Essays on strategic experimentation

Abstract

This dissertation investigates the complexities of strategic experimentation in a variety of contexts, combining insights from two chapters that explore sequential experiments with incomplete information and the interplay between online learning and economic learning costs. By synthesizing these findings, the dissertation provides a comprehensive understanding of the challenges and opportunities presented by strategic experimentation in decision-making processes.

The first chapter source examines sequential experiments in a Bayesian learning framework, focusing on the "entanglement" effect arising from incomplete information about the outcomes of sub-experiments. The optimal experimentation policy is characterized by the three-dimensional state space of entanglement strength and the marginal beliefs about the two bandits. The analysis reveals the significant impact of entanglement on the optimal policy and its dynamics, as well as the positive effect on the continuation value for the decision-maker.

The second chapter source delves into the scenario where online learning meets economic learning costs in a bandit-over-bandit setting with potential endogeneity. The agent faces a dynamic trade-off between learning efficiency and learning costs, which is influenced by the magnitude and frequency of experiments. The optimal experiment scheme is analyzed up to order, providing valuable theoretical results.

Overall, this dissertation offers a novel perspective on strategic experimentation by addressing the challenges and opportunities presented by incomplete information and economic learning costs. The findings contribute to a deeper understanding of how decision-makers can effectively navigate the complexities of strategic experimentation to optimize their decisions in various contexts.