general technological capabilities, product market fragmentation, and markets for technology
TRANSCRIPT
Ga
AD
a
ARRAA
KMLGMS
1
kftiivsn(
WeCu(2Ltont
M
0h
Research Policy 42 (2013) 315– 325
Contents lists available at SciVerse ScienceDirect
Research Policy
jou rn al h om epage: www.elsev ier .com/ locate / respol
eneral technological capabilities, product market fragmentation,nd markets for technology
lfonso Gambardella ∗, Marco S. Giarratanaepartment of Management & Technology, Bocconi University, Italy
r t i c l e i n f o
rticle history:eceived 16 September 2011eceived in revised form 28 May 2012ccepted 9 August 2012vailable online 6 September 2012
a b s t r a c t
The combination of a firm capability (i.e., ability to generate general purpose technologies) and a mar-ket structure condition (i.e., fragmentation of downstream submarkets) may encourage licensing in anindustry. That is, the probability of licensing should increase when product markets are fragmented andtechnologies support general purposes. Evidence consistent with these predictions emerges from a 1993to 2001 panel of 87 firms that owned at least one U.S. software security patent between 1976 and 2001.
eywords:arkets for technology
icensingeneralityarket fragmentation
The analysis uncovers some fundamental characteristics of how external knowledge exploitation func-tions; in particular, technology markets thrive when product markets are fragmented and firms have thecapability to produce general technologies.
© 2012 Elsevier B.V. All rights reserved.
oftware
. Introduction
The commercial exploitation of knowledge is a main tenet ofnowledge management theory (Argote et al., 2003; Teece, 1986),or which the exploitation of external knowledge—particularly inhe so-called market for technology—is critical. From 1980 to 2003n the G8 countries, technology royalty payments and receiptsncreased annually by an average of 10.7% and reached an annualolume of approximately US$190 billion in 2003 (OECD, 2006). Casetudy evidence also has stressed the increasing importance of busi-ess models that focus on the external exploitation of knowledgeArora and Gambardella, 2010; Gans and Stern, 2010).
A mainstream line of research generally follows a classicalilliamson’s framework that hinges on the interplay between
nvironmental transaction costs and licensing decisions (Arora andeccagnoli, 2006; Gans et al., 2002, 2008). However, recent contrib-tions have established some clearer roots in firm-based researchLichtenthaler, 2010; Cassimon et al., 2011; Kani and Motohashi,012; Fosfuri, 2006; Gambardella et al., 2007; Lichtenthaler andichtenthaler, 2009) and shifted the focus from exogenous condi-ions in which firms decide to license to firm-based determinants
f licensing. In this context, we argue that an important determi-ant of licensing is the firm’s capability to produce general-purposeechnologies (GPTs; Bresnahan and Trajtenberg, 1995; Rosenberg,∗ Corresponding author.E-mail addresses: [email protected] (A. Gambardella),
[email protected] (M.S. Giarratana).
048-7333/$ – see front matter © 2012 Elsevier B.V. All rights reserved.ttp://dx.doi.org/10.1016/j.respol.2012.08.002
1976) that can embrace many different product market applica-tions (e.g., Bresnahan and Gambardella, 1998; Maine and Garsney,2006; Von Hippel, 1994). In contrast, a dedicated technology maybe perfectly suited for the application for which it is created, but itis not very useful for other applications.
The central role of GPTs for knowledge exploitation is evi-dent in comments by the CEO of Peregrine Pharmaceuticals(www.peregineinc.com), who acknowledged that the company’s“strategy for clinical development is designed to maximize thelicensing potential of our broad platform technologies [that] givesus the ability to license and collaborate with many partners”(Business Wire, 2000: 12). An increasing number of firms simi-larly recognize the importance of GPTs (Gambardella and McGahan,2010; Palomeras, 2007; Thoma, 2009), which can facilitate licens-ing without encouraging product market competitors (Arora andFosfuri, 2003). Therefore, we consider an environmental conditionthat makes the ability to produce GPTs valuable: when downstreamproduct markets are fragmented in different subniches, licensorscan issue licenses to other firms that operate in market niches inwhich they do not compete directly, though that scenario requiresthe licensor to develop GPTs that can support distant applications.
We provide empirical evidence for this claim by studying thesecurity software industry (SSI), a technology-based industry inwhich innovation plays a major role and that exhibits a clear ver-tical distinction among the market for SSI algorithms, the core
technology of a SSI products, and markets for SSI products orservices. Security algorithms also exhibit different degrees of gen-erality. Those that are more specific to particular domains tend tobe more effective in their realm but not applicable to many other
3 / Rese
dtcro1igr
cs2wFmttoarhc
oot
2
ibeiEiet2(sbIoa
fisnTa2(ttapfiritPt2
16 A. Gambardella, M.S. Giarratana
omains (The Economist, 2007). Moreover, SSI firms often patentheir new software algorithms, which refer to specific technologylasses and allow for greater precision in technology proxies. Ouregressions predict the hazard that a firm will sell its technol-gy from the beginning of the technology market in SSI, using a993–2001 panel data set. We find that the probability of licensing
ncreases with downstream product market fragmentation and theenerality of the licensor’s technology, and that these two factorseinforce each other.
Our article therefore offers two main contributions. First, weontribute to literature on knowledge management, especially thetream dedicated to external knowledge exploitation (Argote et al.,003; Lichtenthaler and Lichtenthaler, 2009). Within this realm,e focus on the role of firm-based determinants. For example,
osfuri (2006) notes the role of firm market shares in a productarket, Gambardella et al. (2007) consider firm size. We introduce
he importance of GPT capabilities (Winter, 2003) and show thathey can explain most heterogeneity in licensing outcomes. Sec-nd, we stress an important link between technological capabilitiesnd downstream industry structure. In this respect, our findingeplicates a classical theorem of the capability-based view thatighlights the co-evolution of firm capabilities and environmentalonditions (Argote et al., 2003; Sorenson, 2003).
In the next section, we provide a literature review, followed byur theory. We then describe the major features of SSI and presentur data and empirical evidence. We conclude with some implica-ions and further research directions.
. Theory
As Teece (1986) has established, knowledge management stud-es should involve not only how to capture value from innovationut also, and more precisely, which conditions make externalxploitation (i.e., technology licensing) more appropriate thannternal exploitation (i.e., technology embedded in final products).arly economics research focused on the role of transaction costsn shaping technology markets. In particular, transaction costsncompass the search costs of finding a partner, fear of oppor-unism, and lack of valid knowledge protections (Cockburn et al.,010). To solve the transaction cost problem, Arora and Ceccagnoli2006), Hall and Ziedonis (2001), and Gans et al. (2002) suggesttronger intellectual propriety rights (IPRs), such that firms maye more likely to sell their technologies if IPRs are well defined.n contrast, if IPRs are weak, firms can earn rents from technologynly by incorporating it into their own final products (McGahannd Silverman, 2006; Dushnitsky and Klueter, 2011).
Management scholars extend the IPR notion by considering howrm characteristics, such as market share, size, and human resourcetrategies, might influence their ability to exploit knowledge exter-ally (Fosfuri, 2006; Gambardella et al., 2007; Lichtenthaler, 2007).hese contributions reflect a classical approach to knowledge man-gement theory that highlights two important facets (Argote et al.,003): firm capabilities and their fit with some exogenous featuree.g., IPR context). We base our theory on a particular capabilityhat explains heterogeneity in firm licensing behavior. Accordingo Helfat and Winter (2011), a capability arises when a firm has thebility to perform a particular activity as an intended purpose withatterned behavior. Capabilities are therefore a key dimension ofrm heterogeneity and a source of competitive advantage. In turn, aepetitive, recognizable pattern of interdependent actions, involv-ng multiple actors—also known as an organizational routine—is
he main mechanism for generating capabilities (Feldman andentland, 2003). In our study context, we focus on a firm’s capabilityo generate GPTs (Bresnahan and Gambardella, 1998; Palomeras,007; Thoma, 2009); in other words, we assume that firmsarch Policy 42 (2013) 315– 325
differ in their R&D organizational routines and that thesedifferences generate heterogeneous capabilities for producingGPTs.
We focus on the interplay between these capabilities, withtheir underlying routines, and the level of downstream marketfragmentation of an industry. In defining downstream market frag-mentation, we follow Klepper and Thompson (2006): Industriescan be differentiated along various dimensions, such as the servicesthey provide, the customer segments they target, or the geographicareas in which they operate. The more an industry is segmented indifferent submarkets, the less homogenous it is, and the more thefirm’s behavior and performance depend on the dynamics of thespecific submarkets in which it operates.
We therefore develop a framework in which we assume thatfirms can own GPT capabilities or not and can compete in frag-mented or homogenous industries. For simplicity, we exclude thecase of pure technology suppliers without products, an assumptionthat fits well with industries in which it is affordable to move fromtechnology to final product (e.g., software). With our framework,we can compare changes in the probability of concluding a licens-ing deal for a dedicated technology versus a GPT, in homogenousand fragmented product markets, respectively.
We start with a homogeneous industry and a firm that hasthe capabilities to produce a dedicated technology. A dedicatedtechnology is well suited for a specific application, but it requiressignificant adaptation costs to be applied to other domains.Therefore, the dedicated technology fits the application to a homo-geneous market, and a potential licensee can use it with minimaladaptation costs. Consider a company that plans to enter the prod-uct market but does not have the technology to operate in it. Thecompany has three options: not enter, enter by investing in thetechnology, or enter by buying the technology. The low adaptationcosts of the licensed technology only affects the third option, byraising the odds of demand for a license rather than the other twooptions.
However, industry homogeneity implies that the licensor maybe reluctant to sell its technology, because doing so would makethe licensee a close product competitor.
Let us define the product market profits of the licensor as �(N),where N is the number of competitors in the product market.Then �(N + 1) represents the product market profits after it licensesto a licensee. Following Arora and Fosfuri (2003), licensing cre-ates a new competitor (i.e., the licensee), and thus the productmarket profits of the licensor become �(N + 1) ≤ �(N), where theinequality occurs because adding a competitor cannot increasethe profits of incumbents. Arora and Fosfuri (2003) point out that�(N + 1) is also the largest revenue that the licensor can obtainfrom licensing, because it cannot extract from the licensee morethan what the licensee earns in the form of product market profits.Thus, the licensor licenses if and only if (1 + ˛) × �(N + 1) ≥ �(N),or ̨ × �(N + 1) ≥ �(N) − �(N + 1), where ̨ is the share of profits ofthe licensee that the licensor earns through royalty rates for thelicense. The share of ˛ depends on several factors, including com-petition across licensors in the technology market, the ability ofthe licensee to develop the technology in-house, other factors thatmight affect the bargaining power of the parties, and the strengthof the IPR that protect the licensor from imitation. We leave thesefactors in the background. We also abstract away from several otherfactors that affect this process, such as whether the licensor licensesto others or the reactions of the other product market competitorsthat may also decide to license. These industry-level factors are dis-cussed extensively by Arora and Fosfuri (2003). Moreover, there are
several other reasons for licensing (or not) that we do not addresshere (e.g., Hellmann and Perotti, 2011). Rather, the intuition that wehighlight, and that it is a key factor in the more general, industry-level model developed by Arora and Fosfuri (2003), is that given ˛,
/ Rese
ti
HlmlcokfiuoliAbowalhf
Gdttanenatmobi
ntccutaisoacibuc
Hct
mttpsm
A. Gambardella, M.S. Giarratana
he decision to license depends on both the level of �(N + 1) and thenequality �(N) − �(N + 1).
As N increases, �(N + 1) decreases, and so does �(N) − �(N + 1).owever, if we assume, as Arora and Fosfuri (2003) do, that the
icensor can only license to a firm that operates in its same productarket, then all else being equal, the distance �(N) − �(N + 1) is the
argest possible difference that results from the introduction of aompetitor. In other words, �(N) − �(N + 1) is the largest differencen the continuum from homogeneous to fully independent mar-ets, where independence implies �(N) = �(N + 1), such that a newrm operating in the market has no impact on the original prod-ct market. The closer �(N + 1) is to �(N), the lower is the degreef substitution among products produced by the licensor and theicensee. Thus, all else being equal, in a homogenous market, licens-ng is limited because it produces the largest profit dissipation.rora and Fosfuri (2003) further show that that when N increases,oth �(N + 1) and �(N) − �(N + 1) decrease; it is not clear whetherne changes faster than the other. That is, we cannot say how theillingness to license changes as N increases, but compared with
more differentiated product market, in which the licensor canicense to a firm that operates in a more distant product market, aomogenous product market creates the greatest profit dissipation
rom licensing and thus the lowest willingness to license.The scenario does not change significantly if the technology is a
PT. A GPT is less distant from many applications than a technologyedicated to a specific application, in the sense of its lower adapta-ion costs for each application. However, the GPT is more distant, inhe sense of higher adaptation costs, from any specific applicationddressed by a particular dedicated technology. Thus, a GPT doesot change the propensity of the licensor to license, because if itnables the licensee to enter, and the product market is homoge-eous, the licensee operates in close competition with the licensornd cannibalizes its product market profits. The key factor thus ishe tight competition between licensor and licensee in the product
arket, which does not change according to whether the technol-gy is dedicated or general. At the same time, the licensee even maye less keen to buy a GPT versus a dedicated technology, because it
nduces higher adaptation costs.In summary, from the licensor’s perspective, whether the tech-
ology is dedicated or general, it is less willing to supply theechnology than it would be in a fragmented market, because itan only be supplied to a close competitor in product space, whichreates a significant threat of cannibalization of the licensor’s prod-ct market profits. From the licensee’s perspective, a more generalechnology reduces its willingness to buy, because of the higherdaptation costs required compared with a technology that is ded-cated to that product market. Because of the low willingness toupply, the effect of the lower willingness to buy on the possibilityf a licensing deal actually is minimized. A low willingness to supplylso mirrors a high royalty rate. Whether the technology is dedi-ated or general in its purpose, most potential licensees (i.e., thosenterested in entering the product market) cannot afford this rate,ecause they are not efficient enough to cover the licensor’s prod-ct market profits. The probability of licensing therefore should nothange significantly. We formulate the following hypothesis.
ypothesis 1. When the industry is homogeneous, firms withapabilities to produce a dedicated or a general technology exhibithe same probability of issuing a license.
A fragmented industry instead is characterized by different sub-arket segments. If the potential licensee operates in a submarket
hat is not in direct competition with the licensor’s submarket,
he licensor is willing to sell its technology, because any resultingroduct market competition is sheltered by differentiation acrossubmarkets. However, if the technology is dedicated to the sub-arket of the licensor, it can be used in the submarket of thearch Policy 42 (2013) 315– 325 317
licensee only after the expenditure of high adaptation costs. Thus,the deal will be closed only if the downstream market potential ofthe licensee is high, enabling the licensee to cover the adaptationcosts. If the licensor owns a GPT, the costs of technology adaptionfor the potential licensees in all other submarket niches instead arelower.
Licensing in a “distant” product market is the key differencebetween the framework we propose and that offered by Aroraand Fosfuri (2003). As we have noted, they consider licensing inthe same product market; de facto, licensing in their model onlyoccurs in homogenous product markets where profit dissipation�(N) − �(N + 1) is highest. In contrast, our framework allows forthe possibility that the product market is fragmented, such thatN competitors are spread across submarkets that are distant fromone another, and the entry of a new competitor, operating in itsown differentiated submarket niche, does not affect the profitsof the other competitors much, particularly those of the licensor.In the inequality ̨ × �(N + 1) ≥ �(N) − �(N + 1), the lower differ-ence �(N) − �(N + 1) increases the licensor’s willingness to license.In addition, a differentiated product market likely implies higher�(N + 1), because the N + 1 competitors in general are more shel-tered from one another’s competitive efforts. Thus, licensing ismore likely, because the licensor also can extract more rents fromthe licensee. Along with the willingness to license, we predict agreater willingness to buy by the licensee. The licensee wants tobuy the technology, because its profits (1 − ˛) × �(N + 1) are higher,due to greater product differentiation. However, the propensity tobuy depends on whether the technology is dedicated or GPT; in thelatter case, it requires less adaptation costs.
In summary, whether the technology is dedicated or general, infragmented markets the licensor is more willing to supply it thanin homogenous markets, because the technology can be suppliedto a distant product market that does not affect the product mar-ket profits of the licensor much. This is the key difference fromthe previous case, where the licensor was unwilling to supply anykind of technology. Furthermore, in the fragmented market, a moregeneral product technology increases willingness to buy, due to itslower adaptation costs, than a technology dedicated to the spe-cific product market of the licensor. We formulate the followinghypothesis.
Hypothesis 2. When the industry is fragmented, a firm with capa-bilities to produce a general technology has a higher probability ofissuing a license than a firm with capabilities to produce a dedicatedtechnology.
In a fragmented market, it also is possible that the licensordecides to enter the distant product market itself, as a producer.This choice is not unusual; Nelson (1959) cites it as a reason forfirms to invest in basic research (which is inherently general pur-pose), that is, to diversify in many product markets. Such entriesalso are possible for firms with a dedicated technology that theydecide to adapt to apply in other markets. We thus argue that, com-pared with a homogenous market, in a fragmented market, thereare cases in which the economies of scope for technology (partic-ularly for the GPT, but possibly for dedicated technologies too) arestronger than the economies of scope in the downstream assetsfor a given set of product submarkets. For example, firms licenseto other companies in distant geographic markets, where domesticfirms have advantages setting up local production and commercial-ization. Similarly, firms in certain industries may prefer to licenseto companies with more established production and commercial-
ization assets in other industries in which the technology can beapplied (e.g., McGahan and Silverman, 2006). Thus, though somelicensors may enter distant submarkets, we can state confidentlythat, all else being equal, in fragmented markets, willingness to
318 A. Gambardella, M.S. Giarratana / Research Policy 42 (2013) 315– 325
Table 1Theory.
Product market
Homogeneous Fragmented
Technology capabilities
Dedicated �L·ıH �H·ıL
Negligible effect on probability that licensing occurs Increase in probability that licensing occurs↓ ↓
GPT �L·ıL �H·ıH
N ing to
f �L(ıL −
li
Claptpltptodiatkcidw
Go�lad0btnl
aaTccetapibfifntlib
Table 2, we list the major SSI product niches that have arisen in thepast 15 years.
Technology embedded in a software product (i.e., the cryptoalgorithm) can be defined easily for sale in a stand-alone
Table 2Product niches in SSI.
Niche Description
Authentication–digitalsignature
Products for the authentication of digitaldocuments with a copyrighted mark
Antivirus Programs that detect and clean viruses fromcomputers
Data and hardwareprotection
Products securing the integrity of sensible datastored in hard drivers
Firewalls A kind of checking door between differentnetworks
Utility software Utility software programs that assureprotection and execution of operating systemsand applications, giving the possibility torecreate the content of some data packages lost
otes: �, ı = probability that, respectively, the potential licensor and licensee are willor licensing); �H > �L, ıH > ıL (H, L = high, low). Hypotheses 1 and 2 → �H(ıH − ıL) >
icense the technology to other parties is more pronounced than its in homogenous markets.
We provide a synthetic representation of this theory in Table 1.onsider a potential licensing deal between a licensor and a
icensee, where ı is the probability that the latter is willing to accept given set of conditions for closing the deal, � is the equivalentrobability for the former, and ı and � represent convenient nota-ions for demand and supply. That is, ı and � are the respectiverobabilities that the reservation values of the licensee and the
icensor are less profitable than the benefits that they can draw fromhe deal. A licensing deal then occurs with probability ı × �. Sup-ose that ıH > ıL and �H > �L, where H and L stand for high and low. Inhe context of Hypothesis 1 (homogenous markets) � = �L and ı = ıLr ıH, depending on whether the licensor has capabilities to pro-uce, respectively, general or dedicated technology. The difference
n the probability of a licensing deal when technology capabilitiesre general or dedicated and the product market is homogenoushus is �L(ıL − ıH). In the context of Hypothesis 2 (fragmented mar-ets), � = �H and ı = ıH or ıL, depending on whether the technologyapability is general or dedicated. The difference in the probabil-ty of a licensing deal when the technology capability is general oredicated and the product market is fragmented then is �H(ıH − ıL),hich is unambiguously higher.
This result also is independent of whether we assume that aPT lowers the benefits of a licensee considering the purchasef a technology in a homogenous market. Whether ı = ıH or ıL,H(ıH − ıL) > �L(ıL − ıi), where i = H or L. Moreover, recall that the
ow � in the homogenous market dampens the effect related to thessumption about the adaptation cost of the GPT compared with theedicated technology: Simply put, low �L makes �L(ıL − ıi) close to
whether i = H or L. In contrast, high �H amplifies the willingness touy difference for dedicated technology versus GPT. In other words,he licensor’s willingness or unwillingness to license in a homoge-ous versus fragmented market is the key factor that boosts the
icensee’s willingness or unwillingness to buy in these settings.We can also unambiguously state that markets for technology
re most likely observed when product markets are fragmentednd firms possess the capabilities to produce GPT technologies. Inable 1, this scenario corresponds to the highest probability for con-luding a licensing deal, that is, �HıH. In the next case, firms haveapabilities to produce dedicated technologies and the market isither homogenous or fragmented, �LıH or �HıL. The former set-ing is what Arora and Fosfuri (2003) consider. However, as soons we relax the assumption that licensing only occurs in the sameroduct market, the potential for licensing increases. For example,
n Arora and Fosfuri (2003), licensing never occurs in a monopoly,ecause once the licensee enters, the two firms earn duopoly pro-ts, and even if the licensor extracts all the product market rents
rom the licensee (i.e., ̨ = 1), 2 × �(2) < �(1), That is, the licensee canever pay a royalty equal to the monopoly profits, which is what
he licensor earns without competition. Although they show thaticensing can occur when N > 1—in which case 2 × �(N + 1) > �(N)s possible—we hold that licensing also is possible in a monopoly,ecause if the monopolist licenses to a distant product market, itconclude a licensing deal at given conditions (� = “supply” of licensing; ı = “demand” ıH).
can still earn the monopoly profits in its own market. Finally, �HıLoccurs when the technology is dedicated and the product marketis fragmented. Licensing can take place when the licensee has theinternal competence and resources to adapt the dedicated technol-ogy from another submarket into technology needed for its ownsubmarket. The least likely case would be if the licensee demandsa GPT to operate in the same product market as the licensor.
3. Industry and patents
3.1. Industry
The SSI has experienced unprecedented growth in recent years.Since its initiation at the end of the 1980s, the SSI world mar-ket reached US$8.9 billon by 2001, up from US$6.3 billion in 2000and US$4.4 billion in 1999 (International Data Corporation, 2002).However, SSI technology is much older than the SSI market. Cryp-tography and encryption studies in computer science researchgrew considerably during the early 1970s, especially thanks tomilitary investments. It was not until the late 1980s that civiliandemand for software security products started rising. The incep-tion of the industry also coincided with the growing market forpersonal computers and the development of the Internet, with itsWeb-based financial transactions and data transmissions. In turn,growing commercial demand exerted different requirements thatbroadened the spectrum and complexity of the products and ser-vices offered. Accordingly, the SSI experienced a proliferation ofproduct submarkets, ranging from basic security software, such asvirtual private networks, firewalls, and virus scanning, to advancedsecurity services such as public key infrastructures, security certi-fication, and penetration testing (Fosfuri and Giarratana, 2007). In
Network security andmanagement
Network security management packages thatguarantee the high performance functioning ofdifferent networks
Source: Fosfuri and Giarratana (2007) and Giarratana (2004).
/ Rese
cafv
3
idM“pbiislsoeMst.r2
aiwtTrtommt
pdevogtTitit“wmtE6apaobnca
A. Gambardella, M.S. Giarratana
ontract. It is therefore not surprising that SSI has developed anctive market for technology. In 2002, software products accountedor more than 50% of SSI sales; an additional 30% came from ser-ices, and the rest derived from licensing algorithms.
.2. Patents
Despite some remaining controversy about the efficacy of IPRsn software industries, substantial academic and anecdotal evi-ence suggests the importance of patents in software. Hall andacGarvie (2007: 30), using a classical Tobin’s q equation, find that
software patents are valued the same by the market as ordinaryatents before the 1994/1995 change in software patent policy,ut they are valued twice as much as ordinary patents follow-
ng the changes.” Mann and Sager (2007), in their analysis of themportance of patents for software start-ups (but not specificallyoftware patents), find significant correlations among patents, firmongevity, and the firm’s ability to obtain late-stage financing. Thepecialized business press also highlights increasing importancef patenting algorithms, especially because reverse engineering ismerging as a major threat to IPR management for software (Riskanagement, 2008). Sanjay Tandon, the CEO and founder of an IT
ecurity company that develops corporate security defenses, claimshat “developers would never think of shipping software without
. . patent protection. It is critical that we patent our sensitive algo-ithms from both competitors and malicious users” (Business Wire,008: 17).
In terms of our reference industry, the technological core of software security product is its crypto-algorithm, which spec-fies mathematical transformations performed on the data. It is
orth noting that it is not the software code that is patented buthe algorithm, or the step-by-step procedure for data processing.he speed of mathematical calculations and security levels rep-esent two main features of SSI products; the time required byhe encryption and decryption processes depends on the lengthf the mathematical algorithms and the power of the computingachines (Giarratana, 2004). A good algorithm attempts to mini-ize the computer time needed to perform the transformation of
he data at a specified security level threshold.Crypto-algorithms typically are the principal object of a firm
atent; SSI patents also usually report step-by-step encryption andecryption routines, as well as detailed descriptions of the math-matical procedures for performing the encryption. Algorithmsary in their degree of technological breadth, which suggests notnly that software security firms patent algorithms or technolo-ies more generally but also that significant variability occurs inhe generality of these technologies and the underlying patents.hey therefore provide a good proxy of the firm-level capabil-ty to develop more or less general technologies. For example,he U.S. Patent and Trademark Office (USPTO) number 7,266,845s a software security patent entitled “Maintaining virus detec-ion software.” The first page of its abstract denotes its specificity:A method of managing a virus signature database associatedith an anti-virus application, both of which are resident in aemory of a mobile wireless.” Other patents instead entail broad
echnologies that can be used in many applications, such as thelliptic Curve Cryptography patent of Certicom (USPTO number,141,420), which has 52 different principal claims and protectsn algorithm that needs only 160 computer bits to perform all itsrocedures (a standard string needs 1024 bits). Standard systemsre based on integer calculus, but the Certicom algorithm reliesn elliptic curves that can be calculated more easily and faster
ut provide the same level of security. Thus, the underlying tech-ology is fundamentally a new method for making the securityheck quicker and more thorough, and its potential applicationsre wider.arch Policy 42 (2013) 315– 325 319
The object of these patents is the algorithm, not the source code;the same algorithm could be written according to different codes.Therefore, open versus closed source business models are not par-ticularly relevant to our research context.
4. Empirical evidence
4.1. Sample and dependent variables
We select our sample according to a technology criterion. Weuse a private data set, the LECG Corptech Patent data set, whichcovers approximately 80,000 software patents granted by USPTOfrom 1976 to 2001. This consulting firm specializing in technologydeals was active until 2011; it set up a team of patent experts thatreviewed software patent documents and applied several selectioncriteria, pertaining especially to the content of the patent abstractsand claims, to create the Corptech Patent data set. According toLECG experts, this procedure helped minimize the potential errorof including a patent that was not a software patent. We selectedall patents in U.S. technological classes 380 (“Cryptology”) and 705,subclasses 50–79 (“Business Processing Using Cryptography”), thatrepresent the sample of pure cryptology algorithms, plus “crypto-graphic apparatus or methods uniquely designed for or utilized inthe practice, administration, or management of an enterprise, theprocessing of financial data, or where a charge for goods or servicesis determined” (USPTO Classification Manual). Eighty-seven firmsown at least one of these patents; they provide the best proxy of anideal sample of all firms that possess an SSI algorithm at USPTO, thecore technology in this industry. On average, LECG Corptech Patentexcludes 73.6% of total patents in the aforementioned classes. Wecombine this information with data from SSI licensing deals fromthe beginning of the industry in 1989 until 2001. Our data source isthe Gale Group’s Infotrac Promt (www.gale.cengage.com) database(Pennings and Harianto, 1992). Promt offers comprehensive, reli-able coverage of companies, products, markets, alliances, and dealsfrom a vast collection of journals, newsletters, news releases, andnewspapers. We construct our database from firm news announce-ments in Promt, classified by standard industrial classification (SIC)code 73726, which corresponds exactly to software security, andfirm events identified as “Licensing Agreements.” We read eachtext to confirm the technological nature of the contract. Before1993, the sample firms did not release licensing contracts, so werestrict our analysis to 1993–2001 and obtain a panel data set of 694observations from 87 firms over nine years. Note that the databasecaptures the first licensing contract of a firm in the SSI field in itshistory.
With a hazard model, we estimate the probability of selling atechnology in year t, conditional on owning the technology andnot having licensed it at t − 1 (hazard rate). We opt for the piece-wise exponential model specification, which does not make strongassumptions about time dependence (Blossfeld and Rohwer, 2002).The hazard model estimates the probability that a firm issues itsfirst license at any point in time; after the firm issues its first license,it exits the sample, so its subsequent licenses cannot affect theestimation (cf. a panel logit approach that determines the prob-ability of licensing at any point in time). We prefer hazard modelsbecause they provide a stricter test of our theory and because we donot observe exclusive licensing in our sample. Therefore, the firstlicense is a signal of the firm’s strategic decision to adopt a licens-ing business model, in which the firm uses technology licensing,along with product development, as its strategy to earn profits. The
shadow dependent variable is LICENSEit, which takes a value of 1if in year t firm i announces at least one licensing deal as an algo-rithm seller and 0 otherwise. Of our 87 firms, 36 signed at least onelicensing contract as a seller.
320 A. Gambardella, M.S. Giarratana / Research Policy 42 (2013) 315– 325
Table 3Descriptive statistics.
Description Mean St. Dev. Min Max
LICENSEit Dummy = 1 if firm has ≥1 license in year t 0.123 0.328 0 1CLAIM PATit-1 Total # of claims in SSIPATENT divided by (1 + SSIPATENT)—average by firm
and year7.017 10.316 0 94.75
GENERALITYit-1 Trajtenberg’s index of generality in SSIPATENT—average by firm and yearcorrected according to Hall (2005)
0.111 0.191 0 1.112
FRAGM MKTt Share of software security products sold by firms that sell products in only oneof the 6 niches in Table 1
0.116 0.094 0.019 0.256
AGEi Firm age in 1993 14.494 22.199 0 117RDit-1 R&D expenditures of the firm ($ ml) 226.013 721.060 0 6522SSIPATENTit-1 Stock of # U.S. software security patents of the firm (see USPTO classes in text),
depreciated at 15%53.605 166.759 0 361.75
COSPEC ASSETSit-1 Firm sales times its share of software trademarks on total trademarks ($ ml) 1024.377 3752.077 0 36451.215SWPATENTit-1 Stock of # U.S. software patents of the firm depreciated at 15% 75.452 180.52 4 1245.70SOFTWAREi Dummy = 1 if firm core business is in SIC 737 0.563 0.496 0 1HARDWAREi Dummy = 1 if firm core business is in SIC 357 0.126 0.333 0 1ELECTRONICSi Dummy = 1 if firm core business is in SIC 359–370 0.138 0.345 0 1
5
t
5
ttrrslwb2s(tsc
oeWodtwbhhicb
neowatm
mark could protect the name of a product, a brand (colors, fonts),or advertising campaigns. Trademarks owners must prove thatthey are selling the products on the market; the marks may be
NORTH AMERICAi Dummy = 1 if firm headquarter is in NA
EUi Dummy = 1 if firm core business is in EuropeTREND Trend variable that starts with 1 in 1993.
. Independent variables
In Table 3, we define all the variables in our analysis, along withhe descriptive statistics.
.1. Generality of the technology
Following consolidated examples in prior literature, we employwo alternative measures. The first, CLAIM PATit-1, is a proxy ofhe annual average number of claims of the firm in software secu-ity patents, lagged by one year. We construct this variable as theatio between (1) the total number of claims by the firm in theoftware security USPTO classes 380 and 705, subclasses 50–79,agged by one year, and (2) 1 plus the total number of patents from
hich the claims are drawn. Several studies argue that the num-er of claims provides a direct measure of scope; Hall et al. (2001:3–24) state that “the number of claims may be indicative of thecope or width of the invention” and Lanjouw and Schankerman2004: 448) note that “the number of claims is also an indicationhat an innovation is broader.” Our confidence in this measure istrong, because USPTO examiners check the consistency of theselaims.
Our second measure, GENERALITYit-1, is the annual averagef patents by the firm in the generality index developed by Hallt al. (2001). We collect this measure from the NBER patent dataeb site (www.nber.org); it equals 1 minus the Herfindhal index
f the forward citations to the patent from the USPTO three-igit patent classes. If a patent is cited by subsequent patentshat belong to a wide range of fields, the measure is higher,hereas if most citations are concentrated in a few fields, it will
e lower. If forward citations indicate the impact of a patent, aigh generality score suggests that the patent presumably hasad widespread impact and influenced subsequent innovations
n various fields. We correct this measure according to the pro-edure in Hall (2005) for accounting for possible small firmiases.
We performed interviews with four managers of SSI compa-ies and four inventors of cryptology algorithms who were notmployed in our sample firms. They confirmed most features ofur analysis. In particular, these respondents stated that patents
ere effective for protecting security algorithms and that patentedlgorithms normally provided firms with a two- to three-year leadime over competitors. Very good algorithms could even persist for
ore than ten years.
0.851 0.357 0 10.069 0.254 0 14.560 2.524 1 9
5.2. Product market fragmentation
Higher fragmentation is typically associated with products withdifferent features, which implies low economies of scope in thedownstream assets across submarket niches. The higher productmarket fragmentation is, the more pronounced are the specializa-tion advantages, and the higher the probability that firms operatein single submarket niches. Therefore, FRAGM MKTt is the annualshare of all products released in SSI by firms that sell productsin only one of SSI product submarkets of Table 1, that is, theshare of products sold by specialized firms.1 Heterogeneity existsin our measure; as we show in Table 3, the minimum value ofFRAGM MKTt is 0.019, and the maximum is 0.259. To give an ideaof the variability over time of this variable we plot FRAGM MKTt
along with other variables of interest.
5.3. Controls
Firm age, calculated as the difference between the current yearand the foundation year (Agei), provides a proxy for firm experi-ence. To control for firm innovations, we use both firm-level totalR&D expenditures (RDit-1) (in millions of USD) and the stocks ofsoftware patents outside software security (SWPATENTIit-1) andwithin software security (SSIPATENTIit-1). The latter equals thenumber of patents granted to each firm in the USPTO classes 380and 705, subclasses 50–79. Both measures are lagged by one year.We construct the patent stock from the perpetual inventory for-mula with a depreciation rate of 0.15, as is typical for constructingR&D stock (Lanjouw and Schankerman, 2004). We use the totalnumber of patents by the firm in 1976 as the initial stock. We exper-imented with other depreciation rates, but the empirical results donot change.
We control for downstream co-specialized assets of the firm,according to the share of live software trademarks as a proportion oftotal firm trademarks, multiplied by firm sales (Cospec assetsit-1),both lagged by one year. Trademarks are combinations of “words,phrases, symbols, or designs that identify and distinguish thesource of the goods (or services)” (USPTO, www.uspto.gov). A trade-
1 We construct this measure from information about all products sold by firms inSSI, not just those of our 87 firms (SSI Data set).
A. Gambardella, M.S. Giarratana / Research Policy 42 (2013) 315– 325 321
Table 4Hazard rates for piecewise exponential model of licensing, 1993–2001.
Model I Model II Model III Model IV
CLAIM PATit-1 1.625**
(0.102)1.155**
(0.148)GENERALITYit-1 6.849**
(1.468)4.778**
(0.302)FRAGM MKTt 4.022**
(0.502)3.065**
(0.564)2.789**
(0.3879)3.251**
(0.536)FRAGM MKTt
* Claim patit-1 1.826**
(0.084)FRAGM MKTt
* Generalityit-1 4.479**
(0.824)Agei 0.601**
(0.131)0.538**
(0.087)0.442**
(0.108)0.500**
(0.097)RDit-1 1.003
(0.176)1.102(0.227)
1.064(0.201)
1.012(0.165)
SSIPATENTit-1 0.844(0.176)
0.812(0.305)
0.952(0.166)
0.825(0.221)
COSPEC ASSETSit-1 1.117**
(0.090)1.265(0.121)
1.070(0.173)
1.104(0.152)
SWPATENTit-1 1.013(0.009)
1.011(0.008)
1.014(0.009)
0.1.014(0.008)
Trend 0.559**
(0.061)0.725**
(0.046)0.648**
(0.072)0.715**
(0.012)Time, core sector and geographical dummies Yes Yes Yes YesNumber of observations 694 694 694 694Log-likelihood 528.91 529.572 529.39 530.24
Notes: Heteroskedastic-consistent standard errors in parenthesis. All variables are log(1 + x). Betas > 1 imply a positive effect.* p < 0.10.
** p < 0.05.
in the
ara2ff
sppWtd5I0
Fig. 1. Trends
bandoned, cancelled, or allowed to expire if they have not beenenewed after a fixed period. Therefore, trademarks offer a proxy for
firm’s downstream assets (e.g., Fosfuri et al., 2008; Krasnikov et al.,009); we proxy for software-specific downstream assets because,
ollowing Teece (1986), we argue that co-specialized assets matteror innovation.22 To find software trademarks, we applied a search algorithm (“computeroftware,” “operating system,” “computer program,” “software algorithm,” “datarocessing,” or “software application”) to the front page of the trademark, whichrovides the description of the good or service trademarked in the USPTO database.e validated our measure with the share of firm sales in software compared with
otal revenues, as published by Software Magazine in the Software 500 List (surveyata, www.softwaremag.com). For four sample firms that appear on the Software00 List, the trademark and revenues shares are very similar: H&P 0.28 versus 0.30;
BM 0.52 versus 0.46; Sun Microsystems 0.32 versus 0.35; and Apple 0.22 versus.20 (2000 data).
main series.
We employ three dummies to control for the firm’s core busi-ness; they take values of 1 if the core business is software (SIC code737, Softwarei), hardware (SIC code 357, Hardwarei), or electron-ics (SIC code 359–370, Electronicsi), and 0 otherwise. We use twogeographical dummies that take a value of 1 if the firm headquar-ters are located in North America (North Americai) or Europe (EUi),and 0 otherwise. We also add a trend variable Trendt to control forpossible confounding effects due to trends in the series (Fig. 1).
All these firm data are drawn from Bureau Van Dijk’s Osiris andLECG Corptech data sets. Finally, given the hazard function, we esti-mate time according to a piecewise-constant model, in which weallow the constant rate to vary within predefined time segments,which we designate as years.
6. Estimation results
In Table 4, we provide the results of four hazard models(I–IV). The first two models use Claim patit-1 as a measure of
322 A. Gambardella, M.S. Giarratana / Research Policy 42 (2013) 315– 325
ntatio
tMmI
ealrruwGoiutdmop2aFhtc
poFcaeGsb
C
Fig. 2. Graphical represe
echnological breadth, whereas the other two use Generalityit-1.odels I and III do not include the interaction term between theeasure of technological breadth and Fragm mktt, whereas Models
I and IV do.We first find that firms with the capability to produce more gen-
ral technology are more likely to license it. Our first hypothesis wasgnostic about whether in homogenous markets firms were moreikely to license a dedicated or a general technology. In this respect,ecall from our previous discussion that a more general technologyaises the probability that a licensing deal occurs when the prod-ct market is fragmented, and it does not have a substantial impacthen the product market is homogenous. We find that, on average,PT increases the probability of licensing: In all models, the impactsf Claim patit-1 and Generalityit-1 are statistically significant. Thiss probably due to the fact that in SSI there is some degree of prod-ct differentiation even within homogenous product markets, andhus general technologies tend to be used by the licensees in suchifferentiated applications. This means that even in homogenousarkets the licensors are more willing to license a general technol-
gy because the licensee will not really impinge on the very sameroducts of the licensor. Second, we corroborate our Hypothesis. As the industry becomes more fragmented, the probability of
technology holder issuing a license increases. The coefficient ofragm mktt, multiplied by Claim patit-1 or Generalityit-1, definesow market fragmentation attenuates or strengthens the effects ofechnology generality. Our estimation shows that the interactionovariate has a significant positive impact on licensing probability.
To clarify our findings, we report the estimates of the multi-lier rate of firm hazard of licensing, conditional on different valuesf fragmentation, claims per patents, and the generality index, inig. 2. Because the hazard model is inherently multiplicative, weannot discuss the absolute relation between an independent vari-ble and the licensing probability; instead, we analyze the partialffects relative to the multiplier rates of Table 3. Using a firm with noPT in a average fragmentation period as a baseline, Fig. 2 shows the
imulation for Claim patit-1 (Panel a) and for Generalityit-1 (Panel).With regard to the controls, we find that the impact ofOSPEC ASSETSit-1 is insignificant in one model. This finding
n of results from Table 4.
replicates Gans et al.’s (2002) evidence that complementary assetownership does not exert any significant effect on the probabilitythat start-up innovator firms license a technology. The total R&Dof the firm and patent stocks do not have sizable or significantimpacts on the hazard rate of licensing; this result is not surpris-ing, because the hazard rate measures the probability that the firmproduces its first license, whereas R&D and patent stock representbetter measures of the scale of licensing. Fosfuri (2006) obtains thesame results in the chemical industry regarding the effect of patentson firm technological licensing. Finally, age has a sizable and sig-nificantly negative impact on the hazard rate, which suggests thatlicensing in SSI is a business of young companies. As expected, thetrend variable exerts a negative impact on licensing decision.
6.1. Is GPT truly a firm capability?
One concern related to our empirical analysis is whether ourcovariates of interest capture the effect of a capability of firms (i.e.,production of GPTs), or their ability to produce one or a few broadpatents. In short, have we discovered a firm or a patent effect? Tostrengthen our evidence in support of the capability view, we runthe same hazard model of Table 5 but substitute the firm averagevalues of CLAIM PATit-1 and GENERALITYit-1 with two alternativemeasures that offer direct proxies of the presence of outliers: theirmaximum values and the means plus the standard deviation. Withthese checks, we test if our results are determined by right-tailedvalues of the distribution of our core variables. If they are, we cannotconfirm whether we were dealing with a true firm capability effector with the ability to produce a few star patents.
In Table 5, we report the results of the hazard model. Consistentwith our arguments, the right-tail proxies do not exert any effect onthe hazard of licensing. Thus the true heterogeneity among firmsoccurs at the level of their average capability to produce GPT, andthis heterogeneity decreases toward the right tail of the variabledistribution. In summary, the statistically important feature is the
ability to produce GPT, not the ability (or luck) to own a very generalpatent.As a further robustness check, we tested whether generalitymatters in the case of product innovations rather than licensing.
A. Gambardella, M.S. Giarratana / Research Policy 42 (2013) 315– 325 323
Table 5Piecewise hazard rates for piecewise exponential model of probability of licensing, 1993–2001.
Model I Model II Model III Model IV
CLAIM PATit-1 (maximum value) 1.025(0.848)
CLAIM PATit-1 (mean + standard deviation) 1.030(0.915)
GENERALITYit-1 (maximum value) 0.952(1.012)
GENERALITYit-1 (mean + standard deviation) 0.947(1.002)6.345
FRAGM MKTt 4.212**
(0.458)3.701**
(1.023)3.644**
(0.857)2.875(0.924)
FRAGM MKTt* CLAIM PATit-1 0.925
(0.726)1.002(0.821)
FRAGM MKTt* GENERALITYit-1 1.022
(0.804)1.03103(0.855)
AGEi 1.071**
(0.031)1.068**
(0.038)1.098**
(0.044)1.125**
(0.056)RDit-1 0.394
(0.852)0.246(0.507)
0.187(0.411)
0.114(0.324)
SSIPATENTit-1 0.854(0.351)
0.894(0.363)
0.821(0.379)
0.827(0.360)
Cospec assetsit-1 1.045(0.452)
1.022(0.424)
1.015(0.452)
0.941(0.491)
SWPATENTit-1 1.005*
(0.003)1.004(0.008)
1.001(0.006)
1.001(0.006)
Trend 0.657**
(0.051)0.602**
(0.049)0.548**
(0.042)0.671**
(0.032)Time, core sector and geographical dummies Yes Yes Yes YesNumber of observations 694 694 694 694Log-likelihood 526.02 527.48 528.12 527.48
N 1 + x).
Toprttnt
TP
N
otes: Heteroskedastic-consistent standard errors in parenthesis. Variables are log(* p < 0.10.
** p < 0.05.
he market for SSI products is separate from the market of technol-gy; algorithms depend on the ability of one or a few researchers toroduce clever mathematical or logical structures, but successfuleleases of new products depend on a more complex set of fac-
ors. In Table 6 we report the results of a hazard model in whichhe shadow dependent variable is the probability of releasing aew product a time t. All our sample firms, at a certain point inime, have released a software product. Table 6 shows that the keyable 6iecewise exponential model for probability of product release, 1993–2001.
Model I Mod
CLAIM PATit-1 1.019(0.176)
0.87(1.46
GENERALITYit-1
FRAGM MKTt 6.690**
(0.259)5.04(0.26
FRAGM MKTt* CLAIM PATit-1 1.19
(2.34FRAGM MKTt
* GENERALITYit-1
Agei 0.613**
(0.085)0.61(0.08
RDit-1 1.185(0.146)
1.18(0.14
SSIPATENTit-1 1.020**
(0.010)1.01(0.01
COSPEC ASSETSit-1 1.151**
(0.074)1.15(0.07
SWPATENTit-1 0.998(0.007)
0.99(0.00
Time, core sector and geographical dummies Yes Yes
# of obs. 664 664
Log-Lik 465.238 465.
otes: Heteroskedastic-consistent standard errors in parenthesis. Variables are log(1 + x).* p < 0.10.
** p < 0.05.
determinants of new product releases are the market fragmenta-tion and co-specialized assets. Technological generality does notmatter at all, whether linearly or in interaction with FRAGM MKTt.Therefore, technological generality and its interaction with frag-
mentation genuinely and selectively capture factors associatedwith the licensing decision, not a trend produced by the growthof the industry in this period. It is also worth noting that soft-ware patents become significant for explaining product release, ael II Model III Model IV
18)
0.387(0.464)
6.711(13.837)
0**
0)6.918**
(0.263)1.735**
(0.091)8*
0)3.320(7.757)
4**
5)0.611**
(0.083)0.609**
(0.084)79)
1.179(0.145)
1.157(0.145)
8*
1)1.035**
(0.012)1.031**
(0.011)2**
4)1.149**
(0.074)1.150**
(0.074)48)
0.945(0.026)
0.966(0.022)
Yes Yes664 664
240 465.461 465.917
3 / Rese
fit
owCfidtp
7
mtgstp
tlrsed
apcseaict(fHppdfitd
etrvScmtairam
lhts
24 A. Gambardella, M.S. Giarratana
nding that confirms the importance of patented algorithms as aool to protect product revenues.
We performed several other checks too. Running regressionsnly for software firms resulted in similar results. As an experiment,e ran the regression using all patents (not just those from theorptech Patent data set) granted by the USPTO in SSI classes to ourrms. The results become mixed in this case, confirmed only for theirect effect of the generality index at a 10% significance level. Thus,here appears to be high heterogeneity in firms’ software patentortfolios.
. Conclusions
This article highlights an important but understudied deter-inant of external knowledge exploitation (Argote et al., 2003):
he combination of a firm’s technology capability (i.e., to produceeneral purpose technologies) and the fragmentation of its down-tream product market. We test our predictions using data fromhe software security industry, in which technologies and productsrovide two important sources of revenues.
Our work extends a recent research trajectory that highlightshe importance of firm-based determinants of the propensity foricensing, compared with exogenous conditions such as the IPRegime (Fosfuri, 2006; Gambardella et al., 2007). We jointly demon-trate the importance of a firm’s technological capability andnvironmental conditions for explaining heterogeneity in licensingecisions (Winter, 2003).
Understanding the link between a technological capability and market feature has remarkable significance for researchers andractitioners. Our argument implies that technology strategy is aomplex construct, in which IPR protection and downstream co-pecialized assets are not the only pillars (Teece, 1986). Technologyntrepreneurs and managers that hope to use external exploitations a strategy to profit from innovation should realize ex ante themportance of the type of innovation they produce. Firms with theapability to produce general technologies are better positionedo profit from innovation through licensing. Many start-up firmsDushnitsky, 2010) use licensing to raise initial cash while waitingor longer-term returns on their investments to produce products.owever, licensing may affect their opportunities to move intoroduct production if their licenses apply to only one or a fewroduct domains in which the licensor and licensee would competeirectly. The capability to develop general technologies implies thatrms can license without jeopardizing their expected returns inhe market for products, because they can sell their technologies inistant product markets.
Our study highlights another novel facet of external knowledgexploitation: a firm’s positioning in product spaces becomes cen-ral to the formulation of its technology strategy. This reasoningeflects growing research describing the different impacts of inno-ation in markets and industries, as highlighted by McGahan andilverman (2006) and Bloom et al. (2007). Technology transactions,oupled with the fragmentation of downstream product markets,ight be mechanisms underlying different trajectories of indus-
ry evolutions, such as shakes-outs versus non-shake-outs (Kleppernd Thompson, 2006). More generally, our analysis highlights themportance of considering both vertical and horizontal linkages andivalry to understand the growth of industries, and it encourages
better understanding of how technology markets and productarket competition are interconnected.This discussion in turn raises interesting questions about the
evel at which this capability could be nurtured within a firm. Weave assumed that it is an exogenously given capability. We thinkhis assumption is likely in the short run, but perhaps firms haveome degree of maneuverability in the long run. Potential avenues
arch Policy 42 (2013) 315– 325
of research therefore might consider antecedents of this capability,especially in the realm of human resource management for inven-tors (Singh and Fleming, 2010), such as recruiting (if a matter ofindividual exogenous capability), contract incentives (if inventorshave the option of creating general technologies), and the organi-zation of labor (team versus individual work).
Our article complements a classical transaction cost view (Ganset al., 2002, 2008; Teece, 1986), in which the effectiveness of IPR is apivotal exogenous determinant of licensing. There could be a rela-tionship between the level of generality of a technological portfolioand the real (or perceived) rate of protection of technology. Addi-tional studies should relate technology generality to the ability towrite well-defined licensing contracts or a firm’s bargaining powerin negotiation (Arora and Ceccagnoli, 2006). In this case, a morefine-grained analysis of the type of claims of each patent, classify-ing for example claims within a product submarket versus claimsacross product submarkets, could add new insights into the truelevel of generality of the technology.
Of course, our study is not without limitations. First, our evi-dence is based on one industry. Therefore, we can test our theorywith only one IPR regime, which is ideal in SSI, where patents mat-ter. Studies in different industries might attempt to corroborate ourtheory in setting marked by greater competition among technologysuppliers. Second, we treat fragmentation as exogenous, similar toKlepper and Thompson (2006). However, firms may make market-ing investments to change fragmentation into standardization, orvice versa, according to whether they possess dedicated or generalcompetencies.
These and other issues offer several avenues for research intothe relationships among the breadth of firm competencies, indus-try structure (both vertical and horizontal), and firm strategies forexploiting external knowledge in technology markets.
Acknowledgements
We thank Ashish Arora, Andrea Fosfuri, Myriam Mariani andGiovanni Valentini. We acknowledge financial support from Euro-pean Commission, Project INNOS&T, Innovative S&T indicatorscombining patent data and surveys: Empirical models and policyanalyses, Contract no 217299. All errors are ours.
References
Argote, L., McEvily, B., Reagans, R., 2003. Managing knowledge in organizations: anintegrative framework and review of emerging themes. Management Science49 (4), 571–582.
Arora, A., Ceccagnoli, M., 2006. Patent protection, complementary assets, and firms:incentives for technology licensing. Management Science 52 (2), 293–308.
Arora, A., Fosfuri, A., 2003. Licensing the market for technology. Journal of EconomicBehavior and Organization 52 (2), 277–295.
Arora, A., Gambardella, A., 2010. Ideas for rent: an overview of markets for technol-ogy. Industrial and Corporate Change 19 (3), 775–804.
Bloom, N., Schankerman, M., Van Reenen, J., 2007. Identifying technology spilloversand product market rivalry. NBER WP # 13060.
Blossfeld, H.P., Rohwer, G., 2002. Techniques in Event History Modeling: NewApproaches to Casual Analysis. Lawrence Erlbaum Associates, Mahwah, NJ.
Bresnahan, T.F., Trajtenberg, M., 1995. General purpose technologies, engines ofgrowth. Journal of Econometrics 65 (1), 83–108.
Bresnahan, T., Gambardella, A., 1998. The division of inventive labor and the extentof the market. In: Helpman, E. (Ed.), General Purpose Technologies and EconomicGrowth. MIT Press, Cambridge, MA, pp. 253–281.
Business Wire, 2008. Who’s Guarding the Guards? April 7, pp. 17–22.Business Wire, 2000. Techniclone Expands VTA Patent Position; Company Receives
Broad Coverage of VTA Compositions, April 19, pp. 12–15.Cassimon, D., De Backer, M., Engelen, P.J., et al., 2011. Incorporating technical risk in
compound real option models to value a pharmaceutical R&D licensing oppor-tunity. Research Policy 40 (9), 1200–1216.
Cockburn, I., MacGarvie, M., Mueller, E., 2010. Patent thickets, licensing and innova-
tive performance. Industrial and Corporate Change 19 (3), 899–926.Dushnitsky, G., Klueter, T., 2011. Is There an eBay for Ideas? Insights from OnlineKnowledge Marketplaces. European Management Review 8 (1), 17–32.
Dushnitsky, G., 2010. Entrepreneurial optimism in the market for technologicalinventions. Organization Science 21 (1), 150–167.
/ Rese
F
F
F
F
G
G
G
G
G
G
H
H
H
H
H
H
IK
K
A. Gambardella, M.S. Giarratana
eldman, M.S., Pentland, B.T., 2003. Reconceptualizing organizational routines asa source of flexibility and change. Administrative Science Quarterly 48 (1),94–118.
osfuri, A., 2006. The licensing dilemma: understanding the determinants of the rateof technology licensing. Strategic Management Journal 27 (12), 1141–1158.
osfuri, A., Giarratana, M.S., 2007. Product strategies and survival in Schumpete-rian environments: evidence from the security software industry. OrganizationStudies 28 (6), 909–929.
osfuri, A., Giarratana, M.S., Luzzi, A., 2008. The penguin has entered the building.The commercialization of open source Software products. Organization Science19 (2), 227–291.
ambardella, A., Giuri, P., Luzzi, A., 2007. The market for patents in Europe. ResearchPolicy 36 (8), 1163–1183.
ambardella, A., McGahan, A., 2010. Business-model innovation, general purposetechnologies, specialization and industry change. Long Range Planning 4 (2–3),45–56.
ans, J., Hsu, D., Stern, S., 2002. When does start up innovation raise the gale ofcreative destruction? Rand Journal of Economics 33 (4), 571–586.
ans, J., Hsu, D., Stern, S., 2008. The impact of uncertain intellectual property rightson the market for ideas: evidence from patent grant delays. Management Science54 (5), 982–997.
ans, J., Stern, S., 2010. Is there any market for ideas? Industrial and CorporateChange 19 (3), 805–838.
iarratana, M.S., 2004. The birth of a new industry: entry by startups and thedrivers of firm growth. The case of encryption software. Research Policy 33 (5),787–806.
all, B.H., Jaffe, A., Trajtenberg, M., 2001. The NBER patent citations data file: lessons,insights and methodological tools. NBER Working Paper 8498.
all, B.H., 2005. A note on the Bias in the Herfindhal based on Count data. Revued’Economie Industrielle 110, 149–156.
all, B.H., MacGarvie, M., 2007. The private value of software patents, NBER WorkingPaper 12195, Cambridge.
all, B.H., Ziedonis, R., 2001. The patent paradox revisited: determinants of patentingin the US semiconductor industry, 1980–1994. Rand Journal of Economics 32 (1),101–128.
elfat, C.E., Winter, S.G., 2011. Untangling dynamic and operational capabilities:strategy for the (n)ever-changing world. Strategic Management Journal 32 (11),1243–1250.
ellmann, T., Perotti, E., 2011. The circulation of ideas in firms and markets. Man-agement Science 57 (10), 1813–1826.
nternational Data Corporation, 2002. Worldwide Software Security Report.
lepper, S., Thompson, P., 2006. Submarkets and the Evolution of Market Structure.Rand Journal of Economics 37 (4), 861–886.rasnikov, A., Mishra, S., Orozco, D., 2009. Evaluating the financial impact of branding
using trademarks: a framework and empirical evidence. Journal of Marketing 73(6), 154–166.
arch Policy 42 (2013) 315– 325 325
Lanjouw, J.O., Schankerman, M., 2004. Patent quality and research productivity:measuring innovation with multiple indicators. Economic Journal 114 (495),441–465.
Lichtenthaler, U., 2010. Determinants of proactive and reactive technology licensing:a contingency perspective. Research Policy 39 (1), 55–66.
Lichtenthaler, U., 2007. The drivers of technology licensing: an industry comparison.California Management Review 49 (4), 67–82.
Lichtenthaler, U., Lichtenthaler, E., 2009. Capability-based framework for open inno-vation: complementing absorptive capacity. Journal of Management Studies 46(8), 1315–1338.
Maine, E., Garsney, E., 2006. Commercializing generic technology: the case ofadvanced materials ventures. Research Policy 35 (3), 375–393.
Mann, R.J., Sager, T.W., 2007. Patents, venture capital, and software start-ups.Research Policy 36 (2), 193–208.
Kani, M., Motohashi, K., 2012. Understanding the technology market for patents:new insights from a licensing survey of Japanese firms. Research Policy 41 (1),226–235.
McGahan, A.M., Silverman, B.S., 2006. Profiting from technological innovation:the effect of competitor patenting on firm value. Research Policy 35 (8),1222–1242.
Nelson, R.R., 1959. The simple economics of basic scientific research. Journal ofPolitical Economy 67 (2), 297–306.
OECD, 2006. OECD Technology Indicators, Technology Balance ofPayment—Payments/Receipts. OECD, Paris.
Palomeras, N., 2007. An analysis of pure-revenue technology licensing. Journal ofEconomics and Management Strategy 16 (4), 971–994.
Pennings, J.M., Harianto, F., 1992. The diffusion of technological innovation in thecommercial banking industry. Strategic Management Journal 13 (1), 29–46.
Risk Management, 2008. Steal this article. 55(5), 25–35.Rosenberg, N., 1976. Technological change in the machine tool industry. In:
Rosenberg, N. (Ed.), Perspectives on Technology. Cambridge University Press,Cambridge, MA, pp. 11–31.
Singh, J., Fleming, L., 2010. Lone inventors as sources of breakthroughs: myth orreality? Management Science 56 (1), 41–56.
Sorenson, O., 2003. Interdependence and adaptability: organizational learn-ing and the long-term effect of integration. Management Science 49 (4),446–463.
Teece, D.J., 1986. Profiting from technological innovation. Research Policy 15 (6),285–305.
The Economist, 2007. Business by Numbers. September 13.Thoma, G., 2009. Striving for a large market: evidence from a general purpose tech-
nology in action. Industrial and Corporate Change 14 (2), 107–138.Von Hippel, E., 1994. ‘Sticky information’ and the locus of problem solving: impli-
cations for innovation. Management Science 40 (4), 429–439.Winter, S.G., 2003. Understanding dynamic capabilities. Strategic Management Jour-
nal 24 (10), 991–995.