B1024
Title: Hypothesis tests for structured rank correlation matrices
Authors: Johanna Neslehova - McGill University (Canada) [presenting]
Samuel Perreault - University of Toronto (Canada)
Thierry Duchesne - Laval University (Canada)
Abstract: Joint modeling of a large number of variables often requires dimension-reduction strategies. Many of them lead to a specific structure of the underlying correlation matrix, and model specification as well as validation calls for formal tests of such structural assumptions. Tests of the hypothesis are discussed that the entries of the Kendall rank correlation matrix are linear combinations of a smaller number of parameters. These tests will be validated through asymptotic arguments both when the dimension is fixed and when it grows with the sample size. Various scalable numerical strategies for the implementation of the proposed procedures and their behavior under local alternatives will also be presented. Simplifications and computational advantages are elaborated that lead to better performance of the tests in the special case of (partial) exchangeability. Finally, the proposed methodology is used to learn about dependence patterns in the sea levels at various locations along the North American coast.
