Incidence and Growth of Patent Thickets - The Impact of Technological Opportunities and Complexity

The term "patent thickets" has recently come to prominence. Our recent study analyzes the impact of technological opportunities and complexity on such thickets. The complexity technology is a well-known reason for patent thickets. Regarding technological opportunities, we think there is a disturbing new finding that indicates that in some areas there are declining technological opportunities, and that this decline may contribute to more patenting as firms race to secure the rights to the remaining realm of opportunities. Leasing aside the multivariate analysis in our study, we have also developed a few new measures of the extent of patent thickets.

The European Patent Office (EPO) commenced operation in 1978. It offers a harmonized application and examination path to patent applicants in Europe. There is no harmonized patent litigation, and further harmonization is currently a much contested area. Over time there has been a relatively strong increase in the number of total patent applications, and the increase has been particularly pronounced in the area of complex technologies. The distinction between complex technology and discrete technology derives from work by Wesley Cohen, Richard Nelson and John Walsh. They saw that with discrete technologies, such as pharmaceuticals, an inventor can pretty much cover with a single patent what is needed for one product, whereas an item like a TV or a communication device would typically be covered by hundreds of patents. Standardization processes and litigation arise from the fact that there is much controversy over the actual coverage of patents in the area of complex technologies.

The increase in the number of patents for discrete technologies is largely commensurate with what has happened in research and development (R&D), whereas the increase in the number for complex technologies outpaces their underlying R&D expenditures. There are some reasons for the increase in patents that have little to do with more inventions, or more R&D, or more productivity gain. Increased patenting is of concern for a number of reasons. Shapiro has pointed out that there is an increased likelihood of patent thickets: sets of interlocking, complimentary patent rights which then lead to relatively high transaction costs, for example in order to secure access to a technology or in order to resolve the issue of how the overall rents are distributed in an economic system. Funds spent on court cases could be spent elsewhere more effectively.

There is also the issue of "strategic patenting," which is used to describe certain types of patenting that could be abusive in terms of the logic of a competition authority. Patent offices are increasingly dealing with more creative applicants who devise new applications strategies, for which the offices are not designed to handle. Multiplication of filings can be a real impediment for competitors. It has been proposed that that the opposition procedure should be set up in such a way that all of these patents (e.g. derived from a single priority) can be challenged in one opposition procedure. If the applicants use strategic patenting, then opponents should have access to a countervailing strategy.

It seems that for a long time, the quality of patent applications at the EPO has been falling, with roughly constant grant rates. Only very recently has it been announced by the EPO that the grant rate is going down. Furthermore, the increase in the number of patents may have implications for competition policy. Decreasing patent quality raises uncertainty about the legal basis of licensing and joint ventures, because it becomes very difficult to distinguish between horizontal and vertical licensing cases.

The aim of the paper I am presenting today is to develop a model of patenting that allows empirical work to be undertaken in a way that is guided by theory. Complexity and technological opportunities are built into the model, and they interact to either reduce or increase patenting incentives. The analysis of panel data capturing the patenting behavior of 2074 firms in 30 technology areas over 15 years suggests that there are patent thickets in nine out of 30 technological areas in the realm administered by the EPO.

In the paper, we try to find out where patent thickets are, and whether they grow. In order to do that, we need some measures: we need to identify patent thickets. The distinction between discrete and complex came out of survey work: it is a classification that that has gained widespread usage and recognition, but it is very hard to look at patent data and write down a recipe that tells you from the patent data, rather than an imposed classification, where the patent thickets are. Following a novel approach, we identify mutually blocking relationships from citation data, and we use these new measures to complement the existing ones.

Our model is built on the following perspective: patenting happens along two dimensions. The first is opportunities: different opportunities means largely independent research paths along which patenting and R&D can occur. The second dimension is the number of facets within an opportunity: the maximum number of facets is an exogenous parameter in our model, and it measures something like the inverse of the breadth of patents. The logic of the model is that if the number of opportunities is increased, some pressure is taken off the firms to obtain patents on a lot of facets. That means that increasing the number of opportunities will reduce total patenting in complex settings. Conversely, our theoretical model indicates that in discrete technologies, an increase in technological opportunities will also lead to an increase in patenting.

We formulate three hypotheses in the paper. The first is that increasing technological opportunities reduces firms' patenting efforts more as technologies become more complex. The second hypothesis is that if complexity increases, firms' patenting efforts will go up, under conditions in which the cost of administering patents are relatively low in comparison to the benefits that the patents give. This seems a reasonable assumption in some technical fields, especially in complex technology areas. The third hypothesis is that increasing technological opportunity leads to more patenting in discrete product technologies.

We measure complexity using our new measure which we call "triples." A triple is a situation in which three patent holders block each other completely, i.e. A and B block each other, B and C block each other, and A and C block each other. We assume that such situations will lead to particularly high transactions costs. Technological opportunity is measured by the number of non-patent references in the search report. The first step that we undertake in our descriptive analysis is checking our measure of triples. Since 1980, the average number of triples in discrete technology areas has been flat and low. The numbers of triples of complex and interlocking relationships has been rising steadily. This comparison suggests that our measure distinguishes nicely between discrete and complex technologies.

We construct a sample taking patent applications between 1987 and 2000, excluding applicants who have filed very few applications. We built a series of time series observations so as to use panel data. The sample represents 25% of the firms, on average, in a particular year and field, and about 56% of the patents. That distribution is not unusual, as we know a relatively small number of firms account for much of the patenting that is going on, because patenting is heavily concentrated in relatively large firms.

In our study we find there is less patenting when the measure of technological opportunities is increased if the number of triples in the technical area is sufficiently high. This effect is more pronounced for larger firms than for smaller firms.

To conclude, our paper uses a theoretical model to investigate how patenting in complex technologies differs from patenting in discrete technologies. It is very important to develop a model that could say something about both categories, so that there could be a broader and sharper empirical test. We developed a model that goes across technologies in order to distinguish between complex and discrete technologies. The model suggests that patenting in complex technology falls as opportunities rise, or more realistically that patenting in complex technology increases as technological opportunities fall. We have found empirical evidence to support this hypothesis.

If this model can be validated further, one of the policy implications that will have to be considered is that patent office reaction or examination practice with respect to granting patents may want to take into account the distinction between discrete and complex. It seems that it may be very worthwhile to dampen the patenting in the complex arena, for example by requiring higher inventive steps in that field, in order to counteract the effects that are visible in this model.

Q: How did you classify the technical data for the 30 areas?

A: This is a classification that has been created by others, including the Organisation for Economic Co-operation and Development (OECD). It is a heuristic classification which works quite well for applied analysis. We have experimented a little bit with variations on this classification. We needed something that was manageable, and 30 seemed to be a good number. Each patent is allocated to one technology class so that we can distribute a firm's activity in any given year into 30 different categories - this leads to our unit of analysis. Very few firms are active in all categories.

Q: Isn't the number of applications determined by complexity?

A: Yes, as the results tell you, complexity has an effect on patent filings. The direct effect is positive, and there is an additional interaction effect of technological opportunity and complexity. You may worry about the endogeneity of the variable. We also try to control for that by treating the variables as endogenous or as predetermined, and by instrumenting them via our generalized method of moments (GMM) estimator. You could write down a model which postulates that for some reason the standards of examination are slackening, more and more patents are filed, then the applicants scramble to get more marginal things in that have fewer non-patent references. We have not modeled it at that level of detail, but we tried to control for endogeneity in the econometrics.

Q: Could you please elaborate on the interaction effects?

A: Perhaps examining the profit function of the firms reveals a little bit more about what is going on. Within each opportunity, the maximum value that you can get out of the opportunity is V. But since several firms may have patent rights within that opportunity because the patents do not belong to one single firm, they somehow have to allocate this V. This is done by our sharing rule: si is the share of the patents of one company, and ω is the sharing rule. This is a short-term notion for the firms coming together to negotiate or even to litigate patents. If one firm's application activity goes up, and its share of patents rises, then it will get a higher share of V. The fewer Vs that are out there, and the higher the complexity of the technological area, the harder firms will fight for the ones that are left. That is the intuition for the interaction effect.

Also, as Rosemarie Ziedonis has argued, the higher the share, the lower may be the legal costs. To get the value V out of any opportunity, it is important that I have, if possible, a high share of those facets. This is the intuitive interpretation of what happens in cross-licensing negotiations, for example: when firms get together, put their patents on the table, and then try to set licensing fees.

Q: Do you have any data about the correlation of triples?

A: We created this indicator because we were searching for something that would capture transaction costs. We need to validate this measure more in terms of behavioral consequences. But from the two figures I presented, two statistical associations are clear: triples are positively correlated with filing growth, and they are negatively related to technological opportunity in complex fields. We are discussing right now what kind of additional research we could undertake to validate the triples measure further.

Q: Can we use pair relationships instead of triples?

A: No. Triples have explanatory power that is not in the pair relationships. Pairs do not explain filings growth to the same extent as triples. Hence, the logic of the variable - that it captures complex bargaining situations with high transactions costs - appears to hold.

Q: How do you relate this to more and more companies moving toward open innovation? Will this new trend affect the findings that you have drawn from these analyses?

A: It is difficult to say exactly what open innovation means. There may be two interpretations. The first could be described as "markets for technology," which means the innovation process is increasingly fragmented along the value chain: suppliers will do more, other firms will provide the complements. In order for that to work we need more intellectual property (IP), and that requires as little uncertainly as possible about the value of that IP. Having lots of patents out there is not necessarily helpful.

The other interpretation of open innovation goes in the direction of open source mechanisms that move away from formal intellectual property, and work more on the basis of reciprocity. This might produce a completely different effect on the data that we are not controlling for, and that is not in our model.

Open innovation is a term that is terribly ill-defined because it has these two possible interpretations. The second interpretation stipulates that a firm is not working with IP, whereas the first one says it is working with more IP. It is odd to have these two definitions under one roof.

Q: How can a "prisoner's dilemma" be discouraged?

A: As soon as you talk about optimality, what we would really need to have is a complete model with a welfare function, with an optimality criterion that would tell us what the optimal policy in dealing with patents would be. No one has that yet. There is no fully developed model that models the institutional levers that a patent office has, and allows us to maximize welfare.

If we had such a model we would be able to answer the following questions. Is a registration system or a high intensity examination system better? We have examination systems, for example, in the area of utility model patterns, and trademarks. Those are not really as contested right now as are patents and patent examination. If it is true that only a few patents are valuable, why should we spend so much money on examining all of them? They could be registered; then an inexpensive arbitration or court mechanism could ensure that the valuable ones are examined in some form. All the other ones where there is no conflict could be left to die. The alternative is intensive examination and very clear delineation, which is very costly and involves thousands of patent examiners working very hard.

If the inventive step requirements for complex technologies were increased, in order to have an administrative dampener on patenting facets, would that lead to welfare increase? I speculate that it would, but we have not developed a full model, so right now this is mere presumption. It is fruitful to think along these terms in order to give, ultimately, patent offices a better and more detailed economic rationale for what they are doing. Many of our economic rationales for patents are very good, but not very detailed, and are 30 or 40 years old. But they do not model the levers that the patent office officials have nowadays. So they cannot say much beyond what is the statutory term of the model.

Q: What about the group of patents examination procedure?

A: If patents are increasingly similar because they are derived from one priority, patent offices should allocate them to one examiner to examine. That examiner would need guidance as to how to handle the portfolio build up, because the current guidelines are written for the examination of individual patents.

Another proposal dealing with portfolio strategies is one that gives rivals the opportunity to oppose each and every one of the patents in the group in a single opposition procedure. At the EPO at least, the opposition procedure is set up to be a one-by-one exercise. While in court proceedings cases can be pooled, in opposition they cannot. I am making these proposals in order to alert patent office officials to the fact that the thinking of the applicants is different from the thinking of those who designed the office rules. It is no longer a world in which the individual patent counts; it is a world where the portfolio counts.

There are further proposals that include a portfolio tax, which would mean an increase in fees commensurate with the growing size of the portfolio. The problem with that is that the size of the portfolio could have many reasons: for instance, it could be full of separate discrete technologies. Before we come to taxation, we have to measure whether there is really a connection between these patents.

I would like to see more discussion among the examiners about what it means to do patent examination in a world in which patents are increasingly linked.

Q: Am I correct in thinking that your model controls for new entries from emerging economies into the patenting activities, especially in the complex technology areas?

A: We have almost none of these new entrants in the empirics, and certainly not in the model. We treat industry structure as fixed in the short run. Korean and Chinese firms command a rapidly increasing share of filings at the EPO. The share is in the single digit percentages, but the growth rate is high. However, new entrants are not a major determinant in our data, because our data starts in the mid-'80s and goes through to 2002. During that time distribution across countries has been relatively stable: the U.S., Japan, and Germany were the largest filing nations at the EPO. Remember, though, that we do our analysis by firm in a technology, so although new entrants will be very important, they do not drive our results at this point.

Q: If you were to do a similar analysis going back further in history, what would your graph look like?

A: I am not sure. The problem here is that the EPO started operation in 1978. Frankly, I have not thought about this, which would entail going to the national data. The problem with going to the national data is that you lose the possibility of making comparisons. It may be possible to use JPO or USPTO data for such an exercise - it is a very interesting proposal.

Q: Only pharmaceuticals have a very big negative for triples. Why?

A: You refer to the data in Table 6 where we compute the marginal effects due to a change in the right-hand side variables, and here we come up with signs in some technological areas which are not always easily interpreted. We are still investigating the reasons for some of these effects. But taken the estimate literally, it suggests that in pharmaceuticals, an increase in the number of triples has a strong negative effect on patenting, while the effect is positive and large in what we would usually refer to as complex technologies. Note that in this case, the results are also confirmed in the fragmentation variable - more fragmentation discourages patenting in pharmaceuticals.