Experimentation with multiple sources of uncertainty




Fears, Jessica Elizabeth

Journal Title

Journal ISSN

Volume Title



I study experimentation under two types of uncertainty--- the quality and profitability of a risky technology. The quality of the technology can be good or bad. If the quality is good, then payoffs arrive at the jump times of the standard Poisson process. If the technology is bad it does not generate payoffs. Payoffs are stochastic and the sizes of the realizations depend on the underlying state of the economy. Some payoffs reveal the state completely, others do not. First, I consider an experimenter who chooses an irreversible exit time. I find that, after the first arrival of a payoff, the optimal stopping policy can be characterized by a cutoff belief about the true state of the economy. Before the first payoff, the stopping region of the experimenter is a subset of the space of beliefs about the technology's quality and profitability which cannot be characterized by cutoff beliefs. I find that the experimenter reacts differently to each source of uncertainty or risk, and that the most cost-effective subsidy for such an experimenter is an increase to his highest possible payoff. That is, the optimal subsidy makes the project riskier and subsidizes the experimenter when he is already performing well. Next, I consider an experimenter facing the same environment who also chooses the rate of the Poisson process after each observation of a payoff. While many of the results from chapter 1 carry through, I find that investment in the project in non-monotonic in the persistence of the states of the economy. Finally, I study an experimenter facing an environment similar to that in the first setting, but who has the option to irreversibly deploy his technology at scale, that is, to augment the payoffs in the long-run by making a large investment today. Here I find a rich set of investment and stopping policies, and that the optimal subsidy depends on the objective of the policy makers. Policy makers can encourage experimentation most efficiently by subsidizing the lowest payoff, and can encourage scaling the project by subsidizing the cost of investment.



LCSH Subject Headings