A Latent Factor Model of Multivariate Conditional Heteroscedasticity

Abstract

This paper examines the joint dynamics of a system of asset returns by describing and implementing a factor multivariate stochastic volatility (factor MSV) model. The foundation for the model discussed here is the work of Doz and Renault (2006). Despite its attractive design, that model has not been adopted widely in the literature, most likely due to the difficulty encountered in its implementation. The main contribution of this paper is to illustrate that this factor MSV model can be implemented easily and with only a few modifications. Specifically, I develop a sequential testing procedure that can account simultaneously for a series of nested hypotheses and structure properly the moment conditions used for estimation. A simulation study suggests that the factor MSV model and estimation strategy presented here is able to recover accurately the number of, and dynamics for, the latent factors that drive the conditional volatility of returns.

Keywords: common features, conditional factor models, generalized method of moments, multivariate conditional heteroscedasticity, stochastic volatility

JEL Classification: C12, C13, C14, C32, C51, G12