Innovation through Competitive Experimentation

This paper studies optimal contest design for competitive experimentation. The principal seeks contributions from multiple agents whose experimentation efforts are unobservable. To induce effort, the contest must provide agents with an information rent that increases in the duration of the contest. In the optimal contest, the principal limits the agents' rent by setting inefficiently tight deadlines. I show that the contest remains optimal when breakthroughs in experimentation are privately observed, and both the principal and the agents can disclose this information strategically. When hiring is costly, then a large contest is desirable only if the value of the innovation is sufficiently high, learning is not too fast, and there is sufficient variance in the prior probability of success.