This article was downloaded by: [194.221.86.206] On: 21 December 2017, At: 01:45Publisher: Institute for Operations Research and the Management Sciences (INFORMS)INFORMS is located in Maryland, USA
Management Science
Publication details, including instructions for authors and subscription information:http://pubsonline.informs.org
On the Effectiveness of Patenting Strategies in InnovationRacesJürgen Mihm, Fabian J. Sting, Tan Wang
To cite this article:Jürgen Mihm, Fabian J. Sting, Tan Wang (2015) On the Effectiveness of Patenting Strategies in Innovation Races. ManagementScience 61(11):2662-2684. https://doi.org/10.1287/mnsc.2014.2128
Full terms and conditions of use: http://pubsonline.informs.org/page/terms-and-conditions
This article may be used only for the purposes of research, teaching, and/or private study. Commercial useor systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisherapproval, unless otherwise noted. For more information, contact [email protected].
The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitnessfor a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, orinclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, orsupport of claims made of that product, publication, or service.
Please scroll down for article—it is on subsequent pages
INFORMS is the largest professional society in the world for professionals in the fields of operations research, managementscience, and analytics.For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
Which, if any, of a firm’s inventions should it patent? Should it patent at all? Many companies engaged in aninnovation race seek a patenting strategy that balances protection of their intellectual property against the
knowledge spillovers resulting from disclosure requirements. Not much is known about factors that determinethe patenting strategy best able to resolve this trade-off. Although scholars in various management, economics,and engineering disciplines have researched patents and patenting regimes, little work has addressed the nor-mative issues that pertain to forming an appropriate firm-level patenting strategy. We develop an inventory ofreal-life patenting strategies and integrate them into a coherent framework. Our simulation model character-izes the optimal patenting choices for different environmental and firm-level contingencies while capturing thedynamics between competing firms. We identify the firm’s research and development strategy as the most salientdeterminant of its optimal patenting strategy. Our research contributes to establishing a contingency theory ofpatenting strategies.
Keywords : patent strategies; innovation races; NK modeling; games on NK landscapes; strategic interaction onNK landscapes
History : Received July 15, 2013; accepted November 21, 2014, by David Hsu, entrepreneurship and innovation.Published online in Articles in Advance July 9, 2015.
1. IntroductionShould a company engaged in an innovation racepatent a certain invention? More generally, whatinventions should the company patent? (All of them?None at all? A select subset?) And upon what fac-tors should the choice of such a patenting strategydepend?
Consider a case from the semiconductor industry.The Dutch company ASML is practically a monopolistprovider of the lithography equipment upon whichevery advanced semiconductor manufacturer (e.g.,Intel and Samsung) relies when developing their next-generation products. Without ASML continuouslyimproving its critical equipment, Moore’s law (Moore1965) would cease to hold. In 2010, ASML facedthe challenge of fundamentally revising the archi-tecture of their lithography tools; in particular, thecompany would need to introduce extreme ultravio-let lithography (EUVL) for its next-generation equip-ment. So for the first time in a decade, ASML wouldhave to exchange its most critical product component:the light source. Two outside companies—Cymer andXtreme Technologies—were competing to becomethe supplier of this light source. Cymer planned to
use laser-produced plasma (LPP), whereas XtremeTechnologies planned to use discharge-producedplasma (DPP) (Adee 2010). Although ASML testedboth technologies simultaneously (Benschop 2010), inthe end only one company would be selected. Xtremeand Cymer were in an innovation race.
This competitive setting resulted in Xtreme Tech-nologies facing a key trade-off when decidingwhether or not to patent its inventions. Filing forpatents would protect the company’s products frombeing imitated by the competition, but at the costof disclosure. That is, Xtreme Technologies wouldhave to disclose—well before the affected productscould be marketed—the technology it was workingon and the kinds of solutions it envisioned. ThenCymer, after analyzing Xtreme Technology’s patentapplication portfolio, could redirect its own researchand development (R&D) efforts and possibly “designaround” Xtreme’s patents before Xtreme’s productswere even available in the market. In this way, the fil-ing of patents by Xtreme Technologies could uninten-tionally accelerate a competitive response that mighteven result in being leapfrogged by its competitor.
The preceding example illustrates the trade-off thatall patenting strategies must negotiate. On the onehand, patents protect technological solutions; thisprevents competitors from commercializing copiedproducts (traditional patent motive) and may alsodissuade competitors from entering a technologicalarea (strategic patent motive). On the other hand, apatenting firm necessarily and clearly discloses itsresearch activities, reveals its location on the tech-nology landscape, and—perhaps most interesting tocompetitors—gives signals about the quality of solu-tions resulting from its R&D efforts. Thus, protect-ing and blocking technologies comes at the costof sacrificing both secrecy and a possible lead-timeadvantage.
The case of Xtreme Technologies raises a sim-ple question: Which of its inventions should a firmengaged in an innovation race patent? Answering thisquestion requires knowledge of three related aspects.First, the firm must determine the goal of its patent-ing activities. Second, it must identify its patentingoptions. The trade-off just described does not requirethe firm to choose only between patenting and notpatenting for each patent; after all, it is possible toconceive of many different patenting strategies. Third,the firm needs to consider its circumstances. Theeffects of patenting strategies are not universal and, inparticular, their benefits may be contingent on differ-ent environmental and firm-level factors. Salient envi-ronmental factors include the technology landscape,industry clock speed, and the competitive situation.As for the firm-level factors, it is sometimes difficultto distinguish between R&D strategy and patentingstrategy.
Starting with the seminal work of Horstmannet al. (1985), economists have analyzed the behav-ior of firms that face the trade-off betweenprotection and disclosure inherent in patenting sys-tems (cf. Scotchmer and Green 1990, Waterson 1990,Gallini 1992). These economists employ stylizedgame-theoretic models to analyze firms’ propensitiesto file for patents. However, their main interest isin solving economy-level problems—that is, givingadvice on effective patent regulations—and hence thecircumstances that surround the patenting decisionsare modeled with a high level of abstraction. Therepresentation of firm-level R&D strategy (as a fullyadjustable and rational dynamic research effort) andalso that of the firm’s patent strategy (as a binary orprobabilistic choice for each invention) are convenientfor modeling but rudimentary at best. So, despitethis rich body of research on patents in innovationraces, we lack a comprehensive and nuanced under-standing of how firms can design patenting strategiesto their competitive advantage while taking environ-mental and firm characteristics into account.
This paper makes three contributions. First, it con-stitutes a first effort to devise a firm-level theory ofpatenting in innovation races: we build a coherentinventory of patenting strategies—one that system-atizes and complements existing knowledge of suchstrategies—and then integrate that knowledge into acommon framework that incorporates as contingen-cies the firm’s R&D strategy, the technological land-scape, and competitor behavior. Second, we providesimulation evidence on the effectiveness of patentstrategies as a function of own-firm and competitorR&D strategies, the competitor’s patenting strategy,and industry conditions. Our study thus helps laythe foundation for a comprehensive contingency the-ory of patenting strategies. Third, we introduce theconcept of dynamically interacting agents to the NKframework and thereby extend existing methodology.Because the NK framework is well suited to representthe search aspects of R&D, we make it the basis of ourmodeling efforts. Yet, our research question demandsthat we explicitly capture the dynamic strategic inter-actions between firms, so we extend the existing NKmethodology accordingly.
Patenting policies are concerned with fostering inno-vation in the economy (Reinganum 1982, Judd 1985,Gilbert and Shapiro 1990, Klemperer 1990). Suchpolicies need to balance protection requirements,which provide incentives for research, with disclo-sure requirements, which lead to spillovers and thusaccelerate technology development (Scotchmer andGreen 1990, Waterson 1990). Hence, when decidingto patent or not, individual firms incorporate thetension between knowledge protection and knowl-edge leakage (Horstmann et al. 1985, Cohen et al.2000). The literature has identified several factors thataffect how firms handle this trade-off: patent breadthand length (Gallini 1992, Denicolò and Franzoni 2004,Kwon 2012), patent novelty requirements (Scotchmerand Green 1990), the possibility of patenting cumula-tive and combinative innovations (Erkal 2005, Ottozand Cugno 2008), the reliability of patent litigation(Choi 1998, Aoki and Hu 1999), imperfect or proba-bilistic patent protection (Waterson 1990; Anton andYao 2004; Kultti et al. 2006, 2007), the possibility ofpatent renewal (Langinier 2004), disclosure require-ments pertaining to patents (Bessen 2005, Aoki andSpiegel 2009), and the quality of the focal invention(Horstmann et al. 1985).
As the factors just listed make clear, extant workin the economics literature has (not surprisingly)addressed the effects of a patenting policy on the
economy as a whole. In shifting this focus from theeconomy to the firm we must lower the level ofmodel abstraction in regard to at least three aspects.First, a firm-level patenting theory needs to incorpo-rate more than the outcome measure of overall firmprofit. Although the profit motive ultimately drivesfirm action, the firm may pursue different patent-ing strategies for a multitude of intermediate goals.A more realistic study of patenting should discuss apatenting strategy’s effect on each of those intermedi-ate goals.
Second, a firm-level patenting theory must builda realistic representation of how patenting decisionsare made. Traditional economics models representfirm interactions using dynamic games of incompleteinformation and hence implicitly assume that firmsadhere to a very specific and intricate ideal of ratio-nality1 that no real firm can potentially live up to.In reality, firms are boundedly rational: they adoptbehavioral strategies and modify those strategies ifthey prove not to be useful (Simon 1969). There-fore, a firm-level model should ground its analyses inthe main classes of documented behavioral patentingstrategies. Moreover, such strategies should allow forrepeated decision making about patenting; two-stagegames simply do not allow one to represent the com-plexities that actual decision makers face.
Third, a firm-level patenting theory needs to cap-ture the main contingency factors. In particular, itmust adequately represent the firm’s R&D process.R&D is fundamentally a search process (Fleming andSorenson 2001). There is uncertainty ex ante about theoutcomes of the search, yet the exploration gives riseto learning in which past searches yield insights thatcan be used in future searches. It seems likely thatthe nature of this search process influences patentingstrategies.
In this section, we describe the three dimensionsthat any firm-level contingency model should con-sider. First, we concentrate on patenting motives andthus on the goals to which a firm may aspire. Second,we identify the different patenting decision options.Third, we focus on contingency factors that determinethe usefulness of those decision options—with respectto patenting motives—under different environmentaland firm circumstances.
Patenting Motives. Why do firms patent? Empiri-cal studies have carefully examined the motives forpatenting—that is, what drives innovating firms toapply for and hold patents (Arundel et al. 1995,
1 The firm must analyze the competitor’s past actions for each indi-vidual patent, after which the firm projects all of its own and itscompetitor’s actions into the (potentially infinite) future. Based onsuch projections, the firm makes a globally optimal dynamic deci-sion; it repeats this process for each of its own and each of itscompetitor’s actions.
Cohen et al. 2000, Macdonald 2004, Sheehan et al.2004, Blind et al. 2006). These studies provide strongevidence that firms do not patent simply to protectagainst imitation. Instead they pursue, often simul-taneously, a number of strategic goals when apply-ing for a patent (Levin et al. 1987, Cohen et al. 2000,Blind et al. 2006). Although the goals of patentingmay appear fairly nuanced in practice, Blind et al.(2006) used factor analysis to condense the multiplic-ity of considerations into five basic motives: protec-tion, blockade, exchange, reputation, and incentive.
Beyond the traditional protection motive, whichremains the principal motive for patenting, firms seekto blockade competitors by anticipating and preempt-ing their use of future technologies (Ceccagnoli 2009).Such a blockade can serve defensive or offensivepurposes: defensive blocking prevents competitorsfrom venturing close to the firm’s own technolog-ical base (Reitzig 2004); offensive blocking aims toimpair competitors’ future R&D trajectories regard-less of their direction (Granstrand 1999, p. 215; seealso Blind et al. 2006). The exchange motive pre-sumes that the firm files for patents in order toimprove its bargaining position in patent licensingdeals (Crama et al. 2008, Aggarwal and Hsu 2009) aswell as in patent exchanges and in the cross-licensingof patents (Grindley and Teece 1997). Cohen et al.(2000) and Kash and Kingston (2001) find that theexchange motive is especially important in indus-tries where inventions by competitors build on eachother. In the electronics industry, for example, inven-tions are a composite of many individual detail solu-tions and so favor cooperation for mutual benefit.Both the reputation and incentive motives—which areusually considered to be less influential than themotives of protection, blockade, and exchange (Blindet al. 2006)—are linked to the standing a firm haswith respect to key stakeholders. The reputationmotive involves external stakeholders, such as poten-tial investors (Hsu and Ziedonis 2008); the incen-tive motive is an internal tool for motivating staff(Oldham and Cummings 1996).
It is important to realize that the three preemi-nent motives (protection, blocking, and exchange) areintermediate goals. Ultimately, they serve the profitmotive by building on a common characteristic ofall patents: establishing a claim in the technologylandscape that the patent owner can prevent oth-ers from exploiting. This claim can be used to pro-tect the firm’s position, to prevent another firm fromentering a technology field, or to establish bargainingpower in an exchange situation. (It is worth remark-ing that secrecy, the alternative to patenting, can atbest achieve only the first of these major patentingmotives.) Hence the patent claim can ultimately be
Table 1 Patenting Strategies on Technology Landscapes
Patent strategy
R&D strategy Everything High quality Close competitor Nothing
Leader Blanketing/flooding
Granstrand (1999),Cohen et al. (2000),Blind et al. (2006, 2009)
Searching for strategic patents
Levin et al. (1987),Granstrand (1999),Blind et al. (2006, 2009)
Fencing
Lerner (1994), Granstrand(1999), Cohen et al. (2000),Arundel and Patel (2003), Blindet al. (2006, 2009)
Nondisclosing leader
Levin et al. (1987),Teece (1986),Cohen et al. (2000),Arundel (2001)
Follower Surrounding
Granstrand (1999), Cohenet al. (2000), Reitzig et al.(2007)
Ad hoc blocking and inventing around
Granstrand (1999),Arundel and Patel (2003)
Surrounding (see text)
Granstrand (1999),Cohen et al. (2000)
Nondisclosing follower
Teece (1986)
used to affect the competitiveness of the firm’s prod-uct offering and thus to create economic value. Forboth protection and blocking, that economic valuemanifests as a relative performance advantage of afinal product as compared with a competitor’s prod-uct. Protection gives a focal firm the right to imple-ment its preferred technological solution—a right thatit denies the opposing firm. Blockade gives a rela-tive performance advantage to the focal firm by pre-venting the competitor from implementing its ownpreferred solution. The exchange motive implies thatboth companies can break a deadlocked situation and,through collaboration, improve the absolute perfor-mance of products offered in the market. Our modelshould therefore account for the roles of protection,blocking, and exchange by examining the perfor-mance consequences of staking out claims in a tech-nology landscape.
Patenting Decisions. Motives are not themselves suf-ficient to establish a patenting strategy. It is onlywhen the firm adopts a consistent decision patternwith respect to patenting that we can speak of itspatenting strategy. The normative literature on patent-ing decisions is sparse, with the Granstrand (1999)typology the rare exception. This book reports thatfirms explore a technology landscape through theirR&D processes and then may use patenting to estab-lish and protect claims in this landscape for theirexclusive economic use. Based on various case stud-ies across nations and industries, Granstrand (1999,pp. 219–222) identifies six so-called generic patentstrategies: ad hoc blocking and inventing around,strategic patent searching, blanketing (or flooding),fencing, surrounding, and combination.
Granstrand’s work is a solid starting point for anyeffort to identify patenting strategies. It is interest-ing that, upon closer examination, his mapped strate-gies turn out to be a combination of two separatefirm decisions. First, the firm’s movement (or trajec-tory) on the technology landscape implies a delib-erate decision about the direction of R&D; second,the choice of whether or not to patent a specific
technological invention presupposes a patenting rule(or, more broadly, a patenting strategy). Consistentlywith Granstrand, we recognize that patenting deci-sions cannot be analyzed without accounting for thefirm’s overarching R&D strategy (see also Arora andCeccagnoli 2006, Arora et al. 2008). In contrast withGranstrand (1999), we make a deliberate distinctionbetween the two; thus we are able to systematize andcomplement current thinking on these strategies.
Table 1 proposes a typological framework for clas-sifying a firm’s patenting strategy along those twocrucial dimensions, and it classifies the extant liter-ature accordingly. The first dimension is the patent-ing rule that dictates when an invention should bepatented. A firm that patents everything seeks patentprotection for every solution generated by its R&Dactivities. A firm that applies the high-quality patent-ing rule files patents only for those solutions thatoffer superior performance. When patenting underthe close-competitor rule, a firm tries to anticipate itscompetitors’ future movements and then to patentonly those inventions that are in those firms’ technol-ogy terrains; this rule is most closely related to theblocking motive. The patent nothing rule reflects thenotion that secrecy will prevent others from exploitingthe focal firm’s inventions (Levin et al. 1987, Cohenet al. 2000, Arundel 2001); in other words, this strat-egy seeks to achieve protection by a means other thanpatenting. The second dimension of our framework isthe firm’s R&D strategy. Research and developmentefforts provide the technological solutions eligible forpatenting. We follow Teece (1986) and distinguishbetween leaders, who autonomously explore the tech-nology landscape, and followers, whose R&D activi-ties are geared to imitate and thus chase competitors’technology trajectories by “inventing around” them.2
2 We assume that followers have enough absorptive capacity tounderstand the leader’s technologies (from patent applications) andcan carry out enough R&D themselves to invent around any leaderpatents: we thus consider able followers. This requirement is relaxedin our robustness analyses (see §4.6).
The combination of what we designate a patent-ing rule and an R&D strategy forms what Granstrandwould call a patenting strategy, and our classifica-tion subsumes his typology. For instance, the “patenteverything” rule followed by an innovation leaderthat files patents for every solution developed on itspath through the technology landscape can be iden-tified as a blanketing/flooding strategy; yet, the patenteverything decision rule when used by an innovationfollower will take the form of a surrounding strategy.To avoid confusion, hereafter we shall use the term“patenting strategy” in the more narrow sense of apatenting rule.
Contingency Factors. The two-dimensional classifi-cation proposed in Table 1 demonstrates that patent-ing decisions take effect contingent on a firm’s R&Dstrategy. The same patenting strategy will lead to dif-ferent outcomes when pursued in combination with adifferent R&D strategy. Hence we view a firm’s R&Dstrategy as an important internal contingency factor forthe effectiveness of any patenting strategy. A firm’sR&D strategy is long term and difficult to change; itspatenting strategy is more malleable. (Of course, R&Dstrategy is itself a decision variable in the long run;the choice of that strategy is discussed in §5.)
Environmental contingency factors may also moder-ate how various patenting strategies affect firm goals.Although many environmental factors affect the suc-cess of a patenting strategy, two are salient in the lit-erature. First, extensive empirical evidence suggeststhat a focal firm’s patenting behavior should notbe analyzed in isolation from the behavior of otherfirms. This is because observed patenting behavioractually follows from the strategic interactions inso-called patent (portfolio) races (Hall and Ziedonis2001, Ziedonis 2004, Hall 2005). If a firm suddenlybegins to flood entire technological areas with patents,then other firms in the industry adapt by intensify-ing their own patenting activities (Jell and Henkel2010). In this case, more patenting simply reduces therisk of being blocked or being sued for infringement(Ziedonis 2004). Thus, competitor behavior is criticalto a contingency model of patenting. Since a com-petitor’s patenting likewise takes effect contingent onits R&D strategy, modeling strategic interaction nec-essarily requires incorporating both the competitor’sR&D strategy and its patenting strategy as environ-mental contingencies for the focal firm. A secondfactor is industry characteristics—in particular, tech-nological complexity (Cohen et al. 2000) and industryclock speed (Nadkarni and Narayanan 2007). Techno-logical complexity affects how easily innovation canbe appropriated in focal markets (Hall and Ziedonis2001, Ziedonis 2004, Ceccagnoli 2009), and indus-try clock speed affects how much time each com-pany has to perform its technology search (Chao and
Kavadias 2008) and patenting activities (Nadkarniand Narayanan 2007). Any model of patenting shouldaccommodate both of these industry characteristics.
A model fulfilling these requirements would fill animportant gap in the literature. It need not accom-modate all facets of patenting activity (unlike manyof the empirical and conceptual works cited in thissection), and neither must it focus on the overall eco-nomic impact of patent races (unlike models that areprimarily economic). Instead, this model aims to cre-ate a framework for making decisions about strate-gic patenting under competition at the firm level. Themodel abstracts sufficiently from individual details tofocus on overall firm patenting strategies, but it is finegrained enough to address questions about how firmsshould choose their patenting strategies.
3. ModelIn §2, we identified critical factors that a contingencymodel for patenting must consider: patenting motivesdriving competitive advantage and ultimately firmprofits as the goal of all patenting activities, patent-ing strategies as the templates for how firms decide onwhat they ultimately protect, and internal contingen-cies (i.e., the firm’s own R&D strategy) and environmen-tal contingencies (i.e., the competitors’ patenting andR&D strategies, industry clock speed and complexity)as mediators between those patenting strategies andthe firm’s achievement of its patenting goals. Thusthe model’s predictions must reflect the competitiveadvantage fostered by patenting motives, the deci-sion variables used in the deployment of patentingstrategies, and the parameters needed to reproducethe most salient internal and environmental contin-gency factors.
A central aspect of our model that helps to formal-ize these requirements is the notion of a general tech-nology space or “technology landscape” (Granstrand1999). The technology landscape links agent decisionsto outcomes: it models the performance outcome ofR&D decisions on a fitness landscape by assigninga fitness value to each of all conceivable technologyconfigurations. The resulting map consists of “coordi-nates” representing particular product solutions and“altitude curves” assessing the overall technologicalfitness of those solutions; this map of the landscapethus establishes a basis for the economic value of pro-tection, since a higher technological fitness impliesa higher economic value of the resulting products.Hence the fitness dimension represents the competi-tive situation that patenting motives seek to affect: theeconomic value embodied in technological fitness canbe protected for exclusive exploitation, competitorscan be blocked from deriving that value, and it canbe used as part of an exchange. Different companies
might search the technology landscape in differentways, thereby implementing different R&D strategies.Contingent on their R&D strategies, firms implementdifferent patenting strategies by reserving (or not) cer-tain locations on the technology landscape for theirexclusive use. The various patent strategies can beused to characterize the focal firm’s behavior and thusserve as our main decision variables; because theycharacterize competitor behavior, too, these strate-gies function also as environmental contingency fac-tors. Finally, the landscape’s structural properties canbe adjusted and so represent additional contingenciespertaining to industry characteristics. Figure 1 pro-vides an overview of our model’s elements and theirinteractions.
Our description of the model begins in §3.1 byexplaining how we represent a problem’s technologysolution as a technology “configuration” and how thetechnology landscape links technology configurationsto fitness outcomes. We continue by assuming the per-spective of a focal firm to represent the strategic sit-uation faced by each firm in our model. In §3.2, weexplain a firm’s implementation of an R&D strategyon a technology landscape; in §3.3, we describe ourrepresentation of its patenting strategy. In this way,the firm’s patenting decisions are linked to its internalcontingency factor (i.e., the firm’s R&D strategy). Wethen show, in §3.4, how an innovation race betweentwo firms plays out, thus connecting the firm’s patent-ing behavior with the most important environmentalcontingency: the competitor’s behavior as shaped by
its own R&D and patenting strategy. After thus imple-menting the main decision variables as well as themain contingency (i.e., competitor behavior), §3.5 isdevoted to the other environmental contingencies ofinterest—namely, industry characteristics. In §3.6, wediscuss some noteworthy implications of the model.We shall employ the following notational conventions:�S� denotes the cardinality of set S (i.e., the numberof its elements); and �s� denotes the Euclidean normof the vector s (i.e., its length).
3.1. The Technology Landscape: Linking theTechnology Decision to Outcomes
We adopt a standard NK approach (Kauffman 1993,Levinthal 1997) when modeling the technology land-scape. Defining a position on this landscape requiresthat N different configuration decisions sl be made.These can be viewed as constituting the set of allpossible design decisions that go into a technolog-ical product. Thus a position—that is, one technol-ogy configuration—is defined by the decision vectors = 4s11 0 0 0 1 sl1 0 0 0 1 sN 5. For simplicity, the NK modelrequires that the individual choices sl be binary (either0 or 1), so there are 2N possible technology configu-rations that an R&D activity can yield. The solutionfitness function V 4s5 maps the N -dimensional tech-nology configuration s into a one-dimensional perfor-mance measure, thereby generating an altitude profileover the N -dimensional technology landscape. Theperformance so measured captures the resulting prod-uct’s competitiveness in the market place.
One benefit of using the NK model is that wecan then structurally alter the technology landscape’scomplexity as follows. Each decision contributes tothe overall fitness. Let Vl denote the contribution ofdecision l to fitness. Then the overall fitness is simplythe average of all the decisions’ contributions:
V 4s5=
∑Nl=1 Vl4sl1 s−l5
N0
Landscape complexity arises because Vl depends notonly on the value of sl but also on the value of K otherdecisions (Levinthal 1997, Levinthal and Warglien1999); we denote these decisions as s−l = 4sl11 0 0 0 1 slK5.For each possible combination of sl and s−l, a randomdraw from a uniform distribution U60117 is assignedto Vl4sl1 s−l5. When K = N − 1, decisions are highlyinterdependent: each Vl is affected by all N − 1 otherdecisions and so the technology landscape is uncor-related, complex, and “rugged.” If one of the N deci-sions changes, then all Vl assume the value of anotherrandom draw, which means that the entire fitnessfunction V takes on the value of a new random draw.As a result, nearly identical technological solutionsmay result in drastically different technological per-formances (Baumann and Siggelkow 2013, Ethiraj andPosen 2013). At the other extreme, if K = 0 then alldecisions are independent because Vl depends exclu-sively on sl; hence the technology landscape is corre-lated, not complex, and “smooth.” In this case, if oneof the N decisions changes then only the correspond-ing Vl undergoes a random change, which means thatV is only marginally affected. Consequently, a slightproduct modification will have a comparably slighteffect on performance.
3.2. R&D StrategiesThe technology landscape links each technology con-figuration to an outcome. We can therefore modelthe R&D search process as a search on the technol-ogy landscape (Mihm et al. 2003, Sting et al. 2011).In the standard NK model, a firm searches the land-scape in isolation and its search heuristics are moti-vated by various behavioral factors (Levinthal 1997,Knudsen and Levinthal 2007). Deviating from stan-dard instances of the NK approach, we model twofirms that explore the technology landscape whileinteracting with each other. For each firm i1 j = 112,i 6= j , the search process starts from a random initialproduct solution s405i and then traverses the landscapevia the search trajectory 4s4t5i 5t∈� = 4s405i 1 s415i 1 s425i 1 0 0 05.
To illustrate the model dynamics and firm interac-tions, we consider an arbitrary period t in which bothfirms have already undertaken some R&D activities.Each firm has investigated the technological field to acertain extent and has thus built a knowledge base ofthat field; furthermore, each firm has protected some
of this knowledge by using patents. In our model, afirm that has built knowledge of a field correspondsto a firm that has explored some technology con-figurations s4�5i and learned about their performanceV 4s4�5i 5 for all past periods � ∈ 811 0 0 0 1 t9. The completestock of knowledge that the firm has assembled priorto the current period t is retained in the solution listS4t5
i , which stores all technology configurations thatthe company has explored and their respective per-formance; a second list, the patent list P4t5
i , containsall solutions for which the firm has filed a patent by tand their respective performance. Clearly, P4t5
i ⊆ S4t5i .
Although S4t5i is private information to firm i—since
we assume that, in general, firms do not reveal theirR&D activities to their competitors—the competingfirm j can observe the subset of patented (and thusdisclosed) solutions P4t5
i .In §2 we established the R&D leader and the R&D
follower as exponents of two archetypical R&D strate-gies. These two strategies are represented by differ-ent heuristics in our model. Consider the beginning ofthe period t + 1, when the firm needs to decide aboutits future R&D direction. It must choose an area ofresearch—that is, a starting point for its R&D efforts.A leader firm chooses its starting technology configu-ration from its solution list S4t5
i , relying exclusively onits own technology. In contrast, a follower focuses oncatching up with the leader; hence a follower choosesits starting technology configuration from its com-petitor’s patent list P4t5
j 0 Neither the leader nor thefollower necessarily choose the best-performing tech-nology available to them as their starting point.Instead they pick a starting configuration randomly,with the selection probabilities proportional to theassessed quality of each solution. Building on the basetechnology chosen, the firm carries out an incrementalresearch step. That is, the firm experiments with onerandomly chosen decision by reversing its state from0 to 1 (or vice versa). This R&D step generates the nextsolution s4t+15
i and its performance V 4s4t+15i 5, updating
the solution list from S4t5i to S4t+15
i .3
3.3. Patenting StrategiesOnce firm i has carried out its R&D efforts with results4t+15i , it must decide whether or not to patent this
technological solution. As described in §2, firms mayapply different patenting strategies.
Under the patent everything rule, the firm files apatent for any solution that is not covered by a pre-vious patent filed by either the firm or its competi-tor; formally, it files a patent when s4t+15
i y 8P4t5i ∪P4t5
j 9.
3 We assume that the follower has full absorptive capacity (Cohenand Levinthal 1990, Lenox and King 2004)—in other words, thatit can readily assess the leader’s patents and easily integrate theminto its own R&D program. This simplifying assumption (whichwe relax in §4.6) renders the model more parsimonious.
Table 2 Patent Strategies: Base Terrains for New R&D Activities and Patent Criteria for Firm i at Time t + 1
Patent strategy
R&D strategy Everything High quality Close competitor Nothing
Leader R&D base: S4t5
i
Patent new solution s4t+15i if
s4t+15i y 8P4t5
i ∪P4t5
j 9
R&D base: S4t5
i
Patent new solution s4t+15i if not
patented andV 4s4t+15
i 5 > max8V 4s5 � s ∈P4t5
i 9
R&D base: S4t5
i
Patent new solution s4t+15i if not
patented andmin8�st+1
i − s� � s ∈P4t5
j 9= 1
R&D base: S4t5
i
Patent nothing
Follower R&D base: P4t5
j
Patent new solution s4t+15i if
s4t+15i y 8P4t5
i ∪P4t5
j 9
R&D base: P4t5
j
Patent new solution s4t+15i if not
patented andV 4s4t+15
i 5 > max8V 4s5 � s ∈P4t5
i 9
R&D base: P4t5
j
Patent new solution s4t+15i if not
patented andmin8�st+1
i − s� � s ∈P4t5
j 9= 1
R&D base: P4t5
j
Patent nothing
Under the “patent high-quality” rule, the firm filesa patent only if the solution is unpatented and ofhigher quality than any previously filed patents bythe firm—formally, only if V 4s4t+15
i 5 > max8V 4s5 � s ∈
P4t5i 9. Under the “close-competitor” patent rule, the
firm files a patent only if the solution is unpatentedand could easily be copied by the firm’s competi-tor. A solution is considered to be “easily copyable”if it is adjacent to one of the competitor’s patents,by which we mean that the latter can be derivedfrom the former via a single mutation. Formally, thefocal firm’s patent is adjacent to a competitor’s patentif min8�s4t+15
i − s� � s ∈ P4t5j 9 = 1. Finally, under the
“patent nothing” rule, the firm applies for no patents.If the firm chooses an active patenting strategy (oneof the first three just described) and if the firm decidesto apply for a patent, then the solution s4t+15
i will beadded to P4t5
i and will yield the updated patent listP4t+15
i . Table 2 summarizes all the combinations ofR&D types and patenting rules.
In addition to deciding whether or not to patenta specific invention, the firm must account for howmuch space its patent can claim in the technologylandscape. In many industries, companies tend to filefor basic patents—especially in industries with “com-pound” or “complex” products (Cohen et al. 2000).Basic patents protect not only a single solution butalso the technological core of that solution in adjacentapplications. For instance, nearly all semiconductormanufacturers had to honor the patent held by TexasInstruments for the CMOS transistor (Weber 1990).However, other companies can file their own patentsas refinements of the basic patents. Virtually allsemiconductor manufacturers patented some detailaspects of the CMOS transistor (or its production).The result was an “interlocking” scenario wherebymanufacturers were able to block each other fromchoosing an optimal transistor design. Even TexasInstruments could not manufacture its own semicon-ductors without using the inventions protected bycompetitors’ patents. In practice, an interlocking sce-nario is an economic incentive for patent exchange.
In the context of our model, basic patents protectnot just one point but instead an entire area in thetechnology landscape. Let parameter B ≥ 0 denote apatent’s breadth, where B <N . The technological coreof a patent is defined by N − B technology deci-sions, which are fixed. Thus, basic patents protect theneighborhood around a specific solution by cover-ing B undefined decisions. Whereas a specific patentgrants the firm the right to exploit a particular solu-tion, a basic patent establishes the firm’s stake ina wider technology space. For example, let N = 5and B = 2, and let the specific solution be the con-figuration s4t+15
i = 41 1 1 0 15. Suppose the technologi-cal base of this specific patent consists of the firstthree decisions; so by filing a basic patent, the firmprotects the technological neighborhood of 41 1 1 · ·5= 841 1 1 0 051 41 1 1 0 151 41 1 1 1 051 41 1 1 1 159. Providedthat B > 0, the corresponding neighborhood of s4t+15
i
will also be included in the updated patent list P4t+15i
as basic patents. Note, however, that both competingfirms will associate the patented neighborhood solu-tions with V 4s4t+15
i 5—that is, with the evaluation forthe specific solution. In line with observations indicat-ing that firms aim to establish patent stakes as broadlyas possible, we assume that the company prefers tofile basic patents with B = 2 (if existing patents donot preclude the firm’s doing so); we choose theN − B basic decisions randomly. If the firm is pre-vented from filing a basic patent by conflicting extantpatents, then the firm narrows the breadth of the focalpatent: first to B = 1 and then, if necessary, to B = 0.A basic patent protects the firm’s stake in ramifiedtechnologies, but it does not grant the firm an exclu-sive right to exploit them. In other words, the com-peting firm may file a specific patent even if thereis already a basic patent that protects the underly-ing area (provided, of course, that there is no spe-cific patent already). This scenario corresponds to thecase of specific patents establishing details of a solu-tion as they build on the core technology of a broaderbasic patent. In our model, as in the real-life exam-ple, either firm can then prevent the other from using
the invention because the basic patent and the spe-cific patent interlock; this standoff sets the stage for apatent exchange.
3.4. Model Dynamics and Model OutcomesHaving delineated the model’s individual elements,we can now describe overall dynamics and ultimatepayoffs. Each firm starts from a random point in thelandscape and then, iteratively, (a) carries out researchby mutating one decision in the N -dimensional deci-sion vector and assessing its performance; (b) decides,based on its patenting strategy, whether or not topatent and then, if necessary, updates its patent list;and (c) identifies, based on its R&D strategy, the start-ing point for carrying out research in the next round.
Each firm searches the landscape by repeatedlygoing through this cycle until the time ends inperiod T . The model parameter T can be viewedas the market clock speed, which is an exogenouslygiven deadline for presenting the product solution.A small T implies substantial time pressure for R&Defforts, whereas a large T corresponds to businesssituations with long market cycles that allow for in-depth R&D exploration (Csaszar and Siggelkow 2010,Mihm et al. 2010).
By the end of period T , firm i receives the payoffassociated with its highest-quality solution s∗
i that is notalready patented by the competitor. In other words,the firm launches the product with the highest perfor-mance that is not protected by the competition. Henceit does not matter whether firm i holds a patent fors∗i ; what matters is whether firm j has already patent-
protected the technology solution s∗i . Formally, firm i
achieves the absolute performance
çi = max8V 4s5 � s ∈S4T+15i \P4T+15
j 91 i = 1121 j 6= i0
However, the firm’s market success may depend lesson this absolute performance than on its relativeperformance—that is, as compared with the com-petitor’s solution. Some strategy profiles yield goodabsolute performance but poor relative performance.We view relative performance as being well proxiedby (relative) market share; we view absolute perfor-mance, which indicates a solution’s overall attractive-ness, as being well proxied by market size. In the restof this paper we report both the absolute performanceand the relative performance achieved by each firm i,where relative performance is formally defined as
çri =çi −çj1 i = 1121 j 6= i0
Previously, we described how, for a given productsolution, one firm’s specific patent may overlap with abasic patent held by another (competing) firm. In suchcases, a product cannot be brought to market with-out the competitor’s consent. The result is a deadlock
in which neither party can exploit the patented prod-uct solution. To examine such situations, we augmentour simulation results by incorporating the potentialvalue of patent exchange. This value is defined asa firm’s expected performance increment that wouldresult from breaking the patent deadlock via mutuallyshared patents. In our simulations, no agent choosesto participate in a patent exchange; we simply indi-cate the value of patent exchange once game out-comes have been realized.
Finally, we consider patenting cost to be linearin the number of patents filed. For c the averagedirect and indirect cost per patent, the accumulatedpatenting cost could be given as Ci = c�P4T+15
i � (i =
112, j 6= i). However, we intentionally refrain fromincluding an explicit cost factor. For some compa-nies, small improvements in competitiveness (relativeperformance) outweigh any patenting cost considera-tions; yet for other companies, even considerable per-formance advantages are quickly outweighed by thecost of acquiring and maintaining patents. We there-fore report the raw number of patents �P4T+15
i �.
3.5. Industry CharacteristicsWe have elaborated on the main internal and envi-ronmental contingencies by embedding both the focalfirm’s R&D strategy as well as the competitor’s R&Dand patenting strategies into the model’s dynamicstructure. Two critical parameters allow us to addressthe contingencies involving industry characteristics.First, we can capture industry clock speed by system-atically varying the time T allotted to a simulationrun. Second, by changing the correlation parameter Kin the NK representation of the technology landscape,we can account for search environments that are rel-atively more or less complex.
3.6. Implications of the ModelIt is worth making explicit one aspect of the modelthat has so far been implicit. What is the rela-tion between our four performance metrics above—absolute and relative performance, exchange potentialand cost—and (what most models assume is) the ulte-rior goal of firm action: financial payoff? Why do werefrain from modeling payoff directly?
From a conceptual perspective, our model (i) ac-counts for the possibility that the firm’s patentingstrategy affects its technology position (and, indi-rectly, the technology position of its competitor)and (ii) views the firm’s technology position (andthat of its competitor’s) as key drivers of financialresults. Obviously, financial payoffs are also affectedby other factors—including firm factors, such as itscommercialization capability (Arora and Ceccagnoli2006), and environmental factors such as demanduncertainty. Some of these other factors (e.g., com-mercialization capabilities) moderate the technology
position’s effect on financial payoffs. One approachto representing this notion formally would be todevise a function that maps all important technol-ogy position parameters and all the “other” factors toa financial payoff. However, such a mapping wouldrequire devising a specific function for how these fac-tors influence payoffs, and we would thus be boundto introduce potentially limiting assumptions. Theless constraining approach chosen in this paper is torefrain from defining a function explicitly; rather, weenumerate the most important aspects of a technologyposition that determine profit—as identified in theempirical literature on patenting motives—and ana-lyze how patenting strategies influence those aspects.This way, readers can draw their own conclusionsabout financial payoffs. The approach developed hereis both parsimonious and yet generalizable.
A second aspect of the model is worth noting. Themodel’s time horizon is T , where T corresponds tothe launch of the product that commercializes a par-ticular technology. Hence, and in line with existingliterature on patents (e.g., Horstmann et al. 1985), ourmodel focuses on the innovation race. It is possiblethat, upon product launch, the product itself causesknowledge spillovers to competitors. When these arelarge, relative differences in performance at the end ofthe innovation race are worth less in financial terms.When spillovers are small, relative differences in per-formance yield a higher payoff.
4. AnalysisIn this section we discuss the effectiveness of patentstrategies. The analysis is organized in terms of ourmain contingency: competitive behavior determinedby the combinations of opposing R&D types. Thuswe report (in §§4.1–4.4, respectively) the results for(1) how a leader should shape its patenting strategywhen competing with another leader, (2) how a leadershould compete against a follower, (3) how a followershould compete against a leader, and (4) how a fol-lower should compete against another follower.
We set the end time T to 100 rounds; we set N = 12and K = 6 so that the reported performance mea-sures reflect a moderately complex environment. (Theeffects of varying these parameters are discussed in§4.6.) Given that simulation-based results are subjectto random fluctuations, we report averages of 10,000runs for each game setting. In each run, the play-ers compete on a new random landscape constructedusing identical parameters. All results mentioned inthe text are significant at the 5% level.
4.1. Best Strategies for an R&D LeaderCompeting with an R&D Leader
First we consider the case of a leader competingagainst a leader. Figure 2 plots all the outcome mea-sures described previously. The figure’s upper left
quadrant shows the relative product performance(i.e., how well the focal firm fares when comparedwith its competitor) at the end of the innovation race:the more positive the relative performance, the morecompetitive the focal firm’s product; the more nega-tive, the less competitive. The upper right quadrantshows the number of patents held by the focal firm atthe end of the innovation race—recall that this num-ber is used as a proxy for patenting cost. The figure’slower left quadrant shows the focal firm’s absoluteproduct performance, which is a proxy for the prod-uct offering’s attractiveness to consumers and thusfor market size. The lower right quadrant illustratesthe product performance gains that the firm couldachieve via patent exchange. The higher the poten-tial gains from patent exchange (for both firms), thegreater the likelihood that such an exchange will takeplace.
In each quadrant, the horizontal axis depicts thefocal firm’s possible strategies and the vertical axismarks out the various performance indicators associ-ated with the (same four) possible competitors’ strate-gies. We identify these competitor strategies by usingdots of different color. Thus the 16 dots in each quad-rant fully characterize each combination of strategiesplayed by the focal firm and its competitor.
We start by concentrating on the relative perfor-mance of the final product. We assume that the firmbegins its analysis of patenting options by estab-lishing whether or not it is capable of beating thecompetition.
The firm tailors its strategy so that it can reactoptimally to competitor actions; in other words, thefirm strategy is designed to be optimal given that thecompetitor chooses a certain strategy. The firm devel-ops such an optimal plan for each possible competi-tor strategy. Then, while assuming that its competitoralso behaves strategically, the firm tries to second-guess its competitor’s actions. We juxtapose the com-petitor’s optimal plan for every given competitivesituation in turn. In other words, the firm devisesbest responses for every contingency (i.e., for eachcompetitor patenting strategy) that maximize its ownposition—relative to the competitor’s. A competitorthat behaves likewise will determine its own strategywhile assuming that the focal firm is also employ-ing a best-response strategy. Neither player has anincentive to deviate from the equilibrium that arisesfrom this dynamic. In fact, we can formally define asimultaneous move game over the behavioral strate-gies identified in previous sections; the action set con-sists of the four archetypical strategies identified bythe empirical literature. This approach allows us toidentify the Nash equilibria of the game. We remarkthat the firms need not be aware of the game in order
Figure 2 Results for a Leader Competing with a Leader
Every-thing
Highquality
Closecompetitor
Nothing
0.30
Rel
ativ
e pe
rfor
man
ce o
f th
e fo
cal
firm
vs.
the
com
petit
or (
%)
0.20
0.10
–0.10
–0.20
–0.30
0.00
Every-thing
Highquality
Closecompetitor
Nothing0.68
0.69
0.70
0.71
Abs
olut
e pe
rfor
man
ce o
f th
e fo
cal f
irm
Every-thing
Highquality
Closecompetitor
Nothing
0.50
0.40
0.30
0.20
0.10
0.00Impr
ovem
ent p
oten
tial t
hrou
ghpa
tent
exc
hang
e (%
)
Every-thing
Highquality
Closecompetitor
Nothing
EverythingHigh qualityClose competitorNothing
0
50
100
150
200
250
300
Competitor strategyto patent
Num
ber
of p
aten
ts f
or th
e fo
cal f
irm
for it to play out as described. If each competitor con-tinues to adjust its patenting strategy in reaction to theother competitor’s move—continually trying to finda better strategy—then this evolutionary process mayreach the same state of equilibrium (see, e.g., Nelsonand Winter 1982).4
Our analysis begins with a discussion of best re-sponses before focusing on the outcomes of vari-ous equilibria. Observe that all the figures representeach competitor patenting strategy using identicallyshaded dots that are linked by a uniquely depicted(dots, dashes, etc.) line. The firm identifies its bestresponse to a competitor’s strategy by choosing as itsown strategy (on each plot’s x-axis) the one with thehighest y-axis value on that line. For instance, if thecompetitor patents nothing (line of black dots) thenthe focal firm’s best response is to patent everything—in other words, to choose the highest (here, the left-most) black dot.
Figure 2 supports several notable conclusions. Inresponse to a competitor that patents nothing (lineof black dots), pursuing that same strategy (right-most black dot) yields poorer performance for thefocal firm than does any of the other, active patentingstrategies (leftmost three black dots). Moreover, given
4 See §5 for an in-depth discussion of our equilibrium concept.
any competitor strategy (any line of dots), the rela-tive performance of patenting nothing (right-columndots) is either inferior to or not significantly differ-ent from any of the other three patenting strategies(corresponding dots in the other columns with iden-tical shading). For example, a strategy of high-qualitypatenting clearly dominates the strategy of filing nopatents. In short, to patent nothing is an inferior strat-egy for a leader competing with a leader. The effect isnot large but it is relevant.
Among the three active patenting strategies, nosingle one dominates. Although patent everythingis the firm’s best response when its competitorpatents nothing, high-quality patenting should beemployed in response to close-competitor patentingand close-competitor patenting is the best response toa competitor that patents everything. So from the rel-ative performance viewpoint, a leader facing a leadershould definitely patent; however, none of the activepatenting strategies is clearly preferable.
In fact, for leader-leader competition, neither abso-lute product performance nor the potential gains froma patent exchange is enough to determine the bestactive patenting strategy. The lower left quadrant ofFigure 2 clearly shows that, with respect to absoluteperformance, no active patenting strategy dominatesany other. The same is true when we analyze patentexchange (lower right quadrant): no active patenting
strategy dominates and, overall, the potential gainsfrom a patent exchange would be small. So as regardsrelative and absolute performance and also patentexchange, all active patenting strategies are consistentwith the goal of achieving good product performancefor leaders that are competing against leaders.
Yet with regard to the number of patents held bythe focal firm at the end of the technology race, thestrategies exhibit substantially different results (upperright quadrant). In particular, the strategy of patent-ing everything naturally results in many more patentsthan do the other two active strategies. So if the cost offiling for—and holding—patents is substantial, thenthe strategies of high-quality and close-competitorpatenting outperform the patent everything strategy,which rational players should therefore avoid. How-ever, if the cost for patenting is negligible and if ahigh number of patents held yields secondary bene-fits, then it is actually preferable to patent everything.
In sum, for an R&D leader engaged in a racewith another technology leader, active patenting is theoptimal strategy. The classical protection motive ofpatenting outweighs any secrecy considerations. Sinceour competitive setup is symmetric, such a patentingstrategy also forms the industry equilibrium. (In equi-librium, the firms’ best responses reciprocally agree;
Figure 3 Results for a Leader Competing with a Follower
Rel
ativ
e pe
rfor
man
ce o
f th
e fo
cal
firm
vs.
the
com
petit
or (
%)
8.00
6.00
4.00
2.00
–2.00
–4.00
0.00
Every-thing
Highquality
Closecompetitor
Nothing
Every-thing
Highquality
Closecompetitor
Nothing0.67
0.68
0.69
0.70
0.71
Abs
olut
e pe
rfor
man
ce o
f th
e fo
cal f
irm
Every-thing
Highquality
Closecompetitor
Nothing
Impr
ovem
ent p
oten
tial t
hrou
ghpa
tent
exc
hang
e (%
)
2.00
1.00
0.00Every-thing
Highquality
Closecompetitor
Nothing
0
50
100
150
200
250
300
EverythingHigh qualityClose competitorNothing
Competitor strategyto patent
Num
ber
of p
aten
ts f
or th
e fo
cal f
irm
no firm benefits by unilaterally deviating from activepatenting.) We therefore expect that, in markets ofcompeting technology leaders, considerable patent-ing activity will be observed. Since the differencesbetween the various active patenting strategies tendto be small, we expect to find a spectrum of differentactive strategies in practice.
Because both competitors adopt a symmetric activepatenting strategy, neither can use patenting to differ-entiate itself in the market; their product performanceis roughly equivalent. Hence both companies are bet-ter off exchanging their patents: doing so increasesthe absolute product performance of each firm with-out affecting the competitive stalemate. We thereforeexpect patent exchange schemes to be prevalent.
4.2. Best Strategies for an R&D LeaderCompeting with an R&D Follower
We now turn our attention to an asymmetric setting inwhich a focal R&D leader competes with a follower.How should the leader protect its inventions againsta follower? How should the leader respond to a fol-lower’s anticipated patenting strategies?
For this leader-follower combination, Figure 3 showsthe relative performance, the number of patents held
at the end of the race, the absolute performance, andthe expected gains from a patent exchange for thefocal R&D leader. If we focus on relative performancein order to identify best responses, then it is clear thatpatent nothing is the leader’s dominant strategy irre-spective of the follower’s strategy; to patent nothingis to outperform any active strategy, and by a consid-erable margin. Therefore, the leader should not revealinformation about its R&D trajectory by patenting,even though not patenting means that the leader can-not formally protect its technology position shouldthe follower happen upon it. The performance advan-tage of a leader competing against a follower stemsfrom secrecy: the follower has no reliable informa-tion concerning where its research efforts should befocused. In contrast, any strategy that includes patent-ing will provide a signal that the follower might useto improve its performance substantially.
Unlike the case of leader-leader competition, inthis case there are subtle but interesting differencesthat allow for distinctions among the active patent-ing strategies. Among these, the strategy of patentingonly high-quality innovations is an especially infe-rior choice. This insight contradicts the common intu-ition that companies should build patent portfoliosconsisting solely of high-quality patents in order tobalance protection and the cost of patenting. To thecontrary, we find that a leader firm adopting the strat-egy of high-quality patents is more easily outper-formed (than are leader firms adopting other activepatent strategies) by its follower: the better the sig-nal for the follower (i.e., the more clearly the leaderidentifies filed patents as superior ones), the easier itwill be for the follower to find desirable and perhapseven superior solutions in the technological vicinity ofthe leader’s inventions. The follower wastes no timeon inferior solutions, focusing instead on the mostpromising technology configurations as devised (andsignaled) by the leader. The implication is that, if aleader must patent, at least it should avoid signalingthe solution quality when competing against a fol-lower. When the leader either patents everything orpatents close to its competitor, much of the patentingdoes not involve superior technological solutions andthus confounds more than facilitates the follower’sR&D efforts.
With the number of patents serving as a proxy forpatent cost, the expenditures required under the dif-ferent strategies reinforce the superiority of the patentnothing strategy, which is cost optimal by defini-tion. It is noteworthy that, among the active patent-ing strategies, the close-competitor patenting strategymay yield the same number of patents as does thepatent everything strategy.
Neither analyzing absolute performance nor ana-lyzing the gains from patent exchange meaning-fully distinguishes the benefits and drawbacks of the
different patenting strategies. The strategy of patent-ing nothing is no less beneficial for absolute perfor-mance than is any active patenting strategy. And eventhough the potential gains from a patent exchangeare naturally higher for the active patenting strate-gies, such gains can at best compensate for the abso-lute performance disadvantages of the correspondingactive strategy in comparison with having patentednothing in the first place. Thus the patent nothingstrategy is unequivocally optimal for an R&D leadercompeting with a follower.
We have established how a leader should respondto any of the follower’s patenting strategies. Next,toward the end of characterizing industry equilibria,we analyze which strategy a follower should choose.
4.3. Best Strategies for an R&D FollowerCompeting with an R&D Leader
In this setting, our focal company is an R&D followerand its competitor is an R&D leader. Thus we reversethe previous setting’s perspective by now assumingthe role of the follower.
As usual, Figure 4 displays all four dimensions ofperformance for the focal firm. With regard to rela-tive performance, the follower clearly benefits froman active patenting strategy regardless of the strat-egy chosen by its competitor: any active strategy is noworse (and usually better) than not patenting at all.The intuition behind this result is that, by choosing anactive patenting strategy, the follower claims technol-ogy positions in the vicinity of the leader’s technologypositions and thus secures positions that the leaderitself might otherwise have claimed via patenting orpicked as a product solution at some later stage; inthis case, the follower carries out “offensive blocking(and inventing around).” The effect of this blocking isamplified if the follower receives high-quality infor-mation about favorable technology regions from theleader’s strategy to patent only high-quality innova-tions, in which case the follower might actually out-perform the leader. But even when the leader hidesits R&D advances by not providing any patent sig-nals, active patenting by the follower entails at leastsome possibility that it will block a potentially supe-rior invention by the leader.
Which strategy (among the active choices) shouldthe follower prefer? Patenting everything and close-competitor patenting exhibit the same relative perfor-mance and cost. This similarity is expected becausethe follower (by definition) expends its researchefforts in the technological vicinity of the leaderand so, simply by virtue of implementing a close-competitor strategy, patents everything. However,
Figure 4 Results for a Follower Competing with a Leader
Rel
ativ
e pe
rfor
man
ce o
f th
e fo
cal
firm
vs.
the
com
petit
or (
%)
0.00
2.00
–2.00
–4.00
–6.00
–8.00
4.00
0
50
100
150
200
250
300
Num
ber
of p
aten
ts f
or th
e fo
cal f
irm
Every-thing
Highquality
Closecompetitor
NothingEvery-thing
Highquality
Closecompetitor
Nothing
0.62
0.64
0.66
0.68
0.70
0.72
Abs
olut
e pe
rfor
man
ce o
f th
e fo
cal f
irm
Every-thing
Highquality
Closecompetitor
Nothing
Impr
ovem
ent p
oten
tial t
hrou
ghpa
tent
exc
hang
e (%
)
2.00
1.00
0.00Every-thing
Highquality
Closecompetitor
Nothing
EverythingHigh qualityClose competitorNothing
Competitor strategyto patent
a patent everything strategy decidedly outperformssolutions based on high-quality patenting—albeit ata substantially higher cost. Hence a trade-off results:if cost is a critical consideration, then the fol-lower should patent only high-quality innovations;if patenting costs are negligible, then the followershould patent everything.
These recommendations remain valid even afterwe take into account both absolute performance andthe potential gains from patent exchange. As com-pared with the strategy of no patenting, none ofthe active strategies exhibits inferior product perfor-mance. Moreover, after accounting for the potentialbenefits of a patent exchange, the absolute perfor-mance of the different strategies becomes virtuallyindistinguishable.
Now that we have analyzed the best responsesfor both leader and follower, the expected equi-librium follows easily. The leader patents nothingbecause this strategy is optimal regardless of thecompeting follower’s choice of strategy. In contrast,the follower either patents everything or selectivelypatents high-quality products. Note that the leader’snot patenting precludes any patent exchange with thefollower.
4.4. Best Strategies for an R&D FollowerCompeting with an R&D Follower
We now consider the final setting, in which one fol-lower competes with another. Overall, as shown inFigure 5, the best responses in terms of relative perfor-mance resemble those of a follower competing witha leader.5 Irrespective of the competitor’s strategy,the strategies of patenting everything and of close-competitor patenting (weakly) dominate the strat-egy of patenting nothing and clearly dominate thestrategy of patenting only high-quality innovations.Accounting for absolute performance reinforces thedominance of these two strategies: the high-qualitypatenting strategy remains slightly inferior both topatenting everything and to close-competitor patent-ing. Considering the potential value from patentexchange at best equalizes absolute performanceamong the three active strategies, but it leaves thepatent nothing strategy as an inferior option. Finally,
5 A paradoxical outcome results from one follower patentingactively while another follower patents nothing. Suppose the sec-ond follower patents nothing and instead builds upon patents ofthe first. But since the first follower does not have a firm to follow,it would linger on the initial (randomly chosen) position while thesecond follower profited from the first’s inventions. We thereforedisregard cases in which only one of the followers patents.
Figure 5 Results for a Follower Competing with a Follower
Every-thing
Highquality
Closecompetitor
Nothing
Every-thing
Highquality
Closecompetitor
Nothing
Every-thing
Highquality
Closecompetitor
Nothing
Rel
ativ
epe
rfor
man
ceof
the
foca
lfi
rm v
s. th
eco
mpe
tito
r (%
)
–1.00
–2.00
–3.00
0.00
1.00
2.00
3.00
0.61
0.63
0.65
0.67
0.69
0.71
0
50
100
150
200
250
300
350
Num
ber
of p
aten
tsfo
rth
efo
cal f
irm
Abs
olut
e pe
rfor
man
ceof
the
foca
l fir
m
Every-thing
Highquality
Closecompetitor
Nothing
Impr
ovem
entp
oten
tial
thro
ugh
pate
ntex
chan
ge (
%)
3.00
2.00
1.00
0.00
EverythingHigh qualityClose competitorNothing
Competitor strategyto patent
as expected, patenting nothing is less costly than anyactive patenting strategy. Among the active strate-gies, high-quality patenting is the cheapest becauseits overall cost is close to that of patenting noth-ing. Therefore, if performance is the firm’s dominantconcern, then it should pursue an active patent-ing strategy—preferably patent everything or close-competitor patenting. Yet if cost outweighs otherconsiderations, the high-quality patenting strategy isevidently the best option.
In equilibrium, we should therefore expect bothplayers to pursue either a strategy of patenting every-thing or a strategy of patenting only high-qualityinventions, depending on whether (respectively) per-formance or cost considerations dominate. Since thepatenting strategies chosen by the competitor aresymmetric, we expect that the two parties will engagein a patent exchange to increase absolute perfor-mance. It is noteworthy that two followers exploring atechnological landscape can produce an absolute levelof performance comparable to that reached in racesinvolving a leader. The reason is that two activelypatenting followers function, in effect, as a researchcartel. Instead of one company searching the land-scape, the two follower companies combine theirresearch efforts and dynamically propel each other’sperformance.
4.5. The Value of Patent AnalysisIn the preceding discussion, each leader pursued itsown R&D agenda and ignored any information aboutthe innovation activities of its competitor; this setupreflected a working assumption that a leader does notexpect technological advances to arise from its com-petitors. Now, however, we shall examine the valueof patent analysis from the leader’s perspective. Howvaluable is the ability to analyze competitors’ patentsand then absorb the embodied innovations into thefirm’s own research program? How do the patentstrategies and R&D types of the players affect thevalue of patent analysis? Will the capacity to analyzepatents affect outcomes of the competition?
For the purpose of this analysis we introduce theanalyzing leader. The analyzing leader bases the choiceof where next to focus its R&D efforts not only onthe portfolio of its own past explorations but alsoon the patented solutions of its competitor. Formally,we have the following: An analyzing leader i thatseeks to investigate a new technological configura-tion chooses a starting point for its R&D effort fromS4t5
i ∪P4t5j ; in contrast, a conventional leader j (which
does not perform patent analysis) bases its choice ofstarting point solely on solutions resulting from itsown R&D activities, S4t5
model (in particular, the functioning of patent andR&D strategies) remain unchanged.
However, we introduce a contingency in this con-text of an analyzing leader. Previously, we fol-lowed the dominant search paradigm (Fleming andSorenson 2001) and implemented a localized incre-mental R&D search in which firms identify new loca-tions for R&D efforts via a random change in onedecision. Yet research is sometimes characterized bya radical departure from existing practice as whenthe company explores completely new avenues (cf.Smith and Tushman 2005, Ceccagnoli 2009). To rep-resent such radical (distant) search behavior, we cre-ate one model variant that allows for “long jumps.”Technically, we follow Levinthal (1997) in supposingthat, whenever the firm carries out research, it eval-uates both a local move (changing one decision at atime) as well as a long jump (changing N decisionssimultaneously) to arrive at a random new position,selecting the move that offers higher performance. Allresults presented in the preceding sections still holdunder this generalization (see §4.6) and hence meritno further discussion. In the context of patent analy-sis, however, conceptualizing R&D as combined localand distant search does affect the results and hencerequires explicit exposition.
Because we now take the perspective of a leaderdeciding whether or not to implement patent analy-sis, we reconsider the outcomes of the “leader versusleader” and “leader versus follower” settings dis-cussed previously. We focus on relative performanceas our key metric in order to establish likely behaviorand equilibrium outcomes. (Taking into account abso-lute performance and the value of patent exchangedoes not alter any of our conclusions.) As a sec-ond metric, we establish the value of patent anal-ysis per se; this value is assessed in terms of thegains expected by an analyzing versus a conventionalleader. Thus we make two comparisons: (i) the caseof an analyzing leader competing with a conventionalleader to the case of a conventional leader competingwith a conventional leader; and (ii) the case of an ana-lyzing leader competing with a follower to the caseof a conventional leader competing with a follower.
For our first analysis, we focus on companiesengaging in incremental search. The upper left panelof Figure 6 plots the relative performance for ananalyzing leader competing against a conventionalleader. This figure indicates that, as a best responseto any active patenting strategy, active patentingstrategies (preferably, patenting everything or close-competitor patenting) dominate patenting nothingand also that, as a response to patenting nothing,active patenting strategies at least weakly dominatepatenting nothing. Hence an analyzing leader should
choose an active patenting strategy and refrain frompatenting nothing.
Choosing an active patenting strategy was the equi-librium outcome for both conventional leaders in thebase setting of §4.1. Although it might therefore seemas if an analyzing leader’s best response is in linewith such an equilibrium outcome, patent analysismodifies the equilibrium fundamentally. To see this,note that gains for the analyzing leader are losses forits competitor, the conventional leader. The conven-tional leader cannot choose a strategy such that itwins, whereas the analyzing leader can realize posi-tive gains for each choice by the conventional leader—that is, for each line in the graph. However, the con-ventional leader can limit the analyzing leader’s gainsby choosing not to patent. Thus the analyzing leader’schoice of an active strategy forces the conventionalleader to adopt the strategy of patenting nothing. Theconventional leader thereby reduces the risk of beingconstrained by a “patent girdle” devised by the ana-lyzing leader. Hence the equilibrium shifts from twoconventional leaders patenting actively to one ana-lyzing leader patenting actively and one conventionalleader patenting nothing.
This analysis of relative payoffs is reinforced by theupper right panel of Figure 6. The graph in this panelcompares the equilibrium of two competing conven-tional leaders with the equilibrium of an analyzingleader competing with a conventional leader. In thesymmetric equilibrium of two conventional leaderspatenting everything (leftmost point of the connect-ing lines), both firms perform at the same level. Ifone of the conventional leaders begins to analyze theother’s actions (and so becomes an analyzing leader),then the equilibrium shifts. As a result, the analyz-ing leader can now harness the conventional leader’sR&D efforts and so enjoys a relative benefit fromintroducing patent analysis. Overall performance forthe analyzing leader climbs to a slightly elevatedlevel.
An important takeaway is that the analyzing leaderis better off for having engaged in patent analysis,which creates a relative benefit and even a small abso-lute benefit for the company.
Examination of an analyzing leader and a conven-tional leader engaging in radical search yields substan-tially different insights. With respect to best-responsefunctions, the lower left panel of Figure 6 is stronglysimilar to the upper left one. Again, the analyzingleader’s best response to any competitor strategy isto choose an active patenting strategy. However, thisequilibrium is fundamentally different than the onefor firms engaging in incremental search. An activestrategy does not guarantee that the analyzing leaderwill dominate the competition. Rather, if the conven-tional leader chooses to patent everything then the
Figure 6 Results for an Analyzing Leader Competing with a Conventional Leader
0.71
0.70
0.69
0.68Conventional
leader/everything vs.conventional
leader/everything
Analyzingleader/everything vs.
conventionalleader/nothing
(%)
2.00
1.00
0.00
–1.00 Abs
olut
e pe
rfor
man
ce f
or b
oth
firm
s
Incremental search
Rel
ativ
e pe
rfor
man
ce o
f th
e an
alyz
ing
lead
er v
s. th
e le
ader
Conventionalleader/everything vs.
conventionalleader/everything
Analyzingleader/everything vs.
conventionalleader/everything
Abs
olut
e pe
rfor
man
ce f
or b
oth
firm
s
0.72
0.70
0.68
Every-thing
Highquality
Closecompetitor
Nothing
1.00
0.00
(%)
–1.00
Radical search
Rel
ativ
e pe
rfor
man
ce o
f th
e an
alyz
ing
lead
er v
s. th
e le
ader
Every-thing
Highquality
Closecompetitor
Nothing
EverythingHigh qualityClose competitorNothing
Competitor strategyto patent
Conventional -> analyzing
Conventional leader
analyzing leader can at best attain a slightly nega-tive relative performance, which means that the con-ventional leader outperforms the analyzing leader. Inequilibrium, of course, the conventional leader willchoose to do so. Hence the equilibrium will be a con-ventional leader patenting everything and the analyz-ing leader choosing one of the active strategies.
Why does the conventional leader win the compe-tition? Both players are capable of performing rad-ical innovations—that is, both can search broadly.The analyzing leader may further broaden its searchby adopting competitor positions as new bases forits own R&D efforts. However, when following andinventing around another player, the analyzing leaderrisks wasting valuable R&D efforts by searching inthe (perhaps densely protected) neighborhood of thecompeting conventional leader. So for a leader whocan already search broadly, it is better not to relyon another leader’s patents; the equilibrium outcomenegates the value of patent analysis in the case of rad-ical search. Patent analysis could actually be harmful
for the company engaging in it, which explains whyno rational firm would perform patent analysis inenvironments characterized by radical search.
The upper (respectively, lower) left panel of Fig-ure 7 plots for the incremental (respectively, radical)search case the relative performance of an analyzingleader competing against a follower. The analyzingleader’s patent analysis capabilities do not qualita-tively alter the best responses of a conventional leaderas described in §4.2. Regardless of the follower’s strat-egy, the analyzing leader should patent nothing; and,as before, the analyzing leader should definitely avoidpatenting only high-quality solutions. The followerresponds by choosing to patent everything. Hence,in the case of an analyzing leader competing with afollower, patent analysis does not alter the prevailingequilibrium.
However, two comparisons reveal a remarkableconsequence of that equilibrium. The equilibriumpoint in the upper left panel of Figure 7 is lower thanthe corresponding point in Figure 3. The upper right
Figure 7 Results for an Analyzing Leader Competing with a Follower
Rel
ativ
e pe
rfor
man
ce o
f th
e an
alyz
ing
lead
er v
s. th
e fo
llow
er (
%)
2.00
–2.00
–4.00
0.00
4.00
6.00
8.00
Radical search
Incremental search
Rel
ativ
e pe
rfor
man
ce o
f th
e an
alyz
ing
lead
er v
s. th
e fo
llow
er (
%)
12.00
10.00
8.00
6.00
4.00
2.00
–2.00
0.00
Every-thing
Highquality
Closecompetitor
Nothing
Every-thing
Highquality
Closecompetitor
Nothing
EverythingHigh qualityClose competitorNothing
Competitor strategyto patent
0.72
0.62
0.64
0.66
0.68
0.70
Conventionalleader/nothing vs.
follower/everything
Analyzingleader/nothing vs.
follower/everything
0.74
0.72
0.70
0.68
0.66
0.64
0.62Abs
olut
e pe
rfor
man
ce f
or b
oth
firm
s
Conventionalleader/nothing vs.
follower/everything
Analyzingleader/nothing vs.
follower/everything
Abs
olut
e pe
rfor
man
ce f
or b
oth
firm
s Conventional -> analyzing
Follower
panel of Figure 7 shows a decline: in equilibrium—under which the leader chooses to patent nothingand the follower chooses to patent everything—theanalyzing leader is actually worse off than a con-ventional leader in terms of both absolute and rela-tive performance. The decline is not large but it isrelevant: the follower does not have enough infor-mation to launch promising R&D initiatives itself.In fact, the follower’s R&D efforts are largely undi-rected. Hence the follower produces substantial quan-tities of low-quality patents. Such patents perturb theanalyzing leader’s patent analysis and so diffuse theleader’s own R&D initiatives. Moreover, the analyz-ing leader’s strategy to patent nothing does not pro-tect against the infringements that are likely to occurwhen novel R&D initiatives are inspired by the fol-lower’s patents. This problem is inherent to patentanalysis unless the analyzing leader is able to dis-tinguish perfectly between better and worse solutions.It would be unrealistic to assume such capacities ina firm so patent analysis is definitively not recom-mended for leaders that compete with followers.
The results of this section are worth highlighting.In contrast to our intuition, patent analysis need notimprove the performance of the analyzing leader—even if it has the capacity to assimilate competi-tor innovations into its R&D program (Cohen andLevinthal 1990). In fact, patent analysis may proveharmful: an analyzing leader that competes with aconventional leader benefits only when both carry outincremental search; the analyzing leader encountersdeteriorating performance when both engage in radi-cal search. In the case of an analyzing leader compet-ing with a follower, patent analysis simply does notadd value for the analyzing leader.
The pattern of when analysis is beneficial versusharmful is quite intricate. Our results may thereforeexplain why competitive patent analysis is neitherwidely promoted in the literature nor used in prac-tice as a standard tool for establishing R&D strategies(Fabry et al. 2006).
4.6. Industry and Additional Firm CharacteristicsWe have implicitly assumed throughout our analy-ses that the most critical moderating influence on the
optimal choice of a patenting strategy is the market’scompetitive dynamics—that is, whether the focal firmis an R&D leader or follower and whether this firm isin competition with a leader or a follower. However,other variables at the industry level and the firm levelmay also affect the choice of patenting strategy.
Industry-Level Influences. At the industry level, wehave systematically varied two of the most central fac-tors: the technology landscape’s complexity and theinnovation race’s search time. Landscape complexityis a proxy for how many acceptable solutions existto a given technological challenge and for how diffi-cult it is to find them. A complex landscape may haveseveral solutions, but they will be difficult to iden-tify. Many different technologies might offer a similarperformance, and competitors can then “avoid” eachother. A simple landscape has relatively few solutions,which are easily identified by surveying the currentconfiguration’s neighborhood. As a result, only a lim-ited set of technologies achieves a reasonable levelof performance and so different companies’ researchefforts will quickly coalesce around those technolo-gies. Search time is our proxy for different industryclock speeds. Some markets require that products beextremely mature; a new technology must undergoseveral iterations before a solution merits marketableproduction (Sommer and Loch 2004). In other, moreexperimental markets, technologies may be quicklytranslated into marketable products without muchiteration. This explains why innovation races quicklyterminate in some markets but span extensive periodsof time in others.
Both of these factors could well affect the optimalstrategy in a given situation. We therefore systemat-ically vary both complexity (using K = 0 and K = 11rather than the main model’s K = 6) and the lengthof the race (10 and 200 steps rather than the mainmodel’s 100 steps). It is remarkable that variationsin complexity and/or in timing do not change ourresults. Although the reported effects become morepronounced for higher complexities and less pro-nounced for shorter times, our qualitative conclusionsremain valid in any case. These analyses imply thatfirms, when choosing their patent strategy, shouldnot focus on industry clock speed or the technologylandscape’s specific shape. Instead, the most criticaldeterminants of an optimal patenting strategy are thecharacteristics of the innovation race’s competitors.
Alternative Firm-Level Factors. Finally, we describethe effects that two additional firm-level character-istics could have on how patent races unfold. Thefirst characteristic is the search behavior of the leader.Firms may engage in both incremental and radicalR&D activities. To accommodate the possibility ofradical innovation, we extend our model to includelong jumps (cf. Levinthal 1997). The results for a
leader competing with a leader or a follower do notchange; however, as laid out in §4.5, the value ofpatent analysis is definitively affected by the intro-duction of long jumps.
As for the second characteristic, the follower maynot be able to (fully) discern the quality of the patentsfiled by a leader; in other words, it may lack absorp-tive capacity. So that our model can accommodatesuch lack of capacity, we have the follower pick atrandom the (leader) patent on which to base its R&Defforts, instead of picking it based on its inherentquality. This modification of the model does not affectthe conclusions that we draw from the analysis, as ourresults remain robust with respect to a follower withless absorptive capacity. It seems that given enoughtime, the follower will run across the leader’s morepromising patents eventually.
The lack of absorptive capacity could alternativelybe conceived as all firms (whether leader or follower)being unable to understand the value of the tech-nology configuration that they have discovered. Suchan error could arise from the randomness associatedwith any commercialization issue, or could more sim-ply be explained by insufficiently capable firms. Weaccommodate this alternative conceptualization in themodel by having the firms base their decisions not onthe technology configuration’s actual value but ratheron a the landscape value plus a white-noise term(normally distributed with N40100055). Once again,the results are affected quantitatively but not qualita-tively: all our conclusions continue to hold.
The last two results underscore an important fact.The leader cannot use the follower’s lack of ability toprotect itself; instead, the leader must actually pursuethe strategy of not patenting anything.
5. DiscussionTheoretical research on patenting has examined thetrade-off between patent protection and disclosure ofinformation at the level of specific innovations (e.g.,Horstmann et al. 1985). These models conceptualizethe decision of patenting versus secrecy either as abinary choice or as continuous substitutes that aretraded off by a “propensity to patent.” Using thisconvenient abstraction, the model-based literature hasestablished the “nature of the innovation” and the“industry with its ruling patent regime” as factorsthat determine optimal resolutions of the trade-offbetween patenting and secrecy. In line with thosepredictions, empirical research on patenting has pro-vided broad evidence that firm-level patenting activ-ities are heterogeneous across industries (Hall et al.2014). Yet, empirical research has also suggested thatpatenting strategies vary even within industries (e.g.,Cohen et al. 2000). In addition, there is evidence
Leader Active patentingPatenting high quality No patentingPatenting close competitor Leader wins
Both perform equally well No patent exchangePatent exchange
Follower Active patenting Active patentingIf cost is a consideration2 Patent high quality If cost is a consideration2 Patent high qualityIf cost is not a consideration2 Patent everything If cost is not a consideration2 Patent everything(or close competitor) (or close competitor)
Leader wins Both perform equally wellNo patent exchange Patent exchange
that patenting activities of firms operating in indus-tries with stable characteristics and unchanged patentregimes can nonetheless change over time (Jell andHenkel 2010). Our model contributes to reconcilingthis apparent inconsistency between theoretical pre-dictions and empirical findings by adopting a morefine-grained and dynamic perspective on patenting:in our coevolutionary framework, the effectiveness ofa patenting strategy emerges as being intricately con-nected to the firm’s own R&D strategy and linkedalso to its competitors’ R&D and patenting strategies.
In this paper we identify contingencies that affectthe choice of a patenting strategy and then char-acterize the optimal choices with respect to thosecontingencies. As do previous models, we incorporatestructural industry characteristics as contingency fac-tors. Yet even though technological complexity andindustry clock speed influence patenting outcomes,they have no appreciable effect on the choice of anoptimal patenting strategy. In contrast, the firm’s strat-egy is substantially affected by competitive dynamicsthat stem from its own and its competitor’s researchstrategy and from its competitor’s patenting strat-egy. Hence the optimal strategy is a function of(1) whether the firm is competing against an R&Dleader or a follower, (2) whether the firm itself aimsto be a leader or a follower, and (3) what patent-ing choices the competitor usually makes. Our modelthus explains heterogeneity in patenting strategies byhitherto omitted contingencies such as variations inthe R&D strategies of incumbent firms. We pay partic-ular attention to the realism of our modeled patentingstrategies by grounding them in the well-documentedbehavior of actual firms. Table 3 summarizes ourmain results.
In sum, we find that if an R&D leader competeswith another R&D leader then both should pursue anactive patenting strategy. A strategy of patenting onlyhigh-quality innovations is preferable for both par-ties, with the strategy of close-competitor patenting a
near second. When an R&D leader is competing witha follower, the leader should patent nothing and thefollower should either patent everything (if patent-ing costs are negligible) or patent only high-qualityinventions (if patenting costs are high). Finally, if twofollowers compete with each other then both shouldemploy either a high-quality patenting strategy (ifpatenting costs are high) or a patent everything strat-egy (if those costs are negligible).
Patent exchange is likely to occur when a leadercompetes against a leader or when a follower com-petes against a follower, since in both of those casesthe exchange is mutually beneficial. However, patentexchange is unlikely when a leader competes with afollower because only the leader would benefit fromthe exchange.
This paper also sheds light on the value of patentanalysis for R&D leaders. Patent analysis makes com-petitive information available to the analyzing leaderand thus could be assumed to give it an edge. In con-trast to that intuition, however, patent analysis neednot actually improve the performance of an R&Dleader: its value depends intricately on the natureof the R&D activity. If the firms are engaged inincremental search of the technology landscape, thenpatent analysis is beneficial for the leader because itcreates a relative benefit and also a small absolutebenefit. But if the firms are engaged in radical search,then a different conclusion emerges: patent analysisdoes not yield benefits and may, in fact, be harmful tothe extent that the leader is thereby induced to chasepotentially harmful competitor ideas. Patent analysisis ambivalent. The overall usefulness of patent analy-sis is thus limited, which hints at the fundamental rea-son why—despite its prima usefulness—such analysishas not become a standard tool in R&D strategizing.
We have explicitly assumed that the R&D strategyis determined by factors external to the model; hencethe company’s only decisions concern its patentingstrategy. In the long run, though, the research strategy
in itself may be a decision variable. Our patentinganalysis can yield some insight on such a combinedR&D and patenting strategy as follows.
The equilibria of the different settings exhibit dif-ferent levels of stability with respect to changes in theR&D strategy. Can any of the players improve theirperformance by changing their R&D strategy? If so,then evolutionary forces will eventually drive them torealize that improvement potential. When a followercompetes against a follower, both perform at aboutthe same level and patent their results. Yet by adopt-ing the behavior of an R&D leader, a follower couldbecome relatively more competitive and thus win thetechnology competition—unless the follower can reapsubstantial cost savings in R&D. Hence the scenario inwhich two followers are competing against each otheris not a stable one in technology-dominated indus-tries. Now consider what happens when a leadercompetes against a follower. The follower patents, butthe leader refrains from patenting and is able to winthe innovation race. Hence this scenario, too, is unsta-ble (again barring R&D cost advantages). The remain-ing follower should therefore attempt to become anR&D leader. In contrast, the equilibrium in whicha leader competes against another leader is a sta-ble one. Both firms perform at about the same level,and both patent using similar strategies. Most impor-tantly, switching to a follower role would entail lossesin relative performance. Overall, then, the followerrole is transitional; it does not exhibit long-run stabil-ity in industries that are dominated by technologicalperformance.
From a methodological point of view, one of ourchoices deserves critical reflection. We conceptualizedthe choice of patenting strategies as a simultaneous-move game over a set of predefined strategies. Alter-natively, we could have formulated a dynamic gamein which players proceed in rounds: (i) analyzing newcompetitor patents and potentially imputing informa-tion about competitor moves that were not patented(e.g., the search location and the quality of the out-come), (ii) deciding on their own search location;and (iii) based on that outcome, deciding whether ornot to patent. Solving such a game would automat-ically yield the optimal patenting behavior as wellas the optimal R&D search behavior. However, suchan equilibrium would require that, at any time thefirm not only build an entire map of the documentedsearch locations of its competitor (the competitor’spatents) but also make inferences about its competi-tor’s unpublished search locations using a preciseprobability distribution over the number of searchlocations, their precise coordinates, and their poten-tial outcomes. The equilibrium would also require that
the firm project its own and its competitors’ actions—including the potential search outcomes into the (pos-sibly infinite) future—and then, based on such projec-tions, make a globally optimal dynamic decision onwhether or not to patent. Although we do not dis-pute the value of an academic effort to characterizesuch equilibria, we believe that the implicit informa-tion processing requirements cannot be met by anyreal-life company. Similar considerations led Simon(1969) to devise the concept of “bounded rational-ity.” In our context, bounded rationality would leadfirms to adopt behavioral strategies and to refine thosestrategies if they prove unsuccessful. Such behavioralstrategies do not need to be static, and may well allowthe firm to adjust its behavior dynamically in responseto competitor moves. However, such adjustments areformulated as contingencies ex ante: the policy con-tains “what if” statements. The realism of such behav-ioral strategies comes at a cost, however. The actionset contains only predefined strategies, so no newstrategies can emerge. Hence it is important for thequality of our analysis that we identify the appropri-ate archetypes among the behavioral strategies. For-tunately, we could ground our work in an extensiveempirical literature (e.g., Granstrand 1999) that hasidentified these archetypes. Yet we acknowledge that,in order to limit complexity, the strategies addressedhere span only the most salient classes of patentingstrategies. There may be strategy variants that wouldyield improvements over the cases we have treated.
A second aspect of our paper requires considerationwhen interpreting the results. The analysis is basedon a formal model that is investigated via computa-tional means; absent an accompanying empirical anal-ysis that verifies the conclusions, our patent strategyadvice should be viewed with an appropriate levelof skepticism. That being said, the recommendationsderived from our model form immediately testablehypothesis about the most likely patenting strategiesin different competitive situations.
In conclusion, this paper has developed a set of nor-mative prescriptions for how companies should makeuse of patents in innovation races. In so doing, ourresearch has taken a meaningful step toward a con-tingency theory of patent strategies.
AcknowledgmentsThe authors gratefully acknowledge comments and sugges-tions by Vikas Aggarwal, Knut Blind, Murat Tarakci, sem-inar participants at Erasmus University and University ofSouthern Denmark, as well as the editorial team.
ReferencesAdee S (2010) EUV’s underdog light source will have its day. IEEE
Aggarwal VA, Hsu DH (2009) Modes of cooperative R&D commer-cialization by start-ups. Strategic Management J. 30(8):835–864.
Anton JJ, Yao DA (2004) Little patents and big secrets: Managingintellectual property. RAND J. Econom. 35(1):1–22.
Aoki R, Hu JL (1999) Licensing vs. litigation: The effect of the legalsystem on incentives to innovate. J. Econom. Management Strat-egy 8(1):133–160.
Aoki R, Spiegel Y (2009) Pre-grant patent publication and cumula-tive innovation. Internat. J. Indust. Organ. 27(3):333–345.
Arora A, Ceccagnoli M (2006) Patent protection, complementaryassets, and firms’ incentives for technology licensing. Manage-ment Sci. 52(2):293–308.
Arora A, Ceccagnoli M, Cohen WM (2008) R&D and the patentpremium. Internat. J. Indust. Organ. 26(5):1153–1179.
Arundel A (2001) The relative effectiveness of patents and secrecyfor appropriation. Res. Policy 30(4):611–624.
Arundel A, Patel P (2003) Strategic patenting. Background Reportfor the Trend Chart Policy Benchmarking Workshop NewTrends in IPR Policy, Luxembourg.
Arundel A, van de Paal G, Soete L (1995) Innovation strategies ofEurope’s largest industrial firms: Results of the PACE surveyfor information sources, public research, protection of innova-tions and government programmes. MERIT Final Report.
Baumann O, Siggelkow N (2013) Dealing with complexity: Inte-grated vs. chunky search processes. Organ. Sci. 24(1):116–132.
Benschop J (2010) EUV: Status and challenges ahead. Keynote talk.2010 International Workshop on EUV Lithography, Maui.
Bessen J (2005) Patents and the diffusion of technical information.Econom. Lett. 86(1):121–128.
Blind K, Cremers K, Mueller E (2009) The influence of strategicpatenting on companies’ patent portfolios. Res. Policy 38(2):428–436.
Blind K, Edler J, Frietsch R, Schmoch U (2006) Motives to patent:Empirical evidence from Germany. Res. Policy 35(5):655–672.
Ceccagnoli M (2009) Appropriability, preemption, and firm perfor-mance. Strategic Management J. 30(1):81–98.
Chao RO, Kavadias S (2008) A theoretical framework for managingthe new product development portfolio: When and how to usestrategic buckets. Management Sci. 54(5):907–921.
Choi JP (1998) Patent litigation as an information-transmissionmechanism. Amer. Econom. Rev. 88(5):1249–1263.
Cohen WM, Levinthal D (1990) Absorptive capacity: A new per-spective on learning and innovation. Admin. Sci. Quart. 35(1):128–152.
Cohen WM, Nelson RR, Walsh JP (2000) Protecting their intellectualassets: Appropriability conditions and why U.S. manufactur-ing firms patent (or not). NBER Working Paper 7552, NationalBureau of Economic Research, Cambridge, MA.
Crama P, De Reyck B, Degraeve Z (2008) Milestone payments orroyalties? contract design for R&D licensing. Oper. Res. 56(6):1539–1552.
Csaszar FA, Siggelkow N (2010) How much to copy? Determinantsof effective imitation breadth. Organ. Sci. 21(3):661–676.
Denicolò V, Franzoni AL (2004) Patents, secrets, and the first-inventor defense. J. Econom. Management Strategy 13(3):517–538.
Erkal N (2005) The decision to patent, cumulative innovation, andoptimal policy. Internat. J. Indust. Organ. 23(7):535–562.
Ethiraj SK, Posen HE (2013) Do product architectures affect inno-vation productivity in complex product ecosystems? AdnerR, Oxley JE, Silverman BS, eds. Collaboration and Competitionin Business Ecosystems (Advances in Strategic Management, Vol-ume 30) (Emerald Group Publishing Limited, Bingley, UK),127–166.
Fabry B, Ernst H, Langholz J, Köster M (2006) Patent portfolioanalysis as a useful tool for identifying R&D and businessopportunities—An empirical application in the nutrition andhealth industry. World Patent Inform. 28(3):215–225.
Fleming L, Sorenson O (2001) Technology as a complex adaptivesystem: Evidence from patent data. Res. Policy 30(7):1019–1039.
Gallini NT (1992) Patent policy and costly imitation. RAND J.Econom. 23(1):52–63.
Gilbert R, Shapiro C (1990) Optimal patent length and breadth.RAND J. Econom. 21(1):106–112.
Granstrand O (1999) The Economics and Management of IntellectualProperty: Towards Intellectual Capitalism (Edward Elgar, Chel-tenham, UK).
Grindley PC, Teece DJ (1997) Managing intellectual capital: Licens-ing and cross—Licensing in semiconductors and electronics.Calif. Management Rev. 39(2):8–41.
Hall BH (2005) Exploring the patent explosion. J. Tech. Transfer30(1):35–48.
Hall BH, Ziedonis RH (2001) The patent paradox revisited: Anempirical study of patenting in the U.S. semiconductor indus-try, 1979–1995. RAND J. Econom. 32(1):101–128.
Hall B, Helmers C, Rogers M, Sena V (2014) The choice betweenformal and informal intellectual property: A review. J. Econom.Literature 52(2):1–50.
Horstmann IJ, MacDonald G, Slivinski A (1985) Patents as informa-tion transfer mechanisms: To patent or (maybe) not to patent.J. Political Econom. 93(5):837–858.
Hsu DH, Ziedonis RH (2008) Patents as quality signals forentrepreneurial ventures. Acad. Management Best Paper Proc.(Academy of Management, Briarcliff Manor, NY), 1–6.
Jell F, Henkel J (2010) Patent portfolio races in concentrated marketsfor technology, Working paper Technical University of Munich.
Judd KL (1985) On the performance of patents. Econometrica53(3):567–586.
Kash DE, Kingston W (2001) Patents in a world of complex tech-nologies. Sci. Public Policy 28(1):11–22.
Kauffman SA (1993) The Origins of Order: Self-Organization and Selec-tion in Evolution (Oxford University Press, New York).
Klemperer P (1990) How broad should the scope of patent protec-tion be? RAND J. Econom. 21(1):113–130.
Knudsen T, Levinthal DA (2007) Two faces of search: Alternativegeneration and alternative evaluation. Organ. Sci. 18(1):39–54.
Kultti K, Takalo T, Toikka J (2006) Simultaneous model of innova-tion, secrecy, and patent policy. Amer. Econom. Rev. 96(2):82–86.
Kwon I (2012) Patent races with secrecy. J. Industrial Econom.60(3):499–516.
Langinier C (2004) Are patents strategic barriers to entry? J. Econom.Bus. 56(5):349–361.
Lenox M, King A (2004) Prospects for developing absorptive capac-ity through internal information provision. Strategic Manage-ment J. 25(4):331–345.
Lerner J (1994) The importance of patent scope: An empirical anal-ysis. RAND J. Econom. 25(2):319–333.
Levin RC, Klevorick AK, Nelson RR, Winter SG, Gilbert R, GrilichesZ (1987) Appropriating the returns from industrial researchand development. Brookings Papers Econom. Activity 1987(3):783–831.
Levinthal DA (1997) Adaptation on rugged landscapes. ManagementSci. 43(7):934–950.
Levinthal DA, Warglien M (1999) Landscape design: Designing forlocal action in complex worlds. Organ. Sci. 10(3):342–357.
Macdonald S (2004) When means become ends: Considering theimpact of patent strategy on innovation. Inform. Econom. Policy16(1):135–158.
Mihm J, Loch CH, Huchzermeier A (2003) Problem-solving oscil-lations in complex engineering projects. Management Sci.49(6):733–750.
Mihm J, Loch CH, Wilkinson D, Huberman B (2010) Hierarchicalstructure and search in complex organizations. Management Sci.56(5):831–848.
Moore GE (1965) Cramming more components onto integrated cir-cuits. Electronics 38(8):114–117.
Nadkarni S, Narayanan VK (2007) Strategic schemas, strategic flex-ibility, and firm performance: The moderating role of industryclockspeed. Strategic Management J. 28(3):243–270.
Nelson R, Winter S (1982) An Evolutionary Theory of Economic Change(Belknap Press, Cambridge, MA).
Oldham GR, Cummings A (1996) Employee creativity: Personaland contextual factors at work. Acad. Management J. 39(3):607–634.
Ottoz E, Cugno F (2008) Patent–secret mix in complex productfirms. Amer. Law Econom. Rev. 10(1):142–158.
Reinganum JF (1982) A dynamic game of R&D: Patent protectionand competitive behavior. Econometrica 50(3):671–688.
Reitzig M (2004) The private values of “thickets” and “fences”:Towards an updated picture of the use of patents across indus-tries. Econom. Innovation and New Tech. 13(5):457–476.
Reitzig M, Henkel J, Heath C (2007) On sharks, trolls, and theirpatent prey: Unrealistic damage awards and firms’ strategiesof being infringed. Res. Policy 36(1):134–154.
Scotchmer S, Green J (1990) Novelty and disclosure in patent law.RAND J. Econom. 21(1):131–146.
Sheehan J, Martinez C, Guellec D (2004) Understanding businesspatenting and licensing: Results of a survey. Patents Innovationand Economic Performance OECD Conf. Proc. (Organisation forEconomic Co-operation and Development Publishing, Paris,France), 89–110.
Simon HA (1969) The Sciences of the Artificial, 2nd ed. (MIT Press,Cambridge, MA).
Smith WK, Tushman ML (2005) Managing strategic contradictions:A top management model for managing innovation streams.Organ. Sci. 16(5):522–536.
Sommer SC, Loch CH (2004) Selectionism and learning in projectswith complexity and unforseeable uncertainty. Management Sci.50(10):1334–1347.
Teece DJ (1986) Profiting from technological innovation: Implica-tions for integration, collaboration, licensing and public policy.Res. Policy 15(6):285–305.
Waterson M (1990) The economics of product patents. Amer.Econom. Rev. 80(4):860–869.
Weber J (1990) Toshiba will pay TI royalty on chip patents. LosAngeles Times (December 12), http://articles.latimes.com/1990-12-12/business/fi-5989_1_japanese-patent.
Ziedonis RH (2004) Don’t fence me in: Fragmented markets fortechnology and the patent acquisition strategies of firms. Man-agement Sci. 50(6):804–820.
