A0852
Title: A realized multi-factor regression model using multivariate realized stochastic volatility
Authors: Tsunehiro Ishihara - Takasaki City University of Economics (Japan) [presenting]
Abstract: Estimation of high-dimensional stochastic volatility models tends to be computationally expensive. A multivariate stochastic volatility model is proposed that can be computed in parallel. It takes reasonable computational time to estimate the parameters and conduct a prediction via MCMC. Realized covariance is computed from indices' high-frequency market, size, and value quasi-factors data. Using them, a time-varying coefficient regression model or a low-dimensional stochastic volatility model is estimated to forecast high-dimensional volatility. As an illustrative example, 33-dimensional Japanese sector indices data are applied.