A0295
Title: A realized multi-factor regression using a multivariate stochastic volatility model
Authors: Tsunehiro Ishihara - Takasaki City University of Economics (Japan) [presenting]
Abstract: A multifactor model is a standard tool in financial econometrics. We introduce information from high-frequency data into the multifactor model using realized measures. We calculate market, size, value quasi-factors, and their realized covariance matrix from intra-daily announced market indices. We propose a time-varying coefficient factor regression model and transform it into the multivariate stochastic volatility model with realized covariance. Bayesian estimation using the Markov chain Monte Carlo method is also proposed. We present empirical illustrations using several sector indices of the Japanese stock market. It is shown that the proposed factors behave similarly to the Fama-French three factors. In addition, the coefficients of the factors were found to change over time. The model is also compared with the realized stochastic volatility model without the factor. The Bayesian predictive likelihood loss function and robust loss function for volatility forecasting show that factor models perform well, for longer forecast horizons.
