**Job Market Paper** Socially-valuable technologies sometimes require complementary innovations. This paper studies the development of innovations that exhibit such complementarity. At each point in time, a unit of attention is allocated across different innovation projects. The projects are completed stochastically in the form of breakthroughs. The social value of the technology depends on the set of completed projects by the time the agent decides to stop the development stage. In some cases it is optimal to develop the innovations in sequence. In others, it is optimal to develop multiple innovations simultaneously. I provide conditions that determine the efficient timing of development: sequential development is efficient when costs are high and there is more uncertainty about the innovations' rate of success. I compare the efficient allocation to the equilibrium outcome with a decentralized industry in which many firms compete for the development of the innovations.
with [Quitzé Valenzuela-Stookey](http://sites.northwestern.edu/qvr919) Decision makers frequently condition their actions on economic outcomes, e.g. asset prices, that they believe convey information about an unknown state. However the decision maker’s action, or expectations thereof, may also influence the outcome. In this paper we study the general problem of choosing decision rules mapping outcomes to actions in the presence of such feedback effects. We characterize the set of joint distributions of outcomes, actions, and states that can be implemented as the unique equilibrium by decision rules which satisfy a minimal notion of robustness to manipulation. Moreover, we show that all such equilibria are robust to model misspecification. This characterization of the feasible set greatly simplifies the problem of choosing decision rules. A simple graphical technique allows us to identify qualitative features of optimal policies. We illustrate the power of this approach with an application to corporate bailouts. The results are also useful for characterizing optimal decision rules when the requirement of unique implementation is relaxed.