A1384
Title: New computational methods for multivariate regression
Authors: Daeyoung Ham - University of Minnesota (United States) [presenting]
Adam Rothman - University of Minnesota (United States)
Brad Price - West Virginia University (United States)
Abstract: New methods are proposed for multivariate regression when the regression coefficient matrix is sparse and the error covariance matrix is dense. It is assumed that the error covariance matrix has equicorrelation across the response variables. Two procedures are proposed; one is based on constant marginal response variance (compound symmetry), and the other is based on general varying marginal response variance. Efficient and fast approximation procedures are also developed for high dimensions. The procedures only require a selection of one tuning parameter, as apposed to the existing joint optimization methods. An approximate Gaussian likelihood minimization is also proposed to guarantee the stability of tuning parameter selection. Through extensive numerical studies, the procedure is showcased to outperform relevant competitors in quality of estimation (prediction) as well as computational cost. With additional comprehensive simulations, the procedures are demonstrated to be robust to model misspecification.
