Title: Robust confidence sets for moment-based models
Authors: Anton Skrobotov - Russian Presidential Academy of National Economy and Public Administration and SPBU (Russia) [presenting]
Artem Prokhorov - University of Sydney (Australia)
Abstract: A new approach is proposed to construct conservative confidence intervals when moment conditions define a large set of models with varying degrees of plausibility as measured by commonly used statistics. The statistics reflect orthogonality, relative efficiency and non-redundancy of the moments for a given model and allow for a multi-criterion search over all available moment combinations. Using the most plausible model for inference requires adjustments to the usual confidence intervals. We show how to obtain the adjustments, and we illustrate that they work very well in Monte Carlo simulations.