Measuring Patent Thickets

Those interested in the patent system have long complained of patent thickets as a barrier to efficient production of new products and services. The more patents in an area, the argument goes, the harder it is to enter. There are several studies that attempt to measure the effect of patent thickets, with some studies arguing that thickets can ease private ordering. I'd like to briefly point out another (new) one. Charles deGrazia (U. London, Royal Holloway College), Jesse Frumkin, Nicholas Pairolero (both of USPTO) have posted a new draft on SSRN, called Embracing Technological Similarity for the Measurement of Complexity and Patent Thickets. Here is the abstract:

Clear and well-defi ned patent rights can incentivize innovation by granting monopoly rights to the inventor for a limited period of time in exchange for public disclosure of the invention. However, when a product draws from intellectual property held across multiple firms (including fragmented intellectual property or patent thickets), contracting failures may lead to suboptimal economic outcomes (Shapiro 2000). Researchers have developed several measures to gauge the extent and impact of patent thickets. This paper contributes to that literature by proposing a new measure of patent thickets that incorporates patent claim similarity to more precisely identify technological similarity, which is shown to increase the information contained in the measurement of patent thickets. Further, the measure is universally computable for all patent systems. These advantages will enable more accurate measurement and allow for novel economic research on technological complexity, fragmentation in intellectual property, and patent thickets within and across all patent jurisdictions.

 The authors use natural language processing to determine overlap in patent claims (and just the claims, arguing that's where the thicket lies) for both backward and forward citations in "triads" - patents that all cite each other. Using this methodology, they compare their results to other attempts to quantify complexity and find greater overlap in more complex technologies - a sign that their method is more accurate. Finally, they validate their results by regressing thickets against examination characteristics, showing that the examination factors more likely to come from thickets (e.g. pendency) are correlated with greater thickets.

This is an interesting study. The use of citations (versus technological class) will always be a limitation because not every patent in a thicket winds up being cited by others. However, the method used here (using forward and backward citations) is better than the alternative, which is using only blocking prior art.

The real question is what to do with all this information. Can it be applied beyond mere study of which areas have thickets? I suppose it could be helpful for portfolio purchases, and maybe to help decisions about whether to enter into a new technology.