Today, an interesting NBER working paper by Deepak Hegde from NYU Stern and coauthors got published:
We provide evidence on the value of patents to startups by leveraging the random assignment of applications to examiners with different propensities to grant patents. Using unique data on all first-time applications filed at the U.S. Patent Office since 2001, we find that startups that win the patent “lottery” by drawing lenient examiners have, on average, 55% higher employment growth and 80% higher sales growth five years later. Patent winners also pursue more, and higher quality, follow-on innovation. Winning a first patent boosts a startup’s subsequent growth and innovation by facilitating access to funding from VCs, banks, and public investors.
They use a, by now, well established instrument to deal with the fact that patents are usually not granted randomly and patentees are therefore a selected group. It turns out that sometimes patents are in fact granted more or less randomly. The authors can show that there are substantial differences between patent examiners in the likelihood of approving a patent. These differences are attributed to personal characteristics and tastes of the persons involved, independent of the actual patent quality. Moreover, in the U.S. system examiners are assigned to a patent applications according to a fixed sequence. This means that sometimes applicants get lucky if they draw an examiner that is known to be quite lenient in her granting decisions, and sometimes it’s the opposite. This is what the authors mean when they talk about a “patent lottery”. It’s not an actual lottery, of course, but it still creates at least some randomness in the process, which helps to estimate the causal effect of having a patent or not.
And the effect seems to be huge. “55% higher employment growth and 80% higher sales growth five years later”, that’s a lot! Most of it is driven by easier access to external finance, such as venture capital funds. This is in line with signaling theory, that patents help to convince financiers to invest in a start-up.
How do patents convey such large and persistent benefits to startups? We find that a patent grant increases a startup’s chances of securing funding from VCs by 47%, and of securing a loan by pledging the patent as collateral by 76%, within three years of the patent decision. A patent grant also more than doubles the odds of the startup raising funding from public investors through an IPO. […] Collectively, these findings suggest that patent grants facilitate startups’ access to external finance in contexts where information frictions, and thus contractual hazards, are high. A patent grant sets a startup on a growth path through funding that helps transform its ideas into products and services that generate jobs, revenues, and follow-on inventions.
However, there is a problem with the study. It lies in the fact that instrumental variables do only identify a local average treatment effect (LATE). Let me try to explain it in easy words. Different granting rates across examiners, i.e., differences in personal attitudes and tastes of individuals, only go so far. They are there—no doubt—and the paper shows this very convincingly. But even the most lenient examiner will not be able to grant all the patents that are put on his desk. Some patents are just of too low quality such that they will be a sure reject. Likewise, also the strictest patent examiner will have to grant some patents with probability one, as their quality is clear-cut. Therefore, personal attitudes and tastes will only matter for patents “at the edge”, where the decision to grant or reject is quite ambiguous.
This group of patent applicants—the so-called compliers—is possibly quite small and for sure not representative for all start-ups out there. Personally, I find it plausible that for these intermediate cases, where the quality of the idea is not very clear, a granted patent can be an extremely positive signal towards investors. For other firms, however, which are more easily judged to be of either very low or very high quality, the effect can be much lower. Venture capital funds might invest in very strong start-ups regardless of whether they have a patent or not. And nobody would invest in obviously bad firms even though they get a (worthless) patent approved.
They authors are aware of this point and discuss it in their paper:
Given that our estimates are local average treatment effects, they should not be generalized to the average startup. It is likely that our LATE estimates overstate the value of patents for the average startup.
But my gut feeling tells me that for this particular instrument the difference between a local average treatment effect and an average treatment effect—the average effect for the entire population that we’re usually interested in—is particularly high. Precisely because I think that personal attitudes of examiners matter, but only for a minority of patents with “so so” quality.
That might help to explain why the authors find such an enormous impact on employment and sales growth from a single granted patent. Other studies, that have used similar instruments (Galasso and Schankerman, 2015), came up with huge estimates too.