Mixed-frequency SV model for stock volatility and macroeconomics

Abstract This paper develops a stochastic volatility-mixed frequency data sampling (SV-MIDAS) model with low frequency macro variables and further extends it to an asymmetric SV-MIDAS model. Empirical study is then implemented on both Chinese and U.S. stock markets. Our results show that the SV-MIDAS model is useful to identify the macroeconomic volatility source of stock volatility and improve the in-sample fitting performance. Moreover, the out-of-sample forecast performances of SV-MIDAS model are significantly superior to that of traditional SV model for both Chinese and U.S. stock markets. In particular, among the macroeconomic variables, the Composite Leading Indicator has the best forecast performance. In addition, we find that the asymmetric SV-MIDAS model is applicable for capturing leverage effects in both stock markets and it outperforms the corresponding benchmark model in the in-sample fitting.