A1280
Title: Break detection in variance-covariance matrix
Authors: Benjamin Poignard - Keio University (Japan) [presenting]
Ying Lin - The University of Hong Kong (Hong Kong)
Abstract: The aim is to estimate a time-varying sparse covariance matrix that evolves in a piecewise constant manner. The proposed procedure is based on the adaptive version of the mixture Group Fused LASSO and LASSO penalties, applied to the squared Frobenius divergence. This approach allows us to estimate both the sparsity patterns and the change-points. We provide the conditions for the consistency of the estimated change-points and sparse estimators, and propose an ADMM-based algorithm to solve the optimization problem.
