Title: Smooth FREE COMFORT and LASSO for GARCH groups
Authors: Marc Paolella - University of Zurich (Switzerland) [presenting]
Abstract: GARCH-based multivariate models for asset returns can outperform their IID counterparts, but often only when not accounting for transaction costs, because of the high turnover induced by the ever-changing dispersion matrix. A setup is considered based on a smooth transition between IID and GARCH-based models, and presents a new method for the latter, termed Fast REduced Estimation (FREE), which cuts estimation time significantly with little or no performance degradation. A further enhancement is introduced that effectively clusters the univariate series into groups with the same GARCH structure. It is conducted using a new LASSO-based shrinkage paradigm, conditional on a filtered latent sequence of mixing random variables, and thus differs completely from related attempts at GARCH clustering.