Abstract

The heterogeneous autoregressive (HAR) model proposed by Corsi (2004) is known to perform well in volatility forecasting. This model formulates RV as a function of past RVs with different frequencies such as daily, weekly and monthly RVs. This paper extends the HAR model such that the coefficients of daily, weekly and monthly RVs and the error variance may change over time. The coefficients and the log of the error variance are assumed to follow first-order autoregressive processes. A Bayesian method using an efficient Markov chain Monte Carlo is developed for the analysis of the proposed model. An empirical application with the RV calculated using the 5-minute returns of the Nikkei 225 stock index is provided. This is a joint work with Jouchi Nakajima.