1 Overview

On an instrumental view of the patent system, the patentee’s incentive to invent should be commensurate with the social value of the invention,Footnote ² and it is widely acknowledged that the social value of a technology is its incremental ex ante value over the next best alternative.Footnote ³

The view that the value of an invention depends on its value over the best alternative is premised on the view that the patent system should incentivize the invention of socially beneficial products. If an already known drug treats pain effectively, and there is a new drug that is equally effective but no better in any respect, then it would be wasteful to spend social resources on the new drug that offers no advantages over the old drug. While this basic starting point is widely accepted, there are a number of details that are open to debate.Footnote ⁴

2 Incremental Value

3 Ex Ante

1 What is “Bargaining Power”?

“Bargaining power” or “bargaining skill” is a term used in two related but distinct ways.Footnote ²⁵ Theoretically, it is used in the context of the solution to the bargaining problem, as initially set out in Nash’s famous paper of that name.Footnote ²⁶ If two parties with the opportunity to collaborate for mutual benefit are rational, they will engage in an exchange that maximizes the total net benefit to the parties jointly.Footnote ²⁷ This net benefit is often referred to as the gains from trade.Footnote ²⁸ The bargaining power discount, β, represents the way in which the parties to a negotiation split the gains from trade.Footnote ²⁹ So, if the patentee appropriates the entire gains from trade we would say β = 1, and if the bargaining power is equal the parties will split the gains equally, then β = 0.5.

Based on this theory, a patent licensing negotiation is often modeled as a bargaining problem in which the gains from trade are the difference between the patentee’s minimum willingness to accept and the implementer’s maximum willingness to pay (also sometimes referred to as the threat point).Footnote ³⁰ In an ex ante negotiation over an ironclad patent, the implementer’s maximum willingness to pay is normally taken to be the value of the invention, ν, as its threat point is to walk away from the negotiation and use the best noninfringing alternative. In assessing a reasonable royalty, the patentee’s minimum willingness to accept is normally taken to be its marginal cost.Footnote ³¹

In light of this model, bargaining power is also often used to mean that part of the difference between the value of the invention and the patentee’s marginal cost that will be captured by the patentee in an actual license negotiation. This practical meaning of bargaining power and the theoretical meaning correspond only if the bargaining model just discussed accurately describes real-world negotiations. In particular, the Nash Bargaining Solution is strictly applicable only to negotiations over pure gains from trade. It is well understood that invention is only the first step toward commercialization, and that an implementer normally must make product specific investments in manufacturing, advertising, and distribution, and so on, in order for a product to be successfully commercialized. A number of authors have suggested that in practice the implementer’s share of the profits is, at least in part, a return to the implementer for its contribution to that joint value.Footnote ³² If that is correct, then the split in profits negotiated by parties to a real-world agreement does not correspond exactly to the split in gains from trade in the theoretical Nash Bargaining Model.Footnote ³³

Returning to the basic theoretical bargaining model, the literature as to what determines bargaining power is thin. Nash’s famous paper setting out what is now known as the Nash Bargaining Solution, took equality of bargaining power as a premise, and did not consider any of the factors that might affect it.Footnote ³⁴ Formal game theory has added little to the concept to give it more real-world content. The main theoretical refinement is by Rubinstein (Reference Rubinstein1982), who shows that under certain conditions, a party with a higher discount rate (higher time value of money) will have less bargaining power. This supports the informal view that resource constraints affect bargaining power. Many other factors doubtless also affect real-world bargaining power, such as the ability to drive a hard bargain by psychological negotiating tactics (which might be termed bargaining skill), or repeat play and reputation effects.

The Nash Bargaining Solution is applicable only when there is some degree of bilateral monopoly, at least in the sense of the object of exchange having unique value to one of the parties.Footnote ³⁵ This means that the market structure is also relevant to bargaining power. For example, a party negotiating with a counterparty in a competitive market will be able to extract the entire surplus by threatening to license to a different counterparty.

On the whole, the elegance and simplicity of the Nash Bargaining Model has made it a very attractive modeling construct, but a richer description of the factors that affect real-world bargaining power would be useful.

2 Justification for Bargaining Power Discount

The main justification provided by Lemley & Shapiro for using θβν as the benchmark is that it reflects the royalty that would be negotiated by parties if they negotiated ex ante, and the return to voluntary market negotiations is theoretically appropriate in the absence of any known market failure.Footnote ³⁶ So, as noted immediately above, the implementer’s share of the profits may represent in part a return to investments made by the implementer to commercialize the invention. If so, voluntary market negotiations will provide an appropriate return to that investment, as will a benchmark that mimics the market.

As well as looking to the incentive to invent, the royalty also affects implementer incentives. To the extent that the gains from trade represent pure economic rents, then the particular split does not in principle affect the incentive to implement because any positive share provides a greater return to the implementer than does the best alternative.Footnote ³⁷ However, to the extent that the implementer’s share of the profits is a return to the implementer’s technology-specific investments, holdup is inefficient, even if the royalty is less than θν, because the implementer may have to pay more for the patented technology than the value the implementer derives from the technology. When that is the case, the implementer may avoid implementing the technology, even though the technology itself would have a net social benefit.Footnote ³⁸ It is the avoidance of efficient investments by implementers that is the real downside of holdup.Footnote ³⁹

3 Criticism of Bargaining Power Discount

4 θ: Patent Strength

The benchmark fair royalty rate requires the value of the patented technology to be discounted by the strength of the patent, which is to say, the probability that it is valid and infringed. Otherwise, the implementer will be paying for a technology that it did not use, or for which no patent should have been granted. In U.S. law, a reasonable royalty is in principle awarded on the assumption that the patent was known to be valid and infringed – that is, without any discount for patent strength, since damages are only awarded if the patent has been held to be valid and infringed. This is not inconsistent with the principle that a fair royalty requires a discount for patent strength; on the contrary, it is necessary to avoid double discounting.Footnote ⁶² While these principles are not controversial, the extent to which the courts appropriately apply or ignore the patent strength discount is another question. The most complete analysis of that issue is Masur (Reference Masur2015), who characterizes existing U.S. law on this point as “both incoherent and backwards.”Footnote ⁶³

1 Sunk Costs Holdup

Farrell et al. (Reference Farrell, Hayes, Shapiro and Sullivan2007) describe “opportunism” or “holdup” as follows: Holdup can arise, in particular, when one party makes investments specific to a relationship before all the terms and conditions of the relationship are agreed upon.Footnote ⁶⁵

They provide the following example of holdup in a case where the patented technology costs $40 to implement, exclusive of any royalty, and the best alternative technology costs $50, so that the inherent advantage of the patented technology is $10, and a benchmark reasonable royalty is any amount less than this:

That is, the fact that the user has made transaction-specific investments prior to negotiating for the right to use the technology means that the patentee can capture part of the user’s sunk costs, in addition to the inherent advantage of the patented technology. (This analysis implicitly assumes that the successful patentee will be granted an injunction.) It is convenient to refer to holdup arising from such transaction-specific investments as “sunk costs holdup” where the transaction specificity is left implicit.Footnote ⁶⁷

Sunk costs holdup relies centrally on the transaction-specific investment being sunk before any negotiations take place. It does not turn on the product being complex, as it can arise when only one patent covers the product. Nor does it turn on the probabilistic nature of the patent, or on the cost of litigation – in the above example, litigation costs are assumed to be zero.

If sunk costs holdup does occur, it has adverse effects on both patentee and implementer behavior. It allows the patentee to capture more than the value of the invention, thus creating an excessive incentive to invest in patented technologies; and the prospect of being held up increases the ex ante risk to the implementer, thus reducing the attractiveness of investments in products that are potentially subject to sunk costs holdup.

2 Network Effect Appropriation

Another type of holdup, applicable primarily in the context of standards, is referred to by Siebrasse & Cotter (Reference Cotter2017a) as network effect appropriation, which they define as follows:

U.S. courts have consistently held that a reasonable royalty should not reflect “any value added by the standardization of that technology.”Footnote ⁶⁹ On its face this appears to say that a patentee should not be able to capture any value arising from network effects, though as Siebrasse & Cotter (Reference Cotter2017a) point out, courts routinely award damages, including ongoing royalties in lieu of an injunction, in the form of running royalties, which do allow the patentee to capture value arising from network effects.Footnote ⁷⁰

From a policy perspective, Siebrasse & Cotter (Reference Cotter2017a) argue that allowing the patentee to capture some part of the value of a patented technology that arises due to network effects is desirable from a dynamic efficiency perspective, because it provides the correct incentive to invent, and it is not undesirable from a static efficiency perspective, as it has no adverse effects on implementer incentives.Footnote ⁷¹ While there are many articles arguing that a patentee should not be able to extract a higher royalty ex post than it could have obtained ex ante, which suggests that the patentee should not be able to capture any value arising from network effects, such articles typically do not distinguish between sunk costs and value arising from network effects. The two often go hand in hand, because adopting a standard and the consequent network effects, is often accompanied by substantial sunk costs. Two exceptions are Swanson & Baumol (Reference Swanson and Baumol2005) and Lee & Melamed (Reference Lee and Melamed2016), which both specifically assert that the patentee should not be able to capture value arising from network effects. However, both treat network effect appropriation equivalently to sunk costs holdup, and they do not offer independent policy justification for not allowing the patentee to capture any of the value arising from network effects.Footnote ⁷² Chao (Reference Chao2016) also takes issue with Siebrasse & Cotter (Reference Cotter2017a) on this point, but his discussion turns on what Siebrasse & Cotter (Reference Cotter2017a) characterize as the distinct problem of apportionment, which is discussed in the next section.Footnote ⁷³

3 The Apportionment Problem

Another type of holdup may arise when a patent claims a relatively minor feature of a complex product. If the patentee can get an injunction that prevents sale of the entire product, it can extract part of the value of the entire product, even though the patented technology contributes little to that value. This point is explored at length by Lemley & Shapiro’s model, which is addressed next.

The apportionment problem, when it exists, has the same adverse incentive effects on patentee and implementer incentives as does sunk costs holdup. It allows the patentee to capture more than the value of the invention, creating an excessive incentive to invest in minor patented technologies. The prospect of being held up in this manner increases the ex ante risk to the implementer, thus reducing the attractiveness of investments in products that are potentially subject to the apportionment problem.

The apportionment problem is exacerbated in the context of a standard. As discussed in more detail below, the excessive royalty that can be extracted by a patentee armed with an injunction is generally capped by the losses that would be suffered while the technology is designed around. This implies that if the technology can quickly and easily be removed or designed around, then the royalty overcharge will be small. However, this is not true in the SEP context because licensing terms of SEPs almost always specify that the SEPs are only licensed for use in the products that comply with the standard.Footnote ⁷⁴ That means that if the technology covered by a SEP could easily be designed around or removed as a technical matter, the product would no longer be compliant with the standard and the other licenses to the truly important SEPs would lapse. This would allow the owner of the unimportant SEP to capture the value of the standard as a whole.Footnote ⁷⁵ It is not enough to design around a SEP technically; instead, the implementer would have to be able to lobby the standards organization to remove the technology in question from the standard. While not necessarily impossible, this will certainly be a very lengthy process.Footnote ⁷⁶ In such a case, the “redesign period” referred to in the discussion below should be interpreted to mean the time needed to get the SEP removed from the standard, rather than the time needed for a technical redesign.

4 Probabilistic Holdup: Lemley & Shapiro Model

Lemley & Shapiro (Reference Shapiro2007a) and Shapiro (Reference Shapiro2010) provide a widely discussed model of holdup.Footnote ⁷⁷ Their model incorporates both sunk costs holdup and the apportionment problem, and additionally addresses the effect of the probabilistic nature of patents; that is, the fact that the validity and scope of granted patents is uncertain until they are litigated.Footnote ⁷⁸ Their model extends the simple sunk costs holdup model in two other respects. First, in the simple sunk costs model the implementer’s option is to license or redesign its product to avoid using the patented technology; Lemley & Shapiro develop a more explicit model of the litigation process in which the implementer may choose to redesign either during the litigation period, or after the end of litigation. Second, in the simple model, the implementer is at risk of being held up for transaction-specific sunk costs that are generally conceptualized as being machinery or other tangible goods. Lemley & Shapiro point out that the implementer is also at risk of being held up for lost profits during the period that its product is being redesigned to avoid infringement. Lemley & Shapiro also focus on redesign costs (sometimes referred to as switching costs), rather than sunk costs. (The relationship between switching costs and sunk costs is discussed below.)

They consider two scenarios: a “surprise” scenario in which the implementer is already selling its product when it learns of the patent, and an “early negotiation” scenario in which negotiations take place before the product is designed.Footnote ⁷⁹ For “ironclad” patents – those that are certainly valid – their model is a variant on the standard sunk costs model of holdup; there is no overcharge when ex ante negotiations are possible, and in the ex post scenario the patentee extracts part of the costs of switching to a noninfringing alternative.

When considering probabilistic patents, Lemley & Shapiro further distinguish between two scenarios. If the patent is relatively weak, it will make more sense for the implementer to refrain from redesigning until after it has lost in litigation, in which case its threat point is determined by the sunk costs plus the lost profits on the entire product during the period of redesign. This is the “Litigate” scenario. On the other hand, if the patent is relatively strong, the implementer’s best negotiating strategy is to threaten to redesign its product during litigation (“Redesign and Litigate”), in which case its threat point is determined by the redesign costs. This means that a weak patent will have a higher relative overcharge because it can extract not just redesign costs, but also lost profits on the entire product during the period of redesign. The overcharge is discounted by the probability of validity, so the absolute overcharge for a weak patent will normally be smaller than for a strong patent covering the same technology.

Lemley & Shapiro show that the probabilistic nature of patents can result in an overcharge even when ex ante negotiations are possible. This is because the implementer’s threat is to avoid using the patented technology entirely, and adopt the best noninfringing alternative instead. This is appropriate for an ironclad patent because it allows the patentee to obtain part of the true value of the invention. The problem is that the implementer’s threat is exactly the same, and so the outcome of the negotiation is exactly the same, even if the patent is potentially invalid. This implies an overcharge, because the royalty should be discounted by the probability of invalidity. The problem, as they put it, is that “the accused infringer has chosen to give up without a fight, effectively agreeing to treat a possibly invalid patent as certainly valid, and so the chance that it would have invalidated the patent will not be reflected in the negotiated royalty.”Footnote ⁸⁰ (Because the overcharge can be extracted even when an ex ante negotiation takes place, it is perhaps not strictly correct to refer to it as “holdup,” which normally implies that a higher royalty can be extracted ex post, than could have been negotiated ex ante.)

To summarize their results:

• Scenario 1 – Surprise – “Litigate” strategy

  • Applicable when patent is weak, redesign costs are high.

  • Overcharge because patentee can extract lost profits on the entire product during redesign, plus redesign costs, both discounted by probability of validity. Because of the discount the absolute value of the overcharge will be small, but because of the lost profits on the entire product, the percentage overcharge will be large.

  • Overcharge increases with (a) redesign costs; (b) lost profits during redesign period; and (c) value of the product relative to the value of the invention.

• Scenario 2 – Surprise – “Litigate & Redesign” strategy

  • Applicable when patent is strong, redesign costs are low.

  • Overcharge because P can extract redesign costs, not discounted.

  • Percentage overcharge (a) increases with redesign costs, and (b) decreases with probability of validity (i.e., is greater for weak patents).

• Scenario 3 – Early Negotiation

  • Either just like surprise case,

  • or

  • Overcharge because implementer’s threat is not to use the invention with certainty, in which case percentage overcharge decreases with the probability of validity.

Their results do not turn directly on the complex nature of the product, but complex products are likely to be subject to a greater overcharge because they are likely to face Scenario 1, in which a weak patent with relatively little value to the product can nonetheless extract a portion of the value of the entire product.

The adverse economic effects of holdup in Lemley & Shapiro’s model in the surprise scenario are the same as for sunk costs holdup (though the mechanism is somewhat different). The economic effects of probabilistic holdup in the early negotiation scenario are slightly different. Again, the patentee is capturing more than the value of its contribution, which creates an excessive incentive to invest in patenting. But in principle the overcharge will not increase the risk to the implementer, because it knows how much it has to pay ex ante. Nor will it cause the implementer to avoid using the patented technology, because the patentee will not charge so much that the implementer would prefer to use the alternative. It will in principle reduce the implementer’s expected profit, thus creating a distortion in the direction of investments. The degree of the distortion will presumably depend on the market structure.

While this model is well-known and influential for its implications respecting injunctive relief, within the context of remedies, and particularly the withholding of injunctive relief, another implication is that additional effort should be devoted to weeding out weak patents before they are licensed or litigated.Footnote ⁸¹

5 Sunk Costs, Switching Costs, and Lock-in

Holdup is sometimes described as involving switching cost, on the view that once one technology is selected, it may be that the cost of switching to the alternative technology is prohibitively expensive.Footnote ⁸² This characterization is used most often in the standards context, where the implementer is said to be “locked in” to the standard once it is chosen, but similar reliance on switching costs as giving rise to holdup is also found in other contexts.Footnote ⁸³ This contrasts with the traditional focus of the general economic holdup literature on sunk costs, in which holdup occurs when a party tries to charge a higher price than it would have been able to before those sunk costs were incurred. The puzzle is that sunk costs were necessarily incurred in the past – a party cannot be held up for costs that it has not yet incurred – while “switching costs” on the other hand, imply costs that would take place in the future, after failed negotiations, to switch to an alternative, nonstandard technology.

Cotter et al. (Reference Cotter, Golden, Menell and Schwartz2018) provide a general framework for reconciling concepts of switching costs and sunk costs. They explain that the threat of adopting the next best alternative always disciplines the royalty that can be extracted by the patentee, but the value of both the patented technology and the alternative may change. After costs are sunk, the selected technology is more profitable going forward, because the costs of implementation have already been incurred. So sunk costs holdup can be thought of as representing holdup due to the differential profitability of the selected technology ex ante versus ex post. The differential profitability of the alternative technology represents a separate source of holdup. If the profitability of the alternative technology changes, either because its costs change, or because its revenues change – as when it is not selected to be the standard – the disciplining value of the user’s threat to switch also changes. Switching costs as such, in the sense of the forward-looking cost of implementing the alternative technology, are irrelevant to holdup. If the cost of implementing the alternative technology is the same ex ante or ex post, any amount that could be extracted by the patentee ex post, because the implementer wants to avoid incurring those costs, could also have been extracted ex ante. Implementers become “locked in” to a standard, not because of the costs of switching, but because the expected revenue from the alternative technology will have been reduced once the original technology was adopted as part of the standard.

This has practical implications. Lemley & Shapiro (Reference Shapiro2007a) recommend that “the court should evaluate the cost that the infringing firm would have to incur to redesign its product to avoid infringing the patent. If this cost is high relative to the value that the patented technology has added to the infringing firm’s product, no permanent injunction should be issued.”Footnote ⁸⁴ But as Denicolò et al. (Reference Denicolò, Geradin, Layne-Farrar and Jorge Padilla2008) point out, the relevant comparison is not the cost of redesign, but the additional cost of adopting the alternative technology ex post as compared with ex ante. Looking only to the cost of redesign risks penalizing “the most valuable patents – precisely, those that are most difficult to circumvent even with full knowledge of the patent.”Footnote ⁸⁵ They note that instead “the policy should indicate that to avoid injunctive relief an infringer must show not only that it is costly to redesign the product in a non-infringing way ex post, but also that it could easily have designed the product in a non-infringing way ex ante if only it had been aware of [the patent holder’s] patent (which again emphasizes the importance of the inadvertent infringement assumption).”Footnote ⁸⁶ The point made by Denicolò et al. (Reference Denicolò, Geradin, Layne-Farrar and Jorge Padilla2008) is consistent with the analysis provided by Cotter et al. (Reference Cotter, Golden, Menell and Schwartz2018); it is not the cost of switching to the alternative that is important, but whether the cost of switching has changed.Footnote ⁸⁷

6 Caveats and Critiques

7 Competing Patentees

Lemley & Shapiro state that their analysis is limited to situations in which the patentee’s predominant commercial interest in bringing a patent infringement case is to obtain licensing revenues and it does not apply to settings in which the patent holder practices the invention and seeks to use the patent to exclude a competitor from the market in order to preserve its profit margins.Footnote ¹¹¹ Golden (Reference Golden2007) and Elhauge (Reference Elhauge2008) argue that Lemley & Shapiro’s distinction is not compelling, and they indicate that even patentees seeking only royalties should be entitled to injunctive relief.Footnote ¹¹²

The holdup problem faced by the implementer is just as severe whether the patentee competes in the market or not.Footnote ¹¹³ The key question is therefore whether there are countervailing considerations that imply that a patentee who competes in the market should be granted injunctive relief notwithstanding these holdup concerns.

Lemley & Shapiro’s model considers only reasonable royalty damages, and they equivocate when considering a patent holder who would be entitled to lost profit damages, saying in cases involving “significant” lost profits, they favor a presumption that the patent holder will be granted a permanent injunction,

This suggests that the costs of denying injunctive relief, in the form of increased error costs of damages calculation, are greater in the context of lost profits.Footnote ¹¹⁵ There are two problems with this argument.

First, while it is no doubt difficult to assess lost profits on an ongoing basis, it is not easy to accurately quantify a reasonable royalty either. It is not evident that lost profit calculations are generally so much more difficult than reasonable royalty calculations, particularly in the case of complex products, as to justify a sharp distinction between cases in which the patentee is seeking lost profits and those in which it is not.Footnote ¹¹⁶

Moreover, Golden (Reference Golden2007) points out that the difficulty in assessing reasonable royalty damages has traditionally been one of the principal rationales for granting permanent injunctions.Footnote ¹¹⁷ Lemley & Shapiro respond by noting that “all that is required for reasonable royalties to play their role in guiding parties to a negotiated settlement in the shadow of litigation is that they be unbiased.”Footnote ¹¹⁸ But this same point undermines their distinction between reasonable royalties and lost profits; even if lost profits are more difficult to assess, that makes no difference so long as the errors are unbiased. The important question is not whether lost profit damages are more difficult to assess than reasonable royalties, but whether they are more likely to be biased against the patentee. There is no obvious reason why errors in lost profit damages are less likely to be unbiased than reasonable royalty damages.

Apart from the relative accuracy of the two types of damages, Elhauge (Reference Elhauge2008), and Denicolò et al. (Reference Denicolò, Geradin, Layne-Farrar and Jorge Padilla2008) suggest that the holdup problem might be worse when the patentee is able to seek lost profit damages because it competes in the downstream market. In that case the patentee may hold up the implementers even more because higher royalties provide it with increased market share in the downstream market, as well as directly benefiting from high royalties itself.Footnote ¹¹⁹ In effect, the patentee has increased bargaining power when it competes in the downstream market; when it only licenses, the royalty is constrained because it will make nothing if the royalties are so high as to unduly restrict sales, but if the patentee competes in the downstream market that constraint is lifted, as the patentee might anticipate capturing those sales itself.

A distinct reason for preferring injunctive relief in the case of a patentee that practices the invention is that in such a case we should expect the patentee to be more efficient, because if the infringer were more efficient than the patentee, the patentee would have been willing to license. Allowing the patentee to exclude the infringer in such circumstances gives the market to the more efficient producer.Footnote ¹²⁰ However, granting an injunction to a nonpracticing patentee should have the same effect, and the patentee would license to the more efficient producer.

As a final point on this issue, Geradin (Reference Geradin2010a) argues that Lemley & Shapiro’s distinction between patentees seeking lost profits and those seeking reasonable royalties “would unduly affect innovators which have opted for a licensing business model for perfectly legitimate reasons, such as for instance the fact that they do not have the skills or the resources to develop and manufacture products embedding their technologies,” and “effectively tip the market in favor of vertically-integrated incumbents … [, which] would impede efficiency-enhancing specialization allowing firms to focus on what they do best and harm innovation.”Footnote ¹²¹ However, it is not clear that the lost profit damages per unit, properly assessed, will be greater than the royalties per unit. That will only be true, in an economic sense, if the vertically integrated firm has the capacity to satisfy the market and is a more efficient producer than the implementer, in which case it is not inefficient to give the patentee extra leverage against the implementer. If the vertically integrated firm is actually worse at commercializing the invention, this implies that its lost profits will be less than the reasonable royalty it could have obtained from licensing to a more efficient implementer; as Geradin (Reference Geradin2010a) points out, the innovator is more likely to opt for a licensing model when it does not have the skills or resources to manufacture the product, and the lower return from a royalty reflects these shortcomings. While this follows as a matter of economic logic, it requires that the lost profits calculation properly accounts for the patentee’s costs of production, including fixed costs. If lost profits calculations are excessively generous to the patentee, then the vertically integrated patentee will have greater leverage because of the excessive damages for past infringement whether or not it is granted an injunction. As discussed above, the implementer’s share of the surplus may be best understood as compensation for its investment in the success of the product, through marketing and distribution, etc., which would represent costs to the patentee.

In summary, despite their protestations, Lemley & Shapiro’s holdup model does prima facie apply to patentees who compete with the infringer. This does not imply that their model should be rejected, but it does suggest that their model is incomplete,Footnote ¹²² and/or that the holdup problem needs to be taken seriously in that context as well.

1 Modified Injunction

1 Inaccuracy v. Biased Damages

It is largely uncontroversial that the assessment of damages for patent infringement is likely to be inaccurate, as Lemley & Shapiro acknowledge in their reply to Golden.Footnote ¹⁵⁸ They say, however, that

This exchange reflects a similar debate in the general literature on property rules versus liability rules. There has been a substantial literature responding to Calabresi & Melamed, arguing that liability rules are superior to injunctive relief in a range of circumstances, to the point that “[p]roperty rules find relatively few defenders among legal economists.”Footnote ¹⁶⁰ Smith (Reference Smith2004) points out that the pro-liability rule literature turns on two basic assumptions: that the risk distribution is known; and that errors in judicial determination of damages are unbiased.Footnote ¹⁶¹ Lemley & Shapiro’s response to Golden reflects the second assumption in particular.

If the courts can reliably award damages that are equal to the loss suffered by the patentee, at least in expectation, the basic argument for injunctive relief would be much weaker, as the assessment of damages would perfectly compensate the patentee while avoiding the holdup problem. The puzzle is that this proves too much: If damages are accurate in expectation, then even a slight possibility of holdup would be enough to warrant denying injunctive relief, given that there will be no impact on the incentive to invent. One response to this puzzle is to say that injunctive relief is justified on the basis of the transaction cost arguments or Smith’s theory, discussed in the preceding sections.

Another response is to posit that damages are systematically undercompensatory, apart from any feedback effects from holdup and the availability of injunctive relief, that would tend to make damages overcompensatory.Footnote ¹⁶² If so, this would not in itself imply that injunctive relief should routinely be granted. Indeed, that would provide a ready explanation, at least in principle, for the observed pattern of injunctive relief in patent cases. Ever since Calabresi & Melamed, the pro-property rights literature has acknowledged that a liability rule is justified when there is a serious risk of holdup. The shift in patent law can be reconciled with the traditional dominance of property rules, and traditional property rights theory, on the basis that shifting realities, such as the rise in patent NPEs and SEPs, and perhaps also a general increase in patents for complex products, have substantially increased the circumstances in which there is a serious risk of holdup.

If damages are systematically undercompensatory, the difficulty is not conceptual, but practical. As Lemley & Shapiro point out, “all advantages are comparative.” They argue that “since, as we have demonstrated, injunctive relief will systematically overcompensate patent owners in component industries, there is a strong reason to prefer damages rules in those cases.”Footnote ¹⁶³ But this observation cuts both ways. Even if it is true that injunctive relief will systematically overcompensate patent owners, that in itself only gives a strong reason to prefer damages rules if there is no counterbalancing reason to prefer property rights. It is not enough to simply point to a risk of undercompensation to justify a property rule, but neither is it enough to simply point to the risk of holdup to justify a liability rule. Instead, the question would turn on whether the problem of undercompensation is outweighed by the holdup problem. If damages are undercompensatory, this kind of balancing inquiry is inherently difficult, as it turns not just on the existence of undercompensation or holdup, but also on an estimate of the relative severity of each.

2 Are Damages Biased?

1 Litigation Costs and Resource Constraints

To isolate the role of litigation costs, suppose that detection is certain and damages are fully compensatory. The basic rejoinder to the argument that implementers will take advantage of high litigation costs to force an unfair settlement is that this strategy is expensive for the implementer as well. If litigation costs are symmetric, costs drop out of most formal models, as the parties will settle in order to avoid them, and symmetric costs do not give either party an advantage in the negotiations. Indeed, it is normally suggested that high litigation costs will encourage early licensing, rather than holdout, in order to avoid the litigation costs.Footnote ¹⁸⁰ This reasoning implies that there must be some kind of asymmetry between the parties before litigation costs can distort the royalty settlement, though when asymmetry does exist, it can result in unfair settlements.Footnote ¹⁸¹ The same is true if there are asymmetries in risk aversion, perhaps because of resource constraints.

Litigation costs may undoubtedly be asymmetric in particular cases, but there seems to be little reason to believe that litigation costs systematically favor the accused infringer either in general, or in the category of patents for complex products. Indeed, it may be that patent holders have a systematic cost advantage because litigation may impose substantial discovery costs on alleged infringers without an equivalent burden on the patent holder.Footnote ¹⁸² Similarly, patent assertion entities likely have a cost advantage over end users.Footnote ¹⁸³ Nor is there any particular reason to believe infringers have a systematic resource advantage over patentees.Footnote ¹⁸⁴

With that said, there is no doubt that cost or resource asymmetries may cause significant distortions in some individual cases, and potentially in some categories of cases. However, it is not clear that granting injunctive relief in such cases will effectively address the problem in those cases where costs consideration favor the implementer. Litigation cost asymmetries tend to lead to unfair settlements when those costs are high relative to the value of the invention, so that the implementer’s main leverage is the threat to impose high litigation costs on the patentee. The prospect of injunctive relief as a remedy, as opposed to an ongoing royalty, will shift that balance only when the extra costs imposed on the implementer by injunctive relief as opposed to an ongoing royalty – that is, the holdup costs – are large enough to counterbalance the litigation cost asymmetry. This means granting injunctive relief would not help the patentee in those cases in which the holdup threat is relatively small. With extremely high costs and delay, injunctive relief becomes entirely irrelevant.Footnote ¹⁸⁵

Injunctive relief might tilt the balance substantially, even in the face of high litigation costs, if there is a very large potential for holdup. But allowing holdup may not be a proportionate response, for example if the implementer was not aware of the patent when it infringed, or if the implementer had a good-faith belief that the patent was invalid. This suggestion is more justifiable where ex ante negotiations were feasible, so that the result of granting injunctive relief is to induce negotiation. But in that case the holdout argument is primarily a supporting rationale for the view that injunctive relief should be preferred when ex ante negotiations are possible.

Most of the analysis of litigation costs in the holdup context has assumed the American rule that each party bears its own costs. Fee shifting may be a more effective way of addressing the holdout problem raised by asymmetric litigation costs, though it raises its own problems.Footnote ¹⁸⁶

2 Asymmetric Stakes

Golden (Reference Golden2007) suggests that there is an inherent asymmetry in the amount at stake in patent litigation because a patentee who is unsuccessful in litigation will lose not just the revenue from that one deal, but also from other potential licenses if the patent is held to be invalid.Footnote ¹⁸⁷ However, this does not reflect undesirable leverage; it merely reflects the point that the negotiated royalty should reflect the probability that the patent is invalid.

1 Theory

The problem of Cournot complements arises in principle whenever multiple independent suppliers with market power sell complementary inputs; Cournot’s example was suppliers of copper and zinc, which is combined to make brass.Footnote ¹⁹⁸ The price decisions of each supplier impose a negative externality on other suppliers; as one supplier raises its price, demand for the product decreases, thus decreasing revenue for the other suppliers. If the suppliers price independently they will not take this externality into account, and the resultant aggregate price will be higher than would be charged by a single supplier that owned all the inputs.Footnote ¹⁹⁹ Consumers will be worse off, and the suppliers (patentees) themselves will also be worse off than if all the inputs were supplied by a single firm.

The Cournot problem does not arise unless there are multiple input owners, and it becomes worse as there are more independent input suppliers. In the patent context, the problem does not turn on the number of complementary patents, but on the number of independent price-setting owners of those patents. This implies that there will be no Cournot complements problem in an industry in which a single entity owns all the complementary patents. By the same token, the problem is mitigated or eliminated if some or all of the input owners coordinate their prices. That is, the extent of the problem depends on the number of patent owners who are independently price-setting.Footnote ²⁰⁰

The Cournot complements problem does not require that the inputs are strict complements; it arises to some degree whenever the demand for one input depends on the demand for the other, so that an increase in the price of one affects demand for the other.Footnote ²⁰¹

Nor does the problem of Cournot complements turn on the presence of sunk costs holdup. However, in the absence of sunk costs holdup, Cournot price-setting alone cannot result in prices greater than the value of the patented technology. If the inputs are strict complements, then the aggregate royalty cannot exceed the combined value of the patented technology to the product to at least some users,Footnote ²⁰² though other users will be priced out of the market. The royalty is nonetheless excessive in the sense that it is higher than the price that would be set by a single firm holding all the relevant patents.

In principle then, the loss to society comes from reduced output, rather than implementers refusing to enter the market at all. However, to the extent that Cournot price-setting exacerbates sunk costs holdup, then it may result in implementers not entering the market at all.

2 Mitigating Mechanisms

1 Overview

Case studies in which arguably excessive royalties were demanded are often advanced as evidence of holdup or royalty stacking. There are two general concerns with case studies. One is that without a sound benchmark for the optimal royalty on the facts of a particular case, it may be difficult to say whether any particular royalty is too much. A second concern is that even if a particular case does illustrate pernicious royalty demands, a single example does not establish that there is a systemic problem. With that said, identifying what might be isolated instances of holdup or royalty stacking remains important as the courts may wish to respond to holdup or royalty stacking if established on the facts of a particular case, even if the problem is not systemic.

2 Distinguishing Holdup and Stacking

In case studies it can be difficult to distinguish holdup from royalty stacking. We may be able to conclude that a particular royalty is excessive because it implies an aggregate royalty for multiple patented technologies that would be excessive. But the royalty may be excessive, even without the Cournot complements problem, because the individual royalties are excessive due to holdup, or it may be excessive even without holdup as a result of the Cournot complements problem. And of course a combination is possible, in which the Cournot complements problem exacerbates individual holdup. The fact that the aggregate royalty is excessive does not in itself allow us to distinguish between these cases.Footnote ²⁰⁹

The main problem in identifying royalty stacking in a particular case is to establish a sound benchmark for what is a reasonable aggregate royalty on the facts. The appropriate benchmark to address the Cournot complements problem is the royalty that would be charged by a single patentee holding all the relevant patents. This benchmark can be approximated by a successful patent pool. Like a single patentee, a pool will seek to maximize its revenue by considering the trade-off between a high royalty and widespread adoption of the standard. But, as noted above, pools face significant hurdles to their formation, and a relevant pool often doesn’t exist. Moreover, patents that are excluded from a pool may be systematically different from those that are included. For example, if the pool in question distributes the royalties to individual patentees purely on a numerical basis, without consideration of the value of the particular patent, patents that are particularly valuable to the standard may not be adequately compensated by the pool rate. When a patentee stays out of a pool and demands a higher than pool rate, this might be because it had an average or weak patent and it was seeking to hold up implementers, but it might also be because it had a particularly valuable patent and the pool rate was not adequate. An assessment of patent quality is needed to distinguish between these possibilities.

3 Case Studies

Lemley & Shapiro (Reference Shapiro2007a) provide two examples of holdup. The first is Rambus charging “a 0.75% royalty rate for patents that do not cover industry standards and 3.50% for patents that do cover industry standards.”Footnote ²¹⁰ However, as Denicolò et al. (Reference Denicolò, Geradin, Layne-Farrar and Jorge Padilla2008) point out, this misunderstands the facts in the Rambus litigation; both sets of patents covered standards, and the difference in royalty rates was due to the fact that the latter incorporated more patented components.Footnote ²¹¹

Their second example, RIM’s settlement with NTP for $612.5 million, is more persuasive.Footnote ²¹² The settlement was eighteen times the jury award, and the parties would have had to anticipate a twelve-fold increase in sales going forward for the settlement to correspond to the reasonable royalty damages awarded by the jury.Footnote ²¹³ Unless an extremely rapid growth in sales was plausible, or if the jury had grossly underestimated the value of the patented technology in its reasonable royalty award, this is very suggestive of holdup.

Lemley & Shapiro (Reference Shapiro2007a) also provide two case studies relating to standards, in addition to Rambus. The first relates to 3G Cellular Technology, in particular the WCDMA (3GPP) and CDMA2000 (3GPP2) standards.Footnote ²¹⁴ They note the large number of patent families associated with each standard, owned by at least forty-one different companies.Footnote ²¹⁵ This indicates that the structural requirements for royalty stacking are satisfied. Lemley & Shapiro then cite estimates in the range of 20 percent of the price of the phone as the total cost of the relevant licenses. Denicolò et al. (Reference Denicolò, Geradin, Layne-Farrar and Jorge Padilla2008) dispute the accuracy of the aggregate rate, citing sources suggesting it is close to 5 percent.Footnote ²¹⁶ More fundamentally, they note that even if 20 percent were the true aggregate rate, this figure in itself does not tell us that the royalty stack is excessive. Much of the value of a cell phone lies in the patented technology, and it is not obvious that 20 percent is too high for the central functionality of a phone. The value of the intellectual property in a book is not excessive simply because it is a multiple of the value of the physical medium in which it is embodied, even if that multiple is very large. The rates themselves, without any objective estimate of the value of the patented technology, are not helpful. Further, the 3G technology at issue was widely licensed and achieved substantial market penetration,Footnote ²¹⁷ which suggests that holdup and stacking did not have serious adverse effects; though, as always with case studies, it might be said that uptake would have been even greater in the absence of stacking.

The second case study provided by Lemley & Shapiro (Reference Shapiro2007a) is of the IEEE 802.11 family of Wi-Fi standards.Footnote ²¹⁸ Again they note that numerous patents held by multiple companies are essential to this standard, which suggests that stacking is potentially a problem, but the only evidence they give that the stacked royalties are actually excessive is that one patentee was awarded a 6 percent royalty after litigation.Footnote ²¹⁹ Geradin et al. (Reference Geradin, Layne-Farrar and Padilla2008) point out that without knowing how important the patent was to the standard, we can’t say from the rate alone whether the royalty was excessive.Footnote ²²⁰ More generally, this verdict may have been an outlier.Footnote ²²¹ Courts, and juries, sometimes make mistakes. As Lemley & Shapiro (Reference Shapiro2007b) point out, only systematic errors will affect negotiating incentives.Footnote ²²² A single error, even if it is a significant outlier, will not substantially affect expected outcomes or negotiated royalties.

Cotter (Reference Cotter2009) provides several other possible examples of “patent ambush,” in which patentees were alleged to have induced an SSO to adopt a standard that incorporated patented or soon-to-be patented technology, “and then, once lock-in has occurred, demanding higher royalties than the patentees would have been able to negotiate ex ante.”Footnote ²²³ It seems clear that in these cases the patentees were attempting to get a higher royalty by negotiating after the standard was adopted, but this does not necessarily illustrate sunk costs holdup, as opposed to network value appropriation.Footnote ²²⁴ As discussed above, the value of a patented technology increases after it is adopted as part of a standard even in the absence of any sunk costs, simply because the technology is more likely to be widely adopted. Siebrasse & Cotter (Reference Cotter2017a) argue that allowing a patentee to capture some part of this network effect value is unobjectionable from a policy perspective. It is, in any event, a distinct effect, as holdup may allow a patentee to capture more than the value of its technology to the implementer, while the network effect does not. Without a more detailed assessment of the facts, we cannot say whether the ex post increase in royalty demanded was due to network effect appropriation or sunk costs holdup.

Other suggestive examples are provided by recent litigation. In Microsoft Corp. v. Motorola, Inc.,Footnote ²²⁵ Motorola had asked for a royalty of 2.25 percent of the end-product selling price for licenses to its patents that were essential to Wi-Fi and video standards. This would have amounted to a royalty of $5.85 for an Xbox, for the Wi-Fi SEPs alone.Footnote ²²⁶ Judge Robart found that a reasonable royalty was only 3.5 cents per unit, and an upper bound on a reasonable rate was 19.5 cents.Footnote ²²⁷ Therefore, Motorola’s demand for the Xbox was a minimum of thirty times greater than what Judge Robart found to be reasonable, and perhaps as much as sixteen hundred times greater.Footnote ²²⁸ Not surprisingly, the U.S. Court of Appeals for the Ninth Circuit said that there was “evidence from which the jury could infer that demanding a 2.25% royalty rate was not a good-faith effort to realize the value of the technology, but rather an attempt to capitalize on the value of the standard itself – that is, to obtain the hold-up value.”Footnote ²²⁹ If we accept Judge Robart’s FRAND rate determination as even roughly accurate, it is difficult not to see this as an instance of holdup of some kind. The video SEPs are a particularly compelling example, because the patents related to interlaced video, which is largely obsolete, and so the technology added very little value to the standard.Footnote ²³⁰

Lemley & Shapiro also note that a patent pool, Via Licensing, has been set up “[i]n an attempt to deal with the problem of patent stacking for 802.11 products.” That is, they cite the existence of a patent pool as evidence of the royalty stacking problem. On the other hand, in their review of the evidence, Geradin et al. (Reference Geradin, Layne-Farrar and Padilla2008) find there is little evidence of systemic problems of royalty stacking within standard setting “that are not already adequately dealt with through existing mechanisms, including … patent pools” among other mechanisms. In effect, Lemley & Shapiro cite the existence of a pool as evidence that there is a problem, and Geradin et al. (Reference Geradin, Layne-Farrar and Padilla2008) cite the existence of a pool as evidence that there is not a problem. More accurately, Geradin et al. (Reference Geradin, Layne-Farrar and Padilla2008) do not deny the existence of the problem,Footnote ²³¹ but they argue it has been adequately addressed.

Even those who are skeptical of whether holdup and royalty stacking are systemic problems generally do not deny that they may occur in individual cases. It is therefore somewhat surprising that there are not more clear-cut individual cases, though that may be in part because the difficulty of assessing whether a royalty is excessive cuts both ways, and because most negotiations remain confidential. With that said, the individual cases taken together are at least strongly suggestive that excessive royalty demands resulting from holdup and/or royalty stacking do occur, at least on occasion.

1 General

Empirical studies generally do not establish that holdup and royalty stacking are serious systemic problems. Geradin et al. (Reference Geradin, Layne-Farrar and Padilla2008) review the empirical evidence relating to the semiconductor, software, and biomedical device industries, and find no clear evidence that anti-commons and royalty stacking are significant problems.Footnote ²³²

2 Holdup

The most important recent study is that of Galetovic et al. (Reference Galetovic, Haber and Levine2015), which examines SEPs in particular. They examine two empirical implications of the SEP holdup hypothesis. First, if holdup in the standards context is slowing the rate of innovation, then products that are highly reliant upon SEPs will experience slower rates of decrease in quality-adjusted prices than similar products that do not. Second, they consider the quasi-natural experiment resulting from the 2006 Supreme Court of the United States decision in eBay Inc. v. MercExchange, LLC,Footnote ²³³ which made it more difficult for SEP holders to obtain injunctions against infringers than for the holders of non-SEP patents. They find no evidence of SEP holdup on either test. With respect to the comparison between industries, they find:

On the second test, they use a difference in differences specification to test whether quality-adjusted prices fall faster in SEP-reliant industries after eBay, while controlling for industry and year effects. Their analysis does not allow them to reject the null hypothesis that eBay did not differentially affect SEP-reliant industries.

These results imply that holdup is not systemically impeding innovation in SEP-reliant industries. There are two caveats to these results that are potentially relevant to remedial issues. First, they do not claim that individual firms never attempt to engage in behavior that can be characterized as holdup.Footnote ²³⁵ Courts may wish to respond to individual instances of holdup, even if it is not a systemic problem.

Secondly, they do not take issue with the view that the theoretical conditions for holdup exist in SEP-reliant industries, which suggests that it is some mitigating mechanism that explains their results. One possibility is that systemic holdup has been avoided as a result of structural factors such as the prevalence of ex ante bargaining or repeat play mechanisms. On the other hand, we have seen that it is sometimes suggested that it is legal constraints, such as the FRAND commitment, that mitigate the effect of holdup. That hypothesis is broadly consistent with the result that the prices of SEP-reliant products have fallen at rates that are fast relative to other patent-intensive products that are not SEP-reliant. It is more difficult to reconcile with the result that eBay has had no observable effect on holdup, but it is possible that eBay was effectively anticipated in the context of SEPs. That is, it may be that even before eBay, implementers understood that the FRAND commitment meant what it said and that they would be able to use standards subject to the FRAND commitment without fear of being held up by injunctions or excessive royalties.

From a remedial perspective, it matters what the particular mechanism might be. If structural factors are at play, this would suggest that the courts should be relatively reluctant to withhold injunctive relief to a successful patentee. On the other hand, if it is the FRAND commitment that is avoiding holdup in SEP-reliant industries, the results of Galetovic et al. (Reference Galetovic, Haber and Levine2015) show that the FRAND system is working, but it might suggest that the courts should continue to apply the FRAND principles relatively aggressively in order to ensure that the system keeps working. This might also suggest that the courts should apply a similar reluctance to grant injunctions even in respect of patents that are not FRAND committed, if the potential for holdup is otherwise present. The other side of that coin is that it is also possible that the FRAND commitment has been applied too aggressively, resulting in an inadequate incentive to invent. There appear to be no systemic studies addressing that possibility, though it is likely too soon for incentive effects to have manifested themselves.

3 Royalty Stacking

Galetovic & Gupta (Reference Galetovic and Gupta2017) empirically investigate royalty stacking, and the Cournot complements problem in particular, in the world mobile wireless industry, focusing on third generation (3G) and fourth generation (4G) wireless cellular standards defined by the third generation partnership project (3GPP). Their paper draws on the fact that the number of SEP holders and the number of SEPs have grown dramatically over the life of this technology: “During the last 20 years the number of SEP holders for 3G and 4G standards grew from 2 in 1994 to 130 in 2013 and the number of SEPs rose from fewer than 150 in 1994 to more than 150,000 in 2013.”Footnote ²³⁶ Cournot complements theory implies that with the increase in the number of SEP holders, royalty stacking would have gotten worse. In particular, they note that the price of phones should increase or (if quality increases demand) at least stagnate; that margins of SEP holders and downstream manufacturers will fall; and that the number of device manufacturers will decrease and industry concentration will rise. They find none of these effects. On price, for example, they find that “between 1994 and 2013 and controlling for technological generation, the real average selling price of a device fell between −11.4% to −24.8% per year. Moreover, the introductory average selling price of successive generations fell.”Footnote ²³⁷ They also find no trend in margins, and that industry concentration fell.Footnote ²³⁸ There are many other variables that might also affect the price of phones. Most obviously, the quality of phones has increased, raising willingness to pay, and manufacturing costs have probably decreased, and other factors such as incomes, substitute prices, and downstream intensity of price competition have also changed.Footnote ²³⁹ However, in their model, such changes cannot explain the price decrease and other observed effects, because when stacking is severe, the stacked royalty will increase to extract any benefit from cost reductions or increased demand.Footnote ²⁴⁰

Galetovic & Gupta portray these results as indicating that royalty stacking has not been a systemic problem in the wireless industry, despite the large number of SEP owners. This raises a puzzle: How is this result to be reconciled with Cournot complements theory? The general Cournot complements model developed by Galetovic & Gupta (Reference Galetovic and Gupta2017) shows that “even with a modest number of SEP holders, the effect of royalty stacking on output is severe and eventually, output collapses.”Footnote ²⁴¹ As they observe, the modern wireless industry has a large number of complementary inputs in the form of SEPs, held by independent owners. This implies that the market should “nearly disappear” and yet, as they also observe, the modern wireless industry is very healthy.

Galetovic & Gupta do not attempt to resolve this puzzle. As discussed above, the Cournot complements problem might be mitigated or solved by wide-scale price coordination, perhaps through patent pools, or possibly by specific pricing strategies or practices, but it is not obvious that such factors can explain the apparent lack of royalty stacking in the wireless industry. If Galetovic & Gupta’s basic results are replicated, it is of pressing interest to explain why the wireless industry is so robust, as this might shed entirely new light on the Cournot complements problem. While Galetovic & Gupta present their work as challenging the claim that royalty stacking is a problem in complex product industries such as cellular phones, their work can also be seen as a challenge to Cournot complements theory itself.