Abstract

Realized volatility, which is the sum of squared intraday returns over a day, is known to be subject to the bias caused by microstructure noise and non-trading hours and some methods such as realized kernel have been proposed for mitigating the bias in realized volatility. Hansen et al. (2012) have proposed a realized GARCH model where they model daily return and realized volatility jointly taking account of the bias in realized volatility. This article extends the realized GARCH model incorporating with a smooth transition model where the mean of the true volatility may change following a logistic transition function of time. The simple and extended realized GARCH models are fitted to the daily return and realized volatility or realized kernel of Nikkei 225 stock index using the maximum likelihood method and the results are compared.