exploration and exploitation alliances in biotechnology: a

Strategic Management JournalStrat. Mgmt. J., 25: 201–221 (2004)

Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/smj.376

EXPLORATION AND EXPLOITATION ALLIANCES INBIOTECHNOLOGY: A SYSTEM OF NEW PRODUCTDEVELOPMENT

FRANK T. ROTHAERMEL1* and DAVID L. DEEDS2

1 DuPree College of Management, Georgia Institute of Technology, Atlanta, Georgia,U.S.A.2 Weatherhead School of Management, Case Western Reserve University, Cleveland,Ohio, U.S.A.

We link the exploration–exploitation framework of organizational learning to a technologyventure’s strategic alliances and argue that the causal relationship between the venture’salliances and its new product development depends on the type of the alliance. In particular, wepropose a product development path beginning with exploration alliances predicting products indevelopment, which in turn predict exploitation alliances, and that concludes with exploitationalliances leading to products on the market. Moreover, we argue that this integrated productdevelopment path is moderated negatively by firm size. As a technology venture grows, it tends towithdraw from this product development path to discover, develop, and commercialize promisingprojects through vertical integration. We test our model on a sample of 325 biotechnology firmsthat entered 2565 alliances over a 25-year period. We find broad support for the hypothesizedproduct development system and the moderating effect of firm size. Copyright 2004 John Wiley& Sons, Ltd.

Strategic alliances are a ubiquitous phenomenon,especially in high-technology industries (Hage-doorn, 1993). Parallel to the rise in interfirm coop-eration, research on strategic alliances has bur-geoned, with one strand focusing on the perfor-mance impact of alliances on the focal firm (Gulati,1998). In this line of inquiry, several scholarshave studied the relationship between a firm’sstrategic alliances and its innovative performanceor new product development (Shan, Walker, andKogut, 1994; Kotabe and Swan, 1995; Deeds andHill, 1996; Baum, Calabrese, and Silverman, 2000;Lerner, Shane, and Tsai, 2003). This is an impor-tant avenue of research since a firm’s innovative-ness and new product development have a directimpact on its continued survival and performance,

Key words: strategic alliances; vertical integration; newproduct development; biotechnology industry*Correspondence to: Frank T. Rothaermel, DuPree Collegeof Management, Georgia Institute of Technology, 800 WestPeachtree Street, NW, Atlanta, GA 30332-0520, U.S.A.

particularly in high-technology industries (Brownand Eisenhardt, 1997).

Prior research provided evidence for the notionthat a firm’s strategic alliances have a positiveimpact upon its innovativeness (Shan et al., 1994),and that the relationship between alliances andnew product development might be characterizedby diminishing marginal returns (Deeds and Hill,1996). Others have shown that a start-up’s con-figuration of alliances impacts its early perfor-mance (Baum et al., 2000), and the nature ofa firm’s cooperative arrangements has a bear-ing on the firm’s level of product innovative-ness (Kotabe and Swan, 1995). More recently,Lerner et al. (2003) found that strategic alliancesentered between small technology ventures andlarge established firms during periods of limitedexternal equity financing tended to be less suc-cessful.

While each of these studies has certainly ad-vanced our understanding of the new product

Copyright 2004 John Wiley & Sons, Ltd. Received 27 December 2000Final revision received 27 August 2003

202 F. T. Rothaermel and D. L. Deeds

development process by establishing a link be-tween a firm’s strategic alliances and an inter-mediate research output or performance indicator,such as patenting propensity (Shan et al., 1994;Baum et al., 2000), level of product innovative-ness (Kotabe and Swan, 1995), products underdevelopment (Deeds and Hill, 1996), and mile-stone stages reached (Lerner et al., 2003), linkingdifferent types of alliances to each distinct stagein the new product development process beginningwith discovery and culminating in commercializa-tion has not yet been undertaken. Understandingmore fully the role of firm allying along the entirenew product development process seems particu-larly salient when considering that most innova-tions will either never reach the market (Griffin,1997; Stevens and Burley, 1997), or if they do,they are not likely to meet financial expectations(Booz-Allen & Hamilton, 1982).

Herein, we build on the exploration–exploitationmodel of organizational learning (March, 1991).Koza and Lewin subsequently applied this modelto a firm’s strategic alliances and argued that afirm’s decision to enter an alliance ‘can be distin-guished in terms of its motivation to exploit anexisting capability or to explore for new oppor-tunities’ (Koza and Lewin, 1998: 256). In theearly stages of a development project, a technol-ogy venture undertakes exploratory search in theattempt to discover something new. This search isfrequently structured through exploration alliances

(Rosenkopf and Nerkar, 2001). Following suc-cessful exploration, the venture’s search processturns to exploiting this new knowledge, often inconjunction with a partner firm through exploita-tion alliances (Rothaermel, 2001a). We proposean integrated product development path where atechnology venture’s exploration alliances predictits products in development, while a venture’sproducts in development predict its exploitationalliances, and where its exploitation alliances inturn lead to products on the market (Figure 1).

Given the fact that high-technology start-upsgenerally face resource constraints, it is likely thatthese ventures rely on alliances with establishedfirms for access to capital (Majewski, 1998), inparticular in tight equity markets (Lerner et al.,2003), and for access to product markets (Hill andRothaermel, 2003). Due to their initially weak bar-gaining position, new technology ventures tend tocede a disproportional amount of control rights tothe financier of the R&D alliance (Aghion andTirole, 1994; Lerner and Merges, 1998). Hence, wealso hypothesize that internal resources are substi-tuted for external alliances as a technology venturegrows, implying that firm size negatively moder-ates the product development path leading fromexploration alliances to products on the market.We empirically test this system of new productdevelopment and the suggested moderating effectof firm size on a sample of 325 biotechnology

ExplorationAlliances

ExploitationAlliances

ControlAge ExploitationAlliances

ControlsPatentsEquity RatioFirm AgeTech. DiversityPublic FirmSubsidiaryU.S. DummyFirm Size

Products onMarket

ControlAge ExplorationAlliances

Products inDevelopment

Figure 1. Firm allying and new product development

Copyright 2004 John Wiley & Sons, Ltd. Strat. Mgmt. J., 25: 201–221 (2004)

Exploration and Exploitation Alliances 203

firms that entered into 2565 alliances in the 25-yearperiod between 1973 and 1997.1

EXPLORATION, EXPLOITATION, ANDNEW PRODUCT DEVELOPMENT:THEORY AND HYPOTHESES

Prior research has argued that learning alliancesallow firms to increase the speed of capabilitydevelopment and minimize uncertainty by acquir-ing and exploiting knowledge developed by oth-ers (Grant and Baden-Fuller, 1995; Lane andLubatkin, 1998; Dussauge, Garrette, and Mitchell,2000). From the generation of new ideas throughthe launch of a new product, the creation andexploitation of knowledge is a core theme of thenew product development process. In fact, theentire new product development process can beviewed as a process of embodying new knowl-edge in a product (Madhavan and Grover, 1998).To more fully understand the relationship betweenfirm allying and R&D outcomes, it is prudent,however, to consider the type of learning imbuedinto the project during the various stages of theproduct development process. Different stages ofthe product development process motivate differ-ent types of searches and thus entry into differenttypes of alliances for firms collaborating in themarket for know-how.

We employ March’s (1991) exploration–ex-ploitation framework to characterize the type ofsearch and the type of alliances firms are pur-suing at different stages of the product devel-opment process. A firm’s choice of the type ofalliance to enter can be distinguished by its moti-vation to either explore for new opportunities orexploit an existing opportunity (Koza and Lewin,1998). Exploration generates discovery of newopportunities and, at the same time, the potentialfor exploitation. Thus, successful exploration alsocreates demand for resources required to exploitnewly discovered opportunities. In the spirit of anevolutionary perspective, we argue that exploita-tion alliances are dependent on the firm’s priorexploration activities. Specifically, we propose aproduct development system in which an R&D

1 While this paper is a large-scale empirical study, it has quali-tative antecedents. Our fieldwork in the biotechnology industryenabled us to ground our model of the new product developmentpath and to obtain qualitative data, some of which we draw onthroughout this paper.

organization’s exploratory alliances motivate sub-sequent exploitative alliances. Further, we proposethat new technology ventures that use explorationand exploitation alliances to organize for innova-tion tend to commercialize more products.

The exploration–exploitation framework devel-oped by March (1991) and refined by Levinthaland March (1993) provides a framework for under-standing the needs of technology ventures at dif-ferent stages of the product development process.March (1991) described exploration as ‘experi-mentation with new alternatives’ having returnsthat ‘are uncertain, distant, and often negative’and exploitation as ‘the refinement and exten-sion of existing competencies, technologies, andparadigms’ exhibiting returns that ‘are positive,proximate, and predictable.’ March thus concludedthat ‘the distance in time and space between thelocus of learning and the locus for realization ofreturns is generally greater in the case of explo-ration than in the case of exploitation, as is theuncertainty’ (March, 1991: 85). Later, Levinthaland March (1993: 105) defined exploration as ‘thepursuit of knowledge, of things that might come tobe known,’ and exploitation as ‘the use and devel-opment of things already known.’

An important distinction between explorationand exploitation processes can be based on theirprecursors. The precursor to exploration is simplydesire, the wish to discover something new. Theprecursor to exploitation, however, is the existenceof an exploitable set of resources, assets, or capa-bilities under the control of the firm. Viewed inthis light, exploitation depends upon prior explo-ration. During the early stages of the new prod-uct development process, a firm is prospecting fornew wealth-creating opportunities. During this dis-covery period, the venture pursues an exploratorysearch involving basic research, invention, risk-taking, and building new capabilities with thegoal of developing new knowledge or capabilitieswhich it can subsequently exploit to create value(Cohen and Levinthal, 1990). Once potentiallyvaluable knowledge and skills have been acquiredthrough exploration, the firm then turns to exploita-tion activities. Thus, the exploration–exploitationmodel implies a sequence for the use of theseprocesses by organizations. Exploitation cannotby definition take place without prior exploration(March, 1991). In reality, most firms engage inboth activities simultaneously because they man-age several concurrent projects at different stages



in the product development process. Yet, from atheoretical viewpoint, the exploration–exploitationmodel implies that a firm’s competency that is cur-rently exploited must have been explored at someearlier time.

Subsequent research has linked the explora-tion–exploitation framework to strategic alliances(Koza and Lewin, 1998; Rothaermel, 2001a). Ap-plying Koza and Lewin’s (1998) conceptual under-standing of alliances as motivated by either explo-ration or exploitation, recent empirical researchindicated that a firm’s propensity to enter explo-ration and exploitation alliances is related to theresource endowments of the firm (Park, Chen,and Gallagher, 2002). Building on these insights,we suggest a product development path beginningwith exploration alliances and continuing throughexploitation alliances, which should enhance theability of technology ventures to discover, develop,and commercialize new products. This path impliesa system of alliance utilization beginning withexploration followed by exploitation. Support forthe entire product development system hinges uponthe simultaneous positive relationship betweeneach of the separate links constituting the model.We extend prior research by moving beyond moti-vations for alliance entry by looking for evi-dence of the effectiveness of an integrated explo-ration–exploitation alliance system in the contextof new product development. Only if the com-plete model holds will we have found evidence forbenefits derived from an exploration–exploitationalliance strategy.

Exploration alliances and products indevelopment

Exploration alliances are entered into with themotivation to discover something new; they focuson the ‘R’ in the research and development process(Koza and Lewin, 1998). Envisioned outcomes andpaybacks are distant in time and generally exhibithigh variance. If we view the new product develop-ment process as a knowledge management process,then the hoped for outcome of the exploration pro-cess is the embodiment of new knowledge learnedthrough exploration into a prototype product thatcan be extended into the testing and developmentprocess. Alternatively, exploration alliances maylead to the codification of new knowledge throughpatenting. In the biotechnology industry, for exam-ple, exploration collaborations are motivated by a

desire to acquire basic knowledge that can be usedto create novel molecular entities which are thenentered into the development and regulatory pro-cess. Hence, we argue that exploration alliancespredict products in development. An example ofan exploration alliance is the collaboration betweenthe biotechnology firm Biogen and the Universityof Zurich. Their cooperation led to the discoveryof Intron A, the first product to enter clinical trialsfor the treatment of certain types of leukemia andhepatitis C.

It is the exploration stage of the new productdevelopment process that most previous studieshave examined when linking firm allying to per-formance (Shan et al., 1994; Deeds and Hill, 1996;Baum et al., 2000). In our proposed model, how-ever, we suggest that it is only the first step ofthe new product development process. Once theknowledge gained through exploration has becomeembodied in a prototype, the firm’s attention thenturns to exploitation processes.

Products in development and exploitationalliances

The filing of a patent or the entry of a productinto the development and regulatory process sig-nals that further new knowledge must be accessedand imbued into the product in order to exploit theknowledge gained through exploration. The com-pletion of a prototype product creates an immediateneed for certain complementary capabilities (e.g.,legal and regulatory competence, manufacturing,marketing, and distribution). At this juncture, anentrepreneurial venture must make the decisionto either go it alone or to collaborate with moreestablished firms that then take on the commercial-ization of the new product. If forward integration iscostly, time consuming, and risky, external fundingthrough capital markets may not be a viable option(Lerner et al., 2003). On the other hand, estab-lished firms that have developed competencies inthe downstream activities of the value chain arewell positioned to collaborate with new ventures tocommercialize new products. This collaboration ismotivated by complementary assets and generatesrents through economies of specialization (Teece,1992).

The scenario above describes interfirm coopera-tion between biotechnology firms and establishedpharmaceutical companies at this point in the newproduct development process (Rothaermel, 2001a).



Biotechnology firms focus on the ‘R’ of theresearch and development process, whereas largepharmaceutical companies focus on the ‘D.’ Priorresearch has shown that most alliances betweennew biotechnology firms and established pharma-ceutical companies were initiated when the newdrug candidate was about to enter clinical trials(Pisano and Mang, 1993). A biotechnology drugcandidate reaching the product development stagesignals a major milestone in the development pro-cess, since more than 95 percent of all drug candi-dates will not make it into clinical trials (Giovan-netti and Morrison, 2000). Uncertainty has beendrastically reduced once a product is ready to enterthe development stage. By embodying knowledgecreated through exploration into a prototype prod-uct, a technology venture is able to distinguishitself and to signal the quality of its project. Thisintermediate success generates alliance opportu-nities for the biotechnology firm. For example,when interviewed, H. Stewart Parker, CEO of thebiotechnology firm Targeted Genetics, indicatedthat every time there is an article about TargetedGenetics’ successful product development in theWall Street Journal, the telephone will ring off thehook with pharmaceutical companies calling andoffering alliance opportunities.

Established pharmaceutical companies havelong-standing routines and competencies to man-age a new drug through the regulatory process andthen to market it via their armies of detail peo-ple, which are often 15,000 strong. In addition,large pharmaceutical companies tend to have theresources to finance this most costly and time-consuming part of the development process, andthey are often short of innovative products in theirown research pipelines. Moreover, some empiricalevidence exists showing that pharmaceutical com-panies possess an informational advantage in eval-uating the research efforts of biotechnology firmsover the capital markets (Majewski, 1998; Lerneret al., 2003). It is argued that the existing pharma-ceutical companies generally see a greater poten-tial in the new biotechnology than capital mar-kets, and thus they apply a smaller discount rateon capital when funding biotechnology research.This in turn implies that pharmaceutical compa-nies tend to be relatively cheaper sources of capitalfor biotechnology firms than the capital markets.Taken together, we argue that a technology ven-ture’s products in development indicate the needfor complementary assets and may create access

to them through exploitation alliances with estab-lished companies.

Exploitation alliances and products on themarket

Exploitation alliances focus on the ‘D’ in theresearch and development process and are enteredinto with the goal to join existing competen-cies across organizational boundaries in order togenerate synergies, which are then shared acrossthe partners (Koza and Lewin, 1998). Exploita-tion alliances can be characterized by the unionof complementary assets (Teece, 1986). Success-ful exploitation enables the firm to commercializethe knowledge gained through exploration. Newbiotechnology firms often focus on creating newdrugs, which are then commercialized by estab-lished pharmaceutical companies.

Above, we described how the biotechnologyfirm Biogen has used an exploration alliance withthe University of Zurich to discover Intron A,the first biotechnology drug for the treatment ofleukemia and hepatitis C to reach clinical devel-opment. Subsequently, Biogen decided to commer-cialize this innovative drug through an exploita-tion alliance with the pharmaceutical companySchering-Plough. In particular, Biogen enteredan exclusive licensing agreement with Schering-Plough, which took on the clinical trials andregulatory activities of the product as well asits marketing, distribution, and sales. This exam-ple shows that Biogen was able to discover anddevelop a new drug through an exploration alliancewith a university, which it then commercializedthrough an exploitation alliance with an establishedpharmaceutical company. In sum, we argue thatexploitation alliances are one possible organizingmode to commercialize new products and thus pre-dict products on the market.

Firm allying and new product development

We presented each individual link of our proposedproduct development path which indicated threeseparate hypotheses. However, we would like toemphasize that we set out to advance an integratedproduct development system. This implies thatthe linkages between each step in the productdevelopment path must hold simultaneously toprovide support for our model, which explains theprocess of embodying knowledge into products



from discovery to market commercialization viainterfirm cooperation. This theoretical model of theproduct development path warrants a system-levelhypothesis.

Hypothesis 1: There exists a system of newproduct development linking exploration alli-ances to products on the market with explorationalliances predicting products in development,products in development predicting exploitationalliances, and exploitation alliances predictingproducts on the market.

Figure 1 depicts our research model and summa-rizes our hypothesized product development pathbeginning with a technology venture’s explorationalliances and ending with products on the mar-ket. The model also depicts a variety of controlvariables that may impact the proposed productdevelopment process. In the methods section, wediscuss each variable in detail.

Moderating effect of firm size on new productdevelopment path

At each stage of the proposed product developmentpath, a technology venture must decide to eitherforwardly integrate or to collaborate. If marketswere efficient, firms would have no reason to ver-tically integrate. Each organization would focus onits respective competencies in a different stage ofthe product development path, and products wouldbe commercialized through transactions in the mar-ket for know-how. All participants could benefitfrom economies of specialization and each organi-zation would fully extract the rents reflecting theirvalue added in the product development process.However, transaction cost economics has advancedtheoretical arguments that propose that marketsmay not always function according to the marketefficiency hypothesis (Williamson, 1985). Follow-ing these arguments, Pisano (1997) has suggestedthat the market for know-how, which emergeswhen different organizations possess complemen-tary competencies, may not always function prop-erly.

When analyzing our proposed product develop-ment path, several potential frictions in the marketfor know-how that can inhibit an efficient marketare noteworthy. Sources of market imperfectionsinclude the challenges of negotiating and enforcingcontracts as well as the risks and costs associated

with making specialized investments in the face ofuncertainty (Williamson, 1985). Frictions can alsoarise from appropriability problems when intellec-tual property rights are not fully specified and suffi-ciently protected (Teece, 1986) or by attempting todevelop and transfer tacit knowledge across orga-nizational boundaries (Lane and Lubatkin, 1998).Empirical work provided support for the notionthat firms tend to vertically integrate when thecosts of transacting in the market exceed those ofvertical integration (Klein, Crawford, and Alchian,1978; Monteverde and Teece, 1982; Pisano, 1990).

Exploration alliances are exposed to the chal-lenge of negotiating and structuring contracts in theface of uncertainty, which frequently causes fric-tions between the partners over intellectual prop-erty. Exploration alliances are further exposed tothe difficulty of coordinating and transferring tacitknowledge across the partner organizations. Manytechnology ventures get their start in universities,which either spin off new ventures, or scientistsbecome entrepreneurs, often in combination withventure capitalists. In their early stages, those newventures generally rely on exploration cooperationswith the organizations they originated from. Dis-putes over intellectual property rights are quitefrequent at this stage as the universities and newtechnology ventures argue over who owns whatrights.

For example, the founding of Genentech in1976, later the first public biotech firm, was basedon the idea of commercializing scientific break-throughs accomplished at Stanford University andthe University of California. Incidentally, HerbertBoyer, one of the scientists at the University ofCalifornia who was involved in the revolutionaryscientific breakthroughs, was also a co-founder ofGenentech. University scientists becoming, in oneform or another, involved in commercially drivenbiotechnology firms is commonplace in this indus-try (Zucker, Darby, and Armstrong, 2002). Suchclose involvement of universities and biotechnol-ogy firms often end up in legal disputes over whoowns the rights to certain patents, as illustrated inthe University of California vs. Genentech legaldispute (Managing Intellectual Property, 1999).Thus, as technology ventures grow, they mightprefer in-house exploration for certain projects toavoid being exposed to the hazards of allying inthe market for know-how.

In a similar manner, exploitation alliances mayalso be exposed to hazards stemming from disputes



over intellectual property rights and to hazardsresulting from necessary investments in specializedassets. Legal disputes over patent infringementsare quite frequent at this stage of the product devel-opment process, as the new ventures license theirtechnology to established firms that take on thecommercialization. For example, the first biotech-nology drug to reach the market was Humulin, ahuman insulin, which was discovered and devel-oped by the biotechnology firm Genentech andcommercialized by the pharmaceutical companyEli Lilly. However, Genentech later sued Lilly,accusing that it misused materials provided byGenentech to commercialize recombinant humaninsulin. Further, necessary investments in special-ized assets to develop the new technology exposesboth the new venture and the firm that commer-cializes the technology to opportunistic behavior(Williamson, 1985).

Aghion and Tirole (1994) developed a theoreti-cal model to analyze the organization of researchand the allocation of ownership for an innova-tion between a research firm and a customer firm,the financier of the research. The Aghion-Tirolemodel led to two important predictions of interestto our research. First, to maximize research effi-ciency, and thus joint value creation, control rightsshould be assigned to the research firm when-ever the value of the final output depends moreon the marginal efficiency of the research effortthan on the marginal impact of the financial invest-ment. Second, a cash constraint on the part of theresearch firm causes an inefficient outcome due tothe financier’s use of its bargaining power to retaingreater ownership. In essence, if research firms,such as biotechnology ventures, face cash con-straints that limit their bargaining power, then theirestablished alliance partners are able to use theirfinancial power opportunistically to drive down theprice of the research and to gain greater ownershipat the expense of the research firm.

Above, we argued that technology ventures, dueto their initial resource constraints, engage in anexploration and exploitation alliance strategy whenembodying new knowledge throughout the productdevelopment process by transforming discoveriesinto commercialized products. However, as pre-dicted by Aghion and Tirole (1994) and empiri-cally supported by Lerner and Merges (1998) andLerner et al. (2003), financially constrained firmstend to give up too much ownership of the innova-tion when entering an alliance. Lerner and Merges

concluded that ‘the most profound effect on theallocation of control rights, at least in technologyalliances . . ., is the financial condition of the R&Dfirm, rather than mutual concern about maximiz-ing joint value’ (Lerner and Merges, 1998: 153).Alliances, as incomplete contracts, inherently posecertain contracting hazards, which appear to beexacerbated by firm financial constraints.

Moreover, technology ventures generally pos-sess an informational advantage in evaluating thequality of their development projects (Lerner et al.,2003). This situation can lead to a lemons prob-lem in the market for know-how (Akerlof, 1970;Pisano, 1997).2 In particular, a technology venturegenerally pursues several concurrent projects, anddue to its intimate familiarity with the projects builtover long periods of time it has a reasonable under-standing of which projects show the most promise.Since the established firm attempting to commer-cialize a project offered by the technology venturehas no such insider information to judge whether aproject offered for collaboration is a promising oneor not, it will discount the project to hedge againstlemons and thus offer a reduced deal to the newventure.

As Akerlof’s example of used cars has demon-strated, information asymmetry can lead to theperverse effect that sellers will only offer lemonsif the price the buyers offer is below the valuethe sellers attach to good cars. Applying this anal-ogy to the market for collaborative know-howin biotechnology, Pisano (1997) found empiricalsupport for a lemons problem in the market fordrug development projects reaching clinical tri-als. As a new venture grows and accrues inter-nal resources to finance its high prospect projects,it will tend to keep them in-house vs. develop-ing them through alliances with a larger partner.Thus, a lemons problem, in combination with thedownward pricing pressure in the market for col-laborative know-how due to the weak bargaining

2 The following briefly summarizes the lemons problem (Akerlof,1970). In the market for used cars, only two types of cars aresold: good cars and bad cars (lemons). Good cars are worth$8000 and bad ones are worth $4000. Moreover, only the sellerknows if his/her car is a good one or a lemon. Assuming themarket supply is split equally between good and bad cars, thenthe ex ante probability of buying a lemon is 50 percent. Buyersare aware of the general possibility of buying a lemon and thuswould like to hedge against it. Therefore, they include a discountand offer $6000 for a used car. This discounting strategy has theperverse effect of crowding out good cars if the sellers perceivetheir value to be above $6000. Assuming that to be the case, allthat are left in the market for used cars will be lemons.



position of the numerous small, under-financedresearch firms, creates a situation in which evena well-financed research firm has an incentive towithhold promising projects, despite its improvedbargaining position.

The gradual tendency to substitute internal re-sources for external ones as the technology ven-ture grows is also supported by the theoreticalpredictions of the optimal capital structure model(Myers, 1984; Myers and Majluf, 1984). Technol-ogy ventures generally find themselves strugglingto acquire resources from skeptical investors whohave difficulty judging the quality of a specificproject. According to the pecking order theory ofoptimal capital structure, a firm’s preferences forobtaining capital will follow a hierarchy with apreference for internal over external sources ofcapital because of asymmetric information and sig-naling problems. Therefore, firms allocate theiravailable internal resources first before seekingexternal resources for the remainder. The strengthof the preference for internal resources to financepromising projects is based on the costs associ-ated with information asymmetries between thefirm and external resource providers (Myers andMajluf, 1984; Shyam-Sunder and Myers, 1999)and the risks of expropriation of knowledge due toopportunistic actions by the partner (Williamson,1985).

Initially, new technology ventures with smallresource endowments and private informationabout valuable projects will be forced to exchangeownership of their projects at a suboptimal price(Aghion and Tirole, 1994; Lerner and Merges,1998). This implies that firms with constraintresources will be forced to be over-reliant onexternal resources in the new product developmentprocess, leaving them open to be undervalued andto the risks of having their core knowledge assetsexpropriated. Recent empirical work providessome evidence for this notion with the finding thata substantial amount of the value created throughsmall firm–large firm alliances was appropriatedby the larger partners (Rothaermel, 2001b).Moreover, alliances between small R&D firmsand large established firms, where more controlrights were given to the large firm that financedthe research, tended to be less successful andwere more likely to be renegotiated when equitymarkets became more favorable (Lerner et al.,2003). Taken together, as a technology venturegrows and acquires more resources, it will tend to

minimize the risk of expropriation by acting on itspreference to retain promising projects for in-housedevelopment, and thus commensurately decreaseits reliance on strategic alliances to discover,develop, and commercialize new products.

Hypothesis 2: The product development pathleading from exploration alliances to productson the market is moderated negatively by firmsize.

METHODS

Research setting

The research setting is the biotechnology indus-try. The emergence of biotechnology can be inter-preted as a radical process innovation that brokethe barriers of entry into the pharmaceutical indus-try, among other industries (Pisano, 1990). Sincethe early 1970s, about 1600 new biotechnologyfirms have emerged to commercialize this tech-nological breakthrough. The commercialization ofbiotechnology is characterized by extensive inter-firm cooperation. Indeed, the biotechnology indus-try has been identified as the industry with thehighest alliance frequency among several indus-tries characterized by high alliance activity (Hage-doorn, 1993).

The drug discovery and development is fraughtwith extremely high uncertainty. The entire pro-cess may extend more than 15 years and can costover $500 million for a single drug.3 The odds ofa discovered molecule succeeding in the develop-ment process are extremely low. For every 10,000compounds screened, 250 (2.5%) so-called leadcandidates make it into preclinical testing. Out ofthose lead candidates, five (2%) enter clinical test-ing, 80 percent pass phase I, 30 percent pass phase

3 The drug discovery and development process can be brokendown into distinct sequential stages (Giovannetti and Morrison,2000: 46–47). The discovery stage can take anywhere between2 and 10 years. In the next stage, which can take up to 4 years,a lead drug candidate is developed and pre-clinical testing isundertaken. A lead candidate then enters phase I of clinicaltesting, which can take up to 2 years. In this phase, the leadcandidate is administered to 20–30 healthy volunteers and itssafety and dosage are evaluated. In phase II, which can take upto 2 years, the drug is given to 100–300 patient volunteers tocheck for efficacy and side effects. In phase III, which can takeup to 3 years, the drug is administered to 1000–5000 patientvolunteers to monitor reactions to long-term drug usage. Thenext stage, FDA review and approval, can take up to 2 years.This is followed by a 2-year post-marketing testing period.



II, and 80 percent pass phase III of the clinical tri-als. Thus, for every 10,000 compounds screened,one drug will be approved by the FDA (Giovan-netti and Morrison, 2000). This implies that the exante probability for a molecule to develop into acommercialized drug is 0.01 percent.

Those numbers also imply that exploration andexploitation alliances both carry uncertainty withrespect to their potential outcome. When studyingthe impact of equity financing cycles on the per-formance of collaborative R&D alliances betweensmall biotechnology firms and their larger part-ners, Lerner et al. (2003: 434) found that only 14percent out of a sample of 200 randomly drawnalliances begun since January 1980 resulted in anapproved drug by December 1998. When consid-ering only alliances that were entered when thedrug development had already progressed to eitherphase I or phase II of clinical trials, the reportedsuccess rate was about 26 percent. These num-bers indicate that, on the average, the majority ofall alliance projects, regardless whether they focuson exploration or exploitation, will not result incommercialized products. They also indicate thatuncertainty declines as the project moves along inthe product development process, thus explorationalliances generally entail higher uncertainty thanexploitation alliances.

Data and sample

We identified all new biotechnology firms fullydedicated to human therapeutics listed in BioScan,4

i.e., all firms that were engaged in developingin vivo therapeutics. This segment of the biotech-nology industry comprises new biotechnology firmsengaged in the research, development, and com-mercialization of therapeutics that are placed insidethe human body (in vivo) as opposed to in vitrotherapeutics that are used outside the human body.We limited our sample to in vivo therapeutics sincethe firms engaged in this segment of biotechnologyare exposed to extensive regulatory requirements(e.g., FDA), which bring with them detailed report-ing of products under development. Focusing onhuman therapeutics also allowed us to create a

4 BioScan, which is published by American Health Consultants,provides one of the most comprehensive publicly availabledirectories covering the global biotechnology industry. It hasbeen used in a number of prior studies (cf. Shan et al., 1994;Powell, Koput, and Smith-Doerr, 1996; Lane and Lubatkin,1998; Rothaermel, 2001a).

homogeneous sample, while controlling for indus-try idiosyncrasies. This process yielded a sampleof 325 biotechnology firms.

In the next step, we obtained each firm’salliance history. BioScan lists detailed qualitativeinformation about each of the firms’ alliances,such as the focal firm’s partners, the month andyear when the alliance was entered, whether thealliance is governed by an equity or contractualarrangement, and what area of the industryvalue chain it covers (research, drug discovery,development, clinical trials, FDA regulatoryprocess, marketing and sales). We based ourclassification scheme concerning different alliancetypes on Koza and Lewin’s notion that a firm’smotivation to enter an alliance is driven bythe desire ‘to exploit an existing capability orto explore for new opportunities’ (Koza andLewin, 1998: 256).5 Firms enter explorationalliances to discover something new jointly withan alliance partner, while exploitation alliancesare ‘associated with increasing the productivityof employed assets—improving and refiningexisting capabilities and technologies’ (Kozaand Lewin, 1998: 256). Thus, in classifying abiotechnology firm’s alliances as either explorationor exploitation, we focused on the motivationfor and activities of the alliance as specified bythe alliance partners. Our classification systemalso corresponds to that employed by Parket al., who, when studying alliance formation bysemiconductor start-ups, categorized explorationalliances as those with a joint research anddevelopment component, and ‘alliances orientedtoward exploiting existing resources . . . wereclassified as exploitation alliances’ (Park et al.,2002: 534). Accordingly, we coded each of thebiotechnology firms’ alliances that focused onbasic research, drug discovery and developmentas exploration alliances, and alliances that weretargeted towards commercialization (clinical trials,

5 While we followed the theoretical work by Koza andLewin (1998), and focused on the intent or motivationof an alliance in categorizing it either as exploration orexploitation, other scholars have advanced different definitions.For example, Rosenkopf and Nerkar (2001) use organizationaland technological boundaries to define four different types ofexploration. Accordingly, alliances go beyond local search asthey are boundary spanning, and thus are considered explorationactivities.



FDA regulatory process, and marketing and sales)were coded as exploitation alliances.6

Measures

Our hypothesized product development path pro-poses linkages between four key variables: explo-ration alliances, products in development, exploita-tion alliances, and products on the market. Weoperationalized a firm’s products in developmentas a count variable of each firm’s biotechnologyproducts in development that have successfullyentered clinical trials but have not yet reachedthe market for pharmaceuticals. A firm’s prod-ucts on the market is a count variable of eachfirm’s biotechnology products that successfullycompleted all stages of the product developmentprocess and are now commercialized. A firm’sexploration alliances is a count variable of itsalliances that focus on the upstream activities ofthe value chain (basic research, drug discoveryand development). Conversely, a firm’s exploita-tion alliances is a count variable of its alliancesthat focus on the downstream activities of the valuechain (clinical trials, FDA regulatory process, andmarketing and sales).

We employed several control variables that mayimpact upon the proposed new product develop-ment path (Figure 1). We controlled for the aver-age age of a firm’s exploration and exploitationalliances in months since older alliances are morelikely to yield products in development and on themarket than younger alliances. We controlled fora firm’s innovativeness through including a countvariable of its patents received between 1991 and1995. Shan et al. (1994) and Baum et al. (2000)proxied a firm’s innovativeness in a similar man-ner.7 A 5-year window for patenting attenuatesannual fluctuations and thus may capture a biotech-nology firm’s patenting propensity more accuratelysince numerous firms in the sample are small firms

6 A second researcher coded independently 10 percent of thesample to assess inter-rater reliability, which we found to be0.94. This is well above the conventional cut-off point of 0.70(James, 1982; Cohen and Cohen, 1983).7 Theoretically, a quality weighted measure of patenting (e.g.,adjusted by citations) provides an alternative to assess a firm’sinnovativeness. However, the recency of the emergence biotech-nology in combination with the patent citation time lag madethis approach infeasible. A raw count of patents provides a rea-sonable proxy since prior research has shown that a firm’s rawpatent count is highly correlated with the quality of its patents(Stuart, 2000).

that do not receive many patents per year, if any.Further, based on the time lag between taking stockof a firm’s patenting and its new product develop-ment, exploitation alliances, and products on themarket, it is reasonable to assume that a firm’spatenting propensity may have an influence onthem. Using a 5-year time window is also con-sistent with prior research attempting to proxy afirm’s innovativeness (Stuart and Podolny, 1996;Ahuja, 2000).8 We obtained the patenting datafrom the U.S. Patent and Trade Mark Office.

We included a ratio of a firm’s equity alliancesover its total alliances to control for the impact of afirm’s preference for equity vs. nonequity allianceson the product development path. We controlledfor firm age, assuming that older firms are morelikely to have more products in development andhave entered more alliances (Sørensen and Stuart,2000). We also controlled for each firm’s degree oftechnological diversity through the inclusion of acount variable representing the number of biotech-nology subfields in which the firm participated(Shan et al., 1994). We further controlled for theownership status of the firm (1 = public firm) andwhether the firm was a subsidiary or independent(1 = subsidiary). We included a dummy variableto distinguish between U.S.-based and non-U.S.biotechnology companies (1 = U.S. firm) to con-trol for institutional differences (Hennart, Roehl,and Zietlow, 1999). Finally, we controlled for firmsize by using the number of employees as a proxy.Firm size is often measured in revenues or marketshare; however, most biotechnology firms do nothave a positive revenue stream at this point. Thus,measuring firm size in terms of employees pro-vides a reasonable alternative (Shan et al., 1994).

Estimation procedure

We propose a theoretical model of an explo-ration–exploitation alliance strategy in new prod-uct development that includes three dependentvariables: products on the market, exploitationalliances, and products in development (Figure 1).Two of the four variables describing the proposedproduct development path (products in develop-ment and exploitation alliances) are at the sametime dependent variables as well as indepen-dent variables. This indicated testing Hypothesis

8 We also conducted a robustness check applying alternative timelags and time windows. All significant relationships remainedrobust.



1 using structural equation modeling (Andersonand Gerbing, 1988). This approach is appropri-ate here since it allows us to test a system ofstructural equations, where a dependent variablein one relationship becomes an independent vari-able in the subsequent relationship. We estimatedthe following recursive system describing the pro-posed product development path by a maximumlikelihood procedure for structural equation mod-els (Bentler, 1995):

Products on the market= f (Exploitation alliances, Controls)

Exploitation alliances= f (Products in development, Controls)

Products in development= f (Exploration alliances, Controls)

To assess the threat of reverse causality, weapplied two separate procedures. First, we esti-mated an alternative recursive model with thestructural equations describing the reverse pathleading from products on the market, via exploita-tion alliances and products in development, toexploration alliances. Here, we found that thismodel did not provide an acceptable overall fit withthe data; moreover, the path itself was not signifi-cant. Second, we applied a two-stage least squaresregression model to assess reverse causality (Shanet al., 1994; Greene, 1997). The results indicatedthat the product development path runs from explo-ration alliances to products in development, whilethe reverse causality was not supported.

The interaction effects stipulated in Hypothesis2 were tested using a negative binomial regres-sion model with a maximum likelihood procedure.The negative binomial regression model treats thedependent variables of interest (products in devel-opment, exploitation alliances, and products on themarket) as count variables, while estimating het-erogeneity. This relaxes the restrictive assumptionof mean and variance equality inherent in the Pois-son model and also accounts for omitted variablebias (Walker, Kogut, and Shan, 1997).

We tested for moderation through the inclusionof the independent variables and the interactionterms between the respective independent variablesin the same regression model. Such a moderatedregression approach is a conservative method forexamining interactions since the interaction term istested for significance after all other direct effects

are controlled (Aiken and West, 1991). Moreover,testing for a negative moderation effect of firmsize on the product development path requires sig-nificant findings for each link and thus supportfor three different interaction terms. We standard-ized the variables, prior to creating the interac-tion terms, to improve their interpretability and toreduce the threat of multicollinearity (Aiken andWest, 1991).

RESULTS

The average biotechnology firm in our sample hasentered into three exploration and five exploita-tion alliances, has one product on the market, fiveproducts in development, holds five patents, has161 employees, is about 10 years old, and par-ticipates in six different technological subfields.Sixty-nine percent of the firms are public, while7 percent of the firms are subsidiaries. Seventy-eight percent of the firms are U.S. based. About4 percent of all alliances are equity arrangements.The 325 firms in our sample entered a total of 2565alliances in the 25-year period between 1973 and1997. These 2565 alliances split into 1072 (42%)exploration alliances and 1493 (58%) exploitationalliances. The number of exploitation alliances issignificantly larger than the number of explorationalliances (p < 0.001). A descriptive statistic ofthe variables as well as a correlation matrix canbe found in Table 1, while Figure 2 and Table 2depict the results.

Figure 2 depicts the structural relationships a-mong the theoretical constructs for testing Hypoth-esis 1. While we estimated the structural equationmodel including all of the relationships presentedin Figure 1, Figure 2 only shows the significantrelationships for the sake of visual clarity. Resultsfrom structural equation models are evaluated withrespect to their model fit (Byrne, 1994). The modelproposed in Figure 1 provides an acceptable over-all model fit with the sample data as indicated inthe following statistics. The likelihood ratio chi-square statistic is 117.92 for 47 degrees of free-dom. This implies that the chi-square statistic forone degree of freedom is 2.51 (= 117.92/47), wellbelow the significant chi-square of 3.84 for p <

0.05 (and also below the significant chi-square of2.71 for p < 0.10). A nonsignificant chi-square isdesired as it indicates that the model is not sig-nificantly different from the underlying data. In



Tabl

e1.

Des

crip

tive

stat

istic

san

dco

rrel

atio

nm

atri

x

Mea

nS.

D.

12

34

56

78

910

1112

13

1.Pr

oduc

tson

mar

ket

1.21

2.24

2.E

xplo

itatio

nal

lianc

es4.

595.

660.

493.

Age

expl

oita

tion

allia

nces

34.6

026

.28

0.12

0.24

4.Pr

oduc

tsin

deve

lopm

ent

5.14

4.06

0.17

0.43

0.05

5.E

xplo

ratio

nal

lianc

es3.

293.

750.

280.

520.

040.

436.

Age

expl

orat

ion

allia

nces

27.0

323

.32

0.01

0.19

0.25

0.16

0.31

7.Pa

tent

s4.

8813

.95

0.50

0.61

0.12

0.34

0.51

0.15

8.E

quity

ratio

0.04

0.16

0.01

−0.0

1−0

.11

0.09

0.04

−0.0

50.

029.

Firm

age

9.60

4.61

0.38

0.33

0.34

0.18

0.14

0.22

0.35

−0.0

310

.Te

chno

logi

cal

dive

rsity

6.22

4.71

0.29

0.45

0.08

0.32

0.45

0.21

0.41

0.01

0.27

11.

Publ

icfir

m0.

690.

460.

040.

180.

020.

220.

100.

030.

10−0

.03

0.17

0.06

12.

Subs

idia

ry0.

070.

270.

040.

010.

16−0

.07

0.02

0.06

−0.0

3−0

.05

0.05

0.03

−0.1

813

.U

.S.

firm

0.78

0.42

−0.0

90.

12−0

.02

0.14

0.09

0.05

0.10

0.05

0.01

0.04

0.11

−0.0

414

.Fi

rmsi

ze16

1.17

573.

190.

550.

580.

090.

360.

580.

070.

700.

030.

230.

400.

090.

010.

06

n=

325.



Public Firm U.S. Firm Firm Size Patents

Equity Ratio Firm Age

1.328**(0.433) 0.749†

(0.460)

0.001†

(0.001)

-0.800***(0.226)

0.001*(0.001)

0.001***(0.000)

0.014†

(0.010)

0.100***(0.028)

0.277***(0.068)

0.909**(0.306)

0.100*(0.047)

2.568†

(1.610)-3.475†

(2.305)

0.092†

(0.061)0.118***(0.022)

ExplorationAlliances


ExploitationAlliances

Products onMarket

TechnologicalDiversity

0.086***(0.023)

Model Fit Assessment: Normed Fit Index (NFI): 0.902;

Comparative Fit Index (CFI): 0.936;

LISREL Goodness of Fit (GFI): 0.951;

Root Mean Error of Approximation (RMSEA): 0.068; † p < 0.10; * p < 0.05; ** p < 0.01; *** p < 0.001;

Standard Errors in Parentheses.

Figure 2. Structural equation modeling results

general, a value for the ratio of chi-square overdegrees of freedom of less than 3.0 indicates agood fit (Carmines and McIver, 1981). In addition,several other evaluations are important in assessingthe fit of the model: the normed fit index (NFI), thecomparative fit index (CFI), and the LISREL good-ness of fit index (GFI) should each be greater than0.90, and the root mean error of approximation(RMSEA) should be smaller than 0.10 for a goodfitting model (Byrne, 1994). Our results reveal thatthe NFI is 0.902, the CFI is 0.936, the GFI is0.951, and the RMSEA is 0.068 with a 90 per-cent confidence interval of (0.053, 0.083). We alsoconducted a Lagrange multiplier test and foundthat no alternative specification of the parameterswould have led to a model that better representedthe data.

Hypothesis 1 suggests an integrated new prod-uct development path in which a high-technologyfirm’s exploration alliances predict its productsin development, which in turn predict the firm’sexploitation alliances, and they in turn predict thefirm’s products on the market. The results depicted

in Figure 2 provide support for the proposed newproduct development path. The results indicatethat a firm’s exploration alliances are significantin predicting the firm’s products in development(p < 0.001), while a firm’s products in develop-ment in turn are a significant predictor of the firm’sexploitation alliances (p < 0.01), and a firm’sexploitation alliances are significant in predictinga firm’s products on the market (p < 0.001).

The results further indicate that a firm’s degreeof technological diversity is positively associatedwith its products in development (p < 0.05) andthat public firms tend to have more products indevelopment (p < 0.01). We also find that firmsthat have more equity alliances, that are U.S.based, and that are larger in size tend to havemore products in development (p < 0.10). Addi-tional results demonstrate that a firm’s patents(p < 0.001) and size (p < 0.05) are significant inpredicting the number of its exploitation alliances.We also find that a firm’s age is marginally signif-icant in predicting its exploitation alliances. Fur-ther, a firm’s preference for equity alliances is



Table 2. Regression results

Dependent variable Products indevelopment

Model 1

Products indevelopment

Model 2

ExploitationalliancesModel 3

ExploitationalliancesModel 4

Products onmarket

Model 5

Products onmarket

Model 6

Intercept 1.587∗∗∗ 1.623∗∗∗ 1.386∗∗∗ 1.169∗∗∗ −0.082 −0.036(0.038) (0.038) (0.041) (0.086) (0.083) (0.089)

Patents 0.042 0.118∗ 0.127∗ 0.111∗ 0.067 0.111(0.052) (0.057) (0.066) (0.061) (0.101) (0.106)

Equity ratio 0.073∗ 0.053† −0.060† −0.086∗ −0.028 −0.028(0.038) (0.036) (0.054) (0.053) (0.088) (0.089)

Firm age 0.017 −0.025 0.157∗∗ 0.136∗∗ 0.455∗∗∗ 0.407∗∗∗

(0.042) (0.043) (0.053) (0.051) (0.085) (0.088)Technological diversity 0.135∗∗∗ 0.056† 0.146∗∗ 0.096∗ 0.024 −0.119†

(0.042) (0.042) (0.049) (0.048) (0.087) (0.095)Public firm 0.137∗∗∗ 0.117∗∗∗ 0.163∗∗∗ 0.138∗∗ −0.058 −0.151∗

(0.041) (0.039) (0.050) (0.049) (0.085) (0.088)Subsidiary −0.036 −0.047† 0.003 0.002 0.082 0.092†

(0.041) (0.039) (0.048) (0.047) (0.079) (0.083)U.S. firm 0.082∗ 0.075∗ 0.083∗ 0.042 −0.244∗∗∗ −0.281∗∗∗

(0.040) (0.038) (0.049) (0.048) (0.077) (0.078)Firm size 0.075† 0.0459∗∗∗ 0.091† 0.654∗∗ 0.240∗∗ 1.002∗∗

(0.051) (0.120) (0.062) (0.267) (0.104) (0.304)Age exploration alliances 0.028

(0.042)Exploration alliances 0.181∗∗∗

(0.048)(Exploration alliances × Size) −0.103∗∗∗

(0.023)Products in development 0.049∗∗∗

(0.013)(Products in development × Size) −0.039∗∗

(0.017)Age exploitation alliances −0.091

(0.088)Exploitation alliances 0.302∗∗

(0.099)(Exploitation alliances × Size) −0.190∗∗

(0.062)Log likelihood −814.36 −795.85 −792.14 −781.21 −442.78 −432.07Chi-square 347.15∗∗∗ 384.19∗∗∗ 876.50∗∗∗ 898.36∗∗∗ 102.14∗∗∗ 123.56∗∗∗

Improvement over base (�χ 2) 37.04∗∗∗ 21.86∗∗∗ 21.42∗∗∗

† p < 0.1; ∗ p < 0.05; ∗∗ p < 0.01; ∗∗∗ p < 0.001; standard errors in parentheses

negatively associated with its exploitation alliances(p < 0.10). The results further indicate that afirm’s age (p < 0.001) and size (p < 0.001) aresignificant positive predictors of a firm’s productson the market, while a firm’s patents are marginallysignificant in predicting a firm’s products on themarket. Finally, U.S. firms tend to have signifi-cantly fewer products on the market (p < 0.001)than their non-U.S. competitors.

Further insights can be gleaned from compar-ing the results for the same parameters across thethree equations. Firms that hold more patents tend

to enter more exploitation alliances and have agreater number of products on the market, althoughpatents are nonsignificant in predicting the firm’sproducts in development. Firms with a highernumber of equity alliances tend to have moreproducts in development but enter fewer exploita-tion alliances. A preference for equity alliances isinsignificant in predicting a firm’s products on themarket. Older firms tend to have more exploita-tion alliances and products on the market, whilefirm age is nonsignificant in explaining a firm’sproducts in development. Firms that diversified in



a larger number of technological subfields tend tohave more products in development, although afirm’s diversity has no significant impact on eitherthe number of its exploitation alliances or its prod-ucts on the market. Public firms tend to have moreproducts in development, but the ownership sta-tus of the firm is not relevant when predicting itsexploitation alliances or products on the market.U.S. firms have on the average more products indevelopment but fewer products on the market.Nationality is insignificant in predicting exploita-tion alliances. Finally, larger firms tend to havemore products in development, more exploitationalliances, and more products on the market.

Hypothesis 2 advances the notion that the pathleading from exploration alliances to products onthe market is moderated negatively by firm size.Testing this hypothesis implied an investigationof three different interaction terms. More specif-ically, we needed to test whether each of thelinks between (1) exploration alliances and prod-ucts in development, (2) products in developmentand exploitation alliances, and (3) exploitationalliances and products on the market are moderatednegatively by firm size. The results are presentedin Table 2. Models 1, 3, and 5 are the three respec-tive baseline models. Models 2, 4, and 6, eachof which represent a significant improvement overtheir respective baseline model (p < 0.001), depictthe results for the size interaction effects. Model2 reveals that the interaction between a technol-ogy venture’s exploration alliances and its sizeis negative and significant (p < 0.001), indicatingthat as the venture becomes larger, its explorationalliances become less important in predicting itsnew product development. Model 4 shows thatthe interaction between a venture’s products indevelopment and its size is negative and signif-icant (p < 0.01). This implies that as the ven-ture grows, its new product development becomesless critical in explaining the venture’s exploita-tion alliances. The results from Model 6 demon-strate that the interaction between a venture’sexploitation alliances and its size is also negativeand significant (p < 0.01), indicating that as theventure becomes larger, its exploitation alliancesbecome less crucial in explaining its products onthe market.

Taken together, we find that all three hypothe-sized size interaction effects are negative and sig-nificant. This provides support for Hypothesis 2and suggests that the product development path

leading from exploration alliances to products onthe market is moderated negatively by firm size.Both types of alliances become less relevant for thefirm’s new product development as the technologyventure accrues more internal resources. More-over, the exploitation–size interaction effect (β =−0.190) is significantly more negative than theexploration–size interaction effect (β = −0.103;p < 0.05), which implies that exploitation alli-ances tend to lose their relevance faster than explo-ration alliances as the technology venture grows insize.

To gain further insights into the nature of themoderation effects between firm size and the dif-ferent types of alliances on products in develop-ment and on the market, we plotted the inter-actions based on the results obtained in Mod-els 2 and 6 (Aiken and West, 1991). Figure 3(a)reveals a negative relationship between firm sizeand products in development for a high numberof exploration alliances. A similar negative rela-tionship is depicted in Figure 3(b) when applyingproducts on the market as a dependent variable.Comparing both figures reveals that exploitationalliances appear to be more critical than explo-ration alliances for smaller firms considering theperformance gap for firms with a low number vs. ahigh number of alliances. Moreover, Figures 3(a)and 3(b) indicate a positive relationship betweenfirm size and products in development or productson the market for a low number of the respectivealliance type. The best-performing larger firms pur-sue only a low number of exploration and exploita-tion alliances. This implies that larger firms tendto use vertical integration for some developmentprojects since the results from the direct effectsrevealed that larger firms have significantly moreproducts in development and on the market.

DISCUSSION

We found support for an integrated product devel-opment path leading from exploration alliances,via products in development and exploitationalliances, to products on the market. On the aver-age, new technology ventures that use an explo-ration–exploitation strategy in their product devel-opment efforts tend to have more products indevelopment and on the market. However, we alsofound that this product development path is moder-ated negatively by firm size. It appears that as the



High


Low

High

Products onMarket

Low

Low ExplorationAlliances

Low ExploitationAlliances

High ExplorationAlliances

High ExploitationAlliances

Low Firm Size(a) High Firm Size

Low Firm Size(b) High Firm Size

Figure 3. Interactions between alliance type, firm size, products in development, and products on market.(a) Exploration alliances, firm size, and products in development. (b) Exploitation alliances, firm size, and products

on market

technology ventures grow, they are in a position toretain their most promising projects for in-houseexploration and exploitation. This interpretationresonates with Pisano’s (1997) finding that collab-orative development projects between biotechnol-ogy and pharmaceutical companies had a higherprobability of termination than projects pursued bythe biotechnology ventures alone through verticalintegration.

The results suggest an understanding of newproduct development as a knowledge managementprocess that requires a firm to imbue theproject with different types of knowledge atthe different stages of the process (Madhavan

and Grover, 1998). Moreover, drawing onthe exploration–exploitation learning frameworkallows us to gain insights when applied to thecontext of strategic alliances in the new productdevelopment process. The evidence supports Kozaand Lewin’s (1998) theoretical notion that firmsenter alliances with different motivations anddifferent goals. The fact that exploration alliancespredict products in development, which in turnpredict exploitation alliances, and that exploitationalliances predict products on the market, indicatesnot only that there are different motivations andgoals for the different types of alliances, butalso that different alliance types precede different



outcomes. These outcomes built on each otherin a sequential manner and thus contribute to aintegrated new product development process.

Biogen’s commercialization of Intron A canbe described by the product development processadvanced in this paper. Biogen used an explo-ration alliance with the University of Zurich andan exploitation alliance with Schering-Plough todiscover, develop, and commercialize Intron A.This product development strategy was successfulfor Biogen as Intron A reached over $1 billion insales in 2001. While Biogen shares this revenuestream with its alliance partners, Biogen wouldhave perhaps not been able to discover and com-mercialize Intron A on its own considering thedifficulty of going it alone. Even if Biogen wouldhave been able to discover, develop, and com-mercialize Intron A alone, it probably would havetaken much longer.

Our results also seem to support the conjec-tures of transaction cost economics (Williamson,1985) and the pecking order hypothesis of the opti-mal capital structure model (Myers, 1984). Whenthe hazards in the market for collaborative know-how become too high, firms tend to verticallyintegrate promising projects and fund them withinternal resources. The fact that firm size nega-tively moderates the impact of firm allying in eachof the hypothesized links in the proposed productdevelopment system seems to indicate that inter-nal resources are preferred to external resourcesby technology ventures to fund promising projects,regardless whether the activity concerns explo-ration or exploitation. Thus, while alliances appearto be one possible organizing mode of developingnew products, the movement away from alliancesas the venture grows indicates that they may be arisky strategy in which the smaller firms may beexposed to the risk of expropriation by their largerpartners (Lerner and Merges, 1998; Rothaermel,2001b; Lerner et al., 2003).

At the end of our study period in 1997, forexample, the old-line pharmaceutical firms mar-keted and distributed seven of the top-10 sell-ing biotechnology drugs via alliances with newbiotechnology firms, even though none of thesenew drugs were developed by the pharmaceuticalcompanies (Morrison and Giovannetti, 1998). Thesix new biotechnology drugs accounted for morethan two thirds of total revenues ($7.5 billion)accrued by the top-10 selling biotechnology drugs.Often the revenue partition is 50/50 between the

large pharmaceutical companies and their biotech-nology counterparts (Rothaermel, 2001a). That thebiotechnology firms have recently begun to for-wardly integrate to capture more value seems tobe indicated by the numbers for 2001: the newbiotechnology firms distributed six of the top-10selling biotechnology drugs on their own, again,all of them were developed by new biotechnol-ogy firms (Standard & Poor’s, 2002). Those sixdrugs alone captured about 50 percent of the over$13 billion revenues accrued by the top-10 newbiotechnology drugs. While these top-performingbiotechnology firms seem to withdraw from ally-ing on a case-by-case basis, we do not expect,however, that the technology ventures withdrawfrom alliances altogether. Rather we expect themto decrease their dependence on an alliance strat-egy and commensurately reduce their exposure tothe risks of opportunism and knowledge expropri-ation.

Anecdotal evidence from our fieldwork alsoseems to suggest that technology ventures maygain considerably when they are able to forwardlyintegrate into the exploitation activities of the prod-uct development process. Both Amgen co-founderRathmann and board member Omenn indicatedthat Amgen’s intended strategy was to originallypursue intensive exploitation collaboration withJohnson & Johnson (J&J) for the new biotech-nology products developed by Amgen. However,the collaboration with J&J went sour and endedin a long, drawn-out litigation. As a consequence,Amgen chose to forwardly integrate after its expe-rience with J&J. One could speculate that Amgenbecame the most successful biotechnology firm todate precisely because it was able to commercial-ize two blockbuster drugs (reaching over $3 billionin annual sales) on its own (Giovannetti and Mor-rison, 2000). Amgen’s vertical integration strategyinto the downstream activities of the value chainseems to resonate with our finding that biotech-nology firms tend to first withdraw from exploita-tion alliances before withdrawing from explorationalliances.

While there may be a liability of unconnected-ness (Baum and Oliver, 1992), there also appearsto be an offsetting liability of connectedness. Man-agers of technology ventures appear to balance thetrade-off between these liabilities as their resourceendowments increase. In fact, our results seemto be in line with the earlier findings that phar-maceutical firms used exploitation alliances with



biotechnology firms to improve their performanceat the expense of the biotechnology firms (Rothaer-mel, 2001b). In contrast to Powell et al. (1996),we find that firms do not opt unconditionally forinterdependence, but rather moderate their inter-dependence with increasing independence throughvertical integration on a project-by-project basis.As more internal resources become available tothe firms, they become less reliant upon externalalliance sources of knowledge and capital.

While many alliances are intended to lead to thecreation of new products, given the high degree ofuncertainty inherent in the new product develop-ment process regardless of industry (Griffin, 1997;Stevens and Burley, 1997), intentions may notreadily translate into new products, much less suc-cessful ones. Given this high degree of uncertaintysurrounding the new product development processin general, it is conceivable that the costs requiredto identify a suitable alliance partner and to nego-tiate, manage, and monitor an alliance may negateany potential benefits from the alliance. Aggre-gated to the firm level, the opportunity costs ofan exploration–exploitation alliance strategy in thenew product development process can potentiallyoutweigh its benefits. Under these circumstances,firms pursuing an alliance strategy should be nomore productive, and perhaps even less produc-tive, than those who pursue a strategy emphasizinginternal development.

Our findings also seem to suggest that as tech-nology ventures grow larger, potential partnersmay need to become increasingly skeptical ofthe technology ventures’ offerings. An expandedresource base allows technology ventures to keepthe highest-quality projects for themselves andonly offer lower-quality projects to potential alli-ance partners (Pisano, 1997; Lerner and Merges,1998). In essence, the tables may turn once atechnology venture is able to achieve a thresh-old size. While technology start-ups may havebeen penalized initially because the problem ofasymmetric information in the market for know-how lowered prices for quality projects, largermore successful technology ventures may be ableto take advantage of their position by gaining aprice premium for lower-quality projects. This isbecause lower quality projects crowd out higher-quality projects according to the lemons hypoth-esis (Akerlof, 1970). In fact, if firm size resultsin increased legitimacy and attributions of qualityto the technology ventures’ projects, these firms

may be able to gain an additional premium forlower-quality projects.

Some of our secondary findings are also worthhighlighting. We found that firms that possess ahigher patenting propensity tend to engage in moreexploitation alliances and have a greater num-ber of products on the market, although patentswere nonsignificant in predicting the firm’s prod-ucts in development. Shan et al. (1994) found thatalliances predicted a biotechnology firm’s patents,while they found no support for a reverse rela-tionship. We presented a more subtle model andfound that patents explained alliances, but only acertain kind of alliances—exploitation alliances.In our research setting, it appears that patents arean output of the discovery and development stageof the product development process rather than aninput to it (Griliches, 1990).

On the aggregate, we find that biotechnologyfirms enter into significantly more exploitationalliances than exploration alliances. This finding isin line with Koza and Lewin’s (1998) theoreticalconjecture that an industry will as a rule be charac-terized by more exploitation alliances than explo-ration alliances. Further, our result that exploita-tion alliances are more frequent than explorationalliances is consistent with Rothaermel’s (2001b)finding when studying large established firms ra-ther than newer technology ventures. It appearsthat exploitation alliances generally exhibit lessuncertainty and thus require fewer resources whichin turn enables firms to manage more exploitationalliances than exploration alliances. However, onecould also speculate that our findings lend supportto the notion that exploitation may drive out explo-ration and that firms could be trapped by their owncompetencies (Levinthal and March, 1993).

Limitations and future research

While were able to assess each firm’s mix ofexploration and exploitation alliances, we did notrelate specific exploration–exploitation ratios tofirm performance. Future research could attemptto specify what mix of exploration and exploita-tion is best when striving for superior perfor-mance, taking the different time horizons forexploration and exploitation learning modes intoaccount. Another path future research could takein analyzing the mix of exploration and exploita-tion activities would be to determine how much



of each activity should be pursuit in-house vs.through alliances.

While our arguments highlighted the costsand risks involved of allying, others haveemphasized value creation and dyadic processesin interfirm allying (Zajac and Olsen, 1993).Future work should begin by investigating factorsthat determine the distribution of the rentsgenerated between the different organizationsinvolved in an exploration–exploitation productdevelopment process. Subsequently, future workcould investigate alliance processes for joint valuecreation and distribution that may mitigate someof the potential problems new technology venturescan encounter when partnering with establishedfirms.

The results provide support for our researchmodel; however, we must also acknowledge thatour focus on biotechnology raises questions aboutthe generalizability of our study beyond this indus-try. Biotechnology has several unique charac-teristics, including a long product developmentand regulatory approval cycle, heavy relianceupon often arcane basic scientific research, anda resource intensive new product developmentprocess. Despite these unique characteristics inour sample, we submit that our results mightbe generalizable beyond the biotechnology indus-try since basic science and interfirm cooperationin high-technology industries appear to be play-ing an increasingly important role in the suc-cess and failure of individual firms (Hagedoorn,1993; Dasgupta and David, 1994). For example,recent alliance announcements between the Cali-fornia Fuel Cells Partnership, MIT, and Dupont,which comprises government laboratories, univer-sities, fuel cell ventures, and automobile as well asenergy companies, seem to indicate that the fuelcell industry might be pursuing a similar devel-opment path as the biotechnology industry. Futureresearch could assess the external validity of ourmodel by testing it in different industry settings.

Another limitation of this study is that we wereonly able to study surviving alliances. In particu-lar, given the higher degree of uncertainty involvedin exploration alliances, one would expect them toexhibit a commensurate higher mortality rate rela-tive to exploitation alliances. However, it is impor-tant to note that alliances in the biotechnologyindustry are characterized by longevity since theproduct development process can take 15 years ormore. For example, Shan et al. (1994) found that

only 15 percent of the alliances entered since theearly 1970s had expired by 1989. Another piece ofevidence that only a small number of biotechnol-ogy alliances are terminated is provided by Green(1997), who reported that the founding to termina-tion ratio for allying was about 9 : 1 in biotechnol-ogy near the end of our study period. Given the lowmortality rate of biotechnology alliances, in con-junction with their longevity, we believe that ourresults are not materially influenced by a potentialsurvivorship bias.

While we were able to find support for an inte-grated product development path beginning withexploration alliances and concluding with prod-ucts on the market, we need to emphasize that theperformance distribution of commercialized prod-ucts, in particular in the biotechnology industry, isheavily skewed. One successful blockbuster drugmay accrue several billion dollars of revenues fordecades, whereas many, if not most drugs, willnot be able to cover their research and develop-ment cost. Thus, while products on the market aremore proximate to firm performance than patentingor products in development used in prior stud-ies (Shan et al., 1994; Deeds and Hill, 1996),future research is needed to illuminate the linkbetween newly commercialized products and firmperformance.

CONCLUSION

We submit that this paper extends our understand-ing of the role of strategic alliances in the newproduct development process by providing a linkbetween alliance participation by technology ven-tures beginning with discovery and culminatingwith products on the market. We submit that wehave added to the field’s understanding of alliancesand new product development by developing andtesting a more subtle and comprehensive modelof the new product development process based onthe exploration–exploitation learning framework,transaction cost economics, pecking order hypoth-esis of optimal capital structure, and informa-tion asymmetry in the market for know-how. Ourmodel provides predictions of managerial behaviorat different stages of the new product developmentprocess including the type of alliances a firm islikely to enter (exploration vs. exploitation), andpredicts an overall trend in a firm’s tendency towithdraw from alliances as the venture expands.Our results appear to support our proposed product



development path suggesting that different types ofalliances are motivated by different goals, achievedifferent outcomes, are best employed at differentstages of development, and that managers preferinternal resources to external resources when fund-ing promising R&D projects. In conclusion, wedo not argue that the technology ventures in oursample see the world as a zero sum game, wherethere are no synergistic returns to collaboration, butrather that given a particular resource endowmentand opportunity set, firms attempt to maximizetheir product development performance by achiev-ing a balance between the risks and rewards ofinterfirm cooperation.

ACKNOWLEDGEMENTS

We thank Associate Editor Edward Zajac, twoanonymous reviewers, Melissa Appleyard, RogerCalantone, Charles Hadlock, Duane Ireland, JagdipSingh, and the seminar participants at Case West-ern Reserve University, the Darden School at theUniversity of Virginia, the University of Min-nesota, and the University of Richmond for helpfulcomments and suggestions on earlier versions ofthis paper. We also thank Gavin Mills for expertresearch assistance and Suzanne Rumsey for edito-rial assistance. All remaining errors and omissionsare entirely our own.

REFERENCES

Aghion P, Tirole J. 1994. On the management ofinnovation. Quarterly Journal of Economics 109:1185–1207.

Ahuja G. 2000. The duality of collaboration: inducementsand opportunities in the formation of interfirmlinkages. Strategic Management Journal 21(3):317–343.

Aiken LS, West G. 1991. Multiple Regression: Testingand Interpreting Interactions . Sage: Newbury Park,CA.

Akerlof GA. 1970. The market for lemons: qualityuncertainty and the market mechanism. QuarterlyJournal of Economics 94: 488–500.

Anderson JC, Gerbing DW. 1988. Structural equationmodeling in practice: a review and recommended two-step approach. Psychological Bulletin 103: 411–423.

Baum JAC, Oliver C. 1992. Institutional embeddednessand the dynamics of organizational populations.American Sociological Review 57: 540–569.

Baum JAC, Calabrese T, Silverman BS. 2000. Don’t goit alone: alliance network composition and startups’performance in Canadian biotechnology. StrategicManagement Journal 21(3): 267–294.

Bentler PM. 1995. EQS Structural Equations ProgramManual . Multivariate Software: Encino, CA.

Booz-Allen & Hamilton. 1982. New Product Managementfor the 1980’s . Booz-Allen & Hamilton: New York.

Brown SL, Eisenhardt KM. 1997. The art of continuouschange: linking complexity theory and time-pacedevolution in relentlessly shifting organizations.Administrative Science Quarterly 42: 1–34.

Byrne B. 1994. Structural Equation Modeling with EQSand EQS/Windows . Sage: Thousand Oaks, CA.

Carmines E, McIver J. 1981. Analyzing models withunobserved variables: analysis of covariance struc-tures. In Social Measurement: Current Issues , Bohrn-stedt G, Borgatta E (eds). Sage: Beverly Hills, CA;65–115.

Cohen J, Cohen P. 1983. Applied Multiple Regres-sion/Correlation Analysis for the Behavioral Sciences .Erlbaum: Hillsdale, NJ.

Cohen WM, Levinthal DA. 1990. Absorptive capacity:a new perspective on learning and innovation.Administrative Science Quarterly 35: 128–152.

Dasgupta P, David PA. 1994. Toward a new economicsof science. Research Policy 23: 487–521.

Deeds DL, Hill CWL. 1996. Strategic alliances and therate of new product development: an empirical studyof entrepreneurial biotechnology firms. Journal ofBusiness Venturing 11: 41–55.

Dussauge P, Garrette B, Mitchell W. 2000. Learningfrom competing partners: outcomes and durations ofscale and link alliances in Europe, North America, andAsia. Strategic Management Journal 21(2): 99–126.

Giovannetti GT, Morrison SW. 2000. Convergence: TheBiotechnology Industry Report . Ernst & Young: PaloAlto, CA.

Grant RM, Baden-Fuller C. 1995. A knowledge-basedtheory of inter-firm collaboration. Academy ofManagement Best Paper Proceedings 17–21.

Green D. 1997. Biotech partners: symbiosis puts downsome roots. Financial Times 24 April: 5.

Greene WH. 1997. Econometric Analysis . Prentice-Hall:Upper Saddle River, NJ.

Griffin A. 1997. Drivers of NPD Success: The 1997PDMA Report . Product Development ManagementAssociation: Chicago, IL.

Griliches Z. 1990. Patent statistics as economic indica-tors: a survey. Journal of Economic Literature 28:1661–1707.

Gulati R. 1998. Alliances and networks. StrategicManagement Journal , Special Issue 19: 293–317.

Hagedoorn J. 1993. Understanding the rationale ofstrategic technology partnering: interorganizationalmodes of cooperation and sectoral differences.Strategic Management Journal 14(5): 371–385.

Hennart J-F, Roehl T, Zietlow DS. 1999. Trojan horseor workhorse? The evolution of U.S.–Japanese jointventures in the United States. Strategic ManagementJournal 20(1): 15–29.

Hill CWL, Rothaermel FT. 2003. The performance ofincumbent firms in the face of radical technologicalinnovation. Academy of Management Review 28:257–274.



James LR. 1982. Aggregation bias in estimates ofperceptual agreement. Journal of Applied Psychology67: 219–229.

Klein B, Crawford R, Alchian A. 1978. Vertical inte-gration, appropriable rents and the competitive con-tracting process. Journal of Law and Economics 21:297–326.

Kotabe M, Swan KS. 1995. The role of strategic alliancesin high-technology new product development. Strate-gic Management Journal 16(8): 621–636.

Koza MP, Lewin AY. 1998. The co-evolution of strategicalliances. Organization Science 9: 255–264.

Lane PJ, Lubatkin M. 1998. Relative absorptive capac-ity and interorganizational learning. Strategic Manage-ment Journal 19(5): 461–477.

Lerner J, Merges RP. 1998. The control of technologyalliances: an empirical analysis of the biotechnologyindustry. Journal of Industrial Economics 46:125–156.

Lerner J, Shane H, Tsai A. 2003. Do equity financingcycles matter? Evidence from biotechnology alliances.Journal of Financial Economics 67: 411–446.

Levinthal DA, March JG. 1993. The myopia of learning.Strategic Management Journal , Winter Special Issue14: 95–112.

Madhavan R, Grover R. 1998. From embedded knowl-edge to embodied knowledge: new product develop-ment as knowledge management. Journal of Marketing62: 1–12.

Majewski SE. 1998. Causes and consequences ofstrategic alliance formation: the case of biotechnology.PhD thesis, University of California, Berkeley.

Managing Intellectual Property. 1999. Genentech settlesfor $200 million. 95: 11.

March JG. 1991. Exploration and exploitation inorganizational learning. Organization Science 2:71–87.

Monteverde K, Teece DJ. 1982. Vertical integration andsupplier switching costs. Bell Journal of Economics13: 1–25.

Morrison SW, Giovannetti GT. 1998. Biotech’99: Bridg-ing the Gap. Ernst & Young: Palo Alto, CA.

Myers SC. 1984. The capital structure puzzle. Journal ofFinance 39: 575–592.

Myers SC, Majluf NS. 1984. Corporate financing andinvestment decisions when firms have informationinvestors do not have. Journal of Financial Economics13: 187–221.

Park SH, Chen R, Gallagher S. 2002. Firm resources asmoderators of the relationship between market growthand strategic alliances in semiconductor start-ups.Academy of Management Journal 45: 527–545.

Pisano GP. 1990. The R&D boundaries of the firm: anempirical analysis. Administrative Science Quarterly35: 153–176.

Pisano GP. 1997. R&D performance, collaborativearrangements, and the market-for-know-how: a testof the ‘lemons’ hypothesis in biotechnology. Workingpaper no. 97–105. Harvard Business School.

Pisano GP, Mang P. 1993. Collaborative product devel-opment and the market for know-how: strategies andstructures in the biotechnology industry. In Research

on Technological Innovation, Management, and Pol-icy, Vol. 5, Rosenbloom R, Burgelman R (eds). JAIPress: Greenwich, CT; 109–136.

Powell WW, Koput KW, Smith-Doerr L. 1996. Interor-ganizational collaboration and the locus of innovation:networks of learning in biotechnology. AdministrativeScience Quarterly 41: 116–145.

Rosenkopf L, Nerkar A. 2001. Beyond local search:boundary spanning, exploration, and impact in theoptical disk industry. Strategic Management Journal22(4): 287–306.

Rothaermel FT. 2001a. Incumbent’s advantage throughexploiting complementary assets via interfirm coop-eration. Strategic Management Journal , Special Issue22: 687–699.

Rothaermel FT. 2001b. Complementary assets, strategicalliances, and the incumbent’s advantage: an empiricalstudy of industry and firm effects in the biopharma-ceutical industry. Research Policy 30: 1235–1251.

Shan W, Walker G, Kogut B. 1994. Interfirm cooperationand startup innovation in the biotechnology industry.Strategic Management Journal 15(5): 387–394.

Shyam-Sunder L, Myers SC. 1999. Testing static tradeoffagainst pecking order models of capital structure.Journal of Financial Economics 51: 219–244.

Sørensen JB, Stuart TE. 2000. Aging, obsolescence,and organizational innovation. Administrative ScienceQuarterly 45: 81–112.

Standard & Poor’s 2002. Biotechnology Industry Report,May.

Stevens GA, Burley J. 1997. 3,000 raw ideas equals onecommercial success. Research Technology Manage-ment 40: 16–27.

Stuart TE. 2000. Interorganizational alliances and theperformance of firms: a study of growth andinnovation rates in a high-technology industry.Strategic Management Journal 21(8): 791–811.

Stuart TE, Podolny JM. 1996. Local search and theevolution of technological capabilities. StrategicManagement Journal , Summer Special Issue 17:21–38.

Teece DJ. 1986. Profiting from technological innovation:implications for integration, collaboration, licensingand public policy. Research Policy 15: 285–305.

Teece DJ. 1992. Competition, cooperation, and innova-tion: organizational arrangements for regimes of rapidtechnological progress. Journal of Economic Behaviorand Organization 18: 1–25.

Walker G, Kogut B, Shan W. 1997. Social capital,structural holes and the formation of an industrynetwork. Organization Science 8: 109–125.

Williamson O. 1985. The Economic Institutions ofCapitalism . Free Press: New York.

Zajac EJ, Olsen CP. 1993. From transaction cost totransactional value analysis: implications for thestudy of interorganizational strategies. Journal ofManagement Studies 30: 131–145.

Zucker LG, Darby MR, Armstrong JS. 2002. Commer-cializing knowledge: university science, knowledgecapture, and firm performance in biotechnology. Man-agement Science 48: 138–153.


exploration and exploitation alliances in biotechnology: a

Documents