A0337
Title: Bayesian variable selection in nonlinear realized HAR-GARCH models for financial markets
Authors: Feng-Chi Liu - Feng Chia University (Taiwan) [presenting]
Yue-Sheng Wu - Feng Chia University (Taiwan)
Abstract: Based on high-frequency intraday data availability, realized volatility (RV) is widely applied to topics in financial modeling. The realized GARCH model is a leading example for integrating the RV into a GARCH model. Furthermore, the realized heterogeneous autoregressive GARCH (realized HAR-GARCH) model extends the realized GARCH model to consider the short- and long-term effects of the RVs in volatility modeling. The first contribution is to consider the spillover effects of external markets by integrating the daily, weekly, and monthly RVs from external markets as exogenous variables in the realized HAR-GARCH model. The second contribution is that a threshold structure is employed for the proposed model to capture asymmetric reactions of financial data. Finally, the issue of variable selection is conducted to identify important variables for the realized HAR-GARCH model of the target market as the third contribution. Under the Bayesian framework, an adaptive Markov chain Monte Carlo sampling method is designed to estimate model parameters and select important variables simultaneously. The simulation study demonstrates that the proposed method can provide accurate parameter estimations and reasonable results of variable selection. In a real example study, there are three stock market indices investigated to examine the spillover effects of the S\&P 500 index. The results can provide more insight into the model building of the short- or long-term effects of RVs.
