Inference for elliptical copula multivariate response regression models (Q2414484)

A multivariate response linear regression model is considered, where only increasing transformations of responses and covariates are observed. The joint dependence between all the observed variables is characterized by an elliptical copula. The paper extends the results of \textit{T. T. Cai} and \textit{L. Zhang} [Stat. Sin. 28, No. 2, 963--993 (2018; Zbl 1390.62099)], where the response was univariate and the observed vector was jointly normal. In the paper under review, estimators of the coefficient matrix $B^*$ are obtained in a high-dimensional setting by assuming that $B^*$ is either (a) element-wise sparse, or (b) row-sparse, and by incorporating the precision matrix $\Omega$ (i.e., the inverse of the variance-covariance matrix) of the error, which is assumed to be sparse as well. It is proven that in case (b), the estimator obtained via a group penalty outperforms the one obtained via simple element-wise penalty. Simulations illustrate this fact and the advantage of incorporating $\Omega$ when the correlation among the components of the error vector is strong. Moreover, for the high-dimensional Gaussian copula models, it is shown that the proposed estimator remains as the optimal one of the copula correlation matrix in this setting.