Bitcoin volatility forecasting with the HAR-type models

We aim to forecast the volatility of bitcoin using high-frequency data in this paper. Based on the HAR-RV model, we construct extensive HAR-type models by including various jump components and analysis their in-sample regression fitness and out-of-sample forecast abilities. We verify the importance...

Full description

Bibliographic Details
Main Author: Zhang, Jiawei
Format: Dissertation (University of Nottingham only)
Language:English
Published: 2019
Online Access:https://eprints.nottingham.ac.uk/57681/
Description
Summary:We aim to forecast the volatility of bitcoin using high-frequency data in this paper. Based on the HAR-RV model, we construct extensive HAR-type models by including various jump components and analysis their in-sample regression fitness and out-of-sample forecast abilities. We verify the importance of the jump component in the in-sample regression analysis. The out-of-sample forecast shows that the HAR-RV-SJ and HAR-RV-J models have the best performance. Overall, our result indicates that the HAR-type models have an advantage at mimicking the long memory and using the jump component is significant to the volatility forecast both in-sample and out-of-sample. Keywords: bitcoin; realized variance; HAR; signed jumps; volatility forecasting; realized semivariance