Learning When to Quit

Abstract:
We propose a model of research and development as a process of experimentation in which researchers repeatedly revise specifications of a project and update their beliefs about the project’s type. Only a good project whose type is learned by researchers can generate value. Researchers abandon a project when the opportunity costs of continuing exceed the expected benefits. We estimate the structural parameters of this dynamic optimization problem using a novel data set with information on both successful and abandoned projects from the Internet Engineering Task Force (IETF), an organization that creates and maintains standards necessary for the functioning of the internet. The structural approach allows us to recover researchers’ unobserved beliefs and opportunity costs, and answer questions about whether specific rules and institutions encourage “efficient abandonment” of ongoing projects. We find that opportunity costs are decreasing over time and feedback and comments from the IETF community at large increase the speed at which developers learn whether a project is worth pursuing.