EcoSta 2018: Registration
View Submission - EcoSta2018
A0270
Title: Factor correlation matrix modelling of large-dimensional portfolio with high-frequency data Authors:  Yingjie Dong - University of International Business and Economics (China) [presenting]
Yiu-Kuen Tse - Singapore Management University (Singapore)
Abstract: A factor correlation matrix approach is proposed to model high-dimensional realized covariance matrix using high-frequency data. We assume the high-dimensional daily realized correlation matrix is driven by a low-dimensional latent process, which is modelled using the principal component method. We adopt the vech representation for the low-dimensional latent process over time. In addition, the return variances are estimated by imposing a long memory structure on the realized volatilities. We conduct Monte Carlo studies to compare the finite sample performance of different methods of estimating the high-dimensional covariances. Our new method is found to perform better by reporting smaller estimation errors. In addition, our empirical studies show that our method provides lower variance in selecting minimum-variance portfolios.