Further papers attempt to provide a stronger theoretical framework to the discussion of trade secrets by modelling either the decision not to patent and/or the decision to use trade secrets. I highlight the distinction because the former approach treats trade secrets as a catch-‐all strategy for actions other than patents, as discussed in Arundel (2001.) The implication is that using no IP protection at all fits into the same category as Trade Secrets; using Trade Secrets is a passive action. The second approach of modelling the decision to use Trade Secrets treats Trade Secrets as an active decision. This approach takes into account the “reasonable steps” required to protect Trade Secrets. However, this approach, while excluding using no IP from using Trade Secrets, fails to offer using no IP as a strategy. Arundel (2001) points out that a number of policy discussions model the decisions to use patents with the assumption that patents are an obvious decision or mutually exclusive from trade secrets; both
assumptions, he argues, that are in conflict with the empirical evidence. This section of the chapter examines the literature’s analysis of the firm’s IP decisions with respect to the decision not to patent and the decision to use trade secrets.
Decision Not to Patent
The decision not to patent, as Arundel (2001) notes, is sometimes modelled as a default decision to use trade secrets. One example is that of Erkal (2005), discussed in Section 2.4.1, in which the author examines the impact of policy regimes on the behaviour of firms in two sequential R&D races. In the first round, depending on the regime, the winner may decide to maintain the first innovation secret in order to delay the progress of its competitors in the second round. Erkal suggests that strong trade secret protection should be accompanied
by broad patent protection and that allowing collusion to encourage disclosure may be optimal. However, Erkal assumes reverse engineering to be easy and that commercialization results in the same amount of information disclosure as patenting. However, reverse engineering can be difficult and/or costly and, consequently, commercialization can result in only limited disclosure. Thus, these two assumptions are fairly strict and reduce the model’s ability to make policy recommendations.
Decision to Use Trade Secrets
The active decision to use trade secrets presents a more complete analysis of the firm’s strategic protection of innovation, as noted in Arundel (2001.) An example is that of Bessen (2004), where the author develops a three-‐stage model with two competing firms. The Bessen model also incorporates issues of disclosure and argues that diffusion of knowledge is not necessarily more likely with a patent system.
In the first stage of the Bessen (2004) model, the innovator decides whether to patent or use trade secrets; in the second, the follower decides whether to develop innovation independently or not (potentially by inventing around); and, in the final stage, the firms produce and compete in the market. The author assumes that patenting costs more than using trade secrecy and concludes that firms use patents when they reduce or eliminate imitation.
If the follower chooses not to imitate, then the innovator will receive monopoly profits under trade secrecy, or, under patent protection, monopoly profits followed by duopoly profits for both firms once the patent expires. (For simplicity, Bessen chooses not to incorporate discounting.) If the follower chooses to imitate or invent around, they will incur R&D costs and their efforts may not result in a successful imitation. If the follower is successful, then both firms will receive duopoly profits under both IP regimes. Unsuccessful imitation results in the innovator retaining monopoly profits. The decisions of each player are determined by the probability of diffusion. Bessen argues that firms use
patents when they serve to reduce or prevent imitation and diffusion from imitation is, therefore, reduced under patents.
Bessen’s analysis focuses on the decision to use secrecy and the subsequent diffusion of technical information. The author argues that technology is more diffused when it is profitable for the follower to imitate and that patent regimes slow down this diffusion. Bessen notes that firms can choose whether to protect inventions by patents or by trade secrecy and predicts that the diffusion of technical information of inventions is not improved by the patent system and may be delayed.
Trade Secret Models and Licensing
The licensing of trade secrets allows for extensions of simpler models and can affect the social surplus effects of trade secrecy.
Bhattacharya and Guriev (2006) develop a two-‐stage model in which a research unit develops an innovative idea, which is then licensed to a development unit to potentially develop into an innovation. The focus of their paper is how the units can use licensing strategically. In their model, licensing occurs either through an open sale, in which the knowledge is protected through patenting, or a closed sale, in which trade secret protection is used. The authors conclude that the closed licensing with trade secrecy is most often used if the knowledge is highly valuable, if intellectual property rights are not well protected and if negotiations involve substantial knowledge leakage.
The Bhattacharya and Guriev approach highlights an important aspect of using trade secrecy in that the leakage during negotiations is paramount. The authors also note that valuable, trade secret protected knowledge is more likely to result in an exclusive license that minimizes leakage. If the monopoly rents created by an exclusive license are high enough, the licensor has no incentive to sell the knowledge to a third party. As the licensing literature acknowledges, self-‐
reinforcing mechanisms are crucial to successful licensing of knowledge.
The Bessen model is extended to include licensing and concludes:
The extent of the market for licenses may actually be greater without patents. The intuition is simple: licensing occurs where there is a credible threat of imitation. Because imitation occurs in more restricted
circumstances with patents than without patents, the extent of licensing is less with patents.
Cugno and Ottoz (2006) also develop a model detailing how the innovator’s choice to use patents or trade secrets as protection affects social surplus, which they term social welfare. In their view, patents represent a temporary monopoly followed by perfect competition, whereas secrecy represents a long-‐term
oligopoly. The introduction of licensing under the secrecy regimes allows for this oligopoly. Cugno and Ottoz conclude that social welfare is greater, ex post, when the innovator chooses secrecy given that transactions costs do not get too costly. Licensing avoids welfare wasting duplication costs (the cost of
reinventing an existing innovation) as these are instead appropriated by the inventor via licenses.
Applying a ratio test, also used in Scotchmer (2005), Cugno and Ottoz (2006) examine the conditions of the deadweight losses and profits under different IP regimes. The ratio test provides an overview of the per-‐period deadweight loss with respect to the per-‐period profit. As Scotchmer notes, this test is used in Gilbert and Shapiro (1990) to argue that substituting lower prices for longer protection is socially beneficial.”37 The ratio examines the ratio of deadweight loss and profits; where the ratio of deadweight loss to profits is lower indicates the better policy.
In Cugno and Ottoz (2006), the authors evaluate this ratio under secrecy and patents. Primarily, they are interested in cases in which the deadweight loss to profit under patents (WP) is greater than the deadweight loss to profit under secrecy (WS):
37 Scotchmer (2005), p. 109.
[2-1]
Assuming that patent duration and secrecy are equally profitable for the
innovator, the authors conclude that social welfare is greater under secrecy. The authors draw this conclusion because duplication costs under trade secrecy are converted into licensing fees. Thus, these duplication costs are not put towards the wasteful task of reinventing the wheel. The introduction of licensing makes trade secrecy a socially beneficial regime by increasing social surplus (which the authors refer to as social welfare.) However, this analysis ignores enforcement costs and the incentive to innovate goal of IP systems.
The licensing of trade secrets does, however, present some practical difficulties.
Primarily, given the intangible nature of trade secrets, it is difficult to determine if the licensee has truly received the trade secret knowledge and difficult to determine if the licensee ceases to use this knowledge upon the license’s
expiration. Nonetheless, as numerous authors examine (Scotchmer, 1991; Cugno and Ottoz, 1991; Choi, 2004; and Anand and Kahnna, 2000), the introduction of licenses to trade secret models can allow for increased social surplus and diffusion of knowledge.
The Anton and Yao Model
However, one weakness of many of the previously discussed models is that they treat trade secrets and patents as mutually exclusive, which, as noted by Arundel (2001), is not necessarily true. An example of an exception to these models is the Anton & Yao (2004) paper. In “Little Patents and Big Secrets,” Anton and Yao (2004) argue, as the title suggests, that firms should patent small innovations and use trade secrets to protect larger innovations. Furthermore, the authors
allow for a mixing of trade secrecy and patent for medium-‐sized innovations.
This widely cited paper provides a relatively comprehensive model in addressing the decision to use patents or trade secrets. Their argument is based on three fundamental assumptions that “innovation creates asymmetric information, innovation often has only limited legal protection and disclosure facilitates imitation.” The authors view the choices of IP protection, and its subsequent disclosure, as an important signalling mechanism. Based on this disclosure and limited legal protection, others determine whether to imitate or not.
Model
The Anton and Yao (2004) model begins with a cost reducing process innovation by an innovator. The innovation reduces the marginal cost (MC) of production to c. In the first stage, the protection and disclosure stage, the innovator chooses the form of IP protection (either secrecy or patent) and, as a consequence, the level of disclosure, which acts as a signal. Disclosure occurs under both Secrecy (S) and Patent (P), but only patent provides protection for disclosure. The disclosure allows the second player, the follower, the option of reducing his costs.
Table 2-2: Key Variables in Anton and Yao (2004) model
€
i=Innovator j=Follower
c =Marginal cost of prior technology c=Marginal cost of new technology
s=Follower's marginal cost with disclosure where
s≥c
The Innovator must decide between Secrecy (S) and Patenting (P) the
innovation. In the second stage, if the Innovator has chosen Patenting, then this stage is the infringement risking imitation stage. In this case, the follower chooses to not imitate (N) or to imitate (I), which is actually a decision to risk losing an infringement lawsuit. If the Innovator has chosen Secrecy, the Follower is assumed to Imitate. The Innovator and Follower have the following decisions:
Table 2-3: Key Decisions in Anton & Yao (2004) model
For the follower, choosing to imitate allows the follower to operate at marginal cost s. However, if the innovator has chosen Patent, the probability of being found to infringe is γ and damages are calculated based on the principle of reasonably royalty at τ.
In the final, competition stages, the firms compete in a Cournot duopoly where:
The model can be summarized in the following game tree:
The best response (BR) functions for each firm are:
[2-2]
[2-3]
Where g = damages rate = γ τ, which is the probability of the Follower being found to infringe, multiplied by the royalty rate that the court would require the Follower to pay the Innovator.
Innovator’s Strategy
Using separating perfect Bayesian equilibrium and backwards induction, the authors determine that the best strategies for the innovator are based on their innovation size. The Innovator chooses to signal its innovation either partially Figure 2-2: Anton and Yao (2004) Game Tree
(via secrecy) or more fully (via patenting.) Anton and Yao define the size of the innovation by the cost reduction it creates; the results are summarized here:
Figure 2-3: Anton and Yao (2004) Marginal Cost and Innovation Size
Table 2-4: Anton and Yao (2004) Model Conclusions Anton and Yao model conclusions
Innovation Size Large Medium Small
Effect Waiver Effect Licensing Effect No Imitation
IP secrecy patents patents
Disclosure partial partial full
Follower’s Action Produces at s Imitates, risks
damages Does not imitate (s ≥ c ≥ c*) Small innovations – No Imitation Effect
Small innovations are always patented because the innovator knows the follower will not imitate. To the follower, because the cost advantage is so small, the risk of paying damages outweighs the benefits of the lower marginal cost. Therefore, the disclosure associated with patents is acceptable for the innovator.
More specifically, a small innovation is defined as one that the marginal cost remains above c*. When c ≥ c*, disclosure is full. The follower could imitate, but the risk of damages outweighs the cost reduction benefits thus the follower remains at s where s ≥ c ≥ c*. The follower then earns (1/9 β)(α – 2c + )2
MC and Innovation Size
Large Medium Small
0 cL c*
c* is cost above which j chooses N, below I
which is dependent on the cost of the prior technology (where the follower remains) and the cost with the innovation (where the innovator operates.) The disclosure of c ≥ c* only affects the follower’s behaviour in that they do not imitate, knowing that damages would outweigh the cost benefit.
Medium innovations – Licensing Effect
For medium innovations, the follower is enticed to imitate, as expected damages no longer outweigh the benefit from infringing and the innovator patents. The follower imitates incurring the expected damages knowingly. These expected damages are transferred to the innovator through litigation and the damages function as a license.
Anton and Yao define the medium innovation as larger, where c<c*. This cost reduction is enough to trigger imitation by the follower. Knowing that types above c* (small innovations) disclose fully, disclosure to signal cost below c* has to be less than c*. Thus, the innovator will patent and disclose partially (e.g. not patent the entire innovation.) It is here that Anton and Yao add the crucial observation that patents and secrecy are not mutually exclusive. The decision to disclose partially and entice innovation means that the innovator will lose in terms of market revenue, but, as the follower is infringing, gain in terms of expected damages. Thus the authors define this area as the “licensing affect”, because the follower chooses to use the innovation in exchange for expected damages payments.
Large innovations - Waiver Effect
As the innovation becomes larger, c≤cL, the innovator faces a trade-‐off, when signalling low cost, between operating revenues and damages (licensing.) Lower costs will reduce the market price and, therefore, reduce expected damages (which are calculated based on reasonable royalty.) Thus, the innovator utilizes the cost advantage in the market as the main source of revenue and elects not to use damages by choosing secrecy.
Thus, the larger innovation uses trade secrecy to minimize disclosure and limit the follower’s ability to imitate. However, the authors highlight some of the practical difficulties with this conclusion. The decision not to patent may be due to the non-‐patentability of the innovation. They also suggest that a patent with a nominal lump sum may be a viable alternative but leave this option for later research.
Anton and Yao make one statement that contradicts the technical definition of a trade secret. They call the decision not to patent a “non-‐action.” The implication is that the decision to use trade secrets is thereby a non-‐action. This is not necessarily true, as discussed in Arundel (2001), as the decision to use trade secrets is an active decision. The decision to use trade secrets is a decision to utilize confidentiality agreements, secrecy and other legal measures for the protection of the innovation.
Furthering Anton and Yao
Encaoua and Lefouili (2006) develop a model similar to Anton and Yao’s.
However, Encaoua and Lefouili focus on the probabilistic nature of patents, rather than the signalling aspect of choosing an IP regime. They focus on three parameters that affect the firm’s IP choice: patent strength (likelihood patent will be held up in court), cost of imitating (the cost to the follower of imitating a patented innovation relative to a secret innovation) and the innovation size (the extent of the cost reduction.)
Both patent strength and innovation size are referred to explicitly in the Anton and Yao model. However, the cost of imitating, in Anton and Yao, is not explicit and is only captured partially in a consequential cost of imitation – that of
expected damages under the patent regime. Instead, the Encaoua model focuses on the costs to reverse engineer or develop the innovation independently, which is typically higher under secrecy than under patent protection. The Encaoua model treats expected damages as a separate cost.
A higher cost of imitation is likely to dissuade a follower from imitating. At the same time, a strong patent regime (patent strength) is also likely to dissuade imitation due to expected damages payments. Lower costs or weaker patent strength, on the other hand, make imitating more attractive. Thus, Encaoua and Lefouili view patent strength and the cost of imitation as strategic substitutes.
The Encaoua and Lefouili (2006) paper also departs significantly from Anton and Yao’s focus on signalling by permitting the innovation (cost reduction) to be directly observable. This is a fairly limiting assumption, as it would require firms to know marginal costs of the innovator directly, before and after the innovation.
The Encaoua and Lefoulli paper focuses on two competing effects: the damage effect and the competition effect. The damage effect is the advantage that patents have over secrecy. Patents allow for the possibility of damages payments, whereas secrecy does not38; thus, ceteris paribus, the firm would choose patents. The competition effect allows for imitation levels to differ between patents and secrecy. The regime that has less imitation will be preferred. The Encaoua and Lefoulli paper highlights the interaction between these two effects.
The Encaoua and Lefoulli (2006) model is very similar to the Anton and Yao model. An innovator creates a cost reducing process innovation and then proceeds to compete with a follower in a Cournot duopoly in three stages: the protection stage, the imitation stage, and, finally, the market competition stage.
However, the new model introduces one important new variable: I (cost of imitating), where imitating a patented innovation costs, at most, as much as under imitating an innovation protected by secrecy.
The results of the Encaoua and Lefoulli model are similar to that of Anton and Yao. They argue that small innovations are always patented. This, however, is due to the low cost of imitating a small innovation coupled with the lack of damages under secrecy. Also like Anton and Yao, the Encaoua model predicts
38 In the case of reverse-‐engineering, which is permitted under trade secrets law.
that large innovations are kept secret because any disclosure reduces I (the imitation cost) and invites imitation.
For medium sized innovations, the two models differ. In the Encaoua and Lefoulli model, medium sized innovations are either patented or kept secret, as opposed to Anton and Yao, where they are patented and partially disclosed.
Imitation occurs partially under secrecy and may occur under patenting. The difference between the two models’ conclusions stems from the fact that patent strength and imitation costs interact to influence the follower’s behaviour in the Encaoua model, whereas signalling by the innovator plays a role in the Anton and Yao model.
Encaoua and Lefoulli (2006) also develop the licensing option hinted at in Anton and Yao. They develop a fixed fee plus royalty licensing scheme which allows for the same equilibrium outcomes as in patenting under the shadow of
infringement. This is akin to Anton and Yao’s “licensing effect” which arises through litigation. Encaoua and Lefoulli frame licensing deliberately as an alternative to litigation. Jointly, these two papers embrace Shankerman and Scotchmer’s (2001) observations on the circularity of damages, in that reasonable royalty is determined by a combination of willing licensing, and coerced licensing through litigation. Encaoua and Lefoulli argue that licensing in lieu of litigation may lead to poor quality patents being licensed and decrease social surplus.
As the Encaoua and Lefoulli paper notes, the introduction of probabilistic patents to theoretical models allows for more thorough analysis of IP policy by opening up new research avenues. The authors claim that the one-‐size-‐fits-‐all approach of patents does not take into account the heterogeneous nature of innovations and patents. Instead, they suggest that:
…some flexibility could be introduced by allowing each innovator to choose a patent inside a menu of characteristics. For instance an innovator may have to choose between a patent with strong property rights and high disclosure requirements and a patent with weak property