Automated Variable Selection in Vector Multiplicative Error Models

Abstract

Multiplicative Error Models (MEM) can be used to trace the dynamics of non- negative valued processes. Interactions between several such processes are accommodated by the vector MEM and estimated by maximum likelihood (Gamma marginals with copula functions) or by Generalized Method of Moments. In choosing the relevant variables one can follow an automated procedure where the full specification is successively pruned in a general-to-specific approach. An efficient and fast algorithm is presented in this paper and evaluated by means of a simulation and a real world example of volatility spillovers in European markets.