Date: 21 March 2018 (Wednesday)
Time: 2:00-4:00pm (2:00-3:00pm,talk and 3:10-4:00pm,discussion)
Speaker: Dr. Anita Wang
Language: English
Title: A dynamic matrix factor model for multivariate realized volatility

Abstract: Modeling multivariate volatility in moderately large dimension with a feasible dynamic structure remains a challenging problem. With the availability of intra-day data, recently study on daily realized covariance matrices constructed form high-frequency data have been receiving much interest.  The existing matrix factor (MFA) model transforms the realized covariance matrices to a lower dimensional subspace and reduce the number of parameters to be forecast. However, the loading matrix of MFA is constant over time. We propose a new method, the dynamic matrix factor (DMF) model, of which the loading matrix is time-varying. The realized covariance matrices are assumed to follow a Wishart distribution, and the scale matrix adopts a spectral decomposition. An loading-driving process with scalar BEKK model are used to capture the dynamics of loading matrix, and the diagonal matrix is modeled by GARCH(1,1). We will further illustrate DMF model with both simulation and real data. Empirical studies of small and medium-size of daily realized covariance matrices process from the New York Stock Exchange demonstrate the benefits of the dynamic structure of DMF, and show that DMF outperforms the existing methods.