Speaker: Prof.Guangming Pan (Nanyang Technological University )

Title:A weighted estimator for the population covariance matrix

Time: Tue, May.29, 2018 , PM:4:20-5:20.

Location: Room 4318, Building No.4, Wushan Campus

Abstract:

    Shrinkage method and Thresholding method are two most popular approaches in estimating covariance matrices. Each method has its pros and cons. By taking the advantages of these two methods, we introduce a novel weighted method in covariance matrices estimation for high dimension datasets. It is applicable to a wider scope of different structures of covariance matrices. In addition, based on the linear shrinkage method, we propose a new shrinkage estimator, which extends the constraints in usual  shrinkage methods, i.e., $p/n\in(0,\infty)$. By the algorithm provided, we see that the proposed approach is efficient. Some theoretical results about new shrinkage method and weighted covariance estimation methods are also given. The competitive finite-sample performance of the proposed methods is illustrated  through extensive simulations and real data analysis.