UFDC Home  myUFDC Home  Help 



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 familymy 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 fiveyearlong 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 INTRASTANDARD 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 3.5inch 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: TimeVarying CV Approach ....................................27 2.4.3 Validation: Stagebased 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 UserBase and FirmBase...............................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 22 Baseline Model: Estimated Competitive Structure ..............................................25 23 Revised Model 1, TimeVarying CV Approach ............................................... 29 24 Revised Model 2, TimeVarying CV Approach ............................................... 31 25 Revised Model 3, StageBased Menu Approach .............................................. 35 31 D escriptiv e Statistics .......................................................... .......... ..................... 55 32 N ested Logit M odel Estim ation ........................................ ......................... 57 33 M market Potential Equation Estim action ........................................ ............... 58 34 The Diffusion Equations Estimation: Category and Standard Level .....................60 35 The Diffusion Equations Estimation: Brand Level ............................................61 LIST OF FIGURES Figure pge 21 The Changes of Conjectural Variation Parameters with Time..............................32 22 Sales: 3.5 inch Floppy Disk Drive vs. Competing Products ..................................39 31 Impacts of InstalledUserBase on Market Potential ............................................59 32 Impacts of SupportingFirmBase on Market Potential .......................................60 33 Impacts of Word of Mouth on Category Diffusion............................. ...............61 34 Impact of Betamax UserBase 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 (interstandards competition) and firms adopting the same standard (intrastandards 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 intrastandardprice 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 interstandard 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 (interstandards competition) and firms adopting the same standard (intrastandards 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 intrastandardprice 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 interstandard competition. In Chapter 2, I analyze the intrastandards 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, installeduserbase and demonstrated its implications to marketing strategies, but underexplored the other characteristic, supportingfirmbase. Although the supportingfirmbase is important to the success of firms in face of standard competition and impacts marketing strategies differently from the installeduserbase, 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 supportingfirmbase and investigates the impacts of supportingfirmbase and installeduserbase 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., intrastandard 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., intrastandard competition). Although they have assumed a fixed competitive structure (e.g., bertrand, leaderfollower, or cooperation) over product life cycle, we argue that the competitive structure of intrastandard competition is likely to be stagedependent. In this study, I apply the new empirical industrial organization framework (NEIO) to examine the supportingfirmbase effects and intrastandard competition in 3.5inch floppy disk drives during the period 19831998. My timevarying 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 supportingfirmbase and installeduserbase with the competitive interactive behavior  a positive relationship between the competitive interaction and the supportingfirmbase and a negative relationship between the competitive interaction and the installeduserbase. The positive relationship suggests that the supportingfirm base effect discourages competitive behavior. In contrast, the negative relationship suggests that installeduserbase 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 intrastandards competition and demonstrating the implications of such intrastandards 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 interstandards competition by incorporating consumer expectation on network effects. Understanding the impact of interstandards 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 interstandard competition affects new technology diffusion at three levels the categorylevel, the standardlevel and the brandlevel. 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: installeduserbase and supportingfirmbase 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 effectsmeasured as installed user base and supportingfirmbase of two competing standards, (2) equations that derive the market share of two incompatible platforms based on the estimated nestedlogit 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 effectscurrent userbase and firmbase 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 userbase and firmbase 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 installeduserbase from the compatible standard increases the consumer's evaluation to a new product, which is consistent with the literature on network effects, the installeduser 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 supportingfirmbase from both of two competing standardsVHS and Betamax positively influence the consumer's evaluation on each brand, and (2) the installeduser 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 installeduserbase from two standards both positively impact the new technology diffusion at the early stage of standard competition, the installeduser 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 installeduserbase on new technology diffusion, the supportingfirmbase from both of two standards impact market potentials seems positively at the early stage of standard competition, but the supportingfirmbase 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 interstandards 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 underexplored research area. CHAPTER 2 A STRUCTURAL ANALYSIS OF INTRASTANDARD 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 intrastandard 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, intrastandard 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 (leaderfollower) 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 supportingfirmbase effect and the intrastandard 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, leaderfollower, or cooperation), we expect that the competitive structure of intrastandard 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 installeduserbase 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 installeduserbase 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 installeduserbase effect motivates competitive behavior in intrastandard competition. Second, network effects may create a supportingfirmbase 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 supportingfirmbase effect will provide the empirical evidence needed to validate the existing theory; in addition, the existence of a supportingfirmbase effect would suggest a negative link between competition and consumer willingness to paycompetition discourages market entry and leads to a small supportingfirmbase, thus a low consumer willingness to pay. This negative link between competition and profit implies that the supportingfirmbase effect discourages competitive behavior but motivates cooperative behavior in intrastandard competition. Since both user base and firm base change with time, the relative strength of the installeduserbase effect and the supportingfirmbase effect may vary across different stages of the product life cycle. As a result, the competitive structure of intrastandard competition may not be fixed but stagespecific. In this paper, we empirically examine the existence of supportingfirmbase and investigate the possible stagedependent pattern of intrastandard 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 19831998. 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 intrastandard competition in the 3.5inch 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 3.5inch FDDs in the same market. These imitating firms were both rivals of and allies to the innovator, Sony. Third, the 3.5inch FDD encountered intensive standards competition when it was first introduced by Sony in 1981. Before its introduction, the 5.25inch FDD had been widely adopted in the microcomputer market. Soon after its introduction, three other new incompatible disk drives (i.e., 3inch, 3.25inch, and 4inch) were also introduced in the FDD market. Hence, the 3.5inch FDD market has the main characteristics we discussed above. To empirically examine the proposed supportingfirmbase effect and the stage dependent competitive interactions, in this paper, we (1) model consumer utility as a function of both the installeduserbase and the supportingfirm base; (2) incorporate an equation of the number of supporting firms in our structural equations; and (3) model stagedependent competitive interactions by introducing a timevarying 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 stagedependent 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 installeduserbase and the supportingfirmbase. 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 intrastandard 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 supportingfirmbase. 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 supportingfirmbase effect has an opposite impact compared with the installeduserbase that encourages competitive behavior. Third, our results show that installeduserbase motivates competitive behavior and supportingfirmbase 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 stagedependent 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 installeduser base but not the impact of supportingfirm 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 supportingfirm base into functions of consumer utility and conjectural parameters and (2) assuming a stagedependent 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 installeduser 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 + 711 + 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 installedbase). 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 bestfitted 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 supportingfirmbase should be incorporated into the consumer utility function and if statedependent 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 firstorder condition for firm j is alrjt aD t D t =D+ +(Ptct+ ) ) =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 bestfitted model. We consider four possible competitive structures: Bertrand, InnovatorLed leaderfollower, ImitatorLed leaderfollower, 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 innovatorled leaderfollower 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 imitatorled leaderfollower 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, supportingfirmbase, into the baseline model. We also allow conjectural variation parameters to be timingvarying. We then use the timevarying CV approach to examine the modified model. Our results show a significant effect of supportingfirm base and suggest that the stagedependent 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 stagebased estimation. The results show that the two approaches, CV and menu approach, lead to consistent results when the modified model and stagedependent competitive structure are used. 2.3 Empirical Analysis of the Baseline Model We estimate the baseline model using data of the 3.5inch floppy disk drive market from 1983 to 1998. 2.3.1 The 3.5inch 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 3.5inch floppy disk 2 From http://en.wikipedia.org/wiki/Floppy disk. drive in 1981. Before its introduction, there were two main floppy disk drives, 8inch and 5.25inch disk drives. The 8inch floppy disk drives, the first floppy disk drive introduced in 1967 by IBM, were mainly used in mainframe computers and minicomputers. The 5.25inch 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: readwrite 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 readwrite 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 steppermotor control circuits used to mover the read write heads to each track, as well as the movement of the readwrite 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 5.25inch floppy disk drives were clearly becoming a limitation to the improvements of microcomputer size. At this time, Sony introduced its 3.5inch disk drive. Sony uses a singlecrystal ferrite head in its 3.5inch floppy disk drive instead of the polycrystal ferrite head used by 8inch and 5.25inch disk drive. The singlecrystal ferrite head is lighter, more stable and more reliable than the polycrystal ferrite head. With this stable and reliable readwrite head, Sony was able to put the same number of 80 tracks as 5.25inch disks on much smaller 3.5inch disks and keep its 3.5inch disk drive with equal or even better reliability than 5.25inch and 8inch 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 3.5inch disk drive, there were two other microfloppy disk drives contending for market standard: the 3inch drive, introduced by Matsushita Electric Industrial, Hitachi and Hitachi/Maxell; and the 3.25inch 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 3.9inch 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 3.5inch FDD in the same market. Sony's 3.5inch 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 3.5inch 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 3.5inch 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 3.5inch 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 installedbase in demand equation that captures the network effects is measured as the cumulative sales of whole 3.5inch 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 19831998. 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 8inch, 5.25inch, 3.5inch, 4inch, 3.25inch, 3inch and high density FDD. We use all of the above five product attributes and one time trend variable in marginal cost equation. Table 21 provides the descriptive statistics of our data. From Table 21, 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 21. Descriptive Statistics Innovator Imitator tstatistic 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 Installedbase (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, installedbase and supporting firmbase), 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 bestfitted model based on the results of the Vuong test.5 The threestage least square (3SLS) method is used in the estimation of both approaches. Table 22 provides the results of the two approaches. As shown in Table 22, 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 loglikelihood value (15.09), while the loglikelihood values for the other three alternative competitive structures (Bertrand, InnovatorLed leader follower, and ImitatorLed leaderfollower) are 24.00, 27.97, and 28.75, respectively. The results of the Vuong test statistics (see Table 22) 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 V In (p q) where In(f) and In(g) are the log likelihood values of the two nonnested 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 viceversa (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 22. Baseline Model: Estimated Competitive Structure Conjectural Variation Approach Conjectural Variation Parameter S22 The loglikelihood n NS 0 2NS Estimation .33.22 (1.24) (0.59) Estimated Bertrand Bertrand Competitive Structure Menu Approach InnovatorLed ImitatorLed Alternative Models Bertrand InvarLed ItatorLed Cooperation LeaderFollower LeaderFollower Loglikelihood 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 tstatistics. 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 22 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 installedusebase effect, but not the supportingfirmbase effect. In the next section, we modify our baseline model by incorporating the supporting firmbase effect into consumer utility function, and modeling the competitive structure to be a function of (1) installeduserbase and supportingfirmbase, 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, supportingfirmbase, into consumer behavior. Let H, denote the number of firms supporting the innovator's technological standard at time t (i.e., a measurement of supportingfirmbase). We formally incorporate the supportingfirmbase into the consumers' utility model: u"t = o + xjktfk Pjt +yNt 1 + Nt 711 1H, + tT + 2Ht T+ ,+t = u jt + t (2.15) k Equation (2.15) assumes that consumer utility is affected not only by the installed userbase, as assumed in the existing literature and in our baseline model, but also by the supportingfirmbase. Furthermore, since both the installeduserbase effect and the supportingfirmbase 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 + q2Ht21 P2t1 +1 (2.16) The quadratic term in Equation (2.16) is intended to capture the saturation effect. Strategic Interactions. To test the possible stagedependent competitive interactions, we relax the assumption of a fixed competitive structure by allowing the conjectural variation parameters in equation (2.8) to be timingvarying. 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 installeduserbase and supportingfirmbase, 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 installeduserbase and supportingfirm base directly influences the competitive structure directly, and whether the competitive structure is stagedependent, while equation (2.18) provides us an overall pattern of the stagedependent competitive structure over time. 2.4.2 Empirical Results: TimeVarying CV Approach We first estimate the modified model using the timevarying conjectural variation approach by modeling the competitive structure to be function of installeduserbase and supportingfirmbase as well as function of time trend, and then validate the key findings using stagebased menu approach. The structural equations for CV approach include the demand equation (2.5), the marginal cost equation (2.7), the supportingfirmbase 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 installeduserbase and supportingfirmbase 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 timevarying CV approach are given in Tables 2.32.4. Impact of SupportingFirmBase on Consumers. As shown in Table 23 (see the results of Demand Equation on the top of Table 23), the installed base is positively related with consumer valuation (p<0.01), suggesting a positive installeduserbase 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 supportingfirm base effect. Table 23 also shows that the parameters of both interaction terms in the demand equation are significant, suggesting that both the positive installeduserbase effect and the positive supportingfirmbase effect vary across time. Impact of Price on SupportingFirmBase. Table 23 shows that the number of supporting firms is positively related to the price of imitators (see the results of SupportingFirmBase Equation in the middle of Table 23). This result combined with the positive effect of supportingfirmbase on consumer valuation discussed above suggests that the supportingfirmbase 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 23. Revised Model 1, TimeVarying 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 Installedbase Installedbase* Time Firmbase Firmbase Time FirmBase at Previous Period Square of FirmBase 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) 0.62E07** (5.03) 0.38E08** (5.15) 0.17* (4.28) 0.032*** (5.53) SupportingFirmBase Equation 1.23*** (5.27) 0.018* (1.70) 0.055*** (3.51) Conjectural Variation Parameters Innovator 0.57 (1.52) UserBase 0.13E05 (2.00) 0.46E07 (3.94) 0.35E08 (3.00) 0.14E08 (1.91) FirmBase 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) UserBase 0.33E04 (0.33) 0.71E07 (2.67) 0.94E08 (3.91) 0.29E08 (1.69) FirmBase 5.51 (0.33) 0.33 (4.10) 0.13 (2.84) 0.07 (1.38) Impacts of UserBase and FirmBase on Competitive Structure. The estimation of the impacts of userbase and firmbase on competitive structure are given in the bottom of Table 23 (see TimeVarying CV Parameter Estimation). As shown in Table 2 3, for both innovating and imitating firms, the coefficient of userbase across four periods are all significantly negative, while the coefficient of firmbase across four periods are all significantly positive. The different signs of the impacts of userbase and firmbase imply that installeduserbase motivates competitive behavior, while the supportingfirmbase motivates cooperative behavior. StageDependent Competitive Interaction. The estimations of timevarying CV parameters where the conjectural variation parameter is defined to be a function of time trend are given in Table 24. As show in Table 24, 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 timevarying conjecture variation parameters given in Table 24 are used. Table 24. Revised Model 2, TimeVarying 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 Installedbase Installedbase* Time Firmbase Firmbase Time FirmBase at Previous Period Square of FirmBase at Previous Period Imitating Firm's Price at Previous Period Loglikelihood value 0.051 (4.21) 0.44E07 (4.86) 0.28E08 (5.22) 0.044* (2.34) 0.016 (3.45) SupportingFirmBase 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) Loglikelihood 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 21, 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 threestage 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 21. The Changes of Conjectural Variation Parameters with Time To evaluate the superiority of the stagedependent competitive structure model over the constant competitive structure model, we test if the former fits data better than the latter by comparing the loglikelihood values of the conjectural variation models given in Table 2 (33.22) with that given in Table 3 (13.88). The loglikelihood ratio test6 suggests a significant superiority of the stagedependent competitive model (%2 = 18.30, p<0.01, d.f. =10). To sum up, the empirical analysis of the modified model using timevarying conjecture variation approach leads to three key findings: (1) a positive effect of supportingfirmbase on consumer valuation, (2) a positive effect of price on the number of supporting firms, and (3) a threestage competitive interaction pattern: cooperation competitioncooperation. To further validate these key findings, we apply a stagebased menu approach to the modified model. 2.4.3 Validation: Stagebased Menu Approach Different from CV approach that captures the stagedependent competitive interaction via a set of timevarying conjecture variation parameters, when stagebased menu approach is used to estimate stagedependent 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 21), we divided the data set into three stages: an early stage (year 15), a middle stage (year 512), and a later stage (year 10 16).7 For each stage, menu approach is used to estimate all four alternative models (i.e., Bertrand, InnovatorLed leaderfollower, ImitatorLed leaderfollower, and Cooperation), 6 We use the loglikelihood ratio test to compare the goodness of fit between these two nested models. The loglikelihood 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 timevarying 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 bestfitted 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 bestfitted models. and the bestfitted 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 25. As shown in Table 25, according to the Vuong statistical test, the bestfitted model is Cooperation in the early stage, InnovatorLed Leaderfollower in the middle stage, and Cooperation again in the later stage. All other alternative models are rejected. Hence, menu approach leads to the same threestage pattern of competitive interaction as CV approach (i.e., cooperationcompetitioncooperation). 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 22). However, when we use the modified model and allow the competitive structure to be stagedependent, the two approaches lead to consistent conclusions. Table 25 also shows the results of parameter estimations of the bestfitted models in each stage. Note that the stagebased 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 25. Revised Model 3, StageBased Menu Approach Model Selection Test InnovatorLed ImitatorLed Best Bestfitted Stage Bertrand Leader Leader Cooperation model model Follower Follower Early 32.52* 32.66* 32.57* 40.07 (Year 15) (3.37) (3.31) (3.35) () InnovatorLed Middle 21.58* 45.03 18.72* 35.84* n r Leader (Year 512) (8.29) () (9.30) (3.25) ollower Follower Late 3.12* 19.56** 3.75* 50.04 (Year 1016) (17.71) (11.50) (17.47) () Parameter Estimation: Demand Model (The Bestfitted Model) Parameter Early Stage Middle Stage Late Stage (Cooperation) (InnovatorLed (Coorperation) LeaderFollower) 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) 0.92E07 0.29E08 0.35E08 Installedbase (1.04) (1.54) (1.82) 0.19 0.032 0.071* Firmbase (2.88) (0.92) (2.42) Parameter Estimation: SupportingFirmBase Model Parameter Early Stage Middle Stage Late Stage FirmBase at Previous 0.74 1.18 0.83 Period (1.28) (10.18) (7.19) Square of FirmBase 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 tstatistics. 2.4.4 Discussion of Key Findings First, previous empirical studies have shown a positive relationship between installeduserbase 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 userbase and supportingfirmbase, we find that both can have a positive effect on consumer adoption decision. More interestingly, our stagebased analyses show that the relative importance of installeduserbase effect and supportingfirmbase effect varies across stages of product life cycle. For example, in the early stage of 3.5inch floppy disk drive, the consumer valuation is positively affected by the supportingfirmbase but not by the installedusebase (see Table 25). 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, 3.5inch disk drive faced intense competition from multiple incompatible technological standards including the old 8inch and 5.25 inch floppy disk drives as well as the new 3inch, 3.25inch and 4inch 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 3.5inch 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 supportingfirmbase but not by the installeduserbase in the early stage suggests that at the beginning of 3.5inch 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 installeduserbase and supporting firmbase. As we will discuss next, the installeduserbase effect and supportingfirm 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 supportingfirmbase 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 userbase 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 firmbase and installeduserbase 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 threestage competitive interaction pattern found in 3.5inch 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., installeduserbase effect and supportingfirmbase effect) to consumers in different stages, and the opposite impacts of the two effects on firms' competitive behavior. In the early stage of 3.5inch disk drive introduction, 3.5inch disk drive faced intense competition from several competing standards (see Figure 22), which drawn consumers' attention toward the supportingfirmbase rather than the current installedusebase (see Table 4). Firms adopted cooperative behavior because cooperation helps but competition obstructs building up supportingfirmbase. 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 installeduserbase nor supportingfirm 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 3.5inch disk drive), new generation technology, highdensity floppy disk drive, enters the market. In this late stage, consumers were concerned about both the installeduserbase and supportingfirmbase, perhaps because both indicate how likely the existing technology would be replaced by the new generation technology. Since the supportingfirmbase was more important to consumers than the installeduserbase (see Table 25), 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 22. 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 installeduserbase effect but also a supportingfirmbase effect on consumer utility. To examine the impact of the supportingfirmbase on intrastandard 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 stagedependent 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 consumerinstalledbase effect and a positive supportingfirmbase effect (on consumer utilities). However, these two effects have opposite impacts on firms' competitive behaviors. To increase consumer installedbase, 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 supportingfirmbase 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 supportingfirmbase effect works in an opposite direction compared with the wellknown installeduserbase effect in terms of their impacts on intrastandard competition. Second, our study provides empirical evidence of a threestage pattern of competitive interaction in intrastandard 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 stagedependent 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 stagedependent intrastandard competition. Finally, this paper illustrates the use of complementary approaches to evaluate different structural models and estimate stagedependent 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 intrastandard 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. installeduserbase effect) on the new technology diffusion, they all focus on the new technology diffusion at categorylevel. There is NO empirical research examines the standard diffusion and how the competition between incompatible technologies impacts the new technology diffusion at three levelscategorylevel, standardlevel and brandlevel. Understanding the new technology diffusion at standardlevel 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 threelevels 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 threelevels 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 installeduserbase and supportingfirmbase. This study sets up a systematic framework by integrating the new technology diffusion at three levels. Specifically, this study builds a choicebase diffusion model by incorporating (1) the standard competitionmeasured as the market share differences between the competing standards, and (2) two important network effect factors: installed userbase and supportingfirmbase of competing standards into a baseline model and a dynamic model. Our choicebased diffusion model involves three components: (1) a consumer utility model that captures the influences of price, product attribute, and network effectsmeasured as installeduser base and supportingfirmbase of two competing standards, (2) equations that derive the market share of two incompatible platforms based on the estimated nestedlogit model, and (3) equations that derive the dynamic diffusion at categorylevel, standardlevel and brandlevel. Different from the baseline model that only captures the current network effectscurrent userbase and firmbase of two competing standards, our dynamic model captures the consumer expectations for the future userbase and firmbase 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 bestknown 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 interstandard competition impacts the new product diffusion at categorylevel, standardlevel and brandlevel. The estimated parameters reveal several interesting results. First, we find that while the installeduserbase from the compatible standard increases the consumer's evaluation to a new product, which is consistent to the literature on network effects, the installed userbase 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 supportingfirmbase from both of two competing standardsVHS and Betamax positively influence the consumer's evaluation on each brand, and (2) the installeduser 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 installeduserbase from two standards both positively impact the new technology diffusion at the early stage of standard competition, the installeduser 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 installeduserbase on new technology diffusion, the supportingfirmbase from both of two standards impact market potentials seems positively at the early stage of standard competition, but the supportingfirmbase 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 interstandards 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 underexplored 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 installeduserbase and the current supportingfirmbase. 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 g1 (3.1) +7 AgNjgt + jgFgt1 + Igrg t1 + 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, Fjgti and Fjgti in the utility function to measure the installed userbase and the supportingfirmbase 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 variablesinstalleduserbase and supportingfirmbase 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 =TVHHexp() (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 UserBase and FirmBase The consumer expectations of the future installeduserbase NV, and supporting firmbase Fg,t are modeled as functions of current installeduserbase Nt, and supportingfirmbase 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(OsgNgt 1 +IfFg,t1)+ 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 singleperiod 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 installeduserbase Ng and supportingfirmbase 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 alternativespecific 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 alternativespecific 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 bestknown 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 interstandard 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 brandsRCA, 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 modellevel 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 31. Table 31. 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 InstalledUserBase SupportingFirmBase 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 32. From table 32, we find that the installeduserbase 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 installeduserbase 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 userbase from compatible technology standard, we find that the supportingfirmbase from the compatible standard also positively influences consumer's evaluation. This reinforces the results in Chapter 2 that supportingfirmbase, 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 userbase from competing standard, our results show that the supportingfirmbase 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 32. Nested Logit Model Estimation Coefficient Estimate tstatistics Constant 4.67 12.33 Head 0.11 1.25 Programming Days 0.00025 0.58 Programming Events 0.73E03 0.29 Price 0.12* 3.54 Userbase for compatible standard 0.043* 5.63 Userbase for competing standard 0.053** 2.32 Firmbase for compatible standard 0.019* 4.03 Firmbase for competing standard 0.034 2.16 In(S, /g) 0.87*** 10.18 Cumulative Sales at t1 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 levelscategorylevel (3.8), standardlevel (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 33. From table 33, 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 firmbase from Betamax standard, the signs of userbase and firmbase from both of two competing standardsVHS and Betamax provide some interesting insights. Table 33. Market Potential Equation Estimation Coefficient Estimation Average Price 0.50E03* (4.38) Head 0.062 (4.69) Programming Days 0.91E03* (4.42) Programming Events 0.14 (1.94) VHS Userbase 0.0040 (0.14) Betamax Userbase 0.29 (2.73) VHS Firmbase 0.26E02 (0.30) Betamax Firmbase 0.19** (6.63) VHS Userbase *Market 0.012 Share Difference (0.058) Betamax Userbase 0.65 *Market Share Difference (0.90) VHS Firmbase *Market 0.078 Share Difference (0.62) Betamax Firmbase 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 userbase 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 userbase of technology Betamax influences the market potential negatively, and this negative impact becomes stronger as it is dominated by its competing standards. Figure 31. Impacts of InstalledUserBase on Market Potential Second, the impacts of firmbase on market potentials are different at two different stages: at the early stage, since the firmbase 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 firmbase of technology Betamax turns to zero at the late years, the impacts of firm base from two technology standards on market potentials reverse: the firmbase of VHS impacts market potentials positively while the firmbase of Betamax impacts market potentials negatively (see table 33 and figure 32). Figure 1. Impacts of InstalledUserBase 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  UserBase of VHS on Market Potential UUserBase of Beta on Market Potential Figure 32. Impacts of SupportingFirmBase 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 34 and 35. Table 34. The Diffusion Equations Estimation: Category and Standard Level Coefficient CategoryLevel StandardLevel Constant 2.42 0.068 (3.37) (1.26) Userbase of VHS 0.014 0.014 (1.38) (1.44) Userbase of Beta 0.53 0.14 (2.64) (2.54) Userbase of VHS*Market 0.27* 0.026* Share Difference (2.20) (1.69) Userbase 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 SupportingFirmBase on Market Potential 1.4 1.2 t; 0.8 0.2 0 0.2 1 2 3 4 :.. F ': 10 11 12 13 14 1 i 1i. 17 10 1' Year Impacts of VHS FirmBase on Market Potential Impacts of Beta FirmBase on Market Potential Table 35. The Diffusion Equations Estimation: Brand Level Coefficient BrandLevel Constant 0.29* (19.58) Userbase of compatible 0.0015* standard (5.37) Userbase of competing 0.0013 standard (1.48) Userbase of compatible 0.0036* standard*Time Trend (13.68) Userbase of Competing 0.058 standard*Time Trend (16.47) Brand own userbase 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 standardsVHS 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 338). Figure 3. Impacts of Word of Mouth from VHS and Betamax on Category Diffusion 5 4 3 2 o. 0 S1 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 33. 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 33. 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 34). 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 UserBase 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 UserBase on Market Diffusion  Impact of Betamax on Standard Diffusion Figure 34. Impact of Betamax UserBase on Category and Standard Diffusion 3.5 Conclusion and Future Research This paper examines how the interstandard competition affects new technology diffusion at three levelsthe categorylevel, the standardlevel and the brandlevel. Specifically, we examine the impact of competition between incompatible standards on new technology diffusion by incorporating (1) the standard competitionmeasured as the market share differences between the competing standards, and (2) two important network effect factors: installeduserbase and supportingfirmbase 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 installeduserbase from the compatible standard increases the consumer's evaluation to a new product, which is consistent to the literature on network effects, the installeduserbase 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 supportingfirmbase from both of two competing standardsVHS and Betamax positively influence the consumer's evaluation on each brand, and (2) the installeduser 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 installeduserbase from two standards both positively impact the new technology diffusion at the early stage of standard competition, the installeduser 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 installeduserbase on new technology diffusion, the supportingfirmbase from both of two standards impact market potentials seems positively at the early stage of standard competition, but the supportingfirmbase 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, Leaderfollower) among firms within a same technological standard (i.e., intrastandard firms) and firms across different technology standards (i.e., interstandard 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 installeduserbase and firmbase 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), 215227. , Trichy V. Krishman and Dipak C. Jain (1994), Why the Bass Model Fits Without Decision Variables, Marketing Science, 13, 203223. Basu, Amiya, Tridib Mazumdar and S. P. Raj (2003), Indirect Network Externality Effects on Product Attributes, Marketing Science, 22 (2), 209221. Bellman, Richard (1957), Dynamic Programming, Princeton, NJ: Princeton University Press. Berry, S. (1994), Estimating DiscreteChoice Models of Product Differentiation, Rand Journal ofEconomics, Vol. 25 (2), 242262. Besanko, D., Gupta S., and Jain D. (1998), Logit Demand Estimation under Competitive Pricing Behavior: An Equilibrium Framework, Management Science, 44 (11), 15331547. 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), 16271647. Christensen, Clayton M., Fernando F. Suarez, and James M. Utterback (1998), Strategies for Survival in FastChanging Industries, Management Science, 44 (2), Part 2 of 2, S207S220. Corner, Kathleen R. (1995), Obtaining Strategic Advantage from Being Imitated: When Can Encouraging "Clones" Pay? Management Science, 41 (2), 209225. 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), 363386. Economides, Nicholas (1996), Network Externalities, Complementarities, and Invitation to Enters, European Journal of Political Economy, 12, 211233. Farrell, Joseph and Garth Saloner (1986), Installed base and Compatibility: Innovation, Product Preannouncement, and Predation, The American Economic Review, 76 (5), 940955. Gallini, Nancy T. (1984), Detterence By Market Sharing: A Strategic Incentive for Licensing, The American Economic Review, 74 (5), 931941. Gandal, N. (1995), Competing Compatibility Standards and Network Externalities in the PC software Market, Review of Economics and Statistics, 77 (4), 599608. , 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, 4361. Garud, Raghu and Arun Kumaraswamy (1993), Changing Competitive Dynamics in Network Industries: An Exploration of Sun Microsystems' Open Systems Strategy, Strategic Management Journal, 14, 351369. 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), 277311. Gowrisankaran, G. and J. Stavins (2004), Network Externalities and Technology Adoption: Lessons from Electronic Payments, Rand Journal ofEconomics, 35 (2), 260276. 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), 396416. Horsky, Dan (1990), A Diffusion Model Incorporating Product Benefits, Price, Income and Information, Marketing Science, 9 (4), 324385. 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), 947966. Kadiyali, Vrinda (1996), Entry, Its Deterrence, and Its Accommodation: A Study of the U.S. Photographic Film Industry, Rand Journal ofEconomics, 27 (3), 452478. , Vrinda, Pradeep Chintagunta, and Naufel Vilcassim (2000), ManufacturerRetailer Channel Interactions and Implications for Market Power: an Empirical Investigation of Pricing in a Local Market, Marketing Science, 19 (2), 127148. , 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, 161186. Katz, Michael and Carl Shapiro (1985), Network Externalities, Competition and Compatibility, The American Economic Review, 75 (3), 424440. Lerner, Josh (1997), an Empirical Exploration of a Technology Race, Rand Journal of Economics, 28 (2), 228247. Mahaj an, Vijay and Eitan Muller (1979), Innovation Diffusion and New Product Growth Models in Market, Journal ofMarketing, 43, 5568.  and  (1996), Timing, Diffusion, and Substitution of Successive Generations of Technological Innovation: The IBM Framework Case, Technological Forecasting and Social Change, 51, 109132. Nair, Harikesh, Pradeep Chintagunta, and JeanPierre Dube (2004), Empirical Analysis of Indirect Network Effects in the Market for Personal Digital Assistants, Quantitative Marketing and Economics, 2, 2358. 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), 10691086.  and  (1992), Evolution of Technological Generations: The Law of Capture, Sloan Management Review, 33 (Winter), 6677. Ohashi, H (2003), The Role of Network Effects in the U.S. VCR Market, 19781986, Journal of Economics andManagement Strategy, Vol. 12 (4), 447494. Park, S. (2003), Quantitative Analysis of Network Externalities in Competing Technologies: The VCR Case, The Review of Economics and Statistics, 86 (4), 937945. Rust, J. (1987), Optimal Replacement of GMC Bus Engines: An Empirical Model of Harold Zurcher, Econometrica, 55 (5), 9991033. 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), 479501. Shankar, Venkatesh and Berry L. Bayus (2003), Network Effects and Competition: An Empirical Analysis of the Home Video Game Industry, Strategic Management Journal, 24, 375384. Shapiro, C. and R. Varian (1999), The Art of Standard Wars, California Management Review, 41 (2), 832.  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), 4158. Sudhir, K. (2001a), Competitive Pricing Behavior in the Auto Market: A Structural Analysis, Marketing Science, 20 (1), 4260.  (200 b), Structural Analysis of Manufacturer Pricing in the Presence of a Strategic Retailer, Marketing Science, 20 (3), 244264. Sun, Baohong, Jinhong Xie, and H. Henry Cao (2004), Product Strategy for Innovators in Markets with Network Effects, Marketing Science, 23 (2), 243254. Vilcassim Naufel J., Vrinda Kadiyali and Pradeep V. Chintagunta (1999), Investigating Dynamic Multifirm Market Interactions in Price and Advertising, Management Science, 45 (4), 499518. Xie, Jinhong and Marvin Sirbu (1995), Price Competition and Compatibility in the Presence of Positive Demand Externalities, Management Science, 41, 909926. Wang, Q. and Jinhong Xie (2005), A Structural Analysis of IntraStandard 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. 