SEMIPARAMETRIC VECTOR MEM

Abstract

SUMMARY Financial time series are often non‐negative‐valued (volumes, trades, durations, realized volatility, daily range) and exhibit clustering. When joint dynamics is of interest, the vector multiplicative error model (vMEM; the element‐by‐element product of a vector of conditionally autoregressive scale factors and a multivariate i.i.d. innovation process) is a suitable strategy. Its parameters can be estimated by generalized method of moments, bypassing the problem of specifying a multivariate distribution for the errors. Simulated results show the gains in efficiency relative to an equation‐by‐equation approach. A vMEM on several measures of volatility justifies a joint approach revealing full interdependence. Copyright © 2012 John Wiley & Sons, Ltd.