Local and Global Inference for High Dimensional Gaussian Copula Graphical Models

In this paper, we propose a unified asymptotic inference framework for Gaussian copula graphical models, to test the presence of a single edge and construct a uniform confidence subgraph for the true graph. To avoid the estimation of the unknown marginal transformation functions, we consider a pseudo likelihood approach. The core of our framework is a decorrelated pseudo score function for the parameter of interest. The proposed pseudo score function has the structure of a nonlinear transformation of U-statistics, and a delta method-type analysis is used to prove its asymptotic normality. Our theory provides asymptotic guarantees on the type I error as well as the local power of the proposed score test. We also propose a global hypothesis testing procedure based on Wald test statistic and multiplier bootstrap, to provide a confidence subgraph for the true graph. The theoretical properties of the proposed inferential methods are verified by thorough numerical experiments.