Sharp minimax tests for large Toeplitz covariance matrices with repeated observations

Cristina Butucea, Rania Zgheib

Abstract: We observe a sample of n<math><mi is="true">n</mi></math> independent p<math><mi is="true">p</mi></math>-dimensional Gaussian vectors with Toeplitz covariance matrix Σ=[σ∣i−j∣]1≤i,j≤p<math><mi is="true">Σ</mi><mo is="true">=</mo><msub is="true"><mrow is="true"><mrow is="true"><mo is="true">[</mo><msub is="true"><mrow is="true"><mi is="true">σ</mi></mrow><mrow is="true"><mo is="true">∣</mo><mi is="true">i</mi><mo is="true">−</mo><mi is="true">j</mi><mo is="true">∣</mo></mrow></msub><mo is="true">]</mo></mrow></mrow><mrow is="true"><mn is="true">1</mn><mo is="true">≤</mo><mi is="true">i</mi><mo is="true">,</mo><mi is="true">j</mi><mo is="true">≤</mo><mi is="true">p</mi></mrow></msub></math> and σ0=1<math><msub is="true"><mrow is="true"><mi is="true">σ</mi></mrow><mrow is="true"><mn is="true">0</mn></mrow></msub><mo is="true">=</mo><mn is="true">1</mn></math>. We consider the problem of testing the hypothesis that Σ<math><mi is="true">Σ</mi></math> is the identity matrix asymptotically when n→∞<math><mi is="true">n</mi><mo is="true">→</mo><mi is="true">∞</mi></math> and p→∞<math><mi is="true">p</mi><mo is="true">→</mo><mi is="true">∞</mi></math>. We suppose that the covariances σk<math><msub is="true"><mrow is="true"><mi is="true">σ</mi></mrow><mrow is="true"><mi is="true">k</mi></mrow></msub></math> decrease either polynomially (∑k≥1k2ασk2≤L<math><msub is="true"><mrow is="true"><mo is="true">∑</mo></mrow><mrow is="true"><mi is="true">k</mi><mo is="true">≥</mo><mn is="true">1</mn></mrow></msub><msup is="true"><mrow is="true"><mi is="true">k</mi></mrow><mrow is="true"><mn is="true">2</mn><mi is="true">α</mi></mrow></msup><msubsup is="true"><mrow is="true"><mi is="true">σ</mi></mrow><mrow is="true"><mi is="true">k</mi></mrow><mrow is="true"><mn is="true">2</mn></mrow></msubsup><mo is="true">≤</mo><mi is="true">L</mi></math> for α>1/4<math><mi is="true">α</mi><mo is="true">&gt;</mo><mn is="true">1</mn><mo is="true">/</mo><mn is="true">4</mn></math> and L>0<math><mi is="true">L</mi><mo is="true">&gt;</mo><mn is="true">0</mn></math>) or exponentially (∑k≥1e2Akσk2≤L<math><msub is="true"><mrow is="true"><mo is="true">∑</mo></mrow><mrow is="true"><mi is="true">k</mi><mo is="true">≥</mo><mn is="true">1</mn></mrow></msub><msup is="true"><mrow is="true"><mi is="true">e</mi></mrow><mrow is="true"><mn is="true">2</mn><mi is="true">A</mi><mi is="true">k</mi></mrow></msup><msubsup is="true"><mrow is="true"><mi is="true">σ</mi></mrow><mrow is="true"><mi is="true">k</mi></mrow><mrow is="true"><mn is="true">2</mn></mrow></msubsup><mo is="true">≤</mo><mi is="true">L</mi></math> for A,L>0<math><mi is="true">A</mi><mo is="true">,</mo><mi is="true">L</mi><mo is="true">&gt;</mo><mn is="true">0</mn></math>).We consider a test procedure based on a weighted U-statistic of order 2, with optimal weights chosen as solution of an extremal problem. We give the asymptotic normality of the test statistic under the null hypothesis for fixed n<math><mi is="true">n</mi></math> and p→+∞<math><mi is="true">p</mi><mo is="true">→</mo><mo is="true">+</mo><mi is="true">∞</mi></math> and the asymptotic behavior of the type I error probability of our test procedure. We also show that the maximal type II error probability, either tend to 0<math><mn is="true">0</mn></math>, or is bounded from above. In the latter case, the upper bound is given using the asymptotic normality of our test statistic under alternatives close to the separation boundary. Our assumptions imply mild conditions: n=o(p2α−1/2)<math><mi is="true">n</mi><mo is="true">=</mo><mi is="true">o</mi><mrow is="true"><mo is="true">(</mo><msup is="true"><mrow is="true"><mi is="true">p</mi></mrow><mrow is="true"><mn is="true">2</mn><mi is="true">α</mi><mo is="true">−</mo><mn is="true">1</mn><mo is="true">/</mo><mn is="true">2</mn></mrow></msup><mo is="true">)</mo></mrow></math> (in the polynomial case), n=o(ep)<math><mi is="true">n</mi><mo is="true">=</mo><mi is="true">o</mi><mrow is="true"><mo is="true">(</mo><msup is="true"><mrow is="true"><mi is="true">e</mi></mrow><mrow is="true"><mi is="true">p</mi></mrow></msup><mo is="true">)</mo></mrow></math> (in the exponential case).We prove both rate optimality and sharp optimality of our results, for α>1<math><mi is="true">α</mi><mo is="true">&gt;</mo><mn is="true">1</mn></math> in the polynomial case and for any A>0<math><mi is="true">A</mi><mo is="true">&gt;</mo><mn is="true">0</mn></math> in the exponential case.A simulation study illustrates the good behavior of our procedure, in particular for small n<math><mi is="true">n</mi></math>, large p<math><mi is="true">p</mi></math>.

External IDs:dblp:journals/ma/ButuceaZ16