I study the problem of a decision maker choosing a policy which allocates treatment to a heterogeneous population on the basis of experimental data that includes only a subset of possible treatment values. The effects of new treatments are partially identified by shape restrictions on treatment response. Policies are compared according to the minimax regret criterion, and I show that the empirical analog of the population decision problem has a tractable linear- and integer-programming formulation. I prove the maximum regret of the estimated policy converges to the lowest possible maximum regret at a rate which is the maximum of N^-1/2 and the rate at which conditional average treatment effects are estimated in the experimental data. In an application to designing targeted subsidies for electrical grid connections in rural Kenya, I find that nearly the entire population should be given a treatment not implemented in the experiment, reducing maximum regret by over 60% compared to the policy that restricts to the treatments implemented in the experiment.
Contextual Bandits in a Survey Experiment on Charitable Giving: Within-Experiment Outcomes versus Policy Learning
Stanford Graduate School of Business(斯坦福大学商学院);Amazon(亚马逊);Stanford Management Science and Engineering(斯坦福大学管理科学与工程系);Coupang(库胖);University of Toronto Computer Science(多伦多大学计算机科学系)
Data Analytics Team, Mizuho–DL Financial Technology, Co., Ltd.(摩根大通数字金融技术公司数据分析团队);Department of Economics, Northwestern University(西北大学经济学系);Graduate School of Economics and Management, Tohoku University(东北大学经济学与管理学院);Department of Economics, Brown University(布朗大学经济学系)