Testing high-dimensional linear hypotheses through spectral shrinkage
Debashis Paul, University of California, Davis
Friday, January 31, 2020 - 3:30pm
We consider the problem of testing linear hypotheses associated with a high-dimensional linear model under the setting where the dimensionality of the response is comparable to the sample size. Classical likelihood ratio tests for such problems suffer from significant loss of power within this asymptotic framework. We propose regularization schemes that modify the likelihood ratio statistic by applying nonlinear shrinkage to the eigenvalues of the empirical covariance matrix of the residuals. We make the structural assumption that the spectral measure of the noise covariance converges to a nontrivial limit. We show that the proposed tests are able to significantly improve on the performance of the classical likelihood ratio test. We address the problem of finding the optimal regularization parameter within a decision-theoretic framework by adopting a probabilistic formulation of the alternatives.
(This is a joint work with Haoran Li, Alexander Aue, Jie Peng and Pei Wang).
Seminars generally take place in 116 Old Chemistry Building on Fridays from 3:30 - 4:30 pm. For additional information contact: firstname.lastname@example.org or phone 919-684-8029. Sorry, but we do not have reprints available. Please feel free to contact the authors by email for follow-up information, articles, etc. Reception following seminar in 203B Old Chemistry.