Citation |

- Permanent Link:
- https://ufdc.ufl.edu/UFE0011601/00001
## Material Information- Title:
- Analysis of markets in the presence of network effects and standards competition
- Creator:
- Wang, Qi (
*Dissertant*) Xie, Jinhong (*Thesis advisor*) - Place of Publication:
- Gainesville, Fla.
- Publisher:
- University of Florida
- Publication Date:
- 2005
- Copyright Date:
- 2005
- Language:
- English
## Subjects- Subjects / Keywords:
- Brands ( jstor )
Economic competition ( jstor ) Emerging technology ( jstor ) Market competition ( jstor ) Market potential ( jstor ) Market share ( jstor ) Network effects ( jstor ) Parametric models ( jstor ) Prices ( jstor ) Product standards ( jstor ) Marketing thesis, Ph. D. Dissertations, Academic -- UF -- Marketing - Genre:
- bibliography ( marcgt )
non-fiction ( marcgt ) theses ( marcgt )
## Notes- Abstract:
- The rapid development of information technology and the digital revolution in recent years have significantly increased the importance of network effects in the success of many new products. When multiple technology standards compete in markets with network effects, consumers's willingness to pay for any given standard depends much more strongly on market size and market share than in markets without network effects. This creates new challenges for marketing strategy. There is a pressing need to understand the marketing implications of technology standards in markets with network effects, for both firms pushing different standards (inter-standards competition) and firms adopting the same standard (intra-standards competition). This dissertation applies theoretical and empirical analyses to address this need. The dissertation consists of two essays. Chapter 2 provides a structural analysis of intra-standard price competition between an innovating firm and imitating firms. Chapter 3 provides a dynamic analysis of new technology diffusion at three levels: product category, platform and brand levels in the presence of inter-standard competition. Data collected from different industries with standards competition are used to test the proposed models.
- Subject:
- competition, effects, empirical, inter, intra, network, standards, structural
- General Note:
- Title from title page of source document.
- General Note:
- Document formatted into pages; contains 81 pages.
- General Note:
- Includes vita.
- Thesis:
- Thesis (Ph. D.)--University of Florida, 2005.
- Bibliography:
- Includes bibliographical references.
- General Note:
- Text (Electronic thesis) in PDF format.
## Record Information- Source Institution:
- University of Florida
- Holding Location:
- University of Florida
- Rights Management:
- Copyright Wang, Qi. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
- Embargo Date:
- 7/30/2007
- Resource Identifier:
- 00744931813398418 ( alepOCLCh )
## UFDC Membership |

Downloads |

## This item has the following downloads:
wang_q ( .pdf )
wang_q_Page_02.txt wang_q_Page_08.txt wang_q_Page_52.txt wang_q_Page_39.txt wang_q_Page_76.txt wang_q_Page_15.txt wang_q_Page_35.txt wang_q_Page_50.txt wang_q_Page_47.txt wang_q_Page_73.txt wang_q_Page_23.txt wang_q_Page_48.txt wang_q_Page_41.txt wang_q_Page_20.txt wang_q_Page_24.txt wang_q_Page_04.txt wang_q_Page_44.txt wang_q_Page_78.txt wang_q_Page_57.txt wang_q_Page_19.txt wang_q_Page_28.txt wang_q_Page_17.txt wang_q_Page_06.txt wang_q_Page_80.txt wang_q_Page_10.txt wang_q_Page_22.txt wang_q_Page_77.txt wang_q_Page_70.txt wang_q_Page_66.txt wang_q_Page_13.txt wang_q_Page_27.txt wang_q_Page_30.txt wang_q_Page_31.txt wang_q_Page_07.txt wang_q_Page_37.txt wang_q_Page_75.txt wang_q_Page_43.txt wang_q_Page_60.txt wang_q_Page_69.txt wang_q_Page_56.txt wang_q_Page_72.txt wang_q_Page_38.txt wang_q_Page_61.txt wang_q_Page_63.txt wang_q_Page_46.txt wang_q_Page_74.txt wang_q_Page_51.txt wang_q_Page_32.txt wang_q_Page_33.txt wang_q_Page_64.txt wang_q_Page_45.txt wang_q_Page_18.txt wang_q_Page_49.txt wang_q_Page_62.txt wang_q_Page_68.txt wang_q_Page_81.txt wang_q_Page_26.txt wang_q_Page_59.txt wang_q_Page_01.txt wang_q_Page_29.txt wang_q_Page_65.txt wang_q_Page_40.txt wang_q_Page_11.txt wang_q_Page_79.txt wang_q_Page_14.txt wang_q_Page_53.txt wang_q_Page_03.txt wang_q_Page_12.txt wang_q_Page_34.txt wang_q_Page_54.txt wang_q_Page_09.txt wang_q_Page_05.txt wang_q_Page_58.txt wang_q_Page_36.txt wang_q_Page_71.txt wang_q_Page_42.txt wang_q_Page_55.txt wang_q_Page_25.txt wang_q_Page_16.txt wang_q_Page_21.txt wang_q_pdf.txt wang_q_Page_67.txt |

Full Text |

ANALYSIS OF MARKETS IN THE PRESENCE OF NETWORK EFFECTS AND STANDARDS COMPETITION By QI WANG A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2005 Copyright 2005 by QI WANG To My Parents ACKNOWLEDGMENTS This dissertation is indebted to many people. First, I would like to thank my advisor, Jinhong Xie, for guiding me through my entire Ph.D. study and the entire dissertation process. She is not only the chair of my dissertation committee, but also a mentor and a lifelong friend to me. Without her constant support, indefatigable encouragement, steady mentoring and patient guidance, I could not have developed strong interests in research, accomplished this dissertation and gone through the difficulties in my graduate study and my personal life. There is no single word in the world that could express my great appreciation to her. Second, I would also like to thank my committee members and other faculty members for their extensive comments and feedback for this dissertation. In particular, I benefited immensely from Barton Weitz, Steven Shugan, Chunrong Ai, Mike Lewis and Debi Mitra. I am also indebted to my friends for their dependable help, especially Yubo Chen, Qiong Wang, Qian Tang, Younghan Bae, Xu Jun, Lucy Liu, Shuman Li and Song Xue. The friendship with them makes my graduate study enjoyable and fun; the help from them made it possible for me to complete my Ph.D. study while bringing up my daughter. Last, but certainly not least, I am extremely grateful from the bottom of my heart to my family--my husband, my daughter, my parents, my brother and sister. I feel extremely lucky to grow up in a family where my parents, my brother and sister are always selflessly with me for every decision I make, even though they are physically far from me. I am thankful to my husband for his understanding and to my daughter for all of the joy that she brought to me in my life, especially in the five-year-long study. It is their unconditional love and support that make all my accomplishment possible. TABLE OF CONTENTS page A C K N O W L E D G M E N T S ................................................................................................. iv LIST OF TABLES .................................................... ....... .. .............. viii LIST OF FIGURES ......... ......................... ...... ........ ............ ix A B STR A C T ................................................. ..................................... .. x CHAPTER 1 IN T R O D U C T IO N ............................................................................. .............. ... 2 A STRUCTURAL ANALYSIS OF INTRA-STANDARD PRICE COMPETITION IN MARKETS WITH NETWORK EFFECTS.............................6 2.1 Introduction..................................................... ................... .. ....... ...... 6 2.2 The B baseline M odel .................. ............................ .... .... .. ........ .... 13 2.2.1 C onsum er B behavior ........................................................ ............. 13 2.2.2 Firm B ehavior................. ............................................ .. ...... .. .......... 14 2.3 Empirical Analysis of the Baseline M odel ............. .............................................18 2.3.1 The 3.5-inch Floppy Disk Drive Market..............................................18 2.3.2 D ata.................................... ........................ ..... ... ......... 21 2.3.3 Identification ................ .................... .............. ............. .. .......... 23 2.3.4 Empirical Results of Baseline M odel ............... ................ ............... 23 2.4 Modified Model and Empirical Results.....................................................26 2.4.1 M odified M odel ........................ ............. .... ........................... 26 2.4.2 Empirical Results: Time-Varying CV Approach ....................................27 2.4.3 Validation: Stage-based Menu Approach.......................................33 2.4.4 D discussion of K ey Findings............................................ .. .............. .35 2.5 C conclusion ..................................... ................................ .......... 39 3 DYNAMIC ANALYSIS OF NEW TECHNOLOGY DIFFUSION IN MAREKTS W ITH COM PETING STANDARDS..................................... ........................ 42 3 .1 Introdu action ......... .... ....... .. .......................................................... 42 3.2 T he B baseline M odel .............................................. .. ...................... .... 47 3.2.1 Consum er U utility Equation..................................... ......... ............... 47 3.2.2 M market Share Equations ........................................ ......... ............... 48 3.2.3 Diffusion Equation at Three Levels ................................. ............... 49 3.3 Dynamic Model with Consumer Expectation.....................................................50 3.3.1 Consumer Expectation of User-Base and Firm-Base...............................51 3.3.2 Consumer Dynamic Decision..................................................................51 3.3.3 Choice Probability ......................................... .... .. .... .. .. ........ .... 53 3 .4 E m pirical A naly sis.......... .......................................................... ... .... .. .... .. 54 3 .4 .1 D ata ....................................................................................................54 3.4.2 Nested Logit Model Estimation....................................... ............... 56 3.4.3 D iffusion Equation Estim ation................... ............................................ 57 Impacts of Network Effects and Standard Competition on Market P potential ................................................. ................ 57 3.5 Conclusion and Future Research ........................................ ....... ............... 62 L IST O F R EFE R E N C E S ............................................................................. ............. 66 B IO G R A PH IC A L SK E TCH ..................................................................... ..................70 LIST OF TABLES Table pge 2 -1 D escriptiv e Statistics ......................................... .. ................................ ............... 22 2-2 Baseline Model: Estimated Competitive Structure ..............................................25 2-3 Revised Model 1, Time-Varying CV Approach ............................................... 29 2-4 Revised Model 2, Time-Varying CV Approach ............................................... 31 2-5 Revised Model 3, Stage-Based Menu Approach .............................................. 35 3-1 D escriptiv e Statistics .......................................................... .......... ..................... 55 3-2 N ested Logit M odel Estim ation ........................................ ......................... 57 3-3 M market Potential Equation Estim action ........................................ ............... 58 3-4 The Diffusion Equations Estimation: Category and Standard Level .....................60 3-5 The Diffusion Equations Estimation: Brand Level ............................................61 LIST OF FIGURES Figure pge 2-1 The Changes of Conjectural Variation Parameters with Time..............................32 2-2 Sales: 3.5 inch Floppy Disk Drive vs. Competing Products ..................................39 3-1 Impacts of Installed-User-Base on Market Potential ............................................59 3-2 Impacts of Supporting-Firm-Base on Market Potential .......................................60 3-3 Impacts of Word of Mouth on Category Diffusion............................. ...............61 3-4 Impact of Betamax User-Base on Category and Standard Diffusion.......................62 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy ANALYSIS OF MARKETS IN THE PRESENCE OF NETWORK EFFECTS AND STANDARDS COMPETITION By Qi Wang August 2005 Chair: Jinhong Xie Major Department: Marketing The rapid development of information technology and the digital revolution in recent years have significantly increased the importance of network effects in the success of many new products. When multiple technology standards compete in markets with network effects, consumers' willingness to pay for any given standard depends much more strongly on market size and market share than in markets without network effects. This creates new challenges for marketing strategy. There is a pressing need to understand the marketing implications of technology standards in markets with network effects, for both firms pushing different standards (inter-standards competition) and firms adopting the same standard (intra-standards competition). This dissertation applies theoretical and empirical analyses to address this need. The dissertation consists of two essays. Chapter 2 provides a structural analysis of intra-standardprice competition between an innovating firm and imitating firms. Chapter 3 provides a dynamic analysis of new technology diffusion at three levels: product category, platform and brand levels in the presence of inter-standard competition. Data collected from different industries with standards competition are used to test the proposed models. CHAPTER 1 INTRODUCTION The rapid development of information technology and the digital revolution in recent years have significantly increased the importance of network effects in the success of many new products. When multiple technology standards compete in markets with network effects, consumers' willingness to pay for any given standard depends much more strongly on market size and market share than in markets without network effects. This creates new challenges for marketing strategy. There is a pressing need to understand the marketing implications of technology standards in markets with network effects, for both firms pushing different standards (inter-standards competition) and firms adopting the same standard (intra-standards competition). This dissertation applies theoretical and empirical analyses to address this need. The dissertation consists of two chapters. Chapter 2 provides a structural analysis of intra-standardprice competition between an innovating firm and imitating firms. Chapter 3 provides a dynamic analysis of new product diffusion at both product category and platform levels in the presence of inter-standard competition. In Chapter 2, I analyze the intra-standards price competition in markets with network effects by using a structural analysis approach. In the literature on network effects, marketing scholars have highlighted one characteristic, installed-user-base and demonstrated its implications to marketing strategies, but under-explored the other characteristic, supporting-firm-base. Although the supporting-firm-base is important to the success of firms in face of standard competition and impacts marketing strategies differently from the installed-user-base, there are NO empirical works that have demonstrated its existence and analyzed its important implications to marketing strategies. This paper empirical examines the existence of supporting-firm-base and investigates the impacts of supporting-firm-base and installed-user-base on the competitive interactions between an innovating firm and its imitators in markets influenced by network effects. The competitive interaction among firms supporting the same technological standard (i.e., intra-standard competition) is also an important but underexplored issue. Theorists differ widely in the assumptions that they have made about the structure of competition among firms supporting the same technological standard (i.e., intra-standard competition). Although they have assumed a fixed competitive structure (e.g., bertrand, leader-follower, or cooperation) over product life cycle, we argue that the competitive structure of intra-standard competition is likely to be stage-dependent. In this study, I apply the new empirical industrial organization framework (NEIO) to examine the supporting-firm-base effects and intra-standard competition in 3.5-inch floppy disk drives during the period 1983-1998. My time-varying structural model estimations yield several interesting results. First, I find empirical evidence that consumers' valuations are not only positively related to the installed base, but also positively related to the number of supporting firms. Second, my results show an opposite relationship of supporting-firm-base and installed-user-base with the competitive interactive behavior --- a positive relationship between the competitive interaction and the supporting-firm-base and a negative relationship between the competitive interaction and the installed-user-base. The positive relationship suggests that the supporting-firm- base effect discourages competitive behavior. In contrast, the negative relationship suggests that installed-user-base motivates competitive behavior. Third, my results show that the competitive structure of the floppy drive market did change over the period studied. Firms behaved collectively in the early stage, competitively in the middle stage, and collectively again in the late stage of the product life cycle. This study fills a gap in the network effects literature by providing empirical evidence on the pattern of intra-standards competition and demonstrating the implications of such intra-standards competition for innovating firms' marketing strategy. The empirical findings of this study raise some important new theoretical and empirical issues for future research on markets with network effects and standards competition. In Chapter 3, I dynamically analyze the diffusion of new technology in markets with inter-standards competition by incorporating consumer expectation on network effects. Understanding the impact of inter-standards competition on new technology diffusion is extremely important to the success of new products in the presence of network effects and competing technological standards. In this study, I empirically examine how inter-standard competition affects new technology diffusion at three levels- -the category-level, the standard-level and the brand-level. Specifically, I examine the impact of competition between incompatible standards on new technology diffusion by incorporating (1) market share differences, which measure the relative strength between competing standards, and (2) two important network effect factors: installed-user-base and supporting-firm-base of competing standards into our model. The network effect factors measure the absolute strength of each competing standard. My model involves three components: (1) a consumer utility function that captures the influences of price, product attribute, and network effects-measured as installed- user- base and supporting-firm-base of two competing standards, (2) equations that derive the market share of two incompatible platforms based on the estimated nested-logit model, and (3) equations that derive the dynamic diffusion of the whole industry, two competing standards and each individual brand. I first estimate a baseline model which captures the impacts of the current network effects--current user-base and firm-base from two competing standards on consumer's utility and consequently on new technology diffusion. I then extend the baseline model by incorporating the consumer expectations for the future user-base and firm-base of competing technological standards into the consumer utility function. A dynamic programming method and the maximum likelihood estimation method are thus used to solve and estimate the extended model. The model is estimated using data from the VCR industry. The estimated parameters reveal several interesting results. First, my results show that while the installed-user-base from the compatible standard increases the consumer's evaluation to a new product, which is consistent with the literature on network effects, the installed-user- base from competing technology standard decreases the consumer's evaluation to a new product, which is new to literature on network effects, and demonstrates the evidence for one of the important impacts of standard competition on new technology diffusion. Second, I find the empirical evidence that standard competition impacts new technology diffusion complimentarily. This can be seen from two results: (1) the supporting-firm-base from both of two competing standards--VHS and Betamax positively influence the consumer's evaluation on each brand, and (2) the installed-user- base from two competing standards generates positive word of mouth effects on the category diffusion as well as the standard and brand diffusion at the early stage of standard competition when none of the two competing standards dominates market. Third, I also find results that the network effects impact the market potential dynamically. While the installed-user-base from two standards both positively impact the new technology diffusion at the early stage of standard competition, the installed-user- base of Betamax turns to impact the diffusion of new technology at three levels negatively after the technology standard VHS dominates the market. Finally, my results suggest that network effects and standard competition not only impact the diffusion on individual consumer's evaluation to each brand, but only on the market potential complementarily and dynamically. Similarly with the word of mouth impact of installed-user-base on new technology diffusion, the supporting-firm-base from both of two standards impact market potentials seems positively at the early stage of standard competition, but the supporting-firm-base of Betamax has no impact on the market potential later on when the VCR market is dominated by VHS format. This study extends research on network effects and provides important implications to markets with network effects by explicitly modeling and estimating the impacts of inter-standards competition on new technology diffusion. It also advances understanding of the formation of consumer expectation for the future installed based of competing technological standards-a very important but under-explored research area. CHAPTER 2 A STRUCTURAL ANALYSIS OF INTRA-STANDARD PRICE COMPETITION IN MARKETS WITH NETWORK EFFECTS 2.1 Introduction The rapid development of information technology and the digital revolution in recent years have significantly increased the importance of network effects and standard competition in the market place (Shapiro and Varian 1999). Marketing scholars are giving increasing attention to the implications of network effects and standards competition for marketing strategies. Several new studies, for example, address such important issues as the evolution of markets with indirect network externalities (Sachin Gupta et al., 1999, Nair, Chintaguta, and Dube 2004), asymmetry network effects (Shankar and Bayus 2003), pioneer survival in network markets (Srinivasan, Lilien, and Rangaswamy 2004), and product line and technology licensing in the presence of network effect (Sun, Xie, and Cao 2004). One important but underexplored issue in the research of network effects is competitive interactions among firms supporting the same technological standard. Understanding competitive interactions among innovating and imitating firms in the presence of network effect is important to practitioners because the strategic behavior of firms is crucial to the success of the technological standard as well as the fate of the individual firms in markets with network effects. Understanding firms' competitive behavior in intra-standard competition is also important from a research perspective: competition among firms adopting the same technology standard in the presence of network effects exhibits certain unique characteristics that do not apply to traditional markets, a phenomenon that raises new, interesting theoretical and empirical issues. Different from traditional markets where competition generally has a negative effect on profit, intra-standard competition in markets with network effect (i.e., competition among firms adopting the same technological standard) can be beneficial. Analyses based on game theory have shown that in markets with network effects, innovating firms can earn a higher profit by having compatible competitors than being a monopolist (Katz and Shapiro 1985, Corer 1995, Xie and Sirbu 1995, Sun, Xie, and Cao 2004) or even benefit from subsidizing compatible entrants (Economides 1996). While theoretical research has provided motivations for innovating firms to encourage compatible entry, the literature has not addressed the issue of how network effects may affect the competition structure in such markets. In the literature on network effects, theorists differ widely in the assumptions that they have made about the structure of competition. For example, some assume that innovator and imitator behave as symmetric Nash game players (e.g., Xie and Sirbu 1995, Economides 1996) while others assume that firms play Stackelberg (leader-follower) games (e.g., Corer 1995). Still other theorists assume a cooperative market structure in which both innovator and imitator behave collectively (Gallini 1984). However, there is little empirical evidence on HOW innovating and imitating firms adopting the same standard actually compete in the market place. In this study, we empirically examine the existence of supporting-firm-base effect and the intra-standard price competition between innovating and imitating firms. We argue that, although theoretical research of network effects has assumed a single fixed competitive structure (e.g., bertrand, leader-follower, or cooperation), we expect that the competitive structure of intra-standard competition is not fixed but varies across different stages of the product life cycle: network externalities create interdependence between consumers and firms and among compatible rivals; as such interdependence changes with the development of markets and firms may change their competitive behavior accordingly. First, network effects create an installed-user-base effect that motivates competitive behavior. Theoretical research has suggested that consumer willingness to pay for a given product increases with the size of the user base of its technology standard (Farrell and Saloner 1986, Katz and Shapiro 1985). This positive relationship between consumer willingness to pay and the installed-user-base has been supported by recent empirical studies based on data from various industries affected by network effects, including software (Gandal 1995, Brynjolfsson and Kemerer 1996,), VCRs (Hiroshi Ohashi 2003), video games (Shankar and Bayus 2003), CD players (Basu, Mazumdar, and Raj 2003), and PDAs (Nair, Chintaguta, and Dube 2004). As suggested by the literature of network effect (e.g., Katz and Shapiro 1985, Corner 1995, Economides 1996), innovating firms can benefit by encouraging competition because competition leads to a lower price and a larger total user base, thus a higher product value. This positive link between competition and profit suggests that the installed-user-base effect motivates competitive behavior in intra-standard competition. Second, network effects may create a supporting-firm-base effect that motivates cooperative behavior. Based on the assumption of fulfilled expectation, several theoretical studies suggest that consumer willingness to pay is positively affected by the expected total market size, which is larger for a competitive market than for a monopoly market (e.g., Katz and Shapiro 1985, Sun, Xie, and Cao 2004) and increases with the number of firms supporting the same technological standards (e.g., Economides 1996). While this theory implies a positive relationship between consumer willingness to pay and the number of supporting firms, such a relationship has not been empirically studied. Testing this possible supporting-firm-base effect will provide the empirical evidence needed to validate the existing theory; in addition, the existence of a supporting-firm-base effect would suggest a negative link between competition and consumer willingness to pay-competition discourages market entry and leads to a small supporting-firm-base, thus a low consumer willingness to pay. This negative link between competition and profit implies that the supporting-firm-base effect discourages competitive behavior but motivates cooperative behavior in intra-standard competition. Since both user base and firm base change with time, the relative strength of the installed-user-base effect and the supporting-firm-base effect may vary across different stages of the product life cycle. As a result, the competitive structure of intra-standard competition may not be fixed but stage-specific. In this paper, we empirically examine the existence of supporting-firm-base and investigate the possible stage-dependent pattern of intra-standard price competition between innovating and imitating firms. Specifically, we apply the new empirical industrial organization framework (NEIO) to examine competitive interactions in the 3.5- inch floppy disk drive (FDD) market using 16 years of data from all manufactures of floppy disk drives during the period 1983-1998. Marketing scholars have recently applied the NEIO framework to examine firms' competitive pricing behavior in several markets without network effects. For example, Sudhir investigates the competitive pricing behavior in the U.S. author market (2001a); he also examines manufacturers' pricing interactions in the presence of strategic retailers, using data from yogurt and peanut butter markets (2001b). Kadiyali (1996) examines the incumbent's strategic behavior (deterrence or accommodation) in face of a new entrant, using data from the photographic film industry. These previous studies based on the NEIO framework have illustrated the importance of understanding competitive interactions to firms' marketing strategies and provided us with useful methodological guidance. We chose to examine intra-standard competition in the 3.5-inch FDD market for several reasons. First, the FDD industry is subject to network effects. Different types of drives in this market are not interchangeable and the value of a given type of disk drive depends on the number of people who have adopted the same drive (i.e., the larger the installed base of a given disk drive, the more people one can share files with). Second, the innovator, Sony, faced competition from other manufactures adopting its technological standard. As Disk Trend Report recorded, Sony was willing to work with U.S. manufacturers to build its technology as the industry standard. As a result, as many as 27 other manufacturers adopted Sony's technology and produced 3.5-inch FDDs in the same market. These imitating firms were both rivals of and allies to the innovator, Sony. Third, the 3.5-inch FDD encountered intensive standards competition when it was first introduced by Sony in 1981. Before its introduction, the 5.25-inch FDD had been widely adopted in the microcomputer market. Soon after its introduction, three other new incompatible disk drives (i.e., 3-inch, 3.25-inch, and 4-inch) were also introduced in the FDD market. Hence, the 3.5-inch FDD market has the main characteristics we discussed above. To empirically examine the proposed supporting-firm-base effect and the stage- dependent competitive interactions, in this paper, we (1) model consumer utility as a function of both the installed-user-base and the supporting-firm base; (2) incorporate an equation of the number of supporting firms in our structural equations; and (3) model stage-dependent competitive interactions by introducing a time-varying structure to the estimation of conduct parameters. Three different approaches ("menu approach," "weighted profit approach," and "conjectural variation") are used complementarily in testing and validating the stage-dependent competitive structure. Our empirical investigation of the floppy disc drive market reveals several interesting findings. First, we find that consumer valuations are not only positively related to the installed base, as suggested by previous empirical studies, but are also positively related to the number of manufactures that adopt the same standard. In other words, consumers care about both the installed-user-base and the supporting-firm-base. As the value of a product increases with both, this result highlights the importance of building up the base of firms adopting the same technological standards, thus the importance of competitive interaction in intra-standard competition. Second, by modeling the interrelationship of the number of supporting firms and the prices of innovating and imitating firms at previous period, we are able to identify factors that affect the supporting-firm-base. Our results show that the number of supporting firms is positively influenced by the mean price level of imitating firms in a previous period, This finding is important because it implies that competitive behavior discourages, but cooperative behavior encourages, market entry and suggests that the supporting-firm-base effect has an opposite impact compared with the installed-user-base that encourages competitive behavior. Third, our results show that installed-user-base motivates competitive behavior and supporting-firm-base motivates cooperative behavior, the competitive structure of the floppy disk drive market changed over the period studied. Firms behaved collectively in the early stage, competitively in the middle stage, and collectively again in the late stage of the product life cycle. Such a stage-dependent competition pattern has never been suggested on either theoretical or empirical grounds, but is certainly plausible for markets where firms supporting the same standard are both rivals and allies. As our results show, consumers' willingness to pay in markets with network effects depends on BOTH the size of the installed base and the number of manufactures adopting the same standard. The latter can motivate the innovating firm to collaborate rather than aggressively compete (on price) with imitating firms at the early and later stages, though for different reasons. The early collaborative behavior helps attract more manufacturers to adopt a new standard when the new standard is still struggling for market acceptance, and the late collaborative behavior discourages manufacturers from abandoning the existing standard when the next generation of technology appears. As a result, firms are more likely to cooperate in the early and late stages, and to compete on price in the middle stage. The rest of this study is organized as follows. We introduce the baseline model in section 2.2. We discuss our data and present the empirical results based on the baseline model in section 2.3. We then introduce the modified model in section 2.4 and report the empirical results and validation results in section 2.5. Finally, we discuss conclusions and future research in Section 2.6. 2.2 The Baseline Model This section defines our baseline model. The baseline model has two characteristics. First, it adopts the specification for consumer behaviors as suggested in the literature of network effects (e.g. Nair, Chintaguta, and Dube 2004), which considers the impact of installed-user base but not the impact of supporting-firm base on consumer utility. Second, as many studies of competitive interactions (e.g. Sudir 2001a) it assumes a constant competitive structure. We modify the baseline model in Section 4 by (1) incorporating supporting-firm base into functions of consumer utility and conjectural parameters and (2) assuming a stage-dependent competition pattern. 2.2.1 Consumer Behavior Following the literature of network effect, we assume that consumers make their purchase decisions by maximizing their utilities and consumer utility is affected by installed-user base and product attributes. Letj denote firms, where j= 1 denotes the innovator and j= 2 denotes imitator.1 The utility of consumer i at time t for firm j's product is defined as "ut = 8o + )XJktk -a + Pt + 71-1 + t +,jt = "jt +,t (2.1) k where xJ, and p, are the kth attribute and price for product j at time t. N, 1 is the cumulative sales at time t -1 (i.e., a measurement of consumer installed-base). J,, is the unobserved characteristic for product j at time t, uj, is the average utility across 1 In real markets, there may be more than one imitator. Following the literature (Suslow (1986) and Putsis and Dhar (1999)), firm 2 in the duopoly model (1) can be interpreted as a representative imitator. 14 consumers for product j at time t, and y is the random term across product j and consumer i at time t. Following the literature (e.g. Sudir 2001b), we assume that s, has double exponential distribution and is independent, identical across products and consumers. We allow consumers to have an option of purchasing outside product j= 0, and normalize the utility of the outside products to zero across times. Then the market share for product j at time t is given by exp(udt) ,t = exp (t) (2.2) 1+ exp( u) Similarly, the market share for outside product j = 0 at time t is given by Sot = +exp (2.3) I+ + exp (u.) where si and so, denote the market share of product j and outside product = 0. From the market share equations (i.e., (2.2) and (2.3)), we can derive In(sj, / s,) = jt,,. (2.4) Since the demand D, for product j at time t can be expressed by Dj = s jMt, where M, is the market potential at time t, the demand equations can hence be written as ln(Djt)= In(Dot) + ui, j 0. (2.5) 2.2.2 Firm Behavior Firms make price decisions by maximizing their profits. The firm j 's profit at time t is given by 7rt = (Pt ct )D1t, j = 1,2. (2.6) where pit, c,, are firm j's price and marginal cost at time t, respectively. Marginal Cost. Firms' marginal costs are defined by cJt = zgzjgt+ jt (2.7) g where ZJgt are the factors that influence firm's marginal cost. Note that, Zjgt includes product attributes xjkt and other influential factors such as time trend. -, is the unobservable cost factors for product j. Strategic Interactions. Two approaches, the conjectural variation approach (CV), and the menu approach, have been used frequently in the empirical studies of firm competitive interactions (e.g., Kadiyali 1996, Vilcassim, Kadiyali and Chintagunta 1999). The CV approach examines the strategic interactions by estimating a set of parameters and inferring the competitive structure from the signs of the set of parameters, while the menu approach does so by separately estimating alternative models and inferring the competitive structure from the selected best-fitted model. It is common for empirical researchers to use only one approach to estimate the competitive interactions because different approaches often lead to consistent results given appropriate model specification (e.g. Bresnahan 1989). Since one of our objectives is to identify more appropriate model specification (e.g., examine if supporting-firm-base should be incorporated into the consumer utility function and if state-dependent competitive structure is more appropriate than a fixed competitive structure), in this paper, we use both approaches complementarily. The CV approach (pioneered by Iwata 1974) captures the competition structure simply by a constant conjectural variation parameter. In the CV approach, a firm is assumed to have conjectures about how its competitors will react to its price changes and incorporate the conjectures into its price decisions (Kadiyali et al. 2001). In the baseline model, the first-order condition for firm j is alrjt aD t D t =D+ +(Pt-ct+ ) )-- =0, j=1,2 (2.8) SDpt pjt 1=12 pclt where 2 = pJt /Pp,, (i= 1,2, i j) are called conjectural variation parameters (i.e., a1, = 9p2t /9p1 and 2 = p9l /0p2t ). The conjectural variation parameters thus tell us how a firm reacts to its competitor's price changes and therefore indicate the competition structure. For example, = 22 = 0 indicates Bertrand competition. When # 0, but 12 = 0, we infer that the innovator is leader and the imitator is follower. Symmetrically, when k2 # 0, but all 0 = 0, we infer that the imitator is leader and the innovator is follower. >,22 >0 implies a cooperative market structure (Kadiyali et al. 2001). Different from CV approach, in menu approach, the competitive structure of an underlying market is inferred by estimating the alternative models (i.e., models with different competitive structure) separately and selecting the best-fitted model. We consider four possible competitive structures: Bertrand, Innovator-Led leader-follower, Imitator-Led leader-follower, and Cooperation. Under the Bertrand competition assumption, each firm chooses its price assuming that its competitor does not react to its price changes, so the first order condition for firm j is S=Dt -c, =0, j=,2. (2.9) ajt = + ( c) jt = j = 1, Under the innovator-led leader-follower assumption, innovator anticipates imitator's reaction to its price and incorporates it in its price decision. Therefore, the first order condition for innovator is D1 = DI + (p t) D, l DIt P2t 0. (2.10) pilt Lpt 8a p 2t pi1t 1 In contrast, the imitator makes its price decision assuming that innovator does not react to its price. That is, the first order condition for imitators is D2, +(2t -c2) =0. (2.11) Taking derivative of imitator's first order condition (2.11) with respect to pit, we derive the expression of p2t /ait and substitute it in the first order condition of innovator (2.10). Symmetrically, we can derive the first order conditions for innovator and imitators under imitator-led leader-follower assumption as t(Plt t), =o. (2.12) '-, ', '/, p -- D2t +(P2t -C2t) D2t + D2t Opt 0. (2.13) OP2t L P2t P pIt OP2t Similarly, we can derive the expression of pit /SP2t in first order condition (2.13) by taking derivative of first order condition (2.12) with respect to p2t and substituting into (2.13). Finally, under the cooperation assumption, both firms act as if they were monopoly selling the differentiated products. In other words, they maximize a total profit of it own and its competitor when making their price decisions. The total profit is n, = (pi, -ct)D, The first order condition for firm j is J =, Dt + (Pt -clt) Dt = 0, j= 1,2. (2.14) apt 1=1,2 apt I#i In the next section, we empirically estimate the baseline model using both CV and menu approach. We show that two approaches lead to different inferred competitive structures. In Section 4, we modify the baseline model by incorporating a new variable, supporting-firm-base, into the baseline model. We also allow conjectural variation parameters to be timing-varying. We then use the time-varying CV approach to examine the modified model. Our results show a significant effect of supporting-firm base and suggest that the stage-dependent competitive structure provides a better fit than the constant competitive structure. Finally, to validate our key results, we apply menu approach to the modified model and perform stage-based estimation. The results show that the two approaches, CV and menu approach, lead to consistent results when the modified model and stage-dependent competitive structure are used. 2.3 Empirical Analysis of the Baseline Model We estimate the baseline model using data of the 3.5-inch floppy disk drive market from 1983 to 1998. 2.3.1 The 3.5-inch Floppy Disk Drive Market A floppy disk is a data storage device that comprises a circular piece of thin, flexible magnetic medium encased in a square or rectangular plastic wallet.2 Floppy disks were ubiquitously used to distribute software, transfer data between computers, and create small backups in the 1980s and 1990s. Sony introduced the 3.5-inch floppy disk 2 From http://en.wikipedia.org/wiki/Floppy disk. drive in 1981. Before its introduction, there were two main floppy disk drives, 8-inch and 5.25-inch disk drives. The 8-inch floppy disk drives, the first floppy disk drive introduced in 1967 by IBM, were mainly used in mainframe computers and minicomputers. The 5.25-inch disk drives, introduced by Shugart in 1976, were mainly used in minicomputers and microcomputers (PCs). Floppy disks are read and written by a floppy disk drive (FDD). A floppy disk drive basically has four principal components: read-write heads, which are located on the both sides of a disk and move together on the same assembly; a drive motor, which is a very small spindle motor driving the rotation of the disks; a stepper motor that makes a precise number of stepped revolutions to move the read-write head assembly to the proper disk track position; and electronic circuitry, which handle the data read from or written to the disk and control the stepper-motor control circuits used to mover the read- write heads to each track, as well as the movement of the read-write heads toward the disk surface. With the increasing popularity of PCs in 1980s, small size, easy to carry and display were becoming the major concerns of computer manufacturers and hence disk drive manufacturers. The 5.25-inch floppy disk drives were clearly becoming a limitation to the improvements of microcomputer size. At this time, Sony introduced its 3.5-inch disk drive. Sony uses a single-crystal ferrite head in its 3.5-inch floppy disk drive instead of the poly-crystal ferrite head used by 8-inch and 5.25-inch disk drive. The single-crystal ferrite head is lighter, more stable and more reliable than the poly-crystal ferrite head. With this stable and reliable read-write head, Sony was able to put the same number of 80 tracks as 5.25-inch disks on much smaller 3.5-inch disks and keep its 3.5-inch disk drive with equal or even better reliability than 5.25-inch and 8-inch disk drive.3 As a result, the disk density was increased from 48 tracks per inch (TPI) to 135 TPI, and disk size was therefore largely decreased. Along with the introduction of Sony's 3.5-inch disk drive, there were two other microfloppy disk drives contending for market standard: the 3-inch drive, introduced by Matsushita Electric Industrial, Hitachi and Hitachi/Maxell; and the 3.25-inch disk drive, introduced by Tabor and Seagate Technology. Besides these two disk drives, IBM, as an industry dominant manufacturer, was also expected to introduce a 3.9-inch disk drive on an unknown schedule. These floppy disk drives with different sizes were incompatible because disks could only be loaded in the correct size of disk drive. As Disk Trend Report recorded, Sony was willing to work with U.S. manufacturers to build its technology as the industry standard. As a result, as many as 27 other manufacturers adopted Sony's technology and produced 3.5-inch FDD in the same market. Sony's 3.5-inch disk drive eventually became the industry standard. To sum up, the 3.5 inch FDD market has the main characteristics in which we are interested. First, it is subject to network effects. Second, the innovator, Sony, faced competition from imitating firms adopting its technological standard. Third, the 3.5-inch FDD encountered intensive standards competition. See Mini Micro Systems, April, 1981. 2.3.2 Data Our annual data on product attributes, sales and market share in the 3.5-inch floppy disk drive market are collected from Disk TrendReport,4 a leading annual market research publication in the disk drive industry. It has been considered as a highly reliable and complete source of disk drive industry data (Josh Lerner 1997, Christensen et al 1998). We compose all imitators into one imitator, and use a composite imitator as representative imitator in estimation as did in Suslow (1986) and Putsis and Dhar (1999). We denote the innovator and imitator hereafter as j, and j2, respectively. Data on the imitator's product attributes are accordingly computed by averaging all imitators' product attributes at each year. Imitator's sales and market share are then derived by subtracting innovator's sales and market share from the sales and market share of 3.5-inch floppy disk drive market. Data on the annual average price of innovator and imitator's products at each year are computed by dividing their annual dollar sales by annual sales volume. Disk TrendReport records the attributes for each disk drive that is introduced by each manufacturer in each year. They are listed according to three categories: (i) disk capacity, which includes total capacity, density per track, data surface per spindle, and tracks per surface; (ii) drive performance, which includes the time spent from track to track, settling time, average rotational delay time and data transfer rate; and (iii) disk drive features, which includes height, width and depth of each disk drive. By analyzing the attributes in each category, we found that the attributes in first and second categories are highly correlated. To reduce the problems of multicollinearity, we conducted step- We would like to thank James Porter for his generous offer ofthe series books of Disk Trend Report. We would like to thank James Porter for his generous offer of the series books of Disk Trend Report. wise regression to choose the most influential attribute variables. Our regression analyses identify five attributes that contribute to the most of variances of price and sales: disk density (Den), track to track time (TT), height (H), width (W) and depth (D). These attributes are used in our utility function (2.1). Following literature on network effects (e.g. Brynjolfsson and kemerer 1996), the consumer installed-base in demand equation that captures the network effects is measured as the cumulative sales of whole 3.5-inch floppy disk drive market at previous period. We also measure the supporting firm base using the total number of supporting firms at previous period. The potential market size (AM,) in the demand equation is measured as the total of the floppy disk drive market sales at each time from 1983-1998. The total of the floppy disk drive market sales is measured by adding together all sales from different sizes of floppy disk drives that were introduced in FDD market. For example, it can be sales from 8-inch, 5.25-inch, 3.5-inch, 4-inch, 3.25-inch, 3-inch and high density FDD. We use all of the above five product attributes and one time trend variable in marginal cost equation. Table 2-1 provides the descriptive statistics of our data. From Table 2-1, we can see that the difference between innovator and imitator's product mainly lies in disk density. And innovator charges higher average price than imitator. Table 2-1. Descriptive Statistics Innovator Imitator t-statistic of Mean Mean Std. Mean Std. Differences Sales (Million) 6.22 4.17 33.94 33.01 Average Price 52.82 34.87 48.07 27.68 Density 135 0 131.73 5.87 2.23** Track to Track 4.23 2.73 3.74 0.58 0.70 Height 1.06 0.34 1.01 0.23 0.49 Width 3.96 0.048 4.05 0.24 1.54 Depth 5.64 0.25 5.69 0.45 -0.36 Installed-base (Million) 142 170.20 Note: **: p<0.05. 2.3.3 Identification We assume that the product attributes xjk are exogenous and orthogonal to the error terms (4, and ,, ); price and sales are endogenous and correlated with the error terms (J, and ,,J). This identification assumption is reasonable and commonly used in the simultaneous equation estimation (e.g. Sudir 200 b). We use product attributes, their squares and the products of innovator's attribute and imitator's attributes as instruments. Since innovator's density remained the same over 16 years, we exclude it as one instrument variable. So we have a total of 23 instruments in the demand equation: ten exogenous variables (a constant, four innovator's attributes, and five imitator's attributes), nine exogenous variables' square, and four products of innovator's and imitator's attributes. We also apply these instruments to the two competitive interaction equations. In the baseline model, we need to estimate 9 parameters in the demand equation (constant, coefficients of five product attributes, price, installed-base and supporting firm-base), 7 parameters in the marginal cost equation (constant, coefficients of five product attributes and time trend), and 2 conjectural variation parameters in the time varying first order conditions if using conjectural variation approach. Therefore, our baseline model is identifiable. 2.3.4 Empirical Results of Baseline Model In the conjectural variation approach, we simultaneously estimate the demand equation (2.5), the marginal cost equation (2.7) and the competitive interaction equation (2.8). The competitive structure is inferred by examining the signs of the two conjectural variation parameters, ) and ), as discussed in Section 2. In the menu approach, we estimate the structural equations under four alternative assumptions of competitive structures separately, i.e., for each assumed competitive structure, we simultaneously estimate the demand equation (2.5), the marginal cost equation (2.7), and one of corresponding competitive interaction equations given in (2.9)- (2.14). We then infer the competitive structure by selecting the best-fitted model based on the results of the Vuong test.5 The three-stage least square (3SLS) method is used in the estimation of both approaches. Table 2-2 provides the results of the two approaches. As shown in Table 2-2, the results of CV approach show that both conjectural variation parameters, \ and ., are not significantly different from zero, indicating a Bertrand competition structure. The results of menu approach show that the cooperation model has a maximum log-likelihood value (-15.09), while the log-likelihood values for the other three alternative competitive structures (Bertrand, Innovator-Led leader- follower, and Imitator-Led leader-follower) are -24.00, -27.97, and -28.75, respectively. The results of the Vuong test statistics (see Table 2-2) indicate that the cooperation model is significantly better than the other three alternatives (p<0.01 for all three comparisons). Hence, our empirical analysis of the baseline model under the assumption of a constant competitive structure leads to different conclusions about the nature of competitive interaction in the studied market-Bertrand based on conjectural variation 5 The Vuong test statistic is defined as V In (p q) where In(f) and In(g) are the log- likelihood values of the two non-nested models, and p and q are the number of parameters estimated in each model, respectively. V is distributed as N(0,1) If V > 0 and V > ritica allue then the model corresponding to f is rejected in favor of the model corresponding to g, and vice-versa (Kadiyali et al. 2000). approach but Cooperation based on menu approach. This indicates that the competitive structures derived from the two approaches are inconsistent when the baseline model is used. Table 2-2. Baseline Model: Estimated Competitive Structure Conjectural Variation Approach Conjectural Variation Parameter S22 The log-likelihood n NS 0 2NS Estimation .-33.22 (1.24) (0.59) Estimated Bertrand Bertrand Competitive Structure Menu Approach Innovator-Led Imitator-Led Alternative Models Bertrand Invar-Led Itator-Led Cooperation Leader-Follower Leader-Follower Log-likelihood -24.00* -27.97* -28.75* -15.09 (Vuong Statistics) (2.23) (3.22) (3.42) (-) Estimated Estimated Cooperation Competitive Structure Coope Note: : p<0.01; N: p>0.1 For conjectural variation approach, the numbers in parentheses are t-statistics. For menu approach, the numbers in parentheses are Vuong statistics. The critical value of Vuong statistics at p=0.05 is 1.64. The Vuong statistics are significant atp<0.01 in all cases (See footnote 7 for discussion of Vuong test). The inconsistency found in Table 2-2 can be the results of model misspecification. As we argued, network effects create two types of interdependences: (1) the interdependence among consumers (i.e., the value of a product to consumers is affected by the number of consumers bought the same product), and (2) the interdependence between consumers and firms (i.e., the value of a product to consumers is affected by the number of the firms supporting the same technological standard). However, as the previous empirical studies of network effects, our baseline model only considers the installed-use-base effect, but not the supporting-firm-base effect. In the next section, we modify our baseline model by incorporating the supporting- firm-base effect into consumer utility function, and modeling the competitive structure to be a function of (1) installed-user-base and supporting-firm-base, and (2) time trend. 2.4 Modified Model and Empirical Results 2.4.1 Modified Model Demand. We modify the baseline model by incorporating a new variable, supporting-firm-base, into consumer behavior. Let H, denote the number of firms supporting the innovator's technological standard at time t (i.e., a measurement of supporting-firm-base). We formally incorporate the supporting-firm-base into the consumers' utility model: u"t = o + xjktfk --Pjt +yNt 1- + Nt 71-1 1-H,- + -tT + 2Ht T+ ,+t = u jt + t (2.15) k Equation (2.15) assumes that consumer utility is affected not only by the installed- user-base, as assumed in the existing literature and in our baseline model, but also by the supporting-firm-base. Furthermore, since both the installed-user-base effect and the supporting-firm-base effect may change with time, we include two interaction terms, (N, 1T) and (H, T), in Equation (2.15) to capture the possible interaction effect, where T donate time trend. To examine how the number of supporting firms may be affected by price competition, we also models the number of supporting firms as a function of imitator's price and the number of supporting firms at the previous period: Ht = (0Ht- + q2Ht-21 P2t-1 +1 (2.16) The quadratic term in Equation (2.16) is intended to capture the saturation effect. Strategic Interactions. To test the possible stage-dependent competitive interactions, we relax the assumption of a fixed competitive structure by allowing the conjectural variation parameters in equation (2.8) to be timing-varying. Specifically, we define the conjectural parameters, A1 and 2 to be two types of functions. We first define the conjectural variation parameters to be a function of installed-user-base and supporting-firm-base, as shown in Equation (2.17): jt = 2J +^NIt- +2 2Ht 1, j = 1,2, (2.17) and second to be a function of time trend T, as shown in Equation (2.18): Jt = 2J + ,T + j22, j = 1,2 (2.18) Equation (2.17) allows us to examine how the installed-user-base and supporting-firm- base directly influences the competitive structure directly, and whether the competitive structure is stage-dependent, while equation (2.18) provides us an overall pattern of the stage-dependent competitive structure over time. 2.4.2 Empirical Results: Time-Varying CV Approach We first estimate the modified model using the time-varying conjectural variation approach by modeling the competitive structure to be function of installed-user-base and supporting-firm-base as well as function of time trend, and then validate the key findings using stage-based menu approach. The structural equations for CV approach include the demand equation (2.5), the marginal cost equation (2.7), the supporting-firm-base equation (2.16), and the strategic interaction equations (2.8). Note that the uJ, in (2.5) is now derived from (2.15) rather than from (2.1) as in Section 2. Also, 2,, in (2.8) is now a function of installed-user-base and supporting-firm-base as defined in (2.17) or function of time trend as defined in (2.18) rather than a constant as in Section 2. To derive the best fitted model, we incorporate four time dummy variables in equation (2.17). The four time dummy variables are defined as dl = 1 if the time period is at years 1 to 4, and otherwise 0; d2 = 1 if the time period is at years 5 to 8 and otherwise 0; d3 = if the time period is at years 9 to 12 and otherwise 0; d4 = 1 if the time period is at years 13 to 16 and otherwise 0. The estimated function of conjectural variation parameters is then written as ~, =J 20 +2 J(N*dl)+-j2(N*d2)+2j3(N*d3)+2j4(N*d4) (2.17') + j5 (H *dl) + 6(H*d2) +j7 (H *d3) +2 (H *d4) where j=1 and 2. The results of the modified model using time-varying CV approach are given in Tables 2.3-2.4. Impact of Supporting-Firm-Base on Consumers. As shown in Table 2-3 (see the results of Demand Equation on the top of Table 2-3), the installed base is positively related with consumer valuation (p<0.01), suggesting a positive installed-user-base effect. This finding is consistent with previous studies (e.g. Brynjolfsson and kemerer 1996). More importantly, our result shows that the number of the supporting firms is also positively related to consumer valuation (p<0.01), suggesting a positive supporting-firm- base effect. Table 2-3 also shows that the parameters of both interaction terms in the demand equation are significant, suggesting that both the positive installed-user-base effect and the positive supporting-firm-base effect vary across time. Impact of Price on Supporting-Firm-Base. Table 2-3 shows that the number of supporting firms is positively related to the price of imitators (see the results of Supporting-Firm-Base Equation in the middle of Table 2-3). This result combined with the positive effect of supporting-firm-base on consumer valuation discussed above suggests that the supporting-firm-base effect reduce firms incentive to aggressively compete on price, because a high price encourage more firms to adopt the same technological standard and in turn increase the value of the product to consumers. Table 2-3. Revised Model 1, Time-Varying CV Approach Demand and Marginal Cost Equations Demand -61.78** Intercept (2.54)a Den 14.14** Den (2.91) T -0.36 (-0.77) 2.31 Height .3 (1.36) Width 28.31** Width (5.36) Deh -25.28*** Depth (-5.91) Time Trend Marginal Cost 3877.86*** (3.89) -278.40 (-1.56) -46.21** (-2.42) 368.85* (3.36) -1919.24* (-2.30) 88.62 (0.73) 6.22* (1.91) Price Installed-base Installed-base* Time Firm-base Firm-base Time Firm-Base at Previous Period Square of Firm-Base at Previous Period Imitating Firm's Price at Previous Period 2,0 (constant) A1 (at first period) K22 (at second period) 2,3 (at third period) 2.4 ( at fourth period) -0.075. (-5.80) 0.62E-07** (5.03) -0.38E-08** (-5.15) 0.17* (4.28) -0.032*** (-5.53) Supporting-Firm-Base Equation 1.23*** (5.27) -0.018* (-1.70) 0.055*** (3.51) Conjectural Variation Parameters Innovator 0.57 (1.52) User-Base -0.13E-05 (-2.00) -0.46E-07 (-3.94) -0.35E-08 (-3.00) -0.14E-08 (-1.91) Firm-Base 0.24 (2.02) 0.13 (4.67) 0.04 (1.67) 0.03 (1.62) Note: The number in parentheses is t statistic. :p<0.01. :p<0.05. Imitator 0.22 (0.28) User-Base -0.33E-04 (-0.33) -0.71E-07 (-2.67) -0.94E-08 (-3.91) -0.29E-08 (-1.69) Firm-Base 5.51 (0.33) 0.33 (4.10) 0.13 (2.84) 0.07 (1.38) Impacts of User-Base and Firm-Base on Competitive Structure. The estimation of the impacts of user-base and firm-base on competitive structure are given in the bottom of Table 2-3 (see Time-Varying CV Parameter Estimation). As shown in Table 2- 3, for both innovating and imitating firms, the coefficient of user-base across four periods are all significantly negative, while the coefficient of firm-base across four periods are all significantly positive. The different signs of the impacts of user-base and firm-base imply that installed-user-base motivates competitive behavior, while the supporting-firm-base motivates cooperative behavior. Stage-Dependent Competitive Interaction. The estimations of time-varying CV parameters where the conjectural variation parameter is defined to be a function of time trend are given in Table 2-4. As show in Table 2-4, for both innovating and imitating firms, the constant, 2j0, is positive (p<0.01), the linear term, 2J,, is negative (p<0.01), and the quadratic term, 2J, is positive (p<0.01). These results suggest that (1) firms behave cooperatively at the beginning of the new product introduction, and (2) the competitive structure may vary across different stages of product life cycle. To graphically illustrate the pattern of competitive behavior in the market examined, Figure 1 shows how the conjecture variation parameters, 2J, (/=1,2), given in (2.18) change with time when the estimated time-varying conjecture variation parameters given in Table 2-4 are used. Table 2-4. Revised Model 2, Time-Varying CV Approach Demand and Marginal Cost Equations Demand -54.16 (-5.23)a 12.80 (5.81) 1.75 (4.56) 3.21 (4.64) 27.57*** (6.85) -26.90 (-7.37) Intercept Den TT Height Width Depth Time Trend Price Installed-base Installed-base* Time Firm-base Firm-base Time Firm-Base at Previous Period Square of Firm-Base at Previous Period Imitating Firm's Price at Previous Period Log-likelihood value -0.051 (-4.21) 0.44E-07 (4.86) -0.28E-08 (-5.22) 0.044* (2.34) -0.016 (-3.45) Supporting-Firm-Base Equation 1.017 (11.99) -0.020 (-5.35) 0.14 (14.60) Conjectural Variation Parameters Innovator 4.56 (7.55) -0.62 (-5.36) 0.023 (3.96) Comparison: Modified vs. Baseline Model Modified model Baseline model -13.88 -33.22 Marginal Cost -20413.3 (-1.20) 4097.15 (1.19) -5.35 (-0.14) -198.08 (-2.39) -1073.70** (-2.18) 1031.99 (2.56) -3.25 (-1.91) Imitator 11.18 (5.36) -1.80 (-4.90) 0.075 (4.67) Log-likelihood ratio 38.68 Note: The number in parentheses is t statistic. : p<0.01. : p<0.05 The loglikelihood ratio has %2 distribution, the critical value %2 =18.30 at p<0.01 with d.f.=10. As shown in Figure 2-1, both conjecture variation parameters, 4~ and ,2, follow a quadratic curve. Both parameters are positive in the early years, indicating a cooperative market structure in the early stage of standard development. Then both 4 and 2, decrease to near zero at around year 12, indicating a competitive market structure in the middle stage. In the late years, both ), and )2 increase to be positive again, indicating that the market moved toward a cooperative structure again in the late stage. Hence, Figure 1 shows a very interesting three-stage pattern of competitive structure in floppy disk drive market: cooperation in the early stage, competition in the middle stage, and cooperation again in the later stage. 12 10 - 10 d 8 E S6 2 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Time -*-lamdal ---lamda2 Figure 2-1. The Changes of Conjectural Variation Parameters with Time To evaluate the superiority of the stage-dependent competitive structure model over the constant competitive structure model, we test if the former fits data better than the latter by comparing the log-likelihood values of the conjectural variation models given in Table 2 (-33.22) with that given in Table 3 (-13.88). The log-likelihood ratio test6 suggests a significant superiority of the stage-dependent competitive model (%2 = 18.30, p<0.01, d.f. =10). To sum up, the empirical analysis of the modified model using time-varying conjecture variation approach leads to three key findings: (1) a positive effect of supporting-firm-base on consumer valuation, (2) a positive effect of price on the number of supporting firms, and (3) a three-stage competitive interaction pattern: cooperation- competition-cooperation. To further validate these key findings, we apply a stage-based menu approach to the modified model. 2.4.3 Validation: Stage-based Menu Approach Different from CV approach that captures the stage-dependent competitive interaction via a set of time-varying conjecture variation parameters, when stage-based menu approach is used to estimate stage-dependent competitive structure, one needs to first decide how stages are divided and then apply menu approach to data in each stages. Based on results from CV approach (see Figure 2-1), we divided the data set into three stages: an early stage (year 1-5), a middle stage (year 5-12), and a later stage (year 10- 16).7 For each stage, menu approach is used to estimate all four alternative models (i.e., Bertrand, Innovator-Led leader-follower, Imitator-Led leader-follower, and Cooperation), 6 We use the log-likelihood ratio test to compare the goodness of fit between these two nested models. The log-likelihood ratio is 38.68, which is larger than the critical value of X2 = 18.30 at p<0.01 with freedom degree= 10, showing that our time-varying conjectural variation model is significantly better than the conjectural variation model with single competitive structure. 7 Duo to the limited data available, these stages partially overlap with each other so that there are sufficient data points within each stage to allow models to be identifiable and to use Vuong test statistics to select the best-fitted models. While different possible cutoff times were considered, our final definition of stages is consistent with the curves observed in Figure 1 and Vuong test statistics are significant in selecting the best-fitted models. and the best-fitted model in each stage is selected using Vuong test statistics. The results of model selecting test for each stages are given in the top part of Table 2-5. As shown in Table 2-5, according to the Vuong statistical test, the best-fitted model is Cooperation in the early stage, Innovator-Led Leader-follower in the middle stage, and Cooperation again in the later stage. All other alternative models are rejected. Hence, menu approach leads to the same three-stage pattern of competitive interaction as CV approach (i.e., cooperation-competition-cooperation). In other words, when we use the baseline model and assume a constant competitive structure, the two different approaches lead to inconsistent conclusions (see Table 2-2). However, when we use the modified model and allow the competitive structure to be stage-dependent, the two approaches lead to consistent conclusions. Table 2-5 also shows the results of parameter estimations of the best-fitted models in each stage. Note that the stage-based menu approach allows not only competitive structure to vary across stages, but also parameters in all structure equations vary across stages. As a result, the estimation results using menu approach reveal more details of the market dynamics, which will be discussed in more details later. Nevertheless, as in the estimation of CV approach, the results of demand model by menu approach show a positive effect of the number of supporting firm on consumer valuation (in the early and later stage), and the results of supporting firm model by menu approach show a positive relationship between the number of supporting firms and price. Hence, our three key findings derived in CV approach are also supported by the menu approach. Table 2-5. Revised Model 3, Stage-Based Menu Approach Model Selection Test Innovator-Led Imitator-Led Best Best-fitted Stage Bertrand Leader- Leader- Cooperation model model Follower Follower Early 32.52* 32.66* 32.57* 40.07 (Year 1-5) (3.37) (3.31) (3.35) (---) Innovator-Led Middle 21.58* 45.03 18.72* 35.84* n r- Leader- (Year 5-12) (8.29) (---) (9.30) (3.25) ollower Follower Late 3.12* 19.56** 3.75* 50.04 (Year 10-16) (17.71) (11.50) (17.47) (---) Parameter Estimation: Demand Model (The Best-fitted Model) Parameter Early Stage Middle Stage Late Stage (Cooperation) (Innovator-Led (Coorperation) Leader-Follower) -26.18 -258.92 527.20 Intercept (-0.99) (-1.94) (1.72) 2.05 52.49 -98.82 (0.34) (1.98) (-1.57) 2.71 1.06** 0.98 (1.45) (2.39) (0.58) -2.50 -1.67** 17.64** g (-0.50) (-1.87) (7.32) 15.72 38.45 24.76 (0.58) (9.81) (5.08) -7.19 -29.85 -43.01 h (-0.35) (-10.22) (-5.77) -0.024 -0.0098 -0.081 Pnrce (-1.40) (-1.65) (-2.40) -0.92E-07 -0.29E-08 0.35E-08 Installed-base (-1.04) (-1.54) (1.82) 0.19 -0.032 0.071* Firm-base (2.88) (-0.92) (2.42) Parameter Estimation: Supporting-Firm-Base Model Parameter Early Stage Middle Stage Late Stage Firm-Base at Previous 0.74 1.18 0.83 Period (1.28) (10.18) (7.19) Square of Firm-Base at -0.0081 -0.018" -0.038* Previous Period (-0.42) (-4.91) (-3.85) Imitating Firm's Price 0.14 0.061 0.39 at Previous Period (4.03) (1.87) (4.48) Note: Sp<0.01, p<0.05. For model selection test, the numbers in parentheses are Vuong statistics. The critical value of Vuong statistics atp=0.05 is 1.64. For parameter estimations, the numbers in parentheses are t-statistics. 2.4.4 Discussion of Key Findings First, previous empirical studies have shown a positive relationship between installed-user-base and consumer valuation (e.g. Nair, Chintaguta, and Dube 2004). Such a positive relationship has been explained by a positive network effect: the larger the user base, the higher the value of the product. By simultaneously examining the installed- user-base and supporting-firm-base, we find that both can have a positive effect on consumer adoption decision. More interestingly, our stage-based analyses show that the relative importance of installed-user-base effect and supporting-firm-base effect varies across stages of product life cycle. For example, in the early stage of 3.5-inch floppy disk drive, the consumer valuation is positively affected by the supporting-firm-base but not by the installed-use-base (see Table 2-5). While this result is not intuitive, it may result from the high uncertainty faced by consumers in the presence of standards competition. In the early stage of product introduction, 3.5-inch disk drive faced intense competition from multiple incompatible technological standards including the old 8-inch and 5.25- inch floppy disk drives as well as the new 3-inch, 3.25-inch and 4-inch floppy disk drives. Since different types of drives were not interchangeable, the adoption utility of a given type of disk drive depends on the outcome of the standards competition. While it is difficult for early buyers to predict if the new 3.5-inch floppy disk technology would be widely adopted by the market, the number of firms supporting the Sony's new innovation in the early stage could be useful information to consumers in evaluating the odds of the new technology because the larger the number of supporting firms, the more likely that the new technology will become industry standards. The fact that consume valuation is affected by the supporting-firm-base but not by the installed-user-base in the early stage suggests that at the beginning of 3.5-inch disk drive introduction, consumers were more concerned about the outcome of ongoing standards battle but less about the current user base of the drive. While conventional wisdom has emphasized the importance of building consumer base in markets with network effects, our finding highlights the importance of simultaneously considering both the installed-user-base and supporting- firm-base. As we will discuss next, the installed-user-base effect and supporting-firm- base effect can have opposite implications on price competition. Second, our empirically analyses show a positive relationship between the number of supporting firms and price. This result implies that the positive supporting-firm-base effect discussed above reduces firms' incentive to cut price because a high price encourages more firms to adopt the innovator's technological standards, which in turn increases consumer valuation. Note that it is commonly agreed that the positive installed- user-base effect motivates firms compete more aggressively because a low price encourages more consumers to adopt the new product, which in turn increases consumer valuation. Take the two effects together, our findings suggest that while both supporting- firm-base and installed-user-base increase consumer valuation, they have opposite impacts on firms' competitive behavior: the former motivates cooperative behavior but the latter motivates competitive behavior. Hence, the nature of firms' strategic interaction is expected to be jointly determined by the two effects and may change across different stages of its product life cycle. Third, our most interesting finding is the three-stage competitive interaction pattern found in 3.5-inch floppy disk drive market: firms cooperate in the early stage, compete in the middle stage, and then cooperate again in the late stage. This pattern can be explained by the relative importance of the two effects (i.e., installed-user-base effect and supporting-firm-base effect) to consumers in different stages, and the opposite impacts of the two effects on firms' competitive behavior. In the early stage of 3.5-inch disk drive introduction, 3.5-inch disk drive faced intense competition from several competing standards (see Figure 2-2), which drawn consumers' attention toward the supporting-firm-base rather than the current installed-use-base (see Table 4). Firms adopted cooperative behavior because cooperation helps but competition obstructs building up supporting-firm-base. In 1990 (i.e., 9 years after the introduction of 3.5 floppy disk drive), the sales of 3.5 floppy disk drive have surpassed the sales of all alternative technologies (see Figure 2), indicating that 3.5 floppy disk drive had established its technology standard position. In the next few years (the middle stage), neither installed-user-base nor supporting-firm- base have a significant effect on consumer valuation (see Table 4), implying that in this stage, the product is evaluated largely based on its performance rather than by network effects. As a result, there would be little incentive for firms to attract more imitators; rather, they had strong incentive to compete aggressively to achieve as much profit as possible. In 1995 (12 years after the introduction of 3.5-inch disk drive), new generation technology, high-density floppy disk drive, enters the market. In this late stage, consumers were concerned about both the installed-user-base and supporting-firm-base, perhaps because both indicate how likely the existing technology would be replaced by the new generation technology. Since the supporting-firm-base was more important to consumers than the installed-user-base (see Table 2-5), firms adopted cooperative behavior again to discourage manufactures from abandoning the existing standard. 120000.0 100000.0 0 80000.0 o 60000.0 40000.0 20000.0 0.0 ..--" , - 1983 1985 1987 1989 1991 1993 1995 1997 Time -- 3.5 inch 8+5.25 inch High Capacity Figure 2-2. Sales: 3.5 inch Floppy Disk Drive vs. Competing Products 2.5 Conclusion In this paper, we use the new empirical industrial organization framework to examine price competition between innovating and imitating firms. Different from the existing research of network effects, we consider not only an installed-user-base effect but also a supporting-firm-base effect on consumer utility. To examine the impact of the supporting-firm-base on intra-standard competition, we model the interrelationship of the number of imitating firms with the prices of innovating and imitating firms. To allow our models to capture the stage-dependent price competition structure, we introduce time- varying parameters to the conjectural variation approach and validate our estimation results using menu approach. Our results provide empirical evidence that markets with network effects generate both a positive consumer-installed-base effect and a positive supporting-firm-base effect (on consumer utilities). However, these two effects have opposite impacts on firms' competitive behaviors. To increase consumer installed-base, firms need to keep price low, which motivates firms to compete. In contrast, to encourage other firms to adopt the innovator's technology, firms need to keep price high, which motivates firms to cooperate. Furthermore, these two opposite effects impact the price competition between firms differently at different stage. Our results using both approaches suggest that the innovator and imitators have more incentive to encourage more manufactures to support a new standard in its early stage and discourage existing manufacturers from abandoning an established standard in its later stage when facing the challenge of next generation technology. As a result, firms are more likely to cooperate in the early and later stage, and compete on price in the middle stage. This paper makes several contributions to the literature of network effects. First, we examine an effect overlooked in the literature, the supporting-firm-base effect. Our results not only provide empirical evidence supporting a positive relationship between the number of firms adopting the same technological standard and consumer's valuation, but also suggest that the supporting-firm-base effect works in an opposite direction compared with the well-known installed-user-base effect in terms of their impacts on intra-standard competition. Second, our study provides empirical evidence of a three-stage pattern of competitive interaction in intra-standard competition. Our results suggest such a pattern in the floppy drive market over the period studied: firms behaved collectively in the early stage, competitively in the middle stage, and collectively again in the late stage. Such a stage-dependent competition pattern is new to the literature on network effects and raises interesting theoretical and empirical questions for future research on the strategic implications of stage-dependent intra-standard competition. Finally, this paper illustrates the use of complementary approaches to evaluate different structural models and estimate stage-dependent competitive structures. This approach can be applied to other situations where a constant competitive structure is questionable. Overall, this study fills a gap in the network effects literature by providing empirical evidence on patterns of intra-standard competition. Our findings raise some important new theoretical and empirical issues for future research on markets with network effects and standards competition. However, since our study only uses data from the floppy disk drive industry, the generality of our findings needs to be tested in other industries with network effects. CHAPTER 3 DYNAMIC ANALYSIS OF NEW TECHNOLOGY DIFFUSION IN MAREKTS WITH COMPETING STANDARDS 3.1 Introduction The diffusion of new technology has been extensively studied since the traditional diffusion model was published (e.g., Bass 1969, Mahajan and Muller 1979, Norton and Bass 1987). Recently, the rapid development of information technology and the digital revolution attract researchers' attention again to the diffusion of new technology, especially in markets with network effects. While studies on the adoption of the new technology such as the ATM machine (Saloner and Shepard 1995), Compact Disc (Gandal, Kende and Rob 2000), DVD player (Inceogle and Park 2003, Dranove and Gandal 2003), and ACH electronic payments system (Gowrisankaran and Stavins 2004) demonstrate the importance of the network effects (i.e. installed-user-base effect) on the new technology diffusion, they all focus on the new technology diffusion at category-level. There is NO empirical research examines the standard diffusion and how the competition between incompatible technologies impacts the new technology diffusion at three levels--category-level, standard-level and brand-level. Understanding the new technology diffusion at standard-level and the impacts of standards competition on the new technology diffusion at three levels is extremely important because the markets with network effects and standards competition exhibit certain unique characteristics that do not apply to the traditional markets and have not been addressed in the literature of new technology diffusion and network effects. First, in markets with network effects and standards competition, the new technology diffusion at three-levels is interdependent: the standard diffusion impacts the market potential for whole category diffusion and the consumer's evaluation for an individual brand, while the category diffusion impacts also the market potentials for each standard's diffusion and the brand diffusion comprises of the standards' diffusion. Studying the interdependence of the new technology diffusion at three-levels requires the marketing scholars to integrate the category diffusion, standard diffusion and brand diffusion into a systematic framework. However, most of the traditional diffusion models (e.g. Bass 1969, Bass, Krishnan and Jain 1994, Horsky 1990) focus only on the category diffusion without considering the impacts of standards' and brands' diffusion on each other. Though there is little diffusion research in recent years that has proposed models to incorporate the competition impact on diffusion (e.g. Norton and Bass 1992, Mahajan and Muller 1996), there is NO research addressing the standard diffusion in a systematic framework. Second, standards competition impacts the new technology diffusion differently from the competitions among competing brands in the traditional markets. The differences exist in two aspects: (1) unlike the traditional markets where the competitions among competing brands may expend the whole category diffusion positively, the standards competition in markets with network effects may impact the market potential negatively. This is because the high uncertainty that consumers have on which standard will dominant market deters consumers from making choice immediately when there is no single technological standard dominating the market. (2) Standards competition generates different word of mouth effects on the new technology diffusion at the stage when the incompatible technologies are competing for market dominance from the stage when the market is dominated by one of the competing technologies. Recent studies about the impacts of network effects on new technology diffusion has revealed the positive impacts of network effects on a particular technology standard (e.g. Saloner and Shepard 1995, Gandal, Kende and Rob 2000, Park 2003 and Ohashi 2003) or the negative impacts of network effects on competing standard (e.g. Dranove and Gandal 2003), but there is NO empirical works demonstrating how the standard competition impacts the new technology diffusion on the market potential and consumers' word of mouth effects. Third, literature has suggested (e.g. Katz and Shapiro 1985 and Shapiro and Varian 1999) that consumer expectation is a major factor in consumer decisions about whether or not to purchase a new technology, and the management of consumer expectation on network effects is a critical factor in standard competition and hence the technology diffusion, but no empirical works demonstrate to managers on how consumers form their expectations on network effects, especially on the future installed-user-base and supporting-firm-base. This study sets up a systematic framework by integrating the new technology diffusion at three levels. Specifically, this study builds a choice-base diffusion model by incorporating (1) the standard competition---measured as the market share differences between the competing standards, and (2) two important network effect factors: installed- user-base and supporting-firm-base of competing standards into a baseline model and a dynamic model. Our choice-based diffusion model involves three components: (1) a consumer utility model that captures the influences of price, product attribute, and network effects-measured as installed-user- base and supporting-firm-base of two competing standards, (2) equations that derive the market share of two incompatible platforms based on the estimated nested-logit model, and (3) equations that derive the dynamic diffusion at category-level, standard-level and brand-level. Different from the baseline model that only captures the current network effects---current user-base and firm-base of two competing standards, our dynamic model captures the consumer expectations for the future user-base and firm-base of competing technological standards, and thus involves using a dynamic programming method to estimate the market share equations for each competing standard and brand. Our model is estimated using data from the VCR industry. The competition in the home VCR market is one of the best-known examples of technology standard competition between two incompatible format Betamax and VHS (Park 2003). Though the Betamax format was first introduced by Sony in February 1976, VHS format, which is introduced one and half year later by JVC in 1977 and incompatible with the Betamax format, was able to surpass the Betamax in the early years of 1980s and eventually dominated the U.S. VCR market in the late years of 1980s. Examining this market would provide valuable insights on how the inter-standard competition impacts the new product diffusion at category-level, standard-level and brand-level. The estimated parameters reveal several interesting results. First, we find that while the installed-user-base from the compatible standard increases the consumer's evaluation to a new product, which is consistent to the literature on network effects, the installed- user-base from competing technology standard decreases the consumer's evaluation to a new product, which is new to literature on network effects, and demonstrates the evidence for one of the important impacts of standard competition on new technology diffusion. Second, we find the empirical evidence that standard competition impacts new technology diffusion complimentarily. This can be seen from two results: (1) the supporting-firm-base from both of two competing standards---VHS and Betamax positively influence the consumer's evaluation on each brand, and (2) the installed-user- base from both of two competing standards generates positive word of mouth effects on the category diffusion as well as the standard and brand diffusion at the early stage of standard competition when none of the two competing standards dominates market. Third, we also find results that the network effects impact the market potential dynamically. While the installed-user-base from two standards both positively impact the new technology diffusion at the early stage of standard competition, the installed-user- base of Betamax turns to impact the diffusion of new technology at three levels negatively after the technology standard VHS dominates the market. Finally, our results suggest that network effects and standard competition not only impact the diffusion on individual consumer's evaluation to each brand, but only on the market potential complementarily and dynamically. Similarly with the word of mouth impact of installed-user-base on new technology diffusion, the supporting-firm-base from both of two standards impact market potentials seems positively at the early stage of standard competition, but the supporting-firm-base of Betamax has no impact on the market potential later on when the VCR market is dominated by VHS format. This study extends research on network effects and provides important implications to markets with network effects by explicitly modeling and estimating the impacts of inter-standards competition on new technology diffusion. It also advances understanding of the formation of consumer expectation for the future installed based of competing technological standards-a very important but under-explored research area. Our paper is organized as follows. We first introduce our baseline model and dynamic model in the section 3.2 and 3.3. Then we present our preliminary results in section 3.4. We discuss our results and future research directions in section 3.5. 3.2 The Baseline Model We start with analyzing the new product diffusion at three levels: category, platform and brand levels by incorporating the impact of network effects and standard competition into (1) the equation of consumer's evaluation of each brand, (2) the market potential equation and (3) the diffusion equations at three levels. The network effect variables that are incorporated in the baseline model are the current installed-user-base and the current supporting-firm-base. The dynamic model by considering consumer expectation aboutfuture network effects will be introduced in the next section. 3.2.1 Consumer Utility Equation We use the nested logit model to estimate the market shares. Let i denote consumer and j denote brand (j = 0,1, 2,.-.,J ), where outside option is denoted as j = 0. Different technology standard groups are denoted as g (g = 0,1,2, G). The outside option j = 0 is the only product in group g = 0. Further, denote the set of brands in each group g as 3, ( 3, = J) and denote the value of brand j as g 5t, = A +i AkXkt aP]t k g-1 (3.1) +7 AgNjgt + jgFgt-1 + Igrg t-1 + gF)g', g +7t ; ZMSD + ,t g' g' g =1 where xJ, and p,, denote the product attribute k and price for brand j; MSD1 denotes the market share difference between the compatible technology standard of brand j and other competing technology standards g'. We incorporate four network effect variables: Njg, N jg't-, Fjgt-i and Fjgt-i in the utility function to measure the installed- user-base and the supporting-firm-base for each brand's compatible technology standard g and competing technology standard g', respectively. Then the utility for consumer i is accordingly modeled as U,t = 6jt + ,;g, + (1 U)E,. (3.2) where s,, is an identically and independently distributed extreme value; and [,, + (1 -o-)s, ] is also an extreme value random variable (i.e., double exponential distribution)). 3.2.2 Market Share Equations If brand j is compatible to technology standard g, then the formula for the market share of brand j as a fraction of the total standard share (see Berry 1994) is s ,ji = [e /, 5(I-)]/Dg (3.3) where Dg = e Similarly, the market share for the technology standard g is J3, D( -) s = (3.4) g Thus, the market share for brand j as a fraction of all brands regardless its technology standard can be derive as e 5j1/(1-(T) s = ~ D(3.5) 1 D, D 9) g With the market share of outside options so -- and g ln(D ) = [ln(s,) ln(so)] /(1- a), we can then derive a simple analytic equation for mean utility levels as ln(s) ln(s)= ,, + o ln(s,, ) (3.6) and the parameters in (3.6) can then be obtained by instrumental estimation methods. 3.2.3 Diffusion Equation at Three Levels To derive the diffusion equation at three levels, we need model the market potential first. Instead of modeling the market potential as TV households as in literature (e.g., Park 2003 and Ohashi 2003), we model it as a function of the average price of all products in market, product quality and network effects at different standard competition stages. Define a parameter 0 as 0 = (o (ap)+ k (max(%zk ))+ I o ig,t 1 + ,g1 g,tl -MSD + goFgt- MSD k g where ap represents the average price of all products in market; max(xjk) is the maximum level of each product attribute among different brands. The Ng,,_i -MSD and Fg,t,- -MSD denote the interactive terms of network effect variables---installed-user-base and supporting-firm-base with the market share difference. These interactive terms allow us to capture the dynamic influences of network effects across different stages of standard competition. The market potential is then modeled as a function of TV households (TVHH) and the defined parameter 0: Mt =TVHH-exp() (3.7) Accordingly, we model the diffusion equations for the whole category as -Xt = (M,-X, )( + gONg, -i ^glNg,,t -MSD) (3.8) g where the industry potential M, is defined in (3.7). Similarly, each standard's and each brand's diffusion is given by Xg,, -,,,1 =Mgt(,,A o +go^,Ng, + ^,,l .MSD+ (gg,,ti +,Ng,,-I .MSD)) (3.9) g' g And X,, -X,,t =J,,t(0 +IX, l + oNgJ g + Ng,t .MSD+ (goNjg, 1 + t,,l .-MSD))(3.10) jg', g where the market potential level for each technology standard and each brand are measured as MJ = s, (Mt Xt-) and M g, = sg,t(M -Xt 1), given the industry potential measure M, and market shares (3.4)-(3.6). 3.3 Dynamic Model with Consumer Expectation We now incorporating consumer expectation about network effects into the logit model and estimate it by using dynamic programming and maximum loglikelihood estimation method. 3.3.1 Consumer Expectation of User-Base and Firm-Base The consumer expectations of the future installed-user-base NV, and supporting- firm-base Fg,t are modeled as functions of current installed-user-base Nt, and supporting-firm-base Fg,t of competing standards. The function is written as G Ng, = (OgiNgt, +OgFg,t -)+ =gut,g 1,2,..-,G. (3.11) /=1 G Fgt =Z(OsgNg-t 1 +IfFg,t-1)+ g,g = 1,2,.-, G. (3.12) /=1 where ug,, and g,, are assumed to follow independent and identical multinomial distribution. 3.3.2 Consumer Dynamic Decision Different from the myopic assumption that a consumer makes purchase decision by only maximizing its single period utility, we assume that a dynamic oriented consumer looks forward T periods ahead and maximize its sum of discounted expected future U, over T periods, given its expectations about the future network effects. The objective function of a dynamic oriented consumer sum of discounted expected future is then defined as T J max Et--' t (UJdJit S,) (3.13) d I rI t [ t ( 3 1 3 ) where u,, is the single-period utility at time t of a consumer i for a product j defined as (3.2); d,, is an indicator that equals to 1 if a consumer i chooses a product j at time t and equals to zero otherwise; and 5 is a discount factor. The E(.) is the mathematical expectation operator and the S, is the state space at time t. The state space S, includes installed-user-base Ng and supporting-firm-base Fgt for different standard g. A dynamic oriented consumer's decision involves making a sequence of decisions on d,, in each period to maximize its expected discounted utility. Denote the maximum value of discounted expected utility over the horizon T as Tr T n t(St)= max Et (tZ(UjtdjSt) (3.14) The value function V,(St) can be rewritten as V,(S(t)) = max [(V, (S)] (3.15) where V,t(S,), the alternative-specific expected value function, is the expected value of consumer i selecting alternative j at time t. According to Bellman (1957), we can derive the alternative specific value function at time t recursively by the following equation t (S) = {E(UJ, St,)+ SE[V+, (St+1 St, t, dt = 1,)] (3.16) where the alternative specific value at last period T is V, (S,) = E(Ur IS,). As seen in (3.16), the alternative-specific expected utility function assumes that decisions are made by maximizing the current period utility plus the expected total future value, given the current choice. The discounted expectation in (3.16) is taken over the distribution of S(t+1) conditional on the current state S(t), decision d,kt, and current error term Ey with the transition density denoted by f(S(t+l) S(t),djt,,t) . 3.3.3 Choice Probability Similarly as in standard choice models that the choice probability of a consumer i for product j is made by comparing the values of different alternatives, the choice probability of a dynamic oriented consumer i for product j is also made by comparing the values of different alternatives. The major difference with the dynamic model is that the values of different alternatives include both the current period utility and the expected discounted utilities at time t over the remaining periods. Assuming that (1) the unobservables c, is conditional independent on observable state variables S(t), and (2) the unobservables s, is distributed as i.i.d. multivariate extreme value. Then we have the choice probability for consumer i to choose alternative j at time t as (Rust 1994): exp f17,(Sol Pr(d,-kt = t) exp t (St) (3.17) exp f,-t (So) J'=0 where V,,(S,) represents the deterministic portion of the alternative specific value function. That is, V,(S,)= UJ, +s8EVI, (S, +s,) . Given the choice probability (3.17) for an individual consumer i to brand j and the observed choice quantities for each brand at each period, we can derive the loglikelihood function by assuming homogenous consumers or a particular heterogeneous function of different consumers. This loglikelihood function hence can be estimated by using maximum loglikelihood estimation methods. Consequently, the market shares and diffusion equations for each technology standard and brands can be finally estimated by using the estimated parameters from the maximum loglikelihood estimation methods. 3.4 Empirical Analysis We estimate the baseline model by using data from the U.S. home VCR market from 1978 to 1996. This time period ranges from the early stage of market emerge, the middle stage of the intensive standard competition and to the late market stable stage. The competition in the home VCR market is one of the best-known examples of technology standard competition between two incompatible format Betamax and VHS (Park 2003). As we know, though the Betamax format was first introduced by Sony in February 1976, VHS format, which is introduced one and half year later by JVC in 1977 and incompatible with the Betamax format, was able to surpass the Betamax in the early years of 1980s and eventually dominated the U.S. VCR market in the late years of 1980s. Examining this market would provide valuable insights on how the inter-standard competition impacts the new product diffusion at the market level, standard level and brand level. 3.4.1 Data Our data set includes annual sales, prices and product attributes on top 20 brands. The annual sales data for the whole VCR market are collected from Consumer Electronics in Review, published by Consumer Electronics Association. The market share data for top 20 brands in this market from 1981 to 1996 are collected from issues of Television Digest / i/h Consumer Electronics, published by Warren Publishing Inc.. The market share data for each brand from 1978 to 1980 are adopted from Ohashi (2003) and Park (2003). The top 20 brands in our original sample from 1978 to 1996 count up to 90% of the total market share. Data on prices and product attributes are collected from New York Times, Orion Blue Book Video and Television (1993, 1997) and Consumer Reports. To obtain a preliminary result about the diffusion at each technology standard and brand, we include the top six brands---RCA, Panasonic, Sony, Zenith, JVC and Magnavox in our estimation. These brands are the major earliest brands in the VCR market, which count for 50% of the total market share on average. RCA, Panasonic and Magnavox adopted VHS format from the beginning when they entered this market. In contrast, Sony and Zenith first adopted format Betamax but switched to format VHS in the late 1980s. Each VCR producer usually introduced multiple models in each year, however since the sales data on model-level are not available, we select a basic model to represent a brand in each year (Park 2003). The product attributes that are included in our preliminary estimation are heads, programming events and programming days. The descriptive statistics for these variables are reported in table 3-1. Table 3-1. Descriptive Statistics Product Attribute Brand Heads Programming Programming Price Sales Market Format Events Days (Million) Share RCA 3.42 6.21 197 754.1 1.04 0.16 VHS (0.96) (2.37) (182) (305) (0.56) (0.0076) Panasonic 3.16 5 18.63 710.3 0.79 0.10 VHS (1.01) (3.11) (12.6) (253.7) (0.45) (0.035) Magnavox 3.16 5.68 125.3 734.5 0.64 0.065 VHS (1.01) (2.67) (167.5) (341.04) (0.56) (0.029) JVC 3.26 5.32 124.2 862.1 0.37 0.036 VHS (0.99) (3.09) (168.2) (399.4) (0.26) (0.018) Sony 3.05 5.32 16.42 819.7 0.42 0.098 Beta to (1.18) (2.94) (12.7) (259.4) (0.25) (0.105) VHS Zenith 3.37 4.84 102.9 699 0.52 0.034 Beta to (1.16) (3.48) (161.1) (285.8) (0.29) (0.017) VHS Network Effect Standard Installed-User-Base Supporting-Firm-Base VHS 19.34 31.21 (18.68) (17.02) Betamax 2.57 4.21 (1.48) (3.61) 3.4.2 Nested Logit Model Estimation As introduced previously, to estimate diffusion equations at three levels, we need first estimate the market share for each brand and technology standard. We use GMM estimation method to estimate the nested logit model (3.6), where the product attributes in our preliminary estimation are Head, Programming Days and Programming Events. The estimation results of the nested logit model are presented in table 3-2. From table 3-2, we find that the installed-user-base from compatible technology standard positively influences consumer's evaluation about a new product, which is consistent to literature on network effects (Gandal 1995, Brynjolfsson and Kemerer 1996). In contrast, we also find that the installed-user-base from competing technology influences consumer's evaluation negatively. This is new to network effects literature and demonstrates the impacts of standard competition on consumer's evaluation and in turn the market share and diffusion of each brand. Similarly as the impacts of user-base from compatible technology standard, we find that the supporting-firm-base from the compatible standard also positively influences consumer's evaluation. This reinforces the results in Chapter 2 that supporting-firm-base, as one of important source of network effects which has been ignored in the literature of network effect, increases consumers' evaluation about a new product. Different from the negative impact of user-base from competing standard, our results show that the supporting-firm-base from competing standard positively impacts the consumer's evaluation. This interesting result suggests that the consumer's evaluation of a brand increases with more entrants in the market no matter which competing standards that an entrant chooses. Table 3-2. Nested Logit Model Estimation Coefficient Estimate t-statistics Constant -4.67 -12.33 Head 0.11 1.25 Programming Days -0.00025 -0.58 Programming Events -0.73E-03 -0.29 Price -0.12* -3.54 User-base for compatible standard 0.043* 5.63 User-base for competing standard -0.053** -2.32 Firm-base for compatible standard 0.019* 4.03 Firm-base for competing standard 0.034 2.16 In(S, /g) 0.87*** 10.18 Cumulative Sales at t-1 0.014 0.63 Market Share Difference 8.48 0.32 : p<0.01, : p<0.05, and *: p<0.1. 3.4.3 Diffusion Equation Estimation With the estimation of the nested logit model, we then derive the estimated market share for each technology standard and brand. Using the estimated market share, the diffusion equations at three levels---category-level (3.8), standard-level (3.9) and brand- level (3.10) are accordingly estimated by using General Linear Estimation method. Impacts of Network Effects and Standard Competition on Market Potential The estimation results of market potential equation (3.7) are reported in table 3-3. From table 3-3, we can see that the market potential is significantly influenced by the market average price and product quality. Specifically, we found that average market price significant decreases the market potential and one of the product quality variable--- Programming Events significantly increases the market potential. In terms of the impact of network effects and standard competition, thought we only find significant results on firm-base from Betamax standard, the signs of user-base and firm-base from both of two competing standards---VHS and Betamax provide some interesting insights. Table 3-3. Market Potential Equation Estimation Coefficient Estimation Average Price -0.50E-03* (-4.38) Head -0.062 (-4.69) Programming Days -0.91E-03* (-4.42) Programming Events 0.14 (1.94) VHS User-base 0.0040 (0.14) Betamax User-base -0.29 (-2.73) VHS Firm-base -0.26E-02 (-0.30) Betamax Firm-base 0.19** (6.63) VHS User-base *Market 0.012 Share Difference (0.058) Betamax User-base -0.65 *Market Share Difference (-0.90) VHS Firm-base *Market 0.078 Share Difference (0.62) Betamax Firm-base -2.98 *Market Share Difference (-6.83) : p<0.01, : p<0.05, and : p<0.1. First, we can see from figure 1 that the user-base of technology VHS influences the market potential positively and this positive impacts becomes stronger as it dominates its competing standards (i.e., the market share difference between VHS and Betamax increases). In contrast, the user-base of technology Betamax influences the market potential negatively, and this negative impact becomes stronger as it is dominated by its competing standards. Figure 3-1. Impacts of Installed-User-Base on Market Potential Second, the impacts of firm-base on market potentials are different at two different stages: at the early stage, since the firm-base of technology Betamax is larger than that of technology VHS due to the first introduction of technology Betamax, the market potential increases as more firms support the technology Betamax and decreases with more firms supporting technology VHS. However, as more firms switch to technology VHS while the firm-base of technology Betamax turns to zero at the late years, the impacts of firm- base from two technology standards on market potentials reverse: the firm-base of VHS impacts market potentials positively while the firm-base of Betamax impacts market potentials negatively (see table 3-3 and figure 3-2). Figure 1. Impacts of Installed-User-Base on Market Potential 0.5 0- 1 ~2 -T~~ 6 E 8 9 10 11 12 13 14 15 16 1. 18 1I -0.5 - -1 -1.5 -2 Year -- User-Base of VHS on Market Potential U--User-Base of Beta on Market Potential Figure 3-2. Impacts of Supporting-Firm-Base on Market Potential Impacts of Word of Mouth and Standard Competition on Diffusion As discussed previously, different from the traditional diffusion models that only study the word of mouth impacts from the whole market's cumulative sales, in our models, we not only incorporate the word of mouth impacts on diffusion from two competing standards separately, but also incorporate the interactive impacts of word of mouth with the stages of standards competition, where the stages of standards competition are measured as the extent of market share difference. The results of the diffusion equation estimations at three levels are reported in table 3-4 and 3-5. Table 3-4. The Diffusion Equations Estimation: Category and Standard Level Coefficient Category-Level Standard-Level Constant -2.42 0.068 (-3.37) (1.26) User-base of VHS 0.014 0.014 (1.38) (1.44) User-base of Beta 0.53 0.14 (2.64) (2.54) User-base of VHS*Market 0.27* 0.026* Share Difference (2.20) (1.69) User-base of Beta*Market -6.29** -0.60** Share Difference (-2.47) (-2.74) : p<0.01, : p<0.05, and : p<0.1. Figure 2. Impacts of Supporting-Firm-Base on Market Potential 1.4 1.2 t; 0.8 0.2 0 -0.2 1 2 3 4 :.. F ': 10 11 12 13 1-4 1 i 1i. 17 10 1' Year --Impacts of VHS Firm-Base on Market Potential --Impacts of Beta Firm-Base on Market Potential Table 3-5. The Diffusion Equations Estimation: Brand Level Coefficient Brand-Level Constant -0.29* (-19.58) User-base of compatible 0.0015* standard (5.37) User-base of competing 0.0013 standard (1.48) User-base of compatible 0.0036* standard*Time Trend (13.68) User-base of Competing -0.058 standard*Time Trend (16.47) Brand own user-base 0.0023 (4.29) ***: p<0.01, **: p<0.05, and : p<0.1. Consistent with the results in the literature of new product diffusion, we find that the word of mouth impacts from two competing standards---VHS and Betamax both positively influence the diffusion of new product at three levels. But differently, we also find that (1) the word of mouth impact of Betamax is larger than that of VHS at the early years, but (2) at the late years, the word of mouth impact of VHS increases while that of Betamax becomes negatively (see figure 3-38). Figure 3. Impacts of Word of Mouth from VHS and Betamax on Category Diffusion 5 4- 3 2 o. 0 S-1 1 2 3 4 5 6 8 10 11 1 17 18 19 -2 -3 -4 -5 Year -*- Impacts of VHS on Market Diffusion --- Impact of Betamax on Market Dffusion Figure 3-3. Impacts of Word of Mouth on Category Diffusion 8 We only present the impact of word of mouth from two competing standards on the whole market diffusion here. The word of mouth impacts from two competing standards on the standards' and brands' diffusion are similar as figure 3-3. More interestingly, we found that as the standard competition appears clear on which standard dominates market (i.e., the market share difference becomes large), the dominated standard (which is the Betamax technology in our data) will no longer generate positive word of mouth, but instead generate negative word of mouth on new product diffusion at three levels (see figure 3-4). This result further suggests that before the standard competition has settled to one of the competing standards, both of the two competing standards contribute to the diffusion at three levels positively and complementary. Once the market settled to one technology standard, the other technology standards would become barriers to the expansion of market, standards and brands. Figure 4. Impact of Betamax User-Base On Category and Standard Diffusion 1 0- -1 1 2 3 5 9 10 11 12 15 -2 -3 -4 -5 Year -Impact of Betamax User-Base on Market Diffusion --- Impact of Betamax on Standard Diffusion Figure 3-4. Impact of Betamax User-Base on Category and Standard Diffusion 3.5 Conclusion and Future Research This paper examines how the inter-standard competition affects new technology diffusion at three levels---the category-level, the standard-level and the brand-level. Specifically, we examine the impact of competition between incompatible standards on new technology diffusion by incorporating (1) the standard competition---measured as the market share differences between the competing standards, and (2) two important network effect factors: installed-user-base and supporting-firm-base of competing standards into (1) the consumer utility function, (2) market potential equation, and (3) the diffusion equations at three levels. The estimation of the diffusion equations at three levels reveals several interesting results. First, our results show that while the installed-user-base from the compatible standard increases the consumer's evaluation to a new product, which is consistent to the literature on network effects, the installed-user-base from competing technology standard decreases the consumer's evaluation to a new product, which is new to literature on network effects, and demonstrates the evidence for one of the important impacts of standard competition on new technology diffusion. Second, we find the empirical evidence that standard competition impacts new technology diffusion complimentarily. This can be seen from two results: (1) the supporting-firm-base from both of two competing standards---VHS and Betamax positively influence the consumer's evaluation on each brand, and (2) the installed-user- base from both of two competing standards generates positive word of mouth effects on the category diffusion as well as the standard and brand diffusion at the early stage of standard competition when none of the two competing standards dominates market. Third, we also find results that the network effects impact the market potential dynamically. While the installed-user-base from two standards both positively impact the new technology diffusion at the early stage of standard competition, the installed-user- base of Betamax turns to impact the diffusion of new technology at three levels negatively after the technology standard VHS dominates the market. Fourth, our results suggest that network effects and standard competition not only impact the diffusion on individual consumer's evaluation to each brand, but only on the market potential complementarily and dynamically. Similarly with the word of mouth impact of installed-user-base on new technology diffusion, the supporting-firm-base from both of two standards impact market potentials seems positively at the early stage of standard competition, but the supporting-firm-base of Betamax has no impact on the market potential later on when the VCR market is dominated by VHS format. Since this is only a preliminary study, there are several directions on which we can work in the future. First, in our preliminary estimation, we only included top 6 brands rather than 20 brands that we have in our original datasets. Future work will include all of the 20 brands in our examination. Second, it would provide additional insights to us on how the competitive behavior among firms impacts the new technology diffusion at three levels. Specifically, we are interested in how the different competitive interactions (i.e., Bertrand, Leader-follower) among firms within a same technological standard (i.e., intra-standard firms) and firms across different technology standards (i.e., inter-standard firms) impact the new technology diffusion differently. Third, we are still working on our dynamic model by taking consumer expectation into consideration. Results from our dynamic model not only provide deep insights on how individual consumer forms its expectation on network effects, particularly on installed-user-base and firm-base for every competing standard, but also enable us to compare the results on new technology diffusion at three levels with our baseline model. The comparison between our baseline model and dynamic model provides 65 methodological implications on whether considering the consumer dynamics improves the prediction accuracy on new technology diffusion at three levels. Finally, the generalization of our study is limited. To see if our results can be generalized across different markets with network effects and standard competition, additional studies are needed on other markets with standard competition. LIST OF REFERENCES Bass, Frank M. (1969), A New Product Growth Model for Consumer Durables, Management Science, 15 (January), 215-227. -----, Trichy V. Krishman and Dipak C. Jain (1994), Why the Bass Model Fits Without Decision Variables, Marketing Science, 13, 203-223. Basu, Amiya, Tridib Mazumdar and S. P. Raj (2003), Indirect Network Externality Effects on Product Attributes, Marketing Science, 22 (2), 209-221. Bellman, Richard (1957), Dynamic Programming, Princeton, NJ: Princeton University Press. Berry, S. (1994), Estimating Discrete-Choice Models of Product Differentiation, Rand Journal ofEconomics, Vol. 25 (2), 242-262. Besanko, D., Gupta S., and Jain D. (1998), Logit Demand Estimation under Competitive Pricing Behavior: An Equilibrium Framework, Management Science, 44 (11), 1533-1547. Bresnahan, T. (1989), Industries and Market Power, In: Schmalensee, R., Willig, R. (Eds), Handbook ofIndustrial Organization, North Holland, Amsterdam. Brynjolfsson, Erik and Chris F. Kemerer (1996), Network Externalities in Microcomputer Software: an Econometric Analysis of the Spreadsheet Market, Management Science, 42 (12), 1627-1647. Christensen, Clayton M., Fernando F. Suarez, and James M. Utterback (1998), Strategies for Survival in Fast-Changing Industries, Management Science, 44 (2), Part 2 of 2, S207-S220. Corner, Kathleen R. (1995), Obtaining Strategic Advantage from Being Imitated: When Can Encouraging "Clones" Pay? Management Science, 41 (2), 209-225. Dranove, D. and Gandel N. (2003), The DVD vs. Divx Standard War: Empirical Evidence of Network Effects and Preannouncemnt Effects, Journal of Economics and Management Science, 12 (3), 363-386. Economides, Nicholas (1996), Network Externalities, Complementarities, and Invitation to Enters, European Journal of Political Economy, 12, 211-233. Farrell, Joseph and Garth Saloner (1986), Installed base and Compatibility: Innovation, Product Preannouncement, and Predation, The American Economic Review, 76 (5), 940-955. Gallini, Nancy T. (1984), Detterence By Market Sharing: A Strategic Incentive for Licensing, The American Economic Review, 74 (5), 931-941. Gandal, N. (1995), Competing Compatibility Standards and Network Externalities in the PC software Market, Review of Economics and Statistics, 77 (4), 599-608. --------, M. Kende and R. Rob (2000), The Dynamics of Technological Adoption in Hardware/Software Systems: The Case of Compact Disc Players, Rand Journal of Economics, 31, 43-61. Garud, Raghu and Arun Kumaraswamy (1993), Changing Competitive Dynamics in Network Industries: An Exploration of Sun Microsystems' Open Systems Strategy, Strategic Management Journal, 14, 351-369. Gasmi, F., Laffont J.J., and Vuong Q. (1992), Econometric Analysis of Collusive Behavior in a Soft Drink Market, Journal of Economics and Management Strategy, 1 (2), 277-311. Gowrisankaran, G. and J. Stavins (2004), Network Externalities and Technology Adoption: Lessons from Electronic Payments, Rand Journal ofEconomics, 35 (2), 260-276. Gupta, Sachin, Dipak C. Jain, and Mohanbir S. Sawhney (1999), Modeling the Evolution of Markets with Indirect Network Externalities; An Application to Digital Television, Marketing Science, 18 (3), 396-416. Horsky, Dan (1990), A Diffusion Model Incorporating Product Benefits, Price, Income and Information, Marketing Science, 9 (4), 324-385. Inceoglu, Firat and Minsoo Park (2003), Diffusion of a New Product under Network Effects: The Case of U.S. DVD Player Market, Boston University, working paper. Iwata, G. (1974), Measurement of Conjectural Variations in Oligopoly, Econometrica, 42 (5), 947-966. Kadiyali, Vrinda (1996), Entry, Its Deterrence, and Its Accommodation: A Study of the U.S. Photographic Film Industry, Rand Journal ofEconomics, 27 (3), 452-478. ------, Vrinda, Pradeep Chintagunta, and Naufel Vilcassim (2000), Manufacturer-Retailer Channel Interactions and Implications for Market Power: an Empirical Investigation of Pricing in a Local Market, Marketing Science, 19 (2), 127-148. -----, K. Sudhir, and Vithala R. Rao (2001), Structural Analysis of Competitive Behavior: New Empirical Industrial Organization Methods in Marketing, International Journal ofResearch in Marketing, 18, 161-186. Katz, Michael and Carl Shapiro (1985), Network Externalities, Competition and Compatibility, The American Economic Review, 75 (3), 424-440. Lerner, Josh (1997), an Empirical Exploration of a Technology Race, Rand Journal of Economics, 28 (2), 228-247. Mahaj an, Vijay and Eitan Muller (1979), Innovation Diffusion and New Product Growth Models in Market, Journal ofMarketing, 43, 55-68. ----------------- and ---------- (1996), Timing, Diffusion, and Substitution of Successive Generations of Technological Innovation: The IBM Framework Case, Technological Forecasting and Social Change, 51, 109-132. Nair, Harikesh, Pradeep Chintagunta, and Jean-Pierre Dube (2004), Empirical Analysis of Indirect Network Effects in the Market for Personal Digital Assistants, Quantitative Marketing and Economics, 2, 23-58. Norton, John A. and Frank M. Bass (1987), A Diffusion Model of Adoption and Substitution for Successive Generations of High Technology Products, Management Science, 33 (September), 1069-1086. --------- and --------- (1992), Evolution of Technological Generations: The Law of Capture, Sloan Management Review, 33 (Winter), 66-77. Ohashi, H (2003), The Role of Network Effects in the U.S. VCR Market, 1978-1986, Journal of Economics andManagement Strategy, Vol. 12 (4), 447-494. Park, S. (2003), Quantitative Analysis of Network Externalities in Competing Technologies: The VCR Case, The Review of Economics and Statistics, 86 (4), 937-945. Rust, J. (1987), Optimal Replacement of GMC Bus Engines: An Empirical Model of Harold Zurcher, Econometrica, 55 (5), 999-1033. Saloner, Garth and Andrea Shepard (1995), Adoption of Technologies with Network Effects: An Empirical Examination of the Adoption of Automated Teller Machines, Rand Journal of Economics, 26 (3), 479-501. Shankar, Venkatesh and Berry L. Bayus (2003), Network Effects and Competition: An Empirical Analysis of the Home Video Game Industry, Strategic Management Journal, 24, 375-384. Shapiro, C. and R. Varian (1999), The Art of Standard Wars, California Management Review, 41 (2), 8-32. ------ and ------ (1999), Information Rules: A Strategic Guide to the Network Economy, Harvard Business School Press, Boston, Massachusetts. Srinivasan, Raji, Gary L. Lilien, and Arvind Rangaswamy (2004), "First in, First out? The Effects of Network Externalities on Pioneer Survival," Journal ofMarketing, 68 (January), 41-58. Sudhir, K. (2001a), Competitive Pricing Behavior in the Auto Market: A Structural Analysis, Marketing Science, 20 (1), 42-60. ----- (200 b), Structural Analysis of Manufacturer Pricing in the Presence of a Strategic Retailer, Marketing Science, 20 (3), 244-264. Sun, Baohong, Jinhong Xie, and H. Henry Cao (2004), Product Strategy for Innovators in Markets with Network Effects, Marketing Science, 23 (2), 243-254. Vilcassim Naufel J., Vrinda Kadiyali and Pradeep V. Chintagunta (1999), Investigating Dynamic Multifirm Market Interactions in Price and Advertising, Management Science, 45 (4), 499-518. Xie, Jinhong and Marvin Sirbu (1995), Price Competition and Compatibility in the Presence of Positive Demand Externalities, Management Science, 41, 909-926. Wang, Q. and Jinhong Xie (2005), A Structural Analysis of Intra-Standard Price Competition in Markets with Network Effects, University of Florida, working paper. BIOGRAPHICAL SKETCH Qi Wang was born in Henan, China, in 1969. She earned a Bachelor of Applied Mathematics at Zhengzhou University with highest honors in 1991, and subsequently a Master of Econometrics at Zhongshan University with highest honors in 1994. Before joining the Ph.D. program in the Marketing Department at the University of Florida, she had been an assistant professor at Zhongshan University for two years and an editor at Zhongshan University Press for five years. During her five years of study at the University of Florida, she taught an undergraduate course in Sales Management in 2003 and 2005. She has now accepted a position as an Assistant Professor of Marketing at State University of New York at Binghamton. |

Full Text |

PAGE 1 ANALYSIS OF MARKETS IN THE PRESENCE OF NETWORK EFFECTS AND STANDARDS COMPETITION By QI WANG A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2005 PAGE 2 Copyright 2005 by QI WANG PAGE 3 To My Parents PAGE 4 ACKNOWLEDGMENTS This dissertation is indebted to many people. First, I would like to thank my advisor, Jinhong Xie, for guiding me through my entire Ph.D. study and the entire dissertation process. She is not only the chair of my dissertation committee, but also a mentor and a lifelong friend to me. Without her constant support, indefatigable encouragement, steady mentoring and patient guidance, I could not have developed strong interests in research, accomplished this dissertation and gone through the difficulties in my graduate study and my personal life. There is no single word in the world that could express my great appreciation to her. Second, I would also like to thank my committee members and other faculty members for their extensive comments and feedback for this dissertation. In particular, I benefited immensely from Barton Weitz, Steven Shugan, Chunrong Ai, Mike Lewis and Debi Mitra. I am also indebted to my friends for their dependable help, especially Yubo Chen, Qiong Wang, Qian Tang, Younghan Bae, Xu Jun, Lucy Liu, Shuman Li and Song Xue. The friendship with them makes my graduate study enjoyable and fun; the help from them made it possible for me to complete my Ph.D. study while bringing up my daughter. Last, but certainly not least, I am extremely grateful from the bottom of my heart to my family--my husband, my daughter, my parents, my brother and sister. I feel extremely lucky to grow up in a family where my parents, my brother and sister are always selflessly with me for every decision I make, even though they are physically far from iv PAGE 5 me. I am thankful to my husband for his understanding and to my daughter for all of the joy that she brought to me in my life, especially in the five-year-long study. It is their unconditional love and support that make all my accomplishment possible. v PAGE 6 TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................................................................................iv LIST OF TABLES...........................................................................................................viii LIST OF FIGURES...........................................................................................................ix ABSTRACT.........................................................................................................................x CHAPTER 1 INTRODUCTION........................................................................................................1 2 A STRUCTURAL ANALYSIS OF INTRA-STANDARD PRICE COMPETITION IN MARKETS WITH NETWORK EFFECTS................................6 2.1 Introduction.............................................................................................................6 2.2 The Baseline Model..............................................................................................13 2.2.1 Consumer Behavior....................................................................................13 2.2.2 Firm Behavior.............................................................................................14 2.3 Empirical Analysis of the Baseline Model...........................................................18 2.3.1 The 3.5-inch Floppy Disk Drive Market....................................................18 2.3.2 Data.............................................................................................................21 2.3.3 Identification...............................................................................................23 2.3.4 Empirical Results of Baseline Model.........................................................23 2.4 Modified Model and Empirical Results................................................................26 2.4.1 Modified Model..........................................................................................26 2.4.2 Empirical Results: Time-Varying CV Approach.......................................27 2.4.3 Validation: Stage-based Menu Approach...................................................33 2.4.4 Discussion of Key Findings........................................................................35 2.5 Conclusion............................................................................................................39 3 DYNAMIC ANALYSIS OF NEW TECHNOLOGY DIFFUSION IN MAREKTS WITH COMPETING STANDARDS.........................................................................42 3.1 Introduction...........................................................................................................42 3.2 The Baseline Model..............................................................................................47 3.2.1 Consumer Utility Equation.........................................................................47 3.2.2 Market Share Equations.............................................................................48 vi PAGE 7 3.2.3 Diffusion Equation at Three Levels...........................................................49 3.3 Dynamic Model with Consumer Expectation.......................................................50 3.3.1 Consumer Expectation of User-Base and Firm-Base.................................51 3.3.2 Consumer Dynamic Decision.....................................................................51 3.3.3 Choice Probability......................................................................................53 3.4 Empirical Analysis................................................................................................54 3.4.1 Data.............................................................................................................54 3.4.2 Nested Logit Model Estimation..................................................................56 3.4.3 Diffusion Equation Estimation...................................................................57 Impacts of Network Effects and Standard Competition on Market Potential..................................................................................................57 3.5 Conclusion and Future Research..........................................................................62 LIST OF REFERENCES...................................................................................................66 BIOGRAPHICAL SKETCH.............................................................................................70 vii PAGE 8 viii LIST OF TABLES Table page 2-1 Descriptive Statistics................................................................................................22 2-2 Baseline Model: Estimated Competitive Structure..................................................25 2-3 Revised Model 1, Time-Varying CV Approach......................................................29 2-4 Revised Model 2, Time-Varying CV Approach......................................................31 2-5 Revised Model 3, Stage-Based Menu Approach.....................................................35 3-1 Descriptive Statistics................................................................................................55 3-2 Nested Logit Model Estimation...............................................................................57 3-3 Market Potential Equation Estimation.....................................................................58 3-4 The Diffusion Equations Estimatio n: Category and Standard Level.......................60 3-5 The Diffusion Equations Estimation: Brand Level..................................................61 PAGE 9 LIST OF FIGURES Figure page 2-1 The Changes of Conjectural Variation Parameters with Time.................................32 2-2 Sales: 3.5 inch Floppy Disk Drive vs Competing Products....................................39 3-1 Impacts of Installed-User-Base on Market Potential...............................................59 3-2 Impacts of Supporting-Firm-Base on Market Potential...........................................60 3-3 Impacts of Word of Mouth on Category Diffusion..................................................61 3-4 Impact of Betamax User-Base on Category and Standard Diffusion.......................62 ix PAGE 10 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy ANALYSIS OF MARKETS IN THE PRESENCE OF NETWORK EFFECTS AND STANDARDS COMPETITION By Qi Wang August 2005 Chair: Jinhong Xie Major Department: Marketing The rapid development of information technology and the digital revolution in recent years have significantly increased the importance of network effects in the success of many new products. When multiple technology standards compete in markets with network effects, consumers willingness to pay for any given standard depends much more strongly on market size and market share than in markets without network effects. This creates new challenges for marketing strategy. There is a pressing need to understand the marketing implications of technology standards in markets with network effects, for both firms pushing different standards (inter-standards competition) and firms adopting the same standard (intra-standards competition). This dissertation applies theoretical and empirical analyses to address this need. The dissertation consists of two essays. Chapter 2 provides a structural analysis of intra-standard price competition between an innovating firm and imitating firms. Chapter 3 provides a dynamic analysis of new technology diffusion at three levels: product category, platform and brand levels in x PAGE 11 the presence of inter-standard competition. Data collected from different industries with standards competition are used to test the proposed models. xi PAGE 12 CHAPTER 1 INTRODUCTION The rapid development of information technology and the digital revolution in recent years have significantly increased the importance of network effects in the success of many new products. When multiple technology standards compete in markets with network effects, consumers willingness to pay for any given standard depends much more strongly on market size and market share than in markets without network effects. This creates new challenges for marketing strategy. There is a pressing need to understand the marketing implications of technology standards in markets with network effects, for both firms pushing different standards (inter-standards competition) and firms adopting the same standard (intra-standards competition). This dissertation applies theoretical and empirical analyses to address this need. The dissertation consists of two chapters. Chapter 2 provides a structural analysis of intra-standard price competition between an innovating firm and imitating firms. Chapter 3 provides a dynamic analysis of new product diffusion at both product category and platform levels in the presence of inter-standard competition. In Chapter 2, I analyze the intra-standards price competition in markets with network effects by using a structural analysis approach. In the literature on network effects, marketing scholars have highlighted one characteristic, installed-user-base and demonstrated its implications to marketing strategies, but under-explored the other characteristic, supporting-firm-base. Although the supporting-firm-base is important to the success of firms in face of standard competition and impacts marketing strategies 1 PAGE 13 2 differently from the installed-user-base, there are NO empirical works that have demonstrated its existence and analyzed its important implications to marketing strategies. This paper empirical examines the existence of supporting-firm-base and investigates the impacts of supporting-firm-base and installed-user-base on the competitive interactions between an innovating firm and its imitators in markets influenced by network effects. The competitive interaction among firms supporting the same technological standard (i.e., intra-standard competition) is also an important but underexplored issue. Theorists differ widely in the assumptions that they have made about the structure of competition among firms supporting the same technological standard (i.e., intra-standard competition). Although they have assumed a fixed competitive structure (e.g., bertrand, leader-follower, or cooperation) over product life cycle, we argue that the competitive structure of intra-standard competition is likely to be stage-dependent. In this study, I apply the new empirical industrial organization framework (NEIO) to examine the supporting-firm-base effects and intra-standard competition in 3.5-inch floppy disk drives during the period 1983-1998. My time-varying structural model estimations yield several interesting results. First, I find empirical evidence that consumers valuations are not only positively related to the installed base, but also positively related to the number of supporting firms. Second, my results show an opposite relationship of supporting-firm-base and installed-user-base with the competitive interactive behavior --a positive relationship between the competitive interaction and the supporting-firm-base and a negative relationship between the competitive interaction and the installed-user-base. The positive relationship suggests that the supporting-firm PAGE 14 3 base effect discourages competitive behavior. In contrast, the negative relationship suggests that installed-user-base motivates competitive behavior. Third, my results show that the competitive structure of the floppy drive market did change over the period studied. Firms behaved collectively in the early stage, competitively in the middle stage, and collectively again in the late stage of the product life cycle. This study fills a gap in the network effects literature by providing empirical evidence on the pattern of intra-standards competition and demonstrating the implications of such intra-standards competition for innovating firms marketing strategy. The empirical findings of this study raise some important new theoretical and empirical issues for future research on markets with network effects and standards competition. In Chapter 3, I dynamically analyze the diffusion of new technology in markets with inter-standards competition by incorporating consumer expectation on network effects. Understanding the impact of inter-standards competition on new technology diffusion is extremely important to the success of new products in the presence of network effects and competing technological standards. In this study, I empirically examine how inter-standard competition affects new technology diffusion at three levels--the category-level, the standard-level and the brand-level. Specifically, I examine the impact of competition between incompatible standards on new technology diffusion by incorporating (1) market share differences, which measure the relative strength between competing standards, and (2) two important network effect factors: installed-user-base and supporting-firm-base of competing standards into our model. The network effect factors measure the absolute strength of each competing standard. PAGE 15 4 My model involves three components: (1) a consumer utility function that captures the influences of price, product attribute, and network effectsmeasured as installed-userbase and supporting-firm-base of two competing standards, (2) equations that derive the market share of two incompatible platforms based on the estimated nested-logit model, and (3) equations that derive the dynamic diffusion of the whole industry, two competing standards and each individual brand. I first estimate a baseline model which captures the impacts of the current network effects--current user-base and firm-base from two competing standards on consumers utility and consequently on new technology diffusion. I then extend the baseline model by incorporating the consumer expectations for the future user-base and firm-base of competing technological standards into the consumer utility function. A dynamic programming method and the maximum likelihood estimation method are thus used to solve and estimate the extended model. The model is estimated using data from the VCR industry. The estimated parameters reveal several interesting results. First, my results show that while the installed-user-base from the compatible standard increases the consumers evaluation to a new product, which is consistent with the literature on network effects, the installed-user-base from competing technology standard decreases the consumers evaluation to a new product, which is new to literature on network effects, and demonstrates the evidence for one of the important impacts of standard competition on new technology diffusion. Second, I find the empirical evidence that standard competition impacts new technology diffusion complimentarily. This can be seen from two results: (1) the supporting-firm-base from both of two competing standards--VHS and Betamax positively influence the consumers evaluation on each brand, and (2) the installed-user PAGE 16 5 base from two competing standards generates positive word of mouth effects on the category diffusion as well as the standard and brand diffusion at the early stage of standard competition when none of the two competing standards dominates market. Third, I also find results that the network effects impact the market potential dynamically. While the installed-user-base from two standards both positively impact the new technology diffusion at the early stage of standard competition, the installed-user-base of Betamax turns to impact the diffusion of new technology at three levels negatively after the technology standard VHS dominates the market. Finally, my results suggest that network effects and standard competition not only impact the diffusion on individual consumers evaluation to each brand, but only on the market potential complementarily and dynamically. Similarly with the word of mouth impact of installed-user-base on new technology diffusion, the supporting-firm-base from both of two standards impact market potentials seems positively at the early stage of standard competition, but the supporting-firm-base of Betamax has no impact on the market potential later on when the VCR market is dominated by VHS format. This study extends research on network effects and provides important implications to markets with network effects by explicitly modeling and estimating the impacts of inter-standards competition on new technology diffusion. It also advances understanding of the formation of consumer expectation for the future installed based of competing technological standardsa very important but under-explored research area. PAGE 17 CHAPTER 2 A STRUCTURAL ANALYSIS OF INTRA-STANDARD PRICE COMPETITION IN MARKETS WITH NETWORK EFFECTS 2.1 Introduction The rapid development of information technology and the digital revolution in recent years have significantly increased the importance of network effects and standard competition in the market place (Shapiro and Varian 1999). Marketing scholars are giving increasing attention to the implications of network effects and standards competition for marketing strategies. Several new studies, for example, address such important issues as the evolution of markets with indirect network externalities (Sachin Gupta et al., 1999, Nair, Chintaguta, and Dube 2004), asymmetry network effects (Shankar and Bayus 2003), pioneer survival in network markets (Srinivasan, Lilien, and Rangaswamy 2004), and product line and technology licensing in the presence of network effect (Sun, Xie, and Cao 2004). One important but underexplored issue in the research of network effects is competitive interactions among firms supporting the same technological standard. Understanding competitive interactions among innovating and imitating firms in the presence of network effect is important to practitioners because the strategic behavior of firms is crucial to the success of the technological standard as well as the fate of the individual firms in markets with network effects. Understanding firms competitive behavior in intra-standard competition is also important from a research perspective: competition among firms adopting the same technology standard in the presence of 6 PAGE 18 7 network effects exhibits certain unique characteristics that do not apply to traditional markets, a phenomenon that raises new, interesting theoretical and empirical issues. Different from traditional markets where competition generally has a negative effect on profit, intra-standard competition in markets with network effect (i.e., competition among firms adopting the same technological standard) can be beneficial. Analyses based on game theory have shown that in markets with network effects, innovating firms can earn a higher profit by having compatible competitors than being a monopolist (Katz and Shapiro 1985, Corner 1995, Xie and Sirbu 1995, Sun, Xie, and Cao 2004) or even benefit from subsidizing compatible entrants (Economides 1996). While theoretical research has provided motivations for innovating firms to encourage compatible entry, the literature has not addressed the issue of how network effects may affect the competition structure in such markets. In the literature on network effects, theorists differ widely in the assumptions that they have made about the structure of competition. For example, some assume that innovator and imitator behave as symmetric Nash game players (e.g., Xie and Sirbu 1995, Economides 1996) while others assume that firms play Stackelberg (leader-follower) games (e.g., Corner 1995). Still other theorists assume a cooperative market structure in which both innovator and imitator behave collectively (Gallini 1984). However, there is little empirical evidence on HOW innovating and imitating firms adopting the same standard actually compete in the market place. In this study, we empirically examine the existence of supporting-firm-base effect and the intra-standard price competition between innovating and imitating firms. We argue that, although theoretical research of network effects has assumed a single fixed PAGE 19 8 competitive structure (e.g., bertrand, leader-follower, or cooperation), we expect that the competitive structure of intra-standard competition is not fixed but varies across different stages of the product life cycle: network externalities create interdependence between consumers and firms and among compatible rivals; as such interdependence changes with the development of markets and firms may change their competitive behavior accordingly. First, network effects create an installed-user-base effect that motivates competitive behavior. Theoretical research has suggested that consumer willingness to pay for a given product increases with the size of the user base of its technology standard (Farrell and Saloner 1986, Katz and Shapiro 1985). This positive relationship between consumer willingness to pay and the installed-user-base has been supported by recent empirical studies based on data from various industries affected by network effects, including software (Gandal 1995, Brynjolfsson and Kemerer 1996,), VCRs (Hiroshi Ohashi 2003), video games (Shankar and Bayus 2003), CD players (Basu, Mazumdar, and Raj 2003), and PDAs (Nair, Chintaguta, and Dube 2004). As suggested by the literature of network effect (e.g., Katz and Shapiro 1985, Corner 1995, Economides 1996), innovating firms can benefit by encouraging competition because competition leads to a lower price and a larger total user base, thus a higher product value. This positive link between competition and profit suggests that the installed-user-base effect motivates competitive behavior in intra-standard competition. Second, network effects may create a supporting-firm-base effect that motivates cooperative behavior. Based on the assumption of fulfilled expectation, several theoretical studies suggest that consumer willingness to pay is positively affected by the PAGE 20 9 expected total market size, which is larger for a competitive market than for a monopoly market (e.g., Katz and Shapiro 1985, Sun, Xie, and Cao 2004) and increases with the number of firms supporting the same technological standards (e.g., Economides 1996). While this theory implies a positive relationship between consumer willingness to pay and the number of supporting firms, such a relationship has not been empirically studied. Testing this possible supporting-firm-base effect will provide the empirical evidence needed to validate the existing theory; in addition, the existence of a supporting-firm-base effect would suggest a negative link between competition and consumer willingness to paycompetition discourages market entry and leads to a small supporting-firm-base, thus a low consumer willingness to pay. This negative link between competition and profit implies that the supporting-firm-base effect discourages competitive behavior but motivates cooperative behavior in intra-standard competition. Since both user base and firm base change with time, the relative strength of the installed-user-base effect and the supporting-firm-base effect may vary across different stages of the product life cycle. As a result, the competitive structure of intra-standard competition may not be fixed but stage-specific. In this paper, we empirically examine the existence of supporting-firm-base and investigate the possible stage-dependent pattern of intra-standard price competition between innovating and imitating firms. Specifically, we apply the new empirical industrial organization framework (NEIO) to examine competitive interactions in the 3.5-inch floppy disk drive (FDD) market using 16 years of data from all manufactures of floppy disk drives during the period 1983-1998. Marketing scholars have recently applied the NEIO framework to examine firms competitive pricing behavior in several markets PAGE 21 10 without network effects. For example, Sudhir investigates the competitive pricing behavior in the U.S. author market (2001a); he also examines manufacturers pricing interactions in the presence of strategic retailers, using data from yogurt and peanut butter markets (2001b). Kadiyali (1996) examines the incumbents strategic behavior (deterrence or accommodation) in face of a new entrant, using data from the photographic film industry. These previous studies based on the NEIO framework have illustrated the importance of understanding competitive interactions to firms marketing strategies and provided us with useful methodological guidance. We chose to examine intra-standard competition in the 3.5-inch FDD market for several reasons. First, the FDD industry is subject to network effects. Different types of drives in this market are not interchangeable and the value of a given type of disk drive depends on the number of people who have adopted the same drive (i.e., the larger the installed base of a given disk drive, the more people one can share files with). Second, the innovator, Sony, faced competition from other manufactures adopting its technological standard. As Disk Trend Report recorded, Sony was willing to work with U.S. manufacturers to build its technology as the industry standard. As a result, as many as 27 other manufacturers adopted Sonys technology and produced 3.5-inch FDDs in the same market. These imitating firms were both rivals of and allies to the innovator, Sony. Third, the 3.5-inch FDD encountered intensive standards competition when it was first introduced by Sony in 1981. Before its introduction, the 5.25-inch FDD had been widely adopted in the microcomputer market. Soon after its introduction, three other new incompatible disk drives (i.e., 3-inch, 3.25-inch, and 4-inch) were also introduced in the PAGE 22 11 FDD market. Hence, the 3.5-inch FDD market has the main characteristics we discussed above. To empirically examine the proposed supporting-firm-base effect and the stage-dependent competitive interactions, in this paper, we (1) model consumer utility as a function of both the installed-user-base and the supporting-firm base; (2) incorporate an equation of the number of supporting firms in our structural equations; and (3) model stage-dependent competitive interactions by introducing a time-varying structure to the estimation of conduct parameters. Three different approaches (menu approach, weighted profit approach, and conjectural variation) are used complementarily in testing and validating the stage-dependent competitive structure. Our empirical investigation of the floppy disc drive market reveals several interesting findings. First, we find that consumer valuations are not only positively related to the installed base, as suggested by previous empirical studies, but are also positively related to the number of manufactures that adopt the same standard. In other words, consumers care about both the installed-user-base and the supporting-firm-base. As the value of a product increases with both, this result highlights the importance of building up the base of firms adopting the same technological standards, thus the importance of competitive interaction in intra-standard competition. Second, by modeling the interrelationship of the number of supporting firms and the prices of innovating and imitating firms at previous period, we are able to identify factors that affect the supporting-firm-base. Our results show that the number of supporting firms is positively influenced by the mean price level of imitating firms in a previous period, This finding is important because it implies that competitive behavior PAGE 23 12 discourages, but cooperative behavior encourages, market entry and suggests that the supporting-firm-base effect has an opposite impact compared with the installed-user-base that encourages competitive behavior. Third, our results show that installed-user-base motivates competitive behavior and supporting-firm-base motivates cooperative behavior, the competitive structure of the floppy disk drive market changed over the period studied. Firms behaved collectively in the early stage, competitively in the middle stage, and collectively again in the late stage of the product life cycle. Such a stage-dependent competition pattern has never been suggested on either theoretical or empirical grounds, but is certainly plausible for markets where firms supporting the same standard are both rivals and allies. As our results show, consumers willingness to pay in markets with network effects depends on BOTH the size of the installed base and the number of manufactures adopting the same standard. The latter can motivate the innovating firm to collaborate rather than aggressively compete (on price) with imitating firms at the early and later stages, though for different reasons. The early collaborative behavior helps attract more manufacturers to adopt a new standard when the new standard is still struggling for market acceptance, and the late collaborative behavior discourages manufacturers from abandoning the existing standard when the next generation of technology appears. As a result, firms are more likely to cooperate in the early and late stages, and to compete on price in the middle stage. The rest of this study is organized as follows. We introduce the baseline model in section 2.2. We discuss our data and present the empirical results based on the baseline model in section 2.3. We then introduce the modified model in section 2.4 and report the PAGE 24 13 empirical results and validation results in section 2.5. Finally, we discuss conclusions and future research in Section 2.6. 2.2 The Baseline Model This section defines our baseline model. The baseline model has two characteristics. First, it adopts the specification for consumer behaviors as suggested in the literature of network effects (e.g. Nair, Chintaguta, and Dube 2004), which considers the impact of installed-user base but not the impact of supporting-firm base on consumer utility. Second, as many studies of competitive interactions (e.g. Sudir 2001a) it assumes a constant competitive structure. We modify the baseline model in Section 4 by (1) incorporating supporting-firm base into functions of consumer utility and conjectural parameters and (2) assuming a stage-dependent competition pattern. 2.2.1 Consumer Behavior Following the literature of network effect, we assume that consumers make their purchase decisions by maximizing their utilities and consumer utility is affected by installed-user base and product attributes. Let j denote firms, where denotes the innovator and denotes imitator.1 The utility of consumer i at time 1j 2j t for firms product is defined as j 011ijtjktkjttjtijtjtijtkuxpNu (2.1) where j kt x and j t p are the attribute and price for product at time is the cumulative sales at time (i.e., a measurement of consumer installed-base). thk j t 1tN 1t jt is the unobserved characteristic for product at time t, j jtu is the average utility across 1 In real markets, there may be more than one imitator. Following the literature (Suslow (1986) and Putsis and Dhar (1999)), firm 2 in the duopoly model (1) can be interpreted as a representative imitator. PAGE 25 14 consumers for product at time t, and j ijt is the random term across product and consumer at time t. Following the literature (e.g. Sudir 2001b), we assume that j i ijt has double exponential distribution and is independent, identical across products and consumers. We allow consumers to have an option of purchasing outside product and normalize the utility of the outside products to zero across times. Then the market share for product at time is given by 0j j t exp1ejtjtjus xpjtu (2.2) Similarly, the market share for outside product 0j at time t is given by 011exptjtjsu (2.3) where j s and 0t s denote the market share of product and outside product j 0j From the market share equations (i.e., (2.2) and (2.3)), we can derive 0ln(/). j ttjt s su (2.4) Since the demand for product at time can be expressed by where jtD j t jtjttDsM t M is the market potential at time the demand equations can hence be written as t 0ln()ln()jttjtDD u 0. j (2.5) 2.2.2 Firm Behavior Firms make price decisions by maximizing their profits. The firm s profit at time is given by j t (), jtjtjtjtpcDj 1,2. (2.6) PAGE 26 15 where jt p are firm s price and marginal cost at time respectively. jtc j t Marginal Cost. Firms marginal costs are defined by jtgjgtjtgcz (2.7) where j gt z are the factors that influence firms marginal cost. Note that, j gt z includes product attributes jkt x and other influential factors such as time trend. j t is the unobservable cost factors for product j Strategic Interactions. Two approaches, the conjectural variation approach (CV), and the menu approach, have been used frequently in the empirical studies of firm competitive interactions (e.g., Kadiyali 1996, Vilcassim, Kadiyali and Chintagunta 1999). The CV approach examines the strategic interactions by estimating a set of parameters and inferring the competitive structure from the signs of the set of parameters, while the menu approach does so by separately estimating alternative models and inferring the competitive structure from the selected best-fitted model. It is common for empirical researchers to use only one approach to estimate the competitive interactions because different approaches often lead to consistent results given appropriate model specification (e.g. Bresnahan 1989). Since one of our objectives is to identify more appropriate model specification (e.g., examine if supporting-firm-base should be incorporated into the consumer utility function and if state-dependent competitive structure is more appropriate than a fixed competitive structure), in this paper, we use both approaches complementarily. The CV approach (pioneered by Iwata 1974) captures the competition structure simply by a constant conjectural variation parameter. In the CV approach, a firm is PAGE 27 16 assumed to have conjectures about how its competitors will react to its price changes and incorporate the conjectures into its price decisions (Kadiyali et al. 2001). In the baseline model, the first-order condition for firm is j 1,2() 0p jtjtjtjtjtjtijtjtltlljDDDpcpp1,2j (2.8) where /ijt it p p 1,2i (, i j ) are called conjectural variation parameters (i.e., 12/tpp 1t and 21/tpp 2t ). The conjectural variation parameters thus tell us how a firm reacts to its competitors price changes and therefore indicate the competition structure. For example, 120 indicates Bertrand competition. When 10 but 20 we infer that the innovator is leader and the imitator is follower. Symmetrically, when 20 but all 10 we infer that the imitator is leader and the innovator is follower. 12, 0 implies a cooperative market structure (Kadiyali et al. 2001). Different from CV approach, in menu approach, the competitive structure of an underlying market is inferred by estimating the alternative models (i.e., models with different competitive structure) separately and selecting the best-fitted model. We consider four possible competitive structures: Bertrand, Innovator-Led leader-follower, Imitator-Led leader-follower, and Cooperation. Under the Bertrand competition assumption, each firm chooses its price assuming that its competitor does not react to its price changes, so the first order condition for firm is j ()0, 1jtjtjtjtjtjtjtDDpcjpp ,2. (2.9) PAGE 28 17 Under the innovator-led leader-follower assumption, innovator anticipates imitators reaction to its price and incorporates it in its price decision. Therefore, the first order condition for innovator is 111111112()tttttttttDDpDpcppp 210.ttp (2.10) In contrast, the imitator makes its price decision assuming that innovator does not react to its price. That is, the first order condition for imitators is 222222() 20tttttttDDpcpp1t (2.11) Taking derivative of imitators first order condition (2.11) with respect to, we derive the expression of and substitute it in the first order condition of innovator (2.10). Symmetrically, we can derive the first order conditions for innovator and imitators under imitator-led leader-follower assumption as 1tp 2/tpp 111111()ttttttDDpcpp 10.t (2.12) 222222221()tttttttttDDpDpcppp 120.ttp2t (2.13) Similarly, we can derive the expression of 1/tpp in first order condition (2.13) by taking derivative of first order condition (2.12) with respect to and substituting into (2.13). 2tp Finally, under the cooperation assumption, both firms act as if they were monopoly selling the differentiated products. In other words, they maximize a total profit of it own and its competitor when making their price decisions. The total profit is ()tjtjtj jt p cD The first order condition for firm is j PAGE 29 18 1,2()0, 1tltjtltltjtjtlljDDpcjpp ,2. (2.14) In the next section, we empirically estimate the baseline model using both CV and menu approach. We show that two approaches lead to different inferred competitive structures. In Section 4, we modify the baseline model by incorporating a new variable, supporting-firm-base, into the baseline model. We also allow conjectural variation parameters to be timing-varying. We then use the time-varying CV approach to examine the modified model. Our results show a significant effect of supporting-firm base and suggest that the stage-dependent competitive structure provides a better fit than the constant competitive structure. Finally, to validate our key results, we apply menu approach to the modified model and perform stage-based estimation. The results show that the two approaches, CV and menu approach, lead to consistent results when the modified model and stage-dependent competitive structure are used. 2.3 Empirical Analysis of the Baseline Model We estimate the baseline model using data of the 3.5-inch floppy disk drive market from 1983 to 1998. 2.3.1 The 3.5-inch Floppy Disk Drive Market A floppy disk is a data storage device that comprises a circular piece of thin, flexible magnetic medium encased in a square or rectangular plastic wallet.2 Floppy disks were ubiquitously used to distribute software, transfer data between computers, and create small backups in the 1980s and 1990s. Sony introduced the 3.5-inch floppy disk 2 From http://en.wikipedia.org/wiki/Floppy_disk PAGE 30 19 drive in 1981. Before its introduction, there were two main floppy disk drives, 8-inch and 5.25-inch disk drives. The 8-inch floppy disk drives, the first floppy disk drive introduced in 1967 by IBM, were mainly used in mainframe computers and minicomputers. The 5.25-inch disk drives, introduced by Shugart in 1976, were mainly used in minicomputers and microcomputers (PCs). Floppy disks are read and written by a floppy disk drive (FDD). A floppy disk drive basically has four principal components: read-write heads, which are located on the both sides of a disk and move together on the same assembly; a drive motor, which is a very small spindle motor driving the rotation of the disks; a stepper motor that makes a precise number of stepped revolutions to move the read-write head assembly to the proper disk track position; and electronic circuitry, which handle the data read from or written to the disk and control the stepper-motor control circuits used to mover the read-write heads to each track, as well as the movement of the read-write heads toward the disk surface. With the increasing popularity of PCs in 1980s, small size, easy to carry and display were becoming the major concerns of computer manufacturers and hence disk drive manufacturers. The 5.25-inch floppy disk drives were clearly becoming a limitation to the improvements of microcomputer size. At this time, Sony introduced its 3.5-inch disk drive. Sony uses a single-crystal ferrite head in its 3.5-inch floppy disk drive instead of the poly-crystal ferrite head used by 8-inch and 5.25-inch disk drive. The single-crystal ferrite head is lighter, more stable and more reliable than the poly-crystal ferrite head. With this stable and reliable read-write head, Sony was able to put the same number of 80 PAGE 31 20 tracks as 5.25-inch disks on much smaller 3.5-inch disks and keep its 3.5-inch disk drive with equal or even better reliability than 5.25-inch and 8-inch disk drive.3 As a result, the disk density was increased from 48 tracks per inch (TPI) to 135 TPI, and disk size was therefore largely decreased. Along with the introduction of Sonys 3.5-inch disk drive, there were two other microfloppy disk drives contending for market standard: the 3-inch drive, introduced by Matsushita Electric Industrial, Hitachi and Hitachi/Maxell; and the 3.25-inch disk drive, introduced by Tabor and Seagate Technology. Besides these two disk drives, IBM, as an industry dominant manufacturer, was also expected to introduce a 3.9-inch disk drive on an unknown schedule. These floppy disk drives with different sizes were incompatible because disks could only be loaded in the correct size of disk drive. As Disk Trend Report recorded, Sony was willing to work with U.S. manufacturers to build its technology as the industry standard. As a result, as many as 27 other manufacturers adopted Sonys technology and produced 3.5-inch FDD in the same market. Sonys 3.5-inch disk drive eventually became the industry standard. To sum up, the 3.5 inch FDD market has the main characteristics in which we are interested. First, it is subject to network effects. Second, the innovator, Sony, faced competition from imitating firms adopting its technological standard. Third, the 3.5-inch FDD encountered intensive standards competition. 3 See Mini Micro Systems, April, 1981. PAGE 32 21 2.3.2 Data Our annual data on product attributes, sales and market share in the 3.5-inch floppy disk drive market are collected from Disk Trend Report,4 a leading annual market research publication in the disk drive industry. It has been considered as a highly reliable and complete source of disk drive industry data (Josh Lerner 1997, Christensen et al 1998). We compose all imitators into one imitator, and use a composite imitator as representative imitator in estimation as did in Suslow (1986) and Putsis and Dhar (1999). We denote the innovator and imitator hereafter as and respectively. Data on the imitators product attributes are accordingly computed by averaging all imitators product attributes at each year. Imitators sales and market share are then derived by subtracting innovators sales and market share from the sales and market share of 3.5-inch floppy disk drive market. Data on the annual average price of innovator and imitators products at each year are computed by dividing their annual dollar sales by annual sales volume. 1j 2j Disk Trend Report records the attributes for each disk drive that is introduced by each manufacturer in each year. They are listed according to three categories: (i) disk capacity, which includes total capacity, density per track, data surface per spindle, and tracks per surface; (ii) drive performance, which includes the time spent from track to track, settling time, average rotational delay time and data transfer rate; and (iii) disk drive features, which includes height, width and depth of each disk drive. By analyzing the attributes in each category, we found that the attributes in first and second categories are highly correlated. To reduce the problems of multicollinearity, we conducted step4 We would like to thank James Porter for his generous offer of the series books of Disk Trend Report. PAGE 33 22 wise regression to choose the most influential attribute variables. Our regression analyses identify five attributes that contribute to the most of variances of price and sales: disk density (Den), track to track time (TT), height (H), width (W) and depth (D). These attributes are used in our utility function (2.1). Following literature on network effects (e.g. Brynjolfsson and kemerer 1996), the consumer installed-base in demand equation that captures the network effects is measured as the cumulative sales of whole 3.5-inch floppy disk drive market at previous period. We also measure the supporting firm base using the total number of supporting firms at previous period. The potential market size ( t M ) in the demand equation is measured as the total of the floppy disk drive market sales at each time from 1983-1998. The total of the floppy disk drive market sales is measured by adding together all sales from different sizes of floppy disk drives that were introduced in FDD market. For example, it can be sales from 8-inch, 5.25-inch, 3.5-inch, 4-inch, 3.25-inch, 3-inch and high density FDD. We use all of the above five product attributes and one time trend variable in marginal cost equation. Table 2-1 provides the descriptive statistics of our data. From Table 2-1, we can see that the difference between innovator and imitators product mainly lies in disk density. And innovator charges higher average price than imitator. Table 2-1. Descriptive Statistics Innovator Imitator Mean Std. Mean Std. t-statistic of Mean Differences Sales (Million) 6.22 4.17 33.94 33.01 Average Price 52.82 34.87 48.07 27.68 Density 135 0 131.73 5.87 2.23** Track to Track 4.23 2.73 3.74 0.58 0.70 Height 1.06 0.34 1.01 0.23 0.49 Width 3.96 0.048 4.05 0.24 1.54 Depth 5.64 0.25 5.69 0.45 -0.36 Installed-base (Million) 142 170.20 Note: **: p<0.05. PAGE 34 23 2.3.3 Identification We assume that the product attributes jkt x are exogenous and orthogonal to the error terms ( jt and jt ); price and sales are endogenous and correlated with the error terms ( jt and jt ). This identification assumption is reasonable and commonly used in the simultaneous equation estimation (e.g. Sudir 2001b). We use product attributes, their squares and the products of innovators attribute and imitators attributes as instruments. Since innovators density remained the same over 16 years, we exclude it as one instrument variable. So we have a total of 23 instruments in the demand equation: ten exogenous variables (a constant, four innovators attributes, and five imitators attributes), nine exogenous variables square, and four products of innovators and imitators attributes. We also apply these instruments to the two competitive interaction equations. In the baseline model, we need to estimate 9 parameters in the demand equation (constant, coefficients of five product attributes, price, installed-base and supporting firm-base), 7 parameters in the marginal cost equation (constant, coefficients of five product attributes and time trend), and 2 conjectural variation parameters in the time varying first order conditions if using conjectural variation approach. Therefore, our baseline model is identifiable. 2.3.4 Empirical Results of Baseline Model In the conjectural variation approach, we simultaneously estimate the demand equation (2.5), the marginal cost equation (2.7) and the competitive interaction equation (2.8). The competitive structure is inferred by examining the signs of the two conjectural variation parameters, 1 and 2 as discussed in Section 2. PAGE 35 24 In the menu approach, we estimate the structural equations under four alternative assumptions of competitive structures separately, i.e., for each assumed competitive structure, we simultaneously estimate the demand equation (2.5), the marginal cost equation (2.7), and one of corresponding competitive interaction equations given in (2.9)-(2.14). We then infer the competitive structure by selecting the best-fitted model based on the results of the Vuong test.5 The three-stage least square (3SLS) method is used in the estimation of both approaches. Table 2-2 provides the results of the two approaches. As shown in Table 2-2, the results of CV approach show that both conjectural variation parameters, 1 and 2 are not significantly different from zero, indicating a Bertrand competition structure. The results of menu approach show that the cooperation model has a maximum log-likelihood value (-15.09), while the log-likelihood values for the other three alternative competitive structures (Bertrand, Innovator-Led leader-follower, and Imitator-Led leader-follower) are .00, -27.97, and .75, respectively. The results of the Vuong test statistics (see Table 2-2) indicate that the cooperation model is significantly better than the other three alternatives (p<0.01 for all three comparisons). Hence, our empirical analysis of the baseline model under the assumption of a constant competitive structure leads to different conclusions about the nature of competitive interaction in the studied marketBertrand based on conjectural variation 5 The Vuong test statistic is defined as 1ln()fVpgT q where and are the log-likelihood values of the two non-nested models, and and are the number of parameters estimated in each model, respectively. V is distributed as If and then the model corresponding to is rejected in favor of the model corresponding to ln()f )ln(g p q )1,0(N 0V critical valueVV f g and vice-versa (Kadiyali et al. 2000). PAGE 36 25 approach but Cooperation based on menu approach. This indicates that the competitive structures derived from the two approaches are inconsistent when the baseline model is used. Table 2-2. Baseline Model: Estimated Competitive Structure Conjectural Variation Approach Conjectural Variation Parameter 1 2 The log-likelihood Estimation 0.34NS (1.24) 0.23NS (0.59) -33.22 Estimated Competitive Structure Bertrand Menu Approach Alternative Models Bertrand Innovator-Led Leader-Follower Imitator-Led Leader-Follower Cooperation Log-likelihood (Vuong Statistics) -24.00*** (2.23) -27.97*** (3.22) -28.75*** (3.42) -15.09 (--) Estimated Competitive Structure Cooperation Note: ***: p<0.01; NS: p>0.1 For conjectural variation approach, the numbers in parentheses are t-statistics. For menu approach, the numbers in parentheses are Vuong statistics. The critical value of Vuong statistics at p=0.05 is 1.64. The Vuong statistics are significant at p<0.01 in all cases (See footnote 7 for discussion of Vuong test). The inconsistency found in Table 2-2 can be the results of model misspecification. As we argued, network effects create two types of interdependences: (1) the interdependence among consumers (i.e., the value of a product to consumers is affected by the number of consumers bought the same product), and (2) the interdependence between consumers and firms (i.e., the value of a product to consumers is affected by the number of the firms supporting the same technological standard). However, as the previous empirical studies of network effects, our baseline model only considers the installed-use-base effect, but not the supporting-firm-base effect. In the next section, we modify our baseline model by incorporating the supporting-firm-base effect into consumer utility function, and modeling the competitive structure to be a function of (1) installed-user-base and supporting-firm-base, and (2) time trend. PAGE 37 26 2.4 Modified Model and Empirical Results 2.4.1 Modified Model Demand. We modify the baseline model by incorporating a new variable, supporting-firm-base, into consumer behavior. Let t H denote the number of firms supporting the innovators technological standard at time t (i.e., a measurement of supporting-firm-base). We formally incorporate the supporting-firm-base into the consumers utility model: 011112121ijtjktkjtttttjtijtjtijtkuxpNHNTHTu (2.15) Equation (2.15) assumes that consumer utility is affected not only by the installed-user-base, as assumed in the existing literature and in our baseline model, but also by the supporting-firm-base. Furthermore, since both the installed-user-base effect and the supporting-firm-base effect may change with time, we include two interaction terms, and 1(tNT )) 1(t H T in Equation (2.15) to capture the possible interaction effect, where T donate time trend. To examine how the number of supporting firms may be affected by price competition, we also models the number of supporting firms as a function of imitators price and the number of supporting firms at the previous period: 2112121ttttHHHp t (2.16) The quadratic term in Equation (2.16) is intended to capture the saturation effect. Strategic Interactions. To test the possible stage-dependent competitive interactions, we relax the assumption of a fixed competitive structure by allowing the conjectural variation parameters in equation (2.8) to be timing-varying. Specifically, we define the conjectural parameters, 1 and 2 to be two types of functions. We first define PAGE 38 27 the conjectural variation parameters to be a function of installed-user-base and supporting-firm-base, as shown in Equation (2.17): 0112 1 jtjjtjtNH 1,2j , (2.17) and second to be a function of time trend T, as shown in Equation (2.18): 2012,jtjjjTT 1,2j (2.18) Equation (2.17) allows us to examine how the installed-user-base and supporting-firm-base directly influences the competitive structure directly, and whether the competitive structure is stage-dependent, while equation (2.18) provides us an overall pattern of the stage-dependent competitive structure over time. 2.4.2 Empirical Results: Time-Varying CV Approach We first estimate the modified model using the time-varying conjectural variation approach by modeling the competitive structure to be function of installed-user-base and supporting-firm-base as well as function of time trend, and then validate the key findings using stage-based menu approach. The structural equations for CV approach include the demand equation (2.5), the marginal cost equation (2.7), the supporting-firm-base equation (2.16), and the strategic interaction equations (2.8). Note that the jtu in (2.5) is now derived from (2.15) rather than from (2.1) as in Section 2. Also, jt in (2.8) is now a function of installed-user-base and supporting-firm-base as defined in (2.17) or function of time trend as defined in (2.18) rather than a constant as in Section 2. To derive the best fitted model, we incorporate four time dummy variables in equation (2.17). The four time dummy variables are defined as 11d if the time period is at years 1 to 4, and otherwise 0; if the time period is at years 5 to 8 and otherwise 0; if the time 21d 31d PAGE 39 28 period is at years 9 to 12 and otherwise 0; 41d if the time period is at years 13 to 16 and otherwise 0. The estimated function of conjectural variation parameters is then written as 012345678(*1)(*2)(*3)(*4) (*1)(*2)(*3)(*4)jjjjjjjjjjNdNdNdNdHdHdHdHd (2.17) where j =1 and 2. The results of the modified model using time-varying CV approach are given in Tables 2.3-2.4. Impact of Supporting-Firm-Base on Consumers. As shown in Table 2-3 (see the results of Demand Equation on the top of Table 2-3), the installed base is positively related with consumer valuation (p<0.01), suggesting a positive installed-user-base effect. This finding is consistent with previous studies (e.g. Brynjolfsson and kemerer 1996). More importantly, our result shows that the number of the supporting firms is also positively related to consumer valuation (p<0.01), suggesting a positive supporting-firm-base effect. Table 2-3 also shows that the parameters of both interaction terms in the demand equation are significant, suggesting that both the positive installed-user-base effect and the positive supporting-firm-base effect vary across time. Impact of Price on Supporting-Firm-Base. Table 2-3 shows that the number of supporting firms is positively related to the price of imitators (see the results of Supporting-Firm-Base Equation in the middle of Table 2-3). This result combined with the positive effect of supporting-firm-base on consumer valuation discussed above suggests that the supporting-firm-base effect reduce firms incentive to aggressively compete on price, because a high price encourage more firms to adopt the same technological standard and in turn increase the value of the product to consumers. PAGE 40 29 Table 2-3. Revised Model 1, Time-Varying CV Approach Demand and Marginal Cost Equations Demand Marginal Cost Intercept -61.78*** (-2.54)a 3877.86*** (3.89) Den 14.14*** (2.91) -278.40 (-1.56) TT -0.36 (-0.77) -46.21*** (-2.42) Height 2.31 (1.36) 368.85** (3.36) Width 28.31*** (5.36) -1919.24** (-2.30) Depth -25.28*** (-5.91) 88.62 (0.73) Time Trend 6.22** (1.91) Price -0.075*** (-5.80) Installed-base 0.62E-07*** (5.03) Installed-base* Time -0.38E-08*** (-5.15) Firm-base 0.17** (4.28) Firm-base Time -0.032*** (-5.53) Supporting-Firm-Base Equation Firm-Base at Previous Period 1.23*** (5.27) Square of Firm-Base at Previous Period -0.018** (-1.70) Imitating Firms Price at Previous Period 0.055*** (3.51) Conjectural Variation Parameters Innovator Imitator 0i (constant) 0.57* (1.52) 0.22 (0.28) User-Base Firm-Base User-Base Firm-Base 1i (at first period) -0.13E-05*** (-2.00) 0.24*** (2.02) -0.33E-04 (-0.33) 5.51 (0.33) 2i (at second period) -0.46E-07*** (-3.94) 0.13*** (4.67) -0.71E-07*** (-2.67) 0.33*** (4.10) 3i (at third period) -0.35E-08*** (-3.00) 0.04** (1.67) -0.94E-08*** (-3.91) 0.13*** (2.84) 4i ( at fourth period) -0.14E-08*** (-1.91) 0.03** (1.62) -0.29E-08** (-1.69) 0.07* (1.38) Note: The number in parentheses is t statistic. ***: p<0.01. **: p<0.05***. PAGE 41 30 Impacts of User-Base and Firm-Base on Competitive Structure. The estimation of the impacts of user-base and firm-base on competitive structure are given in the bottom of Table 2-3 (see Time-Varying CV Parameter Estimation). As shown in Table 2-3, for both innovating and imitating firms, the coefficient of user-base across four periods are all significantly negative, while the coefficient of firm-base across four periods are all significantly positive. The different signs of the impacts of user-base and firm-base imply that installed-user-base motivates competitive behavior, while the supporting-firm-base motivates cooperative behavior. Stage-Dependent Competitive Interaction. The estimations of time-varying CV parameters where the conjectural variation parameter is defined to be a function of time trend are given in Table 2-4. As show in Table 2-4, for both innovating and imitating firms, the constant, 0j is positive (p<0.01), the linear term, 1j is negative (p<0.01), and the quadratic term, 2j is positive (p<0.01). These results suggest that (1) firms behave cooperatively at the beginning of the new product introduction, and (2) the competitive structure may vary across different stages of product life cycle. To graphically illustrate the pattern of competitive behavior in the market examined, Figure 1 shows how the conjecture variation parameters, jt (j=1,2), given in (2.18) change with time when the estimated time-varying conjecture variation parameters given in Table 2-4 are used. PAGE 42 31 Table 2-4. Revised Model 2, Time-Varying CV Approach Demand and Marginal Cost Equations Demand Marginal Cost Intercept -54.16*** (-5.23)a -20413.3 (-1.20) Den 12.80*** (5.81) 4097.15 (1.19) TT 1.75*** (4.56) -5.35 (-0.14) Height 3.21*** (4.64) -198.08** (-2.39) Width 27.57*** (6.85) -1073.70** (-2.18) Depth -26.90*** (-7.37) 1031.99** (2.56) Time Trend -3.25** (-1.91) Price -0.051*** (-4.21) Installed-base 0.44E-07*** (4.86) Installed-base* Time -0.28E-08*** (-5.22) Firm-base 0.044** (2.34) Firm-base Time -0.016*** (-3.45) Supporting-Firm-Base Equation Firm-Base at Previous Period 1.017*** (11.99) Square of Firm-Base at Previous Period -0.020*** (-5.35) Imitating Firms Price at Previous Period 0.14*** (14.60) Conjectural Variation Parameters Innovator Imitator 0i 4.56*** (7.55) 11.18*** (5.36) 1i -0.62*** (-5.36) -1.80*** (-4.90) 2i 0.023*** (3.96) 0.075*** (4.67) Comparison: Modified vs. Baseline Model Modified model Baseline model Log-likelihood ratio Log-likelihood value -13.88 -33.22 38.68*** Note: The number in parentheses is t statistic. ***: p<0.01. **: p<0.05***. The loglikelihood ratio has 2 distribution, the critical valueat p<0.01 with d.f.=10. 218.30 PAGE 43 32 As shown in Figure 2-1, both conjecture variation parameters, 1t and 2t follow a quadratic curve. Both parameters are positive in the early years, indicating a cooperative market structure in the early stage of standard development. Then both 1t and 2t decrease to near zero at around year 12, indicating a competitive market structure in the middle stage. In the late years, both 1and t 2t increase to be positive again, indicating that the market moved toward a cooperative structure again in the late stage. Hence, Figure 1 shows a very interesting three-stage pattern of competitive structure in floppy disk drive market: cooperation in the early stage, competition in the middle stage, and cooperation again in the later stage. 02468101212345678910111213141516TimeCV parameters lamda1 lamda2 Figure 2-1. The Changes of Conjectural Variation Parameters with Time To evaluate the superiority of the stage-dependent competitive structure model over the constant competitive structure model, we test if the former fits data better than the latter by comparing the log-likelihood values of the conjectural variation models given in PAGE 44 33 Table 2 (-33.22) with that given in Table 3 (-13.88). The log-likelihood ratio test6 suggests a significant superiority of the stage-dependent competitive model (, p<0.01, d.f. =10). 218.30 To sum up, the empirical analysis of the modified model using time-varying conjecture variation approach leads to three key findings: (1) a positive effect of supporting-firm-base on consumer valuation, (2) a positive effect of price on the number of supporting firms, and (3) a three-stage competitive interaction pattern: cooperation-competition-cooperation. To further validate these key findings, we apply a stage-based menu approach to the modified model. 2.4.3 Validation: Stage-based Menu Approach Different from CV approach that captures the stage-dependent competitive interaction via a set of time-varying conjecture variation parameters, when stage-based menu approach is used to estimate stage-dependent competitive structure, one needs to first decide how stages are divided and then apply menu approach to data in each stages. Based on results from CV approach (see Figure 2-1), we divided the data set into three stages: an early stage (year 1-5), a middle stage (year 5-12), and a later stage (year 10-16).7 For each stage, menu approach is used to estimate all four alternative models (i.e., Bertrand, Innovator-Led leader-follower, Imitator-Led leader-follower, and Cooperation), 6 We use the log-likelihood ratio test to compare the goodness of fit between these two nested models. The log-likelihood ratio is 38.68, which is larger than the critical value of at p<0.01 with freedom degree=10, showing that our time-varying conjectural variation model is significantly better than the conjectural variation model with single competitive structure. 218.30 7 Duo to the limited data available, these stages partially overlap with each other so that there are sufficient data points within each stage to allow models to be identifiable and to use Vuong test statistics to select the best-fitted models. While different possible cutoff times were considered, our final definition of stages is consistent with the curves observed in Figure 1 and Vuong test statistics are significant in selecting the best-fitted models. PAGE 45 34 and the best-fitted model in each stage is selected using Vuong test statistics. The results of model selecting test for each stages are given in the top part of Table 2-5. As shown in Table 2-5, according to the Vuong statistical test, the best-fitted model is Cooperation in the early stage, Innovator-Led Leader-follower in the middle stage, and Cooperation again in the later stage. All other alternative models are rejected. Hence, menu approach leads to the same three-stage pattern of competitive interaction as CV approach (i.e., cooperation-competition-cooperation). In other words, when we use the baseline model and assume a constant competitive structure, the two different approaches lead to inconsistent conclusions (see Table 2-2). However, when we use the modified model and allow the competitive structure to be stage-dependent, the two approaches lead to consistent conclusions. Table 2-5 also shows the results of parameter estimations of the best-fitted models in each stage. Note that the stage-based menu approach allows not only competitive structure to vary across stages, but also parameters in all structure equations vary across stages. As a result, the estimation results using menu approach reveal more details of the market dynamics, which will be discussed in more details later. Nevertheless, as in the estimation of CV approach, the results of demand model by menu approach show a positive effect of the number of supporting firm on consumer valuation (in the early and later stage), and the results of supporting firm model by menu approach show a positive relationship between the number of supporting firms and price. Hence, our three key findings derived in CV approach are also supported by the menu approach. PAGE 46 35 Table 2-5. Revised Model 3, Stage-Based Menu Approach Model Selection Test Stage Bertrand Innovator-Led Leader-Follower Imitator-Led Leader-Follower Cooperation Best-fitted model Early (Year 1-5) 32.52*** (3.37) 32.66*** (3.31) 32.57*** (3.35) 40.07 (---) Cooperation Middle (Year 5-12) 21.58*** (8.29) 45.03 (---) 18.72*** (9.30) 35.84*** (3.25) Innovator-Led Leader-Follower Late (Year 10-16) 3.12*** (17.71) 19.56*** (11.50) 3.75*** (17.47) 50.04 (---) Cooperation Parameter Estimation: Demand Model (The Best-fitted Model) Parameter Early Stage (Cooperation) Middle Stage (Innovator-Led Leader-Follower) Late Stage (Coorperation) Intercept -26.18 (-0.99) -258.92** (-1.94) 527.20** (1.72) Den 2.05 (0.34) 52.49** (1.98) -98.82 (-1.57) TT 2.71 (1.45) 1.06** (2.39) 0.98 (0.58) Height -2.50 (-0.50) -1.67** (-1.87) 17.64*** (7.32) Width 15.72 (0.58) 38.45*** (9.81) 24.76*** (5.08) Depth -7.19 (-0.35) -29.85*** (-10.22) -43.01*** (-5.77) Price -0.024 (-1.40) -0.0098** (-1.65) -0.081*** (-2.40) Installed-base -0.92E-07 (-1.04) -0.29E-08 (-1.54) 0.35E-08** (1.82) Firm-base 0.19*** (2.88) -0.032 (-0.92) 0.071*** (2.42) Parameter Estimation: Supporting-Firm-Base Model Parameter Early Stage Middle Stage Late Stage Firm-Base at Previous Period 0.74 (1.28) 1.18*** (10.18) 0.83*** (7.19) Square of Firm-Base at Previous Period -0.0081 (-0.42) -0.018*** (-4.91) -0.038*** (-3.85) Imitating Firms Price at Previous Period 0.14*** (4.03) 0.061** (1.87) 0.39*** (4.48) Note: ***: p<0.01, **: p<0.05. For model selection test, the numbers in parentheses are Vuong statistics. The critical value of Vuong statistics at p=0.05 is 1.64. For parameter estimations, the numbers in parentheses are t-statistics. 2.4.4 Discussion of Key Findings First, previous empirical studies have shown a positive relationship between installed-user-base and consumer valuation (e.g. Nair, Chintaguta, and Dube 2004). Such PAGE 47 36 a positive relationship has been explained by a positive network effect: the larger the user base, the higher the value of the product. By simultaneously examining the installed-user-base and supporting-firm-base, we find that both can have a positive effect on consumer adoption decision. More interestingly, our stage-based analyses show that the relative importance of installed-user-base effect and supporting-firm-base effect varies across stages of product life cycle. For example, in the early stage of 3.5-inch floppy disk drive, the consumer valuation is positively affected by the supporting-firm-base but not by the installed-use-base (see Table 2-5). While this result is not intuitive, it may result from the high uncertainty faced by consumers in the presence of standards competition. In the early stage of product introduction, 3.5-inch disk drive faced intense competition from multiple incompatible technological standards including the old 8-inch and 5.25-inch floppy disk drives as well as the new 3-inch, 3.25-inch and 4-inch floppy disk drives. Since different types of drives were not interchangeable, the adoption utility of a given type of disk drive depends on the outcome of the standards competition. While it is difficult for early buyers to predict if the new 3.5-inch floppy disk technology would be widely adopted by the market, the number of firms supporting the Sonys new innovation in the early stage could be useful information to consumers in evaluating the odds of the new technology because the larger the number of supporting firms, the more likely that the new technology will become industry standards. The fact that consume valuation is affected by the supporting-firm-base but not by the installed-user-base in the early stage suggests that at the beginning of 3.5-inch disk drive introduction, consumers were more concerned about the outcome of ongoing standards battle but less about the current user base of the drive. While conventional wisdom has emphasized the importance of PAGE 48 37 building consumer base in markets with network effects, our finding highlights the importance of simultaneously considering both the installed-user-base and supporting-firm-base. As we will discuss next, the installed-user-base effect and supporting-firm-base effect can have opposite implications on price competition. Second, our empirically analyses show a positive relationship between the number of supporting firms and price. This result implies that the positive supporting-firm-base effect discussed above reduces firms incentive to cut price because a high price encourages more firms to adopt the innovators technological standards, which in turn increases consumer valuation. Note that it is commonly agreed that the positive installed-user-base effect motivates firms compete more aggressively because a low price encourages more consumers to adopt the new product, which in turn increases consumer valuation. Take the two effects together, our findings suggest that while both supporting-firm-base and installed-user-base increase consumer valuation, they have opposite impacts on firms competitive behavior: the former motivates cooperative behavior but the latter motivates competitive behavior. Hence, the nature of firms strategic interaction is expected to be jointly determined by the two effects and may change across different stages of its product life cycle. Third, our most interesting finding is the three-stage competitive interaction pattern found in 3.5-inch floppy disk drive market: firms cooperate in the early stage, compete in the middle stage, and then cooperate again in the late stage. This pattern can be explained by the relative importance of the two effects (i.e., installed-user-base effect and supporting-firm-base effect) to consumers in different stages, and the opposite impacts of the two effects on firms competitive behavior. In the early stage of 3.5-inch PAGE 49 38 disk drive introduction, 3.5-inch disk drive faced intense competition from several competing standards (see Figure 2-2), which drawn consumers attention toward the supporting-firm-base rather than the current installed-use-base (see Table 4). Firms adopted cooperative behavior because cooperation helps but competition obstructs building up supporting-firm-base. In 1990 (i.e., 9 years after the introduction of 3.5 floppy disk drive), the sales of 3.5 floppy disk drive have surpassed the sales of all alternative technologies (see Figure 2), indicating that 3.5 floppy disk drive had established its technology standard position. In the next few years (the middle stage), neither installed-user-base nor supporting-firm-base have a significant effect on consumer valuation (see Table 4), implying that in this stage, the product is evaluated largely based on its performance rather than by network effects. As a result, there would be little incentive for firms to attract more imitators; rather, they had strong incentive to compete aggressively to achieve as much profit as possible. In 1995 (12 years after the introduction of 3.5-inch disk drive), new generation technology, high-density floppy disk drive, enters the market. In this late stage, consumers were concerned about both the installed-user-base and supporting-firm-base, perhaps because both indicate how likely the existing technology would be replaced by the new generation technology. Since the supporting-firm-base was more important to consumers than the installed-user-base (see Table 2-5), firms adopted cooperative behavior again to discourage manufactures from abandoning the existing standard. PAGE 50 39 0.020000.040000.060000.080000.0100000.0120000.019831985198719891991199319951997TimeSales (Thousands) 3.5 inch 8+5.25 inch High Capacity Figure 2-2. Sales: 3.5 inch Floppy Disk Drive vs. Competing Products 2.5 Conclusion In this paper, we use the new empirical industrial organization framework to examine price competition between innovating and imitating firms. Different from the existing research of network effects, we consider not only an installed-user-base effect but also a supporting-firm-base effect on consumer utility. To examine the impact of the supporting-firm-base on intra-standard competition, we model the interrelationship of the number of imitating firms with the prices of innovating and imitating firms. To allow our models to capture the stage-dependent price competition structure, we introduce time-varying parameters to the conjectural variation approach and validate our estimation results using menu approach. Our results provide empirical evidence that markets with network effects generate both a positive consumer-installed-base effect and a positive supporting-firm-base effect (on consumer utilities). However, these two effects have opposite impacts on firms PAGE 51 40 competitive behaviors. To increase consumer installed-base, firms need to keep price low, which motivates firms to compete. In contrast, to encourage other firms to adopt the innovators technology, firms need to keep price high, which motivates firms to cooperate. Furthermore, these two opposite effects impact the price competition between firms differently at different stage. Our results using both approaches suggest that the innovator and imitators have more incentive to encourage more manufactures to support a new standard in its early stage and discourage existing manufacturers from abandoning an established standard in its later stage when facing the challenge of next generation technology. As a result, firms are more likely to cooperate in the early and later stage, and compete on price in the middle stage. This paper makes several contributions to the literature of network effects. First, we examine an effect overlooked in the literature, the supporting-firm-base effect. Our results not only provide empirical evidence supporting a positive relationship between the number of firms adopting the same technological standard and consumers valuation, but also suggest that the supporting-firm-base effect works in an opposite direction compared with the well-known installed-user-base effect in terms of their impacts on intra-standard competition. Second, our study provides empirical evidence of a three-stage pattern of competitive interaction in intra-standard competition. Our results suggest such a pattern in the floppy drive market over the period studied: firms behaved collectively in the early stage, competitively in the middle stage, and collectively again in the late stage. Such a stage-dependent competition pattern is new to the literature on network effects and raises PAGE 52 41 interesting theoretical and empirical questions for future research on the strategic implications of stage-dependent intra-standard competition. Finally, this paper illustrates the use of complementary approaches to evaluate different structural models and estimate stage-dependent competitive structures. This approach can be applied to other situations where a constant competitive structure is questionable. Overall, this study fills a gap in the network effects literature by providing empirical evidence on patterns of intra-standard competition. Our findings raise some important new theoretical and empirical issues for future research on markets with network effects and standards competition. However, since our study only uses data from the floppy disk drive industry, the generality of our findings needs to be tested in other industries with network effects. PAGE 53 CHAPTER 3 DYNAMIC ANALYSIS OF NEW TECHNOLOGY DIFFUSION IN MAREKTS WITH COMPETING STANDARDS 3.1 Introduction The diffusion of new technology has been extensively studied since the traditional diffusion model was published (e.g., Bass 1969, Mahajan and Muller 1979, Norton and Bass 1987). Recently, the rapid development of information technology and the digital revolution attract researchers attention again to the diffusion of new technology, especially in markets with network effects. While studies on the adoption of the new technology such as the ATM machine (Saloner and Shepard 1995), Compact Disc (Gandal, Kende and Rob 2000), DVD player (Inceogle and Park 2003, Dranove and Gandal 2003), and ACH electronic payments system (Gowrisankaran and Stavins 2004) demonstrate the importance of the network effects (i.e. installed-user-base effect) on the new technology diffusion, they all focus on the new technology diffusion at category-level. There is NO empirical research examines the standard diffusion and how the competition between incompatible technologies impacts the new technology diffusion at three levels--category-level, standard-level and brand-level. Understanding the new technology diffusion at standard-level and the impacts of standards competition on the new technology diffusion at three levels is extremely important because the markets with network effects and standards competition exhibit 42 PAGE 54 43 certain unique characteristics that do not apply to the traditional markets and have not been addressed in the literature of new technology diffusion and network effects. First, in markets with network effects and standards competition, the new technology diffusion at three-levels is interdependent: the standard diffusion impacts the market potential for whole category diffusion and the consumers evaluation for an individual brand, while the category diffusion impacts also the market potentials for each standards diffusion and the brand diffusion comprises of the standards diffusion. Studying the interdependence of the new technology diffusion at three-levels requires the marketing scholars to integrate the category diffusion, standard diffusion and brand diffusion into a systematic framework. However, most of the traditional diffusion models (e.g. Bass 1969, Bass, Krishnan and Jain 1994, Horsky 1990) focus only on the category diffusion without considering the impacts of standards and brands diffusion on each other. Though there is little diffusion research in recent years that has proposed models to incorporate the competition impact on diffusion (e.g. Norton and Bass 1992, Mahajan and Muller 1996), there is NO research addressing the standard diffusion in a systematic framework. Second, standards competition impacts the new technology diffusion differently from the competitions among competing brands in the traditional markets. The differences exist in two aspects: (1) unlike the traditional markets where the competitions among competing brands may expend the whole category diffusion positively, the standards competition in markets with network effects may impact the market potential negatively. This is because the high uncertainty that consumers have on which standard will dominant market deters consumers from making choice immediately when there is PAGE 55 44 no single technological standard dominating the market. (2) Standards competition generates different word of mouth effects on the new technology diffusion at the stage when the incompatible technologies are competing for market dominance from the stage when the market is dominated by one of the competing technologies. Recent studies about the impacts of network effects on new technology diffusion has revealed the positive impacts of network effects on a particular technology standard (e.g. Saloner and Shepard 1995, Gandal, Kende and Rob 2000, Park 2003 and Ohashi 2003) or the negative impacts of network effects on competing standard (e.g. Dranove and Gandal 2003), but there is NO empirical works demonstrating how the standard competition impacts the new technology diffusion on the market potential and consumers word of mouth effects. Third, literature has suggested (e.g. Katz and Shapiro 1985 and Shapiro and Varian 1999) that consumer expectation is a major factor in consumer decisions about whether or not to purchase a new technology, and the management of consumer expectation on network effects is a critical factor in standard competition and hence the technology diffusion, but no empirical works demonstrate to managers on how consumers form their expectations on network effects, especially on the future installed-user-base and supporting-firm-base. This study sets up a systematic framework by integrating the new technology diffusion at three levels. Specifically, this study builds a choice-base diffusion model by incorporating (1) the standard competition---measured as the market share differences between the competing standards, and (2) two important network effect factors: installed PAGE 56 45 user-base and supporting-firm-base of competing standards into a baseline model and a dynamic model. Our choice-based diffusion model involves three components: (1) a consumer utility model that captures the influences of price, product attribute, and network effectsmeasured as installed-userbase and supporting-firm-base of two competing standards, (2) equations that derive the market share of two incompatible platforms based on the estimated nested-logit model, and (3) equations that derive the dynamic diffusion at category-level, standard-level and brand-level. Different from the baseline model that only captures the current network effects---current user-base and firm-base of two competing standards, our dynamic model captures the consumer expectations for the future user-base and firm-base of competing technological standards, and thus involves using a dynamic programming method to estimate the market share equations for each competing standard and brand. Our model is estimated using data from the VCR industry. The competition in the home VCR market is one of the best-known examples of technology standard competition between two incompatible format Betamax and VHS (Park 2003). Though the Betamax format was first introduced by Sony in February 1976, VHS format, which is introduced one and half year later by JVC in 1977 and incompatible with the Betamax format, was able to surpass the Betamax in the early years of 1980s and eventually dominated the U.S. VCR market in the late years of 1980s. Examining this market would provide valuable insights on how the inter-standard competition impacts the new product diffusion at category-level, standard-level and brand-level. PAGE 57 46 The estimated parameters reveal several interesting results. First, we find that while the installed-user-base from the compatible standard increases the consumers evaluation to a new product, which is consistent to the literature on network effects, the installed-user-base from competing technology standard decreases the consumers evaluation to a new product, which is new to literature on network effects, and demonstrates the evidence for one of the important impacts of standard competition on new technology diffusion. Second, we find the empirical evidence that standard competition impacts new technology diffusion complimentarily. This can be seen from two results: (1) the supporting-firm-base from both of two competing standards---VHS and Betamax positively influence the consumers evaluation on each brand, and (2) the installed-user-base from both of two competing standards generates positive word of mouth effects on the category diffusion as well as the standard and brand diffusion at the early stage of standard competition when none of the two competing standards dominates market. Third, we also find results that the network effects impact the market potential dynamically. While the installed-user-base from two standards both positively impact the new technology diffusion at the early stage of standard competition, the installed-user-base of Betamax turns to impact the diffusion of new technology at three levels negatively after the technology standard VHS dominates the market. Finally, our results suggest that network effects and standard competition not only impact the diffusion on individual consumers evaluation to each brand, but only on the market potential complementarily and dynamically. Similarly with the word of mouth impact of installed-user-base on new technology diffusion, the supporting-firm-base from PAGE 58 47 both of two standards impact market potentials seems positively at the early stage of standard competition, but the supporting-firm-base of Betamax has no impact on the market potential later on when the VCR market is dominated by VHS format. This study extends research on network effects and provides important implications to markets with network effects by explicitly modeling and estimating the impacts of inter-standards competition on new technology diffusion. It also advances understanding of the formation of consumer expectation for the future installed based of competing technological standardsa very important but under-explored research area. Our paper is organized as follows. We first introduce our baseline model and dynamic model in the section 3.2 and 3.3. Then we present our preliminary results in section 3.4. We discuss our results and future research directions in section 3.5. 3.2 The Baseline Model We start with analyzing the new product diffusion at three levels: category, platform and brand levels by incorporating the impact of network effects and standard competition into (1) the equation of consumers evaluation of each brand, (2) the market potential equation and (3) the diffusion equations at three levels. The network effect variables that are incorporated in the baseline model are the current installed-user-base and the current supporting-firm-base. The dynamic model by considering consumer expectation about future network effects will be introduced in the next section. 3.2.1 Consumer Utility Equation We use the nested logit model to estimate the market shares. Let i denote consumer and denote brand ( j 0,1,2,, j J ), where outside option is denoted as 0 j Different technology standard groups are denoted as g ( G g,,2,1,0 ). The outside PAGE 59 48 option is the only product in group 0j 0 g Further, denote the set of brands in each group as () and denote the value of brand as g g Jgg j 01,1,1'',1'',1''1 jtkjktjtkgjgjgtjgjgtjgjgtjgjgtlljtggglxpNFNFMSD (3.1) where and denote the product attribute and price for brand ; jktx jtp k j l M SD denotes the market share difference between the compatible technology standard of brand j and other competing technology standards g We incorporate four network effect variables:, and in the utility function to measure the installed-user-base and the supporting-firm-base for each brands compatible technology standard 1,tjgN 1,'tjgN 1,tjgF 1,'tjgF g and competing technology standard respectively. 'g Then the utility for consumer i is accordingly modeled as (1).ijtjtigtijtU (3.2) where ijt is an identically and independently distributed extreme value; and ])1([ijtigt is also an extreme value random variable (i.e., double exponential distribution)). 3.2.2 Market Share Equations If brand is compatible to technology standard then the formula for the market share of brand as a fraction of the total standard share (see Berry 1994) is j g j ggjDesjt/][)1/(/ (3.3) where Similarly, the market share for the technology standard is gjtjgeD)1/( g PAGE 60 49 ggggDDs)1()1( (3.4) Thus, the market share for brand as a fraction of all brands regardless its technology standard can be derive as j gggjDDesjt)1()1/( (3.5) With the market share of outside options ggDs)1(01 and )1/()]ln()[ln()ln(0 ssDgg we can then derive a simple analytic equation for mean utility levels as )ln()ln()ln(/0gjjtjsss (3.6) and the parameters in (3.6) can then be obtained by instrumental estimation methods. 3.2.3 Diffusion Equation at Three Levels To derive the diffusion equation at three levels, we need model the market potential first. Instead of modeling the market potential as TV households as in literature (e.g., Park 2003 and Ohashi 2003), we model it as a function of the average price of all products in market, product quality and network effects at different standard competition stages. Define a parameter as gtggtggtggkjkjkMSDFMSDNNap1,01,11,00))(max()( where represents the average price of all products in market; is the maximum level of each product attribute among different brands. The ap )(maxjkjx MSDNtg 1, and MSDFtg 1, PAGE 61 50 denote the interactive terms of network effect variables---installed-user-base and supporting-firm-base with the market share difference. These interactive terms allow us to capture the dynamic influences of network effects across different stages of standard competition. The market potential is then modeled as a function of TV households (TVHH) and the defined parameter : )exp( TVHHMt (3.7) Accordingly, we model the diffusion equations for the whole category as 1100,11,1()(ttttggtggtg ) X XMXNNMSD (3.8) where the industry potential t M is defined in (3.7). Similarly, each standards and each brands diffusion is given by ,,1,00,11,1'0',1'1',1'((gtgtgtgggtggtggtggtgg )) X XMNNMSDNNMSD (3.9) And ,,1,0110,11,1'0',1'1',1'((jtjtjtjtgjgtgjgtgjgtgjgtjgg )) X XMXNNMSDNNMSD (3.10) where the market potential level for each technology standard and each brand are measured as 1()jtjtttMsMX and ,,(gtgtttMsMX 1) given the industry potential measure t M and market shares (3.4)-(3.6). 3.3 Dynamic Model with Consumer Expectation We now incorporating consumer expectation about network effects into the logit model and estimate it by using dynamic programming and maximum loglikelihood estimation method. PAGE 62 51 3.3.1 Consumer Expectation of User-Base and Firm-Base The consumer expectations of the future installed-user-base g tN and supporting-firm-base g t F are modeled as functions of current installed-user-base ,1 g tN and supporting-firm-base ,1 g tF of competing standards. The function is written as ,,1,11(),1GgtglgtglgtgutlNNFg ,2,,.G1,2,,. (3.11) ,,1,11(),Ggtgfgtgfgtgftl F NFg G (3.12) where g ut and g ft are assumed to follow independent and identical multinomial distribution. 3.3.2 Consumer Dynamic Decision Different from the myopic assumption that a consumer makes purchase decision by only maximizing its single period utility, we assume that a dynamic oriented consumer looks forward periods ahead and maximize its sum of discounted expected future over T periods, given its expectations about the future network effects. The objective function of a dynamic oriented consumer sum of discounted expected future is then defined as T tU 1max()ijtjJTJttijtijtdtjEUd tS (3.13) where is the single-period utility at time t of a consumer i for a product defined as (3.2); is an indicator that equals to 1 if a consumer i chooses a product at time and equals to zero otherwise; and ijtU j ijtd j t is a discount factor. The is the mathematical expectation operator and the is the state space at time The state space ()E tS t PAGE 63 52 tS includes installed-user-base g tN and supporting-firm-base g t F for different standard g A dynamic oriented consumers decision involves making a sequence of decisions on in each period to maximize its expected discounted utility. Denote the maximum value of discounted expected utility over the horizon T as ijtd ()max()ijtjJTtitttijtijttdtjVSEUdS (3.14) The value function can be rewritten as ()ittVS (3.15) (())max(()itijttjJVStVS where the alternative-specific expected value function, is the expected value of consumer selecting alternative at time According to Bellman (1957), we can derive the alternative specific value function at time t recursively by the following equation ()ijttVS i j t 11()()(,,1,)ijttijttitttijtijtVSEUSEVSSd (3.16) where the alternative specific value at last period is T ()()ijTTijTTVSEUS As seen in (3.16), the alternative-specific expected utility function assumes that decisions are made by maximizing the current period utility plus the expected total future value, given the current choice. The discounted expectation in (3.16) is taken over the distribution of conditional on the current state decision and current error term (1)St ()St iktd ijt with the transition density denoted by ((1)(),,)ijtijtfStStd PAGE 64 53 3.3.3 Choice Probability Similarly as in standard choice models that the choice probability of a consumer for product is made by comparing the values of different alternatives, the choice probability of a dynamic oriented consumer i for product is also made by comparing the values of different alternatives. The major difference with the dynamic model is that the values of different alternatives include both the current period utility and the expected discounted utilities at time t over the remaining periods. Assuming that (1) the unobservables i j j ijt is conditional independent on observable state variables and (2) the unobservables ()St ijt is distributed as i.i.d. multivariate extreme value. Then we have the choice probability for consumer i to choose alternative at time as (Rust 1994): j t ''0exp()Pr(1).exp()jttikttJjttjVSdSVS (3.17) where () j ttVS represents the deterministic portion of the alternative specific value function. That is, 11()() j ttjttttVSUEVSS Given the choice probability (3.17) for an individual consumer i to brand and the observed choice quantities for each brand at each period, we can derive the loglikelihood function by assuming homogenous consumers or a particular heterogeneous function of different consumers. This loglikelihood function hence can be estimated by using maximum loglikelihood estimation methods. Consequently, the market shares and diffusion equations for each technology standard and brands can be finally estimated by using the estimated parameters from the maximum loglikelihood estimation methods. j PAGE 65 54 3.4 Empirical Analysis We estimate the baseline model by using data from the U.S. home VCR market from 1978 to 1996. This time period ranges from the early stage of market emerge, the middle stage of the intensive standard competition and to the late market stable stage. The competition in the home VCR market is one of the best-known examples of technology standard competition between two incompatible format Betamax and VHS (Park 2003). As we know, though the Betamax format was first introduced by Sony in February 1976, VHS format, which is introduced one and half year later by JVC in 1977 and incompatible with the Betamax format, was able to surpass the Betamax in the early years of 1980s and eventually dominated the U.S. VCR market in the late years of 1980s. Examining this market would provide valuable insights on how the inter-standard competition impacts the new product diffusion at the market level, standard level and brand level. 3.4.1 Data Our data set includes annual sales, prices and product attributes on top 20 brands. The annual sales data for the whole VCR market are collected from Consumer Electronics in Review, published by Consumer Electronics Association. The market share data for top 20 brands in this market from 1981 to 1996 are collected from issues of Television Digest with Consumer Electronics, published by Warren Publishing Inc.. The market share data for each brand from 1978 to 1980 are adopted from Ohashi (2003) and Park (2003). The top 20 brands in our original sample from 1978 to 1996 count up to 90% of the total market share. Data on prices and product attributes are collected from New York Times, Orion Blue Book Video and Television (1993, 1997) and Consumer Reports. PAGE 66 55 To obtain a preliminary result about the diffusion at each technology standard and brand, we include the top six brands---RCA, Panasonic, Sony, Zenith, JVC and Magnavox in our estimation. These brands are the major earliest brands in the VCR market, which count for 50% of the total market share on average. RCA, Panasonic and Magnavox adopted VHS format from the beginning when they entered this market. In contrast, Sony and Zenith first adopted format Betamax but switched to format VHS in the late 1980s. Each VCR producer usually introduced multiple models in each year, however since the sales data on model-level are not available, we select a basic model to represent a brand in each year (Park 2003). The product attributes that are included in our preliminary estimation are heads, programming events and programming days. The descriptive statistics for these variables are reported in table 3-1. Table 3-1. Descriptive Statistics Product Attribute Brand Heads Programming Events Programming Days Price Sales (Million) Market Share Format RCA 3.42 (0.96) 6.21 (2.37) 197 (182) 754.1 (305) 1.04 (0.56) 0.16 (0.0076) VHS Panasonic 3.16 (1.01) 5 (3.11) 18.63 (12.6) 710.3 (253.7) 0.79 (0.45) 0.10 (0.035) VHS Magnavox 3.16 (1.01) 5.68 (2.67) 125.3 (167.5) 734.5 (341.04) 0.64 (0.56) 0.065 (0.029) VHS JVC 3.26 (0.99) 5.32 (3.09) 124.2 (168.2) 862.1 (399.4) 0.37 (0.26) 0.036 (0.018) VHS Sony 3.05 (1.18) 5.32 (2.94) 16.42 (12.7) 819.7 (259.4) 0.42 (0.25) 0.098 (0.105) Beta to VHS Zenith 3.37 (1.16) 4.84 (3.48) 102.9 (161.1) 699 (285.8) 0.52 (0.29) 0.034 (0.017) Beta to VHS Network Effect Standard Installed-User-Base Supporting-Firm-Base VHS 19.34 (18.68) 31.21 (17.02) Betamax 2.57 (1.48) 4.21 (3.61) PAGE 67 56 3.4.2 Nested Logit Model Estimation As introduced previously, to estimate diffusion equations at three levels, we need first estimate the market share for each brand and technology standard. We use GMM estimation method to estimate the nested logit model (3.6), where the product attributes in our preliminary estimation are Head, Programming Days and Programming Events. The estimation results of the nested logit model are presented in table 3-2. From table 3-2, we find that the installed-user-base from compatible technology standard positively influences consumers evaluation about a new product, which is consistent to literature on network effects (Gandal 1995, Brynjolfsson and Kemerer 1996). In contrast, we also find that the installed-user-base from competing technology influences consumers evaluation negatively. This is new to network effects literature and demonstrates the impacts of standard competition on consumers evaluation and in turn the market share and diffusion of each brand. Similarly as the impacts of user-base from compatible technology standard, we find that the supporting-firm-base from the compatible standard also positively influences consumers evaluation. This reinforces the results in Chapter 2 that supporting-firm-base, as one of important source of network effects which has been ignored in the literature of network effect, increases consumers evaluation about a new product. Different from the negative impact of user-base from competing standard, our results show that the supporting-firm-base from competing standard positively impacts the consumers evaluation. This interesting result suggests that the consumers evaluation of a brand increases with more entrants in the market no matter which competing standards that an entrant chooses. PAGE 68 57 Table 3-2. Nested Logit Model Estimation Coefficient Estimate t-statistics Constant -4.67*** -12.33 Head 0.11 1.25 Programming Days -0.00025 -0.58 Programming Events -0.73E-03 -0.29 Price -0.12*** -3.54 User-base for compatible standard 0.043*** 5.63 User-base for competing standard -0.053** -2.32 Firm-base for compatible standard 0.019*** 4.03 Firm-base for competing standard 0.034*** 2.16 )/ln(gSj Cumulative Sales at t-1 Market Share Difference 0.87*** 0.014 8.48 10.18 0.63 0.32 ***: p<0.01, **: p<0.05, and *: p<0.1. 3.4.3 Diffusion Equation Estimation With the estimation of the nested logit model, we then derive the estimated market share for each technology standard and brand. Using the estimated market share, the diffusion equations at three levels---category-level (3.8), standard-level (3.9) and brand-level (3.10) are accordingly estimated by using General Linear Estimation method. Impacts of Network Effects and Standard Competition on Market Potential The estimation results of market potential equation (3.7) are reported in table 3-3. From table 3-3, we can see that the market potential is significantly influenced by the market average price and product quality. Specifically, we found that average market price significant decreases the market potential and one of the product quality variable---Programming Events significantly increases the market potential. In terms of the impact of network effects and standard competition, thought we only find significant results on firm-base from Betamax standard, the signs of user-base and firm-base from both of two competing standards---VHS and Betamax provide some interesting insights. PAGE 69 58 Table 3-3. Market Potential Equation Estimation Coefficient Estimation Average Price -0.50E-03*** (-4.38) Head -0.062*** (-4.69) Programming Days -0.91E-03*** (-4.42) Programming Events 0.14** (1.94) VHS User-base 0.0040 (0.14) Betamax User-base -0.29 (-2.73) VHS Firm-base -0.26E-02 (-0.30) Betamax Firm-base 0.19*** (6.63) VHS User-base *Market Share Difference 0.012 (0.058) Betamax User-base *Market Share Difference -0.65 (-0.90) VHS Firm-base *Market Share Difference 0.078 (0.62) Betamax Firm-base *Market Share Difference -2.98*** (-6.83) ***: p<0.01, **: p<0.05, and *: p<0.1. First, we can see from figure 1 that the user-base of technology VHS influences the market potential positively and this positive impacts becomes stronger as it dominates its competing standards (i.e., the market share difference between VHS and Betamax increases). In contrast, the user-base of technology Betamax influences the market potential negatively, and this negative impact becomes stronger as it is dominated by its competing standards. PAGE 70 59 Figure 1. Impacts of Installed-User-Base on Market Potential-2-1.5-1-0.500.512345678910111213141516171819YearImpact User-Base of VHS on Market Potential User-Base of Beta on Market Potential Figure 3-1. Impacts of Installed-User-Base on Market Potential Second, the impacts of firm-base on market potentials are different at two different stages: at the early stage, since the firm-base of technology Betamax is larger than that of technology VHS due to the first introduction of technology Betamax, the market potential increases as more firms support the technology Betamax and decreases with more firms supporting technology VHS. However, as more firms switch to technology VHS while the firm-base of technology Betamax turns to zero at the late years, the impacts of firm-base from two technology standards on market potentials reverse: the firm-base of VHS impacts market potentials positively while the firm-base of Betamax impacts market potentials negatively (see table 3-3 and figure 3-2). PAGE 71 60 Figure 2. Impacts of Supporting-Firm-Base on Market Potential-0.200.20.40.60.811.21.412345678910111213141516171819YearImpact Impacts of VHS Firm-Base on Market Potential Impacts of Beta Firm-Base on Market Potential Figure 3-2. Impacts of Supporting-Firm-Base on Market Potential Impacts of Word of Mouth and Standard Competition on Diffusion As discussed previously, different from the traditional diffusion models that only study the word of mouth impacts from the whole markets cumulative sales, in our models, we not only incorporate the word of mouth impacts on diffusion from two competing standards separately, but also incorporate the interactive impacts of word of mouth with the stages of standards competition, where the stages of standards competition are measured as the extent of market share difference. The results of the diffusion equation estimations at three levels are reported in table 3-4 and 3-5. Table 3-4. The Diffusion Equations Estimation: Category and Standard Level Coefficient Category-Level Standard-Level Constant -2.42*** (-3.37) 0.068 (1.26) User-base of VHS 0.014* (1.38) 0.014 (1.44) User-base of Beta 0.53*** (2.64) 0.14*** (2.54) User-base of VHS*Market Share Difference 0.27*** (2.20) 0.026** (1.69) User-base of Beta*Market Share Difference -6.29** (-2.47) -0.60*** (-2.74) ***: p<0.01, **: p<0.05, and *: p<0.1. PAGE 72 61 Table 3-5. The Diffusion Equations Estimation: Brand Level Coefficient Brand-Level Constant -0.29*** (-19.58) User-base of compatible standard 0.0015*** (5.37) User-base of competing standard 0.0013 (1.48) User-base of compatible standard*Time Trend 0.0036*** (13.68) User-base of Competing standard*Time Trend -0.058*** (16.47) Brand own user-base 0.0023*** (4.29) ***: p<0.01, **: p<0.05, and *: p<0.1. Consistent with the results in the literature of new product diffusion, we find that the word of mouth impacts from two competing standards---VHS and Betamax both positively influence the diffusion of new product at three levels. But differently, we also find that (1) the word of mouth impact of Betamax is larger than that of VHS at the early years, but (2) at the late years, the word of mouth impact of VHS increases while that of Betamax becomes negatively (see figure 3-38). Figure 3. Impacts of Word of Mouth from VHS and Betamax on Category Diffusion-5-4-3-2-101234512345678910111213141516171819YearImpact Impacts of VHS on Market Diffusion Impact of Betamax on Market Diffusion Figure 3-3. Impacts of Word of Mouth on Category Diffusion 8 We only present the impact of word of mouth from two competing standards on the whole market diffusion here. The word of mouth impacts from two competing standards on the standards and brands diffusion are similar as figure 3-3. PAGE 73 62 More interestingly, we found that as the standard competition appears clear on which standard dominates market (i.e., the market share difference becomes large), the dominated standard (which is the Betamax technology in our data) will no longer generate positive word of mouth, but instead generate negative word of mouth on new product diffusion at three levels (see figure 3-4). This result further suggests that before the standard competition has settled to one of the competing standards, both of the two competing standards contribute to the diffusion at three levels positively and complementary. Once the market settled to one technology standard, the other technology standards would become barriers to the expansion of market, standards and brands. Figure 4. Impact of Betamax User-Base On Category and Standard Diffusion-5-4-3-2-10112345678910111213141516171819YearImpact Impact of Betamax User-Base on Market Diffusion Impact of Betamax on Standard Diffusion Figure 3-4. Impact of Betamax User-Base on Category and Standard Diffusion 3.5 Conclusion and Future Research This paper examines how the inter-standard competition affects new technology diffusion at three levels---the category-level, the standard-level and the brand-level. Specifically, we examine the impact of competition between incompatible standards on new technology diffusion by incorporating (1) the standard competition---measured as the market share differences between the competing standards, and (2) two important PAGE 74 63 network effect factors: installed-user-base and supporting-firm-base of competing standards into (1) the consumer utility function, (2) market potential equation, and (3) the diffusion equations at three levels. The estimation of the diffusion equations at three levels reveals several interesting results. First, our results show that while the installed-user-base from the compatible standard increases the consumers evaluation to a new product, which is consistent to the literature on network effects, the installed-user-base from competing technology standard decreases the consumers evaluation to a new product, which is new to literature on network effects, and demonstrates the evidence for one of the important impacts of standard competition on new technology diffusion. Second, we find the empirical evidence that standard competition impacts new technology diffusion complimentarily. This can be seen from two results: (1) the supporting-firm-base from both of two competing standards---VHS and Betamax positively influence the consumers evaluation on each brand, and (2) the installed-user-base from both of two competing standards generates positive word of mouth effects on the category diffusion as well as the standard and brand diffusion at the early stage of standard competition when none of the two competing standards dominates market. Third, we also find results that the network effects impact the market potential dynamically. While the installed-user-base from two standards both positively impact the new technology diffusion at the early stage of standard competition, the installed-user-base of Betamax turns to impact the diffusion of new technology at three levels negatively after the technology standard VHS dominates the market. PAGE 75 64 Fourth, our results suggest that network effects and standard competition not only impact the diffusion on individual consumers evaluation to each brand, but only on the market potential complementarily and dynamically. Similarly with the word of mouth impact of installed-user-base on new technology diffusion, the supporting-firm-base from both of two standards impact market potentials seems positively at the early stage of standard competition, but the supporting-firm-base of Betamax has no impact on the market potential later on when the VCR market is dominated by VHS format. Since this is only a preliminary study, there are several directions on which we can work in the future. First, in our preliminary estimation, we only included top 6 brands rather than 20 brands that we have in our original datasets. Future work will include all of the 20 brands in our examination. Second, it would provide additional insights to us on how the competitive behavior among firms impacts the new technology diffusion at three levels. Specifically, we are interested in how the different competitive interactions (i.e., Bertrand, Leader-follower) among firms within a same technological standard (i.e., intra-standard firms) and firms across different technology standards (i.e., inter-standard firms) impact the new technology diffusion differently. Third, we are still working on our dynamic model by taking consumer expectation into consideration. Results from our dynamic model not only provide deep insights on how individual consumer forms its expectation on network effects, particularly on installed-user-base and firm-base for every competing standard, but also enable us to compare the results on new technology diffusion at three levels with our baseline model. The comparison between our baseline model and dynamic model provides PAGE 76 65 methodological implications on whether considering the consumer dynamics improves the prediction accuracy on new technology diffusion at three levels. Finally, the generalization of our study is limited. To see if our results can be generalized across different markets with network effects and standard competition, additional studies are needed on other markets with standard competition. PAGE 77 LIST OF REFERENCES Bass, Frank M. (1969), A New Product Growth Model for Consumer Durables, Management Science, 15 (January), 215-227. -----, Trichy V. Krishman and Dipak C. Jain (1994), Why the Bass Model Fits Without Decision Variables, Marketing Science, 13, 203-223. Basu, Amiya, Tridib Mazumdar and S P Raj (2003), Indirect Network Externality Effects on Product Attributes, Marketing Science, 22 (2), 209-221. Bellman, Richard (1957), Dynamic Programming, Princeton, NJ: Princeton University Press. Berry, S. (1994), Estimating Discrete-Choice Models of Product Differentiation, Rand Journal of Economics, Vol. 25 (2), 242-262. Besanko, D., Gupta S., and Jain D. (1998), Logit Demand Estimation under Competitive Pricing Behavior: An Equilibrium Framework, Management Science, 44 (11), 1533-1547. Bresnahan, T. (1989), Industries and Market Power, In: Schmalensee, R., Willig, R. (Eds), Handbook of Industrial Organization, North Holland, Amsterdam. Brynjolfsson, Erik and Chris F. Kemerer (1996), Network Externalities in Microcomputer Software: an Econometric Analysis of the Spreadsheet Market, Management Science, 42 (12), 1627-1647. Christensen, Clayton M., Fernando F. Suarez, and James M. Utterback (1998), Strategies for Survival in Fast-Changing Industries, Management Science, 44 (2), Part 2 of 2, S207-S220. Corner, Kathleen R. (1995), Obtaining Strategic Advantage from Being Imitated: When Can Encouraging Clones Pay? Management Science, 41 (2), 209-225. Dranove, D. and Gandel N. (2003), The DVD vs. Divx Standard War: Empirical Evidence of Network Effects and Preannouncemnt Effects, Journal of Economics and Management Science, 12 (3), 363-386. Economides, Nicholas (1996), Network Externalities, Complementarities, and Invitation to Enters, European Journal of Political Economy, 12, 211-233. 66 PAGE 78 67 Farrell, Joseph and Garth Saloner (1986), Installed base and Compatibility: Innovation, Product Preannouncement, and Predation, The American Economic Review, 76 (5), 940-955. Gallini, Nancy T. (1984), Detterence By Market Sharing: A Strategic Incentive for Licensing, The American Economic Review, 74 (5), 931-941. Gandal, N. (1995), Competing Compatibility Standards and Network Externalities in the PC software Market, Review of Economics and Statistics, 77 (4), 599-608. --------, M. Kende and R. Rob (2000), The Dynamics of Technological Adoption in Hardware/Software Systems: The Case of Compact Disc Players, Rand Journal of Economics, 31, 43-61. Garud, Raghu and Arun Kumaraswamy (1993), Changing Competitive Dynamics in Network Industries: An Exploration of Sun Microsystems Open Systems Strategy, Strategic Management Journal, 14, 351-369. Gasmi, F., Laffont J.J., and Vuong Q. (1992), Econometric Analysis of Collusive Behavior in a Soft Drink Market, Journal of Economics and Management Strategy, 1 (2), 277-311. Gowrisankaran, G. and J. Stavins (2004), Network Externalities and Technology Adoption: Lessons from Electronic Payments, Rand Journal of Economics, 35 (2), 260-276. Gupta, Sachin, Dipak C. Jain, and Mohanbir S. Sawhney (1999), Modeling the Evolution of Markets with Indirect Network Externalities; An Application to Digital Television, Marketing Science, 18 (3), 396-416. Horsky, Dan (1990), A Diffusion Model Incorporating Product Benefits, Price, Income and Information, Marketing Science, 9 (4), 324-385. Inceoglu, Firat and Minsoo Park (2003), Diffusion of a New Product under Network Effects: The Case of U.S. DVD Player Market, Boston University, working paper. Iwata, G. (1974), Measurement of Conjectural Variations in Oligopoly, Econometrica, 42 (5), 947-966. Kadiyali, Vrinda (1996), Entry, Its Deterrence, and Its Accommodation: A Study of the U.S. Photographic Film Industry, Rand Journal of Economics, 27 (3), 452-478. ------, Vrinda, Pradeep Chintagunta, and Naufel Vilcassim (2000), Manufacturer-Retailer Channel Interactions and Implications for Market Power: an Empirical Investigation of Pricing in a Local Market, Marketing Science, 19 (2), 127-148. PAGE 79 68 -----, K. Sudhir, and Vithala R. Rao (2001), Structural Analysis of Competitive Behavior: New Empirical Industrial Organization Methods in Marketing, International Journal of Research in Marketing, 18, 161-186. Katz, Michael and Carl Shapiro (1985), Network Externalities, Competition and Compatibility, The American Economic Review, 75 (3), 424-440. Lerner, Josh (1997), an Empirical Exploration of a Technology Race, Rand Journal of Economics, 28 (2), 228-247. Mahajan, Vijay and Eitan Muller (1979), Innovation Diffusion and New Product Growth Models in Market, Journal of Marketing, 43, 55-68. ----------------and ---------------(1996), Timing, Diffusion, and Substitution of Successive Generations of Technological Innovation: The IBM Framework Case, Technological Forecasting and Social Change, 51, 109-132. Nair, Harikesh, Pradeep Chintagunta, and Jean-Pierre Dube (2004), Empirical Analysis of Indirect Network Effects in the Market for Personal Digital Assistants, Quantitative Marketing and Economics, 2, 23-58. Norton, John A. and Frank M. Bass (1987), A Diffusion Model of Adoption and Substitution for Successive Generations of High Technology Products, Management Science, 33 (September), 1069-1086. --------and --------(1992), Evolution of Technological Generations: The Law of Capture, Sloan Management Review, 33 (Winter), 66-77. Ohashi, H (2003), The Role of Network Effects in the U.S. VCR Market, 1978-1986, Journal of Economics and Management Strategy, Vol. 12 (4), 447-494. Park, S. (2003), Quantitative Analysis of Network Externalities in Competing Technologies: The VCR Case, The Review of Economics and Statistics, 86 (4), 937-945. Rust, J. (1987), Optimal Replacement of GMC Bus Engines: An Empirical Model of Harold Zurcher, Econometrica, 55 (5), 999-1033. Saloner, Garth and Andrea Shepard (1995), Adoption of Technologies with Network Effects: An Empirical Examination of the Adoption of Automated Teller Machines, Rand Journal of Economics, 26 (3), 479-501. Shankar, Venkatesh and Berry L. Bayus (2003), Network Effects and Competition: An Empirical Analysis of the Home Video Game Industry, Strategic Management Journal, 24, 375-384. Shapiro, C. and R. Varian (1999), The Art of Standard Wars, California Management Review, 41 (2), 8-32. PAGE 80 69 -----and -----(1999), Information Rules: A Strategic Guide to the Network Economy, Harvard Business School Press, Boston, Massachusetts. Srinivasan, Raji, Gary L. Lilien, and Arvind Rangaswamy (2004), First in, First out? The Effects of Network Externalities on Pioneer Survival, Journal of Marketing, 68 (January), 41-58. Sudhir, K. (2001a), Competitive Pricing Behavior in the Auto Market: A Structural Analysis, Marketing Science, 20 (1), 42-60. ----(2001b), Structural Analysis of Manufacturer Pricing in the Presence of a Strategic Retailer, Marketing Science, 20 (3), 244-264. Sun, Baohong, Jinhong Xie, and H. Henry Cao (2004), Product Strategy for Innovators in Markets with Network Effects, Marketing Science, 23 (2), 243-254. Vilcassim Naufel J., Vrinda Kadiyali and Pradeep V. Chintagunta (1999), Investigating Dynamic Multifirm Market Interactions in Price and Advertising, Management Science, 45 (4), 499-518. Xie, Jinhong and Marvin Sirbu (1995), Price Competition and Compatibility in the Presence of Positive Demand Externalities, Management Science, 41, 909-926. Wang, Q. and Jinhong Xie (2005), A Structural Analysis of Intra-Standard Price Competition in Markets with Network Effects, University of Florida, working paper. PAGE 81 BIOGRAPHICAL SKETCH Qi Wang was born in Henan, China, in 1969. She earned a Bachelor of Applied Mathematics at Zhengzhou University with highest honors in 1991, and subsequently a Master of Econometrics at Zhongshan University with highest honors in 1994. Before joining the Ph.D. program in the Marketing Department at the University of Florida, she had been an assistant professor at Zhongshan University for two years and an editor at Zhongshan University Press for five years. During her five years of study at the University of Florida, she taught an undergraduate course in Sales Management in 2003 and 2005. She has now accepted a position as an Assistant Professor of Marketing at State University of New York at Binghamton. 70 |