A1470
Title: Latent group structures and sparsity analysis in high dimensional panel MIDAS models
Authors: Shahnaz Parsaeian - University of Kansas (United States) [presenting]
Abstract: A method is developed to detect group structures and significant covariates in a high-dimensional panel mixed data sampling (MIDAS) model. The slope coefficients of the model are assumed to be heterogeneous, and group structures exist where the slope coefficients are homogeneous within groups and heterogeneous across groups. A doubly penalized least squares estimator is developed to detect the group structures and the sparsity patterns (i.e., selecting significant/relevant covariates) simultaneously without a-priori knowledge about the number of groups, group structures or sparsity pattern of covariates. It is shown that the proposed penalized estimator enjoys the oracle property and can consistently identify the group structures and the sparsity patterns in large samples. The finite sample performance of the proposed estimator is evaluated through Monte Carlo studies and illustrated with a real dataset in forecasting GDP growth across various countries.
