A0960
Title: Graphical copula GARCH modelling with dynamic conditional dependence
Authors: Shun Hin Chan - The Hong Kong University of Science and Technology (Hong Kong) [presenting]
Amanda Chu - The Education University of Hong Kong (China)
Mike So - The Hong Kong University of Science and Technology (Hong Kong)
Abstract: The aim is to develop a graphical copula GARCH model for volatility modelling. To allow high-dimensional modelling for large portfolios, the complexity of the modelling is greatly reduced by introducing conditional independence among stocks given the market risk factors, such as the S&P500 index in the United States. The market risk factors are modeled using a directed acyclic graph (DAG) model with a pairwise-copula construction to allow flexible distributional modelling. Using the DAG model gives a topological order to the market risk factors, which can be regarded as a list of directions of the flow of information or disturbance. The conditional distributions among stock returns are also modelled through pairwise-copula constructions for flexibility. Dynamic conditional dependence structures are adapted to allow the parameters in the copulas to be time-varying such that the tail dependence can dynamically be modelled between any two stocks. Three-stage estimation is used for estimating parameters in the marginal distributions, the copulas of the DAG of the market risk factors, and the copulas of the stocks. Bayesian inference is used to learn the structure of the DAG. The simulation study shows that these estimation procedures can be used to recover the parameters and the DAG accurately. With Bayesian inference, the structure of the market risk factors can be allowed to be random, and model averaging can be done to obtain robust volatility predictions.