Working Paper On the transaction costs determinants of vertical integration

Transcription

1 econstor Der Open-Access-Publikationsserver der ZBW Leibniz-Informationszentrum Wirtschaft The Open Access Publication Server of the ZBW Leibniz Information Centre for Economics Whinston, Michael D. Working Paper On the transaction costs determinants of vertical integration CSIO working paper, No Provided in Cooperation with: Department of Economics - Center for the Study of Industrial Organization (CSIO), Northwestern University Suggested Citation: Whinston, Michael D. (2000) : On the transaction costs determinants of vertical integration, CSIO working paper, No This Version is available at: Nutzungsbedingungen: Die ZBW räumt Ihnen als Nutzerin/Nutzer das unentgeltliche, räumlich unbeschränkte und zeitlich auf die Dauer des Schutzrechts beschränkte einfache Recht ein, das ausgewählte Werk im Rahmen der unter nachzulesenden vollständigen Nutzungsbedingungen zu vervielfältigen, mit denen die Nutzerin/der Nutzer sich durch die erste Nutzung einverstanden erklärt. Terms of use: The ZBW grants you, the user, the non-exclusive right to use the selected work free of charge, territorially unrestricted and within the time limit of the term of the property rights according to the terms specified at By the first use of the selected work the user agrees and declares to comply with these terms of use. zbw Leibniz-Informationszentrum Wirtschaft Leibniz Information Centre for Economics

2 THE CENTER FOR THE STUDY OF INDUSTRIAL ORGANIZATION AT NORTHWESTERN UNIVERSITY Working Paper #0001 On the Transaction Costs Determinants of Vertical Integration By Michael D. Whinston * Department of Economics, Northwestern University, and NBER ( mwhinston@nwu.edu) First Draft: November, 1997 This Draft: March, 2000 * I thank participants in the November 1997 "Make Versus Buy'' Conference at Columbia Law School and Ilya Segal for their comments on a preliminary draft of this paper. I also gratefully acknowledge the financial support and hospitality of the George J. Stigler Center for the Study of the Economy and the State at the University of Chicago. Visit the CSIO website at: us at: csio@northwestern.edu.

3 1.Introduction The extensive development ofthe transaction cost approach to the organizationof rms(and industries) has been, withoutquestion, among the mostsigni cant advances inindustrial organizationover the last twenty years. This development began with path-breaking work by Williamson[1975,1979,1985],Klein,Crawford,andAlchian[1978],andothersthatpushed forwardthe agendaof explaining rms'boundariesthatwas rstlaidoutbycoase[1937] forty yearsearlier. Theirworkfocusedattentiononthewaysinwhichexpostquasi-rents couldcreatehazardsforlong-termcontractualrelationswhencontractsareincomplete,and the e ectsthatthiscouldhave on rms'choicesofwhetherto integratedistinctstagesin the vertical chain of production and distribution. By doing so, it identi ed a pervasive{and potentially measureable{featureof long-term contracting settings as a critical determinant of the degree to which rms would choose to integrate activities. Just asone would hope, thisconceptual breakthroughwas soonfollowedby empirical workaimedattesting the theory. Veryquicklyitbecameclearthatthetheoryhadsigni cantpredictive power. Indeed,asearlyas1987,Joskow[1988]couldwritethat\Thiswork generally provides strong empirical support for the importance of transactions cost considerations, especially the importance of asset speci city in explaining vertical relationships." Developmentssincehavenotchangedthisconclusion. 1 Atthesame timethatempiricalworkwasprovidingthiscon rmation,acloselyrelated andmore formal theory of vertical integration emerged, beginning with Grossman and Hart's [1986] seminal paper. Like what I shall call the\williamsonian" transactioncosts theory, this \property rights theory" (see Hart [1995]) took the incompleteness of contracts and developmentofexpostquasi-rentsascriticaltounderstandingverticalintegration.itdi ered from the Williamsoniantheory intwo essential respects. First, the property rights theory focused in a very explicit way on the manner in whichintegration could a ect the level ofnon-contractible ex anteinvestments undertakenby contractingparties. Second, in contrast to the Williamsonian theory, which assumed that contracting hazards could be completely avoided by bringing the transaction within the rm (more onthis later), the property rightstheoryassumed thatopportunismwaspresentinallorganizationalmodes, so that theintegrationdecision involveda comparison ofthenatureande±ciency costs of 1 Masten [1996] provides a nice compilation of some of the best work on transaction costs studies of vertical integration and vertical contracting more generally. Shelanski and Klein [1995] survey the area.

4 opportunistic behavior in di ering organizational forms. Because the two theoriesboth focusoncontractualincompletenessandexpost quasirents, it is often presumed that the empirical literature on transaction cost determinants of vertical integration provides support for both. I will arguehere, however, that this is true in only a relatively weak sense. The two theories do share the prediction that integration decisions matter in the presence ofincomplete contracts andex post quasi-rents; and the empirical literature does show that the presence of quasi-rents matters for integration decisions. But this alone is not a very demanding test. Infact, theproperty rights and Williamsonian theories provide quite distinct predictions about the integration decisions of rms. While the Williamsonian theory predicts that any increase in quasi-rents will increase the likelihoodof vertical integration(a nding that is sofar consistent withnearly all of the existing empirical literature), the property rights theory o ers muchmore re ned predictions aboutthe typesofspeci citythatwillmatterforintegrationdecisions,andthedirectionin which various types of speci city will move the likelihood of integration. As a result, I will argue,testsof the propertyrightstheory(and,atthesametime,moredemanding testsof the Williamsonian theory) await further empirical work. Therestof thispaperisorganizedasfollows. InSection2,Ilayoutindetailhow the property rights and Williamsoniantheories di er. For this purpose, I focus on a simple linear-quadratic property rights model and examine the comparative statics predictions that emerge from it. (I also note there some di±culties with using existing property rights theory to explainintegrationpatterns.) InSection3, Idiscuss the ndings of three ofthemost well-known empirical papers on the transaction costs determinants of vertical integration inlightof these observations. My particular focusis on what these papers might tell us abouttheextenttowhichthepropertyrightstheory'spredictionshold. Thegenerallesson tobelearnedfromthisexerciseisthatthesepaperssimplydonottellusenoughaboutthe economic environmentsunder studyto judgetheproperty rightstheory. As aresult,itis relatively easy to construct a variety of reasonable property rights models of these settings that either do or don't accord withthe observed relationships in the data. The implication of this fact is that the existingempirical literature provides little guidance regarding which of the twotheories better explains patterns of integration. Section4 o ers concluding remarks. 2

5 2.TheWilliamsonianTheoryandtheProperty Rights Theory What I shall call the \Williamsonian" theory takes as its starting point two observations: First, that in many exchange relationships it is di±cult to specify fully the contracting parties' obligations and, second, that in many cases the parties become \locked in" to oneanother tosome extent over the course oftheir relationship(i.e. theparties develop \relationship-speci c assets"). Using Klein, Crawford, and Alchian [1978]'s terminology, there are then ex post quasi-rents, because the value of trade withinthe relationshipcomes toexceedthevalueofoutsidetradingopportunities. 2 Whencontractsareincompleteandex post quasi-rents are present, the contracting parties face the hazard of ex post opportunistic behavior: each party to the contract may engageinine±cient behavior inanattempt to \hold-up" the other partyandobtainalargershareoftheavailablequasi-rents. 3 This reductioninthee±ciency ofthetradingrelationship, iflargeenough, may motivate one of thecontracting parties tobring the transaction\inhouse"{ that is, to vertically integrate {toavoidthesehazards. Acentralpredictionofthetheoryisthenthatwhencontractsare incomplete,greaterlevelsofquasi-rentsincreasethelikelihoodofverticalintegration. 4 The property rights theory, developedingrossmanandhart [1986], Hart andmoore [1990] andelsewhere(see Hart[1995] for an overview), starts from the same basic premises, butdi ersfromthewilliamsoniantheoryinitsgreaterlevelofformalization,initsfocuson the e ect of integration on the e±ciency of ex ante choices of non-contractible investments, andinitsviewthatthesametypesofopportunisticbehaviormaycreatee±ciencylosses{ sometimes severe ones{bothwithin integrated rms andacross rm boundaries. Although the \speci city" of investments is important inthis theory, thetypes ofspeci city that are important, andthe predicteddirectionoftheir e ects, candi er substantially from the Williamsoniantheory. To illustrate these points, inthe next subsectioni examine the 2 Williamson cites four kinds of speci city as giving rise to such conditions: site speci city (the parties' have committed immobile physical assets to a \cheek-by-jowl" relationship), physical asset speci city (the parties must make investments in specialized physical assets), human asset speci city (the parties develop speci c human capital), and dedicated assets (one or both parties build productive capacity for which there is insu±cient demand absent their trading relationship). 3 As Masten [1984] puts it: \Idiosyncratic assets, because of their specialized and durable nature, imply that parties to a transaction face only imperfect exchange alternatives for an extended period. The more specialized those assets, the larger will be the quasi-rents at stake over that period, and hence the greater the incentive for agents to attempt to in uence the terms of trade through bargaining or other rent-seeking activities once the investments are in place." 4 Joskow [1988] puts this prediction succinctly as follows: \Other things equal, we expect the parties to more frequently choose vertical integration or a long-term contract as the quasi-rents associated with speci c investments become more important and the associated bene ts of precommitment increase." 3

6 determinants of integration in the context of a simple linear-quadratic property rights model A Simple PropertyRights Model Consider a simple bilateral trade setting. There are two agents who may wish to trade tomorrow,abuyerbandasuppliers: ThesellerS usesanassetfortheproductionofhis product. In whatfollows, I focus ontheincentivesfor buyerintegration; thatis forb to owntheupstreamasset. Idenote bya B =1thecaseinwhichB ownstheupstreamasset (verticalintegration),andbya B =0thecaseinwhichSownsthisasset(non-integration). Since i have in mind a situation in which there are likely to be downstream assets as well, there is also in principle a possibility of seller integration in which S owns both the downstream and the upstream assets. Implicitly, in what follows I am assuming that any observed integration is buyer integration, inwhich B controls both the upstream and downstreamassets. Inlarge part,idothisbecausetheempiricalstudiesidiscusslaterin thepaperfocusonbuyers'decisionsofwhethertoprocureinternally(e.g. GM'sdecisionof whether toownits ownengineplant). Intwo ofthese three studies, this may be justi ed bythefactthatthebuyersaremakingthesedecisionsovertheprocurementofmanyinputs sothatsupplierownershipofthebuyer'sassetsisnotgenerallyaviableoption. 5;6 At time 0, the two parties can decidewho will ownthe upstream asset (e.g. B may purchaseitfromsifsinitiallyownsit). Alsoatthattime,theymayagreeaboutthelevels ofsome contractible investments. For simplicity, we assumethat for therelationship tobe productive, there are certain given levels of these investments that must be made(so that thisisnota choice variable fortheparties,althoughtheappropriatelevelswillneedtobe speci ed contractually exante). As inhart [1995], no otheragreements about tradeare possible, because the good to be traded is not describable in advance. At time 1, each of the parties may make some non-contractible investments. We denote by i B 2 < + thebuyer'snon-contractibleinvestmentlevelandbyi S 2 < + theseller'snon- 5 Of course, this feature would ideally be part of a more complete model of the integration decision than what I present here. (For one such model, see Bolton and Whinston [1993].) 6 In situations in which one does not know on a priori grounds the form that any observed integration takes, an immediate issue in applying the property rights model empirically is whether this di erence in integrated structures is empirically distinguishable. If so, then one would want to allow for achoice between three possible ownership structures. If not, then the property rights model can be used instead to predict the probability that some integrated outcome is observed. (This also raises the theoretical question of the degree to which an integrated rm is able to design the allocation of control, or authority, within the rm; that is, whether \B ownership" and \S ownership" are really the only twopossible forms of authority within the rm.) 4

7 contractible investmentlevel. Theassociatedcostsarec B (i B )andc S (i S ): Attime 2, the buyerandthe sellerbargainovertrade. Denoteby¼(i B ;i S )thepro ts availablefromthee±cienttrade,andbyw B (i B ;i S ja B )thepayo tothebuyerinhisnextbestalternative to trading with S giventhatinvestmentlevelsi B andi S havebeentaken, and that the ownership variable is A B : When A B = 0, this involves procurement from another supplier (or some form of self-production) or shutting down; when A B = 1 this couldinvolveeitherprocurementfromanothersupplier,hiringanothermanagertocomein andproduce theinputusingtheupstreamassetownedbyb,orshuttingdown. Likewise, letw S (i B ;i S ja B )denotethepayo tothesellerinhisnext-bestalternativetotradingwith Bgiventhatinvestmentlevelsi B andi S havebeentaken,andthattheownershipvariableis A B :ThiswillinvolvesellingtoanotherbuyerwhenA B =0orshuttingdown;whenA B =1, thiscouldinvolve eithersellingto anotherbuyerusing atechnologythatdoesnotusethe asset, running another rm, orshutting down. I assumethattheparties engageinnash bargainingand(forsimplicity)thattheyhaveequalbargainingpower;hence,theysplitany available surplus(the quasi-rents) evenly. Note that in contrastto Hart andmoore[1990], I allowforanagent's investmentsto a ectnotonlyhisowntrading alternativesbutalso thoseoftheothercontracting party{ that is, we consider\cross-investments" as well as\self-investments". For example, a seller might invest in improving the quality of the product that can be produced using his asset. This would a ect his trading alternatives if he owned the asset, but would a ect the buyer's alternativesifinsteadthebuyerownedtheasset. 7 Thus,Itaketheessenceofthe\property rights" model in whatfollows to beits focus onthe e ect ofthe allocation ofownership on ex ante non-contractible investment choices. I do this because this generalization seems indispensible if the property rights theory is to be a serious contender in explaining patterns ofintegration. We make the following assumptions about these functions: ¼(i B ;i S ) = 0 + B i B + S i S w B (i B ;i S ja B ) = [ 0+ B0 i B + S0 i S ](1 A B )+[ 1+ B1 i B + S1 i S ]A B w S (i B ;i S ja B ) = [¾ 0 +¾ S0 i S +¾ B0 i B ](1 A B )+[¾ 1 +¾ S1 i S +¾ B1 i B ]A B c B (i B ) = 1 2 (i B) 2 7 As another example, S might invest in training B how to use its input more e ectively. 5

8 c S (i S ) = 1 2 (i S) 2 ; where( B ; S ) 0(sothatpositivelevelsofinvestmentaree±cient)and 0 >Maxf 0+ ¾ 0 ; 1+¾ 1 g, B Maxf B0 +¾ B0 ; B1 +¾ B1 g;and S f¾ S0 + S0 ;¾ S1 + S1 g(thesethree conditions implythatitis alwayse±cientforthetwo partiesto tradeex post;i.e., quasirentsare alwayspresent). Let,,and ¾ denotetherespectiveparametervectors. This linear-quadratic formulationhas theadvantagethat wecanidentify inavery simple way the marginal returns to investment. The assumptions adopted in Chapter 2 of Hart[1995] correspondtothecaseinwhich S0 = S1 =¾ B0 =¾ B1 =0;( B ; S ; B0 ; B1 ;¾ S0 ;¾ S1 ) 0; B > B1 > B0 ;and S >¾ S0 >¾ S The rst-best Consider, rst, the e±cient levels of investments. These solve: max¼(i B ;i S ) c B (i B ) c S (i S )= 0 + B i B + S i S 1 i B ;i S 2 (i B) (i S) 2 : (2.1) Thesolutiontothisproblemis(i B ;i S )=( B; S ),andtheresultingjointsurplusisw = ( B) ( S) 2 : Equilibrium investmentlevels Now consider the equilibrium levels of investment undertaken by the parties given the ownershipvariablea B :Thebuyerchoosesi B tosolve: or maxw B (i B ;i S ja B )+ 1 ib 2 [¼(i B;i S ) w B (i B ;i S ja B ) w S (i B ;i S ja B )] c B (i B ) 1 max i B 2 [¼(i B;i S )+w B (i B ;i S ja B ) w S (i B ;i S ja B )] c B (i B ): Thisgives rst-ordercondition 1 2 [@¼(i B;i S ) B(i B ;i S ja B S(i B ;i S ja B ) B(i B ) B Substituting using our assumed functional forms andsolving we get: i B (A B; ; ; ¾)= 1 2 [ B+( B0 ¾ B0 )(1 A B )+( B1 ¾ B1 )A B ]: (2.2) In a parallel fashion we can get the seller's equilibrium investment level as: i S (A B; ; ; ¾)= 1 2 [ S+(¾ S0 S0 )(1 A B )+(¾ S1 S1 )A B ]: (2.3) We can then denote by W(A B ; ; ; ¾) the equilibrium level of welfare given ownership structurea B andparameters( ; ; ¾)[obtainedbysubstituting(2.2)and(2.3)into(2.1)]. 6

9 Comparativestatics predictions ofthetheory Nowconsiderhowchangesinvariousparameterse ectthelikelihoodofverticalintegration. Toapplythemodeltoactualdata,wewanttoallowforthepresenceofother,unobserved, factors a eecting the integration decision. Here I do this by assuming that the actual joint surplus under ownership structure A B is W(A B ; ; ; ¾)+ " AB, where " AB is a random variable unobserved to the econometrician. Withthis assumption, buyer integration will be observedif W(1; ; ; ¾) W(0; ; ;¾) (" 1 " 0 ). Thus, forcomparativestaticsweare interested in how changes in the parameters a ect the di erence ( ; ; ¾) [W(1; ; ; ¾) W(0; ; ; ¾)]: Note, rst, that the value of ( 0 0 ¾ 0 ) isirrelevant: that is, changes inthelevel of quasi-rents that do not a ect themarginal returnsto investments haveno bearing on the likelihood of vertical integration. As one implication, then, we see that in contrast to thepredictions of thewilliamsonian theory, greater levels ofspeci city of contractible investments have no bearing on the integrationdecisionas long astheydo nota ect the marginalreturnsfromanynon-contractibleinvestments. Wenowstudyhowchangesinthemarginalreturnstonon-contractibleinvestmentsa ect the likelihood of observing vertical integration. Consider rst a change in the marginal productivityofb'sinvestmentwhenb andstrade, B. = [i (1; ) i B (0; )]+[ B i B (1; )]@i B (1; ) [ B i B (0; )]@i B (0; ) B = 1 2 [i B (1; ) i B (0; )] = 1 2 [( B1 ¾ B1) ( B0 ¾ B0 )]: Thus,ifBinvestsmoreunderintegration,thenanincreasein B {whichincreasesthejoint return from B's investment{ increases the probability of integration. Likewise, if S invests more undernonintegration(i.e. whenheownstheupstreamasset)thenanincreasein S, the joint return from S's investment, reduces the probability of = S 2 [i S (1; ) i S (0; )] = 1 2 [( S1 ¾ S1) ( S0 ¾ S0 )]: 7

10 Following similar derivations, we can determine the e ects of changes in the marginal returns to investments on B and S's disagreement payo = B0 2 [ B i B (0; = B1 2 [ B i B (1; = S0 2 [ S i S (0; = S1 2 [ S i S (1; )]: Table 1 summarizes thesee ects for the case inwhichthere is under-investment under eithervertical structure (so that j > i j (A B; ) forj =B;S). Inthiscaseanincreasein themarginale ectofi j onj'sdisagreementpayo underownershipstructurea B increases thelikelihoodofobservingthat ownershipstructure(denotedby a\+" infigure 1), while anincreaseini j 'se ecton j'sdisagreementpayo underownershipstructurea B reduces this likelihood(denoted by a\-"infigure 1). The reasonis straightforward: In a situation of under-investment, anincrease in investment levels under a particular ownershipstructure raises the surplus generated underthatstructure. Whenthemarginale ect ofi j on j'sdisagreementpayo increases under anownershipstructure, thatincreasesj'soptimal investment level under that structure; when the e ect of i j on j's disagreement payo increases,however,thiscausesjtodecreasehisinvestment,loweringthejointsurplusassociated withthat ownership structure. The results are reversed when we are ina situation of over-investment. One interesting aspect of these derivations relates to the impact of the\importance"of an agent's investment in determining the optimal ownership structure. In particular, we see thatincreasesin an agent'smarginalreturnstoinvestmenthaveambiguouse ects,evenif we restrict attention to the case of self-investments and under-investment, as in Hart[1995]. Forexample,itiseasyto verifythatifwehaveequal-sizedincreasesin B, B0,and B1, thenthere is noe ect on the probability ofobserving integration. Thus, by having B0 increase by slightly more or slightly less thanthe other two parameters, we can move the 8

11 probabilityofintegrationineitherdirection. 8;9 Analternative wayto thinkaboutthee ectsofchangesinmarginal returnsisinstead to consider changes in the speci city of marginal returns. Consider rst the case in which i B isaself-investment. Thenwe canthinkofthelevelofmarginal people speci city forb of hisinvestmenti B as being given by thedi erence( B B1 ), whichistheamount by whichthemarginalreturnforbofi B isreducedwhenb hascontroloftheupstreamassets butnolongerdealswiths: Incontrast,thelevelofmarginalassetspeci city forb ofi B is givenby( B1 B0 ),whichisthedegreetowhichb'sreturnsarefurtherreducedifhealso doesnothaveaccesstotheupstreamasset. Achangeinthelevelofmarginalpeoplespeci cityofi B forb holdingthelevelofasset speci city xedcanthereforebecapturedbyequal-sizedreductionsin B1 and B0 andhas adi erentiale ecton equalto 1 2 [i B (1; ) i B (0; )].10 ThisispositiveaslongasB invests more when he owns the upstream asset. Note that increases in marginal people speci city thereforematteronlyifthereissomemarginalassetspeci citypresent{i.e.,if B1 > B0 : To understand why the degree of marginal people speci city matters for the integration decision consider rst the case in which B under-invests regardless of the ownership structure(as in Hart[1995]). Thenthegreateristhedegreeofmarginalpeoplespeci cityofb'sinvestments, the more distortion there is in B's investment decision, and the greater is the value of the increase initslevelthatwouldbe causedbyintegration; hence, thegreater thelikelihood that we would observe integration. When instead we have over-investment by B, greater marginal people speci city reduces the distortioninb's investment level, andso reduces the cost of the increase in B's investment that would accompany integration. Achange inthe levelof marginalassetspeci cityofi B forbiscapturedbya decrease in B0. Thisleadstoadi erentialchangein equalto 1 2 [ B i B (0; )],whichispositive whenweareinacaseofunder-investment. 8 Readers familiar with Hart's [1995] result concerning relatively unproductive investments [Proposition 2(B), p. 45] may be puzzled by this point, since Hart's result asserts that as a party's investments become relatively unproductive that party should not own assets. The cause of the discrepancy is that when Hart makes an agent's investments relatively unproductive he actually increases the marginal returns of investments on disagreement payo s: in particular, because he makes the marginal returns on investments approach the marginal cost of investment, and because at the equilibrium investment level a weighted average of marginal returns under e±cient trade and the disagreement points equals marginal cost, the lower marginal returns on disagreement payo s must actually be increased by this change. Arguably, it is more natural to think of a more important investment as involving an increase in all marginal returns. 9 In contrast, when i B is a cross-investment, an increase in its importance that involves equal-sized changes in B, ¾ B0, ¾ B1 does increase the probability of vertical integration. 10 Alternatively, we can capture this change in marginal people speci city with an increase in B. 9

12 Wheni S isacross-investment,b'sdisagreementpayo willalsobea ectedbys'sinvestmentchoice. Wecanthende ne thelevelofmarginalpeoplespeci cityofi S forbbythe di erence( S S1 ),andthelevelofmarginalassetspeci cityofi S forb by( S1 S0 ). 11 Theincreaseinmarginalpeople speci cityofi S forb leadsto adi erentialchangein of 1 2 [i S (0; ) i S (1; )];whichispositiveaslongass investsmorewhenheownstheupstream asset(i.e. when there is marginal asset speci city present). To understand this result, note that such a change increases S's investments. Inthe case in whichwe have under-investment, thisreducestheextentofdistortionini S,andtherebylowersthecostofbuyerintegration (which reduces S's investment incentives). Hence, the probability of buyer integration rises. Bywayofcontrast,anincreaseinthelevelofmarginalassetspeci cityforb ofi S causesa di erentialchangein of 1 2 [ S i S (0; )],whichisnegativeintheunder-investmentcase. This is so because such a changeraises S'sinvestmentlevel undernon-integration, which raises the probability of observing that structure. Parallelde nitionscan be formulatedforthemarginalspeci cityofinvestmentsfors. Increasesinspeci cityfors allhavetheoppositee ectsfromthosederivedforincreasesin speci city for B. The results for these cases and those derived above are summarized in Table Comparison oftheproperty rights theorywith thewilliamsonian theory The foregoing analysis reveals a number of signi cant distinctions between the property rights theory and the Williamsonian theory. First, and most immediately, while the Williamsonian theory is concerned with the level of quasi-rents, the property rights theory's focus is on the marginal returns to non-contractible investments. These need not be closely related in thedata. Forexample,thelevelofquasi-rentscanvaryacrosssituationswithabsolutelyno variation in marginal returns, or even with marginal returns moving in the opposite direction. Even if marginal returns and the levels of quasi-rents move in the same directions, in the propertyrightstheorythee ectsofsuchchangesdependdelicatelyonanumberoffactors. First,changesinmarginalreturnsunderdi erentownershipstructureshavedi erente ects on the probability of observing an integrated structure: in the property rights theory, an increase inthe marginal return to a self-investment by B under non-integration has the opposite e ect on the probability of integration from the same increase under B ownership. 11 The de nitions of marginal people speci city are perhaps most natural in the cases in which investments are either purely self-investments or purely cross-investments. 10

13 In contrast, such di erences do not arise inthe Williamsonian theory because quasi-rents are assumed to matter only under non-integration: the owner in a vertically integrated structure is assumedto be able to fully direct the outcome within the integrated rm. In addition, inthe property rights theory thee ects of changes in marginal returns reverse completely when we focus on cross-investment e ects instead of self-investment e ects. Similarly, the responses to changes in the speci city of investments depend on precisely the type of speci city we are considering and whose investment returns are becoming more speci c. For example, forself-investmente ects,thee ectofanincreaseineitherpeople orassetspeci cityofb'sinvestmentistheoppositeofthee ectsofsimilarincreasesins's investment. When wethenconsider the speci city ofcross-investment e ects, the e ect of greaterlevelsofassetspeci cityarethesameasfortheself-investmentcase,butthee ects of greater levels of people speci city are the opposite of those in the self-investment case. Theseobservations suggest that thereare in fact a richanddemanding set oftheoretical predictions that in principle can be used to test the property rights theory. However, they also suggestthata greatdealofinformationaboutthetradingenvironmentisnecessaryto doso. 3.Inferences fromthreeprominent Papers In this section Iexamine what we canlearnabout the applicability of the property rights theory from three ofthe more prominent empirical studies of transaction cost determinants ofvertical integration. In turn, I discuss Monteverde and Teece[1982]'s study of automobile assemblers' procurement of automobile components, Masten[1984]'s examination ofprocurement intheaerospaceindustry, and Joskow [1985]'s discussionofvertical integration betweenelectric utilities and coal mines Monteverdeand Teece[1982] Moneteverde andteece[1982](henceforthm&t) providedthe rst econometric study of the transactioncosts determinants of vertical integration. In their path-breakingpaper, M&T soughttoexplainthelevelofinternalversusexternalprocurementfor133componentsused by GM and Ford in M&T focused not on physical asset speci city, but rather on the transaction-speci c know-how that would be generated during thedesign development process: 11

14 \We hypothesize that assemblers will vertically integrate when the production process,broadlyde ned,generatesspecialized,nonpatentableknow-how....the existenceof transaction-speci c know-how andskills and the di±culties of skill transfer mean that it will be costly to switch to an alternative supplier(teece 1977, 1980). Anassembler will tend tochoose vertically integrated component production when high switching costs would otherwise lock the assembler into dependence upon asupplier and thereby expose theassembler toopportunistic recontracting or to the loss of transaction-speci c know-how." M&Tusedanordinalmeasureofthelevelofengineeringe ortthatwentintodesigning thecomponenttoproxyforthelikelyextentofspecializedknow-howacquisition. Inaddition,theyincludedcontrolvariablesforthespeci cityofthecomponenttothemanufacturer inquestion(speci cally, whether the component canbe useddirectly in avariety of cars), an assembler dummy variable, and subsystem dummy variables. They found statistically signi cant evidencethat bothincreased levels ofengineeringe ort andincreasedcomponent speci city raised the likelihood ofintegration and that GM was more likely to procure internallythanford,otherthingsequal. 12 >From the standpoint of the Williamsonian theory the results look like a resounding success: the twovariables designed tocapturethe extent of quasi-rents both indicate that greaterlevelsofquasi-rentsincreasethelikelihoodofintegration. Butwhatdotheytellus aboutthe propertyrightstheory? Thisisnotsoclear. Tobeginwith,M&Ttellusnothing about the extent to which important investments are not contractible: in the absence of non-contractible investments, of course, the property rights theory would predict no e ects should be observed(in contradictionto the evidence). More generally, if all of the speci city inthis situation is people speci city arising from the acquisition of know-how{that is, if there is noasset speci city for these non-contractible investments{then againthe property rightstheorytellsusthatno e ectsshouldbeobserved. Thus,ifwearenottorejectthe property rights theory, we will needto consider models inwhich non-contractible investments are present whose marginal returns exhibit asset speci city. To develop such models, we needto think rst abouthowthepayo sforcomponents withahigherlevelof \know-how" andthosewithahigherlevel ofcomponentspeci city 12 It is somewhat unfortunate, however, that M&T provided no estimates of the economic magnitude of the engineering and component speci city e ects (i.e. the degree towhich changes in these variables altered the probability of internal procurement). 12

15 di erfromthosewithlowlevelsofeachofthesevariables. Inwhatfollows,weshallthinkof componentswithahigherlevelofknow-howashaving: (i)thesamepayo forbintheevent hedoesnotreachanagreementwithsanddoesnotowntheupstreamasset,inwhichcase he mustuse acomponentnotembodyinganyofthisknow-how; (ii)ahigherlevel ofjoint payo if B and S do trade,since theycanthenbene tfromthisknow-how; (iii)possibly a higherpayo tob when he doesnotreachanagreementwithsbutownstheupstream asset, since some of the know-how may be embodied in the asset; and(iv) possibly a higher level of payo for S when he fails toreachanagreementwithb,sincehemay beableto usehisknow-howinsalestootherbuyers. Thus,wemayseeincreasesinsomeorallof B and S [point(ii)], 1, B1, S1 [point(iii)],¾ 0,¾ 1,¾ B0,¾ B1,¾ S0 ;and¾ S1 [point(iv)]. In contrast, we shall think of an increase in the component speci city variable primarily asreducingthepayo stos inthe eventhedoesnotreachanagreementwithb;sincehis componentcannotbereadilysoldtootherbuyerswithoutmodi cation(possiblyleadingto reductionsin 0, 1, B1, S1, B0,and S0 ). Itisalsopossible,however, thatitmaybe associatedwith adecrease inb'spayo sintheeventthathedoesnotreachanagreement with S, since B's investment may be tailored to the speci c component and a more speci c componentmayhavefeweralternativesourcesofsupply(possiblyleading toreductionsin ¾ 0,¾ 1,¾ B1,¾ S1,¾ B0,and¾ S0 ). With these e ects identi ed, wecan imaginea number of plausible property rights models of this situation involving non-contractible investments and asset speci city. As we shall now see, some of these models do produce results that are consistent with the M&T results, while some do not. Consider the following three possibilities: Model MT1: Know-how acquisitionis exogeneous and B makes non-contractible selfinvestments that are complementary to S's acquisition of know-how. As an example, we can imagine that Binvests in marketing and distributing its cars, but thesuccessoftheseinvestmentsdependsonthequalityofthecomponentsthats produces, whichisinturna ectedbys'slevelofknow-how. Torepresentthiscase,assumethatonly B hasanon-contractibleinvestment,that¾ B0 =¾ B1 =0,andthat B > B1 > B0. This last assumption capturesthatidea thatb'smarginalreturnsto investmentareincreasing inthe levelofknow-howthatisincorporatedintothecomponent, andthatthisishighest whenbhasaccessto both S andtheupstreamasset,andislowestwhenbhasaccessto neithersnortheupstreamasset. Notethatinthissituation,Bunder-investsundereither 13

16 ownership structure, and invests more when he owns the upstream asset. Consider rstthe e ectsof increasinglevelsofknow-howinsuchasetting. Thiswould reasonablybeexpectedtoincrease B. Moreover,itmightincrease 1 and B1 ifs'sknowhowispartiallyembodiedintotheupstreamasset(andsobene tsbwhenbownsthisasset andcannotreachanagreementwiths). Finally,itmightincrease¾ 0 and¾ 1 ifscanuseits know-howin dealingswithotherpotentialbuyersintheeventthatbandsdonottrade. In the property rights model, these e ects increase the probability of observing integration, and so are consistent with M&T's ndings. Nowconsiderthee ectsofanincreaseinthelevelofcomponentspeci city. Anincrease incomponentspeci cityislikelyto decrease¾ 0 and¾ 1 byreducingthedegreetowhich S canuseitsknow-howtoserveotherbuyersintheeventofnotreachinganagreementwithb. Thishasno e ectonthe probabilityofintegration. Inaddition,anincreaseincomponent speci citymayreduce 0, 1, B0 and B1 ifb'sinvestmentreturnsaretiedtothespeci c component and more speci c components are less likely to be available from an alternative source. While changes in 0 and 1 have no e ect ontheprobability ofintegration, the changesin B0 and B1 haveo settinge ects. Iftheydecreaseequally{sothatthechange amounts to an increase in the marginal people speci city ofi B without any a ecton its marginal asset speci city{then the probability of integration will increase in the property rights model. Since this reduction in outside sources is more likely to matter when B does notownthe upstreamasset, wemightexpect B0 to fallby morethan B1 ;thatis,asset speci city would increase. This wouldalso leadto an increase in the probability of observing integration. Thus,the predictionsofthis rstpropertyrightsmodelarefullyinaccordwithm&t's ndings. Model MT2: B'sinvestmentscreateknow-howfor S. As anexample, B maybedevoting e ortto helping S understandhowits component canbe more e ectiveinanautomobile, where someof this knowledge is speci c tob 's automobilesand some appliesto all manufacturers' automobiles. To representthis case, assume that only B has a non-contractible investment, that B > ¾ B0 > ¾ B1 > 0 (the bene tstos'sknow-howarelargestwhenbands trade,andsmallestwhens servesother buyerswithoutaccesstotheupstreamasset),andthat B1 B0 =0with B1 >0ifsome ofs'sknow-howisembodiedintheupstreamasset. Undertheseconditions,Bunder-invests 14

17 and Binvests more when he owns the upstream asset. Inthiscase weexpectmoreknow-howtobeassociatedwithincreasesin B,in¾ 0,¾ 1, ¾ B0,and¾ B1 ifknow-howcanbeusedbys inhisdealingswithotherbuyerswhenhedoes notreachanagreementwithb,andin B1 ifknow-howispartiallyembodiedintheasset. The increasesin B and B1 both raisethelikelihoodofanintegratedoutcomeunderthe propertyrightstheory. Thechangesin¾ B0 and¾ B1 haveo settinge ects. If moreknowhowincreases¾ B0 atleastasmuchas¾ B1,aswouldbeexpectedifSismoreabletousehis know-howwhenhe hasaccessto theupstreamasset,thenthenete ectoftheincreasesin these twoparameters is also to increase the probability of integration(whenthe increases in¾ B0 and¾ B1 areequalitamountstoadecreaseinmarginalpeoplespeci cityofi B fors withnochangeinmarginalassetspeci city). Inthisenvironment,morecomponentspeci citywilllimitthereturnsthatscanachieve intheeventhedoesnotreachanagreementwithb. Weexpectthesechangestobeassociated withdecreases in ¾ 0, ¾ 1, ¾ B0, and ¾ B1. It may also reduce 0 and 1 if there are fewer alternativesourcesforamorespeci ccomponent(wedonotexpectittodecrease B1 since if this is positive itmeans thatb nds self-production to be optimal when he owns the upstreamassetanddoesnottradewiths). Oftheseparameterchanges,onlythechanges in¾ B0 and¾ B1 matterforintegrationdecisions. Anequal changeamountsto anincrease inthe marginal people speci city of i B fors, whichreduces the likelihoodofintegration. Moroever,morecomponentspeci citymayalsobeexpectedtodecrease¾ B0 atleastasmuch asitdecreases¾ B1 iftheknow-howistiedtothespeci ccomponentandifhavingaccessto the upstream asset increases S's ability to produce the speci c component. If so, we would also have areductioninthemarginalassetspeci cityofi B fors,whichwouldalsoserveto reduce the likelihoodofintegration. Thus, in this second property rights model, the predicted a ects of more component speci ctyareatoddswithm&t's ndings. Model MT3: S'sinvestmentscreateknow-howfor S. Forexample,S'sacquisitionof know-howmaybedeterminedinsteadbythee ortthat Sputsintounderstanding howto producethecomponent. To representthiscase,assume thatonlyshasanon-contractibleinvestmentandthat S >¾ S0 >¾ S1. Theseinequalities re ectthatfactthats'sknow-howhasthelargestreturnwhenhehasaccesstotheupstream asset andtrades with B, and has the smallest returnwhen he has access to neither the 15

18 upstreamassetnorbasatradingpartner. Inaddition, S1 S0 =0,with S1 non-zeroif some of S's know-how is embodiedinthe upstream asset. These assumptions imply that S under-invests under either ownership structure and that he invests more when he owns the upstreamasset. Inthiscase,we expectmoreknow-howto associatedwithincreasesin S,¾ 0, ¾ 1, ¾ S0, ¾ S1,andpossiblyin S1 ifknow-howgetsembodiedintheupstreamasset. Increasesin S and S1 are associated with decreasesintheprobability ofintegration. Onceagain, the increasesin¾ S0 and¾ S1 haveo settinge ects,butif(aswemayexpect)theincreasein S isatleastaslargeastheincreasein¾ S0,whichisinturnatleastaslargeastheincreasein ¾ S1 ; thenthenete ectofthese threechangesistodecrease theprobability ofintegration (recallthatanequalchangeinallthreeparametershasnoe ect,andsothechangediscussed canbethoughtofasstartingfromsuchachange,andthenfurtherincreasingboth S and ¾ S0 ). Increasesincomponentspeci citywillleadtodecreasesin¾ 0,¾ 1,¾ S0,¾ S1,andpossibly in 0 and 1 if alternative sourcesfor B are more limitedwhencomponent speci city is higher(again,wedon'texpectittoleadtoadecreasein B1 whenthisispositive,sinceifit ispositivethismeansthatb'sbestalternativewhenhedoesnotreachanagreementwiths andheownstheupstreamassetistoself-produce). If¾ S0 and¾ S1 decreaseequallythenwe haveapure increaseinthemarginalpeoplespeci cityofi S,whichlowersthelikelihood of integration. However,wemayexpectthedecreasein¾ S0 tobeatleastaslargeasthedecrease in¾ S1 since hisoutsidealternativesarelikelytobehurtmorebytheincreaseinspeci city when he owns the upstream asset(since he thencan produce the speci c component). The nete ectofthischangetherefore appearstobeambiguous. Thus,thepredictionsof thisthirdpropertyrightsmodelcontradictm&t's ndingsregading the ects of increasedknow-how, and possibly also their ndings concerning increased componentspeci city. The conclusion to be drawn from this discussion seems to be that there is simply not enough information provided in M&T's study to evaluate the predictive power of the property rights model Masten [1984] Masten[1984] providedthe secondeconometric study ofprocurement integrationdecisions. Masten's focuswasontheprocurementdecisionsofa largeaerospacecompany over1,887 16

19 components. Masten includedas explanatory variables the degree of component complexity (as a measure of the di±culty of complete contracting), the degree to which the component was specialized to this aerospace rm(similar to M&T's component speci city variable), andthe degree to which colocation offacilities orprocesses wasthought to be important (as another measure of quasi-rents). He found that the rst two variables had positive and statistically signi cant e ects on the likelihood of integration, while the third variable had a positive but statistically insigni cant e ect onthe likelihoodof integration. He also found thatthepositivee ectofcomplexityseemedtooccuronlyconditionalontherebeingahigh levelofcomponentspeci city,andthatthee ectofcomponentspeci citywasmuchhigher whenthepartwascomplex. 13 Once again, theresults appearto beasuccessforthewilliamsoniantheory, although perhaps a less complete success than the M&T results given the insigni cance of the site speci city variable (although the documented importance of the complexity variable adds a distinct supporting piece of evidence). As we have already seen, however, we could readily describe reasonable property rights models of this setting to get the e ect of increased component speci city on the probability of integration to go in either direction. Moreover, asishalldiscussindetailwhenconsideringjoskow's[1985]studyinthenextsubsection,a similar conclusion can be drawn about the e ects of colocation. Without more detail about the contracting environment, little can be said about the ability of the property rights model to explainthe integration patterns found in Masten's data Joskow [1985] Joskow [1985] studies the coal procurement decisions of electric utilities. In contrast to the previous two papers, Joskow [1985] is not aneconometric exercise, but it does o er its readers a more detailed view of the investment and procurement process thando those papers. Its primary nding regarding vertical integration is that vertical integration appears to be much more likely for\mine-mouth"electric generating plants than for others. This nding is interpreted as being in accord with the Williamsonian theory, since such colocation raises the level of quasi-rents. Whatdoesthepropertyrightstheoryhavetosayaboutthee ectofcolocation(which, it seems clear, is itself a contractible decision) on the probability of integration? To x ideas, 13 One concern, however, is that this nding could be driven by the strong functional form assumptions embodied in the probit model Masten estimated. 17

20 Figure3.1: considerfigure 1. There are ve potentialcoalmines,labeleds ands1tos4:thebuyer, an electric utility plant, serves customers in the city, and the closest coal mine tothe city iss:allroadsgothrough the city. Nowconsiderthee ectofb movingthelocationof its plantfromf (Far)toN (Near);whichisnearertoSandfurtherfromalloftheothercoal mines. 14 Note thatithasnoe ectons'spayo whens cannotreachanagreementwith B (S'sdistanceto otherelectric utilitieshasnotbeena ected). 15 Thus,incontrasttothe component speci city variables in Monteverde and Teece[1982] and in Masten[1984], the colocation variable here a ects only B's disagreement payo s. Asecond di erence from the component speci city variable usedinthose studies is that we can reasonably put much more structure on the e ects of colocation. In particular, for given investment choices by theutility and themine, the e ects of colocationon the payo s when B and S trade, and on their payo s from each alternative trading possibility in the 14 It is worthwhile noting that the e ects derived below would reverse if moving closer to S 1 also moved B closer to the other coal mines. 15 Note that in actual data it could be that colocation decisions do (on average) a ect the distance of the coal mine to other electric utility plants, in contrast to our maintained assumptions here. 18