Patents are traditionally justified by their ability to incentivize innovation while also disseminating information about the invention for use by downstream inventors. There is evidence, however, that secrecy often confers stronger protection than patents. The question, then, is why firms in such a situation patent at all. The authors of this 2007 RandJE show, however, that there is a patent strength that gives more welfare than the no patent state. The basic idea is that firms do not choose between secrecy and patents at the time of innovation, but rather between patenting the idea himself or having another independent discoverer of the innovation patent it. There is something of a prisoner’s dilemma, essentially. When I discover the new product, I do not know whether you have discovered the new product. If you haven’t discovered it also, then I just keep the innovation secret and earn monopoly profits forever under Bertrand competition (because you would earn zero profits by reinventing, and therefore will not invest toward reinvention). If you will discover it also and patent or keep secret, then I want to patent (the model assumed the patent holder is chosen by lottery under simultaneous discovery). Therefore, even if patent protection is weaker than secrecy, I may still want to patent (in order to prevent you from patenting should you simultaneously discover). It turns out that, with some simple assumptions, there is a level of patent protection such that changes in investment in R&D and consumer welfare (from more or fewer monopoly firms) is higher than under the strict secrecy case.
I see only one worrying aspect of this model. In the real world, the choice of patenting versus secrecy is a continuous time problem, not a discrete time problem. Even if there is simultaneous discovery in the sense that R&D on the same problem is simultaneous, that does not mean there is simultaneous decision as to whether to patent or remain secret. In the continuous time case, there is no simultaneous announcement of the results of the R&D process. Whoever successfully finishes the R&D first will immediately announce success, and all competitors will immediately halt R&D work (since even if they can also discover the good, they will only make zero profits). Therefore, secrecy is a dominant strategy. The model, it appears, can be salvaged by assuming Cournot competition or some other model of partial competition, rather than Bertrand competition; it seems clear from the proof derivations that this was not done because stating the equilibrium in the Cournot case will be significantly more difficult.