely to be positive since higher productivity induces higher profit. The effects of time-variant regressors are more straightforward than time-invariant regressors. The Fixed Effects estimator will provide us with consistent estimators even if the firm-specific heterogeneity is cor- related with those regressors. The mean of profit is 7524.01 and its standard error is equal to 52723.44. As profit increases, does supplier invest in R&D more? However, the opposite direction is plau- sible because innovators who invest in R&D have a higher profit. To deal with the 73 simultaneous equation problem, the IV is one possible way. When the unobservable hetereogeneity such as owner’s ability is correlated with profit, the OLS is biased. In this case, the fixed effects procedure could be a solution. 4.3.5 Time-invariant Regressors 4.3.5.1 Degree with length 1 The definition of the degree with length 1 is the number of firms which can be reached within one walk. In other words, it implies the number of partnerships that each supplier has. As I explained in the Chapter 3, I consider two cases: (A) Hyundai and Kia are two separate companies and (B) Hyundai and Kia are a merged company. It is possible to expect both positive and negative effects on R&D investments. The positive effects implies that the more partnerships parts suppliers have, the more R&D they invest in. Similarly, the negative sign shows that suppliers invest less in R&D as they have more partnerships. Positive effects may imply that suppliers risk-averse toward shocks in automobile assemblies. Having more partnerships helps parts suppliers to diversify the automobile assemblies-related risks. Suppose that there are two parts suppliers of air bags. Supplier 1 with partnerships with Hyundai and GM Korea and supplier 2 with a partnership only with Hyundai. When there is a negative shock on sales of Hyundai such as a recall scandal, a negative shock directly affects the sales of both parts suppliers. However, from a sense of risk management, supplier 1 is less likely to be out of business than supplier 2 by virtue of the latter’s ability to sell air bags to GM Korea. The environment in which sensitive information is not perfectly protected may result in negative effects of more partnerships on parts suppliers’ R&D. In other words, suppliers 74 have to face higher risks to spread their own technology to other competitors via their partner as the number of partnerships increases. Automobile assemblies sign a contract with suppliers not to share sensitive informa- tion with other suppliers. However, it is not easy for suppliers to argue their property rights to automobile assemblies when the failure of information protection occurs. Based on the characteristics of producer-drive supply chain, downstream firms have a power for contracts and negotiations in the Korean automotive industry. To parts sup- pliers, having continuous partnerships can be interpreted as the Nash Equilibrium of repeated games compared with the Nash Equilibrium of one-time game. The benefits of current period is smaller than the costs to have higher probabilities to close the rela- tionship with their partner and lose the main customer in the repeated game setting. Hence, the sign of coefficient of degree with length 1 on suppliers’ R&D is determined by which effects dominate. 4.3.5.2 Exclusiveness Exclusiveness can be defined as the suppliers with only one partnership with an automobile assembly. As the data shows, % of supplier have an exclusive contract. In Korea, exclusive contracts were a common and traditional way for automobile assemblies to protect their technology sensitive information from their rivals. However, it has been weaken and weaken. If suppliers were able to receive the information from the assembly, the sign of the exclusiveness on R&D could be negative. 4.3.5.3 Degree with Length 2 The degree with length 2 is the number of firms which can be reached with two walks. The measure is defined based on suppliers who produce the same component. A positive sign implies that suppliers tend to invest more in R&D when they have more connected company. It could be interpreted as a competition effect. A negative 75 sign means that suppliers are less likely to invest in R&D as the risk of spearing their information increases. 4.3.5.4 Number of Competitors The number of competitors is defined as the number of suppliers who produce the same component. The same component implies one of 378 components. Since many suppliers produce more than one component, minimum, maximum and average are used. The number of competitors can be a proxy of the toughness of the price competi- tion. The entrants are less likely to enter the market when price competition is tougher. The toughness of the price competition is an important variable but a difficult factor to observe in empirical analysis. The coefficient shows how suppliers respond to the R&D investment as the tough- ness of competition changes. Both directions of sign are plausible. The coefficient is negative when suppliers are less likely to invest in R&D since their new technology flows directly to their competitors through the arrows between automotive parts sup- pliers and automobile companies. The sign will be positive if the supplier tends to invest in R&D to survive in the highly competitive environment. Empirical analysis will show whether the spillover effects dominate the competition effect or not. The correlation between individual firm’s productivity and the number of competitors is related with the barriers to entry. If the incumbent sets the quality of the product at the higher level, entrants will not be able to get into the market. 76 4.3.6 Characteristics of Components Theory assumes homogeneous products but brake pedals and leather for seats have a different level of underlying technology. I map 378 components into two different spaces. The first space is defined by materials and the list is as follows: Agricultural machin- ery, aluminum, bearing, bolt/nut, casting, electric equipment, electronics, fabricated metal products, forging, glass, leather, mold, other steel products, plastic, pump, rub- ber, steel, textile, valve. The second space is based on the function of the automotive parts and the list is the following: Suspension, power generating system, steering, brake, electronics (batter- ies), parts, lamps, seats, wiper system, air-conditional system. I also define the technology oriented products and metal products. The technology oriented products include agricultural machinery, electric equipment, electronics and pump. The metal products consist of casting, forging, bearing, bolt/nut, fabricated metal product, other steel product and valve. 77 Figure 4.2: Input-Output 78 4.3.6.1 Union Union is a dummy variable. If the supplier joins the union of automotive parts suppliers, the variable is equal to 1. Otherwise, the variable is zero. The data show that 48.5% of suppliers join the union. If the union protects the right of the suppliers’ technology sensitive information, the sign will be positive. 4.3.6.2 Area Figure 4.3: Factories in South Korea 79 Area is defined by the location where the suppliers’ factory is. Figure 4.3 shows the location of the factories of 6 automobile companies. Area dummies are included to control the area specific effects. The geographical clustering could be considered because it is common to start the business close to the automobile assemblies to minimize the transaction costs. However, the small size of South Korea may mitigate the benefits of geographical clustering. The details will be discussed in the later section. As Figure 4.4 shows that suppliers distributed very differently with respect to the location. Figure 4.4: Number of Firms within Area 80 4.4 Results 4.4.1 Network Effects I start to examine the network effects on the R&D investments of automotive parts suppliers by the degree with length 1 and the degree with length 2. Table 4.4 shows the results of Pooled OLS (POLS) and the fixed effects filtered (FEF) estimator with respect to different measures of the degree. Table 4.4 focuses on the network measures. The other coefficients are discussed in Table 4.6 3 . Hyundai and Kia are dealt with as two separate companies in columns (1) and (2) and are considered as one merged company to suppliers in Columns (3) and (4). Regardless of the definition of the degree with length 1, the coefficient of the degree with length 1 is positive and significant at the 1% level in both POLS specifications (Columns (1) and (3)) but not significant in the FEF specification (Columns (2) and (4)). Since the degree with length 1 is the number of partnerships, a positive sign indi- cates that the more partnerships suppliers have, the greater incentives they have to invest in R&D. The size of the coefficient of the degree with length 1 increases as I con- sider Hyundai and Kia as a merged firm. This is a natural result as the variation of the number of partnerships decreases. 3 Tobit (Cohen and Levinthal, 1989) is one way to focus on zero R&D problems since 70% of suppliers have zero R&D. The signs of main regressors are the same with the results that the POLS and the FEF give. 81 Table 4.4: Estimation results: Network effects (A) 82 Table 4.5: Estimation results: Network effects (B) 83 The degree with length 2 is the number of firms which suppliers can reach with two walks and has been defined for each component. Because many suppliers produce mul- tiple components, the measures of the degree with length 2 are defined for minimum, maximum and average. As Columns (3)-(8) show, the coefficients of the degree with length 2 vary from -0.011 to -0.38, and the level of significance differs with respect to which measure is used. In the case of minimum measures, Columns (3) and (4), it is significant at the 1% level in POLS and also significant at the 10% level in FEF. Negative effects of the degree with length 2 on R&D investment are interesting results. The degree with length 2 is the number of competitors which are connected with the supplier via the supplier’s part- ners and both directions are plausible: a positive sign implies the competition effects and a negative sign indicates the suppliers’ reaction to the information spillovers. I find a negative effect which suggests that the information effects overweight the competition effects in the automotive industry. Suppliers are less likely to invest in R&D when more competitors are connected. To suppliers, R&D investment is a huge amount of sunk costs. When the outcomes of R&D are not fully protected, having more competitors decreases the expected gains of R&D investment. Thus, it is easier for firms to be free riders instead of investing in their own R&D. The degree with length 2 can also be interpreted as an already existing level of entry barriers to each supplier. Some may worry about the endogeneity issues by the required level of technology as a source of barriers to entry. When the maximum measure is used (Columns (5) and (6)), the size of the coeffi- cient of the degree with length 2 is one third of Columns (3) and (4) in the case of the minimum measure and the significance drops. Suppliers tend to have more incentives to invest in R&D if they have fewer connected competitors and as multi-component producers, their decisions are more likely to react to the component with a higher mar- ket power. 84 The coefficients of the number of components are negative and significant at least 10% level even if the significance level varies across the specifications. The results sug- gest that suppliers who are specialized in a lower number of components tend to have higher incentives in R&D derived by the characteristics of R&D activities. Suppliers who participate in union are more likely to engage in R&D activities but it is not signif- icant in the FEF specifications. Columns (1)-(6) of Table 4.5 show the results with the number of competitors instead of the degree with length 2. The difference between the degree with length 2 and the number of competitors is whether I focus more on the linkage via partners or the num- ber, itself even if the firm is not connected via the partner. The results are consistent with Table 4.4. Columns (7) and (8) are the results when variables scaled by the number of employees are used. 4.4.2 Exclusive Dealing In this subsection, I estimate the effects of exclusive dealing arrangements. Exclu- sive dealing arrangements are important to automobile assemblies since there is a lower risk of spreading technologically sensitive information to the rivals by their suppliers. Knowledge is more likely to be transferable from automobile assemblies to automotive parts suppliers when the contract is exclusive. Table 17 explores the effects of exclu- siveness and who the exclusive partner is. Columns (1) and (2) show that the exclusive contracts have negative effects on suppliers’ R&D activities. It is significant at 5% in the POLS but not significant in the FEF specification. A negative sign ofexclusiveness shows that suppliers, who deal exclusively, have less incentives to invest in R&D. This can be interpreted as indirect evidence of knowledge transferability in exclusive dealing. If important knowledge about technology is already transferred from their partner, the best strategy for suppliers is to minimize the sunk costs. 85 Table 4.6: Estimation results: Network effects (C) 86 Table 4.7: Estimation results: Exclusiveness 87 Since Hyundai-Kia has distinguished characteristics from GM Korea, Renault- Samsung, Ssangyong, and Tata Daewoo in the R&D activities, I turn to the effects of the identity of the exclusive partner. Columns (3)-(6) reveal that not only exclusiveness matters but also the party with whom suppliers have an exclusive contract affects sup- pliers’ R&D investment decisions. I find that exclusiveness with Hyundai-Kia has a negative effect on supplier’s R&D activities while the exclusiveness with others (Renault-Samsung, Ssangyong, Tata Dae- woo) has a positive effect. The results provide indirect evidence that internal R&D and external R&D are substitutes in the consideration of exclusiveness. 73.35% of patents have been made by Hyundai and Kia and 25.8% by GM Korea 4 . The results are dif- ferent from existing literature (Belderbos et al, 2006; Mohnen and Roller, 2005) that internal and external R&D are complements by synergy effects. The results suggest that the internal R&D and external R&D can be substitutes if partnership structure has been considered. 4.5 Discussion 4.5.1 Geographical Distribution Geographical Clustering is an issue in the automobile industry because suppliers start their business in a place close to the partner. To examine the issue, Table 4.8 shows the number of partnerships and the identity of partnerships within each area. 4 Hyundai owns 36,304 of patents and Kia has 6,565. GM Korea has 11,566 while Ssangyong and Renault Samsung only have 267 and 112, respectively. 88 Table 4.8: Distribution of Networks within Area 89 Because of a small size of Korean peninsula, the benefit of geographical clustering such as minimizing transaction costs and shipping costs may be decreased. However, Table 4.8 describes that there exist geographical clustering. For example, Ulsan is the area that Hyundai has their big factory. In Ulsan, 20 suppliers among 21 have a part- nership with Hyundai. 4.5.2 Endogeneity Issues Empirical analysis has focused on how to deal with firm-specific heterogeneity with time-invariant regressors. As a remedy, FEF has been applied. The best part of FEF esti- mator is to be able to estimate time-variant regressors as well as time-invariant regres- sors without instrumental variables consistently. The assumption that FEF has is that time-invariant regressors are not correlated with firm-specific heterogeneity. If there were correlation, FEF-IV is the next step to apply. Sources of firm-specific heterogeneity would be personality and ability of owner/manager, firm’s productivity and unobserv- able technology which have not been controlled by component specific technology and the number of components each supplier produce. Time-invariant regressors used in the Chapter are as follows: degree with length 1, degree with length 2, the number of components, the characteristics of component, union and area. Degree with length 1 A concern is if the number of partnership is correlated with unobserved firm-specific heterogeneity or not. Some cooperation literature (who?) consider a dummy variable of cooperation (cooperate or not) and in-house R&D as a simultaneous equation problem. It may not be a concern because the data set is constrained by supply chain setting. The sample of this study only includes part suppliers who have at least one partnership. Identity of Partnership The results show that who the partner is more important than just the number. This argument is consistent with Fox (2010)’s setting which suppliers’ revenue function are 90 not additively separable. Degree with length 2 The number of suppliers which are connected by the supplier i’s partner. Less likely correlated with firm-specific heterogeneity: Determined by the other supplier and part- ner. The number of components Historically, firms tend not to change the product by the constraints of technology 4.5.3 Determinats of Exclusiveness: Probit Estimation In the previous subsection, I have considered the possibility of endogeneity. Espe- cially, the exclusiveness is a concern because exclusive contracts could be a choice vari- able to save the R&D expenditure. Table 4.9 shows the probit estimation. The depen- dent variable is the exclusiveness and I use R&D, total revenue, patents which suppli- ers’ partners (automobile assemblies) own, the dummy variable to establish the factory in the same area of the partner, the number of competitors, and the number of employ- ees. 91 Table 4.9: Estimation: Exclusiveness 92 4.5.3.1 R&D Investment All Column (1)-(4) shows that there is a negative effect of R&D investments on exclusiveness. The significance level changes with respect to the specification. As R&D investment increases, they are less likely to be exclusive. The significance level of Col- umn (1) and (4) shows the possibility of endogeneity problem. 4.5.3.2 Patents Patents is defined as the number of patents which supplier’s partner (assembly) owns. Since many suppliers have more than one partner, I used average, minimum and maximum. Column (1)-(3) shows a positive effect on the exclusiveness. 4.5.3.3 Same Area Same Area is a dummy variable. If the supplier has the factory in the same area of its partner’s factory, Same Area is equal to 1. Otherwise, the variable is zero. If the supplier has more than one partnerships, Same area is equal to one if at least one partner’s factory is in the same area. In the column (2), the coefficient of Same Area is -0.452 with 1% significance level. The sign is different from my expectation because I expect that suppliers may have more tendency to be exclusive when they are in the same area. 4.5.3.4 Number of Competitors 4.5.4 Other Issues The following is the possible issues which I consider. Family relationships There exist family relationships between parts suppliers and assemblies. For example, the owner of supplier is a brother in a law of Hyundai’s owner family. Some argue that it could be done by investigating all the suppliers. The government tried to capture the family tree but it is still limited. 93 Accounting variables Stock share by assemblies could be correlated with regressors. If the share does not change over time, the first stage of the fixed effects estimator eliminates the possibility. However, I have to construct stock share variables by annual financial reports for all suppliers if stock shares are changing over time. Subsidiary of Hyundai-Kia There are a few subsidiaries of Hyundai-Kia such as Hyundai Mobis. R&D of subsidies could be interpreted in a different way because of possibilities of money laundry. Sim- ply I can include a dummy variable. If the company is a subsidiary of Hyundai-Kia, the variable is equal to 1. Otherwise, it is zero. Another issue is that Hyudai-Kia do care about profits of their subsidiaries. If Suppliers have subsidiaries of Hyundai-Kia as a competitor, they may feel the pressure more than others who do not. I can include a dummy variable to have any of subsidiary of Hyundai-Kia as a competitor is equal to 1. Merged or not? Hyunai and Kia To define the degree with length 1, I considered both cases: the case that Hyundai and Kia are the same company and the case that they are separate two companies. Some may argue that legally merged firm should have the same brand name. According to Lee’s comments, dealing with Hyundai and Kia as the same firm is a reasonable argu- ment. When Hyudai took Kia during 1997 Asian financila crisis, they thought that the brand name of Kia is valuable and wanted to enjoy benefits from product differentia- tion. Hyundai-Kia should be considered as the same company because Hyudai has the right to determine the CEO/owner of Kia. 94 Chapter 5 Conclusion In this Chapter, I focus on the role of supply chain networks in automotive parts suppliers’ R&D decisions in the Korean automobile industry. I examine the hypothesis that stable supply chain networks are the main source to decrease automotive parts suppliers incentives to invest in R&D. The stability of the supply chain is based on the long-term relationship between automotive parts suppliers and automobile assemblies, especially, Hyundai and Kia. Before testing the hypothesis by panel data techniques, I investigate the importance of networks structures in data analysis and test the stability of the supply chain. First, I present dynamic patterns as evidence that networks structures have to be considered to interpret the theoretical implications from Sutton’s model. Without the network structures, patterns of the Korean automobile industry do not explain the insights from Sutton’s endogenous sunk costs model. As market size increased, the number of automotive parts remained almost the same from 1999 to 2010. However, the automotive parts market did not seem to be R&D intensive. To explain this contradic- tion, I focus on information flows from automobile assemblies to automotive parts sup- pliers based on the stable supply chain. Historically, the Korean government encour- aged automobile assemblies and automotive parts suppliers to be vertically integrated to maximize the efficiency. As a result, it was common that the Korean automobile assemblies provided blueprints of parts/components to their partners based on their long-term partnerships. However, it is not obvious to measure how knowledge is trans- ferable from automobile assemblies to suppliers in the real world. For a proxy of the level of knowledge transferability, I use two measures,exclusiveness and the partnership 95 with Hyundai-Kia and then define four different sectors: Exc-HK,Exc-NonHK,NonExc- HK and NonExc-NonHK. When networks structures are considered, consistent results with Sutton’s model are obtainable. Networks structures are important in the Korean automobile industry because of the stability of the supply chain. As the second step, I test the stability of supply chain networks empirically. I consider distance measures based on the graph spectra. I construct the adjacency matrices for 2008 and 2013. The 2008 adjacency matrix is 806 by 806 since there are 800 suppliers and 6 automobile assemblies. I define the degree matrices and then calculate the Laplacian matrices and normalize them. The Laplacian matrix is defined as a difference between the degree matrix and the adjacency matrix. Eigenvalues of the normalized Laplacian matrices are used to calculate the distance measures such as the Jensen-Shannon measure. The measures show that the supply chain networks of the Korean automobile industry are extremely stable over time. This is the first attempt to test the stability of the networks in Economics. Based on the importance and stability of the network structures, I estimate the net- work effects on automotive parts suppliers’ R&D investments using panel data tech- niques. I define various network measures: the degree with length 1, the degree with length 2,exclusiveness, the number of competitors. The coefficients of network measures are estimated by the Fixed Effects Filtered estimator. The FEF estimator is applied to estimate both time-variant and time-invariant regressors, consistently considering the firm-heterogeneity. When correlations exist between regressors and unobservable firm- heterogeneity, the pooled OLS estimator is inconsistent and biased. Although the fixed effects estimator is consistent, it cannot estimate the effects of time-invariant regressors. Since network structures are stable over time, the network measures are time invariant. I find that suppliers are more likely to decrease R&D investments as they have more competitors. The results imply that free-rider effects are greater than competi- tion effects. The effects of the number of partnerships are not obvious. One of the most 96 interesting results is that the internal R&D and external R&D are substitutes if the con- tract is exclusive. When the suppliers have an exclusive contract with Hyundai-Kia (the biggest R&D investor in Korea), suppliers have less incentives to invest in R&D. If sup- pliers had an exclusive contract with an automobile assemble who has a relatively low level of R&D (Renault-Samsung, Ssangyong, and Tata Daewoo). This study does not contain a direct way to test complementartiy and substitutabil- ity. Using the definition by Milgrom and Roberts (1990) 1 could be one way to test com- plementarity/substitutability. However, applying their methodology directly is diffi- cult because my data set contains multiple partnerships. The econometrical model I used in this article is static. At present, Kripfganz and Schwarz (2013) is the only one that has been developed to estimate time-invariant regressors in the dynamic setting using an the Instrumental variable method. In addition, the data set contains small sample T problem because T is equal to 12. Although Jackknife methods can be consid- ered, that is beyond the scope of this paper. 1 Milgrom and Roberts (1990) defines substitutability by @ 2 f @r i @r j < 0 and complementarity by @ 2 f @r i @r j > 0. Athey and Stern (1997) develop a model to test in the discrete setting. If there are two products,x1 andx2, testing the sign of12 inf(x1;x2) =0 +1x1 +2x2 +12x1x2. 97 Bibliography [1] Asta, Dena and Cosma Rohila Shalizi. (2014). Geometric Network Comparisons, Working Paper. [2] Athey, Susan, and Scott Stern. (1998). “An Empirical Framework for Testing The- ories about Complementarity in Organizational Design.” National Bureau of Eco- nomic Research Working Paper 6600. [3] Banerjee, Anirban. (2012). “Structural Distance and Evolutionary Relationship of Networks.”Biosystems, 107(3): 186-196. [4] Berry Steven, and Joel Waldfogel. (2010). “Product Quality and Market Size.”Journal ofIndustrialEconomics, 58(1): 1-31. [5] Belderbos, Rene, Martin Carree, and Boris Lokshin. (2006). “Complementarity in R&D Cooperation Strategies.”ReviewofIndustrialOrganization, 28: 401-26 [6] Cassiman, Bruno, and Reinhilde Veugelers. (2006). “In Search of Complementarity in Innovation Strategy: Internal R&D and External Knowledge Acquisition.” Man- agementScience, 52(1): 68-82. [7] Cohen, W., and R. Levin. (1989). Empirical studies of innovation and market struc- ture. R. Schmalensee, R. Willig, eds. Handbook of Industrial Organization. North- Holland, Amsterdam, The Netherlands, 1060-107. [8] Cohen, Wesley, and Daniel Levinthal. (1989). “Innovation and learning: The two faces of R&D.”TheEconomicJournal, 99: 569-596. [9] Conley, Timothy, and Christopher Udry. (2010). “Learning about a New Technology: Pineapple in Ghana.”AmericanEconomicReview, 100(1): 35-69. [10] Corchon, Luis, and Simon Wilkie. (1994). “Computers, Productivity, and Market Structure.” Working Paper, Instituto Valenciano de Investigationes Economicas. [11] Ellickson, Paul. 2013. “Supermarkets as a Natural Oligopoly.” Economic Inquiry. 51(2): 1142-1154. 98 [12] Ellickson, Paul. (2007). “Does Sutton apply to supermarkets?.” RAND Journal of Economics. 38(1): 43-59. [13] Foster, Andrew, and Mark Rosenzweig, (1995). “Learning by doing and learning from others: Human capital and technical change in agriculture.”JournalofPolitical Economy, 103(6): 1176-1209. [14] Fox, Jeremy. (2010). “Estimating Matching Games with Transfers. Working Paper: University of Chicago. [15] Gavazza, Alessandro. (2011). “Demand Spillovers and Market Outcomes in the Mutual Fund Industry.”RANDJournalofEconomics. 42(4): 776-804. [16] Grossman, Gene, and Elhanan Helpman. (1991).InnovationandGrowthintheGlobal Economy. Cambridge, MA: MIT Press. [17] Gereffi, Gary. (2001). “Beyond the Producer-Driven/Buyer-Driver Dichotomy: The Evolution of Global Value Chains in the Internet Era.” IDS Bulletin, 32(3): 30-40. [18] Hatfield, John, and Fuhito Kojima. (2008). “Matching with Contracts: Comment.” AmericanEconomicReview, 98(3): 1189-94. [19] Hatfield, John, and Scott Kominers. (2012). “Matching in Networks with Bilateral Contracts.”AmericanEconomicJournal: Microeconomics 4(1): 174-208. [20] Hsiao, Cheng. (2003).AnalysisofPanelData. Cambridge University Press. [21] Jackson, Matthew. (2008).SocialandEconomicNetworks. Princeton University Press. [22] Jurman, Giuseppe, Roberto Visintaniner, and Cesare Furlanello. (2011). “An Intro- duction to Spectral Distances in Networks.” Neural Nets WIRN10: Proceedings of the 20th Italian Workshop on Neural Nets, 227-34. IOS Press Amsterdam. [23] Jo, Hyung Je and Joo Eun Cho. (2012). “Does Hyundai Motor Company’s Pro- duction Management Converge to ”Pull’ Production?: A Study of the Evolution of Demand-driven Production Management through Information System.” Korean Journal of Sociology, 46(5): 233-57. [24] Kenny, Martin, and Richard Florida. (1994). “Japanese Maquiladoras: Production Organization and Global Commodity Chains.” World Development 22: 27-44. [25] Kim, Se-jik, and Hyun Song Shin. (2012). “Sustaining Production Chains Through Financial Linkages.”AmericanEconomicReview, 102(3): 402-406. [26] Lee, Hangkoo. (2011). “Advantages of Vertical Integrations in the Automobile Industry.” KAIBM & KASME Conference, 309-25. [27] Lin, Jianhua. (1991). ”Divergence Measures Based on the Shannon Entropy,” IEEE Transactions on Information Theory, 37(1): 145-51. 99 [28] Milgrom, Paul, and John Roberts. (1990). “The Economics of Modern Manufac- turing: Technology, Strategy, and Organization.” American Economic Review, 80(3): 511-528. [29] Munshi, Laivan, and Jacques Myaux. (2006). “Social Norms and the Fertility Tran- sition.”JournalofDevelopmentEconomics, 80: 1-38. [30] Munshi, Laivan. (2004). “Social Learning in a Heterogeneous Population: Technol- ogy Diffusion in the Indian Green Revolution.”JournalofDevelopmentEconomics, 73: 183-213. [31] Ostrovsky, Michael. (2008). “Stability in Supply Chain Networks.” American Eco- nomicReview, 98(3): 897-923. [32] Park, Minjung. (2013). “Advertising and Market Share Dynamics.” Working Paper. [33] Peel, Leto and Aaron Clauset. (2014). Detecting change points in the large-scale structure of evolving networks, Working Paper. [34] Pesaran, Hashem, and Qiankun Zhou. (2014). “Estimation of Time-invariant Effects in Static Panel Data Models.” Center for Applied Financial Economics Research Paper, University of Southern California. [35] Pincombe, Brandon. (2007). ”Detecting Changes in Time Series of Network Graphs using Minimum Mean Squared Error and Cumulative Summation.” ANZIAM Jour- nal 48: 450-73. [36] Sutton, John. (1991).SunkCostsandMarketStructure: PriceCompetition,Advertising, andtheEvolutionofConcentration (MIT Press). [37] Sutton, John. (2007). Market Structure: Theory and Evidence.HandbookofIndustrial Organization. 100 Chapter A Appendix to Chapter 2 List of Components 1. AIR BAG — 2. AIR BREATHER PIPE — 3. AIR CLEANER (PLASTIC) — 4. AIR CLEANER (PRESS) — 5. AIR COMPRESSOR (DIESEL ENGINE) — 6. AIR CON- DITIONER (CAR COOLER) — 7. AIR CON- DITIONER (COMPRESSOR) — 8. AIR CONDITIONER (CONDENSER) — 9. AIR CONDITIONER (EVAPORATOR) — 10. AIR CONDITIONER (LARGE BUS) — 11. AIR DRYER — 12. AIR HEATER (DIESEL EN- GINE) — 13. AIR SPOILER — 14. AIR SPOILER (TRUCK) — 15. AIR SPRING — 16. AIR VENT DUCT — 17. ALTERNATOR — 18. AN- TENNA (AUTO) — 19. ANTENNA (GLASS) — 20. ARM REST — 21. ASH TRAY — 22. AUTO DOOR CYLIN- DER — 23. AUTO TEMPER- ATURE CONTROLLER — 24. AXLE (FR : BEAM) — 25. AXLE (FR : DRIVE) — 26. AXLE (RR : DRIVE) — 27. AXLE SHAFT — 28. AXLE SPINDLE — 29. BALL JOINT — 30. BATTERY — 31. BATTERY CAR- RIER — 32. BEARING (BALL & ROLLER) — 33. BEARING (CLUTCH RELEASE) — 34. BEAR- ING (NEEDLE) — 35. BEARING (WATER PUMP) — 36. BEARING (WHEEL HUB) — 37. BEARING CAP — 38. BELLOWS (FLEXIBLE TUBE) — 39. BELT (TIMING) — 40. BELT (V) — 41. BOLT & NUT — 42. BRAKE AIR MASTER — 43. BRAKE ASSY (ABS:ANTI-LOCK BRAKE SYSTEM) — 44. BRAKE ASSY (DISC BRAKE SYS- TEM:CALIPER)) — 45. BRAKE ASSY (DRUM BRAKE SYSTEM) — 46. BRAKE ASSY 101 (HYD.UNIT) — 47. BRAKE BOOSTER — 48. BRAKE CHAMBER — 49. BRAKE DISC(A) — 50. BRAKE DISC(B) — 51. BRAKE DRUM & HUB (A) — 52. BRAKE DRUM & HUB (B) — 53. BRAKE DRUM & HUB (HUB BEARING UNIT) — 54. BRAKE LINING & PAD — 55. BRAKE MASTER VAC & CYLINDER — 56. BRAKE SHOE — 57. BRAKE TUBE — 58. BRAKE WHEEL CYLINDER — 59. BULB — 60. BUMPER (PLAS- TIC) — 61. BUMPER (STEEL) — 62. BUMPER GUARD — 63. CAB TILTING PUMP & CYLINDER — 64. CABLE (BATTERY) — 65. CABLE (CONTROL) — 66. CABLE (HIGHTEN- SION/IGNITION) — 67. CABLE (SPEED METER) — 68. CAM SHAFT — 69. CAM SHAFT (SINTERED TYPE) — 70. CAR CLOCK — 71. CAR NAVIGATION SYSTEM — 72. CAR STEREO & AUDIO — 73. CARBON 19 CANISTER — 74. CAR- BURETOR — 75. CATALYTIC CONVERTER — 76. CHECK VALVE — 77. CHIME BELL — 78. CIGAR LIGHTER — 79. CLAMP — 80. CLIP — 81. CLOCK SPRING (ROLL CONNETCOR:AIR BAG) — 82. CLUSTER (COMBINATION METER) — 83. CLUTCH BOOSTER — 84. CLUTCH COVER — 85. CLUTCH DISC — 86. CLUTCH FACING—87. CLUTCH MASTER CYLINDER & RELEASE — 88. CONNECTING ROD — 89. CONNECTING ROD (POWDER FORGING) — 90. CONNECTOR (TUBE &PIPE) — 91. CONNECTOR TERMINAL(WIRING) — 92. CONSOL BOX — 93. CON- TROL UNIT (BLINKER, DOOR LOCK, HEATER, CHIME BELL) — 94. COOLING FAN — 95. COUNTER BALANCE SHAFT — 96. CRANK SHAFT — 97. CRASH PAD (INSTRUMENT PANEL) — 98. CROSS & SIDE MEMBER — 99. CYLINDER HEAD — 100. CYLINDER HEAD COVER — 101. CYLINDER HEAD COVER(PLASTIC) — 102. CYLINDER LINER — 103. DECK — 104. DELIVERY VALVE — 105. DISTRIBU- TOR — 106. DOOR CHECKER — 107. DOOR FRAME/SASH — 108. DOOR HAN- DLE — 109. DOOR HANDLE(ROOF HANDLE) — 110. DOOR LATCH/STRIKER (HOOD & TRUNK)—111. DOOR LOCK—112. DOOR LOCK ACTUATOR(AUTO) — 113. DOOR TRIM — 114. DRAG LINK — 115. DRIVE SHAFT(C.V JOINT) — 116. E.C.U (ELECTRONIC CON- TROL UNIT) — 117. EGR (EHAUST GAS RECIRCULA- TION) VALVE — 118. EMBLEM — 119. ENERGY ABSORBER (BUMPER) — 120. EN- 102 GINE VALVE — 121. EXHAUST BRAKE VALVE/CYLINDER — 122. EXPANSION VALVE(AIR-CON) — 123. FAN CLUTCH — 124. FAN SHROUD(COWLING)—125. FASTENER(SPRAP ,PLUG,CLIP ,PROTECTOR) — 126. FILTER?(OIL,FUEL,AIR) — 127. FLANGE — 128. FLASHER UNIT — 129. FLOOR MAT/CARPET — 130. FLY WHEEL — 131. FORK SHIFT — 132. FREE WHEEL HUB — 133. FUEL DELIVERY PIPE(FUEL, RAIL/RUEL, DISTRIBUTOR) — 134. FUEL DELIVERY PIPE(FUEL RAIL/FUEL, DIS- TRIBUTOR:PLASTIC)—135. FUEL FILLER CAP — 136. FUEL FILLER OPENER — 137. FUEL FILLER TUBE — 138. FUEL INJECTOR — 139. FUEL LEVEL GAUGE(FUEL SENDER) —140. FUEL PRESSURE REGULATOR—141. FUEL PUMP(mechanical) — 142. FUEL TANK — 143. FUEL TANK (PLASTIC) — 144. FUSE — 145. GAS SPRING(GAS DAMPER)—146. GASKET?—147. GEAR?(ENGINE, T/M, AXLE)—148. GEAR PUMP—149. GEAR SHIFT LEVER(CHANGE LEVER) — 150. GEAR SHIFT LEVER (CHANGE LEVER: AUTO T/M?) — 151. GLOVE BOX — 152. GLOW PLUG — 153. GREASE NIPPLE — 154. GROMMET — 155. HEAD LINING(TOP CEALING) — 156. HEAD REST — 157. HEAT PROTECTOR(HEAT SHIEL/EXHINSULATOR) — 158. HEATER UNIT/BLOWER UNIT — 159. HEATER UNIT/BLOWER UNIT(PRE- HEATER) — 160. HINGE(DOOR, TRUNK, HOOD) — 161. HOOD STAY(ROD) — 162. HORN — 163. HOUSING(CLUTCH) — 164. HOUSING(FLY WHELL) — 165. HOUSING(REAR AXLE) — 166. HOUSING(T/M) — 167. HYDRO. LASH ADJUSTER — 168. IDLER ARM — 169. IDLE SPEED ACTUATOR — 170. IGNITION COIL — 171. IMPACT BEAM/BAR — 172. INJECTION PUMP — 173. INSU- LATION NAT/PAD — 174. INTER COLLER(TURBO CHARGER) — 175. JACK(mechanical) — 176. JACK(hydrodynamic) — 177. JUNCTION BOX — 178. KEY SET — 179. KING PIN — 180. KNOB — 181. LADDER — 182. LAMP(COMBINATION) — 183. LAMP (FLOURESCENT) — 184. LAMP (FOG) — 185. LAMP(HEAD) — 186. LAMP(LICENSE) — 187. LAMP (ROOM) — 188. LAMP(TURM SIGNAL) — 189. LANDING GEAR — 190. LATERAL LINK (PARALLER LINK/SUSPENSION ARM) — 191. LIMITED SLIP DIFF(LSD) — 192. LOADSENSING PRESSURE VALVE — 193. 103 LPG KITS (LPG MIXER) — 194. LPG TANK (BOMBE) — 195. MANIFOLD (EXHAUST) — 196. MANIFOLD (EXHAUST:STAINLESS) — 197. MANIFOLD (INTAKE) — 198. MANIFOLD (INTAKE:PLASTIC)— 199. METAL BEARING(CRANK SHAFT BEARING)— 200. METAL BUSHING— 201. METAL BUSHING(SINTERING)— 202. MIRROR(REAR VIEW, OUT SIDE)— 203. MORTOR(ABS)— 204. MORTOR(DOOR LOCK)— 205. MORTOR(FAN)— 206. MORTOR (FUEL STOP)— 207. MORTOR (HEAT/BLOWER) — 208. MORTOR (POWER ANTENNA) — 209. MORTOR (POWER WINDOW)— 210. MOTOR (SEAT)— 211. MOTOR (SUN ROOF)— 212. MOTOR (WASHER)— 213. MOTOR (WIPER) — 214. MOLDING(RUBBER & PLASTIC) — 215. MOLDING(STAINLESS & AL) — 216. MUD GUARD — 217. MUFFLER/PIPE — 218. NOZZLE & HOLDER — 219. OIL COOLER — 220. OIL LEVEL GAUGE & GUIDE — 221. OIL PAN — 222. OIL PAN (AL) — 223. OIL PAN (AUTO T/M) — 224. OIL PUMP(AUTO T/M) — 225. OIL PUMP (ENGINE) — 226. OIL PUMP (POWER STEER- ING) — 227. OIL SCREEN STRAINER — 228. OIL SEAL/O- RING — 229. PACKING TRAY — 230. PARKING BRAKE LEVER — 231. PCV(POSITIVE CRANKCASE VENTI- LATION) VALVE — 232. PEDAL (BRAKE, CLUTCH, ACCELERATOR)— 233. PINTLE HOOK— 234. PIPE/TUBE(FUEL, OIL, AIR)— 235. PIPE/TUBE(FUEL,OIL,AIR: PVC & NYLON)— 236. PISTON — 237. PISTON PIN — 238. PISTON RING — 239. PIT- MAN ARM — 240. PLUNGER(INJECTION PUMP) — 241. POWER SHIFT(LARGE T/M) — 242. POWER TAKE OFF ASS’Y(P .T.O)—243. POWER TRANSITOR— 244. PROPELLER SHAFT— 245. PROPORTIONING CONTROL VALVE (BRAKE) — 246. PULLEY (CASTING)— 247. PULLEY (POWDER METALLURGY) — 248. PUL- LEY (DAMPER/TORSIONAL DAMPER) — 249. PULLEY (PRESS) — 250. PURGE CON- TROL VALVE — 251. RADIATOR— 252. RADIA- TOR CAP — 253. RADIATOR GRILLE(BUMPER GRILLE) — 254. RE- CEIVER DRIER(AIR-CON) — 255. REDUC- TION GEAR — 256. RE- PAY — 257. REMOTE KEYLESS ENTRY — 258. RESERVOIR TANK (AIR/WATER/OIL : PLASTIC)—259. RESERVOIR TANK (AIR/WATER/OIL : PRESS)—260. RESISTER—261. RESONATOR ASS’Y(RESONATOR) — 262. RING 104 GEAR — 263. RING HANDLE (CITY BUS) — 264. ROCKER ARM — 265. ROCKER ARM SHAFT — 266. ROCKER ARM SHAFT (CAM FOLLOWER) — 267. ROOF CARRIER/RACK — 268. RUBBER BUSHING — 269. RUBBER HOSE (BRAKE) — 270. RUBBER HOSE (FUEL, OIL, AIR) — 271. RUBBER HOSE (RADIATOR HOSE) — 272. RUBBER MOUNTING — 273. RUBBER MOUNTING (HYDRO MOUNT- ING) — 274. SAFETY GLASS — 275. SEALANT — 276. SEAL- ING CAP — 277. SEAT — 278. SEAT BELT (SAFETY BELT) — 279. SEAT CLOTH (COVER) — 280. SEAT RECLINDER/ADJUSTER — 281. SENSOR (AIR BAG) — 282. SENSOR (AIR FLOW) — 283. SENSOR (CRANK SHAFT POSITION) — 284. SENSOR (MAP) — 285. SENSOR (OXIGEN) — 286. SENSOR (THERMO & TEMP : WATER,OIL,AIR) — 287. SENSOR (WHEEL SPEED) — 288. SENSOR (SHITF RAIL/SHIFT ROD) — 289. SHIM — 290. SHOCK ABSORBER — 291. SHOCK AB- SORBER (GAS TYPE) — 292. SHOCK ABSORBER (STRUT ASS’Y) — 293. SHOCK ABSORBER (ESC: ELECT SUS- PENSION CONTROL SYS- TEM) — 294. SIDE FRAME — 295. SLACK ADJUSTER — 296. SNAP RING — 297. SOLENOID VALVE (AUTO T/M) — 298. SOLENOID VALVE (EFI SYSTEM) — 299. SOLENOID VALVE(LPG) — 300. SPARE TIRE CAR- RIER — 301. SPARK PLUG — 302. SPEED INDICATOR — 303. SPEED METER DRIVEN GEAR — 304. SPRING?(WIRE) — 305. SPRING?(LEAF) — 306. SPRING? (SUSPENSION COIL) — 307. SPROCKET(CRANK SHAFT & CAM SHAFT) — 308. STABILIZER BAR — 309. STABILIZER LINK — 310. STARTER — 311. STEER- ING COLUMN & SHAFT — 312. STEERING GEAR(MANUAL & POWER SYSTEM) — 313. STEERING GEAR(EPS:ELECTRIC POWER SYSTEM) — 314. STEERING KNUCKLE — 315. STEERING WHEEL — 316. SUN ROOF — 317. SUN VISOR — 318. SURGE TANK — 319. SUSPENSION LOWER/UPPER ARM — 320. SWITCH (BACK UP) — 321. SWITCH (COMBINATION) — 322. SWITCH (OIL PRESSURE) — 323. SWITCH (POWER WINDOW) — 324. SWITCH(STOP LAMP) — 325. SWITCH (ETC) 105 — 326. SYNCHRONIZER RING — 327. SYNCHRONIZER HUB — 328. TACHO- GRAPH(TACHO METER) — 329. TEN- SIONER AUTO/IDLER — 330. THERMO- STAT — 331. THERMOSTAT HOUSING/COVER — 332. THROTTLE BODY — 333. TIE ROD & END — 334. TIMING BELT COVER — 335. TIMING CHAIN — 336. TIRE — 337. TIRE CHAIN — 338. TOOL BOX — 339. TORQUE CON- VERTER(AUTO T/M) — 340. TORQUE ROD — 341. TORQUE ROD BUSHING — 342. TORSION BAR — 343. TRAILING LINK(ARM) — 344. TRANSMISSION(AUTO T/M) — 345. TRANSMIS- SION(REGULAR VEHICLE) — 346. TRANSMISSION(COMMERCIAL VEHICLE) — 347. TRANSMISSION(COMMERCIAL VEHICLE: MULTIPLE STAGE)—348. TRANS- MISSION CONTROL UNIT(T.C.U)—349. TRUNNION BRACKET — 350. TRUNNION SHAFT — 351. TURBO CHARGER — 352. UNI- VERSAL JOINT — 353. VALVE BODY (AUTO T/M) — 354. VALVE COTTER (VALVE KEY) — 355. VALVE GUIDE — 356. VALVE SEAT — 357. VALVE TAPPET — 358. VENTILATION GRILLE(VENTILATION NOZZLE) — 359. VOLTAGE REGULATOR — 360. WASHER — 361. WASHER NOZZLE — 362. WASHER PIPE — 363. WATER PUMP — 364. WEATHER STRIP — 365. WHEEL BALANCE WEIGHT — 366. WHEEL DISC(AL)—367. WHEEL DISC(STEEL)—368. WINCH—369. WINDOWREGULATOR—370. WIPER ARM & BLADE—371. WIPER LINKAGE — 372. WIRE HARNESS — 373. WIRE HAR- NESS(AIR Bag) 106 Chapter B Appendix Process of Supply Chain in the Automobile Industry 107 Figure B.1: Process of Supply Chain in the Automobile Industry 108
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Three essays on the identification and estimation of structural economic models
PDF
Three essays on linear and non-linear econometric dependencies
Asset Metadata
Creator
Song, Hyojin
(author)
Core Title
Three essays on supply chain networks and R&D investments
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Economics
Publication Date
07/28/2015
Defense Date
05/14/2015
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
automobile industry,endogenous sunk costs,OAI-PMH Harvest,supply chain management
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Wilkie, Simon J. (
committee chair
), Hsiao, Cheng (
committee member
), Hsieh, Yu-wei (
committee member
), Jian, Lian (
committee member
)
Creator Email
hyojin338@gmail.com,hyojinso@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-610543
Unique identifier
UC11301010
Identifier
etd-SongHyojin-3718.pdf (filename),usctheses-c3-610543 (legacy record id)
Legacy Identifier
etd-SongHyojin-3718.pdf
Dmrecord
610543
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Song, Hyojin
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
automobile industry
endogenous sunk costs
supply chain management