| This dissertation has two chapters, one on information economics and one on game theory. The first chapter studies the scenarios where an analyst learns about a random variable by observing an ongoing rational experimentation. We assume an optimal stopping exercise with binary signals about a binary state of the world. The analyst observes a public history of experiments, but not an earlier experimentation pre-history of uncertain length. In this setting, a dynamic survivor bias emerges: the naive Bayes-updates ignoring the pre-history is more pessimistic than the sophisticated updates that accounts for all possible pre-histories consistent with an ongoing rational experimentation. We show that this bias is dynamic in the sense that the observation impacts the inference of the un-observed pre-history. In general, we find that the analyst's Bayes-optimal inference critically depends on the ordering of the signal history and the combined knowledge of the signal realizations and the experimenter's actions. My theory has implications for technology adoption in R& D settings, and formally subsumes a class of one armed bandits and the Wald experimentation problem, for instance.;The second chapter studies a static population game with strategic substitutes. I assume one dimensional continuous action with heterogeneous action cost among players. I explore the diminishing cross effect condition on the payoff function, which delivers equilibrium uniqueness and several comparative statics results---1. The equilibrium distribution of actions level rises in the first order stochastic dominance order when the type distribution falls in the first order stochastic dominance order and the dispersion order. 2. The equilibrium distribution of actions rises when the own action effect is larger. My model has applications in games with a p2p network structure and other massive social interactions with a pairwise matching nature. |
