Patent Pool Licensing Project
This page documents the Patent Pool Licensing Project.
The project members are: Ed Egan, David Teece, and Ed Sherry. Access to this page and all sub-pages are restricted to project members.
- 1 Literature Review
- 2 Research Question
- 3 To Do
- 4 Basic Economics
- 5 Empirics
As a first-step towards this project, Ed Egan did a reasonably thorough lit-review. The results are sorted by topic and listed as BibTeX entries on the Patent Pool Literature Review page.
The basic research question is: Are licensing rates lower for patents that are held in pools?
The question is important to practitioners as patent pool licensing rates are used, increasingly frequently, in F/RAND licensing disputes.
And the question is important to economists as:
- There are normative reasons why licensing rates should be lower for patents held in pools.
- Patent licensing is currently poorly understood (what are typical rates?)
- Patent pools are growing in popularity and are an increasingly important part of the licensing ecosystem.
We need to delimit the scope of this project. The single biggest decision is whether or not we want to include an empirical estimation or do some modelling. The paper will be dramatically different accordingly. There are three obvious possibilities:
- A case-based paper, with written theory and a small number of examples.
- An empirical paper, with written theory and solid empirics.
- A theory paper, where we outline each of the arguments and provide formal modelling that captures at least the major thrust of the arguments.
In all cases, we need a literature review done. A solid first pass at one is here: Patent Pool Literature Review. Likewise, in all cases we will need to get the basic economics down. Please read the section below and start fixing and adding material (or email it to me and I'll do it, if you aren't comfortable with the wiki).
The case-based paper is quick and easy, but isn't going to place into a good journal. The empirical paper, if done well (see the product-based sample, below), could place very well. A theory paper is only worthwhile if there is something sufficiently interesting going on to merit a model (most likely in the interaction between complements and network effects, etc).
There are a number of reasons why patents held in pools should charge lower rates than non-pool licensed patents. These include:
- Anti-trust scrutiny
- Cournot complements
- Exogenously given sharing rules
- Vertical (dis-)integration
- Network effects
- Substitution protection
These reasons may apply more or less in 'robust' versus 'fragile' pools. Each is currently a sketch outline - please add or fix each point as appropriate! Likewise, we should consider if there are reasons why patents held in pools might charge higher rates. Possibilities include:
- Enforcement of a defacto monopoly (when the pool is linked to a standard)
Patents in pools must cover complements. Including two substitute patents in a pool would be price-fixing (i.e., using the pool to set a price above that which would result from competition). Furthermore, because the DOJ will be aware of the licensing rates and must 'approve' (i.e., not prosecute) them, one would assume that they are 'competitive' and low.
For a basic model of Cournot complements applied to (blocking) patents, see Shapiro (2001), Navigating the Patent Thicket: Cross Licenses, Patent Pools, and Standard Setting. [(pdf from NBER)].
The essential argument is that by including patents in a pool, multiple entities can avoid charging higher prices (and receiving lower profits) that arise from a failure to coordinate (as a result of 'balkanized rights').
The choice to participate in a patent pool is clearly endogenous. An entity opts to include a patent in a pool (subject to the pool's approval) if and only if this leads to higher profits.
The marginal cost (to a patent holder) of issuing an additional license is effectively zero in a patent pool, as the cost is incurred by the pool administrator (Is there a pass-back of costs?). Note that this excludes 'strategic costs' in the form of lost future rents from the license. On the other hand the marginal cost of issuing a licence to a non-pool patent can be considerable (again ignoring strategic costs). Therefore, in equilibrium, firms will be will to gain less marginal benefit from a patent pool license than a non-patent-pool license, and patents with lower marginal benefits will select into the pool.
Exogenously-given sharing rules
Sharing rules are determined at the formation of the pool. From that point on the sharing rules are essentially exogenously given. A review of 'typical' sharing rules is needed. But the essential premise to this argument is that a weak patent may be able to secure greater revenue from a pool-share than from independent licensing.
A patent might cover anything from a small component (or component process) to an entire product (or entire production process). Patents in pools always cover components. Analogous to Williamson, if, for reasons of first-order economizing, a patented component would be more valuable integrated into the firm, then the firm would not include it in the pool. Including patents in a pool may prevent a hold-up problem, as the pool can act as a cross-licensing agreement, but weakens the patent-holder's protection. Specifically, in the context of cummulative innovation, a firm holding a patent outside of a pool can prevent future hold-up on a technology, whereas a firm that places a patent in the pool can not.
The value of a technology depends on its adoption. A patent pool may aid adoption by allowing developers access to all needed complementary technologies and by assuring consumers that the technology is here to stay (particulary when a pool is linked to a standard). As such a firm may earn higher total rents from including a patent in a pool, even if its marginal rents from each unit are lower.
Furthermore, a firm may gain more rents from the sale of a product that uses a patent than from licensing revenues from the patent itself. With complex products (requiring many diversely-held patents), a patent holder may be better off promoting adoption in order to secure upstream/downstream rents.
Particularly when pools are linked to standards, there is the possibility of later substitution. Standards undergo revision, and substituted patents might not be removed from the pool (this is an empirical question). As such, a firm with a weak patent that will be suffer from substitution in the next generation of the technology may have a strong incentive to join the pool. The firm can choose between higher rents for a short duration, or lower rents for the duration of the patent term (or the pool).
Another reason why rates may be lower in pools concerns the need for 'show-how' with certain patents. Although a patent may embody the 'know-how' of a technology, the implementation of a patent can require technology-specific knowledge that is difficult (and hence costly) to transfer. Patent licenses in pools do not make provision for the transfer of show-how (right?), so patents which require costly knowledge transfer will be more likely to be licensed seperately, and at a rate that compenses the patent-holder for its show-how transfer.
In order to demonstrate that patent pool licensing rates are actually lower than non-pool licensing rates, we need a representative sample of non-pool licensing rates. This is essentially impossible on a population-basis, as licensing terms are confidential and incidences of licensing are unobservable to anyone except the licensee and licensor. The best we can do, therefore, is to find a sample with known properties. I propose the following possibilities:
- Using university licensing
- A product-based sample
- Using a lawyer's sample
BRG may also have some data that is useable providing it is suitably anonymized and processed.
The University of California Technology Licensing Program recently produced a report that to Board (of which David is a member?) that I was forwarded (email subject: PBSI May 30 - revised background material). This report claims that since 1980 UC has:
- 5,000 licensing agreements
- $2.1b in licensing revenue
We could use this data, but:
- 0.2% of licenses (10) account for 80% (1.68b) of revenue, so there isn't a lot of heterogeneity to exploit
- The data will be hard to get. Possibly David can facilitate it, but even then we will almost surely need to pay to have TTO/IPIRA/etc. staff code the data for us, as we won't be able to see the contracts ourselves (they will insist on anonymizing the data to some degree) and we will need richer detail than their summary stats (i.e., rates might vary with volume, etc.)
- It may be the case that no university patents have been included in pools, making this sample questionable (i.e. these may be more 'basic' patents or otherwise be different from those in pools in a way that we can't control for). Furthermore, the TTO explicitly factors social welfare in their licensing decisions, so they aren't a proper profit-maximizing entity. This would understate licensing rates, but might not be 'fatal'.
Regardless of the usefulness of this dataset for this project, we should probably follow up on what we could get (and then worry about what we could use it for). To the best of my knowledge, very, very little is known about univerity licensing rates.
A Product-based sample
Another, much more exciting, option would be to 'build' (i.e., pretend to build) a product that requires a number of patents, including those from one or more pools. Ideally we would want to 'implement' one or more standards and actually negotiate potential licensing agreements (i.e., get terms but don't buy rights).
We would almost certainly need:
- Human Subjects approval (see:http://cphs.berkeley.edu/review.html), which can be difficult and time consuming. But I've seen similar real-world negotiation projects done before (e.g. Antoinette's Schoar's hold-up paper).
- Assistance from an electrical engineer or other field expert (though for some technologies I might be able to do the heavy lifting).
- A lawyer - though we might be able to minimize this.
- Considerable GSI support.
This would not be cheap or easy. But we would have a well-defined sample and it would be fantastic research if we could pull it off.
A Lawyer's Sample
There have been several VC studies that have used data from a single lawyer. There is always the problem of generalizability - Is the IP licensed actually representative? Would another lawyer have negotiated different terms on the same IP? Etc... But with these caveats in mind, this is do-able. Having data from more than one law firm allows some empirical fixes (law firm fixed-effects to suck out unwanted variation). However, these types of relationships are exceeding difficult to set-up and invariably require anonymizing of the data, etc.