Strategic Management Journal Strat. Mgmt. J. (2011) Published online EarlyView in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/smj.971 Received 18 December 2008; Final revision received 28 September 2010

SOME LIKE IT FREE: INNOVATORS’ STRATEGIC USE OF DISCLOSURE TO SLOW DOWN COMPETITION GONCALO ¸ PACHECO-DE-ALMEIDA1 * and PETER B. ZEMSKY2 1 2

HEC Paris, Jouy-en-Josas, France INSEAD, Fontainebleau, France

Why do some innovators freely reveal their intellectual property? This empirical puzzle has been a focal point of debate in the R&D literature. We show that innovators may share proprietary technology with rivals for free—even if it does not directly benefit them—to slow down competition. By disclosing IP, innovators indirectly induce rivals to wait and imitate instead of concurrently investing in innovation, which alleviates competitive pressure. In contrast with the classical strategy view, our paper also shows that imitators may not always benefit from interfirm knowledge spillovers. Specifically, imitators may want to limit the know-how that they can freely appropriate from innovators. Otherwise, innovators have fewer incentives to quickly develop new technologies, which, ultimately, reduces the pace and profits of imitation. Copyright  2011 John Wiley & Sons, Ltd.

INTRODUCTION The classical view in competitive strategy is that interfirm knowledge spillovers are detrimental to innovators but beneficial to imitators. ‘In this classical view, appropriating returns to innovation requires agents to keep the knowledge underlying an innovation secret or to protect it by patents (or other means). (. . .) [N]on-compensated spillovers (. . .) should represent a loss that innovators would seek to avoid’ (Harhoff, Henkel, and von Hippel, 2003: 1754).1 Why, then, do some innovators Keywords: R&D and technology; innovation dynamics; timing games; time compression diseconomies; firm spillovers ∗

Correspondence to: Gon¸calo Pacheco-de-Almeida, HEC Paris, Strategy and Business Policy Department, 1 rue de la Lib´eration, 78351 Jouy en Josas, Cedex, France. E-mail: [email protected] 1 The classical view finds its origins in the seminal work of Arrow (1962) and Spence (1984) on the disincentive effect of spillovers on R&D investments by innovators. Other prominent contributions in this research tradition include Scherer (1959),

Copyright  2011 John Wiley & Sons, Ltd.

freely reveal their intellectual property (IP)? A small but growing literature has widely documented and tried to explain this puzzling empirical regularity. This body of work identified several mechanisms that compensate innovators for the free diffusion of their proprietary information. Examples include gains in firm reputation and market enlargement from collective invention (Allen, 1983), network externalities and standard setting (Harhoff, 1996; Lim, 2009), related innovations induced among and revealed by other firms (De Fraja, 1993), cross-spillovers and agglomeration Mansfield (1985, 1986), Levin et al. (1987), and Cohen, Nelson, and Walsh (2000), which furthered our understanding of innovators’ appropriability problem and how it might be solved through IP protection. The ‘conventional wisdom (. . .) [that firms] will want to keep their spillover rate as low as possible in order not to benefit rivals’ (De Fraja, 1993: 140) and that ‘any uncompensated information transfer that occurs [between firms] (. . .) must be involuntary’ (Harhoff et al., 2003: 1767) has been central not only to the abovementioned literature on innovation and IP protection but also to the literature on imitation and sustainable competitive advantage (e.g., Dierickx and Cool, 1989; Zander and Kogut, 1995).

G. Pacheco-de-Almeida and P. B. Zemsky economies (d’Aspremont and Jacquemin, 1988; Fujita, Krugman, and Venables, 1999; Krugman, 1991; Marshall, 1920), access to other firms’ complementary capabilities (Harhoff et al., 2003), and utility from compliance with industry norms (von Hippel and von Krogh, 2003). The main contribution of our paper is to show that, even without any compensating mechanism, innovators may still find it optimal to freely disclose proprietary knowledge. That is, voluntary spillovers may occur even if innovation diffusion is only ‘bad news’ for the innovator—if the development of the same innovation by other firms brings no benefits to the innovator and simply results in stiffer market competition. Free revealing by innovators may be an effective way of dissuading rivals from concurrently investing in innovation and inducing them to wait and imitate, which alleviates competitive pressure. In other words, by committing to partially disclose private information at the end of the technology development process, innovators may persuade competitors to switch from concurrent to imitative development strategies—thereby slowing down competitive innovation. Free revealing may be particularly useful when what is at stake is not if, but when rivals develop similar technologies, which seems to be the case in most industries (Harhoff et al., 2003; Mansfield, 1985). Thus, innovators should not ‘conceal to reduce diffusion’ (classical view), but often ‘reveal to induce delayed diffusion.’ In this paper, we denote by diffusion-delaying spillovers the minimum optimal level of free revealing by innovators that induce followers to switch from concurrent to imitative technology development. This finding is important because it shows that voluntary spillovers—proprietary technology disclosure, patenting in countries with weak IP protection, or geographical collocation with industry followers—should be an optimal strategy more often than expected. Diffusion-delaying spillovers are an effective anticompetitive strategy seldom subject to antitrust laws. Also in contrast with the classical view, we show that larger spillovers may not always be beneficial to imitators. Imitators may have an incentive to limit the know-how that they freely appropriate from an innovator, as otherwise they may slow down the leader’s development and, thereby, their own imitation. Thus, imitators may be better off with lower levels of absorptive capacity (Cohen and Levinthal, 1990). This finding is true not only Copyright  2011 John Wiley & Sons, Ltd.

under firm symmetry (as in Pacheco-de-Almeida and Zemsky, 2007), but even if innovators have superior capabilities. Finally, we show that technology leader firms are faster to market and have more incentives to invest in capabilities that speed up technology development under concurrent than imitative regimes. Whether follower firms acquire the new technology more rapidly under imitation or concurrent development depends on the level of interfirm spillovers. We also find that followers in new industries are more likely to pursue imitation as leaders’ development capabilities improve. In this paper, we denote firms’ capabilities to accelerate technology development by time compression, or speed, capabilities.

THE TIMING OF INNOVATION The timing of innovation is central to interfirm spillovers for two main reasons. On the one hand, spillovers presuppose that innovation activities of different firms are timed sequentially—otherwise, followers could not learn from precedent. On the other hand, spillovers accelerate followers’ innovation, thereby rendering first-mover advantages more transient. Thus, firm differences in innovation timing are both a precondition for and an outcome of interfirm spillovers. As a result, our paper directly builds on the innovation timing literature, which we review in detail below. Although our focus is obviously on past theoretical contributions, we start by briefly characterizing the broad empirical relevance of timing in innovation. A large body of empirical literature has recognized that timing is a key success factor in the research and development (R&D) of new technologies by profit-seeking firms. Prior studies on first-mover advantages have shown that the payoffs to competitors are typically discontinuous in the timing of entry into new technical subfields. Specifically, this line of work has explored the conditions under which the first firms to introduce new products or technologies earn a substantial premium (e.g., Lieberman and Montgomery, 1988; Mitchell, 1991). The related literature on strategic commitment has hailed timing as the main driver of investment value in new market opportunities. The right innovation timing should be early enough to secure market space and discourage rivals from investing, but not too early to shut Strat. Mgmt. J. (2011) DOI: 10.1002/smj

Innovators’ Strategic Use of Disclosure to Slow down Competition down options before technological uncertainty subsides (Ghemawat, 1991; Trigeorgis, 1996; for a review, see Besanko et al., 2007). Prior work on time pacing has further emphasized the importance of timing as a mechanism through which firms self-impose regular innovation goals to keep an industry lead (Eisenhardt and Brown, 1998). The timing of technology search and transfer has also been put forward as a main driver of firm heterogeneity (Katila and Chen, 2009; Salomon and Martin, 2008), and innovation timing has been at the core of empirical research on hypercompetition and dynamic capabilities (D’Aveni, 1994; Helfat et al., 2007; Wiggins and Ruefli, 2002, 2005). Finally, prior work in strategy and economics has shown that innovation timing is a major determinant of firms’ R&D costs due to the existence of time compression diseconomies—the fact that reducing R&D project duration often raises costs (Dierickx and Cool, 1989; Mansfield, 1971; Scherer, 1967, 1984). Empirical estimates of this acceleration-cost trade-off abound for industries as diverse as chemicals, electrical and instruments, machinery, and software development (Boehm, 1981; Mansfield, 1988; Putnam and Fitzsimmons, 1979). In short, timing matters because it critically influences the expected benefits and costs of innovation in most industries. Theoretical literature Innovation timing has also been extensively studied in the theoretical literature. The standard approach in most formal models has been to assume specific benefit and cost structures associated with innovation and then examine how the strategic interactions among competing firms endogenously determine innovation timing—and other related variables such as firm expenditures over time, the identity of the innovating firm, and the pace of innovation diffusion in an industry (see Reinganum, 1989, for an early review).2 2

Innovation theorists have been interested in timing games because they are strategically more nuanced than comparable quantity or pricing games of innovation. Depending on the structure of the game, firms’ timing decisions may be either strategic substitutes (like quantities in Cournot) or strategic complements (like prices in Bertrand). For example, if changes in innovation strategy are costless, the best response to rivals’ R&D acceleration may be to speed up as well. In this case, firms’ timing decisions are strategic complements and competition may lead to rent equalization (Fudenberg and Tirole, 1985). In contrast, with adjustment costs, firms that are precommitted to

Copyright  2011 John Wiley & Sons, Ltd.

Although prolific, the theoretical literature on R&D has predominantly focused on the timing of technology research—as opposed to technology development. The defining feature of research activities is that the outcomes associated with effort are uncertain as in, for example, the work on patent races. Specifically, in patent race models, time-to-market is highly stochastic, usually with a constant hazard rate (for a review, see Langinier and Moschini, 2002; see also Judd, 2003). However, in reality, after research results in a technological breakthrough, there usually remains a time-consuming and costly—but more certain—period of development before the new technology can be deployed into a market. For example, after the emergence of powerful new Internet technologies in the mid 1990s, economic activity was only transformed when innumerable incumbent and start-up firms completed costly and time-consuming development projects aimed at exploiting the new technologies in a variety of industries such as financial services, travel agencies, and general retail (see Mansfield, 1971, for early estimates of the development lag between invention and innovation in numerous industries). Although development activities can make up a significant fraction of the total time and costs associated with R&D (Mansfield, 1971; Scherer, 1984; Schumpeter, 1934), the literature has largely overlooked their importance. Our paper contributes to a formal theory of the timing of technology development. Our work on technology development is also related to, but distinct from, the literature on the timing of new technology adoption (Fudenberg and Tirole, 1985; Reinganum, 1981; Riordan, 1992; see Hoppe, 2002, for a review). We model technology development using two central features of adoption models. First, we incorporate a trade-off between time-to-market and the cost of acquiring a new technology. Second, we assume that acquiring the technology increases a certain innovation strategies may find it prohibitively expensive to respond to rivals’ R&D acceleration by innovating faster. Since being the first to market may no longer be possible, the focal firm may then prefer to slow down to save costs. In this situation, firms’ timing decisions are often strategic substitutes—and an asymmetric equilibrium with a leader and a follower firm may emerge, even when firms are symmetric ex ante (Reinganum, 1981; Ruiz-Aliseda and Zemsky, 2007). Much of the theoretical literature on innovation timing has revolved around these subtle but important issues that typically find no analog in quantity or pricing games. Strat. Mgmt. J. (2011) DOI: 10.1002/smj

G. Pacheco-de-Almeida and P. B. Zemsky firm’s flow of profits from the market and weakly decreases the profit flow of competitors. Despite these similarities, there is one fundamental difference between the literature on technology adoption and our work: while in the adoption literature firms purchase the new technology from upstream suppliers, in our model firms develop the new technology internally. Each firm engaging in its own time-consuming development process, rather than instantaneously purchasing the technology from a supplier, has important implications. There is now a question of how firms sequence their development activities. One possibility is that firms run concurrent development processes. Another possibility is that some firms take an imitative approach. The use of imitative strategies gives rise to the possibility of spillovers from technology leaders to followers, which is the focus of our paper.3

MOTIVATING EXAMPLES Our theory is well motivated by competition to develop microprocessors for use in personal computers (PC). Three main features make this industry particularly suited to our analysis. First, the development of each new microprocessor generation is time-consuming and costly. For example, the development of a state-of-the-art microprocessor, such as the 80 386 in 1985 or the Pentium II in 1997, took four years and cost several hundred million dollars (reportedly $200 million for 3

Katz and Shapiro (1987) provide an important step toward distinguishing development from adoption. In particular, they allow a follower firm to imitate a technology leader and benefit from spillovers. However, in other respects, their model retains important assumptions from the adoption literature, namely firms face a single, exogenously determined time-cost trade-off and firms instantaneously deploy a technology once they decide to develop it. In contrast to our theory, there is then no distinction between concurrent and imitative development and no heterogeneity in firm capabilities. The latter is an important feature of our model that is central to the field of strategy but rarely captured in formal theory (Sutton, 2005). This paper also builds on prior work by Pacheco-de-Almeida and Zemsky (2007), which uses an analogous micromodel of technology development. However, that paper is silent about (a) any comparisons of competitive dynamics across technology development regimes and (b) firm heterogeneity in time compression capabilities. Both research topics (a) and (b) constitute the focus of the current paper and produce the richness of our results. Overall, the contribution of this study lies not in the novelty of its analytical methodology (which also shares most of the formal features of the adoption literature), but in the new set of results it derives.

Copyright  2011 John Wiley & Sons, Ltd.

the 80 386 platform; cf. FTC Docket no. 9288 by Pitofsky et al., 1998, and Casadesus-Masanell, Yoffie, and Mattu, 2005). Second, the timing of innovation is critical to competitive advantage in this industry, with successive new product introductions doubling microprocessor performance every 18 months, as per Moore’s Law. Third, the PC microprocessor industry has effectively been a duopoly for much of its history: Intel has been the technology leader by typically being the first to introduce next-generation products to the market, whereas Advanced Micro Devices (AMD) has been a technology follower. The evolution of the microprocessor industry provides a rich context in which to illustrate the results of our theory (for a recent summary, please see Shih and Ofek, 2007). Over the years, the competitive dynamics between Intel and AMD have been governed by very distinct technology development regimes. At first, AMD’s strategy was explicitly one of imitation: it waited until Intel released its processors and then developed its own products based on Intel’s specifications. For the first two product generations, the 8086 and the 80 286, AMD had access to Intel’s IP and benefitted from high technological spillovers. Despite the agreement, AMD still faced time-consuming technology development; for example AMD was two years behind Intel to launch its version of the 80 286 in 1984. However, with the development of the 80 386 in 1985, Intel sought to reduce spillovers to AMD, largely by refusing to continue sharing its designs. An Intel lawyer would later claim ‘we don’t have any barriers for competitors, just a few speed bumps that people have to go around’ (CasadesusMasanell et al., 2005: 11). With the need to now reverse engineer Intel’s chips and with the increasing complexity of the technology, AMD was over five years behind Intel in bringing an 80 386 product to market. AMD proved more capable with the 80 486, but it was still four years behind Intel by the time it launched its version in 1993. In response to the reduced spillovers from Intel, AMD ultimately dropped its strategy of imitative development. Instead, AMD started pursuing a strategy of concurrent development where it designed its own next generation processors at the same time that Intel worked on its own processors. While AMD succeeded in bringing out its own next generation designs, including the K5 and K6, Intel retained Strat. Mgmt. J. (2011) DOI: 10.1002/smj

Innovators’ Strategic Use of Disclosure to Slow down Competition its position as the technology leader.4 Nonetheless, it soon became clear that AMD’s shift in strategy from imitative to concurrent development was having a very significant adverse effect on Intel’s profitability (see, for example, ‘The Monkey and the Gorilla’ in The Economist, 5 December, 1998: 71–72). In summary, the Intel-AMD case illustrates the shortcomings of the classical strategy view that innovators should always minimize spillovers to followers. Specifically, Intel made a mistake by reducing spillovers: it forced AMD to ramp up its capabilities and start racing. In contrast, by freely revealing part of its IP pre-1985, Intel kept AMD on an imitative stand. Thus, innovators may better appropriate the returns from innovation by partially sharing it with competitors—as long as, by doing so, innovation diffusion is successfully delayed. Several other examples of free revealing to discourage competitive innovation have also been documented in the business press. In January 2009, IBM ‘pledged to open up more of its inventions than ever before to the public’ (Vance, 2009). IBM, the patent kingpin, committed to freely reveal approximately 3,000 ‘important and innovative’ inventions per year via technical papers and the Internet. This drastic change in the company’s usually aggressive patent strategy has been received by industry experts as a way to prevent concurrent competitive innovation and to encourage imitative technology development. Another example is the ‘Big-Three’ automakers’ formal agreement to freely share pollution control technologies in 1971 (White, 1971). Given General Motors’ strong industry leadership throughout the 1970s (Waring, 1996), this agreement has recently been interpreted as an explicit attempt to diffuse rivalry by giving competitors strong incentives to imitate (Lieberman and Asaba, 2006).5 4

After its shift to concurrent development, AMD has at times sought to challenge Intel’s role as technology leader. See, for example, ‘AMD to Intel: Let’s Rumble’ in BusinessWeek (Hesseldahl, 2005). However, Intel has remained the dominant firm with a market capitalization that has consistently remained significantly greater than that of AMD. 5 ‘In environments where change is more incremental, imitation can defuse rivalry and reduce risk for any given firm. Knowledge that rivals will respond in kind lowers the incentive for any individual firm to act aggressively in an effort to gain competitive advantage. (. . .) One example is the so called ‘smog case’ where the ‘Big-Three’ automakers agreed to share pollution control technology adopted by any one of the firms’ (Lieberman and Asaba, 2006: 367). The agreement between the three car companies eventually gave no incentives also to General Motors Copyright  2011 John Wiley & Sons, Ltd.

Innovators may also have incentives to allow free spillovers in other industries where new product development is costly, time-consuming, and occurs on a regular basis with low levels of technological uncertainty. Markets that exhibit these features include digital cameras, MP3 players, and cell phones. In these settings, both concurrent technology development by multiple firms and reverse engineering of innovators’ products (e.g., Sony, Apple, Samsung) by imitators (e.g., Matsushita, some Chinese companies) are common.

THE MODEL In nontechnical terms, our model can be described as a stylized representation of the development dynamics of a new technology by two competing firms. Internal technology development by firms is subject to time compression diseconomies, whereby faster development is only achieved at higher costs. Firms’ development processes may run concurrently—firms make simultaneous investments in a new technology—or sequentially—one firm waits to start its development until the technology leader has deployed the technology in the market. Sequential development gives rise to the possibility of imitation and interfirm knowledge spillovers, which is central to our analysis. Consistent with the motivation for our paper, we assume that there are no mechanisms that compensate innovators for interfirm spillovers or the free revealing of proprietary knowledge. That is, the development of the same innovation by the follower brings no benefits to the leader and simply results in stiffer competition. In our model, firms are homogeneous except along two dimensions. First, firms may differ in their time compression capabilities. We model such capabilities as shifting a firm’s time-cost trade-off (i.e., for a fixed level of expenditures, a more capable firm develops faster). Second, we assume that one firm has a preexisting reputation as a technology leader while the other firm has a reputation as a follower. These reputations determine the roles firms play. Endogeneizing these roles is certainly of interest, but lies beyond the scope of this paper. Next, we introduce the technical assumptions and the two technology competitive regimes in to develop the needed pollution control technologies and the contract was ultimately deemed anticompetitive by U.S. antitrust authorities. Strat. Mgmt. J. (2011) DOI: 10.1002/smj

G. Pacheco-de-Almeida and P. B. Zemsky our model. Nontechnical readers may skip this section and proceed directly to our first results on ‘Optimal time-to-market.’ Our main findings on ‘Free revealing and the limits to imitation’ appear in our section ‘Knowledge Spillovers and Firm Performance.’ All proofs are in Appendix C. Main assumptions The model is in continuous time indexed by t ≥ 0. We index the two firms that compete to develop the new technology by i = 1, 2. We denote the time at which firm i completes development of the new technology by Ti ≥ 0, which we refer to as firm i’s time-to-market. We focus on equilibria in which the firms have clear expectations as to the order in which firms complete development. As a convention, we can then take T1 ≤ T2 so that firm 1 is the technology leader (if there is one) and firm 2 is the follower. Firm profits are given by the present value of their revenues from the product market net of the present value of their technology development costs for a common discount rate (or cost of capital) r. The present value of revenues for the leader and follower are shown as Expressions 1 and 2:
T1
T2 π00 e−rt dt + π10 e−rt dt R1 (T1 , T2 ) =


+

0 ∞

π11 e−rt dt

T2




T1

R2 (T1 , T2 ) =


+

T1

π00 e

(1) −rt




dt +

0 ∞

π11 e

−rt

T2

(2)

T2

We assume that both firms earn revenue flows of π00 when neither has the new technology (t < T1 ). When both firms have completed technology development (t > T2 ) their revenues are π11 . If only one firm has completed development (T1 < t < T2 ), its revenues are π10 , whereas the firm without the technology earns revenues of π01 . Developing the new technology enhances a firm’s revenues while decreasing those of its competitor, such that π10 > π11 > π00 ≥ π01 ≥ 0. Thus, technology development by a follower yields no benefits to an innovator (its profits always decrease)—and there are no compensating mechanisms for free revealing. New markets are a special Copyright  2011 John Wiley & Sons, Ltd.

C(T ; Zi ) =

rZi2 erT − 1

(3)

This cost function shown in Expression 3 is derived from a micromodel of technology development where firms exert effort over time to bring the new technology to fruition and there are decreasing returns to effort at a point of time (see Appendix A). The level of effort may be interpreted as the number of engineers (and other resources) allocated to technology development. There are decreasing returns to effort, for example, if engineers have to spend an increasing fraction of their time coordinating their activities as the size of the development team grows. The assumption of decreasing returns is key because it gives rise to time compression diseconomies (Dierickx and Cool, 1989; Mansfield, 1971; Scherer, 1967; Teece, 1977), such that costs are a decreasing, convex function of development time T . Costs are also an increasing, convex function of the effort Zi > 0 required to develop the technology.8 Concurrent development

π01 e−rt dt

T1

dt

case of our model where π00 = π01 = 0. Let 1 ≡ π10 − π00 and 2 ≡ π11 − π01 be the increase in revenues for the first firm and the second firm to market, respectively. As is common in the literature, we assume that 1 > 2 (Fudenberg and Tirole, 1985; Reinganum, 1981).6 In our model, firms’ technology development costs are given by the cost function7

Consistent with the introduction to the paper and the motivating examples, we consider two regimes for the timing of firms’ development activities: concurrent vs. imitative. 6

In the new market case, where the new technology is required to have any revenue flows, this restriction is just a requirement that the profits of a monopolist (π10 ) are greater than the profits of a duopolist (π11 ). 7 In earlier versions of the paper, we explored alternative cost functions. A model with a general functional form for costs proved intractable to establish comparisons between firms or across development regimes. Assuming a specific cost function without an explicit technology development process (such as C(T ) = C0 + φKe−T /φ ), while tractable, also had several drawbacks: it required numerous restrictions on the parameter values to rule out corner solutions, it was independent from the cost of capital, and it lacked microfoundations. 8 Faster development may also reduce quality (Cohen, Eliashberg, and Ho, 1996). We assume quality constant for tractability reasons. Strat. Mgmt. J. (2011) DOI: 10.1002/smj

Innovators’ Strategic Use of Disclosure to Slow down Competition With concurrent development, both firms start their development activities at the same time, t = 0 and there are no significant interfirm spillovers. The total effort levels required for each firm to develop the technology in this game are given by Z1 = (1 − d1 )K and Z2 = (1 − d2 )K, where K > 0 reflects the underlying complexity of the technology. The parameters d1 and d2 , di ∈ [0, 1), capture each firm’s time compression capabilities. Firm heterogeneity in capabilities can reflect differences in a general ability to develop new technologies or prior accumulated experience. Importantly, note that with concurrent development there can still be a technology leader if one firm commits greater resources to development and, thereby, completes its technology development first. To assure that both firms have an incentive to √ develop the new technology, we assume that i > rZi for i = 1, 2 (A1). While we can allow for the possibility that the follower is more capable than the leader (i.e., d2 > d1 ), we need to limit the extent of this asymmetry—otherwise, it is not possible to support an equilibrium where firm 1√develops first.√Thus, we assume that (1 − d2 ) 1 > (1 − d1 ) 2 (A2). As 1 > 2 , any d1 ≥ d2 satisfies inequality (A2). Also, the larger 1 is compared to2 , the more can d2 exceed d1 . Imitative development In the imitative regime, firms develop sequentially: the leader first finishes its own development process (i.e., invests from t = 0 until t = T1 ) and then the follower starts investing (from t = T1 until t = T2 ). This regime would arise, for example, if firm 1 were uniquely aware of the opportunity to develop the new technology and firm 2 only became aware of it when firm 1 deployed the technology to the market. Alternatively, imitative development could also be a choice by firm 2 as discussed in the ‘Free revealing and the limits to imitation’ subsection of our ‘Knowledge Spillovers and Firm Performance’ section. Although imitation delays firm 2’s development, it may allow the firm to benefit from the leader’s spillovers, which reduces (but not eliminates) the total effort Z2 required to develop the technology. We assume that Z1 = (1 − d1 )K and Z2I = (1 − s)(1 − d2 )K. The parameters di ∈ [0, 1) and K are as in the concurrent regime. The parameter s ∈ [0, 1) reflects the extent to which the follower benefits from spillovers under imitative development. Copyright  2011 John Wiley & Sons, Ltd.

One would generally expect s to be higher for product innovations, which tend to be visible to customers and competitors and can be reverse engineered, than for process innovations, which tend to be less visible externally. The leader’s effort Z1 is the same across both development regimes. Given the sequential structure of imitative development, it is natural to solve the imitative regime for subgame perfect equilibria. Finally, condition (A1) is not sufficient to assure that the leader develops in this game. A sufficient condition √ to assure that firm 1 develops is π11 − π00 > rZ1 (A3). The follower finds it profitable to develop iff √ 2 > rZ2I but this is implied by (A1) as Z2 ≥ Z2I . We assume that (A1)–(A3) hold throughout the paper.

OPTIMAL TIME-TO-MARKET We now solve for the equilibrium time-to-market in the two development regimes. Firms choose their time-to-market so as to maximize the net present value of their revenues from the product market net of their technology development costs. We define each firm’s regime-specific profit functions below. For concurrent development, we have that firm payoffs are given by revenues minus costs, Ci = Ri (T1C , T2C ) − C(TiC ; Zi ) for i = 1, 2. Here, both firms formulate their technology development strategies simultaneously at time t = 0. As the choice of investment profile over time ci (t) (in Appendix A) has a one-to-one mapping to development times, this is equivalent to a simultaneous choice of concurrent development times (i.e., the time at which firms will launch the new technology), TiC (i = 1, 2). Notice that ∂Ci /∂TiC = C − 1r i e−rTi − ∂C(TiC ; Zi )/∂TiC = 0 is the firstorder condition for optimum for firm i, which has two implications. First, once the order of completion is fixed (and hence the value of i , or the increase in revenues, determined), there are no remaining strategic interactions in the timing choices of the two firms. Second, with 1 > 2 , firm 1 has greater incentive to compress its development process since it expects to be first to market in equilibrium. Firm 1’s expectation that it will be first to complete is self-fulfilling as long as firm 2 does not have a sufficiently greater development capability, which is assured by Assumption (A2). Assumption (A1) assures that Strat. Mgmt. J. (2011) DOI: 10.1002/smj

G. Pacheco-de-Almeida and P. B. Zemsky the equilibrium development times are finite and given by the solution to the two first-order conditions. We have the following result.9 Proposition 1: In concurrent development, the equilibrium time-to-market for firm i is  −1 r Z 1 C i Ti = r ln 1 −  . Firm 1 is the first i to develop the new technology, T1C < T2C . Proposition 1 shows simple and identical closedform expressions for the equilibrium development times of both firms. The comparative statics on the equilibrium time-to-market, or how the exogenous variables in our model affect development times, are trivial given that firms’ development effort is a linear and monotonic function of firms’ capabilities (di ) and the complexity of the technology (K), Zi = (1 − di )K (please see Appendix B for a summary of the comparative statics results). For technical readers interested in the fundamentals of this equilibrium solution, a brief explanation for our results in Proposition 1 follows. Since the work of Fudenberg and Tirole (1985), it is standard in the technology adoption literature to solve a parallel timing game for closed-loop equilibria where firms can costlessly vary their strategies over time.10 This is essentially a requirement of subgame perfection. However, Ruiz-Aliseda and Zemsky (2007) show that open-loop equilibria in our concurrent development regime can satisfy subgame perfection and that the rent-equalizing equilibria identified by Fudenberg and Tirole (1985) break down. Two conditions are required. First, while completion is observable, a firm’s cumulative effort is not. Second, the follower’s incentive to develop the technology cannot be too much lower than the leader’s.11 The key difference with adoption models is that, with internal technology 9

Of course there could also be an equilibrium where firm 2 is the leader and firm 1 is the follower. This√occurs as long √ as an analogue of (A2) holds, namely (1 − d1 ) 1 > (1 − d2 ) 2 . However, by convention we are focusing on the case where firm 1 is the leader. 10 The adoption literature had originally focused on open-loop equilibria where firms committed to development times at the start of the game as in Reinganum (1981) and Quirmbach (1986). Reinganum (1989) had justified the use of precommitment equilibria by evoking the costs of a firm changing its technology strategy. 11 With symmetric capabilities (d1 = d2 ), the outcome T1C , T2C is supported as a subgame perfect equilibirum as long as 1 < 6.852 , which is a fairly weak condition (Ruiz-Aliseda and Zemsky, 2007). Copyright  2011 John Wiley & Sons, Ltd.

development, firms are incurring expenses continuously over time. The instantaneous acquisition of the technology that is required to support the rent equalization of adoption models is prohibitively expensive in our setting where development is a time-consuming activity. Therefore, we simply assume that firms are playing an open-loop equilibria in that they commit to a ci (t) expenditure profile at time t = 0 (see Appendix A) and we then solve for pure strategy Nash equilibria. Hence, in our imitative development regime, the order of moves is taken as fixed and the leader’s optimal development time is not affected by concerns of preemption. We now present the equilibrium time-to-market in imitative development. Let TiI be the timeto-market of firm i in the imitative development regime. It is also useful to define Tˆ2I as the amount of time that the follower spends on development. Thus, the leader is developing the technology during the interval [0, T1I ] and the follower is developing during the interval [T1I , T2I ], where T2I = T1I + Tˆ2I —that is, the total time-tomarket of the follower is the sum of the development times of the leader and the follower. Payoffs in this regime are also given by revenues minus costs, I1 = R1 (T1I , T2I ) − C(T1I ; Z1 ) I and I2 = R2 (T1I , T2I ) − e−rT1 C(Tˆ2I ; Z2I ). The discounting of costs C(Tˆ2I ; Z2I ) for the follower firm arises because the cost function C(T ; Z) gives the present value of development costs at the start of development, which is T1I for the follower (i.e., in the imitative regime the follower only starts development after the leader launches the technology), while firm payoffs are discounted back to time t = 0. As this game has a sequential structure, we work backwards, characterizing first the optimal time spent on development for the follower. The follower’s optimal time spent on development Tˆ2I is a constant independent of how long the leader takes. Thus, we then turn to the leader’s problem. We collect these results in the following proposition. Proposition 2: In imitative development, the leader firm’s equilibrium time-to-market is T1I =   −1

Z1 r  , 1 ln 1 −  where I1 = 1 r I 1 √ −(π10 − π11 )(1 − Z2I r/ 2 ). The equilibrium time spent on development by the follower firm Strat. Mgmt. J. (2011) DOI: 10.1002/smj

Innovators’ Strategic Use of Disclosure to Slow down Competition  −1 I r Z 1 I 2 ˆ is T2 = r ln 1 −  and the total time2 to-market of the follower is T2I = T1I + Tˆ2I .

Proposition 2 presents, again, simple closedform expressions for the equilibrium development times of both firms in the imitative regime. As in the concurrent regime, most comparative statics are trivial (see Appendix B). Note that I1 > π11 − π00 and hence (A3) assures that T1I is the unique, optimal development time for the leader.

TECHNOLOGY DIFFUSION Do firms innovate faster in concurrent or imitative development? This section addresses this question, or how imitation—and the associated spillovers—affect the speed with which a new technology diffuses into a given market. Technology diffusion, a classic topic in the innovation literature, is characterized by comparing firms’ total time-to-market in both development regimes. For the leader, we get the unambiguous result described below. Proposition 3: The leader is faster to market under concurrent development than under imitative development, T1C < T1I . The intuition behind Proposition 3 is as follows. Under concurrent development, the time-to-market of the follower is fixed at T2C and is independent of the speed of the leader. Hence, the incentive for the leader to devote resources to accelerating its development process comes purely from reducing the lower bound on the interval of time [T1C , T2C ] over which the leader enjoys a competitive advantage from exclusive use of the technology. In contrast, under imitation, the time-to-market of the follower falls with the time-to-market of the leader because the follower starts technology development immediately after the leader launches the new technology. This reduces the incentive for the leader to accelerate development since a faster time-tomarket shifts forward but does not lengthen the period of advantage (i.e., decreases in T1I only shift the interval of advantage under imitation [T1I , T1I + Tˆ2I ] earlier). Formally, the result arises from the fact that 1 > I1 , with 1 capturing the leader’s return to acceleration under the concurrent Copyright  2011 John Wiley & Sons, Ltd.

regime and I1 capturing the leader’s returns under the imitative scenario. For the follower firm, the effect of imitation and spillovers on time-to-market is more complicated. A key driver of the follower’s time-to-market is the time it spends on development. Under both imitative and concurrent development, the returns to the follower from deploying the new technology are 2 . When there are spillovers from the leader firm (s > 0), the development problem is easier under imitation (Z2I < Z2 ). Hence, it is straightforward to see that the follower spends less time developing the technology in the imitative than in the concurrent regime, Tˆ2I < T2C . Although the follower develops faster under imitation if there are spillovers, the firm may still take longer to introduce the new technology to the market. This is because, with imitation, the follower only starts investing after the leader has completed its own development process. In other words, in the imitative regime, the follower’s total time-tomarket also varies with the leader’s time-to-market and, thus, we do not have as unambiguous a result as in Proposition 3. The possible nonmonotonicities in the net leader-follower effect are illustrated by considering an example, which we repeatedly use throughout the paper. Example 1: The parameters governing

technology development are d1 = d2 = 1 2 and K = 19. Revenue flows are {π10 , π11 , π00 , π01 } = {4, 3, 2, 1.9}. The cost of capital is r = 0.1. Figure 1 shows the time-to-market for the leader and the follower as a function of spillovers for Example 1. Dashed lines give time-to-market under concurrent development, while the solid lines show time-to-market under imitation. As established in Proposition 3, the leader’s time-to-market is always lower for concurrent than imitative development (T1C < T1I ). The greater the level of spillovers, the longer the leader takes to bring the new technology to market with imitation. This is because the follower catches up faster with more spillovers, which gives the leader increasingly fewer incentives to quickly develop the technology and, thus, its total time-to-market increases. Interestingly, for sufficiently small spillovers (i.e., s close to 0), the follower is faster to market with concurrent development. This is because Strat. Mgmt. J. (2011) DOI: 10.1002/smj

G. Pacheco-de-Almeida and P. B. Zemsky Follower in imitative development

TIME TO MARKET

We collect the general results on when the follower firm is faster to market under concurrent development in part (i) of Proposition 4 below. Part (ii) of the same proposition then identifies conditions under which the follower is faster to market when pursuing imitation strategies. The intuition behind our results is also summarized below.

Follower in concurrent development

30

20 Leader in imitative development 10 Leader in concurrent development 0 0

0.5

1

SPILLOVERS (S)

Figure 1. The time-to-market of leader and follower firms under imitative and concurrent development (Example 1)

imitation delays the start of the follower’s development process and, as spillovers go to zero, the follower is spending as much time on development whether its approach is imitative or concurrent. We show below that this is a general result. Thus, for small spillovers, imitation unambiguously leads to slower diffusion of the technology with both T1C < T1I and T2C < T2I . As spillovers increase, the follower’s time spent on development with imitation, which is given by the gap between T2I and T1I , falls. However, the follower’s start time T1I increases as the leader gets demotivated by the more rapid imitation. In the example, the net effect of increases in spillovers is sufficiently negative for low values of s that there is a reversal: the follower is faster to market under imitation than concurrent development. Although this need not always happen, it establishes that there is no general ranking of T2C and T2I . What happens to the follower’s time-to-market as spillovers become large? In the example, there is a nonmonotonicity and for sufficiently high spillovers the follower is, somewhat surprisingly, again faster to market with a concurrent strategy. As spillovers increase toward the upper limit of 1, the follower’s development problem becomes trivial (i.e., Z2I → 0) so that its time-to-market is just the leader’s development time. The effect of imitation then comes down to whether the follower develops faster under concurrent development than the leader develops under imitative development.12  √ We have that T2C < T1I iff Z2 / 2 < Z1 / I1 and the ranking of time-to-market depends on the incentives to accelerate development relative to capabilities. We have that lims→1 I1 = 12

Copyright  2011 John Wiley & Sons, Ltd.

Proposition 4: (i) The follower is faster to market under concurrent development, T2C < T2I , if either s is sufficiently close to 0 or if s is sufficiently close to 1 when d1 ≤ d2 and π00 > π01 . (ii) For s > 0, the follower is faster to market under imitative development, T2I < T2C√, if either d1 is sufficiently close to 1 or if 2 − rZ2 goes to zero (as long as I1 − Z1 r does not go to zero as well). Intuitively, both parts (i and ii) of Proposition 4 state that the follower is faster to market with imitation when its development problem is much more challenging than the leader’s development problem. This is because, in such circumstances, the main effect of imitation is to reduce the follower’s development time—as demotivation of the leader is relatively less important. The follower’s development problem can be more difficult than that of the leader for several reasons. For example, as the leader’s capabilities improve significantly, d1 → 1, the leader’s development problem√ becomes trivial. Conversely, as Z2 gets close to 2 /r, the follower’s development time under concurrent development goes to infinity because development of the new technology (as a follower without any spillovers) is approaching the point at which it is not economical. As long as the leader is not similarly demotivated in this limit, the follower will be faster to market under imitation. In business strategy, the topic of technology diffusion in an industry is also directly related to the central notion of sustainable competitive advantage (Besanko et al., 2007; Porter, 1985). One can define competitive advantage as an asymmetry among competing firms that allows one firm to outcompete another firm in product markets. Our model is well suited to studying this phenomenon. √ √ π11 − π00 , which is greater than 2 = π11 − π01 if π00 > C I π01 . Then T2 < T1 in the limit as s → 1 as long as Z2 is not too much bigger than Z1 (or equivalently d2 not too much smaller than d1 ). In the example we have d1 = d2 and π00 > π01 . Hence, for s sufficiently large we have that the follower is faster to market under concurrent development. Strat. Mgmt. J. (2011) DOI: 10.1002/smj

Innovators’ Strategic Use of Disclosure to Slow down Competition

Proposition 5: There exists a critical level of spillovers s¯ ∈ (0, 1) such that the leader’s competitive advantage is more sustainable under imitative than concurrent development (i.e., T2C − T1C < T2I − T1I ) if and only if s < s¯ . As spillovers become large, the effort required to develop the technology under imitation becomes small so that the leader’s competitive advantage is more sustainable when the follower uses a strategy of concurrent development.13 In contrast, for small s the main effect of imitation is only to delay the start of the follower’s development activities and, hence, sustainability is greater when the follower is imitating. Although the business strategy literature often treats sustainable competitive advantage as an end in itself, presumably the concept is of interest to the extent to which it is correlated with superior financial performance. It is straightforward to explore the links between technologybased advantages and performance in our model because there are closed-form expressions not only for each firm’s time-to-market, but also for their profits.

13 Note that it is Assumption (A3) that assures that the leader wants to develop the technology even in the limit as s → 1.

Copyright  2011 John Wiley & Sons, Ltd.

Leader-follower profit differential in concurrent development ( )

2

PROFIT DIFFERENTIAL

Our firms are asymmetric in their product market positions precisely when one firm has developed the new technology and the other has not. Thus, we say that the leader in our model has a competitive advantage at a point in time when it has developed the new technology but the follower has not. One can define sustainability as the extent to which the technology asymmetry persists over time. In our model, sustainability is then the difference in the time-to-market of the two firms, T2 − T1 . With both imitation and concurrent development, firms have different time-to-market and, hence, in both cases there is a leader that enjoys a period of competitive advantage. What is the effect of imitation on the sustainability of the leader’s competitive advantage? We find that there is no clear ranking and that spillovers are a key moderating variable.

1 Leader-follower profit differential in ) imitative development ( 0 0

-1

0.5

1

SPILLOVERS (S)

Figure 2. The difference betweeen the leader’s profit and the follower’s profit under imitative and concurrent development (Example 1)

KNOWLEDGE SPILLOVERS AND FIRM PERFORMANCE We now characterize the effect of imitation and spillovers on the profits of technology leaders and followers. We first consider the relative performance of the leader and the follower, or withinindustry firm performance heterogeneity. If the follower is more capable than the leader (i.e., d2 > d1 ) then this alone could drive superior performance by the follower. Thus, we assume that the leader is at least as capable as the follower (d1 ≥ d2 ). Proposition 6: Suppose that d1 ≥ d2 . (i) Under concurrent development, the leader always has a higher payoff than the follower. (ii) Under imitative development, the leader has a higher payoff if and only if spillovers are not too great: there exists a critical level of spillovers sˆ ∈ (0, 1) such that I1 − I2 > 0 iff s < sˆ . Figure 2 below illustrates the results in Proposition 6 for Example 1. For concurrent development, the leader’s competitive advantage unambiguously leads to superior performance (as long as there are not offsetting capability differences): C1 > C2 . The argument is as follows. Were the leader to deviate and develop at the same time as the follower, then both firms would have the same profits with equal capabilities (d1 = d2 )—thus, the leader would have even higher profits than the follower if it had superior capabilities (d1 > d2 ). The decision to use an earlier development time must raise the leader’s profit, while the follower’s payoff is weakly decreasing as the leader develops earlier. Strat. Mgmt. J. (2011) DOI: 10.1002/smj

G. Pacheco-de-Almeida and P. B. Zemsky In contrast, for imitative development, the leader’s period of competitive advantage need not lead to superior performance. In particular, as spillovers become large, the follower acquires the technology almost immediately after the leader and at a fraction of the development cost. Hence, the follower may well have a higher payoff. Thus, there is only a clear link between the leader’s period of competitive advantage and superior performance in the case of concurrent development. We now consider how each firm’s absolute profit level is affected by whether development is concurrent or imitative. We start with the leader. Proposition 7: There exists a critical level of spillovers s¯1 ∈ (0, 1) such that the leader prefers that the follower uses an imitative strategy if and only if s < s¯1 . The leader prefers imitation as long as spillovers are sufficiently small. On the one hand, imitative development has the advantage that delays to the leader’s development also delay the follower (Proposition 3). On the other hand, for large spillovers sustainability is reduced by imitation (Proposition 5). Figure 3, which show the profits of each firm for Example 1, illustrates how the leader’s profits under imitation I1 fall below those under concurrent C1 beyond the critical spillovers s¯1 . Consider the implications of Proposition 7 for the strategy of Intel. Once Intel decided to reduce the spillovers to AMD in 1985 by refusing to share its designs, the proposition suggests that Intel would have benefitted from shifting AMD from

FIRM PROFITS

22

Leader in imitative development

Leader in concurrent development

21 Follower in imitative development Follower in concurrent development

20

0

0.5

1

SPILLOVERS (S)

Figure 3. The profit to the leader and the follower under imitative and concurrent development (Example 1) Copyright  2011 John Wiley & Sons, Ltd.

the concurrent development strategy it started pursuing back to an imitative strategy. Interestingly, Intel did seek to force AMD back to an imitative strategy post-1985 by changing the interfaces for its microprocessors so that AMD could not develop compatible designs until after Intel had released its chips to the market (Shih and Ofek, 2007). In other words, post-1985 AMD had incentives to wait, not because it expected to receive voluntary spillovers from Intel, but because it needed to observe Intel’s designs to then develop a compatible technology. For Intel, this scenario where AMD waited and still had no access to its IP was the most favorable competitive regime possible. Yet instead of waiting, AMD decided to ramp up its technical capabilities and start racing. Is AMD’s decision to shift to a concurrent development strategy in response to the fall in spillovers consistent with our theory? Proposition 8 presents our results for the follower, which we discuss below. Proposition 8: (i) There exists a critical level of spillovers s¯2 ∈ [0, 1) such that the follower prefers imitative development unless s < s¯2 . More (ii) In the new market case √ s¯2 > 0. √ generally, s¯2 > 0 if either 2 > rZ2 ( 5 + 3)/2 or π00 sufficiently close to π01 . We find that for sufficiently high spillovers, the follower prefers imitation. Otherwise, if interfirm spillovers are limited, the follower firm is better off developing the technology concurrently with the leader. This was likely the case of AMD in 1985 when Intel purposefully reduced technology spillovers by refusing to share its designs of the 80 386 microprocessor (and subsequent generations). Indeed, in response to the decline in spillovers, AMD decided to switch from an imitative development regime to concurrent development. AMD’s response was rendered possible by the fact that Intel’s previous strategy of changing chip designs to force AMD to wait before developing compatible technologies could not be endlessly repeated. For the specific parameter values in Example 1 and illustrated in Figure 3, the follower prefers imitation for s > s¯2 ∼ = 0.058.14 14 Interestingly, it is possible that s¯2 = 0 so that the follower prefers imitation for all levels of spillovers. Even without spillovers, the follower has a benefit from imitation, namely that the leader slows down, which delays the onset of the follower’s competitive disadvantage (although this does not arise in the new market case).

Strat. Mgmt. J. (2011) DOI: 10.1002/smj

Innovators’ Strategic Use of Disclosure to Slow down Competition Figure 3 also shows that the follower’s profits with imitation are likely nonmonotonic in s due to the leader’s demotivation that results from being imitated faster with higher spillovers. This is a key observation, which we revisit in the subsection on the limits to imitation below. Now, we simply note that the proof of Proposition 8 establishes that I2 crosses C2 at most once and only from below so that the critical value s¯2 is well defined.

Free revealing and the limits to imitation Up until now we have focused on firms’ choice of time-to-market and the effect on profits. Next, we turn to other firm decisions, especially those affecting the level of spillovers. The standard intuition in the classical view of interfirm spillovers is that knowledge outflows are disadvantageous for the leader and hence the firm should consider a variety of actions to decrease them (Arrow, 1962; Cohen et al., 2000; Harhoff et al., 2003; Levin et al., 1987; Mansfield, 1985, 1986; Spence, 1984). The leader could, for example, seek to reduce the outflow of knowledge about technology development by locating its facilities far from competitors, by taking actions to reduce the turnover of key employees, and by having exclusive contracts with suppliers (to keep information from leaking to competitors via third parties). Conversely, spillovers are generally taken to be beneficial to followers who are seeking to imitate and, thus, they should consider a variety of actions to increase them. Followers could, for example, increase spillovers by building capabilities in reverse engineering, by geographically collocating with the leader, and by aligning strategy with that of the leader firm to increase absorptive capacity and facilitate interorganizational learning (Cohen and Levinthal, 1989, 1990). But to what extent does our model support this classical view of interfirm spillovers? For ease of exposition, we start by discussing the followers’ incentives to limit imitation in Proposition 9 and, then, examine innovators’ free revealing problem in Proposition 10. Proposition 9: (The Limits to Imitation) Suppose there is imitative development. The profits of the follower are optimized for √ spillovers s2∗ < 1 if Z1 r > π11 − π00 Copyright  2011 John Wiley & Sons, Ltd.





 

 4 π10 − π01 . When the follower is com3+ π11 − π01 mitted to imitation, the leader’s profits are maximized for s = 0.

Under imitative development, we find that spillovers are indeed harmful for the leader. This is because spillovers reduce the sustainability of the leader’s competitive advantage.15 In contrast, it is possible for spillovers to be too much of a good thing for the follower: for some parameter values (such as those in Example 1) the follower would not want to maximize spillovers (i.e., push spillovers all the way to s = 1). The greater the spillovers, the faster the leader expects to be imitated and the less its incentives to compress its development process. Since the follower only starts its development once the leader is finished, the total time-to-market for the follower can potentially increase in spillovers (even as the follower’s own development process is unambiguously getting shorter). Not only may the follower’s timeto-market be lengthening, but even factoring in the cost savings from higher spillovers, we have shown that the follower’s profits are potentially falling.16 This finding implies that imitators may sometimes benefit from reductions in their absorptive capacity, which counters most prior theorizing in the strategy literature. In other words, Proposition 9 illustrates the limits to imitation strategies. In the context of AMD-Intel’s duopoly, this result suggests that when AMD was engaging in imitative development prior to 1985, it could have been in its best interest to only have incomplete knowledge about Intel’s in-house technologies. More generally, follower firms may pursue a number of different strategies to credibly commit to limit imitation. First, to the extent that geographical collocation allows firms to benefit from agglomeration economies and interfirm spillovers (Fujita et al., 1999; Krugman, 1991; Marshall, 1920), followers may choose to operate in different geographical sites than industry leaders. Since location decisions typically involve large 15 Pacheco-de-Almeida and Zemsky (2007) show this result for the case of absolute firm symmetry (i.e., d1 = d2 ). √ 16 Because π11 − π00 > Z1 r by (A3), for condition Z1 r > √ 10 − π01 π11 − π00 (4/(3 + π π11 − π01 )) to hold, it must be that π10 is large relative to π11 . The difference π10 − π11 determines the extent to which the leader is demotivated by speedy imitation.

Strat. Mgmt. J. (2011) DOI: 10.1002/smj

G. Pacheco-de-Almeida and P. B. Zemsky irreversible investments (e.g., in productive facilities), this strategy would be a clear commitment to curb spillovers. Second, follower firms may lower their levels of absorptive capacity, which credibly reduces firms’ ability to benefit from interorganizational learning (Cohen and Levinthal, 1989, 1990). For example, followers may choose not to align strategy and organization with that of the leader, avoid hiring the leader’s employees, reduce investments in complementary R&D or in capabilities to reverse engineer the leaders’ products. Followers’ consistent behavior over time along these dimensions will be indicative of their long-term commitment to curb imitation. Finally, the follower may publicly commit to develop its own technology without tapping the leader’s IP. The credibility of such announcements would ultimately rely on the strength of reputational (or legal) penalties in case of infringement. Next, we examine the innovators’ problem and the incentives for free revealing, which we consider an even more surprising result. Is there really no downside to a technology leader that seeks to minimize spillovers, as suggested by the classical view? Consider Figure 3. As spillovers decrease, the follower’s profits under imitation fall to the point where it prefers concurrent development. Thus, if the follower can choose its development strategy based on the level of spillovers, then the leader reducing spillovers to zero will trigger a switch to a concurrent development strategy by the follower. Because the leader prefers that the follower imitates, this introduces a downside to reduced spillovers. Hence, it is possible that the leader’s profits are optimized by allowing the follower just enough spillovers that it is willing to engage in imitative development. Let s1∗ be the optimal level of spillovers for the leader when the follower can choose between concurrent and imitative development.17 In Figure 3, s1∗ = s¯2 > 0 gives this optimum. Although it is possible that s1∗ > 0, this need not be the case. First, it is possible that the follower prefers imitation for all levels of spillovers (Proposition 8) and, second, 17 There are different ways to model the endogeneity of both spillovers and the follower’s development strategy. We are focusing on a particularly tractable approach, where the leader chooses spillovers and then the follower chooses its development strategy, as a way to establish the possibility that a leader can benefit by increasing spillovers. We assume that when the follower is indifferent between modes of development that it chooses imitation, as this makes results easier to state.

Copyright  2011 John Wiley & Sons, Ltd.

it must be the case that the leader still prefers imitation at the level of spillovers where the follower is willing to switch its strategy. Thus, we have the following result. Proposition 10: (Free Revealing) Suppose that the leader commits to a level of spillovers s ∈ [0, 1) and, only then, the follower commits to a concurrent or imitative development strategy. For any parameter values r, d1 ≤ d2 and π11 > π00 , there exist revenue flows π01 and π10 such that the the optimal level of spillovers for the leader is s1∗ = s¯2 > 0. Therefore, a technology leader may be better off when some of its technical know-how freely flows to rivals because this may dissuade them from concurrently developing new technologies, which alleviates competitive pressure. Intuitively, the leader discloses partial information because that is the least bad option when what is at stake is not if, but when competitors develop the same technology. Free revealing may delay competitive innovation. For example, leader firms may prefer to publicly disclose, rather than conceal, proprietary technology after the completion of a development project; they may favor patenting in countries with weak IP protection systems, or decide to geographically collocate with industry followers to increase interfirm spillovers. We denote by diffusion-delaying spillovers this minimum optimal level of knowledge outflows from industry leaders that induces followers to switch from concurrent to imitative development (s1∗ = s¯2 > 0). Proposition 10 is a much stronger result than the well-known possibility finding in Gallini (1984, restated more recently by Kulatilaka and Lin, 2006, and qualitatively discussed by Khazam and Mowery, 1994) that a technology leader may want to license its technology to a competitor to keep it from engaging in its own R&D. We show that the leader firm may want to give up some of its IP even without receiving any licensing payments in return (i.e., for free). In our view, Propositions 9 and 10 constitute important qualifications to the classical strategy view on interfirm knowledge spillovers.

MODEL EXTENSION: INVESTING IN FIRM CAPABILITIES The model extension in this section focuses on a central topic in strategy that has received little Strat. Mgmt. J. (2011) DOI: 10.1002/smj

Innovators’ Strategic Use of Disclosure to Slow down Competition attention in formal theoretical work: firm capabilities (Sutton, 2005). In our paper, time compression capabilities—or speed capabilities—are assumed to determine the level of effort required to bring new technologies to market. The more capable a firm, the faster it can develop a technology for a given cost level, and the cheaper it can develop the technology for a given development time. Since capabilities directly affect firms’ time-cost technology development trade-off, it is important to examine firms’ incentives to invest in capabilities and identify the main drivers of heterogeneity in industry capabilities.18 Definition 1: Let firm i’s incentive to invest in time compression capabilities under concurrent ( θ = C) or imitative ( θ = I ) development be ∂θi /∂di . For i, j ∈ {1, 2} and θ, φ ∈ {C, I }, we say that firm i’s incentive to develop capabilities under θ development is greater than firm j ’s incentive under φ development if ∂θi /∂di > ∂jφ /∂dj for all d1 and d2 that satisfy (A1)–(A3). We first examine the concurrent development case. In this regime, the incentives to invest in capabilities are straightforward (mathematically, √ ∂Ci ∂di = 2K( i − Zi r) > 0 for i = 1, 2). The incentives are naturally increasing in the revenue impact of the new technology, i : firms have more incentives to expand their development expertise if technologies are more valuable. An implication of this result is that the firm that expects to be the leader has, ceteris paribus, more incentives to invest in capabilities than the follower because it expects to receive a larger revenue stream by innovating (1 > 2 ). We also show that a firm’s incentive to invest in capabilities is increasing in its existing capability level because the latter reduces development effort and costs, thereby 18

For a discussion of capability development at Intel in response to increased competitive pressures, see the BusinessWeek cover story ‘Inside Intel—It’s Moving at Double-Time to Head Off Competitors’ (Hof, 1992). In a discussion of Intel’s investments in proprietary design tools, a company executive is quoted as saying ‘People have probably been wondering what we’ve been doing with those 386 profits. They’ve gone into ‘enablers’ that make it possible to design successive chips, each with two or three times as many transistors, at no increase in development time.’ Copyright  2011 John Wiley & Sons, Ltd.

increasing the returns from launching new technologies. These increasing returns to speed capa∂ 2 Ci bilities development (i.e., > 0) suggests that ∂di2 one could observe a positive feedback effect where a leader’s greater incentive to build capabilities can reinforce itself over time. Since we limit the extent to which the follower has superior capabilities—Assumption (A2)—we find that the leader has always greater incentives than the follower to invest in capabilities: Proposition 11: Under concurrent development: (i) The leader has greater incentive to invest in time compression capabilities than the follower; and (ii) each firm’s incentives to invest in capabilities is independent of the other firm’s capabilities: ∂ 2 Ci /∂d1 ∂d2 = 0. It is also useful to note that the returns to capability development in the concurrent regime are nonmonotonic in the complexity of the technology (K). The incentive to invest in capabilities ceases to exist when development is easy, for obvious reasons. The same is true when complexity is too large: in this situation, it takes a long time to develop the technology and the expected returns from innovation—and, thus, from investments in capabilities—are pushed far into the future and become √ more heavily discounted (i.e., as K increases, i − Zi r is reduced). Consider now the effect of imitation on firms’ incentives to invest in capabilities. For the leader (where  ∂I1 I ∂d1 = 2K( 1 − Z1 r) > 0) we have that: Proposition 12: (i) Imitation reduces the leader’s incentive to invest in time compression capabilities. (ii) Under imitation, the more capable the follower, the lower is the leader’s incentive to invest in time compression capabilities: ∂ 2 I1 /∂d1 ∂d2 < 0. Part (i) of Proposition 12 follows from the fact that the leader is in less of a hurry to develop the technology under imitation than in the concurrent regime because the leader knows that the follower will wait until the leader’s technology development process is complete before starting to invest (analytically, the result is due to I1 < 1 ). Since the leader is less motivated to compress its development process, capabilities are necessarily Strat. Mgmt. J. (2011) DOI: 10.1002/smj

G. Pacheco-de-Almeida and P. B. Zemsky less valuable. In the Intel-AMD case, our model would predict that AMD’s switch from an imitative to concurrent development strategy post-1985 led to an increase in Intel’s incentives to invest in capabilities. Part (ii) of Proposition 12 highlights a second contrast with concurrent development as now the follower’s capability investment is a strategic substitute to the leader’s investment. The more capable the follower, the less sustainable is the leader’s technology advantage and, thus, the less the leader invests in speeding up development (formally, this result follows from ∂I1 /∂d2 < 0). We turn now to the follower firm. A simple intuition is that imitation and having one’s own development capabilities are substitutes, which would imply that imitation lowers the incentives of the follower to invest in capabilities. We find that this simple intuition is only partially correct. The incentives of the follower under imitation are considerably more complex than under concurrent development and the leader’s incentives with imitation. This is because the follower’s incentives depend on when it expects to start its development with imitation, which is endogenously determined by the time-to-market of the leader (T1I )—and this, in turn, depends on the follower’s capabilities and the level of spillovers. Therefore, we have an ambiguous result: imitation may increase or decrease the follower’s incentives to invest in capabilities.19 However, can the leader influence the follower’s incentives under imitation? Do leaders’ capabilities induce followers to wait rather than develop the new technology in parallel? Proposition 13: (i) In a new market, an increase in the time compression capabilities of the leader increases the relative payoff to the follower of imitation. (ii) More generally, 19 To see this, we can simplify the problem by considering the limit as d1 → 1 in which case the leader develops almost instantaneously regardless of the follower’s capabilities or the level of spillovers. However, even with this strong simplification, we do not find that imitation and capabilities are necessarily ∂I ZI √ substitutes: limd1 →1 ∂d 2 = 2 1 −2d ( 2 − Z2I r). With Z2I = 2 2 (1 − s)(1 − d2 )K, the right-hand side of the limit expression is the same as the right-hand side of firms’ incentive expressions in the concurrent regime except that K is replaced by (1 − s)K. That is, spillovers shift the level of K in the incentive expression for the follower, but we have already observed that the incentives to invest in capabilities are nonmonotonic in K so that spillovers can either increase or decrease the incentives to invest even in this simplified limit.

Copyright  2011 John Wiley & Sons, Ltd.

an increase in the time compression capabilities of the leader increases the relative payoff to the follower of imitation if π00 is sufficiently close to π01 while it decreases the relative payoff if π00 is sufficiently close to π11 . If the leader’s launch of a new technology has a limited (negative) effect on the follower’s market revenues (i.e., π00 → π01 ), the follower’s incentives to develop the new technology (in both the imitative and concurrent regimes) depend only on its own time-to-market. Since the leader’s capabilities speed the follower’s time-to-market under imitation—and have no effect under concurrent development—they necessarily increase the relative attractiveness of imitation to the follower. This rationale also explains our result for the new market case (as π00 = π01 = 0). Alternatively, if firms have little to gain from a situation where both firms develop the new technology compared to a situation where none has it (i.e., if π00 → π11 ), the follower only really cares about the time interval between these two states—that is, the period in which it has a competitive disadvantage. Thus, the follower’s choice between imitation and concurrent development is driven by its incentive to minimize this disadvantage. Since the leader is less motivated to develop the technology under imitation, capabilities have a greater impact on its time-to-market in this case. As a result, the more capable the leader, the less likely the follower is to prefer imitation.

CONCLUSIONS We seek to elaborate a formal theory of technology development with time compression diseconomies. We characterize and compare firm competitive dynamics across two different technology regimes: concurrent vs. imitative development. In particular, we focus on the effect of imitation and interfirm spillovers on firm performance heterogeneity, the industry rate of technology diffusion, and firms’ incentives to invest in time compression capabilities. We show that technology leaders unambiguously want low spillovers if followers are committed to a strategy of imitation: lower spillovers make leaders’ competitive advantage more sustainable. However, we find that leaders may want to allow Strat. Mgmt. J. (2011) DOI: 10.1002/smj

Innovators’ Strategic Use of Disclosure to Slow down Competition some knowledge spillovers when followers have a choice between concurrent and imitative technology development. Providing a minimal level of spillovers can cause followers to shift from concurrent to imitative development strategies, which allows leaders to slow down their own development activities and, thereby, reduce development costs. This finding directly contributes to the growing literature that tries to explain why innovators often freely reveal their IP (von Hippel and von Krogh, 2006). In contrast with this literature, however, our paper establishes that innovators may have incentives to disclose proprietary knowledge even when spillovers bring no benefits to innovators and simply result in stiffer competition. Free revealing is the least bad option when what is at stake is not if, but when rivals will develop the same technology, which seems to be the case in most industries (Harhoff et al., 2003; Mansfield, 1985). To our knowledge, this is the first paper that shows the optimality of voluntary spillovers in the absence of any mechanism that compensates innovators for the leakage of their private information. We designate by diffusion-delaying spillovers the minimum optimal level of free revealing by innovators that induce follower firms to switch from concurrent to imitative technology development. Our theory advances our understanding of free revealing behavior in several industries and shows the shortcomings of the classical strategy view that innovators should always minimize knowledge spillovers to followers. A case in point is the example we used in our section ‘Motivating Examples’of the competition between Intel and AMD to develop microprocessors for use in PCs. Prior to 1985, Intel pursued an explicit strategy of IP disclosure: it agreed to share its microprocessor designs with AMD. Benefiting from high technological spillovers, AMD imitated: it waited until Intel released its processors and then developed its own products based on Intel’s specifications. However, the nature of competition between both companies fundamentally changed post-1985. At that time, Intel decided to no longer share its chip designs with AMD. In response to the reduced spillovers from Intel, AMD dropped its imitation strategy and started developing its next generation processors at the same time as Intel was working on theirs. This shift from imitative to concurrent competition has reportedly had a very significant adverse effect on Intel’s profitability. In short, Intel’s decision to reduce spillovers was the wrong Copyright  2011 John Wiley & Sons, Ltd.

one: it forced AMD to ramp up its capabilities and start racing. In contrast, by freely revealing part of its IP prior to 1985, Intel kept AMD on an imitative stand and better appropriated the returns from innovation. Our paper also finds that followers may actually benefit from reducing spillovers so as to increase the incentives of leader firms to quickly develop new technologies. Large spillovers may have an adverse effect on followers’ strategies: it may slow down the leader and, thereby, lenghten followers’ imitation processes. This finding is important because it challenges one central idea in conventional theories of innovation: it suggests that imitator firms may sometimes be better off with lower levels of absorptive capacity (Cohen and Levinthal, 1989, 1990). This finding also stands in contrast with the dominant view in international business by suggesting that followers may not always benefit from agglomeration economies or the geographical collocation with industry leaders to facilitate interfirm spillovers (Fujita et al., 1999; Krugman, 1991; Marshall, 1920). Note that this result also extends Pacheco-de-Almeida and Zemsky’s (2007) original analysis of interfirm spillovers to a setting with firm asymmetry. In terms of technology diffusion, we show that imitation always slows down leaders. In other words, leader firms are faster to market under concurrent than imitative regimes. However, there are no clear predictions regarding the relative impact of imitative vs. concurrent development on the speed of technology diffusion to follower firms. Interfirm spillovers have two effects: (a) they reduce the time followers spend on development, but (b) they also demotivate leaders. Hence, the total time-to-market of follower firms may increase or decrease with imitation. Under both concurrent and imitative development regimes, we find that there is a period of technology-based competitive advantage because firms differ in their time-to-market. Whether competitive advantages are more sustainable (i.e., last longer) under concurrent or imitative development depends on the level of interfirm spillovers. Competitive advantage is always associated with superior performance under concurrent development, but not necessarily so under imitation. Finally, we studied firms’ incentives to invest in time compression capabilities. We find that leaders have more incentives to invest in capabilities to accelerate technology development in concurrent Strat. Mgmt. J. (2011) DOI: 10.1002/smj

G. Pacheco-de-Almeida and P. B. Zemsky than imitative regimes. Interestingly, followers in new industries are more likely to pursue imitation as leaders’ development capabilities improve. Future research on time-consuming technology development may take this initial study in several different directions. A natural extension to our theory could be to model asymmetries in productmarket positions that interact with the value of the technology to each of the firms, as in Riordan (1992). Another obvious next step would be to study the effect of uncertainty on the competitive dynamics of our technology development regimes. In terms of empirical work, our theory can be tested by using data on project development times, project complexity, and technology spillovers that is publicly available in several industries such as semiconductors, consumer electronics, oil and gas, and automobiles. In these industries, major projects are often tracked by the trade press, while proxies for interfirm spillovers can be constructed from data on patent citations and cross-licensing agreements for which there usually is news coverage. Firm speed capabilities may also be inferred from firms’ investment projects, as illustrated by Pacheco-de-Almeida, Hawk, and Yeung (2011) in the oil and gas industry. An alternative empirical approach to testing some of the propositions in this paper might be to explore the existing Carnegie Mellon Survey data on imitation lags, complexity, and secrecy (Cohen et al., 2000). The main managerial implications of our paper are twofold. First, leader firms may sometimes better appropriate the returns from innovation by freely revealing part of it to competitors, as long as, by doing so, innovation diffusion is successfully delayed. For example, innovators may (a) publicly pledge to stop defensive patenting and open up more inventions to competitors via technical papers, the Internet, or at industry conferences, (b) patent in countries with weak IP protection regimes, and (c) establish R&D consortia or geographically collocate with followers to facilitate the leakage of private information. Second, imitators may want to limit their ability to appropriate too much of innovators’ know-how by (a) lowering in-house levels of absorptive capacity, (b) publicly committing to a human resources policy of not hiring away employees from leader companies, or by (c) loosely aligning strategy and organization with that of innovators to delay interorganizational learning. Copyright  2011 John Wiley & Sons, Ltd.

ACKNOWLEDGEMENTS We thank SMJ Editor Will Mitchell, April Franco, Michael Katz, Francisco Ruiz-Aliseda, Robert Seamans, Harbir Singh, Scott Stern, two anonymous reviewers, and seminar participants at the Wharton Technology Conference 2008, the Academy of Management Meetings 2008, and the London Business School for their helpful comments. The usual caveat applies.

REFERENCES Allen RC. 1983. Collective invention. Journal of Economic Behavior and Organization 4: 1–24. Arrow K. 1962. Economics welfare and the allocation of resources for inventions. In Rate and Direction of Inventive Activity, Nelson R (ed). Princeton University Press: Princeton, NJ; 609–626. Besanko D, Dranove D, Shanley M, Schaefer S. 2007. Economics of Strategy. Wiley: New York. Boehm BW. 1981. Software Engineering Economics. Prentice Hall: Englewood Cliffs, NJ. Casadesus-Masanell R, Yoffie D, Mattu S. 2005. Intel Corporation: 1968–2003 . HBS Case 9-703-427. Harvard Business School: Boston, MA. Cohen MA, Eliashberg J, Ho T-H. 1996. New product development: the performance and time-to-market tradeoff. Management Science 42(2): 173–186. Cohen WM, Levinthal DA. 1989. Innovation and learning: the two faces of R&D. Economic Journal 99(397): 569–596. Cohen WM, Levinthal DA. 1990. Absorptive capacity: a new perspective on learning and innovation. Administrative Science Quarterly 35(1): 128–152. Cohen WM, Nelson RE, Walsh JP. 2000. Protecting their intellectual assets: appropriability conditions and why U.S. manufacturing firms patent (or not). NBER Working paper no. 7552. National Bureau of Economic Research: Cambridge, MA. d’Aspremont C, Jacquemin A. 1988. Cooperative and noncooperative R&D in duopoly with spillovers. American Economic Review 78(5): 1133–1137. D’Aveni RA. 1994. Hypercompetition. Free Press: New York. De Fraja G. 1993. Strategic spillovers in patent races. International Journal of Industrial Organization 11(1): 139–146. Dierickx I, Cool K. 1989. Asset stock accumulation and sustainability of competitive advantage. Management Science 35(12): 1504–1511. Economist. 1998. Semiconductors. The monkey and the gorilla. 12 May: 71–72. Eisenhardt KM, Brown SL. 1998. Time pacing: competing in markets that won’t stand still. Harvard Business Review 76: 59–69. Fudenberg D, Tirole J. 1985. Preemption and rent equalization in the adoption of new technology. Review of Economic Studies 52: 383–401. Strat. Mgmt. J. (2011) DOI: 10.1002/smj

Innovators’ Strategic Use of Disclosure to Slow down Competition Fujita M, Krugman P, Venables AJ. 1999. The Spatial Economy: Cities, Regions, and International Trade. MIT Press: Cambridge, MA. Gallini N. 1984. Deterrence by market sharing: a strategic incentive for licensing. American Economic Review 74: 931–941. Ghemawat P. 1991. Commitment. Free Press: New York. Harhoff D. 1996. Strategic spillovers and incentives for research and development. Management Science 42(6): 907–925. Harhoff D, Henkel J, von Hippel E. 2003. Profiting from voluntary information spillovers: how users benefit by freely revealing their innovations. Research Policy 32(2003): 1753–1769. Helfat CE, Finkelstein S, Mitchell W, Peteraf MA, Singh H, Teece DJ, Winter SG. 2007. Dynamic Capabilities: Understanding Strategic Change in Organizations. Blackwell Publishing: Oxford, UK. Hesseldahl A. 2005. AMD to Intel: let’s rumble. BusinessWeek 23 August: 1–2. Hof R. 1992. Inside Intel: it’s moving at double-time to head off competitors. BusinessWeek 1 June: 2. Hoppe HC. 2002. The timing of new technology adoption: theoretical models and empirical evidence. Manchester School 70(1): 56–76. Judd KL. 2003. Closed-loop equilibrium in a multi-stage innovation race. Economic Theory 21(2): 673–695. Katila R, Chen EL. 2009. Effects of search timing on product innovation: the value of not being in sync. Administrative Science Quarterly 53(4): 593–625. Katz ML, Shapiro C. 1987. R&D rivalry with licensing or imitation. American Economic Review 77(3): 402–420. Khazam J, Mowery D. 1994. The commercialization of RISC: strategies for the creation of dominant designs. Research Policy 23(1): 89–102. Krugman P. 1991. Geography and Trade. MIT Press: Cambridge, MA. Kulatilaka N, Lin L. 2006. Impact of licensing on investment and financing of technology development. Management Science 52(12): 1824–1837. Langinier C, Moschini G. 2002. The economics of patents: an overview. In Intellectual Property Rights and Patenting in Animal Breeding and Genetics, Newman S, Rothschild M (eds). CABI Publishing: New York; 31–42. Levin RC, Klevorick AK, Nelson RR, Winter SG. 1987. Appropriating the returns from industrial research and development. Brookings Papers of Economic Activity 18(3): 783–831. Lieberman MB, Asaba S. 2006. Why do firms imitate each other? Academy of Management Review 31(2): 366–385. Lieberman MB, Montgomery DB. 1988. First-mover advantages. Strategic Management Journal, Summer Special Issue 9: 41–58. Lim K. 2009. The many faces of absorbtive capacity: spillovers of copper interconnect technology for semiconductor chips. Industrial and Corporate Change 18(6): 1249–1284. Mansfield E. 1971. Industrial Research and Technological Innovation. Norton: New York. Copyright  2011 John Wiley & Sons, Ltd.

Mansfield E. 1985. How rapidly does new industrial technology leak out? Journal of Industrial Economics 34(2): 217–223. Mansfield E. 1986. Patents and innovation: an empirical study. Management Science 32(2): 173–181. Mansfield E. 1988. The speed and cost of industrial innovation in Japan and the United States: external vs. internal technology. Management Science 34(10): 1157–1168. Marshall A. 1920. Principles of Economics. Macmillan: London, UK. Mitchell W. 1991. Dual clocks: entry order influences on incumbent and newcomer market share and survival when specialized assets retain their value. Strategic Management Journal 12(2): 85–100. Pacheco-de-Almeida G, Hawk A, Yeung B. 2011. Speed and Tobin’s q. New York University Working paper, New York. Pacheco-de-Almeida G, Zemsky P. 2007. The timing of resource development and sustainable competitive advantage. Management Science 53(4): 651–666. Pitofsky R, Anthony S, Thompson M, Swindle O. 1998. In the matter of Intel Corporation—complaint. Federal Trade Commission Docket no. 9288: 1–12. Porter ME. 1985. Competitive Advantage. Free Press: New York. Putnam L, Fitzsimmons A. 1979. Estimating software costs. Datamation October: 171–177. Quirmbach H. 1986. The diffusion of new technology and the market for an innovation. RAND Journal of Economics 17: 33–47. Reinganum JF. 1981. On the diffusion of new technology: a game theoretic approach. Review of Economic Studies 48: 395–405. Reinganum JF. 1989. The timing of innovation: research, development, and diffusion. In Handbook of Industrial Organization, Schmalensee R, Willig RD (eds). Elsevier: Amsterdam, the Netherlands; 849–908. Riordan M. 1992. Regulation and preemptive technology adoption. RAND Journal of Economics 23(3): 334–349. Ruiz-Aliseda F, Zemsky PB. 2007. Adoption is not development: first-mover advantages in the diffusion of new technologies. INSEAD Business School Research Paper No. 2007/03/ST. Salomon R, Martin X. 2008. Learning, knowledge transfer, and technology implementation performance: a study of time-to-build in the global semiconductor industry. Management Science 54(7): 1266–1280. Scherer FM. 1959. Patents and the Corporation: A Report on Industrial Technology under Changing Public Policy. Patents and the Corporation: Boston, MA. Scherer FM. 1967. Research and development resource allocation under rivalry. Quarterly Journal of Economics 81(3): 367–391. Scherer FM. 1984. Innovation and Growth. MIT Press: Cambridge, MA. Schumpeter JA. 1934. The Theory of Economic Development: An Enquiry into Profits, Capital, Credit, Interest, and the Business Cycle. Harvard University Press: Cambridge, MA. Strat. Mgmt. J. (2011) DOI: 10.1002/smj

G. Pacheco-de-Almeida and P. B. Zemsky Shih W, Ofek E. 2007. Intel 2006: rising to the graphics challenge. HBS Case 9-607-136. Harvard Business School: Boston, MA. Spence M. 1984. Cost reduction, competition, and industry performance. Econometrica 52(1): 101–121. Sutton J. 2005. Competing in capabilities: an informal overview. London School of Economics Working paper. London: UK. Teece DJ. 1977. Time-cost tradeoffs: elasticity estimates and determinants for international technology transfer projects. Management Science 23(8): 830–837. Trigeorgis L. 1996. Real Options. MIT Press: Cambridge, MA. Vance A. 2009. Patent king I.B.M. will give away more ideas. New York Times. 14 January. von Hippel E, von Krogh G. 2003. Open source software and the ‘private-collective’ innovation model: issues for organization science. Organization Science 14(2): 208–223.

Copyright  2011 John Wiley & Sons, Ltd.

von Hippel E, von Krogh G. 2006. free revealing and the private-collective model for innovation incentives. R&D Management 36(3): 295–306. Waring GF. 1996. Industry differences in the persistence of firm-specific returns. American Economic Review 86(5): 1253–1265. White L. 1971. The Automobile Industry Since 1945 . Harvard University Press: Cambridge, MA. Wiggins RR, Ruefli TW. 2002. Sustained competitive advantage: temporal dynamics and the incidence and persistence of superior economic performance. Organization Science 13(1): 82–105. Wiggins RR, Ruefli TW. 2005. Schumpeter’s ghost: is hypercompetition making the best of times shorter? Strategic Management Journal 26(10): 887–911. Zander U, Kogut B. 1995. Knowledge of the firm and the speed of the transfer and imitation of organizational capabilities: an empirical test. Organization Science 6(1): 76–92.

Strat. Mgmt. J. (2011) DOI: 10.1002/smj

Innovators’ Strategic Use of Disclosure to Slow down Competition APPENDIX A: MICROMODEL OF TECHNOLOGY DEVELOPMENT We derive the cost function C(T ; Zi ) given by Expression 3 in the body of the paper from the following explicit technology development process. Let ci (t) ≥ 0 be the expenditures profile that gives firm i’s flow of development cost at each time √ t. Expenditures translate into effort ci (t), which incorporates decreasat a rate ing returns to instantaneous expenditures. The present value of development a given T

Ti ∗ costs for −rt and Zi is C(T ; Zi ) = 0 ci (t; T , Zi )e dt, where ci∗ (t; T , Zi ) is the expenditure profile that minimizes the net present value of development costs subject to the constraint that all is exerted

T effort √ by the completion time (i.e., 0 ci (t)dt = Zi ). For a given completion time, this cost minimizing expenditure profile is determined by the following trade-off. On the one hand, the decreasing returns to instantaneous expenditures favors spreading effort uniformly over time. On the other hand, discounting favors shifting expenditures toward the end of the development period. Lemma 1 summarizes our results. Lemma 1: Suppose that firm i seeks to develop the technology over the time interval [0, T ]. The cost minimizing expenditure with effort Zi is c∗ (t; T , Zi ) =  profile  2 rZi ert . The minimized cost of technolerT − 1 rZ 2 ogy development is C(T ; Zi ) = rT i . e −1 The cost minimizing expenditure profile is increasing over time and it depends on the cost of capital. As also noted in the body of the paper, the cost function is a decreasing, convex function of the development time T and an increasing, convex function of the total effort Zi .

APPENDIX B: TIME-TO-MARKET COMPARATIVE STATICS This appendix summarizes two non-focal results on the comparative statics of optimal development times that complement the analysis developed in the paper. Corollary 1: Under concurrent development, the equilibrium time-to-market of firm i is increasing in the complexity of the technology Copyright  2011 John Wiley & Sons, Ltd.

( K), the cost of capital ( r) and decreasing in firm i’s capability ( di ) and the increase in revenue flows to the ith firm to deploy the technology ( i ). Corollary 2: In the imitative development regime, the equilibrium time spent on development by the follower ( Tˆ2I ) is decreasing in the follower’s capability ( d2 ), the increase in revenue flows to the follower ( 2 ) and spillovers ( s), and increasing in the cost of capital ( r) and the complexity of the technology ( K). The equilibrium development time of the leader ( T1I ) is decreasing in the leader’s capability ( d1 ) and increasing in the complexity of the technology ( K), the follower’s capability ( d2 ), the cost of capital ( r), and spillovers ( s).

APPENDIX C: PROOFS Due to space limitations, we only present here the proofs for the two main propositions of the paper. All other proofs are available from the authors upon request. Proof of Proposition 9 (The Limits to Imitation): The optimal s for the follower is less than 1 if ∂I2 (s)/∂s < 0 at s = 1. The profits of the follower with imitation are I2 (s) =  √ ( 1r − (1 − d1 )K I1 )(π11 − 2 2 (1 − s) (1 − d2 )Kr + (1 − d2 )2 r 2 K 2 (1 − s)2 ) + π00  ∂I2 (s) (1 − d1 )K I1 Thus, lims→1 ∂s = (1− √ − π01 ) − (1 − d1 )Kr d2 )K(4 π11 − π00 (π11 √ (π √ 10 − 4π01 + 3π11 ))(2 π11 − π00I π11 − π01 ) so that ∂2 (s)/∂s < 0 at s = 1 iff condition Z1 r >  √ π 10 − π01 π11 − π00 4/ 3 + π − π holds. 11 01 Proof of Proposition 10 (Free √ Revealing): √ 2√> rZ2 ( 5 +√3)/2 or, Suppose that 2 < rZ2 ( 5 + 3)/2 alternatively, that but π00 is sufficiently close to π01 . Then, from Proposition 8 we know that there exists a unique critical value s¯2 ∈ (0, 1) such that the follower prefers imitation if and only if s > s¯2 . If π10 is sufficiently close to π11 + (i.e., π10 → π11 ), then from Proposition 7 we know that the leader prefers imitation for all s ∈ [0, 1) and, thus, s1∗ = s¯2 > 0. If π10 is not sufficiently close to π11 but s¯2 < s¯1 , then we still have that s1∗ = s¯2 > 0. Strat. Mgmt. J. (2011) DOI: 10.1002/smj