econstor Make Your Publications Visible.

Transcription

1 econsor Make Your Publicaions Visible. A Service of Wirschaf Cenre zbwleibniz-informaionszenrum Economics Malerba, Franco; Nelson, Richard; Orsenigo, Luigi; Winer, Sidney G. Working Paper Verical inegraion and dis-inegraion of compuer firms: a hisory friendly model of he co-evoluion of he compuer and semiconducor indusries Papers on economics and evoluion, No Provided in Cooperaion wih: Max Planck Insiue of Economics Suggesed Ciaion: Malerba, Franco; Nelson, Richard; Orsenigo, Luigi; Winer, Sidney G. (2006) : Verical inegraion and dis-inegraion of compuer firms: a hisory friendly model of he co-evoluion of he compuer and semiconducor indusries, Papers on economics and evoluion, No This Version is available a: hp://hdl.handle.ne/10419/31831 Sandard-Nuzungsbedingungen: Die Dokumene auf EconSor dürfen zu eigenen wissenschaflichen Zwecken und zum Privagebrauch gespeicher und kopier werden. Sie dürfen die Dokumene nich für öffenliche oder kommerzielle Zwecke vervielfäligen, öffenlich aussellen, öffenlich zugänglich machen, verreiben oder anderweiig nuzen. Sofern die Verfasser die Dokumene uner Open-Conen-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gesell haben sollen, gelen abweichend von diesen Nuzungsbedingungen die in der dor genannen Lizenz gewähren Nuzungsreche. Terms of use: Documens in EconSor may be saved and copied for your personal and scholarly purposes. You are no o copy documens for public or commercial purposes, o exhibi he documens publicly, o make hem publicly available on he inerne, or o disribue or oherwise use he documens in public. If he documens have been made available under an Open Conen Licence (especially Creaive Commons Licences), you may exercise furher usage righs as specified in he indicaed licence.

2 # 0619 Verical Inegraion and Dis-inegraion of Compuer Firms: A Hisory Friendly Model of he Co-evoluion of he Compuer and Semiconducor Indusries by Franco Malerba, Richard Nelson, Luigi Orsenigo and Sidney Winer The Papers on Economics and Evoluion are edied by he Evoluionary Economics Group, MPI Jena. For ediorial correspondence, please conac: evopapers@econ.mpg.de ISSN Max Planck Insiue of Economics Evoluionary Economics Group Kahlaische Sr Jena, Germany Fax: by he auhor

3 Verical Inegraion and Dis-inegraion of Compuer Firms: A Hisory Friendly Model of he Co-evoluion of he Compuer and Semiconducor Indusries Franco Malerba* Richard Nelson** Luigi Orsenigo*** and Sidney Winer**** * CESPRI, Bocconi Universiy, Milan, Ialy ** Columbia Universiy, New York *** Universiy of Brescia and CESPRI, Bocconi Universiy, Milan, Ialy **** Wharon School, Universiy of Pennsylvania, Philadelphia December 2006 Anna De Paol Andea Pozzi and Davide Sgobba provided an invaluable conribuion o he developmen of he model. Suppor from he Ialian Naional Research Council (CNR) and from Bocconi Universiy (Basic Research Program) is graefully acknowledged. 1 1

4 ABSTRACT In his paper we presen a hisory-friendly model of he changing verical scope of compuer firms during he evoluion of he compuer and semiconducor indusries. The model is hisory friendly, in ha i aemps a replicaing some basic, sylized qualiaive feaures of he evoluion of verical inegraion on he basis of he causal mechanisms and processes which we believe can explain he hisory. The specific quesion addressed in he model is se in he conex of dynamic and uncerain echnological and marke environmens, characerized by periods of echnological revoluions puncuaing periods of relaive echnological sabiliy and smooh echnical progress. The model illusraes how he paerns of verical inegraion and specializaion in he compuer indusry change as a funcion of he evolving levels and disribuion of firms capabiliies over ime and how hey depend on he co-evoluion of he upsream and downsream secors. Specific condiions in each of hese markes he size of he exernal marke, he magniude of he echnological disconinuiies, he lock-in effecs in demand exer criical effecs and feedbacks on marke srucure and on he verical scope of firms as ime goes by. 2 2

5 1. Inroducion The hisory of he compuer indusry conains periods when a leas he large compuer firms were verically inegraed, designing and producing heir basic elecronic componens, and periods where even he large compuer firms bough hose componens from specialized suppliers. Over all bu he earlies par of his hisory, he key elecronic componens have been semiconducors. To a large exen he remarkable increases in compuaional power and reducions in he coss of compuaion ha have been achieved over his hisory have been made possible by advances in semiconducor echnology, which on a leas wo occasions has been marked by he developmen of radically new kinds of semiconducor devices. This paper explores how he paern of advance of semiconducor echnology, and he demands for semiconducors, have influenced he srucure of he compuer indusry, and in paricular he exen o which large compuer firms have been verically inegraed or no. The ool of exploraion is a hisory friendly model. In he following secion we describe he paerns, paricularly he dynamic paerns of verical inegraion and disinegraion of large compuer firms, ha we are ineresed in explaining. Then we consider various heories of verical inegraion, and lay ou he heory ha seems mos plausible o us. As we explain nex, hisory friendly modeling is a mehod for exploring he logic of a qualiaive heory and is adequacy o explain he phenomena i purpors o explain. This discussion ses up our presenaion of he model, and is runs. We pull srands ogeher in our concluding secion. 2. The Hisorical Phenomena o be Explained In his paper, we can only briefly illusrae he hisory of he changing verical boundaries of compuer firms in erms of heir verical inegraion ino semiconducors beween he early 1950s and he mid 1980s. For a full discussion of hese issues, see for example Langlois and Roberson (1996), Malerba (1985), Krickx (1995), Bresnahan and Malerba (1999), Langlois and Seinmueller (1999), Bresnahan and Greensein (1999). The hisory of verical inegraion and dis-inegraion of compuer firms summarized here refers o he American case, because he echnological and markes leaders in he indusry were American firms. Therefore we will no focus on Europe or Japan, which presen ineresing bu differen ypes of dynamics. Before we ell he sory, i is necessary o make hree remarks. Firs, we consider verical inegraion and specializaion by compuer firms along a period of hiry years which was quie urbulen in erms of markes and echnologies. The compuer indusry saw he inroducion of mainframes firs (1950s), followed by minicompuer (1960s) and finally by personal compuers (1970s). In he semiconducor indusry ransisors were he main producs during he 1950s, inegraed circuis were inroduced in he 1960s and microprocessor were launched in he 1970s. Second, we will focus on he developmen and producion of sandard semiconducor componens. While large compuer firms have always produced some ypes of cusom componens in-house, his has no been he case for sandard semiconducor componens: in specific sages of he evoluion of he compuer indusry, sandard semiconducor componens have been produced in-house by compuer firms, while in oher periods hey have been purchased on he marke by specialized producers. Third, o keep he analysis simple, in his paper we will no discuss second sourcing or inermediae 3 3

6 organizaional forms such as neworks and parial inegraion. 1 A he very beginning of he indusry, compuer firms produced mainframes and were no verically inegraed. The firs compuer producers IBM, Burroughs, Univac Rand, NCR, Conrol Daa, GE and RCA - mainly purchased receiving ubes componens (he elecronic componens ha preceded ransisors) from he open marke. Afer he inroducion of ransisors (early 1950s) some of he larges firms such as IBM, RCA and GE became parially or oally verically inegraed ino ransisors. Conversely, he smaller mainframe firms purchased semiconducor componens on he marke. 2 In he compuer indusry, IBM began o pull ahead in mid-1950s and i came o dominae he world marke for accouning machines wih he inroducion of he 1401 in 1960, due o is hree pronged invesmens in R- D, producion and markeing. The paern of verical inegraion by he leading mainframe firm - IBM- was reinforced when a new echnological disconinuiy - inegraed circuis - ook place in he early 1960s. IBM verically inegraed ino inegraed circuis developmen and producion (firs wih hybrid inegraed circui echnology (SLT) and hen wih monolihic devices). Three main reasons explain his paern. Firs, inegraed circuis embedded sysem elemens and hus required close co-ordinaion beween he sysem and he componen producer in he design and developmen of boh componens and sysems. Second, semiconducor designs became more and more sraegic for sysem developmen, and herefore heir design, developmen and producion was kep in-house for fears of leakage of sraegic informaion. Third, he rapid growh of he mainframe marke and laer on of he minicompuer marke (1960s and 1970s) generaed fears of shorages of various key semiconducor componens among some of he larges compuer producers. As a verically inegraed company, in he early 1960s IBM launched a new mainframe produc: he sysem 360. Compaible Sysem 360 lines allowed IBM o exploi economies of scale and scope. The Sysem 360 led IBM o mainain he dominance of he world compuer marke during he 1960s. 3 A major change in he paerns of verical inegraion of he leading compuer producer ook place in he 1970s afer he inroducion of radically new semiconducor 1 In he hisory of he indusry hese forms have been used exensively for various reasons. Firs, in order o be secure of he supply of semiconducors in case of large inernal demand of componens, he major compuer producers ofen asked heir semiconducor suppliers o license ou heir semiconducors so ha hey could benefi from a sure second source. Second, in order no o run he risk of overcapaciy in heir in-house producion of componens and o check for he qualiy of heir in-house producion, verically inegraed producers ofen only parially inegraed ino semiconducor producion, producing a fixed amoun of semiconducors inernally and buying he res from he marke. Finally, various forms of neworking in erms of cooperaion in he developmen and of join producion of semiconducors have aken place during he hisory of he indusry. 2 Transisor echnology improved grealy during he 1950s, and was sold o boh compuer producers as well as he miliary and oher markes. Developmens in ransisors enabled significan improvemens in mainframe performance, and some reducion in coss. 3 Inegraed circuis opened he possibiliy of designing compuers - minicompuers - ha had a considerable amoun of power, could be produced a a much lower coss han mainframes and could be direced o a differen marke: medium size firms and laboraories. In he semiconducor indusry, demand coming from compuer producers, oher markes as well as he American miliary and NASA was saisfied in large par by new merchan specialized producers. 4 4

7 componens microprocessors - which opened a new fas expanding compuer marke personal compuers-. The major compuer leader IBM dis-inegraed from he producion of microprocessors, because IBM confroned a major semiconducor indusry leader Inel which emerged in his new quickly developing componen marke. Inel and he oher microprocessor firms could innovae and grow rapidly because hey benefied from a very large and increasing demand coming no only from compuers, bu also form oher final markes - such as elecommunicaions, consumer elecronics, auomobile, and so on. Disinegraion by IBM was accompanied by an increase in concenraion in he semiconducor indusry due o he leadership of Inel. Also he new firms ha had enered he personal compuer marke since he second half of he 1970s - Compuer, Radio Shack and Commodore firs, and a sream of new sar ups laer were all specialized compuer firms. These firms preferred o buy advanced semiconducor devices from large capable and innovaive microprocessor firms, raher han ry o verically inegrae. Thus Inel became he de-faco leader for he microprocessor marke. An episode ha confirms he dynamics of inegraion and specializaion ha we have jus recouned regards he enry by IBM ino personal compuers. When in he early 1980s IBM decided o ener he personal compuer marke, i did ha as a non-inegraed producer, buying is own semiconducor componens - as well as peripherals and sofware - from ouside suppliers. Also in his case, his decision was aken because IBM faced already quie capable producers of sandard semiconducor componens, i needed o speed up personal compuer producion and i did no have advanced inernal capabiliies in his respec. The decision o develop personal compuers joinly wih a leading semiconducor componen producer (Inel) - as well as wih a sofware producer (Microsof) - led IBM o a successful enry, bu no o he dominaion of he personal compuer indusry (Bresnahan and Greensein, 1999). 3. Theorizing Abou Verical Inegraion Wha kind of heory is able o explain he paerns of verical inegraion and specializaion of compuer firms presened in he previous secion? The currenly leading heories of verical inegraion and specializaion are mainly based on some version of he ransacion coss approach. They focus heir aenion on he marke failures ha may emerge in he exchange of goods and services under paricular condiions and propose he view ha hierarchical coordinaion is o be considered as a subsiue for marke ransacions in hose cases. 4 The analysis of hese issues may be framed in differen heoreical seings: for insance, eiher bounded raionaliy (as in Williamson, 1975) or full raionaliy (as in conrac heory) may be assumed. The language of sraegic ineracions is increasingly used in his conex. Choices abou inegraion and he relaive efficiency of markes vs. hierarchies are viewed as deermined by a calculaion ha weighs incenives advanages of markes agains he governance advanages of hierarchical organizaion. 4 As i is well known, ransacion coss are likely o be presen in cases of asse specificiy, asymmeric informaion and unclear definiion of propery righs. Under hese condiions, limiaions on effecive wriing and/or enforcemen of conracs leave room for opporunisic behaviour, ofen leading o sub-opimal soluions. 5 5

8 Wihou denying he clear relevance of ransacion coss, we simply noe here ha his approach has a disinc saic flavor and quie obviously considers ransacions as he main uni and exchange as he primary objec of analysis. Technologies, he properies of he goods and he characerisics of he agens are aken as given and he processes of verical inegraion/ specializaion are commonly examined as a choice a a given momen of ime. In his paper, on he conrary, capabiliies and echnology are a he base of our appreciaive explanaion of verical inegraion and specializaion. This is done in a dynamic, evoluionary seing. Following Nelson and Winer (1982) and he capabiliy-based view of he firm (Teece and Pisano 1994, Teece, Rumel, Dosi and Winer, 1994), we sugges ha a cenral facor explaining he verical scope of firms is he process of accumulaion of capabiliies a he firm and indusry levels. Capabiliies are accumulaed over ime by firms hrough a variey of learning processes in specific echnological, producive and marke domains (Teece and Pisano, 1994). Such compeencies end o be ypically sicky, local ad specific. Heerogeneiy across firms is herefore likely o be a permanen feaure of indusries and he acual disribuion of capabiliies across firms in upsream and downsream indusries is likely o bear a fundamenal influence on he verical srucure of firms. For example he decision o specialize is elicied and criically depends on he acual exisence of upsream suppliers a leas as compeen as he inegraed firm iself. In his framework, when producs are sysems wih various componens and subsysems, he abiliy o coordinae and inegrae he design of such sysems and componens may consiue an imporan compeence in is own righ and a significan source of compeiive advanage (Langlois and Roberson, 1995). Such advanage can be (more han) offse by consideraions relaed o he risk of geing suck in inferior echnological rajecories, especially a imes of rapid and uncerain echnological change, or when suppliers are able o offer significanly superior producs. However, decisions o specialize and o verically inegrae are no enirely symmerical. A firm conemplaing he opion of resoring o exernal sources for he supply of paricular componens can direcly evaluae he relaive qualiy of is inernally produced produc as compared o ha available from he exernal supplier. In he opposie case, such a comparison canno be so direc and i involves expecaions on he abiliy o design and produce in-house. Moreover, if a firm decides o disconinue he developmen and producion of cerain componens, i migh find i difficul o resume such aciviies laer on and in any case ime and effors are required. Thus, hese decisions are no enirely flexible as ime goes by. In addiion, he verical scope of firms is o be analyzed no simply by considering he capabiliies of a firm in isolaion, bu in is relaionships wih he oher paricipans in he relevan indusries (Jacobides and Winer 2005). For example, he degree of heerogeneiy and he disribuion of capabiliies are crucially shaped by he processes of marke selecion, which ends o promoe he growh of more efficien firms and of he relaed organizaional arrangemens and o penalize he laggards. Thus, marke selecion amplifies he impac of differeniaed capabiliies on he verical scope of firms. If specialized firms have superior capabiliies, selecion will push for greaer specializaion; and vice-versa. Or, for insance, he growh of a compeen supplier (or of a vibran indusry) is likely o induce processes of specializaion of he sysem firms, as he supplier becomes able o offer increasingly beer producs. In urn, he process and he loci of capabiliy developmen feed back on he condiions deermining he enry of new firms. Thus, verical inegraion or dis-inegraion can 6 6

9 bear profound effecs on he paerns of compeiion wihin an indusry, creaing he condiions for he enry and growh of new compeiors exploiing capabiliies developed in differen conexs. 5 In sum, a key poin of his paper is ha verical inegraion and specializaion are shaped by he co-evoluion of capabiliies, he size of markes and he srucure of indusries. The growh and dynamics of compeencies in each one of wo verically relaed indusries influence he evoluion of he oher secor and shape verical inegraion and specializaion (Langlois and Roberson, 1995, Jacobides and Winer, 2005). The previous discussion idenifies capabiliies as a cenral facor affecing he verical boundaries of compuer firms. In paricular, in his paper capabiliies refer o he accumulaion of firms compeences in specific echnological and marke realms and o he coordinaion and inegraion capabiliies of componens and sysems in he developmen and producion of new final producs. Wihin his framework, a se of variables affec he level, ype and accumulaion of capabiliies in various ways. In paricular we will concenrae on four of hem: a) The size of firms. Firm size affecs he amoun of R&D effor and hus he accumulaion of compeences (and consequenly he qualiy of sysems and componens produced). In addiion, size affecs verical inegraion because very large sysem firms may need a secure supply of componens. b) The size of he marke. A large marke allows for he enry of new specialized firms and for he growh of capable companies. This is a very Smihian reasoning placed ino a capabiliy perspecive. c) The marke srucures of wo verically relaed indusries. The above discussion abou he need o consider he disribuion of capabiliies among indusry paricipans may be linked o srucure of indusry: for example, a monopolisic sysem indusry ends o become verically inegraed ino componens when confroned wih a fragmened upsream indusry composed by small firms. On he conrary, a fragmened sysem indusry ends o specialize when confroned wih a monopolisic componen indusry. d) Major echnological disconinuiies and he relaed compeence desrucion. Major disconinuiies change he knowledge base and he ype of demand and lead o he enry and growh of firms wih oally new compeences. Under hese condiions, esablished verically inegraed firms face pressures owards dis-inegraion; The discussion developed above and he variables ha have been idenified allow us o recas and explain he dynamics of verical inegraion by compuer firms presened in he hisory in secion 2. A he beginning of he compuer indusry, mainframe firms sared specialized and over ime a company gained he leadership and an almos monopolisic posiion in he sysem indusry. This company verically inegraed, as i was large, needed securiy of supply of componens and once inegraed, is large profis in sysems led o rapid echnological advance in is componens. Conversely, for he independen ransisor producers he exen of he exernal marke was no so big o spur an increase in heir size comparable o 5 Causaion runs in he oher direcion, oo: he process of capabiliy developmen depends very much on he verical scope of an indusry. Le s ake componens developmen and producion. Specialized firms ha compee wih oher specialized firms accumulae knowledge and capabiliies differenly from verically inegraed firms. 7 7

10 ha experienced by he sysem producers. A dominan componen company did no emerge and verical inegraion of sysem firms furher reduced he opporuniies of growh for componen companies. When inegraed circuis were inroduced, he verically inegraed mainframe producer faced pressures owards verical dis-inegraion since he new componen firms were able o produce beer producs. However, since he exernal marke for he new ypes of componens was sill no large enough, specialized componen producers remained relaively small (compared o he larges sysem producer) and did no have massive resources for R&D and innovaion. Afer an iniial period of srong echnological urbulence, when he new echnology sared o maure, he sysem company verically inegraed ino inegraed circuis: verical inegraion allowed close coordinaion beween componens and sysems, componens have become more sraegic for sysem developmen and producion and securiy of supply reasons became sringen. Dis-inegraion of he leading mainframe producers ook place insead when a hird echnological disconinuiy microprocessors - ook place. This ime he verically inegraed mainframe producer faced a new cohor of componen producers ha could benefi from he sharp echnological disconinuiy and hus supply vasly superior producs compared o inegraed circuis, so ha caching up in componens by an esablished verically inegraed mainframe producers was slower. Moreover a much larger exernal marke - no dependen only from mainframe producers allowed hese new semiconducor firms o inves much more in R&D, grow quickly and achieve high levels of componens qualiy. In his process, a leading componen company emerged and dominaed he indusry. Specializaion characerized also he new personal compuer producers, because hese firms remained relaively small as compared o he leading microprocessor company, which supplied he compuer indusry as well as oher large exernal markes. 4. The logic of hisory-friendly models We believe ha he differenial developmen of capabiliies for designing and producing semiconducors among firms, he causal driving facor we highlighed in Secion 3, has been he principal force behind he paern of verical inegraion and disinegraion of compuer firms described in Secion 2. Bu how o es ha heory, or a leas demonsrae is plausibiliy? The vehicle used in his paper is he design and running of a hisory friendly model. Hisory-friendly models are inended o enhance undersanding of paricular ineresing and imporan economic phenomena, in his case he swings in verical inegraion and disinegraion in he American compuer indusry. They are closely focused on hese phenomena, as conrased wih being inended o illuminae broad and general issues, alhough wha one can learn from a successful hisory friendly model may go well beyond he paricular phenomena on which i is focused. Hisory friendly models generally are simulaion models, and his one is. The aim of hisory friendly modeling is no o explain, in he sense of closely maching hrough a simulaion, he quaniaive values observed in he hisorical episode under invesigaion, nor is he specificaion of he model parameers driven by he objecive of geing as close as possible o acual empirical values of variables in he acual conex being modeled. Raher, he objecive is o explore wheher paricular mechanisms and 8 8

11 forces buil ino he model can generae, and in ha sense explain, he paerns in quesion. The design of a hisory friendly model is guided by he heories, generally verbal, ha informed observers and empirically oriened economiss who have analyzed he phenomena have pu forh as heir causal explanaions, and which he model builders find plausible as well as ineresing. Hisory friendly modelers, like ourselves, believe ha much of producive economic heorizing is presened as explanaions of paricular empirical phenomena by hose who know a lo abou he empirical deails. Much of he heorizing discussed in Secion 3 is of his sor. However, we also believe ha i is difficul, someimes impossible, o check ou he logic and he explanaory power of such verbal qualiaive heorizing, wihou formalizing he argumen. A hisory friendly model is buil on a simplified formal represenaion of he heory being considered, and aims o es he consisency and power of ha heory by exploring he performance of he model. The hisory friendly model presened and esed here is focused on he facors affecing firm capabiliies described in Secion 3, and presumes ha a cenral facor deermining wheher compuer firms produce heir own semiconducors or buy hem from a supplier is where he sronges capabiliies of semiconducor design and producion are locaed. In being oriened his way, our model of he deerminans of verical inegraion or specializaion places he emphasis differenly han do he ransacion cos heories ha have been prominen in he indusrial organizaion lieraure. The quesion hen is: how well can such a model do in explaining he observed paern? 5. The model 5.1 An Overview The model we describe in his secion is designed o capure, in highly sylized form, he key elemens of he verbal explanaion ha we find mos persuasive for he observed dynamics of verical inegraion and, laer, verical disinegraion in he American compuer indusry over he period beween he early 1950s and he lae 1980s. The dynamics of he model, as he verbal accoun, is shaped by characerisics of he marke for compuers, he echnologies deermining compuer performance, and exogenous developmens in semiconducor echnology, all of which led o changes over ime in he incenives and capabiliies compuer firms had o inegrae verically.. Firms in he compuer indusry compee for cusomers. A any ime a compuer produced by a paricular company is characerized by wo aribues ha are relevan o is poenial purchasers: performance, and cheapness (he inverse of price). Oher hings equal, he share of marke sales gained by a paricular compuer is a funcion of he meri of is design in erms of hese wo aribues relaive o he meri of oher compuers on he marke, alhough sales are also influenced by brand loyaly and ineria of cusomers. In urn, boh of hese design aribues are deermined by he way he compuer is designed as a sysem, and he qualiy of he semiconducors ha go ino ha sysem. Boh of hese design facors end o improve over ime as a resul of R&D done on sysems, and on semiconducors. Sysems R&D is done by compuer produces. R&D on semiconducors is done by semiconducor firms, and also by compuer firms ha ha v e 9 9

12 verically inegraed ino he producion of heir own semiconducors. Compuer firms ha sell more compuers, on eiher marke, are more profiable han firms ha sell less. High profis induce and permi a firm o grow and as profiable firms expand heir sales, hey increase heir R&D spending. Since on average higher R&D spending enables a firm o make larger improvemens in he compuers i sells, here is a srong poenial in his model for a dominan firm o emerge hrough he dynamics of compeiion. Diminishing reurns o R&D end o damp his endency, as does brand loyaly, a leas in he shor run. The pah of semiconducor echnology plays a paricularly imporan role in he model. In paricular, i is assumed ha here are hree disinc eras. In he firs, all semiconducors are discree ransisors. In he second era, inegraed circuis emerge. In he hird, microprocessors. Wihin each era, here is coninuing improvemen in he basic semiconducors being produced and sold, bu he adven of a new ype of semiconducor is associaed wih a speeding up of he rae of progress. Also, before he adven of microprocessors all compuers are mainframes. The developmen of microprocessors no only permis improvemen in he design meri of mainframes, bu also enables a differen kind of compuer o be designed and produced, PCs, which are significanly cheaper han mainframes alhough having less impressive performance. Mainframes appeal o cusomers who grealy value high performance; we will call hese big firms. PCs appeal o a differen group of cusomers, small users. The adven of new ypes of semiconducors is marked by he enry of new specialized semiconducor firms, who compee wih esablished producers and each oher for sales. Semiconducor firms sell no only o compuer firms, bu o oher users, and he emergence of inegraed circuis and of microprocessors simulaes he developmen of hese oher markes as well as demand from compuer manufacurers. The inroducion of PCs and he opening of a new marke for compuers induces he enry of new compuer firms specializing in PCs. The cenral orienaion of his model is owards facors ha explain wheher compuer firms are specialized, buying on he marke he semiconducors hey employ in he compuers hey produce, or wheher hey are verically inegraed, designing and producing heir own semiconducors. In he model, all compuer firms are born specialized. Bu once in he compuer business hey can decide o inegrae and design and produce heir own semiconducors. And if inegraed, hey can decide o ge rid of heir semiconducor design and producion un and go o he marke again for heir componens. The decision o be inegraed or no parly ress on he advanages of being able o design boh he componens and he sysem ogeher, which can be considerable. I also parly ress on judgmens as o he imporance of being able o conrol he supply of one s inpus, and on wheher he firm believes i can produce semiconducors as or more efficienly han specialized suppliers. In general, large semiconducor firms are more likely han small ones o believe hey can efficienly design and produce heir own semiconducors. This endency o inegrae verically is enhanced if he indusry producing semiconducors for sale is fragmened, and here is considerable experience wih he kind of semiconducors in use. On he oher hand, verical inegraion is deerred if here is a large compeen semiconducor supplier who 1010

13 spends a lo on R&D, or when a new kind of semiconducor is jus emerging and he bes ways o design and produce hem is highly uncerain. In he former case, even a large compuer firm may doub ha i can produce semiconducors as efficienly as he dominan specialized supplier. Wheher or no large semiconducor suppliers come ino exisence is a funcion of he size of he marke for semiconducors ouside of compuers, as well as he porion of he compuer indusry ha purchases raher han makes he semiconducors i uses. This is he model in broad ouline. In he remainder of his secion we describe some of he imporan deails Compuers A he beginning of he simulaion, firms (numbering 12 in he curren parameer seing) sar producing and selling compuers. A compuer is characerized by wo characerisics: is performance, and is cheapness (he inverse of is price). As noed, available componen echnology iniially permis only he producion of mainframes. Laer, he adven of microprocessor echnology enables he developmen of PCs. PCs have less performance han mainframes, bu are cheaper. The wo differen kinds of compuers appeal o wo differen kinds of cusomers. As a consequence of firms R&D invesmen, and he advance of componen echnology, he characerisics of compuers of a given ype improve over ime. The advance o mainframes is along a ray mapping ou he performance cheapness characerisics of ha kind of compuer. The advance of PCs also is along a ray, characerized by a higher raio of cheapness o performance han he mainframe ray. The disance ou along he ray associaed wih is paricular compuer ype defines he meri of design (M, or Mod ) of a paricular compuer. To keep he noaion simple, in wha follows we do no explicily denoe wheher he M refers o a mainframe or a PC compuer. The conex will make ha disincion obvious. In urn, compuers are produced by combining wo main inpus, sysems (s) and componens (c).6 The level of he meri of design, M, is given by a CES funcion: 1 [ ( ) ( )( ) ] C ρ S ρ τ M + τ M ρ M = A (1) 1 wih A > 1, 0 < τ < 1 and ρ > -1. The elasiciy of subsiuion is: δ = 1/ (1 + ρ). In he CES funcions he weigh aribued o he Mod of componens (τ) is always higher han he weigh on he Mod of sysems. PCs have a comparaively higher weigh on componens as compared o MF (i.e. τ PC > τ MF. ) and he elasiciy of subsiuion, δ, is higher in PCs han in MFs. Thus, improvemens in componens are refleced more powerfully in PCs han in MFs. 7 6 Sysems are always produced by compuer firms and canno be sold separaely from compuers. 7 The rajecories followed by firms in he space of he characerisics are assumed o be fixed and equal among firms producing compuers of a given ype. Given he level of he meri of design for a compuer and he slope of he MF and PC rajecories, he values of cheapness (w) and performance (z) ha appear ino he demand funcion are defined using he following rigonomeric formulas: 111

14 Differen compuer ypes, eiher mainframes or PCs, are produced by differen companies. We exclude he possibiliy of diversificaion in his model. 5.3 Demand for compuers Cusomers of compuers are characerized by heir preferences abou he wo aribues ha define a compuer design - performance (z) and cheapness (w). There are wo buyer groups, one consising of big firms and ohers who are especially ineresed in performance, and care less abou cheapness, and he oher of small users who are especially concerned abou cheapness, and who value performance less han do big firms. Big firms buy mainframe compuers. Small users buy PCs, when hese become available afer microprocessors come ino exisence. Wihin each buyer group, here are many individual buyers. 8 While buyers are idenical wihin each group in erms of he preferences ha lead hem o one ype of compuer, hey differ in specific behavior because hey have differen hisories, which are parly randomly deermined in he model. Buyers respond o he compuers offered by differen firms according o he relaive meri of heir producs, bu also according o oher consideraions, including heir specific buying hisory. Markes are represened as being characerized by fricions of various sors, including imperfec informaion and sheer ineria in behavior, brand-loyaly (or lock in) effecs as well as sensiiviy o firms' markeing policies. These facors are capured in a compac form by he share of compuer brands in he overall marke for ha ype ime -1: he larger he share of he marke ha a produc already holds, he greaer he likelihood ha a cusomer will consider ha produc. Finally, here is a sochasic elemen in buyers choices beween differen compuers. We represen he marke process by characerizing he probabiliy disribuion of buyer s choices among he differen compuers of he desired ype. This probabiliy is a renormalized counerpar of a purchase propensiy ha depends on he meri and marke share of a paricular compuer. Formally, he propensiy, L i, for compuer i o be sold o a buyer a ime is given by as: L α 1 β 1 i, = M (1 + s i, 1 ) (2) where is he marke share, β1 is he exponen indicaing he bandwagon effec on s i, 1 compuer marke. The probabiliy Pr of he compuer i being sold o a buyer a ime is given by: w = cos( β ) M z = sin( β ) M where β indicaes an angle expressed in radians, and is differen beween PC and MF. 8 To avoid unnecessary compuaional burdens, we do no scale buyers o a realisic size relaive o firms: he model assumpions simply capure he idea ha buyers are small and numerous. Opionally, each buyer in he formal model could be inerpreed as represening a number of consumers or small firms. 1212

15 L, i Pr = (3) L In shor, he demand for a compuer depends posiively on is Mod and on is marke share in he previous period. The probabiliy ha a compuer from a specific firm is purchased by a buyer is proporional o he propensiy, normalized o sum o one. Then for a compuer firm, he oal compuers sold are equal o M i imes he number of buyers. If α1 = 1, a 1% increase in meri across all firms selling o a buyer group implies a 2% decrease in price/performance, which gives rise o a 1% increase in he number of compuers sold. In his sense, he implied price elasiciy of marke demand for compuers of consan qualiy is.5 α1. i i 5.4 The marke for componens Componens are bough by specialized producers of compuers and also by cusomers in oher markes (i.e., he exernal markes). As described above, here are hree differen componen echnologies, which become available a differen imes. In his version of he model, he ransisor, inegraed circui and microprocessor echnologies become available respecively a periods 1, 40 and 120. We inerpre a period as a calendar quarer. A he beginning of he simulaion and a he ime of each echnological disconinuiy a new cohor of firms (12 in his version of he model) eners he marke, producing componens wih he laes available echnology. The demand for componens, faced by componen firms, comes from wo sources: (i) Demand for componens from users oher han compuer firms. The size of his exernal marke is exogenous, paramerically deermined, and firms gain probabilisically a fracion of i as a funcion of heir meri of design and of heir previous marke share, as in he compuer marke. Exernal demand plays a criical role in he model, since i allows componen producers o survive and grow in he early sages of developmen of a new echnology and o improve he meri of heir componens. (ii) Demand for componens from compuer firms which have decided o ousource componen producion (specialized compuer firms). When a specialized compuer firm seeks componen supplies, i scans he marke for poenial suppliers. Compeiion among componen producers is modeled in a fashion ha parallels he compeiion of compuer producers. A specialized compuer producer conracs wih a componen producer according o a probabiliy funcion ha reflecs he relaive echnical meri of he componens offered by differen suppliers: he higher he meri of a componen, he higher he probabiliy ha is producer signs a conrac wih a compuer producer. Moreover, like he demand for compuers, he demand for componens is influenced by bandwagon and lock-in effecs capured by he previous marke share. 1313

16 Formally: L C α 2 β 2 = M c (1 + s 1 ) (4) and C L C Pr = (5) C L i where M C is he meri of design of he componen, Li C is he propensiy of componen producer i o be seleced, s i, 1 he probabiliy of a supplier o be seleced. is he marke share of firm i in he previous period and C Pr is A componen firm ha signs a conrac sells a number of componens reflecing is cusomer s compuer sales and he number of componens required per compuer; in he curren simulaions ha parameer is se o one. Afer signing he conrac he compuer firm is ied o he componen supplier for a cerain number of periods, which is a parameer of he model. When his period expires, a new supplier migh be seleced, using he same procedure, if he firm sill decides o buy componens on he open marke. The exernal marke is concepualized in he same way as he compuer marke, i.e., i comprises a number of heerogeneous buyer groups or submarkes o which componen firms may sell. However, he submarkes of he exernal componen marke are no modeled C explicily. A firm simply ges a fracion of he oal value of he exernal marke equal o. Pr 5.5 Firms' behavior and echnical progress A he beginning of a simulaion, firms sar wih a given meri of design, hey sar o make operaing profis and inves in R&D. Operaing profis, π, are calculaed in each period as: π (6) = q p q o where q i is he number of compuers sold, depending on he meri of design level and on he number of buyers araced by he firm, p is he price and o is he producion cos of a compuer. Price is obained by adding a mark-up, η, o coss: p o (1 + ) (7) = η Coss are in urn derived from he meri of design achieved by a compuer, considering ha he price of a compuer mus be equal o he inverse of he achieved cheapness. The price of componens charged by componen suppliers is deermined symmerically by adding a fixed mark-up o uni producion coss. 1414

17 For specialized compuer and componen firms, R&D expendiures are calculaed following a simple rule of humb: a consan fracion of profis (100% in his version of he model) is invesed in R&D in each period. By invesing more in R&D, firms buy hemselves higher probabiliies o increase heir meri of design. (The R&D spending of inegraed compuer producers is discussed below.) Technical progress is modeled using he draw scheme used in Nelson and Winer (1982). There are wo such draw schemes, one for he componens and one for sysems, which differ only as regards he means and variances of he normal disribuions involved. Thus, inegraed firms have wo echnical progress funcions, while specialized compuer firms and componen firms have only one echnical progress funcion, respecively for sysems and for componens. According o hese schemes, in each period firms draw possible values of he naural logarihm of Mod from a normal disribuion. The number of draws (d) ha any one firm can ake is se proporional o is R&D spending (R); he parameer of proporionaliy v is he cos of a draw: R d = (8) v When his calculaion fails o reurn an ineger value, a random choice is made beween he adjacen inegers values in such a manner as o make he expeced number of draws equal o he resul of he calculaion in (8). For example, a resul of 2.75 ranslaes ino 3 draws wih probabiliy.75 and 2 draws wih probabiliy.25. The disribuion of poenial new Mod values is more favorable when he curren value is higher; in oher words, echnological change is parly cumulaive a he firm level. The disribuion also becomes more favorable wih he passage of ime, because i reflecs he influence of public knowledge -- a variable exogenous o he indusry and conceived as represening he level of knowledge available in published academic research, specialized journals, or echniques ha have become widely known). The deails of hese assumpions follow. Public knowledge is specific o each basic componen echnology. When a new echnology is inroduced, is corresponding level of public knowledge is lower han ha reached by curren echnology, bu hen i grows faser and surpasses he public knowledge of he older echnology. The rae of growh of public knowledge approaches an asympoic value characerisic of he echnology, wih laer echnologies showing more rapid growh han earlier ones. An inegraed compuer firm is assumed o adop he new echnology meaning ha is R&D resuls subsequenly reflec he public knowledge of he new echnology -- when he mean of is own disribuion for componen draws becomes inferior o he level of he public knowledge of he new echnology. In each period, he values of he Mod obained hrough he firms draws are compared wih he previous value of Mod, and he highes among hese values is kep. Thus, more draws increase he likelihood o ge a higher Mod for boh sysems and componens. The resul of an individual draw is deermined iniially as a value for he naural logarihm of he Mod value drawn from a normal disribuion. The mean of his disribuion is a weighed average of he logarihm of he Mod (ln(m)) a ime -1 of firm i and of he logarihm of he level of publicly available knowledge, K, a ime. The sandard deviaion of he disribuion 1515

18 is a parameer σ. Exponeniaing he resul of he individual draw yields he value o be compared wih he previous Mod (and oher draw resuls, if any). M i, k = h ln( M i, 1 ) + (1 h ) ln( K ) (9) where ( ) k ϕ 1 k K = l K e 1 >c k (10) n c k where l and n are parameers and c k is he dae of inroducion of componen echnology k (k =TR,IC,MP). A higher value of l yields a higher rack for he public knowledge rajecory, and a higher value of n means ha he asympoic growh rae φ k is approached more quickly. Wih n se o a high value, which is ypically he case in our simulaions, a new echnology appears and hen improves raher abruply o quickly surpass he prevailing one (in erms of public knowledge). By seing he parameer h close o zero or one, he model is capable of represening exreme cases where progress is in he indusry is, respecively, science-based or cumulaive. In he former case, firm R&D expendiures are no he ulimae drivers of progress, hey essenially cover he cos of commercializing innovaive possibiliies creaed ouside he indusry. In he laer case, firms are creaing he effecive echnology, in a sepby-sep cumulaive process. Since he cumulaion is firm-specific, whereas public knowledge is a shared resource, i is clear ha firm Mod values should be expeced o diverge from each oher o a greaer exen when h is high han when i is low. We now urn o a quesion ha is cenral o he logic of verical inegraion in he model, he deerminaion of R&D expendiures for inegraed compuer firms. Firs, such firms make he same expendiures on sysems R&D ha a specialized compuer firm of he same Mod would make; hey spend heir profis on sysems R&D. While he producion coss of inegraed compuer firms producers are a funcion of heir achieved meri levels, he producion coss of specialized producers are insead deermined as he coss of he sysem plus he cos of buying he componens on he markeplace, i.e. he price charged by he paricular supplier from which he compuer company is buying. In he model, we assume ha an inegraed and a specialized firm having he same compuer Mod have also he same producion coss for a compuer. For a given componen Mod, he producion cos of inernally produced componens is equal o he producion cos of he exernally produced componens. The addiional componen coss aribuable o he mark-up charged by componen suppliers and which are saved by an inegraed firm - are invesed in componen R&D by inegraed producers and reaed as a cos in he shor run. The R&D, however, confers long-run benefis. If hese R&D benefis are sufficienly large, he inegraed form is favored and ends o persis by virue of is higher Mod values. The exen of he benefis depends on he exen o which inegraed producers enjoy coordinaion advanages as compared o specialized producers, by virue of he fac ha hey can produce componens ailored o heir sysems and, as a consequence, he produciviy of heir R&D effors on componens is enhanced. To represen his effec in he model, we 1616

19 assume ha heir componen R&D expendiures are coslessly augmened by a cerain facor, which we call spillover (f). This facor is cenral o he economic advanages of inegraion vs. specializaion. In sum, componen R&D - COM RErR C - of an inegraed compuer producer is: C C R c + f R = η (11) where c C is he cos of is componen and c C * η is he difference beween he price he componen would command in he open marke and is cos for he producer. Since specialized componen firms are assumed o spend heir operaing profis on R&D, his formulaion means ha he inegraed form is economically equivalen o he specialized form when he spillover, f, is equal o zero, bu is in a superior overall cos posiion if f is posiive. Thus, for any posiive value of f, inegraion is he economically superior form of organizaion for R&D direced o incremenal advances in a given echnology bu he consequences of his for indusry evoluion are qualified by he fac ha i is he specialized componen firms ha pioneer enirely new echnologies. Specialized compuer producers inves all of heir R&D on sysems and obviously do no enjoy he coordinaion advanages. Componen suppliers spend all heir R&D on he developmen of componens. 5.6 Verical Inegraion and Specializaion Specialized compuer producers may decide o verically inegrae ino semiconducors, if hey hink ha hey can design and produce componens ha are comparable in meri o hose offered by specialis suppliers. As noed previously, his is more likely o be he case if compuer producers are large enough compared o exan suppliers, so ha hey can fund a larger flow of R&D expendiures. The decision o verically inegrae also depends probabilisically on he age of he componen echnology. These consideraions are refleced in he following formulaion. The probabiliy of inegraion for each compuer firm is deermined as follows. Le: ϑ1 Ak, q, = min ; 1 i C g q ϑ 2 V (12) where: A k ( k =TR,IC,MP) = (Saring ime of Technology γ); q is he number of compuer sold by he compuer producer; q C is he number of componens sold by he larges componen producer and g is a parameer. Then: Pr ob( Inegrae) b V = (13) 1+ b V 1717

20 where b is a parameer. This formulaion corresponds o he assumpion ha he odds favoring inegraion (P/(1-P)) are proporional o V. Since V is no necessarily below one, i canno direcly serve as a probabiliy. The decision o specialize is no symmerical o he decision o verically inegrae, because is informaion base is significanly differen. I is modeled as depending on a comparison beween he meri of design of he componen produced inernally and he meri of he bes componen available on he marke. Specifically, he probabiliy of specializaion for each firm is deermined as follows: Le: C C max M M Z = max, 0 (14) C M where max M C is he higher componen Mod available on he marke. Then: Pr ob( Specialize) a Z = (15) 1+ a Z This says ha he odds favoring specializaion are proporional o Z. A specialized compuer firm may also decide o change is supplier, if a beer producer has emerged in he marke. The procedure for changing supplier follows he same rule for he specializaion process. Tha is o say, every n periods afer he las decision o specialize or he las change of supplier, a specialized firm checks if a beer supplier han he curren one exiss. If his is he case, a new supplier is chosen using he raing mechanism described in he discussion of he demand module. 5.7 Exi Boh compuer firms and componen suppliers exi he marke when heir marke share falls under a cerain minimum hreshold. Specifically, he exi rule is defined as follows. For each compuer firm and in each period, he variable E ( 1 e) Ei( 1) + e s = (16) is compued, where n is he number of firms acive in he marke a he beginning of he simulaion, s is he marke share of firm i a ime and 0<e<1 is a parameer. Then, if E i < E < 1/n, where E is a consan hreshold (equal o 0.05 in he curren parameerizaion), he firm exis. Thus, for example, if E = 1/(2n), he rule says ha firms exi if hey operae long enough a half or less of heir iniial marke share. A small value of e has he effec of allowing a lo of ime for firms o recover from emporary compeiive sebacks. (As is well known, equaion (16) generaes he E series as a geomeric disribued lag funcion of he s series, wih weighs decreasing ino he pas like e.) E is iniialized a 1/n. 1818