NET Institute*

Transcription

1 NET Insiue* Working Paper #07-43 November 2007 Sofware Exclusiviy and he Scope of Indirec Nework Effecs in he U.S. Home Video Game Marke Kenneh S. Cors and Mara Lederman Roman School of Managemen, Universiy of Torono * The Neworks, Elecronic Commerce, and Telecommunicaions ( NET ) Insiue, hp:// is a non-profi insiuion devoed o research on nework indusries, elecronic commerce, elecommunicaions, he Inerne, virual neworks comprised of compuers ha share he same echnical sandard or operaing sysem, and on nework issues in general.

2 Sofware Exclusiviy and he Scope of Indirec Nework Effecs in he U.S. Home Video Game Marke Kenneh S. Cors Roman School of Managemen, Universiy of Torono Mara Lederman Roman School of Managemen, Universiy of Torono Firs Draf: February 2007 This Draf: November 2007 Absrac This paper invesigaes he scope of indirec nework effecs in he home video game indusry. We argue ha he increasing prevalence of non-exclusive sofware gives rise o indirec nework effecs ha exis beween users of compeing and incompaible hardware plaforms. This is because sofware non-exclusiviy, like hardware compaibiliy, allows a sofware firm o sell o a marke broader han a single plaform s insalled base, leading o a dependence of any paricular plaform s sofware on all firms insalled bases. We look for evidence of hese marke-wide nework effecs by esimaing a model of hardware demand and sofware supply. Our sofware supply equaion allows he supply of games for a paricular plaform o depend no only on he insalled base of ha plaform, bu also on he insalled base of compeing plaforms. Our resuls indicae he presence of boh a plaform-specific nework effec and in recen years a crossplaform (or generaion-wide) nework effec. Our finding ha he scope of indirec nework effecs in his indusry has widened suggess one reason ha his marke, which is ofen cied as a canonical example of one wih srong indirec nework effecs, is no longer dominaed by a single plaform. We hank Avi Goldfarb, Nùria Mas, Tim Simcoe, and seminar paricipans a IESE for helpful commens. The NET Insiue provided financial suppor. Boh auhors: 150 S. George S., Torono, ON M5S 3E6. Cors: kenneh.cors@ roman.uorono.ca; Lederman: mara.lederman@roman.uorono.ca

3 I. Inroducion The home video game marke has long been recognized as one in which nework effecs exis. As in many high-ech indusries, hese nework effecs are indirec. Home video games sysems consis of a console (he hardware) and games ha can be played on ha console (he sofware). Because he value of a console is derived from he games ha can be played on i, consumers prefer o buy a sysem wih a greaer variey of sofware. Because here are fixed coss o developing sofware, game publishers prefer o develop games for a console wih a large base of users. Thus, consumers value of a paricular console depends posiively hough indirecly on he number of oher users of ha console. This paper explores he scope of indirec nework effecs in he home video game indusry. While previous work has assumed ha nework effecs exis only beween users of a given console (for example, Clemens and Ohashi (2005)), we argue ha in recen years nework effecs have also come o exis beween users of compeing plaforms in he same echnological generaion. Ineresingly, his change in he scope of indirec nework effecs has occurred wihou any change in he degree of hardware compaibiliy. Home video games sysems have always been, and coninue o be, incompaible wih one anoher. As Kaz and Shapiro (1985) explain, he scope of indirec nework effecs in an indusry is ypically deermined by he degree of hardware compaibiliy. For example, if all hardware brands are incompaible nework effecs will operae a he brand level, while if all brands are compaible nework effecs will operae a he marke level. Hardware compaibiliy is he relevan consideraion in assessing he scope of indirec effecs because i deermines he size of he poenial marke ha a sofware firm can appeal o when rying o recoup is fixed coss of sofware developmen. Tha is, he degree of hardware compaibiliy deermines he se of echnologies whose users are poenial purchasers of a given piece of sofware. While changes in he degree of compaibiliy beween video game sysems have no been responsible for he change in he scope of nework effecs in his indusry, we believe ha changes in he degree of sofware exclusiviy have. Over he pas 20 years, he fracion of games iles ha are released on more han one console has increased from abou 12% o almos 40%. Jus as compaible hardware allows sofware providers o 1

4 spread he fixed coss of sofware developmen over muliple brands, non-exclusive sofware allows providers o spread he fixed coss of developmen over muliple plaforms. Sofware providers considering a muli-plaform ile will compare hese fixed coss (plus he fixed coss of poring he game across plaforms) o he revenue ha can be earned by selling he game o users of all of he plaforms on which he game will be released. Thus, once non-exclusive sofware is considered, he supply of games for any paricular plaform will clearly depend no only on he number of users of ha plaform, bu also on he number of users of oher plaforms on which hose games could be released. This gives rise o indirec nework effecs beween users of incompaible video game consoles. 1 Following exising work in his area, we will look for evidence of indirec nework effecs by esimaing wo relaionships: (1) he relaionship beween hardware demand and sofware variey; and (2) he relaionship beween sofware availabiliy and he insalled base of hardware. Our analysis uses monhly daa on U.S. hardware sales and sofware availabiliy for all major home video game sysems from We begin by specifying a sandard discree choice model of hardware demand. One imporan benefi of his demand model is ha i allows us o accoun for boh exclusive and non-exclusive sofware in a sraighforward way. In paricular, exclusive and nonexclusive iles need no be disinguished in he uiliy funcion because consumers uiliy from having a paricular game available on a console does no depend on wheher ha game is available on oher consoles. Of course, wheher a game is exclusive will affec he relaive uiliies of differen consoles and, in urn, heir marke shares. The nesed logi model ha we use accouns direcly for his, as each console s marke share is a funcion of he characerisics of all producs in he marke. We can hen use hese marke share expressions o illusrae he differenial effecs ha exclusive and nonexclusive iles have on demand. We hen esimae a reduced-form sofware supply equaion ha accouns for he possibiliy of boh console-specific and cross-console indirec nework effecs. We allow 1 In fac, he heory lieraure (for example, Kaz and Shapiro (1985) and Farrell and Saloner (1992)) has long recognized ha his kind of indirec nework effec can span incompaible plaforms in he presence of echnologies such as adapers and converers. The empirical lieraure has largely ignored his possibiliy, a leas in par because he indusries sudied have ypically been characerized by complee compaibiliy (e.g., CD players) or complee incompaibiliy (e.g., early home video games). 2

5 he supply of games for a paricular console o depend on boh he insalled base of ha console and he insalled base of compeing consoles in he same echnological generaion. By including his addiional insalled base measure in he supply equaion, we allow for he possibiliy ha he insalled base of compeing (and incompaible) hardware plaforms can increase he supply of games for a console because he fixed coss of non-exclusive releases can be spread across users of muliple plaforms. Furhermore, we will allow he coefficien on he compeiors insalled base erm o vary over ime o capure he fac ha sofware publishers incenives o produce nonexclusive sofware have increased. As we explain in greaer deail in he nex secion, we believe ha a rise in he imporance of licensed conen and oher conen coss ha are no plaform-specific, a decrease in poring coss, and a rise in he size and sophisicaion of independen game publishers have increased he araciveness of nonexclusive releases. As he araciveness of muli-plaform releases increases, he scope of indirec nework effecs a work in he indusry broadens. Our sofware supply specificaion allows us o measure his effec. Our empirical resuls suppor he exisence of boh a significan plaform-level indirec nework effec and an increasingly imporan generaion-level nework effec. The resuls of he demand esimaion indicae ha he demand for a paricular console increases wih he availabiliy of sofware for ha console. Furhermore, as expeced, exclusive games have a larger impac on demand han non-exclusive games. The resuls from he sofware supply equaion indicae ha he supply of games for a console depends posiively on he insalled base on ha console. The supply of games for a console also depends on he insalled base of oher consoles, wih his relaionship being negaive early in our daa and posiive in laer generaions. This implies a posiive indirec nework effec operaes a he generaion level by he end of he period we sudy. The video game marke is ofen cied as he canonical example of a ippy marke one in which indirec nework effecs lead o dominance by a single firm. The complee dominance of he indusry by Ninendo s NES sysem in he 1980s and early 1990s is ofen cied as evidence for his claim. 2 However, wih successive echnological generaions, his marke has become significanly less dominaed by any single console. 2 See, for example, Brandenburger and Nalebuff (1996), pp , or Shapiro and Varian (1999), p

6 Indeed, in each of he wo mos recen echnological generaions, hree compeing plaforms (hose of Ninendo, Sony, and Microsof) have reained sizable marke shares (see Figure 1). 3 While we do no esimae a dynamic model and herefore canno use our resuls o illusrae how changes in sofware exclusiviy affec he evoluion of marke shares over ime, we believe ha our empirical resuls provide a leas suggesive evidence as o why his marke has become less prone o ipping. Our resuls indicae ha non-exclusive sofware affecs a marke much in he same way ha compaibiliy does i changes he scope of indirec nework effecs. If nework effecs exis across users of differen plaforms, as our resuls indicae, hen i should be no surprise ha he endency of his marke o ip owards a single plaform has fallen. A he end of Secion V, we use our demand and supply esimaes o carry ou a simple exercise ha illusraes how he presence of generaion-wide nework effecs lowers he benefi ha a console manufacurer ges from simulaing sofware provision. This paper builds on a small bu growing lieraure ha seeks o esimae he role of indirec nework effecs in a variey of echnology indusries. I is mos closely relaed o wo recen papers ha esimae nework effecs in he home video game indusry. Clemens and Ohashi (2005) esimae he effeciveness of console price and sofware variey as alernae ways of simulaing hardware demand. Their empirical analysis implicily reas all sofware as exclusive o a plaform. Thus, hey do no disinguish beween he inroducion of exclusive and non-exclusive games in calculaing demand elasiciies nor do hey allow for he possibiliy of cross-plaform effecs in heir supply equaion. Their daa end several years earlier han ours, before he rend owards nonexclusive sofware had fully manifesed iself (hey sae ha, in heir sample, 17% of iles are available on more han one plaform), so he effec we focus on may no have been operaive in heir daa in any case. Their focus is insead on he dynamics of indirec nework effecs and he evoluion and price and sofware elasiciies over a console s lifecycle. 3 Figure 1 shows he long-run insalled base (IB) marke shares of he major plaforms in each echnological generaion. We define a plaform s long-run IB marke share as is IB marke share in he monh in which he firs major plaform of he nex generaion is launched. We believe ha he launch of he nex generaion acs as a good signal ha he previous generaion has reached a poin of mauraion. 4

7 Prieger and Hu (2006) also esimae indirec nework effecs in he video game indusry. They acknowledge he presence of boh exclusive and non-exclusive sofware and, in fac, are ineresed in esing wheher indirec nework effecs are sronger for exclusive games. However, hey approach his quesion by explicily separaing exclusive and non-exclusive games in he consumer uiliy funcion (which, as described above, we do no believe is he appropriae way o approach his quesion). Perhaps no surprisingly, hey do no ge sensible resuls from his specificaion. Oher papers ha explicily esimae indirec nework effecs in echnology indusries by modeling he complemenariies beween hardware and sofware include Gandal, Kende, and Rob (2000) in he CD marke and Nair, Chinaguna, and Dube (2004) in he personal digial assisan marke. 4 Our paper conribues o his lieraure by being he firs o explicily consider he role of sofware exclusiviy and is impac on he scope of indirec nework effecs. Our analysis indicaes ha he scope of indirec nework effecs depends on more han jus hardware compaibiliy. Raher, he scope of hese effecs is deermined by sofware providers abiliy o share fixed coss across plaforms which is possible if he plaforms are compaible, bu is also possible if he coss of poring sofware across incompaible plaforms are relaively low. This research draws aenion o he fac ha feaures of he sofware marke can also affec he scope of indirec nework effecs and, in urn, he likelihood ha a marke ips owards a single dominan plaform. Moreover, i raises he possibiliy ha hardware providers may have incenives eiher o sign conracs for exclusive sofware or o engineer heir hardware o affec he coss of poring sofware across plaforms. 5 The remainder of his paper is organized as follows. In he nex secion, we provide relevan background informaion on he indusry. Secion III describes he 4 There also exiss a relaed empirical lieraure on wo-sided markes (building on heoreical work by Roche and Tirole (2003) and Armsrong (2005)). For example, Rysman (2004) esimaes equaions for readership and adverising demand in he yellow pages marke, wih muliple direcories compeing for readership and adverising dollars in many ciies. Kaiser and Wrigh (2006) sudy he magazine indusry using a similar approach. 5 In fac, wo recen heory papers (Hogendorn and Yuen (2007) and Manena, Sankaranarayanan, and Viswanahan (2007)) focus on precisely his incenive for he console manufacurer o conrac for he provision of exclusive complemens and illuminae he ension beween he sofware provider s desire o serve a large marke and he console manufacurer s desire o differeniae is produc hrough he provision of unique complemens. 5

8 empirical approach. Secion IV describes he daa. Our resuls are presened in Secion V. A final secion concludes. II. Indusry Background The home video game marke is comprised of a small number of compeing, incompaible video game sysems (or plaforms ). A video game sysem consiss of hardware (a console ha is aached o a elevision se) and sofware (game iles on eiher carridge or CD). Sofware produced for a given hardware plaform canno be played on an alernae plaform; however, as described above, disinc versions of he same sofware ile may be produced for muliple hardware plaforms. Plaforms wih similar echnological characerisics are grouped ino generaions by indusry observers. There have been seven generaions of plaforms in he modern home video game indusry, spanning 1975 o he presen. We focus our analysis on he years 1995 o 2005 inclusive. This ime period covers he launch of mos of he plaforms in generaions five and all of he plaforms in generaion six. I also includes several plaforms from generaions hree and four which were sill acively selling during his period. 6 Table 1 presens he plaforms ha are included in our sample, grouped ino generaions. The able also shows heir dae of inroducion and basic echnological characerisics. Three echnical facors deermine he qualiy of a home video game sysem: (1) insrucion word lengh (in bis) of eiher he cenral processor (CPU) or graphics processor (GPU); (2) clock speed (in MHz); and (3) he amoun of RAM (in MHz). As he able indicaes, plaforms wihin a generaion are ypically quie similar on hese hree characerisics. Each video game plaform is conrolled by wha we call a console manufacurer and, as is eviden from Table 1, many of he same console manufacurers appear in each successive generaion (afer he firm s iniial enry, of course). 7 In addiion o developing he hardware and operaing sysem, a console manufacurer ypically also produces some 6 We ignore handheld game devices and PC games. 7 Noe ha while we use he erm console manufacurer, i is conrol of he operaing sysem, raher han lieral manufacuring of he hardware, ha is relevan for our purposes. For example, Sony is he console manufacurer for he PlaySaion2. This means ha Sony owns he operaing sysem for his plaform and is responsible for he R&D ha goes ino he developmen and mainenance of PlaySaion2 plaform. Wheher Sony ousources manufacuring of some or all of he componens of he hardware is of no consequence for our purposes. 6

9 sofware ha will run on his plaform (so-call in-house or firs-pary iles). The console manufacurer will also ener ino conracs wih independen sofware publishers o provide games for he plaform (known as hird-pary iles). Sofware publishers finance he developmen of he game (including obaining and paying for any licensed conen he game may use) and perform he markeing and disribuion of he ile. Game developmen (he acual programming) may be carried ou by a developmen eam inernal o he publisher or may be conraced ou o an independen game developer. Conracs beween console manufacurers and sofware publishers generally sipulae ha he console manufacurer is o provide sofware developmen ools o he publisher, while he publisher agrees o proec his inellecual propery. The console manufacurer reains he righ o approve games before hey are developed and released for he console. The conrac also specifies he per-uni royalies o be paid by he publisher o he console manufacurer. Finally, he conrac may specify wheher or no he game under developmen is exclusive o he console manufacurer. As menioned in he Inroducion, over he pas 20 years, his indusry has seen a significan increase in he prevalence of non-exclusive sofware. In Tables 2A and 2B, we documen his change in sofware exclusiviy, firs a he generaion level and hen a he plaform level. Noe ha in Table 2A, he level of observaion is he ile raher han he plaform-ile. These rends would be even more pronounced if repored a he plaform-ile level since non-exclusive games would be double- or riple-couned depending on he number of plaforms hey were released for. 8 Table 2A indicaes ha here has been a significan decrease in sofware exclusiviy. 88% of iles released in generaion hree were released for only a single plaform. By generaion six, only 61% were exclusive o a plaform. 9 The rend in hese averages has no been monoonic, however, as a greaer fracion of generaion five games han generaion four games were exclusive. This is due a leas in par o changes in he composiion of console manufacurers from generaion o generaion. Specifically, Sony firs enered he indusry 8 For example, suppose here are hree iles wih he firs being exclusive o one plaform, he second being exclusive o he oher plaform and he hird being available on boh plaforms. Then he fracion of iles ha are exclusive is wo-hirds while he fracion of ile-plaforms ha are exclusive is one-half. 9 When we use plaform-iles as he level of observaion, we calculae ha while 81% of plaform-iles in generaion hree were exclusive, only 37% of plaform-iles in generaion six were. 7

10 in generaion 5 and accouned for he majoriy of games in ha generaion. As Table 2B shows, Sony has also generally had a higher proporion of exclusive games han oher console manufacurers. Since exclusive iles are ofen games ha he console manufacurer publishes iself (i.e. in-house games), one migh wonder wheher he observed decrease in he exen of exclusive sofware is simply reflecing a decrease in he prevalence of in-house games. To invesigae his, he nex row of Table 2A calculaes he fracion of iles ha are exclusive o a single plaform, looking only a iles developed by hird-pary publishers. These numbers indicae ha he fall in exclusive iles is no due o a reducion in in-house publishing. The fracion of hird-pary iles ha are exclusive o a single console is also decreasing over his period. In Table 2B, we show he fracion of each console s games ha are exclusive. This able is useful for highlighing differences in exclusiviy across consoles as well as differences in exclusiviy across successive consoles produced by he same paren (i.e., differences across Ninendo s various sysems). For example, he able indicaes ha boh Ninendo s and Sony s generaions five consoles had a subsanially higher fracion of exclusive iles han heir generaion six consoles. The paerns apparen in Table 2A and 2B clearly beg he quesion of why nonexclusive sofware has become more prevalen in his indusry. While no he focus of his research in he sense ha we are ineresed in esimaing he effecs of changes in exclusiviy on he scope of nework effecs (and no explaining he change in exclusiviy), i is imporan o briefly discuss wha may be causing his rend. To moivae his discussion, we consider he incenive of a sofware publisher o release a game for one or more plaforms. Suppose ha here are wo consoles in he marke. If a sofware publisher releases a game ha is exclusive o one of hese consoles, he earns revenues from he sale of ha game o users of ha console, incurs he fixed coss of developing he game and pays royalies o he console manufacurer for each uni of he game produced. If he publisher insead releases he game on boh consoles, he earns revenues from he sale of he game o users of boh consoles, incurs he fixed coss of he developing he game as well as he fixed coss of poring he game o he second console, and pays royalies o 8

11 boh console manufacurers (which may be higher han he royalies paid on he exclusive game if he console manufacurer offers a lower royaly rae in exchange for exclusiviy). The publisher will prefer o develop he game for boh consoles if he addiional revenue from selling o users of he second plaform exceeds he addiional fixed coss (i.e., he poring coss) and he addiional licensing fees. Noe ha as he fixed coss of developing a game increase and/or he fixed coss of a poring a game decrease, games may increasingly be profiable as muli-plaform releases bu no as exclusive releases. Based on our reading of he rade press and oher indusry sources, we believe ha changes in he cos srucure and echnology of game developmen, an increase in he use of licensed conen in games, and changes in he srucure of he sofware indusry all influence sofware publishers incenives o develop non-exclusive games. Figure 2 shows one indusry analys s esimae of game developmen coss over he pas 25 years. This figure is roughly consisen wih daa we have seen elsewhere ha esimaes he average cos of game developmen in generaion five was around $1 million, while he average cos of game developmen in generaion six was $5-7 million (Lofus, 2006). As illusraed by he discussion above, when he fixed coss of game developmen increase, more projecs become viable only when hey reach a very large audience larger perhaps han any one plaform can provide. In order o recoup hese massive developmen coss, publishers have an incenive o release a ile on muliple plaforms. The composiion of developmen coss has also changed in imporan ways. Wih he rise o dominance of he CD-based console in generaion five, games have become relaively more conen -inensive. CDs make i cheaper o sore vas quaniy of graphical and musical daa in a game, compared o he prior echnology ha used semiconducor chip-based carridges. As a consequence, a larger fracion of he developmen coss has become aribuable o asks like music licensing or composiion and performance, moion-capure sudies, background ar and design (see Lofus (2006) and Reimer (2005)). Since hese coss are no specific o a plaform (for example, music is no operaing sysem-specific), he fracion of iniial developmen coss ha mus be duplicaed in order o por a game o a second plaform have shrunk. In addiion, poring coss have fallen because of he rise of sophisicaed cross-plaform developmen ools called middleware ha can dramaically reduce he coss of wriing a game for muliple 9

12 plaforms compared o he radiional complee rewrie (Reimer, 2005). 10 Middleware allows specific echnical aspecs of he game for example, 3-D animaion o be developed wihin a programming ool ha can provide oupu usable by he operaing sysems of more han one plaform. In sum, he echnology of game developmen has changed so ha more of he iniial coss incurred when wriing a game for is firs plaform are avoidable in poring he game for a second. This reducion in he relaive cos of poring clearly increases he araciveness of nonexclusive releases. 11 Furhermore, as shown in Table 2A, sofware publishers are increasingly relying on licensed conen and sequels. Indusry observers aribue his o an aemp o miigae risk of failure in an environmen wih skyrockeing developmen coss (Reimer, 2005). This move o a blockbuser model mimics an ofen cied developmen in Hollywood filmmaking, which experienced a similar simulaneous run-up in producion budges and increased reliance on sequels. No only may ballooning budges make he relaive predicabiliy of a licensed game seem aracive, bu i also inroduces anoher player ino he game developmen process. If he owner of he relevan inellecual propery (e.g., he Baman franchise) is pursuing a broad, muli-produc or muli-channel sraegy for disseminaing is conen and building/exploiing is brand, and if here are spillovers across markes (e.g., video game sales simulae acion-figure sales), i may provide publishers wih incenives (perhaps wih lower licensing fees) o develop a non-exclusive game based on is conen (even if, in he narrow conex of video games sales alone, i migh be more profiable o license is conen for developmen exclusively on one plaform). One final change ha migh have faciliaed he rise in nonexclusive games is he growh and mauraion of he sofware publishing indusry, which could conribue o he decline in he proporion of (almos always exclusive) in-house games eviden in he hird row of Table 2A. The final row of Table 2A shows ha he average number of iles released by an independen publisher has grown more han hree-fold over his period. As game developmen has increasingly become financed by publishers raher han by developers, and as developmen coss have soared, publishers have grown larger and 10 Also see he Poring in gaming enry in Wikipedia. 11 Manena, Sankaranarayanan and Viswanahan (2007) sae ha indusry figures indicae ha poring coss are now ypically in he range of 15% o 25% of he iniial developmen coss. 10

13 beer capialized, wih a number of he larges becoming publicly raded. This should correc some capial marke imperfecions likely presen in he indusry s earlier days and make in-house publishing less necessary as a means of simulaing sofware developmen due o lack of financing. Of course, here may be oher reasons ha console manufacurers wish o be involved in in-house publishing, bu he relaive increase in he independen provision of games may be in par due o his developmen. And, since inhouse games are almos always exclusive (apparenly because console manufacurers are relucan o share heir IP and developmen ools wih rival plaforms, in addiion o he fac ha i may in heir ineres o simulae plaform-specific demand), a rise in independen publishing may increase he prevalence of non-exclusive games. III. Empirical Approach Indirec nework effecs can be esimaed in wo ways. One can rea hem as direc nework effecs and esimae a direc relaionship beween he demand for a given hardware plaform and is insalled base (see Ohashi (2003) for an example). Or, one can explicily accoun for he feedback beween hardware and sofware by esimaing boh a hardware demand equaion (in which hardware demand depends on sofware availabiliy) and a sofware supply equaion (in which sofware supply depends on he insalled base of hardware). Finding a posiive effec of sofware availabiliy on hardware demand and a posiive effec of hardware insalled base on sofware supply esablishes he (indirec) posiive relaionship beween he demand for a hardware plaform and he exising number of users of ha hardware. This is he basic approach followed in he exising lieraure cied above, and we employ i as well. However, as we describe in he nex wo subsecions, we modify i o accoun for he changing scope of indirec nework effecs in his indusry. III.A. Hardware demand We model a consumer s choice of which (if any) hardware plaform o buy in a given monh as a discree choice problem in which he consumer evaluaes he uiliy ha he would receive from each poenial plaform and chooses he one plaform ha maximizes his uiliy. We include an explici ouside good so ha consumers also have 11

14 he opion of buying none of he plaforms. Following Berry (1994), consumer i s uiliy from purchasing console j, in monh is wrien as, (1) x j u = x β + αp + γsw + ξ + v ij j j j j where is a vecor of observed characerisics for console j in monh, is he price ij p j of console j in monh, SW j is a measure of sofware availabiliy for console j in monh ξ j, is a vecor of unobserved (o he economerician) characerisics for console j in monh (such as markeing or brand image), and consumer is assumed o choose he produc ha maximizes his uiliy. v ij is an idiosyncraic error erm. Each The coefficien γ measures he relaionship beween hardware demand and sofware availabiliy. We expec γ > 0, meaning ha greaer sofware availabiliy on plaform j increases a consumer s uiliy from console j. Because of daa limiaions, in mos of our specificaions we have o assume ha consumers care only abou he number of iles available for he console and no abou he qualiy of hose iles. Bu, because we know ha here is, in fac, significan heerogeneiy in he qualiy of games (and, indeed, only a small number of iles acually become his ), we also employ some qualiy-adjused sofware measures as robusness checks. Coninuing o follow Berry (1994), we le δ = x β + αp + γsw + ξ denoe he j j j j j mean valuaion of console j across all consumers, meaning we can inerpre v ij as he difference beween consumer i s valuaion of console j in monh and he mean valuaion. The disribuion assumed for v ij deermines he choice probabiliies and subsiuion paerns. We adop a nesed logi framework and group all inside goods (i.e.: all consoles) ino one nes and he ouside good ino anoher. This allows for correlaion in v ij across he inside goods, allowing hem o be closer subsiues wih each oher han hey are wih he ouside good. As Berry (1994) shows, wih hese assumpions on and by seing he mean uiliy of he ouside good o zero, he following linear esimaing equaion for one-level nesed logi can be derived, (2) ln( s j ) ln( s0) = x j αp j + γsw j + σ ln( s j / g β ) + ξ j v ij 12

15 where, s j / g is console j s wihin-group share (his is console j s share of all consumers who purchase any console in monh ) and ξ j is he economeric error erm. In our empirical specificaions, we divide ξ j ino a ime invarian componen, which we will esimae as a plaform fixed effec, and a ime-varying componen. 12 We also include dummy variables o conrol for he monh of he year. These capure he fac ha he perceived qualiy of all consoles may be higher in some monhs such as November and December, when parens are purchasing gifs for children. Furhermore, we conrol for he age of he console (wih eiher polynomials or dummy variables) o capure how perceived and acual qualiy changes wih a console age. 13 Finally, in some specificaions we replace he separae plaform and age effecs wih plaform-age fixed effecs. This is our mos flexible specificaion in he sense ha hese fixed effecs conrol for he unobserved qualiy of each plaform in each year of is life. I is worhwhile o highligh he way in which non-exclusive sofware (i.e.: game iles available on muliple compeing plaforms) affecs he demand analysis. Exclusive and nonexclusive sofware iles do no need o be disinguished from one anoher in he daa or in equaion (2). This is because he righ-hand side of equaion (2) is lierally he mean uiliy ha a consumer receives from purchasing produc j. As is eviden from he uiliy funcion in (1), wheher or no a paricular game is exclusive o console j has no impac on he uiliy ha a consumer derives from console j. Tha is, when evaluaing his uiliy from a given console, a consumer will consider he games ha can be played on ha console, bu no wheher or no hose same iles are also available on oher consoles. However, wheher or no sofware is exclusive o a console will affec he relaive uiliies of he differen alernaives and herefore which console a consumer ulimaely chooses. Games ha are available on muliple plaforms will increase he uiliy he consumer ges from each of hose plaforms and will, in urn, have lile effec on he probabiliy ha he consumer chooses one of hose plaforms over anoher. On he oher 12 Because we include plaform fixed effecs, he marginal uiliy of plaforms characerisics which do no change over ime (for example, wheher he plaform is carridge or CD based) will no be separaely idenified. 13 For example, consumers may use console age as a signal for how much longer hey expec sofware wriers o produce sofware for ha console. Alernaively, oher ypes of complemenary producs may emerge for a console (such as, gaming websies or magazines ha offer ips ) as i ages. The age variables will pick up boh of hese hings. 13

16 hand, an exclusive ile will increase he likelihood ha a consumer chooses a paricular plaform. Given his, one way o illusrae he differen effecs ha exclusive and nonexclusive sofware have is o calculae separae derivaives of demand wih respec o exclusive and non-exclusive sofware. 14 When calculaing he change in demand in response o a change in exclusive sofware, he change in sofware will affec only he aribues and uiliy of plaform j. In conras, when calculaing he change in demand wih respec o non-exclusive sofware, he change in sofware will affec boh he aribues of plaform j as well as he aribues of all oher plaform on which his game is available. In a logi model, he demand for any produc depends on he characerisics of all producs in he marke; herefore, an increase in exclusive sofware will clearly have a larger effec on demand han an increase in non-exclusive sofware. III.B. Sofware supply Our supply equaion generally follows he previous lieraure, wih modificaions ha allow us o esimae wheher and how he scope of indirec nework effecs in his indusry has changed. In paricular, we modify he sofware supply equaion so ha we can explicily esimae wheher, in successive generaions, he insalled base of compeing plaforms generaes a posiive spillover in he producion of sofware for plaform j. This would provide evidence ha he increase in non-exclusive sofware acs like an increase in compaibiliy in ha i changes he scope of indirec nework effecs from being plaform-specific o being generaion-specific. We esimae a reduced-form relaionship beween he variey of sofware available on a plaform and ha plaform s insalled base of hardware. 15 Specifically, we esimae he following equaion, (3) ln( SW ) j = j + Z jβ + γ 1IB j + γ 2IB j α + η j 14 Prieger and Hu (2006) ry o esimae wheher indirec nework effecs in he video game indusry are sronger for exclusive han non-exclusive games. They explicily disinguish he wo ypes of sofware in he consumer uiliy funcion find no difference (if anyhing, hey find ha non-unique games have a larger effec). However, given he discussion here, i is clear ha here is no reason o expec he uiliy of he wo ypes of games o differ. Raher, i is heir effecs on demand ha should differ. 15 This is basically he same reduced form model employed by Clemens and Ohashi (2005) and Prieger and Hu (2006) in esimaing he supply of sofware. The primary difference is ha we allow dependence of one plaform s sofware supply on all plaforms insalled base of hardware. 14

17 where, α j is a plaform fixed effec, Z j is a vecor of characerisics of console j in monh ha may affec firms incenive o supply sofware for ha console (such as he age of he console), IB j is he insalled base of console j in monh, IB j is he insalled base of all oher consoles in he same echnological generaion as plaform j, and is an error erm. Because we believe ha here have been several imporan echnological changes in he sofware side of his indusry over he period we sudy, we conrol for ime effecs wih eiher year or monh dummies. 16 η j We conrol plaform age because he diffusion of knowledge and experise relaed o programming for a specific console may increase he supply of developers over ime. The inclusion of plaform and year (or monh) dummies prevens us from also including plaform age dummies (his is he common cohor/age/year problem) so we insead include higher order polynomials of plaform age. 17 γ 1 capures he relaionship beween he supply of sofware for plaform j and is insalled base, while γ 2 capures he relaionship beween he supply of sofware for plaform j and is compeiors (combined) insalled base. We expec γ > 1 0, meaning ha increases in a plaform s insalled base simulae he provision of sofware for ha plaform. This is he source of he radiional plaform-level indirec nework effec. The sign of γ 2 depends on he naure of he echnology of sofware provision. A one exreme, imagine ha all developmen coss were compleely specific o a plaform, so ha wriing a version of he same game o run on a second plaform required replicaion of all he same seps and he same coss. If he supply of inpus o his process was perfecly elasic, hen he decision abou wheher o wrie a game for each poenial plaform would be a compleely independen decision. A sofware firm would simply calculae he required poenial marke size and hen develop he game for all plaforms whose insalled base (or projeced insalled base) exceeded ha hreshold. This would imply γ 2 =0. Alernaively, if some inpus o he sofware developmen process were scarce, or less han perfecly elasically supplied, hen such a model could imply a 16 These are acual monh dummies as opposed o he calendar monh dummies we include in he demand equaion. Obviously, hese monh dummies subsume he calendar monh dummies; however, we do include he calendar monh dummies when we include only year effecs. 17 See Hall, Mairesse and Turner (2005) for example. 15

18 γ 2 <0 ha is, some crowding ou of sofware developmen based on he growh of rival plaforms. In his case, he growh of a rival plaform is bad news for he focal plaform because he consequen increase in sofware developmen for ha plaform divers resources away from sofware developmen for he focal plaform. Under anoher inerpreaion, sofware providers migh use he insalled base of compeiors as a signal abou he likely evoluion of plaform j; for example, if plaform j s compeiors have large insalled bases, sofware providers may infer ha he marke is likely o ip away from plaform j and avoid wriing sofware for he plaform. This would again generae a negaive spillover on sofware developmen from he growh of he insalled base of rival plaforms. Now imagine insead ha he porion of he fixed developmen coss ha mus be replicaed o por he game o anoher plaform falls. This increases he araciveness of muli-plaform releases and inroduces he poenial for a posiive relaionship beween he supply of games for plaform j and he insalled base of plaform k ha is, γ 2 >0. Specifically, for a sofware provider who is conemplaing wriing a game for plaform j, he insalled base of plaform k represens an addiional se of cusomers over which he fixed coss of his game can be spread. As he coss of poring games o addiional plaforms fall and/or as he developmen coss of games increase, games ha may no be profiable if developed only for plaform j migh become profiable if developed for plaforms j and k. If so, he supply of games for plaform j will be direcly affeced by he insalled base of plaform k. This would give rise o generaion-wide indirec nework effec, and i is his relaionship ha we seek o es. Given he evoluion of he gaming indusry described in secion II, we specifically expec γ 2 o become more posiive (or less negaive) over ime. We es his hypohesis by allowing he coefficien γ 2 o vary by generaion so ha we esimae how he relaionship beween compeiors insalled base and he supply of games for plaform j changes over our sample period. We can hen combine he parameers of he demand and supply equaions o esablish he exisence and scope of indirec nework effecs in his indusry. In paricular, a finding of λ > 0 in he demand equaion and γ 0 in he supply equaion esablishes he presence of a plaform-level indirec nework effec. A finding of λ > 0 in 1 > 16

19 he demand equaion and γ 0 in he supply equaion esablishes he presence of a 2 > generaion-level indirec nework effec. Moreover, a finding ha γ 2 increases over successive generaions in our sample would indicae ha he scope of indirec nework effecs has changed from users of he same plaform o users of differen plaforms in he same generaion. Noe ha his change in he scope of indirec nework effecs can ake place even in he absence of physical compaibiliy beween he producs. III.C. Endogeneiy i. Hardware Demand Price, Wihin Group Share and # of Tiles are all poenially endogenous in he demand equaion. Wih he inclusion of plaform fixed effecs, he error erm in he demand equaion can be inerpreed as he deviaion of he unobserved qualiy of plaform j in monh from is average unobserved qualiy. This error erm may capure several sources of variaion in unobserved qualiy. Firs, i may capure changes in perceived qualiy ha resul from adverising campaigns (since hese would occur a differen poins in a plaform s life cycle, hey would no be conrolled for by he plaform fixed effecs). Second, i may capure changes in qualiy ha resul from he release of new informaion abou he plaform for example hrough posiive or negaive produc reviews. Finally, i may capure changes in qualiy ha resul from he emergence of complemenary producs (oher han sofware) ha enhance he value of he plaform for example, a websie ha provides ips on how o solve games. Because hese ypes of changes in unobserved qualiy will be considered by firms when seing prices and by consumers when making choices, hey will likely be correlaed wih boh Price and Wihin Group Share. In addiion, if changes in unobserved qualiy are persisen over ime (i.e.: if here is serial correlaion in he error erm in he demand equaion), hen # of Tiles may be endogenous as well. This logic is perhaps bes illusraed wih an example. Suppose ha in monh, plaform j has a high level of unobserved qualiy because he plaform received posiive reviews from gaming websies. This increase in qualiy will lead o higher demand from plaform j ha monh. This, in urn, will increase he insalled base of plaform j in monh +1 and, hrough he sofware supply relaionship, increase he 17

20 supply of games for plaform j in monh +1. If he increase in he unobserved qualiy of plaform j persiss (i.e., he posiive reviews in monh increase he perceived qualiy of plaform j in monh +1 as well), hen here will be correlaions beween # of Tiles in monh +1 and he error erm in he demand equaion in monh +1. Because we expec ha a leas some of he ime-varying componens of unobserved qualiy may persis for more han one monh, we rea # of Tiles as endogenous. Following Berry, Levinsohn and Pakes (1995), we consruc insrumens ha measure he exen of compeiion faced by a plaform as well as he exen o which ha compeiion comes from oher plaforms owned by he same paren. Specifically, our insrumens include he sum of each hardware characerisic (processor speed, memory, and processor word lengh in bis) over he compeing producs in he marke, he oal number of compeing plaforms in he marke, he number of compeing plaforms from he same generaion, and he number of compeing plaforms from he same manufacurer (i.e., he number of oher Ninendo consoles acively selling in he marke). We expec ha hese variables will be correlaed wih plaform j s price (because hey affec plaform j s abiliy o raise prices), plaform j s wihin group share (because hey affec he relaive uiliies of he differen opions), and plaform j s sofware (because, as we show in our modified supply equaion, hey affec sofware providers incenives o supply games o plaform j). However, hese insrumens should be uncorrelaed wih he error in he demand equaion because ha error erm is lierally he unobserved uiliy of plaform j in a given monh which is independen of he characerisics of oher offerings. Noe ha all of hese insrumens vary over ime for a given plaform (and are herefore no subsumed in he plaform fixed effecs) because he mix of plaforms in he marke changes wih enry and exi. Noe as well ha he endogeneiy concerns discussed here are likely o be considerably less severe in specificaions ha include plaform-age fixed effecs since hese allow he unobserved qualiy of a plaform o change in each year of is life. ii. Sofware Supply The insalled base variables may be endogenous in he sofware supply equaion. As above, his endogeneiy problem will resul if here is serial correlaion in he error 18

21 erm. Specifically, a high value of η j will simulae addiional games for plaform j in monh. This will increase he uiliy and sales of plaform j ha monh which will +1 increase he insalled base of plaform j in monh +1. If is correlaed wih, hen Plaform IB will be endogenous. Because he sofware available on plaform j also affecs he demand for compeing plaforms (and hence heir insalled base), he Generaion IB erms can be endogenous as well. To insrumen for a plaform s insalled base, we consruc a measure of ha plaform s price hisory by aking he average of is price in each monh since is launch. To insrumen for he insalled base of oher plaforms in a generaion, we use he average price hisory of all compeing plaforms in he generaion. In boh cases, we expec ha he price hisories will be correlaed wih he insalled base variables because hey influenced sales in prior periods. However, hey should be unrelaed o he error erm in he sofware supply equaion since sofware providers care only abou curren and fuure insalled base (and hus he only avenue by which pas prices should affec curren sofware availabiliy is hrough heir affec on pas sales). η j η j IV. Daa IV. A. Sources of Daa Our empirical analysis combines several sources of daa. Our daa on hardware prices and quaniies were obained from he NPD Group, a marke research firm. The NPD Group collecs daa from approximaely wo dozen of he larges game reailers in he Unied Saes. These reailers accoun for abou 65% of he U.S. marke. From his daa, NPD formulaes esimaes of figures for he enire U.S. marke. The NPD daa provides monhly uni and dollar sales of each console wih posiive sales. Dollar sales are divided by uni sales o obain an average monhly price for each console. Our hardware daa covers he period , inclusive. We supplemen hese daa wih informaion on he echnological characerisics and release daes of each console, which we colleced from a variey of sources including analys repors, company websies, and rade publicaions. Technological characerisics include processor speed, processor word lengh (8-bi, 16-bi, ec he basis of mos groupings of video game sysems), console memory capaciy, and wheher he sysem was CD-ROM based. While hese 19

22 characerisics are no separaely idenified from he plaform fixed effecs ha we include in our specificaions, hey are imporan because we use hem as insrumens for price. Our main source of daa on sofware is is a websie ha seeks o caalog all relevan informaion abou elecronic games (compuer, console, and arcade) on a game-by-game basis. 18 MobyGames provides a daabase of sofware iles ha includes he release dae of each ile, for each plaform on which i appears and for each counry in which i is released. The daa also conain he name of he publisher, he genre of he game, and an indicaion of wheher i uses conen licensed from anoher pary (like a movie sudio). This daabase in principle goes back indefiniely in ime, and is inended o capure all releases from he beginning of each plaform s life. 19 IV. B. Variables 20 i. Marke Shares and Insalled Base Measures Esimaion of he hardware demand equaion requires ha we consruc measures of each plaform s share of he oal poenial marke for video game consoles as well as is share of he share of he marke capured by all of he inside goods combined. Following Clemens and Ohashi (2005), we define he poenial marke for video games consoles in any monh o be he number of households wih a elevision (aken from he U.S. Census websie) less he combined insalled base of all acive plaforms (i.e.: we wan o subrac a measure of he number of consumer who are no in he marke because hey already own plaforms). The simples such measure would be he sum of each console s pas sales. We sar by consrucing his ype of plaform-level insalled base measure (Plaform IB). Since six of he en plaforms ha we sudy launch wihin he period of our daa, we can consruc heir pas sales simply using he sales figures from 18 See hp:// The informaion conained in MobyGames daabase is provided by he websie s creaors as well as from volunary conribuions. All informaion submied o MobyGames is checked by he websie s creaors. 19 Is earlies enries are in 1972, for which here are six releases for he Odyssey game sysem. 20 Variable names and definiions appear in Table 3a. Table 3b presens summary saisics. 20

23 our daa. 21 For he oher four plaforms, we obained daa on heir year-end 1994 insalled base from oher sources. The year-end insalled base figures for he plaforms in generaions hree and four come from Shankar and Bayus (2003) while he figure for 3DO comes from an analys repor. For hese four plaforms, we consruc Plaform IB by combining hese figures wih he NPD sales daa. Noe ha for each plaform, we runcae he ime series in he monh ha is sales firs fall under 1000, defining his as a plaform s exi from he marke. While we could use his Plaform IB measure in our consrucion of he oal poenial marke, i presens a problem. In paricular, his approach does no allow users of an old plaform o gradually re-ener he marke; raher, when an old plaform exis he marke by having is sales fall below 1000 in a monh, i yields a discree (and poenially huge) influx of new cusomers ino he poenial marke. We solve his problem by modifying he insalled base variable based on a depreciaion rae. Specifically, we calculae each monh s insalled base as a fracion of he previous monh s insalled base plus he previous monh s hardware sales, where he depreciaion rae varies wih he age of he plaform. 22 We call his variable Depreciaed IB. We focus on his paricular formulaion of he depreciaion rae because, pracically speaking, i yields declines in insalled base ha roughly coincides for mos plaforms wih heir exi from he marke defined by curren sales. We also run robusness checks using alernae ways of measuring insalled base. Having calculaed each plaform s depreciaed insalled base, we hen calculae he oal poenial marke for video game sysems in a monh as he number of U.S. households wih a elevision minus he sum of Depreciaed IB over all acive plaforms. We consruc a plaform s marke share in a monh as is hardware sales divided by he oal poenial marke and call his Marke Share. We consruc a plaform s wihingroup share (Wihin Group Share) as is hardware sales ha monh divided by he oal hardware sales of all acive plaforms ha monh. The share of he ouside good (which 21 While Table 2 gives a launch dae for Jaguar ha predaes our sample, is naional launch was no unil he end of 1994, so we ignore sales prior o our daa, which begins in January The monhly depreciaion rae (or rae of reenry ino he poenial marke) is imes he age of he plaform in years. This yields abou a 1% re-enry over he course of he firs year and abou a 7.5% reenry rae over he course of a plaform s enh year. 21

24 is needed o consruc he dependen variable for he demand equaion) is calculaed as one minus he combined marke shares of all of he acive consoles in a monh. Our sofware supply equaion includes measures of boh a plaform s own insalled base and he insalled base of he oher plaforms in is generaion. We use Plaform IB in he sofware supply equaion and calculae Generaion IB as he sum of Plaform IB over he compeing plaform s in a generaion. ii. Console Characerisics Because our hardware demand equaion includes console fixed effecs, non-ime varying console characerisics (such as echnical specificaions) are no separaely idenified. The hree ime-varying characerisics ha we include in he demand model are price, age and, of course, sofware availabiliy. We consruc he average price of each console in each monh (Price) by dividing he console s dollar sales by is uni sales. We measure he age of a plaform as he number of monhs ha have pas since he plaform s U.S. launch monh. We call his variable Plaform Age. In some specificaions, we measure a plaform s age in years since is U.S. launch. While he age of a console may no direcly affec consumers uiliy, we expec ha i maers indirecly. For example, consumers may use console age o predic how many more new games may be released for ha console, or hey may use i o predic when he nex generaion of machines may be launched. As menioned above, age may also capure he availabiliy of complemenary producs oher han sofware, such gaming magazines or websies. We use he MobyGames daa o measure he cumulaive amoun of sofware available on a console in a monh. Our primary measure of sofware availabiliy is simply he sum - from he plaform s release dae up hrough he presen monh - of all iles released on ha plaform in he US. We call his variable # of Tiles. This variable appears on he righ-hand side of he demand equaion and as he dependen variable in he supply equaion. More deailed informaion on wha sofware iles were acually selling a any poin in ime would likely yield a more accurae porrayal of he sofware available o consumers a he ime hey made heir hardware choice. While we do no have his informaion for he enire ime period covered by our hardware daa, we do have 22

25 monhly sales for each game ile exploi his sofware sales daa in wo ways. As we explain in Secion VI, we ry o V. Resuls This secion presens our empirical resuls. I proceeds in several seps. Firs, we presen he resuls of he hardware demand equaion and use he logi marke share equaions o illusrae he differenial impac of exclusive and non-exclusive sofware iles on hardware demand. We hen presen he resuls of he sofware supply equaion and show ha, over ime, posiive cross-plaform (wihin-generaion) nework effecs arise. Tha is, we find ha, conrolling for a plaform s own insalled base, in Generaion 6 he insalled base of compeing plaforms simulaes he supply of games for ha plaform. Togeher, he resuls of he hardware demand and sofware supply equaions indicae ha while plaform-specific indirec nework effecs sill exis, generaion-wide indirec nework effecs exis as well. Finally, we conclude his secion wih an empirical exercise ha illusraes he effecs of having nework effecs beween users of compeing consoles. V.A. Hardware Demand Resuls Tables 4 and 5 presen he resuls of he hardware demand esimaion. All of our models include plaform fixed effecs (excep for he final column of Table 4 which includes plaform-age fixed effecs) and calendar monh fixed effecs. All specificaions rea Price, # of Tiles and Wihin Group Share as endogenous. Recall ha he righhand side of our model is lierally he mean uiliy of produc j in monh and herefore he coefficiens ha we esimae on Price and # of Tiles are marginal uiliies of hese aribues. Thus, he magniudes of hese coefficiens are no, on heir own, informaive; however, he raio of he coefficiens can be used as a way o illusrae he relaive effecs of differen aribues. The firs hree columns of Table 4 esimae he hardware demand model using alernae ways of conrolling for plaform age. Because boh price and sofware 23 This daa is also from he NPD Group. Sofware sales are by plaform. Thus, if a given ile is available on muliple hardware plaforms, we know he sales on each plaform. 23

26 availabiliy will change wih age, i is imporan o carefully conrol for any oher ways in which plaform age may affec he demand for a paricular plaform. In (4-1), we include he linear age variable and is square, in (4-2) we include fixed effecs for age in years and in (4-3) we include fixed effecs for age in monhs. Each of hese specificaions yields quie similar resuls. In paricular, we find ha higher prices lower consumers uiliy from a plaform while greaer sofware availabiliy increase heir uiliy. The relaive magniudes of he coefficiens in (4-3) indicae ha decreasing he hardware price by $10 has roughly he same effec on uiliy as inroducing 20 new exclusive games. As expeced, he coefficien on log(wihin Group Share) lies beween zero and one and ranges from 0.54 o This suggess ha he inside producs are indeed closer subsiues for each oher han for he ouside good. As described in Secion III, if here are unobserved shocks o he qualiy of a plaform ha persis for more han one monh, hen he sofware variable may be endogenous. Of course, a second implicaion of unobserved changes in qualiy ha persis over ime is ha error erms in he hardware demand equaion will be serially correlaed. To accoun for his, we do wo hings. Firs, in he fourh column of Table 4, we re-esimae (4-3) using sandard errors ha are robus o arbirary auocorrelaion. The poin esimaes do no change, bu he sandard errors increase and he significance of he coefficiens on Price and # of Tiles is reduced. Second, in (4-5) we replace he plaform and age fixed effecs wih plaform-age effecs. Inuiively, hese esimae he unobserved qualiy of each plaform in each year of is life (i.e., raher han esimae one level of unobserved qualiy for, say, he Sony PlaySaion, we now esimae en differen levels). These fixed effecs will clearly do a beer job of capuring unobserved and poenially persisen changes in qualiy ha occur a differen poins in a plaform s life; moreover, he error erm in he hardware demand equaion is much less likely o exhibi serial correlaion wih heir inclusion. The disadvanage of hese fixed effecs, however, is ha hey are more demanding of he daa in he sense ha we can only idenify he effecs of he price and sofware variables from variaion in hese variables wihin a year for a given plaform (and here are many plaform-ages in which here is very lile variaion in one or boh of hese variables). When we include hese fixed effecs, we 24

27 again find a posiive and significan effec of sofware availabiliy on hardware demand hough he effec of price is no significan. Having obained esimaes of he marginal uiliy of sofware, we can now illusrae he differenial impac of exclusive and nonexclusive iles on hardware demand. As described in Secion III, here is no need o disinguish exclusive and nonexclusive sofware in he mean uiliy expressions. Raher, he logi demand sysem accouns for he fac ha a non-exclusive game affecs he uiliy of each plaform for which i is available. Using he nesed logi marke share equaions and he esimaes from specificaion (4-3), we compare he effecs on demand of en addiional exclusive games and en addiional non-exclusive games. We do his in he following way. Firs, we back ou he mean uiliy implied by our esimaes for each observaion. We hen use his o calculae he implied shares of he oal marke. Then, we recalculae he implied share for each plaform-monh under he assumpion ha ha plaform s mean uiliy has increased hrough he addiion of 10 sofware iles (i.e., is mean uiliy is increased by 10 imes he coefficien on # of Tiles in specificaion (4-3)). We hen calculae he resuling change in share and average hese changes over all observaions. The average share of he oal marke for all plaform-monh observaions is ; given he roughly 100 million households in he oal poenial marke, his amouns o abou 217,000 unis. The calculaion described above implies ha, on average, he inroducion of en exclusive games increases a plaform s share of he poenial marke by , or abou 30,500 unis per monh. We can compare his wih he esimaed effec of adding en games ha are compaible wih all available hardware plaforms. Such an addiion increases each firm s prediced share of he poenial marke by only , or abou 14,200 unis. Tha is, he relaionship beween hardware demand and exclusive iles is on average a lile over wo imes sronger han he relaionship beween hardware demand and iles available on all plaforms. In Table 5, we esimae specificaion (4-2) of our demand model using wo alernae sofware measures ha aemp o capure differences in he qualiy of games One could also check for robusness o differen assumpions on he iner-generaional compaibiliy beween plaforms from he same provider. In paricular, he PlaySaion 2 can play PlaySaion 1 games, 25

28 Firs, we consruc a weighed sofware measure ha explois he sofware sales daa ha we have for he years Specifically, we esimae a sofware sales equaion for each ile for which we have he sales daa. The righ-hand side variables of his equaion include a series of dummy variables capuring he size of he publisher, a dummy variable for wheher he ile is par of a series (i.e., a sequel o a previously released game), a dummy variable for wheher he ile includes licensed conen, and dummy variables indicaing wheher he game is par of one of four major franchise games (Zelda, Mario Brohers, Donkey Kong, and Sonic he Hedgehog). 25 We use he coefficien esimaes from he sales equaion o calculae he prediced sales of all iles in our MobyGames daabase. We hen scale hese sales projecions by he number of consoles sold for each plaform (his has a similar effec o including plaform fixed effecs, bu has he benefi ha i allows projecions o plaforms no ye inroduced during his period). Thus, our projeced sales variable is measured as dollars per console. 26 These projeced sales are summed up for each plaform o creae a weighed sofware availabiliy measure ha can be used in place of he simple coun of iles described above. We call his variable Weighed # of Tiles. (5-1) replaces he simple # of Tiles variable wih Weighed # of Tiles. When we include his qualiy-adjused measure of sofware, we again find a posiive and saisically significan effec of sofware on uiliy. In (5-2), we ry o capure sofware qualiy in a more direc way. We resric our sample o he period for which we have he sofware sales daa ( ) and consruc a sofware variable ha measures he number of iles on a plaform in a monh ha experience posiive sales. 27 We again find a posiive and saisically significan effec of sofware on uiliy. hough hese are no couned in he # of Tiles variable for PlaySaion 2. However, his is irrelevan since all our specificaions conain plaform fixed effecs, which absorb his fixed increase in iles available on PlaySaion We also ried including genre dummies bu hese proved insignifican. We have replicaed he resuls using publisher dummies in his regression and obained similar resuls. We presen resuls ha use he publisher size dummy variables because his specificaion allows us o projec sales for a larger se of iles (i.e., we can sill projec sales for games ha are published by firms ha do no exis during he period for which we have sofware sales daa). 26 For example, a value of one would represen a game ha was projeced o sell o 2% of he marke (a is size a mauriy) a $50, yielding $1 average revenue per console sold. 27 We esimae his regression using only he plaforms in generaions 5 and 6 because hese are he plaforms launched during or jus before he period for which we have his daa. Sofware launches for exremely maure plaforms are likely o exhibi differen paerns. 26

29 V.B. Sofware Supply Resuls This secion presens he esimaes from our sofware supply equaion. In all specificaions, he dependen variable is # of Tiles. All models include plaform fixed effecs and eiher year or monh fixed effecs (o conrol for general rends on he sofware side of he marke). Since we canno include plaform, year, and age fixed effecs, we conrol for he age of plaform using he square and cube of Age in Monhs. The insalled base variables are reaed as endogenous in all specificaions. In he firs column of Table 6, we esimae he relaionship beween he supply of sofware for a video game plaform and ha plaform s insalled base of hardware. As expeced, we find ha a posiive relaionship. Togeher wih he posiive effec of sofware on hardware demand, he posiive coefficien on Plaform IB esablishes he presence of plaform-specific indirec nework effecs. The mos ineresing and novel of our sofware supply resuls is he relaionship beween he supply of games for a plaform and he insalled base of is compeiors. In he second column of Table 6, we add Generaion IB - recall ha his variable measures he combined insalled base of compeing plaforms in a generaion. When we do no allow he coefficien on his variable o change by generaion, we find ha Generaion IB has a negaive and significan effec on he # of Tiles. Tha is, increasing he insalled base of a plaform s compeiors lowers he supply of games for ha plaform. As discussed in Secion III, he finding of a negaive coefficien on Generaion IB in a reduced-form model should no be surprising. In specificaions (6-3) (6-5) in Table 6, we allow he relaionship beween Generaion IB and # of Tiles o change over successive generaions. We implemen his by ineracing Generaion IB wih dummies for Generaions four hrough six. Since we have daa on only one plaform in Generaion hree (he NES), we canno esimae a coefficien on Generaion IB for ha generaion. 28 We incorporae hese addiional ineracions ino he model of (6-2) and presen he resuls as specificaion (6-3). The coefficiens on all hree of he Generaion IB erms 28 To insrumen for hese ineracions wih Generaion IB, we inerac our insrumens (described in Secion III.C) wih Generaion dummies. 27

30 are significanly differen from zero. Moreover, hey are monoonically increasing over successive generaions. The coefficiens imply ha, in Generaion four, increases in he insalled base of is compeiors lower he supply of games o a plaform. In Generaion five, he effec is sill negaive bu an order of magniude smaller. Finally, in Generaion six, increases in is compeiors insalled base simulae he supply of sofware for a plaform. The magniude of his effec is reasonably large wih he poin esimaes in (6-3) implying ha an increase in rivals combined insalled base is worh, sricly in erms of sofware availabiliy for a plaform, abou one quarer of wha an increase in one s own insalled base is worh. The posiive coefficien on Generaion IB*Generaion-6 esablishes he exisence of indirec nework effec ha are operaing a he generaion level, raher han plaform level. The hardware demand esimaes esablish ha he demand for a plaform is increasing in he availabiliy of sofware for ha plaform, and he sofware supply esimaes esablish ha in Generaion six he availabiliy of sofware for a plaform is increasing in he hardware sales of he enire generaion. Thus, users of compeing video game plaforms creae posiive exernaliies for each oher wihin a generaion. These resuls sugges ha a consumer s uiliy of purchasing a Microsof Xbox is no only increasing in he number of oher users of he Xbox bu also in he number of users of he PlaySaion2 because, wih high coss of developing common conen and low enough poring coss, heir combined size will induce sofware providers o wrie games ha are released on boh plaforms. Thus, he scope of indirec nework effecs in his indusry has changed. The remaining columns of Table 6 esimae sligh variaions on (6-3). Column four replaces he year fixed effecs wih monh fixed effecs and column five adjuss he sandard errors for arbirary auocorrelaion. The supply resuls are quie robus o boh of hese specificaion changes. V.C. Implicaions of he Esimaes We can illusrae he implicaions of he change in he scope of indirec nework effecs in his indusry by racing ou he effec on hardware sales of an exogenous change in sofware availabiliy. In markes wih indirec nework effecs, an exogenous 28

31 increase in sofware for a plaform has boh a direc effec on hardware demand (increasing uiliy and sales) and an indirec effec (since his direc effec simulaes he provision of addiional sofware, which furher increases uiliy and sales). This laer effec which is someimes called he viruous cycle illusraes how, in markes wih nework effecs, he srong ge sronger and he weak ge weaker (Shapiro and Varian (1999)). However, once nework effecs exis across users of compeing plaforms, his cycle is weakened. In paricular, an increase in (exclusive or non-exclusive) sofware for a plaform sill increases he insalled base of ha plaform, which sill increases he supply of games for ha plaform, bu now i also increases he supply of games for compeing plaforms. Thus, in his scenario, one could say ha he srong may ge sronger bu he weak may no ge as weak as hey would have absen his cross-plaform posiive spillover. Alhough we do no esimae a dynamic model, we can use he resuls of our demand and supply equaions o illusrae one ieraion of his cycle under he assumpion ha nework effecs do no exis across plaforms (i.e., using our Generaion four esimae of he coefficien on Generaion IB) and under he assumpion ha nework effecs do exis across plaforms (i.e., using our Generaion six coefficien on Generaion IB). We do his by calculaing he effecs of a hypoheical addiion of 10 exclusive iles for PlaySaion 2. We do his separaely (no cumulaively) for each monh from January 2004 onward, which is he period in which PS2, GameCube, and Xbox are he hree compeing plaforms in generaion 6, and hen presen he average effecs over hese monhs. We calculae hese effecs as follows. We begin wih he same calculaion employed in he illusraion of he demand esimaes, in which we back ou implied mean uiliies and calculae prediced shares of he oal marke. We hen increase PlaySaion2 s mean uiliy o reflec he addiion of en new games. We make no changes o sofware measures of he oher plaforms, consisen wih he inerpreaion ha hese sofware iles are exclusive o PlaySaion2. We hen recalculae he firms prediced marke shares based on hese new mean uiliies, calculae he difference from he original implied share, and muliply his by he size of he poenial marke. We refer o his increase in marke share as he direc effec of he increase in sofware for he 29

32 PlaySaion2, and we repor i in he firs row of Table 7. This row demonsraes wo consequences of an increase in sofware for PS2. Firs, he increase in sofware divers sales from boh of he oher plaforms o PS2. Second, oal sales grow since he sum of he numbers in his row is posiive; ha is, PS2 gains more sales han XBox and GameCube lose. We hen use he esimaes from our sofware equaion o predic how he number of iles provided on each plaform changes as a resul of he insalled base changes ha resul from he sales changes repored in he firs row. This akes ino accoun boh he change in each plaform s own insalled base and he change in oher plaforms insalled base. Thus, he implied effec on sofware provision will vary depending on wheher we use he Generaion 4 or Generaion 6 esimaes. Finally, we use his implied change in sofware availabiliy o calculae he indirec effec on hardware sales of he iniial hypoheical increase in exclusive iles for PS2. The second and hird row of Table 7 follow his exercise hrough using he Generaion 4 esimaes ha is, he esimaes ha show a negaive cross-plaform spillover. The fourh and fifh rows of Table 7 follow he exercise hrough wih he Generaion 6 esimaes, which show a posiive cross-plaform spillover. The conras in hese calculaions using Generaion 4 and Generaion 6 esimaes helps make concree he implicaions of he change in he scope of indirec nework effecs ha we find. Noe ha in boh scenarios (rows wo and four) he effec of he diversion of hardware sales o PS2 is o simulae addiional sofware provision for PS2 and o reduce sofware availabiliy for he oher wo plaforms. This is one sense in which he indirec nework effecs lead he srong o ge sronger and he weak o ge weaker an exogenous increase in sofware for one plaform leads o boh addiional sofware for ha plaform and a decrease in sofware for oher plaforms. However, he magniudes of he effecs are quie differen beween he wo scenarios. The broadening of he nework effec in moving from Generaion 4 esimaes o Generaion 6 esimaes cus he implied follow-on increase in sofware for PS2 in half, from 0.35 iles o 0.18 iles. However, he negaive effec on he oher plaforms is cu in boh cases by more han a facor of 15 (-0.32 o and o ). In his sense, he broadening of he nework effecs leads o a scenario in which he srong sill ge sronger, bu he weak do no suffer nearly 30

33 he same penaly from furher increases in he leader s sales. This is, of course, precisely because of he posiive spillover ha he leader s increased sales creaes in he way of sofware provision for he smaller plaforms. The hird and fifh rows of Table 7 show how his effec on follow-on sofware provision ranslaes ino an indirec effec on hardware sales. The numbers here exhibi a similar paern: he broadening of nework effecs in Generaion 6 somewha weakens he indirec benefi o he leader bu dramaically reduces he penaly o he weaker firms. VI. Conclusion In his paper we invesigae he scope of indirec nework effecs in he home video game indusry and argue ha, despie he fac ha all home video game sysems are incompaible, indirec nework effecs increasingly exis beween users of rival sysems. This is because muli-plaform releases (or non-exclusive sofware iles) allow sofware publishers o spread he fixed coss of game developmen over users of compeing plaforms. Over ime, sofware publishers incenives for muli-plaform releases have increased as he fixed developmen coss of common conen have soared, while he fixed coss of poring a game o addiional plaforms have fallen. Boh of hese increase he relaive profiabiliy of non-exclusive sofware releases, which in urn lead o he presence of cross-plaform indirec nework effecs. Our empirical resuls suppor his inerpreaion. We esimae a model of hardware demand and sofware supply ha indicaes ha while plaform-specific indirec nework effecs exis, in recen years generaion-wide indirec nework effecs have also come o exis. Thus, he scope of indirec nework effecs has changed. We also carry ou a simple illusraive exercise ha shows why he presence of posiive cross-plaform spillovers weakens he benefi ha a plaform ges from a sraegy aimed a simulaing demand, such as he inroducion of new sofware. While we do no esimae a dynamic model, we believe ha our findings are suggesive of why, in recen generaions, his indusry has no been dominaed by any single console. Furhermore, our resuls indicae ha hardware compaibiliy is no he only facor affecing he scope of nework effecs in an indusry. Raher, non-exclusive sofware creaes anoher avenue by which crossplaform spillovers can arise, as changes in sofware developmen and poring echnology 31

34 increase he araciveness of expanding he poenial marke hrough muli-plaform releases. 32

35 References Armsrong, Mark, 2005, Compeiion in Two-sided Markes, working paper. Berry, S. (1994), Empirical Models of Produc Differeniaion, RAND Journal of Economics, 25, Berry, S., J. Levinsohn, and A. Pakes, 1995, Auomobile Prices in Equilibrium, Economerica 63, Brandenburger, A. and B. Nalebuff, 1996, Co-opeiion, New York: Currency Doubleday. Clemens, M. and H. Ohashi (2005), Indirec Nework Effecs and he Produc Cycle: Video Games in he U.S., , The Journal of Indusrial Economics, 53, Farrell, J. and G. Saloner, 1992, Converer, Compaibiliy, and he Conrol of Inerfaces, Journal of Indusrial Economics 40, Gandal, N., M. Kende, and R. Rob (2000), The Dynamics of Technological Adopion in Hardware/Sofware Sysems: The Case of Compac Disc Players, RAND Journal of Economics, 31, Hall B., J. Mairesse, L. Turner, 2007, Idenifying Age, Cohor and Period Effecs in Scienific Research Produciviy: Discussion and Illusraion Using Simulaed and Acual Daa on French Physiciss, Economics of Innovaion and New Technology, 16(2), Hogendorn, C. and K.Y. Yuen, 2007, Plaform Compeiion wih Mus-Have Componens, working paper. Kaiser, U. and J. Wrigh, 2006, Price Srucure in Two-Sided Markes: Evidence from he Magazine Indusry, Inernaional Journal of Indusrial Organizaion, 24, Kaz, M., and C. Shapiro (1985), Nework Exernaliies, Compeiion and Compaibiliy, American Economic Review, 75(3), Lofus, Tom, 2006, Top Video Games May Soon Cos More, MSNBC.com, hp:// Manena, R., R. Sankaranarayanan, S. Viswanahan, 2007, "Exclusive Licensing in Complemenary Nework Indusries," NET Insiue Working Paper No Nair, H., P. Chinaguna, and J.P Dube (2004), Empirical Analysis of Indirec Nework Effecs in he Marke for Personal Digial Assisans, Quaniaive Markeing and Economics, 2,

36 Ohashi, H., 2003, "The Role of Nework Effecs in he US VCR Marke, ," Journal of Economics & Managemen Sraegy 12, Prieger, J. and W. Hu, 2006, An Empirical Analysis of Indirec Nework Effecs in he Home Video Game Marke, NET Insiue Working Paper Reimer, Jeremy, 2005, Cross-Plaform Game Developmen and he Nex Generaion of Consoles, Arsechnica, hp://arsechnica.com/aricles/paedia/hardware/crossplaform.ars. Roche, Jean-Charles and Jean Tirole, 2003, Plaform Compeiion in Two-Sided Markes, Journal of he European Economic Associaion, 1, Rysman, Mark, 2004, Compeiion Beween Neworks: A Sudy of he Marke for Yellow Pages, Review of Economic Sudies, 71, Shankar, Venkaesh and Barry Bayus, 2003, Nework Effecs and Compeiion: An Empirical Analysis of he Home Video Game Indusry, Sraegic Managemen Journal, 24, Shapiro, C. and H. Varian, 1999, Informaion Rules, Boson: Harvard Business School Press. 34

37 Figure 1 Long-Run Insalled Base Marke Share, by Generaion NES Oher SNES Genesis Oher N64 Saurn PlaySaion 3DO Jaguar GameCube Dreamcas PlaySaion2 Xbox Source: Generaion 3, Power Play (A); Generaion 4, Power Play (B); Generaions 5 and 6, NPD daa. 35

38 Figure 2 Game Developmen Coss Source: hp://arsechnica.com/aricles/paedia/hardware/crossplaform.ars/2 36