Spinoff Entry in High-tech Industries: Motives and Consequences

Transcription

1 Spinoff Enry in High-ech Indusries: Moives and Consequences Seven Klepper* Carnegie Mellon Universy Peer Thompson** Florida Inernaional Universy February 005 Various heories have been advanced for why employees leave incumben firms o found firms in he same indusry, which we call spinoffs. We review he accumulaing evidence abou spinoffs in various high-ech indusries, highlighing he cenral role ofen played by disagreemens. Because exising heories have ignored hem, we develop he foundaions of a model of spinoff formaion driven by disagreemens. Doing so proves o be raher challenging, because disagreemens are no possible among raional acors ha alk o each oher. We inroduce a minimal degree of non-raionaly, based on he concep of solipsism, and ask wheher such a concep is capable of generaing predicions consisen wh he empirical leraure. * Dep. of Social & Decision Sciences, Carnegie Mellon Universy, Psburgh, PA 1513, sk3f@andrew.cmu.edu. Klepper graefully acknowledges suppor from he Economics Program of he Naional Science Foundaion, Gran No. SES ** Deparmen of Economics, Florida Inernaional Universy, Miami, FL 33199, peer.hompson@fiu.edu.

2 I. Inroducion In recen years ineres has grown in he phenomenon of enrepreneurship. One does no have o look beyond Silicon Valley o see he imporance of new enerprises, which seemingly have played a key role in he region s valy. Bu where do new enerprises come from? Surprisingly, lle is known abou he origin of enrans, especially new enerprises. This is perhaps a legacy of he way enry is ypically modeled in heories of compeion. I has always been assumed ha if enry is profable, will occur. I is no a all clear, hough, wheher such confidence is jusified (Geroski [1995]). Recen work suggess ha enrans are que diverse a birh, and heir pre-enry experience persisenly affecs heir performance (Carroll e al. [1996], Geroski, Maa, and Porugal [00], Klepper [00a, 00b], Klepper and Simons [000], Thompson [004]). One class of enrans ha perform disincly well in some indusries are firms founded by employees of incumben firms in he same indusry (Klepper [00b], Agarwal e al. [004], Walsh, Kirchhoff and Boylan [1996]). We shall call hese firms spinoffs. While in some insances spinoffs are sponsored or linked o heir paren firm, generally he founders of spinoffs do no mainain any link o heir prior employers. In some indusries, spinoffs are legion. Indeed, in he semiconducor indusry so many spinoffs can be raced back o one firm alone, Fairchild Semiconducor, hey have been dubbed Fairchildren. Opinions differ grealy abou he conribuion of spinoffs o innovaion and economic growh. Some perceive spinoffs as parases, feeding off he innovaive effors of heir unwing parens. 1 Scholars who inerpre spinoffs as parases fear ha spinoffs can undermine he abily of heir parens o appropriae he reurns of heir innovaive effors, hereby undermining he incenives of incumbens o innovae. Ohers see spinoffs as he fon of innovaion, compensaing for he ineria ha plagues many incumbens. To hem, he Fairchildren jumped a sinking ship and led he semiconducor indusry o new glory, fueling he juggernau known as Silicon Valley. Where does he ruh abou spinoffs lie? The answer presumably lies in a beer undersanding of he moives of spinoffs in innovaive, high-ech indusries and he process governing heir formaion. Why, in fac, do employees of high-ech firms leave o found firms in he same indusry? Is mainly o explo innovaions hey worked on for heir employers? Is mainly because of he inabily of heir employers o perceive and/or ac upon promising 1 Ironically, his is he view of Inel, Fairchild s mos famous spinoff. Inel goes o grea lenghs o harass employees ha leave o sar heir own firms (Jackson [1998, pp ]). 1

3 echnological developmens in heir indusry? Alernaively, migh spinoffs be a naural oucome of a world in which people have differen percepions abou he bes pahs for organizaions o follow? The main purpose of his paper is o explore he moivaions behind spinoffs and he role hey play in economic growh. In Secion II, we review empirical sudies of spinoffs, many of which are que recen, and exrac a se of common paerns from he sudies. One paricularly prominen heme of he empirical leraure is ha spinoffs arise ou of disagreemens whin exising firms ha lead frusraed employees o pursue heir ideas in heir own firms. The exising heoreical leraure, reviewed in Secion III, has ignored he role of disagreemens. In Secion IV, we herefore develop he foundaions of a model of spinoff formaion driven by disagreemens. Doing so proves o be raher challenging, because Aumann [1976] all bu precludes disagreemens among raional acors ha alk o each oher. We inroduce a minimal degree of non-raionaly, based on he concep of solipsism, and ask wheher such a concep is capable of generaing predicions consisen wh he empirical leraure. We show ha he concep is indeed able o accoun for a number of disincive empirical regularies concerning spinoffs. However, some predicions of he model are a odds wh he daa. In Secion V, we herefore conclude wh a discussion of new direcions for developmen of our model. II. Empirical Regularies Concerning Spinoffs New sudies of high-ech spinoffs in he auomobile (Klepper [003b, 004], laser (Klepper and Sleeper [004]), and disk drive (Franco and Filson [000], Agarwal e al. [004]) indusries have added grealy o our knowledge abou spinoffs. Using unique sources o idenify all indusry enrans and heir characerisics, including heir pre-enry backgrounds, hese sudies analyze he facors influencing he rae a which firms spawn spinoffs and he performance of he spinoffs. Anoher high-ech indusry where spinoffs were prominen is semiconducors, and Brain and Freeman [1986] sudy he facors influencing he rae of spinoffs from semiconducor firms in Silicon Valley. The only oher high-ech indusry where spinoffs have been considered is bioech. Suar and Sorenson [003] explo daa on he locaion of all bioech sarups and on bioech firms ha were acquired or engaged in IPOs o make inferences abou he impeus for spinoffs whou having o race he herage of he bioech enrans. Mon [1990] also sudies he origin of bioech sarups in San Diego. Ouside of he high-ech secor, Phillips [00] sudies spinoffs from Silicon Valley law firms. Cuing across indusries, Gompers e al. [003] use daa on venure capal (VC) financed sarups o analyze he rae a which publicly raded firms spawned VC-financed spinoffs. We review he main findings from hese sudies.

4 We consider firs he auomobile indusry, which began in Through 1966 here were 75 enrans ino he indusry, nearly all of which enered before 196. Spinoffs accouned for 0% of he enrans, wh he percenage of spinoff enrans rising over ime. Spinoffs performed comparably o enrans wh pre-enry experience in indusries relaed o auos and subsanially beer han he majory of enrans whou any relevan pre-enry experience, and by he 1910s spinoffs produced a majory of he leading makes of auomobiles. The op four firms in erms of he number of spinoffs spawned were he four early leaders of he indusry, Olds Moor Works, Cadillac, Ford, and Buick (which was he cornersone of General Moors when was formed). Among all firms, he spinoff rae was greaer in firms ha produced leading makes of auomobiles and ha survived longer. The firm spinoff rae also increased wh age hrough age 14 and hen declined, was greaer in firms ha were acquired by eher auo or nonauo firms (in a shor window around he ime of he acquision), and was greaer in firms locaed in he Dero area, where he indusry was heavily agglomeraed (Klepper [004]). The performance of he spinoffs in erms of heir longevy was posively relaed o he performance of heir parens, wh 11 of he 13 spinoffs ha produced leading makes of auomobiles descended from Olds, Cadillac, Ford, and Buick/GM (Klepper [004]). In a deailed sudy of hese 13 spinoffs, Klepper [003b] found ha nine were formed by op level engineers and managers as he resul of disagreemens whin he paren firm abou he kinds of cars o produce or abou he managemen of he firm, wh he spinoffs someimes coninuing effors heir paren iniaed bu hen abandoned. The 13 spinoffs played an imporan role in he echnological advance of he indusry, accouning for a majory of he 50 major innovaions in he indusry from ha were no inroduced by he wo leading firms, Ford and Buick/General Moors The laser indusry began in 1961, and hrough 1994 spinoffs accouned for 69 or 17% of he 465 enrans whose backgrounds could be raced. The spinoffs survived much longer han oher sarups and comparably o diversifying enrans wh prior experience in indusrial elecronics, who were he longes-lived diversifying enrans. Dividing lasers ino eigh main caegories (and a residual), Klepper and Sleeper [004] noe ha spinoffs ypically specialized inially in a ype of laser produced by heir paren firm. Firms ended o remain specialized and produce a narrow range of laser ypes, wh spinoffs accouning for many of he leading producers of each ype of laser as well as he op wo firms in he indusry, Specra Physics and Coheren. Similar o auos, for each ype of laser he rae a which firms spawned spinoffs rose o age 14 and hen declined, was greaer in firms acquired by laser or nonlaser firms (in a shor window around he ime of he acquision), and was greaer in Silicon Valley firms, where he indusry was modesly agglomeraed. Using repors from a monhly rade journal and supplemened by inerviews wh founders, Klepper and Sleeper [004] discuss he impeus for eigh spinoffs ha were illusraive 3

5 of he facors underlying spinoffs in each of he eigh main laser ypes. Each had a founding eam wh a leas one high level echnical manager and some also had founders wh high-level managerial backgrounds in markeing and operaions. In five of he eigh spinoffs, he founders lef o develop a echnology hey worked on in heir paren firm bu he paren chose no o develop; wh hree of he spinoffs licensing echnology from heir paren. Two of he oher hree spinoffs were formed afer he paren was acquired, in one insance o service cusomers he paren abandoned afer being moved and in he oher o compee direcly wh he paren. In he disk drive indusry, of he 153 enrans from 1977 o 1997, 6% were spinoffs. Five major archecural innovaions ha reduced he size of disk drives and opened up new markes servicing smaller compuers were inroduced in he period All five were pioneered by spinoffs, who displaced he indusry leaders and survived longer han enrans wh oher backgrounds (Agarwal e al. [004]). All spinoffs had a leas one founder wh a high level echnical background and someimes oher founders wh a high level markeing or producion background, similar o auos and lasers. The rae of spinoffs was greaer in firms wh beer disk drives and ha were quicker o produce he new drives. These firms in urn had spinoffs wh beer disk drives and ha were quicker o ener subsequen new markes, consisen wh beer firms having beer spinoffs. Older firms ha enered by 1976 had a lower rae of spinoffs, bu oherwise age did no affec he firm spinoff rae. Chrisensen s [1993] analysis of he slowness of incumbens o inroduce he smaller disk drives is revealing abou he impeus for he leading spinoffs. Based on over 60 inerviews wh execuives, Chrisensen [1993, pp ] found ha leading incumben firms conceived and developed prooypes of he smaller disk drives bu hen abandoned hem when heir cusomers showed lle ineres in hem. Engineers ha worked on he smaller drives hen lef in frusraion o sar heir own firms, which ended up pioneering he drives. Judging from King and Tucci s [00] analysis of enry ino new disk drive markes, hough, his was no a general endency. They found ha more experienced firms were acually more likely o ener new markes a every poin in ime, consisen wh heir finding ha he sales of enrans in he new markes increased wh heir prior experience. In Silicon Valley semiconducor firms sudied by Brain and Freeman [1986], firms ha produced a wider array of semiconducor devices had higher spinoff raes, similar o he findings for lasers. Firms ha were earlier enrans ino new produc groups also had higher spinoff raes, similar o disk drives. Boh findings are consisen wh beer firms having higher spinoff raes. The spinoff rae was greaer in firms whose growh had slowed and in firms ha were acquired by nonsemiconducor firms or ha hired a new CEO from ouside he semiconducor indusry. Suar and Sorenson [003] analyze he effec of acquisions and IPOs on he rae of formaion of bioech firms. They found higher sarup raes in regions near where bioech firms were 4

6 acquired or engaged in IPOs, which hey presumed was due o spinoffs. These effecs were presen only in saes wh greaer resricions on he enforcemen of noncompee covenans and he effec of acquisions on sarups was resriced o acquisions where he acquirer came from ouside of he bioech indusry. Mon [1990] documens how in San Diego bioech firms, conrol changes resuling from acquisions by nonbioech firms led o culural differences ha spurred op level managers o leave o sar heir own bioech spinoffs. Mon also found ha mos of he San Diego bioech spinoffs were formed o develop echnologies heir parens declined o pursue. The oldes parens of spinoffs in Mon s sudy were 10 years old, and hrough age 10 he rae a which hey spawned spinoffs increased wh age, consisen wh he findings for auos and lasers Phillips [00] analysis of spinoffs from Silicon Valley law firms focused on how spinoffs affeced he performance of heir parens, bu he also analyzed facors influencing he performance of spinoffs and briefly he facors influencing he rae a which firms spawned spinoffs. Firm spinoff raes increased hrough he age bracke 9-15 afer which hey declined. The lengh of survival was greaer for spinoffs whose founders had greaer saus in heir paren firm and less for spinoffs from failing firms. He also found ha firms had higher hazards afer spinoffs han comparable firms whou spinoffs. Gompers e al. [003] found ha publicly raded firms in Silicon Valley and Massachuses, boh hobeds of enrepreneurial sarups, and firms ha were hemselves VC-financed had higher raes of VC-financed spinoffs. The former resul is consisen wh he higher rae of auo spinoffs in Dero and laser spinoffs in Silicon Valley and may simply reflec he greaer ease of forming a founding eam and securing advice and financial suppor in regions wh a larger number of relaed sarups. The spinoffs of Silicon Valley and Massachuses firms and VCbacked firms were less likely o be engaged in echnologies relaed o heir parens han he spinoffs of oher firms. I was also found ha less diversified firms had a higher spinoff rae and ha slowed growh heighened firm spinoff raes, similar o he Silicon Valley semiconducor firms. Firm spinoff raes also declined wh age, which is consisen wh he findings for auos and lasers if publicly raded firms were generally over 14 years old. Cerain paerns consisenly emerge from he various indusries sudied. Around 0% of all enrans were spinoffs, and he spinoffs were disincly good performers. They generally had a leas one founder who was a high level echnical manager and someimes also had founders wh high level markeing and operaional experience. Beer firms had a higher spinoff rae and heir In conras o auos, lasers, and disk drives, however, 40% of he sarups were financed in par by heir parens. 5

7 spinoffs were beer performers. Disagreemens over wha echnologies o develop and someimes abou managemen pracices were he principal impeus for he leading spinoffs. Spinoffs were less likely o occur in older firms, bu appears ha inially he spinoff rae increased wh firm age. Acquisions induced an increase in he likelihood of spinoffs during a shor window around he ime of he acquision, especially when he acquirer was from anoher indusry. A new CEO from anoher indusry, an IPO, and slowed growh also appear o have increased he rae of firm spinoffs. III. Exising Theories of Spinoff Formaion How well do exising heories explain he common paerns in he various sudies of highech spinoffs? Wh he growing ineres in spinoffs, various models have been proposed o explain hem. 3 These models end o fall ino hree camps. In he firs, an employee makes a serendipous discovery of some economic value. This discovery is in principle more valuable o he incumben firm han would be o a sarup, bu informaion asymmeries of one form or anoher frequenly persuade he employee o implemen he discovery hrough his own sarup raher han reveal o his employer. 4 In he second ype of model, he discovery is common knowledge whin he firm bu is less valuable o he incumben han would be o a sar-up, because s implemenaion would cannibalize exising rens or because he firm has limed compeency o evaluae he idea, paricularly when he idea is angenial o he firm s main acivies. 5 In he hird ype of model, employees learn from heir employers abou how o profably compee in heir indusry, especially when heir employer is successful. They explo his knowledge by seing up heir own firm in he same indusry. 6 3 The radional explanaion for who becomes an enrepreneur is based on abily (Lucas [1978], Holmes and Schmz [1990]). While is obvious ha abily may be enhanced by working for an incumben in he indusry (Irigoyen [00]), his leraure does no explain why only a small fracion of he many employees wh he requise experience and enrepreneurial abily leave heir employers o found a new firm. 4 See Anon and Yao [1995], Wiggins [1995], Bankman and Gilson [1999], Gromb and Scharfsein [00], Hellman [00], and Amador and Landier [003]. Common hemes are (i) firms canno comm o a coningen conrac ha adequaely rewards he employee for a discovery and he subsequen employee effor needed o implemen, and (ii) non-coningen conracs ha are ex ane accepable o he firm will no always be sufficien o preven a deparure by he employee. 5 See Pakes and Nzan [1983], Tushman and Anderson [1986], Henderson and Clark [1990], Chrisensen [1993], Klepper and Sleeper [004], and Cassiman and Ueda [00]. 6 See Franco and Filson [000] and Agarwal e al. [004]. 6

8 None of hese heories seems o capure he process underlying mos spinoffs. The firs ype of heory predics ha employees ha found spinoffs will no reveal heir ideas o heir employers. However, he evidence suggess ha a leas among he leading spinoffs, he employer ofen knows precisely wha he employee s idea is bu does no wan o pursue. Indeed, a common heme of sudies ha focus on he moives of spinoffs is ha hey arise from frusraion by employees over rejecion of heir ideas by heir employers (Garvin [1983, p. 6], Lindholm [1994, p. 163]). 7 Moreover, is no clear how he firs ype of heory can explain why acquisions, IPOs, and slowed growh should heighen he chance of spinoffs or why age should have a non-monoonic effec on he firm spinoff rae. The second ype of model is consisen wh firms rejecing he ideas of heir employees. This could happen if esablished firms are unable o evaluae cerain ypes of ideas ha are no in heir core areas. Wh he success of firms no doub dependen on heir abily o evaluae ideas of heir employees, migh be expeced such spinoffs would be more likely in less successful firms. Ye he evidence srongly poins in he oppose direcion, wh more successful firms having higher spinoff raes. Alernaively, in he second ype of model he ideas migh be rejeced because hey would cannibalize he firm s sales and hence profs. Bu many of he ideas ha spinoffs from he leading auo and laser firms pursued were acually iniaed and firs worked on by heir parens, which is consisen wh cannibalizaion fears only if he firms could no anicipae where he ideas would lead. This is hard o rule ou from he limed evidence on spinoffs. Bu if his were imporan, one would imagine ha mos employees would be able o undersand why heir ideas were rejeced and would no be so frusraed wh heir employers. I is also no clear why fears of cannibalizaion would be heighened when a firm was acquired or would be relaed o he age of he firm, hus providing no explanaion for hese findings. The hird ype of model feauring learning is consisen wh one aspec of he findings abou spinoffs, namely he endency for beer firms o have higher spinoff raes and beer-performing spinoffs. Bu hese heories imply ha spinoffs should do similar hings o heir parens based on heir common knowledge. Ye he evidence indicaes ha he leading spinoffs commonly pursue ideas heir parens rejeced. So wha exacly are he spinoffs learning from heir parens, and why should disagreemens be he impeus for so many spinoffs? Moreover, if learning is he moive for spinoffs, why would spinoffs be more likely when firms are acquired? Acquisions conceivably could promoe learning. However, acquisions by firms in he same indusry migh 7 I is possible ha employees ge heir ideas rejeced because hey do no wan o fully reveal hem in order o proec hem from being copied by heir employers. However, he frusraion expressed by so many employees when heir ideas are rejeced suggess ha parial revelaion is no he problem. 7

9 be expeced o promoe he mos learning, ye spinoffs seem especially likely when firms are acquired by firms in oher indusries. Moreover, he increase in spinoff formaion associaed wh acquisions seems o be concenraed in oo shor a window around he ime of acquisions o be consisen wh learning heories. Thus, all hree ypes of explanaions for spinoffs come up shor in imporan ways. The fac ha many leading spinoffs arise ou of disagreemens whin heir paren firms is difficul for exising heories o accommodae. Exising heories also do no address why acquisions, paricularly by firms from anoher indusry, increase spinoff raes or why age should affec spinoff raes. In he nex secion we propose a new model of spinoffs based on disagreemens, and we explore s abily o accommodae he empirical evidence. IV. Disagreemens and Spinoffs: A New Model To model disagreemens, we need o confron he fac ha firms are no unary acors bu are composed of decision makers wh poenially differen views abou wha he firm should do. Accordingly, we assume ha a firm is composed of muliple individuals, each of which has some influence on he firm s decisions. A any given momen, opimal choices are no known, bu over ime hey are slowly learned as employees receive signals abou he rue environmen facing he firm. The signals are noisy and differ across members of he firm, leading o disagreemens abou wha he firm should do. The firm s choice is a weighed average of he choices favored by each of s decision makers, wh weighs based on he posions of individuals and perhaps heir ownership share of he firm. Employees leave o sar heir own firms when heir view of wha he firm should do differs sufficienly from he firm s choice abou how o proceed. The essence of he model is ha employees wan o be involved wh a firm ha does wha hey believe o be he righ hing. In our view, sufficien aenion has already been paid o conracing difficulies caused by employees being all oo willing o screw heir employers by no revealing discoveries hey were paid o make, and by employers being unable o comm o sae-coningen rewards ha would persuade employees o be hones. We assume in conras ha individuals a higher echelons of firms are concerned wh he value of he firm s acivies, raher han wh how hey can manipulae heir own payoff hrough decepion or omission. On he oher side of he same coin, we assume firms have no ineres in ripping off heir senior employees. 8 We assume here is considerable uncerainy abou wha he righ hing is, and his 8 This is no a radical approach. I simply applies o he higher echelons of he firm Akerlof s [198] widely-admired bu widely-ignored sociological characerizaion of he workplace, where he average worker works harder han is 8

10 generaes genuine disagreemens abou he choice of sraegy for a firm. These disagreemens need no be predicaed upon any paricular discovery, wheher privae or common knowledge. There is, however, an inellecual challenge in modeling disagreemens. Presumably each individual mus, in he language of Bayesian learning, reveal his poserior mean o his colleagues if he wans o paricipae in he decision-making process. Bu his revelaion also allows each individual o infer precisely his colleagues privae signals. Efficienly incorporaing hese signals ino his own beliefs, he revised poserior mean will hen be he same for everyone. Aumann [1976] has shown under very general condions ha if he poseriors of wo Bayesians wh common priors are common knowledge, heir poseriors mus be he same; Geanakoplos and Polemarcharkis [198] show furher ha if wo agens wh common priors exchange heir efficien poseriors back and forh hey will arrive a he common knowledge poserior; and McKelvey and Page [1986] have exended hese resuls o n individuals. These resuls leave wo ways in which disagreemens can persis. Firs, one can drop he common prior assumpion. Alhough has a subsanial radion in behavioral finance, beginning wh Harrison and Kreps [1978], here has been some debae abou wheher doing so is reasonable (Aumann [1987, 1988], Gul [1988], Morris [1995]). 9 The consensus seems o be agains dropping he common prior. Moreover, in our seing seems reasonable o assume ha firms are formed by individuals ha wan o work ogeher because hey hold similar beliefs. Second, one can drop he efficiency of individuals updaing algorhms. Some auhors have assumed ha individuals are overconfiden in he sense ha he poserior mean is a biased esimae of he rue mean. 10 Ohers have assumed ha decision-makers overweigh he informaion conen of heir privae signals relaive o publicly available informaion. This second approach has also been dubbed a form of overconfidence. Our approach is of he second ype. However, we propose a somewha differen nomenclaure. We reserve he erm overconfiden o refer o individuals who underesimae he noise of any signals, wheher heir own or hose inferred from heir colleagues. We erm asymmeric weighing of privae and non-privae signals (in favor of he former) as solipsism. necessary according o he firm s work rules, and in reurn for his donaion of goodwill and effor, he expecs a fair wage from he firm. 9 See Van den Seen [001] for an exensive review. 10 This is he cenral assumpion in Amador and Landier [003]. However, he assumpion is subsequenly jusified by appeal o issues of asymmeric informaion and moral hazard. See Malmendier and Tae [00, 003] for applicaions of he assumpion o corporae invesmen and acquisions. 9

11 The disincion beween overconfidence and solipsism has subsance: solipsism is a necessary condion for disagreemen; overconfidence simply magnifies he size of disagreemen. There is a large empirical leraure supporing he assumpion of overconfidence and, o a lesser exen, wha we have called solipsism. De Bond and Thaler [1995] have gone so far as o claim ha perhaps he mos robus finding in he psychology of judgmen is ha people are overconfiden. Evidence of overconfidence has been repored among diverse professions, including enrepreneurs (Cooper, Woo, and Dunkelberg [1988]) and managers (Russo and Schoemaker [199]), alhough enrepreneurs exhib much more overconfidence han managers (Busenz and Barney [1997]). Odean [1998] and Daniel, Hirshleifer, and Subrahmanyam [1998] ce many oher examples, and differen forms of overconfidence. Our assumpion ha individuals overweigh privae informaion relaive o public informaion has found suppor in he laboraory (Anderson and Hol [1996]) and among financial analyss (Chen and Jiang [003]). Their findings are consisen wh he broader noion ha people expec good hings (e.g. receiving accurae signals) o happen o hem more ofen han hey do o ohers (Weinsein [1980], Kunda [1987]). The Dynamics of Disagreemen Our formal model is an exension of he Jovanovic and Nyarko [1995] Bayesian learning model o eams of decision makers. Suppose a firm is operaed by n individuals, each of which has some degree of decision-making auhory or influence. Each individual is concerned wh maximizing he expeced value of he firm, raher han his own privae reurns. All individuals know ha firm value is given by v= ( x), where x is he acivy underaken by he firm and is a arge. No one knows he arge, bu a ime individual i=1,,..., n, believes is a draw from a normal disribuion wh mean and variance σ. Given his beliefs, i calculaes ha he opimal sraegy is x=, yielding an expeced payoff of v = σ. The acivy acually chosen i n by he firm is a compromise, x = = i= 1φ, of everyone's beliefs. The parameers φ i are ime- n invarian weighs aached o individual expecaions, wh 1 1 inerpreed as i's decision-making influence. Individual i's expeced value of he compromise decision is ( ) E[] v = E φ i= i =. The weigh φ i can be 10

12 (( ) ( )) = E + ( ) ( ) = E ( ) = σ, (1) where in he hird line he fac ha any Bayesian poserior is unbiased implies E ( ) = 0. From i's perspecive he firm will do worse he more he disagrees wh he firm s decision, and he may wan o do somehing abou. We focus aenion on i's possible deparure o form his own sarup. Doing so will be aracive if he cos is no oo high and if by so doing he can operae a firm using a sraegy closer o his own beliefs. Le he cos be k (we shall call his he enry cos), and assume ha i can form a new eam consising of individuals holding exacly he same subjecive beliefs as his. 11 The expeced value of he spinoff is hen prefers o srike ou on his own whenever z ( ) E [ w] = k σ, so i k. () φ For any se of weighs, { i} i 1 n =, he decision o leave depends only on he subjecive means, and no a all on he precision of hose beliefs. Undersanding he spinoff process is herefore reduced o he ask of undersanding how expeced values come o differ sufficienly o induce individuals o srike ou on heir own. This difference will, however, be relaed o he precision of beliefs. To see how subjecive means come o differ over ime, assume ha he firm is founded a ime 0 by a group of n individuals, all of whom share he same prior ha is a random draw from N σ (0, ). Once each period, hese individuals receive privae and noisy signals, s = +, where he are random draws from a normal disribuion wh zero mean and variance σ. Alhough all signals have variance σ, each individual believes his own signals o have variance γσ and his colleagues signals o have variance γβσ. Individuals are labeled overconfiden if γ<1, and solipsisic if β>1. In he lim as β individuals only respond o heir own privae signals. Individual i s poserior afer receiving privae signals is normal wh mean 11 The assumpion reflecs he noion ha firms are formed by groups of individuals wh a common prior. 11

13 and variance σ s = σ, (3) + γσ γσ σ σ =, (4) σ + γσ where 1 s τ s = 1 i τ = is he mean of i's privae signals o dae. As β and γ are common o all decision-makers, he common-knowledge beliefs arrived a afer repeaedly exchanging poseriors are he same as would be obained if each individual s privae signals were direcly observable o his colleagues. In period, herefore, individual i forms beliefs as hough he has observed privae signals and (n 1) signals from his colleagues. Sandard Bayesian formulae for normal conjugaes hen imply ha i s expecaion of he arge is σ, (5) + + = βs + sj ( β n 1) σ βγσ j i wh poserior variance βγσ σ σ =, (6) ( β + n 1) σ + βγσ The firm s decision is a weighed average of each eam member s subjecive mean: Hence, σ β φ φ + + n n = is + i sj ( β n 1) σ βγσ i= 1 i= 1 j i n σ = ( 1 + ( β 1) φi) s. (7) ( β + n 1) σ + βγσ = i 1 ( β 1) σ z s s + +. (8) n = φi ( β n 1) σ βγσ i= 1 If β=1, hen z 0. Tha is, whou solipsism, disagreemen is no possible. 1

14 The Hazard of Spinoff Formaion The empirical evidence shows ha he likelihood ha a firm spawns a spinoff inially increases wh age, bu declines wh age for older firms. In his subsecion, we show ha his age profile for spinoff formaion is prediced by he model. The mean signals s, i=1,,.., n, are normally disribued and independen across individuals, each wh (unknown) mean and (rue) variance σ / normal wh zero mean and variance where var( z ) = n i= 1 i ( H) 4 ( 1) 1 i β σ σ φ + (( β + n 1) σ + βγσ ). I hen follows ha z is, (9) H = φ is he Herfindahl index of decision-making auhories. 1 Hence, i prefers o srike ou on his own if 4 ( H) χ ( β 1) σ σ 1 φ + (( β + n 1) σ + βγσ ) i k, where χ is a chi-squared random variable wh one degree of freedom. Rearranging, ex is preferred by i if (( β + 1) σ ) + βγσ k n χ β σ σ φ + 4 ( 1) (1 i H). (10) Figure 1 illusraes inequaly (10). The curve AA races ou he righ hand side as a funcion of. I is u-shaped, wh a minimum a βγσ /(( β n 1) σ ) = +, and lims of + a =0 and as. The unbounded lims reflec he facs ha all individuals begin wh he same prior and ha hey will evenually learn he common rue parameer value if hey receive a sufficien number of signals. Thus, for small enough and large enough, is unlikely ha a draw from he χ (1) disribuion will be large enough o induce a spinoff. Noe also ha for β=1, he righ hand side of (10) is infine. Figure 1 also plos sequences of he lef hand side of (10) for wo individuals. The sequence for individual i firs exceeds AA a poin a, which is when he depars 1 As (1 ) (1 ) j i φ + H φ φ i i +, is clearly posive. j 13

15 A χ a A χ j 0 βγσ ( β + n 1) σ FIGURE 1. The dynamics of spinoff formaion. o form his own firm. The sequence for individual j never crosses AA, so his individual does no depar he firm. I is clear from Figure 1 ha he hazard of i forming a spinoff mus sar ou a zero, hen rise o some sricly posive value (because a χ random variable has unbounded suppor), and evenually decline back o zero as he firm ages. PROPOSITION 1. The hazard of spinoff formaion inially rises from zero, bu evenually asympoically declines o zero. Assuming β>1, mos parameer changes induce an unambiguously-signed change in he probabily ha a any given inequaly (10) is saisfied. The righ-hand side of (10) is sricly increasing in γ, k and φ i for all posive, and sricly decreasing in H and σ for all posive. In conras, changes in σ have an ambiguous effec on he righ hand side of (10), which is increasing [decreasing] in σ for < [ > ] σ / σ. Le p denoe he uncondional probabily ha a any given inequaly (10) is saisfied. Then he hazard of spinoff formaion in period is ( 1 i = 1 ) h = p p i, and has he following properies: PROPOSITION. For any =1,, 3,..., and β>1, he spinoff hazard is sricly decreasing in γ, k, and φ i, and sricly increasing in H and σ. The hazard is sricly decreasing in σ for σ / σ, and is ambiguously relaed o σ hereafer. 14

16 Inuively, spinoffs are more likely in indusries wh low enry coss and high uncerainy abou he appropriae acivy. Low enry coss jus make easier o leave. The greaer he uncerainy abou, he more aenion individuals pay o heir privae signals and he more likely hey are o disagree. Overconfidence maers only if β>1, in which case spinoffs are decreasing in γ. Condional on he concenraion of decision-making auhory, individuals wh less auhory are more likely o leave. Finally, condional on φ i, greaer concenraion of decisionmaking auhory induces higher spinoff formaion raes. Concenraion of decision-making auhory maers even if individual i has no auhory, because increased concenraion makes he compromise decision more erraic. Changes in σ have an ambiguous effec on spinoff formaion. 13 The Effec of Acquisions A robus resul from he empirical sudies is ha spinoffs are more likely o occur around he ime ha a firm is acquired, and his is paricularly he case if he acquirer comes from ouside he indusry. This phenomenon is a naural consequences of our heory. Acquisions change he disribuion of decision-making auhory. New individuals are brough from ouside he acquired firm and reorganizaions ake place inside. Consider a suaion in which an individual i is currenly no planning o leave a firm ha is hen acquired. If he reorienaion of he firm moves he firm s decision oward i s beliefs, he sill will no leave and here is no consequence for he spinoff hazard. Bu if i s decision-making influence is reduced, he firm s decision moves away from i s beliefs, possibly enough o induce immediae ex. PROPOSITION 3. Acquisions ha change he disribuion of decision-making auhory induce a shor-erm spike in he hazard of spinoff formaion. 13 The ambiguy arises from wo counervailing effecs of an increase in he precision of he signals. On he one hand, more precise signals induce greaer sensivy of he poserior mean o signals and increase he likelihood ha poseriors diverge. On he oher hand, more precise signals increase he rae of learning, so ha he poseriors of all individuals converge on he rue parameer value more quickly. The former effec dominaes for small, while he laer dominaes for larger values of. Thus, when learning is more difficul, in he sense ha σ is larger, spinoffs are likely o occur laer in a paren firm s life. 15

17 In some insance an acquision may eliminae a maverick CEO and preven impending deparures. Bu acquisions ha accomplish his mus be underaken before frusraed employees have lef, requiring some remarkable prescience on he par of he acquiring firm. Such cases are likely o be rare. Spinoff Qualy Assume ha a ime individual i forms a spinoff, which implies ha here is a sufficienly large disance beween and o jusify he enry cos. The value of he spinoff is hen w = ( ) k. Bu even ignoring he enry cos, urns ou ha spinoffs will on average be bad ideas, in he sense ha he expeced inial qualy of i s firm is lower han he curren qualy of s paren. The inuion is simply ha he mean of (n 1) poseriors is likely o be closer o he rue value han a single poserior ha diverges from his mean. More formally, recall ha k is he smalles disance beween i s poserior mean and he compromise decision necessary o induce i o form a spinoff. Then he expeced qualy of he k Ev, if spinoff, E[ w ] +, is less han he expeced qualy of he paren, [ ] E ( ) ( ) + k < 0. (11) I is easy o show ha inequaly (11) holds for k=0, because in his special case we have E ( ) ( ) + k k = 0 = E E 4 σσ = + + β σ βγσ i= 1 = (( + n 1) + ) ( β 1) ( H 1) σ σ 4 (( β + n 1) σ + βγσ ) n ( 1 ( β 1) φi ) ( β ( n 1) ) < 0, (1) 16

18 where he second line explos he fac ha E()=0. Proving he same for he general case appears o be infeasible, because we need o furher condion on he difference beween and. 14 In Appendix A we herefore evaluae he properies of (11) numerically. The resuls are summarized in Figure. In all cases, he expeced value of he spinoff is less han he expeced value of he paren, alhough for some parameer values he difference is small. PROPOSITION 4. For β>1, (a) The inial qualy of a spinoff is on average lower han he qualy of s paren. (b) The average gap beween he qualy of a paren and s spinoff a he ime he spinoff is formed is increasing in β and σ, decreasing in γ and k, and exhibs a u- shape as φ i, n,, and σ increase. As one migh anicipae, greaer degrees of solipsism and overconfidence lead o spinoffs of relaively poor qualy. Spinoffs also have lower qualy relaive o heir parens in indusries where he variance of he prior, σ, is large. As Jovanovic and Nyarko [1995] have shown, σ is posively relaed o he amoun here is o learn over he life of a firm or echnology, he amoun ha is acually learned over any inerval of ime, as well as o he amoun of inequaly in efficiency among a cohor of firms. Hence, ceeris paribus, in indusries wh greaer inequaly and more learning, spinoffs are also likely o have lower relaive qualy. Recall ha high enry barriers require large disagreemens o induce spinoffs, and large disagreemens also imply large differences in he sraegies chosen by paren and spinoff. One migh herefore expec he relaive qualy of spinoffs o be declining in he size of enry barriers. Somewha surprisingly, he oppose is rue. I urns ou ha he effec of high enry barriers on he minimum difference in beliefs necessary o induce a spinoff is more han offse by he requiremen ha deparing individuals mus also be more confiden abou heir beliefs when enry barriers are high. Thus, he possibily of highly misleading signals is lower he higher are enry barriers. 14 The erm consiss of sums of producs of posively-correlaed normal variables. Some analyical resuls and approximaions are known for he disribuion of such producs only in special cases (Craig [1936], Aroian [1947]), and modern work has resored o numerical and Mone Carlo mehods (e.g. Ware and Lad [003]). 17

19 β σ γ σ k n φ i FIGURE. Expeced qualy of spinoff minus expeced qualy of paren. Diamond symbol on each graph indicaes he baseline value used o generae all he oher figures. The forms of he curves were verified for parameer ranges wider han hose shown here. 18

20 Greaer solipsism, greaer overconfidence, greaer prior uncerainy abou he righ hing o do, and lower enry coss all raise he hazard of spinoff formaion and reduce qualy, suggesing ha we would expec o observe a negaive associaion beween raes of spinoff formaion and spinoff qualy. However, his associaion mus be empered by he presence of several non-monoonic comparaive saics resuls. Increases in i s decision-making auhory, in he size of he managerial eam, in he signal noise, and in he age of he paren all have nonmonoonic effecs on spinoff qualy, causing firs o fall and hen o rise as each parameer increases. Mos noably, because his variable is mos readily measured, he spinoffs wh he lowes relaive qualy are hose ha form when he paren is of inermediae age. Spinoffs from boh young and old parens are on average higher qualy. These comparaive saics are on he whole inuive, and hey are relaed o he non-monoonic relaionship beween hese parameers and inequaly in efficiency among firms explored by Jovanovic and Nyarko [1995] in heir simpler seing. The non-monoonic relaionship beween he relaive qualy of he spinoff and he number of decision makers in he paren firm represens he radeoff beween wo opposing forces. On he one hand, he paren firm s decision is a weighed sum of n poseriors and hence he expeced qualy of he decision improves as n increases. On he oher hand, he poserior of individual i is also a weighed sum of n poseriors, so ha he average qualy of any spinoff formed also increases in n. For small values of n, he former effec dominaes, so he relaive qualy of he spinoff declines as n rises. For larger values of n he laer effec dominaes. Of course, increasing n has a smaller effec on he qualy of he spinoff he greaer he degree of solipsism. Thus, one would expec ha increasing β no only increases he average gap beween he qualies of paren and spinoff, also pospones o higher values of n he poin a which he second effec dominaes. Figure 3, which repeas he plo for n from Figure wh differen values of β, confirms ha his is he case. For our baseline value of β=4, he expeced qualy difference is greaes a n =3. For β=5, is greaes a n=5, while for β=100 is greaes a n=8. In he lim as β, i s poserior does no benef from he signals of his colleagues, and in his case he difference in relaive qualy asympoically aains s greaes value as n. Our resul ha on average spinoffs will perform worse han heir parens is a disincive predicion of he model. Oher heories envision spinoffs as eher exploing especially valuable ideas developed whin heir paren firm, pursuing ideas heir paren rejeced for parochial reasons or because of bureaucraic ineria, or compeing on even erms wh heir paren based on knowledge gleaned from working for heir paren. All of hese heories sugges ha on average spinoffs will perform as well if no beer han heir paren firms. Our model cerainly leaves open he possibily ha some spinoffs will ouperform heir parens, bu predics on average ha 19

21 expced difference in firm values β=4 β=5 β= number of decision makers, n FIGURE 3. Expeced qualy of spinoff minus expeced qualy of paren. his will no occur. Noe ha if here are echnological spillovers, which seems inevable, spinoffs could urn ou o be socially producive even if on average hey perform worse han heir parens. This could help explain he observaion by Klepper [003b] ha spinoffs played an imporan role in advancing he echnology of he U.S. auomobile indusry. We can explo he daa colleced in Klepper [004] on all auomobile firms o compare he performance of spinoffs and heir parens. For each firm, we can compue he number of years produced auomobiles, which is a kind of all-purpose measure of firm performance. By definion, a firm has o produce auomobiles for some amoun of ime in order for o be a paren, guaraneeing a minimal number of years of survival for parens. On he oher hand, spinoffs could fail a any poin. Consequenly, migh be expeced ha a greaer percenage of spinoffs would fail a young ages han heir parens, bu if spinoffs performed comparably o heir parens hen a older ages spinoffs and parens would have comparable survival raes. To es his, we consruc Kaplan Meier survival curves for auomobile spinoffs and heir parens, which are repored in Figure For each age, he curves indicae he naural log of he percenage of spinoffs and parens surviving o ha age (he slope of he curves a each age reflecs he hazard of ex). As expeced, spinoffs had lower survival raes han heir parens a young ages. Consisen wh our model, hese lower survival raes persised a older ages. For example, a age 0 he percen of survivors was 9% for spinoffs and 6% for heir parens, and a age 40 he spinoff survival rae was 3% versus 11% for heir parens. No doub here were many 15 There were 145 spinoffs and 88 firms ha accouned for he 145 spinoffs. The bes firms spawned more spinoffs (Klepper [004]), bu we included only one observaion for each paren so ha our es would be conservaive. 0

22 ln % parens surviving ln % spinoffs surviving FIGURE 4: Survival curves for spinoffs and parens in he auomobile indusry.. reasons parens survived longer han heir spinoffs, bu he resuls are encouraging regarding he model. Paren Qualy and he Spinoff Hazard A robus regulary idenified in he empirical leraure is ha beer-qualy parens produce more spinoffs. I urns ou ha our absrac model predics he oppose. A formal proof is 1 difficul. We resric aenion o he special case in which φ = (1 φ )(1 ) for all j i, and j i n provide a numerical evaluaion. Figure 5 provides represenaive plos of he hazard ha individual i forms a spinoff for given values of he unknown parameer,, and he firm s decision,. 16 Panel A plos he case in which i has less decision-making influence han he remainder of he eam, while panel B plos he case in which he has more influence. Boh graphs are symmeric around an axis where he decision,, equals he unknown arge,, and hey have a minimum along ha axis. Tha is, he probabily ha individual i chooses o leave he paren firm is lowes for high-qualy firms. 17 PROPOSITION 5. The hazard of spinoff formaion is decreasing in firm qualy. 16 Appendix B provides he formal derivaions. Afer suable scaling of he axes, differen ses of parameer values produce very similar graphs Because his resul holds for φ i > n and φ i < n, is noeworhy ha parenal qualy does no maer in he 1 symmeric case, where φ i = n (see appendix ). 1

23 PANEL A: φ i <1/n. Pr { z k } Increasing values of _ Increasing values of = PANEL B: φ i >1/n. Pr { z k } FIGURE 5. Represenaive plos of probabily ha individual i chooses o form a spinoff, as a funcion of he unknown arge value,, and he firm s decision,.

24 Facors oher han disagreemens appear o be he driving force behind he empirical finding ha beer-qualy firms spawn more spinoffs. There is, however, a very simple exension o he model ha can overurn proposion 5. Assume ha firm value is given by v= ζ ( x), where ζ is a parameer varying from firm o firm ha measures he qualy of he decision-making group, and assume ha spinoffs inher heir paren s ζ. Then, individual i will choose o form a spinoff he firs insance ha ( ) / ζ. Tha is, an increase inζ acs jus like a reducion in k he enry-cos, which is posively-relaed o he spinoff hazard. Consequenly, if ζ varies sufficienly across firms, he model would predic a posive relaionship beween firm qualy and he spinoff hazard. V. Discussion Our model of spinoffs is based on he idea ha disagreemens naurally arise in he course of doing business, and under cerain circumsances hey will lead o spinoffs. The model is preliminary and raher absrac. Noneheless, already generaes some disincive predicions ha resonae wh prior findings, suggesing ha his is a promising line of inquiry. Firs, he model predics ha he hazard rae of spinoffs inially rises wh he age of he prospecive paren firm, bu evenually declines o zero. This accords wh findings for auos (Klepper [004]) and lasers (Klepper and Sleeper [003]). Second, he model predics ha spinoffs are more likely in indusries where here is considerable uncerainy abou he arge, which seems consisen wh he prevalence of spinoffs in high-ech and younger indusries (Garvin [1983]). Third, he model predics ha spinoffs have on average lower qualy han heir parens and his accords wh some preliminary evidence for auos repored in his paper. The model also predics ha spinoffs are more likely when here is a srongly hierarchical srucure in decision-making, as measured by he Herfindahl index of concenraion of decisionmaking auhory, and ha individuals wh less decision-making influence will be more likely o sar spinoffs. These las wo resuls seem o be a he hear of why acquisions increase he chance of spinoffs. They shif conrol of decision making o he acquirer. If he acquirer has a disincive culure hen managers of he incumben firm will have lle influence on he decisions of he acquirer, raising he prospec of a culure clash in which employees leave o pursue heir ideas in heir own firms. This seems consisen wh Mon s [1990] observaions abou how acquisions of San Diego bioech firms by nonbioech firms led o spinoffs. Our model needs o be developed furher o accommodae he full range of common paerns ha we deailed abou spinoffs. The model does no allow for heerogeney across firms, hence 3

25 canno address why beer firms have more and beer spinoffs. In fac, our model predics ha beer firms have fewer spinoffs, while he relaionship beween paren and firm qualy is ambiguous. We suspec he explanaion lies in spinoffs inhering an aspec of qualy form heir parens. We are undecided abou wheher his is because beer firms are able o hire more able eam members, some of whom subsequenly form spinoffs, or wheher employees of beer firms are able o learn more. Perhaps boh facors are in play, bu a leas one such mechanism needs o be buil ino he model. We also need o draw ou and es oher disincive implicaions of our model. One promising direcion concerns he effec of spinoffs on indusry performance. Mos models of spinoffs sugges ha spinoffs will harm heir parens, which accords wh Phillips [00] findings for Silicon Valley law firms. Our heory suggess a more benign view of spinoffs. Alhough on average hey do worse han heir parens, some spinoffs will inevably do much beer. Spinoffs herefore provide parens, and que possibly he indusry as a whole, he opporuny o observe he oucome of decisions ha incumbens had chosen no o make. When such spillovers exis, all firms in he indusry may learn much more quickly when spinoffs are prevalen. We suspec ha his benign role of spinoffs may be quanaively impor (his seems o be he case for auos, where spinoffs played a major role in he developmen and diffusion of new echnology), bu his mechanism is absen from he model. While here is much o do, our findings o dae are encouraging abou he model. We inend o srucure he model furher o accommodae he full range of findings abou spinoffs and hen use daa from muliple indusries o es s disincive implicaions. If we wan o sor ou he welfare implicaions of spinoffs and appropriae public policies o deal wh hem, we will need o push all heories, including ours, much harder. Given how well spinoffs performed in he highech indusries where hey have been sudied, he agenda seems well worh engaging. 4