Volatility forecasting with realized measures:

The use of estimators of volatility based on high-frequency data has greatly improved our ability to measure and model financial market volatility. Over the past few years there has been an abundance of research on the subject of modeling return volatility. This paper compares two of the more successful models, being the HEAVY and HAR models. The HEAVY model framework as developed in Shephard and Sheppard (2009) uses two separate equations and realized measures to model the volatility. The HAR model introduced by Corsi (2009) uses an additive cascade model of volatility components defined over different time periods. This paper also uses ideas from Patton and Sheppard (2009b) for the implementation of the HAR model. To compare the models we use a predictive ability test and compare their forecast performance at different horizons. Results of these tests show that the HAR models outperform HEAVY models when using a lot of observations to estimate. But we see the opposite when using an estimation window of four years, in that case the HEAVY models perform better at almost every horizon.