Seminars

Estimation and inference procedures for single-index conditional distribution model

98
reads

Ming-Yueh Huang

2012-05-18
13:00:00 - 14:40:00

103 , Mathematics Research Center Building (ori. New Math. Bldg.)

In this talk, I will introduce a flexible single-index conditional distribution model, which was first employed by Hall and Yao (2005) to approximate the distribution function of a real-valued response conditional on multi-dimensional covariates. Based on an accumulation of square differences between the joint probability and the expected conditional probability, these authors provided a criterion to search the optimal direction. Instead of estimating the joint distribution, we use an induced counting process and its estimated conditional distribution to develop a pseudo least integrated squares estimation procedure. As one can see, both of these estimators are root-n consistent and asymptotically normal. However, our estimation procedure is easily implemented and has better performance in our comprehensive simulations. In addition, we proposed an asymptotic variance estimator, established a model checking rule, and applied the adaptive Lasso to detect insignificant covariates.