vertical integration and firm boundaries: the...

Journal of Economic LiteratureVol. XLV (September 2007), pp. 629–685

Vertical Integration and FirmBoundaries: The Evidence

FRANCINE LAFONTAINE AND MARGARET SLADE∗

Since Ronald H. Coase’s (1937) seminal paper, a rich set of theories has been devel-oped that deal with firm boundaries in vertical or input–output structures. In the lasttwenty-five years, empirical evidence that can shed light on those theories also hasbeen accumulating. We review the findings of empirical studies that have addressedtwo main interrelated questions: First, what types of transactions are best broughtwithin the firm and, second, what are the consequences of vertical integration deci-sions for economic outcomes such as prices, quantities, investment, and profits.Throughout, we highlight areas of potential cross-fertilization and promising areasfor future work.

629

1. Introduction

Understanding what determines firmboundaries and the choice between

interacting in a firm or a market is not onlythe fundamental concern of the theory ofthe firm, but it is also one of the most impor-tant issues in economics. Data on valueadded, for example, reveal that, in theUnited States, transactions that occur infirms are roughly equal in value to those thatoccur in markets.1 The economics profes-sion, however, has devoted much moreattention to the workings of markets than to

the study of firms, and even less attention tothe interface between the two. Nevertheless,since Ronald H. Coase’s (1937) seminalpaper on the subject, a rich set of theorieshas been developed that deal with firmboundaries in vertical or input–output struc-tures. Furthermore, in the last twenty-fiveyears, empirical evidence that can shed lighton those theories has been accumulating.

The empirical literature on vertical inte-gration has focused on two main, interrelatedquestions: First, what types of transactionsare best brought within the firm or, put dif-ferently, under what circumstances do weobserve that an input or service is producedin house? And second, what are the conse-quences of vertical integration for economicoutcomes such as prices, quantities, invest-ment, and profits? The answers to thosequestions are important in that they caninform managers’ decisions directly. But theyare also important ultimately as input into the

∗ Lafontaine: University of Michigan. Slade: Universityof Warwick. We would like to thank Roger Gordon andtwo reviewers for helpful comments, and Desmond (Ho-Fu) Lo for his assistance. Margaret Slade would like toacknowledge financial support from the LeverhulmeFoundation.

1 For manufacturing the ratio is about one third,whereas for services it is twice that. Calculated by authorsfrom Census bureau data.

sep07_Article2 8/17/07 4:29 PM Page 629

development of sensible vertical merger pol-icy and related government intervention invertical relationships.

In this paper, we review the findings ofempirical studies that have examined eitheror both questions. Recent articles have sur-veyed the theories of vertical integration(see, e.g., Bengt Holmstrom and JohnRoberts (1998), Michael D. Whinston(2003), and Robert Gibbons (2005)) with aneye toward highlighting similarities and dif-ferences among the theories, while othershave surveyed the evidence that relates to aparticular theory (see, e.g., Peter G. Klein(2005) or Howard A. Shelanski and Klein(1995), and Francine Lafontaine andMargaret E. Slade (1997, 2001) respectivelyfor surveys of tests of transaction-cost andmoral-hazard models). However, we areaware of no prior survey of the evidence onvertical integration that encompasses thedifferent approaches and tests of variousmodels.2

Our reasons for reviewing the evidence onthis topic at this time are twofold. First, webelieve that enough evidence has accumulat-ed by now and it is time to assess what theempirical regularities can tell us about thepredictive power of existing theories, as wellas how they can guide the development offuture theories. In particular, we are inter-ested in highlighting areas of potential cross-fertilization, namely how tests of one type oftheory might be relied upon to inform us asto the validity of other theoretical approach-es. Second, we also believe that it is impor-tant to examine what the evidence can tell usabout the efficacy of public policy toward

vertical mergers and divestitures, as well ashow it can guide future competition policy.

The paper is organized as follows. In thenext section, we present theories and evi-dence about the decisions that firms makeconcerning their boundaries. We begin withdecisions to integrate forward into retailingand then discuss backward integration intoinput production. We treat these separatelylargely because the models that authors haverelied upon to derive testable implicationshave been different for these two sets ofdecisions. Specifically, most of the empiricalliterature on firms’ decisions to integrate for-ward into retailing relies on incentive andmoral-hazard type arguments, whereas theempirical literature on backward integration,otherwise known as the “make or buy” deci-sion, mostly tests predictions derived fromtransaction-cost arguments. In our treat-ment of both forward and backward integra-tion, we begin with an overview of somestylized facts from the literature and thenpresent simple versions of the relevant mod-els that highlight the predictions that havebeen taken to data.3 We then organize ourpresentation of the evidence around the pre-dictions derived from the models. In partic-ular, the empirical studies are organized intotables according to model tested (e.g.,moral-hazard) and issue addressed (e.g.,riskiness of transactions). Each table con-tains information on the industry examined,the empirical technique used, and theauthor’s interpretation of the findings. Weconclude this section with some thoughts onthe potential for cross fertilization, that is,how evidence relating to one model or con-text can also shed light on other models ofvertical integration. In section 3, we reviewthe theories and evidence concerning theconsequences of vertical integration for anumber of different outcomes, including

630 Journal of Economic Literature, Vol. XLV (September 2007)

2 See James C. Cooper et al. (2005) and Lafontaine andSlade (2006) for reviews of the empirical literature on ver-tical restraints as opposed to vertical integration. Notethat we focus mostly on empirical research published ineconomics. TCE-based empirical research published inmarketing and in management or strategy journals in par-ticular is quite voluminous and, although we discuss someof these studies, a complete overview of this literature isbeyond the scope of the present article. But see Klein(2005) for example for a broader coverage of TCE-basedresearch.

3 Some might say that our models are simplistic.Nevertheless, the simple models, while neglecting muchof the richness of the theories, are capable of capturingmost of the comparative statics that empiricists havefocused on.


4 We very briefly mention potential contractual alter-natives to vertical integration in section 3, where we pres-ent market power arguments for vertical integration. Thenotion that contracting and vertical integration are equiv-alent is a recurring theme in particular in the literature onvertical restraints. However, in the models of vertical inte-gration we focus on, what is and is not vertically integrat-ed is well defined, but differs across theories. Forexample, agency theory focuses on differences in residualclaims—with vertical integration involving no such claimsfor the agent—while property right theory defines verticalintegration as common ownership of assets and associatedcontrol rights. A complete treatment of contracting and

when it is similar or not to vertical integration is beyondthe scope of the present paper. However, see Cooper et al.(2005), Lafontaine and Slade (forthcoming-a), andLafontaine and Slade (forthcoming-b) for reviews of theempirical literature on the effects of vertical restraints andon interfirm contracts respectively.

5 Of course, decisions to integrate forward into retailingare also a form of “make or buy” decision in that the firmis choosing to make or buy downstream distribution serv-ices. However, the literature typically reserves the expres-sion “make or buy” to contexts where firms integratebackward, and we follow this convention here.

prices, costs, profits, and investment. Thesections on both incidence and conse-quences include brief discussions of themain econometric problems that authorsface, as econometric and data issues havebeen major challenges in this literature.Finally, in the last section, we draw somegeneral lessons from the body of evidence,focusing in particular on what it can tell usabout the theories of firm boundaries aswell as about public policy toward verticalmergers.

2. The Vertical-Integration Decision

In this section, we examine the evidencethat relates to circumstances under whichfirms choose to integrate vertically. This inturn requires that we define precisely whatwe mean by vertical integration and markettransaction. The difference that we empha-size is that, under the former, ownership isjoint and control rights are integrated,whereas under the latter, they are separate.As will become clear, we do not distinguishbetween the entrepreneurial firm and themodern corporation nor do we discuss issuesof governance within firms or modes of mar-ket organization. In particular, while we rec-ognize the important role of contracts aspotential ways to achieve “almost integra-tion,” we rely on authors’ institutionalknowledge and, as such, do not question thedefinition of vertical integration and marketsthat is used in the empirical studies. In mostcases, this implies that we equate contractswith arms length transactions and contrast

firms’ decisions to rely on such transactionsversus vertical integration.4

The empirical literature on the vertical-integration decision is easily divided into twomajor segments: those papers that considerthe decision whether to integrate forwardinto retailing and those that examine the“make or buy” decision, which is the decisionwhether to integrate backwards.5 We discussthe evidence on these below in that order. Inboth cases, we begin by describing some styl-ized facts, followed by a simple version of thetype of model that authors have relied uponto derive the hypotheses that they test. Ourmotive for discussing the theories is not toproduce a comprehensive survey of theirrichness but rather to provide us with aframework within which to present the evi-dence. In particular, the evidence that wepresent is organized around the predictionsof simple bare-bones models.

Throughout our presentation, we discusssome of the measurement challenges thatauthors face but mostly ignore econometricproblems. We do this to keep the overviewtractable and of reasonable length. This isnot to say that the econometric problems areunimportant. To make this point clear, ineach of the two segments we highlight someof the econometric issues that researchersmust confront. Since those issues are notalways dealt with satisfactorily, one can beskeptical about some of the conclusions thatauthors have reached. Nevertheless, taken asa body, the evidence is often so strong that itcan overcome much of our skepticism.

631Lafontaine and Slade: Vertical Integration and Firm Boundaries


2.1 Forward Integration into Retailing

The empirical literature on forward inte-gration generally considers a manufactur-er’s decision to sell her outputs toconsumers herself—that is reaching cus-tomers through premises she owns andoperates directly—versus using independentretailers. This question, in turn, arises in con-texts where manufacturers produce a set ofoutputs that can be sold by themselves inbranded stores. For that reason, this litera-ture has been concerned with distributionunder exclusive dealing, as in the case of fran-chising, rather than common agency, such assales through department or grocery stores.

Franchising commonly takes two forms:traditional and business format. The formerinvolves an upstream manufacturer and adownstream retailer (e.g., gasoline or auto-mobile sales), whereas the latter does notinvolve upstream production. Instead, thefranchisor sells a business format—a way ofdoing business—to the franchisee and allowshim to use the trademark (e.g., fast-foodsales or hotel services).

A franchise, whether traditional or busi-ness format, is an independent businessunder the law and is thus not vertically inte-grated with the upstream firm. Nevertheless,transactions are often not completely arm’slength. Indeed, business-format franchisecontracts are normally long term, and involvethe payment of royalties (ρ) and fixed fran-chise fees ( f) to the principal. The agentthen obtains (1 − ρ)q − f where q is the valueof output, as compensation for his effort.Traditional franchises, on the other hand,are dealer networks where franchisors,instead of charging fees to franchiseesdirectly, earn a return on the products theysell to them.6

Most franchisors operate some outletsdirectly, while they franchise others. Due todata constraints, many empirical studies ofbusiness-format franchising have focused onthe proportion of company owned, or verti-cally integrated, outlets across chains as theirmain dependent variable.7 Studies of tradi-tional franchise relationships more often havelooked at the vertical integration decisionoutlet by outlet.

The literature has revealed a number ofconsistent patterns. In particular, there issystematic evidence that franchisors andmanufacturers rely on independent retailersor franchisees to a greater extent the moreimportant is the effort of the franchisee orthe more geographically dispersed the oper-ations of the firm are. Authors have alsofound a positive relationship between risk orsales variability and the use of franchising.On the other hand, these firms verticallyintegrate more when the inputs provided bythe franchisor, namely the value of thebrand, is greater. They also integrate agreater proportion of their outlets whentheir outlets are larger.

In what follows, we provide the detailedresults behind these stylized facts. In themajority of cases, authors in this literaturehave relied on agency theoretic argumentsand, more specifically, incentive or moral-hazard models of franchise relationships, toderive predictions to take to their data, andperhaps most importantly, organize and inter-pret their results. In the next section, we pres-ent a simple moral-hazard (MH) model thatgenerates many of these predictions. At thesame time, the model helps in pinpointing thetype of empirical model of vertical integrationthat arises from this approach, and the simi-larities and differences between this approachand others we discuss further below.


6 The distinction between business-format and tradi-tional franchising is partly a matter of degree as somebusiness-format franchisors, e.g., Baskin Robbins, also sellinputs to their franchisees and earn a return on such sales,and the payment scheme in traditional franchising mayalso include some payments beside per unit markups (e.g.,rental fees in gasoline retailing).

7 A few studies examine also, or instead, factors thataffect the proportion of revenues to the principal (royaltyrate). Given our focus on vertical integration, we do notreview those results, but simply note that they are general-ly consistent with those found for the proportion of outletsthat are integrated.


2.1.1 The Moral-Hazard Model

The idea that risk and uncertainty areimportant determinants of firm size andscope dates at least as far back as FrankKnight (1921), who emphasized the need toinsure workers and consolidate managerialdecision making. The problem with insur-ance, however, is that workers who are fullyinsured do not necessarily have appropriateincentives to supply effort. Moral-hazardarguments for firm boundaries thusemphasize the trade-off between providingworkers with insurance, which firms do well,and with effort incentives, which marketsdo well.

To illustrate, it is common for firms topay workers fixed wages that are independ-ent of performance, at least in the shortrun.8 Within the firm, therefore, incentivestend to be low powered whereas insuranceis high, since worker pay does not fluctu-ate. Independent contractors, in contrast,are entrepreneurs who receive the profitsthat remain after variable costs have beenpaid. In other words, they are residualclaimants. When a transaction occurs in amarket, such as when an input is procuredvia an independent contractor, incentivesthus tend to be high powered. However,the independent contractor also bearsmuch risk, since his pay fluctuates inresponse to both demand and productionshocks.

We use a standard principal–agent moral-hazard model of worker compensation toderive some testable hypotheses.9 A slightmodification of that model yields a theory ofvertical integration.

Since our goal is to make predictionsabout forward integration, in our model, the

principal is a manufacturer (M) while theagent is a retailer (R). Assume that bothprincipal and agent must exert effort, aM andaR, respectively. Examples of such effortswould include advertising the brand, usinghigh-quality inputs, and performing servicesat the point of sale. Output is produced (salesare realized) according to the productionfunction

(1) q = f(aM,aR,u),

where u is a random variable that capturesuncertainty in the production process. Forsimplicity, we assume that this function islinear, which implies that retailer and manu-facturer effort are additively separable, andthat u is normally distributed with zero meanand constant variance, so that

(2) q = β0 + βMaM + βRaR + u, u ~ N(0,� 2).

In this production function, βM and βR,which are the marginal products or returnsto manufacturer and retailer efforts, areassumed to be nonnegative. In other words,effort is not unproductive.10

The principal would like to design an opti-mal payment scheme for the agent.However, agent effort is not observable and,due to the presence of u, it cannot beinferred by the principal.11 We assume thatthe compensation scheme is based on real-ized output, q, which we take to be observ-able.12 We also assume that the agentcompensation scheme is linear, i.e.,s(q) = �q + W, where � is a parameter thatdetermines the intensity of incentive pay,and W is a fixed wage that is independent of


8 Clearly there are other payment schemes, such aspiece rates, that are available to firms. However, fixedwages are more common, at least for nonmanagerialstaff. Of course, career concerns provide incentives toforward-looking workers.

9 Variants of this model can be found in Lafontaine andSlade (2001).

10 With this formulation, q can be negative and hencemay best be thought of as profit or returns than sales.However, the probability that q < 0 can be made arbitrarilysmall by our choice of β0.

11 In other words, this a moral-hazard model.12 Difficulties in either measuring output or inferring

effort from output will reduce its appeal as a compensa-tion basis, and increase the appeal of alternatives such asdirect quality monitoring or other signals of effort. See thediscussion of costly monitoring in the next subsection onthis issue.


effort.13 Finally, the private cost of effort isc(ai) = 1−2 (ai)2, i = M,R, and there are no othercosts.

The parameter, �, plays a key role in theanalysis as it determines the agent’s share ofresidual claims. Two limit cases are of inter-est. When � = 0, the agent is a salariedemployee who is perfectly insured, whereaswhen � = 1, the agent is the residualclaimant who bears all of the risk. Oneexpects that, in general, 0 ≤ � ≤ 1, but, as wediscuss below, the firm may not find it optimalto use nonlimit values for �. Still, we identify� with the power of the agent’s incentives.

We assume that the principal is risk neu-tral, whereas the agent, who is risk averse,receives utility from income y according tothe constant absolute risk aversion, orCARA, utility function, U(y) = − e−ry, where ris his coefficient of absolute risk aversion.Note that agent risk aversion was required togenerate a sharing arrangement in earlyagency models. More recently, and partlybecause of some of the evidence that wesummarize below, models that do not requireagent risk aversion have become more com-monplace. Moreover, even our simple modelyields output sharing without risk aversion.Nevertheless we introduce agent risk aver-sion to generate many of the predictions thathave been tested in the literature.

The first-best solution is the set of effortlevels that maximize the joint surplus. Underour assumptions, the first-best efforts area∗

i∗ = βi, i = M,R.The second-best problem has two incen-

tive constraints: given the payment scheme,the principal chooses effort to maximize herexpected income, E(π), and the agentchooses effort to maximize his certainty-equivalent income, E(y) − r−2 VAR(y), where

E(•) and VAR(•) are the expectation and vari-ance functions, and the term r−2 VAR(y) is theagent’s risk premium.14

The first-order conditions for those maxi-mizations are a∗

M = βM(1 − �) and a∗R = βR�.

Note that, in general, the situation is one ofunderinvestment in effort by both partiesrelative to the first best. Furthermore, as �increases (falls), the agent’s (principal’s)effort moves toward first best, but the prin-cipal’s (agent’s) effort moves toward zero.15

Finally, the principal chooses � to maxi-mize the joint surplus, taking into accountboth incentive constraints and the agent’sparticipation constraint. This maximizationyields16

β 2R(3) �∗ = ⎯⎯⎯⎯⎯⎯ .

β 2R + β 2

M + r� 2

One can transform equation (3) into anempirically tractable model of the shareparameter, �, as a function of a set of variablesthat capture the fundamentals of the technol-ogy and the agent’s utility, namely the β’s, r,and �, by appending a random variable � withcumulative distribution function F(•) to (3).17

We interpret that variable as representingthose factors that affect desired compensationbut are unobserved by the econometrician.18

The model thus far has focused on agentcompensation, specifically incentive pay. Aslight modification yields a theory of verticalintegration. For this version, we use the par-ticipation constraint, which will bind in equi-librium (i.e., U(y) = 0), to obtain W as afunction of the model parameters. This


13 The assumption of a linear compensation scheme ismotivated by an empirical regularity: a large fraction ofreal-world contracts take this form. Optimal contracts arein general more complex. However, see SugatoBhattacharyya and Lafontaine (1995) for a discussion of theoptimality of linear contracts in cases where both principalsand agents exert unobservable effort, as is the case here.

14 Given our assumptions on functional forms, thisexpression for the risk premium is exact.

15 We interpret zero effort as some minimal level.16 See Lafontaine and Slade (2001) for this and other

calculations.17 We assume that F is differentiable and that F′ > 0.

Furthermore, in deriving our model of vertical integra-tion, we assume that F(•) is symmetric, an assumption thatcan easily be relaxed.

18 See, for example, Lafontaine (1992) and Lafontaineand Kathryn L. Shaw (1999) for empirical analyses offranchise-contract sharing terms that rely on thisapproach.

sep07_Article2 8/17/07 4:29 PM 34

function is substituted into the principal’sexpected profit under vertical separation,which becomes

(4) E(π)VS = (1 − �)E(q) − c(aM) − W =1 1−β 2

M + −β 2R(�∗ − �).2 2

Profit under integration (� = 0) is

1(5) E(π)VI = −β 2M.2

Suppose that a fixed cost is associated withwriting and administering a contract, a trans-action cost T. Then the principal will chooseintegration if

(6) E(π)VI − [E(π)VS − T] =1− −β 2

R(�∗ − �) + T ≥ 0.2

This will be true if

2T β 2R(7) ⎯⎯ − ⎯⎯⎯⎯⎯⎯ ≥ − �.

β 2R β 2

R + β 2M + r� 2

The probability of observing integrationthen is

2T β 2R(8) PROB[VI] = F�⎯⎯ − ⎯⎯⎯⎯⎯⎯�,β 2

R β 2R + β 2

M + r� 2

which is our discrete-choice model of verticalintegration. The model predicts that, in caseswhere the costs of incentive contracting out-weigh the benefits, the principal will offerthe agent a uniform-wage contract (� = 0 orVI). Furthermore, when contracting is desir-able, the model can be used to predict thepower of the incentives that will be offered.Finally, in some circumstances, arm’s lengthtransactions (� = 1) will be chosen.

This moral-hazard model of vertical inte-gration yields a number of testable predic-tions.19 First, equation (8) implies that the

probability of vertical integration should belower when the retailer’s effort is more pro-ductive (i.e., the partial derivative of the right-hand side of (8) with respect to βR isnegative). On the other hand, that probabilityis expected to be higher when the manufac-turer’s effort is more productive, when risk(� 2) and/or retailer risk aversion (r) is greater,and when the cost of contracting is higher.

To rephrase these predictions, when themarginal return to an individual’s effortbecomes larger, that individual should begiven higher powered incentives (a higherfraction of residual claims). In our context,this will mean more or less vertical integrationdepending on whose effort we are consider-ing. However, when risk or agent risk aversionincreases, insurance considerations becomemore important and the agent—who is theonly risk averse party in the model—shouldbe given lower powered incentives, whichimplies that we should see a greater tendencytoward vertical integration in the data.20

Finally, if contracting were costless (T = 0),vertical integration would be a limit case(�∗ = 0) that was rarely observed.

Now consider adding another variable, x,to the production function. For example, xmight be the size of the retail outlet, thesize of the market in which it is located, orany other characteristic of the principal, theagent, the outlet, or the market. Whetheror not x makes a difference to the vertical-integration decision depends on how itenters equation (2). The following is a fairlygeneral formulation,

(9) q = β0 + (βM + βMxx)aM +(βR + βRxx)aR + (γ + u)x.


19 Except for the effect of T, the same predictions canbe obtained from (3) or from (8). This means that we cansimultaneously discuss the power of an agent’s incentivesin, for example, a revenue-sharing contract and the choicebetween interacting in a firm or a market.

20 But see Lafontaine and Bhattacharyya (1995) andCanice Prendergast (2002) who point out that an agent’soptimal use of private information may lead to a positiveassociation between observed risk and agent incentives.Also see Daniel A. Ackerberg and Maristella Botticini(2002) for an argument that less risk averse agents may beattracted to riskier contracts, thereby negating the expect-ed correlation between risk and incentives in our simplemodel.


Under our assumptions above, which arefairly typical in this literature, γ has no effecton agent incentives or the decision to inte-grate.21 Indeed, unless x affects the margin-al return to the effort of the principal oragent in the model, or unless it affects risk, itis irrelevant. This means that x must interactwith aM, aR, or u to become relevant.Furthermore, the comparative statics for theinteraction terms are the same as those dis-cussed above. Specifically, if x, for example,increases the marginal return to the agent’seffort while leaving the return to the princi-pal’s effort unchanged, then increases in xwill be associated with a greater tendency touse markets rather than firms.

The model that we have developed can beextended to accommodate many complica-tions. For example, we have considered onlyone agent and can thus say nothing aboutteam production. Yet the notion of team pro-duction, which occurs when individualsworking together are more productive thanwhen each works alone, is central to themoral-hazard model of the firm (see, e.g.,Armen A. Alchian and Harold Demsetz1972 and Holmstrom 1982). Team produc-tion is a technological characteristic of theproduction process—similar to thoseemphasized in neoclassical theories of verti-cal integration—whereby individual margin-al products are enhanced by the efforts ofothers. This, in turn, makes interaction with-in a firm more desirable. Unfortunately, italso makes the allocation of rewards moredifficult. Indeed, if wages are based on mar-ginal productivities, a shirking worker canlower the wages of everyone in the team.

In our simple model, team productioncould be introduced by adding a secondagent and allowing agent efforts to interactin equation (2). In other words, the margin-al product of each agent’s effort coulddepend on the effort of the other one.

Unfortunately, the departure from linearitycaused by allowing efforts to interact wouldmake the solution of the model more diffi-cult. A similar difficulty would arise if weallowed synergies between the efforts ofprincipal and agent.

In addition, we have not allowed for thepossibility that the principal could monitorthe agent’s activities at a cost. Whether ornot costly monitoring leads to more integra-tion depends on the type of information thatthe principal can gather via this monitoring(see Lafontaine and Slade 1996).

Finally, in our simple model, the agentperforms only one task. A multitask model ismuch richer, but comparative-static deriva-tives can be signed only in special cases (seeHolmstrom and Paul Milgrom 1991, 1994).In other words, designing reward systemsthat provide high-powered incentives formultiple tasks is quite difficult. This difficul-ty can lead to more vertical integration,where internally firms rely on subjectiveperformance evaluations instead of explicittask specific incentives (see Holmstrom1999 and Pierre Azoulay 2004).

In spite of these limitations and its sim-plicity overall, the model embodied in(1)–(9) is useful as it yields the types of pre-dictions that are most often tested in theempirical moral-hazard literature.

2.1.2 Evidence on Predictions from Moral-Hazard Models

The theoretical moral-hazard modelabove identified a number of factors thatshould affect the vertical-integration deci-sion. Unfortunately, some of those factors donot easily lend themselves to empiricalassessment (e.g., the degree of risk aversion,r). In what follows, we limit attention to fac-tors that can be assessed more readily—theimportance of: risk (� 2), downstream effort(βR), upstream effort (βM), and outlet size (anx). In addition, we discuss factors that requireslight modifications to the basic model: mon-itoring difficulty, spillovers within a chain,and multitasking.


21 Note that this would still be true if γ x were replacedby an arbitrary function g(x).


We summarize the empirical evidence in aseries of tables that are organized by factor(e.g., risk). For each study, the relevant tableindicates the author’s name, the year of pub-lication, the industry studied, the data typeor empirical technique used, the way thefactor of interest is measured, and theauthor’s conclusion concerning the effect ofthat factor. These conclusions are summa-rized in the final column, where a + (−) indi-cates that the factor encourages(discourages) vertical integration, and a ∗

indicates that the finding is significant, usinga two-sided test and a 5 percent confidencelevel.22 Parentheses in the last column indi-cate that the variable that is examined is aninverse measure of the factor of interest andis therefore expected to have the oppositeeffect on vertical integration from a directmeasure. For example, “outlet density,”which is an inverse measure of monitoringcosts, is expected to have a sign that is oppo-site from “distance from headquarters,”which is a direct measure. Finally, given thatmeasurement has proved challenging inmuch of this literature, our discussion ofeach factor considers measurement issues aswell as empirical findings.

Risk

The standard agency model of retail con-tracting suggests that, as the level of uncer-tainty increases, so does the need for agentinsurance and thus the desirability of verticalintegration with the presumably less riskaverse upstream firm. In other words, thelower-powered incentives that are typicallyused inside the firm protect the agent fromthe vagaries of the market, a protection thatbecomes all the more valuable as uncertaintyrises.

The notion of uncertainty or risk that isrelevant in this context is the risk that is

borne by the agent, namely the risk at theoutlet or downstream level. Unfortunately,data that measure outlet risk are virtuallynonexistent. For this reason, imperfect prox-ies are employed. The two most common aremeasures of variation in detrended sales peroutlet and measures of failure rates such asthe fraction of outlets that were discontin-ued in a particular period of time.Furthermore, data are often available only atthe level of the sector rather than at the levelof the franchisor or the retail outlet.

Table 1 gives details of studies that assessthe role of risk in determining the tendencytowards integration. In all but two of thosestudies, contrary to prediction, increasedrisk is associated with less integration.Moreover, the two positive findings are notsignificant. These results suggest a robustpattern that is unsupportive of the basicagency model. Interestingly, allowing effortto interact with risk in the basic model onlymakes matters worse. In particular, if aR isinteracted with u in equation (2), higherpowered incentives become even more cost-ly, since, by increasing the agent’s effort, theyalso increase the risk that he must bear.

The finding that risk is negatively ratherthan positively associated with integration isa puzzle that, as noted earlier, has attractedsome attention already in the literature.Some authors have concluded from the evi-dence that franchisors shed risk onto fran-chisees (e.g., Robert E. Martin 1988). Thiswould be optimal, however, only if fran-chisors were more risk averse than fran-chisees. Unfortunately, if agents wereindeed less risk averse than their principals,there also would be less need to balance theprovision of incentives and insurance tothose agents. At the extreme, franchisorswould simply sell outlets to franchisees out-right for a fixed price, a situation that israrely observed.

Several alternative, and we believe moresatisfactory, explanations for the observednegative risk/integration relationship havesurfaced in the literature. The first stems


22 Note that while we follow the author in assessing theimportance and significance of different factors, and ininterpreting their findings more generally, we do notalways interpret their measures as they intended.


from the fact that market uncertainty can beendogenous and that the power of incentivescan influence sales variability. Indeed, retail-ers often have superior information concern-ing local-market conditions. Moreover, sinceseparation gives agents greater incentives toreact to those conditions, one is likely to findmore sales variability in separated than inintegrated units.23 A second possibility is thatdifferences in risk aversion, which typicallyare not controlled for in the empirical analy-ses due to the absence of data, can explainthe correlation. With this interpretation,more risk averse agents select safer marketsas well as contracts with lower-poweredincentives.24 Finally, we come back to theanomalous effect of risk on the extent of ver-tical integration below in our discussion ofevidence relating to property-rights theory,

as this theory provides yet another potentialexplanation for the empirical regularity thatappears in table 1.

Downstream Effort

The moral-hazard model predicts thatincreases in the importance of the retailer’sinput should be associated with less integra-tion and higher-powered incentive contracts.In other words, when the agent’s job is moreentrepreneurial in nature, his compensationshould reflect that fact.

From a practical point of view, proxiesfor the importance of the agent’s effort (orits inverse) have included measures oflabor intensity (either employee/sales orcapital/labor ratios) as the agent is the onewho must oversee the provision of labor.Researchers also have used a measure of theagent’s value added, or discretion over inputchoices, and a variable that captures whetherprevious experience in the business isrequired. In the context of banking, sincemanagers in rural settings must offer a more


23 See Lafontaine and Bhattacharrya (1995) andPrendergast (2002) for more on this.

24 See Ackerberg and Botticini (2002) for an explana-tion based on selection and some corroborating evidencein the context of sharecropping.

TABLE 1THE EFFECT OF RISK ON FORWARD INTEGRATION

Author Year Industry Data/Technique Variable Examined Effecton VI

Anderson and Schmittlein 1984 Electronic components Cross section; Logit % forecast error, −and sales sales

John and Weitz 1988 Industrial goods Cross section; Environmental +and sales OLS, Logit uncertainty index

Martin 1988 Retail and services Panel; Weighted Dispersion in −∗Least Sq. detrended sales

Norton 1988 Restaurants and motels Cross section; Dispersion inOLS, 2SLS detrended sales

-Restaurants +-Motels −-Refreshment Places −∗

Lafontaine 1992 Retail and services Cross section; Tobit Proportion of outlets −∗discontinued

Lafontaine and 1995 Retail and services Descriptive Sales dispersion −Bhattacharyya Rate of outlet −

discontinuationWoodruff 2002 Footwear and sales Cross section; Frequent fashion −∗

Regressions change

∗ denotes significance at 5 percent using a two-tailed test.


complete set of services, locational dummieshave been used to capture levels of responsi-bility. Finally, two studies of gasoline retail-ing rely on a dummy variable thatdistinguishes full from self service.

Table 2 summarizes the results from stud-ies that assess the effect of the importanceof agent effort. In every case where thecoefficient of the agent-importance variableis statistically significant, its relationshipwith integration with the upstream compa-ny is negative, as predicted by standardagency considerations and other incentive-based arguments. In other words, when theagent’s effort plays a more significant role indetermining sales, integration is less likely.

Upstream Effort

It is common for MH models to be basedon the assumption that only one party, the

agent, provides effort in the production (orsales-generation) process. Our modelabove incorporates the possibility that theprincipal also provides some effortbecause, in reality, success at the retaillevel often depends importantly on thebehavior of the upstream firm or principal.For example, franchisees expect their fran-chisors to maintain the value of the trade-name under which they operate (viaadvertising and other forms of promotion),and to screen and police other franchiseesin the chain as well as managers of corpo-rate stores. If this behavior is not easilyassessed, there is moral hazard on bothsides—up and downstream—and the fran-chisor, like the franchisee, must be givenincentives to perform. Not surprisingly,when the effort of the principal increases inimportance, it is the share of output that she


TABLE 2THE EFFECT OF THE IMPORTANCE OF AGENT EFFORT ON VERTICAL INTEGRATION

Author Year Industry Data/Technique VariableExamined Effect on VI

Caves and 1976 Retail and services Cross section; Personalized service −∗Murphy Regressions dummy Norton 1988 Restaurants and Cross section; Employee to sales ratio

motels OLS, 2SLS -Restaurants −∗-Motels +-Refreshment places −∗

Lafontaine 1992 Retail and services Cross section; Tobit Sales minus franchisor −inputsFranchisee experience +required

Shepard 1993 Gasoline refining Cross section; Full service dummy −and sales Regressions

Scott 1995 Retail and services Cross section; Capital to labor ratio (+∗)Regressions

Maness 1996 Various chains Descriptive Control over costs −Slade 1996 Gasoline refining Cross section; Probit Full service dummy −∗

and salesWoodruff 2002 Footwear and sales Cross section; Probit Frequent fashion change −∗

Brickley, Linck, 2003 Banks and offices Cross section; Logit Rural location −∗and Smith

∗ denotes significance at 5 percent using a two-tailed test. Parentheses in the last column indicate that the vari-able examined is an inverse measure of the construct and is therefore expected to have the opposite effect onthe extent of vertical integration.


receives, or the extent of vertical integration,that must rise.

Table 3 shows results from studies thathave considered how the importance of thefranchisor’s effort affects the probability ofintegration. The importance of upstreameffort is measured by the value of the trade-name (proxied by the amount of advertising,the number of outlets in the chain, or thedifference between the market and the bookvalue of equity), the amount of training pro-vided by the franchisor, or the number ofyears spent developing the business formatprior to franchising. The table shows that, inall cases, when franchisor inputs are moreimportant, more vertical integration isobserved, as predicted.

One proxy for the importance of the fran-chisor’s input that has been used in the liter-ature but is not included in table 3 is thechain’s number of years of franchising (orbusiness experience). The idea is that moreyears in franchising (or business) lead to abetter known and, thus, more valuable

tradename. However, that variable is also aproxy for the extent to which franchisorshave access to capital as well as for learningand reputation effects. Furthermore, cross-sectional evidence relating to this variable isaffected by the adjustment process all fran-chisors go through as they first begin toexpand the franchised side of their business.Using panel data at the franchisor level,Lafontaine and Shaw (2005) show that, afterthe first few years in franchising, the propor-tion of corporate units within chains levelsoff —at levels that differ across chains—andthen remains stable. They conclude that afirm’s years in franchising is not a majordeterminant of the “stable” extent of verticalintegration in these chains.

Outlet Size

Modeling the effect of outlet size is lessstraightforward than modeling the previoustwo factors, and model predictions are moresensitive to specification as a consequence.In particular, in the context of equation (9),


TABLE 3THE EFFECT OF THE IMPORTANCE OF UPSTREAM EFFORT ON VERTICAL INTEGRATION

Author Year Industry Data/Technique Variable Examined Effect on VILafontaine 1992 Retail and services Cross section; Tobit Weeks of training +∗

Lagged chain size +∗Years before franchising +∗

Muris, Scheffman, 1992 Soft-drink bottling Descriptive National accounts +and SpillerMinkler and Park 1994 Retail and services Panel; Grouped Logit Market minus book value +∗

Thompson 1994 Retail and services Cross section; Years before franchising +∗Regressions

Scott 1995 Retail and services Cross section; Days of training +Regressions

Nickerson and 2003 Trucking services Cross section; Tobit Advertising expenditures +∗SilvermanPénard, Raynaud, 2003 Retail and services Panel; Tobit Years before franchising +∗and SaussierLafontaine 2005 Retail and services Panel; Tobit Advertising expenditures +∗and Shaw Advertising fee +∗

Years before franchising +∗



size is a characteristic of the outlet (i.e., an x)that can enter linearly or multiplicatively orin some other form. Unfortunately, we canachieve any prediction, as noted above, forsuch an x variable depending on how weincorporate it in the model. We quite pur-posely choose a specification whose predic-tions are consistent with the empiricalregularity that we present below. In particu-lar, we model size as interacting with risk (u),in addition to having its own direct effect onoutput through γ. This interaction with riskcaptures the idea that the franchisee hasmore at stake in a larger outlet—the marketis not riskier per se, but more capital is nowsubject to the same degree of risk. As wenoted earlier, were it not for this interactionwith u, x would have no effect on the optimalcontract. With this interaction, it is predict-ed to have the same effect as risk does. Thisvariant of the MH model thus predicts thatvertical integration becomes more likelywhen the size of the capital outlay increases.Furthermore, vertical integration in thiscontext has the added advantage that it sub-stitutes the principal’s capital for the agent’s.

Unlike the factors discussed earlier, theempirical measurement of outlet size is fair-ly straightforward. Common measures areaverage sales per outlet and the initial invest-ment required. Table 4 shows that, in all butone study, greater size leads to increasedcompany ownership or integration. In otherwords, people responsible for large outletstend to be company employees who receivelow-powered incentives, as predicted.

While our specification ensures that themodel and evidence agree, it is nonethelesspossible to argue for the opposite relationshipin an equally convincing manner. Indeed,when an outlet is large, the agent has moreresponsibility. For this reason, outlet size hasbeen interpreted as a measure of the impor-tance of the agent’s input in the literature.25

Not surprisingly then, it is often claimed thatan agency model should predict that anincrease in size will be associated with lessintegration and higher-powered incentives.The data, however, contradict that prediction.

Costly Monitoring

The idea that monitoring the agent’s effortcan be costly or difficult for the principal iscentral to the incentive-based contractingliterature. In fact, if monitoring were cost-less and effort were contractible, therewould be no need for incentive pay.

Given the centrality of the notion of cost-ly monitoring, it is somewhat surprising thatthere exists confusion in the literature con-cerning the effect of an increase in monitor-ing cost on the tendency toward verticalintegration. Indeed, one can find statementsthat imply that monitoring difficultiesshould, on the one hand, encourage and, onthe other hand, discourage integration.26

To reconcile those discrepancies,Lafontaine and Slade (1996) modify thestandard agency model to include the possi-bility that the principal can use not only out-come (i.e., sales) information to infersomething about the agent’s effort, but also adirect signal of effort. Furthermore, theprincipal is allowed to base the agent’s com-pensation on both signals. We consider twotypes of signals because, in most real-worldmanufacturer–retailer relationships, it ispossible to supervise the actions of a retailerdirectly by, for example, testing food quality,assessing the cleanliness of the unit, anddetermining work hours. This direct supervi-sion provides the manufacturer with infor-mation on retailer effort that supplementsthe information contained in sales data.


25 In terms of our agency model (9), this is equivalentto interacting size with aR. If it were interacted with aM,predictions would be reversed.

26 For example, consider the following statementsfrom the empirical literature: “The likelihood of integra-tion should increase with the difficulty of monitoring per-formance” (Erin Anderson and David C. Schmittlein1984, p. 388). “Franchised units (as opposed to verticalintegration) will be observed where the cost of monitoringis high.” (James A. Brickley and Frederick H. Dark 1987,p. 408, text in parentheses added).


To model this situation in the simplestpossible way, we follow Lafontaine and Slade(1996) and replace the effort–sales relation-ship in a basic agency model—one thatinvolves only franchisee moral hazard—withtwo functions to denote the fact that theprincipal receives two noisy signals of theagent’s effort, aR. In particular, the principalobserves retail sales, q, and a direct signal, e.We assume that the vector of signals is unbi-ased and normally distributed with covari-ance matrix ∑, where ∑ = [�i j].

The contract that the principal offers theagent is amended to include, in addition to

the fixed wage W, not only an outcome-based or sales commission rate, �1, but also abehavior-based commission rate, �2, thatrelates to the direct signal of effort. With thesimplest version of the model, the two sig-nals are uncorrelated (�i j = 0, i ≠ j). Underthat assumption, solution of the two first-order conditions yields

1(10) �∗i = ⎯⎯⎯⎯⎯⎯⎯ ,1 + r�i i + �i i /�j j

where r is again the agent’s coefficient ofabsolute risk aversion. This solution is to becompared to the optimal contract from the


TABLE 4THE EFFECT OF OUTLET SIZE ON VERTICAL INTEGRATION

Author Year Industry Data/Technique Variable Examined Effect on VI

Brickley and Dark 1987 Retail and services Cross section; Initial investment +∗Regressions by franchisee

Martin 1988 Retail and services Panel; Weighted LS Average sales +∗

Norton 1988 Restaurants and Cross section; Average salesmotels OLS, 2SLS -Restaurants −

-Motels −∗-Refreshment Places −∗

Brickley, Dark, 1991 Retail and services Cross section; Tobit Initial investment +∗and Weisbach by franchiseeLafontaine 1992 Retail and services Cross section; Tobit Initial investment +∗

by franchisee +∗Average sales

Ohanian 1994 Pulp and Cross section; Logit Paper capacity +∗paper mills and Tobit

Thompson 1994 Retail and services Panel; Regressions Initial investment +∗by franchisee

Scott 1995 Retail and services Cross section; Initial investment +Regressions by franchisee

Kehoe 1996 Hotels Cross section; Tobit, Number of rooms +∗Logit

Brickley 1999 Retail and services Cross section; Logit Initial investment +∗by franchisee

Lafontaine 2005 Retail and services Panel analyses; Tobit Employees per outlet +∗and ShawHortaçsu and 2007b Manufacturing Panel analyses Value of plant shipments +∗Syversona plants

∗ denotes significance at 5 percent using a two-tailed test.a Results relate to plants that are brought into vertical structures, whether they are upstream or downstream

components of this structure.


basic agency model, which yields�∗ = 1/(1 + r� 2). Equation (10) then showsthat with two signals of effort, the optimalcontract must be modified to account for therelative precisions of the signals. In otherwords, the compensation package places rel-atively more weight on the signal with thesmaller variance.

We are interested in the effect of increas-es in the two sorts of uncertainty on the sizeof �∗

1, since this is the incentive-based paythat appears in the data. It is straightforwardto show that increases in the precision ofsales data (1/�11) lead to a higher reliance onoutcome-based compensation (higher �∗

1),which corresponds to less vertical integra-tion. However, increases in the precision ofthe direct signal of effort (1/�22) lead to lessoutcome-based compensation (lower �∗

1) ormore vertical integration.

While the above model does not explicitlyinclude monitoring costs, it should be clearthat, when the cost of increasing the preci-sion of sales data as an indicator of effort islow, we should observe more reliance onsales data in the compensation scheme,which means less vertical integration. On theother hand, when the cost of behavior mon-itoring is low, the firm will perform more ofthat type of monitoring. A low �22 will leadthe firm to choose a lower �∗

1, whichamounts to more vertical integration.

To summarize, our comparative staticsshow that the effect of monitoring on thedegree of vertical integration depends on thetype of information garnered by the firm inthe process. If this information gives a betterdirect signal of effort, it reduces the need touse sales-based incentive contracting andincreases the likelihood of integration. If, onthe other hand, monitoring increases thevalue of sales data by increasing its precision,it makes integration less attractive.

Turning to the empirical evidence, thefirst part of table 5, under “OutcomeMonitoring,” shows results obtained in thesales-force compensation literature, wherethe focus has been on the usefulness of

observed sales data in assessing agent effort.In the first two studies, researchers askedmanagers to respond to statements such as“it is very difficult to measure equitably theresults of individual salespeople” or “teamsales are common.” Other measures of theusefulness of outcome measures of effortinclude the length of the selling cycle (on thebasis that a long lag between actions andmarket responses makes it difficult to attrib-ute output to effort), as well as a measure ofenvironmental uncertainty that captures theextent to which agents control sales out-comes. Using scores thus obtained as meas-ures of the cost of monitoring sales andinferring effort from it, researchers foundthat higher monitoring costs lead to morevertical integration, as predicted by ourmodel.

The second part of table 5, which islabeled “Behavior Monitoring,” containsempirical results that come mostly from thefranchising literature. Here, authors havefocused on the cost of direct monitoring ofbehavior, that is information that is used tosupplement data on sales outcomes.Frequently used measures of behavior-mon-itoring costs include some notion of geo-graphic dispersion or of distance frommonitoring headquarters. Those measuresare proxies for the cost of sending a compa-ny representative to visit the unit to obtaindata on cleanliness, product quality, etc.Other measures are inversely related tocosts. These include outlet density and, inthe case of trucking, the presence of onboardcomputers. The table shows that, regardlessof whether behavior-monitoring costs aremeasured directly or inversely, in all caseswhere coefficients are significant, highermonitoring costs lead to less vertical integra-tion.27 Again the evidence is consistent withthe model.

In sum, the two types of measures thatauthors have relied upon in the empirical


27 Recall that the inverse measures (i.e., of the ease ofmonitoring) should have the opposite sign.


literature have captured different types ofmonitoring costs: the fit of sales data toindividual effort versus direct monitoring ofbehavior that is a substitute for sales data.Taking this difference into account, the“contradictory” results obtained and claimsmade by researchers are in fact consistent

with each other as well as with standardincentive arguments.

Spillovers Within a Chain

One reason for the prevalence of chainsrather than single outlets in the retail andservice sectors is that there are externalities


TABLE 5THE EFFECT OF MONITORING COST ON VERTICAL INTEGRATION


Part 1 OutcomeMonitoring

Anderson and 1984 Electronic Cross section; Logit Index of +∗Schmittlein components measurement difficulties

and salesAnderson 1985 Electronic Cross section; Logit Index of +∗

components evaluation difficultiesand sales Importance of +∗

nonselling activitiesJohn and Weitz 1988 Industrial Goods Cross section; OLS, Length of selling cycle +∗

and sales Logit

Part 2 BehaviorMonitoring

Brickley and Dark 1987 Retail and services Cross section; Distance from headquarters −∗Regressions

Norton 1988 Restaurants Cross section; Rural population in state (%) −∗and motels OLS, 2SLS

Minkler 1990 Taco Bell Cross section; Logit, Distance from headquarters −∗restaurants Probit, Linear Prob. Outlet density (−)

Brickley, Dark, 1991 Retail and services Cross section; Tobit Outlet density (+∗)and WeisbachCarney and 1991 Retail and services Cross section; Outlet density (+∗)Gedajlovic DescriptiveLafontaine 1992 Retail and services Cross section; Tobit Number of states −∗

in which operatesScott 1995 Retail and services Cross section; Number of states −∗

Regressions in which operatesKehoe 1996 Hotels Cross section; Tobit, Number of same-chain (+∗)

Logit hotels in cityBaker and 2003 Shipping and Panel; First Dif., IV Presence of (+∗)Hubbard trucking on–board computerBaker and 2004 Trucking and Panel; First Dif., IV Adoption of (+∗)Hubbard trucks on–board computerBrickley, Linck, 2003 Banks and offices Cross section; Logit Rural location −∗and SmithLafontaine 2005 Retail and services Panel; Tobit Number of states −∗and Shaw in which operates

∗ denotes significance at 5 percent using a two-tailed test. Parentheses in the last column indicate that the variable examined is an inverse measure of the construct and is therefore expected to have the opposite effect onthe extent of vertical integration.


that are associated with the brand or chainname. Although such spillovers are meant tobe beneficial, they can also create problemsfor both up and downstream firms. Forexample, one form that a spillover can take isa brand-loyalty demand externality. Withthat sort of spillover, a low price at one outletin a chain increases demand not only at thatoutlet but also for other retailers in the samechain. Conversely, a high price at one outletcan cause customers to switch their businessto another chain rather than merely seek adifferent unit of the same chain. When thissort of externality is important, integrationbecomes more desirable. The reason is thatthe chain internalizes the spillover that isexternal to the individual unit.

Franchisee free riding can take a variety offorms. For example, franchisees can uselower quality inputs or not abide by variousrules—such as a requirement that bakedgoods be disposed of if not sold within a cer-tain time period—that are good for the chainas a whole but impose costs on individualfranchisees. Indeed, once an agent is givenhigh-powered incentives via a franchise con-tract, he can shirk and free ride on the valueof the tradename (see, e.g., Benjamin Klein1980, 1995 and Brickley and Dark 1987). Theproblem is that the cost of the agent’s effortto maintain the quality of the trademark isprivate, whereas the benefits of his activitiesaccrue, at least partially, to all members ofthe chain. In this case, the spillover worksthrough effort or product quality, not price.

Whether the externality works throughprice or effort, the free-riding problem isexacerbated in situations where consumersdo not impose sufficient discipline on retail-ers, namely in cases of nonrepeat business.The franchisor, unlike the franchisee, caninternalize spillovers that damage the trade-mark by operating units in transient-customerlocations, such as freeway exits, herself.

Table 6 summarizes the evidence fromstudies that have examined the effect of non-repeat business on the propensity to inte-grate. This table shows that the evidence on

this effect is mixed. One explanation for thelack of strong evidence that vertical integra-tion is used to overcome the free-ridingproblem is that franchisors can find othermethods of controlling retail behavior by, forexample, using approved-supplier require-ments and imposing minimum advertisingrequirements. The lack of “highway” effectin particular probably reflects the fact thatfranchisors often contract with very largecompanies to operate units along freeways.These large franchisees, in turn, have incen-tives to maintain quality to the extent thatthey also internalize spillovers among alltheir units.28

Multiple Tasks

In many retailing situations, the agent per-forms more than one task. For example, aservice-station operator might repair cars aswell as sell gasoline, a publican might offerfood services as well as beer, and a truckermight perform cargo-handling services aswell as drive a truck. Generally, when this isthe case, the optimal contract for one task(and thus the propensity to integrate)depends on the characteristics of the others(see Holmstrom and Milgrom 1991, 1994).

There are many possible variants of multi-task models. We discuss a very simple ver-sion that illustrates our point. Since thisversion abstracts from any moral-hazardissue on the part of the principal, we com-pare the solution here to that of the basicone-sided moral-hazard version of ourmodel, which as noted earlier, yields�∗ = 1/(1 + r� 2).

Suppose that there are two tasks and thatthe agent exerts effort, aRi

, on the ith task.Output on each task, qi, is a noisy signal ofeffort, aRi

. Suppose further that the signalsare unbiased, they have covariance matrix ∑,


28 See, e.g., Brickley (1999) and Arturs Kalnins andLafontaine (2004) for evidence that franchisors grant mul-tiple units within the same markets to the same fran-chisees. In both studies, authors argue that this is done atleast in part so franchisees internalize more of the demandexternalities.


and the agent’s cost of effort is (aRTaR) /2. As

before, the principal chooses the vector ofcommissions, �, to maximize the total sur-plus subject to the incentive constraints. Inthe symmetric case where �11 = �22 = � 2, thefirst-order conditions can be manipulated toyield

1(11) �∗i = ⎯⎯⎯⎯⎯⎯ , i = 1, 2.

1 + r(� 2 + �12)

If one compares (11) to the solution of thebasic model, it is clear that, when a secondtask is added, the propensity to integraterises (falls) if the associated risks are posi-tively (negatively) correlated. This occursfor pure insurance reasons. In other words,positive correlation means higher risk,whereas negative correlation is a source ofrisk diversification for the agent.

In this simple model, tasks are linked onlythrough covariation in uncertainty. Thereare, however, many other possible linkages.For example, the level of effort devoted toone task can affect the marginal cost of per-forming the other and, when prices areendogenous, nonzero cross-price elasticities

of demand for the outputs can link thereturns to effort.

A model that incorporates these threeeffects is developed in Slade (1996). Sheshows that, when an agent has full residual-claimancy rights on outcomes for a secondtask, the power of incentives for a first taskshould be lower when the tasks are morecomplementary. Intuitively, since the secondtask already has high-powered incentives, iftasks are substitutes and incentives are lowfor the first, the agent will spend most of histime on his own activity (working in the backcourt). Her empirical application to gasolineretailing supports the model’s prediction.Specifically, she finds that, when the secondactivity—the one for which the agent is a fullresidual claimant—is repairing cars, anactivity that is less complementary with sell-ing gasoline than is managing a conveniencestore, then vertical integration of the gaso-line-selling task is less likely.

Similarly, George P. Baker and Thomas N.Hubbard (2003) look at multitasking in for-hire trucking, where the two tasks are ship-ping and cargo handling. Since onboard


TABLE 6THE EFFECT OF NON-REPEAT BUSINESS ON VERTICAL INTEGRATION


Brickley 1987 Retail and services Cross section; Nonrepeat sector dummy +∗and Dark Regressions Highway dummy (outlet) −∗

Norton 1988 Restaurants Cross section; Household tripsand motels OLS, 2SLS in the state

-Restaurants +-Motels −∗-Refreshment Places +

Brickley, Dark, 1991 Retail and services Cross section; Tobit Nonrepeat industry dummy −and WeisbachMinkler 1990 Taco Bell Cross section; Logit, Highway dummy +

restaurants Probit, Linear Prob.Brickley 1999 Retail and services Cross section; Logit Nonrepeat industry dummy −

Index: how local are −your customers?



computers (OBC) in trucking facilitate coor-dination and lower the cost of multitasking,the comparative-static predictions theyderive work through costs rather than risk ordemand. They find that adoption of OBCresults in more integration, particularly insituations where multitasking is important.

For his part, Azoulay (2004) provides evi-dence that, in the pharmaceutical industry,firms outsource clinical trials that are dataintensive while they keep in house those tri-als that are also knowledge intensive. Heargues that firms thereby ensure that incen-tives on both data and knowledge produc-tion are balanced for those projects thatreally involve both tasks. For those projectsinvolving mostly data production, in con-trast, outsourcing provides appropriate high-powered incentives for that main task. Giventhat pharmaceutical firms rely on flat explic-it incentives, combined with some subjectiveperformance evaluations, to evaluate workerperformance internally, he interprets thisevidence as consistent with multitask agencymodels (see Holmstrom 1999).

Finally, our double-sided moral-hazardmodel above assumes that franchisees areresponsible for local service provision whilefranchisors manage the brand and its value.Instead of viewing principals as active in thegoodwill-production process, it is possible tomodel agents as those who put effort intoboth local service and brand value. In thiscontext, free-riding is a situation in whichthe principal would like her agents to engagein various activities to support the brand, orat least not reduce its value, whereas thoseagents, when paid residual claims, choose toput too much effort into increasing theirown profits.29 The results from Jack A.Nickerson and Brian S. Silverman (2003)and Lafontaine and Shaw (2005), describedin table 3, where higher brand values are

associated with more vertical integration,can thus be interpreted as supporting a mul-titasking view of incentive provision foragents. In particular, vertical integration,which corresponds to lower-powered incen-tives for local effort, is appropriate whenbrand protection is more important.Similarly, Clarissa A. Yeap (2006) finds evi-dence that more complex production activi-ties in restaurant chains, in particular onsitefood production and table service, are asso-ciated with more company ownership. Sheinterprets her results in terms of multitask-ing, arguing that the chains do not want torely on high-powered incentives for agentswhen, if some of their tasks are not tended toproperly, this can have a large detrimentalimpact on the chain as a whole.

Summary

Two central predictions of the moral-haz-ard model of forward integration have beenconfirmed by the empirical evidence. Theseare that as the importance of local or down-stream effort grows, integration becomesless likely, whereas as the importance ofcompanywide or upstream effort grows,integration becomes more likely, whereimportance is measured by the marginalproductivity of effort. Moreover, the ideathat monitoring the agent is costly is alsocentral to the moral-hazard model of con-tracting. Nevertheless, there has been someconfusion in the literature concerning theeffect of higher monitoring cost on verticalintegration. We showed that once one recog-nizes that there are two sorts of monitoringthat the principal can perform—outcomeand behavior monitoring—the evidenceagain is highly supportive of the agencymodel. On all these fronts, the moral-hazardmodel performs very well.

Model predictions concerning the effectsof other factors, such as outlet size, spillovers,and multitasking, are more sensitive to thespecification of how those factors enter theoutput/effort relationship. However, thereare reasonable model formulations that lead


29 Chong-en Bai and Zhigang Tao (2000) propose amultitask model in this spirit where outlet managers areresponsible for both local service provision and goodwillvalue.


to predictions that are supported by the evi-dence. Moreover, here again the evidenceleads one to conclude that aligning incentivesis a central concern.

One important prediction from basicagency models, however—that increased riskmakes integration more likely as insuranceconsiderations begin to dominate—is notsupported by the data. One possible explana-tion for the discrepancy between theory andevidence is that output variability is likely tobe endogenous in situations where agentshave private information about local marketconditions. However, a similar finding sur-faces in the sharecropping literature (seeDouglas W. Allen and Dean Lueck 1995 fora survey), a context where exogenous outputfluctuations are more apt to dominate. Allenand Lueck suggest that measurement costs,i.e., the possibility that tenants might try tounderreport output, may explain the anom-alous risk effect in sharecropping. Anotherpossible explanation mentioned above relieson selectivity in a situation of heterogeneousrisk preferences.30 We suggest a third possi-ble explanation below in our discussion ofevidence pertinent to the property-rightsmodel.

Finally, many of the predictions from theMH model have received significant atten-tion in the empirical literature, allowing us todraw conclusions from several studies anddifferent contexts. But although theory andevidence seem well integrated in this area,there remains a need for further detailedempirical analyses to test implications derivedfrom various extensions of the basic model insimilar and new institutional contexts.

2.2 Backward Integration into InputProduction

The empirical literature on backward inte-gration is concerned with a manufacturer’s

decision to integrate either partially or com-pletely with its suppliers of parts or equip-ment or, put differently, the decision tomake or buy its supplies. Most of this litera-ture has addressed predictions derived fromtransaction-cost economics (TCE) eventhough property-rights theories also haveaimed to explain when firms might integratebackward. As noted by Paul L. Joskow(2005) “the TCE framework has stimulatedmuch more empirical work than [. . .] themore recent property-rights literature. Thisis to the credit of the scholars who havedone theoretical work in the TCE traditionsince they have produced testable hypothe-ses and endeavored to provide guidance toempirical researchers regarding how tomeasure relevant attributes of transactionsaffecting market contracting and internalorganization.”

The large body of empirical research inthe area has found considerable support forthe notion, derived from TCE, that specificinvestments are economically and statistical-ly important when it comes to the decision toorganize the production of a given inputinternally or externally. It also has estab-lished that backward integration is morelikely for more complex inputs and when theenvironment within which the firms operateis more uncertain. In some cases, this sameevidence has been interpreted as providingsupport for property-rights models of verti-cal integration. Whinston (2003) has shown,however, that the property-rights approachgenerates a distinct set of predictions. Wediscuss this in some detail below.

In what follows, we review the empiricalliterature on the make-or-buy decision,organizing the evidence along the lines sug-gested by the theories. With this in mind,we first present the theoretical arguments,starting with transaction-cost economics,followed by the property-rights approach.Moreover, since transaction-cost argumentsare usually informal, our overview of that setof arguments is also informal. Also, like oursimple moral-hazard model above, the


30 Ackerberg and Botticini (2002) find that risk is asso-ciated with more sharing relative to fixed rent contractonce they control for endogenous matching in their data,a result that is consistent with predictions from basicagency theory.


property-rights model that we present is abare-bones or skeleton version of the theorythat it represents, and we do not attempt toportray the richness that this class of modelscan embody. Nevertheless, we believe thatthe simple skeleton that we discuss capturesthe main predictions needed to organize ourdiscussion of the evidence.

2.2.1 The Transaction-Cost Model

Transaction costs (TC) are the costs ofestablishing and administering business rela-tionships within and between firms or indi-viduals, including those costs associated withopportunistic behavior and haggling ex post.TC theories of firm boundaries can betraced back to Coase (1937), who focused onthe costs of transacting under different orga-nizational forms, particularly the costs ofwriting and enforcing contracts. Those theo-ries have been developed further, notably byOliver E. Williamson (1971, 1975, 1979,1985), Klein, Robert G. Crawford, andAlchian (1978), and others.

The fundamental insight of TCE concern-ing vertical integration is as follows. Partiesto a transaction often make investments thathave greater value inside than outside therelationship. In other words, the value of theassets in their intended use is higher thantheir value in alternative uses. Examplesinclude specialized tools that can only beused to produce the products of one manu-facturer, training that increases worker pro-ductivity exclusively in using those tools, andsupplier facilities that have been located inclose geographic proximity to purchasers.Specific investments give each party to arelationship a degree of monopoly ormonopsony power. Indeed, even when thereare many potential trading parties ex ante,when investments are specific, parties arelocked in ex post.

When specific assets are involved, partiescan write long-term contacts to protectthemselves and their assets. If such contractswere complete, specificity would not createproblems. The complete contract would

specify exactly what will occur and who willcontrol the assets under all possible contin-gencies. However, writing complete con-tracts is costly and not all contingencies canbe foreseen. Thus, real-world contracts arenormally incomplete. In that context, specif-ic investments generate quasi rents and eachof the parties to a contract has incentives toendeavor to capture those rents. This meansthat they are likely to haggle with one anoth-er, thereby increasing the costs of writingand administering the contract, as well asattempt to renegotiate the contract or, moregenerally, engage in opportunistic behaviorex post.31 These possibilities, which are theessence of the hold-up problem, clearly poseproblems for long-term contracting, andthose problems are exacerbated in volatileenvironments.

TC theories of firm boundaries usuallyassume that the hold-up problem is mitigat-ed inside the firm. However, they are oftensilent as to just how that mitigation occurs.Yet even inside firms, workers who havereceived specialized training can attempt tohold their employers up and vice versa.Moreover, employees also can engage ininfluence activities that are designed to cap-ture quasi rents. Nevertheless, it is probablytrue that, even if mitigation is not complete,the problem is lessened inside firms.Indeed, relative to markets, firms are moreclosely related to command economies.32

Transactions in which opportunistic behav-ior is known to cause large problems aretherefore more apt to occur in house.

In sum, when the problems that are asso-ciated with transaction costs are important,TC models suggest that firms will choosegovernance structures—including vertical


31 See Steven Tadelis (2002) for a model that empha-sizes the role of transaction complexity in affecting hag-gling and ex post adaptation.

32 See Scott E. Masten (1988) for a discussion of thedifferent legal rules that apply inside and outside the firmand how those differences explain the different capacitiesto mitigate hold up. Also see Baker, Gibbons, and Kevin J.Murphy (1999, 2002) on this topic.


integration or separation—to reduce thelikelihood and cost of haggling and exploita-tion. The theory provides a number of impli-cations concerning circumstances underwhich firms are likely to choose vertical inte-gration. Specifically, firms are expected torely on in-house production when transac-tions are complex, when they involve specif-ic investments, when those specific assetsare durable, when the quality of those assetsis difficult to verify, when the environment isuncertain, and when the quasi rents that aregenerated by a relationship are large.

2.2.2 The Property-Rights Model

Property-rights (PR) theories, which aremore recent and more formal than transac-tion-costs arguments, were developed bySanford J. Grossman and Oliver Hart (1986),Hart and John Moore (1990), Hart (1995)and others. Those theories emphasize howasset ownership can change investmentincentives. More specifically, they demon-strate how the allocation of property rights,which confer the rights to make decisionsconcerning the use of an asset when contin-gencies arise that were not foreseen or notspecified in a contract, changes ex anteinvestment incentives.

Unlike the TC literature, the PR literaturedoes not focus on ex post haggling, renegoti-ation, and opportunistic behavior. Instead, itstresses contractual incompleteness anddevelops formal models that show how expost bargaining affects ex ante investment innoncontractible assets. Nevertheless, sincePR theories deal with relationship-specificassets, incomplete contracts, and ex postbargaining,33 they are often thought to beclosely related to TC models. Whinston(2003), however, shows how the predictionsfrom the two classes of theories can be verydifferent. Indeed, unlike TC predictions,with the PR model the problems associatedwith specificity need not be mitigated by

bringing a transaction inside the firm. Infact, in the PR literature, vertical integrationcan exacerbate the problem by reducinginvestment incentives to levels that are evenlower than those provided by markets.Finally, again relative to the TC literature,PR models provide a more rigorous set ofpredictions concerning the determinants offirm boundaries. Unfortunately, those pre-dictions are also more fragile and thus moredifficult to take to data.

Interestingly, although the moral-hazardliterature is concerned with residual claims,whereas the property-rights literature is con-cerned with residual decision rights, theirpredictions concerning the effects of mar-ginal-productivity changes are frequentlysimilar, bearing in mind that MH modelsdeal with incentives to exert effort and theproductivity of those efforts, whereas PRmodels deal with incentives to invest in phys-ical assets or human capital and the produc-tivity of those investments. We demonstratethe similarities more formally below.

We follow Whinston’s (2003) modellingapproach to derive some comparative staticsfrom a very simple version of PR theory.However, we specialize his assumptions tothe case that most closely resembles Hart’s(1995) model.34 The model is concernedwith a manufacturer (M) who must decidewhether to buy an input from an independ-ent supplier (S) or to produce it herself (toacquire S ).

Assume that integration decisions areagreed upon at t = 0. At t = 1, (which wedenote ex ante) each party to the transactionmust make a noncontractible investment, ij,at a cost c(ij) = 1−2 (ij)2, j = S, M. Note that thisoccurs whether or not the manufacturer hasdecided to vertically integrate. In otherwords, under both integration and separa-tion, someone who is not the manufacturermakes the upstream investment decision.


33 PR theories can therefore also be traced back toCoase’s (1937) seminal contribution.

34 Specifically, we do not consider cross investments,and we assume that there is underinvestment under eithervertical structure.


One way to think about these investmentsunder integration then is that the upstreamdivision manager decides what equipment topurchase or how much effort is put intomaintaining the equipment.35

Finally, at t = 2 (which we denote ex postsince investments have been sunk), the play-ers bargain over the surplus generated bythose investments. Thus even though the expost bargaining game is efficient (e.g., theNash bargaining solution is often used), sinceinvestments are sunk, the outcome of bar-gaining is determined by relative bargainingstrengths that are often only loosely related torelative investment levels. Furthermore,investment shifts not only the frontier—thesize of the pie that is to be split —but also thethreat point—the outcome that occurs whenagreement cannot be reached. Players there-fore have incentives to behave strategicallyand use their investments to better their posi-tions in the ex post game. Underinvestment isthe most usual outcome of this process.36

In the ex post game, either a bargain isstruck, in which case the joint surplus isindependent of asset ownership, or agree-ment cannot be reached, in which case pay-offs differ depending on asset ownership.37

Suppose that the ex post surplus when bar-gaining is successful is given by

(12) π(i) = �0 + �MiM + �SiS,

where i = (iM, iS) is the vector of invest-ments. We assume that �M > 0 and �S > 0. Inother words, both types of investments areproductive.

First-best investments are those that max-imize W(i) = π (i) − c(iM) − c(iS), whichimplies that i∗

j∗ = �j. The first-best ex ante

surplus is then W∗∗ = W(i∗∗). This level ofsurplus, however, simply cannot be attained.The only options available to the manufac-turer are integration or separation, and nei-ther of these yields the first-best levels ofinvestment or surplus.

To see this, we focus on backward integra-tion (i.e., ownership of the upstream asset).Let A be an indicator of asset ownershipwith A = 1 denoting manufacturer owner-ship of iS (vertical integration) and A = 0denoting supplier ownership (nonintegra-tion or market transaction). Manufacturerand supplier disagreement payoffs (i.e., theirpayoffs in the next best alternative to tradingwith each other) are38

(13) wM(i | A) = (�0 + �M0iM)(1 − A) +(�1 + �M1iM)A

and

(14) wS(i | A) = (�0 + �S0iS)(1 − A) +(�1 + �S1iS)A,

where the subscripts 0 and 1 indicate that anoutcome is associated with supplier or man-ufacturer ownership of iS respectively (i.e.A = 0 or 1). Assume further that�M > �M1 > �M0 ≥ 0, and �S > �S0 > �S1 ≥ 0. Inother words, assets are most productivewhen an agreement is reached (i.e., they arespecific). However, when no agreement canbe struck, M’s asset is more productive when


35 For example, in the case of GM and Fisher body, theassumption amounts to the Fisher brothers still getting tomake investment decisions in say equipment used oreffort put into maintenance in the plant even though theplant and equipment now belong to GM.

36 For a formal treatment of the underinvestmentresult, see Paul A. Grout (1984). Note, however, thatoverinvestment can also occur, as, for example, wheninfluence activities are involved. Moreover, in very simplesituations, contracts that specify bargaining strengths canovercome the underinvestment problem. To illustrate, ifonly one party must invest, the problem is overcome by

giving all bargaining power to the investing party.However, when the situation is more complex, such aswhen both parties must invest, this simple solution doesnot apply.

37 The fact that ownership only matters when partiesdisagree is an assumption of the Hart (1995) model.However, in a different context, Baker, Gibbons, andMurphy (2002) derive this result endogenously.

38 Whinston allows wj, j = S,M, to depend on bothinvestments. We adopt instead the Hart (1995) assump-tion that disagreement payoffs depend only on own invest-ment as this simpler model is sufficient for our purposes.


she owns both assets (integration), while S ’sasset is more productive when he owns it(separation).

With the second-best situation, weassume, as is typical in this literature, that theNash bargaining solution with side paymentsis used in the bargaining game with (wM,wS)as the threat point. As is well known, this isequivalent to each player receiving his threatpayoff plus one half of the gains from trade.Since there are gains from trade, a bargainwill always be struck. Nevertheless, despitethe fact that players never receive theirthreats, a distortion in investment decisionsoccurs because each party’s objective func-tion assigns positive weight to his threat. Exante, players therefore choose their threatsstrategically, to better position themselves inthe ex post game.

Second-best investment for the manufac-turer is given by39

(15) i∗M(A) =

1−[�M + �M0(1 − A) + �M1A] < i∗M

∗.2

The solution for the supplier’s investmentis similar. Finally, the second-best ex antesurplus, W∗(A), is obtained by substitutingsecond-best investments into W(•). At timet = 0, the manufacturer decides whether tovertically integrate or not by comparingthese values.

To obtain a model of vertical integrationthat is also an estimating equation, we againappend an unobserved (by the econometri-cian) zero-mean random variable, �A, to thesecond-best surpluses. Vertical integrationwill be chosen if

(16) W∗(1) + �1 > W∗(0) + �0

or

(17) Δ = W∗(1) − W∗(0) > �0 − �1.

Finally, if �0 − �1 has cdf F(•), the probabilityof vertical integration is

(18) PROB[VI] = PROB[A = 1] = F(Δ).

One can use (18) and the related equa-tions above to derive some testable hypothe-ses. First, an increase in the marginal returnto the manufacturer’s (supplier’s) investmentin the joint surplus, �M(�S), makes backwardintegration more (less) likely.40

Second, consider an increase in a margin-al return to one party’s investment in his dis-agreement payoff. If that increase occursunder asset ownership A, it makes that formof asset ownership more likely. For example,if the party is the manufacturer, an increaseunder integration (i.e., in �M1) makes inte-gration more likely, but an increase undernonintegration (�M0) makes integration lesslikely.41 This occurs because such changes inmarginal returns make disagreement underA more profitable while leaving payoffs inother situations unchanged.

Third, if we add a variable x to either thejoint surplus (12) or to the disagreement pay-offs (13) and (14), it will not affect invest-ment decisions unless it affects somemarginal return to investment. Furthermore,when such a variable affects one of thosereturns, its comparative statics are the sameas those outlined above for the marginalreturn that it affects.

Note the similarities with the moral-hazardmodel. In both models, increases in the man-ufacturer’s marginal productivity of invest-ment make integration more likely,42 whereasincreases in the other party’s marginal pro-ductivity make integration less likely.


39 The manufacturer’s objective function is

wM(i|A) + 1/2[π(i) − wM(i|A) − wS(i|A)] − c(iM)= 1/2[π(i) + wM(i|A) − wS(i|A)] − c(iM).

Maximizing this objective by choice of iM yields theinvestment level defined by equation (15).

40 This result depends on the assumptions �M1 > �M0

and �S0 > �S1 (see Whinston 2003, p. 8).41 Whinston (2003) derives other comparative statics

for marginal-productivity increases.42 Unless the increase is to the productivity of invest-

ment under nonintegration.


Furthermore, with both models, addingexogenous variables such as characteristics ofthe parties, the product, or the market makesno difference unless those variables affectmarginal productivities.43 Finally, in bothmodels, some form of underinvestment—beit in terms of effort level or assets—occursunder both integration and separation. Ofcourse, there are also important differencesbetween the models. For example, sincethere is no risk in this version of the PRmodel, there are no predictions concerningchanges in risk or risk aversion.

In contrast, the implications of the PRmodel can be quite different from thosegenerated by TC analysis. Specifically, withthe PR model, changes that make the gapbetween π and wM or wS larger can be inter-preted as increases in quasi rents or speci-ficity. TC arguments predict that suchincreases will make integration more likely.PR theories imply instead that the outcomeof such changes in specificity will depend onthe source of the increase. For example, anincrease in the marginal productivity of theagent’s investment in the joint surplus (12),which implies an increase in quasi rent,makes integration less likely in the PRmodel.

2.2.3 Evidence on Predictions fromTransaction-Cost Models

Predictions from transaction-cost models,while rather informal, are still well under-stood. Indeed, asset specificity generates aflow of quasi rents that are associated with expost haggling and opportunism, whereascomplexity and uncertainty lead to contrac-tual incompleteness. Thus, vertical integra-tion is predicted to be more likely whenassets are specific, when transactions arecomplex, and when uncertainty is important.

Following Williamson (1983), it is com-mon to divide asset specificity into fourmain categories based on the source of thespecificity:

(1) Physical capital specificity stems frominvestments that involve tools or otherphysical assets that have higher value intheir intended use.

(2) Human capital specificity results whenindividuals undergo training or on-the-job learning that is more valuable insidethan outside a relationship.

(3) Dedicated assets are ones that would notbe acquired if a specific buyer were notintending to purchase a significant frac-tion of their product.

(4) Site specificity results from colocation.In other words, the flow of quasi rents isgenerated by savings in inventory andtransport costs.

A fifth type of asset specificity, namelytemporal specificity, has also received someattention in the literature (Masten, JamesW. Meehan Jr., and Edward A. Snyder1991; Stephen Craig Pirrong 1993;Williamson 1991). This type of specificityrefers to assets that must be used in a givenorder, or on a particular schedule, such thattheir unavailability at a point in time canhold up production.

We use these categories to organize ourdiscussion of the evidence on specificity, andthen present the evidence on other factorssuggested by the theory, namely complexityand uncertainty. As with the tests of moral-hazard models, the evidence is summarizedin a set of tables, one for each factor of inter-est.44 Although we focus on backward inte-gration, we include some studies of forwardintegration in the tables when those studiestest TC predictions.

In principle, the notion of asset specificityis fairly straightforward, but the measure-ment of such a concept, and of other factors


43 Or, in the case of the moral-hazard model, risk.

44 We focus, in the tables, mostly on papers publishedin economics. TCE-based research published in market-ing and in management or strategy journals is quite volu-minous, and though we discuss some of these studies, acomplete overview of this literature is beyond the scope ofthe present article. We also do not include the many arti-cles on the GM–Fisher Body case as these are mostlyqualitative.


influencing the make-or-buy decision accord-ing to theory, can be quite problematic.Indeed, publicly available data rarely containuseful information concerning such things asspecificity or complexity. For that reason,most studies rely on qualitative data, obtaineddirectly by the author(s) from inspection, or,more frequently, through interviews, ques-tionnaires, or postal surveys of firm man-agers. In addition, some measures areaverages of standardized variables that havebeen constructed from answers to question-naires. The measures also vary importantlyacross studies based on context. In our dis-cussion of the evidence, and in the tablesbelow, we discuss some of the measurementchallenges and describe the measuresauthors have relied upon.

Physical Capital Specificity

Authors have used several different meas-ures of physical specificity, usually tailoringtheir measure to the context. To illustrate, insome instances the measure is a dichoto-mous variable that equals one if a respon-dent thought that physical specificity wasimportant. In other cases, it is an index thatranges from 0 to n, depending on the degreeof physical specificity. For example, 0 mightcorrespond to “relatively standard” whereasn might denote “design specific.” In still

other studies, the measure represents a par-ticular feature of an input. For example, itcould be a dummy variable that equals one ifan input is a gas (which requires pipelinesand storage tanks).

Table 7, which summarizes the evidenceconcerning physical capital specificity,shows that its effect on vertical integration isalways positive and usually significant. Inother words, consistent with the predictionsof TC analysis, the presence of this sort ofspecificity makes integration more likely.

Human Capital Specificity

The most common measure of humancapital specificity involves some notion ofthe amount of training that is required toproduce or use an input. In some cases, it isa direct measure of training. In others, how-ever, it might be a measure of engineeringdesign cost, which is a proxy for the amountof technical know how that must beacquired. This means that the measure canalso be a proxy for complexity, and it is diffi-cult to disentangle the two effects. For thisreason, in some cases we present the samefinding in both tables.

Table 8 summarizes the evidence concern-ing human capital specificity. It shows that,with one exception, the effect on verticalintegration is positive and significant, which is


TABLE 7THE EFFECT OF PHYSICAL CAPITAL SPECIFICITY ON VERTICAL INTEGRATION


Masten 1984 Parts and aerospace Cross section; Probit Highly specialized dummy +∗

Masten, Meehan, 1989 Parts and Cross section; 1–10 scale of specificity +and Snyder automobiles RegressionsLieberman 1991 Inputs to Cross section; Logit Input is gas +∗

chemical productsMasten, Meehan, 1991 Naval shipbuilding Cross section; Index of specificity +and Snyder RegressionsLyons 1995 Inputs to Cross section; Logit Survey index of specificity +∗

engineering firms



evidence in favor of the TC model. It is inter-esting to note that the single negative effect,which is obtained by Christopher Woodruff(2002), involves forward integration intoretailing. The other studies involve eitherbackward integration with suppliers or for-ward integration of an industrial sales force.

Dedicated Assets

A few researchers have examined theeffect of asset dedication on vertical integra-tion, and the measures used, such as thedownstream firm’s share of purchases or adichotomous variable that equals one if onlyone firm buys the input, are relativelystraightforward. Table 9, which summarizesthis evidence, shows that when assets arededicated, vertical integration is more likely.These findings are also supportive of TCarguments.

Site Specificity

The importance of site specificity or colo-cation also has been assessed. Sometimes thespecificity measure is a qualitative scale vari-able constructed from answers to questionsconcerning the importance of proximity. In

other situations it is a 0/1 variable that equalsone if the two facilities are located close toone another (for example, if a plant that gen-erates electricity is located at the mouth of acoal mine that supplies its fuel).

The evidence concerning site specificity,which is summarized in table 10, is not veryconclusive. However, the only significanteffect on vertical integration is positive,which is consistent with TC predictions.

Temporal Specificity

The importance of temporal specificity, orthe need to integrate transactions that candelay other aspects of production, has alsobeen assessed. The measures used includean index that captures how important it isthat a given component be available onschedule, or indicators of how atypical afirm’s needs may be. The argument behindthese measures is that the firm is likely tohave difficulty finding alternative supplies atthe last minute, i.e. these input markets arethin and thus more subject to potential holdup by suppliers.

The evidence concerning temporal speci-ficity, as summarized in table 11, shows that


TABLE 8THE EFFECT OF HUMAN CAPITAL SPECIFICITY ON VERTICAL INTEGRATION


Monteverde 1982 Parts and Cross section; Probit Engineering design effort +∗and Teece automobilesAnderson and 1984 Electronic Cross section; Logit Index of specialized +∗Schmittlein components knowledge

and salesJohn and Weitz 1988 Industrial goods Cross section; Firm-specific training +∗

and distribution RegressionsMasten, Meehan, 1989 Parts and Cross section; 1–10 scale of know how +∗and Snyder automobiles RegressionsMasten, Meehan, 1991 Naval Cross section; Index of skill and +∗and Snyder shipbuilding Regressions knowledge specificityHanson 1995 Apparel Cross section; Degree of standardization +∗

OLS and TobitWoodruff 2002 Footwear and sales Cross section; Probit Frequent fashion change −∗



vertical integration is more likely when alter-native timely sources of supply are likely tobe rare, a result that is consistent with TCpredictions.

General Specificity

Some empirical tests are not designed toidentify the precise nature of specificity.

Instead they test for its presence in moregeneral terms. To illustrate, one study (AviWeiss 1992) assesses residual correlation ofshare-price returns, under the hypothesisthat, when specificity is important, shocks toone firm will affect the other in the samedirection. Another study (Federico Ciliberto2006) assesses how health maintenance


TABLE 9THE EFFECT OF DEDICATED ASSETS ON VERTICAL INTEGRATION


Monteverde 1982 Parts and Cross section; Probit Part specific to firm +∗and Teece automobilesLieberman 1991 Inputs to Cross section; Logit Firm share of purchases +

chemical productsAcemoglu, Aghion, 2005 Manufacturing Cross section; Plant share of purchases +∗Griffith, and plants Discrete ChoiceZilibotti


TABLE 10THE EFFECT OF SITE SPECIFICITY ON VERTICAL INTEGRATION


Masten 1984 Parts and aerospace Cross section; Probit Importance of colocation +Joskow 1985 Coal and electricity Descriptive Mine–mouth plant +∗

Masten, Meehan, 1989 Parts and Cross section; Importance of colocation −and Snyder automobiles Regressions


TABLE 11THE EFFECT OF TEMPORAL SPECIFICITY ON VERTICAL INTEGRATION


Masten, Meehan, 1991 Naval shipbuilding Cross section; Importance of on-time +∗and Snyder Regressions availability indexPirrong 1993 Bulk ocean Descriptive Market thinness potential +

shippingNickerson and 2003 Trucking and Cross section; Atypical haul +∗Silverman subcontractors Regressions weight measuresArruñada, 2004 Trucking Cross section; Specialized freight dummy +∗González-Díaz, OLS, Logitand Fernández



organizations (HMOs), which tie physiciansto hospitals, affect integration decisions.Finally, Manuel González-Díaz, BenitoArruñada, and Alberto Fernández (2000)examine how the extent of subcontracting byfirms in the construction industry relates tothe specificity of their product offering,measured by some weighted sum of theother firms offering the same product in thesame market each period. Those studies, assummarized in table 12, also show that verti-cal integration is more likely when assets arespecific, as predicted.

Complexity

All forms of specificity are associated withquasi rents that can lead to disputes as eachparty to a transaction attempts to appropri-ate those rents. However, problems wouldnot occur if contracts were complete. Thenext two factors, asset complexity and trans-action uncertainty, exacerbate the problembecause they increase the difficulties thatare associated with writing complete con-tracts. Although it is the interaction of thetwo groups of factors—specificity and con-tractual incompleteness—that is important,the effect of each factor has usually beenconsidered on its own. We follow the litera-ture in this respect and present the effect ofeach factor separately here and in the tables.For reasons that will become clear later, wepostpone our discussion of the few cases

where authors have examined the interac-tion between specificity and complexity—oruncertainty—to the next section.

As with specificity, complexity measuresare often based on qualitative informationthat has been collected through interviewsor surveys. For example, respondents mightbe asked to rank the complexity of an inputon a scale from 1 to n, or some notion ofdesign cost or product heterogeneity mightbe constructed. Moreover, quantitativemeasures, such as R&D intensity and rene-gotiation frequency, have also been used.Table 13, which summarizes the evidenceconcerning complexity, shows that, with oneexception, its effect on vertical integration isboth positive and significant. The exception(Daron Acemoglu et al. 2005) finds that sup-plier R&D intensity in U.K. manufacturingis associated with less integration. As before,however, most of the evidence is supportiveof the importance of transaction costs.

Uncertainty

Uncertainty also increases the difficultiesthat are associated with, and lessens thedesirability of, writing complete contracts.There are many measures of uncertaintyand most can apply to either up or down-stream products or markets. For example,uncertainty can be proxied by the varianceof sales or of forecasting errors or by theinstability of shares in either market. It can


TABLE 12THE EFFECT OF GENERAL SPECIFICITY ON VERTICAL INTEGRATION


Weiss 1992 Many vertical Descriptive Residual correlation +∗mergers of returns

González-Díaz, 2000 Construction firms Panel; Regressions Index capturing how +∗Arruñada, and and subcontractors many firms offer sameFernández productCiliberto 2005 Physicians and Panel; Multinomial Percent of local patients +∗

hospitals logit in HMO



also be captured by an indicator of the fre-quency of specification or design changesfor inputs or outputs.

In table 14, which summarizes the evi-dence from studies of backward integrationthat have considered this factor,45 a U (D)indicates that the measure of uncertaintyapplies to the upstream or supplier (down-stream or buyer) market. This table showsthat, whenever the effect is significant, high-er uncertainty leads to more vertical integra-tion. Furthermore, this conclusion isindependent of the market in which theuncertainty occurs. The evidence is thereforeconsistent with TCE predictions.

Summary

The weight of the evidence is overwhelm-ing. Indeed, virtually all predictions fromtransaction-cost analysis appear to be borneout by the data. In particular, when the rela-tionship that is assessed involves backward

integration between a manufacturer andher suppliers, there are almost no statisti-cally significant results that contradict TCpredictions.

2.2.4 Evidence on Predictions fromProperty-Rights Models

In contrast to the abundance of work thatattempts to assess the validity of the predic-tions from moral-hazard and transaction-cost models, there are very few studies thatdeal directly with property-rights predic-tions. This is perhaps due to the fact that thePR models are newer and their predictionsare more fragile. Moreover, as noted earlier,many researchers make little distinctionbetween TC and PR models and interprettests of one as tests of both. Nevertheless, asWhinston (2003) and our discussion abovehave shown, the predictions from the twoclasses of models can be very different.

A quick reading of tables 7–14 might leadone to conclude that, at least when TC andPR model predictions do not agree, the evi-dence is not supportive of property-rights


45 We consider only backward integration here to dis-tinguish this table from table 1.

TABLE 13THE EFFECT OF COMPLEXITY ON VERTICAL INTEGRATION


Monteverde 1982 Parts and Cross section; Probit Engineering design effort +∗and Teece automobilesMasten 1984 Parts and aerospace Cross section; Probit Dummy based on +∗

firm’s classificationMasten, Meehan, 1991 Naval shipbuilding Cross section; Index of complexity U-shaped ∗and Snyder RegressionsWoodruff 2002 Footwear and sales Cross section; Probit Product heterogeneity +∗

Forbes and 2005 Major and regional Cross section; Logit Major hub +∗Lederman airlines Weather +∗

Acemoglu Aghion, 2005 Manufacturing Cross section; R&D intensity: Griffith, and plants Discrete Choice Upstream −∗Zilibotti Downstream +∗

Gil 2007 Movie distribution Cross section; Renegotiation frequency +∗Linear Prob. Model

Hortaçsu and 2007b Manufacturing Panel Complex inputs +∗Syverson plants (Upstream value added)



theories of vertical integration. Indeed, thelevel of specificity and quasi rents of any formappear to foster vertical integration. However,this need not be the correct conclusion todraw from the data.

First, the predictions from the PR and TCmodel agree in some cases, and there the evi-dence supports both. More importantly fromour perspective, there are two cases wherethe data contradict transaction cost argu-ments. The first (Acemoglu et al. 2005) findsthat, whereas downstream product complexi-ty leads to more integration, technologicalintensity upstream has the opposite effect, afinding that is very supportive of PR models.It is also interesting that the other significantevidence that is at odds with TC analysiscomes from a study of integration betweenmanufacturing and retailing (Woodruff 2002),which is the typical setting of tests of moral-hazard models. As we have shown, the pre-dictions from PR and MH models have muchin common. It might therefore be possible tolearn something about the empirical rele-vance of PR models from the regularities thatsurfaced in studies of manufacturer/retailer

or franchisor/franchisee relationships.In particular, table 2 shows that, in all

cases where the importance of the agent’seffort is a significant determinant of integra-tion, it leads to less, not more integration.With the MH model, effort is normally inter-preted as the marginal productivity of theagent’s work. However, the variables thathave been used to measure effort, such asthe need for personalized service or previousexperience dummies, can equally be viewedas measures of the marginal productivity ofthe agent’s human-capital investment.Moreover, as Nancy A. Lutz (1995) pointsout, a franchisee differs from a companymanager not only because he is a residualclaimant, but also because he owns thefuture profits of the outlet and the right tosell that outlet, at least as long as he satisfiesthe constraints that are imposed by the fran-chisor. Hence, a franchisee’s effort in makingthe business successful can be seen as aninvestment. To the extent that that invest-ment is specific, TCE implies that it will beassociated with more vertical integration.Thus, not only is the evidence in table 2


TABLE 14THE EFFECT OF UNCERTAINTY ON BACKWARD INTEGRATION


Walker and Weber 1984 Parts and Cross section; Index of volume and +∗automobiles Regressions specification uncertainty (U)a

Lieberman 1991 Inputs to chemical Cross section; Variance ofproducts Logit detrended sales:

Upstream +∗Downstream −Uncorrelatedc +∗

Hanson 1995 Apparel Cross section; Frequent style change (D)b +∗manufacturers OLS, Tobitand suppliers

González-Diaz, 2000 Construction firms Panel; Variation in number −Arruñada, and and subcontractors Regressions of workers (U)a

Fernández

∗ denotes significance at 5 percent using a two-tailed test.a (U) denotes upstream uncertainty. b (D) denotes downstream uncertainty. c Uncorrelated denotes upstream uncertainty that is uncorrelated with downstream sales.


supportive of the moral-hazard model, but also it supports property-rights vis-a-vis transaction-cost arguments (see, e.g.,Woodruff 2002 for more on this).

We can also learn something about therelevance of PR models from an examinationof the effect of risk on vertical integration.The empirical evidence that is summarizedin table 1 appears to contradict a fundamen-tal prediction of the moral-hazard model.Indeed, more downstream risk leads to less,not more, integration and therefore lessinsurance for the agent. This finding, whichis puzzling in the context of moral-hazardtheories, can be explained using property-rights arguments. Specifically, when down-stream risk increases, the agent’s ability to beflexible in the face of unforeseen contingen-cies becomes more important. For example,when faced with unpredictable fluctuationsin sales, the retailer needs to be more skill-ful at managing inventories and employees;or, when customer tastes become more fick-le and style changes more frequent, heneeds a better understanding of customerneeds. In other words, in the presence ofincreased uncertainty, his investmentsbecome more productive and PR modelspredict less integration as a consequence.46

Finally, a notion that is central to both TCand PR models but is more strongly empha-sized by PR theory is that, in the absence ofcontractual incompleteness, problemsshould not surface. For example, there is noreason to believe that simply because a partis used by only one firm (a dedicated asset),it must also be noncontractible. Given theimportance of the interaction between speci-ficity and contractual incompleteness in thetheory, it is surprising that it is rarely testeddirectly. Exceptions include Masten (1984),who assesses the interaction of specificityand complexity, and Anderson (1985) whoconsiders the interaction of specificity andenvironmental uncertainty. It is surprising

that the findings from the empirical TC liter-ature are so robust given that the effect ofeach factor is considered independently.Still, in those cases where interaction effectswere considered, authors found support forthe theories.

Summary

Although property-rights models havebeen around for two decades, empirical test-ing of predictions derived from those mod-els lags behind. Nevertheless, as arguedabove, we can glean some insights into thevalidity of PR theories through a reinterpre-tation of tests of MH and TC predictions. Inparticular, the evidence that comes fromsupplier–manufacturer relationships, whichis the typical setting of TC tests, is not verysupportive of PR arguments, at least whenthe two sets of predictions disagree.However, the evidence that comes frommanufacturer–retailer or franchisor–fran-chisee relationships is much more positive.Not only is it consistent with many PR pre-dictions, but also PR ideas provide insightsinto and suggest a solution to a puzzle thatsurfaces in the MH literature, namely thenegative relationship between risk at theretail level and vertical integration. Still,much further work is needed before the rel-ative lack of direct tests of PR predictionscan be adequately addressed, and, perhapsmore importantly, the potential for cross-fertilization among tests of different modelscan be fully realized.

2.3 Some Econometric Issues

We have thus far ignored econometricissues, most of which are not unique to thestudies that we summarize. Nevertheless,many of the problems that arise in the liter-ature that we survey are related to the dis-creteness of the choices that firms make. Wetherefore conclude this subsection with abrief discussion of some of those issues.

Many empirical studies that assess theincidence of vertical integration use transac-tion or outlet-level data. For example, one


46 This explanation relates also to Prendergast’s (2002)argument.


might have observations on franchisedchains, each of which has many retail outlets,and be interested in modeling whether anoutlet is operated by the franchisor or by aquasi-independent franchisee. In that case,the dependent variable in the estimatingequation is discrete. Methods of dealingwith discrete dependent variables are wellknown.47 There are, however, a number ofproblems that are apt to surface in discrete-choice studies of the sort that we have inmind, problems whose solutions are morecomplex than when the dependent variableis continuous.

First, there is the ubiquitous endogeneityproblem—this problem is endemic in empir-ical research in industrial organization and iscompounded by the absence of valid instru-ments. To illustrate, firm age and size are tosome extent the result of managerial deci-sions that can be based on an underlying fac-tor that can also lead them to integrate aparticular transaction. Similarly, in the studiesthat we discuss below, outlet characteristicsare included among the “exogenous” explana-tory variables that determine the method oftransacting between manufacturer and retail-er. However, when an upstream firm decidesto change the nature of its relationship with aretailer, she might decide to change some ofthe outlet’s characteristics and vice versa. Forexample, this is often the case with gasolineretailing—stations that are changed from fullto self service are often changed from inde-pendent dealer to company operation at thesame time. A simple method of overcomingthe endogeneity problem is to estimate a lin-ear probability model by two-stage leastsquares.48 Unfortunately, the linear-probabil-ity (LP) model has other undesirable fea-tures.49 Other solutions to the endogeneity

problem in the presence of limited depend-ent variables normally require strongassumptions (see, e.g., Wooldridge 2002, pp.472–77). Moreover, when the endogenousexplanatory variable is itself binary, furthercomplications arise. These issues make struc-tural estimation methods particularly appeal-ing. As we discuss in the next section,however, those methods suffer from severallimitations of their own, especially in the con-text of empirical studies of vertical relation-ships.

Second, errors in a cross section, the typeof data that one must often rely on in thistype of study, are apt to be heteroskedastic.For example, outlets can be of very differentsizes, which normally induces heteroskedas-ticity. OLS estimates in the presence of het-eroskedasticity are inefficient. With a probit,in contrast, they are inconsistent. Indeed,heteroskedasticity changes the functionalform for PROB(y = 1|x), which is no longernormal. As before, the simplest remedy is toestimate a linear probability model with acorrection for heteroskedasticity. However, ifthe true model is a probit, the LP estimateswill still be inconsistent.

Finally, the errors in a discrete-choicemodel are apt to be spatially correlated inthe sense that the off-diagonal entries in thevariance–covariance matrix at a point in timeare nonzero.50 For example, outlets that arelocated in the city center might experiencecommon shocks that are not experienced byones that are located in the suburbs; or out-lets that sell brands of a common manufac-turer might have common privateinformation. One possible remedy is to usethe correction for spatial and time-seriescorrelation of an unknown form that isdeveloped in Joris Pinkse, Slade, and LihongShen (2006).51


47 For example, see Jeffrey M. Wooldridge (2002),chapter 15.

48 This method is simple provided that valid instrumentscan be found. However, the problem of finding instrumentsis just as acute here as in the continuous-choice situation.

49 For example, it is usually not possible to constrainPROB(y = 1|x) to lie between 0 and 1.

50 We use the term spatial to denote either geographicor characteristic space.

51 This correction is similar to the one developed byWhitney K. Newey and Kenneth D. West (1987) in atime-series context. Note that the spatial procedure alsocorrects for heteroskedasticity.


Many of the studies that we present suf-fer from one or more of these types of prob-lems, as well as major measurementproblems as described above, and theextent to which authors have tried toaddress these varies importantly acrossstudies. In the end, however, we believethat a preponderance of evidence, garneredacross numerous studies using differentapproaches in various institutional andindustry contexts, is most apt to yield con-vincing evidence on the validity of the theo-ries. We offer our summary of the evidencein that spirit.

3. The Consequences of Vertical Integration

Having considered factors that can leadfirms to integrate vertically, we now turn toan examination of the empirical evidenceconcerning the effects of vertical integrationon economic outcomes such as prices, prof-its, quantities, and costs. We do this becausesuch evidence can shed light on two majorquestions: first, can we identify the purport-ed benefits from vertical integration in afirm’s profits or its choices of prices andquantities? Second, if we can find evidenceof those benefits, can we identify the win-ners and losers? The first question is impor-tant because it can shed light on thecircumstances under which vertical integra-tion is apt to benefit the firms involved, andthus lead to better informed decision mak-ing. Furthermore, answers to both questionsare important because they can help us dis-tinguish cases where vertical integration ismore apt to benefit consumers from thosewhere it is more likely to be detrimental.Such an understanding is an important pre-condition for the design of sensible publicpolicy towards vertical mergers and otherforms of vertical-market transactions.

Why there should be any public-policydebate about vertical integration is unclearfrom our discussion so far. The motives forvertical integration that are associated withthe theories that we have presented

emphasize that, when firms choose verticalintegration, it is efficient for them to do so.Moreover, by highlighting the importance ofthe different efficiency motives, the empiri-cal evidence that we have reviewed suggeststhat vertical-merger policy should be deminimus if it exists at all. After all, both firmsand consumers can benefit when firms real-ize efficiencies. Yet in reality, attitudestowards vertical relations and mergers haveundergone important reversals in antitrust-policy circles, being sometimes restrictiveand other times permissive.

Consider, for example, the history of theU.S. Department of Justice’s (the DOJ’s)position towards vertical mergers. The firstDOJ merger guidelines, which were pub-lished in 1968, were relatively hostile towardvertical integration. Indeed, they viewed withsuspicion vertical mergers between firmsthat accounted for as little as 10 percent oftheir respective markets. The replacementguidelines, published in 1982 and 1984, incontrast, regarded nonhorizontal mergers tobe of interest for antitrust policy only if theyhad substantial horizontal consequences.Finally, the 1992 guidelines were renamed“Horizontal Merger Guidelines,” as verticalmergers were essentially forgotten.52

The existence of, at times stringent, con-trols on vertical mergers suggests that theremust exist motives for vertical integrationthat are not as innocuous as the efficiencyarguments we have discussed so far. Andindeed this is the case. In this section, webriefly discuss this alternative set of motivesfor vertical integration, all of which have todo with the creation and exploitation of mar-ket power. As should become clear shortly,however, not all motives for vertical integra-tion that are associated with the exercise ofmarket power imply that vertical mergersare necessarily detrimental to consumers. Infact, efficiencies can be generated when


52 See Robert Pitofsky (1997) on this, and Frederick R.Warren-Boulton (2003) for a more in-depth discussion ofthe history and interpretation of the DOJ guidelines.


firms integrate to, for example, eliminatedouble margins or input-choice distortions.The theories of vertical mergers thatantitrust authorities are most concernedwith in reality focus on the horizontalaspects of the merger, namely exclusion andcollusion. In other words, when a manufac-turer operates in an imperfectly competitivemarket, her interactions with her competi-tors—and in particular her capacity to col-lude with or exclude her rivals—can provideadditional motives for vertical integration orseparation.

Not surprisingly then, authors have lookedfor detrimental effects from vertical mergersmostly in concentrated markets (e.g., cableTV). As we will see below, however, eventhough authors typically choose marketswhere they expect to find evidence of exclu-sion, half of the studies find no sign of it.And where they find evidence of exclusionor foreclosure, they also at times documentefficiencies that arise from the same merger.Thus, although foreclosure may occur someof the time, the end result is not necessarilydetrimental to consumers. In fact, consistentwith the large set of efficiency motives forvertical mergers that we have described sofar, the evidence on the consequences ofvertical mergers suggests that consumersmostly benefit from mergers that firmsundertake voluntarily. On the other hand,divorcement requirements, which are sepa-ration requirements that are imposed bylocal authorities, often to protect local deal-ers, typically lead to higher prices and lowerservice levels for consumers. In other words,consumers are often worse off when govern-ments require vertical separation in marketswhere firms would have chosen otherwise.

We begin by presenting the various argu-ments for vertical mergers that have givenrise to public-policy concerns, which are themore traditional motives for vertical integra-tion that arise in contexts where firms havemarket power. Since there are many suchmotives and many variants of each model,we review the arguments only briefly and do

not present formal models. We then discusssome methodological issues that are particu-larly severe in this context and the methodsthat have been used to identify the effects ofinterest. Finally, we present the evidence onthe effects of vertical integration on eco-nomic outcomes, evidence that sheds lighton the motives behind vertical integrationand ultimately should inform public-policydecisions in the area of vertical mergers.

3.1 Market-Power Based Theories ofVertical Integration

3.1.1 Double Marginalization

Double marginalization occurs whenthere are successive stages of monopoly (oroligopoly) and the firms at each stage are notvertically integrated (Joseph J. Spengler1950; M. L. Greenhut and H. Ohta 1979).Unintegrated firms ignore the reduction inprofits that they inflict on other stages ofproduction when their prices increase,whereas vertically integrated firms capturethat externality. As a result, prices are lowerunder integration.53

The name double marginalization refersto the fact that monopoly profits are extract-ed at each stage of production (e.g., thereare multiple margins applied, each time toraise price above marginal cost).54 Undervertical integration, in contrast, there is asingle marginalization. Indeed, the verticallyintegrated monopolist maximizes the jointsurplus, up and downstream, as this maxi-mizes her profits. Furthermore, consumersalso are better off under integration in thiscase, as they pay less for the product thanunder successive markups.

This situation is analogous to one inwhich two goods are perfect complementsin downstream production or in ultimate


53 We are assuming a fixed-proportion technology. Seeour discussion below on how ambiguities arise under vari-able proportions.

54 As the number of marginalizations increases withoutbound, profits and sales go to zero.


consumption. With the latter situation, inte-gration between the two producers alsoresults in lower prices for consumers andhigher joint profits as the pricing externalityis internalized.

3.1.2 Variable Factor Proportions

When inputs are used in variable propor-tions, specifically, when they are substitutes,an upstream monopolist selling one of theinputs can have an incentive to integrateforward to prevent downstream firms fromsubstituting away from her product. In par-ticular, suppose firms in two upstreamindustries supply inputs to a competitivedownstream industry. If the industry thatsupplies x1 is monopolized, the monopolistwill set a price for x1 that exceeds marginalcost. Since the inputs are substitutes, down-stream firms will use too much of the com-petitively supplied input, x2, and too little ofthe monopolist’s input, relative to the situa-tion that would occur if both were sold atmarginal cost. In other words, the down-stream firms will substitute x2 for x1. Thisproduction inefficiency gives the upstreammonopolist a motive for acquiring the down-stream firms. Indeed, in so doing, x1 can betransferred internally at marginal cost, andthe inputs can be used in the correct proportions.

Superficially, it might seem that, in thevariable-proportions situation, vertical inte-gration makes everyone better off.However, that need not be the case. Theproblem is that the monopoly distortionpersists; it has simply been moved from upto downstream. The effect of integration onthe price of the final product is ambiguousand will depend on the parameters of theproblem.55

3.1.3 Foreclosure and Raising Rival’s Costs

Foreclosure occurs when practices areadopted that reduce buyers’ access to sup-pliers (upstream foreclosure) or sellersaccess to buyers (downstream foreclosure).Foreclosure is an important concept.Indeed, the main worry of antitrust author-ities when it comes to vertical relationshipsis the possibility that integration will fore-close entry by competitors at some level ofthe vertical chain, will cause competitors toexit, or will disadvantage them in somemanner. For example, a manufacturer whoacquires a large network involving mostretailers might prevent competitors fromgaining access to customers at reasonablecost, if at all. This in turn could prevententry of potential competitors upstream, orperhaps even lead rivals to exit theupstream industry.

Early theories of the problems associatedwith foreclosure were not based on rigorousmodels. Furthermore, Chicago School econ-omists (e.g., Wesley J. Liebeler 1968; RobertH. Bork 1969; and Samuel Peltzman 1969)argued that those theories were spurious andthat vertically integrated firms have noincentive to foreclose since they can achievethe same outcome whether or not they inte-grate. They thus concluded that verticalintegration can have no pernicious effect.

Later, economists began to model the ver-tical-merger/foreclosure issue in a strategicsetting (see, e.g., Michael A. Salinger 1988;Hart and Jean Tirole 1990; and Janusz A.Ordover, Garth Saloner, and Steven C. Salop1990). For example, Salinger (1988) showsthat the effect of a vertical merger on pricesin an industry with Cournot oligopolists ateach stage is ambiguous. Indeed, there aretwo opposing forces at work: first, a mergercan raise the costs of unintegrated down-stream firms, a factor that can cause retailprices to rise; and second, a merger caneliminate double marginalization that exist-ed in the preintegrated situation, a factorthat can cause retail prices to fall.


55 See, e.g., Richard Schmalensee (1973) and Warren-Boulton (1974). See also Klein and Kevin M. Murphy(1997) for a different type of variable proportion problemthat can give rise to vertical integration, namely one thatoperates through dealer services that are complementaryto the manufacturer’s product.


The Salinger (1988) model also demon-strates that vertical mergers can be benefi-cial to manufacturers even if the integratedmanufacturer does not refuse to sell or com-pletely foreclose access to facilities to unin-tegrated producers. In fact, it is oftenadvantageous to simply raise rivals’ costs.The incentive to raise the costs of uninte-grated downstream competitors is easy tosee. An increase in the wholesale price to adownstream competitor will cause that rival’sretail price to rise, which will lead some ofthe rival’s customers to switch to the inte-grated firm’s retail facilities. This point is thefocus of several papers on raising rivals’ costs(see, e.g., Salop and David T. Scheffman1987 and Thomas G. Krattenmaker andSalop 1986). In these models, in the absenceof double marginalization in the unintegrat-ed situation (e.g., if manufacturers use two-part tariffs), vertical mergers will result inincreased prices to consumers.

3.1.4 Strategic Delegation and Collusion

When an industry is oligopolistic, verticalseparation is often modeled as a two-stagegame in which contracts are written in thefirst stage (wholesale prices, w, and fixedfees, F, are set), and retail prices (p) are cho-sen in the second. This setup implies that, ifrival contracts can be observed, downstreamagents will condition their retail-price choic-es on those contracts. Under vertical inte-gration, in contrast, the product istransferred internally at transfer prices thatnormally cannot be observed by rivals. Theintegrated situation is therefore usuallymodeled as a one-shot game.

The idea that upstream firms can softenthe intensity with which they compete bydelegating the pricing decision to independ-ent retailers is by now well understood. Themodels in this case focus on interbrand com-petition across vertical structures. When ver-tical structures compete directly with eachother (i.e., when manufacturers set retailprices themselves as they do under verticalintegration), the resulting Nash-equilibrium

prices are lower than joint-profit-maximizingprices. However, when manufacturers sell toretailers who have some market power, ifmanufacturers delegate the pricing decisionto those retailers (as they typically do undervertical separation), the equilibrium pricesthat result are higher than under integration(see, e.g., Patrick Rey and Joseph Stiglitz1995). A softening of competition occursbecause prices are normally strategic com-plements (i.e., price reaction functions nor-mally slope up). An increase in amanufacturer’s wholesale price is thereforeassociated not only with higher own-dealerprices, but also with higher competitor retailprices. Furthermore, with two-part tariffs,equilibrium prices will not exceed monopolyprices.56

The above argument is premised on theassumption that retailers or distributors havemarket power. Spatial separation is one—butnot the only—factor that can lead to pricingpower. The argument also relies on theassumption of price competition at the retaillevel, which is apt to be valid in most con-texts. However, if downstream firms engagein quantity competition, delegation will notbenefit the vertical chain.57

It is straightforward to show that, underagent risk neutrality, delegation of the pric-ing decision (vertical separation) is a domi-nant strategy. However, as risk or riskaversion increases, the advantages of delega-tion fall. This occurs because a higher retailprice is accompanied by an increase in theproportion of the retailer’s income that isvariable, thereby increasing the risk that hemust bear. At some level of risk and/or riskaversion, the retailer’s need for compensa-tion for bearing increased risk makes verticalseparation unattractive, and the firm choos-es to vertically integrate instead. On theother hand, the more substitutable are the


56 In the absence of fixed fees, delegation can lead tohigher upstream profits but it is not guaranteed to do so.

57 This is true because quantities are strategic substi-tutes.


products of rival retailers, the more the firmsbenefit from delegation (separation), andthus the more likely it will be chosen. Overallthen, vertical integration will be preferredwhen there is substantial risk or risk aversionand products are not highly substitutable.58

3.1.5 Backward Integration by aMonopsonist

The above theories deal with forwardintegration by a monopolist. However, back-ward integration by a monopsonist can alsooccur, a situation that is considered byMartin K. Perry (1978). In his model, thevertical chain consists of an upstream com-petitive industry that produces an inputunder conditions of increasing marginal costdue to the presence of a fixed factor. Thecompetitive industry thus earns rent. Theinput is purchased by a monopsonist whowould like to capture the upstream rent.However, in order for there to be a motivefor backward integration, it must be the casethat the scarce input is worth more to themonopsonist than to the competitive firms.

This motive is similar to the incentive forforward integration in the variable-proportionscase. In particular, the monopsonist’s incen-tive for backward integration can stem fromthe desire to internalize the efficiency lossthat is due to under utilization of the inputwhose supply is upward sloping. There is,however, also a rent effect that enables themonopsonist to reduce the sum of rent pay-ments to independent suppliers plus thecosts of acquiring integrated suppliers.

With Perry’s model, full backward integra-tion eliminates efficiency losses due tomonopsony behavior and lowers prices toconsumers. With certain acquisition-costfunctions, however, the monopsonist will notchoose to integrate fully.3.1.6 Price Discrimination

The last imperfectly competitive motivewe discuss, price discrimination, can best be

explained with the use of a simple stylizedmodel. Consider an upstream monopolistthat supplies an input to two competitivedownstream industries with different elastic-ities of demand for the input.59 If arbitrage isnot possible, that is if the input cannot bepurchased in one downstream market andsold in the other, the monopolist will be ableto price discriminate. In this case, she willcharge a higher price to the industry withthe less elastic demand. However, if arbi-trage is possible, a single price will prevail,and the monopolist’s profit will be lowerthan in the no-arbitrage situation.

To remedy this problem, it suffices for themonopolist to acquire the buyers with moreelastic demands and to suppress that market,say market one. Since customers in marketone buy at a lower price, absent integration,arbitrageurs will purchase the input in thatmarket and sell it to the buyers in markettwo, who are willing to pay more. Verticalintegration suppresses the low-price market,which is the one that is causing the monopo-list’s problem, and enables her to engage insuccessful price discrimination.

As with most imperfectly competitivemotives, the outcome for ultimate con-sumers under the price-discriminationmotive is ambiguous. Indeed, relative to auniform price for the input, consumers inmarket one pay lower prices under integra-tion, whereas consumers in market two payhigher prices.

3.1.7 Summary

There are several general conclusions thatcan be drawn from our discussion of imper-fectly competitive motives. First, there arefew unambiguous results. Ambiguity in thetheories makes an analysis of the data evenmore important. Second, even when themotive for a merger stems from imperfectcompetition in horizontal markets, verticalmergers can be unambiguously beneficial.Such is the case when the merger motive is


58 For a formal model that embodies all of these fea-tures, see Lafontaine and Slade (2001). 59 See, e.g., J. R. Gould (1977).


to eliminate double marginalization undersuccessive stages of monopoly. Third, it isalways the link in the chain that has marketpower, whether it be monopoly or monop-sony power, that benefits from integration.Thus absent market power at some stage inthe chain, the above motives cannot be reliedupon to explain the data. Finally, in manycases theory suggests that firms with marketpower are able to obtain the same resultswith various forms of vertical restraintsrather than integration. For example, firmscan eliminate double marginalization withtwo-part tariffs, maximum resale prices, orquantity forcing. They can also either achieveforeclosure or address variable-proportionsissues via tying. To the extent that firms havethese alternative mechanisms at their dispos-al to address the issues that are raised by themodels, it is unclear whether researchers canexpect to find evidence that firms opt for ver-tical integration to solve the problems thatthose models describe. Put differently, therewould be no way to identify the conse-quences of vertical integration if nonintegrat-ed firms could achieve similar results withcontractual restraints. On the other hand, iffirms do use vertical mergers rather thancontracts to foreclose rivals or facilitate collu-sion, it is important to recognize that publicpolicy aimed only at preventing verticalmergers would prove ineffective as it wouldsimply lead firms towards those alternativemechanisms.

3.2 Some Methodological Issues

We have already mentioned some of theimportant econometric problems that authorsconfront when conducting research into thefactors that drive the vertical-integrationdecision. The same problems arise in studiesof the effects or consequences of vertical inte-gration. Furthermore, the issues are moreserious in this literature as vertical integrationdecisions are clearly endogenous in analysesof the consequences of such decisions. Thisproblem is compounded here as in otherareas of empirical IO by the difficulty of

obtaining valid instruments. These difficul-ties partly explain the appeal of structuralestimation methods in the study of firmbehavior. As we discuss below, however,those methods suffer from limitations oftheir own in the context of empirical studiesof vertical relationships.

The most straightforward way to evaluatethe effects of vertical mergers or divestituresis to present some persuasive descriptive sta-tistics. For example, one can compile infor-mation on retail prices before and after suchmergers. Descriptive statistics are useful inso far as they convince the reader that thereis an empirical regularity that needs to beexplained. The obvious problem, however, isthat there can be many explanations for thatregularity. For this reason, most researcherscombine descriptive statistics with someform of econometric analysis. In what fol-lows, we describe some of the main methodsthat have been used and their principal limitations.

3.2.1 Cross-Section, Time-Series, and PanelMethods

For the purpose of presentation, assumethat the data consist of a set of firms in agiven industry, where a subset of the firmsis vertically integrated (v) and another sub-set transacts with independent suppliers orretailers in a market (m). We are interestedin the consequences of that difference forsome measure of average performance y(e.g., profits, sales, prices, or costs). We callthe first set of observations the treatmentgroup and the second the control group. Inother words, we think of vertical integra-tion as a treatment that the firms undergo.This is a classic example of policy evalua-tion, and the techniques that we describeare used to assess many different policyissues.

Let Δ be the difference in average per-formance that we wish to measure. Our idealmeasure would be

(19) Δ∗ = yv,T − yvm,T,



where T is some time period, yv,T is the aver-age performance of v firms in that period,and yvm,T is the hypothetical average perform-ance that the treated observations wouldhave experienced had they not been treated(had they been m firms). Unfortunately, wecannot observe yvm,T and must use someproxy to measure Δ. Ultimately, the mostuseful data set will be a panel that includesboth cross-sectional and time-series variationin firm organization. In other words, v and mfirms can coexist at a point in time, and someof those firms may change their method oftransacting with suppliers or retailers overtime.

There are many panel-data methods forestimating treatment effects.60 We focus onone here—a difference-in-difference (DD)estimator. Suppose that in period T1 all firmsare m types. However, a subset of thosefirms (m1) undergoes an organizationalchange between periods (i.e., they become vfirms), whereas the remaining firms (m2)remain untreated. With a DD estimator, Δ isapproximated by

(20) ΔDD = (yv,T2− ym1,T1

) − (ym2,T2− ym2,T1

).

The first difference, yv,T2− ym1,T1

in equation(20) measures the change in the average per-formance of the treated firms, whereas thesecond difference, ym2,T2

− ym2,T1, measures the

change in average performance among theuntreated. Finally, the difference in differ-ence measures the relative change—theamount by which the performance changesdiffer across the two groups. Since the DDestimator removes both firm and time-periodfixed effects, it is common to attribute thefinal difference to the treatment. Of course,average performance is usually measuredconditional on a vector of explanatory vari-ables. With a DD estimator, however, onlyexplanatory variables that differ over bothfirms and time are relevant.

More generally, an advantage of paneldata is that it is possible to include both firmand time-period fixed effects in the estimat-ing equation. The firm dummies remove theinfluence of firm characteristics that aretime invariant, whereas the time dummiesremove the influence of factors that arecommon to all firms at a point in time.Furthermore, if a potential endogeneityproblem arises due to unobserved character-istics (common causal factors) that differ byfirm (over time) but not over time (by firm),the firm (time) fixed effects will purge theequation of that problem.

Unfortunately, in many of the empiricalcontributions discussed below, researchersdid not have access to panel data. They theneither exploited a time-series data set andperformed a before-and-after study, or moretimes than not, they used a cross-sectionaldata set to exploit the variation in organiza-tional form across firms. The obvious prob-lem with using only time-series variation isthat many other things can change betweentime periods in addition to the vertical struc-ture of the firms. Similarly, in purely cross-sectional data, the set of firms that arevertically integrated and those that are notare not random draws from an underlyingpopulation. Indeed, the method of organizingtransactions is usually an endogenous choice.In both time-series and cross-sectional data,the endogeneity problem can be partiallyovercome by including a vector of controlvariables (e.g., variables that measuredemand and supply conditions in a timeseries, and variables that measure firmcharacteristics in a cross section).61

Unfortunately, it is rarely possible to obtaindata on all relevant control variables.

In some cases, however, the organization-al choice comes from outside the verticalstructure, and thus can be considered exoge-nous. This might be the case, for example,


60 See, e.g., Wooldridge (2002), chapter 18, for a gen-eral discussion.

61 When control variables are included in time seriesand other models, the researcher compares conditionalmeans.


when certain vertical arrangements are pro-hibited by law (see our discussion of divorce-ment in subsection 3.3) and there istime-series or cross-sectional variation inlegal practice. Although the fact that thefirms did not choose their organizationalform in such cases mitigates the endogeneityproblem, it does not resolve the issue entire-ly. Indeed, the firms might be targeted bythe law exactly because the problem that thegovernment agency was trying to remedywas thought to be more acute for them.

3.2.2 Event Studies

In our discussion of the evaluation of theeffects of vertical-structure decisions so far,we have been concerned with the realizedconsequences of changes. One can also esti-mate market forecasts of the effects of thosechanges (e.g., mergers and divestitures) onfirm value. The tool that is commonly usedto perform such evaluation is an event study,which requires that the firms whose data areused in the analyses be publicly traded.

An event study is based on the assumptionthat stock markets are efficient and that shareprices reflect all currently available informa-tion at every point in time. In other words, itis assumed that the current stock price equalsthe expected value that accrues to the holderof the share—the expected discountedstream of capital gains and dividends—whereexpectations are formed efficiently andrationally. With efficient markets, when a“surprise” occurs, the associated change inthe share price is an estimate of the expectedvalue of the change in that flow.62

It is common to base an event study on theWilliam F. Sharpe (1963) market model thatrelates the return on asset i in period t, Rit,63

to the market return, RMt, where the market

return is the return on a broad portfolio oftraded assets,64

(21) Rit = �i + βiRMt + uit,i = 1, . . . , n, t = 1, . . . ,T.

When using the market model to assess anevent such as a merger, however, it is impor-tant that the event be a “surprise.”Unfortunately, it is often the case that newsof an impending merger leaks out prior tothe event. In addition, the market might notreact instantaneously to the news. For thisreason, it is common to focus on a windowthat surrounds the event (e.g., the merger).The goal of the analysis is to compare whatwould have occurred in the event windowhad the event not taken place to what actu-ally took place, in other words to assess theabnormal returns that are due to the merger.

With this in mind, let t = 0 denote theperiod in which the event occurs,65 so thatt < 0 (t > 0) denotes time before (after) theevent, and let t = t1, . . . , t2 < 0 be periodsbefore the event—periods in the estimationwindow. Periods t = t3 . . . , t4 ≥ 0, t3 > t2, arethen chosen to be in the event window.

The estimation proceeds as follows: first,the market model is estimated using obser-vations in the estimation window, i.e. fort = t1, . . . , t2 . The estimated equation is thenused to forecast returns inside the eventwindow, R̂it =�̂i +β̂iRMt,t = t3. . . , t4. Abnormalreturns are then calculated for observationsin the event window, where abnormalreturns are realized minus forecast returns,ARit = Rit − R̂it. One can plot and performstatistical tests on individual abnormalreturns. However, more commonly abnor-mal returns are summed over the observa-tions in the window to find the overall effectof the merger for each firm, and averaged


62 See A. Craig MacKinlay (1997) for a general dis-cussion of the use of event studies in economics andfinance.

63 The return on asset i in period t is the capital gainearned plus dividends issued between t − 1 and t dividedby the share price at t − 1.

64 The market model can be augmented to includeother financial and nonfinancial assets, as in the APTmodel of Stephen A. Ross (1976).

65 Note that t in this case does not represent calendartime, and t = 0 is often a different calendar date for eachevent.


across firms to find the average effect.Obviously, the choice of event window,which can be one or several periods, is cru-cial here. Indeed, if the window is too nar-row, the event can be missed, and if it is toobroad, averaging can remove the effect.Finally, standard errors of each estimate canbe calculated using well-known formulas.66

Clearly, positive (negative) abnormalreturns imply that the market values thenews as profitable (unprofitable) for a firmthat is involved in the merger. However,notice that there are three firms involved,the acquiring firm, the acquired firm, andthe merged firm, and it is possible for newsto be good for the acquired but bad for theacquiring firms, or vice versa.

Event studies can be used to evaluate theconsequences of vertical mergers. They canalso be used to distinguish between efficien-cy and anticompetitive motives for merg-ers.67 In the case of horizontal mergers, theprocedure is straightforward—a merger formarket power is good for rivals, whereas onefor efficiency is bad. One therefore looks atthe effect of the event on rival share prices(see B. Espen Eckbo 1983). However, withvertical mergers, things are more complex.Indeed, a vertical merger can harm down-stream rivals either because it lowers theintegrated firm’s costs (an efficient merger)or because it raises unintegrated costs due toforeclosure (an anticompetitive merger).68

One remedy is to look at share-price effectsfor buyers of the downstream product (seeJoseph C. Mullin and Wallace P. Mullin1997). However, in many contexts, this

effect can be far removed and is apt to bequite weak.

3.2.3 Computer Simulations and StructuralModels

The econometric methods that we havediscussed thus far involve estimatingreduced-form equations. In particular, thereis no way to recover the structural parame-ters that characterize tastes and technologyfrom such models. This is not a criticism initself, but it does mean that certain types ofanalyses cannot be performed. In particular,it is not possible to use reduced-form equa-tions for ex ante forecasts of the conse-quences of changes in policy.69 There aremany circumstances, however, both academ-ic and practical, in which it is desirable toassess the consequences of changes in verti-cal structures that have not yet occurred. Amerger simulation is a tool that could beused for that purpose, and this tool requiresa structural model.

The goal of a merger simulation is to pre-dict the equilibrium prices charged andquantities sold under the new, post-merger,market structure using only the informationavailable pre merger. Of course, the advan-tage of such an approach is that, if the simu-lation can forecast accurately, performing anex ante evaluation is much less wasteful thanwaiting for an ex post assessment. In partic-ular, the possibility of costly divestitures islessened by methods that accurately forecastmerger effects before the fact.

To illustrate the horizontal-merger tech-nique, consider the case of K firms that pro-duce n branded products with K ≤ n. Thebrands are assumed to be substitutes, butthe strength of substitutability can vary bybrand pair. It is standard to assume that thefirms are engaged in a static pricing game. Amarket structure in that game consists of apartition of the product space into K subsets,where each subset is controlled by one firm


66 See, e.g., John Y. Campbell, Andrew W. Lo, andMacKinlay (1997), chapter 4.

67 Since this technique assesses how the market eval-uates a particular vertical merger, it addresses the ques-tion of incidence only in the sense that, if abnormalreturns are positive, we can conclude that the managersmade the right decision when they chose to bring thissupplier or retailer within the firm. It does not, however,consider what characteristics of the transaction or firmsmade integration desirable.

68 On the other hand, foreclosure could benefit inte-grated and unintegrated upstream rivals

69 This is just another example of the Robert E. LucasJr. (1976) critique.


or decision maker. Specifically, each firm canchoose the prices of the products that are inits subset. A merger then involves combin-ing two or more of the subsets and allowingone player to choose the prices that wereformerly chosen by two or more players.

Consider a typical player’s choice. Whenthe price of product i increases, the demandfor brand j shifts out. If both brands areowned by the same firm, that firm will cap-ture the pricing externality. However, if theyare owned by different firms, the externalitywill be ignored. After a merger involvingsubstitute products, therefore, prices shouldincrease, or at least not fall. The questionthat horizontal-merger simulations aim toanswer is by how much. Clearly the answerdepends on the matrix of cross-price elastic-ities. Merger simulations have thereforefocused on modeling and estimatingdemand.

Whereas it is becoming increasingly com-mon to supplement traditional horizontal-merger analysis with a merger simulationalong the lines just described, this method ofevaluation is not yet common for verticalmergers. We are aware only of work byKenneth Hendricks and R. Preston McAfee(1999) and McAfee et al. (2001), who focuson homogeneous, intermediate-goods mar-kets in which both buyers and sellers havemarket power.70 In their work, mergers occurbetween firms (refiners) that are already par-tially vertically integrated. Their mergersthus have both horizontal and vertical com-ponents. Unfortunately, many of the firmsthat we are concerned with do not fit theassumptions of the McAfee et al. model. Inparticular, many produce differentiated retailproducts, not homogenous intermediategoods.

For a number of reasons, we are not opti-mistic that a “generic” vertical-mergermodel that could be used in a wide variety of

contexts can be designed. Some aspects ofthe problem are shared with horizontal-merger analysis and some are unique to ver-tical mergers. The former include thedifficulty of capturing changes in efficiency(cost-lowering effects) and coordinatedeffects (changes in the ability to collude).Cost savings that arise from a vertical mergerare particularly hard to handle in a simula-tion because they are often based on motiva-tional factors (i.e., better alignment ofincentives) rather than arising from techno-logical considerations. In addition, one mustconsider strategic interactions among hori-zontal rivals in a vertical context.71 Equationsfor rival brands can be included; however acomplete model would be very complex.Indeed, it would require assumptions con-cerning the horizontal games that are playedboth up and downstream as well as the bar-gaining games between members of the ver-tical structures. Moreover, in common withhorizontal-merger simulations, if theassumptions that underlie the simulationmodel are inaccurate, the forecasts will alsobe inaccurate. We therefore feel that,although this is a fruitful area for futureresearch, routine use of simulation methodsto assess vertical mergers is unlikely in thenear future. In fact, none of the evidence onthe effect of vertical mergers that we presentbelow is derived using a structural or simulation approach.

3.3 Evidence on the Consequences ofVertical Mergers

The research reported in tables 1–14 isdevoted to an assessment of the incidence ofvertical integration. In other words, the vari-able that is explained in most studies is ameasure of whether a transaction takes place(or has a tendency to take place in moreaggregate studies) inside a firm or in a mar-ket. The research that is reported in tables


70 See also John Asker (2004) and Randy Brenkers andFrank Verboven (2006), who adopt a structural approachto assessing the effect of vertical restraints.

71 It is common to ignore vertical considerations whenmodeling horizontal mergers, but that does not justify thepractice.


15–17, in contrast, assesses consequences.In other words, the latter group of studiesevaluates the effect of vertical integration onown or rival price, cost, profits, productofferings, survival, or some other economicvariable. The information reported for eachstudy is generally the same as that includedin the earlier tables. Notice however, thatsince the dependent variable pertains to oneof several outcomes that are not comparableacross studies, the penultimate column intables 15 and 17 indicates the outcome ofinterest. The final column, in this case, indi-cates the estimated effect of vertical integra-tion as interpreted by the author. In table 16,we add one more column at the end, tosummarize conclusions relative to consumerwell being.

3.3.1 Foreclosure and Raising Rival Costs

Competition authorities have focused mostattention on foreclosure and raising-rival-costmotives for mergers.72 It is therefore not sur-prising that empiricists have also devotedconsiderable attention to testing whethervertical mergers give rise to foreclosure.

The industries that have been examinedtend to be those that have received the mostscrutiny from authorities; for example,cement and concrete, cable TV program-ming and distribution, and oil refining anddistribution. In all these cases, the industries,which are natural monopolies or oligopolies,have little in common with the fast-food and


72 See Eric S. Rosengren and Meehan (1994) for a listof challenged mergers.

TABLE 15ASSESSMENT OF FORECLOSURE AND RAISING RIVALS COSTS

Author Year Industry Data/Technique Variable Examined Finding

Allen 1971 Cement and concrete Descriptive Acquisitions ForeclosureReiffen and Kleit 1990 Railroads and Descriptive Access to No foreclosure

terminals railroad terminalsRosengren and 1994 Challenged mergers Event study Returns, unintegrated No foreclosureMeehan downstream rivalsWaterman 1996 Cable TV Cross section; Program offerings Fewer rivaland Weiss programming Regressions programs carried

and distribution ForeclosureSnyder 1996 Crude oil Event study Returns, integrated Foreclosure

and refining rivalsMullin and Mullin 1997 Iron ore and steel Event study Returns, downstream No foreclosure

consumers Efficiency gainsFord and Jackson 1997 Cable TV Cross section; Subscription price Foreclosure

programming IV regressions Program cost Lower program costand distribution No welfare change

Chipty 2001 Cable TV Cross section; Program offerings, Fewer rivalprogramming and IV regressions price, and programs carrieddistribution subscriptions Foreclosure

Efficiency gainsoutweigh losses

Hastings and 2005 Gasoline refining Panel; Difference Wholesale price to ForeclosureGilbert and sales in difference unintegrated rivalsHortaçsu and 2007a Cement and concrete Panel; Difference Concrete price No foreclosureSyverson in difference, Concrete production Efficiency gains

Probit Plant survival



other franchise chains that have typicallybeen studied in the empirical moral-hazardliterature. In particular, the chances ofuncovering anticompetitive behavior is muchhigher with the former than with the latter.

Table 15 lists articles that test for foreclo-sure effects. In the table, we do not distin-guish between foreclosure and raisingrivals’ costs. Instead, we include studiesthat consider imperfectly competitiveindustries in which some firms are vertical-ly integrated and some are not and wherethe authors attempt to assess the conse-quences of that difference. Some of thestudies look for tendencies to exclude theproducts of unintegrated rivals (e.g., rivalprograms in the case of cable TV), othersassess whether unintegrated rivals payhigher prices for the upstream product(e.g., wholesale prices for gasoline), where-as still others evaluate stock-market reac-tions to vertical-merger announcements(e.g., changes in returns to holding sharesin either rival or downstream consumerfirms).

It is clear from the table that someauthors have uncovered evidence of fore-closure. However, the existence of foreclo-sure is, by itself, insufficient to concludethat vertical integration is pernicious.Indeed, recall that Salinger’s (1988) modelshows that there are two countervailing fac-tors associated with vertical mergers: anincrease in foreclosure or other practicesthat disadvantage rivals and a lessening ofdouble marginalization or other practicesthat are inefficient. One must thereforebalance the two.

Two of the papers in the table attempt toassess that trade-off (i.e., Mullin and Mullin1997 and Tasneem Chipty 2001), and bothconclude that efficiency gains outweighforeclosure costs. The evidence in favor ofanticompetitive foreclosure is therefore, atbest weak, particularly when one considersthat the industries studied were chosenbecause their vertical practices have beenthe subject of antitrust investigations.

3.3.2 Strategic Delegation

Next, we examine evidence concerningthe principal’s incentive to delegate the pric-ing decision (vertical separation) in a strate-gic setting. Recall that under retailer riskneutrality, principals prefer the separatedsituation. However, the strategic agencymodel of price competition predicts that ver-tical integration will gain advantage as risk orrisk aversion gains importance and as prod-ucts become less substitutable. One can testthose hypotheses individually but, to ourknowledge, this has not been done.Alternatively, a joint test can be constructedfrom the observation that integration is lessapt to occur when rival reaction functionsare steep, since the slope of the reactionfunction determines the strength of rivalresponse to own price increases. This sort oftest requires information about each unit’scompetitors. Slade (1998b), who has suchdata, finds that, in the context of retail-gaso-line sales, integration is indeed less likelywhen rival reaction functions are steep. Thisfinding is consistent with the idea that pricesshould be higher under separation, whichmeans that, although firms might preferthat arrangement, consumers will preferintegration.

3.3.3 Other Consequences

Table 16 summarizes the evidence fromseveral other articles devoted to the conse-quences of vertical integration. Thisresearch is more heterogeneous and moredifficult to put in neat pigeonholes. In par-ticular, the consequences of vertical integra-tion for price, cost, investment, profit, profitstability, stock ratings, and, for the work thatis based on the capital-asset pricing model,abnormal returns and systematic risk, haveall been the subject of investigation. We donot attempt to discuss each article in thetable. However, one can get a fairly goodidea of the bare bones of each study from thetable itself. As mentioned above, this tableagain includes information on the variable



that is assessed (denoted y in the table toindicate that it is the dependent variable),and the effect that integration has on y. Forexample, if y were cost, and if cost werefound to fall with vertical integration, the

penultimate column would contain a minus.If, in addition, the negative effect were signif-icant at 5 percent, using a two-tailed test, thatcolumn would contain a ∗. The final columnin table 16 shows the effect that integration


TABLE 16THE CONSEQUENCES OF VERTICAL INTEGRATION

Author Year Industry Data/Technique Variable Effect EffectExamined (y) on y on W

Shelton 1967 Restaurant Panel; Costs + +Description Profit +

Levin 1981 Crude oil Panel; Regressions Profit −∗ +and refining Stability of profit −

McBridea 1983 Cement and Regional panel; Delivered price −∗ +concrete Regressions

Spiller 1985 Various Cross section; Financial gains +∗ +Regressions Systematic risk −

Helfat and 1987 Various Paired samples; Systematic risk −∗ +Teece Difference in differenceAnderson 1988 Electronic Cross section; Index of opportunism −∗ +

component sales RegressionsKerkvliet 1991 Coal and electricity Panel; Regressions Cost efficiency +∗ +

Exercise of monopsony −∗power

Muris, 1992 Soft drinks Panel; Regressions Retail price −∗ +Scheffman, and bottlersand SpillerShepardb 1993 Gasoline refining Cross section; Regressions Retail price −∗ +

and salesFord and 1997 Cable TV Cross section; Regressions Program cost −∗ ?Jackson programming Price +∗

and distributionEdwards, 2000 Crude oil and Panel; Ordered probit Stock rating +∗ ?Jackson, and refining andThompson pipelinesCorts 2001 Film production Cross section; Tobit Release date clustering −∗ +

and distributionMullainathan 2001 Chemical Panel; Regressions Investment −∗ ?and Scharfstein responsivenessCiliberto 2005 Physicians and Panel; Regressions Investment in health +∗ +

hospitals care servicesJin and Leslie 2005 Restaurant chains Panel; Regressions Quality (health scores) +∗ +Gil 2006 Movie distribution Cross section; OLS, Movie run length +∗ +

Duration Analysis

∗ denotes significance at 5 percent using a two-tailed test.a Johnson and Parkman (1987) note that the introduction of time trends in the regressions renders effects

documented here insignificant.b results significant for unleaded sold full service only.


has on consumer well being (W). Whenmore than one consequence is examined(e.g., profit and profit variability), the effecton well being is determined by the combina-tion of the consequences. For this reason,we indicate the overall effect of vertical inte-gration rather than the effect on each conse-quence. Not surprisingly, some of the wellbeing results are ambiguous (denoted by ? inthe table). For example, if profits increaseafter integration, we cannot say if consumersare better or worse off, since the changecould be due either to higher prices or tolower costs.

To give an idea of the variety of the work,we discuss two somewhat arbitrarily chosenstudies. The first, which is by Anderson(1988) examines the effect of vertical inte-gration on opportunism. The setting is oneof industrial sales force, which can be eitherdirect (vertically integrated) or manufactur-ers’ reps (vertically separated). In order toget a measure of opportunism, Andersonasked sales managers to answer questionsthat reflected the behavior and attitudes oftheir sales forces. For example, one questionasked managers to rank the validity of thestatement that sales people distort informa-tion to the company in order to protect theirown interests. The “opportunism” variable,which was constructed as an index of theanswers to several such questions, was thenregressed against a vertical-integrationdummy, as well as variables that captureasset specificity, environmental uncertainty,and other relevant factors. Anderson foundthat integration significantly reduced oppor-tunistic behavior. In addition, she foundthat opportunism was positively related tospecificity and uncertainty.

The second study, which is by JoeKerkvliet (1991), involves mine-mouth elec-tric-generating plants. In other words, theplants studied are located in close proximityto coal mines, which is a classic example ofsite specificity (see, e.g., Joskow 1985).Kerkvliet estimated a neoclassical cost func-tion. However, instead of using market

prices of inputs, as would be common incompetitive environments, he allowed inputshadow prices to differ systematically frommarket prices. The factors that he consideredmight cause distortions, or wedges betweenthe two sets of prices, are regulatory vari-ables, monopsony power, and verticalarrangement (integration or separation). Hefound that integration led to increased alloca-tive and technical efficiency. Furthermore,although site specificity endowed all plantswith monopsony power, the tendency toexercise that power was significantly reducedby vertical integration.

Like the studies just discussed, the bodyof research that is reported in table 16 ishighly supportive of the efficiency of verticalintegration and mergers. In particular, thereare no minus signs in the final column of thetable, which indicates that integration bene-fits consumers, or at least does not harmthem. In addition, almost all of the positivefindings are statistically significant. Finally,as in table 15, many of the horizontal mar-kets examined (e.g., ready-mix concrete) arehighly concentrated. Since these are exactlythe type of markets where one might expectto find negative welfare effects from verticalmergers, it is particularly informative thatthe set of results in table 16 shows no suchnegative effects.

3.3.4 Divorcement

The mergers and divestitures whoseeffects we have considered so far were vol-untarily undertaken by the parties to thetransaction.73 Not all changes in verticalstructures, however, come from within theupstream/downstream relationship. Indeed,it is not uncommon for government agenciesto mandate structural changes, usuallydivestitures. This is most apt to occur whenthe agency believes that the vertical structureis exacerbating horizontal market power.


73 We do not mean that the table excludes hostiletakeovers. Instead, it excludes mergers or divestitures thathave been mandated by public authorities.


For example, gasoline stations can beowned and operated by the oil company (CCcontracts), owned by the company but oper-ated by the dealer (CD contracts), or ownedand operated by the dealer (DD contracts).In other words, transactions can occur with-in a vertically integrated firm (CC), in anarm’s length market (DD), or under an inter-mediate arrangement (CD), and there aremany efficiency considerations that motivatethe choice among those possibilities.Nevertheless, in a number of instances,competition authorities or regional govern-ing bodies have alleged that anticompetitivemotives outweigh efficiency considerationswhen firms make that choice. In particular, anumber of U.S. states have outlawed CCcontracts on the grounds that integrated oilcompanies would attempt to disadvantageunintegrated downstream competitors.

Table 17 contains details of five studiesthat relate to the issue of divorcement. Thethree studies that assess gasoline divorce-ment directly (John M. Barron and John R.Umbeck 1984; Michael G. Vita 2000; andAsher A. Blass and Dennis W. Carlton 2001)conclude that retail prices and costs werehigher and hours were shorter after itoccurred. In other words, they are unani-mous in concluding that the policy was mis-guided. The fourth study of the gasoline

market, Justine S. Hastings (2004), looks at aslightly different issue. She finds that,although retail prices are higher at verticallyintegrated stations than at unintegrated inde-pendents, there is no difference betweenprices at CC and CD stations. Given that therationale behind divorcement was that CCarrangements gave oil companies incentivesto charge higher wholesale prices to CD sta-tions, her finding is unsupportive of thatmotive.

Finally, the contracts that are writtenbetween brewers and publicans in the U.K.beer market are almost identical to thosebetween oil companies and service stations inthe United States. Moreover, those contractshave also been the subject of investigationsthat eventually led to divorcement. However,in that market, divorcement involvedchanges in ownership not mode of operation.In other words, CD contracts were changedto DD. Slade (1998a) finds that that changealso led to higher retail prices, probably as aresult of double marginalization.

The logic that led to divorcement regula-tions thus seems to have been flawed. In par-ticular, the forced move from CC to CDcontracts for gasoline appears to haveignored the fact that integrated oil compa-nies owned the affected stations and chosewhether to operate them under CC or CD


TABLE 17EMPIRICAL ASSESSMENT OF DIVORCEMENT

Author Year Industry Data/Technique Variable Examined Effect of Divorcement

Barron and 1984 Gasoline refining Panel; Difference Retail price Price higherUmbeck and sales in difference Station hours Hours shorterSlade 1998 Beer brewing Panel; Difference Retail price Price higher

and sales in differenceVita 2000 Gasoline refining Panel; Retail price Price higher

and sales Regressions, IVBlass and 2001 Gasoline refining Cross section; Retail cost Cost higherCarlton and sales ProbitHastings 2004 Gasoline refining Panel; Difference Retail price No difference

and sales in difference between CC and CD



arrangements. Having made a profit-maxi-mizing decision to operate some of theirowned stations internally and allow dealers tooperate the others under rental contracts(presumably based on efficiency considera-tions), it would be perverse for those compa-nies to turn around and attempt todisadvantage their affiliated CD retailers anddrive them out of the market. After all, the oilcompany could have chosen closure or selfoperation for those outlets in the first place.

The thinking that led to the move fromCD to DD contracts in the beer market, incontrast, appears to have ignored the factthat divestiture is associated with counter-vailing factors—the introduction of doublemarginalization and the elimination offoreclosure—and that the former costs canoutweigh the latter benefits.

3.3.5 Summary

The literature on the consequences of ver-tical integration appears to be much morefragmented than that on incidence. This isperhaps due to the fact that it is less wellintegrated with the theory. In particular,there is a need for a simple model that canencompass the various predictions that havebeen tested in the empirical literature.Instead, we have a set of theoretical models,each one concentrating on a single aspect ofthe problem. This is clearly an area wherefuture research could help us understandand better interpret the body of evidence.

In spite of the lack of unified theory, over-all a fairly clear empirical picture emerges.The data appear to be telling us that effi-ciency considerations overwhelm anticom-petitive motives in most contexts.Furthermore, even when we limit attentionto natural monopolies or tight oligopolies,the evidence of anticompetitive harm is notstrong.

4. Conclusions

In our attempt to organize what is now avery large empirical literature on the vertical

boundaries of the firm, we have covered a lotof ground. Of course, to keep the papermanageable, we have also made a number ofchoices. First, we have decided not to coverthe neoclassical approach to integration inwhich a firm is seen as a production func-tion—a set of feasible input/output relation-ships. With that class, integration ismotivated by technological considerations ofeconomies of scale and scope, including ver-tical economies. Although the empirical lit-erature in that area is vast, we chose to notcover it as we do not feel that technologicalfactors are especially complementary to theincentive-based motives that have been ourfocus.

Second, we have made no attempt toreview the large body of work that deals withvertical integration in the management liter-ature. Much of that work relies on TCEarguments, but significant portions also relyon alternative views, focusing on ideas likeorganizational capabilities or resources,where firms hire people or develop process-es that make them good at certain things butnot good at others. Considerations in thatvein imply that firms will integrate into areasthat are consistent with their capabilities andwill shy away from areas that are not. Weview those arguments as similar to the tradi-tional arguments described above and, inthat sense, outside the scope of this survey.

Third, we have not discussed models thatare too new to have produced a body ofempirical evidence. In particular, we havenot touched upon dynamic models such asthose that are based on the notion of rela-tional contracts (e.g., Baker, Gibbons, andMurphy 2002) or the notion of self-enforcingcontracts (Klein and Keith B. Leffler 1981;Clive Bull 1987; Klein and Murphy 1997).74

With those dynamic models, interaction,whether it be in a firm or a market, is modeled


74 Other models, such as the PR model of Grossmanand Hart (1986) and Birger Wernerfelt’s (1997) adjust-ment-cost model are dynamic in that they involve sequen-tial actions. However, they do not involve the threat ofpunishment to sustain cooperation.


as a repeated game with spot transactions(again in a firm or market) as the punishmentfor reneging. This is clearly a fruitful area forfuture empirical research that has remainedrelatively unexplored.75

Other important areas that are ripe forempirical assessment include the relation-ship between vertical integration and devel-opment, particularly the development of theinstitutions that make contracting a feasiblealternative (see, e.g., John McMillan andWoodruff 1999 and Acemoglu, SimonJohnson, and Todd Mitton 2007), and therelationship between vertical integration andtrade (see, e.g., Pol Antras 2003).

Fourth, we have tried to provide simpleversions of the theoretical models thatunderlie the material that we cover in orderto derive predictions around which we couldorganize the findings. Of course, by defini-tion, those simple models neglect manyimportant issues and extensions. For exam-ple, our moral-hazard model does not con-sider the rich set of tools that can be used toprovide incentives inside firms, the transac-tion-cost model glosses over the hold-upproblems that can occur within firms, andthe property-rights model does not explainwhy firms, rather than individuals, ownassets.76 A further limitation is that the theo-ries, at least as we have presented them, aremore applicable to the entrepreneurial firm.In particular, we have modeled the manufac-turer as both decisionmaker and assetowner. This means that we have not consid-ered the important conflicts that can occurbetween managers and shareholders in the

modern corporation—problems that resultfrom divorce of ownership and control.77 Itis therefore perhaps surprising that, as wehave seen, the models’ predictive powers areso high.

Finally, we have partitioned the theoriesthat we discuss into distinct groups, whichwe have called moral-hazard, transaction-cost, property-rights, and market-powerarguments, and we have used that partition-ing to organize the evidence. The partitionthat we chose, however, is somewhat arbi-trary, and it is often difficult to fit empiricalstudies into neat nonoverlapping classes. As aresult, there are a number of instances wherewe have included a single study in more thanone pigeonhole and, in other instances, wehave not included studies in pigeonholes thatthe authors might find appropriate. We usethe study by Baker and Hubbard (2003) to illustrate the difficulties involved in categorizing. In their setting—for-hire trucking—relationship specific assets are notparticularly important, whereas incentivesand job design are. In that sense, therefore,their study fits into the moral-hazard, particu-larly the multitasking paradigm. However, theissue at stake is residual decision rights—theability to determine asset use in contingenciesnot specified in contracts—and not residualclaimancy. In that sense, therefore, theirstudy fits into the property-rights paradigm.

The problem with any attempt to catego-rize the evidence is that the world is morecomplex than the simple models might leadone to believe. The advantage, however, isthat the possibilities for cross fertilization areabundant. To illustrate, although there arefew direct tests of the property-rights model,we have been able to gain insights into itspredictive power by considering evidencethat comes from the moral-hazard and trans-action-cost literature. Moreover, we havefound that, at least when TC and PR modelpredictions are at variance with one another,


75 Most of the empirical literature in this area takes theform of case studies, as in the many articles on theGM–Fisher Body case, Roy W. Kenney and Klein (1983)and the related literature on the Paramount case, whichbrought studio ownership of theaters under fire, andPatrick J. Kaufmann and Lafontaine (1994), onMcDonald’s, for example. But see Ricard Gil and WesleyR. Hartmann (2007) for an example of more quantitativeanalysis of the effect of reputation on vertical integrationdecisions.

76 This last point is made by Holmstrom (1999) whoprovides a good discussion of the relationship betweenMH and PR models.

77 The issue of divorce of ownership and control wasemphasized by Adolf Berle and Gardiner Means (1933).


there is little support for the PR theory in evi-dence that can be gleaned from make-or-buydecisions. However, there is much more sup-port for the PR model in evidence that can beobtained from manufacturer/retailer andfranchisor/franchisee integration decisions.The relationship among predictions and evi-dence from these two sets of models, and theopportunities for further cross fertilization,clearly deserves further thought.

We have emphasized—and perhapsoveremphasized—some of the differencesbetween TC and PR models. We did this tounderline the fact that one is not merely aformalization of the other, as is sometimesclaimed. Nevertheless, the similarities arestrong. Indeed, both are based on the ideathat relationship-specific assets and noncon-tractability lead to ex post problems whenquasi rents must be allocated.

The links between MH and PR modelshave received much less attention in the lit-erature. Yet the fact that their predictionsare similar should not be surprising. Indeed,although the first looks at the division of rev-enues and the second at the division of con-trol, if these two important aspects of thefirm are divorced from one another, newagency costs will arise. Similarities also stemfrom the fact that both are concerned withmechanisms that can alleviate the problemsassociated with double-sided moral-hazard,or more generally, double-sided incentiveissues.

We have discussed some of the economet-ric problems that are involved in identifyingthe effects of interest. We have also stressedthat many of the tests that are summarizedin the tables are incomplete in the sense thatthey look at one factor that is predicted toaffect vertical integration, holding the otherfactors constant, whereas it is often a combi-nation of factors that ought to be assessed.For example, asset specificity does not cre-ate problems on its own but only in conjunc-tion with noncontractibility. Yet only a fewstudies have considered such interactions.Moreover, as Sharon Novak and Scott Stern

(2003) note, vertical integration decisions forparts of a system—in their application sys-tems within cars—are apt to be related toone another rather than taken separately. Toour knowledge, theirs is the only study tohave considered such interactions.Empirical work that accounts for comple-mentarities among factors and decisions isan under exploited area that is ripe for futureresearch.

We have described measurement difficul-ties throughout, starting with the difficultyof measuring vertical integration itself as dis-tinct from various types of contractualarrangements. A related problem, that wehave thus far ignored, arises from the differ-ence between marginal and average. Theempirical studies are mostly concerned withaverages (e.g., the importance of the agent’seffort). Moreover, the bare-bones modelsthat we have presented make no distinctionbetween the two. However, in so far as thetwo differ in real-world contexts, the dataoften measure the wrong one.

One must bear these and other caveats inmind when evaluating the evidence.Nevertheless, the empirical regularities areboth consistent and strong. In other words,when we compare the evidence concerningthe effect of a particular factor on verticalintegration taken from studies of differentindustries, time periods, and geographicregions, we find that the sign of the effect isalmost always consistent across studies, atleast in cases where it is significant. The evi-dence is therefore stronger than one mightexpect given the difficulties involved. In fact,the degree of consistency suggests that per-haps some form of publication bias exists.Specifically, it might be easier to publishpapers that confirm theories. Yet it is ourview that much can be learned from studiesthat find evidence contradicting certainassumptions or predictions from theory.

Of course, we showed that not all of the evi-dence is consistent with the theory that moti-vates a test. A striking example of this is thenegative relationship between downstream



risk and vertical integration, which is incon-sistent with the trade-off between incentiveand insurance concerns that is fundamentalto the basic moral-hazard model. We havenoted that the regularity could result fromendogenous risk (i.e., agents with higherpowered incentives respond more stronglyto shocks), from self selection (i.e., agentswith lower risk aversion choose riskier activ-ities), or from increased agent importance(i.e., when conditions are more volatile,agent investments are more valuable, as sug-gested by PR theories). Nevertheless, thestrength of the finding is puzzling, and weare encouraged by the fact that theoristshave responded to this contrary evidenceand provided new models to explain it.

As to what the data reveal in relation topublic policy, we did not have a particularconclusion in mind when we began to col-lect the evidence, and we have tried to befair in presenting the empirical regularities.We are therefore somewhat surprised atwhat the weight of the evidence is telling us.It says that, under most circumstances, profit-maximizing vertical-integration decisionsare efficient, not just from the firms’ butalso from the consumers’ points of view.Although there are isolated studies that con-tradict this claim, the vast majority supportit. Moreover, even in industries that arehighly concentrated so that horizontal con-siderations assume substantial importance,the net effect of vertical integration appearsto be positive in many instances. We there-fore conclude that, faced with a verticalarrangement, the burden of evidenceshould be placed on competition authoritiesto demonstrate that that arrangement isharmful before the practice is attacked.Furthermore, we have found clear evidencethat restrictions on vertical integration thatare imposed, often by local authorities, onowners of retail networks are usually detri-mental to consumers. Given the weight ofthe evidence, it behooves government agen-cies to reconsider the validity of suchrestrictions.

REFERENCES

Acemoglu, Daron, Philippe Aghion, Rachel Griffith,and Fabrizio Zilibotti. 2005. “Vertical Integrationand Technology: Theory and Evidence.” CEPRDiscussion Paper, no. 5258.

Acemoglu, Daron, Simon Johnson, and Todd Mitton.2007. “Determinants of Vertical Integration:Financial Development and Contracting Costs.”Unpublished.

Ackerberg, Daniel A., and Maristella Botticini. 2002.“Endogenous Matching and the EmpiricalDeterminants of Contract Form.” Journal of PoliticalEconomy, 110(3): 564–91.

Alchian, Armen A., and Harold Demsetz. 1972.“Production, Information Costs, and EconomicOrganization.” American Economic Review, 62(5):777–95.

Allen, Bruce T. 1971. “Vertical Integration and MarketForeclosure: The Case of Cement and Concrete.”Journal of Law and Economics, 14(1): 251–74.

Allen, Douglas W., and Dean Lueck. 1995. “RiskPreferences and the Economics of Contracts.”American Economic Review, 85(2): 447–51.

Anderson, Erin. 1985. “The Salesperson as OutsideAgent or Employee: A Transaction Cost Analysis.”Marketing Science, 4(3): 234–54.

Anderson, Erin. 1988. “Transaction Costs asDeterminants of Opportunism in Integrated andIndependent Sales Forces.” Journal of EconomicBehavior and Organization, 9(3): 247–64.

Anderson, Erin, and David C. Schmittlein. 1984.“Integration of the Sales Force: An EmpiricalExamination.” RAND Journal of Economics, 15(3):385–95.

Antras, Pol. 2003. “Firms, Contracts, and TradeStructure.” Quarterly Journal of Economics, 118(4):1375–1418.

Arruñada, Benito, Manuel González-Díaz, and AlbertoFernández. 2004. “Determinants of OrganizationalForm: Transaction Costs and Institutions in theEuropean Trucking Industry.” Industrial andCorporate Change, 13(6): 867–82.

Asker, John. 2004. “Measuring Cost Advantages fromExclusive Dealing: An Empirical Study of BeerDistribution.” Unpublished.

Azoulay, Pierre. 2004. “Capturing Knowledge withinand across Firm Boundaries: Evidence from ClinicalDevelopment.” American Economic Review, 94(5):1591–1612.

Bai, Chong-en, and Zhigang Tao. 2000. “ContractMixing in Franchising as a Mechanism for Public-Good Provision.” Journal of Economics andManagement Strategy, 9(1): 85–113.

Baker, George P., Robert Gibbons, and Kevin J.Murphy. 1999. “Informal Authority in Organizations.”Journal of Law, Economics, and Organization, 15(1):56–73.

Baker, George P., Robert Gibbons, and Kevin J.Murphy. 2002. “Relational Contracts and the Theoryof the Firm.” Quarterly Journal of Economics,117(1): 39–84.

Baker, George P., and Thomas N. Hubbard. 2003.“Make versus Buy in Trucking: Asset Ownership, Job



Design, and Information.” American EconomicReview, 93(3): 551–72.

Baker, George P., and Thomas N. Hubbard. 2004.“Contractibility and Asset Ownership: On-BoardComputers and Governance in U.S. Trucking.”Quarterly Journal of Economics, 119(4): 1443–79.

Barron, John M., and John R. Umbeck. 1984. “TheEffects of Different Contractual Arrangements: TheCase of Retail Gasoline Markets.” Journal of Lawand Economics, 27(2): 313–28.

Berle, Adolf, and Gardiner Means. 1933. The ModernCorporation and Private Property. New York:Macmillan.

Bhattacharyya, Sugato, and Francine Lafontaine. 1995.“Double-Sided Moral Hazard and the Nature ofShare Contracts.” RAND Journal of Economics,26(4): 761–81.

Blass, Asher A., and Dennis W. Carlton. 2001. “TheChoice of Organizational Form in Gasoline Retailingand the Cost of Laws That Limit That Choice.”Journal of Law and Economics, 44(2): 511–24.

Bork, Robert H. 1969. “Vertical Integration andCompetitive Processes.” In Public Policy towardsMergers, ed. F. Weston and S. Peltzman. PacificPalisades: Goodyear Publishing, 139–49.

Brenkers, Randy, and Frank Verboven. 2006.“Liberalizing a Distribution System: The EuropeanCar Market.” Journal of the European EconomicAssociation, 4(1): 216–51.

Brickley, James A. 1999. “Incentive Conflicts andContractual Restraints: Evidence from Franchising.”Journal of Law and Economics, 42(2): 745–74.

Brickley, James A., and Frederick H. Dark. 1987. “TheChoice of Organizational Form: The Case ofFranchising.” Journal of Financial Economics, 18(2):401–20.

Brickley, James A., Frederick H. Dark, and Michael S.Weisbach. 1991. “An Agency Perspective onFranchising.” Financial Management, 20(1): 27–35.

Brickley, James A., James S. Linck, and Clifford W.Smith Jr. 2003. “Boundaries of the Firm: Evidencefrom the Banking Industry.” Journal of FinancialEconomics, 70(3): 351–83.

Bull, Clive. 1987. “The Existence of Self-EnforcingImplicit Contracts.” Quarterly Journal of Economics,102(1): 147–59.

Campbell, John Y., Andrew W. Lo, and A. CraigMacKinlay. 1997. The Econometrics of FinancialMarkets. Princeton: Princeton University Press.

Carney, Mick, and Eric Gedajlovic. 1991. “VerticalIntegration and Franchise Systems: Agency Theoryand Resource Explanations.” Strategic ManagementJournal, 12(8): 607–29.

Caves, Richard E., and William F. Murphy, II. 1976.“Franchising: Firms, Markets, and IntangibleAssets.” Southern Economic Journal, 42(4): 572–86.

Chipty, Tasneem. 2001. “Vertical Integration, MarketForeclosure, and Consumer Welfare in the CableTelevision Industry.” American Economic Review,91(3): 428–53.

Ciliberto, Federico. 2006. “Does Organizational FormAffect Investment Decisions?” Journal of IndustrialEconomics, 54(1): 63–93.

Coase, Ronald H. 1937. “The Nature of the Firm.”

Economica, 4(16): 386–405.Cooper, James C., Luke M. Froeb, Dan O’Brien, and

Michael G. Vita. 2005. “Vertical Antitrust Policy as aProblem of Inference.” International Journal ofIndustrial Organization, 23(7–8): 639–64.

Corts, Kenneth S. 2001. “The Strategic Effects ofVertical Market Structure: Common Agency andDivisionalization in the US Motion PictureIndustry.” Journal of Economics and ManagementStrategy, 10(4): 509–28.

Eckbo, B. Espen. 1983. “Horizontal Mergers,Collusion, and Stockholder Wealth.” Journal ofFinancial Economics, 11(1–4): 241–73.

Edwards, Kenneth, John D. Jackson, and Henry L.Thompson. 2000. “A Note on Vertical Integrationand Stock Ratings of Oil Companies in the U.S.”Energy Journal, 21(2): 145–51.

Forbes, Silke J., and Mara Lederman. 2006. “ControlRights, Network Structure and Vertical Integration:Evidence from Regional Airlines.” Unpublished.

Ford, George S., and John D. Jackson. 1997.“Horizontal Concentration and Vertical Integrationin the Cable Television Industry.” Review ofIndustrial Organization, 12(4): 501–18.

Gibbons, Robert. 2005. “Four Formal(izable) Theoriesof the Firm?” Journal of Economic Behavior andOrganization, 58(2): 200–245.

Gil, Ricard. 2006. “Revenue Sharing Distortions andVertical Integration in the Movie Industry.”Unpublished.

Gil, Ricard. 2007. “‘Make-or-Buy’ in Movies: Integrationand Ex-post Renegotiation.” International Journal ofIndustrial Organization, 25(4): 643–55.

Gil, Ricard, and Wesley R. Hartmann. 2007. “AiringYour Dirty Laundry: Vertical Integration,Reputational Capital and Social Networks.”Unpublished.

González-Díaz, Manuel, Benito Arruñada, and AlbertoFernández. 2000. “Causes of Subcontracting:Evidence from Panel Data on Construction Firms.”Journal of Economic Behavior and Organization,42(2): 167–87.

Gould, J. R. 1977. “Price Discrimination and VerticalControl: A Note.” Journal of Political Economy,85(5): 1063–71.

Greenhut, M. L., and H. Ohta. 1979. “VerticalIntegration of Successive Oligopolists.” AmericanEconomic Review, 69(1): 137–41.

Grossman, Sanford J., and Oliver Hart. 1986. “TheCosts and Benefits of Ownership: A Theory ofVertical and Lateral Integration.” Journal of PoliticalEconomy, 94(4): 691–719.

Grout, Paul A. 1984. “Investment and Wages in theAbsence of Binding Contracts: A Nash BarginingApproach.” Econometrica, 52(2): 449–60.

Hanson, Gordon H. 1995. “Incomplete Contracts,Risk, and Ownership.” International EconomicReview, 36(2): 341–63.

Hart, Oliver. 1995. Firms, Contracts, and FinancialStructure. Clarendon Lectures in Economics.Oxford and New York: Oxford University Press,Clarendon Press.

Hart, Oliver, and John Moore. 1990. “Property Rightsand the Nature of the Firm.” Journal of Political



Economy, 98(6): 1119–58.Hart, Oliver, and Jean Tirole. 1990. “Vertical

Integration and Market Foreclosure.” BrookingsPapers on Economic Activity: Microeconomics,205–76.

Hastings, Justine S. 2004. “Vertical Relationships andCompetition in Retail Gasoline Markets: EmpiricalEvidence from Contract Changes in SouthernCalifornia.” American Economic Review, 94(1):317–28.

Hastings, Justine S., and Richard J. Gilbert. 2005.“Market Power, Vertical Integration and theWholesale Price of Gasoline.” Journal of IndustrialEconomics, 53(4): 469–92.

Helfat, Constance E., and David J. Teece. 1987.“Vertical Integration and Risk Reduction.” Journal ofLaw, Economics, and Organization, 3(1): 47–67.

Hendricks, Kenneth, and R. Preston McAfee. 1999.“Measuring Industry Concentration in IntermediateGoods.” Unpublished.

Holmstrom, Bengt. 1982. “Moral Hazard in Teams.”Bell Journal of Economics, 13(2): 324–40.

Holmstrom, Bengt. 1999. “The Firm as aSubeconomy.” Journal of Law, Economics, andOrganization, 15(1): 74–102.

Holmstrom, Bengt, and Paul Milgrom. 1991.“Multitask Principal–Agent Analyses: IncentiveContracts, Asset Ownership, and Job Design.”Journal of Law, Economics, and Organization, 7:24–52.

Holmstrom, Bengt, and Paul Milgrom. 1994. “TheFirm as an Incentive System.” American EconomicReview, 84(4): 972–91.

Holmstrom, Bengt, and John Roberts. 1998. “TheBoundaries of the Firm Revisited.” Journal ofEconomic Perspectives, 12(4): 73–94.

Hortaçsu, Ali, and Chad Syverson. 2007a. “CementingRelationships: Vertical Integration, Foreclosure,Productivity, and Prices.” Journal of PoliticalEconomy, 115(2): 250–301.

Hortaçsu, Ali, and Chad Syverson. 2007b. “VerticalIntegration and Production: Some Plant-LevelEvidence.” Unpublished.

Jin, Ginger, and Phillip Leslie. 2005. “ReputationIncentives for Restaurant Hygiene.” Unpublished.

John, George, and Barton A. Weitz. 1988. “ForwardIntegration into Distribution: An Empirical Test ofTransaction Cost Analysis.” Journal of Law,Economics, and Organization, 4(2): 337–55.

Johnson, Ronald N., and Allen M. Parkman. 1987.“Spatial Competition and Vertical Integration;Cement and Concrete Revisited: Comment.”American Economic Review, 77(4): 750–53.

Joskow, Paul L. 1985. “Vertical Integration and Long-Term Contracts: The Case of Coal-Burning ElectricGenerating Plants.” Journal of Law, Economics, andOrganization, 1(1): 33–80.

Joskow, Paul L. 2005. “Vertical Integration.” InHandbook of New Institutional Economics, ed. C.Menard and M. M. Shirley. Dordrecht and NewYork: Springer, 319–48.

Kalnins, Arturs, and Francine Lafontaine. 2004.“Multi-unit Ownership in Franchising: Evidencefrom the Fast-Food Industry in Texas.” RAND

Journal of Economics, 35(4): 747–61.Kaufmann, Patrick J., and Francine Lafontaine. 1994.

“Costs of Control: The Source of Economic Rentsfor McDonald’s Franchisees.” Journal of Law andEconomics, 37(2): 417–53.

Kehoe, Michael R. 1996. “Franchising, AgencyProblems, and the Cost of Capital.” AppliedEconomics, 28(11): 1485–93.

Kenney, Roy W., and Benjamin Klein. 1983. “TheEconomics of Block Booking.” Journal of Law andEconomics, 26(3): 497–540.

Kerkvliet, Joe. 1991. “Efficiency and VerticalIntegration: The Case of Mine-Mouth ElectricGenerating Plants.” Journal of Industrial Economics,39(5): 467–82.

Klein, Benjamin. 1980. “Transaction Cost Determinantsof “Unfair” Contractual Arrangements.” AmericanEconomic Review, 70(2): 356–62.

Klein, Benjamin. 1995. “The Economics of FranchiseContracts.” Journal of Corporate Finance, 2(1–2):9–37.

Klein, Benjamin, Robert G. Crawford, and Armen A.Alchian. 1978. “Vertical Integration, AppropriableRents, and the Competitive Contracting Process.”Journal of Law and Economics, 21(2): 297–326.

Klein, Benjamin, and Keith B. Leffler. 1981. “The Roleof Market Forces in Assuring ContractualPerformance.” Journal of Political Economy, 89(4):615–41.

Klein, Benjamin, and Kevin M. Murphy. 1997.“Vertical Integration as a Self-Enforcing ContractualArrangement.” American Economic Review, 87(2):415–20.

Klein, Peter G. 2005. “The Make-or-Buy Decision:Lessons from Empirical Studies.” In Handbook ofNew Institutional Economics, ed. C. Menard and M.M. Shirley. Dordrecht and New York: Springer,435–64.

Knight, Frank. 1921. Risk, Uncertainty, and Profit.Boston: Houghton Mifflin.

Krattenmaker, Thomas G., and Steven C. Salop. 1986.“Anticompetitive Exclusion: Raising Rivals’ Costs toAchieve Power over Price.” Yale Law Journal, 96(2):209–93.

Lafontaine, Francine. 1992. “Agency Theory andFranchising: Some Empirical Results.” RANDJournal of Economics, 23(2): 263–83.

Lafontaine, Francine, and Sugato Bhattacharyya. 1995.“The Role of Risk in Franchising.” Journal ofCorporate Finance, 2(1–2): 39–74.

Lafontaine, Francine, and Emmanuel Raynaud. 2002.“Residual Claims and Self-Enforcement as IncentiveMechanisms in Franchise Contracts: Substitutes orComplements?” In The Economics of Contracts:Theories and Applications, ed. E. Brousseau and J.-M. Glachant. Cambridge; New York and Melbourne:Cambridge University Press, 315–36.

Lafontaine, Francine, and Kathryn L. Shaw. 1999.“The Dynamics of Franchise Contracting: Evidencefrom Panel Data.” Journal of Political Economy,107(5): 1041–80.

Lafontaine, Francine, and Kathryn L. Shaw. 2005.“Targeting Managerial Control: Evidence fromFranchising.” RAND Journal of Economics, 36(1):



131–50.Lafontaine, Francine, and Margaret E. Slade. 1996.

“Retail Contracting and Costly Monitoring: Theoryand Evidence.” European Economic Review,40(3–5): 923–32.

Lafontaine, Francine, and Margaret E. Slade. 1997.“Retail Contracting: Theory and Practice.” Journal ofIndustrial Economics, 45(1): 1–25.

Lafontaine, Francine, and Margaret E. Slade. 2001.“Incentive Contracting and the Franchise Decision.”In Game Theory and Business Applications, ed. K.Chatterjee and W. Samuelson. Norwell, Mass.:Kluwer Academic Press, 133–88.

Lafontaine, Francine, and Margaret E. Slade.Forthcoming-a. “Exclusive Contracts and VerticalRestraints: Empirical Evidence and Public Policy.”In Handbook of Antitrust Economics, ed. P.Buccirossi. Cambridge, Mass.: MIT Press.

Lafontaine, Francine, and Margaret E. Slade.Forthcoming-b. “Inter-firm Contracts: TheEvidence.” In Handbook of OrganizationalEconomics, ed. R. Gibbons and J. Roberts.Princeton: Princeton University Press.

Levin, Richard C. 1981. “Vertical Integration andProfitability in the Oil Industry.” Journal ofEconomic Behavior and Organization, 2(3): 215–35.

Liebeler, Wesley J. 1968. “Towards a Consumer’sAntitrust Law: The Federal Trade Commission andVertical Mergers in the Cement Industry.” UCLALaw Review, 15: 1153–1202.

Lieberman, Marvin B. 1991. “Determinants of VerticalIntegration: An Empirical Test.” Journal ofIndustrial Economics, 39(5): 451–66.

Lucas, Robert E., Jr. 1976. “Econometric PolicyEvaluation: A Critique.” Carnegie–RochesterConference Series on Public Policy, 1: 19–46.

Lutz, Nancy A. 1995. “Ownership Rights andIncentives in Franchising.” Journal of CorporateFinance, 2(1–2): 103–31.

Lyons, Bruce R. 1995. “Specific Investment,Economies of Scale, and the Make-or-Buy Decision:A Test of Transaction Cost Theory.” Journal ofEconomic Behavior and Organization, 26(3):431–43.

MacKinlay, A. Craig. 1997. “Event Studies inEconomics and Finance.” Journal of EconomicLiterature, 35(1): 13–39.

Maness, Robert. 1996. “Incomplete Contracts and theChoice between Vertical Integration andFranchising.” Journal of Economic Behavior andOrganization, 31(1): 101–15.

Martin, Robert E. 1988. “Franchising and RiskManagement.” American Economic Review, 78(5):954–68.

Masten, Scott E. 1984. “The Organization ofProduction: Evidence from the Aerospace Industry.”Journal of Law and Economics, 27(2): 403–17.

Masten, Scott E. 1988. “A Legal Basis for the Firm.”Journal of Law, Economics, and Organization, 4(1):181–98.

Masten, Scott E., James W. Meehan Jr., and Edward A.Snyder. 1989. “Vertical Integration in the U.S. AutoIndustry: A Note on the Influence of TransactionSpecific Assets.” Journal of Economic Behavior and

Organization, 12(2): 265–73.Masten, Scott E., James W. Meehan Jr., and Edward A.

Snyder. 1991. “The Costs of Organization.” Journalof Law, Economics, and Organization, 7(1): 1–25.

McAfee, R. Preston, Kenneth Hendricks, Joshua M.Fried, Michael A. Williams, and Melanie StallingsWilliams. 2001. “Measuring Anticompetitive Effectsof Mergers When Buyer Power Is Concentrated.”Texas Law Review, 79(6): 1621–39.

McBride, Mark E. 1983. “Spatial Competition andVertical Integration: Cement and ConcreteRevisited.” American Economic Review, 73(5):1011–22.

McMillan, John, and Christopher Woodruff. 1999.“Dispute Prevention without Courts in Vietnam.”Journal of Law, Economics, and Organization, 15(3):637–58.

Minkler, Alanson P. 1990. “An Empirical Analysis of aFirm’s Decision to Franchise.” Economics Letters,34(1): 77–82.

Minkler, Alanson P., and Timothy A. Park. 1994. “AssetSpecificity and Vertical Integration in Franchising.”Review of Industrial Organization, 9(4): 409–23.

Monteverde, Kirk, and David J. Teece. 1982. “SupplierSwitching Costs and Vertical Integration in theAutomobile Industry.” Bell Journal of Economics,13(1): 206–13.

Mullainathan, Sendhil, and David Scharfstein. 2001.“Do Firm Boundaries Matter?” American EconomicReview, 91(2): 195–99.

Mullin, Joseph C., and Wallace P. Mullin. 1997.“United States Steel’s Acquisition of the GreatNorthern Ore Properties: Vertical Foreclosure orEfficient Contractual Governance?” Journal of Law,Economics, and Organization, 13(1): 74–100.

Muris, Timothy J., David T. Scheffman, and Pablo T.Spiller. 1992. “Strategy and Transaction Costs: TheOrganization of Distribution in the Carbonated SoftDrink Industry.” Journal of Economics andManagement Strategy, 1(1): 83–128.

Newey, Whitney K., and Kenneth D. West. 1987. “ASimple, Positive Semi-Definite, Heteroskedasticityand Autocorrelation Consistent Covariance Matrix.”Econometrica, 55(3): 703–08.

Nickerson, Jack A., and Brian S. Silverman. 2003.“Why Aren’t All Truck Drivers Owner-Operators?Asset Ownership and the Employment Relation inInterstate For-Hire Trucking.” Journal of Economicsand Management Strategy, 12(1): 91–118.

Norton, Seth W. 1988. “An Empirical Look atFranchising as an Organizational Form.” Journal ofBusiness, 61(2): 197–218.

Novak, Sharon, and Scott Stern. 2003.“Complementarity among Vertical IntegrationDecisions: Evidence from Automobile ProductDevelopment.” Unpublished.

Ohanian, Nancy Kane. 1994. “Vertical Integration inthe U.S. Pulp and Paper Industry, 1900–1940.”Review of Economics and Statistics, 76(1): 202–07.

Ordover, Janusz A., Garth Saloner, and Steven C.Salop. 1990. “Equilibrium Vertical Foreclosure.”American Economic Review, 80(1): 127–42.

Peltzman, Samuel. 1969. “Issues in Vertical IntegrationPolicy.” In Public Policy towards Mergers, ed. F.



Weston and S. Peltzman. Pacific Palisades: GoodyearPublishing, 167–76.

Pénard, Thierry, Emmanuel Raynaud, and StephaneSaussier. 2003. “Dual Distribution and Royalty Ratesin Franchised Chains: An Empirical Analysis UsingFrench Data.” Journal of Marketing Channels,10(3–4): 5–31.

Perry, Martin K. 1978. “Vertical Integration: TheMonopsony Case.” American Economic Review,68(4): 561–70.

Pinkse, Joris, Margaret E. Slade, and Craig Brett. 2002.“Spatial Price Competition: A SemiparametricApproach.” Econometrica, 70(3): 1111–53.

Pinkse, Joris, Margaret E. Slade, and Lihong Shen.2006. “Dynamic Spatial Discrete Choice Using One-Step GMM: An Application to Mine OperatingDecisions.” Spatial Economic Analysis, 1(1): 53–99.

Pirrong, Stephen Craig. 1993. “Contracting Practicesin Bulk Shipping Markets: A Transactions CostExplanation.” Journal of Law and Economics, 36(2):937–76.

Pitofsky, Robert. 1997. “Vertical Restraints and VerticalAspects of Mergers—A U.S. Perspective.” Speechdelivered to the Fordham Corporate Law Institute.

Prendergast, Canice. 2002. “The Tenuous Trade-Offbetween Risk and Incentives.” Journal of PoliticalEconomy, 110(5): 1071–1102.

Reiffen, David, and Andrew N. Kleit. 1990. “TerminalRailroad Revisited: Foreclosure of an EssentialFacility or Simple Horizontal Monopoly?” Journal ofLaw and Economics, 33(2): 419–38.

Rey, Patrick, and Joseph Stiglitz. 1995. “The Role ofExclusive Territories in Producers’ Competition.”RAND Journal of Economics, 26(3): 431–51.

Rosengren, Eric S., and James W. Meehan Jr. 1994.“Empirical Evidence on Vertical Foreclosure.”Economic Inquiry, 32(2): 303–17.

Ross, Stephen A. 1976. “The Arbitrage Theory ofCapital Asset Pricing.” Journal of Economic Theory,13(3): 341–60.

Salinger, Michael A. 1988. “Vertical Mergers andMarket Foreclosure.” Quarterly Journal ofEconomics, 103(2): 345–56.

Salop, Steven C., and David T. Scheffman. 1987. “Cost-Raising Strategies.” Journal of Industrial Economics,36(1): 19–34.

Schmalensee, Richard. 1973. “A Note on the Theory ofVertical Integration.” Journal of Political Economy,81(2): 442–49.

Scott, Frank A., Jr. 1995. “Franchising vs. CompanyOwnership as a Decision Variable of the Firm.”Review of Industrial Organization, 10(1): 69–81.

Sharpe, William F. 1963. “A Simplified Model forPortfolio Analysis.” Management Science, 9(2):277–93.

Shelanski, Howard A., and Peter G. Klein. 1995.“Empirical Research in Transaction Cost Economics:A Review and Assessment.” Journal of Law,Economics, and Organization, 11(2): 335–61.

Shelton, John P. 1967. “Allocative Efficiency vs. ‘X-Efficiency’: Comment.” American Economic Review,57(5): 1252–58.

Shepard, Andrea. 1993. “Contractual Form, RetailPrice, and Asset Characteristics in Gasoline

Retailing.” RAND Journal of Economics, 24(1): 58–77.Slade, Margaret E. 1996. “Multitask Agency and

Contract Choice: An Empirical Exploration.”International Economic Review, 37(2): 465–86.

Slade, Margaret E. 1998a. “Beer and the Tie: DidDivestiture of Brewer-Owned Public Houses Leadto Higher Beer Prices?” Economic Journal,108(448): 565–602.

Slade, Margaret E. 1998b. “Strategic Motives forVertical Separation: Evidence from Retail GasolineMarkets.” Journal of Law, Economics, andOrganization, 14(1): 84–113.

Snyder, Christopher M. 1996. “Vertical Integration forEfficiency or Market Power? Event Studies of theU.S. Oil Industry.” Unpublished.

Spengler, Joseph J. 1950. “Vertical Integration andAntitrust Policy.” Journal of Political Economy,58(4): 347–52.

Spiller, Pablo T. 1985. “On Vertical Mergers.” Journal ofLaw, Economics, and Organization, 1(2): 285–312.

Tadelis, Steven. 2002. “Complexity, Flexibility, and theMake-or-Buy Decision.” American EconomicReview, 92(2): 433–37.

Thompson, R. Steve. 1994. “The Franchise Life Cycleand the Penrose Effect.” Journal of EconomicBehavior and Organization, 24(2): 207–18.

Vita, Michael G. 2000. “Regulatory Restrictions onVertical Integration and Control: The CompetitiveImpact of Gasoline Divorcement Policies.” Journalof Regulatory Economics, 18(3): 217–33.

Walker, Gordon, and David Weber. 1984. “A TransactionCost Approach to Make-or-Buy Decisions.”Administrative Science Quarterly, 29(3): 373–91.

Warren-Boulton, Frederick R. 1974. “Vertical Controlwith Variable Proportions.” Journal of PoliticalEconomy, 82(4): 783–802.

Warren-Boulton, Frederick R. 2003. “TheContribution of the Merger Guidelines to theAnalysis of Non-horizontal Mergers.” Unpublished.

Waterman, David, and Andrew A. Weiss. 1996. “TheEffects of Vertical Integration between CableTelevision Systems and Pay Cable Networks.”Journal of Econometrics, 72(1–2): 357–95.

Weiss, Avi. 1992. “The Role of Firm-Specific Capital inVertical Mergers.” Journal of Law and Economics,35(1): 71–88.

Wernerfelt, Birger. 1997. “On the Nature and Scope ofthe Firm: An Adjustment-Cost Theory.” Journal ofBusiness, 70(4): 489–514.

Whinston, Michael D. 2003. “On the Transaction CostDeterminants of Vertical Integration.” Journal ofLaw, Economics, and Organization, 19(1): 1–23.

Williamson, Oliver E. 1971. “The Vertical Integrationof Production: Market Failure Considerations.”American Economic Review, 61(2): 112–23.

Williamson, Oliver E. 1975. Markets and Hierarchies:Analysis and Antitrust Implications. New York: FreePress.

Williamson, Oliver E. 1979. “Transaction-CostEconomics: The Governance of ContracturalRelations.” Journal of Law and Economics, 22(2):233–61.

Williamson, Oliver E. 1983. “Credible Commitments:Using Hostages to Support Exchange.” American



Economic Review, 73(4): 519–40.Williamson, Oliver E. 1985. The Economic Institutions

of Capitalism. New York: Free Press.Williamson, Oliver E. 1991. “Comparative Economic

Organization: The Analysis of Discrete StructuralAlternatives.” Administrative Science Quarterly,36(2): 269–96.

Woodruff, Christopher. 2002. “Non-contractibleInvestments and Vertical Integration in the Mexican

Footwear Industry.” International Journal ofIndustrial Organization, 20(8): 1197–1224.

Wooldridge, Jeffrey M. 2002. Econometric Analysis ofCross Section and Panel Data. Cambridge andLondon: MIT Press.

Yeap, Clarissa A. 2006. “Residual Claims andIncentives in Restaurant Chains.” U.S. CensusBureau, Center for Economic Studies WorkingPaper, no. CES 06-18.



This article has been cited by:

1. Albert Jolink, Eva Niesten. 2012. Recent qualitative advances on hybrid organizations: Takingstock, looking ahead. Scandinavian Journal of Management . [CrossRef]

2. Giorgio Zanarone. 2012. Coase (1937) revisited: Endogenous fiat in firms and markets.International Review of Economics . [CrossRef]

3. G. Zanarone. 2012. Contract Adaptation under Legal Constraints. Journal of Law, Economics, andOrganization . [CrossRef]

4. J. Luo, C. Y. Baldwin, D. E. Whitney, C. L. Magee. 2012. The architecture of transaction networks:a comparative analysis of hierarchy in two sectors. Industrial and Corporate Change . [CrossRef]

5. Sharon Novak, Birger Wernerfelt. 2012. On the Grouping of Tasks into Firms: Make-or-Buy withInterdependent Parts*. Journal of Economics & Management Strategy 21:1, 53-77. [CrossRef]

6. Renáta Kosová, Francine Lafontaine. 2012. Much Ado about Chains: A Research Agenda.International Journal of Industrial Organization . [CrossRef]

7. Wei Guan, Jakob Rehme. 2012. Vertical integration in supply chains: driving forces andconsequences for a manufacturer's downstream integration. Supply Chain Management: AnInternational Journal 17:2, 187-201. [CrossRef]

8. Simi Kedia, S. Abraham Ravid, Vicente Pons. 2011. When Do Vertical Mergers Create Value?.Financial Management 40:4, 845-877. [CrossRef]

9. Thierry Pénard, Emmanuel Raynaud, Stéphane Saussier. 2011. Monitoring Policy andOrganizational Forms in Franchised Chains. International Journal of the Economics of Business18:3, 399-417. [CrossRef]

10. Guanming Shi, Kyle W. Stiegert, Jean P. Chavas. 2011. An analysis of bundle pricing in horizontaland vertical markets: The case of the U.S. cottonseed market. Agricultural Economics 42, 77-88.[CrossRef]

11. Michele Moretto, Gianpaolo Rossini. 2011. Flexible Outsourcing. Managerial and DecisionEconomics n/a-n/a. [CrossRef]

12. Fabio Pieri, Enrico Zaninotto. 2011. Vertical integration and efficiency: an application to the Italianmachine tool industry. Small Business Economics . [CrossRef]

13. Asta Dis Oladottir, Bersant Hobdari, Marina Papanastassiou, Robert Pearce, Evis Sinani. 2011.Strategic complexity and global expansion: An empirical study of newcomer MultinationalCorporations from small economies. Journal of World Business . [CrossRef]

14. Shiqiang Li. 2011. The Firm as a Means of Concealing Information. LABOUR no-no. [CrossRef]

15. James A. Brickley, James S. Linck, Clifford W. Smith. 2011. Vertical integration to avoidcontracting with potential competitors: Evidence from bankers’ banks. Journal of FinancialEconomics . [CrossRef]

16. Desmond (Ho-Fu) Lo, Mrinal Ghosh, Francine Lafontaine. 2011. The Incentive and SelectionRoles of Sales Force Compensation Contracts. Journal of Marketing Research 48:4, 781-798.[CrossRef]

17. Roberto Hernán González, Praveen Kujal. 2011. Vertical integration, market foreclosure andquality investment. Portuguese Economic Journal . [CrossRef]

http://dx.doi.org/10.1016/j.scaman.2012.03.001

http://dx.doi.org/10.1007/s12232-012-0155-z

http://dx.doi.org/10.1093/jleo/ews005

http://dx.doi.org/10.1093/icc/dts007

http://dx.doi.org/10.1111/j.1530-9134.2011.00322.x

http://dx.doi.org/10.1016/j.ijindorg.2011.12.010

http://dx.doi.org/10.1108/13598541211212915

http://dx.doi.org/10.1111/j.1755-053X.2011.01164.x

http://dx.doi.org/10.1080/13571516.2011.618613

http://dx.doi.org/10.1111/j.1574-0862.2011.00553.x

http://dx.doi.org/10.1002/mde.1560

http://dx.doi.org/10.1007/s11187-011-9367-y

http://dx.doi.org/10.1016/j.jwb.2011.08.003

http://dx.doi.org/10.1111/j.1467-9914.2011.00529.x

http://dx.doi.org/10.1016/j.jfineco.2011.09.001

http://dx.doi.org/10.1509/jmkr.48.4.781

http://dx.doi.org/10.1007/s10258-011-0074-z

18. D. Geradin, A. Layne-Farrar, A. J. Padilla. 2011. Elves or Trolls? The role of nonpracticing patentowners in the innovation economy. Industrial and Corporate Change . [CrossRef]

19. Bruce M. Owen. 2011. Antitrust and Vertical Integration in “New Economy” Industries withApplication to Broadband Access. Review of Industrial Organization . [CrossRef]

20. Philippe Aghion, , Richard Holden. 2011. Incomplete Contracts and the Theory of the Firm: WhatHave We Learned over the Past 25 Years?. Journal of Economic Perspectives 25:2, 181-197.[Abstract] [View PDF article] [PDF with links]

21. Nuno Garoupa, Anthony Ogus, Andrew Sanders. 2011. THE INVESTIGATION ANDPROSECUTION OF REGULATORY OFFENCES: IS THERE AN ECONOMIC CASE FORINTEGRATION?. The Cambridge Law Journal 70:01, 229-259. [CrossRef]

22. Oliver Hart. 2011. Thinking about the Firm: A Review of Daniel Spulber's The Theory of the Firm.Journal of Economic Literature 49:1, 101-113. [Abstract] [View PDF article] [PDF with links]

23. Bharat A. Jain, Omesh Kini, Jaideep Shenoy. 2011. Vertical divestitures through equity carve-outsand spin-offs: A product markets perspective#. Journal of Financial Economics . [CrossRef]

24. Jérôme Barthélemy. 2011. Agency and institutional influences on franchising decisions. Journalof Business Venturing 26:1, 93-103. [CrossRef]

25. Silke J. Forbes, Mara Lederman. 2010. Does vertical integration affect firm performance?Evidence from the airline industry. The RAND Journal of Economics 41:4, 765-790. [CrossRef]

26. Robert Gibbons. 2010. Transaction-Cost Economics: Past, Present, and Future?. ScandinavianJournal of Economics 112:2, 263-288. [CrossRef]

27. Mitsuru Sunada. 2010. Vertical Integration in the Japanese Movie Industry. Journal of Industry,Competition and Trade 10:2, 135-150. [CrossRef]

28. Luca Lambertini. 2010. Make vs buy in a monopoly with demand or cost uncertainty#. Researchin Economics 64:2, 101-109. [CrossRef]

29. Forbes Silke Januszewski, Lederman Mara. 2009. Adaptation and Vertical Integration in theAirline Industry. American Economic Review 99:5, 1831-1849. [Abstract] [View PDF article][PDF with links]

30. Giulio Cainelli, Donato Iacobucci. 2009. Do Agglomeration and Technology Affect VerticalIntegration? Evidence from Italian Business Groups. International Journal of the Economics ofBusiness 16:3, 305-322. [CrossRef]

31. Stéphane Saussier, Carine Staropoli, Anne Yvrande-Billon. 2009. Public–Private Agreements,Institutions, and Competition: When Economic Theory Meets Facts. Review of IndustrialOrganization 35:1-2, 1-18. [CrossRef]

32. A. Volokh. 2009. Property Rights and Contract Form in Medieval Europe. American Law andEconomics Review 11:2, 399-450. [CrossRef]

33. M. Cave, K. Hatta. 2009. Transforming telecommunications technologies--policy and regulation.Oxford Review of Economic Policy 25:3, 488-505. [CrossRef]

34. Thomas Hutzschenreuter, Florian Gröne. 2009. Changing Vertical Integration Strategies underPressure from Foreign Competition: The Case of US and German Multinationals. Journal ofManagement Studies 46:2, 269-307. [CrossRef]

http://dx.doi.org/10.1093/icc/dtr031

http://dx.doi.org/10.1007/s11151-011-9291-y

http://dx.doi.org/10.1257/jep.25.2.181

http://pubs.aeaweb.org/doi/pdf/10.1257/jep.25.2.181

http://pubs.aeaweb.org/doi/pdfplus/10.1257/jep.25.2.181

http://dx.doi.org/10.1017/S0008197311000237

http://dx.doi.org/10.1257/jel.49.1.101

http://pubs.aeaweb.org/doi/pdf/10.1257/jel.49.1.101

http://pubs.aeaweb.org/doi/pdfplus/10.1257/jel.49.1.101

http://dx.doi.org/10.1016/j.jfineco.2011.01.001

http://dx.doi.org/10.1016/j.jbusvent.2009.05.002

http://dx.doi.org/10.1111/j.1756-2171.2010.00120.x

http://dx.doi.org/10.1111/j.1467-9442.2010.01609.x

http://dx.doi.org/10.1007/s10842-009-0059-0

http://dx.doi.org/10.1016/j.rie.2010.01.001

http://dx.doi.org/10.1257/aer.99.5.1831

http://pubs.aeaweb.org/doi/pdf/10.1257/aer.99.5.1831

http://pubs.aeaweb.org/doi/pdfplus/10.1257/aer.99.5.1831

http://dx.doi.org/10.1080/13571510903227049

http://dx.doi.org/10.1007/s11151-009-9226-z

http://dx.doi.org/10.1093/aler/ahp012

http://dx.doi.org/10.1093/oxrep/grp023

http://dx.doi.org/10.1111/j.1467-6486.2008.00811.x

35. Emil D. Attanasi. 2008. Volatility of Bitumen Prices and Implications for the Industry. NaturalResources Research 17:4, 205-213. [CrossRef]

36. Michael R. Baye, Matias Barenstein, Debra J. Holt, Pauline M. Ippolito, James M. Lacko, JesseB. Leary, Janis K. Pappalardo, Paul A. Pautler, Michael G. Vita. 2008. Economics at the FTC:The Google-DoubleClick Merger, Resale Price Maintenance, Mortgage Disclosures, and CreditScoring in Auto Insurance. Review of Industrial Organization 33:3, 211-230. [CrossRef]

37. Sebastian K. Fixson, Jin-Kyu Park. 2008. The power of integrality: Linkages between productarchitecture, innovation, and industry structure. Research Policy 37:8, 1296-1316. [CrossRef]

38. OLIVER HART. 2008. Economica Coase Lecture: Reference Points and the Theory of the Firm.Economica 75:299, 404-411. [CrossRef]

39. Thomas N. Hubbard. 2008. Viewpoint: Empirical research on firms' boundaries. Canadian Journalof Economics/Revue canadienne d'économique 41:2, 341-359. [CrossRef]

40. OLIVER E. WILLIAMSON. 2008. OUTSOURCING: TRANSACTION COST ECONOMICSAND SUPPLY CHAIN MANAGEMENT. The Journal of Supply Chain Management 44:2, 5-16.[CrossRef]

http://dx.doi.org/10.1007/s11053-008-9078-5

http://dx.doi.org/10.1007/s11151-008-9183-y

http://dx.doi.org/10.1016/j.respol.2008.04.026

http://dx.doi.org/10.1111/j.1468-0335.2007.00659.x

http://dx.doi.org/10.1111/j.1540-5982.2008.00466.x

http://dx.doi.org/10.1111/j.1745-493X.2008.00051.x

vertical integration and firm boundaries: the...

Documents