Adaptive test for large covariance matrices with missing observations

We observe independent dimensional Gaussian vectors with missing coordinates, that is each value (which is assumed standardized) is observed with probability . We investigate the problem of minimax nonparametric testing that the high-dimensional covariance matrix of the underlying Gaussian distribution is the identity matrix, using these partially observed vectors. Here, and tend to infinity and tends to 0, asymptotically. We assume that belongs to a Sobolev-type ellipsoid with parameter . When is known, we give asymptotically minimax consistent test procedure and find the minimax separation rates , under some additional constraints on and . We show that, in the particular case of Toeplitz covariance matrices,the minimax separation rates are faster, . We note how the "missingness" parameter deteriorates the rates with respect to the case of fully observed vectors (). We also propose adaptive test procedures, that is free of the parameter in some interval, and show that the loss of rate is and for Toeplitz covariance matrices, respectively.
View on arXiv