Constructing a Realized DCC Model with Measurement Errors

Abstract

We propose novel conditional autoregressive Wishart (CAW) models for high-dimensional realized covariance matrices of asset returns. We incorporate measurement errors into realized variance dynamics of Realized Dynamic Conditional Correlation (Re-DCC) model. It is well known that the measurement errors make the realized volatility less persistent than the latent volatility process. Therefore, by introducing measurement errors into realized volatility, the persistence of the realized volatility based on the magnitude of the corresponding measurement errors can be incorporated into the multivariate model. Our empirical analysis performs in- and out-of-sample evaluations for 100 stocks on the Tokyo Stock Exchange from January 1, 2014, through December 31, 2020. Our model based on logarithmic realized volatility shows the best forecast performance across test periods and loss functions.