Testing Separability of High-Dimensional Covariance Matrices

Abstract: Due to their parsimony, separable covariance models have been popular in modeling matrix-variate data. However, the inference from such a model may be misleading if the population covariance matrix $\Sigma$ is actually non-separable, motivating the use of statistical tests of separability. Likelihood ratio tests have tractable null distributions and good power when the sample size $n$ is larger than the number of variables $p$, but are not well-defined otherwise. Other existing separability tests for the $p>n$ case have low power for small sample sizes, and have null distributions that depend on unknown parameters, preventing exact error rate control. To address these issues, we propose novel invariant tests leveraging the core covariance matrix, a complementary notion to a separable covariance matrix. We show that testing separability of $\Sigma$ is equivalent to testing sphericity of its core component. With this insight, we construct test statistics that are well-defined in high-dimensional settings and have distributions that are invariant under the null hypothesis of separability, allowing for exact simulation of null distributions. We study asymptotic null distributions and prove consistency of our tests in a $p/n\rightarrow\gamma\in(0,\infty)$ asymptotic regime. The large power of our proposed tests relative to existing procedures is demonstrated numerically.