統計学輪講(第25回) 日時 2014年01月14日(火) 15時40分~16時30分 場所 経済学部新棟3階第3教室 講演者 朴 海香 (経済M2) 演題 Realized Cholesky Stochastic Volatility Model 概要 Multivariate stochastic volatility has many important applications in finance, including asset pricing and risk management, but it is not easy to estimate. One of the reasons is because for $p$ time series, there are $p(p+1)/2$ elements we need to estimate. The second reason is that the covariance matrix must to be positive definite for all time point. In this paper, we use Cholesky decomposition to solve the positiveness problem. First, we decompose covariance matrix by Cholesky decomposition to ensure the positivity of the covariance matrix. Second, we introduce realized covariance matrix and construct the Realized Cholesky Stochastic Volatility (RCSV) model to make model more accurate. Moreover, we propose priors and we estimate variables and parameters by Bayesian approach. At the end of the paper, we apply our approach to stocks' price data and construct portfolios by using the estimated parameters. We compare the variance of our portfolio returns with that of a simple model that ignores the correlation between stocks.