A dynamic factor model with stylized facts to forecast volatility for an optimal portfolio

Abstract

In this paper, we have integrated numerous stylized facts, namely co-jumps, long memory, measurement error and leverage effects, in dynamic factor multivariate stochastic volatility (DfMSV) model for realized covariance measures using high frequency data. The aim is to examine the performance of these new models in reproducing characteristic features of foreign currency exchange series, which gather these stylized facts, and capturing the most of the information provided by financial time series. The dynamic conditional correlation model (DCC-DfMSV) combining Wishart autoregressive model (WAR) was applied to capture the selected stylized fact. The results proved that the DCC-DfMSV outperforms the asymmetric-DCC (ADCC), and fractionally integrated matrix-exponential-DCC (FIEDCC) models in forecasting future volatilities series. Along with the DfMSV model, the DCC-DfMSV combined with the overall stylized facts have proved to be better for forecasting portfolio volatility and providing better portfolio optimization, via efficient frontier, than the basic model.