A0201
Title: Model averaging for high-dimensional linear regression models with dependent observations
Authors: Henghsiu Tsai - Academia Sinica (Taiwan) [presenting]
Abstract: The orthogonal greedy algorithm (OGA) is introduced to screen out the nested set of signal variables under a high-dimensional linear regression framework with dependent observations. To gain the prediction performance, we propose the high-dimensional Mallow model averaging (HDMMA) criteria to determine the weight for averaging these nested high-dimensional linear regression models. We further analyze rates of convergence of prediction error for the averaging model under different sparsity conditions. The contribution has three folds. First, we show that the procedure, named OGA+HDMMA, can achieve optimal convergence rates of prediction error. Second, we use simulation to show that the out-sample prediction of OGA+HAMMA can outperform the MCV method when the covariates are highly correlated or contain time-series effects. Third, the out-sample prediction of OGA+HDMMA performs comparably or even better than many well-known high-dimensional variable selection methods in some scenarios.
