A0912
Title: An estimator of a jump discontinuity in regression based on generated observations
Authors: Feng Yao - West Virginia University (United States) [presenting]
Abstract: A new class of estimators is proposed for a jump discontinuity on nonparametric regression. While a vast amount of econometrics literature addresses this issue, the main approach in these studies is to use local polynomial (linear) estimators on both sides of the discontinuity to produce an estimator for the jump with desirable boundary properties. The approach extends the regression with generated observations from both sides of the discontinuity using a theorem of Hestenes. The extended regressions' generated observations are estimated and used to construct an estimator for the jump discontinuity that solves the boundary problems normally associated with classical kernel estimators. Asymptotic characterizations are provided for the jump estimators, including bias and variance orders and asymptotic distributions after suitable centering and normalization. Monte Carlo simulations show that the estimator for the jump can outperform that based on local polynomial (linear) regression.
