A0366
Title: HAR-Ito models and their high-dimensional statistical inference
Authors: Kexin Lu - University of Hong Kong (Hong Kong) [presenting]
Huiling Yuan - University of Hong Kong (Hong Kong)
Yifeng Guo - The University of Hong kong (China)
Guodong Li - University of Hong Kong (Hong Kong)
Abstract: Modelling realized volatilities for high-frequency data is an important task in finance and economics. The heterogeneous autoregressive (HAR) model is one of the most popular models used in this area. However, it has three limitations. Firstly, the linear combinations of daily realized volatilities with fixed weights limit its flexibility for different types of assets. Secondly, the high-frequency probabilistic structure of this model, as well as other HAR-type models in the literature, is still unknown. Thirdly, there is no high-dimensional inference tool available for HAR modelling, even though real applications often involve multiple assets. To address these issues, a multilinear low-rank HAR model is proposed using tensor techniques. A data-driven approach is adopted to select the heterogeneous components automatically. HAR-Ito models are also introduced to interpret the corresponding high-frequency dynamics of the proposed model and other HAR-type models. Theoretical properties of high-dimensional HAR modelling are established, and a projected gradient descent algorithm is suggested to search for estimates. The analysis is performed on real data to illustrate the performance of the proposed method.
