Statistics Seminar（2013-04）

Topic:Inferences in Two-Step Semiparametric Partially Identi.ed Models

Speaker:Lin Zhu, Tsinghua University

Time:Thursday,11 April, 14:00-15:30

Location:Room 217, Guanghua Building 2

Abstract：This paper provides inference tools for partially identified semiparametric models. The main working assumption is that the finite-dimensional parameter of interest and the infinite-dimensional nuisance parameters are identified conditionally on other nuisance parameters being known. This structure arises in numerous applications and permits the use of standard tools from empirical processes theory. We develop uniform convergence for the set of estimates and use the uniformity to establish set inference. We allow for the identified set of nuisance parameters to be unknown and estimated, which in general requires to establish inference under misspecification. The latter is also useful in a formal development of sensitivity analysis. Inference is implemented with a multiplier-type bootstrap. Several examples illustrate the wide applicability of our results.