Title: Empirical likelihood based covariance matrix estimation
Authors: Sanjay Chaudhuri - National University of Singapore (Singapore) [presenting]
Abstract: Empirical likelihood based methods are discussed for estimating covariance matrices from data. It is well known that classical estimates cannot be easily modified to include equality based constraints among its entries while keeping it non-negative definite. It is known that empirical likelihood based methods can estimate parameters with additional equality constraints on the expectation of various functionals of the data. We employ empirical likelihood to estimate covariance matrix with equality constraints on its diagonal and off-diagonal entries. We show that the proposed estimates are non-negative definite and does not depend on the underlying distribution of the observations. Such methods can also be extended to rank based constraints and to estimation of high dimensional covariances.