A0393
Title: Multivariate realized stochastic volatility model using time varying coefficient characteristic factor regression
Authors: Tsunehiro Ishihara - Takasaki City University of Economics (Japan) [presenting]
Abstract: The computational cost to estimate a high-dimensional time-varying correlation volatility model is often expensive. A multivariate stochastic volatility model is proposed with observed characteristic factors and their realized covariance. To reduce the computational time, the high-dimensional model is split into conditional univariate models and low-dimensional characteristic factor multivariate models and estimated in parallel. For conditional univariate models, the time-varying coefficient characteristic factor regression model is proposed with stochastic volatility, and their realized measurements are introduced into the model. For the low-dimensional multivariate stochastic volatility model for characteristic factors, the matrix exponential realized stochastic volatility model is used. As an illustrative example, the model to Japanese sector indices and market are applied, as well as value and size factors. Model comparison is conducted with other multivariate models.
