A1981
Title: Fast variational approximation for multivariate factor stochastic volatility models
Authors: David Gunawan - University of Wollongong (Australia) [presenting]
David Nott - National University of Singapore (Singapore)
Robert Kohn - University of New South Wales (Australia)
Matias Quiroz - University of Technology Sydney (Australia)
Abstract: Estimating and predicting the densities and time-varying correlation matrices for high dimensional time series are important and an active area of research. One of the main challenges is the curse of dimensionality, where the number of parameters in the covariance matrix grows quadratically with the number of time series considered. The factor SV model is an important multivariate time series model for parsimoniously modeling a vector of time series. It imposes a much lower dimensional latent factors that are allowed to exhibit stochastic volatility and govern the comovement of the time series over time. One of the main challenges of estimating this model is that it has both large number of latent states and large number of parameters. This model is usually estimated by using MCMC or particle MCMC method, which can be slow for high dimensional and long time series. Fast sequential and batch variational estimation methods are proposed to approximate the posterior distribution of the states and parameters in a multivariate factor stochastic volatility (SV) model and obtain one-step and multiple step-ahead forecast distribution. We apply our method to simulated and real datasets.
