A0915
Title: Sparse matrix estimation based on greedy algorithms and information criteria
Authors: Hsueh-Han Huang - Academia Sinica (Taiwan) [presenting]
Abstract: The problem of estimating the covariance matrix of serially correlated vectors whose dimension is allowed to be much larger than the sample size is considered. It is proposed using the orthogonal greedy algorithm (OGA) and a high-dimensional Akaike information criterion (HDAIC) to estimate the matrix, showing that the proposed estimate is rate optimal under a sparsity condition more flexible than the existing literature. When the covariance matrix is bandable, a banding/tapering estimate whose parameters are chosen by a novel information criterion is introduced. The rate optimality of the latter estimate is also established.
