Events and Seminars : Upcoming Seminars

Nonparametric Threshold Regression: Estimation and Inference

CHRISTOPHER PARMETER, PH.D.
Associate Professor
Department of Economics
University of Miami
TUESDAY, OCTOBER 6, 2015
11:00 a.m.–12:00 p.m., CRB 692

Threshold models provide useful references in the medical sciences, examples including blood pressure medication if a patient’s level is higher then a prescribed cutoff. Much is known about estimation and inference in threshold models when the underlying relationship is parametric.The present work describes a simple approach to estimating the location of a threshold/change point in a nonparametric regression. This model has connections both to the time-series and regression discontinuity literatures as well. The estimator leverages a simple decomposition, giving it the form of a semiparametric smooth coecientmodel. Further, we provide general results for weakly identified semiparametric M-estimators, for which our estimator is contained. We discuss resampling methods toobtain confidence bounds to perform inference and present Monte Carlo results to demonstrate the merit of the method.