|
Activity Number:
|
338
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, July 31, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #309965 |
|
Title:
|
An Empirical Likelihood-Based Approach To Estimate Covariance Matrices with Structural Zeroes
|
|
Author(s):
|
Sanjay Chaudhuri*+
|
|
Companies:
|
National University of Singapore
|
|
Address:
|
Dept of Statistics and Appl Prob, Singapore, 117546, Singapore
|
|
Keywords:
|
empirical likelihood ; covariance matrices ; covariance graphs ; constrained maximisation
|
|
Abstract:
|
Zeroes in the covariance matrices of a multivariate random vector indicates linear independence between the corresponding components. Due to the presence of structural zeroes a maximum likelihood estimator cannot be expressed analytically and can only be conveniently iteratively computed, if the underlying distribution is Gaussian. Currently same or related algorithms are used to compute an estimator if the underlying distribution is non-Gaussian. In this talk we present an empirical likelihood based estimate of the covariance matrices, which preserves the structural zeroes. This estimator is unique, consistent and convenient to apply in a variety of situations. Moreover we shall show that, it is more efficient than the other estimators for non-Gaussian case and slightly less efficient than the mle for the Gaussian case. Some generalizations of the methodology will also be discussed.
|
