 |
High-dimensional penalized ARCH processes
1 : Centre de Recherche en Économie et Statistique (CREST)
* : Corresponding author
Centre de Recherche en Économie et STatistique (CREST)
We introduce several families of multidimensional ARCH models, possibly with
a very large number of parameters. The corresponding conditions of stationarity
and of positive deniteness are studied. Through penalized OLS methods (sparse
group-lasso), we consistently estimate such models. We evaluate the relevance of
such strategies by simulation.
