A0711
Title: Penalized empirical likelihood estimation for low-rank vector autoregressive models
Authors: Fumiya Akashi - University of Tokyo (Japan) [presenting]
Abstract: In real data analysis, heavy-tailed data are often encountered, making it unsuitable to use classical statistical methods directly. On the other hand, fully unrestricted models can be redundant in practice, as some data exhibit low-rank structures. To address these issues, an estimation problem is considered for vector autoregressive models with possibly infinite variance innovations under a low-rank structure on the coefficient matrices. A spatial median-based, self-weighted, and penalized empirical likelihood estimator is proposed, and its asymptotic distribution is derived. A smoothing technique is also incorporated to overcome a typical problem appearing in the context of spatial median regression. The proposed estimator is shown to be more efficient than the conventional spatial median-based estimator. Moreover, the oracle property of the penalized version is established. Some simulation studies demonstrate the advantages of the method.
